Update version.h to 7.10.2

Summary:
Fix a bug in the calculation of the input buffer address/offset in log_reader.cc. The bug is when consecutive fragments of a compressed record are located at the same offset in the log reader buffer, the second fragment input buffer is treated as a leftover from the previous input buffer. As a result, the offset in the `ZSTD_inBuffer` is not reset.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/11198

Test Plan: Add a unit test in log_test.cc that fails without the fix and passes with it.

Reviewed By: ajkr, cbi42

Differential Revision: D43102692

Pulled By: anand1976

fbshipit-source-id: aa2648f4802c33991b76a3233c5a58d4cc9e77fd

.circleci/config.yml

@@ -1,7 +1,7 @@
 version: 2.1
 
 orbs:
-  win: circleci/windows@2.4.0
+  win: circleci/windows@5.0.0
 
 commands:
   install-cmake-on-macos:
@@ -45,6 +45,38 @@ commands:
             echo "export LZ4_DOWNLOAD_BASE=https://rocksdb-deps.s3.us-west-2.amazonaws.com/pkgs/lz4" >> $BASH_ENV
             echo "export ZSTD_DOWNLOAD_BASE=https://rocksdb-deps.s3.us-west-2.amazonaws.com/pkgs/zstd" >> $BASH_ENV
 
+  windows-build-steps:
+    steps:
+      - checkout
+      - run:
+          name: "Install thirdparty dependencies"
+          command: |
+            echo "Installing CMake..."
+            choco install cmake --installargs 'ADD_CMAKE_TO_PATH=System' -y
+            mkdir $Env:THIRDPARTY_HOME
+            cd $Env:THIRDPARTY_HOME
+            echo "Building Snappy dependency..."
+            curl https://github.com/google/snappy/archive/refs/tags/1.1.8.zip -O snappy-1.1.8.zip
+            unzip -q snappy-1.1.8.zip
+            cd snappy-1.1.8
+            mkdir build
+            cd build
+            & $Env:CMAKE_BIN -G "$Env:CMAKE_GENERATOR" ..
+            msbuild.exe Snappy.sln -maxCpuCount -property:Configuration=Debug -property:Platform=x64
+      - run:
+          name: "Build RocksDB"
+          command: |
+            mkdir build
+            cd build
+            & $Env:CMAKE_BIN -G "$Env:CMAKE_GENERATOR" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DJNI=1 ..
+            cd ..
+            echo "Building with VS version: $Env:CMAKE_GENERATOR"
+            msbuild.exe build/rocksdb.sln -maxCpuCount -property:Configuration=Debug -property:Platform=x64
+      - run:
+          name: "Test RocksDB"
+          shell: powershell.exe
+          command: |
+            build_tools\run_ci_db_test.ps1 -SuiteRun arena_test,db_basic_test,db_test,db_test2,db_merge_operand_test,bloom_test,c_test,coding_test,crc32c_test,dynamic_bloom_test,env_basic_test,env_test,hash_test,random_test -Concurrency 16
   pre-steps-macos:
       steps:
         - pre-steps
@@ -72,6 +104,14 @@ commands:
           path: /tmp/core_dumps
           when: on_fail
 
+  upgrade-cmake:
+    steps:
+      - run:
+          name: Upgrade cmake
+          command: |
+            sudo apt remove --purge cmake
+            sudo snap install cmake --classic
+
   install-gflags:
     steps:
       - run:
@@ -93,6 +133,13 @@ commands:
           command: |
             make checkout_folly
 
+  build-folly:
+    steps:
+      - run:
+          name: Build folly and dependencies
+          command: |
+            make build_folly
+
   build-for-benchmarks:
     steps:
       - pre-steps
@@ -109,10 +156,19 @@ commands:
           environment:
             LD_LIBRARY_PATH: /usr/local/lib
             # How long to run parts of the test(s)
-            DURATION_RO: 450
-            DURATION_RW: 450
+            DURATION_RO: 400
+            DURATION_RW: 700
+            # Keep threads within physical capacity of server (much lower than default)
+            NUM_THREADS: 1
+            MAX_BACKGROUND_JOBS: 3
+            # Don't run a couple of "optional" initial tests
+            CI_TESTS_ONLY: "true"
+            # Reduce configured size of levels to ensure more levels in the leveled compaction LSM tree
+            WRITE_BUFFER_SIZE_MB: 16
+            TARGET_FILE_SIZE_BASE_MB: 16
+            MAX_BYTES_FOR_LEVEL_BASE_MB: 64
             # The benchmark host has 32GB memory
-            # The folllowing values are tailored to work with that
+            # The following values are tailored to work with that
             # Note, tests may not exercise the targeted issues if the memory is increased on new test hosts.
 
 
@@ -130,11 +186,6 @@ commands:
             true
 
 executors:
-  windows-2xlarge:
-    machine:
-      image: 'windows-server-2019-vs2019:stable'
-      resource_class: windows.2xlarge
-      shell: bash.exe
   linux-docker:
     docker:
       # The image configuration is build_tools/ubuntu20_image/Dockerfile
@@ -149,7 +200,7 @@ executors:
       # to run the CI image environment locally:
       #  $ docker run --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -it zjay437/rocksdb:0.5 bash
       # option `--cap-add=SYS_PTRACE --security-opt seccomp=unconfined` is used to enable gdb to attach an existing process
-      - image: zjay437/rocksdb:0.5
+      - image: zjay437/rocksdb:0.6
 
 jobs:
   build-macos:
@@ -215,7 +266,7 @@ jobs:
       - pre-steps
       - run: ENCRYPTED_ENV=1 ROCKSDB_DISABLE_SNAPPY=1 ROCKSDB_DISABLE_ZLIB=1 ROCKSDB_DISABLE_BZIP=1 ROCKSDB_DISABLE_LZ4=1 ROCKSDB_DISABLE_ZSTD=1 make V=1 J=32 -j32 check
       - run: |
-          ./sst_dump --help | egrep -q 'Supported compression types: kNoCompression$' # Verify no compiled in compression
+          ./sst_dump --help | grep -E -q 'Supported compression types: kNoCompression$' # Verify no compiled in compression
       - post-steps
 
   build-linux-shared_lib-alt_namespace-status_checked:
@@ -346,7 +397,17 @@ jobs:
     steps:
       - pre-steps
       - setup-folly
-      - run: (mkdir build && cd build && cmake -DUSE_FOLLY=1 -DWITH_GFLAGS=1 .. && make V=1 -j20 && ctest -j20)
+      - build-folly
+      - run: (mkdir build && cd build && cmake -DUSE_FOLLY=1 -DWITH_GFLAGS=1 -DROCKSDB_BUILD_SHARED=0 .. && make V=1 -j20 && ctest -j20)
+      - post-steps
+
+  build-linux-cmake-with-folly-lite-no-test:
+    executor: linux-docker
+    resource_class: 2xlarge
+    steps:
+      - pre-steps
+      - setup-folly
+      - run: (mkdir build && cd build && cmake -DUSE_FOLLY_LITE=1 -DWITH_GFLAGS=1 .. && make V=1 -j20)
       - post-steps
 
   build-linux-cmake-with-benchmark:
@@ -376,9 +437,19 @@ jobs:
     steps:
       - pre-steps
       - setup-folly
+      - build-folly
       - run: USE_FOLLY=1 CC=gcc-7 CXX=g++-7 V=1 make -j32 check
       - post-steps
 
+  build-linux-gcc-7-with-folly-lite-no-test:
+    executor: linux-docker
+    resource_class: 2xlarge
+    steps:
+      - pre-steps
+      - setup-folly
+      - run: USE_FOLLY_LITE=1 CC=gcc-7 CXX=g++-7 V=1 make -j32 all
+      - post-steps
+
   build-linux-gcc-8-no_test_run:
     executor: linux-docker
     resource_class: 2xlarge
@@ -387,6 +458,19 @@ jobs:
       - run: CC=gcc-8 CXX=g++-8 V=1 make -j32 all
       - post-steps
 
+  build-linux-cmake-with-folly-coroutines:
+    executor: linux-docker
+    resource_class: 2xlarge
+    environment:
+      CC: gcc-10
+      CXX: g++-10
+    steps:
+      - pre-steps
+      - setup-folly
+      - build-folly
+      - run: (mkdir build && cd build && cmake -DUSE_COROUTINES=1 -DWITH_GFLAGS=1 -DROCKSDB_BUILD_SHARED=0 .. && make V=1 -j20 && ctest -j20)
+      - post-steps
+
   build-linux-gcc-10-cxx20-no_test_run:
     executor: linux-docker
     resource_class: 2xlarge
@@ -418,6 +502,7 @@ jobs:
     steps:
       - pre-steps
       - setup-folly
+      - build-folly
       - run: CC=clang-13 CXX=clang++-13 USE_CLANG=1 USE_FOLLY=1 COMPILE_WITH_UBSAN=1 COMPILE_WITH_ASAN=1 make -j32 check
       - post-steps
 
@@ -435,7 +520,7 @@ jobs:
     resource_class: large
     steps:
       - pre-steps
-      - run: ulimit -S -n `ulimit -H -n` && make V=1 -j8 CRASH_TEST_EXT_ARGS=--duration=960 blackbox_crash_test_with_atomic_flush
+      - run: ulimit -S -n `ulimit -H -n` && make V=1 -j8 CRASH_TEST_EXT_ARGS='--duration=960 --max_key=2500000' blackbox_crash_test_with_atomic_flush
       - post-steps
 
   build-linux-crashtest-tiered-storage-bb:
@@ -460,73 +545,35 @@ jobs:
           no_output_timeout: 100m
       - post-steps
 
-  build-windows:
-    executor: windows-2xlarge
-    parameters:
-      extra_cmake_opt:
-        default: ""
-        type: string
-      vs_year:
-        default: "2019"
-        type: string
-      cmake_generator:
-        default: "Visual Studio 16 2019"
-        type: string
+  build-windows-vs2022:
+    executor:
+      name: win/server-2022
+      size: 2xlarge
     environment:
       THIRDPARTY_HOME: C:/Users/circleci/thirdparty
-      CMAKE_HOME: C:/Users/circleci/thirdparty/cmake-3.16.4-win64-x64
-      CMAKE_BIN: C:/Users/circleci/thirdparty/cmake-3.16.4-win64-x64/bin/cmake.exe
-      SNAPPY_HOME: C:/Users/circleci/thirdparty/snappy-1.1.7
-      SNAPPY_INCLUDE: C:/Users/circleci/thirdparty/snappy-1.1.7;C:/Users/circleci/thirdparty/snappy-1.1.7/build
-      SNAPPY_LIB_DEBUG: C:/Users/circleci/thirdparty/snappy-1.1.7/build/Debug/snappy.lib
-      VS_YEAR: <<parameters.vs_year>>
-      CMAKE_GENERATOR: <<parameters.cmake_generator>>
+      CMAKE_HOME: C:/Program Files/CMake
+      CMAKE_BIN: C:/Program Files/CMake/bin/cmake.exe
+      SNAPPY_HOME: C:/Users/circleci/thirdparty/snappy-1.1.8
+      SNAPPY_INCLUDE: C:/Users/circleci/thirdparty/snappy-1.1.8;C:/Users/circleci/thirdparty/snappy-1.1.8/build
+      SNAPPY_LIB_DEBUG: C:/Users/circleci/thirdparty/snappy-1.1.8/build/Debug/snappy.lib
+      CMAKE_GENERATOR: Visual Studio 17 2022
     steps:
-      - checkout
-      - run:
-          name: "Setup VS"
-          command: |
-            if [[ "${VS_YEAR}" == "2019" ]]; then
-              echo "VS2019 already present."
-            elif [[ "${VS_YEAR}" == "2017" ]]; then
-              echo "Installing VS2017..."
-              powershell .circleci/vs2017_install.ps1
-            elif [[ "${VS_YEAR}" == "2015" ]]; then
-              echo "Installing VS2015..."
-              powershell .circleci/vs2015_install.ps1
-            fi
-      - store_artifacts:
-          path: \Users\circleci\AppData\Local\Temp\vslogs.zip
-      - run:
-          name: "Install thirdparty dependencies"
-          command: |
-            mkdir ${THIRDPARTY_HOME}
-            cd ${THIRDPARTY_HOME}
-            echo "Installing CMake..."
-            curl --fail --silent --show-error --output cmake-3.16.4-win64-x64.zip --location https://github.com/Kitware/CMake/releases/download/v3.16.4/cmake-3.16.4-win64-x64.zip
-            unzip -q cmake-3.16.4-win64-x64.zip
-            echo "Building Snappy dependency..."
-            curl --fail --silent --show-error --output snappy-1.1.7.zip --location https://github.com/google/snappy/archive/1.1.7.zip
-            unzip -q snappy-1.1.7.zip
-            cd snappy-1.1.7
-            mkdir build
-            cd build
-            ${CMAKE_BIN} -G "${CMAKE_GENERATOR}" ..
-            msbuild.exe Snappy.sln -maxCpuCount -property:Configuration=Debug -property:Platform=x64
-      - run:
-          name: "Build RocksDB"
-          command: |
-            mkdir build
-            cd build
-            ${CMAKE_BIN} -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DJNI=1 << parameters.extra_cmake_opt >> ..
-            cd ..
-            echo "Building with VS version: ${CMAKE_GENERATOR}"
-            msbuild.exe build/rocksdb.sln -maxCpuCount -property:Configuration=Debug -property:Platform=x64
-      - run:
-          name: "Test RocksDB"
-          shell: powershell.exe
-          command: |
-            build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test,db_test2,db_merge_operand_test,bloom_test,c_test,coding_test,crc32c_test,dynamic_bloom_test,env_basic_test,env_test,hash_test,random_test -Concurrency 16
+      - windows-build-steps
+
+  build-windows-vs2019:
+    executor:
+      name: win/server-2019
+      size: 2xlarge
+    environment:
+      THIRDPARTY_HOME: C:/Users/circleci/thirdparty
+      CMAKE_HOME: C:/Program Files/CMake
+      CMAKE_BIN: C:/Program Files/CMake/bin/cmake.exe
+      SNAPPY_HOME: C:/Users/circleci/thirdparty/snappy-1.1.8
+      SNAPPY_INCLUDE: C:/Users/circleci/thirdparty/snappy-1.1.8;C:/Users/circleci/thirdparty/snappy-1.1.8/build
+      SNAPPY_LIB_DEBUG: C:/Users/circleci/thirdparty/snappy-1.1.8/build/Debug/snappy.lib
+      CMAKE_GENERATOR: Visual Studio 16 2019
+    steps:
+      - windows-build-steps
 
   build-linux-java:
     executor: linux-docker
@@ -584,6 +631,7 @@ jobs:
       - run:
           name: "Test RocksDBJava"
           command: make V=1 J=16 -j16 jtest
+          no_output_timeout: 20m
       - post-steps
 
   build-macos-java-static:
@@ -608,6 +656,7 @@ jobs:
       - run:
           name: "Build RocksDBJava x86 and ARM Static Libraries"
           command: make V=1 J=16 -j16 rocksdbjavastaticosx
+          no_output_timeout: 20m
       - post-steps
 
   build-macos-java-static-universal:
@@ -632,6 +681,7 @@ jobs:
       - run:
           name: "Build RocksDBJava Universal Binary Static Library"
           command: make V=1 J=16 -j16 rocksdbjavastaticosx_ub
+          no_output_timeout: 20m
       - post-steps
 
   build-examples:
@@ -765,7 +815,10 @@ workflows:
     jobs:
       - build-linux
       - build-linux-cmake-with-folly
+      - build-linux-cmake-with-folly-lite-no-test
       - build-linux-gcc-7-with-folly
+      - build-linux-gcc-7-with-folly-lite-no-test
+      - build-linux-cmake-with-folly-coroutines
       - build-linux-cmake-with-benchmark
       - build-linux-encrypted_env-no_compression
       - build-linux-lite
@@ -795,15 +848,8 @@ workflows:
       - build-linux-mini-crashtest
   jobs-windows:
     jobs:
-      - build-windows:
-          name: "build-windows-vs2019"
-      - build-windows:
-          name: "build-windows-vs2019-cxx20"
-          extra_cmake_opt: -DCMAKE_CXX_STANDARD=20
-      - build-windows:
-          name: "build-windows-vs2017"
-          vs_year: "2017"
-          cmake_generator: "Visual Studio 15 Win64"
+      - build-windows-vs2022
+      - build-windows-vs2019
       - build-cmake-mingw
   jobs-java:
     jobs:

.circleci/vs2015_install.ps1

@@ -1,24 +0,0 @@
-$VS_DOWNLOAD_LINK = "https://go.microsoft.com/fwlink/?LinkId=691126"
-$COLLECT_DOWNLOAD_LINK = "https://aka.ms/vscollect.exe"
-curl.exe --retry 3 -kL $VS_DOWNLOAD_LINK --output vs_installer.exe
-if ($LASTEXITCODE -ne 0) {
-    echo "Download of the VS 2015 installer failed"
-    exit 1
-}
-$VS_INSTALL_ARGS = @("/Quiet", "/NoRestart")
-$process = Start-Process "${PWD}\vs_installer.exe" -ArgumentList $VS_INSTALL_ARGS -NoNewWindow -Wait -PassThru
-Remove-Item -Path vs_installer.exe -Force
-$exitCode = $process.ExitCode
-if (($exitCode -ne 0) -and ($exitCode -ne 3010)) {
-    echo "VS 2015 installer exited with code $exitCode, which should be one of [0, 3010]."
-    curl.exe --retry 3 -kL $COLLECT_DOWNLOAD_LINK --output Collect.exe
-    if ($LASTEXITCODE -ne 0) {
-        echo "Download of the VS Collect tool failed."
-        exit 1
-    }
-    Start-Process "${PWD}\Collect.exe" -NoNewWindow -Wait -PassThru
-    New-Item -Path "C:\w\build-results" -ItemType "directory" -Force
-    Copy-Item -Path "C:\Users\circleci\AppData\Local\Temp\vslogs.zip" -Destination "C:\w\build-results\"
-    exit 1
-}
-echo "VS 2015 installed."

.circleci/vs2017_install.ps1

@@ -1,35 +0,0 @@
-$VS_DOWNLOAD_LINK = "https://aka.ms/vs/15/release/vs_buildtools.exe"
-$COLLECT_DOWNLOAD_LINK = "https://aka.ms/vscollect.exe"
-$VS_INSTALL_ARGS = @("--nocache","--quiet","--wait", "--add Microsoft.VisualStudio.Workload.VCTools",
-                                                     "--add Microsoft.VisualStudio.Component.VC.Tools.14.13",
-                                                     "--add Microsoft.Component.MSBuild",
-                                                     "--add Microsoft.VisualStudio.Component.Roslyn.Compiler",
-                                                     "--add Microsoft.VisualStudio.Component.TextTemplating",
-                                                     "--add Microsoft.VisualStudio.Component.VC.CoreIde",
-                                                     "--add Microsoft.VisualStudio.Component.VC.Redist.14.Latest",
-                                                     "--add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Core",
-                                                     "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64",
-                                                     "--add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Win81")
-
-curl.exe --retry 3 -kL $VS_DOWNLOAD_LINK --output vs_installer.exe
-if ($LASTEXITCODE -ne 0) {
-    echo "Download of the VS 2017 installer failed"
-    exit 1
-}
-
-$process = Start-Process "${PWD}\vs_installer.exe" -ArgumentList $VS_INSTALL_ARGS -NoNewWindow -Wait -PassThru
-Remove-Item -Path vs_installer.exe -Force
-$exitCode = $process.ExitCode
-if (($exitCode -ne 0) -and ($exitCode -ne 3010)) {
-    echo "VS 2017 installer exited with code $exitCode, which should be one of [0, 3010]."
-    curl.exe --retry 3 -kL $COLLECT_DOWNLOAD_LINK --output Collect.exe
-    if ($LASTEXITCODE -ne 0) {
-        echo "Download of the VS Collect tool failed."
-        exit 1
-    }
-    Start-Process "${PWD}\Collect.exe" -NoNewWindow -Wait -PassThru
-    New-Item -Path "C:\w\build-results" -ItemType "directory" -Force
-    Copy-Item -Path "C:\Users\circleci\AppData\Local\Temp\vslogs.zip" -Destination "C:\w\build-results\"
-    exit 1
-}
-echo "VS 2017 installed."

.github/workflows/sanity_check.yml

@@ -1,5 +1,8 @@
 name: Check buck targets and code format
 on: [push, pull_request]
+permissions:
+  contents: read
+
 jobs:
   check:
     name: Check TARGETS file and code format

.gitignore

@@ -56,6 +56,7 @@ io_tracer_parser
 .DS_Store
 .vs
 .vscode
+.clangd
 
 java/out
 java/target

CMakeLists.txt

@@ -74,6 +74,7 @@ option(WITH_WINDOWS_UTF8_FILENAMES "use UTF8 as characterset for opening files,
 if (WITH_WINDOWS_UTF8_FILENAMES)
   add_definitions(-DROCKSDB_WINDOWS_UTF8_FILENAMES)
 endif()
+option(ROCKSDB_BUILD_SHARED "Build shared versions of the RocksDB libraries" ON)
 
 if ($ENV{CIRCLECI})
   message(STATUS "Build for CircieCI env, a few tests may be disabled")
@@ -244,11 +245,21 @@ if(CMAKE_SYSTEM_PROCESSOR MATCHES "s390x")
   endif(HAS_S390X_MARCH_NATIVE)
 endif(CMAKE_SYSTEM_PROCESSOR MATCHES "s390x")
 
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64")
+  CHECK_C_COMPILER_FLAG("-march=loongarch64" HAS_LOONGARCH64)
+  if(HAS_LOONGARCH64)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mcpu=loongarch64 -mtune=loongarch64")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcpu=loongarch64 -mtune=loongarch64")
+  endif(HAS_LOONGARCH64)
+endif(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64")
+
 option(PORTABLE "build a portable binary" OFF)
 option(FORCE_SSE42 "force building with SSE4.2, even when PORTABLE=ON" OFF)
 option(FORCE_AVX "force building with AVX, even when PORTABLE=ON" OFF)
 option(FORCE_AVX2 "force building with AVX2, even when PORTABLE=ON" OFF)
 if(PORTABLE)
+  add_definitions(-DROCKSDB_PORTABLE)
+
   # MSVC does not need a separate compiler flag to enable SSE4.2; if nmmintrin.h
   # is available, it is available by default.
   if(FORCE_SSE42 AND NOT MSVC)
@@ -272,6 +283,9 @@ if(PORTABLE)
     if(CMAKE_SYSTEM_PROCESSOR MATCHES "^s390x")
       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z196")
     endif()
+    if(CMAKE_SYSTEM_PROCESSOR MATCHES "^loongarch64")
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=loongarch64")
+    endif()
   endif()
 else()
   if(MSVC)
@@ -291,8 +305,7 @@ if(NOT MSVC)
   set(CMAKE_REQUIRED_FLAGS "-msse4.2 -mpclmul")
 endif()
 
-if (NOT PORTABLE OR FORCE_SSE42)
-  CHECK_CXX_SOURCE_COMPILES("
+CHECK_CXX_SOURCE_COMPILES("
 #include <cstdint>
 #include <nmmintrin.h>
 #include <wmmintrin.h>
@@ -304,12 +317,11 @@ int main() {
   auto d = _mm_cvtsi128_si64(c);
 }
 " HAVE_SSE42)
-  if(HAVE_SSE42)
-    add_definitions(-DHAVE_SSE42)
-    add_definitions(-DHAVE_PCLMUL)
-  elseif(FORCE_SSE42)
-    message(FATAL_ERROR "FORCE_SSE42=ON but unable to compile with SSE4.2 enabled")
-  endif()
+if(HAVE_SSE42)
+  add_definitions(-DHAVE_SSE42)
+  add_definitions(-DHAVE_PCLMUL)
+elseif(FORCE_SSE42)
+  message(FATAL_ERROR "FORCE_SSE42=ON but unable to compile with SSE4.2 enabled")
 endif()
 
 # Check if -latomic is required or not
@@ -584,10 +596,61 @@ endif()
 
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
+
+if(USE_COROUTINES)
+  if(USE_FOLLY OR USE_FOLLY_LITE)
+    message(FATAL_ERROR "Please specify exactly one of USE_COROUTINES,"
+    " USE_FOLLY, and USE_FOLLY_LITE")
+  endif()
+  set(CMAKE_CXX_STANDARD 20)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fcoroutines -Wno-maybe-uninitialized")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-deprecated")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-redundant-move")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-invalid-memory-model")
+  add_compile_definitions(USE_COROUTINES)
+  set(USE_FOLLY 1)
+endif()
+
 if(USE_FOLLY)
-  include_directories(${PROJECT_SOURCE_DIR}/third-party/folly)
-  add_definitions(-DUSE_FOLLY -DFOLLY_NO_CONFIG)
-  list(APPEND THIRDPARTY_LIBS glog)
+  if(USE_FOLLY_LITE)
+    message(FATAL_ERROR "Please specify one of USE_FOLLY or USE_FOLLY_LITE")
+  endif()
+  if(ROCKSDB_BUILD_SHARED)
+    message(FATAL_ERROR "Cannot build RocksDB shared library with folly")
+  endif()
+  set(ROCKSDB_BUILD_SHARED OFF)
+  set(GFLAGS_SHARED FALSE)
+  find_package(folly)
+  # If cmake could not find the folly-config.cmake file, fall back
+  # to looking in third-party/folly for folly and its dependencies
+  if(NOT FOLLY_LIBRARIES)
+    exec_program(python3 ${PROJECT_SOURCE_DIR}/third-party/folly ARGS
+    build/fbcode_builder/getdeps.py show-inst-dir OUTPUT_VARIABLE
+    FOLLY_INST_PATH)
+    exec_program(ls ARGS -d ${FOLLY_INST_PATH}/../boost* OUTPUT_VARIABLE
+    BOOST_INST_PATH)
+    exec_program(ls ARGS -d ${FOLLY_INST_PATH}/../fmt* OUTPUT_VARIABLE
+    FMT_INST_PATH)
+    exec_program(ls ARGS -d ${FOLLY_INST_PATH}/../gflags* OUTPUT_VARIABLE
+    GFLAGS_INST_PATH)
+    set(Boost_DIR ${BOOST_INST_PATH}/lib/cmake/Boost-1.78.0)
+    if(EXISTS ${FMT_INST_PATH}/lib64)
+      set(fmt_DIR ${FMT_INST_PATH}/lib64/cmake/fmt)
+    else()
+      set(fmt_DIR ${FMT_INST_PATH}/lib/cmake/fmt)
+    endif()
+    set(gflags_DIR ${GFLAGS_INST_PATH}/lib/cmake/gflags)
+
+    exec_program(sed ARGS -i 's/gflags_shared//g'
+    ${FOLLY_INST_PATH}/lib/cmake/folly/folly-targets.cmake)
+
+    include(${FOLLY_INST_PATH}/lib/cmake/folly/folly-config.cmake)
+  endif()
+
+  add_compile_definitions(USE_FOLLY FOLLY_NO_CONFIG HAVE_CXX11_ATOMIC)
+  list(APPEND THIRDPARTY_LIBS Folly::folly)
+  set(FOLLY_LIBS Folly::folly)
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--copy-dt-needed-entries")
 endif()
 find_package(Threads REQUIRED)
 
@@ -597,14 +660,16 @@ set(SOURCES
         cache/cache.cc
         cache/cache_entry_roles.cc
         cache/cache_key.cc
+        cache/cache_helpers.cc
         cache/cache_reservation_manager.cc
         cache/charged_cache.cc
         cache/clock_cache.cc
         cache/compressed_secondary_cache.cc
-        cache/fast_lru_cache.cc
         cache/lru_cache.cc
+        cache/secondary_cache.cc
         cache/sharded_cache.cc
         db/arena_wrapped_db_iter.cc
+        db/blob/blob_contents.cc
         db/blob/blob_fetcher.cc
         db/blob/blob_file_addition.cc
         db/blob/blob_file_builder.cc
@@ -667,7 +732,7 @@ set(SOURCES
         db/merge_helper.cc
         db/merge_operator.cc
         db/output_validator.cc
-        db/periodic_work_scheduler.cc
+        db/periodic_task_scheduler.cc
         db/range_del_aggregator.cc
         db/range_tombstone_fragmenter.cc
         db/repair.cc
@@ -743,6 +808,7 @@ set(SOURCES
         options/options.cc
         options/options_helper.cc
         options/options_parser.cc
+        port/mmap.cc
         port/stack_trace.cc
         table/adaptive/adaptive_table_factory.cc
         table/block_based/binary_search_index_reader.cc
@@ -752,6 +818,7 @@ set(SOURCES
         table/block_based/block_based_table_iterator.cc
         table/block_based/block_based_table_reader.cc
         table/block_based/block_builder.cc
+        table/block_based/block_cache.cc
         table/block_based/block_prefetcher.cc
         table/block_based/block_prefix_index.cc
         table/block_based/data_block_hash_index.cc
@@ -827,6 +894,7 @@ set(SOURCES
         util/slice.cc
         util/file_checksum_helper.cc
         util/status.cc
+        util/stderr_logger.cc
         util/string_util.cc
         util/thread_local.cc
         util/threadpool_imp.cc
@@ -924,6 +992,12 @@ if ( ROCKSDB_PLUGINS )
         plugin/${plugin}/${src}
         PROPERTIES COMPILE_FLAGS "${${plugin}_COMPILE_FLAGS}")
     endforeach()
+    foreach (test ${${plugin}_TESTS})
+      list(APPEND PLUGIN_TESTS plugin/${plugin}/${test})
+      set_source_files_properties(
+        plugin/${plugin}/${test}
+        PROPERTIES COMPILE_FLAGS "${${plugin}_COMPILE_FLAGS}")
+    endforeach()
     foreach (path ${${plugin}_INCLUDE_PATHS})
       include_directories(${path})
     endforeach()
@@ -981,7 +1055,7 @@ else()
     env/io_posix.cc)
 endif()
 
-if(USE_FOLLY)
+if(USE_FOLLY_LITE)
   list(APPEND SOURCES
     third-party/folly/folly/container/detail/F14Table.cpp
     third-party/folly/folly/detail/Futex.cpp
@@ -991,13 +1065,14 @@ if(USE_FOLLY)
     third-party/folly/folly/synchronization/AtomicNotification.cpp
     third-party/folly/folly/synchronization/DistributedMutex.cpp
     third-party/folly/folly/synchronization/ParkingLot.cpp)
+  include_directories(${PROJECT_SOURCE_DIR}/third-party/folly)
+  add_definitions(-DUSE_FOLLY -DFOLLY_NO_CONFIG)
+  list(APPEND THIRDPARTY_LIBS glog)
 endif()
 
 set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
 set(ROCKSDB_SHARED_LIB rocksdb-shared${ARTIFACT_SUFFIX})
 
-option(ROCKSDB_BUILD_SHARED "Build shared versions of the RocksDB libraries" ON)
-
 
 if(WIN32)
   set(SYSTEM_LIBS ${SYSTEM_LIBS} shlwapi.lib rpcrt4.lib)
@@ -1008,31 +1083,36 @@ endif()
 set(ROCKSDB_PLUGIN_EXTERNS "")
 set(ROCKSDB_PLUGIN_BUILTINS "")
 message(STATUS "ROCKSDB PLUGINS TO BUILD ${ROCKSDB_PLUGINS}")
-list(APPEND PLUGINS ${ROCKSDB_PLUGINS})
 foreach(PLUGIN IN LISTS PLUGINS)
   set(PLUGIN_ROOT "${CMAKE_CURRENT_SOURCE_DIR}/plugin/${PLUGIN}/")
-  message("including rocksb plugin ${PLUGIN_ROOT}")
+  message(STATUS "PLUGIN ${PLUGIN} including rocksb plugin ${PLUGIN_ROOT}")
   set(PLUGINMKFILE "${PLUGIN_ROOT}${PLUGIN}.mk")
   if (NOT EXISTS ${PLUGINMKFILE})
-    message(FATAL_ERROR "Missing plugin makefile: ${PLUGINMKFILE}")
+    message(FATAL_ERROR "PLUGIN ${PLUGIN} Missing plugin makefile: ${PLUGINMKFILE}")
   endif()
   file(READ ${PLUGINMKFILE} PLUGINMK)
+
   string(REGEX MATCH "SOURCES = ([^\n]*)" FOO ${PLUGINMK})
   set(MK_SOURCES ${CMAKE_MATCH_1})
   separate_arguments(MK_SOURCES)
   foreach(MK_FILE IN LISTS MK_SOURCES)
     list(APPEND SOURCES "${PLUGIN_ROOT}${MK_FILE}")
+    message(STATUS "PLUGIN ${PLUGIN} Appending ${PLUGIN_ROOT}${MK_FILE} to SOURCES")
   endforeach()
+
   string(REGEX MATCH "_FUNC = ([^\n]*)" FOO ${PLUGINMK})
   if (NOT ${CMAKE_MATCH_1} STREQUAL "")
     string(APPEND ROCKSDB_PLUGIN_BUILTINS "{\"${PLUGIN}\", " ${CMAKE_MATCH_1} "},")
     string(APPEND ROCKSDB_PLUGIN_EXTERNS "int " ${CMAKE_MATCH_1} "(ROCKSDB_NAMESPACE::ObjectLibrary&, const std::string&); ")
   endif()
+
   string(REGEX MATCH "_LIBS = ([^\n]*)" FOO ${PLUGINMK})
-  if (NOT ${CMAKE_MATCH_1} STREQUAL "")
-    list(APPEND THIRDPARTY_LIBS "${CMAKE_MATCH_1}")
-  endif()
-  message("THIRDPARTY_LIBS=${THIRDPARTY_LIBS}")
+  separate_arguments(CMAKE_MATCH_1)
+  foreach(MK_LIB IN LISTS CMAKE_MATCH_1)
+    list(APPEND THIRDPARTY_LIBS "${MK_LIB}")
+  endforeach()
+  message(STATUS "PLUGIN ${PLUGIN} THIRDPARTY_LIBS=${THIRDPARTY_LIBS}")
+
   #TODO: We need to set any compile/link-time flags and add any link libraries
 endforeach()
 
@@ -1297,7 +1377,7 @@ if(WITH_TESTS)
         db/merge_test.cc
         db/options_file_test.cc
         db/perf_context_test.cc
-        db/periodic_work_scheduler_test.cc
+        db/periodic_task_scheduler_test.cc
         db/plain_table_db_test.cc
         db/seqno_time_test.cc
         db/prefix_test.cc
@@ -1409,6 +1489,7 @@ if(WITH_TESTS)
         utilities/ttl/ttl_test.cc
         utilities/util_merge_operators_test.cc
         utilities/write_batch_with_index/write_batch_with_index_test.cc
+	${PLUGIN_TESTS}
     )
   endif()
 
@@ -1424,7 +1505,7 @@ if(WITH_TESTS)
   add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND})
   set(TESTUTILLIB testutillib${ARTIFACT_SUFFIX})
   add_library(${TESTUTILLIB} STATIC ${TESTUTIL_SOURCE})
-  target_link_libraries(${TESTUTILLIB} ${ROCKSDB_LIB})
+  target_link_libraries(${TESTUTILLIB} ${ROCKSDB_LIB} ${FOLLY_LIBS})
   if(MSVC)
     set_target_properties(${TESTUTILLIB} PROPERTIES COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/testutillib${ARTIFACT_SUFFIX}.pdb")
   endif()
@@ -1482,45 +1563,45 @@ if(WITH_BENCHMARK_TOOLS)
     cache/cache_bench.cc
     cache/cache_bench_tool.cc)
   target_link_libraries(cache_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(memtablerep_bench${ARTIFACT_SUFFIX}
     memtable/memtablerep_bench.cc)
   target_link_libraries(memtablerep_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(range_del_aggregator_bench${ARTIFACT_SUFFIX}
     db/range_del_aggregator_bench.cc)
   target_link_libraries(range_del_aggregator_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(table_reader_bench${ARTIFACT_SUFFIX}
     table/table_reader_bench.cc)
   target_link_libraries(table_reader_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} testharness ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} testharness ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(filter_bench${ARTIFACT_SUFFIX}
     util/filter_bench.cc)
   target_link_libraries(filter_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(hash_table_bench${ARTIFACT_SUFFIX}
     utilities/persistent_cache/hash_table_bench.cc)
   target_link_libraries(hash_table_bench${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 endif()
 
 option(WITH_TRACE_TOOLS "build with trace tools" ON)
 if(WITH_TRACE_TOOLS)
-  add_executable(block_cache_trace_analyzer_tool${ARTIFACT_SUFFIX}
+  add_executable(block_cache_trace_analyzer${ARTIFACT_SUFFIX}
     tools/block_cache_analyzer/block_cache_trace_analyzer_tool.cc)
-  target_link_libraries(block_cache_trace_analyzer_tool${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+  target_link_libraries(block_cache_trace_analyzer${ARTIFACT_SUFFIX}
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(trace_analyzer${ARTIFACT_SUFFIX}
     tools/trace_analyzer.cc)
   target_link_libraries(trace_analyzer${ARTIFACT_SUFFIX}
-    ${ROCKSDB_LIB} ${GFLAGS_LIB})
+    ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
 endif()
 

HISTORY.md

@@ -1,5 +1,148 @@
 # Rocksdb Change Log
-## Unreleased
+## 7.10.2 (02/10/2023)
+### Bug Fixes
+* Fixed a bug in DB open/recovery from a compressed WAL that was caused due to incorrect handling of certain record fragments with the same offset within a WAL block.
+
+## 7.10.1 (02/01/2023)
+### Bug Fixes
+* Fixed a data race on `ColumnFamilyData::flush_reason` caused by concurrent flushes.
+* Fixed `DisableManualCompaction()` and `CompactRangeOptions::canceled` to cancel compactions even when they are waiting on conflicting compactions to finish
+* Fixed a bug in which a successful `GetMergeOperands()` could transiently return `Status::MergeInProgress()`
+* Return the correct error (Status::NotSupported()) to MultiGet caller when ReadOptions::async_io flag is true and IO uring is not enabled. Previously, Status::Corruption() was being returned when the actual failure was lack of async IO support.
+
+## 7.10.0 (01/23/2023)
+### Behavior changes
+* Make best-efforts recovery verify SST unique ID before Version construction (#10962)
+* Introduce `epoch_number` and sort L0 files by `epoch_number` instead of `largest_seqno`. `epoch_number` represents the order of a file being flushed or ingested/imported. Compaction output file will be assigned with the minimum `epoch_number` among input files'. For L0, larger `epoch_number` indicates newer L0 file.
+
+### Bug Fixes
+* Fixed a regression in iterator where range tombstones after `iterate_upper_bound` is processed.
+* Fixed a memory leak in MultiGet with async_io read option, caused by IO errors during table file open
+* Fixed a bug that multi-level FIFO compaction deletes one file in non-L0 even when `CompactionOptionsFIFO::max_table_files_size` is no exceeded since #10348 or 7.8.0.
+* Fixed a bug caused by `DB::SyncWAL()` affecting `track_and_verify_wals_in_manifest`. Without the fix, application may see "open error: Corruption: Missing WAL with log number" while trying to open the db. The corruption is a false alarm but prevents DB open (#10892).
+* Fixed a BackupEngine bug in which RestoreDBFromLatestBackup would fail if the latest backup was deleted and there is another valid backup available.
+* Fix L0 file misorder corruption caused by ingesting files of overlapping seqnos with memtable entries' through introducing `epoch_number`. Before the fix, `force_consistency_checks=true` may catch the corruption before it's exposed to readers, in which case writes returning `Status::Corruption` would be expected. Also replace the previous incomplete fix (#5958) to the same corruption with this new and more complete fix.
+* Fixed a bug in LockWAL() leading to re-locking mutex (#11020).
+* Fixed a heap use after free bug in async scan prefetching when the scan thread and another thread try to read and load the same seek block into cache.
+* Fixed a heap use after free in async scan prefetching if dictionary compression is enabled, in which case sync read of the compression dictionary gets mixed with async prefetching
+* Fixed a data race bug of `CompactRange()` under `change_level=true` acts on overlapping range with an ongoing file ingestion for level compaction. This will either result in overlapping file ranges corruption at a certain level caught by `force_consistency_checks=true` or protentially two same keys both with seqno 0 in two different levels (i.e, new data ends up in lower/older level). The latter will be caught by assertion in debug build but go silently and result in read returning wrong result in release build. This fix is general so it also replaced previous fixes to a similar problem for `CompactFiles()` (#4665), general `CompactRange()` and auto compaction (commit 5c64fb6 and 87dfc1d).
+* Fixed a bug in compaction output cutting where small output files were produced due to TTL file cutting states were not being updated (#11075).
+
+### New Features
+* When an SstPartitionerFactory is configured, CompactRange() now automatically selects for compaction any files overlapping a partition boundary that is in the compaction range, even if no actual entries are in the requested compaction range. With this feature, manual compaction can be used to (re-)establish SST partition points when SstPartitioner changes, without a full compaction.
+* Add BackupEngine feature to exclude files from backup that are known to be backed up elsewhere, using `CreateBackupOptions::exclude_files_callback`. To restore the DB, the excluded files must be provided in alternative backup directories using `RestoreOptions::alternate_dirs`.
+
+### Public API Changes
+* Substantial changes have been made to the Cache class to support internal development goals. Direct use of Cache class members is discouraged and further breaking modifications are expected in the future. SecondaryCache has some related changes and implementations will need to be updated. (Unlike Cache, SecondaryCache is still intended to support user implementations, and disruptive changes will be avoided.) (#10975)
+* Add `MergeOperationOutput::op_failure_scope` for merge operator users to control the blast radius of merge operator failures. Existing merge operator users do not need to make any change to preserve the old behavior
+
+### Performance Improvements
+* Updated xxHash source code, which should improve kXXH3 checksum speed, at least on ARM (#11098).
+* Improved CPU efficiency of DB reads, from block cache access improvements (#10975).
+
+## 7.9.0 (11/21/2022)
+### Performance Improvements
+* Fixed an iterator performance regression for delete range users when scanning through a consecutive sequence of range tombstones (#10877).
+
+### Bug Fixes
+* Fix memory corruption error in scans if async_io is enabled. Memory corruption happened if there is IOError while reading the data leading to empty buffer and other buffer already in progress of async read goes again for reading.
+* Fix failed memtable flush retry bug that could cause wrongly ordered updates, which would surface to writers as `Status::Corruption` in case of `force_consistency_checks=true` (default). It affects use cases that enable both parallel flush (`max_background_flushes > 1` or `max_background_jobs >= 8`) and non-default memtable count (`max_write_buffer_number > 2`).
+* Fixed an issue where the `READ_NUM_MERGE_OPERANDS` ticker was not updated when the base key-value or tombstone was read from an SST file.
+* Fixed a memory safety bug when using a SecondaryCache with `block_cache_compressed`. `block_cache_compressed` no longer attempts to use SecondaryCache features.
+* Fixed a regression in scan for async_io. During seek, valid buffers were getting cleared causing a regression.
+* Tiered Storage: fixed excessive keys written to penultimate level in non-debug builds.
+
+### New Features
+* Add basic support for user-defined timestamp to Merge (#10819).
+* Add stats for ReadAsync time spent and async read errors.
+* Basic support for the wide-column data model is now available. Wide-column entities can be stored using the `PutEntity` API, and retrieved using `GetEntity` and the new `columns` API of iterator. For compatibility, the classic APIs `Get` and `MultiGet`, as well as iterator's `value` API return the value of the anonymous default column of wide-column entities; also, `GetEntity` and iterator's `columns` return any plain key-values in the form of an entity which only has the anonymous default column. `Merge` (and `GetMergeOperands`) currently also apply to the default column; any other columns of entities are unaffected by `Merge` operations. Note that some features like compaction filters, transactions, user-defined timestamps, and the SST file writer do not yet support wide-column entities; also, there is currently no `MultiGet`-like API to retrieve multiple entities at once. We plan to gradually close the above gaps and also implement new features like column-level operations (e.g. updating or querying only certain columns of an entity).
+* Marked HyperClockCache as a production-ready alternative to LRUCache for the block cache. HyperClockCache greatly improves hot-path CPU efficiency under high parallel load or high contention, with some documented caveats and limitations. As much as 4.5x higher ops/sec vs. LRUCache has been seen in db_bench under high parallel load.
+* Add periodic diagnostics to info_log (LOG file) for HyperClockCache block cache if performance is degraded by bad `estimated_entry_charge` option.
+
+### Public API Changes
+* Marked `block_cache_compressed` as a deprecated feature. Use SecondaryCache instead.
+* Added a `SecondaryCache::InsertSaved()` API, with default implementation depending on `Insert()`. Some implementations might need to add a custom implementation of `InsertSaved()`. (Details in API comments.)
+
+## 7.8.0 (10/22/2022)
+### New Features
+* `DeleteRange()` now supports user-defined timestamp.
+* Provide support for async_io with tailing iterators when ReadOptions.tailing is enabled during scans.
+* Tiered Storage: allow data moving up from the last level to the penultimate level if the input level is penultimate level or above.
+* Added `DB::Properties::kFastBlockCacheEntryStats`, which is similar to `DB::Properties::kBlockCacheEntryStats`, except returns cached (stale) values in more cases to reduce overhead.
+* FIFO compaction now supports migrating from a multi-level DB via DB::Open(). During the migration phase, FIFO compaction picker will:
+* picks the sst file with the smallest starting key in the bottom-most non-empty level.
+* Note that during the migration phase, the file purge order will only be an approximation of "FIFO" as files in lower-level might sometime contain newer keys than files in upper-level.
+* Added an option `ignore_max_compaction_bytes_for_input` to ignore max_compaction_bytes limit when adding files to be compacted from input level. This should help reduce write amplification. The option is enabled by default.
+* Tiered Storage: allow data moving up from the last level even if it's a last level only compaction, as long as the penultimate level is empty.
+* Add a new option IOOptions.do_not_recurse that can be used by underlying file systems to skip recursing through sub directories and list only files in GetChildren API.
+* Add option `preserve_internal_time_seconds` to preserve the time information for the latest data. Which can be used to determine the age of data when `preclude_last_level_data_seconds` is enabled. The time information is attached with SST in table property `rocksdb.seqno.time.map` which can be parsed by tool ldb or sst_dump.
+
+### Bug Fixes
+* Fix a bug in io_uring_prep_cancel in AbortIO API for posix which expects sqe->addr to match with read request submitted and wrong paramter was being passed.
+* Fixed a regression in iterator performance when the entire DB is a single memtable introduced in #10449. The fix is in #10705 and #10716.
+* Fixed an optimistic transaction validation bug caused by DBImpl::GetLatestSequenceForKey() returning non-latest seq for merge (#10724).
+* Fixed a bug in iterator refresh which could segfault for DeleteRange users (#10739).
+* Fixed a bug causing manual flush with `flush_opts.wait=false` to stall when database has stopped all writes (#10001).
+* Fixed a bug in iterator refresh that was not freeing up SuperVersion, which could cause excessive resource pinniung (#10770).
+* Fixed a bug where RocksDB could be doing compaction endlessly when allow_ingest_behind is true and the bottommost level is not filled (#10767).
+* Fixed a memory safety bug in experimental HyperClockCache (#10768)
+* Fixed some cases where `ldb update_manifest` and `ldb unsafe_remove_sst_file` are not usable because they were requiring the DB files to match the existing manifest state (before updating the manifest to match a desired state).
+
+### Performance Improvements
+* Try to align the compaction output file boundaries to the next level ones, which can reduce more than 10% compaction load for the default level compaction. The feature is enabled by default, to disable, set `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size` to false. As a side effect, it can create SSTs larger than the target_file_size (capped at 2x target_file_size) or smaller files.
+* Improve RoundRobin TTL compaction, which is going to be the same as normal RoundRobin compaction to move the compaction cursor.
+* Fix a small CPU regression caused by a change that UserComparatorWrapper was made Customizable, because Customizable itself has small CPU overhead for initialization.
+
+### Behavior Changes
+* Sanitize min_write_buffer_number_to_merge to 1 if atomic flush is enabled to prevent unexpected data loss when WAL is disabled in a multi-column-family setting (#10773).
+* With periodic stat dumper waits up every options.stats_dump_period_sec seconds, it won't dump stats for a CF if it has no change in the period, unless 7 periods have been skipped.
+* Only periodic stats dumper triggered by options.stats_dump_period_sec will update stats interval. Ones triggered by DB::GetProperty() will not update stats interval and will report based on an interval since the last time stats dump period.
+
+### Public API changes
+* Make kXXH3 checksum the new default, because it is faster on common hardware, especially with kCRC32c affected by a performance bug in some versions of clang (https://github.com/facebook/rocksdb/issues/9891). DBs written with this new setting can be read by RocksDB 6.27 and newer.
+* Refactor the classes, APIs and data structures for block cache tracing to allow a user provided trace writer to be used. Introduced an abstract BlockCacheTraceWriter class that takes a structured BlockCacheTraceRecord. The BlockCacheTraceWriter implementation can then format and log the record in whatever way it sees fit. The default BlockCacheTraceWriterImpl does file tracing using a user provided TraceWriter. More details in rocksdb/includb/block_cache_trace_writer.h.
+
+## 7.7.0 (09/18/2022)
+### Bug Fixes
+* Fixed a hang when an operation such as `GetLiveFiles` or `CreateNewBackup` is asked to trigger and wait for memtable flush on a read-only DB. Such indirect requests for memtable flush are now ignored on a read-only DB.
+* Fixed bug where `FlushWAL(true /* sync */)` (used by `GetLiveFilesStorageInfo()`, which is used by checkpoint and backup) could cause parallel writes at the tail of a WAL file to never be synced.
+* Fix periodic_task unable to re-register the same task type, which may cause `SetOptions()` fail to update periodical_task time like: `stats_dump_period_sec`, `stats_persist_period_sec`.
+* Fixed a bug in the rocksdb.prefetched.bytes.discarded stat. It was counting the prefetch buffer size, rather than the actual number of bytes discarded from the buffer.
+* Fix bug where the directory containing CURRENT can left unsynced after CURRENT is updated to point to the latest MANIFEST, which leads to risk of unsync data loss of CURRENT.
+* Update rocksdb.multiget.io.batch.size stat in non-async MultiGet as well.
+* Fix a bug in key range overlap checking with concurrent compactions when user-defined timestamp is enabled. User-defined timestamps should be EXCLUDED when checking if two ranges overlap.
+* Fixed a bug where the blob cache prepopulating logic did not consider the secondary cache (see #10603).
+* Fixed the rocksdb.num.sst.read.per.level, rocksdb.num.index.and.filter.blocks.read.per.level and rocksdb.num.level.read.per.multiget stats in the MultiGet coroutines
+
+### Public API changes
+* Add `rocksdb_column_family_handle_get_id`, `rocksdb_column_family_handle_get_name` to get name, id of column family in C API
+* Add a new stat rocksdb.async.prefetch.abort.micros to measure time spent waiting for async prefetch reads to abort
+
+### Java API Changes
+* Add CompactionPriority.RoundRobin.
+* Revert to using the default metadata charge policy when creating an LRU cache via the Java API.
+
+### Behavior Change
+* DBOptions::verify_sst_unique_id_in_manifest is now an on-by-default feature that verifies SST file identity whenever they are opened by a DB, rather than only at DB::Open time.
+* Right now, when the option migration tool (OptionChangeMigration()) migrates to FIFO compaction, it compacts all the data into one single SST file and move to L0. This might create a problem for some users: the giant file may be soon deleted to satisfy max_table_files_size, and might cayse the DB to be almost empty. We change the behavior so that the files are cut to be smaller, but these files might not follow the data insertion order. With the change, after the migration, migrated data might not be dropped by insertion order by FIFO compaction.
+* When a block is firstly found from `CompressedSecondaryCache`, we just insert a dummy block into the primary cache and don’t erase the block from `CompressedSecondaryCache`. A standalone handle is returned to the caller. Only if the block is found again from `CompressedSecondaryCache` before the dummy block is evicted, we erase the block from `CompressedSecondaryCache` and insert it into the primary cache.
+* When a block is firstly evicted from the primary cache to `CompressedSecondaryCache`, we just insert a dummy block in `CompressedSecondaryCache`. Only if it is evicted again before the dummy block is evicted from the cache, it is treated as a hot block and is inserted into `CompressedSecondaryCache`.
+* Improved the estimation of memory used by cached blobs by taking into account the size of the object owning the blob value and also the allocator overhead if `malloc_usable_size` is available (see #10583).
+* Blob values now have their own category in the cache occupancy statistics, as opposed to being lumped into the "Misc" bucket (see #10601).
+* Change the optimize_multiget_for_io experimental ReadOptions flag to default on.
+
+### New Features
+*  RocksDB does internal auto prefetching if it notices 2 sequential reads if readahead_size is not specified. New option `num_file_reads_for_auto_readahead` is added in BlockBasedTableOptions which indicates after how many sequential reads internal auto prefetching should be start (default is 2).
+* Added new perf context counters `block_cache_standalone_handle_count`, `block_cache_real_handle_count`,`compressed_sec_cache_insert_real_count`, `compressed_sec_cache_insert_dummy_count`, `compressed_sec_cache_uncompressed_bytes`, and `compressed_sec_cache_compressed_bytes`.
+* Memory for blobs which are to be inserted into the blob cache is now allocated using the cache's allocator (see #10628 and #10647).
+* HyperClockCache is an experimental, lock-free Cache alternative for block cache that offers much improved CPU efficiency under high parallel load or high contention, with some caveats. As much as 4.5x higher ops/sec vs. LRUCache has been seen in db_bench under high parallel load.
+* `CompressedSecondaryCacheOptions::enable_custom_split_merge` is added for enabling the custom split and merge feature, which split the compressed value into chunks so that they may better fit jemalloc bins.
+
+### Performance Improvements
+* Iterator performance is improved for `DeleteRange()` users. Internally, iterator will skip to the end of a range tombstone when possible, instead of looping through each key and check individually if a key is range deleted.
+* Eliminated some allocations and copies in the blob read path. Also, `PinnableSlice` now only points to the blob value and pins the backing resource (cache entry or buffer) in all cases, instead of containing a copy of the blob value. See #10625 and #10647.
+* In case of scans with async_io enabled, few optimizations have been added to issue more asynchronous requests in parallel in order to avoid synchronous prefetching.
+* `DeleteRange()` users should see improvement in get/iterator performance from mutable memtable (see #10547).
 
 ## 7.6.0 (08/19/2022)
 ### New Features
@@ -41,15 +184,13 @@
 * If an error is hit when writing to a file (append, sync, etc), RocksDB is more strict with not issuing more operations to it, except closing the file, with exceptions of some WAL file operations in error recovery path.
 * A `WriteBufferManager` constructed with `allow_stall == false` will no longer trigger write stall implicitly by thrashing until memtable count limit is reached. Instead, a column family can continue accumulating writes while that CF is flushing, which means memory may increase. Users who prefer stalling writes must now explicitly set `allow_stall == true`.
 * Add `CompressedSecondaryCache` into the stress tests.
+* Block cache keys have changed, which will cause any persistent caches to miss between versions.
 
 ### Performance Improvements
 * Instead of constructing `FragmentedRangeTombstoneList` during every read operation, it is now constructed once and stored in immutable memtables. This improves speed of querying range tombstones from immutable memtables.
 * When using iterators with the integrated BlobDB implementation, blob cache handles are now released immediately when the iterator's position changes.
 * MultiGet can now do more IO in parallel by reading data blocks from SST files in multiple levels, if the optimize_multiget_for_io ReadOption flag is set.
 
-## Behavior Change
-* Block cache keys have changed, which will cause any persistent caches to miss between versions.
-
 ## 7.5.0 (07/15/2022)
 ### New Features
 * Mempurge option flag `experimental_mempurge_threshold` is now a ColumnFamilyOptions and can now be dynamically configured using `SetOptions()`.

LANGUAGE-BINDINGS.md

@@ -6,7 +6,9 @@ This is the list of all known third-party language bindings for RocksDB. If some
     * http://pyrocksdb.readthedocs.org/en/latest/ (unmaintained)
 * Perl - https://metacpan.org/pod/RocksDB
 * Node.js - https://npmjs.org/package/rocksdb
-* Go - https://github.com/tecbot/gorocksdb
+* Go 
+  * https://github.com/linxGnu/grocksdb
+  * https://github.com/tecbot/gorocksdb (unmaintained)
 * Ruby - http://rubygems.org/gems/rocksdb-ruby
 * Haskell - https://hackage.haskell.org/package/rocksdb-haskell
 * PHP - https://github.com/Photonios/rocksdb-php

Makefile

@@ -139,9 +139,14 @@ endif
 GIT_COMMAND ?= git
 ifeq ($(USE_COROUTINES), 1)
 	USE_FOLLY = 1
-	OPT += -DUSE_COROUTINES
+	# glog/logging.h requires HAVE_CXX11_ATOMIC
+	OPT += -DUSE_COROUTINES -DHAVE_CXX11_ATOMIC
 	ROCKSDB_CXX_STANDARD = c++2a
 	USE_RTTI = 1
+ifneq ($(USE_CLANG), 1)
+	ROCKSDB_CXX_STANDARD = c++20
+	PLATFORM_CXXFLAGS += -fcoroutines
+endif
 endif
 
 # if we're compiling for release, compile without debug code (-DNDEBUG)
@@ -185,6 +190,16 @@ ifeq ($(USE_LTO), 1)
 	LDFLAGS += -flto -fuse-linker-plugin
 endif
 
+# `COERCE_CONTEXT_SWITCH=1` will inject spurious wakeup and
+# random length of sleep or context switch at critical
+# points (e.g, before acquring db mutex) in RocksDB.
+# In this way, it coerces as many excution orders as possible in the hope of
+# exposing the problematic excution order
+COERCE_CONTEXT_SWITCH ?= 0
+ifeq ($(COERCE_CONTEXT_SWITCH), 1)
+OPT += -DCOERCE_CONTEXT_SWITCH
+endif
+
 #-----------------------------------------------
 include src.mk
 
@@ -235,6 +250,7 @@ dummy := $(shell (export ROCKSDB_ROOT="$(CURDIR)"; \
                   export USE_CLANG="$(USE_CLANG)"; \
                   export LIB_MODE="$(LIB_MODE)"; \
 		  export ROCKSDB_CXX_STANDARD="$(ROCKSDB_CXX_STANDARD)"; \
+		  export USE_FOLLY="$(USE_FOLLY)"; \
                   "$(CURDIR)/build_tools/build_detect_platform" "$(CURDIR)/make_config.mk"))
 # this file is generated by the previous line to set build flags and sources
 include make_config.mk
@@ -250,6 +266,7 @@ ROCKSDB_PLUGIN_EXTERNS = $(foreach p, $(ROCKSDB_PLUGIN_W_FUNCS), int $($(p)_FUNC
 ROCKSDB_PLUGIN_BUILTINS = $(foreach p, $(ROCKSDB_PLUGIN_W_FUNCS), {\"$(p)\"\, $($(p)_FUNC)}\,)
 ROCKSDB_PLUGIN_LDFLAGS = $(foreach plugin, $(ROCKSDB_PLUGINS), $($(plugin)_LDFLAGS))
 ROCKSDB_PLUGIN_PKGCONFIG_REQUIRES = $(foreach plugin, $(ROCKSDB_PLUGINS), $($(plugin)_PKGCONFIG_REQUIRES))
+ROCKSDB_PLUGIN_TESTS = $(foreach p, $(ROCKSDB_PLUGINS), $(foreach test, $($(p)_TESTS), plugin/$(p)/$(test)))
 
 CXXFLAGS += $(foreach plugin, $(ROCKSDB_PLUGINS), $($(plugin)_CXXFLAGS))
 PLATFORM_LDFLAGS += $(ROCKSDB_PLUGIN_LDFLAGS)
@@ -415,6 +432,10 @@ ifndef DISABLE_JEMALLOC
 			PLATFORM_CXXFLAGS += -DUSE_JEMALLOC
 			PLATFORM_CCFLAGS  += -DUSE_JEMALLOC
 		endif
+		ifeq ($(USE_FOLLY_LITE),1)
+			PLATFORM_CXXFLAGS += -DUSE_JEMALLOC
+			PLATFORM_CCFLAGS  += -DUSE_JEMALLOC
+		endif
 	endif
 	ifdef WITH_JEMALLOC_FLAG
 		PLATFORM_LDFLAGS += -ljemalloc
@@ -448,10 +469,57 @@ endif
 
 # This provides a Makefile simulation of a Meta-internal folly integration.
 # It is not validated for general use.
+#
+# USE_FOLLY links the build targets with libfolly.a. The latter could be
+# built using 'make build_folly', or built externally and specified in
+# the CXXFLAGS and EXTRA_LDFLAGS env variables. The build_detect_platform
+# script tries to detect if an external folly dependency has been specified.
+# If not, it exports FOLLY_PATH to the path of the installed Folly and
+# dependency libraries.
+#
+# USE_FOLLY_LITE cherry picks source files from Folly to include in the
+# RocksDB library. Its faster and has fewer dependencies on 3rd party
+# libraries, but with limited functionality. For example, coroutine
+# functionality is not available.
 ifeq ($(USE_FOLLY),1)
-	ifeq (,$(FOLLY_DIR))
-		FOLLY_DIR = ./third-party/folly
+ifeq ($(USE_FOLLY_LITE),1)
+$(error Please specify only one of USE_FOLLY and USE_FOLLY_LITE)
+endif
+ifneq ($(strip $(FOLLY_PATH)),)
+	BOOST_PATH = $(shell (ls -d $(FOLLY_PATH)/../boost*))
+	DBL_CONV_PATH = $(shell (ls -d $(FOLLY_PATH)/../double-conversion*))
+	GFLAGS_PATH = $(shell (ls -d $(FOLLY_PATH)/../gflags*))
+	GLOG_PATH = $(shell (ls -d $(FOLLY_PATH)/../glog*))
+	LIBEVENT_PATH = $(shell (ls -d $(FOLLY_PATH)/../libevent*))
+	XZ_PATH = $(shell (ls -d $(FOLLY_PATH)/../xz*))
+	LIBSODIUM_PATH = $(shell (ls -d $(FOLLY_PATH)/../libsodium*))
+	FMT_PATH = $(shell (ls -d $(FOLLY_PATH)/../fmt*))
+
+	# For some reason, glog and fmt libraries are under either lib or lib64
+	GLOG_LIB_PATH = $(shell (ls -d $(GLOG_PATH)/lib*))
+	FMT_LIB_PATH = $(shell (ls -d $(FMT_PATH)/lib*))
+
+	# AIX: pre-defined system headers are surrounded by an extern "C" block
+	ifeq ($(PLATFORM), OS_AIX)
+		PLATFORM_CCFLAGS += -I$(BOOST_PATH)/include -I$(DBL_CONV_PATH)/include -I$(GLOG_PATH)/include -I$(LIBEVENT_PATH)/include -I$(XZ_PATH)/include -I$(LIBSODIUM_PATH)/include -I$(FOLLY_PATH)/include -I$(FMT_PATH)/include
+		PLATFORM_CXXFLAGS += -I$(BOOST_PATH)/include -I$(DBL_CONV_PATH)/include -I$(GLOG_PATH)/include -I$(LIBEVENT_PATH)/include -I$(XZ_PATH)/include -I$(LIBSODIUM_PATH)/include -I$(FOLLY_PATH)/include -I$(FMT_PATH)/include
+	else
+		PLATFORM_CCFLAGS += -isystem $(BOOST_PATH)/include -isystem $(DBL_CONV_PATH)/include -isystem $(GLOG_PATH)/include -isystem $(LIBEVENT_PATH)/include -isystem $(XZ_PATH)/include -isystem $(LIBSODIUM_PATH)/include -isystem $(FOLLY_PATH)/include -isystem $(FMT_PATH)/include
+		PLATFORM_CXXFLAGS += -isystem $(BOOST_PATH)/include -isystem $(DBL_CONV_PATH)/include -isystem $(GLOG_PATH)/include -isystem $(LIBEVENT_PATH)/include -isystem $(XZ_PATH)/include -isystem $(LIBSODIUM_PATH)/include -isystem $(FOLLY_PATH)/include -isystem $(FMT_PATH)/include
 	endif
+
+	# Add -ldl at the end as gcc resolves a symbol in a library by searching only in libraries specified later
+	# in the command line
+	PLATFORM_LDFLAGS += $(FOLLY_PATH)/lib/libfolly.a $(BOOST_PATH)/lib/libboost_context.a $(BOOST_PATH)/lib/libboost_filesystem.a $(BOOST_PATH)/lib/libboost_atomic.a $(BOOST_PATH)/lib/libboost_program_options.a $(BOOST_PATH)/lib/libboost_regex.a $(BOOST_PATH)/lib/libboost_system.a $(BOOST_PATH)/lib/libboost_thread.a $(DBL_CONV_PATH)/lib/libdouble-conversion.a $(FMT_LIB_PATH)/libfmt.a $(GLOG_LIB_PATH)/libglog.so $(GFLAGS_PATH)/lib/libgflags.so.2.2 $(LIBEVENT_PATH)/lib/libevent-2.1.so -ldl
+	PLATFORM_LDFLAGS += -Wl,-rpath=$(GFLAGS_PATH)/lib -Wl,-rpath=$(GLOG_LIB_PATH) -Wl,-rpath=$(LIBEVENT_PATH)/lib -Wl,-rpath=$(LIBSODIUM_PATH)/lib -Wl,-rpath=$(LIBEVENT_PATH)/lib
+endif
+	PLATFORM_CCFLAGS += -DUSE_FOLLY -DFOLLY_NO_CONFIG
+	PLATFORM_CXXFLAGS += -DUSE_FOLLY -DFOLLY_NO_CONFIG
+endif
+
+ifeq ($(USE_FOLLY_LITE),1)
+	# Path to the Folly source code and include files
+	FOLLY_DIR = ./third-party/folly
 	# AIX: pre-defined system headers are surrounded by an extern "C" block
 	ifeq ($(PLATFORM), OS_AIX)
 		PLATFORM_CCFLAGS += -I$(FOLLY_DIR)
@@ -550,7 +618,7 @@ LIB_OBJECTS += $(patsubst %.c, $(OBJ_DIR)/%.o, $(LIB_SOURCES_C))
 LIB_OBJECTS += $(patsubst %.S, $(OBJ_DIR)/%.o, $(LIB_SOURCES_ASM))
 endif
 
-ifeq ($(USE_FOLLY),1)
+ifeq ($(USE_FOLLY_LITE),1)
   LIB_OBJECTS += $(patsubst %.cpp, $(OBJ_DIR)/%.o, $(FOLLY_SOURCES))
 endif
 
@@ -580,10 +648,12 @@ STRESS_OBJECTS =  $(patsubst %.cc, $(OBJ_DIR)/%.o, $(STRESS_LIB_SOURCES))
 ALL_SOURCES  = $(filter-out util/build_version.cc, $(LIB_SOURCES)) $(TEST_LIB_SOURCES) $(MOCK_LIB_SOURCES) $(GTEST_DIR)/gtest/gtest-all.cc
 ALL_SOURCES += $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(CACHE_BENCH_LIB_SOURCES) $(ANALYZER_LIB_SOURCES) $(STRESS_LIB_SOURCES)
 ALL_SOURCES += $(TEST_MAIN_SOURCES) $(TOOL_MAIN_SOURCES) $(BENCH_MAIN_SOURCES)
-ALL_SOURCES += $(ROCKSDB_PLUGIN_SOURCES)
+ALL_SOURCES += $(ROCKSDB_PLUGIN_SOURCES) $(ROCKSDB_PLUGIN_TESTS)
 
+PLUGIN_TESTS = $(patsubst %.cc, %, $(notdir $(ROCKSDB_PLUGIN_TESTS)))
 TESTS = $(patsubst %.cc, %, $(notdir $(TEST_MAIN_SOURCES)))
 TESTS += $(patsubst %.c, %, $(notdir $(TEST_MAIN_SOURCES_C)))
+TESTS += $(PLUGIN_TESTS)
 
 # `make check-headers` to very that each header file includes its own
 # dependencies
@@ -591,7 +661,7 @@ ifneq ($(filter check-headers, $(MAKECMDGOALS)),)
 # TODO: add/support JNI headers
 	DEV_HEADER_DIRS := $(sort include/ $(dir $(ALL_SOURCES)))
 # Some headers like in port/ are platform-specific
-	DEV_HEADERS := $(shell $(FIND) $(DEV_HEADER_DIRS) -type f -name '*.h' | egrep -v 'port/|plugin/|lua/|range_tree/')
+	DEV_HEADERS := $(shell $(FIND) $(DEV_HEADER_DIRS) -type f -name '*.h' | grep -E -v 'port/|plugin/|lua/|range_tree/')
 else
 	DEV_HEADERS :=
 endif
@@ -635,6 +705,7 @@ NON_PARALLEL_TEST = \
 	env_test \
 	deletefile_test \
 	db_bloom_filter_test \
+	$(PLUGIN_TESTS) \
 
 PARALLEL_TEST = $(filter-out $(NON_PARALLEL_TEST), $(TESTS))
 
@@ -728,9 +799,9 @@ TOOLS_LIBRARY=$(STATIC_TOOLS_LIBRARY)
 endif
 STRESS_LIBRARY=$(STATIC_STRESS_LIBRARY)
 
-ROCKSDB_MAJOR = $(shell egrep "ROCKSDB_MAJOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
-ROCKSDB_MINOR = $(shell egrep "ROCKSDB_MINOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
-ROCKSDB_PATCH = $(shell egrep "ROCKSDB_PATCH.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
+ROCKSDB_MAJOR = $(shell grep -E "ROCKSDB_MAJOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
+ROCKSDB_MINOR = $(shell grep -E "ROCKSDB_MINOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
+ROCKSDB_PATCH = $(shell grep -E "ROCKSDB_PATCH.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
 
 # If NO_UPDATE_BUILD_VERSION is set we don't update util/build_version.cc, but
 # the file needs to already exist or else the build will fail
@@ -1288,6 +1359,14 @@ db_sanity_test: $(OBJ_DIR)/tools/db_sanity_test.o $(LIBRARY)
 db_repl_stress: $(OBJ_DIR)/tools/db_repl_stress.o $(LIBRARY)
 	$(AM_LINK)
 
+define MakeTestRule
+$(notdir $(1:%.cc=%)): $(1:%.cc=$$(OBJ_DIR)/%.o) $$(TEST_LIBRARY) $$(LIBRARY)
+	$$(AM_LINK)
+endef
+
+# For each PLUGIN test, create a rule to generate the test executable
+$(foreach test, $(ROCKSDB_PLUGIN_TESTS), $(eval $(call MakeTestRule, $(test))))
+
 arena_test: $(OBJ_DIR)/memory/arena_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
@@ -1765,7 +1844,7 @@ memtable_list_test: $(OBJ_DIR)/db/memtable_list_test.o $(TEST_LIBRARY) $(LIBRARY
 write_callback_test: $(OBJ_DIR)/db/write_callback_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
-heap_test: $(OBJ_DIR)/util/heap_test.o $(GTEST)
+heap_test: $(OBJ_DIR)/util/heap_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
 point_lock_manager_test: utilities/transactions/lock/point/point_lock_manager_test.o $(TEST_LIBRARY) $(LIBRARY)
@@ -1882,7 +1961,7 @@ blob_garbage_meter_test: $(OBJ_DIR)/db/blob/blob_garbage_meter_test.o $(TEST_LIB
 timer_test: $(OBJ_DIR)/util/timer_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
-periodic_work_scheduler_test: $(OBJ_DIR)/db/periodic_work_scheduler_test.o $(TEST_LIBRARY) $(LIBRARY)
+periodic_task_scheduler_test: $(OBJ_DIR)/db/periodic_task_scheduler_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
 testutil_test: $(OBJ_DIR)/test_util/testutil_test.o $(TEST_LIBRARY) $(LIBRARY)
@@ -1991,7 +2070,7 @@ JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
 ifeq ($(PLATFORM), OS_SOLARIS)
 	ARCH := $(shell isainfo -b)
 else ifeq ($(PLATFORM), OS_OPENBSD)
-	ifneq (,$(filter amd64 ppc64 ppc64le s390x arm64 aarch64 sparc64, $(MACHINE)))
+	ifneq (,$(filter amd64 ppc64 ppc64le s390x arm64 aarch64 sparc64 loongarch64, $(MACHINE)))
 		ARCH := 64
 	else
 		ARCH := 32
@@ -2012,7 +2091,7 @@ ifneq ($(origin JNI_LIBC), undefined)
 endif
 
 ifeq (,$(ROCKSDBJNILIB))
-ifneq (,$(filter ppc% s390x arm64 aarch64 sparc64, $(MACHINE)))
+ifneq (,$(filter ppc% s390x arm64 aarch64 sparc64 loongarch64, $(MACHINE)))
 	ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE)$(JNI_LIBC_POSTFIX).so
 else
 	ROCKSDBJNILIB = librocksdbjni-linux$(ARCH)$(JNI_LIBC_POSTFIX).so
@@ -2025,8 +2104,8 @@ ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(ROCKSDB_JAVA_VERSION)-javadoc.jar
 ROCKSDB_SOURCES_JAR = rocksdbjni-$(ROCKSDB_JAVA_VERSION)-sources.jar
 SHA256_CMD = sha256sum
 
-ZLIB_VER ?= 1.2.12
-ZLIB_SHA256 ?= 91844808532e5ce316b3c010929493c0244f3d37593afd6de04f71821d5136d9
+ZLIB_VER ?= 1.2.13
+ZLIB_SHA256 ?= b3a24de97a8fdbc835b9833169501030b8977031bcb54b3b3ac13740f846ab30
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.8
 BZIP2_SHA256 ?= ab5a03176ee106d3f0fa90e381da478ddae405918153cca248e682cd0c4a2269
@@ -2371,13 +2450,26 @@ checkout_folly:
 	@# need to worry about folly breaking our integration. Update periodically
 	cd third-party/folly && git reset --hard beacd86d63cd71c904632262e6c36f60874d78ba
 	@# A hack to remove boost dependency.
-	@# NOTE: this hack is not needed if using FBCODE compiler config
+	@# NOTE: this hack is only needed if building using USE_FOLLY_LITE
 	perl -pi -e 's/^(#include <boost)/\/\/$$1/' third-party/folly/folly/functional/Invoke.h
 	@# NOTE: this hack is required for clang in some cases
 	perl -pi -e 's/int rv = syscall/int rv = (int)syscall/' third-party/folly/folly/detail/Futex.cpp
 	@# NOTE: this hack is required for gcc in some cases
 	perl -pi -e 's/(__has_include.<experimental.memory_resource>.)/__cpp_rtti && $$1/' third-party/folly/folly/memory/MemoryResource.h
 
+build_folly:
+	FOLLY_INST_PATH=`cd third-party/folly; $(PYTHON) build/fbcode_builder/getdeps.py show-inst-dir`; \
+	if [ "$$FOLLY_INST_PATH" ]; then \
+		rm -rf $${FOLLY_INST_PATH}/../../*; \
+	else \
+		echo "Please run checkout_folly first"; \
+		false; \
+	fi
+	# Restore the original version of Invoke.h with boost dependency
+	cd third-party/folly && ${GIT_COMMAND} checkout folly/functional/Invoke.h
+	cd third-party/folly && MAYBE_AVX2=`echo $(CXXFLAGS) | grep -o -- -DHAVE_AVX2 | sed 's/-DHAVE_AVX2/-mavx2/g' || true` && \
+		CXXFLAGS=" $$MAYBE_AVX2 -DHAVE_CXX11_ATOMIC " $(PYTHON) build/fbcode_builder/getdeps.py build --no-tests
+
 # ---------------------------------------------------------------------------
 #   Build size testing
 # ---------------------------------------------------------------------------
@@ -2462,7 +2554,7 @@ endif
 ifneq ($(SKIP_DEPENDS), 1)
 DEPFILES = $(patsubst %.cc, $(OBJ_DIR)/%.cc.d, $(ALL_SOURCES))
 DEPFILES+ = $(patsubst %.c, $(OBJ_DIR)/%.c.d, $(LIB_SOURCES_C) $(TEST_MAIN_SOURCES_C))
-ifeq ($(USE_FOLLY),1)
+ifeq ($(USE_FOLLY_LITE),1)
   DEPFILES +=$(patsubst %.cpp, $(OBJ_DIR)/%.cpp.d, $(FOLLY_SOURCES))
 endif
 endif

TARGETS

@@ -11,15 +11,17 @@ load("//rocks/buckifier:defs.bzl", "cpp_library_wrapper","rocks_cpp_library_wrap
 cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "cache/cache.cc",
         "cache/cache_entry_roles.cc",
+        "cache/cache_helpers.cc",
         "cache/cache_key.cc",
         "cache/cache_reservation_manager.cc",
         "cache/charged_cache.cc",
         "cache/clock_cache.cc",
         "cache/compressed_secondary_cache.cc",
-        "cache/fast_lru_cache.cc",
         "cache/lru_cache.cc",
+        "cache/secondary_cache.cc",
         "cache/sharded_cache.cc",
         "db/arena_wrapped_db_iter.cc",
+        "db/blob/blob_contents.cc",
         "db/blob/blob_fetcher.cc",
         "db/blob/blob_file_addition.cc",
         "db/blob/blob_file_builder.cc",
@@ -82,7 +84,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "db/merge_helper.cc",
         "db/merge_operator.cc",
         "db/output_validator.cc",
-        "db/periodic_work_scheduler.cc",
+        "db/periodic_task_scheduler.cc",
         "db/range_del_aggregator.cc",
         "db/range_tombstone_fragmenter.cc",
         "db/repair.cc",
@@ -162,6 +164,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "options/options.cc",
         "options/options_helper.cc",
         "options/options_parser.cc",
+        "port/mmap.cc",
         "port/port_posix.cc",
         "port/stack_trace.cc",
         "port/win/env_default.cc",
@@ -178,6 +181,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "table/block_based/block_based_table_iterator.cc",
         "table/block_based/block_based_table_reader.cc",
         "table/block_based/block_builder.cc",
+        "table/block_based/block_cache.cc",
         "table/block_based/block_prefetcher.cc",
         "table/block_based/block_prefix_index.cc",
         "table/block_based/data_block_footer.cc",
@@ -252,6 +256,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "util/ribbon_config.cc",
         "util/slice.cc",
         "util/status.cc",
+        "util/stderr_logger.cc",
         "util/string_util.cc",
         "util/thread_local.cc",
         "util/threadpool_imp.cc",
@@ -348,15 +353,17 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
 cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "cache/cache.cc",
         "cache/cache_entry_roles.cc",
+        "cache/cache_helpers.cc",
         "cache/cache_key.cc",
         "cache/cache_reservation_manager.cc",
         "cache/charged_cache.cc",
         "cache/clock_cache.cc",
         "cache/compressed_secondary_cache.cc",
-        "cache/fast_lru_cache.cc",
         "cache/lru_cache.cc",
+        "cache/secondary_cache.cc",
         "cache/sharded_cache.cc",
         "db/arena_wrapped_db_iter.cc",
+        "db/blob/blob_contents.cc",
         "db/blob/blob_fetcher.cc",
         "db/blob/blob_file_addition.cc",
         "db/blob/blob_file_builder.cc",
@@ -419,7 +426,7 @@ cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "db/merge_helper.cc",
         "db/merge_operator.cc",
         "db/output_validator.cc",
-        "db/periodic_work_scheduler.cc",
+        "db/periodic_task_scheduler.cc",
         "db/range_del_aggregator.cc",
         "db/range_tombstone_fragmenter.cc",
         "db/repair.cc",
@@ -499,6 +506,7 @@ cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "options/options.cc",
         "options/options_helper.cc",
         "options/options_parser.cc",
+        "port/mmap.cc",
         "port/port_posix.cc",
         "port/stack_trace.cc",
         "port/win/env_default.cc",
@@ -515,6 +523,7 @@ cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "table/block_based/block_based_table_iterator.cc",
         "table/block_based/block_based_table_reader.cc",
         "table/block_based/block_builder.cc",
+        "table/block_based/block_cache.cc",
         "table/block_based/block_prefetcher.cc",
         "table/block_based/block_prefix_index.cc",
         "table/block_based/data_block_footer.cc",
@@ -589,6 +598,7 @@ cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "util/ribbon_config.cc",
         "util/slice.cc",
         "util/status.cc",
+        "util/stderr_logger.cc",
         "util/string_util.cc",
         "util/thread_local.cc",
         "util/threadpool_imp.cc",
@@ -5598,8 +5608,8 @@ cpp_unittest_wrapper(name="perf_context_test",
             extra_compiler_flags=[])
 
 
-cpp_unittest_wrapper(name="periodic_work_scheduler_test",
-            srcs=["db/periodic_work_scheduler_test.cc"],
+cpp_unittest_wrapper(name="periodic_task_scheduler_test",
+            srcs=["db/periodic_task_scheduler_test.cc"],
             deps=[":rocksdb_test_lib"],
             extra_compiler_flags=[])
 

USERS.md

@@ -79,6 +79,9 @@ quasardb uses a heavily tuned RocksDB as its persistence layer.
 ## TiKV
 [TiKV](https://github.com/pingcap/tikv) is a GEO-replicated, high-performance, distributed, transactional key-value database. TiKV is powered by Rust and Raft. TiKV uses RocksDB as its persistence layer.
 
+## Apache Spark
+[Spark Structured Streaming](https://docs.databricks.com/structured-streaming/rocksdb-state-store.html) uses RocksDB as the local state store.
+
 ## Apache Flink
 [Apache Flink](https://flink.apache.org/news/2016/03/08/release-1.0.0.html) uses RocksDB to store state locally on a machine.
 

buckifier/buckify_rocksdb.py

@@ -1,19 +1,18 @@
 #!/usr/bin/env python3
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
+from __future__ import absolute_import, division, print_function, unicode_literals
+
 try:
     from builtins import str
 except ImportError:
     from __builtin__ import str
-from targets_builder import TARGETSBuilder
+import fnmatch
 import json
 import os
-import fnmatch
 import sys
 
+from targets_builder import TARGETSBuilder
+
 from util import ColorString
 
 # This script generates TARGETS file for Buck.
@@ -44,13 +43,13 @@ def parse_src_mk(repo_path):
     src_files = {}
     for line in open(src_mk):
         line = line.strip()
-        if len(line) == 0 or line[0] == '#':
+        if len(line) == 0 or line[0] == "#":
             continue
-        if '=' in line:
-            current_src = line.split('=')[0].strip()
+        if "=" in line:
+            current_src = line.split("=")[0].strip()
             src_files[current_src] = []
-        elif '.c' in line:
-            src_path = line.split('\\')[0].strip()
+        elif ".c" in line:
+            src_path = line.split("\\")[0].strip()
             src_files[current_src].append(src_path)
     return src_files
 
@@ -58,14 +57,16 @@ def parse_src_mk(repo_path):
 # get all .cc / .c files
 def get_cc_files(repo_path):
     cc_files = []
-    for root, dirnames, filenames in os.walk(repo_path):  # noqa: B007 T25377293 Grandfathered in
-        root = root[(len(repo_path) + 1):]
+    for root, _dirnames, filenames in os.walk(
+        repo_path
+    ):  # noqa: B007 T25377293 Grandfathered in
+        root = root[(len(repo_path) + 1) :]
         if "java" in root:
             # Skip java
             continue
-        for filename in fnmatch.filter(filenames, '*.cc'):
+        for filename in fnmatch.filter(filenames, "*.cc"):
             cc_files.append(os.path.join(root, filename))
-        for filename in fnmatch.filter(filenames, '*.c'):
+        for filename in fnmatch.filter(filenames, "*.c"):
             cc_files.append(os.path.join(root, filename))
     return cc_files
 
@@ -93,14 +94,10 @@ def get_non_parallel_tests(repo_path):
 
     return s
 
+
 # Parse extra dependencies passed by user from command line
 def get_dependencies():
-    deps_map = {
-        '': {
-            'extra_deps': [],
-            'extra_compiler_flags': []
-        }
-    }
+    deps_map = {"": {"extra_deps": [], "extra_compiler_flags": []}}
     if len(sys.argv) < 2:
         return deps_map
 
@@ -111,6 +108,7 @@ def get_dependencies():
                 v = encode_dict(v)
             rv[k] = v
         return rv
+
     extra_deps = json.loads(sys.argv[1], object_hook=encode_dict)
     for target_alias, deps in extra_deps.items():
         deps_map[target_alias] = deps
@@ -143,73 +141,73 @@ def generate_targets(repo_path, deps_map):
         "rocksdb_lib",
         src_mk["LIB_SOURCES"] +
         # always add range_tree, it's only excluded on ppc64, which we don't use internally
-        src_mk["RANGE_TREE_SOURCES"] +
-        src_mk["TOOL_LIB_SOURCES"],
+        src_mk["RANGE_TREE_SOURCES"] + src_mk["TOOL_LIB_SOURCES"],
         deps=[
             "//folly/container:f14_hash",
             "//folly/experimental/coro:blocking_wait",
             "//folly/experimental/coro:collect",
             "//folly/experimental/coro:coroutine",
             "//folly/experimental/coro:task",
-            "//folly/synchronization:distributed_mutex"])
+            "//folly/synchronization:distributed_mutex",
+        ],
+    )
     # rocksdb_whole_archive_lib
     TARGETS.add_library(
         "rocksdb_whole_archive_lib",
         src_mk["LIB_SOURCES"] +
         # always add range_tree, it's only excluded on ppc64, which we don't use internally
-        src_mk["RANGE_TREE_SOURCES"] +
-        src_mk["TOOL_LIB_SOURCES"],
+        src_mk["RANGE_TREE_SOURCES"] + src_mk["TOOL_LIB_SOURCES"],
         deps=[
             "//folly/container:f14_hash",
             "//folly/experimental/coro:blocking_wait",
             "//folly/experimental/coro:collect",
             "//folly/experimental/coro:coroutine",
             "//folly/experimental/coro:task",
-            "//folly/synchronization:distributed_mutex"],
+            "//folly/synchronization:distributed_mutex",
+        ],
         headers=None,
         extra_external_deps="",
-        link_whole=True)
+        link_whole=True,
+    )
     # rocksdb_test_lib
     TARGETS.add_library(
         "rocksdb_test_lib",
-        src_mk.get("MOCK_LIB_SOURCES", []) +
-        src_mk.get("TEST_LIB_SOURCES", []) +
-        src_mk.get("EXP_LIB_SOURCES", []) +
-        src_mk.get("ANALYZER_LIB_SOURCES", []),
+        src_mk.get("MOCK_LIB_SOURCES", [])
+        + src_mk.get("TEST_LIB_SOURCES", [])
+        + src_mk.get("EXP_LIB_SOURCES", [])
+        + src_mk.get("ANALYZER_LIB_SOURCES", []),
         [":rocksdb_lib"],
-        extra_test_libs=True
-        )
+        extra_test_libs=True,
+    )
     # rocksdb_tools_lib
     TARGETS.add_library(
         "rocksdb_tools_lib",
-        src_mk.get("BENCH_LIB_SOURCES", []) +
-        src_mk.get("ANALYZER_LIB_SOURCES", []) +
-        ["test_util/testutil.cc"],
-        [":rocksdb_lib"])
+        src_mk.get("BENCH_LIB_SOURCES", [])
+        + src_mk.get("ANALYZER_LIB_SOURCES", [])
+        + ["test_util/testutil.cc"],
+        [":rocksdb_lib"],
+    )
     # rocksdb_cache_bench_tools_lib
     TARGETS.add_library(
         "rocksdb_cache_bench_tools_lib",
         src_mk.get("CACHE_BENCH_LIB_SOURCES", []),
-        [":rocksdb_lib"])
+        [":rocksdb_lib"],
+    )
     # rocksdb_stress_lib
     TARGETS.add_rocksdb_library(
         "rocksdb_stress_lib",
         src_mk.get("ANALYZER_LIB_SOURCES", [])
-        + src_mk.get('STRESS_LIB_SOURCES', [])
-        + ["test_util/testutil.cc"])
+        + src_mk.get("STRESS_LIB_SOURCES", [])
+        + ["test_util/testutil.cc"],
+    )
     # db_stress binary
-    TARGETS.add_binary("db_stress",
-                       ["db_stress_tool/db_stress.cc"],
-                       [":rocksdb_stress_lib"])
+    TARGETS.add_binary(
+        "db_stress", ["db_stress_tool/db_stress.cc"], [":rocksdb_stress_lib"]
+    )
     # bench binaries
     for src in src_mk.get("MICROBENCH_SOURCES", []):
-        name =  src.rsplit('/',1)[1].split('.')[0] if '/' in src else src.split('.')[0]
-        TARGETS.add_binary(
-            name,
-            [src],
-            [],
-            extra_bench_libs=True
-            )
+        name = src.rsplit("/", 1)[1].split(".")[0] if "/" in src else src.split(".")[0]
+        TARGETS.add_binary(name, [src], [], extra_bench_libs=True)
     print("Extra dependencies:\n{0}".format(json.dumps(deps_map)))
 
     # Dictionary test executable name -> relative source file path
@@ -219,7 +217,7 @@ def generate_targets(repo_path, deps_map):
     # are more than one .c test file, we need to extend
     # TARGETS.add_c_test() to include other C tests too.
     for test_src in src_mk.get("TEST_MAIN_SOURCES_C", []):
-        if test_src != 'db/c_test.c':
+        if test_src != "db/c_test.c":
             print("Don't know how to deal with " + test_src)
             return False
     TARGETS.add_c_test()
@@ -229,7 +227,7 @@ def generate_targets(repo_path, deps_map):
             fast_fancy_bench_config_list = json.load(json_file)
             for config_dict in fast_fancy_bench_config_list:
                 clean_benchmarks = {}
-                benchmarks = config_dict['benchmarks']
+                benchmarks = config_dict["benchmarks"]
                 for binary, benchmark_dict in benchmarks.items():
                     clean_benchmarks[binary] = {}
                     for benchmark, overloaded_metric_list in benchmark_dict.items():
@@ -237,13 +235,20 @@ def generate_targets(repo_path, deps_map):
                         for metric in overloaded_metric_list:
                             if not isinstance(metric, dict):
                                 clean_benchmarks[binary][benchmark].append(metric)
-                TARGETS.add_fancy_bench_config(config_dict['name'], clean_benchmarks, False, config_dict['expected_runtime_one_iter'], config_dict['sl_iterations'], config_dict['regression_threshold'])
+                TARGETS.add_fancy_bench_config(
+                    config_dict["name"],
+                    clean_benchmarks,
+                    False,
+                    config_dict["expected_runtime_one_iter"],
+                    config_dict["sl_iterations"],
+                    config_dict["regression_threshold"],
+                )
 
         with open(f"{repo_path}/buckifier/bench-slow.json") as json_file:
             slow_fancy_bench_config_list = json.load(json_file)
             for config_dict in slow_fancy_bench_config_list:
                 clean_benchmarks = {}
-                benchmarks = config_dict['benchmarks']
+                benchmarks = config_dict["benchmarks"]
                 for binary, benchmark_dict in benchmarks.items():
                     clean_benchmarks[binary] = {}
                     for benchmark, overloaded_metric_list in benchmark_dict.items():
@@ -252,7 +257,14 @@ def generate_targets(repo_path, deps_map):
                             if not isinstance(metric, dict):
                                 clean_benchmarks[binary][benchmark].append(metric)
             for config_dict in slow_fancy_bench_config_list:
-                TARGETS.add_fancy_bench_config(config_dict['name']+"_slow", clean_benchmarks, True, config_dict['expected_runtime_one_iter'], config_dict['sl_iterations'], config_dict['regression_threshold'])
+                TARGETS.add_fancy_bench_config(
+                    config_dict["name"] + "_slow",
+                    clean_benchmarks,
+                    True,
+                    config_dict["expected_runtime_one_iter"],
+                    config_dict["sl_iterations"],
+                    config_dict["regression_threshold"],
+                )
     # it is better servicelab experiments break
     # than rocksdb github ci
     except Exception:
@@ -261,7 +273,7 @@ def generate_targets(repo_path, deps_map):
     TARGETS.add_test_header()
 
     for test_src in src_mk.get("TEST_MAIN_SOURCES", []):
-        test = test_src.split('.c')[0].strip().split('/')[-1].strip()
+        test = test_src.split(".c")[0].strip().split("/")[-1].strip()
         test_source_map[test] = test_src
         print("" + test + " " + test_src)
 
@@ -271,23 +283,29 @@ def generate_targets(repo_path, deps_map):
                 print(ColorString.warning("Failed to get test name for %s" % test_src))
                 continue
 
-            test_target_name = \
-                test if not target_alias else test + "_" + target_alias
+            test_target_name = test if not target_alias else test + "_" + target_alias
 
             if test in _EXPORTED_TEST_LIBS:
                 test_library = "%s_lib" % test_target_name
-                TARGETS.add_library(test_library, [test_src], deps=[":rocksdb_test_lib"], extra_test_libs=True)
+                TARGETS.add_library(
+                    test_library,
+                    [test_src],
+                    deps=[":rocksdb_test_lib"],
+                    extra_test_libs=True,
+                )
                 TARGETS.register_test(
                     test_target_name,
                     test_src,
-                    deps = json.dumps(deps['extra_deps'] + [':'+test_library]),
-                    extra_compiler_flags = json.dumps(deps['extra_compiler_flags']))
+                    deps=json.dumps(deps["extra_deps"] + [":" + test_library]),
+                    extra_compiler_flags=json.dumps(deps["extra_compiler_flags"]),
+                )
             else:
                 TARGETS.register_test(
                     test_target_name,
                     test_src,
-                    deps = json.dumps(deps['extra_deps'] + [":rocksdb_test_lib"] ),
-                    extra_compiler_flags = json.dumps(deps['extra_compiler_flags']))
+                    deps=json.dumps(deps["extra_deps"] + [":rocksdb_test_lib"]),
+                    extra_compiler_flags=json.dumps(deps["extra_compiler_flags"]),
+                )
 
     print(ColorString.info("Generated TARGETS Summary:"))
     print(ColorString.info("- %d libs" % TARGETS.total_lib))
@@ -300,8 +318,7 @@ def get_rocksdb_path():
     # rocksdb = {script_dir}/..
     script_dir = os.path.dirname(sys.argv[0])
     script_dir = os.path.abspath(script_dir)
-    rocksdb_path = os.path.abspath(
-        os.path.join(script_dir, "../"))
+    rocksdb_path = os.path.abspath(os.path.join(script_dir, "../"))
 
     return rocksdb_path
 
@@ -318,5 +335,6 @@ def main():
     if not ok:
         exit_with_error("Failed to generate TARGETS files")
 
+
 if __name__ == "__main__":
     main()

buckifier/targets_builder.py

@@ -1,113 +1,150 @@
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
+from __future__ import absolute_import, division, print_function, unicode_literals
+
 try:
-    from builtins import object
-    from builtins import str
+    from builtins import object, str
 except ImportError:
-    from __builtin__ import object
-    from __builtin__ import str
-import targets_cfg
+    from __builtin__ import object, str
 import pprint
 
+import targets_cfg
+
+
 def pretty_list(lst, indent=8):
     if lst is None or len(lst) == 0:
         return ""
 
     if len(lst) == 1:
-        return "\"%s\"" % lst[0]
+        return '"%s"' % lst[0]
 
-    separator = "\",\n%s\"" % (" " * indent)
+    separator = '",\n%s"' % (" " * indent)
     res = separator.join(sorted(lst))
-    res = "\n" + (" " * indent) + "\"" + res + "\",\n" + (" " * (indent - 4))
+    res = "\n" + (" " * indent) + '"' + res + '",\n' + (" " * (indent - 4))
     return res
 
 
 class TARGETSBuilder(object):
     def __init__(self, path, extra_argv):
         self.path = path
-        self.targets_file = open(path, 'wb')
         header = targets_cfg.rocksdb_target_header_template.format(
-            extra_argv=extra_argv)
-        self.targets_file.write(header.encode("utf-8"))
+            extra_argv=extra_argv
+        )
+        with open(path, "wb") as targets_file:
+            targets_file.write(header.encode("utf-8"))
         self.total_lib = 0
         self.total_bin = 0
         self.total_test = 0
         self.tests_cfg = ""
 
-    def __del__(self):
-        self.targets_file.close()
-
-    def add_library(self, name, srcs, deps=None, headers=None,
-                    extra_external_deps="", link_whole=False,
-                     external_dependencies=None, extra_test_libs=False):
+    def add_library(
+        self,
+        name,
+        srcs,
+        deps=None,
+        headers=None,
+        extra_external_deps="",
+        link_whole=False,
+        external_dependencies=None,
+        extra_test_libs=False,
+    ):
         if headers is not None:
             headers = "[" + pretty_list(headers) + "]"
-        self.targets_file.write(targets_cfg.library_template.format(
-            name=name,
-            srcs=pretty_list(srcs),
-            headers=headers,
-            deps=pretty_list(deps),
-            extra_external_deps=extra_external_deps,
-            link_whole=link_whole,
-            external_dependencies=pretty_list(external_dependencies),
-            extra_test_libs=extra_test_libs
-            ).encode("utf-8"))
-        self.total_lib = self.total_lib + 1
-
-    def add_rocksdb_library(self, name, srcs, headers=None,
-                            external_dependencies=None):
-        if headers is not None:
-            headers = "[" + pretty_list(headers) + "]"
-        self.targets_file.write(targets_cfg.rocksdb_library_template.format(
-            name=name,
-            srcs=pretty_list(srcs),
-            headers=headers,
-            external_dependencies=pretty_list(external_dependencies)
-            ).encode("utf-8")
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                targets_cfg.library_template.format(
+                    name=name,
+                    srcs=pretty_list(srcs),
+                    headers=headers,
+                    deps=pretty_list(deps),
+                    extra_external_deps=extra_external_deps,
+                    link_whole=link_whole,
+                    external_dependencies=pretty_list(external_dependencies),
+                    extra_test_libs=extra_test_libs,
+                ).encode("utf-8")
             )
         self.total_lib = self.total_lib + 1
 
-    def add_binary(self, name, srcs, deps=None, extra_preprocessor_flags=None,extra_bench_libs=False):
-        self.targets_file.write(targets_cfg.binary_template.format(
-            name=name,
-            srcs=pretty_list(srcs),
-            deps=pretty_list(deps),
-            extra_preprocessor_flags=pretty_list(extra_preprocessor_flags),
-            extra_bench_libs=extra_bench_libs,
-            ).encode("utf-8"))
+    def add_rocksdb_library(self, name, srcs, headers=None, external_dependencies=None):
+        if headers is not None:
+            headers = "[" + pretty_list(headers) + "]"
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                targets_cfg.rocksdb_library_template.format(
+                    name=name,
+                    srcs=pretty_list(srcs),
+                    headers=headers,
+                    external_dependencies=pretty_list(external_dependencies),
+                ).encode("utf-8")
+            )
+        self.total_lib = self.total_lib + 1
+
+    def add_binary(
+        self,
+        name,
+        srcs,
+        deps=None,
+        extra_preprocessor_flags=None,
+        extra_bench_libs=False,
+    ):
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                targets_cfg.binary_template.format(
+                    name=name,
+                    srcs=pretty_list(srcs),
+                    deps=pretty_list(deps),
+                    extra_preprocessor_flags=pretty_list(extra_preprocessor_flags),
+                    extra_bench_libs=extra_bench_libs,
+                ).encode("utf-8")
+            )
         self.total_bin = self.total_bin + 1
 
     def add_c_test(self):
-        self.targets_file.write(b"""
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                b"""
 add_c_test_wrapper()
-""")
+"""
+            )
 
     def add_test_header(self):
-        self.targets_file.write(b"""
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                b"""
         # Generate a test rule for each entry in ROCKS_TESTS
         # Do not build the tests in opt mode, since SyncPoint and other test code
         # will not be included.
-""")
+"""
+            )
 
-    def add_fancy_bench_config(self, name, bench_config, slow, expected_runtime, sl_iterations, regression_threshold):
-        self.targets_file.write(targets_cfg.fancy_bench_template.format(
+    def add_fancy_bench_config(
+        self,
+        name,
+        bench_config,
+        slow,
+        expected_runtime,
+        sl_iterations,
+        regression_threshold,
+    ):
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                targets_cfg.fancy_bench_template.format(
                     name=name,
                     bench_config=pprint.pformat(bench_config),
                     slow=slow,
                     expected_runtime=expected_runtime,
                     sl_iterations=sl_iterations,
-                    regression_threshold=regression_threshold
-                    ).encode("utf-8"))
+                    regression_threshold=regression_threshold,
+                ).encode("utf-8")
+            )
 
-    def register_test(self,
-                      test_name,
-                      src,
-                      deps,
-                      extra_compiler_flags):
-
-        self.targets_file.write(targets_cfg.unittests_template.format(test_name=test_name,test_cc=str(src),deps=deps,
-            extra_compiler_flags=extra_compiler_flags).encode("utf-8"))
+    def register_test(self, test_name, src, deps, extra_compiler_flags):
+        with open(self.path, "ab") as targets_file:
+            targets_file.write(
+                targets_cfg.unittests_template.format(
+                    test_name=test_name,
+                    test_cc=str(src),
+                    deps=deps,
+                    extra_compiler_flags=extra_compiler_flags,
+                ).encode("utf-8")
+            )
         self.total_test = self.total_test + 1

buckifier/targets_cfg.py

@@ -1,11 +1,7 @@
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
+from __future__ import absolute_import, division, print_function, unicode_literals
 
-rocksdb_target_header_template = \
-    """# This file \100generated by:
+rocksdb_target_header_template = """# This file \100generated by:
 #$ python3 buckifier/buckify_rocksdb.py{extra_argv}
 # --> DO NOT EDIT MANUALLY <--
 # This file is a Facebook-specific integration for buck builds, so can
@@ -27,7 +23,6 @@ rocks_cpp_library_wrapper(name="{name}", srcs=[{srcs}], headers={headers})
 """
 
 
-
 binary_template = """
 cpp_binary_wrapper(name="{name}", srcs=[{srcs}], deps=[{deps}], extra_preprocessor_flags=[{extra_preprocessor_flags}], extra_bench_libs={extra_bench_libs})
 """

buckifier/util.py

@@ -2,37 +2,35 @@
 """
 This module keeps commonly used components.
 """
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
+from __future__ import absolute_import, division, print_function, unicode_literals
+
 try:
     from builtins import object
 except ImportError:
     from __builtin__ import object
+import os
 import subprocess
 import sys
-import os
 import time
 
+
 class ColorString(object):
-    """ Generate colorful strings on terminal """
-    HEADER = '\033[95m'
-    BLUE = '\033[94m'
-    GREEN = '\033[92m'
-    WARNING = '\033[93m'
-    FAIL = '\033[91m'
-    ENDC = '\033[0m'
+    """Generate colorful strings on terminal"""
+
+    HEADER = "\033[95m"
+    BLUE = "\033[94m"
+    GREEN = "\033[92m"
+    WARNING = "\033[93m"
+    FAIL = "\033[91m"
+    ENDC = "\033[0m"
 
     @staticmethod
     def _make_color_str(text, color):
         # In Python2, default encoding for unicode string is ASCII
         if sys.version_info.major <= 2:
-            return "".join(
-                [color, text.encode('utf-8'), ColorString.ENDC])
+            return "".join([color, text.encode("utf-8"), ColorString.ENDC])
         # From Python3, default encoding for unicode string is UTF-8
-        return "".join(
-            [color, text, ColorString.ENDC])
+        return "".join([color, text, ColorString.ENDC])
 
     @staticmethod
     def ok(text):
@@ -68,37 +66,38 @@ class ColorString(object):
 
 
 def run_shell_command(shell_cmd, cmd_dir=None):
-    """ Run a single shell command.
-        @returns a tuple of shell command return code, stdout, stderr """
+    """Run a single shell command.
+    @returns a tuple of shell command return code, stdout, stderr"""
 
     if cmd_dir is not None and not os.path.exists(cmd_dir):
         run_shell_command("mkdir -p %s" % cmd_dir)
 
     start = time.time()
     print("\t>>> Running: " + shell_cmd)
-    p = subprocess.Popen(shell_cmd,
-                         shell=True,
-                         stdout=subprocess.PIPE,
-                         stderr=subprocess.PIPE,
-                         cwd=cmd_dir)
+    p = subprocess.Popen(  # noqa
+        shell_cmd,
+        shell=True,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        cwd=cmd_dir,
+    )
     stdout, stderr = p.communicate()
     end = time.time()
 
     # Report time if we spent more than 5 minutes executing a command
     execution_time = end - start
     if execution_time > (60 * 5):
-        mins = (execution_time / 60)
-        secs = (execution_time % 60)
+        mins = execution_time / 60
+        secs = execution_time % 60
         print("\t>time spent: %d minutes %d seconds" % (mins, secs))
 
-
     return p.returncode, stdout, stderr
 
 
 def run_shell_commands(shell_cmds, cmd_dir=None, verbose=False):
-    """ Execute a sequence of shell commands, which is equivalent to
-        running `cmd1 && cmd2 && cmd3`
-        @returns boolean indication if all commands succeeds.
+    """Execute a sequence of shell commands, which is equivalent to
+    running `cmd1 && cmd2 && cmd3`
+    @returns boolean indication if all commands succeeds.
     """
 
     if cmd_dir:

build_tools/amalgamate.py

@@ -28,14 +28,15 @@
 from __future__ import print_function
 
 import argparse
-from os import path
 import re
 import sys
+from os import path
 
 include_re = re.compile('^[ \t]*#include[ \t]+"(.*)"[ \t]*$')
 included = set()
 excluded = set()
 
+
 def find_header(name, abs_path, include_paths):
     samedir = path.join(path.dirname(abs_path), name)
     if path.exists(samedir):
@@ -46,17 +47,31 @@ def find_header(name, abs_path, include_paths):
             return include_path
     return None
 
-def expand_include(include_path, f, abs_path, source_out, header_out, include_paths, public_include_paths):
+
+def expand_include(
+    include_path,
+    f,
+    abs_path,
+    source_out,
+    header_out,
+    include_paths,
+    public_include_paths,
+):
     if include_path in included:
         return False
 
     included.add(include_path)
     with open(include_path) as f:
         print('#line 1 "{}"'.format(include_path), file=source_out)
-        process_file(f, include_path, source_out, header_out, include_paths, public_include_paths)
+        process_file(
+            f, include_path, source_out, header_out, include_paths, public_include_paths
+        )
     return True
 
-def process_file(f, abs_path, source_out, header_out, include_paths, public_include_paths):
+
+def process_file(
+    f, abs_path, source_out, header_out, include_paths, public_include_paths
+):
     for (line, text) in enumerate(f):
         m = include_re.match(text)
         if m:
@@ -68,7 +83,15 @@ def process_file(f, abs_path, source_out, header_out, include_paths, public_incl
                     source_out.write(text)
                     expanded = False
                 else:
-                    expanded = expand_include(include_path, f, abs_path, source_out, header_out, include_paths, public_include_paths)
+                    expanded = expand_include(
+                        include_path,
+                        f,
+                        abs_path,
+                        source_out,
+                        header_out,
+                        include_paths,
+                        public_include_paths,
+                    )
             else:
                 # now try public headers
                 include_path = find_header(filename, abs_path, public_include_paths)
@@ -78,23 +101,52 @@ def process_file(f, abs_path, source_out, header_out, include_paths, public_incl
                     if include_path in excluded:
                         source_out.write(text)
                     else:
-                        expand_include(include_path, f, abs_path, header_out, None, public_include_paths, [])
+                        expand_include(
+                            include_path,
+                            f,
+                            abs_path,
+                            header_out,
+                            None,
+                            public_include_paths,
+                            [],
+                        )
                 else:
-                    sys.exit("unable to find {}, included in {} on line {}".format(filename, abs_path, line))
+                    sys.exit(
+                        "unable to find {}, included in {} on line {}".format(
+                            filename, abs_path, line
+                        )
+                    )
 
             if expanded:
-                print('#line {} "{}"'.format(line+1, abs_path), file=source_out)
+                print('#line {} "{}"'.format(line + 1, abs_path), file=source_out)
         elif text != "#pragma once\n":
             source_out.write(text)
 
+
 def main():
-    parser = argparse.ArgumentParser(description="Transform a unity build into an amalgamation")
+    parser = argparse.ArgumentParser(
+        description="Transform a unity build into an amalgamation"
+    )
     parser.add_argument("source", help="source file")
-    parser.add_argument("-I", action="append", dest="include_paths", help="include paths for private headers")
-    parser.add_argument("-i", action="append", dest="public_include_paths", help="include paths for public headers")
-    parser.add_argument("-x", action="append", dest="excluded", help="excluded header files")
+    parser.add_argument(
+        "-I",
+        action="append",
+        dest="include_paths",
+        help="include paths for private headers",
+    )
+    parser.add_argument(
+        "-i",
+        action="append",
+        dest="public_include_paths",
+        help="include paths for public headers",
+    )
+    parser.add_argument(
+        "-x", action="append", dest="excluded", help="excluded header files"
+    )
     parser.add_argument("-o", dest="source_out", help="output C++ file", required=True)
-    parser.add_argument("-H", dest="header_out", help="output C++ header file", required=True)
+    parser.add_argument(
+        "-H", dest="header_out", help="output C++ header file", required=True
+    )
     args = parser.parse_args()
 
     include_paths = list(map(path.abspath, args.include_paths or []))
@@ -102,10 +154,15 @@ def main():
     excluded.update(map(path.abspath, args.excluded or []))
     filename = args.source
     abs_path = path.abspath(filename)
-    with open(filename) as f, open(args.source_out, 'w') as source_out, open(args.header_out, 'w') as header_out:
+    with open(filename) as f, open(args.source_out, "w") as source_out, open(
+        args.header_out, "w"
+    ) as header_out:
         print('#line 1 "{}"'.format(filename), file=source_out)
         print('#include "{}"'.format(header_out.name), file=source_out)
-        process_file(f, abs_path, source_out, header_out, include_paths, public_include_paths)
+        process_file(
+            f, abs_path, source_out, header_out, include_paths, public_include_paths
+        )
+
 
 if __name__ == "__main__":
     main()

build_tools/benchmark_log_tool.py

@@ -4,23 +4,27 @@
 #  COPYING file in the root directory) and Apache 2.0 License
 #  (found in the LICENSE.Apache file in the root directory).
 
-'''Access the results of benchmark runs
+"""Access the results of benchmark runs
 Send these results on to OpenSearch graphing service
-'''
+"""
 
 import argparse
 import itertools
+import logging
 import os
 import re
 import sys
+
 import requests
 from dateutil import parser
-import logging
 
 logging.basicConfig(level=logging.DEBUG)
+
+
 class Configuration:
-    opensearch_user = os.environ['ES_USER']
-    opensearch_pass = os.environ['ES_PASS']
+    opensearch_user = os.environ["ES_USER"]
+    opensearch_pass = os.environ["ES_PASS"]
+
 
 class BenchmarkResultException(Exception):
     def __init__(self, message, content):
@@ -30,45 +34,71 @@ class BenchmarkResultException(Exception):
 
 class BenchmarkUtils:
 
-    expected_keys = ['ops_sec', 'mb_sec',  'lsm_sz',  'blob_sz', 'c_wgb',   'w_amp',
-    'c_mbps',  'c_wsecs', 'c_csecs', 'b_rgb',   'b_wgb',   'usec_op',
-    'p50',     'p99',     'p99.9',   'p99.99', 'pmax',
-    'uptime',  'stall%', 'Nstall',  'u_cpu',   's_cpu',   'rss', 'test', 'date', 'version', 'job_id']
+    expected_keys = [
+        "ops_sec",
+        "mb_sec",
+        "lsm_sz",
+        "blob_sz",
+        "c_wgb",
+        "w_amp",
+        "c_mbps",
+        "c_wsecs",
+        "c_csecs",
+        "b_rgb",
+        "b_wgb",
+        "usec_op",
+        "p50",
+        "p99",
+        "p99.9",
+        "p99.99",
+        "pmax",
+        "uptime",
+        "stall%",
+        "Nstall",
+        "u_cpu",
+        "s_cpu",
+        "rss",
+        "test",
+        "date",
+        "version",
+        "job_id",
+    ]
 
     def sanity_check(row):
-        if not 'test' in row:
+        if "test" not in row:
             logging.debug(f"not 'test' in row: {row}")
             return False
-        if row['test'] == '':
+        if row["test"] == "":
             logging.debug(f"row['test'] == '': {row}")
             return False
-        if not 'date' in row:
+        if "date" not in row:
             logging.debug(f"not 'date' in row: {row}")
             return False
-        if not 'ops_sec' in row:
+        if "ops_sec" not in row:
             logging.debug(f"not 'ops_sec' in row: {row}")
             return False
         try:
-            v = int(row['ops_sec'])
+            _ = int(row["ops_sec"])
         except (ValueError, TypeError):
             logging.debug(f"int(row['ops_sec']): {row}")
             return False
         try:
-            (_, _) = parser.parse(row['date'], fuzzy_with_tokens=True)
+            (_, _) = parser.parse(row["date"], fuzzy_with_tokens=True)
         except (parser.ParserError):
-            logging.error(f"parser.parse((row['date']): not a valid format for date in row: {row}")
+            logging.error(
+                f"parser.parse((row['date']): not a valid format for date in row: {row}"
+            )
             return False
         return True
 
     def conform_opensearch(row):
-        (dt, _) = parser.parse(row['date'], fuzzy_with_tokens=True)
+        (dt, _) = parser.parse(row["date"], fuzzy_with_tokens=True)
         # create a test_date field, which was previously what was expected
         # repair the date field, which has what can be a WRONG ISO FORMAT, (no leading 0 on single-digit day-of-month)
         # e.g. 2022-07-1T00:14:55 should be 2022-07-01T00:14:55
-        row['test_date'] = dt.isoformat()
-        row['date'] = dt.isoformat()
-        return dict((key.replace('.', '_'), value)
-                    for (key, value) in row.items())
+        row["test_date"] = dt.isoformat()
+        row["date"] = dt.isoformat()
+        return {key.replace(".", "_"): value for key, value in row.items()}
 
 
 class ResultParser:
@@ -80,24 +110,24 @@ class ResultParser:
     def ignore(self, l_in: str):
         if len(l_in) == 0:
             return True
-        if l_in[0:1] == '#':
+        if l_in[0:1] == "#":
             return True
         return False
 
-    def line(self, l_in: str):
-        '''Parse a line into items
+    def line(self, line_in: str):
+        """Parse a line into items
         Being clever about separators
-        '''
-        l = l_in
+        """
+        line = line_in
         row = []
-        while l != '':
-            match_item = self.field.match(l)
+        while line != "":
+            match_item = self.field.match(line)
             if match_item:
                 item = match_item.group(0)
                 row.append(item)
-                l = l[len(item):]
+                line = line[len(item) :]
             else:
-                match_intra = self.intra.match(l)
+                match_intra = self.intra.match(line)
                 if match_intra:
                     intra = match_intra.group(0)
                     # Count the separators
@@ -107,26 +137,29 @@ class ResultParser:
                     sep_count = len(tabbed) - 1
                     if sep_count == 0:
                         sep_count = 1
-                    for i in range(sep_count-1):
-                        row.append('')
-                    l = l[len(intra):]
+                    for _ in range(sep_count - 1):
+                        row.append("")
+                    line = line[len(intra) :]
                 else:
                     raise BenchmarkResultException(
-                        'Invalid TSV line', f"{l_in} at {l}")
+                        "Invalid TSV line", f"{line_in} at {line}"
+                    )
         return row
 
     def parse(self, lines):
-        '''Parse something that iterates lines'''
+        """Parse something that iterates lines"""
         rows = [self.line(line) for line in lines if not self.ignore(line)]
         header = rows[0]
         width = len(header)
-        records = [{k: v for (k, v) in itertools.zip_longest(
-            header, row[:width])} for row in rows[1:]]
+        records = [
+            {k: v for (k, v) in itertools.zip_longest(header, row[:width])}
+            for row in rows[1:]
+        ]
         return records
 
 
 def load_report_from_tsv(filename: str):
-    file = open(filename, 'r')
+    file = open(filename, "r")
     contents = file.readlines()
     file.close()
     parser = ResultParser()
@@ -136,52 +169,70 @@ def load_report_from_tsv(filename: str):
 
 
 def push_report_to_opensearch(report, esdocument):
-    sanitized = [BenchmarkUtils.conform_opensearch(row)
-                 for row in report if BenchmarkUtils.sanity_check(row)]
-    logging.debug(f"upload {len(sanitized)} sane of {len(report)} benchmarks to opensearch")
+    sanitized = [
+        BenchmarkUtils.conform_opensearch(row)
+        for row in report
+        if BenchmarkUtils.sanity_check(row)
+    ]
+    logging.debug(
+        f"upload {len(sanitized)} sane of {len(report)} benchmarks to opensearch"
+    )
     for single_benchmark in sanitized:
         logging.debug(f"upload benchmark: {single_benchmark}")
         response = requests.post(
             esdocument,
-            json=single_benchmark, auth=(os.environ['ES_USER'], os.environ['ES_PASS']))
+            json=single_benchmark,
+            auth=(os.environ["ES_USER"], os.environ["ES_PASS"]),
+        )
         logging.debug(
-            f"Sent to OpenSearch, status: {response.status_code}, result: {response.text}")
+            f"Sent to OpenSearch, status: {response.status_code}, result: {response.text}"
+        )
         response.raise_for_status()
 
+
 def push_report_to_null(report):
-    
+
     for row in report:
         if BenchmarkUtils.sanity_check(row):
             logging.debug(f"row {row}")
             conformed = BenchmarkUtils.conform_opensearch(row)
             logging.debug(f"conformed row {conformed}")
 
+
 def main():
-    '''Tool for fetching, parsing and uploading benchmark results to OpenSearch / ElasticSearch
+    """Tool for fetching, parsing and uploading benchmark results to OpenSearch / ElasticSearch
     This tool will
 
     (1) Open a local tsv benchmark report file
     (2) Upload to OpenSearch document, via https/JSON
-    '''
+    """
 
-    parser = argparse.ArgumentParser(
-        description='CircleCI benchmark scraper.')
+    parser = argparse.ArgumentParser(description="CircleCI benchmark scraper.")
 
     # --tsvfile is the name of the file to read results from
     # --esdocument is the ElasticSearch document to push these results into
     #
-    parser.add_argument('--tsvfile', default='build_tools/circle_api_scraper_input.txt',
-                        help='File from which to read tsv report')
-    parser.add_argument('--esdocument', help='ElasticSearch/OpenSearch document URL to upload report into')
-    parser.add_argument('--upload', choices=['opensearch', 'none'], default='opensearch')
+    parser.add_argument(
+        "--tsvfile",
+        default="build_tools/circle_api_scraper_input.txt",
+        help="File from which to read tsv report",
+    )
+    parser.add_argument(
+        "--esdocument",
+        help="ElasticSearch/OpenSearch document URL to upload report into",
+    )
+    parser.add_argument(
+        "--upload", choices=["opensearch", "none"], default="opensearch"
+    )
 
     args = parser.parse_args()
     logging.debug(f"Arguments: {args}")
     reports = load_report_from_tsv(args.tsvfile)
-    if (args.upload == 'opensearch'):
+    if args.upload == "opensearch":
         push_report_to_opensearch(reports, args.esdocument)
     else:
         push_report_to_null(reports)
 
-if __name__ == '__main__':
+
+if __name__ == "__main__":
     sys.exit(main())

build_tools/build_detect_platform

@@ -277,6 +277,7 @@ if [ "$CROSS_COMPILE" = "true" -o "$FBCODE_BUILD" = "true" ]; then
     if [ "$FBCODE_BUILD" = "true" ]; then
       # Enable backtrace on fbcode since the necessary libraries are present
       COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+      FOLLY_DIR="third-party/folly"
     fi
     true
 else
@@ -601,6 +602,17 @@ EOF
             PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbenchmark"
         fi
     fi
+    if test $USE_FOLLY; then
+        # Test whether libfolly library is installed
+        $CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <folly/synchronization/DistributedMutex.h>
+          int main() {}
+EOF
+        if [ "$?" != 0 ]; then
+          FOLLY_DIR="./third-party/folly"
+        fi
+    fi
+
 fi
 
 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -827,6 +839,13 @@ fi
 
 rm -f test.o test_dl.o
 
+# Get the path for the folly installation dir
+if [ "$USE_FOLLY" ]; then
+  if [ "$FOLLY_DIR" ]; then
+    FOLLY_PATH=`cd $FOLLY_DIR && $PYTHON build/fbcode_builder/getdeps.py show-inst-dir folly`
+  fi
+fi
+
 PLATFORM_CCFLAGS="$PLATFORM_CCFLAGS $COMMON_FLAGS"
 PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS $COMMON_FLAGS"
 
@@ -866,6 +885,8 @@ echo "CLANG_ANALYZER=$CLANG_ANALYZER" >> "$OUTPUT"
 echo "PROFILING_FLAGS=$PROFILING_FLAGS" >> "$OUTPUT"
 echo "FIND=$FIND" >> "$OUTPUT"
 echo "WATCH=$WATCH" >> "$OUTPUT"
+echo "FOLLY_PATH=$FOLLY_PATH" >> "$OUTPUT"
+
 # This will enable some related identifiers for the preprocessor
 if test -n "$JEMALLOC"; then
   echo "JEMALLOC=1" >> "$OUTPUT"

build_tools/dependencies_platform009.sh

@@ -19,8 +19,4 @@ BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/08634589372fa5f237bfd374e8c644a836
 VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/6ae525939ad02e5e676855082fbbc7828dbafeac/3.15.0/platform009/7f3b187
 LUA_BASE=/mnt/gvfs/third-party2/lua/162efd9561a3d21f6869f4814011e9cf1b3ff4dc/5.3.4/platform009/a6271c4
 BENCHMARK_BASE=/mnt/gvfs/third-party2/benchmark/30bf49ad6414325e17f3425b0edcb64239427ae3/1.6.1/platform009/7f3b187
-BOOST_BASE=/mnt/gvfs/third-party2/boost/201b7d74941e54b436dfa364a063aa6d2cd7de4c/1.69.0/platform009/8a7ffdf
-GLOG_BASE=/mnt/gvfs/third-party2/glog/32d751bd5673375b438158717ab6a57c1cc57e3d/0.3.2_fb/platform009/10a364d/
-FMT_BASE=/mnt/gvfs/third-party2/fmt/ce0c25f67165f4d2c22a29b8ef50f5600d7873ca/6.1.1/platform009/7f3b187/
-DBL_CONV_BASE=/mnt/gvfs/third-party2/double_conversion/109b3d9696d71f1048678cd7da1e22505470543d/20141126/platform009/7f3b187/
-LIBEVENT_BASE=/mnt/gvfs/third-party2/libevent/4a4d3a79a76c2439b6bd471bf3586b3481dde75e/1.4.14b_hphp/platform009/7f3b187/
+GLOG_BASE=/mnt/gvfs/third-party2/glog/32d751bd5673375b438158717ab6a57c1cc57e3d/0.3.2_fb/platform009/10a364d

build_tools/error_filter.py

@@ -3,16 +3,13 @@
 #  COPYING file in the root directory) and Apache 2.0 License
 #  (found in the LICENSE.Apache file in the root directory).
 
-'''Filter for error messages in test output:
+"""Filter for error messages in test output:
     - Receives merged stdout/stderr from test on stdin
     - Finds patterns of known error messages for test name (first argument)
     - Prints those error messages to stdout
-'''
+"""
 
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
+from __future__ import absolute_import, division, print_function, unicode_literals
 
 import re
 import sys
@@ -20,23 +17,24 @@ import sys
 
 class ErrorParserBase(object):
     def parse_error(self, line):
-        '''Parses a line of test output. If it contains an error, returns a
+        """Parses a line of test output. If it contains an error, returns a
         formatted message describing the error; otherwise, returns None.
         Subclasses must override this method.
-        '''
+        """
         raise NotImplementedError
 
 
 class GTestErrorParser(ErrorParserBase):
-    '''A parser that remembers the last test that began running so it can print
+    """A parser that remembers the last test that began running so it can print
     that test's name upon detecting failure.
-    '''
-    _GTEST_NAME_PATTERN = re.compile(r'\[ RUN      \] (\S+)$')
+    """
+
+    _GTEST_NAME_PATTERN = re.compile(r"\[ RUN      \] (\S+)$")
     # format: '<filename or "unknown file">:<line #>: Failure'
-    _GTEST_FAIL_PATTERN = re.compile(r'(unknown file|\S+:\d+): Failure$')
+    _GTEST_FAIL_PATTERN = re.compile(r"(unknown file|\S+:\d+): Failure$")
 
     def __init__(self):
-        self._last_gtest_name = 'Unknown test'
+        self._last_gtest_name = "Unknown test"
 
     def parse_error(self, line):
         gtest_name_match = self._GTEST_NAME_PATTERN.match(line)
@@ -45,14 +43,13 @@ class GTestErrorParser(ErrorParserBase):
             return None
         gtest_fail_match = self._GTEST_FAIL_PATTERN.match(line)
         if gtest_fail_match:
-            return '%s failed: %s' % (
-                    self._last_gtest_name, gtest_fail_match.group(1))
+            return "%s failed: %s" % (self._last_gtest_name, gtest_fail_match.group(1))
         return None
 
 
 class MatchErrorParser(ErrorParserBase):
-    '''A simple parser that returns the whole line if it matches the pattern.
-    '''
+    """A simple parser that returns the whole line if it matches the pattern."""
+
     def __init__(self, pattern):
         self._pattern = re.compile(pattern)
 
@@ -69,97 +66,104 @@ class CompilerErrorParser(MatchErrorParser):
         # format (link error):
         #   '<filename>:<line #>: error: <error msg>'
         # The below regex catches both
-        super(CompilerErrorParser, self).__init__(r'\S+:\d+: error:')
+        super(CompilerErrorParser, self).__init__(r"\S+:\d+: error:")
 
 
 class ScanBuildErrorParser(MatchErrorParser):
     def __init__(self):
-        super(ScanBuildErrorParser, self).__init__(
-                r'scan-build: \d+ bugs found.$')
+        super(ScanBuildErrorParser, self).__init__(r"scan-build: \d+ bugs found.$")
 
 
 class DbCrashErrorParser(MatchErrorParser):
     def __init__(self):
-        super(DbCrashErrorParser, self).__init__(r'\*\*\*.*\^$|TEST FAILED.')
+        super(DbCrashErrorParser, self).__init__(r"\*\*\*.*\^$|TEST FAILED.")
 
 
 class WriteStressErrorParser(MatchErrorParser):
     def __init__(self):
         super(WriteStressErrorParser, self).__init__(
-                r'ERROR: write_stress died with exitcode=\d+')
+            r"ERROR: write_stress died with exitcode=\d+"
+        )
 
 
 class AsanErrorParser(MatchErrorParser):
     def __init__(self):
-        super(AsanErrorParser, self).__init__(
-                r'==\d+==ERROR: AddressSanitizer:')
+        super(AsanErrorParser, self).__init__(r"==\d+==ERROR: AddressSanitizer:")
 
 
 class UbsanErrorParser(MatchErrorParser):
     def __init__(self):
         # format: '<filename>:<line #>:<column #>: runtime error: <error msg>'
-        super(UbsanErrorParser, self).__init__(r'\S+:\d+:\d+: runtime error:')
+        super(UbsanErrorParser, self).__init__(r"\S+:\d+:\d+: runtime error:")
 
 
 class ValgrindErrorParser(MatchErrorParser):
     def __init__(self):
         # just grab the summary, valgrind doesn't clearly distinguish errors
         # from other log messages.
-        super(ValgrindErrorParser, self).__init__(r'==\d+== ERROR SUMMARY:')
+        super(ValgrindErrorParser, self).__init__(r"==\d+== ERROR SUMMARY:")
 
 
 class CompatErrorParser(MatchErrorParser):
     def __init__(self):
-        super(CompatErrorParser, self).__init__(r'==== .*[Ee]rror.* ====$')
+        super(CompatErrorParser, self).__init__(r"==== .*[Ee]rror.* ====$")
 
 
 class TsanErrorParser(MatchErrorParser):
     def __init__(self):
-        super(TsanErrorParser, self).__init__(r'WARNING: ThreadSanitizer:')
+        super(TsanErrorParser, self).__init__(r"WARNING: ThreadSanitizer:")
 
 
 _TEST_NAME_TO_PARSERS = {
-    'punit': [CompilerErrorParser, GTestErrorParser],
-    'unit': [CompilerErrorParser, GTestErrorParser],
-    'release': [CompilerErrorParser, GTestErrorParser],
-    'unit_481': [CompilerErrorParser, GTestErrorParser],
-    'release_481': [CompilerErrorParser, GTestErrorParser],
-    'clang_unit': [CompilerErrorParser, GTestErrorParser],
-    'clang_release': [CompilerErrorParser, GTestErrorParser],
-    'clang_analyze': [CompilerErrorParser, ScanBuildErrorParser],
-    'code_cov': [CompilerErrorParser, GTestErrorParser],
-    'unity': [CompilerErrorParser, GTestErrorParser],
-    'lite': [CompilerErrorParser],
-    'lite_test': [CompilerErrorParser, GTestErrorParser],
-    'stress_crash': [CompilerErrorParser, DbCrashErrorParser],
-    'stress_crash_with_atomic_flush': [CompilerErrorParser, DbCrashErrorParser],
-    'stress_crash_with_txn': [CompilerErrorParser, DbCrashErrorParser],
-    'write_stress': [CompilerErrorParser, WriteStressErrorParser],
-    'asan': [CompilerErrorParser, GTestErrorParser, AsanErrorParser],
-    'asan_crash': [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
-    'asan_crash_with_atomic_flush': [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
-    'asan_crash_with_txn': [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
-    'ubsan': [CompilerErrorParser, GTestErrorParser, UbsanErrorParser],
-    'ubsan_crash': [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
-    'ubsan_crash_with_atomic_flush': [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
-    'ubsan_crash_with_txn': [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
-    'valgrind': [CompilerErrorParser, GTestErrorParser, ValgrindErrorParser],
-    'tsan': [CompilerErrorParser, GTestErrorParser, TsanErrorParser],
-    'format_compatible': [CompilerErrorParser, CompatErrorParser],
-    'run_format_compatible': [CompilerErrorParser, CompatErrorParser],
-    'no_compression': [CompilerErrorParser, GTestErrorParser],
-    'run_no_compression': [CompilerErrorParser, GTestErrorParser],
-    'regression': [CompilerErrorParser],
-    'run_regression': [CompilerErrorParser],
+    "punit": [CompilerErrorParser, GTestErrorParser],
+    "unit": [CompilerErrorParser, GTestErrorParser],
+    "release": [CompilerErrorParser, GTestErrorParser],
+    "unit_481": [CompilerErrorParser, GTestErrorParser],
+    "release_481": [CompilerErrorParser, GTestErrorParser],
+    "clang_unit": [CompilerErrorParser, GTestErrorParser],
+    "clang_release": [CompilerErrorParser, GTestErrorParser],
+    "clang_analyze": [CompilerErrorParser, ScanBuildErrorParser],
+    "code_cov": [CompilerErrorParser, GTestErrorParser],
+    "unity": [CompilerErrorParser, GTestErrorParser],
+    "lite": [CompilerErrorParser],
+    "lite_test": [CompilerErrorParser, GTestErrorParser],
+    "stress_crash": [CompilerErrorParser, DbCrashErrorParser],
+    "stress_crash_with_atomic_flush": [CompilerErrorParser, DbCrashErrorParser],
+    "stress_crash_with_txn": [CompilerErrorParser, DbCrashErrorParser],
+    "write_stress": [CompilerErrorParser, WriteStressErrorParser],
+    "asan": [CompilerErrorParser, GTestErrorParser, AsanErrorParser],
+    "asan_crash": [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
+    "asan_crash_with_atomic_flush": [
+        CompilerErrorParser,
+        AsanErrorParser,
+        DbCrashErrorParser,
+    ],
+    "asan_crash_with_txn": [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
+    "ubsan": [CompilerErrorParser, GTestErrorParser, UbsanErrorParser],
+    "ubsan_crash": [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
+    "ubsan_crash_with_atomic_flush": [
+        CompilerErrorParser,
+        UbsanErrorParser,
+        DbCrashErrorParser,
+    ],
+    "ubsan_crash_with_txn": [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
+    "valgrind": [CompilerErrorParser, GTestErrorParser, ValgrindErrorParser],
+    "tsan": [CompilerErrorParser, GTestErrorParser, TsanErrorParser],
+    "format_compatible": [CompilerErrorParser, CompatErrorParser],
+    "run_format_compatible": [CompilerErrorParser, CompatErrorParser],
+    "no_compression": [CompilerErrorParser, GTestErrorParser],
+    "run_no_compression": [CompilerErrorParser, GTestErrorParser],
+    "regression": [CompilerErrorParser],
+    "run_regression": [CompilerErrorParser],
 }
 
 
 def main():
     if len(sys.argv) != 2:
-        return 'Usage: %s <test name>' % sys.argv[0]
+        return "Usage: %s <test name>" % sys.argv[0]
     test_name = sys.argv[1]
     if test_name not in _TEST_NAME_TO_PARSERS:
-        return 'Unknown test name: %s' % test_name
+        return "Unknown test name: %s" % test_name
 
     error_parsers = []
     for parser_cls in _TEST_NAME_TO_PARSERS[test_name]:
@@ -173,5 +177,5 @@ def main():
                 print(error_msg)
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     sys.exit(main())

build_tools/fbcode_config_platform009.sh

@@ -68,20 +68,9 @@ CFLAGS+=" -DGFLAGS=gflags"
 BENCHMARK_INCLUDE=" -I $BENCHMARK_BASE/include/"
 BENCHMARK_LIBS=" $BENCHMARK_BASE/lib/libbenchmark${MAYBE_PIC}.a"
 
-BOOST_INCLUDE=" -I $BOOST_BASE/include/"
-
 GLOG_INCLUDE=" -I $GLOG_BASE/include/"
 GLOG_LIBS=" $GLOG_BASE/lib/libglog${MAYBE_PIC}.a"
 
-FMT_INCLUDE=" -I $FMT_BASE/include/"
-FMT_LIBS=" $FMT_BASE/lib/libfmt${MAYBE_PIC}.a"
-
-DBL_CONV_INCLUDE=" -I $DBL_CONV_BASE/include/"
-DBL_CONV_LIBS=" $DBL_CONV_BASE/lib/libdouble-conversion${MAYBE_PIC}.a"
-
-LIBEVENT_INCLUDE=" -I $LIBEVENT_BASE/include/"
-LIBEVENT_LIBS=" $LIBEVENT_BASE/lib/libevent${MAYBE_PIC}.a"
-
 # location of jemalloc
 JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
 JEMALLOC_LIB=" $JEMALLOC_BASE/lib/libjemalloc${MAYBE_PIC}.a"
@@ -113,7 +102,7 @@ BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"
 AS="$BINUTILS/as"
 
-DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE $LIBURING_INCLUDE $BENCHMARK_INCLUDE $BOOST_INCLUDE $GLOG_INCLUDE $FMT_INCLUDE $DBL_CONV_INCLUDE $LIBEVENT_INCLUDE"
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE $LIBURING_INCLUDE $BENCHMARK_INCLUDE $GLOG_INCLUDE"
 
 STDLIBS="-L $GCC_BASE/lib64"
 

build_tools/gnu_parallel

@@ -1170,7 +1170,7 @@ sub parse_env_var {
     # Check if any variables contain \n
     if(my @v = map { s/BASH_FUNC_(.*)\(\)/$1/; $_ } grep { $ENV{$_}=~/\n/ } @vars) {
 	# \n is bad for csh and will cause it to fail.
-	$Global::envwarn = ::shell_quote_scalar(q{echo $SHELL | egrep "/t?csh" > /dev/null && echo CSH/TCSH DO NOT SUPPORT newlines IN VARIABLES/FUNCTIONS. Unset }."@v".q{ && exec false;}."\n\n") . $Global::envwarn;
+	$Global::envwarn = ::shell_quote_scalar(q{echo $SHELL | grep -E "/t?csh" > /dev/null && echo CSH/TCSH DO NOT SUPPORT newlines IN VARIABLES/FUNCTIONS. Unset }."@v".q{ && exec false;}."\n\n") . $Global::envwarn;
     }
 
     if(not @qcsh) { push @qcsh, "true"; }
@@ -1916,7 +1916,8 @@ sub drain_job_queue {
                 } elsif (not $ps_reported and (time() - $last_progress_time) >= 60) {
                     # No progress in at least 60 seconds: run ps
                     print $Global::original_stderr "\n";
-                    system("ps", "-wf");
+                    my $script_dir = ::dirname($0);
+                    system("$script_dir/ps_with_stack || ps -wwf");
                     $ps_reported = 1;
                 }
                 $last_left = $Global::left;

build_tools/ps_with_stack

@@ -0,0 +1,38 @@
+#!/usr/bin/env perl
+
+use strict;
+
+open(my $ps, "-|", "ps -wwf");
+my $cols_known = 0;
+my $cmd_col = 0;
+my $pid_col = 0;
+while (<$ps>) {
+  print;
+  my @cols = split(/\s+/);
+
+  if (!$cols_known && /CMD/) {
+    # Parse relevant ps column headers
+    for (my $i = 0; $i <= $#cols; $i++) {
+      if ($cols[$i] eq "CMD") {
+        $cmd_col = $i;
+      }
+      if ($cols[$i] eq "PID") {
+        $pid_col = $i;
+      }
+    }
+    $cols_known = 1;
+  } else {
+    my $pid = $cols[$pid_col];
+    my $cmd = $cols[$cmd_col];
+    # Match numeric PID and relative path command
+    # -> The intention is only to dump stack traces for hangs in code under
+    # test, which means we probably just built it and are executing by
+    # relative path (e.g. ./my_test or foo/bar_test) rather then by absolute
+    # path (e.g. /usr/bin/time) or PATH search (e.g. grep).
+    if ($pid =~ /^[0-9]+$/ && $cmd =~ /^[^\/ ]+[\/]/) {
+      print "Dumping stacks for $pid...\n";
+      system("pstack $pid || gdb -batch -p $pid -ex 'thread apply all bt'");
+    }
+  }
+}
+close $ps;

build_tools/ubuntu20_image/Dockerfile

@@ -16,6 +16,7 @@ RUN apt-get install -y libgflags-dev libtbb-dev
 RUN apt-get install -y libsnappy-dev zlib1g-dev libbz2-dev liblz4-dev libzstd-dev
 # install cmake
 RUN apt-get install -y cmake
+RUN apt-get install -y libssl-dev
 # install clang-13
 WORKDIR /root
 RUN wget https://apt.llvm.org/llvm.sh

cache/cache.cc

@@ -58,6 +58,11 @@ static std::unordered_map<std::string, OptionTypeInfo>
                    compress_format_version),
           OptionType::kUInt32T, OptionVerificationType::kNormal,
           OptionTypeFlags::kMutable}},
+        {"enable_custom_split_merge",
+         {offsetof(struct CompressedSecondaryCacheOptions,
+                   enable_custom_split_merge),
+          OptionType::kBoolean, OptionVerificationType::kNormal,
+          OptionTypeFlags::kMutable}},
 };
 #endif  // ROCKSDB_LITE
 

cache/cache_bench_tool.cc

@@ -13,8 +13,6 @@
 #include <set>
 #include <sstream>
 
-#include "cache/clock_cache.h"
-#include "cache/fast_lru_cache.h"
 #include "db/db_impl/db_impl.h"
 #include "monitoring/histogram.h"
 #include "port/port.h"
@@ -228,7 +226,7 @@ struct KeyGen {
   }
 };
 
-char* createValue(Random64& rnd) {
+Cache::ObjectPtr createValue(Random64& rnd) {
   char* rv = new char[FLAGS_value_bytes];
   // Fill with some filler data, and take some CPU time
   for (uint32_t i = 0; i < FLAGS_value_bytes; i += 8) {
@@ -238,28 +236,33 @@ char* createValue(Random64& rnd) {
 }
 
 // Callbacks for secondary cache
-size_t SizeFn(void* /*obj*/) { return FLAGS_value_bytes; }
+size_t SizeFn(Cache::ObjectPtr /*obj*/) { return FLAGS_value_bytes; }
 
-Status SaveToFn(void* obj, size_t /*offset*/, size_t size, void* out) {
-  memcpy(out, obj, size);
+Status SaveToFn(Cache::ObjectPtr from_obj, size_t /*from_offset*/,
+                size_t length, char* out) {
+  memcpy(out, from_obj, length);
   return Status::OK();
 }
 
-// Different deleters to simulate using deleter to gather
-// stats on the code origin and kind of cache entries.
-void deleter1(const Slice& /*key*/, void* value) {
-  delete[] static_cast<char*>(value);
-}
-void deleter2(const Slice& /*key*/, void* value) {
-  delete[] static_cast<char*>(value);
-}
-void deleter3(const Slice& /*key*/, void* value) {
+Status CreateFn(const Slice& data, Cache::CreateContext* /*context*/,
+                MemoryAllocator* /*allocator*/, Cache::ObjectPtr* out_obj,
+                size_t* out_charge) {
+  *out_obj = new char[data.size()];
+  memcpy(*out_obj, data.data(), data.size());
+  *out_charge = data.size();
+  return Status::OK();
+};
+
+void DeleteFn(Cache::ObjectPtr value, MemoryAllocator* /*alloc*/) {
   delete[] static_cast<char*>(value);
 }
 
-Cache::CacheItemHelper helper1(SizeFn, SaveToFn, deleter1);
-Cache::CacheItemHelper helper2(SizeFn, SaveToFn, deleter2);
-Cache::CacheItemHelper helper3(SizeFn, SaveToFn, deleter3);
+Cache::CacheItemHelper helper1(CacheEntryRole::kDataBlock, DeleteFn, SizeFn,
+                               SaveToFn, CreateFn);
+Cache::CacheItemHelper helper2(CacheEntryRole::kIndexBlock, DeleteFn, SizeFn,
+                               SaveToFn, CreateFn);
+Cache::CacheItemHelper helper3(CacheEntryRole::kFilterBlock, DeleteFn, SizeFn,
+                               SaveToFn, CreateFn);
 }  // namespace
 
 class CacheBench {
@@ -292,17 +295,12 @@ class CacheBench {
     }
 
     if (FLAGS_cache_type == "clock_cache") {
-      cache_ = ExperimentalNewClockCache(
-          FLAGS_cache_size, FLAGS_value_bytes, FLAGS_num_shard_bits,
-          false /*strict_capacity_limit*/, kDefaultCacheMetadataChargePolicy);
-      if (!cache_) {
-        fprintf(stderr, "Clock cache not supported.\n");
-        exit(1);
-      }
-    } else if (FLAGS_cache_type == "fast_lru_cache") {
-      cache_ = NewFastLRUCache(
-          FLAGS_cache_size, FLAGS_value_bytes, FLAGS_num_shard_bits,
-          false /*strict_capacity_limit*/, kDefaultCacheMetadataChargePolicy);
+      fprintf(stderr, "Old clock cache implementation has been removed.\n");
+      exit(1);
+    } else if (FLAGS_cache_type == "hyper_clock_cache") {
+      cache_ = HyperClockCacheOptions(FLAGS_cache_size, FLAGS_value_bytes,
+                                      FLAGS_num_shard_bits)
+                   .MakeSharedCache();
     } else if (FLAGS_cache_type == "lru_cache") {
       LRUCacheOptions opts(FLAGS_cache_size, FLAGS_num_shard_bits,
                            false /* strict_capacity_limit */,
@@ -441,7 +439,9 @@ class CacheBench {
     uint64_t total_key_size = 0;
     uint64_t total_charge = 0;
     uint64_t total_entry_count = 0;
-    std::set<Cache::DeleterFn> deleters;
+    uint64_t table_occupancy = 0;
+    uint64_t table_size = 0;
+    std::set<const Cache::CacheItemHelper*> helpers;
     StopWatchNano timer(clock);
 
     for (;;) {
@@ -456,6 +456,9 @@ class CacheBench {
             std::ostringstream ostr;
             ostr << "Most recent cache entry stats:\n"
                  << "Number of entries: " << total_entry_count << "\n"
+                 << "Table occupancy: " << table_occupancy << " / "
+                 << table_size << " = "
+                 << (100.0 * table_occupancy / table_size) << "%\n"
                  << "Total charge: " << BytesToHumanString(total_charge) << "\n"
                  << "Average key size: "
                  << (1.0 * total_key_size / total_entry_count) << "\n"
@@ -463,7 +466,7 @@ class CacheBench {
                  << BytesToHumanString(static_cast<uint64_t>(
                         1.0 * total_charge / total_entry_count))
                  << "\n"
-                 << "Unique deleters: " << deleters.size() << "\n";
+                 << "Unique helpers: " << helpers.size() << "\n";
             *stats_report = ostr.str();
             return;
           }
@@ -479,19 +482,21 @@ class CacheBench {
       total_key_size = 0;
       total_charge = 0;
       total_entry_count = 0;
-      deleters.clear();
-      auto fn = [&](const Slice& key, void* /*value*/, size_t charge,
-                    Cache::DeleterFn deleter) {
+      helpers.clear();
+      auto fn = [&](const Slice& key, Cache::ObjectPtr /*value*/, size_t charge,
+                    const Cache::CacheItemHelper* helper) {
         total_key_size += key.size();
         total_charge += charge;
         ++total_entry_count;
-        // Something slightly more expensive as in (future) stats by category
-        deleters.insert(deleter);
+        // Something slightly more expensive as in stats by category
+        helpers.insert(helper);
       };
       timer.Start();
       Cache::ApplyToAllEntriesOptions opts;
       opts.average_entries_per_lock = FLAGS_gather_stats_entries_per_lock;
       shared->GetCacheBench()->cache_->ApplyToAllEntries(fn, opts);
+      table_occupancy = shared->GetCacheBench()->cache_->GetOccupancyCount();
+      table_size = shared->GetCacheBench()->cache_->GetTableAddressCount();
       stats_hist->Add(timer.ElapsedNanos() / 1000);
     }
   }
@@ -533,14 +538,6 @@ class CacheBench {
     for (uint64_t i = 0; i < FLAGS_ops_per_thread; i++) {
       Slice key = gen.GetRand(thread->rnd, max_key_, max_log_);
       uint64_t random_op = thread->rnd.Next();
-      Cache::CreateCallback create_cb = [](const void* buf, size_t size,
-                                           void** out_obj,
-                                           size_t* charge) -> Status {
-        *out_obj = reinterpret_cast<void*>(new char[size]);
-        memcpy(*out_obj, buf, size);
-        *charge = size;
-        return Status::OK();
-      };
 
       timer.Start();
 
@@ -550,8 +547,8 @@ class CacheBench {
           handle = nullptr;
         }
         // do lookup
-        handle = cache_->Lookup(key, &helper2, create_cb, Cache::Priority::LOW,
-                                true);
+        handle = cache_->Lookup(key, &helper2, /*context*/ nullptr,
+                                Cache::Priority::LOW, true);
         if (handle) {
           if (!FLAGS_lean) {
             // do something with the data
@@ -579,8 +576,8 @@ class CacheBench {
           handle = nullptr;
         }
         // do lookup
-        handle = cache_->Lookup(key, &helper2, create_cb, Cache::Priority::LOW,
-                                true);
+        handle = cache_->Lookup(key, &helper2, /*context*/ nullptr,
+                                Cache::Priority::LOW, true);
         if (handle) {
           if (!FLAGS_lean) {
             // do something with the data

cache/cache_entry_roles.cc

@@ -23,6 +23,7 @@ std::array<std::string, kNumCacheEntryRoles> kCacheEntryRoleToCamelString{{
     "FilterConstruction",
     "BlockBasedTableReader",
     "FileMetadata",
+    "BlobValue",
     "BlobCache",
     "Misc",
 }};
@@ -39,6 +40,7 @@ std::array<std::string, kNumCacheEntryRoles> kCacheEntryRoleToHyphenString{{
     "filter-construction",
     "block-based-table-reader",
     "file-metadata",
+    "blob-value",
     "blob-cache",
     "misc",
 }};
@@ -99,34 +101,4 @@ std::string BlockCacheEntryStatsMapKeys::UsedPercent(CacheEntryRole role) {
   return GetPrefixedCacheEntryRoleName(kPrefix, role);
 }
 
-namespace {
-
-struct Registry {
-  std::mutex mutex;
-  UnorderedMap<Cache::DeleterFn, CacheEntryRole> role_map;
-  void Register(Cache::DeleterFn fn, CacheEntryRole role) {
-    std::lock_guard<std::mutex> lock(mutex);
-    role_map[fn] = role;
-  }
-  UnorderedMap<Cache::DeleterFn, CacheEntryRole> Copy() {
-    std::lock_guard<std::mutex> lock(mutex);
-    return role_map;
-  }
-};
-
-Registry& GetRegistry() {
-  STATIC_AVOID_DESTRUCTION(Registry, registry);
-  return registry;
-}
-
-}  // namespace
-
-void RegisterCacheDeleterRole(Cache::DeleterFn fn, CacheEntryRole role) {
-  GetRegistry().Register(fn, role);
-}
-
-UnorderedMap<Cache::DeleterFn, CacheEntryRole> CopyCacheDeleterRoleMap() {
-  return GetRegistry().Copy();
-}
-
 }  // namespace ROCKSDB_NAMESPACE

cache/cache_entry_roles.h

@@ -7,11 +7,8 @@
 
 #include <array>
 #include <cstdint>
-#include <memory>
-#include <type_traits>
 
 #include "rocksdb/cache.h"
-#include "util/hash_containers.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -20,84 +17,4 @@ extern std::array<std::string, kNumCacheEntryRoles>
 extern std::array<std::string, kNumCacheEntryRoles>
     kCacheEntryRoleToHyphenString;
 
-// To associate cache entries with their role, we use a hack on the
-// existing Cache interface. Because the deleter of an entry can authenticate
-// the code origin of an entry, we can elaborate the choice of deleter to
-// also encode role information, without inferring false role information
-// from entries not choosing to encode a role.
-//
-// The rest of this file is for handling mappings between deleters and
-// roles.
-
-// To infer a role from a deleter, the deleter must be registered. This
-// can be done "manually" with this function. This function is thread-safe,
-// and the registration mappings go into private but static storage. (Note
-// that DeleterFn is a function pointer, not std::function. Registrations
-// should not be too many.)
-void RegisterCacheDeleterRole(Cache::DeleterFn fn, CacheEntryRole role);
-
-// Gets a copy of the registered deleter -> role mappings. This is the only
-// function for reading the mappings made with RegisterCacheDeleterRole.
-// Why only this interface for reading?
-// * This function has to be thread safe, which could incur substantial
-// overhead. We should not pay this overhead for every deleter look-up.
-// * This is suitable for preparing for batch operations, like with
-// CacheEntryStatsCollector.
-// * The number of mappings should be sufficiently small (dozens).
-UnorderedMap<Cache::DeleterFn, CacheEntryRole> CopyCacheDeleterRoleMap();
-
-// ************************************************************** //
-// An automatic registration infrastructure. This enables code
-// to simply ask for a deleter associated with a particular type
-// and role, and registration is automatic. In a sense, this is
-// a small dependency injection infrastructure, because linking
-// in new deleter instantiations is essentially sufficient for
-// making stats collection (using CopyCacheDeleterRoleMap) aware
-// of them.
-
-namespace cache_entry_roles_detail {
-
-template <typename T, CacheEntryRole R>
-struct RegisteredDeleter {
-  RegisteredDeleter() { RegisterCacheDeleterRole(Delete, R); }
-
-  // These have global linkage to help ensure compiler optimizations do not
-  // break uniqueness for each <T,R>
-  static void Delete(const Slice& /* key */, void* value) {
-    // Supports T == Something[], unlike delete operator
-    std::default_delete<T>()(
-        static_cast<typename std::remove_extent<T>::type*>(value));
-  }
-};
-
-template <CacheEntryRole R>
-struct RegisteredNoopDeleter {
-  RegisteredNoopDeleter() { RegisterCacheDeleterRole(Delete, R); }
-
-  static void Delete(const Slice& /* key */, void* /* value */) {
-    // Here was `assert(value == nullptr);` but we can also put pointers
-    // to static data in Cache, for testing at least.
-  }
-};
-
-}  // namespace cache_entry_roles_detail
-
-// Get an automatically registered deleter for value type T and role R.
-// Based on C++ semantics, registration is invoked exactly once in a
-// thread-safe way on first call to this function, for each <T, R>.
-template <typename T, CacheEntryRole R>
-Cache::DeleterFn GetCacheEntryDeleterForRole() {
-  static cache_entry_roles_detail::RegisteredDeleter<T, R> reg;
-  return reg.Delete;
-}
-
-// Get an automatically registered no-op deleter (value should be nullptr)
-// and associated with role R. This is used for Cache "reservation" entries
-// such as for WriteBufferManager.
-template <CacheEntryRole R>
-Cache::DeleterFn GetNoopDeleterForRole() {
-  static cache_entry_roles_detail::RegisteredNoopDeleter<R> reg;
-  return reg.Delete;
-}
-
 }  // namespace ROCKSDB_NAMESPACE

cache/cache_entry_stats.h

@@ -10,8 +10,8 @@
 #include <memory>
 #include <mutex>
 
-#include "cache/cache_helpers.h"
 #include "cache/cache_key.h"
+#include "cache/typed_cache.h"
 #include "port/lang.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/status.h"
@@ -111,11 +111,14 @@ class CacheEntryStatsCollector {
   // Gets or creates a shared instance of CacheEntryStatsCollector in the
   // cache itself, and saves into `ptr`. This shared_ptr will hold the
   // entry in cache until all refs are destroyed.
-  static Status GetShared(Cache *cache, SystemClock *clock,
+  static Status GetShared(Cache *raw_cache, SystemClock *clock,
                           std::shared_ptr<CacheEntryStatsCollector> *ptr) {
-    const Slice &cache_key = GetCacheKey();
+    assert(raw_cache);
+    BasicTypedCacheInterface<CacheEntryStatsCollector, CacheEntryRole::kMisc>
+        cache{raw_cache};
 
-    Cache::Handle *h = cache->Lookup(cache_key);
+    const Slice &cache_key = GetCacheKey();
+    auto h = cache.Lookup(cache_key);
     if (h == nullptr) {
       // Not yet in cache, but Cache doesn't provide a built-in way to
       // avoid racing insert. So we double-check under a shared mutex,
@@ -123,15 +126,15 @@ class CacheEntryStatsCollector {
       STATIC_AVOID_DESTRUCTION(std::mutex, static_mutex);
       std::lock_guard<std::mutex> lock(static_mutex);
 
-      h = cache->Lookup(cache_key);
+      h = cache.Lookup(cache_key);
       if (h == nullptr) {
-        auto new_ptr = new CacheEntryStatsCollector(cache, clock);
+        auto new_ptr = new CacheEntryStatsCollector(cache.get(), clock);
         // TODO: non-zero charge causes some tests that count block cache
         // usage to go flaky. Fix the problem somehow so we can use an
         // accurate charge.
         size_t charge = 0;
-        Status s = cache->Insert(cache_key, new_ptr, charge, Deleter, &h,
-                                 Cache::Priority::HIGH);
+        Status s =
+            cache.Insert(cache_key, new_ptr, charge, &h, Cache::Priority::HIGH);
         if (!s.ok()) {
           assert(h == nullptr);
           delete new_ptr;
@@ -140,11 +143,11 @@ class CacheEntryStatsCollector {
       }
     }
     // If we reach here, shared entry is in cache with handle `h`.
-    assert(cache->GetDeleter(h) == Deleter);
+    assert(cache.get()->GetCacheItemHelper(h) == &cache.kBasicHelper);
 
     // Build an aliasing shared_ptr that keeps `ptr` in cache while there
     // are references.
-    *ptr = MakeSharedCacheHandleGuard<CacheEntryStatsCollector>(cache, h);
+    *ptr = cache.SharedGuard(h);
     return Status::OK();
   }
 
@@ -157,10 +160,6 @@ class CacheEntryStatsCollector {
         cache_(cache),
         clock_(clock) {}
 
-  static void Deleter(const Slice &, void *value) {
-    delete static_cast<CacheEntryStatsCollector *>(value);
-  }
-
   static const Slice &GetCacheKey() {
     // For each template instantiation
     static CacheKey ckey = CacheKey::CreateUniqueForProcessLifetime();

cache/cache_helpers.cc

@@ -0,0 +1,40 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "cache/cache_helpers.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+void ReleaseCacheHandleCleanup(void* arg1, void* arg2) {
+  Cache* const cache = static_cast<Cache*>(arg1);
+  assert(cache);
+
+  Cache::Handle* const cache_handle = static_cast<Cache::Handle*>(arg2);
+  assert(cache_handle);
+
+  cache->Release(cache_handle);
+}
+
+Status WarmInCache(Cache* cache, const Slice& key, const Slice& saved,
+                   Cache::CreateContext* create_context,
+                   const Cache::CacheItemHelper* helper,
+                   Cache::Priority priority, size_t* out_charge) {
+  assert(helper);
+  assert(helper->create_cb);
+  Cache::ObjectPtr value;
+  size_t charge;
+  Status st = helper->create_cb(saved, create_context,
+                                cache->memory_allocator(), &value, &charge);
+  if (st.ok()) {
+    st =
+        cache->Insert(key, value, helper, charge, /*handle*/ nullptr, priority);
+    if (out_charge) {
+      *out_charge = charge;
+    }
+  }
+  return st;
+}
+
+}  // namespace ROCKSDB_NAMESPACE

cache/cache_helpers.h

@@ -17,22 +17,17 @@ template <typename T>
 T* GetFromCacheHandle(Cache* cache, Cache::Handle* handle) {
   assert(cache);
   assert(handle);
-
   return static_cast<T*>(cache->Value(handle));
 }
 
-// Simple generic deleter for Cache (to be used with Cache::Insert).
-template <typename T>
-void DeleteCacheEntry(const Slice& /* key */, void* value) {
-  delete static_cast<T*>(value);
-}
-
 // Turns a T* into a Slice so it can be used as a key with Cache.
 template <typename T>
-Slice GetSlice(const T* t) {
+Slice GetSliceForKey(const T* t) {
   return Slice(reinterpret_cast<const char*>(t), sizeof(T));
 }
 
+void ReleaseCacheHandleCleanup(void* arg1, void* arg2);
+
 // Generic resource management object for cache handles that releases the handle
 // when destroyed. Has unique ownership of the handle, so copying it is not
 // allowed, while moving it transfers ownership.
@@ -88,7 +83,7 @@ class CacheHandleGuard {
     if (cleanable) {
       if (handle_ != nullptr) {
         assert(cache_);
-        cleanable->RegisterCleanup(&ReleaseCacheHandle, cache_, handle_);
+        cleanable->RegisterCleanup(&ReleaseCacheHandleCleanup, cache_, handle_);
       }
     }
     ResetFields();
@@ -115,16 +110,6 @@ class CacheHandleGuard {
     value_ = nullptr;
   }
 
-  static void ReleaseCacheHandle(void* arg1, void* arg2) {
-    Cache* const cache = static_cast<Cache*>(arg1);
-    assert(cache);
-
-    Cache::Handle* const cache_handle = static_cast<Cache::Handle*>(arg2);
-    assert(cache_handle);
-
-    cache->Release(cache_handle);
-  }
-
  private:
   Cache* cache_ = nullptr;
   Cache::Handle* handle_ = nullptr;
@@ -139,7 +124,16 @@ template <typename T>
 std::shared_ptr<T> MakeSharedCacheHandleGuard(Cache* cache,
                                               Cache::Handle* handle) {
   auto wrapper = std::make_shared<CacheHandleGuard<T>>(cache, handle);
-  return std::shared_ptr<T>(wrapper, static_cast<T*>(cache->Value(handle)));
+  return std::shared_ptr<T>(wrapper, GetFromCacheHandle<T>(cache, handle));
 }
 
+// Given the persistable data (saved) for a block cache entry, parse that
+// into a cache entry object and insert it into the given cache. The charge
+// of the new entry can be returned to the caller through `out_charge`.
+Status WarmInCache(Cache* cache, const Slice& key, const Slice& saved,
+                   Cache::CreateContext* create_context,
+                   const Cache::CacheItemHelper* helper,
+                   Cache::Priority priority = Cache::Priority::LOW,
+                   size_t* out_charge = nullptr);
+
 }  // namespace ROCKSDB_NAMESPACE

cache/cache_reservation_manager.cc

@@ -13,7 +13,6 @@
 #include <cstring>
 #include <memory>
 
-#include "cache/cache_entry_roles.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
@@ -41,17 +40,17 @@ CacheReservationManagerImpl<
 template <CacheEntryRole R>
 CacheReservationManagerImpl<R>::CacheReservationManagerImpl(
     std::shared_ptr<Cache> cache, bool delayed_decrease)
-    : delayed_decrease_(delayed_decrease),
+    : cache_(cache),
+      delayed_decrease_(delayed_decrease),
       cache_allocated_size_(0),
       memory_used_(0) {
   assert(cache != nullptr);
-  cache_ = cache;
 }
 
 template <CacheEntryRole R>
 CacheReservationManagerImpl<R>::~CacheReservationManagerImpl() {
   for (auto* handle : dummy_handles_) {
-    cache_->Release(handle, true);
+    cache_.ReleaseAndEraseIfLastRef(handle);
   }
 }
 
@@ -115,8 +114,7 @@ Status CacheReservationManagerImpl<R>::IncreaseCacheReservation(
   Status return_status = Status::OK();
   while (new_mem_used > cache_allocated_size_.load(std::memory_order_relaxed)) {
     Cache::Handle* handle = nullptr;
-    return_status = cache_->Insert(GetNextCacheKey(), nullptr, kSizeDummyEntry,
-                                   GetNoopDeleterForRole<R>(), &handle);
+    return_status = cache_.Insert(GetNextCacheKey(), kSizeDummyEntry, &handle);
 
     if (return_status != Status::OK()) {
       return return_status;
@@ -141,7 +139,7 @@ Status CacheReservationManagerImpl<R>::DecreaseCacheReservation(
          cache_allocated_size_.load(std::memory_order_relaxed)) {
     assert(!dummy_handles_.empty());
     auto* handle = dummy_handles_.back();
-    cache_->Release(handle, true);
+    cache_.ReleaseAndEraseIfLastRef(handle);
     dummy_handles_.pop_back();
     cache_allocated_size_ -= kSizeDummyEntry;
   }
@@ -169,8 +167,9 @@ Slice CacheReservationManagerImpl<R>::GetNextCacheKey() {
 }
 
 template <CacheEntryRole R>
-Cache::DeleterFn CacheReservationManagerImpl<R>::TEST_GetNoopDeleterForRole() {
-  return GetNoopDeleterForRole<R>();
+const Cache::CacheItemHelper*
+CacheReservationManagerImpl<R>::TEST_GetCacheItemHelperForRole() {
+  return &CacheInterface::kHelper;
 }
 
 template class CacheReservationManagerImpl<

cache/cache_reservation_manager.h

@@ -18,7 +18,7 @@
 
 #include "cache/cache_entry_roles.h"
 #include "cache/cache_key.h"
-#include "rocksdb/cache.h"
+#include "cache/typed_cache.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
 #include "util/coding.h"
@@ -197,10 +197,10 @@ class CacheReservationManagerImpl
 
   static constexpr std::size_t GetDummyEntrySize() { return kSizeDummyEntry; }
 
-  // For testing only - it is to help ensure the NoopDeleterForRole<R>
+  // For testing only - it is to help ensure the CacheItemHelperForRole<R>
   // accessed from CacheReservationManagerImpl and the one accessed from the
   // test are from the same translation units
-  static Cache::DeleterFn TEST_GetNoopDeleterForRole();
+  static const Cache::CacheItemHelper *TEST_GetCacheItemHelperForRole();
 
  private:
   static constexpr std::size_t kSizeDummyEntry = 256 * 1024;
@@ -211,7 +211,8 @@ class CacheReservationManagerImpl
   Status IncreaseCacheReservation(std::size_t new_mem_used);
   Status DecreaseCacheReservation(std::size_t new_mem_used);
 
-  std::shared_ptr<Cache> cache_;
+  using CacheInterface = PlaceholderSharedCacheInterface<R>;
+  CacheInterface cache_;
   bool delayed_decrease_;
   std::atomic<std::size_t> cache_allocated_size_;
   std::size_t memory_used_;

cache/cache_reservation_manager_test.cc

@@ -463,6 +463,7 @@ TEST(CacheReservationHandleTest, HandleTest) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

cache/cache_test.cc

@@ -15,16 +15,15 @@
 #include <string>
 #include <vector>
 
-#include "cache/clock_cache.h"
-#include "cache/fast_lru_cache.h"
 #include "cache/lru_cache.h"
+#include "cache/typed_cache.h"
 #include "port/stack_trace.h"
 #include "test_util/testharness.h"
 #include "util/coding.h"
 #include "util/string_util.h"
 
-// FastLRUCache and ClockCache only support 16-byte keys, so some of
-// the tests originally wrote for LRUCache do not work on the other caches.
+// HyperClockCache only supports 16-byte keys, so some of the tests
+// originally written for LRUCache do not work on the other caches.
 // Those tests were adapted to use 16-byte keys. We kept the original ones.
 // TODO: Remove the original tests if they ever become unused.
 
@@ -57,27 +56,34 @@ int DecodeKey32Bits(const Slice& k) {
   return DecodeFixed32(k.data());
 }
 
-void* EncodeValue(uintptr_t v) { return reinterpret_cast<void*>(v); }
+Cache::ObjectPtr EncodeValue(uintptr_t v) {
+  return reinterpret_cast<Cache::ObjectPtr>(v);
+}
 
 int DecodeValue(void* v) {
   return static_cast<int>(reinterpret_cast<uintptr_t>(v));
 }
 
-void DumbDeleter(const Slice& /*key*/, void* /*value*/) {}
+const Cache::CacheItemHelper kDumbHelper{
+    CacheEntryRole::kMisc,
+    [](Cache::ObjectPtr /*value*/, MemoryAllocator* /*alloc*/) {}};
 
-void EraseDeleter1(const Slice& /*key*/, void* value) {
-  Cache* cache = reinterpret_cast<Cache*>(value);
-  cache->Erase("foo");
-}
+const Cache::CacheItemHelper kEraseOnDeleteHelper1{
+    CacheEntryRole::kMisc,
+    [](Cache::ObjectPtr value, MemoryAllocator* /*alloc*/) {
+      Cache* cache = static_cast<Cache*>(value);
+      cache->Erase("foo");
+    }};
 
-void EraseDeleter2(const Slice& /*key*/, void* value) {
-  Cache* cache = reinterpret_cast<Cache*>(value);
-  cache->Erase(EncodeKey16Bytes(1234));
-}
+const Cache::CacheItemHelper kEraseOnDeleteHelper2{
+    CacheEntryRole::kMisc,
+    [](Cache::ObjectPtr value, MemoryAllocator* /*alloc*/) {
+      Cache* cache = static_cast<Cache*>(value);
+      cache->Erase(EncodeKey16Bytes(1234));
+    }};
 
 const std::string kLRU = "lru";
-const std::string kClock = "clock";
-const std::string kFast = "fast";
+const std::string kHyperClock = "hyper_clock";
 
 }  // anonymous namespace
 
@@ -86,14 +92,11 @@ class CacheTest : public testing::TestWithParam<std::string> {
   static CacheTest* current_;
   static std::string type_;
 
-  static void Deleter(const Slice& key, void* v) {
-    if (type_ == kFast || type_ == kClock) {
-      current_->deleted_keys_.push_back(DecodeKey16Bytes(key));
-    } else {
-      current_->deleted_keys_.push_back(DecodeKey32Bits(key));
-    }
+  static void Deleter(Cache::ObjectPtr v, MemoryAllocator*) {
     current_->deleted_values_.push_back(DecodeValue(v));
   }
+  static constexpr Cache::CacheItemHelper kHelper{CacheEntryRole::kMisc,
+                                                  &Deleter};
 
   static const int kCacheSize = 1000;
   static const int kNumShardBits = 4;
@@ -101,11 +104,12 @@ class CacheTest : public testing::TestWithParam<std::string> {
   static const int kCacheSize2 = 100;
   static const int kNumShardBits2 = 2;
 
-  std::vector<int> deleted_keys_;
   std::vector<int> deleted_values_;
   std::shared_ptr<Cache> cache_;
   std::shared_ptr<Cache> cache2_;
 
+  size_t estimated_value_size_ = 1;
+
   CacheTest()
       : cache_(NewCache(kCacheSize, kNumShardBits, false)),
         cache2_(NewCache(kCacheSize2, kNumShardBits2, false)) {
@@ -120,15 +124,10 @@ class CacheTest : public testing::TestWithParam<std::string> {
     if (type == kLRU) {
       return NewLRUCache(capacity);
     }
-    if (type == kClock) {
-      return ExperimentalNewClockCache(
-          capacity, 1 /*estimated_value_size*/, -1 /*num_shard_bits*/,
-          false /*strict_capacity_limit*/, kDefaultCacheMetadataChargePolicy);
-    }
-    if (type == kFast) {
-      return NewFastLRUCache(
-          capacity, 1 /*estimated_value_size*/, -1 /*num_shard_bits*/,
-          false /*strict_capacity_limit*/, kDefaultCacheMetadataChargePolicy);
+    if (type == kHyperClock) {
+      return HyperClockCacheOptions(
+                 capacity, estimated_value_size_ /*estimated_value_size*/)
+          .MakeSharedCache();
     }
     return nullptr;
   }
@@ -146,27 +145,22 @@ class CacheTest : public testing::TestWithParam<std::string> {
       co.metadata_charge_policy = charge_policy;
       return NewLRUCache(co);
     }
-    if (type == kClock) {
-      return ExperimentalNewClockCache(capacity, 1 /*estimated_value_size*/,
-                                       num_shard_bits, strict_capacity_limit,
-                                       charge_policy);
-    }
-    if (type == kFast) {
-      return NewFastLRUCache(capacity, 1 /*estimated_value_size*/,
-                             num_shard_bits, strict_capacity_limit,
-                             charge_policy);
+    if (type == kHyperClock) {
+      return HyperClockCacheOptions(capacity, 1 /*estimated_value_size*/,
+                                    num_shard_bits, strict_capacity_limit,
+                                    nullptr /*allocator*/, charge_policy)
+          .MakeSharedCache();
     }
     return nullptr;
   }
 
   // These functions encode/decode keys in tests cases that use
   // int keys.
-  // Currently, FastLRUCache requires keys to be 16B long, whereas
-  // LRUCache and ClockCache don't, so the encoding depends on
-  // the cache type.
+  // Currently, HyperClockCache requires keys to be 16B long, whereas
+  // LRUCache doesn't, so the encoding depends on the cache type.
   std::string EncodeKey(int k) {
     auto type = GetParam();
-    if (type == kFast || type == kClock) {
+    if (type == kHyperClock) {
       return EncodeKey16Bytes(k);
     } else {
       return EncodeKey32Bits(k);
@@ -175,7 +169,7 @@ class CacheTest : public testing::TestWithParam<std::string> {
 
   int DecodeKey(const Slice& k) {
     auto type = GetParam();
-    if (type == kFast || type == kClock) {
+    if (type == kHyperClock) {
       return DecodeKey16Bytes(k);
     } else {
       return DecodeKey32Bits(k);
@@ -193,37 +187,29 @@ class CacheTest : public testing::TestWithParam<std::string> {
 
   void Insert(std::shared_ptr<Cache> cache, int key, int value,
               int charge = 1) {
-    EXPECT_OK(cache->Insert(EncodeKey(key), EncodeValue(value), charge,
-                            &CacheTest::Deleter));
+    EXPECT_OK(
+        cache->Insert(EncodeKey(key), EncodeValue(value), &kHelper, charge));
   }
 
   void Erase(std::shared_ptr<Cache> cache, int key) {
     cache->Erase(EncodeKey(key));
   }
 
-  int Lookup(int key) {
-    return Lookup(cache_, key);
-  }
+  int Lookup(int key) { return Lookup(cache_, key); }
 
   void Insert(int key, int value, int charge = 1) {
     Insert(cache_, key, value, charge);
   }
 
-  void Erase(int key) {
-    Erase(cache_, key);
-  }
+  void Erase(int key) { Erase(cache_, key); }
 
-  int Lookup2(int key) {
-    return Lookup(cache2_, key);
-  }
+  int Lookup2(int key) { return Lookup(cache2_, key); }
 
   void Insert2(int key, int value, int charge = 1) {
     Insert(cache2_, key, value, charge);
   }
 
-  void Erase2(int key) {
-    Erase(cache2_, key);
-  }
+  void Erase2(int key) { Erase(cache2_, key); }
 };
 
 CacheTest* CacheTest::current_;
@@ -239,7 +225,10 @@ TEST_P(CacheTest, UsageTest) {
   auto cache = NewCache(kCapacity, 8, false, kDontChargeCacheMetadata);
   auto precise_cache = NewCache(kCapacity, 0, false, kFullChargeCacheMetadata);
   ASSERT_EQ(0, cache->GetUsage());
-  ASSERT_EQ(0, precise_cache->GetUsage());
+  size_t baseline_meta_usage = precise_cache->GetUsage();
+  if (type != kHyperClock) {
+    ASSERT_EQ(0, baseline_meta_usage);
+  }
 
   size_t usage = 0;
   char value[10] = "abcdef";
@@ -252,19 +241,21 @@ TEST_P(CacheTest, UsageTest) {
       key = EncodeKey(i);
     }
     auto kv_size = key.size() + 5;
-    ASSERT_OK(cache->Insert(key, reinterpret_cast<void*>(value), kv_size,
-                            DumbDeleter));
-    ASSERT_OK(precise_cache->Insert(key, reinterpret_cast<void*>(value),
-                                    kv_size, DumbDeleter));
+    ASSERT_OK(cache->Insert(key, value, &kDumbHelper, kv_size));
+    ASSERT_OK(precise_cache->Insert(key, value, &kDumbHelper, kv_size));
     usage += kv_size;
     ASSERT_EQ(usage, cache->GetUsage());
-    ASSERT_LT(usage, precise_cache->GetUsage());
+    if (type == kHyperClock) {
+      ASSERT_EQ(baseline_meta_usage + usage, precise_cache->GetUsage());
+    } else {
+      ASSERT_LT(usage, precise_cache->GetUsage());
+    }
   }
 
   cache->EraseUnRefEntries();
   precise_cache->EraseUnRefEntries();
   ASSERT_EQ(0, cache->GetUsage());
-  ASSERT_EQ(0, precise_cache->GetUsage());
+  ASSERT_EQ(baseline_meta_usage, precise_cache->GetUsage());
 
   // make sure the cache will be overloaded
   for (size_t i = 1; i < kCapacity; ++i) {
@@ -274,17 +265,23 @@ TEST_P(CacheTest, UsageTest) {
     } else {
       key = EncodeKey(static_cast<int>(1000 + i));
     }
-    ASSERT_OK(cache->Insert(key, reinterpret_cast<void*>(value), key.size() + 5,
-                            DumbDeleter));
-    ASSERT_OK(precise_cache->Insert(key, reinterpret_cast<void*>(value),
-                                    key.size() + 5, DumbDeleter));
+    ASSERT_OK(cache->Insert(key, value, &kDumbHelper, key.size() + 5));
+    ASSERT_OK(precise_cache->Insert(key, value, &kDumbHelper, key.size() + 5));
   }
 
   // the usage should be close to the capacity
   ASSERT_GT(kCapacity, cache->GetUsage());
   ASSERT_GT(kCapacity, precise_cache->GetUsage());
   ASSERT_LT(kCapacity * 0.95, cache->GetUsage());
-  ASSERT_LT(kCapacity * 0.95, precise_cache->GetUsage());
+  if (type != kHyperClock) {
+    ASSERT_LT(kCapacity * 0.95, precise_cache->GetUsage());
+  } else {
+    // estimated value size of 1 is weird for clock cache, because
+    // almost all of the capacity will be used for metadata, and due to only
+    // using power of 2 table sizes, we might hit strict occupancy limit
+    // before hitting capacity limit.
+    ASSERT_LT(kCapacity * 0.80, precise_cache->GetUsage());
+  }
 }
 
 // TODO: This test takes longer than expected on ClockCache. This is
@@ -301,6 +298,10 @@ TEST_P(CacheTest, PinnedUsageTest) {
   const size_t kCapacity = 200000;
   auto cache = NewCache(kCapacity, 8, false, kDontChargeCacheMetadata);
   auto precise_cache = NewCache(kCapacity, 8, false, kFullChargeCacheMetadata);
+  size_t baseline_meta_usage = precise_cache->GetUsage();
+  if (type != kHyperClock) {
+    ASSERT_EQ(0, baseline_meta_usage);
+  }
 
   size_t pinned_usage = 0;
   char value[10] = "abcdef";
@@ -320,11 +321,9 @@ TEST_P(CacheTest, PinnedUsageTest) {
     auto kv_size = key.size() + 5;
     Cache::Handle* handle;
     Cache::Handle* handle_in_precise_cache;
-    ASSERT_OK(cache->Insert(key, reinterpret_cast<void*>(value), kv_size,
-                            DumbDeleter, &handle));
+    ASSERT_OK(cache->Insert(key, value, &kDumbHelper, kv_size, &handle));
     assert(handle);
-    ASSERT_OK(precise_cache->Insert(key, reinterpret_cast<void*>(value),
-                                    kv_size, DumbDeleter,
+    ASSERT_OK(precise_cache->Insert(key, value, &kDumbHelper, kv_size,
                                     &handle_in_precise_cache));
     assert(handle_in_precise_cache);
     pinned_usage += kv_size;
@@ -365,10 +364,8 @@ TEST_P(CacheTest, PinnedUsageTest) {
     } else {
       key = EncodeKey(static_cast<int>(1000 + i));
     }
-    ASSERT_OK(cache->Insert(key, reinterpret_cast<void*>(value), key.size() + 5,
-                            DumbDeleter));
-    ASSERT_OK(precise_cache->Insert(key, reinterpret_cast<void*>(value),
-                                    key.size() + 5, DumbDeleter));
+    ASSERT_OK(cache->Insert(key, value, &kDumbHelper, key.size() + 5));
+    ASSERT_OK(precise_cache->Insert(key, value, &kDumbHelper, key.size() + 5));
   }
   ASSERT_EQ(pinned_usage, cache->GetPinnedUsage());
   ASSERT_EQ(precise_cache_pinned_usage, precise_cache->GetPinnedUsage());
@@ -390,7 +387,7 @@ TEST_P(CacheTest, PinnedUsageTest) {
   cache->EraseUnRefEntries();
   precise_cache->EraseUnRefEntries();
   ASSERT_EQ(0, cache->GetUsage());
-  ASSERT_EQ(0, precise_cache->GetUsage());
+  ASSERT_EQ(baseline_meta_usage, precise_cache->GetUsage());
 }
 
 TEST_P(CacheTest, HitAndMiss) {
@@ -398,25 +395,38 @@ TEST_P(CacheTest, HitAndMiss) {
 
   Insert(100, 101);
   ASSERT_EQ(101, Lookup(100));
-  ASSERT_EQ(-1,  Lookup(200));
-  ASSERT_EQ(-1,  Lookup(300));
+  ASSERT_EQ(-1, Lookup(200));
+  ASSERT_EQ(-1, Lookup(300));
 
   Insert(200, 201);
   ASSERT_EQ(101, Lookup(100));
   ASSERT_EQ(201, Lookup(200));
-  ASSERT_EQ(-1,  Lookup(300));
+  ASSERT_EQ(-1, Lookup(300));
 
   Insert(100, 102);
-  ASSERT_EQ(102, Lookup(100));
+  if (GetParam() == kHyperClock) {
+    // ClockCache usually doesn't overwrite on Insert
+    ASSERT_EQ(101, Lookup(100));
+  } else {
+    ASSERT_EQ(102, Lookup(100));
+  }
   ASSERT_EQ(201, Lookup(200));
-  ASSERT_EQ(-1,  Lookup(300));
+  ASSERT_EQ(-1, Lookup(300));
 
-  ASSERT_EQ(1U, deleted_keys_.size());
-  ASSERT_EQ(100, deleted_keys_[0]);
-  ASSERT_EQ(101, deleted_values_[0]);
+  ASSERT_EQ(1U, deleted_values_.size());
+  if (GetParam() == kHyperClock) {
+    ASSERT_EQ(102, deleted_values_[0]);
+  } else {
+    ASSERT_EQ(101, deleted_values_[0]);
+  }
 }
 
 TEST_P(CacheTest, InsertSameKey) {
+  if (GetParam() == kHyperClock) {
+    ROCKSDB_GTEST_BYPASS(
+        "ClockCache doesn't guarantee Insert overwrite same key.");
+    return;
+  }
   Insert(1, 1);
   Insert(1, 2);
   ASSERT_EQ(2, Lookup(1));
@@ -424,24 +434,28 @@ TEST_P(CacheTest, InsertSameKey) {
 
 TEST_P(CacheTest, Erase) {
   Erase(200);
-  ASSERT_EQ(0U, deleted_keys_.size());
+  ASSERT_EQ(0U, deleted_values_.size());
 
   Insert(100, 101);
   Insert(200, 201);
   Erase(100);
-  ASSERT_EQ(-1,  Lookup(100));
+  ASSERT_EQ(-1, Lookup(100));
   ASSERT_EQ(201, Lookup(200));
-  ASSERT_EQ(1U, deleted_keys_.size());
-  ASSERT_EQ(100, deleted_keys_[0]);
+  ASSERT_EQ(1U, deleted_values_.size());
   ASSERT_EQ(101, deleted_values_[0]);
 
   Erase(100);
-  ASSERT_EQ(-1,  Lookup(100));
+  ASSERT_EQ(-1, Lookup(100));
   ASSERT_EQ(201, Lookup(200));
-  ASSERT_EQ(1U, deleted_keys_.size());
+  ASSERT_EQ(1U, deleted_values_.size());
 }
 
 TEST_P(CacheTest, EntriesArePinned) {
+  if (GetParam() == kHyperClock) {
+    ROCKSDB_GTEST_BYPASS(
+        "ClockCache doesn't guarantee Insert overwrite same key.");
+    return;
+  }
   Insert(100, 101);
   Cache::Handle* h1 = cache_->Lookup(EncodeKey(100));
   ASSERT_EQ(101, DecodeValue(cache_->Value(h1)));
@@ -450,23 +464,21 @@ TEST_P(CacheTest, EntriesArePinned) {
   Insert(100, 102);
   Cache::Handle* h2 = cache_->Lookup(EncodeKey(100));
   ASSERT_EQ(102, DecodeValue(cache_->Value(h2)));
-  ASSERT_EQ(0U, deleted_keys_.size());
+  ASSERT_EQ(0U, deleted_values_.size());
   ASSERT_EQ(2U, cache_->GetUsage());
 
   cache_->Release(h1);
-  ASSERT_EQ(1U, deleted_keys_.size());
-  ASSERT_EQ(100, deleted_keys_[0]);
+  ASSERT_EQ(1U, deleted_values_.size());
   ASSERT_EQ(101, deleted_values_[0]);
   ASSERT_EQ(1U, cache_->GetUsage());
 
   Erase(100);
   ASSERT_EQ(-1, Lookup(100));
-  ASSERT_EQ(1U, deleted_keys_.size());
+  ASSERT_EQ(1U, deleted_values_.size());
   ASSERT_EQ(1U, cache_->GetUsage());
 
   cache_->Release(h2);
-  ASSERT_EQ(2U, deleted_keys_.size());
-  ASSERT_EQ(100, deleted_keys_[1]);
+  ASSERT_EQ(2U, deleted_values_.size());
   ASSERT_EQ(102, deleted_values_[1]);
   ASSERT_EQ(0U, cache_->GetUsage());
 }
@@ -474,10 +486,9 @@ TEST_P(CacheTest, EntriesArePinned) {
 TEST_P(CacheTest, EvictionPolicy) {
   Insert(100, 101);
   Insert(200, 201);
-
   // Frequently used entry must be kept around
   for (int i = 0; i < 2 * kCacheSize; i++) {
-    Insert(1000+i, 2000+i);
+    Insert(1000 + i, 2000 + i);
     ASSERT_EQ(101, Lookup(100));
   }
   ASSERT_EQ(101, Lookup(100));
@@ -503,6 +514,12 @@ TEST_P(CacheTest, ExternalRefPinsEntries) {
     for (int j = 0; j < 2 * kCacheSize + 100; j++) {
       Insert(1000 + j, 2000 + j);
     }
+    // Clock cache is even more stateful and needs more churn to evict
+    if (GetParam() == kHyperClock) {
+      for (int j = 0; j < kCacheSize; j++) {
+        Insert(11000 + j, 11000 + j);
+      }
+    }
     if (i < 2) {
       ASSERT_EQ(101, Lookup(100));
     }
@@ -564,9 +581,9 @@ TEST_P(CacheTest, EvictEmptyCache) {
   // Insert item large than capacity to trigger eviction on empty cache.
   auto cache = NewCache(1, 0, false);
   if (type == kLRU) {
-    ASSERT_OK(cache->Insert("foo", nullptr, 10, DumbDeleter));
+    ASSERT_OK(cache->Insert("foo", nullptr, &kDumbHelper, 10));
   } else {
-    ASSERT_OK(cache->Insert(EncodeKey(1000), nullptr, 10, DumbDeleter));
+    ASSERT_OK(cache->Insert(EncodeKey(1000), nullptr, &kDumbHelper, 10));
   }
 }
 
@@ -577,19 +594,19 @@ TEST_P(CacheTest, EraseFromDeleter) {
   // the cache at that point.
   std::shared_ptr<Cache> cache = NewCache(10, 0, false);
   std::string foo, bar;
-  Cache::DeleterFn erase_deleter;
+  const Cache::CacheItemHelper* erase_helper;
   if (type == kLRU) {
     foo = "foo";
     bar = "bar";
-    erase_deleter = EraseDeleter1;
+    erase_helper = &kEraseOnDeleteHelper1;
   } else {
     foo = EncodeKey(1234);
     bar = EncodeKey(5678);
-    erase_deleter = EraseDeleter2;
+    erase_helper = &kEraseOnDeleteHelper2;
   }
 
-  ASSERT_OK(cache->Insert(foo, nullptr, 1, DumbDeleter));
-  ASSERT_OK(cache->Insert(bar, cache.get(), 1, erase_deleter));
+  ASSERT_OK(cache->Insert(foo, nullptr, &kDumbHelper, 1));
+  ASSERT_OK(cache->Insert(bar, cache.get(), erase_helper, 1));
 
   cache->Erase(bar);
   ASSERT_EQ(nullptr, cache->Lookup(foo));
@@ -642,7 +659,7 @@ TEST_P(CacheTest, HeavyEntries) {
       ASSERT_EQ(1000 + i, r);
     }
   }
-  ASSERT_LE(cached_weight, kCacheSize + kCacheSize/10);
+  ASSERT_LE(cached_weight, kCacheSize + kCacheSize / 10);
 }
 
 TEST_P(CacheTest, NewId) {
@@ -651,55 +668,56 @@ TEST_P(CacheTest, NewId) {
   ASSERT_NE(a, b);
 }
 
-class Value {
- public:
-  explicit Value(int v) : v_(v) {}
-
-  int v_;
-};
-
-namespace {
-void deleter(const Slice& /*key*/, void* value) {
-  delete static_cast<Value *>(value);
-}
-}  // namespace
-
 TEST_P(CacheTest, ReleaseAndErase) {
   std::shared_ptr<Cache> cache = NewCache(5, 0, false);
   Cache::Handle* handle;
-  Status s = cache->Insert(EncodeKey(100), EncodeValue(100), 1,
-                           &CacheTest::Deleter, &handle);
+  Status s =
+      cache->Insert(EncodeKey(100), EncodeValue(100), &kHelper, 1, &handle);
   ASSERT_TRUE(s.ok());
   ASSERT_EQ(5U, cache->GetCapacity());
   ASSERT_EQ(1U, cache->GetUsage());
-  ASSERT_EQ(0U, deleted_keys_.size());
+  ASSERT_EQ(0U, deleted_values_.size());
   auto erased = cache->Release(handle, true);
   ASSERT_TRUE(erased);
   // This tests that deleter has been called
-  ASSERT_EQ(1U, deleted_keys_.size());
+  ASSERT_EQ(1U, deleted_values_.size());
 }
 
 TEST_P(CacheTest, ReleaseWithoutErase) {
   std::shared_ptr<Cache> cache = NewCache(5, 0, false);
   Cache::Handle* handle;
-  Status s = cache->Insert(EncodeKey(100), EncodeValue(100), 1,
-                           &CacheTest::Deleter, &handle);
+  Status s =
+      cache->Insert(EncodeKey(100), EncodeValue(100), &kHelper, 1, &handle);
   ASSERT_TRUE(s.ok());
   ASSERT_EQ(5U, cache->GetCapacity());
   ASSERT_EQ(1U, cache->GetUsage());
-  ASSERT_EQ(0U, deleted_keys_.size());
+  ASSERT_EQ(0U, deleted_values_.size());
   auto erased = cache->Release(handle);
   ASSERT_FALSE(erased);
   // This tests that deleter is not called. When cache has free capacity it is
   // not expected to immediately erase the released items.
-  ASSERT_EQ(0U, deleted_keys_.size());
+  ASSERT_EQ(0U, deleted_values_.size());
 }
 
+namespace {
+class Value {
+ public:
+  explicit Value(int v) : v_(v) {}
+
+  int v_;
+
+  static constexpr auto kCacheEntryRole = CacheEntryRole::kMisc;
+};
+
+using SharedCache = BasicTypedSharedCacheInterface<Value>;
+using TypedHandle = SharedCache::TypedHandle;
+}  // namespace
+
 TEST_P(CacheTest, SetCapacity) {
   auto type = GetParam();
-  if (type == kFast || type == kClock) {
+  if (type == kHyperClock) {
     ROCKSDB_GTEST_BYPASS(
-        "FastLRUCache and ClockCache don't support arbitrary capacity "
+        "FastLRUCache and HyperClockCache don't support arbitrary capacity "
         "adjustments.");
     return;
   }
@@ -707,19 +725,19 @@ TEST_P(CacheTest, SetCapacity) {
   // lets create a cache with capacity 5,
   // then, insert 5 elements, then increase capacity
   // to 10, returned capacity should be 10, usage=5
-  std::shared_ptr<Cache> cache = NewCache(5, 0, false);
-  std::vector<Cache::Handle*> handles(10);
+  SharedCache cache{NewCache(5, 0, false)};
+  std::vector<TypedHandle*> handles(10);
   // Insert 5 entries, but not releasing.
   for (int i = 0; i < 5; i++) {
     std::string key = EncodeKey(i + 1);
-    Status s = cache->Insert(key, new Value(i + 1), 1, &deleter, &handles[i]);
+    Status s = cache.Insert(key, new Value(i + 1), 1, &handles[i]);
     ASSERT_TRUE(s.ok());
   }
-  ASSERT_EQ(5U, cache->GetCapacity());
-  ASSERT_EQ(5U, cache->GetUsage());
-  cache->SetCapacity(10);
-  ASSERT_EQ(10U, cache->GetCapacity());
-  ASSERT_EQ(5U, cache->GetUsage());
+  ASSERT_EQ(5U, cache.get()->GetCapacity());
+  ASSERT_EQ(5U, cache.get()->GetUsage());
+  cache.get()->SetCapacity(10);
+  ASSERT_EQ(10U, cache.get()->GetCapacity());
+  ASSERT_EQ(5U, cache.get()->GetUsage());
 
   // test2: decrease capacity
   // insert 5 more elements to cache, then release 5,
@@ -727,176 +745,148 @@ TEST_P(CacheTest, SetCapacity) {
   // and usage should be 7
   for (int i = 5; i < 10; i++) {
     std::string key = EncodeKey(i + 1);
-    Status s = cache->Insert(key, new Value(i + 1), 1, &deleter, &handles[i]);
+    Status s = cache.Insert(key, new Value(i + 1), 1, &handles[i]);
     ASSERT_TRUE(s.ok());
   }
-  ASSERT_EQ(10U, cache->GetCapacity());
-  ASSERT_EQ(10U, cache->GetUsage());
+  ASSERT_EQ(10U, cache.get()->GetCapacity());
+  ASSERT_EQ(10U, cache.get()->GetUsage());
   for (int i = 0; i < 5; i++) {
-    cache->Release(handles[i]);
+    cache.Release(handles[i]);
   }
-  ASSERT_EQ(10U, cache->GetCapacity());
-  ASSERT_EQ(10U, cache->GetUsage());
-  cache->SetCapacity(7);
-  ASSERT_EQ(7, cache->GetCapacity());
-  ASSERT_EQ(7, cache->GetUsage());
+  ASSERT_EQ(10U, cache.get()->GetCapacity());
+  ASSERT_EQ(10U, cache.get()->GetUsage());
+  cache.get()->SetCapacity(7);
+  ASSERT_EQ(7, cache.get()->GetCapacity());
+  ASSERT_EQ(7, cache.get()->GetUsage());
 
   // release remaining 5 to keep valgrind happy
   for (int i = 5; i < 10; i++) {
-    cache->Release(handles[i]);
+    cache.Release(handles[i]);
   }
 
   // Make sure this doesn't crash or upset ASAN/valgrind
-  cache->DisownData();
+  cache.get()->DisownData();
 }
 
 TEST_P(LRUCacheTest, SetStrictCapacityLimit) {
-  auto type = GetParam();
-  if (type == kFast) {
-    ROCKSDB_GTEST_BYPASS(
-        "FastLRUCache only supports a limited number of "
-        "inserts beyond "
-        "capacity.");
-    return;
-  }
   // test1: set the flag to false. Insert more keys than capacity. See if they
   // all go through.
-  std::shared_ptr<Cache> cache = NewCache(5, 0, false);
-  std::vector<Cache::Handle*> handles(10);
+  SharedCache cache{NewCache(5, 0, false)};
+  std::vector<TypedHandle*> handles(10);
   Status s;
   for (int i = 0; i < 10; i++) {
     std::string key = EncodeKey(i + 1);
-    s = cache->Insert(key, new Value(i + 1), 1, &deleter, &handles[i]);
+    s = cache.Insert(key, new Value(i + 1), 1, &handles[i]);
     ASSERT_OK(s);
     ASSERT_NE(nullptr, handles[i]);
   }
-  ASSERT_EQ(10, cache->GetUsage());
+  ASSERT_EQ(10, cache.get()->GetUsage());
 
   // test2: set the flag to true. Insert and check if it fails.
   std::string extra_key = EncodeKey(100);
   Value* extra_value = new Value(0);
-  cache->SetStrictCapacityLimit(true);
-  Cache::Handle* handle;
-  s = cache->Insert(extra_key, extra_value, 1, &deleter, &handle);
+  cache.get()->SetStrictCapacityLimit(true);
+  TypedHandle* handle;
+  s = cache.Insert(extra_key, extra_value, 1, &handle);
   ASSERT_TRUE(s.IsMemoryLimit());
   ASSERT_EQ(nullptr, handle);
-  ASSERT_EQ(10, cache->GetUsage());
+  ASSERT_EQ(10, cache.get()->GetUsage());
 
   for (int i = 0; i < 10; i++) {
-    cache->Release(handles[i]);
+    cache.Release(handles[i]);
   }
 
   // test3: init with flag being true.
-  std::shared_ptr<Cache> cache2 = NewCache(5, 0, true);
+  SharedCache cache2{NewCache(5, 0, true)};
   for (int i = 0; i < 5; i++) {
     std::string key = EncodeKey(i + 1);
-    s = cache2->Insert(key, new Value(i + 1), 1, &deleter, &handles[i]);
+    s = cache2.Insert(key, new Value(i + 1), 1, &handles[i]);
     ASSERT_OK(s);
     ASSERT_NE(nullptr, handles[i]);
   }
-  s = cache2->Insert(extra_key, extra_value, 1, &deleter, &handle);
+  s = cache2.Insert(extra_key, extra_value, 1, &handle);
   ASSERT_TRUE(s.IsMemoryLimit());
   ASSERT_EQ(nullptr, handle);
   // test insert without handle
-  s = cache2->Insert(extra_key, extra_value, 1, &deleter);
+  s = cache2.Insert(extra_key, extra_value, 1);
   // AS if the key have been inserted into cache but get evicted immediately.
   ASSERT_OK(s);
-  ASSERT_EQ(5, cache2->GetUsage());
-  ASSERT_EQ(nullptr, cache2->Lookup(extra_key));
+  ASSERT_EQ(5, cache2.get()->GetUsage());
+  ASSERT_EQ(nullptr, cache2.Lookup(extra_key));
 
   for (int i = 0; i < 5; i++) {
-    cache2->Release(handles[i]);
+    cache2.Release(handles[i]);
   }
 }
 
 TEST_P(CacheTest, OverCapacity) {
-  auto type = GetParam();
-  if (type == kClock) {
-    ROCKSDB_GTEST_BYPASS("Requires LRU eviction policy.");
-    return;
-  }
   size_t n = 10;
 
   // a LRUCache with n entries and one shard only
-  std::shared_ptr<Cache> cache = NewCache(n, 0, false);
-
-  std::vector<Cache::Handle*> handles(n+1);
+  SharedCache cache{NewCache(n, 0, false)};
+  std::vector<TypedHandle*> handles(n + 1);
 
   // Insert n+1 entries, but not releasing.
   for (int i = 0; i < static_cast<int>(n + 1); i++) {
     std::string key = EncodeKey(i + 1);
-    Status s = cache->Insert(key, new Value(i + 1), 1, &deleter, &handles[i]);
+    Status s = cache.Insert(key, new Value(i + 1), 1, &handles[i]);
     ASSERT_TRUE(s.ok());
   }
 
   // Guess what's in the cache now?
   for (int i = 0; i < static_cast<int>(n + 1); i++) {
     std::string key = EncodeKey(i + 1);
-    auto h = cache->Lookup(key);
+    auto h = cache.Lookup(key);
     ASSERT_TRUE(h != nullptr);
-    if (h) cache->Release(h);
+    if (h) cache.Release(h);
   }
 
   // the cache is over capacity since nothing could be evicted
-  ASSERT_EQ(n + 1U, cache->GetUsage());
+  ASSERT_EQ(n + 1U, cache.get()->GetUsage());
   for (int i = 0; i < static_cast<int>(n + 1); i++) {
-    cache->Release(handles[i]);
+    cache.Release(handles[i]);
   }
-  // Make sure eviction is triggered.
-  cache->SetCapacity(n);
 
-  // cache is under capacity now since elements were released
-  ASSERT_EQ(n, cache->GetUsage());
+  if (GetParam() == kHyperClock) {
+    // Make sure eviction is triggered.
+    ASSERT_OK(cache.Insert(EncodeKey(-1), nullptr, 1, &handles[0]));
 
-  // element 0 is evicted and the rest is there
-  // This is consistent with the LRU policy since the element 0
-  // was released first
-  for (int i = 0; i < static_cast<int>(n + 1); i++) {
-    std::string key = EncodeKey(i + 1);
-    auto h = cache->Lookup(key);
-    if (h) {
-      ASSERT_NE(static_cast<size_t>(i), 0U);
-      cache->Release(h);
-    } else {
-      ASSERT_EQ(static_cast<size_t>(i), 0U);
+    // cache is under capacity now since elements were released
+    ASSERT_GE(n, cache.get()->GetUsage());
+
+    // clean up
+    cache.Release(handles[0]);
+  } else {
+    // LRUCache checks for over-capacity in Release.
+
+    // cache is exactly at capacity now with minimal eviction
+    ASSERT_EQ(n, cache.get()->GetUsage());
+
+    // element 0 is evicted and the rest is there
+    // This is consistent with the LRU policy since the element 0
+    // was released first
+    for (int i = 0; i < static_cast<int>(n + 1); i++) {
+      std::string key = EncodeKey(i + 1);
+      auto h = cache.Lookup(key);
+      if (h) {
+        ASSERT_NE(static_cast<size_t>(i), 0U);
+        cache.Release(h);
+      } else {
+        ASSERT_EQ(static_cast<size_t>(i), 0U);
+      }
     }
   }
 }
 
-namespace {
-std::vector<std::pair<int, int>> legacy_callback_state;
-void legacy_callback(void* value, size_t charge) {
-  legacy_callback_state.push_back(
-      {DecodeValue(value), static_cast<int>(charge)});
-}
-};
-
-TEST_P(CacheTest, ApplyToAllCacheEntriesTest) {
-  std::vector<std::pair<int, int>> inserted;
-  legacy_callback_state.clear();
-
-  for (int i = 0; i < 10; ++i) {
-    Insert(i, i * 2, i + 1);
-    inserted.push_back({i * 2, i + 1});
-  }
-  cache_->ApplyToAllCacheEntries(legacy_callback, true);
-
-  std::sort(inserted.begin(), inserted.end());
-  std::sort(legacy_callback_state.begin(), legacy_callback_state.end());
-  ASSERT_EQ(inserted.size(), legacy_callback_state.size());
-  for (int i = 0; i < static_cast<int>(inserted.size()); ++i) {
-    EXPECT_EQ(inserted[i], legacy_callback_state[i]);
-  }
-}
-
 TEST_P(CacheTest, ApplyToAllEntriesTest) {
   std::vector<std::string> callback_state;
-  const auto callback = [&](const Slice& key, void* value, size_t charge,
-                            Cache::DeleterFn deleter) {
+  const auto callback = [&](const Slice& key, Cache::ObjectPtr value,
+                            size_t charge,
+                            const Cache::CacheItemHelper* helper) {
     callback_state.push_back(std::to_string(DecodeKey(key)) + "," +
                              std::to_string(DecodeValue(value)) + "," +
                              std::to_string(charge));
-    assert(deleter == &CacheTest::Deleter);
+    assert(helper == &CacheTest::kHelper);
   };
 
   std::vector<std::string> inserted;
@@ -935,8 +925,8 @@ TEST_P(CacheTest, ApplyToAllEntriesDuringResize) {
 
   // For callback
   int special_count = 0;
-  const auto callback = [&](const Slice&, void*, size_t charge,
-                            Cache::DeleterFn) {
+  const auto callback = [&](const Slice&, Cache::ObjectPtr, size_t charge,
+                            const Cache::CacheItemHelper*) {
     if (charge == static_cast<size_t>(kSpecialCharge)) {
       ++special_count;
     }
@@ -966,19 +956,31 @@ TEST_P(CacheTest, ApplyToAllEntriesDuringResize) {
 }
 
 TEST_P(CacheTest, DefaultShardBits) {
-  // test1: set the flag to false. Insert more keys than capacity. See if they
-  // all go through.
-  std::shared_ptr<Cache> cache = NewCache(16 * 1024L * 1024L);
-  ShardedCache* sc = dynamic_cast<ShardedCache*>(cache.get());
+  // Prevent excessive allocation (to save time & space)
+  estimated_value_size_ = 100000;
+  // Implementations use different minimum shard sizes
+  size_t min_shard_size =
+      (GetParam() == kHyperClock ? 32U * 1024U : 512U) * 1024U;
+
+  std::shared_ptr<Cache> cache = NewCache(32U * min_shard_size);
+  ShardedCacheBase* sc = dynamic_cast<ShardedCacheBase*>(cache.get());
   ASSERT_EQ(5, sc->GetNumShardBits());
 
-  cache = NewLRUCache(511 * 1024L, -1, true);
-  sc = dynamic_cast<ShardedCache*>(cache.get());
+  cache = NewCache(min_shard_size / 1000U * 999U);
+  sc = dynamic_cast<ShardedCacheBase*>(cache.get());
   ASSERT_EQ(0, sc->GetNumShardBits());
 
-  cache = NewLRUCache(1024L * 1024L * 1024L, -1, true);
-  sc = dynamic_cast<ShardedCache*>(cache.get());
+  cache = NewCache(3U * 1024U * 1024U * 1024U);
+  sc = dynamic_cast<ShardedCacheBase*>(cache.get());
+  // current maximum of 6
   ASSERT_EQ(6, sc->GetNumShardBits());
+
+  if constexpr (sizeof(size_t) > 4) {
+    cache = NewCache(128U * min_shard_size);
+    sc = dynamic_cast<ShardedCacheBase*>(cache.get());
+    // current maximum of 6
+    ASSERT_EQ(6, sc->GetNumShardBits());
+  }
 }
 
 TEST_P(CacheTest, GetChargeAndDeleter) {
@@ -986,17 +988,13 @@ TEST_P(CacheTest, GetChargeAndDeleter) {
   Cache::Handle* h1 = cache_->Lookup(EncodeKey(1));
   ASSERT_EQ(2, DecodeValue(cache_->Value(h1)));
   ASSERT_EQ(1, cache_->GetCharge(h1));
-  ASSERT_EQ(&CacheTest::Deleter, cache_->GetDeleter(h1));
+  ASSERT_EQ(&CacheTest::kHelper, cache_->GetCacheItemHelper(h1));
   cache_->Release(h1);
 }
 
-std::shared_ptr<Cache> (*new_clock_cache_func)(size_t, size_t, int, bool,
-                                               CacheMetadataChargePolicy) =
-    ExperimentalNewClockCache;
 INSTANTIATE_TEST_CASE_P(CacheTestInstance, CacheTest,
-                        testing::Values(kLRU, kClock, kFast));
-INSTANTIATE_TEST_CASE_P(CacheTestInstance, LRUCacheTest,
-                        testing::Values(kLRU, kFast));
+                        testing::Values(kLRU, kHyperClock));
+INSTANTIATE_TEST_CASE_P(CacheTestInstance, LRUCacheTest, testing::Values(kLRU));
 
 }  // namespace ROCKSDB_NAMESPACE
 

cache/charged_cache.cc

@@ -17,25 +17,10 @@ ChargedCache::ChargedCache(std::shared_ptr<Cache> cache,
               CacheReservationManagerImpl<CacheEntryRole::kBlobCache>>(
               block_cache))) {}
 
-Status ChargedCache::Insert(const Slice& key, void* value, size_t charge,
-                            DeleterFn deleter, Handle** handle,
-                            Priority priority) {
-  Status s = cache_->Insert(key, value, charge, deleter, handle, priority);
-  if (s.ok()) {
-    // Insert may cause the cache entry eviction if the cache is full. So we
-    // directly call the reservation manager to update the total memory used
-    // in the cache.
-    assert(cache_res_mgr_);
-    cache_res_mgr_->UpdateCacheReservation(cache_->GetUsage())
-        .PermitUncheckedError();
-  }
-  return s;
-}
-
-Status ChargedCache::Insert(const Slice& key, void* value,
+Status ChargedCache::Insert(const Slice& key, ObjectPtr obj,
                             const CacheItemHelper* helper, size_t charge,
                             Handle** handle, Priority priority) {
-  Status s = cache_->Insert(key, value, helper, charge, handle, priority);
+  Status s = cache_->Insert(key, obj, helper, charge, handle, priority);
   if (s.ok()) {
     // Insert may cause the cache entry eviction if the cache is full. So we
     // directly call the reservation manager to update the total memory used
@@ -47,22 +32,21 @@ Status ChargedCache::Insert(const Slice& key, void* value,
   return s;
 }
 
-Cache::Handle* ChargedCache::Lookup(const Slice& key, Statistics* stats) {
-  return cache_->Lookup(key, stats);
-}
-
 Cache::Handle* ChargedCache::Lookup(const Slice& key,
                                     const CacheItemHelper* helper,
-                                    const CreateCallback& create_cb,
+                                    CreateContext* create_context,
                                     Priority priority, bool wait,
                                     Statistics* stats) {
-  auto handle = cache_->Lookup(key, helper, create_cb, priority, wait, stats);
+  auto handle =
+      cache_->Lookup(key, helper, create_context, priority, wait, stats);
   // Lookup may promote the KV pair from the secondary cache to the primary
   // cache. So we directly call the reservation manager to update the total
   // memory used in the cache.
-  assert(cache_res_mgr_);
-  cache_res_mgr_->UpdateCacheReservation(cache_->GetUsage())
-      .PermitUncheckedError();
+  if (helper && helper->create_cb) {
+    assert(cache_res_mgr_);
+    cache_res_mgr_->UpdateCacheReservation(cache_->GetUsage())
+        .PermitUncheckedError();
+  }
   return handle;
 }
 

cache/charged_cache.h

@@ -23,16 +23,14 @@ class ChargedCache : public Cache {
                std::shared_ptr<Cache> block_cache);
   ~ChargedCache() override = default;
 
-  Status Insert(const Slice& key, void* value, size_t charge, DeleterFn deleter,
-                Handle** handle, Priority priority) override;
-  Status Insert(const Slice& key, void* value, const CacheItemHelper* helper,
+  Status Insert(const Slice& key, ObjectPtr obj, const CacheItemHelper* helper,
                 size_t charge, Handle** handle = nullptr,
                 Priority priority = Priority::LOW) override;
 
-  Cache::Handle* Lookup(const Slice& key, Statistics* stats) override;
   Cache::Handle* Lookup(const Slice& key, const CacheItemHelper* helper,
-                        const CreateCallback& create_cb, Priority priority,
-                        bool wait, Statistics* stats = nullptr) override;
+                        CreateContext* create_context,
+                        Priority priority = Priority::LOW, bool wait = true,
+                        Statistics* stats = nullptr) override;
 
   bool Release(Cache::Handle* handle, bool useful,
                bool erase_if_last_ref = false) override;
@@ -56,7 +54,9 @@ class ChargedCache : public Cache {
     return cache_->HasStrictCapacityLimit();
   }
 
-  void* Value(Cache::Handle* handle) override { return cache_->Value(handle); }
+  ObjectPtr Value(Cache::Handle* handle) override {
+    return cache_->Value(handle);
+  }
 
   bool IsReady(Cache::Handle* handle) override {
     return cache_->IsReady(handle);
@@ -84,22 +84,17 @@ class ChargedCache : public Cache {
     return cache_->GetCharge(handle);
   }
 
-  Cache::DeleterFn GetDeleter(Cache::Handle* handle) const override {
-    return cache_->GetDeleter(handle);
+  const CacheItemHelper* GetCacheItemHelper(Handle* handle) const override {
+    return cache_->GetCacheItemHelper(handle);
   }
 
   void ApplyToAllEntries(
-      const std::function<void(const Slice& key, void* value, size_t charge,
-                               Cache::DeleterFn deleter)>& callback,
+      const std::function<void(const Slice& key, ObjectPtr value, size_t charge,
+                               const CacheItemHelper* helper)>& callback,
       const Cache::ApplyToAllEntriesOptions& opts) override {
     cache_->ApplyToAllEntries(callback, opts);
   }
 
-  void ApplyToAllCacheEntries(void (*callback)(void* value, size_t charge),
-                              bool thread_safe) override {
-    cache_->ApplyToAllCacheEntries(callback, thread_safe);
-  }
-
   std::string GetPrintableOptions() const override {
     return cache_->GetPrintableOptions();
   }

cache/compressed_secondary_cache.cc

@@ -10,6 +10,7 @@
 #include <memory>
 
 #include "memory/memory_allocator.h"
+#include "monitoring/perf_context_imp.h"
 #include "util/compression.h"
 #include "util/string_util.h"
 
@@ -20,11 +21,13 @@ CompressedSecondaryCache::CompressedSecondaryCache(
     double high_pri_pool_ratio, double low_pri_pool_ratio,
     std::shared_ptr<MemoryAllocator> memory_allocator, bool use_adaptive_mutex,
     CacheMetadataChargePolicy metadata_charge_policy,
-    CompressionType compression_type, uint32_t compress_format_version)
+    CompressionType compression_type, uint32_t compress_format_version,
+    bool enable_custom_split_merge)
     : cache_options_(capacity, num_shard_bits, strict_capacity_limit,
-                     high_pri_pool_ratio, memory_allocator, use_adaptive_mutex,
-                     metadata_charge_policy, compression_type,
-                     compress_format_version, low_pri_pool_ratio) {
+                     high_pri_pool_ratio, low_pri_pool_ratio, memory_allocator,
+                     use_adaptive_mutex, metadata_charge_policy,
+                     compression_type, compress_format_version,
+                     enable_custom_split_merge) {
   cache_ =
       NewLRUCache(capacity, num_shard_bits, strict_capacity_limit,
                   high_pri_pool_ratio, memory_allocator, use_adaptive_mutex,
@@ -34,26 +37,43 @@ CompressedSecondaryCache::CompressedSecondaryCache(
 CompressedSecondaryCache::~CompressedSecondaryCache() { cache_.reset(); }
 
 std::unique_ptr<SecondaryCacheResultHandle> CompressedSecondaryCache::Lookup(
-    const Slice& key, const Cache::CreateCallback& create_cb, bool /*wait*/,
+    const Slice& key, const Cache::CacheItemHelper* helper,
+    Cache::CreateContext* create_context, bool /*wait*/, bool advise_erase,
     bool& is_in_sec_cache) {
+  assert(helper);
   std::unique_ptr<SecondaryCacheResultHandle> handle;
   is_in_sec_cache = false;
   Cache::Handle* lru_handle = cache_->Lookup(key);
   if (lru_handle == nullptr) {
-    return handle;
+    return nullptr;
   }
 
-  CacheValueChunk* handle_value =
-      reinterpret_cast<CacheValueChunk*>(cache_->Value(lru_handle));
+  void* handle_value = cache_->Value(lru_handle);
+  if (handle_value == nullptr) {
+    cache_->Release(lru_handle, /*erase_if_last_ref=*/false);
+    return nullptr;
+  }
+
+  CacheAllocationPtr* ptr{nullptr};
+  CacheAllocationPtr merged_value;
   size_t handle_value_charge{0};
-  CacheAllocationPtr merged_value =
-      MergeChunksIntoValue(handle_value, handle_value_charge);
+  if (cache_options_.enable_custom_split_merge) {
+    CacheValueChunk* value_chunk_ptr =
+        reinterpret_cast<CacheValueChunk*>(handle_value);
+    merged_value = MergeChunksIntoValue(value_chunk_ptr, handle_value_charge);
+    ptr = &merged_value;
+  } else {
+    ptr = reinterpret_cast<CacheAllocationPtr*>(handle_value);
+    handle_value_charge = cache_->GetCharge(lru_handle);
+  }
+  MemoryAllocator* allocator = cache_options_.memory_allocator.get();
 
   Status s;
-  void* value{nullptr};
+  Cache::ObjectPtr value{nullptr};
   size_t charge{0};
   if (cache_options_.compression_type == kNoCompression) {
-    s = create_cb(merged_value.get(), handle_value_charge, &value, &charge);
+    s = helper->create_cb(Slice(ptr->get(), handle_value_charge),
+                          create_context, allocator, &value, &charge);
   } else {
     UncompressionContext uncompression_context(cache_options_.compression_type);
     UncompressionInfo uncompression_info(uncompression_context,
@@ -61,32 +81,57 @@ std::unique_ptr<SecondaryCacheResultHandle> CompressedSecondaryCache::Lookup(
                                          cache_options_.compression_type);
 
     size_t uncompressed_size{0};
-    CacheAllocationPtr uncompressed;
-    uncompressed = UncompressData(uncompression_info, (char*)merged_value.get(),
-                                  handle_value_charge, &uncompressed_size,
-                                  cache_options_.compress_format_version,
-                                  cache_options_.memory_allocator.get());
+    CacheAllocationPtr uncompressed = UncompressData(
+        uncompression_info, (char*)ptr->get(), handle_value_charge,
+        &uncompressed_size, cache_options_.compress_format_version, allocator);
 
     if (!uncompressed) {
-      cache_->Release(lru_handle, /* erase_if_last_ref */ true);
-      return handle;
+      cache_->Release(lru_handle, /*erase_if_last_ref=*/true);
+      return nullptr;
     }
-    s = create_cb(uncompressed.get(), uncompressed_size, &value, &charge);
+    s = helper->create_cb(Slice(uncompressed.get(), uncompressed_size),
+                          create_context, allocator, &value, &charge);
   }
 
   if (!s.ok()) {
-    cache_->Release(lru_handle, /* erase_if_last_ref */ true);
-    return handle;
+    cache_->Release(lru_handle, /*erase_if_last_ref=*/true);
+    return nullptr;
   }
 
-  cache_->Release(lru_handle, /* erase_if_last_ref */ true);
+  if (advise_erase) {
+    cache_->Release(lru_handle, /*erase_if_last_ref=*/true);
+    // Insert a dummy handle.
+    cache_
+        ->Insert(key, /*obj=*/nullptr,
+                 GetHelper(cache_options_.enable_custom_split_merge),
+                 /*charge=*/0)
+        .PermitUncheckedError();
+  } else {
+    is_in_sec_cache = true;
+    cache_->Release(lru_handle, /*erase_if_last_ref=*/false);
+  }
   handle.reset(new CompressedSecondaryCacheResultHandle(value, charge));
-
   return handle;
 }
 
-Status CompressedSecondaryCache::Insert(const Slice& key, void* value,
+Status CompressedSecondaryCache::Insert(const Slice& key,
+                                        Cache::ObjectPtr value,
                                         const Cache::CacheItemHelper* helper) {
+  if (value == nullptr) {
+    return Status::InvalidArgument();
+  }
+
+  Cache::Handle* lru_handle = cache_->Lookup(key);
+  auto internal_helper = GetHelper(cache_options_.enable_custom_split_merge);
+  if (lru_handle == nullptr) {
+    PERF_COUNTER_ADD(compressed_sec_cache_insert_dummy_count, 1);
+    // Insert a dummy handle if the handle is evicted for the first time.
+    return cache_->Insert(key, /*obj=*/nullptr, internal_helper,
+                          /*charge=*/0);
+  } else {
+    cache_->Release(lru_handle, /*erase_if_last_ref=*/false);
+  }
+
   size_t size = (*helper->size_cb)(value);
   CacheAllocationPtr ptr =
       AllocateBlock(size, cache_options_.memory_allocator.get());
@@ -99,6 +144,7 @@ Status CompressedSecondaryCache::Insert(const Slice& key, void* value,
 
   std::string compressed_val;
   if (cache_options_.compression_type != kNoCompression) {
+    PERF_COUNTER_ADD(compressed_sec_cache_uncompressed_bytes, size);
     CompressionOptions compression_opts;
     CompressionContext compression_context(cache_options_.compression_type);
     uint64_t sample_for_compression{0};
@@ -115,16 +161,42 @@ Status CompressedSecondaryCache::Insert(const Slice& key, void* value,
     }
 
     val = Slice(compressed_val);
+    size = compressed_val.size();
+    PERF_COUNTER_ADD(compressed_sec_cache_compressed_bytes, size);
+
+    if (!cache_options_.enable_custom_split_merge) {
+      ptr = AllocateBlock(size, cache_options_.memory_allocator.get());
+      memcpy(ptr.get(), compressed_val.data(), size);
+    }
   }
 
-  size_t charge{0};
-  CacheValueChunk* value_chunks_head =
-      SplitValueIntoChunks(val, cache_options_.compression_type, charge);
-  return cache_->Insert(key, value_chunks_head, charge, DeletionCallback);
+  PERF_COUNTER_ADD(compressed_sec_cache_insert_real_count, 1);
+  if (cache_options_.enable_custom_split_merge) {
+    size_t charge{0};
+    CacheValueChunk* value_chunks_head =
+        SplitValueIntoChunks(val, cache_options_.compression_type, charge);
+    return cache_->Insert(key, value_chunks_head, internal_helper, charge);
+  } else {
+    CacheAllocationPtr* buf = new CacheAllocationPtr(std::move(ptr));
+    return cache_->Insert(key, buf, internal_helper, size);
+  }
 }
 
 void CompressedSecondaryCache::Erase(const Slice& key) { cache_->Erase(key); }
 
+Status CompressedSecondaryCache::SetCapacity(size_t capacity) {
+  MutexLock l(&capacity_mutex_);
+  cache_options_.capacity = capacity;
+  cache_->SetCapacity(capacity);
+  return Status::OK();
+}
+
+Status CompressedSecondaryCache::GetCapacity(size_t& capacity) {
+  MutexLock l(&capacity_mutex_);
+  capacity = cache_options_.capacity;
+  return Status::OK();
+}
+
 std::string CompressedSecondaryCache::GetPrintableOptions() const {
   std::string ret;
   ret.reserve(20000);
@@ -141,9 +213,9 @@ std::string CompressedSecondaryCache::GetPrintableOptions() const {
 }
 
 CompressedSecondaryCache::CacheValueChunk*
-CompressedSecondaryCache::SplitValueIntoChunks(
-    const Slice& value, const CompressionType compression_type,
-    size_t& charge) {
+CompressedSecondaryCache::SplitValueIntoChunks(const Slice& value,
+                                               CompressionType compression_type,
+                                               size_t& charge) {
   assert(!value.empty());
   const char* src_ptr = value.data();
   size_t src_size{value.size()};
@@ -210,28 +282,44 @@ CacheAllocationPtr CompressedSecondaryCache::MergeChunksIntoValue(
   return ptr;
 }
 
-void CompressedSecondaryCache::DeletionCallback(const Slice& /*key*/,
-                                                void* obj) {
-  CacheValueChunk* chunks_head = reinterpret_cast<CacheValueChunk*>(obj);
-  while (chunks_head != nullptr) {
-    CacheValueChunk* tmp_chunk = chunks_head;
-    chunks_head = chunks_head->next;
-    tmp_chunk->Free();
+const Cache::CacheItemHelper* CompressedSecondaryCache::GetHelper(
+    bool enable_custom_split_merge) const {
+  if (enable_custom_split_merge) {
+    static const Cache::CacheItemHelper kHelper{
+        CacheEntryRole::kMisc,
+        [](Cache::ObjectPtr obj, MemoryAllocator* /*alloc*/) {
+          CacheValueChunk* chunks_head = static_cast<CacheValueChunk*>(obj);
+          while (chunks_head != nullptr) {
+            CacheValueChunk* tmp_chunk = chunks_head;
+            chunks_head = chunks_head->next;
+            tmp_chunk->Free();
+            obj = nullptr;
+          };
+        }};
+    return &kHelper;
+  } else {
+    static const Cache::CacheItemHelper kHelper{
+        CacheEntryRole::kMisc,
+        [](Cache::ObjectPtr obj, MemoryAllocator* /*alloc*/) {
+          delete static_cast<CacheAllocationPtr*>(obj);
+          obj = nullptr;
+        }};
+    return &kHelper;
   }
-  obj = nullptr;
 }
 
 std::shared_ptr<SecondaryCache> NewCompressedSecondaryCache(
     size_t capacity, int num_shard_bits, bool strict_capacity_limit,
-    double high_pri_pool_ratio,
+    double high_pri_pool_ratio, double low_pri_pool_ratio,
     std::shared_ptr<MemoryAllocator> memory_allocator, bool use_adaptive_mutex,
     CacheMetadataChargePolicy metadata_charge_policy,
     CompressionType compression_type, uint32_t compress_format_version,
-    double low_pri_pool_ratio) {
+    bool enable_custom_split_merge) {
   return std::make_shared<CompressedSecondaryCache>(
       capacity, num_shard_bits, strict_capacity_limit, high_pri_pool_ratio,
       low_pri_pool_ratio, memory_allocator, use_adaptive_mutex,
-      metadata_charge_policy, compression_type, compress_format_version);
+      metadata_charge_policy, compression_type, compress_format_version,
+      enable_custom_split_merge);
 }
 
 std::shared_ptr<SecondaryCache> NewCompressedSecondaryCache(
@@ -240,9 +328,10 @@ std::shared_ptr<SecondaryCache> NewCompressedSecondaryCache(
   assert(opts.secondary_cache == nullptr);
   return NewCompressedSecondaryCache(
       opts.capacity, opts.num_shard_bits, opts.strict_capacity_limit,
-      opts.high_pri_pool_ratio, opts.memory_allocator, opts.use_adaptive_mutex,
-      opts.metadata_charge_policy, opts.compression_type,
-      opts.compress_format_version, opts.low_pri_pool_ratio);
+      opts.high_pri_pool_ratio, opts.low_pri_pool_ratio, opts.memory_allocator,
+      opts.use_adaptive_mutex, opts.metadata_charge_policy,
+      opts.compression_type, opts.compress_format_version,
+      opts.enable_custom_split_merge);
 }
 
 }  // namespace ROCKSDB_NAMESPACE

cache/compressed_secondary_cache.h

@@ -15,14 +15,15 @@
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
 #include "util/compression.h"
+#include "util/mutexlock.h"
 
 namespace ROCKSDB_NAMESPACE {
 
 class CompressedSecondaryCacheResultHandle : public SecondaryCacheResultHandle {
  public:
-  CompressedSecondaryCacheResultHandle(void* value, size_t size)
+  CompressedSecondaryCacheResultHandle(Cache::ObjectPtr value, size_t size)
       : value_(value), size_(size) {}
-  virtual ~CompressedSecondaryCacheResultHandle() override = default;
+  ~CompressedSecondaryCacheResultHandle() override = default;
 
   CompressedSecondaryCacheResultHandle(
       const CompressedSecondaryCacheResultHandle&) = delete;
@@ -33,19 +34,33 @@ class CompressedSecondaryCacheResultHandle : public SecondaryCacheResultHandle {
 
   void Wait() override {}
 
-  void* Value() override { return value_; }
+  Cache::ObjectPtr Value() override { return value_; }
 
   size_t Size() override { return size_; }
 
  private:
-  void* value_;
+  Cache::ObjectPtr value_;
   size_t size_;
 };
 
 // The CompressedSecondaryCache is a concrete implementation of
 // rocksdb::SecondaryCache.
 //
-// Users can also cast a pointer to it and call methods on
+// When a block is found from CompressedSecondaryCache::Lookup, we check whether
+// there is a dummy block with the same key in the primary cache.
+// 1. If the dummy block exits, we erase the block from
+//    CompressedSecondaryCache and insert it into the primary cache.
+// 2. If not, we just insert a dummy block into the primary cache
+//    (charging the actual size of the block) and don not erase the block from
+//    CompressedSecondaryCache. A standalone handle is returned to the caller.
+//
+// When a block is evicted from the primary cache, we check whether
+// there is a dummy block with the same key in CompressedSecondaryCache.
+// 1. If the dummy block exits, the block is inserted into
+//    CompressedSecondaryCache.
+// 2. If not, we just insert a dummy block (size 0) in CompressedSecondaryCache.
+//
+// Users can also cast a pointer to CompressedSecondaryCache and call methods on
 // it directly, especially custom methods that may be added
 // in the future.  For example -
 // std::unique_ptr<rocksdb::SecondaryCache> cache =
@@ -62,30 +77,36 @@ class CompressedSecondaryCache : public SecondaryCache {
       CacheMetadataChargePolicy metadata_charge_policy =
           kDefaultCacheMetadataChargePolicy,
       CompressionType compression_type = CompressionType::kLZ4Compression,
-      uint32_t compress_format_version = 2);
-  virtual ~CompressedSecondaryCache() override;
+      uint32_t compress_format_version = 2,
+      bool enable_custom_split_merge = false);
+  ~CompressedSecondaryCache() override;
 
   const char* Name() const override { return "CompressedSecondaryCache"; }
 
-  Status Insert(const Slice& key, void* value,
+  Status Insert(const Slice& key, Cache::ObjectPtr value,
                 const Cache::CacheItemHelper* helper) override;
 
   std::unique_ptr<SecondaryCacheResultHandle> Lookup(
-      const Slice& key, const Cache::CreateCallback& create_cb, bool /*wait*/,
+      const Slice& key, const Cache::CacheItemHelper* helper,
+      Cache::CreateContext* create_context, bool /*wait*/, bool advise_erase,
       bool& is_in_sec_cache) override;
 
+  bool SupportForceErase() const override { return true; }
+
   void Erase(const Slice& key) override;
 
   void WaitAll(std::vector<SecondaryCacheResultHandle*> /*handles*/) override {}
 
+  Status SetCapacity(size_t capacity) override;
+
+  Status GetCapacity(size_t& capacity) override;
+
   std::string GetPrintableOptions() const override;
 
  private:
   friend class CompressedSecondaryCacheTest;
-  static constexpr std::array<uint16_t, 33> malloc_bin_sizes_{
-      32,   64,   96,   128,  160,  192,  224,   256,   320,   384,   448,
-      512,  640,  768,  896,  1024, 1280, 1536,  1792,  2048,  2560,  3072,
-      3584, 4096, 5120, 6144, 7168, 8192, 10240, 12288, 14336, 16384, 32768};
+  static constexpr std::array<uint16_t, 8> malloc_bin_sizes_{
+      128, 256, 512, 1024, 2048, 4096, 8192, 16384};
 
   struct CacheValueChunk {
     // TODO try "CacheAllocationPtr next;".
@@ -101,7 +122,7 @@ class CompressedSecondaryCache : public SecondaryCache {
   // are stored in CacheValueChunk and extra charge is needed for each chunk,
   // so the cache charge is recalculated here.
   CacheValueChunk* SplitValueIntoChunks(const Slice& value,
-                                        const CompressionType compression_type,
+                                        CompressionType compression_type,
                                         size_t& charge);
 
   // After merging chunks, the extra charge for each chunk is removed, so
@@ -109,10 +130,11 @@ class CompressedSecondaryCache : public SecondaryCache {
   CacheAllocationPtr MergeChunksIntoValue(const void* chunks_head,
                                           size_t& charge);
 
-  // An implementation of Cache::DeleterFn.
-  static void DeletionCallback(const Slice& /*key*/, void* obj);
+  // TODO: clean up to use cleaner interfaces in typed_cache.h
+  const Cache::CacheItemHelper* GetHelper(bool enable_custom_split_merge) const;
   std::shared_ptr<Cache> cache_;
   CompressedSecondaryCacheOptions cache_options_;
+  mutable port::Mutex capacity_mutex_;
 };
 
 }  // namespace ROCKSDB_NAMESPACE

cache/fast_lru_cache.cc

@@ -1,612 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-//
-// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file. See the AUTHORS file for names of contributors.
-
-#include "cache/fast_lru_cache.h"
-
-#include <cassert>
-#include <cstdint>
-#include <cstdio>
-#include <functional>
-
-#include "monitoring/perf_context_imp.h"
-#include "monitoring/statistics.h"
-#include "port/lang.h"
-#include "util/distributed_mutex.h"
-#include "util/hash.h"
-#include "util/math.h"
-#include "util/random.h"
-
-namespace ROCKSDB_NAMESPACE {
-
-namespace fast_lru_cache {
-
-LRUHandleTable::LRUHandleTable(int hash_bits)
-    : length_bits_(hash_bits),
-      length_bits_mask_((uint32_t{1} << length_bits_) - 1),
-      occupancy_(0),
-      occupancy_limit_(static_cast<uint32_t>((uint32_t{1} << length_bits_) *
-                                             kStrictLoadFactor)),
-      array_(new LRUHandle[size_t{1} << length_bits_]) {
-  assert(hash_bits <= 32);
-}
-
-LRUHandleTable::~LRUHandleTable() {
-  ApplyToEntriesRange([](LRUHandle* h) { h->FreeData(); }, 0, GetTableSize());
-}
-
-LRUHandle* LRUHandleTable::Lookup(const Slice& key, uint32_t hash) {
-  int probe = 0;
-  int slot = FindVisibleElement(key, hash, probe, 0);
-  return (slot == -1) ? nullptr : &array_[slot];
-}
-
-LRUHandle* LRUHandleTable::Insert(LRUHandle* h, LRUHandle** old) {
-  int probe = 0;
-  int slot = FindVisibleElementOrAvailableSlot(h->key(), h->hash, probe,
-                                               1 /*displacement*/);
-  *old = nullptr;
-  if (slot == -1) {
-    // TODO(Guido) Don't we need to roll back displacements here?
-    return nullptr;
-  }
-
-  if (array_[slot].IsEmpty() || array_[slot].IsTombstone()) {
-    bool empty = array_[slot].IsEmpty();
-    Assign(slot, h);
-    LRUHandle* new_entry = &array_[slot];
-    if (empty) {
-      // This used to be an empty slot.
-      return new_entry;
-    }
-    // It used to be a tombstone, so there may already be a copy of the
-    // key in the table.
-    slot = FindVisibleElement(h->key(), h->hash, probe, 0 /*displacement*/);
-    if (slot == -1) {
-      // No existing copy of the key.
-      return new_entry;
-    }
-    *old = &array_[slot];
-    return new_entry;
-  } else {
-    // There is an existing copy of the key.
-    *old = &array_[slot];
-    // Find an available slot for the new element.
-    array_[slot].displacements++;
-    slot = FindAvailableSlot(h->key(), probe, 1 /*displacement*/);
-    if (slot == -1) {
-      // No available slots. Roll back displacements.
-      probe = 0;
-      slot = FindVisibleElement(h->key(), h->hash, probe, -1);
-      array_[slot].displacements--;
-      FindAvailableSlot(h->key(), probe, -1);
-      return nullptr;
-    }
-    Assign(slot, h);
-    return &array_[slot];
-  }
-}
-
-void LRUHandleTable::Remove(LRUHandle* h) {
-  assert(h->next == nullptr &&
-         h->prev == nullptr);  // Already off the LRU list.
-  int probe = 0;
-  FindSlot(
-      h->key(), [&h](LRUHandle* e) { return e == h; }, probe,
-      -1 /*displacement*/);
-  h->SetIsVisible(false);
-  h->SetIsElement(false);
-  occupancy_--;
-}
-
-void LRUHandleTable::Assign(int slot, LRUHandle* h) {
-  LRUHandle* dst = &array_[slot];
-  uint32_t disp = dst->displacements;
-  *dst = *h;
-  dst->displacements = disp;
-  dst->SetIsVisible(true);
-  dst->SetIsElement(true);
-  occupancy_++;
-}
-
-void LRUHandleTable::Exclude(LRUHandle* h) { h->SetIsVisible(false); }
-
-int LRUHandleTable::FindVisibleElement(const Slice& key, uint32_t hash,
-                                       int& probe, int displacement) {
-  return FindSlot(
-      key,
-      [&](LRUHandle* h) { return h->Matches(key, hash) && h->IsVisible(); },
-      probe, displacement);
-}
-
-int LRUHandleTable::FindAvailableSlot(const Slice& key, int& probe,
-                                      int displacement) {
-  return FindSlot(
-      key, [](LRUHandle* h) { return h->IsEmpty() || h->IsTombstone(); }, probe,
-      displacement);
-}
-
-int LRUHandleTable::FindVisibleElementOrAvailableSlot(const Slice& key,
-                                                      uint32_t hash, int& probe,
-                                                      int displacement) {
-  return FindSlot(
-      key,
-      [&](LRUHandle* h) {
-        return h->IsEmpty() || h->IsTombstone() ||
-               (h->Matches(key, hash) && h->IsVisible());
-      },
-      probe, displacement);
-}
-
-inline int LRUHandleTable::FindSlot(const Slice& key,
-                                    std::function<bool(LRUHandle*)> cond,
-                                    int& probe, int displacement) {
-  uint32_t base = ModTableSize(Hash(key.data(), key.size(), kProbingSeed1));
-  uint32_t increment =
-      ModTableSize((Hash(key.data(), key.size(), kProbingSeed2) << 1) | 1);
-  uint32_t current = ModTableSize(base + probe * increment);
-  while (true) {
-    LRUHandle* h = &array_[current];
-    probe++;
-    if (current == base && probe > 1) {
-      // We looped back.
-      return -1;
-    }
-    if (cond(h)) {
-      return current;
-    }
-    if (h->IsEmpty()) {
-      // We check emptyness after the condition, because
-      // the condition may be emptyness.
-      return -1;
-    }
-    h->displacements += displacement;
-    current = ModTableSize(current + increment);
-  }
-}
-
-LRUCacheShard::LRUCacheShard(size_t capacity, size_t estimated_value_size,
-                             bool strict_capacity_limit,
-                             CacheMetadataChargePolicy metadata_charge_policy)
-    : capacity_(capacity),
-      strict_capacity_limit_(strict_capacity_limit),
-      table_(
-          CalcHashBits(capacity, estimated_value_size, metadata_charge_policy)),
-      usage_(0),
-      lru_usage_(0) {
-  set_metadata_charge_policy(metadata_charge_policy);
-  // Make empty circular linked list.
-  lru_.next = &lru_;
-  lru_.prev = &lru_;
-  lru_low_pri_ = &lru_;
-}
-
-void LRUCacheShard::EraseUnRefEntries() {
-  autovector<LRUHandle> last_reference_list;
-  {
-    DMutexLock l(mutex_);
-    while (lru_.next != &lru_) {
-      LRUHandle* old = lru_.next;
-      // LRU list contains only elements which can be evicted.
-      assert(old->IsVisible() && !old->HasRefs());
-      LRU_Remove(old);
-      table_.Remove(old);
-      assert(usage_ >= old->total_charge);
-      usage_ -= old->total_charge;
-      last_reference_list.push_back(*old);
-    }
-  }
-
-  // Free the entries here outside of mutex for performance reasons.
-  for (auto& h : last_reference_list) {
-    h.FreeData();
-  }
-}
-
-void LRUCacheShard::ApplyToSomeEntries(
-    const std::function<void(const Slice& key, void* value, size_t charge,
-                             DeleterFn deleter)>& callback,
-    uint32_t average_entries_per_lock, uint32_t* state) {
-  // The state is essentially going to be the starting hash, which works
-  // nicely even if we resize between calls because we use upper-most
-  // hash bits for table indexes.
-  DMutexLock l(mutex_);
-  uint32_t length_bits = table_.GetLengthBits();
-  uint32_t length = table_.GetTableSize();
-
-  assert(average_entries_per_lock > 0);
-  // Assuming we are called with same average_entries_per_lock repeatedly,
-  // this simplifies some logic (index_end will not overflow).
-  assert(average_entries_per_lock < length || *state == 0);
-
-  uint32_t index_begin = *state >> (32 - length_bits);
-  uint32_t index_end = index_begin + average_entries_per_lock;
-  if (index_end >= length) {
-    // Going to end
-    index_end = length;
-    *state = UINT32_MAX;
-  } else {
-    *state = index_end << (32 - length_bits);
-  }
-
-  table_.ApplyToEntriesRange(
-      [callback,
-       metadata_charge_policy = metadata_charge_policy_](LRUHandle* h) {
-        callback(h->key(), h->value, h->GetCharge(metadata_charge_policy),
-                 h->deleter);
-      },
-      index_begin, index_end);
-}
-
-void LRUCacheShard::LRU_Remove(LRUHandle* h) {
-  assert(h->next != nullptr);
-  assert(h->prev != nullptr);
-  h->next->prev = h->prev;
-  h->prev->next = h->next;
-  h->prev = h->next = nullptr;
-  assert(lru_usage_ >= h->total_charge);
-  lru_usage_ -= h->total_charge;
-}
-
-void LRUCacheShard::LRU_Insert(LRUHandle* h) {
-  assert(h->next == nullptr);
-  assert(h->prev == nullptr);
-  // Insert h to head of LRU list.
-  h->next = &lru_;
-  h->prev = lru_.prev;
-  h->prev->next = h;
-  h->next->prev = h;
-  lru_usage_ += h->total_charge;
-}
-
-void LRUCacheShard::EvictFromLRU(size_t charge,
-                                 autovector<LRUHandle>* deleted) {
-  while ((usage_ + charge) > capacity_ && lru_.next != &lru_) {
-    LRUHandle* old = lru_.next;
-    // LRU list contains only elements which can be evicted.
-    assert(old->IsVisible() && !old->HasRefs());
-    LRU_Remove(old);
-    table_.Remove(old);
-    assert(usage_ >= old->total_charge);
-    usage_ -= old->total_charge;
-    deleted->push_back(*old);
-  }
-}
-
-size_t LRUCacheShard::CalcEstimatedHandleCharge(
-    size_t estimated_value_size,
-    CacheMetadataChargePolicy metadata_charge_policy) {
-  LRUHandle h;
-  h.CalcTotalCharge(estimated_value_size, metadata_charge_policy);
-  return h.total_charge;
-}
-
-int LRUCacheShard::CalcHashBits(
-    size_t capacity, size_t estimated_value_size,
-    CacheMetadataChargePolicy metadata_charge_policy) {
-  size_t handle_charge =
-      CalcEstimatedHandleCharge(estimated_value_size, metadata_charge_policy);
-  assert(handle_charge > 0);
-  uint32_t num_entries =
-      static_cast<uint32_t>(capacity / (kLoadFactor * handle_charge)) + 1;
-  assert(num_entries <= uint32_t{1} << 31);
-  return FloorLog2((num_entries << 1) - 1);
-}
-
-void LRUCacheShard::SetCapacity(size_t capacity) {
-  autovector<LRUHandle> last_reference_list;
-  {
-    DMutexLock l(mutex_);
-    if (capacity > capacity_) {
-      assert(false);  // Not supported.
-    }
-    capacity_ = capacity;
-    EvictFromLRU(0, &last_reference_list);
-  }
-
-  // Free the entries here outside of mutex for performance reasons.
-  for (auto& h : last_reference_list) {
-    h.FreeData();
-  }
-}
-
-void LRUCacheShard::SetStrictCapacityLimit(bool strict_capacity_limit) {
-  DMutexLock l(mutex_);
-  strict_capacity_limit_ = strict_capacity_limit;
-}
-
-Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
-                             size_t charge, Cache::DeleterFn deleter,
-                             Cache::Handle** handle,
-                             Cache::Priority /*priority*/) {
-  if (key.size() != kCacheKeySize) {
-    return Status::NotSupported("FastLRUCache only supports key size " +
-                                std::to_string(kCacheKeySize) + "B");
-  }
-
-  LRUHandle tmp;
-  tmp.value = value;
-  tmp.deleter = deleter;
-  tmp.hash = hash;
-  tmp.CalcTotalCharge(charge, metadata_charge_policy_);
-  for (int i = 0; i < kCacheKeySize; i++) {
-    tmp.key_data[i] = key.data()[i];
-  }
-
-  Status s = Status::OK();
-  autovector<LRUHandle> last_reference_list;
-  {
-    DMutexLock l(mutex_);
-    assert(table_.GetOccupancy() <= table_.GetOccupancyLimit());
-
-    // Free the space following strict LRU policy until enough space
-    // is freed or the lru list is empty.
-    EvictFromLRU(tmp.total_charge, &last_reference_list);
-    if ((usage_ + tmp.total_charge > capacity_ &&
-         (strict_capacity_limit_ || handle == nullptr)) ||
-        table_.GetOccupancy() == table_.GetOccupancyLimit()) {
-      // There are two measures of capacity:
-      // - Space (or charge) capacity: The maximum possible sum of the charges
-      //    of the elements.
-      // - Table capacity: The number of slots in the hash table.
-      // These are incomparable, in the sense that one doesn't imply the other.
-      // Typically we will reach space capacity before table capacity---
-      // if the user always inserts values with size equal to
-      // estimated_value_size, then at most a kLoadFactor fraction of slots
-      // will ever be occupied. But in some cases we may reach table capacity
-      // before space capacity---if the user initially claims a very large
-      // estimated_value_size but then inserts tiny values, more elements than
-      // initially estimated will be inserted.
-
-      // TODO(Guido) Some tests (at least two from cache_test, as well as the
-      // stress tests) currently assume the table capacity is unbounded.
-      if (handle == nullptr) {
-        // Don't insert the entry but still return ok, as if the entry inserted
-        // into cache and get evicted immediately.
-        last_reference_list.push_back(tmp);
-      } else {
-        if (table_.GetOccupancy() == table_.GetOccupancyLimit()) {
-          // TODO: Consider using a distinct status for this case, but usually
-          // it will be handled the same way as reaching charge capacity limit
-          s = Status::MemoryLimit(
-              "Insert failed because all slots in the hash table are full.");
-        } else {
-          s = Status::MemoryLimit(
-              "Insert failed because the total charge has exceeded the "
-              "capacity.");
-        }
-      }
-    } else {
-      // Insert into the cache. Note that the cache might get larger than its
-      // capacity if not enough space was freed up.
-      LRUHandle* old;
-      LRUHandle* h = table_.Insert(&tmp, &old);
-      assert(h != nullptr);  // We're below occupancy, so this insertion should
-                             // never fail.
-      usage_ += h->total_charge;
-      if (old != nullptr) {
-        s = Status::OkOverwritten();
-        assert(old->IsVisible());
-        table_.Exclude(old);
-        if (!old->HasRefs()) {
-          // old is on LRU because it's in cache and its reference count is 0.
-          LRU_Remove(old);
-          table_.Remove(old);
-          assert(usage_ >= old->total_charge);
-          usage_ -= old->total_charge;
-          last_reference_list.push_back(*old);
-        }
-      }
-      if (handle == nullptr) {
-        LRU_Insert(h);
-      } else {
-        // If caller already holds a ref, no need to take one here.
-        if (!h->HasRefs()) {
-          h->Ref();
-        }
-        *handle = reinterpret_cast<Cache::Handle*>(h);
-      }
-    }
-  }
-
-  // Free the entries here outside of mutex for performance reasons.
-  for (auto& h : last_reference_list) {
-    h.FreeData();
-  }
-
-  return s;
-}
-
-Cache::Handle* LRUCacheShard::Lookup(const Slice& key, uint32_t hash) {
-  LRUHandle* h = nullptr;
-  {
-    DMutexLock l(mutex_);
-    h = table_.Lookup(key, hash);
-    if (h != nullptr) {
-      assert(h->IsVisible());
-      if (!h->HasRefs()) {
-        // The entry is in LRU since it's in hash and has no external
-        // references.
-        LRU_Remove(h);
-      }
-      h->Ref();
-    }
-  }
-  return reinterpret_cast<Cache::Handle*>(h);
-}
-
-bool LRUCacheShard::Ref(Cache::Handle* h) {
-  LRUHandle* e = reinterpret_cast<LRUHandle*>(h);
-  DMutexLock l(mutex_);
-  // To create another reference - entry must be already externally referenced.
-  assert(e->HasRefs());
-  e->Ref();
-  return true;
-}
-
-bool LRUCacheShard::Release(Cache::Handle* handle, bool erase_if_last_ref) {
-  if (handle == nullptr) {
-    return false;
-  }
-  LRUHandle* h = reinterpret_cast<LRUHandle*>(handle);
-  LRUHandle copy;
-  bool last_reference = false;
-  {
-    DMutexLock l(mutex_);
-    last_reference = h->Unref();
-    if (last_reference && h->IsVisible()) {
-      // The item is still in cache, and nobody else holds a reference to it.
-      if (usage_ > capacity_ || erase_if_last_ref) {
-        // The LRU list must be empty since the cache is full.
-        assert(lru_.next == &lru_ || erase_if_last_ref);
-        // Take this opportunity and remove the item.
-        table_.Remove(h);
-      } else {
-        // Put the item back on the LRU list, and don't free it.
-        LRU_Insert(h);
-        last_reference = false;
-      }
-    }
-    // If it was the last reference, then decrement the cache usage.
-    if (last_reference) {
-      assert(usage_ >= h->total_charge);
-      usage_ -= h->total_charge;
-      copy = *h;
-    }
-  }
-
-  // Free the entry here outside of mutex for performance reasons.
-  if (last_reference) {
-    copy.FreeData();
-  }
-  return last_reference;
-}
-
-void LRUCacheShard::Erase(const Slice& key, uint32_t hash) {
-  LRUHandle copy;
-  bool last_reference = false;
-  {
-    DMutexLock l(mutex_);
-    LRUHandle* h = table_.Lookup(key, hash);
-    if (h != nullptr) {
-      table_.Exclude(h);
-      if (!h->HasRefs()) {
-        // The entry is in LRU since it's in cache and has no external
-        // references.
-        LRU_Remove(h);
-        table_.Remove(h);
-        assert(usage_ >= h->total_charge);
-        usage_ -= h->total_charge;
-        last_reference = true;
-        copy = *h;
-      }
-    }
-  }
-  // Free the entry here outside of mutex for performance reasons.
-  // last_reference will only be true if e != nullptr.
-  if (last_reference) {
-    copy.FreeData();
-  }
-}
-
-size_t LRUCacheShard::GetUsage() const {
-  DMutexLock l(mutex_);
-  return usage_;
-}
-
-size_t LRUCacheShard::GetPinnedUsage() const {
-  DMutexLock l(mutex_);
-  assert(usage_ >= lru_usage_);
-  return usage_ - lru_usage_;
-}
-
-std::string LRUCacheShard::GetPrintableOptions() const { return std::string{}; }
-
-LRUCache::LRUCache(size_t capacity, size_t estimated_value_size,
-                   int num_shard_bits, bool strict_capacity_limit,
-                   CacheMetadataChargePolicy metadata_charge_policy)
-    : ShardedCache(capacity, num_shard_bits, strict_capacity_limit) {
-  assert(estimated_value_size > 0 ||
-         metadata_charge_policy != kDontChargeCacheMetadata);
-  num_shards_ = 1 << num_shard_bits;
-  shards_ = reinterpret_cast<LRUCacheShard*>(
-      port::cacheline_aligned_alloc(sizeof(LRUCacheShard) * num_shards_));
-  size_t per_shard = (capacity + (num_shards_ - 1)) / num_shards_;
-  for (int i = 0; i < num_shards_; i++) {
-    new (&shards_[i])
-        LRUCacheShard(per_shard, estimated_value_size, strict_capacity_limit,
-                      metadata_charge_policy);
-  }
-}
-
-LRUCache::~LRUCache() {
-  if (shards_ != nullptr) {
-    assert(num_shards_ > 0);
-    for (int i = 0; i < num_shards_; i++) {
-      shards_[i].~LRUCacheShard();
-    }
-    port::cacheline_aligned_free(shards_);
-  }
-}
-
-CacheShard* LRUCache::GetShard(uint32_t shard) {
-  return reinterpret_cast<CacheShard*>(&shards_[shard]);
-}
-
-const CacheShard* LRUCache::GetShard(uint32_t shard) const {
-  return reinterpret_cast<CacheShard*>(&shards_[shard]);
-}
-
-void* LRUCache::Value(Handle* handle) {
-  return reinterpret_cast<const LRUHandle*>(handle)->value;
-}
-
-size_t LRUCache::GetCharge(Handle* handle) const {
-  CacheMetadataChargePolicy metadata_charge_policy = kDontChargeCacheMetadata;
-  if (num_shards_ > 0) {
-    metadata_charge_policy = shards_[0].metadata_charge_policy_;
-  }
-  return reinterpret_cast<const LRUHandle*>(handle)->GetCharge(
-      metadata_charge_policy);
-}
-
-Cache::DeleterFn LRUCache::GetDeleter(Handle* handle) const {
-  auto h = reinterpret_cast<const LRUHandle*>(handle);
-  return h->deleter;
-}
-
-uint32_t LRUCache::GetHash(Handle* handle) const {
-  return reinterpret_cast<const LRUHandle*>(handle)->hash;
-}
-
-void LRUCache::DisownData() {
-  // Leak data only if that won't generate an ASAN/valgrind warning.
-  if (!kMustFreeHeapAllocations) {
-    shards_ = nullptr;
-    num_shards_ = 0;
-  }
-}
-
-}  // namespace fast_lru_cache
-
-std::shared_ptr<Cache> NewFastLRUCache(
-    size_t capacity, size_t estimated_value_size, int num_shard_bits,
-    bool strict_capacity_limit,
-    CacheMetadataChargePolicy metadata_charge_policy) {
-  if (num_shard_bits >= 20) {
-    return nullptr;  // The cache cannot be sharded into too many fine pieces.
-  }
-  if (num_shard_bits < 0) {
-    num_shard_bits = GetDefaultCacheShardBits(capacity);
-  }
-  return std::make_shared<fast_lru_cache::LRUCache>(
-      capacity, estimated_value_size, num_shard_bits, strict_capacity_limit,
-      metadata_charge_policy);
-}
-
-}  // namespace ROCKSDB_NAMESPACE

cache/fast_lru_cache.h

@@ -1,474 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-//
-// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file. See the AUTHORS file for names of contributors.
-#pragma once
-
-#include <array>
-#include <memory>
-#include <string>
-
-#include "cache/cache_key.h"
-#include "cache/sharded_cache.h"
-#include "port/lang.h"
-#include "port/malloc.h"
-#include "port/port.h"
-#include "rocksdb/secondary_cache.h"
-#include "util/autovector.h"
-#include "util/distributed_mutex.h"
-
-namespace ROCKSDB_NAMESPACE {
-
-namespace fast_lru_cache {
-
-// Forward declaration of friend class.
-class FastLRUCacheTest;
-
-// LRU cache implementation using an open-address hash table.
-//
-// Every slot in the hash table is an LRUHandle. Because handles can be
-// referenced externally, we can't discard them immediately once they are
-// deleted (via a delete or an LRU eviction) or replaced by a new version
-// (via an insert of the same key). The state of an element is defined by
-// the following two properties:
-// (R) Referenced: An element can be referenced externally (refs > 0), or not.
-//    Importantly, an element can be evicted if and only if it's not
-//    referenced. In particular, when an element becomes referenced, it's
-//    temporarily taken out of the LRU list until all references to it
-//    are dropped.
-// (V) Visible: An element can visible for lookups (IS_VISIBLE set), or not.
-//    Initially, every element is visible. An element that is not visible is
-//    called a ghost.
-// These properties induce 4 different states, with transitions defined as
-// follows:
-// - V --> not V: When a visible element is deleted or replaced by a new
-//    version.
-// - Not V --> V: This cannot happen. A ghost remains in that state until it's
-//    not referenced any more, at which point it's ready to be removed from the
-//    hash table. (A ghost simply waits to transition to the afterlife---it will
-//    never be visible again.)
-// - R --> not R: When all references to an element are dropped.
-// - Not R --> R: When an unreferenced element becomes referenced. This can only
-//    happen if the element is V, since references to an element can only be
-//    created when it's visible.
-//
-// Internally, the cache uses an open-addressed hash table to index the handles.
-// We use tombstone counters to keep track of displacements.
-// Because of the tombstones and the two possible visibility states of an
-// element, the table slots can be in 4 different states:
-// 1. Visible element (IS_ELEMENT set and IS_VISIBLE set): The slot contains a
-//    key-value element.
-// 2. Ghost element (IS_ELEMENT set and IS_VISIBLE unset): The slot contains an
-//    element that has been removed, but it's still referenced. It's invisible
-//    to lookups.
-// 3. Tombstone (IS_ELEMENT unset and displacements > 0): The slot contains a
-//    tombstone.
-// 4. Empty (IS_ELEMENT unset and displacements == 0): The slot is unused.
-//    A slot that is an element can further have IS_VISIBLE set or not.
-// When a ghost is removed from the table, it can either transition to being a
-// tombstone or an empty slot, depending on the number of displacements of the
-// slot. In any case, the slot becomes available. When a handle is inserted
-// into that slot, it becomes a visible element again.
-
-// The load factor p is a real number in (0, 1) such that at all
-// times at most a fraction p of all slots, without counting tombstones,
-// are occupied by elements. This means that the probability that a
-// random probe hits an empty slot is at most p, and thus at most 1/p probes
-// are required on average. For example, p = 70% implies that between 1 and 2
-// probes are needed on average (bear in mind that this reasoning doesn't
-// consider the effects of clustering over time).
-// Because the size of the hash table is always rounded up to the next
-// power of 2, p is really an upper bound on the actual load factor---the
-// actual load factor is anywhere between p/2 and p. This is a bit wasteful,
-// but bear in mind that slots only hold metadata, not actual values.
-// Since space cost is dominated by the values (the LSM blocks),
-// overprovisioning the table with metadata only increases the total cache space
-// usage by a tiny fraction.
-constexpr double kLoadFactor = 0.35;
-
-// The user can exceed kLoadFactor if the sizes of the inserted values don't
-// match estimated_value_size, or if strict_capacity_limit == false. To
-// avoid performance to plunge, we set a strict upper bound on the load factor.
-constexpr double kStrictLoadFactor = 0.7;
-
-// Arbitrary seeds.
-constexpr uint32_t kProbingSeed1 = 0xbc9f1d34;
-constexpr uint32_t kProbingSeed2 = 0x7a2bb9d5;
-
-// An experimental (under development!) alternative to LRUCache
-
-struct LRUHandle {
-  void* value;
-  Cache::DeleterFn deleter;
-  LRUHandle* next;
-  LRUHandle* prev;
-  size_t total_charge;  // TODO(opt): Only allow uint32_t?
-  // The hash of key(). Used for fast sharding and comparisons.
-  uint32_t hash;
-  // The number of external refs to this entry.
-  uint32_t refs;
-
-  enum Flags : uint8_t {
-    // Whether the handle is visible to Lookups.
-    IS_VISIBLE = (1 << 0),
-    // Whether the slot is in use by an element.
-    IS_ELEMENT = (1 << 1),
-  };
-  uint8_t flags;
-
-  // The number of elements that hash to this slot or a lower one,
-  // but wind up in a higher slot.
-  uint32_t displacements;
-
-  std::array<char, kCacheKeySize> key_data;
-
-  LRUHandle() {
-    value = nullptr;
-    deleter = nullptr;
-    next = nullptr;
-    prev = nullptr;
-    total_charge = 0;
-    hash = 0;
-    refs = 0;
-    flags = 0;
-    displacements = 0;
-    key_data.fill(0);
-  }
-
-  Slice key() const { return Slice(key_data.data(), kCacheKeySize); }
-
-  // Increase the reference count by 1.
-  void Ref() { refs++; }
-
-  // Just reduce the reference count by 1. Return true if it was last reference.
-  bool Unref() {
-    assert(refs > 0);
-    refs--;
-    return refs == 0;
-  }
-
-  // Return true if there are external refs, false otherwise.
-  bool HasRefs() const { return refs > 0; }
-
-  bool IsVisible() const { return flags & IS_VISIBLE; }
-
-  void SetIsVisible(bool is_visible) {
-    if (is_visible) {
-      flags |= IS_VISIBLE;
-    } else {
-      flags &= ~IS_VISIBLE;
-    }
-  }
-
-  bool IsElement() const { return flags & IS_ELEMENT; }
-
-  void SetIsElement(bool is_element) {
-    if (is_element) {
-      flags |= IS_ELEMENT;
-    } else {
-      flags &= ~IS_ELEMENT;
-    }
-  }
-
-  void FreeData() {
-    assert(refs == 0);
-    if (deleter) {
-      (*deleter)(key(), value);
-    }
-  }
-
-  // Calculate the memory usage by metadata.
-  inline size_t CalcMetaCharge(
-      CacheMetadataChargePolicy metadata_charge_policy) const {
-    if (metadata_charge_policy != kFullChargeCacheMetadata) {
-      return 0;
-    } else {
-      // #ifdef ROCKSDB_MALLOC_USABLE_SIZE
-      //       return malloc_usable_size(
-      //           const_cast<void*>(static_cast<const void*>(this)));
-      // #else
-      // TODO(Guido) malloc_usable_size only works when we call it on
-      // a pointer allocated with malloc. Because our handles are all
-      // allocated in a single shot as an array, the user can't call
-      // CalcMetaCharge (or CalcTotalCharge or GetCharge) on a handle
-      // pointer returned by the cache. Moreover, malloc_usable_size
-      // expects a heap-allocated handle, but sometimes in our code we
-      // wish to pass a stack-allocated handle (this is only a performance
-      // concern).
-      // What is the right way to compute metadata charges with pre-allocated
-      // handles?
-      return sizeof(LRUHandle);
-      // #endif
-    }
-  }
-
-  inline void CalcTotalCharge(
-      size_t charge, CacheMetadataChargePolicy metadata_charge_policy) {
-    total_charge = charge + CalcMetaCharge(metadata_charge_policy);
-  }
-
-  inline size_t GetCharge(
-      CacheMetadataChargePolicy metadata_charge_policy) const {
-    size_t meta_charge = CalcMetaCharge(metadata_charge_policy);
-    assert(total_charge >= meta_charge);
-    return total_charge - meta_charge;
-  }
-
-  inline bool IsEmpty() {
-    return !this->IsElement() && this->displacements == 0;
-  }
-
-  inline bool IsTombstone() {
-    return !this->IsElement() && this->displacements > 0;
-  }
-
-  inline bool Matches(const Slice& some_key, uint32_t some_hash) {
-    return this->IsElement() && this->hash == some_hash &&
-           this->key() == some_key;
-  }
-};
-
-
-class LRUHandleTable {
- public:
-  explicit LRUHandleTable(int hash_bits);
-  ~LRUHandleTable();
-
-  // Returns a pointer to a visible element matching the key/hash, or
-  // nullptr if not present.
-  LRUHandle* Lookup(const Slice& key, uint32_t hash);
-
-  // Inserts a copy of h into the hash table.
-  // Returns a pointer to the inserted handle, or nullptr if no slot
-  // available was found. If an existing visible element matching the
-  // key/hash is already present in the hash table, the argument old
-  // is set to pointe to it; otherwise, it's set to nullptr.
-  LRUHandle* Insert(LRUHandle* h, LRUHandle** old);
-
-  // Removes h from the hash table. The handle must already be off
-  // the LRU list.
-  void Remove(LRUHandle* h);
-
-  // Turns a visible element h into a ghost (i.e., not visible).
-  void Exclude(LRUHandle* h);
-
-  // Assigns a copy of h to the given slot.
-  void Assign(int slot, LRUHandle* h);
-
-  template <typename T>
-  void ApplyToEntriesRange(T func, uint32_t index_begin, uint32_t index_end) {
-    for (uint32_t i = index_begin; i < index_end; i++) {
-      LRUHandle* h = &array_[i];
-      if (h->IsVisible()) {
-        func(h);
-      }
-    }
-  }
-
-  uint32_t GetTableSize() const { return uint32_t{1} << length_bits_; }
-
-  int GetLengthBits() const { return length_bits_; }
-
-  uint32_t GetOccupancyLimit() const { return occupancy_limit_; }
-
-  uint32_t GetOccupancy() const { return occupancy_; }
-
-  // Returns x mod 2^{length_bits_}.
-  uint32_t ModTableSize(uint32_t x) { return x & length_bits_mask_; }
-
- private:
-  int FindVisibleElement(const Slice& key, uint32_t hash, int& probe,
-                         int displacement);
-
-  int FindAvailableSlot(const Slice& key, int& probe, int displacement);
-
-  int FindVisibleElementOrAvailableSlot(const Slice& key, uint32_t hash,
-                                        int& probe, int displacement);
-
-  // Returns the index of the first slot probed (hashing with
-  // the given key) with a handle e such that cond(e) is true.
-  // Otherwise, if no match is found, returns -1.
-  // For every handle e probed except the final slot, updates
-  // e->displacements += displacement.
-  // The argument probe is modified such that consecutive calls
-  // to FindSlot continue probing right after where the previous
-  // call left.
-  int FindSlot(const Slice& key, std::function<bool(LRUHandle*)> cond,
-               int& probe, int displacement);
-
-  // Number of hash bits used for table index.
-  // The size of the table is 1 << length_bits_.
-  int length_bits_;
-
-  const uint32_t length_bits_mask_;
-
-  // Number of elements in the table.
-  uint32_t occupancy_;
-
-  // Maximum number of elements the user can store in the table.
-  uint32_t occupancy_limit_;
-
-  std::unique_ptr<LRUHandle[]> array_;
-};
-
-// A single shard of sharded cache.
-class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
- public:
-  LRUCacheShard(size_t capacity, size_t estimated_value_size,
-                bool strict_capacity_limit,
-                CacheMetadataChargePolicy metadata_charge_policy);
-  ~LRUCacheShard() override = default;
-
-  // Separate from constructor so caller can easily make an array of LRUCache
-  // if current usage is more than new capacity, the function will attempt to
-  // free the needed space.
-  void SetCapacity(size_t capacity) override;
-
-  // Set the flag to reject insertion if cache if full.
-  void SetStrictCapacityLimit(bool strict_capacity_limit) override;
-
-  // Like Cache methods, but with an extra "hash" parameter.
-  // Insert an item into the hash table and, if handle is null, insert into
-  // the LRU list. Older items are evicted as necessary. If the cache is full
-  // and free_handle_on_fail is true, the item is deleted and handle is set to
-  // nullptr.
-  Status Insert(const Slice& key, uint32_t hash, void* value, size_t charge,
-                Cache::DeleterFn deleter, Cache::Handle** handle,
-                Cache::Priority priority) override;
-
-  Status Insert(const Slice& key, uint32_t hash, void* value,
-                const Cache::CacheItemHelper* helper, size_t charge,
-                Cache::Handle** handle, Cache::Priority priority) override {
-    return Insert(key, hash, value, charge, helper->del_cb, handle, priority);
-  }
-
-  Cache::Handle* Lookup(const Slice& key, uint32_t hash,
-                        const Cache::CacheItemHelper* /*helper*/,
-                        const Cache::CreateCallback& /*create_cb*/,
-                        Cache::Priority /*priority*/, bool /*wait*/,
-                        Statistics* /*stats*/) override {
-    return Lookup(key, hash);
-  }
-  Cache::Handle* Lookup(const Slice& key, uint32_t hash) override;
-
-  bool Release(Cache::Handle* handle, bool /*useful*/,
-               bool erase_if_last_ref) override {
-    return Release(handle, erase_if_last_ref);
-  }
-  bool IsReady(Cache::Handle* /*handle*/) override { return true; }
-  void Wait(Cache::Handle* /*handle*/) override {}
-
-  bool Ref(Cache::Handle* handle) override;
-  bool Release(Cache::Handle* handle, bool erase_if_last_ref = false) override;
-  void Erase(const Slice& key, uint32_t hash) override;
-
-  size_t GetUsage() const override;
-  size_t GetPinnedUsage() const override;
-
-  void ApplyToSomeEntries(
-      const std::function<void(const Slice& key, void* value, size_t charge,
-                               DeleterFn deleter)>& callback,
-      uint32_t average_entries_per_lock, uint32_t* state) override;
-
-  void EraseUnRefEntries() override;
-
-  std::string GetPrintableOptions() const override;
-
- private:
-  friend class LRUCache;
-  friend class FastLRUCacheTest;
-
-  void LRU_Remove(LRUHandle* e);
-  void LRU_Insert(LRUHandle* e);
-
-  // Free some space following strict LRU policy until enough space
-  // to hold (usage_ + charge) is freed or the LRU list is empty
-  // This function is not thread safe - it needs to be executed while
-  // holding the mutex_.
-  void EvictFromLRU(size_t charge, autovector<LRUHandle>* deleted);
-
-  // Returns the charge of a single handle.
-  static size_t CalcEstimatedHandleCharge(
-      size_t estimated_value_size,
-      CacheMetadataChargePolicy metadata_charge_policy);
-
-  // Returns the number of bits used to hash an element in the hash
-  // table.
-  static int CalcHashBits(size_t capacity, size_t estimated_value_size,
-                          CacheMetadataChargePolicy metadata_charge_policy);
-
-  // Initialized before use.
-  size_t capacity_;
-
-  // Whether to reject insertion if cache reaches its full capacity.
-  bool strict_capacity_limit_;
-
-  // Dummy head of LRU list.
-  // lru.prev is newest entry, lru.next is oldest entry.
-  // LRU contains items which can be evicted, ie reference only by cache
-  LRUHandle lru_;
-
-  // Pointer to head of low-pri pool in LRU list.
-  LRUHandle* lru_low_pri_;
-
-  // ------------^^^^^^^^^^^^^-----------
-  // Not frequently modified data members
-  // ------------------------------------
-  //
-  // We separate data members that are updated frequently from the ones that
-  // are not frequently updated so that they don't share the same cache line
-  // which will lead into false cache sharing
-  //
-  // ------------------------------------
-  // Frequently modified data members
-  // ------------vvvvvvvvvvvvv-----------
-  LRUHandleTable table_;
-
-  // Memory size for entries residing in the cache.
-  size_t usage_;
-
-  // Memory size for entries residing only in the LRU list.
-  size_t lru_usage_;
-
-  // mutex_ protects the following state.
-  // We don't count mutex_ as the cache's internal state so semantically we
-  // don't mind mutex_ invoking the non-const actions.
-  mutable DMutex mutex_;
-};
-
-class LRUCache
-#ifdef NDEBUG
-    final
-#endif
-    : public ShardedCache {
- public:
-  LRUCache(size_t capacity, size_t estimated_value_size, int num_shard_bits,
-           bool strict_capacity_limit,
-           CacheMetadataChargePolicy metadata_charge_policy =
-               kDontChargeCacheMetadata);
-  ~LRUCache() override;
-  const char* Name() const override { return "LRUCache"; }
-  CacheShard* GetShard(uint32_t shard) override;
-  const CacheShard* GetShard(uint32_t shard) const override;
-  void* Value(Handle* handle) override;
-  size_t GetCharge(Handle* handle) const override;
-  uint32_t GetHash(Handle* handle) const override;
-  DeleterFn GetDeleter(Handle* handle) const override;
-  void DisownData() override;
-
- private:
-  LRUCacheShard* shards_ = nullptr;
-  int num_shards_ = 0;
-};
-}  // namespace fast_lru_cache
-
-std::shared_ptr<Cache> NewFastLRUCache(
-    size_t capacity, size_t estimated_value_size, int num_shard_bits,
-    bool strict_capacity_limit,
-    CacheMetadataChargePolicy metadata_charge_policy);
-
-}  // namespace ROCKSDB_NAMESPACE

cache/lru_cache.cc

@@ -12,6 +12,7 @@
 #include <cassert>
 #include <cstdint>
 #include <cstdio>
+#include <cstdlib>
 
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/statistics.h"
@@ -21,20 +22,31 @@
 namespace ROCKSDB_NAMESPACE {
 namespace lru_cache {
 
-LRUHandleTable::LRUHandleTable(int max_upper_hash_bits)
+namespace {
+// A distinct pointer value for marking "dummy" cache entries
+struct DummyValue {
+  char val[12] = "kDummyValue";
+};
+DummyValue kDummyValue{};
+}  // namespace
+
+LRUHandleTable::LRUHandleTable(int max_upper_hash_bits,
+                               MemoryAllocator* allocator)
     : length_bits_(/* historical starting size*/ 4),
       list_(new LRUHandle* [size_t{1} << length_bits_] {}),
       elems_(0),
-      max_length_bits_(max_upper_hash_bits) {}
+      max_length_bits_(max_upper_hash_bits),
+      allocator_(allocator) {}
 
 LRUHandleTable::~LRUHandleTable() {
+  auto alloc = allocator_;
   ApplyToEntriesRange(
-      [](LRUHandle* h) {
+      [alloc](LRUHandle* h) {
         if (!h->HasRefs()) {
-          h->Free();
+          h->Free(alloc);
         }
       },
-      0, uint32_t{1} << length_bits_);
+      0, size_t{1} << length_bits_);
 }
 
 LRUHandle* LRUHandleTable::Lookup(const Slice& key, uint32_t hash) {
@@ -109,12 +121,15 @@ void LRUHandleTable::Resize() {
   length_bits_ = new_length_bits;
 }
 
-LRUCacheShard::LRUCacheShard(
-    size_t capacity, bool strict_capacity_limit, double high_pri_pool_ratio,
-    double low_pri_pool_ratio, bool use_adaptive_mutex,
-    CacheMetadataChargePolicy metadata_charge_policy, int max_upper_hash_bits,
-    const std::shared_ptr<SecondaryCache>& secondary_cache)
-    : capacity_(0),
+LRUCacheShard::LRUCacheShard(size_t capacity, bool strict_capacity_limit,
+                             double high_pri_pool_ratio,
+                             double low_pri_pool_ratio, bool use_adaptive_mutex,
+                             CacheMetadataChargePolicy metadata_charge_policy,
+                             int max_upper_hash_bits,
+                             MemoryAllocator* allocator,
+                             SecondaryCache* secondary_cache)
+    : CacheShardBase(metadata_charge_policy),
+      capacity_(0),
       high_pri_pool_usage_(0),
       low_pri_pool_usage_(0),
       strict_capacity_limit_(strict_capacity_limit),
@@ -122,12 +137,11 @@ LRUCacheShard::LRUCacheShard(
       high_pri_pool_capacity_(0),
       low_pri_pool_ratio_(low_pri_pool_ratio),
       low_pri_pool_capacity_(0),
-      table_(max_upper_hash_bits),
+      table_(max_upper_hash_bits, allocator),
       usage_(0),
       lru_usage_(0),
       mutex_(use_adaptive_mutex),
       secondary_cache_(secondary_cache) {
-  set_metadata_charge_policy(metadata_charge_policy);
   // Make empty circular linked list.
   lru_.next = &lru_;
   lru_.prev = &lru_;
@@ -154,44 +168,42 @@ void LRUCacheShard::EraseUnRefEntries() {
   }
 
   for (auto entry : last_reference_list) {
-    entry->Free();
+    entry->Free(table_.GetAllocator());
   }
 }
 
 void LRUCacheShard::ApplyToSomeEntries(
-    const std::function<void(const Slice& key, void* value, size_t charge,
-                             DeleterFn deleter)>& callback,
-    uint32_t average_entries_per_lock, uint32_t* state) {
+    const std::function<void(const Slice& key, Cache::ObjectPtr value,
+                             size_t charge,
+                             const Cache::CacheItemHelper* helper)>& callback,
+    size_t average_entries_per_lock, size_t* state) {
   // The state is essentially going to be the starting hash, which works
   // nicely even if we resize between calls because we use upper-most
   // hash bits for table indexes.
   DMutexLock l(mutex_);
-  uint32_t length_bits = table_.GetLengthBits();
-  uint32_t length = uint32_t{1} << length_bits;
+  int length_bits = table_.GetLengthBits();
+  size_t length = size_t{1} << length_bits;
 
   assert(average_entries_per_lock > 0);
   // Assuming we are called with same average_entries_per_lock repeatedly,
   // this simplifies some logic (index_end will not overflow).
   assert(average_entries_per_lock < length || *state == 0);
 
-  uint32_t index_begin = *state >> (32 - length_bits);
-  uint32_t index_end = index_begin + average_entries_per_lock;
+  size_t index_begin = *state >> (sizeof(size_t) * 8u - length_bits);
+  size_t index_end = index_begin + average_entries_per_lock;
   if (index_end >= length) {
     // Going to end
     index_end = length;
-    *state = UINT32_MAX;
+    *state = SIZE_MAX;
   } else {
-    *state = index_end << (32 - length_bits);
+    *state = index_end << (sizeof(size_t) * 8u - length_bits);
   }
 
   table_.ApplyToEntriesRange(
       [callback,
        metadata_charge_policy = metadata_charge_policy_](LRUHandle* h) {
-        DeleterFn deleter = h->IsSecondaryCacheCompatible()
-                                ? h->info_.helper->del_cb
-                                : h->info_.deleter;
         callback(h->key(), h->value, h->GetCharge(metadata_charge_policy),
-                 deleter);
+                 h->helper);
       },
       index_begin, index_end);
 }
@@ -329,6 +341,19 @@ void LRUCacheShard::EvictFromLRU(size_t charge,
   }
 }
 
+void LRUCacheShard::TryInsertIntoSecondaryCache(
+    autovector<LRUHandle*> evicted_handles) {
+  for (auto entry : evicted_handles) {
+    if (secondary_cache_ && entry->IsSecondaryCacheCompatible() &&
+        !entry->IsInSecondaryCache()) {
+      secondary_cache_->Insert(entry->key(), entry->value, entry->helper)
+          .PermitUncheckedError();
+    }
+    // Free the entries here outside of mutex for performance reasons.
+    entry->Free(table_.GetAllocator());
+  }
+}
+
 void LRUCacheShard::SetCapacity(size_t capacity) {
   autovector<LRUHandle*> last_reference_list;
   {
@@ -339,16 +364,7 @@ void LRUCacheShard::SetCapacity(size_t capacity) {
     EvictFromLRU(0, &last_reference_list);
   }
 
-  // Try to insert the evicted entries into tiered cache.
-  // Free the entries outside of mutex for performance reasons.
-  for (auto entry : last_reference_list) {
-    if (secondary_cache_ && entry->IsSecondaryCacheCompatible() &&
-        !entry->IsInSecondaryCache()) {
-      secondary_cache_->Insert(entry->key(), entry->value, entry->info_.helper)
-          .PermitUncheckedError();
-    }
-    entry->Free();
-  }
+  TryInsertIntoSecondaryCache(last_reference_list);
 }
 
 void LRUCacheShard::SetStrictCapacityLimit(bool strict_capacity_limit) {
@@ -356,7 +372,7 @@ void LRUCacheShard::SetStrictCapacityLimit(bool strict_capacity_limit) {
   strict_capacity_limit_ = strict_capacity_limit;
 }
 
-Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
+Status LRUCacheShard::InsertItem(LRUHandle* e, LRUHandle** handle,
                                  bool free_handle_on_fail) {
   Status s = Status::OK();
   autovector<LRUHandle*> last_reference_list;
@@ -377,7 +393,7 @@ Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
         last_reference_list.push_back(e);
       } else {
         if (free_handle_on_fail) {
-          delete[] reinterpret_cast<char*>(e);
+          free(e);
           *handle = nullptr;
         }
         s = Status::MemoryLimit("Insert failed due to LRU cache being full.");
@@ -406,21 +422,12 @@ Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
         if (!e->HasRefs()) {
           e->Ref();
         }
-        *handle = reinterpret_cast<Cache::Handle*>(e);
+        *handle = e;
       }
     }
   }
 
-  // Try to insert the evicted entries into the secondary cache.
-  // Free the entries here outside of mutex for performance reasons.
-  for (auto entry : last_reference_list) {
-    if (secondary_cache_ && entry->IsSecondaryCacheCompatible() &&
-        !entry->IsInSecondaryCache()) {
-      secondary_cache_->Insert(entry->key(), entry->value, entry->info_.helper)
-          .PermitUncheckedError();
-    }
-    entry->Free();
-  }
+  TryInsertIntoSecondaryCache(last_reference_list);
 
   return s;
 }
@@ -429,75 +436,135 @@ void LRUCacheShard::Promote(LRUHandle* e) {
   SecondaryCacheResultHandle* secondary_handle = e->sec_handle;
 
   assert(secondary_handle->IsReady());
-  e->SetIncomplete(false);
-  e->SetInCache(true);
+  // e is not thread-shared here; OK to modify "immutable" fields as well as
+  // "mutable" (normally requiring mutex)
+  e->SetIsPending(false);
   e->value = secondary_handle->Value();
-  e->CalcTotalCharge(secondary_handle->Size(), metadata_charge_policy_);
+  assert(e->total_charge == 0);
+  size_t value_size = secondary_handle->Size();
   delete secondary_handle;
 
-  // This call could fail if the cache is over capacity and
-  // strict_capacity_limit_ is true. In such a case, we don't want
-  // InsertItem() to free the handle, since the item is already in memory
-  // and the caller will most likely just read from disk if we erase it here.
   if (e->value) {
-    Cache::Handle* handle = reinterpret_cast<Cache::Handle*>(e);
-    Status s = InsertItem(e, &handle, /*free_handle_on_fail=*/false);
+    e->CalcTotalCharge(value_size, metadata_charge_policy_);
+    Status s;
+    if (e->IsStandalone()) {
+      assert(secondary_cache_ && secondary_cache_->SupportForceErase());
+
+      // Insert a dummy handle and return a standalone handle to caller.
+      // Charge the standalone handle.
+      autovector<LRUHandle*> last_reference_list;
+      bool free_standalone_handle{false};
+      {
+        DMutexLock l(mutex_);
+
+        // Free the space following strict LRU policy until enough space
+        // is freed or the lru list is empty.
+        EvictFromLRU(e->total_charge, &last_reference_list);
+
+        if ((usage_ + e->total_charge) > capacity_ && strict_capacity_limit_) {
+          free_standalone_handle = true;
+        } else {
+          usage_ += e->total_charge;
+        }
+      }
+
+      TryInsertIntoSecondaryCache(last_reference_list);
+      if (free_standalone_handle) {
+        e->Unref();
+        e->Free(table_.GetAllocator());
+        e = nullptr;
+      } else {
+        PERF_COUNTER_ADD(block_cache_standalone_handle_count, 1);
+      }
+
+      // Insert a dummy handle into the primary cache. This dummy handle is
+      // not IsSecondaryCacheCompatible().
+      // FIXME? This should not overwrite an existing non-dummy entry in the
+      // rare case that one exists
+      Cache::Priority priority =
+          e->IsHighPri() ? Cache::Priority::HIGH : Cache::Priority::LOW;
+      s = Insert(e->key(), e->hash, &kDummyValue, &kNoopCacheItemHelper,
+                 /*charge=*/0,
+                 /*handle=*/nullptr, priority);
+    } else {
+      e->SetInCache(true);
+      LRUHandle* handle = e;
+      // This InsertItem() could fail if the cache is over capacity and
+      // strict_capacity_limit_ is true. In such a case, we don't want
+      // InsertItem() to free the handle, since the item is already in memory
+      // and the caller will most likely just read it from disk if we erase it
+      // here.
+      s = InsertItem(e, &handle, /*free_handle_on_fail=*/false);
+      if (s.ok()) {
+        PERF_COUNTER_ADD(block_cache_real_handle_count, 1);
+      }
+    }
+
     if (!s.ok()) {
       // Item is in memory, but not accounted against the cache capacity.
       // When the handle is released, the item should get deleted.
       assert(!e->InCache());
     }
   } else {
-    // Since the secondary cache lookup failed, mark the item as not in cache
-    // Don't charge the cache as its only metadata that'll shortly be released
-    DMutexLock l(mutex_);
-    // TODO
-    e->CalcTotalCharge(0, metadata_charge_policy_);
-    e->SetInCache(false);
+    // Secondary cache lookup failed. The caller will take care of detecting
+    // this and eventually releasing e.
+    assert(!e->value);
+    assert(!e->InCache());
   }
 }
 
-Cache::Handle* LRUCacheShard::Lookup(
-    const Slice& key, uint32_t hash,
-    const ShardedCache::CacheItemHelper* helper,
-    const ShardedCache::CreateCallback& create_cb, Cache::Priority priority,
-    bool wait, Statistics* stats) {
+LRUHandle* LRUCacheShard::Lookup(const Slice& key, uint32_t hash,
+                                 const Cache::CacheItemHelper* helper,
+                                 Cache::CreateContext* create_context,
+                                 Cache::Priority priority, bool wait,
+                                 Statistics* stats) {
   LRUHandle* e = nullptr;
+  bool found_dummy_entry{false};
   {
     DMutexLock l(mutex_);
     e = table_.Lookup(key, hash);
     if (e != nullptr) {
       assert(e->InCache());
-      if (!e->HasRefs()) {
-        // The entry is in LRU since it's in hash and has no external references
-        LRU_Remove(e);
+      if (e->value == &kDummyValue) {
+        // For a dummy handle, if it was retrieved from secondary cache,
+        // it may still exist in secondary cache.
+        // If the handle exists in secondary cache, the value should be
+        // erased from sec cache and be inserted into primary cache.
+        found_dummy_entry = true;
+        // Let the dummy entry be overwritten
+        e = nullptr;
+      } else {
+        if (!e->HasRefs()) {
+          // The entry is in LRU since it's in hash and has no external
+          // references.
+          LRU_Remove(e);
+        }
+        e->Ref();
+        e->SetHit();
       }
-      e->Ref();
-      e->SetHit();
     }
   }
 
-  // If handle table lookup failed, then allocate a handle outside the
-  // mutex if we're going to lookup in the secondary cache.
-  // Only support synchronous for now.
-  // TODO: Support asynchronous lookup in secondary cache
-  if (!e && secondary_cache_ && helper && helper->saveto_cb) {
-    // For objects from the secondary cache, we expect the caller to provide
-    // a way to create/delete the primary cache object. The only case where
-    // a deleter would not be required is for dummy entries inserted for
-    // accounting purposes, which we won't demote to the secondary cache
-    // anyway.
-    assert(create_cb && helper->del_cb);
+  // If handle table lookup failed or the handle is a dummy one, allocate
+  // a handle outside the mutex if we re going to lookup in the secondary cache.
+  //
+  // When a block is firstly Lookup from CompressedSecondaryCache, we just
+  // insert a dummy block into the primary cache (charging the actual size of
+  // the block) and don't erase the block from CompressedSecondaryCache. A
+  // standalone handle is returned to the caller. Only if the block is hit
+  // again, we erase it from CompressedSecondaryCache and add it into the
+  // primary cache.
+  if (!e && secondary_cache_ && helper && helper->create_cb) {
     bool is_in_sec_cache{false};
     std::unique_ptr<SecondaryCacheResultHandle> secondary_handle =
-        secondary_cache_->Lookup(key, create_cb, wait, is_in_sec_cache);
+        secondary_cache_->Lookup(key, helper, create_context, wait,
+                                 found_dummy_entry, is_in_sec_cache);
     if (secondary_handle != nullptr) {
-      e = reinterpret_cast<LRUHandle*>(
-          new char[sizeof(LRUHandle) - 1 + key.size()]);
+      e = static_cast<LRUHandle*>(malloc(sizeof(LRUHandle) - 1 + key.size()));
 
-      e->flags = 0;
-      e->SetSecondaryCacheCompatible(true);
-      e->info_.helper = helper;
+      e->m_flags = 0;
+      e->im_flags = 0;
+      e->helper = helper;
       e->key_length = key.size();
       e->hash = hash;
       e->refs = 0;
@@ -509,37 +576,46 @@ Cache::Handle* LRUCacheShard::Lookup(
       e->total_charge = 0;
       e->Ref();
       e->SetIsInSecondaryCache(is_in_sec_cache);
+      e->SetIsStandalone(secondary_cache_->SupportForceErase() &&
+                         !found_dummy_entry);
 
       if (wait) {
         Promote(e);
-        if (!e->value) {
-          // The secondary cache returned a handle, but the lookup failed.
-          e->Unref();
-          e->Free();
-          e = nullptr;
-        } else {
-          PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
-          RecordTick(stats, SECONDARY_CACHE_HITS);
+        if (e) {
+          if (!e->value) {
+            // The secondary cache returned a handle, but the lookup failed.
+            e->Unref();
+            e->Free(table_.GetAllocator());
+            e = nullptr;
+          } else {
+            PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
+            RecordTick(stats, SECONDARY_CACHE_HITS);
+          }
         }
       } else {
         // If wait is false, we always return a handle and let the caller
         // release the handle after checking for success or failure.
-        e->SetIncomplete(true);
+        e->SetIsPending(true);
         // This may be slightly inaccurate, if the lookup eventually fails.
         // But the probability is very low.
         PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
         RecordTick(stats, SECONDARY_CACHE_HITS);
       }
+    } else {
+      // Caller will most likely overwrite the dummy entry with an Insert
+      // after this Lookup fails
+      assert(e == nullptr);
     }
   }
-  return reinterpret_cast<Cache::Handle*>(e);
+  return e;
 }
 
-bool LRUCacheShard::Ref(Cache::Handle* h) {
-  LRUHandle* e = reinterpret_cast<LRUHandle*>(h);
+bool LRUCacheShard::Ref(LRUHandle* e) {
   DMutexLock l(mutex_);
   // To create another reference - entry must be already externally referenced.
   assert(e->HasRefs());
+  // Pending handles are not for sharing
+  assert(!e->IsPending());
   e->Ref();
   return true;
 }
@@ -558,12 +634,15 @@ void LRUCacheShard::SetLowPriorityPoolRatio(double low_pri_pool_ratio) {
   MaintainPoolSize();
 }
 
-bool LRUCacheShard::Release(Cache::Handle* handle, bool erase_if_last_ref) {
-  if (handle == nullptr) {
+bool LRUCacheShard::Release(LRUHandle* e, bool /*useful*/,
+                            bool erase_if_last_ref) {
+  if (e == nullptr) {
     return false;
   }
-  LRUHandle* e = reinterpret_cast<LRUHandle*>(handle);
   bool last_reference = false;
+  // Must Wait or WaitAll first on pending handles. Otherwise, would leak
+  // a secondary cache handle.
+  assert(!e->IsPending());
   {
     DMutexLock l(mutex_);
     last_reference = e->Unref();
@@ -581,12 +660,8 @@ bool LRUCacheShard::Release(Cache::Handle* handle, bool erase_if_last_ref) {
         last_reference = false;
       }
     }
-    // If it was the last reference, and the entry is either not secondary
-    // cache compatible (i.e a dummy entry for accounting), or is secondary
-    // cache compatible and has a non-null value, then decrement the cache
-    // usage. If value is null in the latter case, that means the lookup
-    // failed and we didn't charge the cache.
-    if (last_reference && (!e->IsSecondaryCacheCompatible() || e->value)) {
+    // If it was the last reference, then decrement the cache usage.
+    if (last_reference) {
       assert(usage_ >= e->total_charge);
       usage_ -= e->total_charge;
     }
@@ -594,33 +669,28 @@ bool LRUCacheShard::Release(Cache::Handle* handle, bool erase_if_last_ref) {
 
   // Free the entry here outside of mutex for performance reasons.
   if (last_reference) {
-    e->Free();
+    e->Free(table_.GetAllocator());
   }
   return last_reference;
 }
 
-Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
-                             size_t charge,
-                             void (*deleter)(const Slice& key, void* value),
+Status LRUCacheShard::Insert(const Slice& key, uint32_t hash,
+                             Cache::ObjectPtr value,
                              const Cache::CacheItemHelper* helper,
-                             Cache::Handle** handle, Cache::Priority priority) {
+                             size_t charge, LRUHandle** handle,
+                             Cache::Priority priority) {
+  assert(helper);
+
   // Allocate the memory here outside of the mutex.
   // If the cache is full, we'll have to release it.
   // It shouldn't happen very often though.
-  LRUHandle* e = reinterpret_cast<LRUHandle*>(
-      new char[sizeof(LRUHandle) - 1 + key.size()]);
+  LRUHandle* e =
+      static_cast<LRUHandle*>(malloc(sizeof(LRUHandle) - 1 + key.size()));
 
   e->value = value;
-  e->flags = 0;
-  if (helper) {
-    e->SetSecondaryCacheCompatible(true);
-    e->info_.helper = helper;
-  } else {
-#ifdef __SANITIZE_THREAD__
-    e->is_secondary_cache_compatible_for_tsan = false;
-#endif  // __SANITIZE_THREAD__
-    e->info_.deleter = deleter;
-  }
+  e->m_flags = 0;
+  e->im_flags = 0;
+  e->helper = helper;
   e->key_length = key.size();
   e->hash = hash;
   e->refs = 0;
@@ -630,6 +700,10 @@ Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
   memcpy(e->key_data, key.data(), key.size());
   e->CalcTotalCharge(charge, metadata_charge_policy_);
 
+  // value == nullptr is reserved for indicating failure for when secondary
+  // cache compatible
+  assert(!(e->IsSecondaryCacheCompatible() && value == nullptr));
+
   return InsertItem(e, handle, /* free_handle_on_fail */ true);
 }
 
@@ -655,13 +729,11 @@ void LRUCacheShard::Erase(const Slice& key, uint32_t hash) {
   // Free the entry here outside of mutex for performance reasons.
   // last_reference will only be true if e != nullptr.
   if (last_reference) {
-    e->Free();
+    e->Free(table_.GetAllocator());
   }
 }
 
-bool LRUCacheShard::IsReady(Cache::Handle* handle) {
-  LRUHandle* e = reinterpret_cast<LRUHandle*>(handle);
-  DMutexLock l(mutex_);
+bool LRUCacheShard::IsReady(LRUHandle* e) {
   bool ready = true;
   if (e->IsPending()) {
     assert(secondary_cache_);
@@ -682,7 +754,17 @@ size_t LRUCacheShard::GetPinnedUsage() const {
   return usage_ - lru_usage_;
 }
 
-std::string LRUCacheShard::GetPrintableOptions() const {
+size_t LRUCacheShard::GetOccupancyCount() const {
+  DMutexLock l(mutex_);
+  return table_.GetOccupancyCount();
+}
+
+size_t LRUCacheShard::GetTableAddressCount() const {
+  DMutexLock l(mutex_);
+  return size_t{1} << table_.GetLengthBits();
+}
+
+void LRUCacheShard::AppendPrintableOptions(std::string& str) const {
   const int kBufferSize = 200;
   char buffer[kBufferSize];
   {
@@ -692,7 +774,7 @@ std::string LRUCacheShard::GetPrintableOptions() const {
     snprintf(buffer + strlen(buffer), kBufferSize - strlen(buffer),
              "    low_pri_pool_ratio: %.3lf\n", low_pri_pool_ratio_);
   }
-  return std::string(buffer);
+  str.append(buffer);
 }
 
 LRUCache::LRUCache(size_t capacity, int num_shard_bits,
@@ -701,88 +783,45 @@ LRUCache::LRUCache(size_t capacity, int num_shard_bits,
                    std::shared_ptr<MemoryAllocator> allocator,
                    bool use_adaptive_mutex,
                    CacheMetadataChargePolicy metadata_charge_policy,
-                   const std::shared_ptr<SecondaryCache>& secondary_cache)
+                   std::shared_ptr<SecondaryCache> _secondary_cache)
     : ShardedCache(capacity, num_shard_bits, strict_capacity_limit,
-                   std::move(allocator)) {
-  num_shards_ = 1 << num_shard_bits;
-  shards_ = reinterpret_cast<LRUCacheShard*>(
-      port::cacheline_aligned_alloc(sizeof(LRUCacheShard) * num_shards_));
-  size_t per_shard = (capacity + (num_shards_ - 1)) / num_shards_;
-  for (int i = 0; i < num_shards_; i++) {
-    new (&shards_[i]) LRUCacheShard(
+                   std::move(allocator)),
+      secondary_cache_(std::move(_secondary_cache)) {
+  size_t per_shard = GetPerShardCapacity();
+  SecondaryCache* secondary_cache = secondary_cache_.get();
+  MemoryAllocator* alloc = memory_allocator();
+  InitShards([=](LRUCacheShard* cs) {
+    new (cs) LRUCacheShard(
         per_shard, strict_capacity_limit, high_pri_pool_ratio,
         low_pri_pool_ratio, use_adaptive_mutex, metadata_charge_policy,
-        /* max_upper_hash_bits */ 32 - num_shard_bits, secondary_cache);
-  }
-  secondary_cache_ = secondary_cache;
+        /* max_upper_hash_bits */ 32 - num_shard_bits, alloc, secondary_cache);
+  });
 }
 
-LRUCache::~LRUCache() {
-  if (shards_ != nullptr) {
-    assert(num_shards_ > 0);
-    for (int i = 0; i < num_shards_; i++) {
-      shards_[i].~LRUCacheShard();
-    }
-    port::cacheline_aligned_free(shards_);
-  }
-}
-
-CacheShard* LRUCache::GetShard(uint32_t shard) {
-  return reinterpret_cast<CacheShard*>(&shards_[shard]);
-}
-
-const CacheShard* LRUCache::GetShard(uint32_t shard) const {
-  return reinterpret_cast<CacheShard*>(&shards_[shard]);
-}
-
-void* LRUCache::Value(Handle* handle) {
-  return reinterpret_cast<const LRUHandle*>(handle)->value;
+Cache::ObjectPtr LRUCache::Value(Handle* handle) {
+  auto h = reinterpret_cast<const LRUHandle*>(handle);
+  assert(!h->IsPending() || h->value == nullptr);
+  assert(h->value != &kDummyValue);
+  return h->value;
 }
 
 size_t LRUCache::GetCharge(Handle* handle) const {
-  CacheMetadataChargePolicy metadata_charge_policy = kDontChargeCacheMetadata;
-  if (num_shards_ > 0) {
-    metadata_charge_policy = shards_[0].metadata_charge_policy_;
-  }
   return reinterpret_cast<const LRUHandle*>(handle)->GetCharge(
-      metadata_charge_policy);
+      GetShard(0).metadata_charge_policy_);
 }
 
-Cache::DeleterFn LRUCache::GetDeleter(Handle* handle) const {
+const Cache::CacheItemHelper* LRUCache::GetCacheItemHelper(
+    Handle* handle) const {
   auto h = reinterpret_cast<const LRUHandle*>(handle);
-  if (h->IsSecondaryCacheCompatible()) {
-    return h->info_.helper->del_cb;
-  } else {
-    return h->info_.deleter;
-  }
-}
-
-uint32_t LRUCache::GetHash(Handle* handle) const {
-  return reinterpret_cast<const LRUHandle*>(handle)->hash;
-}
-
-void LRUCache::DisownData() {
-  // Leak data only if that won't generate an ASAN/valgrind warning.
-  if (!kMustFreeHeapAllocations) {
-    shards_ = nullptr;
-    num_shards_ = 0;
-  }
+  return h->helper;
 }
 
 size_t LRUCache::TEST_GetLRUSize() {
-  size_t lru_size_of_all_shards = 0;
-  for (int i = 0; i < num_shards_; i++) {
-    lru_size_of_all_shards += shards_[i].TEST_GetLRUSize();
-  }
-  return lru_size_of_all_shards;
+  return SumOverShards([](LRUCacheShard& cs) { return cs.TEST_GetLRUSize(); });
 }
 
 double LRUCache::GetHighPriPoolRatio() {
-  double result = 0.0;
-  if (num_shards_ > 0) {
-    result = shards_[0].GetHighPriPoolRatio();
-  }
-  return result;
+  return GetShard(0).GetHighPriPoolRatio();
 }
 
 void LRUCache::WaitAll(std::vector<Handle*>& handles) {
@@ -808,22 +847,17 @@ void LRUCache::WaitAll(std::vector<Handle*>& handles) {
       if (!lru_handle->IsPending()) {
         continue;
       }
-      uint32_t hash = GetHash(handle);
-      LRUCacheShard* shard = static_cast<LRUCacheShard*>(GetShard(Shard(hash)));
-      shard->Promote(lru_handle);
+      GetShard(lru_handle->hash).Promote(lru_handle);
     }
   }
 }
 
-std::string LRUCache::GetPrintableOptions() const {
-  std::string ret;
-  ret.reserve(20000);
-  ret.append(ShardedCache::GetPrintableOptions());
+void LRUCache::AppendPrintableOptions(std::string& str) const {
+  ShardedCache::AppendPrintableOptions(str);  // options from shard
   if (secondary_cache_) {
-    ret.append("  secondary_cache:\n");
-    ret.append(secondary_cache_->GetPrintableOptions());
+    str.append("  secondary_cache:\n");
+    str.append(secondary_cache_->GetPrintableOptions());
   }
-  return ret;
 }
 
 }  // namespace lru_cache
@@ -843,7 +877,7 @@ std::shared_ptr<Cache> NewLRUCache(
     return nullptr;
   }
   if (low_pri_pool_ratio < 0.0 || low_pri_pool_ratio > 1.0) {
-    // Invalid high_pri_pool_ratio
+    // Invalid low_pri_pool_ratio
     return nullptr;
   }
   if (low_pri_pool_ratio + high_pri_pool_ratio > 1.0) {

cache/lru_cache.h

@@ -13,6 +13,7 @@
 
 #include "cache/sharded_cache.h"
 #include "port/lang.h"
+#include "port/likely.h"
 #include "port/malloc.h"
 #include "port/port.h"
 #include "rocksdb/secondary_cache.h"
@@ -38,27 +39,18 @@ namespace lru_cache {
 //    (refs == 0 && in_cache == true)
 // 3. Referenced externally AND not in hash table.
 //    In that case the entry is not in the LRU list and not in hash table.
-//    The entry can be freed when refs becomes 0.
+//    The entry must be freed if refs becomes 0 in this state.
 //    (refs >= 1 && in_cache == false)
-//
-// All newly created LRUHandles are in state 1. If you call
-// LRUCacheShard::Release on entry in state 1, it will go into state 2.
-// To move from state 1 to state 3, either call LRUCacheShard::Erase or
-// LRUCacheShard::Insert with the same key (but possibly different value).
-// To move from state 2 to state 1, use LRUCacheShard::Lookup.
-// Before destruction, make sure that no handles are in state 1. This means
-// that any successful LRUCacheShard::Lookup/LRUCacheShard::Insert have a
-// matching LRUCache::Release (to move into state 2) or LRUCacheShard::Erase
-// (to move into state 3).
+// If you call LRUCacheShard::Release enough times on an entry in state 1, it
+// will go into state 2. To move from state 1 to state 3, either call
+// LRUCacheShard::Erase or LRUCacheShard::Insert with the same key (but
+// possibly different value). To move from state 2 to state 1, use
+// LRUCacheShard::Lookup.
+// While refs > 0, public properties like value and deleter must not change.
 
 struct LRUHandle {
-  void* value;
-  union Info {
-    Info() {}
-    ~Info() {}
-    Cache::DeleterFn deleter;
-    const ShardedCache::CacheItemHelper* helper;
-  } info_;
+  Cache::ObjectPtr value;
+  const Cache::CacheItemHelper* helper;
   // An entry is not added to the LRUHandleTable until the secondary cache
   // lookup is complete, so its safe to have this union.
   union {
@@ -74,42 +66,45 @@ struct LRUHandle {
   // The number of external refs to this entry. The cache itself is not counted.
   uint32_t refs;
 
-  enum Flags : uint16_t {
+  // Mutable flags - access controlled by mutex
+  // The m_ and M_ prefixes (and im_ and IM_ later) are to hopefully avoid
+  // checking an M_ flag on im_flags or an IM_ flag on m_flags.
+  uint8_t m_flags;
+  enum MFlags : uint8_t {
     // Whether this entry is referenced by the hash table.
-    IN_CACHE = (1 << 0),
-    // Whether this entry is high priority entry.
-    IS_HIGH_PRI = (1 << 1),
-    // Whether this entry is in high-pri pool.
-    IN_HIGH_PRI_POOL = (1 << 2),
+    M_IN_CACHE = (1 << 0),
     // Whether this entry has had any lookups (hits).
-    HAS_HIT = (1 << 3),
-    // Can this be inserted into the secondary cache.
-    IS_SECONDARY_CACHE_COMPATIBLE = (1 << 4),
-    // Is the handle still being read from a lower tier.
-    IS_PENDING = (1 << 5),
-    // Whether this handle is still in a lower tier
-    IS_IN_SECONDARY_CACHE = (1 << 6),
-    // Whether this entry is low priority entry.
-    IS_LOW_PRI = (1 << 7),
+    M_HAS_HIT = (1 << 1),
+    // Whether this entry is in high-pri pool.
+    M_IN_HIGH_PRI_POOL = (1 << 2),
     // Whether this entry is in low-pri pool.
-    IN_LOW_PRI_POOL = (1 << 8),
+    M_IN_LOW_PRI_POOL = (1 << 3),
   };
 
-  uint16_t flags;
-
-#ifdef __SANITIZE_THREAD__
-  // TSAN can report a false data race on flags, where one thread is writing
-  // to one of the mutable bits and another thread is reading this immutable
-  // bit. So precisely suppress that TSAN warning, we separate out this bit
-  // during TSAN runs.
-  bool is_secondary_cache_compatible_for_tsan;
-#endif  // __SANITIZE_THREAD__
+  // "Immutable" flags - only set in single-threaded context and then
+  // can be accessed without mutex
+  uint8_t im_flags;
+  enum ImFlags : uint8_t {
+    // Whether this entry is high priority entry.
+    IM_IS_HIGH_PRI = (1 << 0),
+    // Whether this entry is low priority entry.
+    IM_IS_LOW_PRI = (1 << 1),
+    // Is the handle still being read from a lower tier.
+    IM_IS_PENDING = (1 << 2),
+    // Whether this handle is still in a lower tier
+    IM_IS_IN_SECONDARY_CACHE = (1 << 3),
+    // Marks result handles that should not be inserted into cache
+    IM_IS_STANDALONE = (1 << 4),
+  };
 
   // Beginning of the key (MUST BE THE LAST FIELD IN THIS STRUCT!)
   char key_data[1];
 
   Slice key() const { return Slice(key_data, key_length); }
 
+  // For HandleImpl concept
+  uint32_t GetHash() const { return hash; }
+
   // Increase the reference count by 1.
   void Ref() { refs++; }
 
@@ -123,110 +118,98 @@ struct LRUHandle {
   // Return true if there are external refs, false otherwise.
   bool HasRefs() const { return refs > 0; }
 
-  bool InCache() const { return flags & IN_CACHE; }
-  bool IsHighPri() const { return flags & IS_HIGH_PRI; }
-  bool InHighPriPool() const { return flags & IN_HIGH_PRI_POOL; }
-  bool IsLowPri() const { return flags & IS_LOW_PRI; }
-  bool InLowPriPool() const { return flags & IN_LOW_PRI_POOL; }
-  bool HasHit() const { return flags & HAS_HIT; }
-  bool IsSecondaryCacheCompatible() const {
-#ifdef __SANITIZE_THREAD__
-    return is_secondary_cache_compatible_for_tsan;
-#else
-    return flags & IS_SECONDARY_CACHE_COMPATIBLE;
-#endif  // __SANITIZE_THREAD__
+  bool InCache() const { return m_flags & M_IN_CACHE; }
+  bool IsHighPri() const { return im_flags & IM_IS_HIGH_PRI; }
+  bool InHighPriPool() const { return m_flags & M_IN_HIGH_PRI_POOL; }
+  bool IsLowPri() const { return im_flags & IM_IS_LOW_PRI; }
+  bool InLowPriPool() const { return m_flags & M_IN_LOW_PRI_POOL; }
+  bool HasHit() const { return m_flags & M_HAS_HIT; }
+  bool IsSecondaryCacheCompatible() const { return helper->size_cb != nullptr; }
+  bool IsPending() const { return im_flags & IM_IS_PENDING; }
+  bool IsInSecondaryCache() const {
+    return im_flags & IM_IS_IN_SECONDARY_CACHE;
   }
-  bool IsPending() const { return flags & IS_PENDING; }
-  bool IsInSecondaryCache() const { return flags & IS_IN_SECONDARY_CACHE; }
+  bool IsStandalone() const { return im_flags & IM_IS_STANDALONE; }
 
   void SetInCache(bool in_cache) {
     if (in_cache) {
-      flags |= IN_CACHE;
+      m_flags |= M_IN_CACHE;
     } else {
-      flags &= ~IN_CACHE;
+      m_flags &= ~M_IN_CACHE;
     }
   }
 
   void SetPriority(Cache::Priority priority) {
     if (priority == Cache::Priority::HIGH) {
-      flags |= IS_HIGH_PRI;
-      flags &= ~IS_LOW_PRI;
+      im_flags |= IM_IS_HIGH_PRI;
+      im_flags &= ~IM_IS_LOW_PRI;
     } else if (priority == Cache::Priority::LOW) {
-      flags &= ~IS_HIGH_PRI;
-      flags |= IS_LOW_PRI;
+      im_flags &= ~IM_IS_HIGH_PRI;
+      im_flags |= IM_IS_LOW_PRI;
     } else {
-      flags &= ~IS_HIGH_PRI;
-      flags &= ~IS_LOW_PRI;
+      im_flags &= ~IM_IS_HIGH_PRI;
+      im_flags &= ~IM_IS_LOW_PRI;
     }
   }
 
   void SetInHighPriPool(bool in_high_pri_pool) {
     if (in_high_pri_pool) {
-      flags |= IN_HIGH_PRI_POOL;
+      m_flags |= M_IN_HIGH_PRI_POOL;
     } else {
-      flags &= ~IN_HIGH_PRI_POOL;
+      m_flags &= ~M_IN_HIGH_PRI_POOL;
     }
   }
 
   void SetInLowPriPool(bool in_low_pri_pool) {
     if (in_low_pri_pool) {
-      flags |= IN_LOW_PRI_POOL;
+      m_flags |= M_IN_LOW_PRI_POOL;
     } else {
-      flags &= ~IN_LOW_PRI_POOL;
+      m_flags &= ~M_IN_LOW_PRI_POOL;
     }
   }
 
-  void SetHit() { flags |= HAS_HIT; }
+  void SetHit() { m_flags |= M_HAS_HIT; }
 
-  void SetSecondaryCacheCompatible(bool compat) {
-    if (compat) {
-      flags |= IS_SECONDARY_CACHE_COMPATIBLE;
+  void SetIsPending(bool pending) {
+    if (pending) {
+      im_flags |= IM_IS_PENDING;
     } else {
-      flags &= ~IS_SECONDARY_CACHE_COMPATIBLE;
-    }
-#ifdef __SANITIZE_THREAD__
-    is_secondary_cache_compatible_for_tsan = compat;
-#endif  // __SANITIZE_THREAD__
-  }
-
-  void SetIncomplete(bool incomp) {
-    if (incomp) {
-      flags |= IS_PENDING;
-    } else {
-      flags &= ~IS_PENDING;
+      im_flags &= ~IM_IS_PENDING;
     }
   }
 
   void SetIsInSecondaryCache(bool is_in_secondary_cache) {
     if (is_in_secondary_cache) {
-      flags |= IS_IN_SECONDARY_CACHE;
+      im_flags |= IM_IS_IN_SECONDARY_CACHE;
     } else {
-      flags &= ~IS_IN_SECONDARY_CACHE;
+      im_flags &= ~IM_IS_IN_SECONDARY_CACHE;
     }
   }
 
-  void Free() {
-    assert(refs == 0);
-#ifdef __SANITIZE_THREAD__
-    // Here we can safely assert they are the same without a data race reported
-    assert(((flags & IS_SECONDARY_CACHE_COMPATIBLE) != 0) ==
-           is_secondary_cache_compatible_for_tsan);
-#endif  // __SANITIZE_THREAD__
-    if (!IsSecondaryCacheCompatible() && info_.deleter) {
-      (*info_.deleter)(key(), value);
-    } else if (IsSecondaryCacheCompatible()) {
-      if (IsPending()) {
-        assert(sec_handle != nullptr);
-        SecondaryCacheResultHandle* tmp_sec_handle = sec_handle;
-        tmp_sec_handle->Wait();
-        value = tmp_sec_handle->Value();
-        delete tmp_sec_handle;
-      }
-      if (value) {
-        (*info_.helper->del_cb)(key(), value);
-      }
+  void SetIsStandalone(bool is_standalone) {
+    if (is_standalone) {
+      im_flags |= IM_IS_STANDALONE;
+    } else {
+      im_flags &= ~IM_IS_STANDALONE;
     }
-    delete[] reinterpret_cast<char*>(this);
+  }
+
+  void Free(MemoryAllocator* allocator) {
+    assert(refs == 0);
+
+    if (UNLIKELY(IsPending())) {
+      assert(sec_handle != nullptr);
+      SecondaryCacheResultHandle* tmp_sec_handle = sec_handle;
+      tmp_sec_handle->Wait();
+      value = tmp_sec_handle->Value();
+      delete tmp_sec_handle;
+    }
+    assert(helper);
+    if (helper->del_cb) {
+      helper->del_cb(value, allocator);
+    }
+
+    free(this);
   }
 
   inline size_t CalcuMetaCharge(
@@ -265,10 +248,7 @@ struct LRUHandle {
 // 4.4.3's builtin hashtable.
 class LRUHandleTable {
  public:
-  // If the table uses more hash bits than `max_upper_hash_bits`,
-  // it will eat into the bits used for sharding, which are constant
-  // for a given LRUHandleTable.
-  explicit LRUHandleTable(int max_upper_hash_bits);
+  explicit LRUHandleTable(int max_upper_hash_bits, MemoryAllocator* allocator);
   ~LRUHandleTable();
 
   LRUHandle* Lookup(const Slice& key, uint32_t hash);
@@ -276,8 +256,8 @@ class LRUHandleTable {
   LRUHandle* Remove(const Slice& key, uint32_t hash);
 
   template <typename T>
-  void ApplyToEntriesRange(T func, uint32_t index_begin, uint32_t index_end) {
-    for (uint32_t i = index_begin; i < index_end; i++) {
+  void ApplyToEntriesRange(T func, size_t index_begin, size_t index_end) {
+    for (size_t i = index_begin; i < index_end; i++) {
       LRUHandle* h = list_[i];
       while (h != nullptr) {
         auto n = h->next_hash;
@@ -290,6 +270,10 @@ class LRUHandleTable {
 
   int GetLengthBits() const { return length_bits_; }
 
+  size_t GetOccupancyCount() const { return elems_; }
+
+  MemoryAllocator* GetAllocator() const { return allocator_; }
+
  private:
   // Return a pointer to slot that points to a cache entry that
   // matches key/hash.  If there is no such cache entry, return a
@@ -311,26 +295,38 @@ class LRUHandleTable {
 
   // Set from max_upper_hash_bits (see constructor).
   const int max_length_bits_;
+
+  // From Cache, needed for delete
+  MemoryAllocator* const allocator_;
 };
 
 // A single shard of sharded cache.
-class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
+class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShardBase {
  public:
   LRUCacheShard(size_t capacity, bool strict_capacity_limit,
                 double high_pri_pool_ratio, double low_pri_pool_ratio,
                 bool use_adaptive_mutex,
                 CacheMetadataChargePolicy metadata_charge_policy,
-                int max_upper_hash_bits,
-                const std::shared_ptr<SecondaryCache>& secondary_cache);
-  virtual ~LRUCacheShard() override = default;
+                int max_upper_hash_bits, MemoryAllocator* allocator,
+                SecondaryCache* secondary_cache);
+
+ public:  // Type definitions expected as parameter to ShardedCache
+  using HandleImpl = LRUHandle;
+  using HashVal = uint32_t;
+  using HashCref = uint32_t;
+
+ public:  // Function definitions expected as parameter to ShardedCache
+  static inline HashVal ComputeHash(const Slice& key) {
+    return Lower32of64(GetSliceNPHash64(key));
+  }
 
   // Separate from constructor so caller can easily make an array of LRUCache
   // if current usage is more than new capacity, the function will attempt to
   // free the needed space.
-  virtual void SetCapacity(size_t capacity) override;
+  void SetCapacity(size_t capacity);
 
   // Set the flag to reject insertion if cache if full.
-  virtual void SetStrictCapacityLimit(bool strict_capacity_limit) override;
+  void SetStrictCapacityLimit(bool strict_capacity_limit);
 
   // Set percentage of capacity reserved for high-pri cache entries.
   void SetHighPriorityPoolRatio(double high_pri_pool_ratio);
@@ -339,56 +335,39 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
   void SetLowPriorityPoolRatio(double low_pri_pool_ratio);
 
   // Like Cache methods, but with an extra "hash" parameter.
-  virtual Status Insert(const Slice& key, uint32_t hash, void* value,
-                        size_t charge, Cache::DeleterFn deleter,
-                        Cache::Handle** handle,
-                        Cache::Priority priority) override {
-    return Insert(key, hash, value, charge, deleter, nullptr, handle, priority);
-  }
-  virtual Status Insert(const Slice& key, uint32_t hash, void* value,
-                        const Cache::CacheItemHelper* helper, size_t charge,
-                        Cache::Handle** handle,
-                        Cache::Priority priority) override {
-    assert(helper);
-    return Insert(key, hash, value, charge, nullptr, helper, handle, priority);
-  }
-  // If helper_cb is null, the values of the following arguments don't matter.
-  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash,
-                                const ShardedCache::CacheItemHelper* helper,
-                                const ShardedCache::CreateCallback& create_cb,
-                                ShardedCache::Priority priority, bool wait,
-                                Statistics* stats) override;
-  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash) override {
-    return Lookup(key, hash, nullptr, nullptr, Cache::Priority::LOW, true,
-                  nullptr);
-  }
-  virtual bool Release(Cache::Handle* handle, bool /*useful*/,
-                       bool erase_if_last_ref) override {
-    return Release(handle, erase_if_last_ref);
-  }
-  virtual bool IsReady(Cache::Handle* /*handle*/) override;
-  virtual void Wait(Cache::Handle* /*handle*/) override {}
-  virtual bool Ref(Cache::Handle* handle) override;
-  virtual bool Release(Cache::Handle* handle,
-                       bool erase_if_last_ref = false) override;
-  virtual void Erase(const Slice& key, uint32_t hash) override;
+  Status Insert(const Slice& key, uint32_t hash, Cache::ObjectPtr value,
+                const Cache::CacheItemHelper* helper, size_t charge,
+                LRUHandle** handle, Cache::Priority priority);
+
+  LRUHandle* Lookup(const Slice& key, uint32_t hash,
+                    const Cache::CacheItemHelper* helper,
+                    Cache::CreateContext* create_context,
+                    Cache::Priority priority, bool wait, Statistics* stats);
+
+  bool Release(LRUHandle* handle, bool useful, bool erase_if_last_ref);
+  bool IsReady(LRUHandle* /*handle*/);
+  void Wait(LRUHandle* /*handle*/) {}
+  bool Ref(LRUHandle* handle);
+  void Erase(const Slice& key, uint32_t hash);
 
   // Although in some platforms the update of size_t is atomic, to make sure
   // GetUsage() and GetPinnedUsage() work correctly under any platform, we'll
   // protect them with mutex_.
 
-  virtual size_t GetUsage() const override;
-  virtual size_t GetPinnedUsage() const override;
+  size_t GetUsage() const;
+  size_t GetPinnedUsage() const;
+  size_t GetOccupancyCount() const;
+  size_t GetTableAddressCount() const;
 
-  virtual void ApplyToSomeEntries(
-      const std::function<void(const Slice& key, void* value, size_t charge,
-                               DeleterFn deleter)>& callback,
-      uint32_t average_entries_per_lock, uint32_t* state) override;
+  void ApplyToSomeEntries(
+      const std::function<void(const Slice& key, Cache::ObjectPtr value,
+                               size_t charge,
+                               const Cache::CacheItemHelper* helper)>& callback,
+      size_t average_entries_per_lock, size_t* state);
 
-  virtual void EraseUnRefEntries() override;
-
-  virtual std::string GetPrintableOptions() const override;
+  void EraseUnRefEntries();
 
+ public:  // other function definitions
   void TEST_GetLRUList(LRUHandle** lru, LRUHandle** lru_low_pri,
                        LRUHandle** lru_bottom_pri);
 
@@ -402,17 +381,16 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
   // Retrieves low pri pool ratio
   double GetLowPriPoolRatio();
 
+  void AppendPrintableOptions(std::string& /*str*/) const;
+
  private:
   friend class LRUCache;
   // Insert an item into the hash table and, if handle is null, insert into
   // the LRU list. Older items are evicted as necessary. If the cache is full
   // and free_handle_on_fail is true, the item is deleted and handle is set to
   // nullptr.
-  Status InsertItem(LRUHandle* item, Cache::Handle** handle,
+  Status InsertItem(LRUHandle* item, LRUHandle** handle,
                     bool free_handle_on_fail);
-  Status Insert(const Slice& key, uint32_t hash, void* value, size_t charge,
-                DeleterFn deleter, const Cache::CacheItemHelper* helper,
-                Cache::Handle** handle, Cache::Priority priority);
   // Promote an item looked up from the secondary cache to the LRU cache.
   // The item may be still in the secondary cache.
   // It is only inserted into the hash table and not the LRU list, and only
@@ -435,6 +413,9 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
   // holding the mutex_.
   void EvictFromLRU(size_t charge, autovector<LRUHandle*>* deleted);
 
+  // Try to insert the evicted handles into the secondary cache.
+  void TryInsertIntoSecondaryCache(autovector<LRUHandle*> evicted_handles);
+
   // Initialized before use.
   size_t capacity_;
 
@@ -496,14 +477,15 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
   // don't mind mutex_ invoking the non-const actions.
   mutable DMutex mutex_;
 
-  std::shared_ptr<SecondaryCache> secondary_cache_;
+  // Owned by LRUCache
+  SecondaryCache* secondary_cache_;
 };
 
 class LRUCache
 #ifdef NDEBUG
     final
 #endif
-    : public ShardedCache {
+    : public ShardedCache<LRUCacheShard> {
  public:
   LRUCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit,
            double high_pri_pool_ratio, double low_pri_pool_ratio,
@@ -511,27 +493,21 @@ class LRUCache
            bool use_adaptive_mutex = kDefaultToAdaptiveMutex,
            CacheMetadataChargePolicy metadata_charge_policy =
                kDontChargeCacheMetadata,
-           const std::shared_ptr<SecondaryCache>& secondary_cache = nullptr);
-  virtual ~LRUCache();
-  virtual const char* Name() const override { return "LRUCache"; }
-  virtual CacheShard* GetShard(uint32_t shard) override;
-  virtual const CacheShard* GetShard(uint32_t shard) const override;
-  virtual void* Value(Handle* handle) override;
-  virtual size_t GetCharge(Handle* handle) const override;
-  virtual uint32_t GetHash(Handle* handle) const override;
-  virtual DeleterFn GetDeleter(Handle* handle) const override;
-  virtual void DisownData() override;
-  virtual void WaitAll(std::vector<Handle*>& handles) override;
-  std::string GetPrintableOptions() const override;
+           std::shared_ptr<SecondaryCache> secondary_cache = nullptr);
+  const char* Name() const override { return "LRUCache"; }
+  ObjectPtr Value(Handle* handle) override;
+  size_t GetCharge(Handle* handle) const override;
+  const CacheItemHelper* GetCacheItemHelper(Handle* handle) const override;
+  void WaitAll(std::vector<Handle*>& handles) override;
 
   // Retrieves number of elements in LRU, for unit test purpose only.
   size_t TEST_GetLRUSize();
   // Retrieves high pri pool ratio.
   double GetHighPriPoolRatio();
 
+  void AppendPrintableOptions(std::string& str) const override;
+
  private:
-  LRUCacheShard* shards_ = nullptr;
-  int num_shards_ = 0;
   std::shared_ptr<SecondaryCache> secondary_cache_;
 };
 

cache/secondary_cache.cc

@@ -0,0 +1,41 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "rocksdb/secondary_cache.h"
+
+#include "cache/cache_entry_roles.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace {
+
+void NoopDelete(Cache::ObjectPtr, MemoryAllocator*) {}
+
+size_t SliceSize(Cache::ObjectPtr obj) {
+  return static_cast<Slice*>(obj)->size();
+}
+
+Status SliceSaveTo(Cache::ObjectPtr from_obj, size_t from_offset, size_t length,
+                   char* out) {
+  const Slice& slice = *static_cast<Slice*>(from_obj);
+  std::memcpy(out, slice.data() + from_offset, length);
+  return Status::OK();
+}
+
+Status FailCreate(const Slice&, Cache::CreateContext*, MemoryAllocator*,
+                  Cache::ObjectPtr*, size_t*) {
+  return Status::NotSupported("Only for dumping data into SecondaryCache");
+}
+
+}  // namespace
+
+Status SecondaryCache::InsertSaved(const Slice& key, const Slice& saved) {
+  static Cache::CacheItemHelper helper{CacheEntryRole::kMisc, &NoopDelete,
+                                       &SliceSize, &SliceSaveTo, &FailCreate};
+  // NOTE: depends on Insert() being synchronous, not keeping pointer `&saved`
+  return Insert(key, const_cast<Slice*>(&saved), &helper);
+}
+
+}  // namespace ROCKSDB_NAMESPACE

cache/sharded_cache.cc

@@ -19,184 +19,49 @@
 
 namespace ROCKSDB_NAMESPACE {
 
-namespace {
-
-inline uint32_t HashSlice(const Slice& s) {
-  return Lower32of64(GetSliceNPHash64(s));
-}
-
-}  // namespace
-
-ShardedCache::ShardedCache(size_t capacity, int num_shard_bits,
-                           bool strict_capacity_limit,
-                           std::shared_ptr<MemoryAllocator> allocator)
+ShardedCacheBase::ShardedCacheBase(size_t capacity, int num_shard_bits,
+                                   bool strict_capacity_limit,
+                                   std::shared_ptr<MemoryAllocator> allocator)
     : Cache(std::move(allocator)),
+      last_id_(1),
       shard_mask_((uint32_t{1} << num_shard_bits) - 1),
-      capacity_(capacity),
       strict_capacity_limit_(strict_capacity_limit),
-      last_id_(1) {}
+      capacity_(capacity) {}
 
-void ShardedCache::SetCapacity(size_t capacity) {
+size_t ShardedCacheBase::ComputePerShardCapacity(size_t capacity) const {
   uint32_t num_shards = GetNumShards();
-  const size_t per_shard = (capacity + (num_shards - 1)) / num_shards;
-  MutexLock l(&capacity_mutex_);
-  for (uint32_t s = 0; s < num_shards; s++) {
-    GetShard(s)->SetCapacity(per_shard);
-  }
-  capacity_ = capacity;
+  return (capacity + (num_shards - 1)) / num_shards;
 }
 
-void ShardedCache::SetStrictCapacityLimit(bool strict_capacity_limit) {
-  uint32_t num_shards = GetNumShards();
-  MutexLock l(&capacity_mutex_);
-  for (uint32_t s = 0; s < num_shards; s++) {
-    GetShard(s)->SetStrictCapacityLimit(strict_capacity_limit);
-  }
-  strict_capacity_limit_ = strict_capacity_limit;
+size_t ShardedCacheBase::GetPerShardCapacity() const {
+  return ComputePerShardCapacity(GetCapacity());
 }
 
-Status ShardedCache::Insert(const Slice& key, void* value, size_t charge,
-                            DeleterFn deleter, Handle** handle,
-                            Priority priority) {
-  uint32_t hash = HashSlice(key);
-  return GetShard(Shard(hash))
-      ->Insert(key, hash, value, charge, deleter, handle, priority);
-}
-
-Status ShardedCache::Insert(const Slice& key, void* value,
-                            const CacheItemHelper* helper, size_t charge,
-                            Handle** handle, Priority priority) {
-  uint32_t hash = HashSlice(key);
-  if (!helper) {
-    return Status::InvalidArgument();
-  }
-  return GetShard(Shard(hash))
-      ->Insert(key, hash, value, helper, charge, handle, priority);
-}
-
-Cache::Handle* ShardedCache::Lookup(const Slice& key, Statistics* /*stats*/) {
-  uint32_t hash = HashSlice(key);
-  return GetShard(Shard(hash))->Lookup(key, hash);
-}
-
-Cache::Handle* ShardedCache::Lookup(const Slice& key,
-                                    const CacheItemHelper* helper,
-                                    const CreateCallback& create_cb,
-                                    Priority priority, bool wait,
-                                    Statistics* stats) {
-  uint32_t hash = HashSlice(key);
-  return GetShard(Shard(hash))
-      ->Lookup(key, hash, helper, create_cb, priority, wait, stats);
-}
-
-bool ShardedCache::IsReady(Handle* handle) {
-  uint32_t hash = GetHash(handle);
-  return GetShard(Shard(hash))->IsReady(handle);
-}
-
-void ShardedCache::Wait(Handle* handle) {
-  uint32_t hash = GetHash(handle);
-  GetShard(Shard(hash))->Wait(handle);
-}
-
-bool ShardedCache::Ref(Handle* handle) {
-  uint32_t hash = GetHash(handle);
-  return GetShard(Shard(hash))->Ref(handle);
-}
-
-bool ShardedCache::Release(Handle* handle, bool erase_if_last_ref) {
-  uint32_t hash = GetHash(handle);
-  return GetShard(Shard(hash))->Release(handle, erase_if_last_ref);
-}
-
-bool ShardedCache::Release(Handle* handle, bool useful,
-                           bool erase_if_last_ref) {
-  uint32_t hash = GetHash(handle);
-  return GetShard(Shard(hash))->Release(handle, useful, erase_if_last_ref);
-}
-
-void ShardedCache::Erase(const Slice& key) {
-  uint32_t hash = HashSlice(key);
-  GetShard(Shard(hash))->Erase(key, hash);
-}
-
-uint64_t ShardedCache::NewId() {
+uint64_t ShardedCacheBase::NewId() {
   return last_id_.fetch_add(1, std::memory_order_relaxed);
 }
 
-size_t ShardedCache::GetCapacity() const {
-  MutexLock l(&capacity_mutex_);
+size_t ShardedCacheBase::GetCapacity() const {
+  MutexLock l(&config_mutex_);
   return capacity_;
 }
 
-bool ShardedCache::HasStrictCapacityLimit() const {
-  MutexLock l(&capacity_mutex_);
+bool ShardedCacheBase::HasStrictCapacityLimit() const {
+  MutexLock l(&config_mutex_);
   return strict_capacity_limit_;
 }
 
-size_t ShardedCache::GetUsage() const {
-  // We will not lock the cache when getting the usage from shards.
-  uint32_t num_shards = GetNumShards();
-  size_t usage = 0;
-  for (uint32_t s = 0; s < num_shards; s++) {
-    usage += GetShard(s)->GetUsage();
-  }
-  return usage;
-}
-
-size_t ShardedCache::GetUsage(Handle* handle) const {
+size_t ShardedCacheBase::GetUsage(Handle* handle) const {
   return GetCharge(handle);
 }
 
-size_t ShardedCache::GetPinnedUsage() const {
-  // We will not lock the cache when getting the usage from shards.
-  uint32_t num_shards = GetNumShards();
-  size_t usage = 0;
-  for (uint32_t s = 0; s < num_shards; s++) {
-    usage += GetShard(s)->GetPinnedUsage();
-  }
-  return usage;
-}
-
-void ShardedCache::ApplyToAllEntries(
-    const std::function<void(const Slice& key, void* value, size_t charge,
-                             DeleterFn deleter)>& callback,
-    const ApplyToAllEntriesOptions& opts) {
-  uint32_t num_shards = GetNumShards();
-  // Iterate over part of each shard, rotating between shards, to
-  // minimize impact on latency of concurrent operations.
-  std::unique_ptr<uint32_t[]> states(new uint32_t[num_shards]{});
-
-  uint32_t aepl_in_32 = static_cast<uint32_t>(
-      std::min(size_t{UINT32_MAX}, opts.average_entries_per_lock));
-  aepl_in_32 = std::min(aepl_in_32, uint32_t{1});
-
-  bool remaining_work;
-  do {
-    remaining_work = false;
-    for (uint32_t s = 0; s < num_shards; s++) {
-      if (states[s] != UINT32_MAX) {
-        GetShard(s)->ApplyToSomeEntries(callback, aepl_in_32, &states[s]);
-        remaining_work |= states[s] != UINT32_MAX;
-      }
-    }
-  } while (remaining_work);
-}
-
-void ShardedCache::EraseUnRefEntries() {
-  uint32_t num_shards = GetNumShards();
-  for (uint32_t s = 0; s < num_shards; s++) {
-    GetShard(s)->EraseUnRefEntries();
-  }
-}
-
-std::string ShardedCache::GetPrintableOptions() const {
+std::string ShardedCacheBase::GetPrintableOptions() const {
   std::string ret;
   ret.reserve(20000);
   const int kBufferSize = 200;
   char buffer[kBufferSize];
   {
-    MutexLock l(&capacity_mutex_);
+    MutexLock l(&config_mutex_);
     snprintf(buffer, kBufferSize, "    capacity : %" ROCKSDB_PRIszt "\n",
              capacity_);
     ret.append(buffer);
@@ -210,12 +75,12 @@ std::string ShardedCache::GetPrintableOptions() const {
   snprintf(buffer, kBufferSize, "    memory_allocator : %s\n",
            memory_allocator() ? memory_allocator()->Name() : "None");
   ret.append(buffer);
-  ret.append(GetShard(0)->GetPrintableOptions());
+  AppendPrintableOptions(ret);
   return ret;
 }
-int GetDefaultCacheShardBits(size_t capacity) {
+
+int GetDefaultCacheShardBits(size_t capacity, size_t min_shard_size) {
   int num_shard_bits = 0;
-  size_t min_shard_size = 512L * 1024L;  // Every shard is at least 512KB.
   size_t num_shards = capacity / min_shard_size;
   while (num_shards >>= 1) {
     if (++num_shard_bits >= 6) {
@@ -226,8 +91,10 @@ int GetDefaultCacheShardBits(size_t capacity) {
   return num_shard_bits;
 }
 
-int ShardedCache::GetNumShardBits() const { return BitsSetToOne(shard_mask_); }
+int ShardedCacheBase::GetNumShardBits() const {
+  return BitsSetToOne(shard_mask_);
+}
 
-uint32_t ShardedCache::GetNumShards() const { return shard_mask_ + 1; }
+uint32_t ShardedCacheBase::GetNumShards() const { return shard_mask_ + 1; }
 
 }  // namespace ROCKSDB_NAMESPACE

cache/sharded_cache.h

@@ -10,123 +10,298 @@
 #pragma once
 
 #include <atomic>
+#include <cstdint>
 #include <string>
 
+#include "port/lang.h"
 #include "port/port.h"
 #include "rocksdb/cache.h"
+#include "util/hash.h"
+#include "util/mutexlock.h"
 
 namespace ROCKSDB_NAMESPACE {
 
-// Single cache shard interface.
-class CacheShard {
+// Optional base class for classes implementing the CacheShard concept
+class CacheShardBase {
  public:
-  CacheShard() = default;
-  virtual ~CacheShard() = default;
+  explicit CacheShardBase(CacheMetadataChargePolicy metadata_charge_policy)
+      : metadata_charge_policy_(metadata_charge_policy) {}
 
   using DeleterFn = Cache::DeleterFn;
-  virtual Status Insert(const Slice& key, uint32_t hash, void* value,
-                        size_t charge, DeleterFn deleter,
-                        Cache::Handle** handle, Cache::Priority priority) = 0;
-  virtual Status Insert(const Slice& key, uint32_t hash, void* value,
-                        const Cache::CacheItemHelper* helper, size_t charge,
-                        Cache::Handle** handle, Cache::Priority priority) = 0;
-  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash) = 0;
-  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash,
-                                const Cache::CacheItemHelper* helper,
-                                const Cache::CreateCallback& create_cb,
-                                Cache::Priority priority, bool wait,
-                                Statistics* stats) = 0;
-  virtual bool Release(Cache::Handle* handle, bool useful,
-                       bool erase_if_last_ref) = 0;
-  virtual bool IsReady(Cache::Handle* handle) = 0;
-  virtual void Wait(Cache::Handle* handle) = 0;
-  virtual bool Ref(Cache::Handle* handle) = 0;
-  virtual bool Release(Cache::Handle* handle, bool erase_if_last_ref) = 0;
-  virtual void Erase(const Slice& key, uint32_t hash) = 0;
-  virtual void SetCapacity(size_t capacity) = 0;
-  virtual void SetStrictCapacityLimit(bool strict_capacity_limit) = 0;
-  virtual size_t GetUsage() const = 0;
-  virtual size_t GetPinnedUsage() const = 0;
+
+  // Expected by concept CacheShard (TODO with C++20 support)
+  // Some Defaults
+  std::string GetPrintableOptions() const { return ""; }
+  using HashVal = uint64_t;
+  using HashCref = uint64_t;
+  static inline HashVal ComputeHash(const Slice& key) {
+    return GetSliceNPHash64(key);
+  }
+  static inline uint32_t HashPieceForSharding(HashCref hash) {
+    return Lower32of64(hash);
+  }
+  void AppendPrintableOptions(std::string& /*str*/) const {}
+
+  // Must be provided for concept CacheShard (TODO with C++20 support)
+  /*
+  struct HandleImpl {  // for concept HandleImpl
+    HashVal hash;
+    HashCref GetHash() const;
+    ...
+  };
+  Status Insert(const Slice& key, HashCref hash, Cache::ObjectPtr value,
+                const Cache::CacheItemHelper* helper, size_t charge,
+                HandleImpl** handle, Cache::Priority priority) = 0;
+  HandleImpl* Lookup(const Slice& key, HashCref hash,
+                        const Cache::CacheItemHelper* helper,
+                        Cache::CreateContext* create_context,
+                        Cache::Priority priority, bool wait,
+                        Statistics* stats) = 0;
+  bool Release(HandleImpl* handle, bool useful, bool erase_if_last_ref) = 0;
+  bool IsReady(HandleImpl* handle) = 0;
+  void Wait(HandleImpl* handle) = 0;
+  bool Ref(HandleImpl* handle) = 0;
+  void Erase(const Slice& key, HashCref hash) = 0;
+  void SetCapacity(size_t capacity) = 0;
+  void SetStrictCapacityLimit(bool strict_capacity_limit) = 0;
+  size_t GetUsage() const = 0;
+  size_t GetPinnedUsage() const = 0;
+  size_t GetOccupancyCount() const = 0;
+  size_t GetTableAddressCount() const = 0;
   // Handles iterating over roughly `average_entries_per_lock` entries, using
   // `state` to somehow record where it last ended up. Caller initially uses
-  // *state == 0 and implementation sets *state = UINT32_MAX to indicate
+  // *state == 0 and implementation sets *state = SIZE_MAX to indicate
   // completion.
-  virtual void ApplyToSomeEntries(
-      const std::function<void(const Slice& key, void* value, size_t charge,
-                               DeleterFn deleter)>& callback,
-      uint32_t average_entries_per_lock, uint32_t* state) = 0;
-  virtual void EraseUnRefEntries() = 0;
-  virtual std::string GetPrintableOptions() const { return ""; }
-  void set_metadata_charge_policy(
-      CacheMetadataChargePolicy metadata_charge_policy) {
-    metadata_charge_policy_ = metadata_charge_policy;
-  }
+  void ApplyToSomeEntries(
+      const std::function<void(const Slice& key, ObjectPtr value,
+                               size_t charge,
+                               const Cache::CacheItemHelper* helper)>& callback,
+      size_t average_entries_per_lock, size_t* state) = 0;
+  void EraseUnRefEntries() = 0;
+  */
 
  protected:
-  CacheMetadataChargePolicy metadata_charge_policy_ = kDontChargeCacheMetadata;
+  const CacheMetadataChargePolicy metadata_charge_policy_;
 };
 
-// Generic cache interface which shards cache by hash of keys. 2^num_shard_bits
-// shards will be created, with capacity split evenly to each of the shards.
-// Keys are sharded by the highest num_shard_bits bits of hash value.
-class ShardedCache : public Cache {
+// Portions of ShardedCache that do not depend on the template parameter
+class ShardedCacheBase : public Cache {
  public:
-  ShardedCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit,
-               std::shared_ptr<MemoryAllocator> memory_allocator = nullptr);
-  virtual ~ShardedCache() = default;
-  virtual CacheShard* GetShard(uint32_t shard) = 0;
-  virtual const CacheShard* GetShard(uint32_t shard) const = 0;
-
-  virtual uint32_t GetHash(Handle* handle) const = 0;
-
-  virtual void SetCapacity(size_t capacity) override;
-  virtual void SetStrictCapacityLimit(bool strict_capacity_limit) override;
-
-  virtual Status Insert(const Slice& key, void* value, size_t charge,
-                        DeleterFn deleter, Handle** handle,
-                        Priority priority) override;
-  virtual Status Insert(const Slice& key, void* value,
-                        const CacheItemHelper* helper, size_t charge,
-                        Handle** handle = nullptr,
-                        Priority priority = Priority::LOW) override;
-  virtual Handle* Lookup(const Slice& key, Statistics* stats) override;
-  virtual Handle* Lookup(const Slice& key, const CacheItemHelper* helper,
-                         const CreateCallback& create_cb, Priority priority,
-                         bool wait, Statistics* stats = nullptr) override;
-  virtual bool Release(Handle* handle, bool useful,
-                       bool erase_if_last_ref = false) override;
-  virtual bool IsReady(Handle* handle) override;
-  virtual void Wait(Handle* handle) override;
-  virtual bool Ref(Handle* handle) override;
-  virtual bool Release(Handle* handle, bool erase_if_last_ref = false) override;
-  virtual void Erase(const Slice& key) override;
-  virtual uint64_t NewId() override;
-  virtual size_t GetCapacity() const override;
-  virtual bool HasStrictCapacityLimit() const override;
-  virtual size_t GetUsage() const override;
-  virtual size_t GetUsage(Handle* handle) const override;
-  virtual size_t GetPinnedUsage() const override;
-  virtual void ApplyToAllEntries(
-      const std::function<void(const Slice& key, void* value, size_t charge,
-                               DeleterFn deleter)>& callback,
-      const ApplyToAllEntriesOptions& opts) override;
-  virtual void EraseUnRefEntries() override;
-  virtual std::string GetPrintableOptions() const override;
+  ShardedCacheBase(size_t capacity, int num_shard_bits,
+                   bool strict_capacity_limit,
+                   std::shared_ptr<MemoryAllocator> memory_allocator);
+  virtual ~ShardedCacheBase() = default;
 
   int GetNumShardBits() const;
   uint32_t GetNumShards() const;
 
- protected:
-  inline uint32_t Shard(uint32_t hash) { return hash & shard_mask_; }
+  uint64_t NewId() override;
 
- private:
+  bool HasStrictCapacityLimit() const override;
+  size_t GetCapacity() const override;
+
+  using Cache::GetUsage;
+  size_t GetUsage(Handle* handle) const override;
+  std::string GetPrintableOptions() const override;
+
+ protected:  // fns
+  virtual void AppendPrintableOptions(std::string& str) const = 0;
+  size_t GetPerShardCapacity() const;
+  size_t ComputePerShardCapacity(size_t capacity) const;
+
+ protected:                        // data
+  std::atomic<uint64_t> last_id_;  // For NewId
   const uint32_t shard_mask_;
-  mutable port::Mutex capacity_mutex_;
-  size_t capacity_;
+
+  // Dynamic configuration parameters, guarded by config_mutex_
   bool strict_capacity_limit_;
-  std::atomic<uint64_t> last_id_;
+  size_t capacity_;
+  mutable port::Mutex config_mutex_;
 };
 
-extern int GetDefaultCacheShardBits(size_t capacity);
+// Generic cache interface that shards cache by hash of keys. 2^num_shard_bits
+// shards will be created, with capacity split evenly to each of the shards.
+// Keys are typically sharded by the lowest num_shard_bits bits of hash value
+// so that the upper bits of the hash value can keep a stable ordering of
+// table entries even as the table grows (using more upper hash bits).
+// See CacheShardBase above for what is expected of the CacheShard parameter.
+template <class CacheShard>
+class ShardedCache : public ShardedCacheBase {
+ public:
+  using HashVal = typename CacheShard::HashVal;
+  using HashCref = typename CacheShard::HashCref;
+  using HandleImpl = typename CacheShard::HandleImpl;
+
+  ShardedCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit,
+               std::shared_ptr<MemoryAllocator> allocator)
+      : ShardedCacheBase(capacity, num_shard_bits, strict_capacity_limit,
+                         allocator),
+        shards_(reinterpret_cast<CacheShard*>(port::cacheline_aligned_alloc(
+            sizeof(CacheShard) * GetNumShards()))),
+        destroy_shards_in_dtor_(false) {}
+
+  virtual ~ShardedCache() {
+    if (destroy_shards_in_dtor_) {
+      ForEachShard([](CacheShard* cs) { cs->~CacheShard(); });
+    }
+    port::cacheline_aligned_free(shards_);
+  }
+
+  CacheShard& GetShard(HashCref hash) {
+    return shards_[CacheShard::HashPieceForSharding(hash) & shard_mask_];
+  }
+
+  const CacheShard& GetShard(HashCref hash) const {
+    return shards_[CacheShard::HashPieceForSharding(hash) & shard_mask_];
+  }
+
+  void SetCapacity(size_t capacity) override {
+    MutexLock l(&config_mutex_);
+    capacity_ = capacity;
+    auto per_shard = ComputePerShardCapacity(capacity);
+    ForEachShard([=](CacheShard* cs) { cs->SetCapacity(per_shard); });
+  }
+
+  void SetStrictCapacityLimit(bool s_c_l) override {
+    MutexLock l(&config_mutex_);
+    strict_capacity_limit_ = s_c_l;
+    ForEachShard(
+        [s_c_l](CacheShard* cs) { cs->SetStrictCapacityLimit(s_c_l); });
+  }
+
+  Status Insert(const Slice& key, ObjectPtr value,
+                const CacheItemHelper* helper, size_t charge,
+                Handle** handle = nullptr,
+                Priority priority = Priority::LOW) override {
+    assert(helper);
+    HashVal hash = CacheShard::ComputeHash(key);
+    auto h_out = reinterpret_cast<HandleImpl**>(handle);
+    return GetShard(hash).Insert(key, hash, value, helper, charge, h_out,
+                                 priority);
+  }
+
+  Handle* Lookup(const Slice& key, const CacheItemHelper* helper = nullptr,
+                 CreateContext* create_context = nullptr,
+                 Priority priority = Priority::LOW, bool wait = true,
+                 Statistics* stats = nullptr) override {
+    HashVal hash = CacheShard::ComputeHash(key);
+    HandleImpl* result = GetShard(hash).Lookup(
+        key, hash, helper, create_context, priority, wait, stats);
+    return reinterpret_cast<Handle*>(result);
+  }
+
+  void Erase(const Slice& key) override {
+    HashVal hash = CacheShard::ComputeHash(key);
+    GetShard(hash).Erase(key, hash);
+  }
+
+  bool Release(Handle* handle, bool useful,
+               bool erase_if_last_ref = false) override {
+    auto h = reinterpret_cast<HandleImpl*>(handle);
+    return GetShard(h->GetHash()).Release(h, useful, erase_if_last_ref);
+  }
+  bool IsReady(Handle* handle) override {
+    auto h = reinterpret_cast<HandleImpl*>(handle);
+    return GetShard(h->GetHash()).IsReady(h);
+  }
+  void Wait(Handle* handle) override {
+    auto h = reinterpret_cast<HandleImpl*>(handle);
+    GetShard(h->GetHash()).Wait(h);
+  }
+  bool Ref(Handle* handle) override {
+    auto h = reinterpret_cast<HandleImpl*>(handle);
+    return GetShard(h->GetHash()).Ref(h);
+  }
+  bool Release(Handle* handle, bool erase_if_last_ref = false) override {
+    return Release(handle, true /*useful*/, erase_if_last_ref);
+  }
+  using ShardedCacheBase::GetUsage;
+  size_t GetUsage() const override {
+    return SumOverShards2(&CacheShard::GetUsage);
+  }
+  size_t GetPinnedUsage() const override {
+    return SumOverShards2(&CacheShard::GetPinnedUsage);
+  }
+  size_t GetOccupancyCount() const override {
+    return SumOverShards2(&CacheShard::GetPinnedUsage);
+  }
+  size_t GetTableAddressCount() const override {
+    return SumOverShards2(&CacheShard::GetTableAddressCount);
+  }
+  void ApplyToAllEntries(
+      const std::function<void(const Slice& key, ObjectPtr value, size_t charge,
+                               const CacheItemHelper* helper)>& callback,
+      const ApplyToAllEntriesOptions& opts) override {
+    uint32_t num_shards = GetNumShards();
+    // Iterate over part of each shard, rotating between shards, to
+    // minimize impact on latency of concurrent operations.
+    std::unique_ptr<size_t[]> states(new size_t[num_shards]{});
+
+    size_t aepl = opts.average_entries_per_lock;
+    aepl = std::min(aepl, size_t{1});
+
+    bool remaining_work;
+    do {
+      remaining_work = false;
+      for (uint32_t i = 0; i < num_shards; i++) {
+        if (states[i] != SIZE_MAX) {
+          shards_[i].ApplyToSomeEntries(callback, aepl, &states[i]);
+          remaining_work |= states[i] != SIZE_MAX;
+        }
+      }
+    } while (remaining_work);
+  }
+
+  virtual void EraseUnRefEntries() override {
+    ForEachShard([](CacheShard* cs) { cs->EraseUnRefEntries(); });
+  }
+
+  void DisownData() override {
+    // Leak data only if that won't generate an ASAN/valgrind warning.
+    if (!kMustFreeHeapAllocations) {
+      destroy_shards_in_dtor_ = false;
+    }
+  }
+
+ protected:
+  inline void ForEachShard(const std::function<void(CacheShard*)>& fn) {
+    uint32_t num_shards = GetNumShards();
+    for (uint32_t i = 0; i < num_shards; i++) {
+      fn(shards_ + i);
+    }
+  }
+
+  inline size_t SumOverShards(
+      const std::function<size_t(CacheShard&)>& fn) const {
+    uint32_t num_shards = GetNumShards();
+    size_t result = 0;
+    for (uint32_t i = 0; i < num_shards; i++) {
+      result += fn(shards_[i]);
+    }
+    return result;
+  }
+
+  inline size_t SumOverShards2(size_t (CacheShard::*fn)() const) const {
+    return SumOverShards([fn](CacheShard& cs) { return (cs.*fn)(); });
+  }
+
+  // Must be called exactly once by derived class constructor
+  void InitShards(const std::function<void(CacheShard*)>& placement_new) {
+    ForEachShard(placement_new);
+    destroy_shards_in_dtor_ = true;
+  }
+
+  void AppendPrintableOptions(std::string& str) const override {
+    shards_[0].AppendPrintableOptions(str);
+  }
+
+ private:
+  CacheShard* const shards_;
+  bool destroy_shards_in_dtor_;
+};
+
+// 512KB is traditional minimum shard size.
+int GetDefaultCacheShardBits(size_t capacity,
+                             size_t min_shard_size = 512U * 1024U);
 
 }  // namespace ROCKSDB_NAMESPACE

cache/typed_cache.h

@@ -0,0 +1,339 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+// APIs for accessing Cache in a type-safe and convenient way. Cache is kept
+// at a low, thin level of abstraction so that different implementations can
+// be plugged in, but these wrappers provide clean, convenient access to the
+// most common operations.
+//
+// A number of template classes are needed for sharing common structure. The
+// key classes are these:
+//
+// * PlaceholderCacheInterface - Used for making cache reservations, with
+// entries that have a charge but no value.
+// * BasicTypedCacheInterface<TValue> - Used for primary cache storage of
+// objects of type TValue.
+// * FullTypedCacheHelper<TValue, TCreateContext> - Used for secondary cache
+// compatible storage of objects of type TValue.
+// * For each of these, there's a "Shared" version
+// (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache,
+// rather than assuming external ownership by holding only a raw `Cache*`.
+
+#pragma once
+
+#include <algorithm>
+#include <cstdint>
+#include <memory>
+#include <type_traits>
+
+#include "cache/cache_helpers.h"
+#include "rocksdb/advanced_options.h"
+#include "rocksdb/cache.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+// For future consideration:
+// * Pass in value to Insert with std::unique_ptr& to simplify ownership
+//   transfer logic in callers
+// * Make key type a template parameter (e.g. useful for table cache)
+// * Closer integration with CacheHandleGuard (opt-in, so not always
+//   paying the extra overhead)
+
+#define CACHE_TYPE_DEFS()                     \
+  using Priority = Cache::Priority;           \
+  using Handle = Cache::Handle;               \
+  using ObjectPtr = Cache::ObjectPtr;         \
+  using CreateContext = Cache::CreateContext; \
+  using CacheItemHelper = Cache::CacheItemHelper /* caller ; */
+
+template <typename CachePtr>
+class BaseCacheInterface {
+ public:
+  CACHE_TYPE_DEFS();
+
+  /*implicit*/ BaseCacheInterface(CachePtr cache) : cache_(std::move(cache)) {}
+
+  inline void Release(Handle* handle) { cache_->Release(handle); }
+
+  inline void ReleaseAndEraseIfLastRef(Handle* handle) {
+    cache_->Release(handle, /*erase_if_last_ref*/ true);
+  }
+
+  inline void RegisterReleaseAsCleanup(Handle* handle, Cleanable& cleanable) {
+    cleanable.RegisterCleanup(&ReleaseCacheHandleCleanup, get(), handle);
+  }
+
+  inline Cache* get() const { return &*cache_; }
+
+  explicit inline operator bool() const noexcept { return cache_ != nullptr; }
+
+ protected:
+  CachePtr cache_;
+};
+
+// PlaceholderCacheInterface - Used for making cache reservations, with
+// entries that have a charge but no value. CacheEntryRole is required as
+// a template parameter.
+template <CacheEntryRole kRole, typename CachePtr = Cache*>
+class PlaceholderCacheInterface : public BaseCacheInterface<CachePtr> {
+ public:
+  CACHE_TYPE_DEFS();
+  using BaseCacheInterface<CachePtr>::BaseCacheInterface;
+
+  inline Status Insert(const Slice& key, size_t charge, Handle** handle) {
+    return this->cache_->Insert(key, /*value=*/nullptr, &kHelper, charge,
+                                handle);
+  }
+
+  static constexpr Cache::CacheItemHelper kHelper{kRole};
+};
+
+template <CacheEntryRole kRole>
+using PlaceholderSharedCacheInterface =
+    PlaceholderCacheInterface<kRole, std::shared_ptr<Cache>>;
+
+template <class TValue>
+class BasicTypedCacheHelperFns {
+ public:
+  CACHE_TYPE_DEFS();
+  // E.g. char* for char[]
+  using TValuePtr = std::remove_extent_t<TValue>*;
+
+ protected:
+  inline static ObjectPtr UpCastValue(TValuePtr value) { return value; }
+  inline static TValuePtr DownCastValue(ObjectPtr value) {
+    return static_cast<TValuePtr>(value);
+  }
+
+  static void Delete(ObjectPtr value, MemoryAllocator* allocator) {
+    // FIXME: Currently, no callers actually allocate the ObjectPtr objects
+    // using the custom allocator, just subobjects that keep a reference to
+    // the allocator themselves (with CacheAllocationPtr).
+    if (/*DISABLED*/ false && allocator) {
+      if constexpr (std::is_destructible_v<TValue>) {
+        DownCastValue(value)->~TValue();
+      }
+      allocator->Deallocate(value);
+    } else {
+      // Like delete but properly handles TValue=char[] etc.
+      std::default_delete<TValue>{}(DownCastValue(value));
+    }
+  }
+};
+
+// In its own class to try to minimize the number of distinct CacheItemHelper
+// instances (e.g. don't vary by CachePtr)
+template <class TValue, CacheEntryRole kRole>
+class BasicTypedCacheHelper : public BasicTypedCacheHelperFns<TValue> {
+ public:
+  static constexpr Cache::CacheItemHelper kBasicHelper{
+      kRole, &BasicTypedCacheHelper::Delete};
+};
+
+// BasicTypedCacheInterface - Used for primary cache storage of objects of
+// type TValue, which can be cleaned up with std::default_delete<TValue>. The
+// role is provided by TValue::kCacheEntryRole or given in an optional
+// template parameter.
+template <class TValue, CacheEntryRole kRole = TValue::kCacheEntryRole,
+          typename CachePtr = Cache*>
+class BasicTypedCacheInterface : public BaseCacheInterface<CachePtr>,
+                                 public BasicTypedCacheHelper<TValue, kRole> {
+ public:
+  CACHE_TYPE_DEFS();
+  using typename BasicTypedCacheHelperFns<TValue>::TValuePtr;
+  struct TypedHandle : public Handle {};
+  using BasicTypedCacheHelper<TValue, kRole>::kBasicHelper;
+  // ctor
+  using BaseCacheInterface<CachePtr>::BaseCacheInterface;
+
+  inline Status Insert(const Slice& key, TValuePtr value, size_t charge,
+                       TypedHandle** handle = nullptr,
+                       Priority priority = Priority::LOW) {
+    auto untyped_handle = reinterpret_cast<Handle**>(handle);
+    return this->cache_->Insert(
+        key, BasicTypedCacheHelperFns<TValue>::UpCastValue(value),
+        &kBasicHelper, charge, untyped_handle, priority);
+  }
+
+  inline TypedHandle* Lookup(const Slice& key, Statistics* stats = nullptr) {
+    return reinterpret_cast<TypedHandle*>(
+        this->cache_->BasicLookup(key, stats));
+  }
+
+  inline CacheHandleGuard<TValue> Guard(TypedHandle* handle) {
+    if (handle) {
+      return CacheHandleGuard<TValue>(&*this->cache_, handle);
+    } else {
+      return {};
+    }
+  }
+
+  inline std::shared_ptr<TValue> SharedGuard(TypedHandle* handle) {
+    if (handle) {
+      return MakeSharedCacheHandleGuard<TValue>(&*this->cache_, handle);
+    } else {
+      return {};
+    }
+  }
+
+  inline TValuePtr Value(TypedHandle* handle) {
+    return BasicTypedCacheHelperFns<TValue>::DownCastValue(
+        this->cache_->Value(handle));
+  }
+};
+
+// BasicTypedSharedCacheInterface - Like BasicTypedCacheInterface but with a
+// shared_ptr<Cache> for keeping Cache alive.
+template <class TValue, CacheEntryRole kRole = TValue::kCacheEntryRole>
+using BasicTypedSharedCacheInterface =
+    BasicTypedCacheInterface<TValue, kRole, std::shared_ptr<Cache>>;
+
+// TValue must implement ContentSlice() and ~TValue
+// TCreateContext must implement Create(std::unique_ptr<TValue>*, ...)
+template <class TValue, class TCreateContext>
+class FullTypedCacheHelperFns : public BasicTypedCacheHelperFns<TValue> {
+ public:
+  CACHE_TYPE_DEFS();
+
+ protected:
+  using typename BasicTypedCacheHelperFns<TValue>::TValuePtr;
+  using BasicTypedCacheHelperFns<TValue>::DownCastValue;
+  using BasicTypedCacheHelperFns<TValue>::UpCastValue;
+
+  static size_t Size(ObjectPtr v) {
+    TValuePtr value = DownCastValue(v);
+    auto slice = value->ContentSlice();
+    return slice.size();
+  }
+
+  static Status SaveTo(ObjectPtr v, size_t from_offset, size_t length,
+                       char* out) {
+    TValuePtr value = DownCastValue(v);
+    auto slice = value->ContentSlice();
+    assert(from_offset < slice.size());
+    assert(from_offset + length <= slice.size());
+    std::copy_n(slice.data() + from_offset, length, out);
+    return Status::OK();
+  }
+
+  static Status Create(const Slice& data, CreateContext* context,
+                       MemoryAllocator* allocator, ObjectPtr* out_obj,
+                       size_t* out_charge) {
+    std::unique_ptr<TValue> value = nullptr;
+    if constexpr (sizeof(TCreateContext) > 0) {
+      TCreateContext* tcontext = static_cast<TCreateContext*>(context);
+      tcontext->Create(&value, out_charge, data, allocator);
+    } else {
+      TCreateContext::Create(&value, out_charge, data, allocator);
+    }
+    *out_obj = UpCastValue(value.release());
+    return Status::OK();
+  }
+};
+
+// In its own class to try to minimize the number of distinct CacheItemHelper
+// instances (e.g. don't vary by CachePtr)
+template <class TValue, class TCreateContext, CacheEntryRole kRole>
+class FullTypedCacheHelper
+    : public FullTypedCacheHelperFns<TValue, TCreateContext> {
+ public:
+  static constexpr Cache::CacheItemHelper kFullHelper{
+      kRole, &FullTypedCacheHelper::Delete, &FullTypedCacheHelper::Size,
+      &FullTypedCacheHelper::SaveTo, &FullTypedCacheHelper::Create};
+};
+
+// FullTypedCacheHelper - Used for secondary cache compatible storage of
+// objects of type TValue. In addition to BasicTypedCacheInterface constraints,
+// we require TValue::ContentSlice() to return persistable data. This
+// simplifies usage for the normal case of simple secondary cache compatibility
+// (can give you a Slice to the data already in memory). In addition to
+// TCreateContext performing the role of Cache::CreateContext, it is also
+// expected to provide a function Create(std::unique_ptr<TValue>* value,
+// size_t* out_charge, const Slice& data, MemoryAllocator* allocator) for
+// creating new TValue.
+template <class TValue, class TCreateContext,
+          CacheEntryRole kRole = TValue::kCacheEntryRole,
+          typename CachePtr = Cache*>
+class FullTypedCacheInterface
+    : public BasicTypedCacheInterface<TValue, kRole, CachePtr>,
+      public FullTypedCacheHelper<TValue, TCreateContext, kRole> {
+ public:
+  CACHE_TYPE_DEFS();
+  using typename BasicTypedCacheInterface<TValue, kRole, CachePtr>::TypedHandle;
+  using typename BasicTypedCacheHelperFns<TValue>::TValuePtr;
+  using BasicTypedCacheHelper<TValue, kRole>::kBasicHelper;
+  using FullTypedCacheHelper<TValue, TCreateContext, kRole>::kFullHelper;
+  using BasicTypedCacheHelperFns<TValue>::UpCastValue;
+  using BasicTypedCacheHelperFns<TValue>::DownCastValue;
+  // ctor
+  using BasicTypedCacheInterface<TValue, kRole,
+                                 CachePtr>::BasicTypedCacheInterface;
+
+  // Insert with SecondaryCache compatibility (subject to CacheTier).
+  // (Basic Insert() also inherited.)
+  inline Status InsertFull(
+      const Slice& key, TValuePtr value, size_t charge,
+      TypedHandle** handle = nullptr, Priority priority = Priority::LOW,
+      CacheTier lowest_used_cache_tier = CacheTier::kNonVolatileBlockTier) {
+    auto untyped_handle = reinterpret_cast<Handle**>(handle);
+    auto helper = lowest_used_cache_tier == CacheTier::kNonVolatileBlockTier
+                      ? &kFullHelper
+                      : &kBasicHelper;
+    return this->cache_->Insert(key, UpCastValue(value), helper, charge,
+                                untyped_handle, priority);
+  }
+
+  // Like SecondaryCache::InsertSaved, with SecondaryCache compatibility
+  // (subject to CacheTier).
+  inline Status InsertSaved(
+      const Slice& key, const Slice& data, TCreateContext* create_context,
+      Priority priority = Priority::LOW,
+      CacheTier lowest_used_cache_tier = CacheTier::kNonVolatileBlockTier,
+      size_t* out_charge = nullptr) {
+    ObjectPtr value;
+    size_t charge;
+    Status st = kFullHelper.create_cb(data, create_context,
+                                      this->cache_->memory_allocator(), &value,
+                                      &charge);
+    if (out_charge) {
+      *out_charge = charge;
+    }
+    if (st.ok()) {
+      st = InsertFull(key, DownCastValue(value), charge, nullptr /*handle*/,
+                      priority, lowest_used_cache_tier);
+    } else {
+      kFullHelper.del_cb(value, this->cache_->memory_allocator());
+    }
+    return st;
+  }
+
+  // Lookup with SecondaryCache support (subject to CacheTier).
+  // (Basic Lookup() also inherited.)
+  inline TypedHandle* LookupFull(
+      const Slice& key, TCreateContext* create_context = nullptr,
+      Priority priority = Priority::LOW, bool wait = true,
+      Statistics* stats = nullptr,
+      CacheTier lowest_used_cache_tier = CacheTier::kNonVolatileBlockTier) {
+    if (lowest_used_cache_tier == CacheTier::kNonVolatileBlockTier) {
+      return reinterpret_cast<TypedHandle*>(this->cache_->Lookup(
+          key, &kFullHelper, create_context, priority, wait, stats));
+    } else {
+      return BasicTypedCacheInterface<TValue, kRole, CachePtr>::Lookup(key,
+                                                                       stats);
+    }
+  }
+};
+
+// FullTypedSharedCacheInterface - Like FullTypedCacheInterface but with a
+// shared_ptr<Cache> for keeping Cache alive.
+template <class TValue, class TCreateContext,
+          CacheEntryRole kRole = TValue::kCacheEntryRole>
+using FullTypedSharedCacheInterface =
+    FullTypedCacheInterface<TValue, TCreateContext, kRole,
+                            std::shared_ptr<Cache>>;
+
+#undef CACHE_TYPE_DEFS
+
+}  // namespace ROCKSDB_NAMESPACE

coverage/parse_gcov_output.py

@@ -47,35 +47,39 @@ def parse_gcov_report(gcov_input):
 
     return per_file_coverage, total_coverage
 
+
 def get_option_parser():
-    usage = "Parse the gcov output and generate more human-readable code " +\
-            "coverage report."
+    usage = (
+        "Parse the gcov output and generate more human-readable code "
+        + "coverage report."
+    )
     parser = optparse.OptionParser(usage)
 
     parser.add_option(
-        "--interested-files", "-i",
+        "--interested-files",
+        "-i",
         dest="filenames",
-        help="Comma separated files names. if specified, we will display " +
-             "the coverage report only for interested source files. " +
-             "Otherwise we will display the coverage report for all " +
-             "source files."
+        help="Comma separated files names. if specified, we will display "
+        + "the coverage report only for interested source files. "
+        + "Otherwise we will display the coverage report for all "
+        + "source files.",
     )
     return parser
 
+
 def display_file_coverage(per_file_coverage, total_coverage):
     # To print out auto-adjustable column, we need to know the longest
     # length of file names.
-    max_file_name_length = max(
-        len(fname) for fname in per_file_coverage.keys()
-    )
+    max_file_name_length = max(len(fname) for fname in per_file_coverage.keys())
 
     # -- Print header
     # size of separator is determined by 3 column sizes:
     # file name, coverage percentage and lines.
-    header_template = \
-        "%" + str(max_file_name_length) + "s\t%s\t%s"
+    header_template = "%" + str(max_file_name_length) + "s\t%s\t%s"
     separator = "-" * (max_file_name_length + 10 + 20)
-    print(header_template % ("Filename", "Coverage", "Lines"))  # noqa: E999 T25377293 Grandfathered in
+    print(
+        header_template % ("Filename", "Coverage", "Lines")
+    )  # noqa: E999 T25377293 Grandfathered in
     print(separator)
 
     # -- Print body
@@ -91,13 +95,14 @@ def display_file_coverage(per_file_coverage, total_coverage):
         print(separator)
         print(record_template % ("Total", total_coverage[0], total_coverage[1]))
 
+
 def report_coverage():
     parser = get_option_parser()
     (options, args) = parser.parse_args()
 
     interested_files = set()
     if options.filenames is not None:
-        interested_files = set(f.strip() for f in options.filenames.split(','))
+        interested_files = {f.strip() for f in options.filenames.split(",")}
 
     # To make things simple, right now we only read gcov report from the input
     per_file_coverage, total_coverage = parse_gcov_report(sys.stdin)
@@ -105,7 +110,8 @@ def report_coverage():
     # Check if we need to display coverage info for interested files.
     if len(interested_files):
         per_file_coverage = dict(
-            (fname, per_file_coverage[fname]) for fname in interested_files
+            (fname, per_file_coverage[fname])
+            for fname in interested_files
             if fname in per_file_coverage
         )
         # If we only interested in several files, it makes no sense to report
@@ -117,5 +123,6 @@ def report_coverage():
         return
     display_file_coverage(per_file_coverage, total_coverage)
 
+
 if __name__ == "__main__":
     report_coverage()

crash_test.mk

@@ -8,7 +8,7 @@ DB_STRESS_CMD?=./db_stress
 include common.mk
 
 CRASHTEST_MAKE=$(MAKE) -f crash_test.mk
-CRASHTEST_PY=$(PYTHON) -u tools/db_crashtest.py --stress_cmd=$(DB_STRESS_CMD)
+CRASHTEST_PY=$(PYTHON) -u tools/db_crashtest.py --stress_cmd=$(DB_STRESS_CMD) --cleanup_cmd='$(DB_CLEANUP_CMD)'
 
 .PHONY: crash_test crash_test_with_atomic_flush crash_test_with_txn \
 	crash_test_with_best_efforts_recovery crash_test_with_ts \
@@ -78,7 +78,7 @@ blackbox_crash_test_with_multiops_wp_txn: $(DB_STRESS_CMD)
 	$(CRASHTEST_PY) --test_multiops_txn --write_policy write_prepared blackbox $(CRASH_TEST_EXT_ARGS)
 
 blackbox_crash_test_with_tiered_storage: $(DB_STRESS_CMD)
-	$(CRASHTEST_PY) --enable_tiered_storage blackbox $(CRASH_TEST_EXT_ARGS)
+	$(CRASHTEST_PY) --test_tiered_storage blackbox $(CRASH_TEST_EXT_ARGS)
 
 ifeq ($(CRASH_TEST_KILL_ODD),)
   CRASH_TEST_KILL_ODD=888887
@@ -103,5 +103,5 @@ whitebox_crash_test_with_ts: $(DB_STRESS_CMD)
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
 
 whitebox_crash_test_with_tiered_storage: $(DB_STRESS_CMD)
-	$(CRASHTEST_PY) --enable_tiered_storage whitebox --random_kill_odd \
+	$(CRASHTEST_PY) --test_tiered_storage whitebox --random_kill_odd \
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)

db/arena_wrapped_db_iter.cc

@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "db/arena_wrapped_db_iter.h"
+
 #include "memory/arena.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -45,6 +46,7 @@ void ArenaWrappedDBIter::Init(
   sv_number_ = version_number;
   read_options_ = read_options;
   allow_refresh_ = allow_refresh;
+  memtable_range_tombstone_iter_ = nullptr;
 }
 
 Status ArenaWrappedDBIter::Refresh() {
@@ -58,41 +60,68 @@ Status ArenaWrappedDBIter::Refresh() {
   uint64_t cur_sv_number = cfd_->GetSuperVersionNumber();
   TEST_SYNC_POINT("ArenaWrappedDBIter::Refresh:1");
   TEST_SYNC_POINT("ArenaWrappedDBIter::Refresh:2");
+  auto reinit_internal_iter = [&]() {
+    Env* env = db_iter_->env();
+    db_iter_->~DBIter();
+    arena_.~Arena();
+    new (&arena_) Arena();
+
+    SuperVersion* sv = cfd_->GetReferencedSuperVersion(db_impl_);
+    SequenceNumber latest_seq = db_impl_->GetLatestSequenceNumber();
+    if (read_callback_) {
+      read_callback_->Refresh(latest_seq);
+    }
+    Init(env, read_options_, *(cfd_->ioptions()), sv->mutable_cf_options,
+         sv->current, latest_seq,
+         sv->mutable_cf_options.max_sequential_skip_in_iterations,
+         cur_sv_number, read_callback_, db_impl_, cfd_, expose_blob_index_,
+         allow_refresh_);
+
+    InternalIterator* internal_iter = db_impl_->NewInternalIterator(
+        read_options_, cfd_, sv, &arena_, latest_seq,
+        /* allow_unprepared_value */ true, /* db_iter */ this);
+    SetIterUnderDBIter(internal_iter);
+  };
   while (true) {
     if (sv_number_ != cur_sv_number) {
-      Env* env = db_iter_->env();
-      db_iter_->~DBIter();
-      arena_.~Arena();
-      new (&arena_) Arena();
-
-      SuperVersion* sv = cfd_->GetReferencedSuperVersion(db_impl_);
-      SequenceNumber latest_seq = db_impl_->GetLatestSequenceNumber();
-      if (read_callback_) {
-        read_callback_->Refresh(latest_seq);
-      }
-      Init(env, read_options_, *(cfd_->ioptions()), sv->mutable_cf_options,
-           sv->current, latest_seq,
-           sv->mutable_cf_options.max_sequential_skip_in_iterations,
-           cur_sv_number, read_callback_, db_impl_, cfd_, expose_blob_index_,
-           allow_refresh_);
-
-      InternalIterator* internal_iter = db_impl_->NewInternalIterator(
-          read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator(),
-          latest_seq, /* allow_unprepared_value */ true);
-      SetIterUnderDBIter(internal_iter);
+      reinit_internal_iter();
       break;
     } else {
       SequenceNumber latest_seq = db_impl_->GetLatestSequenceNumber();
       // Refresh range-tombstones in MemTable
       if (!read_options_.ignore_range_deletions) {
         SuperVersion* sv = cfd_->GetThreadLocalSuperVersion(db_impl_);
-        ReadRangeDelAggregator* range_del_agg =
-            db_iter_->GetRangeDelAggregator();
-        std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter;
-        range_del_iter.reset(sv->mem->NewRangeTombstoneIterator(
-            read_options_, latest_seq, false /* immutable_memtable */));
-        range_del_agg->AddTombstones(std::move(range_del_iter));
-        cfd_->ReturnThreadLocalSuperVersion(sv);
+        TEST_SYNC_POINT_CALLBACK("ArenaWrappedDBIter::Refresh:SV", nullptr);
+        auto t = sv->mem->NewRangeTombstoneIterator(
+            read_options_, latest_seq, false /* immutable_memtable */);
+        if (!t || t->empty()) {
+          // If memtable_range_tombstone_iter_ points to a non-empty tombstone
+          // iterator, then it means sv->mem is not the memtable that
+          // memtable_range_tombstone_iter_ points to, so SV must have changed
+          // after the sv_number_ != cur_sv_number check above. We will fall
+          // back to re-init the InternalIterator, and the tombstone iterator
+          // will be freed during db_iter destruction there.
+          if (memtable_range_tombstone_iter_) {
+            assert(!*memtable_range_tombstone_iter_ ||
+                   sv_number_ != cfd_->GetSuperVersionNumber());
+          }
+          delete t;
+        } else {  // current mutable memtable has range tombstones
+          if (!memtable_range_tombstone_iter_) {
+            delete t;
+            db_impl_->ReturnAndCleanupSuperVersion(cfd_, sv);
+            // The memtable under DBIter did not have range tombstone before
+            // refresh.
+            reinit_internal_iter();
+            break;
+          } else {
+            delete *memtable_range_tombstone_iter_;
+            *memtable_range_tombstone_iter_ = new TruncatedRangeDelIterator(
+                std::unique_ptr<FragmentedRangeTombstoneIterator>(t),
+                &cfd_->internal_comparator(), nullptr, nullptr);
+          }
+        }
+        db_impl_->ReturnAndCleanupSuperVersion(cfd_, sv);
       }
       // Refresh latest sequence number
       db_iter_->set_sequence(latest_seq);

db/arena_wrapped_db_iter.h

@@ -9,7 +9,9 @@
 
 #pragma once
 #include <stdint.h>
+
 #include <string>
+
 #include "db/db_impl/db_impl.h"
 #include "db/db_iter.h"
 #include "db/range_del_aggregator.h"
@@ -44,9 +46,7 @@ class ArenaWrappedDBIter : public Iterator {
   // Get the arena to be used to allocate memory for DBIter to be wrapped,
   // as well as child iterators in it.
   virtual Arena* GetArena() { return &arena_; }
-  virtual ReadRangeDelAggregator* GetRangeDelAggregator() {
-    return db_iter_->GetRangeDelAggregator();
-  }
+
   const ReadOptions& GetReadOptions() { return read_options_; }
 
   // Set the internal iterator wrapped inside the DB Iterator. Usually it is
@@ -55,6 +55,10 @@ class ArenaWrappedDBIter : public Iterator {
     db_iter_->SetIter(iter);
   }
 
+  void SetMemtableRangetombstoneIter(TruncatedRangeDelIterator** iter) {
+    memtable_range_tombstone_iter_ = iter;
+  }
+
   bool Valid() const override { return db_iter_->Valid(); }
   void SeekToFirst() override { db_iter_->SeekToFirst(); }
   void SeekToLast() override { db_iter_->SeekToLast(); }
@@ -68,6 +72,7 @@ class ArenaWrappedDBIter : public Iterator {
   void Prev() override { db_iter_->Prev(); }
   Slice key() const override { return db_iter_->key(); }
   Slice value() const override { return db_iter_->value(); }
+  const WideColumns& columns() const override { return db_iter_->columns(); }
   Status status() const override { return db_iter_->status(); }
   Slice timestamp() const override { return db_iter_->timestamp(); }
   bool IsBlob() const { return db_iter_->IsBlob(); }
@@ -104,6 +109,9 @@ class ArenaWrappedDBIter : public Iterator {
   ReadCallback* read_callback_;
   bool expose_blob_index_ = false;
   bool allow_refresh_ = true;
+  // If this is nullptr, it means the mutable memtable does not contain range
+  // tombstone when added under this DBIter.
+  TruncatedRangeDelIterator** memtable_range_tombstone_iter_ = nullptr;
 };
 
 // Generate the arena wrapped iterator class.

db/blob/blob_contents.cc

@@ -0,0 +1,42 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "db/blob/blob_contents.h"
+
+#include <cassert>
+
+#include "cache/cache_entry_roles.h"
+#include "cache/cache_helpers.h"
+#include "port/malloc.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+size_t BlobContents::ApproximateMemoryUsage() const {
+  size_t usage = 0;
+
+  if (allocation_) {
+    MemoryAllocator* const allocator = allocation_.get_deleter().allocator;
+
+    if (allocator) {
+      usage += allocator->UsableSize(allocation_.get(), data_.size());
+    } else {
+#ifdef ROCKSDB_MALLOC_USABLE_SIZE
+      usage += malloc_usable_size(allocation_.get());
+#else
+      usage += data_.size();
+#endif
+    }
+  }
+
+#ifdef ROCKSDB_MALLOC_USABLE_SIZE
+  usage += malloc_usable_size(const_cast<BlobContents*>(this));
+#else
+  usage += sizeof(*this);
+#endif
+
+  return usage;
+}
+
+}  // namespace ROCKSDB_NAMESPACE

db/blob/blob_contents.h

@@ -0,0 +1,59 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#include <memory>
+
+#include "memory/memory_allocator.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/rocksdb_namespace.h"
+#include "rocksdb/slice.h"
+#include "rocksdb/status.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+// A class representing a single uncompressed value read from a blob file.
+class BlobContents {
+ public:
+  BlobContents(CacheAllocationPtr&& allocation, size_t size)
+      : allocation_(std::move(allocation)), data_(allocation_.get(), size) {}
+
+  BlobContents(const BlobContents&) = delete;
+  BlobContents& operator=(const BlobContents&) = delete;
+
+  BlobContents(BlobContents&&) = default;
+  BlobContents& operator=(BlobContents&&) = default;
+
+  ~BlobContents() = default;
+
+  const Slice& data() const { return data_; }
+  size_t size() const { return data_.size(); }
+
+  size_t ApproximateMemoryUsage() const;
+
+  // For TypedCacheInterface
+  const Slice& ContentSlice() const { return data_; }
+  static constexpr CacheEntryRole kCacheEntryRole = CacheEntryRole::kBlobValue;
+
+ private:
+  CacheAllocationPtr allocation_;
+  Slice data_;
+};
+
+class BlobContentsCreator : public Cache::CreateContext {
+ public:
+  static void Create(std::unique_ptr<BlobContents>* out, size_t* out_charge,
+                     const Slice& contents, MemoryAllocator* alloc) {
+    auto raw = new BlobContents(AllocateAndCopyBlock(contents, alloc),
+                                contents.size());
+    out->reset(raw);
+    if (out_charge) {
+      *out_charge = raw->ApproximateMemoryUsage();
+    }
+  }
+};
+
+}  // namespace ROCKSDB_NAMESPACE

db/blob/blob_counting_iterator_test.cc

@@ -321,6 +321,7 @@ TEST(BlobCountingIteratorTest, CorruptBlobIndex) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_file_addition_test.cc

@@ -205,6 +205,7 @@ TEST_F(BlobFileAdditionTest, ForwardIncompatibleCustomField) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_file_builder.cc

@@ -7,11 +7,13 @@
 
 #include <cassert>
 
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_file_addition.h"
 #include "db/blob/blob_file_completion_callback.h"
 #include "db/blob/blob_index.h"
 #include "db/blob/blob_log_format.h"
 #include "db/blob/blob_log_writer.h"
+#include "db/blob/blob_source.h"
 #include "db/event_helpers.h"
 #include "db/version_set.h"
 #include "file/filename.h"
@@ -392,7 +394,7 @@ Status BlobFileBuilder::PutBlobIntoCacheIfNeeded(const Slice& blob,
                                                  uint64_t blob_offset) const {
   Status s = Status::OK();
 
-  auto blob_cache = immutable_options_->blob_cache;
+  BlobSource::SharedCacheInterface blob_cache{immutable_options_->blob_cache};
   auto statistics = immutable_options_->statistics.get();
   bool warm_cache =
       prepopulate_blob_cache_ == PrepopulateBlobCache::kFlushOnly &&
@@ -406,22 +408,12 @@ Status BlobFileBuilder::PutBlobIntoCacheIfNeeded(const Slice& blob,
 
     const Cache::Priority priority = Cache::Priority::BOTTOM;
 
-    // Objects to be put into the cache have to be heap-allocated and
-    // self-contained, i.e. own their contents. The Cache has to be able to
-    // take unique ownership of them. Therefore, we copy the blob into a
-    // string directly, and insert that into the cache.
-    std::unique_ptr<std::string> buf = std::make_unique<std::string>();
-    buf->assign(blob.data(), blob.size());
+    s = blob_cache.InsertSaved(key, blob, nullptr /*context*/, priority,
+                               immutable_options_->lowest_used_cache_tier);
 
-    // TODO: support custom allocators and provide a better estimated memory
-    // usage using malloc_usable_size.
-    s = blob_cache->Insert(key, buf.get(), buf->size(),
-                           &DeleteCacheEntry<std::string>,
-                           nullptr /* cache_handle */, priority);
     if (s.ok()) {
       RecordTick(statistics, BLOB_DB_CACHE_ADD);
-      RecordTick(statistics, BLOB_DB_CACHE_BYTES_WRITE, buf->size());
-      buf.release();
+      RecordTick(statistics, BLOB_DB_CACHE_BYTES_WRITE, blob.size());
     } else {
       RecordTick(statistics, BLOB_DB_CACHE_ADD_FAILURES);
     }

db/blob/blob_file_builder_test.cc

@@ -674,6 +674,7 @@ TEST_P(BlobFileBuilderIOErrorTest, IOError) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_file_cache.cc

@@ -42,13 +42,13 @@ Status BlobFileCache::GetBlobFileReader(
   assert(blob_file_reader);
   assert(blob_file_reader->IsEmpty());
 
-  const Slice key = GetSlice(&blob_file_number);
+  const Slice key = GetSliceForKey(&blob_file_number);
 
   assert(cache_);
 
-  Cache::Handle* handle = cache_->Lookup(key);
+  TypedHandle* handle = cache_.Lookup(key);
   if (handle) {
-    *blob_file_reader = CacheHandleGuard<BlobFileReader>(cache_, handle);
+    *blob_file_reader = cache_.Guard(handle);
     return Status::OK();
   }
 
@@ -57,9 +57,9 @@ Status BlobFileCache::GetBlobFileReader(
   // Check again while holding mutex
   MutexLock lock(mutex_.get(key));
 
-  handle = cache_->Lookup(key);
+  handle = cache_.Lookup(key);
   if (handle) {
-    *blob_file_reader = CacheHandleGuard<BlobFileReader>(cache_, handle);
+    *blob_file_reader = cache_.Guard(handle);
     return Status::OK();
   }
 
@@ -84,8 +84,7 @@ Status BlobFileCache::GetBlobFileReader(
   {
     constexpr size_t charge = 1;
 
-    const Status s = cache_->Insert(key, reader.get(), charge,
-                                    &DeleteCacheEntry<BlobFileReader>, &handle);
+    const Status s = cache_.Insert(key, reader.get(), charge, &handle);
     if (!s.ok()) {
       RecordTick(statistics, NO_FILE_ERRORS);
       return s;
@@ -94,7 +93,7 @@ Status BlobFileCache::GetBlobFileReader(
 
   reader.release();
 
-  *blob_file_reader = CacheHandleGuard<BlobFileReader>(cache_, handle);
+  *blob_file_reader = cache_.Guard(handle);
 
   return Status::OK();
 }

db/blob/blob_file_cache.h

@@ -7,7 +7,8 @@
 
 #include <cinttypes>
 
-#include "cache/cache_helpers.h"
+#include "cache/typed_cache.h"
+#include "db/blob/blob_file_reader.h"
 #include "rocksdb/rocksdb_namespace.h"
 #include "util/mutexlock.h"
 
@@ -18,7 +19,6 @@ struct ImmutableOptions;
 struct FileOptions;
 class HistogramImpl;
 class Status;
-class BlobFileReader;
 class Slice;
 class IOTracer;
 
@@ -36,7 +36,10 @@ class BlobFileCache {
                            CacheHandleGuard<BlobFileReader>* blob_file_reader);
 
  private:
-  Cache* cache_;
+  using CacheInterface =
+      BasicTypedCacheInterface<BlobFileReader, CacheEntryRole::kMisc>;
+  using TypedHandle = CacheInterface::TypedHandle;
+  CacheInterface cache_;
   // Note: mutex_ below is used to guard against multiple threads racing to open
   // the same file.
   Striped<port::Mutex, Slice> mutex_;

db/blob/blob_file_cache_test.cc

@@ -263,6 +263,7 @@ TEST_F(BlobFileCacheTest, GetBlobFileReader_CacheFull) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_file_garbage_test.cc

@@ -168,6 +168,7 @@ TEST_F(BlobFileGarbageTest, ForwardIncompatibleCustomField) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_file_reader.cc

@@ -8,6 +8,7 @@
 #include <cassert>
 #include <string>
 
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_log_format.h"
 #include "file/file_prefetch_buffer.h"
 #include "file/filename.h"
@@ -283,14 +284,12 @@ BlobFileReader::BlobFileReader(
 
 BlobFileReader::~BlobFileReader() = default;
 
-Status BlobFileReader::GetBlob(const ReadOptions& read_options,
-                               const Slice& user_key, uint64_t offset,
-                               uint64_t value_size,
-                               CompressionType compression_type,
-                               FilePrefetchBuffer* prefetch_buffer,
-                               PinnableSlice* value,
-                               uint64_t* bytes_read) const {
-  assert(value);
+Status BlobFileReader::GetBlob(
+    const ReadOptions& read_options, const Slice& user_key, uint64_t offset,
+    uint64_t value_size, CompressionType compression_type,
+    FilePrefetchBuffer* prefetch_buffer, MemoryAllocator* allocator,
+    std::unique_ptr<BlobContents>* result, uint64_t* bytes_read) const {
+  assert(result);
 
   const uint64_t key_size = user_key.size();
 
@@ -361,8 +360,8 @@ Status BlobFileReader::GetBlob(const ReadOptions& read_options,
   const Slice value_slice(record_slice.data() + adjustment, value_size);
 
   {
-    const Status s = UncompressBlobIfNeeded(value_slice, compression_type,
-                                            clock_, statistics_, value);
+    const Status s = UncompressBlobIfNeeded(
+        value_slice, compression_type, allocator, clock_, statistics_, result);
     if (!s.ok()) {
       return s;
     }
@@ -375,16 +374,18 @@ Status BlobFileReader::GetBlob(const ReadOptions& read_options,
   return Status::OK();
 }
 
-void BlobFileReader::MultiGetBlob(const ReadOptions& read_options,
-                                  autovector<BlobReadRequest*>& blob_reqs,
-                                  uint64_t* bytes_read) const {
+void BlobFileReader::MultiGetBlob(
+    const ReadOptions& read_options, MemoryAllocator* allocator,
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>&
+        blob_reqs,
+    uint64_t* bytes_read) const {
   const size_t num_blobs = blob_reqs.size();
   assert(num_blobs > 0);
   assert(num_blobs <= MultiGetContext::MAX_BATCH_SIZE);
 
 #ifndef NDEBUG
   for (size_t i = 0; i < num_blobs - 1; ++i) {
-    assert(blob_reqs[i]->offset <= blob_reqs[i + 1]->offset);
+    assert(blob_reqs[i].first->offset <= blob_reqs[i + 1].first->offset);
   }
 #endif  // !NDEBUG
 
@@ -393,16 +394,21 @@ void BlobFileReader::MultiGetBlob(const ReadOptions& read_options,
   uint64_t total_len = 0;
   read_reqs.reserve(num_blobs);
   for (size_t i = 0; i < num_blobs; ++i) {
-    const size_t key_size = blob_reqs[i]->user_key->size();
-    const uint64_t offset = blob_reqs[i]->offset;
-    const uint64_t value_size = blob_reqs[i]->len;
+    BlobReadRequest* const req = blob_reqs[i].first;
+    assert(req);
+    assert(req->user_key);
+    assert(req->status);
+
+    const size_t key_size = req->user_key->size();
+    const uint64_t offset = req->offset;
+    const uint64_t value_size = req->len;
 
     if (!IsValidBlobOffset(offset, key_size, value_size, file_size_)) {
-      *blob_reqs[i]->status = Status::Corruption("Invalid blob offset");
+      *req->status = Status::Corruption("Invalid blob offset");
       continue;
     }
-    if (blob_reqs[i]->compression != compression_type_) {
-      *blob_reqs[i]->status =
+    if (req->compression != compression_type_) {
+      *req->status =
           Status::Corruption("Compression type mismatch when reading a blob");
       continue;
     }
@@ -411,12 +417,12 @@ void BlobFileReader::MultiGetBlob(const ReadOptions& read_options,
         read_options.verify_checksums
             ? BlobLogRecord::CalculateAdjustmentForRecordHeader(key_size)
             : 0;
-    assert(blob_reqs[i]->offset >= adjustment);
+    assert(req->offset >= adjustment);
     adjustments.push_back(adjustment);
 
     FSReadRequest read_req = {};
-    read_req.offset = blob_reqs[i]->offset - adjustment;
-    read_req.len = blob_reqs[i]->len + adjustment;
+    read_req.offset = req->offset - adjustment;
+    read_req.len = req->len + adjustment;
     read_reqs.emplace_back(read_req);
     total_len += read_req.len;
   }
@@ -450,8 +456,11 @@ void BlobFileReader::MultiGetBlob(const ReadOptions& read_options,
     for (auto& req : read_reqs) {
       req.status.PermitUncheckedError();
     }
-    for (auto& req : blob_reqs) {
+    for (auto& blob_req : blob_reqs) {
+      BlobReadRequest* const req = blob_req.first;
+      assert(req);
       assert(req->status);
+
       if (!req->status->IsCorruption()) {
         // Avoid overwriting corruption status.
         *req->status = s;
@@ -464,38 +473,42 @@ void BlobFileReader::MultiGetBlob(const ReadOptions& read_options,
 
   uint64_t total_bytes = 0;
   for (size_t i = 0, j = 0; i < num_blobs; ++i) {
-    assert(blob_reqs[i]->status);
-    if (!blob_reqs[i]->status->ok()) {
+    BlobReadRequest* const req = blob_reqs[i].first;
+    assert(req);
+    assert(req->user_key);
+    assert(req->status);
+
+    if (!req->status->ok()) {
       continue;
     }
 
     assert(j < read_reqs.size());
-    auto& req = read_reqs[j++];
-    const auto& record_slice = req.result;
-    if (req.status.ok() && record_slice.size() != req.len) {
-      req.status = IOStatus::Corruption("Failed to read data from blob file");
+    auto& read_req = read_reqs[j++];
+    const auto& record_slice = read_req.result;
+    if (read_req.status.ok() && record_slice.size() != read_req.len) {
+      read_req.status =
+          IOStatus::Corruption("Failed to read data from blob file");
     }
 
-    *blob_reqs[i]->status = req.status;
-    if (!blob_reqs[i]->status->ok()) {
+    *req->status = read_req.status;
+    if (!req->status->ok()) {
       continue;
     }
 
     // Verify checksums if enabled
     if (read_options.verify_checksums) {
-      *blob_reqs[i]->status =
-          VerifyBlob(record_slice, *blob_reqs[i]->user_key, blob_reqs[i]->len);
-      if (!blob_reqs[i]->status->ok()) {
+      *req->status = VerifyBlob(record_slice, *req->user_key, req->len);
+      if (!req->status->ok()) {
         continue;
       }
     }
 
     // Uncompress blob if needed
-    Slice value_slice(record_slice.data() + adjustments[i], blob_reqs[i]->len);
-    *blob_reqs[i]->status =
-        UncompressBlobIfNeeded(value_slice, compression_type_, clock_,
-                               statistics_, blob_reqs[i]->result);
-    if (blob_reqs[i]->status->ok()) {
+    Slice value_slice(record_slice.data() + adjustments[i], req->len);
+    *req->status =
+        UncompressBlobIfNeeded(value_slice, compression_type_, allocator,
+                               clock_, statistics_, &blob_reqs[i].second);
+    if (req->status->ok()) {
       total_bytes += record_slice.size();
     }
   }
@@ -549,16 +562,14 @@ Status BlobFileReader::VerifyBlob(const Slice& record_slice,
   return Status::OK();
 }
 
-Status BlobFileReader::UncompressBlobIfNeeded(const Slice& value_slice,
-                                              CompressionType compression_type,
-                                              SystemClock* clock,
-                                              Statistics* statistics,
-                                              PinnableSlice* value) {
-  assert(value);
+Status BlobFileReader::UncompressBlobIfNeeded(
+    const Slice& value_slice, CompressionType compression_type,
+    MemoryAllocator* allocator, SystemClock* clock, Statistics* statistics,
+    std::unique_ptr<BlobContents>* result) {
+  assert(result);
 
   if (compression_type == kNoCompression) {
-    SaveValue(value_slice, value);
-
+    BlobContentsCreator::Create(result, nullptr, value_slice, allocator);
     return Status::OK();
   }
 
@@ -568,7 +579,6 @@ Status BlobFileReader::UncompressBlobIfNeeded(const Slice& value_slice,
 
   size_t uncompressed_size = 0;
   constexpr uint32_t compression_format_version = 2;
-  constexpr MemoryAllocator* allocator = nullptr;
 
   CacheAllocationPtr output;
 
@@ -587,19 +597,9 @@ Status BlobFileReader::UncompressBlobIfNeeded(const Slice& value_slice,
     return Status::Corruption("Unable to uncompress blob");
   }
 
-  SaveValue(Slice(output.get(), uncompressed_size), value);
+  result->reset(new BlobContents(std::move(output), uncompressed_size));
 
   return Status::OK();
 }
 
-void BlobFileReader::SaveValue(const Slice& src, PinnableSlice* dst) {
-  assert(dst);
-
-  if (dst->IsPinned()) {
-    dst->Reset();
-  }
-
-  dst->PinSelf(src);
-}
-
 }  // namespace ROCKSDB_NAMESPACE

db/blob/blob_file_reader.h

@@ -23,7 +23,7 @@ class HistogramImpl;
 struct ReadOptions;
 class Slice;
 class FilePrefetchBuffer;
-class PinnableSlice;
+class BlobContents;
 class Statistics;
 
 class BlobFileReader {
@@ -44,13 +44,17 @@ class BlobFileReader {
   Status GetBlob(const ReadOptions& read_options, const Slice& user_key,
                  uint64_t offset, uint64_t value_size,
                  CompressionType compression_type,
-                 FilePrefetchBuffer* prefetch_buffer, PinnableSlice* value,
+                 FilePrefetchBuffer* prefetch_buffer,
+                 MemoryAllocator* allocator,
+                 std::unique_ptr<BlobContents>* result,
                  uint64_t* bytes_read) const;
 
   // offsets must be sorted in ascending order by caller.
-  void MultiGetBlob(const ReadOptions& read_options,
-                    autovector<BlobReadRequest*>& blob_reqs,
-                    uint64_t* bytes_read) const;
+  void MultiGetBlob(
+      const ReadOptions& read_options, MemoryAllocator* allocator,
+      autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>&
+          blob_reqs,
+      uint64_t* bytes_read) const;
 
   CompressionType GetCompressionType() const { return compression_type_; }
 
@@ -89,11 +93,10 @@ class BlobFileReader {
 
   static Status UncompressBlobIfNeeded(const Slice& value_slice,
                                        CompressionType compression_type,
+                                       MemoryAllocator* allocator,
                                        SystemClock* clock,
                                        Statistics* statistics,
-                                       PinnableSlice* value);
-
-  static void SaveValue(const Slice& src, PinnableSlice* dst);
+                                       std::unique_ptr<BlobContents>* result);
 
   std::unique_ptr<RandomAccessFileReader> file_reader_;
   uint64_t file_size_;

db/blob/blob_file_reader_test.cc

@@ -8,6 +8,7 @@
 #include <cassert>
 #include <string>
 
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_log_format.h"
 #include "db/blob/blob_log_writer.h"
 #include "env/mock_env.h"
@@ -180,15 +181,17 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
   read_options.verify_checksums = false;
 
   constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
+  constexpr MemoryAllocator* allocator = nullptr;
 
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_OK(reader->GetBlob(read_options, keys[0], blob_offsets[0],
                               blob_sizes[0], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read));
-    ASSERT_EQ(value, blobs[0]);
+                              allocator, &value, &bytes_read));
+    ASSERT_NE(value, nullptr);
+    ASSERT_EQ(value->data(), blobs[0]);
     ASSERT_EQ(bytes_read, blob_sizes[0]);
 
     // MultiGetBlob
@@ -196,22 +199,25 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
     size_t total_size = 0;
 
     std::array<Status, num_blobs> statuses_buf;
-    std::array<PinnableSlice, num_blobs> value_buf;
     std::array<BlobReadRequest, num_blobs> requests_buf;
-    autovector<BlobReadRequest*> blob_reqs;
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>
+        blob_reqs;
 
     for (size_t i = 0; i < num_blobs; ++i) {
       requests_buf[i] =
           BlobReadRequest(keys[i], blob_offsets[i], blob_sizes[i],
-                          kNoCompression, &value_buf[i], &statuses_buf[i]);
-      blob_reqs.push_back(&requests_buf[i]);
+                          kNoCompression, nullptr, &statuses_buf[i]);
+      blob_reqs.emplace_back(&requests_buf[i], std::unique_ptr<BlobContents>());
     }
 
-    reader->MultiGetBlob(read_options, blob_reqs, &bytes_read);
+    reader->MultiGetBlob(read_options, allocator, blob_reqs, &bytes_read);
 
     for (size_t i = 0; i < num_blobs; ++i) {
+      const auto& result = blob_reqs[i].second;
+
       ASSERT_OK(statuses_buf[i]);
-      ASSERT_EQ(value_buf[i], blobs[i]);
+      ASSERT_NE(result, nullptr);
+      ASSERT_EQ(result->data(), blobs[i]);
       total_size += blob_sizes[i];
     }
     ASSERT_EQ(bytes_read, total_size);
@@ -220,13 +226,14 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
   read_options.verify_checksums = true;
 
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_OK(reader->GetBlob(read_options, keys[1], blob_offsets[1],
                               blob_sizes[1], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read));
-    ASSERT_EQ(value, blobs[1]);
+                              allocator, &value, &bytes_read));
+    ASSERT_NE(value, nullptr);
+    ASSERT_EQ(value->data(), blobs[1]);
 
     const uint64_t key_size = keys[1].size();
     ASSERT_EQ(bytes_read,
@@ -236,47 +243,50 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
 
   // Invalid offset (too close to start of file)
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(read_options, keys[0], blob_offsets[0] - 1,
                               blob_sizes[0], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read)
+                              allocator, &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
   }
 
   // Invalid offset (too close to end of file)
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(read_options, keys[2], blob_offsets[2] + 1,
                               blob_sizes[2], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read)
+                              allocator, &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
   }
 
   // Incorrect compression type
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(read_options, keys[0], blob_offsets[0],
-                              blob_sizes[0], kZSTD, prefetch_buffer, &value,
-                              &bytes_read)
+                              blob_sizes[0], kZSTD, prefetch_buffer, allocator,
+                              &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
   }
 
   // Incorrect key size
   {
     constexpr char shorter_key[] = "k";
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
@@ -284,8 +294,9 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
                               blob_offsets[0] -
                                   (keys[0].size() - sizeof(shorter_key) + 1),
                               blob_sizes[0], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read)
+                              allocator, &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
 
     // MultiGetBlob
@@ -302,18 +313,18 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
         blob_offsets[2]};
 
     std::array<Status, num_blobs> statuses_buf;
-    std::array<PinnableSlice, num_blobs> value_buf;
     std::array<BlobReadRequest, num_blobs> requests_buf;
-    autovector<BlobReadRequest*> blob_reqs;
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>
+        blob_reqs;
 
     for (size_t i = 0; i < num_blobs; ++i) {
       requests_buf[i] =
           BlobReadRequest(key_refs[i], offsets[i], blob_sizes[i],
-                          kNoCompression, &value_buf[i], &statuses_buf[i]);
-      blob_reqs.push_back(&requests_buf[i]);
+                          kNoCompression, nullptr, &statuses_buf[i]);
+      blob_reqs.emplace_back(&requests_buf[i], std::unique_ptr<BlobContents>());
     }
 
-    reader->MultiGetBlob(read_options, blob_reqs, &bytes_read);
+    reader->MultiGetBlob(read_options, allocator, blob_reqs, &bytes_read);
 
     for (size_t i = 0; i < num_blobs; ++i) {
       if (i == 1) {
@@ -327,14 +338,15 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
   // Incorrect key
   {
     constexpr char incorrect_key[] = "foo1";
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(read_options, incorrect_key, blob_offsets[0],
                               blob_sizes[0], kNoCompression, prefetch_buffer,
-                              &value, &bytes_read)
+                              allocator, &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
 
     // MultiGetBlob
@@ -346,19 +358,18 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
     key_refs[2] = std::cref(wrong_key_slice);
 
     std::array<Status, num_blobs> statuses_buf;
-    std::array<PinnableSlice, num_blobs> value_buf;
     std::array<BlobReadRequest, num_blobs> requests_buf;
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>
+        blob_reqs;
 
     for (size_t i = 0; i < num_blobs; ++i) {
       requests_buf[i] =
           BlobReadRequest(key_refs[i], blob_offsets[i], blob_sizes[i],
-                          kNoCompression, &value_buf[i], &statuses_buf[i]);
+                          kNoCompression, nullptr, &statuses_buf[i]);
+      blob_reqs.emplace_back(&requests_buf[i], std::unique_ptr<BlobContents>());
     }
 
-    autovector<BlobReadRequest*> blob_reqs = {
-        &requests_buf[0], &requests_buf[1], &requests_buf[2]};
-
-    reader->MultiGetBlob(read_options, blob_reqs, &bytes_read);
+    reader->MultiGetBlob(read_options, allocator, blob_reqs, &bytes_read);
 
     for (size_t i = 0; i < num_blobs; ++i) {
       if (i == num_blobs - 1) {
@@ -371,14 +382,15 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
 
   // Incorrect value size
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(read_options, keys[1], blob_offsets[1],
                               blob_sizes[1] + 1, kNoCompression,
-                              prefetch_buffer, &value, &bytes_read)
+                              prefetch_buffer, allocator, &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
 
     // MultiGetBlob
@@ -388,23 +400,26 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
     }
 
     std::array<Status, num_blobs> statuses_buf;
-    std::array<PinnableSlice, num_blobs> value_buf;
     std::array<BlobReadRequest, num_blobs> requests_buf;
 
     requests_buf[0] =
         BlobReadRequest(key_refs[0], blob_offsets[0], blob_sizes[0],
-                        kNoCompression, &value_buf[0], &statuses_buf[0]);
+                        kNoCompression, nullptr, &statuses_buf[0]);
     requests_buf[1] =
         BlobReadRequest(key_refs[1], blob_offsets[1], blob_sizes[1] + 1,
-                        kNoCompression, &value_buf[1], &statuses_buf[1]);
+                        kNoCompression, nullptr, &statuses_buf[1]);
     requests_buf[2] =
         BlobReadRequest(key_refs[2], blob_offsets[2], blob_sizes[2],
-                        kNoCompression, &value_buf[2], &statuses_buf[2]);
+                        kNoCompression, nullptr, &statuses_buf[2]);
 
-    autovector<BlobReadRequest*> blob_reqs = {
-        &requests_buf[0], &requests_buf[1], &requests_buf[2]};
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>
+        blob_reqs;
 
-    reader->MultiGetBlob(read_options, blob_reqs, &bytes_read);
+    for (size_t i = 0; i < num_blobs; ++i) {
+      blob_reqs.emplace_back(&requests_buf[i], std::unique_ptr<BlobContents>());
+    }
+
+    reader->MultiGetBlob(read_options, allocator, blob_reqs, &bytes_read);
 
     for (size_t i = 0; i < num_blobs; ++i) {
       if (i != 1) {
@@ -665,14 +680,17 @@ TEST_F(BlobFileReaderTest, BlobCRCError) {
   SyncPoint::GetInstance()->EnableProcessing();
 
   constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
-  PinnableSlice value;
+  constexpr MemoryAllocator* allocator = nullptr;
+
+  std::unique_ptr<BlobContents> value;
   uint64_t bytes_read = 0;
 
   ASSERT_TRUE(reader
                   ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
-                            kNoCompression, prefetch_buffer, &value,
+                            kNoCompression, prefetch_buffer, allocator, &value,
                             &bytes_read)
                   .IsCorruption());
+  ASSERT_EQ(value, nullptr);
   ASSERT_EQ(bytes_read, 0);
 
   SyncPoint::GetInstance()->DisableProcessing();
@@ -720,28 +738,31 @@ TEST_F(BlobFileReaderTest, Compression) {
   read_options.verify_checksums = false;
 
   constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
+  constexpr MemoryAllocator* allocator = nullptr;
 
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_OK(reader->GetBlob(read_options, key, blob_offset, blob_size,
-                              kSnappyCompression, prefetch_buffer, &value,
-                              &bytes_read));
-    ASSERT_EQ(value, blob);
+                              kSnappyCompression, prefetch_buffer, allocator,
+                              &value, &bytes_read));
+    ASSERT_NE(value, nullptr);
+    ASSERT_EQ(value->data(), blob);
     ASSERT_EQ(bytes_read, blob_size);
   }
 
   read_options.verify_checksums = true;
 
   {
-    PinnableSlice value;
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_OK(reader->GetBlob(read_options, key, blob_offset, blob_size,
-                              kSnappyCompression, prefetch_buffer, &value,
-                              &bytes_read));
-    ASSERT_EQ(value, blob);
+                              kSnappyCompression, prefetch_buffer, allocator,
+                              &value, &bytes_read));
+    ASSERT_NE(value, nullptr);
+    ASSERT_EQ(value->data(), blob);
 
     constexpr uint64_t key_size = sizeof(key) - 1;
     ASSERT_EQ(bytes_read,
@@ -799,14 +820,17 @@ TEST_F(BlobFileReaderTest, UncompressionError) {
   SyncPoint::GetInstance()->EnableProcessing();
 
   constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
-  PinnableSlice value;
+  constexpr MemoryAllocator* allocator = nullptr;
+
+  std::unique_ptr<BlobContents> value;
   uint64_t bytes_read = 0;
 
   ASSERT_TRUE(reader
                   ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
-                            kSnappyCompression, prefetch_buffer, &value,
-                            &bytes_read)
+                            kSnappyCompression, prefetch_buffer, allocator,
+                            &value, &bytes_read)
                   .IsCorruption());
+  ASSERT_EQ(value, nullptr);
   ASSERT_EQ(bytes_read, 0);
 
   SyncPoint::GetInstance()->DisableProcessing();
@@ -885,14 +909,17 @@ TEST_P(BlobFileReaderIOErrorTest, IOError) {
     ASSERT_OK(s);
 
     constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
-    PinnableSlice value;
+    constexpr MemoryAllocator* allocator = nullptr;
+
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
-                              kNoCompression, prefetch_buffer, &value,
-                              &bytes_read)
+                              kNoCompression, prefetch_buffer, allocator,
+                              &value, &bytes_read)
                     .IsIOError());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
   }
 
@@ -970,14 +997,17 @@ TEST_P(BlobFileReaderDecodingErrorTest, DecodingError) {
     ASSERT_OK(s);
 
     constexpr FilePrefetchBuffer* prefetch_buffer = nullptr;
-    PinnableSlice value;
+    constexpr MemoryAllocator* allocator = nullptr;
+
+    std::unique_ptr<BlobContents> value;
     uint64_t bytes_read = 0;
 
     ASSERT_TRUE(reader
                     ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
-                              kNoCompression, prefetch_buffer, &value,
-                              &bytes_read)
+                              kNoCompression, prefetch_buffer, allocator,
+                              &value, &bytes_read)
                     .IsCorruption());
+    ASSERT_EQ(value, nullptr);
     ASSERT_EQ(bytes_read, 0);
   }
 
@@ -988,6 +1018,7 @@ TEST_P(BlobFileReaderDecodingErrorTest, DecodingError) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_garbage_meter_test.cc

@@ -191,6 +191,7 @@ TEST(BlobGarbageMeterTest, InlinedTTLBlobIndex) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/blob_log_format.h

@@ -22,6 +22,8 @@ constexpr uint32_t kVersion1 = 1;
 
 using ExpirationRange = std::pair<uint64_t, uint64_t>;
 
+// clang-format off
+
 // Format of blob log file header (30 bytes):
 //
 //    +--------------+---------+---------+-------+-------------+-------------------+
@@ -35,6 +37,9 @@ using ExpirationRange = std::pair<uint64_t, uint64_t>;
 //
 // Expiration range in the header is a rough range based on
 // blob_db_options.ttl_range_secs.
+
+// clang-format on
+
 struct BlobLogHeader {
   static constexpr size_t kSize = 30;
 
@@ -57,6 +62,8 @@ struct BlobLogHeader {
   Status DecodeFrom(Slice slice);
 };
 
+// clang-format off
+
 // Format of blob log file footer (32 bytes):
 //
 //    +--------------+------------+-------------------+------------+
@@ -69,6 +76,9 @@ struct BlobLogHeader {
 //
 // Unlike the same field in file header, expiration range in the footer is the
 // range of smallest and largest expiration of the data in this file.
+
+// clang-format on
+
 struct BlobLogFooter {
   static constexpr size_t kSize = 32;
 
@@ -81,6 +91,8 @@ struct BlobLogFooter {
   Status DecodeFrom(Slice slice);
 };
 
+// clang-format off
+
 // Blob record format (32 bytes header + key + value):
 //
 //    +------------+--------------+------------+------------+----------+---------+-----------+
@@ -100,6 +112,9 @@ struct BlobLogFooter {
 //
 // We could use variable length encoding (Varint64) to save more space, but it
 // make reader more complicated.
+
+// clang-format on
+
 struct BlobLogRecord {
   // header include fields up to blob CRC
   static constexpr size_t kHeaderSize = 32;

db/blob/blob_source.cc

@@ -10,6 +10,7 @@
 
 #include "cache/cache_reservation_manager.h"
 #include "cache/charged_cache.h"
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_file_reader.h"
 #include "db/blob/blob_log_format.h"
 #include "monitoring/statistics.h"
@@ -35,115 +36,68 @@ BlobSource::BlobSource(const ImmutableOptions* immutable_options,
   if (bbto &&
       bbto->cache_usage_options.options_overrides.at(CacheEntryRole::kBlobCache)
               .charged == CacheEntryRoleOptions::Decision::kEnabled) {
-    blob_cache_ = std::make_shared<ChargedCache>(immutable_options->blob_cache,
-                                                 bbto->block_cache);
+    blob_cache_ = SharedCacheInterface{std::make_shared<ChargedCache>(
+        immutable_options->blob_cache, bbto->block_cache)};
   }
 #endif  // ROCKSDB_LITE
 }
 
 BlobSource::~BlobSource() = default;
 
-Status BlobSource::GetBlobFromCache(const Slice& cache_key,
-                                    CacheHandleGuard<std::string>* blob) const {
-  assert(blob);
-  assert(blob->IsEmpty());
+Status BlobSource::GetBlobFromCache(
+    const Slice& cache_key, CacheHandleGuard<BlobContents>* cached_blob) const {
   assert(blob_cache_);
   assert(!cache_key.empty());
+  assert(cached_blob);
+  assert(cached_blob->IsEmpty());
 
   Cache::Handle* cache_handle = nullptr;
   cache_handle = GetEntryFromCache(cache_key);
   if (cache_handle != nullptr) {
-    *blob = CacheHandleGuard<std::string>(blob_cache_.get(), cache_handle);
+    *cached_blob =
+        CacheHandleGuard<BlobContents>(blob_cache_.get(), cache_handle);
+
+    assert(cached_blob->GetValue());
+
+    PERF_COUNTER_ADD(blob_cache_hit_count, 1);
+    RecordTick(statistics_, BLOB_DB_CACHE_HIT);
+    RecordTick(statistics_, BLOB_DB_CACHE_BYTES_READ,
+               cached_blob->GetValue()->size());
+
     return Status::OK();
   }
 
-  assert(blob->IsEmpty());
+  RecordTick(statistics_, BLOB_DB_CACHE_MISS);
 
   return Status::NotFound("Blob not found in cache");
 }
 
-Status BlobSource::PutBlobIntoCache(const Slice& cache_key,
-                                    CacheHandleGuard<std::string>* cached_blob,
-                                    PinnableSlice* blob) const {
-  assert(blob);
-  assert(!cache_key.empty());
+Status BlobSource::PutBlobIntoCache(
+    const Slice& cache_key, std::unique_ptr<BlobContents>* blob,
+    CacheHandleGuard<BlobContents>* cached_blob) const {
   assert(blob_cache_);
+  assert(!cache_key.empty());
+  assert(blob);
+  assert(*blob);
+  assert(cached_blob);
+  assert(cached_blob->IsEmpty());
 
-  Status s;
-  const Cache::Priority priority = Cache::Priority::BOTTOM;
-
-  // Objects to be put into the cache have to be heap-allocated and
-  // self-contained, i.e. own their contents. The Cache has to be able to take
-  // unique ownership of them. Therefore, we copy the blob into a string
-  // directly, and insert that into the cache.
-  std::unique_ptr<std::string> buf = std::make_unique<std::string>();
-  buf->assign(blob->data(), blob->size());
-
-  // TODO: support custom allocators and provide a better estimated memory
-  // usage using malloc_usable_size.
-  Cache::Handle* cache_handle = nullptr;
-  s = InsertEntryIntoCache(cache_key, buf.get(), buf->size(), &cache_handle,
-                           priority);
+  TypedHandle* cache_handle = nullptr;
+  const Status s = InsertEntryIntoCache(cache_key, blob->get(),
+                                        &cache_handle, Cache::Priority::BOTTOM);
   if (s.ok()) {
-    buf.release();
+    blob->release();
+
     assert(cache_handle != nullptr);
     *cached_blob =
-        CacheHandleGuard<std::string>(blob_cache_.get(), cache_handle);
-  }
+        CacheHandleGuard<BlobContents>(blob_cache_.get(), cache_handle);
 
-  return s;
-}
+    assert(cached_blob->GetValue());
 
-Cache::Handle* BlobSource::GetEntryFromCache(const Slice& key) const {
-  Cache::Handle* cache_handle = nullptr;
-
-  if (lowest_used_cache_tier_ == CacheTier::kNonVolatileBlockTier) {
-    Cache::CreateCallback create_cb = [&](const void* buf, size_t size,
-                                          void** out_obj,
-                                          size_t* charge) -> Status {
-      std::string* blob = new std::string();
-      blob->assign(static_cast<const char*>(buf), size);
-      *out_obj = blob;
-      *charge = size;
-      return Status::OK();
-    };
-    cache_handle = blob_cache_->Lookup(key, GetCacheItemHelper(), create_cb,
-                                       Cache::Priority::BOTTOM,
-                                       true /* wait_for_cache */, statistics_);
-  } else {
-    cache_handle = blob_cache_->Lookup(key, statistics_);
-  }
-
-  if (cache_handle != nullptr) {
-    PERF_COUNTER_ADD(blob_cache_hit_count, 1);
-    RecordTick(statistics_, BLOB_DB_CACHE_HIT);
-    RecordTick(statistics_, BLOB_DB_CACHE_BYTES_READ,
-               blob_cache_->GetUsage(cache_handle));
-  } else {
-    RecordTick(statistics_, BLOB_DB_CACHE_MISS);
-  }
-  return cache_handle;
-}
-
-Status BlobSource::InsertEntryIntoCache(const Slice& key, std::string* value,
-                                        size_t charge,
-                                        Cache::Handle** cache_handle,
-                                        Cache::Priority priority) const {
-  Status s;
-
-  if (lowest_used_cache_tier_ == CacheTier::kNonVolatileBlockTier) {
-    s = blob_cache_->Insert(key, value, GetCacheItemHelper(), charge,
-                            cache_handle, priority);
-  } else {
-    s = blob_cache_->Insert(key, value, charge, &DeleteCacheEntry<std::string>,
-                            cache_handle, priority);
-  }
-
-  if (s.ok()) {
-    assert(*cache_handle != nullptr);
     RecordTick(statistics_, BLOB_DB_CACHE_ADD);
     RecordTick(statistics_, BLOB_DB_CACHE_BYTES_WRITE,
-               blob_cache_->GetUsage(*cache_handle));
+               cached_blob->GetValue()->size());
+
   } else {
     RecordTick(statistics_, BLOB_DB_CACHE_ADD_FAILURES);
   }
@@ -151,6 +105,57 @@ Status BlobSource::InsertEntryIntoCache(const Slice& key, std::string* value,
   return s;
 }
 
+BlobSource::TypedHandle* BlobSource::GetEntryFromCache(const Slice& key) const {
+  return blob_cache_.LookupFull(
+      key, nullptr /* context */, Cache::Priority::BOTTOM,
+      true /* wait_for_cache */, statistics_, lowest_used_cache_tier_);
+}
+
+void BlobSource::PinCachedBlob(CacheHandleGuard<BlobContents>* cached_blob,
+                               PinnableSlice* value) {
+  assert(cached_blob);
+  assert(cached_blob->GetValue());
+  assert(value);
+
+  // To avoid copying the cached blob into the buffer provided by the
+  // application, we can simply transfer ownership of the cache handle to
+  // the target PinnableSlice. This has the potential to save a lot of
+  // CPU, especially with large blob values.
+
+  value->Reset();
+
+  constexpr Cleanable* cleanable = nullptr;
+  value->PinSlice(cached_blob->GetValue()->data(), cleanable);
+
+  cached_blob->TransferTo(value);
+}
+
+void BlobSource::PinOwnedBlob(std::unique_ptr<BlobContents>* owned_blob,
+                              PinnableSlice* value) {
+  assert(owned_blob);
+  assert(*owned_blob);
+  assert(value);
+
+  BlobContents* const blob = owned_blob->release();
+  assert(blob);
+
+  value->Reset();
+  value->PinSlice(
+      blob->data(),
+      [](void* arg1, void* /* arg2 */) {
+        delete static_cast<BlobContents*>(arg1);
+      },
+      blob, nullptr);
+}
+
+Status BlobSource::InsertEntryIntoCache(const Slice& key, BlobContents* value,
+                                        TypedHandle** cache_handle,
+                                        Cache::Priority priority) const {
+  return blob_cache_.InsertFull(key, value, value->ApproximateMemoryUsage(),
+                                cache_handle, priority,
+                                lowest_used_cache_tier_);
+}
+
 Status BlobSource::GetBlob(const ReadOptions& read_options,
                            const Slice& user_key, uint64_t file_number,
                            uint64_t offset, uint64_t file_size,
@@ -164,7 +169,7 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
 
   const CacheKey cache_key = GetCacheKey(file_number, file_size, offset);
 
-  CacheHandleGuard<std::string> blob_handle;
+  CacheHandleGuard<BlobContents> blob_handle;
 
   // First, try to get the blob from the cache
   //
@@ -172,24 +177,8 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
   if (blob_cache_) {
     Slice key = cache_key.AsSlice();
     s = GetBlobFromCache(key, &blob_handle);
-    if (s.ok() && blob_handle.GetValue()) {
-      {
-        value->Reset();
-        // To avoid copying the cached blob into the buffer provided by the
-        // application, we can simply transfer ownership of the cache handle to
-        // the target PinnableSlice. This has the potential to save a lot of
-        // CPU, especially with large blob values.
-        value->PinSlice(
-            *blob_handle.GetValue(),
-            [](void* arg1, void* arg2) {
-              Cache* const cache = static_cast<Cache*>(arg1);
-              Cache::Handle* const handle = static_cast<Cache::Handle*>(arg2);
-              cache->Release(handle);
-            },
-            blob_handle.GetCache(), blob_handle.GetCacheHandle());
-        // Make the CacheHandleGuard relinquish ownership of the handle.
-        blob_handle.TransferTo(nullptr);
-      }
+    if (s.ok()) {
+      PinCachedBlob(&blob_handle, value);
 
       // For consistency, the size of on-disk (possibly compressed) blob record
       // is assigned to bytes_read.
@@ -218,6 +207,8 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
 
   // Can't find the blob from the cache. Since I/O is allowed, read from the
   // file.
+  std::unique_ptr<BlobContents> blob_contents;
+
   {
     CacheHandleGuard<BlobFileReader> blob_file_reader;
     s = blob_file_cache_->GetBlobFileReader(file_number, &blob_file_reader);
@@ -231,10 +222,15 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
       return Status::Corruption("Compression type mismatch when reading blob");
     }
 
+    MemoryAllocator* const allocator =
+        (blob_cache_ && read_options.fill_cache)
+            ? blob_cache_.get()->memory_allocator()
+            : nullptr;
+
     uint64_t read_size = 0;
     s = blob_file_reader.GetValue()->GetBlob(
         read_options, user_key, offset, value_size, compression_type,
-        prefetch_buffer, value, &read_size);
+        prefetch_buffer, allocator, &blob_contents, &read_size);
     if (!s.ok()) {
       return s;
     }
@@ -247,10 +243,14 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
     // If filling cache is allowed and a cache is configured, try to put the
     // blob to the cache.
     Slice key = cache_key.AsSlice();
-    s = PutBlobIntoCache(key, &blob_handle, value);
+    s = PutBlobIntoCache(key, &blob_contents, &blob_handle);
     if (!s.ok()) {
       return s;
     }
+
+    PinCachedBlob(&blob_handle, value);
+  } else {
+    PinOwnedBlob(&blob_contents, value);
   }
 
   assert(s.ok());
@@ -310,33 +310,17 @@ void BlobSource::MultiGetBlobFromOneFile(const ReadOptions& read_options,
     for (size_t i = 0; i < num_blobs; ++i) {
       auto& req = blob_reqs[i];
 
-      CacheHandleGuard<std::string> blob_handle;
+      CacheHandleGuard<BlobContents> blob_handle;
       const CacheKey cache_key = base_cache_key.WithOffset(req.offset);
       const Slice key = cache_key.AsSlice();
 
       const Status s = GetBlobFromCache(key, &blob_handle);
 
-      if (s.ok() && blob_handle.GetValue()) {
+      if (s.ok()) {
         assert(req.status);
         *req.status = s;
 
-        {
-          req.result->Reset();
-          // To avoid copying the cached blob into the buffer provided by the
-          // application, we can simply transfer ownership of the cache handle
-          // to the target PinnableSlice. This has the potential to save a lot
-          // of CPU, especially with large blob values.
-          req.result->PinSlice(
-              *blob_handle.GetValue(),
-              [](void* arg1, void* arg2) {
-                Cache* const cache = static_cast<Cache*>(arg1);
-                Cache::Handle* const handle = static_cast<Cache::Handle*>(arg2);
-                cache->Release(handle);
-              },
-              blob_handle.GetCache(), blob_handle.GetCacheHandle());
-          // Make the CacheHandleGuard relinquish ownership of the handle.
-          blob_handle.TransferTo(nullptr);
-        }
+        PinCachedBlob(&blob_handle, req.result);
 
         // Update the counter for the number of valid blobs read from the cache.
         ++cached_blob_count;
@@ -367,8 +351,10 @@ void BlobSource::MultiGetBlobFromOneFile(const ReadOptions& read_options,
   if (no_io) {
     for (size_t i = 0; i < num_blobs; ++i) {
       if (!(cache_hit_mask & (Mask{1} << i))) {
-        assert(blob_reqs[i].status);
-        *blob_reqs[i].status =
+        BlobReadRequest& req = blob_reqs[i];
+        assert(req.status);
+
+        *req.status =
             Status::Incomplete("Cannot read blob(s): no disk I/O allowed");
       }
     }
@@ -377,12 +363,13 @@ void BlobSource::MultiGetBlobFromOneFile(const ReadOptions& read_options,
 
   {
     // Find the rest of blobs from the file since I/O is allowed.
-    autovector<BlobReadRequest*> _blob_reqs;
+    autovector<std::pair<BlobReadRequest*, std::unique_ptr<BlobContents>>>
+        _blob_reqs;
     uint64_t _bytes_read = 0;
 
     for (size_t i = 0; i < num_blobs; ++i) {
       if (!(cache_hit_mask & (Mask{1} << i))) {
-        _blob_reqs.push_back(&blob_reqs[i]);
+        _blob_reqs.emplace_back(&blob_reqs[i], std::unique_ptr<BlobContents>());
       }
     }
 
@@ -391,32 +378,51 @@ void BlobSource::MultiGetBlobFromOneFile(const ReadOptions& read_options,
         blob_file_cache_->GetBlobFileReader(file_number, &blob_file_reader);
     if (!s.ok()) {
       for (size_t i = 0; i < _blob_reqs.size(); ++i) {
-        assert(_blob_reqs[i]->status);
-        *_blob_reqs[i]->status = s;
+        BlobReadRequest* const req = _blob_reqs[i].first;
+        assert(req);
+        assert(req->status);
+
+        *req->status = s;
       }
       return;
     }
 
     assert(blob_file_reader.GetValue());
 
-    blob_file_reader.GetValue()->MultiGetBlob(read_options, _blob_reqs,
-                                              &_bytes_read);
+    MemoryAllocator* const allocator =
+        (blob_cache_ && read_options.fill_cache)
+            ? blob_cache_.get()->memory_allocator()
+            : nullptr;
+
+    blob_file_reader.GetValue()->MultiGetBlob(read_options, allocator,
+                                              _blob_reqs, &_bytes_read);
 
     if (blob_cache_ && read_options.fill_cache) {
       // If filling cache is allowed and a cache is configured, try to put
       // the blob(s) to the cache.
-      for (size_t i = 0; i < _blob_reqs.size(); ++i) {
-        if (_blob_reqs[i]->status->ok()) {
-          CacheHandleGuard<std::string> blob_handle;
-          const CacheKey cache_key =
-              base_cache_key.WithOffset(_blob_reqs[i]->offset);
+      for (auto& [req, blob_contents] : _blob_reqs) {
+        assert(req);
+
+        if (req->status->ok()) {
+          CacheHandleGuard<BlobContents> blob_handle;
+          const CacheKey cache_key = base_cache_key.WithOffset(req->offset);
           const Slice key = cache_key.AsSlice();
-          s = PutBlobIntoCache(key, &blob_handle, _blob_reqs[i]->result);
+          s = PutBlobIntoCache(key, &blob_contents, &blob_handle);
           if (!s.ok()) {
-            *_blob_reqs[i]->status = s;
+            *req->status = s;
+          } else {
+            PinCachedBlob(&blob_handle, req->result);
           }
         }
       }
+    } else {
+      for (auto& [req, blob_contents] : _blob_reqs) {
+        assert(req);
+
+        if (req->status->ok()) {
+          PinOwnedBlob(&blob_contents, req->result);
+        }
+      }
     }
 
     total_bytes += _bytes_read;
@@ -427,39 +433,28 @@ void BlobSource::MultiGetBlobFromOneFile(const ReadOptions& read_options,
 }
 
 bool BlobSource::TEST_BlobInCache(uint64_t file_number, uint64_t file_size,
-                                  uint64_t offset) const {
+                                  uint64_t offset, size_t* charge) const {
   const CacheKey cache_key = GetCacheKey(file_number, file_size, offset);
   const Slice key = cache_key.AsSlice();
 
-  CacheHandleGuard<std::string> blob_handle;
+  CacheHandleGuard<BlobContents> blob_handle;
   const Status s = GetBlobFromCache(key, &blob_handle);
 
   if (s.ok() && blob_handle.GetValue() != nullptr) {
+    if (charge) {
+      const Cache* const cache = blob_handle.GetCache();
+      assert(cache);
+
+      Cache::Handle* const handle = blob_handle.GetCacheHandle();
+      assert(handle);
+
+      *charge = cache->GetUsage(handle);
+    }
+
     return true;
   }
 
   return false;
 }
 
-// Callbacks for secondary blob cache
-size_t BlobSource::SizeCallback(void* obj) {
-  assert(obj != nullptr);
-  return static_cast<const std::string*>(obj)->size();
-}
-
-Status BlobSource::SaveToCallback(void* from_obj, size_t from_offset,
-                                  size_t length, void* out) {
-  assert(from_obj != nullptr);
-  const std::string* buf = static_cast<const std::string*>(from_obj);
-  assert(buf->size() >= from_offset + length);
-  memcpy(out, buf->data() + from_offset, length);
-  return Status::OK();
-}
-
-Cache::CacheItemHelper* BlobSource::GetCacheItemHelper() {
-  static Cache::CacheItemHelper cache_helper(SizeCallback, SaveToCallback,
-                                             &DeleteCacheEntry<std::string>);
-  return &cache_helper;
-}
-
 }  // namespace ROCKSDB_NAMESPACE

db/blob/blob_source.h

@@ -6,9 +6,11 @@
 #pragma once
 
 #include <cinttypes>
+#include <memory>
 
-#include "cache/cache_helpers.h"
 #include "cache/cache_key.h"
+#include "cache/typed_cache.h"
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_file_cache.h"
 #include "db/blob/blob_read_request.h"
 #include "rocksdb/cache.h"
@@ -102,20 +104,34 @@ class BlobSource {
   inline Cache* GetBlobCache() const { return blob_cache_.get(); }
 
   bool TEST_BlobInCache(uint64_t file_number, uint64_t file_size,
-                        uint64_t offset) const;
+                        uint64_t offset, size_t* charge = nullptr) const;
+
+  // For TypedSharedCacheInterface
+  void Create(BlobContents** out, const char* buf, size_t size,
+              MemoryAllocator* alloc);
+
+  using SharedCacheInterface =
+      FullTypedSharedCacheInterface<BlobContents, BlobContentsCreator>;
+  using TypedHandle = SharedCacheInterface::TypedHandle;
 
  private:
   Status GetBlobFromCache(const Slice& cache_key,
-                          CacheHandleGuard<std::string>* blob) const;
+                          CacheHandleGuard<BlobContents>* cached_blob) const;
 
   Status PutBlobIntoCache(const Slice& cache_key,
-                          CacheHandleGuard<std::string>* cached_blob,
-                          PinnableSlice* blob) const;
+                          std::unique_ptr<BlobContents>* blob,
+                          CacheHandleGuard<BlobContents>* cached_blob) const;
 
-  Cache::Handle* GetEntryFromCache(const Slice& key) const;
+  static void PinCachedBlob(CacheHandleGuard<BlobContents>* cached_blob,
+                            PinnableSlice* value);
 
-  Status InsertEntryIntoCache(const Slice& key, std::string* value,
-                              size_t charge, Cache::Handle** cache_handle,
+  static void PinOwnedBlob(std::unique_ptr<BlobContents>* owned_blob,
+                           PinnableSlice* value);
+
+  TypedHandle* GetEntryFromCache(const Slice& key) const;
+
+  Status InsertEntryIntoCache(const Slice& key, BlobContents* value,
+                              TypedHandle** cache_handle,
                               Cache::Priority priority) const;
 
   inline CacheKey GetCacheKey(uint64_t file_number, uint64_t /*file_size*/,
@@ -124,14 +140,6 @@ class BlobSource {
     return base_cache_key.WithOffset(offset);
   }
 
-  // Callbacks for secondary blob cache
-  static size_t SizeCallback(void* obj);
-
-  static Status SaveToCallback(void* from_obj, size_t from_offset,
-                               size_t length, void* out);
-
-  static Cache::CacheItemHelper* GetCacheItemHelper();
-
   const std::string& db_id_;
   const std::string& db_session_id_;
 
@@ -141,7 +149,7 @@ class BlobSource {
   BlobFileCache* blob_file_cache_;
 
   // A cache to store uncompressed blobs.
-  std::shared_ptr<Cache> blob_cache_;
+  mutable SharedCacheInterface blob_cache_;
 
   // The control option of how the cache tiers will be used. Currently rocksdb
   // support block/blob cache (volatile tier) and secondary cache (this tier

db/blob/blob_source_test.cc

@@ -13,6 +13,7 @@
 
 #include "cache/charged_cache.h"
 #include "cache/compressed_secondary_cache.h"
+#include "db/blob/blob_contents.h"
 #include "db/blob/blob_file_cache.h"
 #include "db/blob/blob_file_reader.h"
 #include "db/blob/blob_log_format.h"
@@ -218,6 +219,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                     kNoCompression, prefetch_buffer, &values[i],
                                     &bytes_read));
       ASSERT_EQ(values[i], blobs[i]);
+      ASSERT_TRUE(values[i].IsPinned());
       ASSERT_EQ(bytes_read,
                 BlobLogRecord::kHeaderSize + keys[i].size() + blob_sizes[i]);
 
@@ -255,6 +257,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                     kNoCompression, prefetch_buffer, &values[i],
                                     &bytes_read));
       ASSERT_EQ(values[i], blobs[i]);
+      ASSERT_TRUE(values[i].IsPinned());
       ASSERT_EQ(bytes_read,
                 BlobLogRecord::kHeaderSize + keys[i].size() + blob_sizes[i]);
 
@@ -298,6 +301,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                     kNoCompression, prefetch_buffer, &values[i],
                                     &bytes_read));
       ASSERT_EQ(values[i], blobs[i]);
+      ASSERT_TRUE(values[i].IsPinned());
       ASSERT_EQ(bytes_read,
                 BlobLogRecord::kHeaderSize + keys[i].size() + blob_sizes[i]);
 
@@ -336,6 +340,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                     kNoCompression, prefetch_buffer, &values[i],
                                     &bytes_read));
       ASSERT_EQ(values[i], blobs[i]);
+      ASSERT_TRUE(values[i].IsPinned());
       ASSERT_EQ(bytes_read,
                 BlobLogRecord::kHeaderSize + keys[i].size() + blob_sizes[i]);
 
@@ -382,6 +387,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                &bytes_read)
                       .IsIncomplete());
       ASSERT_TRUE(values[i].empty());
+      ASSERT_FALSE(values[i].IsPinned());
       ASSERT_EQ(bytes_read, 0);
 
       ASSERT_FALSE(blob_source.TEST_BlobInCache(blob_file_number, file_size,
@@ -423,6 +429,7 @@ TEST_F(BlobSourceTest, GetBlobsFromCache) {
                                &bytes_read)
                       .IsIOError());
       ASSERT_TRUE(values[i].empty());
+      ASSERT_FALSE(values[i].IsPinned());
       ASSERT_EQ(bytes_read, 0);
 
       ASSERT_FALSE(blob_source.TEST_BlobInCache(file_number, file_size,
@@ -855,6 +862,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
       if (i % 2 == 0) {
         ASSERT_OK(statuses_buf[i]);
         ASSERT_EQ(value_buf[i], blobs[i]);
+        ASSERT_TRUE(value_buf[i].IsPinned());
         fs_read_bytes +=
             blob_sizes[i] + keys[i].size() + BlobLogRecord::kHeaderSize;
         ASSERT_TRUE(blob_source.TEST_BlobInCache(blob_file_number, file_size,
@@ -863,6 +871,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
       } else {
         statuses_buf[i].PermitUncheckedError();
         ASSERT_TRUE(value_buf[i].empty());
+        ASSERT_FALSE(value_buf[i].IsPinned());
         ASSERT_FALSE(blob_source.TEST_BlobInCache(blob_file_number, file_size,
                                                   blob_offsets[i]));
       }
@@ -895,6 +904,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
                                     kNoCompression, prefetch_buffer,
                                     &value_buf[i], &bytes_read));
       ASSERT_EQ(value_buf[i], blobs[i]);
+      ASSERT_TRUE(value_buf[i].IsPinned());
       ASSERT_EQ(bytes_read,
                 BlobLogRecord::kHeaderSize + keys[i].size() + blob_sizes[i]);
 
@@ -920,6 +930,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
     for (size_t i = 0; i < num_blobs; ++i) {
       ASSERT_OK(statuses_buf[i]);
       ASSERT_EQ(value_buf[i], blobs[i]);
+      ASSERT_TRUE(value_buf[i].IsPinned());
       ASSERT_TRUE(blob_source.TEST_BlobInCache(blob_file_number, file_size,
                                                blob_offsets[i]));
       blob_bytes += blob_sizes[i];
@@ -968,6 +979,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
     for (size_t i = 0; i < num_blobs; ++i) {
       ASSERT_TRUE(statuses_buf[i].IsIncomplete());
       ASSERT_TRUE(value_buf[i].empty());
+      ASSERT_FALSE(value_buf[i].IsPinned());
       ASSERT_FALSE(blob_source.TEST_BlobInCache(blob_file_number, file_size,
                                                 blob_offsets[i]));
     }
@@ -1011,6 +1023,7 @@ TEST_F(BlobSourceTest, MultiGetBlobsFromCache) {
     for (size_t i = 0; i < num_blobs; ++i) {
       ASSERT_TRUE(statuses_buf[i].IsIOError());
       ASSERT_TRUE(value_buf[i].empty());
+      ASSERT_FALSE(value_buf[i].IsPinned());
       ASSERT_FALSE(blob_source.TEST_BlobInCache(non_existing_file_number,
                                                 file_size, blob_offsets[i]));
     }
@@ -1040,16 +1053,17 @@ class BlobSecondaryCacheTest : public DBTestBase {
 
     // Set a small cache capacity to evict entries from the cache, and to test
     // that secondary cache is used properly.
-    lru_cache_ops_.capacity = 1024;
-    lru_cache_ops_.num_shard_bits = 0;
-    lru_cache_ops_.strict_capacity_limit = true;
-    lru_cache_ops_.metadata_charge_policy = kDontChargeCacheMetadata;
-    lru_cache_ops_.high_pri_pool_ratio = 0.2;
-    lru_cache_ops_.low_pri_pool_ratio = 0.2;
+    lru_cache_opts_.capacity = 1024;
+    lru_cache_opts_.num_shard_bits = 0;
+    lru_cache_opts_.strict_capacity_limit = true;
+    lru_cache_opts_.metadata_charge_policy = kDontChargeCacheMetadata;
+    lru_cache_opts_.high_pri_pool_ratio = 0.2;
+    lru_cache_opts_.low_pri_pool_ratio = 0.2;
 
     secondary_cache_opts_.capacity = 8 << 20;  // 8 MB
     secondary_cache_opts_.num_shard_bits = 0;
-    secondary_cache_opts_.metadata_charge_policy = kDontChargeCacheMetadata;
+    secondary_cache_opts_.metadata_charge_policy =
+        kDefaultCacheMetadataChargePolicy;
 
     // Read blobs from the secondary cache if they are not in the primary cache
     options_.lowest_used_cache_tier = CacheTier::kNonVolatileBlockTier;
@@ -1060,7 +1074,7 @@ class BlobSecondaryCacheTest : public DBTestBase {
 
   Options options_;
 
-  LRUCacheOptions lru_cache_ops_;
+  LRUCacheOptions lru_cache_opts_;
   CompressedSecondaryCacheOptions secondary_cache_opts_;
 
   std::string db_id_;
@@ -1073,9 +1087,9 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
   }
 
   secondary_cache_opts_.compression_type = kSnappyCompression;
-  lru_cache_ops_.secondary_cache =
+  lru_cache_opts_.secondary_cache =
       NewCompressedSecondaryCache(secondary_cache_opts_);
-  options_.blob_cache = NewLRUCache(lru_cache_ops_);
+  options_.blob_cache = NewLRUCache(lru_cache_opts_);
 
   options_.cf_paths.emplace_back(
       test::PerThreadDBPath(
@@ -1098,8 +1112,8 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
   Random rnd(301);
 
   std::vector<std::string> key_strs{"key0", "key1"};
-  std::vector<std::string> blob_strs{rnd.RandomString(1010),
-                                     rnd.RandomString(1020)};
+  std::vector<std::string> blob_strs{rnd.RandomString(512),
+                                     rnd.RandomString(768)};
 
   std::vector<Slice> keys{key_strs[0], key_strs[1]};
   std::vector<Slice> blobs{blob_strs[0], blob_strs[1]};
@@ -1134,17 +1148,7 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
   read_options.verify_checksums = true;
 
   auto blob_cache = options_.blob_cache;
-  auto secondary_cache = lru_cache_ops_.secondary_cache;
-
-  Cache::CreateCallback create_cb = [&](const void* buf, size_t size,
-                                        void** out_obj,
-                                        size_t* charge) -> Status {
-    std::string* blob = new std::string();
-    blob->assign(static_cast<const char*>(buf), size);
-    *out_obj = blob;
-    *charge = size;
-    return Status::OK();
-  };
+  auto secondary_cache = lru_cache_opts_.secondary_cache;
 
   {
     // GetBlob
@@ -1154,6 +1158,25 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
     get_perf_context()->Reset();
 
     // key0 should be filled to the primary cache from the blob file.
+    ASSERT_OK(blob_source.GetBlob(read_options, keys[0], file_number,
+                                  blob_offsets[0], file_size, blob_sizes[0],
+                                  kNoCompression, nullptr /* prefetch_buffer */,
+                                  &values[0], nullptr /* bytes_read */));
+    // Release cache handle
+    values[0].Reset();
+
+    // key0 should be evicted and key0's dummy item is inserted into secondary
+    // cache. key1 should be filled to the primary cache from the blob file.
+    ASSERT_OK(blob_source.GetBlob(read_options, keys[1], file_number,
+                                  blob_offsets[1], file_size, blob_sizes[1],
+                                  kNoCompression, nullptr /* prefetch_buffer */,
+                                  &values[1], nullptr /* bytes_read */));
+
+    // Release cache handle
+    values[1].Reset();
+
+    // key0 should be filled to the primary cache from the blob file. key1
+    // should be evicted and key1's dummy item is inserted into secondary cache.
     ASSERT_OK(blob_source.GetBlob(read_options, keys[0], file_number,
                                   blob_offsets[0], file_size, blob_sizes[0],
                                   kNoCompression, nullptr /* prefetch_buffer */,
@@ -1162,8 +1185,11 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
     ASSERT_TRUE(
         blob_source.TEST_BlobInCache(file_number, file_size, blob_offsets[0]));
 
-    // key0 should be demoted to the secondary cache, and key1 should be filled
-    // to the primary cache from the blob file.
+    // Release cache handle
+    values[0].Reset();
+
+    // key0 should be evicted and is inserted into secondary cache.
+    // key1 should be filled to the primary cache from the blob file.
     ASSERT_OK(blob_source.GetBlob(read_options, keys[1], file_number,
                                   blob_offsets[1], file_size, blob_sizes[1],
                                   kNoCompression, nullptr /* prefetch_buffer */,
@@ -1172,6 +1198,9 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
     ASSERT_TRUE(
         blob_source.TEST_BlobInCache(file_number, file_size, blob_offsets[1]));
 
+    // Release cache handle
+    values[1].Reset();
+
     OffsetableCacheKey base_cache_key(db_id_, db_session_id_, file_number);
 
     // blob_cache here only looks at the primary cache since we didn't provide
@@ -1181,37 +1210,43 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
     {
       CacheKey cache_key = base_cache_key.WithOffset(blob_offsets[0]);
       const Slice key0 = cache_key.AsSlice();
-      auto handle0 = blob_cache->Lookup(key0, statistics);
+      auto handle0 = blob_cache->BasicLookup(key0, statistics);
       ASSERT_EQ(handle0, nullptr);
 
-      // key0 should be in the secondary cache. After looking up key0 in the
-      // secondary cache, it will be erased from the secondary cache.
+      // key0's item should be in the secondary cache.
       bool is_in_sec_cache = false;
-      auto sec_handle0 =
-          secondary_cache->Lookup(key0, create_cb, true, is_in_sec_cache);
+      auto sec_handle0 = secondary_cache->Lookup(
+          key0, &BlobSource::SharedCacheInterface::kFullHelper,
+          /*context*/ nullptr, true,
+          /*advise_erase=*/true, is_in_sec_cache);
       ASSERT_FALSE(is_in_sec_cache);
       ASSERT_NE(sec_handle0, nullptr);
       ASSERT_TRUE(sec_handle0->IsReady());
-      auto value = static_cast<std::string*>(sec_handle0->Value());
-      ASSERT_EQ(*value, blobs[0]);
+      auto value = static_cast<BlobContents*>(sec_handle0->Value());
+      ASSERT_NE(value, nullptr);
+      ASSERT_EQ(value->data(), blobs[0]);
       delete value;
 
-      // key0 doesn't exist in the blob cache
+      // key0 doesn't exist in the blob cache although key0's dummy
+      // item exist in the secondary cache.
       ASSERT_FALSE(blob_source.TEST_BlobInCache(file_number, file_size,
                                                 blob_offsets[0]));
     }
 
-    // key1 should exist in the primary cache.
+    // key1 should exists in the primary cache. key1's dummy item exists
+    // in the secondary cache.
     {
       CacheKey cache_key = base_cache_key.WithOffset(blob_offsets[1]);
       const Slice key1 = cache_key.AsSlice();
-      auto handle1 = blob_cache->Lookup(key1, statistics);
+      auto handle1 = blob_cache->BasicLookup(key1, statistics);
       ASSERT_NE(handle1, nullptr);
       blob_cache->Release(handle1);
 
       bool is_in_sec_cache = false;
-      auto sec_handle1 =
-          secondary_cache->Lookup(key1, create_cb, true, is_in_sec_cache);
+      auto sec_handle1 = secondary_cache->Lookup(
+          key1, &BlobSource::SharedCacheInterface::kFullHelper,
+          /*context*/ nullptr, true,
+          /*advise_erase=*/true, is_in_sec_cache);
       ASSERT_FALSE(is_in_sec_cache);
       ASSERT_EQ(sec_handle1, nullptr);
 
@@ -1221,46 +1256,61 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
 
     {
       // fetch key0 from the blob file to the primary cache.
+      // key1 is evicted and inserted into the secondary cache.
       ASSERT_OK(blob_source.GetBlob(
           read_options, keys[0], file_number, blob_offsets[0], file_size,
           blob_sizes[0], kNoCompression, nullptr /* prefetch_buffer */,
           &values[0], nullptr /* bytes_read */));
       ASSERT_EQ(values[0], blobs[0]);
 
+      // Release cache handle
+      values[0].Reset();
+
       // key0 should be in the primary cache.
       CacheKey cache_key0 = base_cache_key.WithOffset(blob_offsets[0]);
       const Slice key0 = cache_key0.AsSlice();
-      auto handle0 = blob_cache->Lookup(key0, statistics);
+      auto handle0 = blob_cache->BasicLookup(key0, statistics);
       ASSERT_NE(handle0, nullptr);
-      auto value = static_cast<std::string*>(blob_cache->Value(handle0));
-      ASSERT_EQ(*value, blobs[0]);
+      auto value = static_cast<BlobContents*>(blob_cache->Value(handle0));
+      ASSERT_NE(value, nullptr);
+      ASSERT_EQ(value->data(), blobs[0]);
       blob_cache->Release(handle0);
 
-      // key1 is not in the primary cache, and it should be demoted to the
-      // secondary cache.
+      // key1 is not in the primary cache and is in the secondary cache.
       CacheKey cache_key1 = base_cache_key.WithOffset(blob_offsets[1]);
       const Slice key1 = cache_key1.AsSlice();
-      auto handle1 = blob_cache->Lookup(key1, statistics);
+      auto handle1 = blob_cache->BasicLookup(key1, statistics);
       ASSERT_EQ(handle1, nullptr);
 
       // erase key0 from the primary cache.
       blob_cache->Erase(key0);
-      handle0 = blob_cache->Lookup(key0, statistics);
+      handle0 = blob_cache->BasicLookup(key0, statistics);
       ASSERT_EQ(handle0, nullptr);
 
       // key1 promotion should succeed due to the primary cache being empty. we
       // did't call secondary cache's Lookup() here, because it will remove the
       // key but it won't be able to promote the key to the primary cache.
-      // Instead we use the end-to-end blob source API to promote the key to
-      // the primary cache.
+      // Instead we use the end-to-end blob source API to read key1.
+      // In function TEST_BlobInCache, key1's dummy item is inserted into the
+      // primary cache and a standalone handle is checked by GetValue().
       ASSERT_TRUE(blob_source.TEST_BlobInCache(file_number, file_size,
                                                blob_offsets[1]));
 
-      // key1 should be in the primary cache.
-      handle1 = blob_cache->Lookup(key1, statistics);
+      // key1's dummy handle is in the primary cache and key1's item is still
+      // in the secondary cache. So, the primary cache's Lookup() without
+      // secondary cache support cannot see it. (NOTE: The dummy handle used
+      // to be a leaky abstraction but not anymore.)
+      handle1 = blob_cache->BasicLookup(key1, statistics);
+      ASSERT_EQ(handle1, nullptr);
+
+      // But after another access, it is promoted to primary cache
+      ASSERT_TRUE(blob_source.TEST_BlobInCache(file_number, file_size,
+                                               blob_offsets[1]));
+
+      // And Lookup() can find it (without secondary cache support)
+      handle1 = blob_cache->BasicLookup(key1, statistics);
       ASSERT_NE(handle1, nullptr);
-      value = static_cast<std::string*>(blob_cache->Value(handle1));
-      ASSERT_EQ(*value, blobs[1]);
+      ASSERT_NE(blob_cache->Value(handle1), nullptr);
       blob_cache->Release(handle1);
     }
   }
@@ -1385,11 +1435,11 @@ TEST_F(BlobSourceCacheReservationTest, SimpleCacheReservation) {
   ReadOptions read_options;
   read_options.verify_checksums = true;
 
-  std::vector<PinnableSlice> values(keys_.size());
-
   {
     read_options.fill_cache = false;
 
+    std::vector<PinnableSlice> values(keys_.size());
+
     for (size_t i = 0; i < kNumBlobs; ++i) {
       ASSERT_OK(blob_source.GetBlob(
           read_options, keys_[i], kBlobFileNumber, blob_offsets[i],
@@ -1403,6 +1453,8 @@ TEST_F(BlobSourceCacheReservationTest, SimpleCacheReservation) {
   {
     read_options.fill_cache = true;
 
+    std::vector<PinnableSlice> values(keys_.size());
+
     // num_blobs is 16, so the total blob cache usage is less than a single
     // dummy entry. Therefore, cache reservation manager only reserves one dummy
     // entry here.
@@ -1412,7 +1464,12 @@ TEST_F(BlobSourceCacheReservationTest, SimpleCacheReservation) {
           read_options, keys_[i], kBlobFileNumber, blob_offsets[i],
           blob_file_size_, blob_sizes[i], kNoCompression,
           nullptr /* prefetch_buffer */, &values[i], nullptr /* bytes_read */));
-      blob_bytes += blob_sizes[i];
+
+      size_t charge = 0;
+      ASSERT_TRUE(blob_source.TEST_BlobInCache(kBlobFileNumber, blob_file_size_,
+                                               blob_offsets[i], &charge));
+
+      blob_bytes += charge;
       ASSERT_EQ(cache_res_mgr->GetTotalReservedCacheSize(), kSizeDummyEntry);
       ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(), blob_bytes);
       ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(),
@@ -1425,19 +1482,23 @@ TEST_F(BlobSourceCacheReservationTest, SimpleCacheReservation) {
     size_t blob_bytes = options_.blob_cache->GetUsage();
 
     for (size_t i = 0; i < kNumBlobs; ++i) {
+      size_t charge = 0;
+      ASSERT_TRUE(blob_source.TEST_BlobInCache(kBlobFileNumber, blob_file_size_,
+                                               blob_offsets[i], &charge));
+
       CacheKey cache_key = base_cache_key.WithOffset(blob_offsets[i]);
       // We didn't call options_.blob_cache->Erase() here, this is because
       // the cache wrapper's Erase() method must be called to update the
       // cache usage after erasing the cache entry.
       blob_source.GetBlobCache()->Erase(cache_key.AsSlice());
       if (i == kNumBlobs - 1) {
-        // The last blob is not in the cache. cache_res_mgr should not reserve
-        // any space for it.
+        // All the blobs got removed from the cache. cache_res_mgr should not
+        // reserve any space for them.
         ASSERT_EQ(cache_res_mgr->GetTotalReservedCacheSize(), 0);
       } else {
         ASSERT_EQ(cache_res_mgr->GetTotalReservedCacheSize(), kSizeDummyEntry);
       }
-      blob_bytes -= blob_sizes[i];
+      blob_bytes -= charge;
       ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(), blob_bytes);
       ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(),
                 options_.blob_cache->GetUsage());
@@ -1495,11 +1556,11 @@ TEST_F(BlobSourceCacheReservationTest, IncreaseCacheReservationOnFullCache) {
   ReadOptions read_options;
   read_options.verify_checksums = true;
 
-  std::vector<PinnableSlice> values(keys_.size());
-
   {
     read_options.fill_cache = false;
 
+    std::vector<PinnableSlice> values(keys_.size());
+
     for (size_t i = 0; i < kNumBlobs; ++i) {
       ASSERT_OK(blob_source.GetBlob(
           read_options, keys_[i], kBlobFileNumber, blob_offsets[i],
@@ -1513,21 +1574,33 @@ TEST_F(BlobSourceCacheReservationTest, IncreaseCacheReservationOnFullCache) {
   {
     read_options.fill_cache = true;
 
-    // Since we resized each blob to be kSizeDummyEntry / (num_blobs/ 2), we
-    // should observe cache eviction for the second half blobs.
+    std::vector<PinnableSlice> values(keys_.size());
+
+    // Since we resized each blob to be kSizeDummyEntry / (num_blobs / 2), we
+    // can't fit all the blobs in the cache at the same time, which means we
+    // should observe cache evictions once we reach the cache's capacity.
+    // Due to the overhead of the cache and the BlobContents objects, as well as
+    // jemalloc bin sizes, this happens after inserting seven blobs.
     uint64_t blob_bytes = 0;
     for (size_t i = 0; i < kNumBlobs; ++i) {
       ASSERT_OK(blob_source.GetBlob(
           read_options, keys_[i], kBlobFileNumber, blob_offsets[i],
           blob_file_size_, blob_sizes[i], kNoCompression,
           nullptr /* prefetch_buffer */, &values[i], nullptr /* bytes_read */));
-      blob_bytes += blob_sizes[i];
-      ASSERT_EQ(cache_res_mgr->GetTotalReservedCacheSize(), kSizeDummyEntry);
-      if (i >= kNumBlobs / 2) {
-        ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(), kSizeDummyEntry);
-      } else {
-        ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(), blob_bytes);
+
+      // Release cache handle
+      values[i].Reset();
+
+      if (i < kNumBlobs / 2 - 1) {
+        size_t charge = 0;
+        ASSERT_TRUE(blob_source.TEST_BlobInCache(
+            kBlobFileNumber, blob_file_size_, blob_offsets[i], &charge));
+
+        blob_bytes += charge;
       }
+
+      ASSERT_EQ(cache_res_mgr->GetTotalReservedCacheSize(), kSizeDummyEntry);
+      ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(), blob_bytes);
       ASSERT_EQ(cache_res_mgr->GetTotalMemoryUsed(),
                 options_.blob_cache->GetUsage());
     }
@@ -1538,6 +1611,7 @@ TEST_F(BlobSourceCacheReservationTest, IncreaseCacheReservationOnFullCache) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/blob/db_blob_basic_test.cc

@@ -7,6 +7,7 @@
 #include <sstream>
 #include <string>
 
+#include "cache/compressed_secondary_cache.h"
 #include "db/blob/blob_index.h"
 #include "db/blob/blob_log_format.h"
 #include "db/db_test_util.h"
@@ -53,7 +54,7 @@ TEST_F(DBBlobBasicTest, GetBlobFromCache) {
   Options options = GetDefaultOptions();
 
   LRUCacheOptions co;
-  co.capacity = 2048;
+  co.capacity = 2 << 20;  // 2MB
   co.num_shard_bits = 2;
   co.metadata_charge_policy = kDontChargeCacheMetadata;
   auto backing_cache = NewLRUCache(co);
@@ -121,7 +122,7 @@ TEST_F(DBBlobBasicTest, IterateBlobsFromCache) {
   Options options = GetDefaultOptions();
 
   LRUCacheOptions co;
-  co.capacity = 2048;
+  co.capacity = 2 << 20;  // 2MB
   co.num_shard_bits = 2;
   co.metadata_charge_policy = kDontChargeCacheMetadata;
   auto backing_cache = NewLRUCache(co);
@@ -452,7 +453,7 @@ TEST_F(DBBlobBasicTest, MultiGetBlobsFromCache) {
   Options options = GetDefaultOptions();
 
   LRUCacheOptions co;
-  co.capacity = 2048;
+  co.capacity = 2 << 20;  // 2MB
   co.num_shard_bits = 2;
   co.metadata_charge_policy = kDontChargeCacheMetadata;
   auto backing_cache = NewLRUCache(co);
@@ -778,7 +779,7 @@ TEST_F(DBBlobBasicTest, MultiGetBlobsFromMultipleFiles) {
   Options options = GetDefaultOptions();
 
   LRUCacheOptions co;
-  co.capacity = 2 << 20; // 2MB
+  co.capacity = 2 << 20;  // 2MB
   co.num_shard_bits = 2;
   co.metadata_charge_policy = kDontChargeCacheMetadata;
   auto backing_cache = NewLRUCache(co);
@@ -1701,6 +1702,83 @@ TEST_F(DBBlobBasicTest, DynamicallyWarmCacheDuringFlush) {
 }
 #endif  // !ROCKSDB_LITE
 
+TEST_F(DBBlobBasicTest, WarmCacheWithBlobsSecondary) {
+  CompressedSecondaryCacheOptions secondary_cache_opts;
+  secondary_cache_opts.capacity = 1 << 20;
+  secondary_cache_opts.num_shard_bits = 0;
+  secondary_cache_opts.metadata_charge_policy = kDontChargeCacheMetadata;
+  secondary_cache_opts.compression_type = kNoCompression;
+
+  LRUCacheOptions primary_cache_opts;
+  primary_cache_opts.capacity = 1024;
+  primary_cache_opts.num_shard_bits = 0;
+  primary_cache_opts.metadata_charge_policy = kDontChargeCacheMetadata;
+  primary_cache_opts.secondary_cache =
+      NewCompressedSecondaryCache(secondary_cache_opts);
+
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.statistics = CreateDBStatistics();
+  options.enable_blob_files = true;
+  options.blob_cache = NewLRUCache(primary_cache_opts);
+  options.prepopulate_blob_cache = PrepopulateBlobCache::kFlushOnly;
+
+  DestroyAndReopen(options);
+
+  // Note: only one of the two blobs fit in the primary cache at any given time.
+  constexpr char first_key[] = "foo";
+  constexpr size_t first_blob_size = 512;
+  const std::string first_blob(first_blob_size, 'a');
+
+  constexpr char second_key[] = "bar";
+  constexpr size_t second_blob_size = 768;
+  const std::string second_blob(second_blob_size, 'b');
+
+  // First blob is inserted into primary cache during flush.
+  ASSERT_OK(Put(first_key, first_blob));
+  ASSERT_OK(Flush());
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_ADD), 1);
+
+  // Second blob is inserted into primary cache during flush,
+  // First blob is evicted but only a dummy handle is inserted into secondary
+  // cache.
+  ASSERT_OK(Put(second_key, second_blob));
+  ASSERT_OK(Flush());
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_ADD), 1);
+
+  // First blob is inserted into primary cache.
+  // Second blob is evicted but only a dummy handle is inserted into secondary
+  // cache.
+  ASSERT_EQ(Get(first_key), first_blob);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_MISS), 1);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_HIT), 0);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(SECONDARY_CACHE_HITS),
+            0);
+  // Second blob is inserted into primary cache,
+  // First blob is evicted and is inserted into secondary cache.
+  ASSERT_EQ(Get(second_key), second_blob);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_MISS), 1);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_HIT), 0);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(SECONDARY_CACHE_HITS),
+            0);
+
+  // First blob's dummy item is inserted into primary cache b/c of lookup.
+  // Second blob is still in primary cache.
+  ASSERT_EQ(Get(first_key), first_blob);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_MISS), 0);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_HIT), 1);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(SECONDARY_CACHE_HITS),
+            1);
+
+  // First blob's item is inserted into primary cache b/c of lookup.
+  // Second blob is evicted and inserted into secondary cache.
+  ASSERT_EQ(Get(first_key), first_blob);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_MISS), 0);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(BLOB_DB_CACHE_HIT), 1);
+  ASSERT_EQ(options.statistics->getAndResetTickerCount(SECONDARY_CACHE_HITS),
+            1);
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/builder.cc

@@ -71,8 +71,9 @@ Status BuildTable(
     int job_id, const Env::IOPriority io_priority,
     TableProperties* table_properties, Env::WriteLifeTimeHint write_hint,
     const std::string* full_history_ts_low,
-    BlobFileCompletionCallback* blob_callback, uint64_t* num_input_entries,
-    uint64_t* memtable_payload_bytes, uint64_t* memtable_garbage_bytes) {
+    BlobFileCompletionCallback* blob_callback, Version* version,
+    uint64_t* num_input_entries, uint64_t* memtable_payload_bytes,
+    uint64_t* memtable_garbage_bytes) {
   assert((tboptions.column_family_id ==
           TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
          tboptions.column_family_name.empty());
@@ -90,7 +91,7 @@ Status BuildTable(
   iter->SeekToFirst();
   std::unique_ptr<CompactionRangeDelAggregator> range_del_agg(
       new CompactionRangeDelAggregator(&tboptions.internal_comparator,
-                                       snapshots));
+                                       snapshots, full_history_ts_low));
   uint64_t num_unfragmented_tombstones = 0;
   uint64_t total_tombstone_payload_bytes = 0;
   for (auto& range_del_iter : range_del_iters) {
@@ -175,10 +176,10 @@ Status BuildTable(
       builder = NewTableBuilder(tboptions, file_writer.get());
     }
 
+    auto ucmp = tboptions.internal_comparator.user_comparator();
     MergeHelper merge(
-        env, tboptions.internal_comparator.user_comparator(),
-        ioptions.merge_operator.get(), compaction_filter.get(), ioptions.logger,
-        true /* internal key corruption is not ok */,
+        env, ucmp, ioptions.merge_operator.get(), compaction_filter.get(),
+        ioptions.logger, true /* internal key corruption is not ok */,
         snapshots.empty() ? 0 : snapshots.back(), snapshot_checker);
 
     std::unique_ptr<BlobFileBuilder> blob_file_builder(
@@ -196,9 +197,8 @@ Status BuildTable(
 
     const std::atomic<bool> kManualCompactionCanceledFalse{false};
     CompactionIterator c_iter(
-        iter, tboptions.internal_comparator.user_comparator(), &merge,
-        kMaxSequenceNumber, &snapshots, earliest_write_conflict_snapshot,
-        job_snapshot, snapshot_checker, env,
+        iter, ucmp, &merge, kMaxSequenceNumber, &snapshots,
+        earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
         ShouldReportDetailedTime(env, ioptions.stats),
         true /* internal key corruption is not ok */, range_del_agg.get(),
         blob_file_builder.get(), ioptions.allow_data_in_errors,
@@ -241,14 +241,28 @@ Status BuildTable(
 
     if (s.ok()) {
       auto range_del_it = range_del_agg->NewIterator();
+      Slice last_tombstone_start_user_key{};
       for (range_del_it->SeekToFirst(); range_del_it->Valid();
            range_del_it->Next()) {
         auto tombstone = range_del_it->Tombstone();
         auto kv = tombstone.Serialize();
         builder->Add(kv.first.Encode(), kv.second);
-        meta->UpdateBoundariesForRange(kv.first, tombstone.SerializeEndKey(),
-                                       tombstone.seq_,
+        InternalKey tombstone_end = tombstone.SerializeEndKey();
+        meta->UpdateBoundariesForRange(kv.first, tombstone_end, tombstone.seq_,
                                        tboptions.internal_comparator);
+        if (version) {
+          if (last_tombstone_start_user_key.empty() ||
+              ucmp->CompareWithoutTimestamp(last_tombstone_start_user_key,
+                                            range_del_it->start_key()) < 0) {
+            SizeApproximationOptions approx_opts;
+            approx_opts.files_size_error_margin = 0.1;
+            meta->compensated_range_deletion_size += versions->ApproximateSize(
+                approx_opts, version, kv.first.Encode(), tombstone_end.Encode(),
+                0 /* start_level */, -1 /* end_level */,
+                TableReaderCaller::kFlush);
+          }
+          last_tombstone_start_user_key = range_del_it->start_key();
+        }
       }
     }
 
@@ -281,7 +295,8 @@ Status BuildTable(
       meta->fd.file_size = file_size;
       meta->marked_for_compaction = builder->NeedCompact();
       assert(meta->fd.GetFileSize() > 0);
-      tp = builder->GetTableProperties(); // refresh now that builder is finished
+      tp = builder
+               ->GetTableProperties();  // refresh now that builder is finished
       if (memtable_payload_bytes != nullptr &&
           memtable_garbage_bytes != nullptr) {
         const CompactionIterationStats& ci_stats = c_iter.iter_stats();

db/builder.h

@@ -13,6 +13,7 @@
 #include "db/range_tombstone_fragmenter.h"
 #include "db/seqno_to_time_mapping.h"
 #include "db/table_properties_collector.h"
+#include "db/version_set.h"
 #include "logging/event_logger.h"
 #include "options/cf_options.h"
 #include "rocksdb/comparator.h"
@@ -70,7 +71,7 @@ extern Status BuildTable(
     Env::WriteLifeTimeHint write_hint = Env::WLTH_NOT_SET,
     const std::string* full_history_ts_low = nullptr,
     BlobFileCompletionCallback* blob_callback = nullptr,
-    uint64_t* num_input_entries = nullptr,
+    Version* version = nullptr, uint64_t* num_input_entries = nullptr,
     uint64_t* memtable_payload_bytes = nullptr,
     uint64_t* memtable_garbage_bytes = nullptr);
 

db/c_test.c

@@ -49,32 +49,32 @@ static void StartPhase(const char* name) {
 }
 #ifdef _MSC_VER
 #pragma warning(push)
-#pragma warning (disable: 4996) // getenv security warning
+#pragma warning(disable : 4996)  // getenv security warning
 #endif
 static const char* GetTempDir(void) {
-    const char* ret = getenv("TEST_TMPDIR");
-    if (ret == NULL || ret[0] == '\0')
+  const char* ret = getenv("TEST_TMPDIR");
+  if (ret == NULL || ret[0] == '\0')
 #ifdef OS_WIN
-      ret = getenv("TEMP");
+    ret = getenv("TEMP");
 #else
-      ret = "/tmp";
+    ret = "/tmp";
 #endif
-    return ret;
+  return ret;
 }
 #ifdef _MSC_VER
 #pragma warning(pop)
 #endif
 
-#define CheckNoError(err)                                               \
-  if ((err) != NULL) {                                                  \
+#define CheckNoError(err)                                                 \
+  if ((err) != NULL) {                                                    \
     fprintf(stderr, "%s:%d: %s: %s\n", __FILE__, __LINE__, phase, (err)); \
-    abort();                                                            \
+    abort();                                                              \
   }
 
-#define CheckCondition(cond)                                            \
-  if (!(cond)) {                                                        \
+#define CheckCondition(cond)                                              \
+  if (!(cond)) {                                                          \
     fprintf(stderr, "%s:%d: %s: %s\n", __FILE__, __LINE__, phase, #cond); \
-    abort();                                                            \
+    abort();                                                              \
   }
 
 static void CheckEqual(const char* expected, const char* v, size_t n) {
@@ -98,21 +98,15 @@ static void Free(char** ptr) {
   }
 }
 
-static void CheckValue(
-    char* err,
-    const char* expected,
-    char** actual,
-    size_t actual_length) {
+static void CheckValue(char* err, const char* expected, char** actual,
+                       size_t actual_length) {
   CheckNoError(err);
   CheckEqual(expected, *actual, actual_length);
   Free(actual);
 }
 
-static void CheckGet(
-    rocksdb_t* db,
-    const rocksdb_readoptions_t* options,
-    const char* key,
-    const char* expected) {
+static void CheckGet(rocksdb_t* db, const rocksdb_readoptions_t* options,
+                     const char* key, const char* expected) {
   char* err = NULL;
   size_t val_len;
   char* val;
@@ -122,12 +116,9 @@ static void CheckGet(
   Free(&val);
 }
 
-static void CheckGetCF(
-    rocksdb_t* db,
-    const rocksdb_readoptions_t* options,
-    rocksdb_column_family_handle_t* handle,
-    const char* key,
-    const char* expected) {
+static void CheckGetCF(rocksdb_t* db, const rocksdb_readoptions_t* options,
+                       rocksdb_column_family_handle_t* handle, const char* key,
+                       const char* expected) {
   char* err = NULL;
   size_t val_len;
   char* val;
@@ -174,8 +165,8 @@ static void CheckMultiGetValues(size_t num_keys, char** values,
   }
 }
 
-static void CheckIter(rocksdb_iterator_t* iter,
-                      const char* key, const char* val) {
+static void CheckIter(rocksdb_iterator_t* iter, const char* key,
+                      const char* val) {
   size_t len;
   const char* str;
   str = rocksdb_iter_key(iter, &len);
@@ -185,10 +176,9 @@ static void CheckIter(rocksdb_iterator_t* iter,
 }
 
 // Callback from rocksdb_writebatch_iterate()
-static void CheckPut(void* ptr,
-                     const char* k, size_t klen,
-                     const char* v, size_t vlen) {
-  int* state = (int*) ptr;
+static void CheckPut(void* ptr, const char* k, size_t klen, const char* v,
+                     size_t vlen) {
+  int* state = (int*)ptr;
   CheckCondition(*state < 2);
   switch (*state) {
     case 0:
@@ -205,7 +195,7 @@ static void CheckPut(void* ptr,
 
 // Callback from rocksdb_writebatch_iterate()
 static void CheckDel(void* ptr, const char* k, size_t klen) {
-  int* state = (int*) ptr;
+  int* state = (int*)ptr;
   CheckCondition(*state == 2);
   CheckEqual("bar", k, klen);
   (*state)++;
@@ -213,14 +203,16 @@ static void CheckDel(void* ptr, const char* k, size_t klen) {
 
 static void CmpDestroy(void* arg) { (void)arg; }
 
-static int CmpCompare(void* arg, const char* a, size_t alen,
-                      const char* b, size_t blen) {
+static int CmpCompare(void* arg, const char* a, size_t alen, const char* b,
+                      size_t blen) {
   (void)arg;
   size_t n = (alen < blen) ? alen : blen;
   int r = memcmp(a, b, n);
   if (r == 0) {
-    if (alen < blen) r = -1;
-    else if (alen > blen) r = +1;
+    if (alen < blen)
+      r = -1;
+    else if (alen > blen)
+      r = +1;
   }
   return r;
 }
@@ -405,11 +397,9 @@ static const char* MergeOperatorName(void* arg) {
   return "TestMergeOperator";
 }
 static char* MergeOperatorFullMerge(
-    void* arg,
-    const char* key, size_t key_length,
-    const char* existing_value, size_t existing_value_length,
-    const char* const* operands_list, const size_t* operands_list_length,
-    int num_operands,
+    void* arg, const char* key, size_t key_length, const char* existing_value,
+    size_t existing_value_length, const char* const* operands_list,
+    const size_t* operands_list_length, int num_operands,
     unsigned char* success, size_t* new_value_length) {
   (void)arg;
   (void)key;
@@ -425,12 +415,12 @@ static char* MergeOperatorFullMerge(
   memcpy(result, "fake", 4);
   return result;
 }
-static char* MergeOperatorPartialMerge(
-    void* arg,
-    const char* key, size_t key_length,
-    const char* const* operands_list, const size_t* operands_list_length,
-    int num_operands,
-    unsigned char* success, size_t* new_value_length) {
+static char* MergeOperatorPartialMerge(void* arg, const char* key,
+                                       size_t key_length,
+                                       const char* const* operands_list,
+                                       const size_t* operands_list_length,
+                                       int num_operands, unsigned char* success,
+                                       size_t* new_value_length) {
   (void)arg;
   (void)key;
   (void)key_length;
@@ -444,18 +434,16 @@ static char* MergeOperatorPartialMerge(
   return result;
 }
 
-static void CheckTxnGet(
-        rocksdb_transaction_t* txn,
-        const rocksdb_readoptions_t* options,
-        const char* key,
-        const char* expected) {
-        char* err = NULL;
-        size_t val_len;
-        char* val;
-        val = rocksdb_transaction_get(txn, options, key, strlen(key), &val_len, &err);
-        CheckNoError(err);
-        CheckEqual(expected, val, val_len);
-        Free(&val);
+static void CheckTxnGet(rocksdb_transaction_t* txn,
+                        const rocksdb_readoptions_t* options, const char* key,
+                        const char* expected) {
+  char* err = NULL;
+  size_t val_len;
+  char* val;
+  val = rocksdb_transaction_get(txn, options, key, strlen(key), &val_len, &err);
+  CheckNoError(err);
+  CheckEqual(expected, val, val_len);
+  Free(&val);
 }
 
 static void CheckTxnGetCF(rocksdb_transaction_t* txn,
@@ -502,11 +490,9 @@ static void CheckTxnPinGetCF(rocksdb_transaction_t* txn,
   rocksdb_pinnableslice_destroy(p);
 }
 
-static void CheckTxnDBGet(
-        rocksdb_transactiondb_t* txn_db,
-        const rocksdb_readoptions_t* options,
-        const char* key,
-        const char* expected) {
+static void CheckTxnDBGet(rocksdb_transactiondb_t* txn_db,
+                          const rocksdb_readoptions_t* options, const char* key,
+                          const char* expected) {
   char* err = NULL;
   size_t val_len;
   char* val;
@@ -632,7 +618,7 @@ int main(int argc, char** argv) {
   rocksdb_t* db;
   rocksdb_comparator_t* cmp;
   rocksdb_cache_t* cache;
-  rocksdb_dbpath_t *dbpath;
+  rocksdb_dbpath_t* dbpath;
   rocksdb_env_t* env;
   rocksdb_options_t* options;
   rocksdb_compactoptions_t* coptions;
@@ -649,30 +635,20 @@ int main(int argc, char** argv) {
   char* err = NULL;
   int run = -1;
 
-  snprintf(dbname, sizeof(dbname),
-           "%s/rocksdb_c_test-%d",
-           GetTempDir(),
-           ((int) geteuid()));
-
-  snprintf(dbbackupname, sizeof(dbbackupname),
-           "%s/rocksdb_c_test-%d-backup",
-           GetTempDir(),
-           ((int) geteuid()));
-
-  snprintf(dbcheckpointname, sizeof(dbcheckpointname),
-           "%s/rocksdb_c_test-%d-checkpoint",
-           GetTempDir(),
-           ((int) geteuid()));
-
-  snprintf(sstfilename, sizeof(sstfilename),
-           "%s/rocksdb_c_test-%d-sst",
-           GetTempDir(),
+  snprintf(dbname, sizeof(dbname), "%s/rocksdb_c_test-%d", GetTempDir(),
            ((int)geteuid()));
 
-  snprintf(dbpathname, sizeof(dbpathname),
-           "%s/rocksdb_c_test-%d-dbpath",
-           GetTempDir(),
-           ((int) geteuid()));
+  snprintf(dbbackupname, sizeof(dbbackupname), "%s/rocksdb_c_test-%d-backup",
+           GetTempDir(), ((int)geteuid()));
+
+  snprintf(dbcheckpointname, sizeof(dbcheckpointname),
+           "%s/rocksdb_c_test-%d-checkpoint", GetTempDir(), ((int)geteuid()));
+
+  snprintf(sstfilename, sizeof(sstfilename), "%s/rocksdb_c_test-%d-sst",
+           GetTempDir(), ((int)geteuid()));
+
+  snprintf(dbpathname, sizeof(dbpathname), "%s/rocksdb_c_test-%d-dbpath",
+           GetTempDir(), ((int)geteuid()));
 
   StartPhase("create_objects");
   cmp = rocksdb_comparator_create(NULL, CmpDestroy, CmpCompare, CmpName);
@@ -746,7 +722,8 @@ int main(int argc, char** argv) {
     rocksdb_destroy_db(options, dbbackupname, &err);
     CheckNoError(err);
 
-    rocksdb_backup_engine_t *be = rocksdb_backup_engine_open(options, dbbackupname, &err);
+    rocksdb_backup_engine_t* be =
+        rocksdb_backup_engine_open(options, dbbackupname, &err);
     CheckNoError(err);
 
     rocksdb_backup_engine_create_new_backup(be, db, &err);
@@ -759,7 +736,8 @@ int main(int argc, char** argv) {
     rocksdb_backup_engine_create_new_backup(be, db, &err);
     CheckNoError(err);
 
-    const rocksdb_backup_engine_info_t* bei = rocksdb_backup_engine_get_backup_info(be);
+    const rocksdb_backup_engine_info_t* bei =
+        rocksdb_backup_engine_get_backup_info(be);
     CheckCondition(rocksdb_backup_engine_info_count(bei) > 1);
     rocksdb_backup_engine_info_destroy(bei);
 
@@ -778,9 +756,11 @@ int main(int argc, char** argv) {
     rocksdb_destroy_db(options, dbname, &err);
     CheckNoError(err);
 
-    rocksdb_restore_options_t *restore_options = rocksdb_restore_options_create();
+    rocksdb_restore_options_t* restore_options =
+        rocksdb_restore_options_create();
     rocksdb_restore_options_set_keep_log_files(restore_options, 0);
-    rocksdb_backup_engine_restore_db_from_latest_backup(be, dbname, dbname, restore_options, &err);
+    rocksdb_backup_engine_restore_db_from_latest_backup(be, dbname, dbname,
+                                                        restore_options, &err);
     CheckNoError(err);
     rocksdb_restore_options_destroy(restore_options);
 
@@ -799,7 +779,8 @@ int main(int argc, char** argv) {
     rocksdb_destroy_db(options, dbcheckpointname, &err);
     CheckNoError(err);
 
-    rocksdb_checkpoint_t* checkpoint = rocksdb_checkpoint_object_create(db, &err);
+    rocksdb_checkpoint_t* checkpoint =
+        rocksdb_checkpoint_object_create(db, &err);
     CheckNoError(err);
 
     rocksdb_checkpoint_create(checkpoint, dbcheckpointname, 0, &err);
@@ -976,10 +957,10 @@ int main(int argc, char** argv) {
   StartPhase("writebatch_vectors");
   {
     rocksdb_writebatch_t* wb = rocksdb_writebatch_create();
-    const char* k_list[2] = { "z", "ap" };
-    const size_t k_sizes[2] = { 1, 2 };
-    const char* v_list[3] = { "x", "y", "z" };
-    const size_t v_sizes[3] = { 1, 1, 1 };
+    const char* k_list[2] = {"z", "ap"};
+    const size_t k_sizes[2] = {1, 2};
+    const char* v_list[3] = {"x", "y", "z"};
+    const size_t v_sizes[3] = {1, 1, 1};
     rocksdb_writebatch_putv(wb, 2, k_list, k_sizes, 3, v_list, v_sizes);
     rocksdb_write(db, woptions, wb, &err);
     CheckNoError(err);
@@ -1041,13 +1022,17 @@ int main(int argc, char** argv) {
     CheckCondition(count == 3);
     size_t size;
     char* value;
-    value = rocksdb_writebatch_wi_get_from_batch(wbi, options, "box", 3, &size, &err);
+    value = rocksdb_writebatch_wi_get_from_batch(wbi, options, "box", 3, &size,
+                                                 &err);
     CheckValue(err, "c", &value, size);
-    value = rocksdb_writebatch_wi_get_from_batch(wbi, options, "bar", 3, &size, &err);
+    value = rocksdb_writebatch_wi_get_from_batch(wbi, options, "bar", 3, &size,
+                                                 &err);
     CheckValue(err, NULL, &value, size);
-    value = rocksdb_writebatch_wi_get_from_batch_and_db(wbi, db, roptions, "foo", 3, &size, &err);
+    value = rocksdb_writebatch_wi_get_from_batch_and_db(wbi, db, roptions,
+                                                        "foo", 3, &size, &err);
     CheckValue(err, "hello", &value, size);
-    value = rocksdb_writebatch_wi_get_from_batch_and_db(wbi, db, roptions, "box", 3, &size, &err);
+    value = rocksdb_writebatch_wi_get_from_batch_and_db(wbi, db, roptions,
+                                                        "box", 3, &size, &err);
     CheckValue(err, "c", &value, size);
     rocksdb_write_writebatch_wi(db, woptions, wbi, &err);
     CheckNoError(err);
@@ -1064,10 +1049,10 @@ int main(int argc, char** argv) {
   StartPhase("writebatch_wi_vectors");
   {
     rocksdb_writebatch_wi_t* wb = rocksdb_writebatch_wi_create(0, 1);
-    const char* k_list[2] = { "z", "ap" };
-    const size_t k_sizes[2] = { 1, 2 };
-    const char* v_list[3] = { "x", "y", "z" };
-    const size_t v_sizes[3] = { 1, 1, 1 };
+    const char* k_list[2] = {"z", "ap"};
+    const size_t k_sizes[2] = {1, 2};
+    const char* v_list[3] = {"x", "y", "z"};
+    const size_t v_sizes[3] = {1, 1, 1};
     rocksdb_writebatch_wi_putv(wb, 2, k_list, k_sizes, 3, v_list, v_sizes);
     rocksdb_write_writebatch_wi(db, woptions, wb, &err);
     CheckNoError(err);
@@ -1156,13 +1141,14 @@ int main(int argc, char** argv) {
 
   StartPhase("multiget");
   {
-    const char* keys[3] = { "box", "foo", "notfound" };
-    const size_t keys_sizes[3] = { 3, 3, 8 };
+    const char* keys[3] = {"box", "foo", "notfound"};
+    const size_t keys_sizes[3] = {3, 3, 8};
     char* vals[3];
     size_t vals_sizes[3];
     char* errs[3];
     const char* expected[3] = {"c", "hello", NULL};
-    rocksdb_multi_get(db, roptions, 3, keys, keys_sizes, vals, vals_sizes, errs);
+    rocksdb_multi_get(db, roptions, 3, keys, keys_sizes, vals, vals_sizes,
+                      errs);
     CheckMultiGetValues(3, vals, vals_sizes, errs, expected);
   }
 
@@ -1180,10 +1166,10 @@ int main(int argc, char** argv) {
     char keybuf[100];
     char valbuf[100];
     uint64_t sizes[2];
-    const char* start[2] = { "a", "k00000000000000010000" };
-    size_t start_len[2] = { 1, 21 };
-    const char* limit[2] = { "k00000000000000010000", "z" };
-    size_t limit_len[2] = { 21, 1 };
+    const char* start[2] = {"a", "k00000000000000010000"};
+    size_t start_len[2] = {1, 21};
+    const char* limit[2] = {"k00000000000000010000", "z"};
+    size_t limit_len[2] = {21, 1};
     rocksdb_writeoptions_set_sync(woptions, 0);
     for (i = 0; i < n; i++) {
       snprintf(keybuf, sizeof(keybuf), "k%020d", i);
@@ -1393,8 +1379,8 @@ int main(int argc, char** argv) {
                                                   factory);
     db = CheckCompaction(db, options_with_filter_factory, roptions, woptions);
 
-    rocksdb_options_set_compaction_filter_factory(
-        options_with_filter_factory, NULL);
+    rocksdb_options_set_compaction_filter_factory(options_with_filter_factory,
+                                                  NULL);
     rocksdb_options_destroy(options_with_filter_factory);
   }
 
@@ -1449,7 +1435,8 @@ int main(int argc, char** argv) {
     rocksdb_close(db);
 
     size_t cflen;
-    char** column_fams = rocksdb_list_column_families(db_options, dbname, &cflen, &err);
+    char** column_fams =
+        rocksdb_list_column_families(db_options, dbname, &cflen, &err);
     CheckNoError(err);
     CheckEqual("default", column_fams[0], 7);
     CheckEqual("cf1", column_fams[1], 3);
@@ -1465,7 +1452,8 @@ int main(int argc, char** argv) {
     LoadAndCheckLatestOptions(dbname, env, false, cache, NULL, 2, cf_names,
                               NULL);
 
-    db = rocksdb_open_column_families(db_options, dbname, 2, cf_names, cf_opts, handles, &err);
+    db = rocksdb_open_column_families(db_options, dbname, 2, cf_names, cf_opts,
+                                      handles, &err);
     CheckNoError(err);
 
     rocksdb_put_cf(db, woptions, handles[1], "foo", 3, "hello", 5, &err);
@@ -1483,11 +1471,10 @@ int main(int argc, char** argv) {
                                      &err);
     CheckNoError(err);
 
-    rocksdb_flushoptions_t *flush_options = rocksdb_flushoptions_create();
+    rocksdb_flushoptions_t* flush_options = rocksdb_flushoptions_create();
     rocksdb_flushoptions_set_wait(flush_options, 1);
     rocksdb_flush_cf(db, flush_options, handles[1], &err);
-    CheckNoError(err)
-    rocksdb_flushoptions_destroy(flush_options);
+    CheckNoError(err) rocksdb_flushoptions_destroy(flush_options);
 
     CheckGetCF(db, roptions, handles[1], "foo", "hello");
     CheckPinGetCF(db, roptions, handles[1], "foo", "hello");
@@ -1524,27 +1511,29 @@ int main(int argc, char** argv) {
     rocksdb_flush_wal(db, 1, &err);
     CheckNoError(err);
 
-    const char* keys[3] = { "box", "box", "barfooxx" };
-    const rocksdb_column_family_handle_t* get_handles[3] = { handles[0], handles[1], handles[1] };
-    const size_t keys_sizes[3] = { 3, 3, 8 };
+    const char* keys[3] = {"box", "box", "barfooxx"};
+    const rocksdb_column_family_handle_t* get_handles[3] = {
+        handles[0], handles[1], handles[1]};
+    const size_t keys_sizes[3] = {3, 3, 8};
     char* vals[3];
     size_t vals_sizes[3];
     char* errs[3];
-    rocksdb_multi_get_cf(db, roptions, get_handles, 3, keys, keys_sizes, vals, vals_sizes, errs);
+    rocksdb_multi_get_cf(db, roptions, get_handles, 3, keys, keys_sizes, vals,
+                         vals_sizes, errs);
 
     int i;
     for (i = 0; i < 3; i++) {
       CheckEqual(NULL, errs[i], 0);
       switch (i) {
-      case 0:
-        CheckEqual(NULL, vals[i], vals_sizes[i]); // wrong cf
-        break;
-      case 1:
-        CheckEqual("c", vals[i], vals_sizes[i]); // bingo
-        break;
-      case 2:
-        CheckEqual(NULL, vals[i], vals_sizes[i]); // normal not found
-        break;
+        case 0:
+          CheckEqual(NULL, vals[i], vals_sizes[i]);  // wrong cf
+          break;
+        case 1:
+          CheckEqual("c", vals[i], vals_sizes[i]);  // bingo
+          break;
+        case 2:
+          CheckEqual(NULL, vals[i], vals_sizes[i]);  // normal not found
+          break;
       }
       Free(&vals[i]);
     }
@@ -1592,7 +1581,8 @@ int main(int argc, char** argv) {
       }
     }
 
-    rocksdb_iterator_t* iter = rocksdb_create_iterator_cf(db, roptions, handles[1]);
+    rocksdb_iterator_t* iter =
+        rocksdb_create_iterator_cf(db, roptions, handles[1]);
     CheckCondition(!rocksdb_iter_valid(iter));
     rocksdb_iter_seek_to_first(iter);
     CheckCondition(rocksdb_iter_valid(iter));
@@ -1605,9 +1595,11 @@ int main(int argc, char** argv) {
     CheckNoError(err);
     rocksdb_iter_destroy(iter);
 
-    rocksdb_column_family_handle_t* iters_cf_handles[2] = { handles[0], handles[1] };
+    rocksdb_column_family_handle_t* iters_cf_handles[2] = {handles[0],
+                                                           handles[1]};
     rocksdb_iterator_t* iters_handles[2];
-    rocksdb_create_iterators(db, roptions, iters_cf_handles, iters_handles, 2, &err);
+    rocksdb_create_iterators(db, roptions, iters_cf_handles, iters_handles, 2,
+                             &err);
     CheckNoError(err);
 
     iter = iters_handles[0];
@@ -1652,7 +1644,8 @@ int main(int argc, char** argv) {
   {
     // Create new database
     rocksdb_options_set_allow_mmap_reads(options, 1);
-    rocksdb_options_set_prefix_extractor(options, rocksdb_slicetransform_create_fixed_prefix(3));
+    rocksdb_options_set_prefix_extractor(
+        options, rocksdb_slicetransform_create_fixed_prefix(3));
     rocksdb_options_set_hash_skip_list_rep(options, 5000, 4, 4);
     rocksdb_options_set_plain_table_factory(options, 4, 10, 0.75, 16);
     rocksdb_options_set_allow_concurrent_memtable_write(options, 0);
@@ -1747,8 +1740,9 @@ int main(int argc, char** argv) {
     // amount of memory used within memtables should grow
     CheckCondition(rocksdb_approximate_memory_usage_get_mem_table_total(mu2) >=
                    rocksdb_approximate_memory_usage_get_mem_table_total(mu1));
-    CheckCondition(rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu2) >=
-                   rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu1));
+    CheckCondition(
+        rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu2) >=
+        rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu1));
 
     rocksdb_memory_consumers_destroy(consumers);
     rocksdb_approximate_memory_usage_destroy(mu1);
@@ -2578,6 +2572,9 @@ int main(int argc, char** argv) {
     rocksdb_readoptions_set_io_timeout(ro, 400);
     CheckCondition(400 == rocksdb_readoptions_get_io_timeout(ro));
 
+    rocksdb_readoptions_set_async_io(ro, 1);
+    CheckCondition(1 == rocksdb_readoptions_get_async_io(ro));
+
     rocksdb_readoptions_destroy(ro);
   }
 
@@ -2839,53 +2836,57 @@ int main(int argc, char** argv) {
     db = rocksdb_open(options, dbname, &err);
     CheckNoError(err);
 
-    rocksdb_put(db, woptions, "a",    1, "0",    1, &err); CheckNoError(err);
-    rocksdb_put(db, woptions, "foo",  3, "bar",  3, &err); CheckNoError(err);
-    rocksdb_put(db, woptions, "foo1", 4, "bar1", 4, &err); CheckNoError(err);
-    rocksdb_put(db, woptions, "g1",   2, "0",    1, &err); CheckNoError(err);
+    rocksdb_put(db, woptions, "a", 1, "0", 1, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo", 3, "bar", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo1", 4, "bar1", 4, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "g1", 2, "0", 1, &err);
+    CheckNoError(err);
 
     // testing basic case with no iterate_upper_bound and no prefix_extractor
     {
-       rocksdb_readoptions_set_iterate_upper_bound(roptions, NULL, 0);
-       rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
+      rocksdb_readoptions_set_iterate_upper_bound(roptions, NULL, 0);
+      rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
 
-       rocksdb_iter_seek(iter, "foo", 3);
-       CheckCondition(rocksdb_iter_valid(iter));
-       CheckIter(iter, "foo", "bar");
+      rocksdb_iter_seek(iter, "foo", 3);
+      CheckCondition(rocksdb_iter_valid(iter));
+      CheckIter(iter, "foo", "bar");
 
-       rocksdb_iter_next(iter);
-       CheckCondition(rocksdb_iter_valid(iter));
-       CheckIter(iter, "foo1", "bar1");
+      rocksdb_iter_next(iter);
+      CheckCondition(rocksdb_iter_valid(iter));
+      CheckIter(iter, "foo1", "bar1");
 
-       rocksdb_iter_next(iter);
-       CheckCondition(rocksdb_iter_valid(iter));
-       CheckIter(iter, "g1", "0");
+      rocksdb_iter_next(iter);
+      CheckCondition(rocksdb_iter_valid(iter));
+      CheckIter(iter, "g1", "0");
 
-       rocksdb_iter_destroy(iter);
+      rocksdb_iter_destroy(iter);
     }
 
     // testing iterate_upper_bound and forward iterator
     // to make sure it stops at bound
     {
-       // iterate_upper_bound points beyond the last expected entry
-       rocksdb_readoptions_set_iterate_upper_bound(roptions, "foo2", 4);
+      // iterate_upper_bound points beyond the last expected entry
+      rocksdb_readoptions_set_iterate_upper_bound(roptions, "foo2", 4);
 
-       rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
+      rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
 
-       rocksdb_iter_seek(iter, "foo", 3);
-       CheckCondition(rocksdb_iter_valid(iter));
-       CheckIter(iter, "foo", "bar");
+      rocksdb_iter_seek(iter, "foo", 3);
+      CheckCondition(rocksdb_iter_valid(iter));
+      CheckIter(iter, "foo", "bar");
 
-       rocksdb_iter_next(iter);
-       CheckCondition(rocksdb_iter_valid(iter));
-       CheckIter(iter, "foo1", "bar1");
+      rocksdb_iter_next(iter);
+      CheckCondition(rocksdb_iter_valid(iter));
+      CheckIter(iter, "foo1", "bar1");
 
-       rocksdb_iter_next(iter);
-       // should stop here...
-       CheckCondition(!rocksdb_iter_valid(iter));
+      rocksdb_iter_next(iter);
+      // should stop here...
+      CheckCondition(!rocksdb_iter_valid(iter));
 
-       rocksdb_iter_destroy(iter);
-       rocksdb_readoptions_set_iterate_upper_bound(roptions, NULL, 0);
+      rocksdb_iter_destroy(iter);
+      rocksdb_readoptions_set_iterate_upper_bound(roptions, NULL, 0);
     }
   }
 
@@ -3009,7 +3010,7 @@ int main(int argc, char** argv) {
     snapshot = rocksdb_transactiondb_create_snapshot(txn_db);
     rocksdb_readoptions_set_snapshot(roptions, snapshot);
 
-    rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hey", 3,  &err);
+    rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hey", 3, &err);
     CheckNoError(err);
 
     CheckTxnDBGet(txn_db, roptions, "foo", "hello");
@@ -3021,7 +3022,8 @@ int main(int argc, char** argv) {
 
     // iterate
     rocksdb_transaction_put(txn, "bar", 3, "hi", 2, &err);
-    rocksdb_iterator_t* iter = rocksdb_transaction_create_iterator(txn, roptions);
+    rocksdb_iterator_t* iter =
+        rocksdb_transaction_create_iterator(txn, roptions);
     CheckCondition(!rocksdb_iter_valid(iter));
     rocksdb_iter_seek_to_first(iter);
     CheckCondition(rocksdb_iter_valid(iter));

db/column_family.cc

@@ -192,8 +192,7 @@ Status CheckCFPathsSupported(const DBOptions& db_options,
       return Status::NotSupported(
           "More than one CF paths are only supported in "
           "universal and level compaction styles. ");
-    } else if (cf_options.cf_paths.empty() &&
-               db_options.db_paths.size() > 1) {
+    } else if (cf_options.cf_paths.empty() && db_options.db_paths.size() > 1) {
       return Status::NotSupported(
           "More than one DB paths are only supported in "
           "universal and level compaction styles. ");
@@ -205,7 +204,7 @@ Status CheckCFPathsSupported(const DBOptions& db_options,
 namespace {
 const uint64_t kDefaultTtl = 0xfffffffffffffffe;
 const uint64_t kDefaultPeriodicCompSecs = 0xfffffffffffffffe;
-}  // namespace
+}  // anonymous namespace
 
 ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                     const ColumnFamilyOptions& src) {
@@ -233,6 +232,21 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
     result.min_write_buffer_number_to_merge = 1;
   }
 
+  if (db_options.atomic_flush && result.min_write_buffer_number_to_merge > 1) {
+    ROCKS_LOG_WARN(
+        db_options.logger,
+        "Currently, if atomic_flush is true, then triggering flush for any "
+        "column family internally (non-manual flush) will trigger flushing "
+        "all column families even if the number of memtables is smaller "
+        "min_write_buffer_number_to_merge. Therefore, configuring "
+        "min_write_buffer_number_to_merge > 1 is not compatible and should "
+        "be satinized to 1. Not doing so will lead to data loss and "
+        "inconsistent state across multiple column families when WAL is "
+        "disabled, which is a common setting for atomic flush");
+
+    result.min_write_buffer_number_to_merge = 1;
+  }
+
   if (result.num_levels < 1) {
     result.num_levels = 1;
   }
@@ -276,7 +290,6 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
   }
 
   if (result.compaction_style == kCompactionStyleFIFO) {
-    result.num_levels = 1;
     // since we delete level0 files in FIFO compaction when there are too many
     // of them, these options don't really mean anything
     result.level0_slowdown_writes_trigger = std::numeric_limits<int>::max();
@@ -339,7 +352,8 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
   // were not deleted yet, when we open the DB we will find these .trash files
   // and schedule them to be deleted (or delete immediately if SstFileManager
   // was not used)
-  auto sfm = static_cast<SstFileManagerImpl*>(db_options.sst_file_manager.get());
+  auto sfm =
+      static_cast<SstFileManagerImpl*>(db_options.sst_file_manager.get());
   for (size_t i = 0; i < result.cf_paths.size(); i++) {
     DeleteScheduler::CleanupDirectory(db_options.env, sfm,
                                       result.cf_paths[i].path)
@@ -543,7 +557,6 @@ ColumnFamilyData::ColumnFamilyData(
       next_(nullptr),
       prev_(nullptr),
       log_number_(0),
-      flush_reason_(FlushReason::kOthers),
       column_family_set_(column_family_set),
       queued_for_flush_(false),
       queued_for_compaction_(false),
@@ -551,7 +564,8 @@ ColumnFamilyData::ColumnFamilyData(
       allow_2pc_(db_options.allow_2pc),
       last_memtable_id_(0),
       db_paths_registered_(false),
-      mempurge_used_(false) {
+      mempurge_used_(false),
+      next_epoch_number_(1) {
   if (id_ != kDummyColumnFamilyDataId) {
     // TODO(cc): RegisterDbPaths can be expensive, considering moving it
     // outside of this constructor which might be called with db mutex held.
@@ -596,8 +610,8 @@ ColumnFamilyData::ColumnFamilyData(
       compaction_picker_.reset(
           new FIFOCompactionPicker(ioptions_, &internal_comparator_));
     } else if (ioptions_.compaction_style == kCompactionStyleNone) {
-      compaction_picker_.reset(new NullCompactionPicker(
-          ioptions_, &internal_comparator_));
+      compaction_picker_.reset(
+          new NullCompactionPicker(ioptions_, &internal_comparator_));
       ROCKS_LOG_WARN(ioptions_.logger,
                      "Column family %s does not use any background compaction. "
                      "Compactions can only be done via CompactFiles\n",
@@ -864,7 +878,7 @@ int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
     return static_cast<int>(res);
   }
 }
-}  // namespace
+}  // anonymous namespace
 
 std::pair<WriteStallCondition, ColumnFamilyData::WriteStallCause>
 ColumnFamilyData::GetWriteStallConditionAndCause(
@@ -905,7 +919,7 @@ ColumnFamilyData::GetWriteStallConditionAndCause(
 }
 
 WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
-      const MutableCFOptions& mutable_cf_options) {
+    const MutableCFOptions& mutable_cf_options) {
   auto write_stall_condition = WriteStallCondition::kNormal;
   if (current_ != nullptr) {
     auto* vstorage = current_->storage_info();
@@ -998,7 +1012,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
           mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
           (compaction_needed_bytes -
            mutable_cf_options.soft_pending_compaction_bytes_limit) >
-              3 * (mutable_cf_options.hard_pending_compaction_bytes_limit -
+              3 *
+                  (mutable_cf_options.hard_pending_compaction_bytes_limit -
                    mutable_cf_options.soft_pending_compaction_bytes_limit) /
                   4;
 
@@ -1113,12 +1128,9 @@ bool ColumnFamilyData::NeedsCompaction() const {
 Compaction* ColumnFamilyData::PickCompaction(
     const MutableCFOptions& mutable_options,
     const MutableDBOptions& mutable_db_options, LogBuffer* log_buffer) {
-  SequenceNumber earliest_mem_seqno =
-      std::min(mem_->GetEarliestSequenceNumber(),
-               imm_.current()->GetEarliestSequenceNumber(false));
   auto* result = compaction_picker_->PickCompaction(
       GetName(), mutable_options, mutable_db_options, current_->storage_info(),
-      log_buffer, earliest_mem_seqno);
+      log_buffer);
   if (result != nullptr) {
     result->SetInputVersion(current_);
   }
@@ -1144,7 +1156,8 @@ Status ColumnFamilyData::RangesOverlapWithMemtables(
   MergeIteratorBuilder merge_iter_builder(&internal_comparator_, &arena);
   merge_iter_builder.AddIterator(
       super_version->mem->NewIterator(read_opts, &arena));
-  super_version->imm->AddIterators(read_opts, &merge_iter_builder);
+  super_version->imm->AddIterators(read_opts, &merge_iter_builder,
+                                   false /* add_range_tombstone_iter */);
   ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
 
   auto read_seq = super_version->current->version_set()->LastSequence();
@@ -1204,6 +1217,7 @@ Compaction* ColumnFamilyData::CompactRange(
   if (result != nullptr) {
     result->SetInputVersion(current_);
   }
+  TEST_SYNC_POINT("ColumnFamilyData::CompactRange:Return");
   return result;
 }
 
@@ -1287,8 +1301,8 @@ bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
   return false;
 }
 
-void ColumnFamilyData::InstallSuperVersion(
-    SuperVersionContext* sv_context, InstrumentedMutex* db_mutex) {
+void ColumnFamilyData::InstallSuperVersion(SuperVersionContext* sv_context,
+                                           InstrumentedMutex* db_mutex) {
   db_mutex->AssertHeld();
   return InstallSuperVersion(sv_context, mutable_cf_options_);
 }
@@ -1415,6 +1429,14 @@ Status ColumnFamilyData::ValidateOptions(
         "FIFO compaction only supported with max_open_files = -1.");
   }
 
+  std::vector<uint32_t> supported{0, 1, 2, 4, 8};
+  if (std::find(supported.begin(), supported.end(),
+                cf_options.memtable_protection_bytes_per_key) ==
+      supported.end()) {
+    return Status::NotSupported(
+        "Memtable per key-value checksum protection only supports 0, 1, 2, 4 "
+        "or 8 bytes per key.");
+  }
   return s;
 }
 
@@ -1457,8 +1479,8 @@ Env::WriteLifeTimeHint ColumnFamilyData::CalculateSSTWriteHint(int level) {
     // than base_level.
     return Env::WLTH_MEDIUM;
   }
-  return static_cast<Env::WriteLifeTimeHint>(level - base_level +
-                            static_cast<int>(Env::WLTH_MEDIUM));
+  return static_cast<Env::WriteLifeTimeHint>(
+      level - base_level + static_cast<int>(Env::WLTH_MEDIUM));
 }
 
 Status ColumnFamilyData::AddDirectories(
@@ -1496,6 +1518,13 @@ FSDirectory* ColumnFamilyData::GetDataDir(size_t path_id) const {
   return data_dirs_[path_id].get();
 }
 
+void ColumnFamilyData::RecoverEpochNumbers() {
+  assert(current_);
+  auto* vstorage = current_->storage_info();
+  assert(vstorage);
+  vstorage->RecoverEpochNumbers(this);
+}
+
 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                  const ImmutableDBOptions* db_options,
                                  const FileOptions& file_options,
@@ -1554,8 +1583,8 @@ ColumnFamilyData* ColumnFamilySet::GetColumnFamily(uint32_t id) const {
   }
 }
 
-ColumnFamilyData* ColumnFamilySet::GetColumnFamily(const std::string& name)
-    const {
+ColumnFamilyData* ColumnFamilySet::GetColumnFamily(
+    const std::string& name) const {
   auto cfd_iter = column_families_.find(name);
   if (cfd_iter != column_families_.end()) {
     auto cfd = GetColumnFamily(cfd_iter->second);

db/column_family.h

@@ -163,8 +163,8 @@ extern const double kIncSlowdownRatio;
 class ColumnFamilyHandleImpl : public ColumnFamilyHandle {
  public:
   // create while holding the mutex
-  ColumnFamilyHandleImpl(
-      ColumnFamilyData* cfd, DBImpl* db, InstrumentedMutex* mutex);
+  ColumnFamilyHandleImpl(ColumnFamilyData* cfd, DBImpl* db,
+                         InstrumentedMutex* mutex);
   // destroy without mutex
   virtual ~ColumnFamilyHandleImpl();
   virtual ColumnFamilyData* cfd() const { return cfd_; }
@@ -189,7 +189,8 @@ class ColumnFamilyHandleImpl : public ColumnFamilyHandle {
 class ColumnFamilyHandleInternal : public ColumnFamilyHandleImpl {
  public:
   ColumnFamilyHandleInternal()
-      : ColumnFamilyHandleImpl(nullptr, nullptr, nullptr), internal_cfd_(nullptr) {}
+      : ColumnFamilyHandleImpl(nullptr, nullptr, nullptr),
+        internal_cfd_(nullptr) {}
 
   void SetCFD(ColumnFamilyData* _cfd) { internal_cfd_ = _cfd; }
   virtual ColumnFamilyData* cfd() const override { return internal_cfd_; }
@@ -309,10 +310,6 @@ class ColumnFamilyData {
   void SetLogNumber(uint64_t log_number) { log_number_ = log_number; }
   uint64_t GetLogNumber() const { return log_number_; }
 
-  void SetFlushReason(FlushReason flush_reason) {
-    flush_reason_ = flush_reason;
-  }
-  FlushReason GetFlushReason() const { return flush_reason_; }
   // thread-safe
   const FileOptions* soptions() const;
   const ImmutableOptions* ioptions() const { return &ioptions_; }
@@ -357,7 +354,7 @@ class ColumnFamilyData {
   Version* current() { return current_; }
   Version* dummy_versions() { return dummy_versions_; }
   void SetCurrent(Version* _current);
-  uint64_t GetNumLiveVersions() const;  // REQUIRE: DB mutex held
+  uint64_t GetNumLiveVersions() const;    // REQUIRE: DB mutex held
   uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
   uint64_t GetLiveSstFilesSize() const;   // REQUIRE: DB mutex held
   uint64_t GetTotalBlobFileSize() const;  // REQUIRE: DB mutex held
@@ -532,6 +529,24 @@ class ColumnFamilyData {
   void SetMempurgeUsed() { mempurge_used_ = true; }
   bool GetMempurgeUsed() { return mempurge_used_; }
 
+  // Allocate and return a new epoch number
+  uint64_t NewEpochNumber() { return next_epoch_number_.fetch_add(1); }
+
+  // Get the next epoch number to be assigned
+  uint64_t GetNextEpochNumber() const { return next_epoch_number_.load(); }
+
+  // Set the next epoch number to be assigned
+  void SetNextEpochNumber(uint64_t next_epoch_number) {
+    next_epoch_number_.store(next_epoch_number);
+  }
+
+  // Reset the next epoch number to be assigned
+  void ResetNextEpochNumber() { next_epoch_number_.store(1); }
+
+  // Recover the next epoch number of this CF and epoch number
+  // of its files (if missing)
+  void RecoverEpochNumbers();
+
  private:
   friend class ColumnFamilySet;
   ColumnFamilyData(uint32_t id, const std::string& name,
@@ -552,7 +567,7 @@ class ColumnFamilyData {
   Version* dummy_versions_;  // Head of circular doubly-linked list of versions.
   Version* current_;         // == dummy_versions->prev_
 
-  std::atomic<int> refs_;      // outstanding references to ColumnFamilyData
+  std::atomic<int> refs_;  // outstanding references to ColumnFamilyData
   std::atomic<bool> initialized_;
   std::atomic<bool> dropped_;  // true if client dropped it
 
@@ -597,8 +612,6 @@ class ColumnFamilyData {
   // recovered from
   uint64_t log_number_;
 
-  std::atomic<FlushReason> flush_reason_;
-
   // An object that keeps all the compaction stats
   // and picks the next compaction
   std::unique_ptr<CompactionPicker> compaction_picker_;
@@ -633,6 +646,8 @@ class ColumnFamilyData {
   // a Version associated with this CFD
   std::shared_ptr<CacheReservationManager> file_metadata_cache_res_mgr_;
   bool mempurge_used_;
+
+  std::atomic<uint64_t> next_epoch_number_;
 };
 
 // ColumnFamilySet has interesting thread-safety requirements
@@ -656,8 +671,7 @@ class ColumnFamilySet {
   // ColumnFamilySet supports iteration
   class iterator {
    public:
-    explicit iterator(ColumnFamilyData* cfd)
-        : current_(cfd) {}
+    explicit iterator(ColumnFamilyData* cfd) : current_(cfd) {}
     // NOTE: minimum operators for for-loop iteration
     iterator& operator++() {
       current_ = current_->next_;

db/column_family_test.cc

@@ -39,9 +39,7 @@ class EnvCounter : public SpecialEnv {
  public:
   explicit EnvCounter(Env* base)
       : SpecialEnv(base), num_new_writable_file_(0) {}
-  int GetNumberOfNewWritableFileCalls() {
-    return num_new_writable_file_;
-  }
+  int GetNumberOfNewWritableFileCalls() { return num_new_writable_file_; }
   Status NewWritableFile(const std::string& f, std::unique_ptr<WritableFile>* r,
                          const EnvOptions& soptions) override {
     ++num_new_writable_file_;
@@ -187,7 +185,7 @@ class ColumnFamilyTestBase : public testing::Test {
   }
 
   Status OpenReadOnly(std::vector<std::string> cf,
-                         std::vector<ColumnFamilyOptions> options = {}) {
+                      std::vector<ColumnFamilyOptions> options = {}) {
     std::vector<ColumnFamilyDescriptor> column_families;
     names_.clear();
     for (size_t i = 0; i < cf.size(); ++i) {
@@ -201,20 +199,17 @@ class ColumnFamilyTestBase : public testing::Test {
 
 #ifndef ROCKSDB_LITE  // ReadOnlyDB is not supported
   void AssertOpenReadOnly(std::vector<std::string> cf,
-                    std::vector<ColumnFamilyOptions> options = {}) {
+                          std::vector<ColumnFamilyOptions> options = {}) {
     ASSERT_OK(OpenReadOnly(cf, options));
   }
 #endif  // !ROCKSDB_LITE
 
-
   void Open(std::vector<std::string> cf,
             std::vector<ColumnFamilyOptions> options = {}) {
     ASSERT_OK(TryOpen(cf, options));
   }
 
-  void Open() {
-    Open({"default"});
-  }
+  void Open() { Open({"default"}); }
 
   DBImpl* dbfull() { return static_cast_with_check<DBImpl>(db_); }
 
@@ -253,7 +248,7 @@ class ColumnFamilyTestBase : public testing::Test {
   }
 
   void Destroy(const std::vector<ColumnFamilyDescriptor>& column_families =
-                  std::vector<ColumnFamilyDescriptor>()) {
+                   std::vector<ColumnFamilyDescriptor>()) {
     Close();
     ASSERT_OK(DestroyDB(dbname_, Options(db_options_, column_family_options_),
                         column_families));
@@ -335,9 +330,7 @@ class ColumnFamilyTestBase : public testing::Test {
     ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[cf]));
   }
 
-  void WaitForCompaction() {
-    ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  }
+  void WaitForCompaction() { ASSERT_OK(dbfull()->TEST_WaitForCompact()); }
 
   uint64_t MaxTotalInMemoryState() {
     return dbfull()->TEST_MaxTotalInMemoryState();
@@ -354,9 +347,7 @@ class ColumnFamilyTestBase : public testing::Test {
   Status Merge(int cf, const std::string& key, const std::string& value) {
     return db_->Merge(WriteOptions(), handles_[cf], Slice(key), Slice(value));
   }
-  Status Flush(int cf) {
-    return db_->Flush(FlushOptions(), handles_[cf]);
-  }
+  Status Flush(int cf) { return db_->Flush(FlushOptions(), handles_[cf]); }
 
   std::string Get(int cf, const std::string& key) {
     ReadOptions options;
@@ -409,8 +400,8 @@ class ColumnFamilyTestBase : public testing::Test {
 #ifndef ROCKSDB_LITE
     ASSERT_EQ(value, FilesPerLevel(cf));
 #else
-    (void) value;
-    (void) cf;
+    (void)value;
+    (void)cf;
 #endif
   }
 
@@ -426,7 +417,7 @@ class ColumnFamilyTestBase : public testing::Test {
 #ifndef ROCKSDB_LITE
     ASSERT_EQ(expected_value, CountLiveFiles());
 #else
-    (void) expected_value;
+    (void)expected_value;
 #endif
   }
 
@@ -476,7 +467,7 @@ class ColumnFamilyTestBase : public testing::Test {
 #ifndef ROCKSDB_LITE  // GetSortedWalFiles is not supported
     ASSERT_EQ(value, CountLiveLogFiles());
 #else
-    (void) value;
+    (void)value;
 #endif  // !ROCKSDB_LITE
   }
 
@@ -521,14 +512,14 @@ class ColumnFamilyTestBase : public testing::Test {
     return static_cast<int>(files.size());
   }
 
-  void RecalculateWriteStallConditions(ColumnFamilyData* cfd,
-      const MutableCFOptions& mutable_cf_options)  {
+  void RecalculateWriteStallConditions(
+      ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options) {
     // add lock to avoid race condition between
     // `RecalculateWriteStallConditions` which writes to CFStats and
     // background `DBImpl::DumpStats()` threads which read CFStats
     dbfull()->TEST_LockMutex();
     cfd->RecalculateWriteStallConditions(mutable_cf_options);
-    dbfull()-> TEST_UnlockMutex();
+    dbfull()->TEST_UnlockMutex();
   }
 
   std::vector<ColumnFamilyHandle*> handles_;
@@ -970,8 +961,7 @@ TEST_P(ColumnFamilyTest, FlushTest) {
     }
 
     for (int i = 0; i < 3; ++i) {
-      uint64_t max_total_in_memory_state =
-          MaxTotalInMemoryState();
+      uint64_t max_total_in_memory_state = MaxTotalInMemoryState();
       ASSERT_OK(Flush(i));
       AssertMaxTotalInMemoryState(max_total_in_memory_state);
     }
@@ -1209,7 +1199,7 @@ TEST_P(ColumnFamilyTest, DifferentWriteBufferSizes) {
   WaitForFlush(2);
   AssertNumberOfImmutableMemtables({0, 0, 0, 0});
   AssertCountLiveLogFiles(12);
-  PutRandomData(1, 2*200, 1000);
+  PutRandomData(1, 2 * 200, 1000);
   WaitForFlush(1);
   AssertNumberOfImmutableMemtables({0, 0, 0, 0});
   AssertCountLiveLogFiles(7);
@@ -2123,7 +2113,6 @@ TEST_P(ColumnFamilyTest, ReadOnlyDBTest) {
   ASSERT_EQ("bla", Get(1, "foo"));
   ASSERT_EQ("blablablabla", Get(2, "foo"));
 
-
   // test newiterators
   {
     std::vector<Iterator*> iterators;
@@ -2488,7 +2477,7 @@ void DropSingleColumnFamily(ColumnFamilyTest* cf_test, int cf_id,
   }
   test_stage = kChildThreadFinishDroppingColumnFamily;
 }
-}  // namespace
+}  // anonymous namespace
 
 TEST_P(ColumnFamilyTest, CreateAndDropRace) {
   const int kCfCount = 5;
@@ -3433,6 +3422,27 @@ TEST(ColumnFamilyTest, ValidateBlobGCForceThreshold) {
                   .IsInvalidArgument());
 }
 
+TEST(ColumnFamilyTest, ValidateMemtableKVChecksumOption) {
+  DBOptions db_options;
+
+  ColumnFamilyOptions cf_options;
+  ASSERT_OK(ColumnFamilyData::ValidateOptions(db_options, cf_options));
+
+  cf_options.memtable_protection_bytes_per_key = 5;
+  ASSERT_TRUE(ColumnFamilyData::ValidateOptions(db_options, cf_options)
+                  .IsNotSupported());
+
+  cf_options.memtable_protection_bytes_per_key = 1;
+  ASSERT_OK(ColumnFamilyData::ValidateOptions(db_options, cf_options));
+
+  cf_options.memtable_protection_bytes_per_key = 16;
+  ASSERT_TRUE(ColumnFamilyData::ValidateOptions(db_options, cf_options)
+                  .IsNotSupported());
+
+  cf_options.memtable_protection_bytes_per_key = 0;
+  ASSERT_OK(ColumnFamilyData::ValidateOptions(db_options, cf_options));
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/compact_files_test.cc

@@ -348,9 +348,7 @@ TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
       return true;
     }
 
-    void SetDB(DB* db) {
-      db_ = db;
-    }
+    void SetDB(DB* db) { db_ = db; }
 
     const char* Name() const override { return "FilterWithGet"; }
 
@@ -358,7 +356,6 @@ TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
     DB* db_;
   };
 
-
   std::shared_ptr<FilterWithGet> cf(new FilterWithGet());
 
   Options options;
@@ -385,7 +382,6 @@ TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
         db->CompactFiles(ROCKSDB_NAMESPACE::CompactionOptions(), {fname}, 0));
   }
 
-
   delete db;
 }
 
@@ -400,10 +396,9 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
   }
   // Check that passing `CompressionType::kDisableCompressionOption` to
   // `CompactFiles` causes it to use the column family compression options.
-  for (auto compaction_style :
-       {CompactionStyle::kCompactionStyleLevel,
-        CompactionStyle::kCompactionStyleUniversal,
-        CompactionStyle::kCompactionStyleNone}) {
+  for (auto compaction_style : {CompactionStyle::kCompactionStyleLevel,
+                                CompactionStyle::kCompactionStyleUniversal,
+                                CompactionStyle::kCompactionStyleNone}) {
     ASSERT_OK(DestroyDB(db_name_, Options()));
     Options options;
     options.compaction_style = compaction_style;
@@ -490,6 +485,7 @@ TEST_F(CompactFilesTest, GetCompactionJobInfo) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/compaction/clipping_iterator_test.cc

@@ -253,6 +253,7 @@ INSTANTIATE_TEST_CASE_P(
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/compaction/compaction.cc

@@ -221,7 +221,7 @@ Compaction::Compaction(
     : input_vstorage_(vstorage),
       start_level_(_inputs[0].level),
       output_level_(_output_level),
-      max_output_file_size_(_target_file_size),
+      target_output_file_size_(_target_file_size),
       max_compaction_bytes_(_max_compaction_bytes),
       max_subcompactions_(_max_subcompactions),
       immutable_options_(_immutable_options),
@@ -238,12 +238,19 @@ Compaction::Compaction(
       inputs_(PopulateWithAtomicBoundaries(vstorage, std::move(_inputs))),
       grandparents_(std::move(_grandparents)),
       score_(_score),
-      bottommost_level_(IsBottommostLevel(output_level_, vstorage, inputs_)),
+      bottommost_level_(
+          // For simplicity, we don't support the concept of "bottommost level"
+          // with
+          // `CompactionReason::kExternalSstIngestion` and
+          // `CompactionReason::kRefitLevel`
+          (_compaction_reason == CompactionReason::kExternalSstIngestion ||
+           _compaction_reason == CompactionReason::kRefitLevel)
+              ? false
+              : IsBottommostLevel(output_level_, vstorage, inputs_)),
       is_full_compaction_(IsFullCompaction(vstorage, inputs_)),
       is_manual_compaction_(_manual_compaction),
       trim_ts_(_trim_ts),
       is_trivial_move_(false),
-
       compaction_reason_(_compaction_reason),
       notify_on_compaction_completion_(false),
       enable_blob_garbage_collection_(
@@ -258,8 +265,15 @@ Compaction::Compaction(
                   _blob_garbage_collection_age_cutoff > 1
               ? mutable_cf_options()->blob_garbage_collection_age_cutoff
               : _blob_garbage_collection_age_cutoff),
-      penultimate_level_(EvaluatePenultimateLevel(
-          immutable_options_, start_level_, output_level_)) {
+      penultimate_level_(
+          // For simplicity, we don't support the concept of "penultimate level"
+          // with `CompactionReason::kExternalSstIngestion` and
+          // `CompactionReason::kRefitLevel`
+          _compaction_reason == CompactionReason::kExternalSstIngestion ||
+                  _compaction_reason == CompactionReason::kRefitLevel
+              ? Compaction::kInvalidLevel
+              : EvaluatePenultimateLevel(vstorage, immutable_options_,
+                                         start_level_, output_level_)) {
   MarkFilesBeingCompacted(true);
   if (is_manual_compaction_) {
     compaction_reason_ = CompactionReason::kManualCompaction;
@@ -268,6 +282,14 @@ Compaction::Compaction(
     max_subcompactions_ = _mutable_db_options.max_subcompactions;
   }
 
+  // for the non-bottommost levels, it tries to build files match the target
+  // file size, but not guaranteed. It could be 2x the size of the target size.
+  max_output_file_size_ =
+      bottommost_level_ || grandparents_.empty() ||
+              !_immutable_options.level_compaction_dynamic_file_size
+          ? target_output_file_size_
+          : 2 * target_output_file_size_;
+
 #ifndef NDEBUG
   for (size_t i = 1; i < inputs_.size(); ++i) {
     assert(inputs_[i].level > inputs_[i - 1].level);
@@ -314,9 +336,40 @@ void Compaction::PopulatePenultimateLevelOutputRange() {
     return;
   }
 
+  // exclude the last level, the range of all input levels is the safe range
+  // of keys that can be moved up.
+  int exclude_level = number_levels_ - 1;
+  penultimate_output_range_type_ = PenultimateOutputRangeType::kNonLastRange;
+
+  // For universal compaction, the penultimate_output_range could be extended if
+  // all penultimate level files are included in the compaction (which includes
+  // the case that the penultimate level is empty).
+  if (immutable_options_.compaction_style == kCompactionStyleUniversal) {
+    exclude_level = kInvalidLevel;
+    penultimate_output_range_type_ = PenultimateOutputRangeType::kFullRange;
+    std::set<uint64_t> penultimate_inputs;
+    for (const auto& input_lvl : inputs_) {
+      if (input_lvl.level == penultimate_level_) {
+        for (const auto& file : input_lvl.files) {
+          penultimate_inputs.emplace(file->fd.GetNumber());
+        }
+      }
+    }
+    auto penultimate_files = input_vstorage_->LevelFiles(penultimate_level_);
+    for (const auto& file : penultimate_files) {
+      if (penultimate_inputs.find(file->fd.GetNumber()) ==
+          penultimate_inputs.end()) {
+        exclude_level = number_levels_ - 1;
+        penultimate_output_range_type_ =
+            PenultimateOutputRangeType::kNonLastRange;
+        break;
+      }
+    }
+  }
+
   GetBoundaryKeys(input_vstorage_, inputs_,
                   &penultimate_level_smallest_user_key_,
-                  &penultimate_level_largest_user_key_, number_levels_ - 1);
+                  &penultimate_level_largest_user_key_, exclude_level);
 }
 
 Compaction::~Compaction() {
@@ -334,6 +387,8 @@ bool Compaction::SupportsPerKeyPlacement() const {
 
 int Compaction::GetPenultimateLevel() const { return penultimate_level_; }
 
+// smallest_key and largest_key include timestamps if user-defined timestamp is
+// enabled.
 bool Compaction::OverlapPenultimateLevelOutputRange(
     const Slice& smallest_key, const Slice& largest_key) const {
   if (!SupportsPerKeyPlacement()) {
@@ -342,21 +397,30 @@ bool Compaction::OverlapPenultimateLevelOutputRange(
   const Comparator* ucmp =
       input_vstorage_->InternalComparator()->user_comparator();
 
-  return ucmp->Compare(smallest_key, penultimate_level_largest_user_key_) <=
-             0 &&
-         ucmp->Compare(largest_key, penultimate_level_smallest_user_key_) >= 0;
+  return ucmp->CompareWithoutTimestamp(
+             smallest_key, penultimate_level_largest_user_key_) <= 0 &&
+         ucmp->CompareWithoutTimestamp(
+             largest_key, penultimate_level_smallest_user_key_) >= 0;
 }
 
+// key includes timestamp if user-defined timestamp is enabled.
 bool Compaction::WithinPenultimateLevelOutputRange(const Slice& key) const {
   if (!SupportsPerKeyPlacement()) {
     return false;
   }
 
+  if (penultimate_level_smallest_user_key_.empty() ||
+      penultimate_level_largest_user_key_.empty()) {
+    return false;
+  }
+
   const Comparator* ucmp =
       input_vstorage_->InternalComparator()->user_comparator();
 
-  return ucmp->Compare(key, penultimate_level_smallest_user_key_) >= 0 &&
-         ucmp->Compare(key, penultimate_level_largest_user_key_) <= 0;
+  return ucmp->CompareWithoutTimestamp(
+             key, penultimate_level_smallest_user_key_) >= 0 &&
+         ucmp->CompareWithoutTimestamp(
+             key, penultimate_level_largest_user_key_) <= 0;
 }
 
 bool Compaction::InputCompressionMatchesOutput() const {
@@ -730,7 +794,18 @@ uint64_t Compaction::MinInputFileOldestAncesterTime(
   return min_oldest_ancester_time;
 }
 
+uint64_t Compaction::MinInputFileEpochNumber() const {
+  uint64_t min_epoch_number = std::numeric_limits<uint64_t>::max();
+  for (const auto& inputs_per_level : inputs_) {
+    for (const auto& file : inputs_per_level.files) {
+      min_epoch_number = std::min(min_epoch_number, file->epoch_number);
+    }
+  }
+  return min_epoch_number;
+}
+
 int Compaction::EvaluatePenultimateLevel(
+    const VersionStorageInfo* vstorage,
     const ImmutableOptions& immutable_options, const int start_level,
     const int output_level) {
   // TODO: currently per_key_placement feature only support level and universal
@@ -745,7 +820,19 @@ int Compaction::EvaluatePenultimateLevel(
 
   int penultimate_level = output_level - 1;
   assert(penultimate_level < immutable_options.num_levels);
-  if (penultimate_level <= 0 || penultimate_level < start_level) {
+  if (penultimate_level <= 0) {
+    return kInvalidLevel;
+  }
+
+  // If the penultimate level is not within input level -> output level range
+  // check if the penultimate output level is empty, if it's empty, it could
+  // also be locked for the penultimate output.
+  // TODO: ideally, it only needs to check if there's a file within the
+  //  compaction output key range. For simplicity, it just check if there's any
+  //  file on the penultimate level.
+  if (start_level == immutable_options.num_levels - 1 &&
+      (immutable_options.compaction_style != kCompactionStyleUniversal ||
+       !vstorage->LevelFiles(penultimate_level).empty())) {
     return kInvalidLevel;
   }
 

db/compaction/compaction.h

@@ -87,6 +87,15 @@ class Compaction {
                  BlobGarbageCollectionPolicy::kUseDefault,
              double blob_garbage_collection_age_cutoff = -1);
 
+  // The type of the penultimate level output range
+  enum class PenultimateOutputRangeType : int {
+    kNotSupported,  // it cannot output to the penultimate level
+    kFullRange,     // any data could be output to the penultimate level
+    kNonLastRange,  // only the keys within non_last_level compaction inputs can
+                    // be outputted to the penultimate level
+    kDisabled,      // no data can be outputted to the penultimate level
+  };
+
   // No copying allowed
   Compaction(const Compaction&) = delete;
   void operator=(const Compaction&) = delete;
@@ -163,6 +172,9 @@ class Compaction {
   // Maximum size of files to build during this compaction.
   uint64_t max_output_file_size() const { return max_output_file_size_; }
 
+  // Target output file size for this compaction
+  uint64_t target_output_file_size() const { return target_output_file_size_; }
+
   // What compression for output
   CompressionType output_compression() const { return output_compression_; }
 
@@ -307,6 +319,18 @@ class Compaction {
 
   Slice GetLargestUserKey() const { return largest_user_key_; }
 
+  Slice GetPenultimateLevelSmallestUserKey() const {
+    return penultimate_level_smallest_user_key_;
+  }
+
+  Slice GetPenultimateLevelLargestUserKey() const {
+    return penultimate_level_largest_user_key_;
+  }
+
+  PenultimateOutputRangeType GetPenultimateOutputRangeType() const {
+    return penultimate_output_range_type_;
+  }
+
   // Return true if the compaction supports per_key_placement
   bool SupportsPerKeyPlacement() const;
 
@@ -316,6 +340,8 @@ class Compaction {
 
   // Return true if the given range is overlap with penultimate level output
   // range.
+  // Both smallest_key and largest_key include timestamps if user-defined
+  // timestamp is enabled.
   bool OverlapPenultimateLevelOutputRange(const Slice& smallest_key,
                                           const Slice& largest_key) const;
 
@@ -325,6 +351,7 @@ class Compaction {
   // If per_key_placement is not supported, always return false.
   // TODO: currently it doesn't support moving data from the last level to the
   //  penultimate level
+  //  key includes timestamp if user-defined timestamp is enabled.
   bool WithinPenultimateLevelOutputRange(const Slice& key) const;
 
   CompactionReason compaction_reason() const { return compaction_reason_; }
@@ -351,6 +378,9 @@ class Compaction {
   // This is used to filter out some input files' ancester's time range.
   uint64_t MinInputFileOldestAncesterTime(const InternalKey* start,
                                           const InternalKey* end) const;
+  // Return the minimum epoch number among
+  // input files' associated with this compaction
+  uint64_t MinInputFileEpochNumber() const;
 
   // Called by DBImpl::NotifyOnCompactionCompleted to make sure number of
   // compaction begin and compaction completion callbacks match.
@@ -363,11 +393,18 @@ class Compaction {
   }
 
   static constexpr int kInvalidLevel = -1;
+
   // Evaluate penultimate output level. If the compaction supports
   // per_key_placement feature, it returns the penultimate level number.
   // Otherwise, it's set to kInvalidLevel (-1), which means
   // output_to_penultimate_level is not supported.
-  static int EvaluatePenultimateLevel(const ImmutableOptions& immutable_options,
+  // Note: even the penultimate level output is supported (PenultimateLevel !=
+  // kInvalidLevel), some key range maybe unsafe to be outputted to the
+  // penultimate level. The safe key range is populated by
+  // `PopulatePenultimateLevelOutputRange()`.
+  // Which could potentially disable all penultimate level output.
+  static int EvaluatePenultimateLevel(const VersionStorageInfo* vstorage,
+                                      const ImmutableOptions& immutable_options,
                                       const int start_level,
                                       const int output_level);
 
@@ -384,11 +421,6 @@ class Compaction {
   // populate penultimate level output range, which will be used to determine if
   // a key is safe to output to the penultimate level (details see
   // `Compaction::WithinPenultimateLevelOutputRange()`.
-  // TODO: Currently the penultimate level output range is the min/max keys of
-  //  non-last-level input files. Which is only good if there's no key moved
-  //  from the last level to the penultimate level. For a more complicated per
-  //  key placement which may move data from the last level to the penultimate
-  //  level, it needs extra check.
   void PopulatePenultimateLevelOutputRange();
 
   // Get the atomic file boundaries for all files in the compaction. Necessary
@@ -412,6 +444,7 @@ class Compaction {
 
   const int start_level_;   // the lowest level to be compacted
   const int output_level_;  // levels to which output files are stored
+  uint64_t target_output_file_size_;
   uint64_t max_output_file_size_;
   uint64_t max_compaction_bytes_;
   uint32_t max_subcompactions_;
@@ -470,9 +503,11 @@ class Compaction {
   TablePropertiesCollection output_table_properties_;
 
   // smallest user keys in compaction
+  // includes timestamp if user-defined timestamp is enabled.
   Slice smallest_user_key_;
 
   // largest user keys in compaction
+  // includes timestamp if user-defined timestamp is enabled.
   Slice largest_user_key_;
 
   // Reason for compaction
@@ -493,8 +528,12 @@ class Compaction {
   const int penultimate_level_;
 
   // Key range for penultimate level output
+  // includes timestamp if user-defined timestamp is enabled.
+  // penultimate_output_range_type_ shows the range type
   Slice penultimate_level_smallest_user_key_;
   Slice penultimate_level_largest_user_key_;
+  PenultimateOutputRangeType penultimate_output_range_type_ =
+      PenultimateOutputRangeType::kNotSupported;
 };
 
 #ifndef NDEBUG

db/compaction/compaction_iterator.cc

@@ -34,7 +34,8 @@ CompactionIterator::CompactionIterator(
     const std::atomic<bool>* shutting_down,
     const std::shared_ptr<Logger> info_log,
     const std::string* full_history_ts_low,
-    const SequenceNumber penultimate_level_cutoff_seqno)
+    const SequenceNumber preserve_time_min_seqno,
+    const SequenceNumber preclude_last_level_min_seqno)
     : CompactionIterator(
           input, cmp, merge_helper, last_sequence, snapshots,
           earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
@@ -44,7 +45,7 @@ CompactionIterator::CompactionIterator(
           std::unique_ptr<CompactionProxy>(
               compaction ? new RealCompaction(compaction) : nullptr),
           compaction_filter, shutting_down, info_log, full_history_ts_low,
-          penultimate_level_cutoff_seqno) {}
+          preserve_time_min_seqno, preclude_last_level_min_seqno) {}
 
 CompactionIterator::CompactionIterator(
     InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
@@ -61,7 +62,8 @@ CompactionIterator::CompactionIterator(
     const std::atomic<bool>* shutting_down,
     const std::shared_ptr<Logger> info_log,
     const std::string* full_history_ts_low,
-    const SequenceNumber penultimate_level_cutoff_seqno)
+    const SequenceNumber preserve_time_min_seqno,
+    const SequenceNumber preclude_last_level_min_seqno)
     : input_(input, cmp,
              !compaction || compaction->DoesInputReferenceBlobFiles()),
       cmp_(cmp),
@@ -105,8 +107,10 @@ CompactionIterator::CompactionIterator(
       current_key_committed_(false),
       cmp_with_history_ts_low_(0),
       level_(compaction_ == nullptr ? 0 : compaction_->level()),
-      penultimate_level_cutoff_seqno_(penultimate_level_cutoff_seqno) {
+      preserve_time_min_seqno_(preserve_time_min_seqno),
+      preclude_last_level_min_seqno_(preclude_last_level_min_seqno) {
   assert(snapshots_ != nullptr);
+  assert(preserve_time_min_seqno_ <= preclude_last_level_min_seqno_);
 
   if (compaction_ != nullptr) {
     level_ptrs_ = std::vector<size_t>(compaction_->number_levels(), 0);
@@ -120,6 +124,9 @@ CompactionIterator::CompactionIterator(
          timestamp_size_ == full_history_ts_low_->size());
 #endif
   input_.SetPinnedItersMgr(&pinned_iters_mgr_);
+  // The default `merge_until_status_` does not need to be checked since it is
+  // overwritten as soon as `MergeUntil()` is called
+  merge_until_status_.PermitUncheckedError();
   TEST_SYNC_POINT_CALLBACK("CompactionIterator:AfterInit", compaction_.get());
 }
 
@@ -164,11 +171,30 @@ void CompactionIterator::Next() {
       }
 
       // Keep current_key_ in sync.
-      current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
+      if (0 == timestamp_size_) {
+        current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
+      } else {
+        Slice ts = ikey_.GetTimestamp(timestamp_size_);
+        current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type, &ts);
+      }
       key_ = current_key_.GetInternalKey();
       ikey_.user_key = current_key_.GetUserKey();
       validity_info_.SetValid(ValidContext::kMerge1);
     } else {
+      if (merge_until_status_.IsMergeInProgress()) {
+        // `Status::MergeInProgress()` tells us that the previous `MergeUntil()`
+        // produced only merge operands. Those merge operands were accessed and
+        // written out using `merge_out_iter_`. Since `merge_out_iter_` is
+        // exhausted at this point, all merge operands have been written out.
+        //
+        // Still, there may be a base value (PUT, DELETE, SINGLEDEL, etc.) that
+        // needs to be written out. Normally, `CompactionIterator` would skip it
+        // on the basis that it has already output something in the same
+        // snapshot stripe. To prevent this, we reset `has_current_user_key_` to
+        // trick the future iteration from finding out the snapshot stripe is
+        // unchanged.
+        has_current_user_key_ = false;
+      }
       // We consumed all pinned merge operands, release pinned iterators
       pinned_iters_mgr_.ReleasePinnedData();
       // MergeHelper moves the iterator to the first record after the merged
@@ -821,8 +847,8 @@ void CompactionIterator::NextFromInput() {
                  cmp_with_history_ts_low_ < 0)) &&
                bottommost_level_) {
       // Handle the case where we have a delete key at the bottom most level
-      // We can skip outputting the key iff there are no subsequent puts for this
-      // key
+      // We can skip outputting the key iff there are no subsequent puts for
+      // this key
       assert(!compaction_ || compaction_->KeyNotExistsBeyondOutputLevel(
                                  ikey_.user_key, &level_ptrs_));
       ParsedInternalKey next_ikey;
@@ -849,8 +875,8 @@ void CompactionIterator::NextFromInput() {
               DefinitelyNotInSnapshot(next_ikey.sequence, prev_snapshot))) {
         AdvanceInputIter();
       }
-      // If you find you still need to output a row with this key, we need to output the
-      // delete too
+      // If you find you still need to output a row with this key, we need to
+      // output the delete too
       if (input_.Valid() &&
           (ParseInternalKey(input_.key(), &next_ikey, allow_data_in_errors_)
                .ok()) &&
@@ -871,14 +897,15 @@ void CompactionIterator::NextFromInput() {
       // have hit (A)
       // We encapsulate the merge related state machine in a different
       // object to minimize change to the existing flow.
-      Status s = merge_helper_->MergeUntil(
+      merge_until_status_ = merge_helper_->MergeUntil(
           &input_, range_del_agg_, prev_snapshot, bottommost_level_,
-          allow_data_in_errors_, blob_fetcher_.get(), prefetch_buffers_.get(),
-          &iter_stats_);
+          allow_data_in_errors_, blob_fetcher_.get(), full_history_ts_low_,
+          prefetch_buffers_.get(), &iter_stats_);
       merge_out_iter_.SeekToFirst();
 
-      if (!s.ok() && !s.IsMergeInProgress()) {
-        status_ = s;
+      if (!merge_until_status_.ok() &&
+          !merge_until_status_.IsMergeInProgress()) {
+        status_ = merge_until_status_;
         return;
       } else if (merge_out_iter_.Valid()) {
         // NOTE: key, value, and ikey_ refer to old entries.
@@ -914,8 +941,17 @@ void CompactionIterator::NextFromInput() {
     } else {
       // 1. new user key -OR-
       // 2. different snapshot stripe
-      bool should_delete = range_del_agg_->ShouldDelete(
-          key_, RangeDelPositioningMode::kForwardTraversal);
+      // If user-defined timestamp is enabled, we consider keys for GC if they
+      // are below history_ts_low_. CompactionRangeDelAggregator::ShouldDelete()
+      // only considers range deletions that are at or below history_ts_low_ and
+      // trim_ts_. We drop keys here that are below history_ts_low_ and are
+      // covered by a range tombstone that is at or below history_ts_low_ and
+      // trim_ts.
+      bool should_delete = false;
+      if (!timestamp_size_ || cmp_with_history_ts_low_ < 0) {
+        should_delete = range_del_agg_->ShouldDelete(
+            key_, RangeDelPositioningMode::kForwardTraversal);
+      }
       if (should_delete) {
         ++iter_stats_.num_record_drop_hidden;
         ++iter_stats_.num_record_drop_range_del;
@@ -1079,6 +1115,7 @@ void CompactionIterator::GarbageCollectBlobIfNeeded() {
 }
 
 void CompactionIterator::DecideOutputLevel() {
+  assert(compaction_->SupportsPerKeyPlacement());
 #ifndef NDEBUG
   // Could be overridden by unittest
   PerKeyPlacementContext context(level_, ikey_.user_key, value_,
@@ -1086,11 +1123,13 @@ void CompactionIterator::DecideOutputLevel() {
   TEST_SYNC_POINT_CALLBACK("CompactionIterator::PrepareOutput.context",
                            &context);
   output_to_penultimate_level_ = context.output_to_penultimate_level;
-#endif /* !NDEBUG */
+#else
+  output_to_penultimate_level_ = false;
+#endif  // NDEBUG
 
   // if the key is newer than the cutoff sequence or within the earliest
   // snapshot, it should output to the penultimate level.
-  if (ikey_.sequence > penultimate_level_cutoff_seqno_ ||
+  if (ikey_.sequence > preclude_last_level_min_seqno_ ||
       ikey_.sequence > earliest_snapshot_) {
     output_to_penultimate_level_ = true;
   }
@@ -1152,7 +1191,7 @@ void CompactionIterator::PrepareOutput() {
         DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
         ikey_.type != kTypeMerge && current_key_committed_ &&
         !output_to_penultimate_level_ &&
-        ikey_.sequence < penultimate_level_cutoff_seqno_) {
+        ikey_.sequence < preserve_time_min_seqno_) {
       if (ikey_.type == kTypeDeletion ||
           (ikey_.type == kTypeSingleDeletion && timestamp_size_ == 0)) {
         ROCKS_LOG_FATAL(
@@ -1198,8 +1237,8 @@ inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
     ROCKS_LOG_FATAL(info_log_,
                     "No snapshot left in findEarliestVisibleSnapshot");
   }
-  auto snapshots_iter = std::lower_bound(
-      snapshots_->begin(), snapshots_->end(), in);
+  auto snapshots_iter =
+      std::lower_bound(snapshots_->begin(), snapshots_->end(), in);
   assert(prev_snapshot != nullptr);
   if (snapshots_iter == snapshots_->begin()) {
     *prev_snapshot = 0;
@@ -1214,8 +1253,8 @@ inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
     }
   }
   if (snapshot_checker_ == nullptr) {
-    return snapshots_iter != snapshots_->end()
-      ? *snapshots_iter : kMaxSequenceNumber;
+    return snapshots_iter != snapshots_->end() ? *snapshots_iter
+                                               : kMaxSequenceNumber;
   }
   bool has_released_snapshot = !released_snapshots_.empty();
   for (; snapshots_iter != snapshots_->end(); ++snapshots_iter) {

db/compaction/compaction_iterator.h

@@ -88,6 +88,7 @@ class CompactionIterator {
 
     virtual int number_levels() const = 0;
 
+    // Result includes timestamp if user-defined timestamp is enabled.
     virtual Slice GetLargestUserKey() const = 0;
 
     virtual bool allow_ingest_behind() const = 0;
@@ -108,6 +109,7 @@ class CompactionIterator {
 
     virtual bool SupportsPerKeyPlacement() const = 0;
 
+    // `key` includes timestamp if user-defined timestamp is enabled.
     virtual bool WithinPenultimateLevelOutputRange(const Slice& key) const = 0;
   };
 
@@ -133,6 +135,7 @@ class CompactionIterator {
 
     int number_levels() const override { return compaction_->number_levels(); }
 
+    // Result includes timestamp if user-defined timestamp is enabled.
     Slice GetLargestUserKey() const override {
       return compaction_->GetLargestUserKey();
     }
@@ -173,6 +176,7 @@ class CompactionIterator {
 
     // Check if key is within penultimate level output range, to see if it's
     // safe to output to the penultimate level for per_key_placement feature.
+    // `key` includes timestamp if user-defined timestamp is enabled.
     bool WithinPenultimateLevelOutputRange(const Slice& key) const override {
       return compaction_->WithinPenultimateLevelOutputRange(key);
     }
@@ -196,7 +200,8 @@ class CompactionIterator {
       const std::atomic<bool>* shutting_down = nullptr,
       const std::shared_ptr<Logger> info_log = nullptr,
       const std::string* full_history_ts_low = nullptr,
-      const SequenceNumber penultimate_level_cutoff_seqno = kMaxSequenceNumber);
+      const SequenceNumber preserve_time_min_seqno = kMaxSequenceNumber,
+      const SequenceNumber preclude_last_level_min_seqno = kMaxSequenceNumber);
 
   // Constructor with custom CompactionProxy, used for tests.
   CompactionIterator(
@@ -214,7 +219,8 @@ class CompactionIterator {
       const std::atomic<bool>* shutting_down = nullptr,
       const std::shared_ptr<Logger> info_log = nullptr,
       const std::string* full_history_ts_low = nullptr,
-      const SequenceNumber penultimate_level_cutoff_seqno = kMaxSequenceNumber);
+      const SequenceNumber preserve_time_min_seqno = kMaxSequenceNumber,
+      const SequenceNumber preclude_last_level_min_seqno = kMaxSequenceNumber);
 
   ~CompactionIterator();
 
@@ -426,6 +432,7 @@ class CompactionIterator {
   bool clear_and_output_next_key_ = false;
 
   MergeOutputIterator merge_out_iter_;
+  Status merge_until_status_;
   // PinnedIteratorsManager used to pin input_ Iterator blocks while reading
   // merge operands and then releasing them after consuming them.
   PinnedIteratorsManager pinned_iters_mgr_;
@@ -466,9 +473,12 @@ class CompactionIterator {
   // output to.
   bool output_to_penultimate_level_{false};
 
-  // any key later than this sequence number should have
-  // output_to_penultimate_level_ set to true
-  const SequenceNumber penultimate_level_cutoff_seqno_ = kMaxSequenceNumber;
+  // min seqno for preserving the time information.
+  const SequenceNumber preserve_time_min_seqno_ = kMaxSequenceNumber;
+
+  // min seqno to preclude the data from the last level, if the key seqno larger
+  // than this, it will be output to penultimate level
+  const SequenceNumber preclude_last_level_min_seqno_ = kMaxSequenceNumber;
 
   void AdvanceInputIter() { input_.Next(); }
 

db/compaction/compaction_iterator_test.cc

@@ -203,7 +203,8 @@ class TestSnapshotChecker : public SnapshotChecker {
  public:
   explicit TestSnapshotChecker(
       SequenceNumber last_committed_sequence,
-      const std::unordered_map<SequenceNumber, SequenceNumber>& snapshots = {{}})
+      const std::unordered_map<SequenceNumber, SequenceNumber>& snapshots =
+          {{}})
       : last_committed_sequence_(last_committed_sequence),
         snapshots_(snapshots) {}
 
@@ -811,6 +812,8 @@ TEST_P(PerKeyPlacementCompIteratorTest, SplitLastLevelData) {
   c_iter_->Next();
   ASSERT_OK(c_iter_->status());
   ASSERT_FALSE(c_iter_->Valid());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
 TEST_P(PerKeyPlacementCompIteratorTest, SnapshotData) {
@@ -876,6 +879,8 @@ TEST_P(PerKeyPlacementCompIteratorTest, ConflictWithSnapshot) {
   // output_to_penultimate_level.
   c_iter_->Next();
   ASSERT_TRUE(c_iter_->status().IsCorruption());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
 INSTANTIATE_TEST_CASE_P(PerKeyPlacementCompIteratorTest,
@@ -1038,7 +1043,7 @@ TEST_F(CompactionIteratorWithSnapshotCheckerTest,
 
 TEST_F(CompactionIteratorWithSnapshotCheckerTest,
        NotRemoveDeletionIfValuePresentToEarlierSnapshot) {
-  AddSnapshot(2,1);
+  AddSnapshot(2, 1);
   RunTest({test::KeyStr("a", 4, kTypeDeletion),
            test::KeyStr("a", 1, kTypeValue), test::KeyStr("b", 3, kTypeValue)},
           {"", "", ""},
@@ -1250,6 +1255,31 @@ TEST_P(CompactionIteratorTsGcTest, NoKeyEligibleForGC) {
   }
 }
 
+TEST_P(CompactionIteratorTsGcTest, NoMergeEligibleForGc) {
+  constexpr char user_key[] = "a";
+  const std::vector<std::string> input_keys = {
+      test::KeyStr(10002, user_key, 102, kTypeMerge),
+      test::KeyStr(10001, user_key, 101, kTypeMerge),
+      test::KeyStr(10000, user_key, 100, kTypeValue)};
+  const std::vector<std::string> input_values = {"2", "1", "a0"};
+  std::shared_ptr<MergeOperator> merge_op =
+      MergeOperators::CreateStringAppendTESTOperator();
+  const std::vector<std::string>& expected_keys = input_keys;
+  const std::vector<std::string>& expected_values = input_values;
+  const std::vector<std::pair<bool, bool>> params = {
+      {false, false}, {false, true}, {true, true}};
+  for (const auto& param : params) {
+    const bool bottommost_level = param.first;
+    const bool key_not_exists_beyond_output_level = param.second;
+    RunTest(input_keys, input_values, expected_keys, expected_values,
+            /*last_committed_seq=*/kMaxSequenceNumber, merge_op.get(),
+            /*compaction_filter=*/nullptr, bottommost_level,
+            /*earliest_write_conflict_snapshot=*/kMaxSequenceNumber,
+            key_not_exists_beyond_output_level,
+            /*full_history_ts_low=*/nullptr);
+  }
+}
+
 TEST_P(CompactionIteratorTsGcTest, AllKeysOlderThanThreshold) {
   constexpr char user_key[][2] = {{'a', '\0'}, {'b', '\0'}};
   const std::vector<std::string> input_keys = {
@@ -1303,6 +1333,91 @@ TEST_P(CompactionIteratorTsGcTest, AllKeysOlderThanThreshold) {
   }
 }
 
+TEST_P(CompactionIteratorTsGcTest, SomeMergesOlderThanThreshold) {
+  constexpr char user_key[][2] = {"a", "f"};
+  const std::vector<std::string> input_keys = {
+      test::KeyStr(/*ts=*/25000, user_key[0], /*seq=*/2500, kTypeMerge),
+      test::KeyStr(/*ts=*/19000, user_key[0], /*seq=*/2300, kTypeMerge),
+      test::KeyStr(/*ts=*/18000, user_key[0], /*seq=*/1800, kTypeMerge),
+      test::KeyStr(/*ts=*/16000, user_key[0], /*seq=*/1600, kTypeValue),
+      test::KeyStr(/*ts=*/19000, user_key[1], /*seq=*/2000, kTypeMerge),
+      test::KeyStr(/*ts=*/17000, user_key[1], /*seq=*/1700, kTypeMerge),
+      test::KeyStr(/*ts=*/15000, user_key[1], /*seq=*/1600,
+                   kTypeDeletionWithTimestamp)};
+  const std::vector<std::string> input_values = {"25", "19", "18", "16",
+                                                 "19", "17", ""};
+  std::shared_ptr<MergeOperator> merge_op =
+      MergeOperators::CreateStringAppendTESTOperator();
+  std::string full_history_ts_low;
+  PutFixed64(&full_history_ts_low, 20000);
+
+  const std::vector<std::pair<bool, bool>> params = {
+      {false, false}, {false, true}, {true, true}};
+
+  {
+    AddSnapshot(1600);
+    AddSnapshot(1900);
+    const std::vector<std::string> expected_keys = {
+        test::KeyStr(/*ts=*/25000, user_key[0], /*seq=*/2500, kTypeMerge),
+        test::KeyStr(/*ts=*/19000, user_key[0], /*seq=*/2300, kTypeMerge),
+        test::KeyStr(/*ts=*/18000, user_key[0], /*seq=*/1800, kTypeMerge),
+        test::KeyStr(/*ts=*/16000, user_key[0], /*seq=*/1600, kTypeValue),
+        test::KeyStr(/*ts=*/19000, user_key[1], /*seq=*/2000, kTypeMerge),
+        test::KeyStr(/*ts=*/17000, user_key[1], /*seq=*/1700, kTypeMerge),
+        test::KeyStr(/*ts=*/15000, user_key[1], /*seq=*/1600,
+                     kTypeDeletionWithTimestamp)};
+    const std::vector<std::string> expected_values = {"25", "19", "18", "16",
+                                                      "19", "17", ""};
+    for (const auto& param : params) {
+      const bool bottommost_level = param.first;
+      const bool key_not_exists_beyond_output_level = param.second;
+      auto expected_keys_copy = expected_keys;
+      auto expected_values_copy = expected_values;
+      if (bottommost_level || key_not_exists_beyond_output_level) {
+        // the kTypeDeletionWithTimestamp will be dropped
+        expected_keys_copy.pop_back();
+        expected_values_copy.pop_back();
+        if (bottommost_level) {
+          // seq zero
+          expected_keys_copy[3] =
+              test::KeyStr(/*ts=*/0, user_key[0], /*seq=*/0, kTypeValue);
+        }
+      }
+      RunTest(input_keys, input_values, expected_keys_copy,
+              expected_values_copy,
+              /*last_committed_seq=*/kMaxSequenceNumber, merge_op.get(),
+              /*compaction_filter=*/nullptr, bottommost_level,
+              /*earliest_write_conflict_snapshot=*/kMaxSequenceNumber,
+              key_not_exists_beyond_output_level, &full_history_ts_low);
+    }
+    ClearSnapshots();
+  }
+
+  // No snapshots
+  {
+    const std::vector<std::string> expected_keys = {
+        test::KeyStr(/*ts=*/25000, user_key[0], /*seq=*/2500, kTypeValue),
+        test::KeyStr(/*ts=*/19000, user_key[1], /*seq=*/2000, kTypeValue)};
+    const std::vector<std::string> expected_values = {"16,18,19,25", "17,19"};
+    for (const auto& param : params) {
+      const bool bottommost_level = param.first;
+      const bool key_not_exists_beyond_output_level = param.second;
+      auto expected_keys_copy = expected_keys;
+      auto expected_values_copy = expected_values;
+      if (bottommost_level) {
+        expected_keys_copy[1] =
+            test::KeyStr(/*ts=*/0, user_key[1], /*seq=*/0, kTypeValue);
+      }
+      RunTest(input_keys, input_values, expected_keys_copy,
+              expected_values_copy,
+              /*last_committed_seq=*/kMaxSequenceNumber, merge_op.get(),
+              /*compaction_filter=*/nullptr, bottommost_level,
+              /*earliest_write_conflict_snapshot=*/kMaxSequenceNumber,
+              key_not_exists_beyond_output_level, &full_history_ts_low);
+    }
+  }
+}
+
 TEST_P(CompactionIteratorTsGcTest, NewHidesOldSameSnapshot) {
   constexpr char user_key[] = "a";
   const std::vector<std::string> input_keys = {
@@ -1497,6 +1612,7 @@ INSTANTIATE_TEST_CASE_P(CompactionIteratorTsGcTestInstance,
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/compaction/compaction_job.cc

@@ -97,6 +97,10 @@ const char* GetCompactionReasonString(CompactionReason compaction_reason) {
       return "ChangeTemperature";
     case CompactionReason::kForcedBlobGC:
       return "ForcedBlobGC";
+    case CompactionReason::kRoundRobinTtl:
+      return "RoundRobinTtl";
+    case CompactionReason::kRefitLevel:
+      return "RefitLevel";
     case CompactionReason::kNumOfReasons:
       // fall through
     default:
@@ -105,6 +109,23 @@ const char* GetCompactionReasonString(CompactionReason compaction_reason) {
   }
 }
 
+const char* GetCompactionPenultimateOutputRangeTypeString(
+    Compaction::PenultimateOutputRangeType range_type) {
+  switch (range_type) {
+    case Compaction::PenultimateOutputRangeType::kNotSupported:
+      return "NotSupported";
+    case Compaction::PenultimateOutputRangeType::kFullRange:
+      return "FullRange";
+    case Compaction::PenultimateOutputRangeType::kNonLastRange:
+      return "NonLastRange";
+    case Compaction::PenultimateOutputRangeType::kDisabled:
+      return "Disabled";
+    default:
+      assert(false);
+      return "Invalid";
+  }
+}
+
 CompactionJob::CompactionJob(
     int job_id, Compaction* compaction, const ImmutableDBOptions& db_options,
     const MutableDBOptions& mutable_db_options, const FileOptions& file_options,
@@ -254,7 +275,7 @@ void CompactionJob::Prepare() {
       // timestamp part).
       assert(i == 0 || i == boundaries_.size() ||
              cfd->user_comparator()->CompareWithoutTimestamp(
-                 boundaries_[i - 1], true, boundaries_[i], true) < 0);
+                 boundaries_[i - 1], boundaries_[i]) < 0);
     }
     RecordInHistogram(stats_, NUM_SUBCOMPACTIONS_SCHEDULED,
                       compact_->sub_compact_states.size());
@@ -263,11 +284,15 @@ void CompactionJob::Prepare() {
                                               /*sub_job_id*/ 0);
   }
 
-  if (c->immutable_options()->preclude_last_level_data_seconds > 0) {
-    // TODO(zjay): move to a function
-    seqno_time_mapping_.SetMaxTimeDuration(
-        c->immutable_options()->preclude_last_level_data_seconds);
+  // collect all seqno->time information from the input files which will be used
+  // to encode seqno->time to the output files.
+  uint64_t preserve_time_duration =
+      std::max(c->immutable_options()->preserve_internal_time_seconds,
+               c->immutable_options()->preclude_last_level_data_seconds);
+
+  if (preserve_time_duration > 0) {
     // setup seqno_time_mapping_
+    seqno_time_mapping_.SetMaxTimeDuration(preserve_time_duration);
     for (const auto& each_level : *c->inputs()) {
       for (const auto& fmd : each_level.files) {
         std::shared_ptr<const TableProperties> tp;
@@ -293,10 +318,27 @@ void CompactionJob::Prepare() {
       ROCKS_LOG_WARN(db_options_.info_log,
                      "Failed to get current time in compaction: Status: %s",
                      status.ToString().c_str());
-      penultimate_level_cutoff_seqno_ = 0;
+      // preserve all time information
+      preserve_time_min_seqno_ = 0;
+      preclude_last_level_min_seqno_ = 0;
     } else {
-      penultimate_level_cutoff_seqno_ =
-          seqno_time_mapping_.TruncateOldEntries(_current_time);
+      seqno_time_mapping_.TruncateOldEntries(_current_time);
+      uint64_t preserve_time =
+          static_cast<uint64_t>(_current_time) > preserve_time_duration
+              ? _current_time - preserve_time_duration
+              : 0;
+      preserve_time_min_seqno_ =
+          seqno_time_mapping_.GetOldestSequenceNum(preserve_time);
+      if (c->immutable_options()->preclude_last_level_data_seconds > 0) {
+        uint64_t preclude_last_level_time =
+            static_cast<uint64_t>(_current_time) >
+                    c->immutable_options()->preclude_last_level_data_seconds
+                ? _current_time -
+                      c->immutable_options()->preclude_last_level_data_seconds
+                : 0;
+        preclude_last_level_min_seqno_ =
+            seqno_time_mapping_.GetOldestSequenceNum(preclude_last_level_time);
+      }
     }
   }
 }
@@ -321,6 +363,7 @@ void CompactionJob::AcquireSubcompactionResources(
               ->write_controller()
               ->NeedSpeedupCompaction())
           .max_compactions;
+  InstrumentedMutexLock l(db_mutex_);
   // Apply min function first since We need to compute the extra subcompaction
   // against compaction limits. And then try to reserve threads for extra
   // subcompactions. The actual number of reserved threads could be less than
@@ -329,7 +372,6 @@ void CompactionJob::AcquireSubcompactionResources(
       std::max(max_db_compactions - *bg_compaction_scheduled_ -
                    *bg_bottom_compaction_scheduled_,
                0);
-  db_mutex_->Lock();
   // Reservation only supports backgrdoun threads of which the priority is
   // between BOTTOM and HIGH. Need to degrade the priority to HIGH if the
   // origin thread_pri_ is higher than that. Similar to ReleaseThreads().
@@ -346,7 +388,6 @@ void CompactionJob::AcquireSubcompactionResources(
   } else {
     *bg_compaction_scheduled_ += extra_num_subcompaction_threads_reserved_;
   }
-  db_mutex_->Unlock();
 }
 
 void CompactionJob::ShrinkSubcompactionResources(uint64_t num_extra_resources) {
@@ -380,17 +421,20 @@ void CompactionJob::ReleaseSubcompactionResources() {
   if (extra_num_subcompaction_threads_reserved_ == 0) {
     return;
   }
-  // The number of reserved threads becomes larger than 0 only if the
-  // compaction prioity is round robin and there is no sufficient
-  // sub-compactions available
+  {
+    InstrumentedMutexLock l(db_mutex_);
+    // The number of reserved threads becomes larger than 0 only if the
+    // compaction prioity is round robin and there is no sufficient
+    // sub-compactions available
 
-  // The scheduled compaction must be no less than 1 + extra number
-  // subcompactions using acquired resources since this compaction job has not
-  // finished yet
-  assert(*bg_bottom_compaction_scheduled_ >=
-             1 + extra_num_subcompaction_threads_reserved_ ||
-         *bg_compaction_scheduled_ >=
-             1 + extra_num_subcompaction_threads_reserved_);
+    // The scheduled compaction must be no less than 1 + extra number
+    // subcompactions using acquired resources since this compaction job has not
+    // finished yet
+    assert(*bg_bottom_compaction_scheduled_ >=
+               1 + extra_num_subcompaction_threads_reserved_ ||
+           *bg_compaction_scheduled_ >=
+               1 + extra_num_subcompaction_threads_reserved_);
+  }
   ShrinkSubcompactionResources(extra_num_subcompaction_threads_reserved_);
 }
 
@@ -462,7 +506,7 @@ void CompactionJob::GenSubcompactionBoundaries() {
         FileMetaData* f = flevel->files[i].file_metadata;
         std::vector<TableReader::Anchor> my_anchors;
         Status s = cfd->table_cache()->ApproximateKeyAnchors(
-            ReadOptions(), icomp, f->fd, my_anchors);
+            ReadOptions(), icomp, *f, my_anchors);
         if (!s.ok() || my_anchors.empty()) {
           my_anchors.emplace_back(f->largest.user_key(), f->fd.GetFileSize());
         }
@@ -484,8 +528,8 @@ void CompactionJob::GenSubcompactionBoundaries() {
   std::sort(
       all_anchors.begin(), all_anchors.end(),
       [cfd_comparator](TableReader::Anchor& a, TableReader::Anchor& b) -> bool {
-        return cfd_comparator->CompareWithoutTimestamp(a.user_key, true,
-                                                       b.user_key, true) < 0;
+        return cfd_comparator->CompareWithoutTimestamp(a.user_key, b.user_key) <
+               0;
       });
 
   // Remove duplicated entries from boundaries.
@@ -494,7 +538,7 @@ void CompactionJob::GenSubcompactionBoundaries() {
                   [cfd_comparator](TableReader::Anchor& a,
                                    TableReader::Anchor& b) -> bool {
                     return cfd_comparator->CompareWithoutTimestamp(
-                               a.user_key, true, b.user_key, true) == 0;
+                               a.user_key, b.user_key) == 0;
                   }),
       all_anchors.end());
 
@@ -672,11 +716,12 @@ Status CompactionJob::Run() {
           break;
         }
         // Verify that the table is usable
-        // We set for_compaction to false and don't OptimizeForCompactionTableRead
-        // here because this is a special case after we finish the table building
-        // No matter whether use_direct_io_for_flush_and_compaction is true,
-        // we will regard this verification as user reads since the goal is
-        // to cache it here for further user reads
+        // We set for_compaction to false and don't
+        // OptimizeForCompactionTableRead here because this is a special case
+        // after we finish the table building No matter whether
+        // use_direct_io_for_flush_and_compaction is true, we will regard this
+        // verification as user reads since the goal is to cache it here for
+        // further user reads
         ReadOptions read_options;
         InternalIterator* iter = cfd->table_cache()->NewIterator(
             read_options, file_options_, cfd->internal_comparator(),
@@ -722,8 +767,8 @@ Status CompactionJob::Run() {
       }
     };
     for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
-      thread_pool.emplace_back(verify_table,
-                               std::ref(compact_->sub_compact_states[i].status));
+      thread_pool.emplace_back(
+          verify_table, std::ref(compact_->sub_compact_states[i].status));
     }
     verify_table(compact_->sub_compact_states[0].status);
     for (auto& thread : thread_pool) {
@@ -1033,7 +1078,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   NotifyOnSubcompactionBegin(sub_compact);
 
   auto range_del_agg = std::make_unique<CompactionRangeDelAggregator>(
-      &cfd->internal_comparator(), existing_snapshots_);
+      &cfd->internal_comparator(), existing_snapshots_, &full_history_ts_low_,
+      &trim_ts_);
 
   // TODO: since we already use C++17, should use
   // std::optional<const Slice> instead.
@@ -1083,13 +1129,34 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   Slice start_slice;
   Slice end_slice;
 
+  static constexpr char kMaxTs[] =
+      "\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff";
+  Slice ts_slice;
+  std::string max_ts;
+  if (ts_sz > 0) {
+    if (ts_sz <= strlen(kMaxTs)) {
+      ts_slice = Slice(kMaxTs, ts_sz);
+    } else {
+      max_ts = std::string(ts_sz, '\xff');
+      ts_slice = Slice(max_ts);
+    }
+  }
+
   if (start.has_value()) {
     start_ikey.SetInternalKey(start.value(), kMaxSequenceNumber,
                               kValueTypeForSeek);
+    if (ts_sz > 0) {
+      start_ikey.UpdateInternalKey(kMaxSequenceNumber, kValueTypeForSeek,
+                                   &ts_slice);
+    }
     start_slice = start_ikey.GetInternalKey();
   }
   if (end.has_value()) {
     end_ikey.SetInternalKey(end.value(), kMaxSequenceNumber, kValueTypeForSeek);
+    if (ts_sz > 0) {
+      end_ikey.UpdateInternalKey(kMaxSequenceNumber, kValueTypeForSeek,
+                                 &ts_slice);
+    }
     end_slice = end_ikey.GetInternalKey();
   }
 
@@ -1110,7 +1177,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   std::unique_ptr<InternalIterator> trim_history_iter;
-  if (cfd->user_comparator()->timestamp_size() > 0 && !trim_ts_.empty()) {
+  if (ts_sz > 0 && !trim_ts_.empty()) {
     trim_history_iter = std::make_unique<HistoryTrimmingIterator>(
         input, cfd->user_comparator(), trim_ts_);
     input = trim_history_iter.get();
@@ -1190,21 +1257,14 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
       blob_file_builder.get(), db_options_.allow_data_in_errors,
       db_options_.enforce_single_del_contracts, manual_compaction_canceled_,
       sub_compact->compaction, compaction_filter, shutting_down_,
-      db_options_.info_log, full_history_ts_low,
-      penultimate_level_cutoff_seqno_);
+      db_options_.info_log, full_history_ts_low, preserve_time_min_seqno_,
+      preclude_last_level_min_seqno_);
   c_iter->SeekToFirst();
 
   // Assign range delete aggregator to the target output level, which makes sure
   // it only output to single level
   sub_compact->AssignRangeDelAggregator(std::move(range_del_agg));
 
-  if (c_iter->Valid() && sub_compact->compaction->output_level() != 0) {
-    sub_compact->FillFilesToCutForTtl();
-    // ShouldStopBefore() maintains state based on keys processed so far. The
-    // compaction loop always calls it on the "next" key, thus won't tell it the
-    // first key. So we do that here.
-    sub_compact->ShouldStopBefore(c_iter->key());
-  }
   const auto& c_iter_stats = c_iter->iter_stats();
 
   // define the open and close functions for the compaction files, which will be
@@ -1221,6 +1281,10 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
       };
 
   Status status;
+  TEST_SYNC_POINT_CALLBACK(
+      "CompactionJob::ProcessKeyValueCompaction()::Processing",
+      reinterpret_cast<void*>(
+          const_cast<Compaction*>(sub_compact->compaction)));
   while (status.ok() && !cfd->IsDropped() && c_iter->Valid()) {
     // Invariant: c_iter.status() is guaranteed to be OK if c_iter->Valid()
     // returns true.
@@ -1253,16 +1317,6 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
     if (c_iter->status().IsManualCompactionPaused()) {
       break;
     }
-
-    // TODO: Support earlier file cut for the penultimate level files. Maybe by
-    //  moving `ShouldStopBefore()` to `CompactionOutputs` class. Currently
-    //  the penultimate level output is only cut when it reaches the size limit.
-    if (!sub_compact->Current().IsPendingClose() &&
-        sub_compact->compaction->output_level() != 0 &&
-        !sub_compact->compaction->SupportsPerKeyPlacement() &&
-        sub_compact->ShouldStopBefore(c_iter->key())) {
-      sub_compact->Current().SetPendingClose();
-    }
   }
 
   sub_compact->compaction_job_stats.num_blobs_read =
@@ -1449,7 +1503,7 @@ Status CompactionJob::FinishCompactionOutputFile(
                                          : nullptr,
           sub_compact->end.has_value() ? &(sub_compact->end.value()) : nullptr,
           range_del_out_stats, bottommost_level_, cfd->internal_comparator(),
-          earliest_snapshot, next_table_min_key);
+          earliest_snapshot, next_table_min_key, full_history_ts_low_);
     }
     RecordDroppedKeys(range_del_out_stats, &sub_compact->compaction_job_stats);
     TEST_SYNC_POINT("CompactionJob::FinishCompactionOutputFile1");
@@ -1655,7 +1709,9 @@ Status CompactionJob::InstallCompactionResults(
                              stats.GetBytes());
   }
 
-  if (compaction->compaction_reason() == CompactionReason::kLevelMaxLevelSize &&
+  if ((compaction->compaction_reason() ==
+           CompactionReason::kLevelMaxLevelSize ||
+       compaction->compaction_reason() == CompactionReason::kRoundRobinTtl) &&
       compaction->immutable_options()->compaction_pri == kRoundRobin) {
     int start_level = compaction->start_level();
     if (start_level > 0) {
@@ -1780,12 +1836,14 @@ Status CompactionJob::OpenCompactionOutputFile(SubcompactionState* sub_compact,
   }
 
   // Initialize a SubcompactionState::Output and add it to sub_compact->outputs
+  uint64_t epoch_number = sub_compact->compaction->MinInputFileEpochNumber();
   {
     FileMetaData meta;
     meta.fd = FileDescriptor(file_number,
                              sub_compact->compaction->output_path_id(), 0);
     meta.oldest_ancester_time = oldest_ancester_time;
     meta.file_creation_time = current_time;
+    meta.epoch_number = epoch_number;
     meta.temperature = temperature;
     assert(!db_id_.empty());
     assert(!db_session_id_.empty());
@@ -1944,7 +2002,7 @@ void CompactionJob::LogCompaction() {
         compaction->InputLevelSummary(&inputs_summary), compaction->score());
     char scratch[2345];
     compaction->Summary(scratch, sizeof(scratch));
-    ROCKS_LOG_INFO(db_options_.info_log, "[%s] Compaction start summary: %s\n",
+    ROCKS_LOG_INFO(db_options_.info_log, "[%s]: Compaction start summary: %s\n",
                    cfd->GetName().c_str(), scratch);
     // build event logger report
     auto stream = event_logger_->Log();
@@ -1965,6 +2023,23 @@ void CompactionJob::LogCompaction() {
            << (existing_snapshots_.empty()
                    ? int64_t{-1}  // Use -1 for "none"
                    : static_cast<int64_t>(existing_snapshots_[0]));
+    if (compaction->SupportsPerKeyPlacement()) {
+      stream << "preclude_last_level_min_seqno"
+             << preclude_last_level_min_seqno_;
+      stream << "penultimate_output_level" << compaction->GetPenultimateLevel();
+      stream << "penultimate_output_range"
+             << GetCompactionPenultimateOutputRangeTypeString(
+                    compaction->GetPenultimateOutputRangeType());
+
+      if (compaction->GetPenultimateOutputRangeType() ==
+          Compaction::PenultimateOutputRangeType::kDisabled) {
+        ROCKS_LOG_WARN(
+            db_options_.info_log,
+            "[%s] [JOB %d] Penultimate level output is disabled, likely "
+            "because of the range conflict in the penultimate level",
+            cfd->GetName().c_str(), job_id_);
+      }
+    }
   }
 }
 

db/compaction/compaction_job.h

@@ -261,12 +261,12 @@ class CompactionJob {
   Status OpenCompactionOutputFile(SubcompactionState* sub_compact,
                                   CompactionOutputs& outputs);
   void UpdateCompactionJobStats(
-    const InternalStats::CompactionStats& stats) const;
+      const InternalStats::CompactionStats& stats) const;
   void RecordDroppedKeys(const CompactionIterationStats& c_iter_stats,
                          CompactionJobStats* compaction_job_stats = nullptr);
 
-  void UpdateCompactionInputStatsHelper(
-      int* num_files, uint64_t* bytes_read, int input_level);
+  void UpdateCompactionInputStatsHelper(int* num_files, uint64_t* bytes_read,
+                                        int input_level);
 
   void NotifyOnSubcompactionBegin(SubcompactionState* sub_compact);
 
@@ -335,12 +335,15 @@ class CompactionJob {
   // it also collects the smallest_seqno -> oldest_ancester_time from the SST.
   SeqnoToTimeMapping seqno_time_mapping_;
 
-  // cutoff sequence number for penultimate level, only set when
-  // per_key_placement feature is enabled.
-  // If a key with sequence number larger than penultimate_level_cutoff_seqno_,
-  // it will be placed on the penultimate_level and seqnuence number won't be
-  // zeroed out.
-  SequenceNumber penultimate_level_cutoff_seqno_ = kMaxSequenceNumber;
+  // Minimal sequence number for preserving the time information. The time info
+  // older than this sequence number won't be preserved after the compaction and
+  // if it's bottommost compaction, the seq num will be zeroed out.
+  SequenceNumber preserve_time_min_seqno_ = kMaxSequenceNumber;
+
+  // Minimal sequence number to preclude the data from the last level. If the
+  // key has bigger (newer) sequence number than this, it will be precluded from
+  // the last level (output to penultimate level).
+  SequenceNumber preclude_last_level_min_seqno_ = kMaxSequenceNumber;
 
   // Get table file name in where it's outputting to, which should also be in
   // `output_directory_`.
@@ -399,6 +402,7 @@ struct CompactionServiceOutputFile {
   std::string largest_internal_key;
   uint64_t oldest_ancester_time;
   uint64_t file_creation_time;
+  uint64_t epoch_number;
   uint64_t paranoid_hash;
   bool marked_for_compaction;
   UniqueId64x2 unique_id;
@@ -408,8 +412,8 @@ struct CompactionServiceOutputFile {
       const std::string& name, SequenceNumber smallest, SequenceNumber largest,
       std::string _smallest_internal_key, std::string _largest_internal_key,
       uint64_t _oldest_ancester_time, uint64_t _file_creation_time,
-      uint64_t _paranoid_hash, bool _marked_for_compaction,
-      UniqueId64x2 _unique_id)
+      uint64_t _epoch_number, uint64_t _paranoid_hash,
+      bool _marked_for_compaction, UniqueId64x2 _unique_id)
       : file_name(name),
         smallest_seqno(smallest),
         largest_seqno(largest),
@@ -417,6 +421,7 @@ struct CompactionServiceOutputFile {
         largest_internal_key(std::move(_largest_internal_key)),
         oldest_ancester_time(_oldest_ancester_time),
         file_creation_time(_file_creation_time),
+        epoch_number(_epoch_number),
         paranoid_hash(_paranoid_hash),
         marked_for_compaction(_marked_for_compaction),
         unique_id(std::move(_unique_id)) {}

db/compaction/compaction_job_stats_test.cc

@@ -155,9 +155,8 @@ class CompactionJobStatsTest : public testing::Test,
     ASSERT_OK(TryReopenWithColumnFamilies(cfs, options));
   }
 
-  Status TryReopenWithColumnFamilies(
-      const std::vector<std::string>& cfs,
-      const std::vector<Options>& options) {
+  Status TryReopenWithColumnFamilies(const std::vector<std::string>& cfs,
+                                     const std::vector<Options>& options) {
     Close();
     EXPECT_EQ(cfs.size(), options.size());
     std::vector<ColumnFamilyDescriptor> column_families;
@@ -175,9 +174,7 @@ class CompactionJobStatsTest : public testing::Test,
     return TryReopenWithColumnFamilies(cfs, v_opts);
   }
 
-  void Reopen(const Options& options) {
-    ASSERT_OK(TryReopen(options));
-  }
+  void Reopen(const Options& options) { ASSERT_OK(TryReopen(options)); }
 
   void Close() {
     for (auto h : handles_) {
@@ -226,9 +223,7 @@ class CompactionJobStatsTest : public testing::Test,
     return db_->Put(wo, handles_[cf], k, v);
   }
 
-  Status Delete(const std::string& k) {
-    return db_->Delete(WriteOptions(), k);
-  }
+  Status Delete(const std::string& k) { return db_->Delete(WriteOptions(), k); }
 
   Status Delete(int cf, const std::string& k) {
     return db_->Delete(WriteOptions(), handles_[cf], k);
@@ -338,21 +333,21 @@ class CompactionJobStatsTest : public testing::Test,
     }
   }
 
-  static void SetDeletionCompactionStats(
-      CompactionJobStats *stats, uint64_t input_deletions,
-      uint64_t expired_deletions, uint64_t records_replaced) {
+  static void SetDeletionCompactionStats(CompactionJobStats* stats,
+                                         uint64_t input_deletions,
+                                         uint64_t expired_deletions,
+                                         uint64_t records_replaced) {
     stats->num_input_deletion_records = input_deletions;
     stats->num_expired_deletion_records = expired_deletions;
     stats->num_records_replaced = records_replaced;
   }
 
-  void MakeTableWithKeyValues(
-    Random* rnd, uint64_t smallest, uint64_t largest,
-    int key_size, int value_size, uint64_t interval,
-    double ratio, int cf = 0) {
+  void MakeTableWithKeyValues(Random* rnd, uint64_t smallest, uint64_t largest,
+                              int key_size, int value_size, uint64_t interval,
+                              double ratio, int cf = 0) {
     for (auto key = smallest; key < largest; key += interval) {
       ASSERT_OK(Put(cf, Slice(Key(key, key_size)),
-                        Slice(RandomString(rnd, value_size, ratio))));
+                    Slice(RandomString(rnd, value_size, ratio))));
     }
     ASSERT_OK(Flush(cf));
   }
@@ -361,9 +356,9 @@ class CompactionJobStatsTest : public testing::Test,
   // rounds of keys are inserted into the database, as per the behavior
   // of the DeletionStatsTest.
   void SelectivelyDeleteKeys(uint64_t smallest, uint64_t largest,
-    uint64_t interval, int deletion_interval, int key_size,
-    uint64_t cutoff_key_num, CompactionJobStats* stats, int cf = 0) {
-
+                             uint64_t interval, int deletion_interval,
+                             int key_size, uint64_t cutoff_key_num,
+                             CompactionJobStats* stats, int cf = 0) {
     // interval needs to be >= 2 so that deletion entries can be inserted
     // that are intended to not result in an actual key deletion by using
     // an offset of 1 from another existing key
@@ -387,20 +382,19 @@ class CompactionJobStatsTest : public testing::Test,
 
     // Insert some deletions for keys that don't exist that
     // are both in and out of the key range
-    ASSERT_OK(Delete(cf, Key(smallest+1, key_size)));
+    ASSERT_OK(Delete(cf, Key(smallest + 1, key_size)));
     deletions_made++;
 
-    ASSERT_OK(Delete(cf, Key(smallest-1, key_size)));
+    ASSERT_OK(Delete(cf, Key(smallest - 1, key_size)));
     deletions_made++;
     num_expired++;
 
-    ASSERT_OK(Delete(cf, Key(smallest-9, key_size)));
+    ASSERT_OK(Delete(cf, Key(smallest - 9, key_size)));
     deletions_made++;
     num_expired++;
 
     ASSERT_OK(Flush(cf));
-    SetDeletionCompactionStats(stats, deletions_made, num_expired,
-      num_deleted);
+    SetDeletionCompactionStats(stats, deletions_made, num_expired, num_deleted);
   }
 };
 
@@ -440,25 +434,20 @@ class CompactionJobStatsChecker : public EventListener {
   // use ASSERT_GE and ASSERT_LE with a reasonable bias ---
   // 10% in uncompressed case and 20% when compression is used.
   virtual void Verify(const CompactionJobStats& current_stats,
-              const CompactionJobStats& stats) {
+                      const CompactionJobStats& stats) {
     // time
     ASSERT_GT(current_stats.elapsed_micros, 0U);
 
-    ASSERT_EQ(current_stats.num_input_records,
-        stats.num_input_records);
-    ASSERT_EQ(current_stats.num_input_files,
-        stats.num_input_files);
+    ASSERT_EQ(current_stats.num_input_records, stats.num_input_records);
+    ASSERT_EQ(current_stats.num_input_files, stats.num_input_files);
     ASSERT_EQ(current_stats.num_input_files_at_output_level,
-        stats.num_input_files_at_output_level);
+              stats.num_input_files_at_output_level);
 
-    ASSERT_EQ(current_stats.num_output_records,
-        stats.num_output_records);
-    ASSERT_EQ(current_stats.num_output_files,
-        stats.num_output_files);
+    ASSERT_EQ(current_stats.num_output_records, stats.num_output_records);
+    ASSERT_EQ(current_stats.num_output_files, stats.num_output_files);
 
     ASSERT_EQ(current_stats.is_full_compaction, stats.is_full_compaction);
-    ASSERT_EQ(current_stats.is_manual_compaction,
-        stats.is_manual_compaction);
+    ASSERT_EQ(current_stats.is_manual_compaction, stats.is_manual_compaction);
 
     // file size
     double kFileSizeBias = compression_enabled_ ? 0.20 : 0.10;
@@ -475,18 +464,14 @@ class CompactionJobStatsChecker : public EventListener {
     ASSERT_EQ(current_stats.total_input_raw_value_bytes,
               stats.total_input_raw_value_bytes);
 
-    ASSERT_EQ(current_stats.num_records_replaced,
-        stats.num_records_replaced);
+    ASSERT_EQ(current_stats.num_records_replaced, stats.num_records_replaced);
 
-    ASSERT_EQ(current_stats.num_corrupt_keys,
-        stats.num_corrupt_keys);
+    ASSERT_EQ(current_stats.num_corrupt_keys, stats.num_corrupt_keys);
 
-    ASSERT_EQ(
-        std::string(current_stats.smallest_output_key_prefix),
-        std::string(stats.smallest_output_key_prefix));
-    ASSERT_EQ(
-        std::string(current_stats.largest_output_key_prefix),
-        std::string(stats.largest_output_key_prefix));
+    ASSERT_EQ(std::string(current_stats.smallest_output_key_prefix),
+              std::string(stats.smallest_output_key_prefix));
+    ASSERT_EQ(std::string(current_stats.largest_output_key_prefix),
+              std::string(stats.largest_output_key_prefix));
   }
 
   // Add an expected compaction stats, which will be used to
@@ -497,9 +482,7 @@ class CompactionJobStatsChecker : public EventListener {
     expected_stats_.push(stats);
   }
 
-  void EnableCompression(bool flag) {
-    compression_enabled_ = flag;
-  }
+  void EnableCompression(bool flag) { compression_enabled_ = flag; }
 
   bool verify_next_comp_io_stats() const { return verify_next_comp_io_stats_; }
 
@@ -517,45 +500,37 @@ class CompactionJobDeletionStatsChecker : public CompactionJobStatsChecker {
   // Verifies whether two CompactionJobStats match.
   void Verify(const CompactionJobStats& current_stats,
               const CompactionJobStats& stats) override {
-    ASSERT_EQ(
-      current_stats.num_input_deletion_records,
-      stats.num_input_deletion_records);
-    ASSERT_EQ(
-        current_stats.num_expired_deletion_records,
-        stats.num_expired_deletion_records);
-    ASSERT_EQ(
-        current_stats.num_records_replaced,
-        stats.num_records_replaced);
+    ASSERT_EQ(current_stats.num_input_deletion_records,
+              stats.num_input_deletion_records);
+    ASSERT_EQ(current_stats.num_expired_deletion_records,
+              stats.num_expired_deletion_records);
+    ASSERT_EQ(current_stats.num_records_replaced, stats.num_records_replaced);
 
-    ASSERT_EQ(current_stats.num_corrupt_keys,
-        stats.num_corrupt_keys);
+    ASSERT_EQ(current_stats.num_corrupt_keys, stats.num_corrupt_keys);
   }
 };
 
 namespace {
 
-uint64_t EstimatedFileSize(
-    uint64_t num_records, size_t key_size, size_t value_size,
-    double compression_ratio = 1.0,
-    size_t block_size = 4096,
-    int bloom_bits_per_key = 10) {
+uint64_t EstimatedFileSize(uint64_t num_records, size_t key_size,
+                           size_t value_size, double compression_ratio = 1.0,
+                           size_t block_size = 4096,
+                           int bloom_bits_per_key = 10) {
   const size_t kPerKeyOverhead = 8;
   const size_t kFooterSize = 512;
 
-  uint64_t data_size =
-    static_cast<uint64_t>(
-      num_records * (key_size + value_size * compression_ratio +
-                     kPerKeyOverhead));
+  uint64_t data_size = static_cast<uint64_t>(
+      num_records *
+      (key_size + value_size * compression_ratio + kPerKeyOverhead));
 
-  return data_size + kFooterSize
-         + num_records * bloom_bits_per_key / 8      // filter block
+  return data_size + kFooterSize +
+         num_records * bloom_bits_per_key / 8        // filter block
          + data_size * (key_size + 8) / block_size;  // index block
 }
 
 namespace {
 
-void CopyPrefix(
-    const Slice& src, size_t prefix_length, std::string* dst) {
+void CopyPrefix(const Slice& src, size_t prefix_length, std::string* dst) {
   assert(prefix_length > 0);
   size_t length = src.size() > prefix_length ? prefix_length : src.size();
   dst->assign(src.data(), length);
@@ -581,28 +556,24 @@ CompactionJobStats NewManualCompactionJobStats(
   stats.num_output_files = num_output_files;
 
   stats.total_input_bytes =
-      EstimatedFileSize(
-          num_input_records / num_input_files,
-          key_size, value_size, compression_ratio) * num_input_files;
+      EstimatedFileSize(num_input_records / num_input_files, key_size,
+                        value_size, compression_ratio) *
+      num_input_files;
   stats.total_output_bytes =
-      EstimatedFileSize(
-          num_output_records / num_output_files,
-          key_size, value_size, compression_ratio) * num_output_files;
-  stats.total_input_raw_key_bytes =
-      num_input_records * (key_size + 8);
-  stats.total_input_raw_value_bytes =
-      num_input_records * value_size;
+      EstimatedFileSize(num_output_records / num_output_files, key_size,
+                        value_size, compression_ratio) *
+      num_output_files;
+  stats.total_input_raw_key_bytes = num_input_records * (key_size + 8);
+  stats.total_input_raw_value_bytes = num_input_records * value_size;
 
   stats.is_full_compaction = is_full;
   stats.is_manual_compaction = is_manual;
 
   stats.num_records_replaced = num_records_replaced;
 
-  CopyPrefix(smallest_key,
-             CompactionJobStats::kMaxPrefixLength,
+  CopyPrefix(smallest_key, CompactionJobStats::kMaxPrefixLength,
              &stats.smallest_output_key_prefix);
-  CopyPrefix(largest_key,
-             CompactionJobStats::kMaxPrefixLength,
+  CopyPrefix(largest_key, CompactionJobStats::kMaxPrefixLength,
              &stats.largest_output_key_prefix);
 
   return stats;
@@ -662,13 +633,11 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
 
     // 1st Phase: generate "num_L0_files" L0 files.
     int num_L0_files = 0;
-    for (uint64_t start_key = key_base;
-                  start_key <= key_base * kTestScale;
-                  start_key += key_base) {
-      MakeTableWithKeyValues(
-          &rnd, start_key, start_key + key_base - 1,
-          kKeySize, kValueSize, key_interval,
-          compression_ratio, 1);
+    for (uint64_t start_key = key_base; start_key <= key_base * kTestScale;
+         start_key += key_base) {
+      MakeTableWithKeyValues(&rnd, start_key, start_key + key_base - 1,
+                             kKeySize, kValueSize, key_interval,
+                             compression_ratio, 1);
       snprintf(buf, kBufSize, "%d", ++num_L0_files);
       ASSERT_EQ(std::string(buf), FilesPerLevel(1));
     }
@@ -684,13 +653,9 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
          start_key += key_base, count++) {
       smallest_key = Key(start_key, 10);
       largest_key = Key(start_key + key_base - key_interval, 10);
-      stats_checker->AddExpectedStats(
-          NewManualCompactionJobStats(
-              smallest_key, largest_key,
-              1, 0, num_keys_per_L0_file,
-              kKeySize, kValueSize,
-              1, num_keys_per_L0_file,
-              compression_ratio, 0));
+      stats_checker->AddExpectedStats(NewManualCompactionJobStats(
+          smallest_key, largest_key, 1, 0, num_keys_per_L0_file, kKeySize,
+          kValueSize, 1, num_keys_per_L0_file, compression_ratio, 0));
       ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 1U);
       TEST_Compact(0, 1, smallest_key, largest_key);
       snprintf(buf, kBufSize, "%d,%d", num_L0_files - count, count);
@@ -701,14 +666,10 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
     int num_remaining_L0 = num_L0_files - L0_compaction_count;
     smallest_key = Key(key_base * (L0_compaction_count + 1), 10);
     largest_key = Key(key_base * (kTestScale + 1) - key_interval, 10);
-    stats_checker->AddExpectedStats(
-        NewManualCompactionJobStats(
-            smallest_key, largest_key,
-            num_remaining_L0,
-            0, num_keys_per_L0_file * num_remaining_L0,
-            kKeySize, kValueSize,
-            1, num_keys_per_L0_file * num_remaining_L0,
-            compression_ratio, 0));
+    stats_checker->AddExpectedStats(NewManualCompactionJobStats(
+        smallest_key, largest_key, num_remaining_L0, 0,
+        num_keys_per_L0_file * num_remaining_L0, kKeySize, kValueSize, 1,
+        num_keys_per_L0_file * num_remaining_L0, compression_ratio, 0));
     ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 1U);
     TEST_Compact(0, 1, smallest_key, largest_key);
 
@@ -719,13 +680,11 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
 
     // 3rd Phase: generate sparse L0 files (wider key-range, same num of keys)
     int sparseness = 2;
-    for (uint64_t start_key = key_base;
-                  start_key <= key_base * kTestScale;
-                  start_key += key_base * sparseness) {
+    for (uint64_t start_key = key_base; start_key <= key_base * kTestScale;
+         start_key += key_base * sparseness) {
       MakeTableWithKeyValues(
-          &rnd, start_key, start_key + key_base * sparseness - 1,
-          kKeySize, kValueSize,
-          key_base * sparseness / num_keys_per_L0_file,
+          &rnd, start_key, start_key + key_base * sparseness - 1, kKeySize,
+          kValueSize, key_base * sparseness / num_keys_per_L0_file,
           compression_ratio, 1);
       snprintf(buf, kBufSize, "%d,%d", ++num_L0_files, num_L1_files);
       ASSERT_EQ(std::string(buf), FilesPerLevel(1));
@@ -737,21 +696,15 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
     // output files without coordinating to see if the output could fit into
     // a smaller number of files like it does when it runs sequentially
     int num_output_files = options.max_subcompactions > 1 ? 2 : 1;
-    for (uint64_t start_key = key_base;
-         num_L0_files > 1;
+    for (uint64_t start_key = key_base; num_L0_files > 1;
          start_key += key_base * sparseness) {
       smallest_key = Key(start_key, 10);
-      largest_key =
-          Key(start_key + key_base * sparseness - key_interval, 10);
-      stats_checker->AddExpectedStats(
-          NewManualCompactionJobStats(
-              smallest_key, largest_key,
-              3, 2, num_keys_per_L0_file * 3,
-              kKeySize, kValueSize,
-              num_output_files,
-              num_keys_per_L0_file * 2,  // 1/3 of the data will be updated.
-              compression_ratio,
-              num_keys_per_L0_file));
+      largest_key = Key(start_key + key_base * sparseness - key_interval, 10);
+      stats_checker->AddExpectedStats(NewManualCompactionJobStats(
+          smallest_key, largest_key, 3, 2, num_keys_per_L0_file * 3, kKeySize,
+          kValueSize, num_output_files,
+          num_keys_per_L0_file * 2,  // 1/3 of the data will be updated.
+          compression_ratio, num_keys_per_L0_file));
       ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 1U);
       Compact(1, smallest_key, largest_key);
       if (options.max_subcompactions == 1) {
@@ -766,14 +719,10 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
     // In the first sub-compaction, we expect L0 compaction.
     smallest_key = Key(key_base, 10);
     largest_key = Key(key_base * (kTestScale + 1) - key_interval, 10);
-    stats_checker->AddExpectedStats(
-        NewManualCompactionJobStats(
-            Key(key_base * (kTestScale + 1 - sparseness), 10), largest_key,
-            2, 1, num_keys_per_L0_file * 3,
-            kKeySize, kValueSize,
-            1, num_keys_per_L0_file * 2,
-            compression_ratio,
-            num_keys_per_L0_file));
+    stats_checker->AddExpectedStats(NewManualCompactionJobStats(
+        Key(key_base * (kTestScale + 1 - sparseness), 10), largest_key, 2, 1,
+        num_keys_per_L0_file * 3, kKeySize, kValueSize, 1,
+        num_keys_per_L0_file * 2, compression_ratio, num_keys_per_L0_file));
     ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 1U);
     Compact(1, smallest_key, largest_key);
 
@@ -869,7 +818,7 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
   Options options;
   options.listeners.emplace_back(stats_checker);
   options.create_if_missing = true;
-  options.level0_file_num_compaction_trigger = kTestScale+1;
+  options.level0_file_num_compaction_trigger = kTestScale + 1;
   options.num_levels = 3;
   options.compression = kNoCompression;
   options.max_bytes_for_level_multiplier = 2;
@@ -881,13 +830,10 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
   // Stage 1: Generate several L0 files and then send them to L2 by
   // using CompactRangeOptions and CompactRange(). These files will
   // have a strict subset of the keys from the full key-range
-  for (uint64_t start_key = key_base;
-                start_key <= key_base * kTestScale / 2;
-                start_key += key_base) {
-    MakeTableWithKeyValues(
-        &rnd, start_key, start_key + key_base - 1,
-        kKeySize, kValueSize, key_interval,
-        compression_ratio, 1);
+  for (uint64_t start_key = key_base; start_key <= key_base * kTestScale / 2;
+       start_key += key_base) {
+    MakeTableWithKeyValues(&rnd, start_key, start_key + key_base - 1, kKeySize,
+                           kValueSize, key_interval, compression_ratio, 1);
   }
 
   CompactRangeOptions cr_options;
@@ -897,13 +843,10 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
   ASSERT_GT(NumTableFilesAtLevel(2, 1), 0);
 
   // Stage 2: Generate files including keys from the entire key range
-  for (uint64_t start_key = key_base;
-                start_key <= key_base * kTestScale;
-                start_key += key_base) {
-    MakeTableWithKeyValues(
-        &rnd, start_key, start_key + key_base - 1,
-        kKeySize, kValueSize, key_interval,
-        compression_ratio, 1);
+  for (uint64_t start_key = key_base; start_key <= key_base * kTestScale;
+       start_key += key_base) {
+    MakeTableWithKeyValues(&rnd, start_key, start_key + key_base - 1, kKeySize,
+                           kValueSize, key_interval, compression_ratio, 1);
   }
 
   // Send these L0 files to L1
@@ -912,16 +855,16 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
 
   // Add a new record and flush so now there is a L0 file
   // with a value too (not just deletions from the next step)
-  ASSERT_OK(Put(1, Key(key_base-6, kKeySize), "test"));
+  ASSERT_OK(Put(1, Key(key_base - 6, kKeySize), "test"));
   ASSERT_OK(Flush(1));
 
   // Stage 3: Generate L0 files with some deletions so now
   // there are files with the same key range in L0, L1, and L2
   int deletion_interval = 3;
   CompactionJobStats first_compaction_stats;
-  SelectivelyDeleteKeys(key_base, largest_key_num,
-      key_interval, deletion_interval, kKeySize, cutoff_key_num,
-      &first_compaction_stats, 1);
+  SelectivelyDeleteKeys(key_base, largest_key_num, key_interval,
+                        deletion_interval, kKeySize, cutoff_key_num,
+                        &first_compaction_stats, 1);
 
   stats_checker->AddExpectedStats(first_compaction_stats);
 
@@ -932,8 +875,7 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
 namespace {
 int GetUniversalCompactionInputUnits(uint32_t num_flushes) {
   uint32_t compaction_input_units;
-  for (compaction_input_units = 1;
-       num_flushes >= compaction_input_units;
+  for (compaction_input_units = 1; num_flushes >= compaction_input_units;
        compaction_input_units *= 2) {
     if ((num_flushes & compaction_input_units) != 0) {
       return compaction_input_units > 1 ? compaction_input_units : 0;
@@ -998,13 +940,10 @@ TEST_P(CompactionJobStatsTest, UniversalCompactionTest) {
   }
   ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 3U);
 
-  for (uint64_t start_key = key_base;
-                start_key <= key_base * kTestScale;
-                start_key += key_base) {
-    MakeTableWithKeyValues(
-        &rnd, start_key, start_key + key_base - 1,
-        kKeySize, kValueSize, key_interval,
-        compression_ratio, 1);
+  for (uint64_t start_key = key_base; start_key <= key_base * kTestScale;
+       start_key += key_base) {
+    MakeTableWithKeyValues(&rnd, start_key, start_key + key_base - 1, kKeySize,
+                           kValueSize, key_interval, compression_ratio, 1);
     ASSERT_OK(static_cast_with_check<DBImpl>(db_)->TEST_WaitForCompact());
   }
   ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 0U);

db/compaction/compaction_outputs.cc

@@ -76,6 +76,272 @@ IOStatus CompactionOutputs::WriterSyncClose(const Status& input_status,
   return io_s;
 }
 
+bool CompactionOutputs::UpdateFilesToCutForTTLStates(
+    const Slice& internal_key) {
+  if (!files_to_cut_for_ttl_.empty()) {
+    const InternalKeyComparator* icmp =
+        &compaction_->column_family_data()->internal_comparator();
+    if (cur_files_to_cut_for_ttl_ != -1) {
+      // Previous key is inside the range of a file
+      if (icmp->Compare(internal_key,
+                        files_to_cut_for_ttl_[cur_files_to_cut_for_ttl_]
+                            ->largest.Encode()) > 0) {
+        next_files_to_cut_for_ttl_ = cur_files_to_cut_for_ttl_ + 1;
+        cur_files_to_cut_for_ttl_ = -1;
+        return true;
+      }
+    } else {
+      // Look for the key position
+      while (next_files_to_cut_for_ttl_ <
+             static_cast<int>(files_to_cut_for_ttl_.size())) {
+        if (icmp->Compare(internal_key,
+                          files_to_cut_for_ttl_[next_files_to_cut_for_ttl_]
+                              ->smallest.Encode()) >= 0) {
+          if (icmp->Compare(internal_key,
+                            files_to_cut_for_ttl_[next_files_to_cut_for_ttl_]
+                                ->largest.Encode()) <= 0) {
+            // With in the current file
+            cur_files_to_cut_for_ttl_ = next_files_to_cut_for_ttl_;
+            return true;
+          }
+          // Beyond the current file
+          next_files_to_cut_for_ttl_++;
+        } else {
+          // Still fall into the gap
+          break;
+        }
+      }
+    }
+  }
+  return false;
+}
+
+size_t CompactionOutputs::UpdateGrandparentBoundaryInfo(
+    const Slice& internal_key) {
+  size_t curr_key_boundary_switched_num = 0;
+  const std::vector<FileMetaData*>& grandparents = compaction_->grandparents();
+
+  if (grandparents.empty()) {
+    return curr_key_boundary_switched_num;
+  }
+  assert(!internal_key.empty());
+  InternalKey ikey;
+  ikey.DecodeFrom(internal_key);
+  assert(ikey.Valid());
+
+  const Comparator* ucmp = compaction_->column_family_data()->user_comparator();
+
+  // Move the grandparent_index_ to the file containing the current user_key.
+  // If there are multiple files containing the same user_key, make sure the
+  // index points to the last file containing the key.
+  while (grandparent_index_ < grandparents.size()) {
+    if (being_grandparent_gap_) {
+      if (sstableKeyCompare(ucmp, ikey,
+                            grandparents[grandparent_index_]->smallest) < 0) {
+        break;
+      }
+      if (seen_key_) {
+        curr_key_boundary_switched_num++;
+        grandparent_overlapped_bytes_ +=
+            grandparents[grandparent_index_]->fd.GetFileSize();
+        grandparent_boundary_switched_num_++;
+      }
+      being_grandparent_gap_ = false;
+    } else {
+      int cmp_result = sstableKeyCompare(
+          ucmp, ikey, grandparents[grandparent_index_]->largest);
+      // If it's same key, make sure grandparent_index_ is pointing to the last
+      // one.
+      if (cmp_result < 0 ||
+          (cmp_result == 0 &&
+           (grandparent_index_ == grandparents.size() - 1 ||
+            sstableKeyCompare(ucmp, ikey,
+                              grandparents[grandparent_index_ + 1]->smallest) <
+                0))) {
+        break;
+      }
+      if (seen_key_) {
+        curr_key_boundary_switched_num++;
+        grandparent_boundary_switched_num_++;
+      }
+      being_grandparent_gap_ = true;
+      grandparent_index_++;
+    }
+  }
+
+  // If the first key is in the middle of a grandparent file, adding it to the
+  // overlap
+  if (!seen_key_ && !being_grandparent_gap_) {
+    assert(grandparent_overlapped_bytes_ == 0);
+    grandparent_overlapped_bytes_ =
+        GetCurrentKeyGrandparentOverlappedBytes(internal_key);
+  }
+
+  seen_key_ = true;
+  return curr_key_boundary_switched_num;
+}
+
+uint64_t CompactionOutputs::GetCurrentKeyGrandparentOverlappedBytes(
+    const Slice& internal_key) const {
+  // no overlap with any grandparent file
+  if (being_grandparent_gap_) {
+    return 0;
+  }
+  uint64_t overlapped_bytes = 0;
+
+  const std::vector<FileMetaData*>& grandparents = compaction_->grandparents();
+  const Comparator* ucmp = compaction_->column_family_data()->user_comparator();
+  InternalKey ikey;
+  ikey.DecodeFrom(internal_key);
+#ifndef NDEBUG
+  // make sure the grandparent_index_ is pointing to the last files containing
+  // the current key.
+  int cmp_result =
+      sstableKeyCompare(ucmp, ikey, grandparents[grandparent_index_]->largest);
+  assert(
+      cmp_result < 0 ||
+      (cmp_result == 0 &&
+       (grandparent_index_ == grandparents.size() - 1 ||
+        sstableKeyCompare(
+            ucmp, ikey, grandparents[grandparent_index_ + 1]->smallest) < 0)));
+  assert(sstableKeyCompare(ucmp, ikey,
+                           grandparents[grandparent_index_]->smallest) >= 0);
+#endif
+  overlapped_bytes += grandparents[grandparent_index_]->fd.GetFileSize();
+
+  // go backwards to find all overlapped files, one key can overlap multiple
+  // files. In the following example, if the current output key is `c`, and one
+  // compaction file was cut before `c`, current `c` can overlap with 3 files:
+  //  [a b]               [c...
+  // [b, b] [c, c] [c, c] [c, d]
+  for (int64_t i = static_cast<int64_t>(grandparent_index_) - 1;
+       i >= 0 && sstableKeyCompare(ucmp, ikey, grandparents[i]->largest) == 0;
+       i--) {
+    overlapped_bytes += grandparents[i]->fd.GetFileSize();
+  }
+
+  return overlapped_bytes;
+}
+
+bool CompactionOutputs::ShouldStopBefore(const CompactionIterator& c_iter) {
+  assert(c_iter.Valid());
+  const Slice& internal_key = c_iter.key();
+  const uint64_t previous_overlapped_bytes = grandparent_overlapped_bytes_;
+  const InternalKeyComparator* icmp =
+      &compaction_->column_family_data()->internal_comparator();
+  size_t num_grandparent_boundaries_crossed = 0;
+  bool should_stop_for_ttl = false;
+  // Always update grandparent information like overlapped file number, size
+  // etc., and TTL states.
+  // If compaction_->output_level() == 0, there is no need to update grandparent
+  // info, and that `grandparent` should be empty.
+  if (compaction_->output_level() > 0) {
+    num_grandparent_boundaries_crossed =
+        UpdateGrandparentBoundaryInfo(internal_key);
+    should_stop_for_ttl = UpdateFilesToCutForTTLStates(internal_key);
+  }
+
+  if (!HasBuilder()) {
+    return false;
+  }
+
+  if (should_stop_for_ttl) {
+    return true;
+  }
+
+  // If there's user defined partitioner, check that first
+  if (partitioner_ && partitioner_->ShouldPartition(PartitionerRequest(
+                          last_key_for_partitioner_, c_iter.user_key(),
+                          current_output_file_size_)) == kRequired) {
+    return true;
+  }
+
+  // files output to Level 0 won't be split
+  if (compaction_->output_level() == 0) {
+    return false;
+  }
+
+  // reach the max file size
+  if (current_output_file_size_ >= compaction_->max_output_file_size()) {
+    return true;
+  }
+
+  // Check if it needs to split for RoundRobin
+  // Invalid local_output_split_key indicates that we do not need to split
+  if (local_output_split_key_ != nullptr && !is_split_) {
+    // Split occurs when the next key is larger than/equal to the cursor
+    if (icmp->Compare(internal_key, local_output_split_key_->Encode()) >= 0) {
+      is_split_ = true;
+      return true;
+    }
+  }
+
+  // only check if the current key is going to cross the grandparents file
+  // boundary (either the file beginning or ending).
+  if (num_grandparent_boundaries_crossed > 0) {
+    // Cut the file before the current key if the size of the current output
+    // file + its overlapped grandparent files is bigger than
+    // max_compaction_bytes. Which is to prevent future bigger than
+    // max_compaction_bytes compaction from the current output level.
+    if (grandparent_overlapped_bytes_ + current_output_file_size_ >
+        compaction_->max_compaction_bytes()) {
+      return true;
+    }
+
+    // Cut the file if including the key is going to add a skippable file on
+    // the grandparent level AND its size is reasonably big (1/8 of target file
+    // size). For example, if it's compacting the files L0 + L1:
+    //  L0:  [1,   21]
+    //  L1:    [3,   23]
+    //  L2: [2, 4] [11, 15] [22, 24]
+    // Without this break, it will output as:
+    //  L1: [1,3, 21,23]
+    // With this break, it will output as (assuming [11, 15] at L2 is bigger
+    // than 1/8 of target size):
+    //  L1: [1,3] [21,23]
+    // Then for the future compactions, [11,15] won't be included.
+    // For random datasets (either evenly distributed or skewed), it rarely
+    // triggers this condition, but if the user is adding 2 different datasets
+    // without any overlap, it may likely happen.
+    // More details, check PR #1963
+    const size_t num_skippable_boundaries_crossed =
+        being_grandparent_gap_ ? 2 : 3;
+    if (compaction_->immutable_options()->compaction_style ==
+            kCompactionStyleLevel &&
+        compaction_->immutable_options()->level_compaction_dynamic_file_size &&
+        num_grandparent_boundaries_crossed >=
+            num_skippable_boundaries_crossed &&
+        grandparent_overlapped_bytes_ - previous_overlapped_bytes >
+            compaction_->target_output_file_size() / 8) {
+      return true;
+    }
+
+    // Pre-cut the output file if it's reaching a certain size AND it's at the
+    // boundary of a grandparent file. It can reduce the future compaction size,
+    // the cost is having smaller files.
+    // The pre-cut size threshold is based on how many grandparent boundaries
+    // it has seen before. Basically, if it has seen no boundary at all, then it
+    // will pre-cut at 50% target file size. Every boundary it has seen
+    // increases the threshold by 5%, max at 90%, which it will always cut.
+    // The idea is based on if it has seen more boundaries before, it will more
+    // likely to see another boundary (file cutting opportunity) before the
+    // target file size. The test shows it can generate larger files than a
+    // static threshold like 75% and has a similar write amplification
+    // improvement.
+    if (compaction_->immutable_options()->compaction_style ==
+            kCompactionStyleLevel &&
+        compaction_->immutable_options()->level_compaction_dynamic_file_size &&
+        current_output_file_size_ >=
+            ((compaction_->target_output_file_size() + 99) / 100) *
+                (50 + std::min(grandparent_boundary_switched_num_ * 5,
+                               size_t{40}))) {
+      return true;
+    }
+  }
+
+  return false;
+}
+
 Status CompactionOutputs::AddToOutput(
     const CompactionIterator& c_iter,
     const CompactionFileOpenFunc& open_file_func,
@@ -83,33 +349,28 @@ Status CompactionOutputs::AddToOutput(
   Status s;
   const Slice& key = c_iter.key();
 
-  if (!pending_close_ && c_iter.Valid() && partitioner_ && HasBuilder() &&
-      partitioner_->ShouldPartition(
-          PartitionerRequest(last_key_for_partitioner_, c_iter.user_key(),
-                             current_output_file_size_)) == kRequired) {
-    pending_close_ = true;
-  }
-
-  if (pending_close_) {
+  if (ShouldStopBefore(c_iter) && HasBuilder()) {
     s = close_file_func(*this, c_iter.InputStatus(), key);
-    pending_close_ = false;
-  }
-  if (!s.ok()) {
-    return s;
+    if (!s.ok()) {
+      return s;
+    }
+    // reset grandparent information
+    grandparent_boundary_switched_num_ = 0;
+    grandparent_overlapped_bytes_ =
+        GetCurrentKeyGrandparentOverlappedBytes(key);
   }
 
   // Open output file if necessary
   if (!HasBuilder()) {
     s = open_file_func(*this);
-  }
-  if (!s.ok()) {
-    return s;
+    if (!s.ok()) {
+      return s;
+    }
   }
 
-  Output& curr = current_output();
   assert(builder_ != nullptr);
   const Slice& value = c_iter.value();
-  s = curr.validator.Add(key, value);
+  s = current_output().validator.Add(key, value);
   if (!s.ok()) {
     return s;
   }
@@ -130,19 +391,6 @@ Status CompactionOutputs::AddToOutput(
   s = current_output().meta.UpdateBoundaries(key, value, ikey.sequence,
                                              ikey.type);
 
-  // Close output file if it is big enough. Two possibilities determine it's
-  // time to close it: (1) the current key should be this file's last key, (2)
-  // the next key should not be in this file.
-  //
-  // TODO(aekmekji): determine if file should be closed earlier than this
-  // during subcompactions (i.e. if output size, estimated by input size, is
-  // going to be 1.2MB and max_output_file_size = 1MB, prefer to have 0.6MB
-  // and 0.6MB instead of 1MB and 0.2MB)
-  if (compaction_->output_level() != 0 &&
-      current_output_file_size_ >= compaction_->max_output_file_size()) {
-    pending_close_ = true;
-  }
-
   if (partitioner_) {
     last_key_for_partitioner_.assign(c_iter.user_key().data_,
                                      c_iter.user_key().size_);
@@ -152,10 +400,10 @@ Status CompactionOutputs::AddToOutput(
 }
 
 Status CompactionOutputs::AddRangeDels(
-    const Slice* comp_start, const Slice* comp_end,
+    const Slice* comp_start_user_key, const Slice* comp_end_user_key,
     CompactionIterationStats& range_del_out_stats, bool bottommost_level,
     const InternalKeyComparator& icmp, SequenceNumber earliest_snapshot,
-    const Slice& next_table_min_key) {
+    const Slice& next_table_min_key, const std::string& full_history_ts_low) {
   assert(HasRangeDel());
   FileMetaData& meta = current_output().meta;
   const Comparator* ucmp = icmp.user_comparator();
@@ -165,11 +413,22 @@ Status CompactionOutputs::AddRangeDels(
   const Slice *lower_bound, *upper_bound;
   bool lower_bound_from_sub_compact = false;
 
+  // The following example does not happen since
+  // CompactionOutput::ShouldStopBefore() always return false for the first
+  // point key. But we should consider removing this dependency. Suppose for the
+  // first compaction output file,
+  //  - next_table_min_key.user_key == comp_start_user_key
+  //  - no point key is in the output file
+  //  - there is a range tombstone @seqno to be added that covers
+  //  comp_start_user_key
+  // Then meta.smallest will be set to comp_start_user_key@seqno
+  // and meta.largest will be set to comp_start_user_key@kMaxSequenceNumber
+  // which violates the assumption that meta.smallest should be <= meta.largest.
   size_t output_size = outputs_.size();
   if (output_size == 1) {
     // For the first output table, include range tombstones before the min
     // key but after the subcompaction boundary.
-    lower_bound = comp_start;
+    lower_bound = comp_start_user_key;
     lower_bound_from_sub_compact = true;
   } else if (meta.smallest.size() > 0) {
     // For subsequent output tables, only include range tombstones from min
@@ -189,21 +448,22 @@ Status CompactionOutputs::AddRangeDels(
     // use the smaller key as the upper bound of the output file, to ensure
     // that there is no overlapping between different output files.
     upper_bound_guard = ExtractUserKey(next_table_min_key);
-    if (comp_end != nullptr &&
-        ucmp->Compare(upper_bound_guard, *comp_end) >= 0) {
-      upper_bound = comp_end;
+    if (comp_end_user_key != nullptr &&
+        ucmp->CompareWithoutTimestamp(upper_bound_guard, *comp_end_user_key) >=
+            0) {
+      upper_bound = comp_end_user_key;
     } else {
       upper_bound = &upper_bound_guard;
     }
   } else {
     // This is the last file in the subcompaction, so extend until the
     // subcompaction ends.
-    upper_bound = comp_end;
+    upper_bound = comp_end_user_key;
   }
   bool has_overlapping_endpoints;
   if (upper_bound != nullptr && meta.largest.size() > 0) {
-    has_overlapping_endpoints =
-        ucmp->Compare(meta.largest.user_key(), *upper_bound) == 0;
+    has_overlapping_endpoints = ucmp->CompareWithoutTimestamp(
+                                    meta.largest.user_key(), *upper_bound) == 0;
   } else {
     has_overlapping_endpoints = false;
   }
@@ -212,8 +472,8 @@ Status CompactionOutputs::AddRangeDels(
   // bound. If the end of subcompaction is null or the upper bound is null,
   // it means that this file is the last file in the compaction. So there
   // will be no overlapping between this file and others.
-  assert(comp_end == nullptr || upper_bound == nullptr ||
-         ucmp->Compare(*upper_bound, *comp_end) <= 0);
+  assert(comp_end_user_key == nullptr || upper_bound == nullptr ||
+         ucmp->CompareWithoutTimestamp(*upper_bound, *comp_end_user_key) <= 0);
   auto it = range_del_agg_->NewIterator(lower_bound, upper_bound,
                                         has_overlapping_endpoints);
   // Position the range tombstone output iterator. There may be tombstone
@@ -224,23 +484,48 @@ Status CompactionOutputs::AddRangeDels(
   } else {
     it->SeekToFirst();
   }
+  Slice last_tombstone_start_user_key{};
   for (; it->Valid(); it->Next()) {
     auto tombstone = it->Tombstone();
     if (upper_bound != nullptr) {
-      int cmp = ucmp->Compare(*upper_bound, tombstone.start_key_);
-      if ((has_overlapping_endpoints && cmp < 0) ||
-          (!has_overlapping_endpoints && cmp <= 0)) {
-        // Tombstones starting after upper_bound only need to be included in
-        // the next table. If the current SST ends before upper_bound, i.e.,
-        // `has_overlapping_endpoints == false`, we can also skip over range
-        // tombstones that start exactly at upper_bound. Such range
-        // tombstones will be included in the next file and are not relevant
-        // to the point keys or endpoints of the current file.
+      int cmp =
+          ucmp->CompareWithoutTimestamp(*upper_bound, tombstone.start_key_);
+      // Tombstones starting after upper_bound only need to be included in
+      // the next table.
+      // If the current SST ends before upper_bound, i.e.,
+      // `has_overlapping_endpoints == false`, we can also skip over range
+      // tombstones that start exactly at upper_bound. Such range
+      // tombstones will be included in the next file and are not relevant
+      // to the point keys or endpoints of the current file.
+      // If the current SST ends at the same user key at upper_bound,
+      // i.e., `has_overlapping_endpoints == true`, AND the tombstone has
+      // the same start key as upper_bound, i.e., cmp == 0, then
+      // the tombstone is relevant only if the tombstone's sequence number
+      // is no larger than this file's largest key's sequence number. This
+      // is because the upper bound to truncate this file's range tombstone
+      // will be meta.largest in this case, and any tombstone that starts after
+      // it will not be relevant.
+      if (cmp < 0) {
         break;
+      } else if (cmp == 0) {
+        if (!has_overlapping_endpoints ||
+            tombstone.seq_ < GetInternalKeySeqno(meta.largest.Encode())) {
+          break;
+        }
       }
     }
 
-    if (bottommost_level && tombstone.seq_ <= earliest_snapshot) {
+    const size_t ts_sz = ucmp->timestamp_size();
+    // Garbage collection for range tombstones.
+    // If user-defined timestamp is enabled, range tombstones are dropped if
+    // they are at bottommost_level, below full_history_ts_low and not visible
+    // in any snapshot. trim_ts_ is passed to the constructor for
+    // range_del_agg_, and range_del_agg_ internally drops tombstones above
+    // trim_ts_.
+    if (bottommost_level && tombstone.seq_ <= earliest_snapshot &&
+        (ts_sz == 0 ||
+         (!full_history_ts_low.empty() &&
+          ucmp->CompareTimestamp(tombstone.ts_, full_history_ts_low) < 0))) {
       // TODO(andrewkr): tombstones that span multiple output files are
       // counted for each compaction output file, so lots of double
       // counting.
@@ -251,38 +536,50 @@ Status CompactionOutputs::AddRangeDels(
 
     auto kv = tombstone.Serialize();
     assert(lower_bound == nullptr ||
-           ucmp->Compare(*lower_bound, kv.second) < 0);
+           ucmp->CompareWithoutTimestamp(*lower_bound, kv.second) < 0);
     // Range tombstone is not supported by output validator yet.
     builder_->Add(kv.first.Encode(), kv.second);
-    InternalKey smallest_candidate = std::move(kv.first);
+    InternalKey tombstone_start = std::move(kv.first);
+    InternalKey smallest_candidate{tombstone_start};
     if (lower_bound != nullptr &&
-        ucmp->Compare(smallest_candidate.user_key(), *lower_bound) <= 0) {
+        ucmp->CompareWithoutTimestamp(smallest_candidate.user_key(),
+                                      *lower_bound) <= 0) {
       // Pretend the smallest key has the same user key as lower_bound
       // (the max key in the previous table or subcompaction) in order for
       // files to appear key-space partitioned.
-      //
-      // When lower_bound is chosen by a subcompaction, we know that
-      // subcompactions over smaller keys cannot contain any keys at
-      // lower_bound. We also know that smaller subcompactions exist,
-      // because otherwise the subcompaction woud be unbounded on the left.
-      // As a result, we know that no other files on the output level will
-      // contain actual keys at lower_bound (an output file may have a
-      // largest key of lower_bound@kMaxSequenceNumber, but this only
-      // indicates a large range tombstone was truncated). Therefore, it is
-      // safe to use the tombstone's sequence number, to ensure that keys at
-      // lower_bound at lower levels are covered by truncated tombstones.
-      //
-      // If lower_bound was chosen by the smallest data key in the file,
-      // choose lowest seqnum so this file's smallest internal key comes
-      // after the previous file's largest. The fake seqnum is OK because
-      // the read path's file-picking code only considers user key.
-      smallest_candidate = InternalKey(
-          *lower_bound, lower_bound_from_sub_compact ? tombstone.seq_ : 0,
-          kTypeRangeDeletion);
+      if (lower_bound_from_sub_compact) {
+        // When lower_bound is chosen by a subcompaction
+        // (lower_bound_from_sub_compact), we know that subcompactions over
+        // smaller keys cannot contain any keys at lower_bound. We also know
+        // that smaller subcompactions exist, because otherwise the
+        // subcompaction woud be unbounded on the left. As a result, we know
+        // that no other files on the output level will contain actual keys at
+        // lower_bound (an output file may have a largest key of
+        // lower_bound@kMaxSequenceNumber, but this only indicates a large range
+        // tombstone was truncated). Therefore, it is safe to use the
+        // tombstone's sequence number, to ensure that keys at lower_bound at
+        // lower levels are covered by truncated tombstones.
+        if (ts_sz) {
+          assert(tombstone.ts_.size() == ts_sz);
+          smallest_candidate = InternalKey(*lower_bound, tombstone.seq_,
+                                           kTypeRangeDeletion, tombstone.ts_);
+        } else {
+          smallest_candidate =
+              InternalKey(*lower_bound, tombstone.seq_, kTypeRangeDeletion);
+        }
+      } else {
+        // If lower_bound was chosen by the smallest data key in the file,
+        // choose lowest seqnum so this file's smallest internal key comes
+        // after the previous file's largest. The fake seqnum is OK because
+        // the read path's file-picking code only considers user key.
+        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+      }
     }
-    InternalKey largest_candidate = tombstone.SerializeEndKey();
+    InternalKey tombstone_end = tombstone.SerializeEndKey();
+    InternalKey largest_candidate{tombstone_end};
     if (upper_bound != nullptr &&
-        ucmp->Compare(*upper_bound, largest_candidate.user_key()) <= 0) {
+        ucmp->CompareWithoutTimestamp(*upper_bound,
+                                      largest_candidate.user_key()) <= 0) {
       // Pretend the largest key has the same user key as upper_bound (the
       // min key in the following table or subcompaction) in order for files
       // to appear key-space partitioned.
@@ -296,26 +593,247 @@ Status CompactionOutputs::AddRangeDels(
       // kMaxSequenceNumber), but with kTypeDeletion (0x7) instead of
       // kTypeRangeDeletion (0xF), so the range tombstone comes before the
       // Seek() key in InternalKey's ordering. So Seek() will look in the
-      // next file for the user key.
-      largest_candidate =
-          InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      // next file for the user key
+      if (ts_sz) {
+        static constexpr char kTsMax[] = "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
+        if (ts_sz <= strlen(kTsMax)) {
+          largest_candidate =
+              InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion,
+                          Slice(kTsMax, ts_sz));
+        } else {
+          largest_candidate =
+              InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion,
+                          std::string(ts_sz, '\xff'));
+        }
+      } else {
+        largest_candidate =
+            InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      }
     }
-#ifndef NDEBUG
-    SequenceNumber smallest_ikey_seqnum = kMaxSequenceNumber;
-    if (meta.smallest.size() > 0) {
-      smallest_ikey_seqnum = GetInternalKeySeqno(meta.smallest.Encode());
-    }
-#endif
     meta.UpdateBoundariesForRange(smallest_candidate, largest_candidate,
                                   tombstone.seq_, icmp);
-    // The smallest key in a file is used for range tombstone truncation, so
-    // it cannot have a seqnum of 0 (unless the smallest data key in a file
-    // has a seqnum of 0). Otherwise, the truncated tombstone may expose
-    // deleted keys at lower levels.
-    assert(smallest_ikey_seqnum == 0 ||
-           ExtractInternalKeyFooter(meta.smallest.Encode()) !=
-               PackSequenceAndType(0, kTypeRangeDeletion));
+    if (!bottommost_level) {
+      bool start_user_key_changed =
+          last_tombstone_start_user_key.empty() ||
+          ucmp->CompareWithoutTimestamp(last_tombstone_start_user_key,
+                                        it->start_key()) < 0;
+      last_tombstone_start_user_key = it->start_key();
+      // Range tombstones are truncated at file boundaries
+      if (icmp.Compare(tombstone_start, meta.smallest) < 0) {
+        tombstone_start = meta.smallest;
+      }
+      if (icmp.Compare(tombstone_end, meta.largest) > 0) {
+        tombstone_end = meta.largest;
+      }
+      // this assertion validates invariant (2) in the comment below.
+      assert(icmp.Compare(tombstone_start, tombstone_end) <= 0);
+      if (start_user_key_changed) {
+        // if tombstone_start >= tombstone_end, then either no key range is
+        // covered, or that they have the same user key. If they have the same
+        // user key, then the internal key range should only be within this
+        // level, and no keys from older levels is covered.
+        if (ucmp->CompareWithoutTimestamp(tombstone_start.user_key(),
+                                          tombstone_end.user_key()) < 0) {
+          SizeApproximationOptions approx_opts;
+          approx_opts.files_size_error_margin = 0.1;
+          auto approximate_covered_size =
+              compaction_->input_version()->version_set()->ApproximateSize(
+                  approx_opts, compaction_->input_version(),
+                  tombstone_start.Encode(), tombstone_end.Encode(),
+                  compaction_->output_level() + 1 /* start_level */,
+                  -1 /* end_level */, kCompaction);
+          meta.compensated_range_deletion_size += approximate_covered_size;
+        }
+      }
+    }
+    // TODO: show invariants that ensure all necessary range tombstones are
+    // added
+    //  and that file boundaries ensure no coverage is lost.
+    // Each range tombstone with internal key range [tombstone_start,
+    // tombstone_end] is being added to the current compaction output file here.
+    // The range tombstone is going to be truncated at range [meta.smallest,
+    // meta.largest] during reading/scanning. We should maintain invariants
+    // (1) meta.smallest <= meta.largest and,
+    // (2) [tombstone_start, tombstone_end] and [meta.smallest, meta.largest]
+    // overlaps, as there is no point adding range tombstone with a range
+    // outside the file's range.
+    // Since `tombstone_end` is always some user_key@kMaxSeqno, it is okay to
+    // use either open or closed range. Using closed range here to make
+    // reasoning easier, and it is more consistent with an ongoing work that
+    // tries to simplify this method.
+    //
+    // There are two cases:
+    // Case 1. Output file has no point key:
+    //   First we show this case only happens when the entire compaction output
+    //   is range tombstone only. This is true if CompactionIterator does not
+    //   emit any point key. Suppose CompactionIterator emits some point key.
+    //   Based on the assumption that CompactionOutputs::ShouldStopBefore()
+    //   always return false for the first point key, the first compaction
+    //   output file always contains a point key. Each new compaction output
+    //   file is created if there is a point key for which ShouldStopBefore()
+    //   returns true, and the point key would be added to the new compaction
+    //   output file. So each new compaction file always contains a point key.
+    //   So Case 1 only happens when CompactionIterator does not emit any
+    //   point key.
+    //
+    //   To show (1) meta.smallest <= meta.largest:
+    //   Since the compaction output is range tombstone only, `lower_bound` and
+    //   `upper_bound` are either null or comp_start/end_user_key respectively.
+    //   According to how UpdateBoundariesForRange() is implemented, it blindly
+    //   updates meta.smallest and meta.largest to smallest_candidate and
+    //   largest_candidate the first time it is called. Subsequently, it
+    //   compares input parameter with meta.smallest and meta.largest and only
+    //   updates them when input is smaller/larger. So we only need to show
+    //   smallest_candidate <= largest_candidate the first time
+    //   UpdateBoundariesForRange() is called. Here we show something stronger
+    //   that smallest_candidate.user_key < largest_candidate.user_key always
+    //   hold for Case 1.
+    //   We assume comp_start_user_key < comp_end_user_key, if provided. We
+    //   assume that tombstone_start < tombstone_end. This assumption is based
+    //   on that each fragment in FragmentedTombstoneList has
+    //   start_key < end_key (user_key) and that
+    //   FragmentedTombstoneIterator::Tombstone() returns the pair
+    //   (start_key@tombstone_seqno with op_type kTypeRangeDeletion, end_key).
+    //   The logic in this loop sets smallest_candidate to
+    //   max(tombstone_start.user_key, comp_start_user_key)@tombstone.seq_ with
+    //   op_type kTypeRangeDeletion, largest_candidate to
+    //   min(tombstone_end.user_key, comp_end_user_key)@kMaxSequenceNumber with
+    //   op_type kTypeRangeDeletion. When a bound is null, there is no
+    //   truncation on that end. To show that smallest_candidate.user_key <
+    //   largest_candidate.user_key, it suffices to show
+    //   tombstone_start.user_key < comp_end_user_key (if not null) AND
+    //   comp_start_user_key (if not null) < tombstone_end.user_key.
+    //   Since the file has no point key, `has_overlapping_endpoints` is false.
+    //   In the first sanity check of this for-loop, we compare
+    //   tombstone_start.user_key against upper_bound = comp_end_user_key,
+    //   and only proceed if tombstone_start.user_key < comp_end_user_key.
+    //   We assume FragmentedTombstoneIterator::Seek(k) lands
+    //   on a tombstone with end_key > k. So the call it->Seek(*lower_bound)
+    //   above implies compact_start_user_key < tombstone_end.user_key.
+    //
+    //   To show (2) [tombstone_start, tombstone_end] and [meta.smallest,
+    //   meta.largest] overlaps (after the call to UpdateBoundariesForRange()):
+    //   In the proof for (1) we have shown that
+    //   smallest_candidate <= largest_candidate. Since tombstone_start <=
+    //   smallest_candidate <= largest_candidate <= tombstone_end, for (2) to
+    //   hold, it suffices to show that [smallest_candidate, largest_candidate]
+    //   overlaps with [meta.smallest, meta.largest]. too.
+    //   Given meta.smallest <= meta.largest shown above, we need to show
+    //   that it is impossible to have largest_candidate < meta.smallest or
+    //   meta.largest < smallest_candidate. If the above
+    //   meta.UpdateBoundariesForRange(smallest_candidate, largest_candidate)
+    //   updates meta.largest or meta.smallest, then the two ranges overlap.
+    //   So we assume meta.UpdateBoundariesForRange(smallest_candidate,
+    //   largest_candidate) did not update meta.smallest nor meta.largest, which
+    //   means meta.smallest < smallest_candidate and largest_candidate <
+    //   meta.largest.
+    //
+    // Case 2. Output file has >= 1 point key. This means meta.smallest and
+    // meta.largest are not empty when AddRangeDels() is called.
+    //   To show (1) meta.smallest <= meta.largest:
+    //   Assume meta.smallest <= meta.largest when AddRangeDels() is called,
+    //   this follow from how UpdateBoundariesForRange() is implemented where it
+    //   takes min or max to update meta.smallest or meta.largest.
+    //
+    //   To show (2) [tombstone_start, tombstone_end] and [meta.smallest,
+    //   meta.largest] overlaps (after the call to UpdateBoundariesForRange()):
+    //   When smallest_candidate <= largest_candidate, the proof in Case 1
+    //   applies, so we only need to show (2) holds when smallest_candidate >
+    //   largest_candidate. When both bounds are either null or from
+    //   subcompaction boundary, the proof in Case 1 applies, so we only need to
+    //   show (2) holds when at least one bound is from a point key (either
+    //   meta.smallest for lower bound or next_table_min_key for upper bound).
+    //
+    //   Suppose lower bound is meta.smallest.user_key. The call
+    //   it->Seek(*lower_bound) implies tombstone_end.user_key >
+    //   meta.smallest.user_key. We have smallest_candidate.user_key =
+    //   max(tombstone_start.user_key, meta.smallest.user_key). For
+    //   smallest_candidate to be > largest_candidate, we need
+    //   largest_candidate.user_key = upper_bound = smallest_candidate.user_key,
+    //   where tombstone_end is truncated to largest_candidate.
+    //   Subcase 1:
+    //   Suppose largest_candidate.user_key = comp_end_user_key (there is no
+    //   next point key). Subcompaction ensures any point key from this
+    //   subcompaction has a user_key < comp_end_user_key, so 1)
+    //   meta.smallest.user_key < comp_end_user_key, 2)
+    //   `has_overlapping_endpoints` is false, and the first if condition in
+    //   this for-loop ensures tombstone_start.user_key < comp_end_user_key. So
+    //   smallest_candidate.user_key < largest_candidate.user_key. This case
+    //   cannot happen when smallest > largest_candidate.
+    //   Subcase 2:
+    //   Suppose largest_candidate.user_key = next_table_min_key.user_key.
+    //   The first if condition in this for-loop together with
+    //   smallest_candidate.user_key = next_table_min_key.user_key =
+    //   upper_bound implies `has_overlapping_endpoints` is true (so meta
+    //   largest.user_key = upper_bound) and
+    //   tombstone.seq_ < meta.largest.seqno. So
+    //   tombstone_start < meta.largest < tombstone_end.
+    //
+    //   Suppose lower bound is comp_start_user_key and upper_bound is
+    //   next_table_min_key. The call it->Seek(*lower_bound) implies we have
+    //   tombstone_end_key.user_key > comp_start_user_key. So
+    //   tombstone_end_key.user_key > smallest_candidate.user_key. For
+    //   smallest_candidate to be > largest_candidate, we need
+    //   tombstone_start.user_key = largest_candidate.user_key = upper_bound =
+    //   next_table_min_key.user_key. This means `has_overlapping_endpoints` is
+    //   true (so meta.largest.user_key = upper_bound) and tombstone.seq_ <
+    //   meta.largest.seqno. So tombstone_start < meta.largest < tombstone_end.
   }
   return Status::OK();
 }
+
+void CompactionOutputs::FillFilesToCutForTtl() {
+  if (compaction_->immutable_options()->compaction_style !=
+          kCompactionStyleLevel ||
+      compaction_->immutable_options()->compaction_pri !=
+          kMinOverlappingRatio ||
+      compaction_->mutable_cf_options()->ttl == 0 ||
+      compaction_->num_input_levels() < 2 || compaction_->bottommost_level()) {
+    return;
+  }
+
+  // We define new file with the oldest ancestor time to be younger than 1/4
+  // TTL, and an old one to be older than 1/2 TTL time.
+  int64_t temp_current_time;
+  auto get_time_status =
+      compaction_->immutable_options()->clock->GetCurrentTime(
+          &temp_current_time);
+  if (!get_time_status.ok()) {
+    return;
+  }
+
+  auto current_time = static_cast<uint64_t>(temp_current_time);
+  if (current_time < compaction_->mutable_cf_options()->ttl) {
+    return;
+  }
+
+  uint64_t old_age_thres =
+      current_time - compaction_->mutable_cf_options()->ttl / 2;
+  const std::vector<FileMetaData*>& olevel =
+      *(compaction_->inputs(compaction_->num_input_levels() - 1));
+  for (FileMetaData* file : olevel) {
+    // Worth filtering out by start and end?
+    uint64_t oldest_ancester_time = file->TryGetOldestAncesterTime();
+    // We put old files if they are not too small to prevent a flood
+    // of small files.
+    if (oldest_ancester_time < old_age_thres &&
+        file->fd.GetFileSize() >
+            compaction_->mutable_cf_options()->target_file_size_base / 2) {
+      files_to_cut_for_ttl_.push_back(file);
+    }
+  }
+}
+
+CompactionOutputs::CompactionOutputs(const Compaction* compaction,
+                                     const bool is_penultimate_level)
+    : compaction_(compaction), is_penultimate_level_(is_penultimate_level) {
+  partitioner_ = compaction->output_level() == 0
+                     ? nullptr
+                     : compaction->CreateSstPartitioner();
+
+  if (compaction->output_level() != 0) {
+    FillFilesToCutForTtl();
+  }
+}
+
 }  // namespace ROCKSDB_NAMESPACE

db/compaction/compaction_outputs.h

@@ -45,12 +45,7 @@ class CompactionOutputs {
   CompactionOutputs() = delete;
 
   explicit CompactionOutputs(const Compaction* compaction,
-                             const bool is_penultimate_level)
-      : compaction_(compaction), is_penultimate_level_(is_penultimate_level) {
-    partitioner_ = compaction->output_level() == 0
-                       ? nullptr
-                       : compaction->CreateSstPartitioner();
-  }
+                             const bool is_penultimate_level);
 
   // Add generated output to the list
   void AddOutput(FileMetaData&& meta, const InternalKeyComparator& icmp,
@@ -173,17 +168,16 @@ class CompactionOutputs {
   }
 
   // Add range-dels from the aggregator to the current output file
-  Status AddRangeDels(const Slice* comp_start, const Slice* comp_end,
+  // @param comp_start_user_key and comp_end_user_key include timestamp if
+  // user-defined timestamp is enabled.
+  // @param full_history_ts_low used for range tombstone garbage collection.
+  Status AddRangeDels(const Slice* comp_start_user_key,
+                      const Slice* comp_end_user_key,
                       CompactionIterationStats& range_del_out_stats,
                       bool bottommost_level, const InternalKeyComparator& icmp,
                       SequenceNumber earliest_snapshot,
-                      const Slice& next_table_min_key);
-
-  // Is the current file is already pending for close
-  bool IsPendingClose() const { return pending_close_; }
-
-  // Current file should close before adding a new key
-  void SetPendingClose() { pending_close_ = true; }
+                      const Slice& next_table_min_key,
+                      const std::string& full_history_ts_low);
 
   // if the outputs have range delete, range delete is also data
   bool HasRangeDel() const {
@@ -193,6 +187,32 @@ class CompactionOutputs {
  private:
   friend class SubcompactionState;
 
+  void FillFilesToCutForTtl();
+
+  void SetOutputSlitKey(const std::optional<Slice> start,
+                        const std::optional<Slice> end) {
+    const InternalKeyComparator* icmp =
+        &compaction_->column_family_data()->internal_comparator();
+
+    const InternalKey* output_split_key = compaction_->GetOutputSplitKey();
+    // Invalid output_split_key indicates that we do not need to split
+    if (output_split_key != nullptr) {
+      // We may only split the output when the cursor is in the range. Split
+      if ((!end.has_value() ||
+           icmp->user_comparator()->Compare(
+               ExtractUserKey(output_split_key->Encode()), end.value()) < 0) &&
+          (!start.has_value() || icmp->user_comparator()->Compare(
+                                     ExtractUserKey(output_split_key->Encode()),
+                                     start.value()) > 0)) {
+        local_output_split_key_ = output_split_key;
+      }
+    }
+  }
+
+  // Returns true iff we should stop building the current output
+  // before processing the current key in compaction iterator.
+  bool ShouldStopBefore(const CompactionIterator& c_iter);
+
   void Cleanup() {
     if (builder_ != nullptr) {
       // May happen if we get a shutdown call in the middle of compaction
@@ -201,11 +221,25 @@ class CompactionOutputs {
     }
   }
 
-  uint64_t GetCurrentOutputFileSize() const {
-    return current_output_file_size_;
-  }
+  // Updates states related to file cutting for TTL.
+  // Returns a boolean value indicating whether the current
+  // compaction output file should be cut before `internal_key`.
+  //
+  // @param internal_key the current key to be added to output.
+  bool UpdateFilesToCutForTTLStates(const Slice& internal_key);
 
-  // Add curent key from compaction_iterator to the output file. If needed
+  // update tracked grandparents information like grandparent index, if it's
+  // in the gap between 2 grandparent files, accumulated grandparent files size
+  // etc.
+  // It returns how many boundaries it crosses by including current key.
+  size_t UpdateGrandparentBoundaryInfo(const Slice& internal_key);
+
+  // helper function to get the overlapped grandparent files size, it's only
+  // used for calculating the first key's overlap.
+  uint64_t GetCurrentKeyGrandparentOverlappedBytes(
+      const Slice& internal_key) const;
+
+  // Add current key from compaction_iterator to the output file. If needed
   // close and open new compaction output with the functions provided.
   Status AddToOutput(const CompactionIterator& c_iter,
                      const CompactionFileOpenFunc& open_file_func,
@@ -255,10 +289,6 @@ class CompactionOutputs {
 
   const Compaction* compaction_;
 
-  // The current file is pending close, which needs to run `close_file_func()`
-  // first to add a new key.
-  bool pending_close_ = false;
-
   // current output builder and writer
   std::unique_ptr<TableBuilder> builder_;
   std::unique_ptr<WritableFileWriter> file_writer_;
@@ -282,6 +312,39 @@ class CompactionOutputs {
   // partitioner information
   std::string last_key_for_partitioner_;
   std::unique_ptr<SstPartitioner> partitioner_;
+
+  // A flag determines if this subcompaction has been split by the cursor
+  bool is_split_ = false;
+
+  // We also maintain the output split key for each subcompaction to avoid
+  // repetitive comparison in ShouldStopBefore()
+  const InternalKey* local_output_split_key_ = nullptr;
+
+  // Some identified files with old oldest ancester time and the range should be
+  // isolated out so that the output file(s) in that range can be merged down
+  // for TTL and clear the timestamps for the range.
+  std::vector<FileMetaData*> files_to_cut_for_ttl_;
+  int cur_files_to_cut_for_ttl_ = -1;
+  int next_files_to_cut_for_ttl_ = 0;
+
+  // An index that used to speed up ShouldStopBefore().
+  size_t grandparent_index_ = 0;
+
+  // if the output key is being grandparent files gap, so:
+  //  key > grandparents[grandparent_index_ - 1].largest &&
+  //  key < grandparents[grandparent_index_].smallest
+  bool being_grandparent_gap_ = true;
+
+  // The number of bytes overlapping between the current output and
+  // grandparent files used in ShouldStopBefore().
+  uint64_t grandparent_overlapped_bytes_ = 0;
+
+  // A flag determines whether the key has been seen in ShouldStopBefore()
+  bool seen_key_ = false;
+
+  // for the current output file, how many file boundaries has it crossed,
+  // basically number of files overlapped * 2
+  size_t grandparent_boundary_switched_num_ = 0;
 };
 
 // helper struct to concatenate the last level and penultimate level outputs

db/compaction/compaction_picker.cc

@@ -27,44 +27,20 @@
 
 namespace ROCKSDB_NAMESPACE {
 
-namespace {
-uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) {
-  uint64_t sum = 0;
-  for (size_t i = 0; i < files.size() && files[i]; i++) {
-    sum += files[i]->compensated_file_size;
-  }
-  return sum;
-}
-}  // anonymous namespace
-
 bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
                            size_t min_files_to_compact,
                            uint64_t max_compact_bytes_per_del_file,
                            uint64_t max_compaction_bytes,
-                           CompactionInputFiles* comp_inputs,
-                           SequenceNumber earliest_mem_seqno) {
-  // Do not pick ingested file when there is at least one memtable not flushed
-  // which of seqno is overlap with the sst.
+                           CompactionInputFiles* comp_inputs) {
   TEST_SYNC_POINT("FindIntraL0Compaction");
+
   size_t start = 0;
-  for (; start < level_files.size(); start++) {
-    if (level_files[start]->being_compacted) {
-      return false;
-    }
-    // If there is no data in memtable, the earliest sequence number would the
-    // largest sequence number in last memtable.
-    // Because all files are sorted in descending order by largest_seqno, so we
-    // only need to check the first one.
-    if (level_files[start]->fd.largest_seqno <= earliest_mem_seqno) {
-      break;
-    }
-  }
-  if (start >= level_files.size()) {
+
+  if (level_files.size() == 0 || level_files[start]->being_compacted) {
     return false;
   }
+
   size_t compact_bytes = static_cast<size_t>(level_files[start]->fd.file_size);
-  uint64_t compensated_compact_bytes =
-      level_files[start]->compensated_file_size;
   size_t compact_bytes_per_del_file = std::numeric_limits<size_t>::max();
   // Compaction range will be [start, limit).
   size_t limit;
@@ -73,11 +49,10 @@ bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
   size_t new_compact_bytes_per_del_file = 0;
   for (limit = start + 1; limit < level_files.size(); ++limit) {
     compact_bytes += static_cast<size_t>(level_files[limit]->fd.file_size);
-    compensated_compact_bytes += level_files[limit]->compensated_file_size;
     new_compact_bytes_per_del_file = compact_bytes / (limit - start);
     if (level_files[limit]->being_compacted ||
         new_compact_bytes_per_del_file > compact_bytes_per_del_file ||
-        compensated_compact_bytes > max_compaction_bytes) {
+        compact_bytes > max_compaction_bytes) {
       break;
     }
     compact_bytes_per_del_file = new_compact_bytes_per_del_file;
@@ -288,8 +263,10 @@ bool CompactionPicker::RangeOverlapWithCompaction(
   const Comparator* ucmp = icmp_->user_comparator();
   for (Compaction* c : compactions_in_progress_) {
     if (c->output_level() == level &&
-        ucmp->Compare(smallest_user_key, c->GetLargestUserKey()) <= 0 &&
-        ucmp->Compare(largest_user_key, c->GetSmallestUserKey()) >= 0) {
+        ucmp->CompareWithoutTimestamp(smallest_user_key,
+                                      c->GetLargestUserKey()) <= 0 &&
+        ucmp->CompareWithoutTimestamp(largest_user_key,
+                                      c->GetSmallestUserKey()) >= 0) {
       // Overlap
       return true;
     }
@@ -305,13 +282,12 @@ bool CompactionPicker::RangeOverlapWithCompaction(
 }
 
 bool CompactionPicker::FilesRangeOverlapWithCompaction(
-    const std::vector<CompactionInputFiles>& inputs, int level) const {
+    const std::vector<CompactionInputFiles>& inputs, int level,
+    int penultimate_level) const {
   bool is_empty = true;
-  int start_level = -1;
   for (auto& in : inputs) {
     if (!in.empty()) {
       is_empty = false;
-      start_level = in.level;  // inputs are sorted by level
       break;
     }
   }
@@ -320,17 +296,24 @@ bool CompactionPicker::FilesRangeOverlapWithCompaction(
     return false;
   }
 
+  // TODO: Intra L0 compactions can have the ranges overlapped, but the input
+  //  files cannot be overlapped in the order of L0 files.
   InternalKey smallest, largest;
   GetRange(inputs, &smallest, &largest, Compaction::kInvalidLevel);
-  int penultimate_level =
-      Compaction::EvaluatePenultimateLevel(ioptions_, start_level, level);
   if (penultimate_level != Compaction::kInvalidLevel) {
-    InternalKey penultimate_smallest, penultimate_largest;
-    GetRange(inputs, &penultimate_smallest, &penultimate_largest, level);
-    if (RangeOverlapWithCompaction(penultimate_smallest.user_key(),
-                                   penultimate_largest.user_key(),
-                                   penultimate_level)) {
-      return true;
+    if (ioptions_.compaction_style == kCompactionStyleUniversal) {
+      if (RangeOverlapWithCompaction(smallest.user_key(), largest.user_key(),
+                                     penultimate_level)) {
+        return true;
+      }
+    } else {
+      InternalKey penultimate_smallest, penultimate_largest;
+      GetRange(inputs, &penultimate_smallest, &penultimate_largest, level);
+      if (RangeOverlapWithCompaction(penultimate_smallest.user_key(),
+                                     penultimate_largest.user_key(),
+                                     penultimate_level)) {
+        return true;
+      }
     }
   }
 
@@ -354,11 +337,25 @@ Compaction* CompactionPicker::CompactFiles(
     const std::vector<CompactionInputFiles>& input_files, int output_level,
     VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options, uint32_t output_path_id) {
+#ifndef NDEBUG
   assert(input_files.size());
   // This compaction output should not overlap with a running compaction as
   // `SanitizeCompactionInputFiles` should've checked earlier and db mutex
   // shouldn't have been released since.
-  assert(!FilesRangeOverlapWithCompaction(input_files, output_level));
+  int start_level = Compaction::kInvalidLevel;
+  for (const auto& in : input_files) {
+    // input_files should already be sorted by level
+    if (!in.empty()) {
+      start_level = in.level;
+      break;
+    }
+  }
+  assert(output_level == 0 ||
+         !FilesRangeOverlapWithCompaction(
+             input_files, output_level,
+             Compaction::EvaluatePenultimateLevel(vstorage, ioptions_,
+                                                  start_level, output_level)));
+#endif /* !NDEBUG */
 
   CompressionType compression_type;
   if (compact_options.compression == kDisableCompressionOption) {
@@ -505,8 +502,8 @@ bool CompactionPicker::SetupOtherInputs(
   if (!output_level_inputs->empty()) {
     const uint64_t limit = mutable_cf_options.max_compaction_bytes;
     const uint64_t output_level_inputs_size =
-        TotalCompensatedFileSize(output_level_inputs->files);
-    const uint64_t inputs_size = TotalCompensatedFileSize(inputs->files);
+        TotalFileSize(output_level_inputs->files);
+    const uint64_t inputs_size = TotalFileSize(inputs->files);
     bool expand_inputs = false;
 
     CompactionInputFiles expanded_inputs;
@@ -525,13 +522,13 @@ bool CompactionPicker::SetupOtherInputs(
                                      &expanded_inputs.files, base_index,
                                      nullptr);
     }
-    uint64_t expanded_inputs_size =
-        TotalCompensatedFileSize(expanded_inputs.files);
+    uint64_t expanded_inputs_size = TotalFileSize(expanded_inputs.files);
     if (!ExpandInputsToCleanCut(cf_name, vstorage, &expanded_inputs)) {
       try_overlapping_inputs = false;
     }
     if (try_overlapping_inputs && expanded_inputs.size() > inputs->size() &&
-        output_level_inputs_size + expanded_inputs_size < limit &&
+        (mutable_cf_options.ignore_max_compaction_bytes_for_input ||
+         output_level_inputs_size + expanded_inputs_size < limit) &&
         !AreFilesInCompaction(expanded_inputs.files)) {
       InternalKey new_start, new_limit;
       GetRange(expanded_inputs, &new_start, &new_limit);
@@ -552,9 +549,10 @@ bool CompactionPicker::SetupOtherInputs(
       vstorage->GetCleanInputsWithinInterval(input_level, &all_start,
                                              &all_limit, &expanded_inputs.files,
                                              base_index, nullptr);
-      expanded_inputs_size = TotalCompensatedFileSize(expanded_inputs.files);
+      expanded_inputs_size = TotalFileSize(expanded_inputs.files);
       if (expanded_inputs.size() > inputs->size() &&
-          output_level_inputs_size + expanded_inputs_size < limit &&
+          (mutable_cf_options.ignore_max_compaction_bytes_for_input ||
+           output_level_inputs_size + expanded_inputs_size < limit) &&
           !AreFilesInCompaction(expanded_inputs.files)) {
         expand_inputs = true;
       }
@@ -655,7 +653,10 @@ Compaction* CompactionPicker::CompactRange(
 
     // 2 non-exclusive manual compactions could run at the same time producing
     // overlaping outputs in the same level.
-    if (FilesRangeOverlapWithCompaction(inputs, output_level)) {
+    if (FilesRangeOverlapWithCompaction(
+            inputs, output_level,
+            Compaction::EvaluatePenultimateLevel(vstorage, ioptions_,
+                                                 start_level, output_level))) {
       // This compaction output could potentially conflict with the output
       // of a currently running compaction, we cannot run it.
       *manual_conflict = true;
@@ -722,18 +723,18 @@ Compaction* CompactionPicker::CompactRange(
       }
       largest = &inputs[i]->largest;
 
-      uint64_t s = inputs[i]->compensated_file_size;
+      uint64_t input_file_size = inputs[i]->fd.GetFileSize();
       uint64_t output_level_total = 0;
       if (output_level < vstorage->num_non_empty_levels()) {
         std::vector<FileMetaData*> files;
         vstorage->GetOverlappingInputsRangeBinarySearch(
             output_level, smallest, largest, &files, hint_index, &hint_index);
         for (const auto& file : files) {
-          output_level_total += file->compensated_file_size;
+          output_level_total += file->fd.GetFileSize();
         }
       }
 
-      input_level_total += s;
+      input_level_total += input_file_size;
 
       if (input_level_total + output_level_total >= limit) {
         covering_the_whole_range = false;
@@ -834,7 +835,10 @@ Compaction* CompactionPicker::CompactRange(
 
   // 2 non-exclusive manual compactions could run at the same time producing
   // overlaping outputs in the same level.
-  if (FilesRangeOverlapWithCompaction(compaction_inputs, output_level)) {
+  if (FilesRangeOverlapWithCompaction(
+          compaction_inputs, output_level,
+          Compaction::EvaluatePenultimateLevel(vstorage, ioptions_, input_level,
+                                               output_level))) {
     // This compaction output could potentially conflict with the output
     // of a currently running compaction, we cannot run it.
     *manual_conflict = true;
@@ -878,21 +882,21 @@ namespace {
 // Test whether two files have overlapping key-ranges.
 bool HaveOverlappingKeyRanges(const Comparator* c, const SstFileMetaData& a,
                               const SstFileMetaData& b) {
-  if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
-    if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
+  if (c->CompareWithoutTimestamp(a.smallestkey, b.smallestkey) >= 0) {
+    if (c->CompareWithoutTimestamp(a.smallestkey, b.largestkey) <= 0) {
       // b.smallestkey <= a.smallestkey <= b.largestkey
       return true;
     }
-  } else if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
+  } else if (c->CompareWithoutTimestamp(a.largestkey, b.smallestkey) >= 0) {
     // a.smallestkey < b.smallestkey <= a.largestkey
     return true;
   }
-  if (c->Compare(a.largestkey, b.largestkey) <= 0) {
-    if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
+  if (c->CompareWithoutTimestamp(a.largestkey, b.largestkey) <= 0) {
+    if (c->CompareWithoutTimestamp(a.largestkey, b.smallestkey) >= 0) {
       // b.smallestkey <= a.largestkey <= b.largestkey
       return true;
     }
-  } else if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
+  } else if (c->CompareWithoutTimestamp(a.smallestkey, b.largestkey) <= 0) {
     // a.smallestkey <= b.largestkey < a.largestkey
     return true;
   }
@@ -931,15 +935,16 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
     // identify the first and the last compaction input files
     // in the current level.
     for (size_t f = 0; f < current_files.size(); ++f) {
-      if (input_files->find(TableFileNameToNumber(current_files[f].name)) !=
-          input_files->end()) {
-        first_included = std::min(first_included, static_cast<int>(f));
-        last_included = std::max(last_included, static_cast<int>(f));
-        if (is_first == false) {
-          smallestkey = current_files[f].smallestkey;
-          largestkey = current_files[f].largestkey;
-          is_first = true;
-        }
+      const uint64_t file_number = TableFileNameToNumber(current_files[f].name);
+      if (input_files->find(file_number) == input_files->end()) {
+        continue;
+      }
+      first_included = std::min(first_included, static_cast<int>(f));
+      last_included = std::max(last_included, static_cast<int>(f));
+      if (is_first == false) {
+        smallestkey = current_files[f].smallestkey;
+        largestkey = current_files[f].largestkey;
+        is_first = true;
       }
     }
     if (last_included == kNotFound) {
@@ -947,21 +952,22 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
     }
 
     if (l != 0) {
-      // expend the compaction input of the current level if it
+      // expand the compaction input of the current level if it
       // has overlapping key-range with other non-compaction input
       // files in the same level.
       while (first_included > 0) {
-        if (comparator->Compare(current_files[first_included - 1].largestkey,
-                                current_files[first_included].smallestkey) <
-            0) {
+        if (comparator->CompareWithoutTimestamp(
+                current_files[first_included - 1].largestkey,
+                current_files[first_included].smallestkey) < 0) {
           break;
         }
         first_included--;
       }
 
       while (last_included < static_cast<int>(current_files.size()) - 1) {
-        if (comparator->Compare(current_files[last_included + 1].smallestkey,
-                                current_files[last_included].largestkey) > 0) {
+        if (comparator->CompareWithoutTimestamp(
+                current_files[last_included + 1].smallestkey,
+                current_files[last_included].largestkey) > 0) {
           break;
         }
         last_included++;
@@ -977,27 +983,29 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
                                current_files[f].name +
                                " is currently being compacted.");
       }
+
       input_files->insert(TableFileNameToNumber(current_files[f].name));
     }
 
     // update smallest and largest key
     if (l == 0) {
       for (int f = first_included; f <= last_included; ++f) {
-        if (comparator->Compare(smallestkey, current_files[f].smallestkey) >
-            0) {
+        if (comparator->CompareWithoutTimestamp(
+                smallestkey, current_files[f].smallestkey) > 0) {
           smallestkey = current_files[f].smallestkey;
         }
-        if (comparator->Compare(largestkey, current_files[f].largestkey) < 0) {
+        if (comparator->CompareWithoutTimestamp(
+                largestkey, current_files[f].largestkey) < 0) {
           largestkey = current_files[f].largestkey;
         }
       }
     } else {
-      if (comparator->Compare(smallestkey,
-                              current_files[first_included].smallestkey) > 0) {
+      if (comparator->CompareWithoutTimestamp(
+              smallestkey, current_files[first_included].smallestkey) > 0) {
         smallestkey = current_files[first_included].smallestkey;
       }
-      if (comparator->Compare(largestkey,
-                              current_files[last_included].largestkey) < 0) {
+      if (comparator->CompareWithoutTimestamp(
+              largestkey, current_files[last_included].largestkey) < 0) {
         largestkey = current_files[last_included].largestkey;
       }
     }
@@ -1116,8 +1124,13 @@ void CompactionPicker::RegisterCompaction(Compaction* c) {
   }
   assert(ioptions_.compaction_style != kCompactionStyleLevel ||
          c->output_level() == 0 ||
-         !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level()));
-  if (c->start_level() == 0 ||
+         !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level(),
+                                          c->GetPenultimateLevel()));
+  // CompactionReason::kExternalSstIngestion's start level is just a placeholder
+  // number without actual meaning as file ingestion technically does not have
+  // an input level like other compactions
+  if ((c->start_level() == 0 &&
+       c->compaction_reason() != CompactionReason::kExternalSstIngestion) ||
       ioptions_.compaction_style == kCompactionStyleUniversal) {
     level0_compactions_in_progress_.insert(c);
   }

db/compaction/compaction_picker.h

@@ -51,14 +51,15 @@ class CompactionPicker {
   virtual ~CompactionPicker();
 
   // Pick level and inputs for a new compaction.
+  //
   // Returns nullptr if there is no compaction to be done.
   // Otherwise returns a pointer to a heap-allocated object that
   // describes the compaction.  Caller should delete the result.
-  virtual Compaction* PickCompaction(
-      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-      const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
-      LogBuffer* log_buffer,
-      SequenceNumber earliest_memtable_seqno = kMaxSequenceNumber) = 0;
+  virtual Compaction* PickCompaction(const std::string& cf_name,
+                                     const MutableCFOptions& mutable_cf_options,
+                                     const MutableDBOptions& mutable_db_options,
+                                     VersionStorageInfo* vstorage,
+                                     LogBuffer* log_buffer) = 0;
 
   // Return a compaction object for compacting the range [begin,end] in
   // the specified level.  Returns nullptr if there is nothing in that
@@ -91,6 +92,7 @@ class CompactionPicker {
 // files.  If it's not possible to conver an invalid input_files
 // into a valid one by adding more files, the function will return a
 // non-ok status with specific reason.
+//
 #ifndef ROCKSDB_LITE
   Status SanitizeCompactionInputFiles(std::unordered_set<uint64_t>* input_files,
                                       const ColumnFamilyMetaData& cf_meta,
@@ -182,7 +184,8 @@ class CompactionPicker {
   // Returns true if the key range that `inputs` files cover overlap with the
   // key range of a currently running compaction.
   bool FilesRangeOverlapWithCompaction(
-      const std::vector<CompactionInputFiles>& inputs, int level) const;
+      const std::vector<CompactionInputFiles>& inputs, int level,
+      int penultimate_level) const;
 
   bool SetupOtherInputs(const std::string& cf_name,
                         const MutableCFOptions& mutable_cf_options,
@@ -254,12 +257,11 @@ class NullCompactionPicker : public CompactionPicker {
   virtual ~NullCompactionPicker() {}
 
   // Always return "nullptr"
-  Compaction* PickCompaction(
-      const std::string& /*cf_name*/,
-      const MutableCFOptions& /*mutable_cf_options*/,
-      const MutableDBOptions& /*mutable_db_options*/,
-      VersionStorageInfo* /*vstorage*/, LogBuffer* /* log_buffer */,
-      SequenceNumber /* earliest_memtable_seqno */) override {
+  Compaction* PickCompaction(const std::string& /*cf_name*/,
+                             const MutableCFOptions& /*mutable_cf_options*/,
+                             const MutableDBOptions& /*mutable_db_options*/,
+                             VersionStorageInfo* /*vstorage*/,
+                             LogBuffer* /* log_buffer */) override {
     return nullptr;
   }
 
@@ -303,11 +305,11 @@ class NullCompactionPicker : public CompactionPicker {
 //                                        files. Cannot be nullptr.
 //
 // @return                                true iff compaction was found.
-bool FindIntraL0Compaction(
-    const std::vector<FileMetaData*>& level_files, size_t min_files_to_compact,
-    uint64_t max_compact_bytes_per_del_file, uint64_t max_compaction_bytes,
-    CompactionInputFiles* comp_inputs,
-    SequenceNumber earliest_mem_seqno = kMaxSequenceNumber);
+bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
+                           size_t min_files_to_compact,
+                           uint64_t max_compact_bytes_per_del_file,
+                           uint64_t max_compaction_bytes,
+                           CompactionInputFiles* comp_inputs);
 
 CompressionType GetCompressionType(const VersionStorageInfo* vstorage,
                                    const MutableCFOptions& mutable_cf_options,

db/compaction/compaction_picker_fifo.cc

@@ -121,20 +121,45 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(
   return c;
 }
 
+// The size-based compaction picker for FIFO.
+//
+// When the entire column family size exceeds max_table_files_size, FIFO will
+// try to delete the oldest sst file(s) until the resulting column family size
+// is smaller than max_table_files_size.
+//
+// This function also takes care the case where a DB is migrating from level /
+// universal compaction to FIFO compaction.  During the migration, the column
+// family will also have non-L0 files while FIFO can only create L0 files.
+// In this case, this function will first purge the sst files in the bottom-
+// most non-empty level first, and the DB will eventually converge to the
+// regular FIFO case where there're only L0 files.  Note that during the
+// migration case, the purge order will only be an approximation of "FIFO"
+// as entries inside lower-level files might sometimes be newer than some
+// entries inside upper-level files.
 Compaction* FIFOCompactionPicker::PickSizeCompaction(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
     LogBuffer* log_buffer) {
-  const int kLevel0 = 0;
-  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
-  uint64_t total_size = GetTotalFilesSize(level_files);
+  // compute the total size and identify the last non-empty level
+  int last_level = 0;
+  uint64_t total_size = 0;
+  for (int level = 0; level < vstorage->num_levels(); ++level) {
+    auto level_size = GetTotalFilesSize(vstorage->LevelFiles(level));
+    total_size += level_size;
+    if (level_size > 0) {
+      last_level = level;
+    }
+  }
+  const std::vector<FileMetaData*>& last_level_files =
+      vstorage->LevelFiles(last_level);
 
-  if (total_size <=
-          mutable_cf_options.compaction_options_fifo.max_table_files_size ||
-      level_files.size() == 0) {
-    // total size not exceeded
+  if (last_level == 0 &&
+      total_size <=
+          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+    // total size not exceeded, try to find intra level 0 compaction if enabled
+    const std::vector<FileMetaData*>& level0_files = vstorage->LevelFiles(0);
     if (mutable_cf_options.compaction_options_fifo.allow_compaction &&
-        level_files.size() > 0) {
+        level0_files.size() > 0) {
       CompactionInputFiles comp_inputs;
       // try to prevent same files from being compacted multiple times, which
       // could produce large files that may never TTL-expire. Achieve this by
@@ -146,7 +171,7 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
               static_cast<uint64_t>(mutable_cf_options.write_buffer_size),
               1.1));
       if (FindIntraL0Compaction(
-              level_files,
+              level0_files,
               mutable_cf_options
                   .level0_file_num_compaction_trigger /* min_files_to_compact */
               ,
@@ -187,27 +212,71 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
 
   std::vector<CompactionInputFiles> inputs;
   inputs.emplace_back();
-  inputs[0].level = 0;
+  inputs[0].level = last_level;
 
-  for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
-    auto f = *ritr;
-    total_size -= f->fd.file_size;
-    inputs[0].files.push_back(f);
-    char tmp_fsize[16];
-    AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
-    ROCKS_LOG_BUFFER(log_buffer,
-                     "[%s] FIFO compaction: picking file %" PRIu64
-                     " with size %s for deletion",
-                     cf_name.c_str(), f->fd.GetNumber(), tmp_fsize);
-    if (total_size <=
-        mutable_cf_options.compaction_options_fifo.max_table_files_size) {
-      break;
+  if (last_level == 0) {
+    // In L0, right-most files are the oldest files.
+    for (auto ritr = last_level_files.rbegin(); ritr != last_level_files.rend();
+         ++ritr) {
+      auto f = *ritr;
+      total_size -= f->fd.file_size;
+      inputs[0].files.push_back(f);
+      char tmp_fsize[16];
+      AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
+      ROCKS_LOG_BUFFER(log_buffer,
+                       "[%s] FIFO compaction: picking file %" PRIu64
+                       " with size %s for deletion",
+                       cf_name.c_str(), f->fd.GetNumber(), tmp_fsize);
+      if (total_size <=
+          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+        break;
+      }
     }
+  } else if (total_size >
+             mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+    // If the last level is non-L0, we actually don't know which file is
+    // logically the oldest since the file creation time only represents
+    // when this file was compacted to this level, which is independent
+    // to when the entries in this file were first inserted.
+    //
+    // As a result, we delete files from the left instead.  This means the sst
+    // file with the smallest key will be deleted first.  This design decision
+    // better serves a major type of FIFO use cases where smaller keys are
+    // associated with older data.
+    for (const auto& f : last_level_files) {
+      total_size -= f->fd.file_size;
+      inputs[0].files.push_back(f);
+      char tmp_fsize[16];
+      AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
+      ROCKS_LOG_BUFFER(
+          log_buffer,
+          "[%s] FIFO compaction: picking file %" PRIu64
+          " with size %s for deletion under total size %" PRIu64
+          " vs max table files size %" PRIu64,
+          cf_name.c_str(), f->fd.GetNumber(), tmp_fsize, total_size,
+          mutable_cf_options.compaction_options_fifo.max_table_files_size);
+
+      if (total_size <=
+          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+        break;
+      }
+    }
+  } else {
+    ROCKS_LOG_BUFFER(
+        log_buffer,
+        "[%s] FIFO compaction: nothing to do. Total size %" PRIu64
+        ", max size %" PRIu64 "\n",
+        cf_name.c_str(), total_size,
+        mutable_cf_options.compaction_options_fifo.max_table_files_size);
+    return nullptr;
   }
 
   Compaction* c = new Compaction(
       vstorage, ioptions_, mutable_cf_options, mutable_db_options,
-      std::move(inputs), 0, 0, 0, 0, kNoCompression,
+      std::move(inputs), last_level,
+      /* target_file_size */ 0,
+      /* max_compaction_bytes */ 0,
+      /* output_path_id */ 0, kNoCompression,
       mutable_cf_options.compression_opts, Temperature::kUnknown,
       /* max_subcompactions */ 0, {}, /* is manual */ false,
       /* trim_ts */ "", vstorage->CompactionScore(0),
@@ -224,6 +293,13 @@ Compaction* FIFOCompactionPicker::PickCompactionToWarm(
     return nullptr;
   }
 
+  // PickCompactionToWarm is only triggered if there is no non-L0 files.
+  for (int level = 1; level < vstorage->num_levels(); ++level) {
+    if (GetTotalFilesSize(vstorage->LevelFiles(level)) > 0) {
+      return nullptr;
+    }
+  }
+
   const int kLevel0 = 0;
   const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
 
@@ -326,9 +402,7 @@ Compaction* FIFOCompactionPicker::PickCompactionToWarm(
 Compaction* FIFOCompactionPicker::PickCompaction(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
-    LogBuffer* log_buffer, SequenceNumber /*earliest_memtable_seqno*/) {
-  assert(vstorage->num_levels() == 1);
-
+    LogBuffer* log_buffer) {
   Compaction* c = nullptr;
   if (mutable_cf_options.ttl > 0) {
     c = PickTTLCompaction(cf_name, mutable_cf_options, mutable_db_options,

db/compaction/compaction_picker_fifo.h

@@ -19,11 +19,11 @@ class FIFOCompactionPicker : public CompactionPicker {
                        const InternalKeyComparator* icmp)
       : CompactionPicker(ioptions, icmp) {}
 
-  virtual Compaction* PickCompaction(
-      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-      const MutableDBOptions& mutable_db_options, VersionStorageInfo* version,
-      LogBuffer* log_buffer,
-      SequenceNumber earliest_memtable_seqno = kMaxSequenceNumber) override;
+  virtual Compaction* PickCompaction(const std::string& cf_name,
+                                     const MutableCFOptions& mutable_cf_options,
+                                     const MutableDBOptions& mutable_db_options,
+                                     VersionStorageInfo* version,
+                                     LogBuffer* log_buffer) override;
 
   virtual Compaction* CompactRange(
       const std::string& cf_name, const MutableCFOptions& mutable_cf_options,

db/compaction/compaction_picker_level.cc

@@ -50,7 +50,6 @@ class LevelCompactionBuilder {
  public:
   LevelCompactionBuilder(const std::string& cf_name,
                          VersionStorageInfo* vstorage,
-                         SequenceNumber earliest_mem_seqno,
                          CompactionPicker* compaction_picker,
                          LogBuffer* log_buffer,
                          const MutableCFOptions& mutable_cf_options,
@@ -58,7 +57,6 @@ class LevelCompactionBuilder {
                          const MutableDBOptions& mutable_db_options)
       : cf_name_(cf_name),
         vstorage_(vstorage),
-        earliest_mem_seqno_(earliest_mem_seqno),
         compaction_picker_(compaction_picker),
         log_buffer_(log_buffer),
         mutable_cf_options_(mutable_cf_options),
@@ -122,7 +120,6 @@ class LevelCompactionBuilder {
 
   const std::string& cf_name_;
   VersionStorageInfo* vstorage_;
-  SequenceNumber earliest_mem_seqno_;
   CompactionPicker* compaction_picker_;
   LogBuffer* log_buffer_;
   int start_level_ = -1;
@@ -196,7 +193,10 @@ void LevelCompactionBuilder::SetupInitialFiles() {
       }
       output_level_ =
           (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1;
-      if (PickFileToCompact()) {
+      bool picked_file_to_compact = PickFileToCompact();
+      TEST_SYNC_POINT_CALLBACK("PostPickFileToCompact",
+                               &picked_file_to_compact);
+      if (picked_file_to_compact) {
         // found the compaction!
         if (start_level_ == 0) {
           // L0 score = `num L0 files` / `level0_file_num_compaction_trigger`
@@ -254,6 +254,25 @@ void LevelCompactionBuilder::SetupInitialFiles() {
   }
 
   // TTL Compaction
+  if (ioptions_.compaction_pri == kRoundRobin &&
+      !vstorage_->ExpiredTtlFiles().empty()) {
+    auto expired_files = vstorage_->ExpiredTtlFiles();
+    // the expired files list should already be sorted by level
+    start_level_ = expired_files.front().first;
+#ifndef NDEBUG
+    for (const auto& file : expired_files) {
+      assert(start_level_ <= file.first);
+    }
+#endif
+    if (start_level_ > 0) {
+      output_level_ = start_level_ + 1;
+      if (PickFileToCompact()) {
+        compaction_reason_ = CompactionReason::kRoundRobinTtl;
+        return;
+      }
+    }
+  }
+
   PickFileToCompact(vstorage_->ExpiredTtlFiles(), true);
   if (!start_level_inputs_.empty()) {
     compaction_reason_ = CompactionReason::kTtl;
@@ -318,7 +337,7 @@ void LevelCompactionBuilder::SetupOtherFilesWithRoundRobinExpansion() {
   // Constraint 3 (pre-calculate the ideal max bytes to compact)
   for (auto f : level_files) {
     if (!f->being_compacted) {
-      start_lvl_bytes_no_compacting += f->compensated_file_size;
+      start_lvl_bytes_no_compacting += f->fd.GetFileSize();
     }
   }
   if (start_lvl_bytes_no_compacting >
@@ -341,7 +360,7 @@ void LevelCompactionBuilder::SetupOtherFilesWithRoundRobinExpansion() {
     }
   }
   // Constraint 3
-  if (start_level_inputs_[0]->compensated_file_size >=
+  if (start_level_inputs_[0]->fd.GetFileSize() >=
       start_lvl_max_bytes_to_compact) {
     return;
   }
@@ -360,7 +379,9 @@ void LevelCompactionBuilder::SetupOtherFilesWithRoundRobinExpansion() {
     if (!compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
                                                     &tmp_start_level_inputs) ||
         compaction_picker_->FilesRangeOverlapWithCompaction(
-            {tmp_start_level_inputs}, output_level_)) {
+            {tmp_start_level_inputs}, output_level_,
+            Compaction::EvaluatePenultimateLevel(
+                vstorage_, ioptions_, start_level_, output_level_))) {
       // Constraint 1a
       tmp_start_level_inputs.clear();
       return;
@@ -368,7 +389,7 @@ void LevelCompactionBuilder::SetupOtherFilesWithRoundRobinExpansion() {
 
     curr_bytes_to_compact = 0;
     for (auto start_lvl_f : tmp_start_level_inputs.files) {
-      curr_bytes_to_compact += start_lvl_f->compensated_file_size;
+      curr_bytes_to_compact += start_lvl_f->fd.GetFileSize();
     }
 
     // Check whether any output level files are locked
@@ -385,7 +406,7 @@ void LevelCompactionBuilder::SetupOtherFilesWithRoundRobinExpansion() {
 
     uint64_t start_lvl_curr_bytes_to_compact = curr_bytes_to_compact;
     for (auto output_lvl_f : output_level_inputs.files) {
-      curr_bytes_to_compact += output_lvl_f->compensated_file_size;
+      curr_bytes_to_compact += output_lvl_f->fd.GetFileSize();
     }
     if (curr_bytes_to_compact > mutable_cf_options_.max_compaction_bytes) {
       // Constraint 2
@@ -426,19 +447,21 @@ bool LevelCompactionBuilder::SetupOtherInputsIfNeeded() {
       compaction_inputs_.push_back(output_level_inputs_);
     }
 
+    // In some edge cases we could pick a compaction that will be compacting
+    // a key range that overlap with another running compaction, and both
+    // of them have the same output level. This could happen if
+    // (1) we are running a non-exclusive manual compaction
+    // (2) AddFile ingest a new file into the LSM tree
+    // We need to disallow this from happening.
+    if (compaction_picker_->FilesRangeOverlapWithCompaction(
+            compaction_inputs_, output_level_,
+            Compaction::EvaluatePenultimateLevel(
+                vstorage_, ioptions_, start_level_, output_level_))) {
+      // This compaction output could potentially conflict with the output
+      // of a currently running compaction, we cannot run it.
+      return false;
+    }
     if (!is_l0_trivial_move_) {
-      // In some edge cases we could pick a compaction that will be compacting
-      // a key range that overlap with another running compaction, and both
-      // of them have the same output level. This could happen if
-      // (1) we are running a non-exclusive manual compaction
-      // (2) AddFile ingest a new file into the LSM tree
-      // We need to disallow this from happening.
-      if (compaction_picker_->FilesRangeOverlapWithCompaction(
-              compaction_inputs_, output_level_)) {
-        // This compaction output could potentially conflict with the output
-        // of a currently running compaction, we cannot run it.
-        return false;
-      }
       compaction_picker_->GetGrandparents(vstorage_, start_level_inputs_,
                                           output_level_inputs_, &grandparents_);
     }
@@ -662,9 +685,14 @@ bool LevelCompactionBuilder::TryExtendNonL0TrivialMove(int start_index) {
         break;
       }
       if (i < static_cast<int>(level_files.size()) - 1 &&
-          compaction_picker_->icmp()->user_comparator()->Compare(
-              next_file->largest.user_key(),
-              level_files[i + 1]->smallest.user_key()) == 0) {
+          compaction_picker_->icmp()
+                  ->user_comparator()
+                  ->CompareWithoutTimestamp(
+                      next_file->largest.user_key(),
+                      level_files[i + 1]->smallest.user_key()) == 0) {
+        TEST_SYNC_POINT_CALLBACK(
+            "LevelCompactionBuilder::TryExtendNonL0TrivialMove:NoCleanCut",
+            nullptr);
         // Not a clean up after adding the next file. Skip.
         break;
       }
@@ -731,7 +759,9 @@ bool LevelCompactionBuilder::PickFileToCompact() {
     if (!compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
                                                     &start_level_inputs_) ||
         compaction_picker_->FilesRangeOverlapWithCompaction(
-            {start_level_inputs_}, output_level_)) {
+            {start_level_inputs_}, output_level_,
+            Compaction::EvaluatePenultimateLevel(
+                vstorage_, ioptions_, start_level_, output_level_))) {
       // A locked (pending compaction) input-level file was pulled in due to
       // user-key overlap.
       start_level_inputs_.clear();
@@ -795,16 +825,16 @@ bool LevelCompactionBuilder::PickIntraL0Compaction() {
   return FindIntraL0Compaction(level_files, kMinFilesForIntraL0Compaction,
                                std::numeric_limits<uint64_t>::max(),
                                mutable_cf_options_.max_compaction_bytes,
-                               &start_level_inputs_, earliest_mem_seqno_);
+                               &start_level_inputs_);
 }
 }  // namespace
 
 Compaction* LevelCompactionPicker::PickCompaction(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
-    LogBuffer* log_buffer, SequenceNumber earliest_mem_seqno) {
-  LevelCompactionBuilder builder(cf_name, vstorage, earliest_mem_seqno, this,
-                                 log_buffer, mutable_cf_options, ioptions_,
+    LogBuffer* log_buffer) {
+  LevelCompactionBuilder builder(cf_name, vstorage, this, log_buffer,
+                                 mutable_cf_options, ioptions_,
                                  mutable_db_options);
   return builder.PickCompaction();
 }

db/compaction/compaction_picker_level.h

@@ -20,11 +20,11 @@ class LevelCompactionPicker : public CompactionPicker {
   LevelCompactionPicker(const ImmutableOptions& ioptions,
                         const InternalKeyComparator* icmp)
       : CompactionPicker(ioptions, icmp) {}
-  virtual Compaction* PickCompaction(
-      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-      const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
-      LogBuffer* log_buffer,
-      SequenceNumber earliest_memtable_seqno = kMaxSequenceNumber) override;
+  virtual Compaction* PickCompaction(const std::string& cf_name,
+                                     const MutableCFOptions& mutable_cf_options,
+                                     const MutableDBOptions& mutable_db_options,
+                                     VersionStorageInfo* vstorage,
+                                     LogBuffer* log_buffer) override;
 
   virtual bool NeedsCompaction(
       const VersionStorageInfo* vstorage) const override;