Fix bug in WAL streaming uncompression (#11198)

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 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
@@ -120,7 +152,7 @@ commands:
     steps:
       - run:
           name: "Test low-variance benchmarks"
-          command: ./tools/benchmark_ci.py --db_dir /tmp/rocksdb-benchmark-datadir --output_dir /tmp/benchmark-results --num_keys 50000
+          command: ./tools/benchmark_ci.py --db_dir /tmp/rocksdb-benchmark-datadir --output_dir /tmp/benchmark-results --num_keys 10000000
           environment:
             LD_LIBRARY_PATH: /usr/local/lib
             # How long to run parts of the test(s)
@@ -154,11 +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
@@ -239,7 +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:
@@ -518,73 +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
@@ -859,15 +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

@@ -250,6 +250,8 @@ 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)
@@ -292,8 +294,7 @@ 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>
@@ -305,12 +306,11 @@ 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
@@ -653,8 +653,8 @@ set(SOURCES
         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
@@ -1065,28 +1065,34 @@ set(ROCKSDB_PLUGIN_BUILTINS "")
 message(STATUS "ROCKSDB PLUGINS TO BUILD ${ROCKSDB_PLUGINS}")
 foreach(PLUGIN IN LISTS PLUGINS)
   set(PLUGIN_ROOT "${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()
 
@@ -1566,9 +1572,9 @@ 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${ARTIFACT_SUFFIX}
+  target_link_libraries(block_cache_trace_analyzer${ARTIFACT_SUFFIX}
     ${ROCKSDB_LIB} ${GFLAGS_LIB} ${FOLLY_LIBS})
 
   add_executable(trace_analyzer${ARTIFACT_SUFFIX}

HISTORY.md

@@ -1,6 +1,89 @@
 # Rocksdb Change Log
 ## Unreleased
 ### 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.
+
+## 7.9.3 (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
+
+## 7.9.2 (12/21/2022)
+### Bug Fixes
+* 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.
+
+## 7.9.1 (12/8/2022)
+### 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
+
+### Behavior changes
+* Make best-efforts recovery verify SST unique ID before Version construction (#10962)
+
+## 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`.

Makefile

@@ -509,7 +509,7 @@ ifneq ($(strip $(FOLLY_PATH)),)
 
 	# 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 += $(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 $(FOLLY_PATH)/lib/libfolly.a $(FMT_LIB_PATH)/libfmt.a $(GLOG_LIB_PATH)/libglog.so $(GFLAGS_PATH)/lib/libgflags.so.2.2 -ldl
+	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
@@ -658,7 +658,7 @@ 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) $(DEV_HEADER_DIRS) -type f -name '*.h' | grep -E -v 'port/|plugin/|lua/|range_tree/')
 else
 	DEV_HEADERS :=
 endif
@@ -795,9 +795,9 @@ TOOLS_LIBRARY=$(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
@@ -1832,7 +1832,7 @@ 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) $(LIBRARY)
@@ -2092,8 +2092,8 @@ ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(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

TARGETS

@@ -16,8 +16,8 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "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",
@@ -355,8 +355,8 @@ cpp_library_wrapper(name="rocksdb_whole_archive_lib", srcs=[
         "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",

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/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/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_bench_tool.cc

@@ -13,7 +13,6 @@
 #include <set>
 #include <sstream>
 
-#include "cache/fast_lru_cache.h"
 #include "db/db_impl/db_impl.h"
 #include "monitoring/histogram.h"
 #include "port/port.h"
@@ -297,10 +296,6 @@ class CacheBench {
       cache_ = HyperClockCacheOptions(FLAGS_cache_size, FLAGS_value_bytes,
                                       FLAGS_num_shard_bits)
                    .MakeSharedCache();
-    } 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);
     } else if (FLAGS_cache_type == "lru_cache") {
       LRUCacheOptions opts(FLAGS_cache_size, FLAGS_num_shard_bits,
                            false /* strict_capacity_limit */,

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

@@ -15,15 +15,14 @@
 #include <string>
 #include <vector>
 
-#include "cache/fast_lru_cache.h"
 #include "cache/lru_cache.h"
 #include "port/stack_trace.h"
 #include "test_util/testharness.h"
 #include "util/coding.h"
 #include "util/string_util.h"
 
-// FastLRUCache and HyperClockCache 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.
 
@@ -76,7 +75,6 @@ void EraseDeleter2(const Slice& /*key*/, void* value) {
 
 const std::string kLRU = "lru";
 const std::string kHyperClock = "hyper_clock";
-const std::string kFast = "fast";
 
 }  // anonymous namespace
 
@@ -86,7 +84,7 @@ class CacheTest : public testing::TestWithParam<std::string> {
   static std::string type_;
 
   static void Deleter(const Slice& key, void* v) {
-    if (type_ == kFast || type_ == kHyperClock) {
+    if (type_ == kHyperClock) {
       current_->deleted_keys_.push_back(DecodeKey16Bytes(key));
     } else {
       current_->deleted_keys_.push_back(DecodeKey32Bits(key));
@@ -126,11 +124,6 @@ class CacheTest : public testing::TestWithParam<std::string> {
                  capacity, estimated_value_size_ /*estimated_value_size*/)
           .MakeSharedCache();
     }
-    if (type == kFast) {
-      return NewFastLRUCache(
-          capacity, estimated_value_size_, -1 /*num_shard_bits*/,
-          false /*strict_capacity_limit*/, kDefaultCacheMetadataChargePolicy);
-    }
     return nullptr;
   }
 
@@ -153,11 +146,6 @@ class CacheTest : public testing::TestWithParam<std::string> {
                                     nullptr /*allocator*/, charge_policy)
           .MakeSharedCache();
     }
-    if (type == kFast) {
-      return NewFastLRUCache(capacity, 1 /*estimated_value_size*/,
-                             num_shard_bits, strict_capacity_limit,
-                             charge_policy);
-    }
     return nullptr;
   }
 
@@ -167,7 +155,7 @@ class CacheTest : public testing::TestWithParam<std::string> {
   // LRUCache doesn't, so the encoding depends on the cache type.
   std::string EncodeKey(int k) {
     auto type = GetParam();
-    if (type == kFast || type == kHyperClock) {
+    if (type == kHyperClock) {
       return EncodeKey16Bytes(k);
     } else {
       return EncodeKey32Bits(k);
@@ -176,7 +164,7 @@ class CacheTest : public testing::TestWithParam<std::string> {
 
   int DecodeKey(const Slice& k) {
     auto type = GetParam();
-    if (type == kFast || type == kHyperClock) {
+    if (type == kHyperClock) {
       return DecodeKey16Bytes(k);
     } else {
       return DecodeKey32Bits(k);
@@ -202,29 +190,21 @@ class CacheTest : public testing::TestWithParam<std::string> {
     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_;
@@ -418,13 +398,13 @@ 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);
   if (GetParam() == kHyperClock) {
@@ -434,7 +414,7 @@ TEST_P(CacheTest, HitAndMiss) {
     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]);
@@ -463,14 +443,14 @@ TEST_P(CacheTest, Erase) {
   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(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());
 }
@@ -515,7 +495,7 @@ TEST_P(CacheTest, EvictionPolicy) {
   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));
@@ -686,7 +666,7 @@ 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) {
@@ -704,7 +684,7 @@ class Value {
 
 namespace {
 void deleter(const Slice& /*key*/, void* value) {
-  delete static_cast<Value *>(value);
+  delete static_cast<Value*>(value);
 }
 }  // namespace
 
@@ -741,7 +721,7 @@ TEST_P(CacheTest, ReleaseWithoutErase) {
 
 TEST_P(CacheTest, SetCapacity) {
   auto type = GetParam();
-  if (type == kFast || type == kHyperClock) {
+  if (type == kHyperClock) {
     ROCKSDB_GTEST_BYPASS(
         "FastLRUCache and HyperClockCache don't support arbitrary capacity "
         "adjustments.");
@@ -795,14 +775,6 @@ TEST_P(CacheTest, SetCapacity) {
 }
 
 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);
@@ -859,7 +831,7 @@ TEST_P(CacheTest, OverCapacity) {
   // 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);
+  std::vector<Cache::Handle*> handles(n + 1);
 
   // Insert n+1 entries, but not releasing.
   for (int i = 0; i < static_cast<int>(n + 1); i++) {
@@ -919,7 +891,7 @@ void legacy_callback(void* value, size_t charge) {
   legacy_callback_state.push_back(
       {DecodeValue(value), static_cast<int>(charge)});
 }
-};
+};  // namespace
 
 TEST_P(CacheTest, ApplyToAllCacheEntriesTest) {
   std::vector<std::pair<int, int>> inserted;
@@ -1023,21 +995,21 @@ TEST_P(CacheTest, DefaultShardBits) {
       (GetParam() == kHyperClock ? 32U * 1024U : 512U) * 1024U;
 
   std::shared_ptr<Cache> cache = NewCache(32U * min_shard_size);
-  ShardedCache* sc = dynamic_cast<ShardedCache*>(cache.get());
+  ShardedCacheBase* sc = dynamic_cast<ShardedCacheBase*>(cache.get());
   ASSERT_EQ(5, sc->GetNumShardBits());
 
   cache = NewCache(min_shard_size / 1000U * 999U);
-  sc = dynamic_cast<ShardedCache*>(cache.get());
+  sc = dynamic_cast<ShardedCacheBase*>(cache.get());
   ASSERT_EQ(0, sc->GetNumShardBits());
 
   cache = NewCache(3U * 1024U * 1024U * 1024U);
-  sc = dynamic_cast<ShardedCache*>(cache.get());
+  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<ShardedCache*>(cache.get());
+    sc = dynamic_cast<ShardedCacheBase*>(cache.get());
     // current maximum of 6
     ASSERT_EQ(6, sc->GetNumShardBits());
   }
@@ -1053,9 +1025,8 @@ TEST_P(CacheTest, GetChargeAndDeleter) {
 }
 
 INSTANTIATE_TEST_CASE_P(CacheTestInstance, CacheTest,
-                        testing::Values(kLRU, kHyperClock, 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/clock_cache.h

@@ -27,12 +27,13 @@
 
 namespace ROCKSDB_NAMESPACE {
 
-namespace hyper_clock_cache {
+namespace clock_cache {
 
 // Forward declaration of friend class.
 class ClockCacheTest;
 
-// HyperClockCache is an experimental alternative to LRUCache.
+// HyperClockCache is an alternative to LRUCache specifically tailored for
+// use as BlockBasedTableOptions::block_cache
 //
 // Benefits
 // --------
@@ -303,30 +304,30 @@ constexpr double kLoadFactor = 0.7;
 // strict upper bound on the load factor.
 constexpr double kStrictLoadFactor = 0.84;
 
-using CacheKeyBytes = std::array<char, kCacheKeySize>;
-
 struct ClockHandleBasicData {
   void* value = nullptr;
   Cache::DeleterFn deleter = nullptr;
-  CacheKeyBytes key = {};
+  // A lossless, reversible hash of the fixed-size (16 byte) cache key. This
+  // eliminates the need to store a hash separately.
+  UniqueId64x2 hashed_key = kNullUniqueId64x2;
   size_t total_charge = 0;
 
-  Slice KeySlice() const { return Slice(key.data(), kCacheKeySize); }
+  // For total_charge_and_flags
+  // "Detached" means the handle is allocated separately from hash table.
+  static constexpr uint64_t kFlagDetached = uint64_t{1} << 63;
+  // Extract just the total charge
+  static constexpr uint64_t kTotalChargeMask = kFlagDetached - 1;
 
-  void FreeData() const {
-    if (deleter) {
-      (*deleter)(KeySlice(), value);
-    }
-  }
+  inline size_t GetTotalCharge() const { return total_charge; }
+
+  // Calls deleter (if non-null) on cache key and value
+  void FreeData() const;
+
+  // Required by concept HandleImpl
+  const UniqueId64x2& GetHash() const { return hashed_key; }
 };
 
-struct ClockHandleMoreData : public ClockHandleBasicData {
-  uint32_t hash = 0;
-};
-
-// Target size to be exactly a common cache line size (see static_assert in
-// clock_cache.cc)
-struct ALIGN_AS(64U) ClockHandle : public ClockHandleMoreData {
+struct ClockHandle : public ClockHandleBasicData {
   // Constants for handling the atomic `meta` word, which tracks most of the
   // state of the handle. The meta word looks like this:
   // low bits                                                     high bits
@@ -378,47 +379,69 @@ struct ALIGN_AS(64U) ClockHandle : public ClockHandleMoreData {
 
   // See above
   std::atomic<uint64_t> meta{};
-  // The number of elements that hash to this slot or a lower one, but wind
-  // up in this slot or a higher one.
-  std::atomic<uint32_t> displacements{};
 
-  // True iff the handle is allocated separately from hash table.
-  bool detached = false;
+  // Anticipating use for SecondaryCache support
+  void* reserved_for_future_use = nullptr;
 };  // struct ClockHandle
 
-class ClockHandleTable {
+class HyperClockTable {
  public:
-  explicit ClockHandleTable(int hash_bits, bool initial_charge_metadata);
-  ~ClockHandleTable();
+  // Target size to be exactly a common cache line size (see static_assert in
+  // clock_cache.cc)
+  struct ALIGN_AS(64U) HandleImpl : public ClockHandle {
+    // The number of elements that hash to this slot or a lower one, but wind
+    // up in this slot or a higher one.
+    std::atomic<uint32_t> displacements{};
 
-  Status Insert(const ClockHandleMoreData& proto, ClockHandle** handle,
+    // Whether this is a "deteched" handle that is independently allocated
+    // with `new` (so must be deleted with `delete`).
+    // TODO: ideally this would be packed into some other data field, such
+    // as upper bits of total_charge, but that incurs a measurable performance
+    // regression.
+    bool detached = false;
+
+    inline bool IsDetached() const { return detached; }
+
+    inline void SetDetached() { detached = true; }
+  };  // struct HandleImpl
+
+  struct Opts {
+    size_t estimated_value_size;
+  };
+
+  HyperClockTable(size_t capacity, bool strict_capacity_limit,
+                  CacheMetadataChargePolicy metadata_charge_policy,
+                  const Opts& opts);
+  ~HyperClockTable();
+
+  Status Insert(const ClockHandleBasicData& proto, HandleImpl** handle,
                 Cache::Priority priority, size_t capacity,
                 bool strict_capacity_limit);
 
-  ClockHandle* Lookup(const CacheKeyBytes& key, uint32_t hash);
+  HandleImpl* Lookup(const UniqueId64x2& hashed_key);
 
-  bool Release(ClockHandle* handle, bool useful, bool erase_if_last_ref);
+  bool Release(HandleImpl* handle, bool useful, bool erase_if_last_ref);
 
-  void Ref(ClockHandle& handle);
+  void Ref(HandleImpl& handle);
 
-  void Erase(const CacheKeyBytes& key, uint32_t hash);
+  void Erase(const UniqueId64x2& hashed_key);
 
-  void ConstApplyToEntriesRange(std::function<void(const ClockHandle&)> func,
-                                uint32_t index_begin, uint32_t index_end,
+  void ConstApplyToEntriesRange(std::function<void(const HandleImpl&)> func,
+                                size_t index_begin, size_t index_end,
                                 bool apply_if_will_be_deleted) const;
 
   void EraseUnRefEntries();
 
-  uint32_t GetTableSize() const { return uint32_t{1} << length_bits_; }
+  size_t GetTableSize() const { return size_t{1} << length_bits_; }
 
   int GetLengthBits() const { return length_bits_; }
 
-  uint32_t GetOccupancyLimit() const { return occupancy_limit_; }
-
-  uint32_t GetOccupancy() const {
+  size_t GetOccupancy() const {
     return occupancy_.load(std::memory_order_relaxed);
   }
 
+  size_t GetOccupancyLimit() const { return occupancy_limit_; }
+
   size_t GetUsage() const { return usage_.load(std::memory_order_relaxed); }
 
   size_t GetDetachedUsage() const {
@@ -426,18 +449,20 @@ class ClockHandleTable {
   }
 
   // Acquire/release N references
-  void TEST_RefN(ClockHandle& handle, size_t n);
-  void TEST_ReleaseN(ClockHandle* handle, size_t n);
+  void TEST_RefN(HandleImpl& handle, size_t n);
+  void TEST_ReleaseN(HandleImpl* handle, size_t n);
 
  private:  // functions
   // Returns x mod 2^{length_bits_}.
-  uint32_t ModTableSize(uint32_t x) { return x & length_bits_mask_; }
+  inline size_t ModTableSize(uint64_t x) {
+    return static_cast<size_t>(x) & length_bits_mask_;
+  }
 
   // Runs the clock eviction algorithm trying to reclaim at least
   // requested_charge. Returns how much is evicted, which could be less
   // if it appears impossible to evict the requested amount without blocking.
-  void Evict(size_t requested_charge, size_t* freed_charge,
-             uint32_t* freed_count);
+  inline void Evict(size_t requested_charge, size_t* freed_charge,
+                    size_t* freed_count);
 
   // Returns the first slot in the probe sequence, starting from the given
   // probe number, with a handle e such that match(e) is true. At every
@@ -450,15 +475,54 @@ class ClockHandleTable {
   // value of probe is one more than the last non-aborting probe during the
   // call. This is so that that the variable can be used to keep track of
   // progress across consecutive calls to FindSlot.
-  inline ClockHandle* FindSlot(uint32_t hash,
-                               std::function<bool(ClockHandle*)> match,
-                               std::function<bool(ClockHandle*)> stop,
-                               std::function<void(ClockHandle*)> update,
-                               uint32_t& probe);
+  inline HandleImpl* FindSlot(const UniqueId64x2& hashed_key,
+                              std::function<bool(HandleImpl*)> match,
+                              std::function<bool(HandleImpl*)> stop,
+                              std::function<void(HandleImpl*)> update,
+                              size_t& probe);
 
   // Re-decrement all displacements in probe path starting from beginning
   // until (not including) the given handle
-  void Rollback(uint32_t hash, const ClockHandle* h);
+  inline void Rollback(const UniqueId64x2& hashed_key, const HandleImpl* h);
+
+  // Subtracts `total_charge` from `usage_` and 1 from `occupancy_`.
+  // Ideally this comes after releasing the entry itself so that we
+  // actually have the available occupancy/usage that is claimed.
+  // However, that means total_charge has to be saved from the handle
+  // before releasing it so that it can be provided to this function.
+  inline void ReclaimEntryUsage(size_t total_charge);
+
+  // Helper for updating `usage_` for new entry with given `total_charge`
+  // and evicting if needed under strict_capacity_limit=true rules. This
+  // means the operation might fail with Status::MemoryLimit. If
+  // `need_evict_for_occupancy`, then eviction of at least one entry is
+  // required, and the operation should fail if not possible.
+  // NOTE: Otherwise, occupancy_ is not managed in this function
+  inline Status ChargeUsageMaybeEvictStrict(size_t total_charge,
+                                            size_t capacity,
+                                            bool need_evict_for_occupancy);
+
+  // Helper for updating `usage_` for new entry with given `total_charge`
+  // and evicting if needed under strict_capacity_limit=false rules. This
+  // means that updating `usage_` always succeeds even if forced to exceed
+  // capacity. If `need_evict_for_occupancy`, then eviction of at least one
+  // entry is required, and the operation should return false if such eviction
+  // is not possible. `usage_` is not updated in that case. Otherwise, returns
+  // true, indicating success.
+  // NOTE: occupancy_ is not managed in this function
+  inline bool ChargeUsageMaybeEvictNonStrict(size_t total_charge,
+                                             size_t capacity,
+                                             bool need_evict_for_occupancy);
+
+  // Creates a "detached" handle for returning from an Insert operation that
+  // cannot be completed by actually inserting into the table.
+  // Updates `detached_usage_` but not `usage_` nor `occupancy_`.
+  inline HandleImpl* DetachedInsert(const ClockHandleBasicData& proto);
+
+  // 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);
 
  private:  // data
   // Number of hash bits used for table index.
@@ -466,13 +530,13 @@ class ClockHandleTable {
   const int length_bits_;
 
   // For faster computation of ModTableSize.
-  const uint32_t length_bits_mask_;
+  const size_t length_bits_mask_;
 
   // Maximum number of elements the user can store in the table.
-  const uint32_t occupancy_limit_;
+  const size_t occupancy_limit_;
 
   // Array of slots comprising the hash table.
-  const std::unique_ptr<ClockHandle[]> array_;
+  const std::unique_ptr<HandleImpl[]> array_;
 
   // We partition the following members into different cache lines
   // to avoid false sharing among Lookup, Release, Erase and Insert
@@ -484,96 +548,121 @@ class ClockHandleTable {
 
   ALIGN_AS(CACHE_LINE_SIZE)
   // Number of elements in the table.
-  std::atomic<uint32_t> occupancy_{};
+  std::atomic<size_t> occupancy_{};
 
   // Memory usage by entries tracked by the cache (including detached)
   std::atomic<size_t> usage_{};
 
   // Part of usage by detached entries (not in table)
   std::atomic<size_t> detached_usage_{};
-};  // class ClockHandleTable
+};  // class HyperClockTable
 
 // A single shard of sharded cache.
-class ALIGN_AS(CACHE_LINE_SIZE) ClockCacheShard final : public CacheShard {
+template <class Table>
+class ALIGN_AS(CACHE_LINE_SIZE) ClockCacheShard final : public CacheShardBase {
  public:
-  ClockCacheShard(size_t capacity, size_t estimated_value_size,
-                  bool strict_capacity_limit,
-                  CacheMetadataChargePolicy metadata_charge_policy);
-  ~ClockCacheShard() override = default;
+  ClockCacheShard(size_t capacity, bool strict_capacity_limit,
+                  CacheMetadataChargePolicy metadata_charge_policy,
+                  const typename Table::Opts& opts);
 
-  // TODO: document limitations
-  void SetCapacity(size_t capacity) override;
+  // For CacheShard concept
+  using HandleImpl = typename Table::HandleImpl;
+  // Hash is lossless hash of 128-bit key
+  using HashVal = UniqueId64x2;
+  using HashCref = const HashVal&;
+  static inline uint32_t HashPieceForSharding(HashCref hash) {
+    return Upper32of64(hash[0]);
+  }
+  static inline HashVal ComputeHash(const Slice& key) {
+    assert(key.size() == kCacheKeySize);
+    HashVal in;
+    HashVal out;
+    // NOTE: endian dependence
+    // TODO: use GetUnaligned?
+    std::memcpy(&in, key.data(), kCacheKeySize);
+    BijectiveHash2x64(in[1], in[0], &out[1], &out[0]);
+    return out;
+  }
 
-  void SetStrictCapacityLimit(bool strict_capacity_limit) override;
+  // For reconstructing key from hashed_key. Requires the caller to provide
+  // backing storage for the Slice in `unhashed`
+  static inline Slice ReverseHash(const UniqueId64x2& hashed,
+                                  UniqueId64x2* unhashed) {
+    BijectiveUnhash2x64(hashed[1], hashed[0], &(*unhashed)[1], &(*unhashed)[0]);
+    // NOTE: endian dependence
+    return Slice(reinterpret_cast<const char*>(unhashed), kCacheKeySize);
+  }
 
-  Status Insert(const Slice& key, uint32_t hash, void* value, size_t charge,
-                Cache::DeleterFn deleter, Cache::Handle** handle,
-                Cache::Priority priority) override;
+  // Although capacity is dynamically changeable, the number of table slots is
+  // not, so growing capacity substantially could lead to hitting occupancy
+  // limit.
+  void SetCapacity(size_t capacity);
 
-  Cache::Handle* Lookup(const Slice& key, uint32_t hash) override;
+  void SetStrictCapacityLimit(bool strict_capacity_limit);
 
-  bool Release(Cache::Handle* handle, bool useful,
-               bool erase_if_last_ref) override;
+  Status Insert(const Slice& key, const UniqueId64x2& hashed_key, void* value,
+                size_t charge, Cache::DeleterFn deleter, HandleImpl** handle,
+                Cache::Priority priority);
 
-  bool Release(Cache::Handle* handle, bool erase_if_last_ref = false) override;
+  HandleImpl* Lookup(const Slice& key, const UniqueId64x2& hashed_key);
 
-  bool Ref(Cache::Handle* handle) override;
+  bool Release(HandleImpl* handle, bool useful, bool erase_if_last_ref);
 
-  void Erase(const Slice& key, uint32_t hash) override;
+  bool Release(HandleImpl* handle, bool erase_if_last_ref = false);
 
-  size_t GetUsage() const override;
+  bool Ref(HandleImpl* handle);
 
-  size_t GetPinnedUsage() const override;
+  void Erase(const Slice& key, const UniqueId64x2& hashed_key);
 
-  size_t GetOccupancyCount() const override;
+  size_t GetCapacity() const;
 
-  size_t GetTableAddressCount() const override;
+  size_t GetUsage() const;
+
+  size_t GetDetachedUsage() const;
+
+  size_t GetPinnedUsage() const;
+
+  size_t GetOccupancyCount() const;
+
+  size_t GetOccupancyLimit() const;
+
+  size_t GetTableAddressCount() const;
 
   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;
+      size_t average_entries_per_lock, size_t* state);
 
-  void EraseUnRefEntries() override;
+  void EraseUnRefEntries();
 
-  std::string GetPrintableOptions() const override { return std::string{}; }
+  std::string GetPrintableOptions() const { return std::string{}; }
 
   // SecondaryCache not yet supported
-  Status Insert(const Slice& key, uint32_t hash, void* value,
+  Status Insert(const Slice& key, const UniqueId64x2& hashed_key, 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);
+                HandleImpl** handle, Cache::Priority priority) {
+    return Insert(key, hashed_key, 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);
+  HandleImpl* Lookup(const Slice& key, const UniqueId64x2& hashed_key,
+                     const Cache::CacheItemHelper* /*helper*/,
+                     const Cache::CreateCallback& /*create_cb*/,
+                     Cache::Priority /*priority*/, bool /*wait*/,
+                     Statistics* /*stats*/) {
+    return Lookup(key, hashed_key);
   }
 
-  bool IsReady(Cache::Handle* /*handle*/) override { return true; }
+  bool IsReady(HandleImpl* /*handle*/) { return true; }
 
-  void Wait(Cache::Handle* /*handle*/) override {}
+  void Wait(HandleImpl* /*handle*/) {}
 
   // Acquire/release N references
-  void TEST_RefN(Cache::Handle* handle, size_t n);
-  void TEST_ReleaseN(Cache::Handle* handle, size_t n);
-
- private:  // functions
-  friend class ClockCache;
-  friend class ClockCacheTest;
-
-  ClockHandle* DetachedInsert(const ClockHandleMoreData& h);
-
-  // 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);
+  void TEST_RefN(HandleImpl* handle, size_t n);
+  void TEST_ReleaseN(HandleImpl* handle, size_t n);
 
  private:  // data
-  ClockHandleTable table_;
+  Table table_;
 
   // Maximum total charge of all elements stored in the table.
   std::atomic<size_t> capacity_;
@@ -586,37 +675,27 @@ class HyperClockCache
 #ifdef NDEBUG
     final
 #endif
-    : public ShardedCache {
+    : public ShardedCache<ClockCacheShard<HyperClockTable>> {
  public:
+  using Shard = ClockCacheShard<HyperClockTable>;
+
   HyperClockCache(size_t capacity, size_t estimated_value_size,
                   int num_shard_bits, bool strict_capacity_limit,
-                  CacheMetadataChargePolicy metadata_charge_policy =
-                      kDontChargeCacheMetadata);
-
-  ~HyperClockCache() override;
+                  CacheMetadataChargePolicy metadata_charge_policy,
+                  std::shared_ptr<MemoryAllocator> memory_allocator);
 
   const char* Name() const override { return "HyperClockCache"; }
 
-  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:
-  ClockCacheShard* shards_ = nullptr;
-
-  int num_shards_;
+  void ReportProblems(
+      const std::shared_ptr<Logger>& /*info_log*/) const override;
 };  // class HyperClockCache
 
-}  // namespace hyper_clock_cache
+}  // namespace clock_cache
 
 }  // namespace ROCKSDB_NAMESPACE

cache/compressed_secondary_cache.cc

@@ -178,6 +178,19 @@ Status CompressedSecondaryCache::Insert(const Slice& key, void* value,
 
 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);
@@ -194,9 +207,9 @@ 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()};

cache/compressed_secondary_cache.h

@@ -15,6 +15,7 @@
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
 #include "util/compression.h"
+#include "util/mutexlock.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -22,7 +23,7 @@ class CompressedSecondaryCacheResultHandle : public SecondaryCacheResultHandle {
  public:
   CompressedSecondaryCacheResultHandle(void* value, size_t size)
       : value_(value), size_(size) {}
-  virtual ~CompressedSecondaryCacheResultHandle() override = default;
+  ~CompressedSecondaryCacheResultHandle() override = default;
 
   CompressedSecondaryCacheResultHandle(
       const CompressedSecondaryCacheResultHandle&) = delete;
@@ -78,7 +79,7 @@ class CompressedSecondaryCache : public SecondaryCache {
       CompressionType compression_type = CompressionType::kLZ4Compression,
       uint32_t compress_format_version = 2,
       bool enable_custom_split_merge = false);
-  virtual ~CompressedSecondaryCache() override;
+  ~CompressedSecondaryCache() override;
 
   const char* Name() const override { return "CompressedSecondaryCache"; }
 
@@ -95,6 +96,10 @@ class CompressedSecondaryCache : public SecondaryCache {
 
   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:
@@ -116,7 +121,7 @@ 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
@@ -128,6 +133,7 @@ class CompressedSecondaryCache : public SecondaryCache {
   static Cache::DeleterFn GetDeletionCallback(bool enable_custom_split_merge);
   std::shared_ptr<Cache> cache_;
   CompressedSecondaryCacheOptions cache_options_;
+  mutable port::Mutex capacity_mutex_;
 };
 
 }  // namespace ROCKSDB_NAMESPACE

cache/compressed_secondary_cache_test.cc

@@ -5,28 +5,21 @@
 
 #include "cache/compressed_secondary_cache.h"
 
-#include <algorithm>
-#include <cstdint>
 #include <iterator>
+#include <memory>
 #include <tuple>
 
-#include "cache/lru_cache.h"
 #include "memory/jemalloc_nodump_allocator.h"
-#include "memory/memory_allocator.h"
-#include "rocksdb/compression_type.h"
 #include "rocksdb/convenience.h"
-#include "rocksdb/secondary_cache.h"
 #include "test_util/testharness.h"
 #include "test_util/testutil.h"
-#include "util/compression.h"
-#include "util/random.h"
 
 namespace ROCKSDB_NAMESPACE {
 
 class CompressedSecondaryCacheTest : public testing::Test {
  public:
   CompressedSecondaryCacheTest() : fail_create_(false) {}
-  ~CompressedSecondaryCacheTest() {}
+  ~CompressedSecondaryCacheTest() override = default;
 
  protected:
   class TestItem {
@@ -34,10 +27,10 @@ class CompressedSecondaryCacheTest : public testing::Test {
     TestItem(const char* buf, size_t size) : buf_(new char[size]), size_(size) {
       memcpy(buf_.get(), buf, size);
     }
-    ~TestItem() {}
+    ~TestItem() = default;
 
     char* Buf() { return buf_.get(); }
-    size_t Size() { return size_; }
+    [[nodiscard]] size_t Size() const { return size_; }
 
    private:
     std::unique_ptr<char[]> buf_;
@@ -50,7 +43,7 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
   static Status SaveToCallback(void* from_obj, size_t from_offset,
                                size_t length, void* out) {
-    TestItem* item = reinterpret_cast<TestItem*>(from_obj);
+    auto item = reinterpret_cast<TestItem*>(from_obj);
     const char* buf = item->Buf();
     EXPECT_EQ(length, item->Size());
     EXPECT_EQ(from_offset, 0);
@@ -315,12 +308,12 @@ class CompressedSecondaryCacheTest : public testing::Test {
     get_perf_context()->Reset();
     Random rnd(301);
     std::string str1 = rnd.RandomString(1001);
-    TestItem* item1_1 = new TestItem(str1.data(), str1.length());
+    auto item1_1 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert(
         "k1", item1_1, &CompressedSecondaryCacheTest::helper_, str1.length()));
 
     std::string str2 = rnd.RandomString(1012);
-    TestItem* item2_1 = new TestItem(str2.data(), str2.length());
+    auto item2_1 = new TestItem(str2.data(), str2.length());
     // After this Insert, primary cache contains k2 and secondary cache contains
     // k1's dummy item.
     ASSERT_OK(cache->Insert(
@@ -330,7 +323,7 @@ class CompressedSecondaryCacheTest : public testing::Test {
     ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
 
     std::string str3 = rnd.RandomString(1024);
-    TestItem* item3_1 = new TestItem(str3.data(), str3.length());
+    auto item3_1 = new TestItem(str3.data(), str3.length());
     // After this Insert, primary cache contains k3 and secondary cache contains
     // k1's dummy item and k2's dummy item.
     ASSERT_OK(cache->Insert(
@@ -339,14 +332,14 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     // After this Insert, primary cache contains k1 and secondary cache contains
     // k1's dummy item, k2's dummy item, and k3's dummy item.
-    TestItem* item1_2 = new TestItem(str1.data(), str1.length());
+    auto item1_2 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert(
         "k1", item1_2, &CompressedSecondaryCacheTest::helper_, str1.length()));
     ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 3);
 
     // After this Insert, primary cache contains k2 and secondary cache contains
     // k1's item, k2's dummy item, and k3's dummy item.
-    TestItem* item2_2 = new TestItem(str2.data(), str2.length());
+    auto item2_2 = new TestItem(str2.data(), str2.length());
     ASSERT_OK(cache->Insert(
         "k2", item2_2, &CompressedSecondaryCacheTest::helper_, str2.length()));
     ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 1);
@@ -362,7 +355,7 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     // After this Insert, primary cache contains k3 and secondary cache contains
     // k1's item and k2's item.
-    TestItem* item3_2 = new TestItem(str3.data(), str3.length());
+    auto item3_2 = new TestItem(str3.data(), str3.length());
     ASSERT_OK(cache->Insert(
         "k3", item3_2, &CompressedSecondaryCacheTest::helper_, str3.length()));
     ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 2);
@@ -381,7 +374,7 @@ class CompressedSecondaryCacheTest : public testing::Test {
                            test_item_creator, Cache::Priority::LOW, true,
                            stats.get());
     ASSERT_NE(handle, nullptr);
-    TestItem* val3 = static_cast<TestItem*>(cache->Value(handle));
+    auto val3 = static_cast<TestItem*>(cache->Value(handle));
     ASSERT_NE(val3, nullptr);
     ASSERT_EQ(memcmp(val3->Buf(), item3_2->Buf(), item3_2->Size()), 0);
     cache->Release(handle);
@@ -399,18 +392,20 @@ class CompressedSecondaryCacheTest : public testing::Test {
                            stats.get());
     ASSERT_NE(handle, nullptr);
     ASSERT_EQ(get_perf_context()->block_cache_standalone_handle_count, 1);
-    TestItem* val1_1 = static_cast<TestItem*>(cache->Value(handle));
+    auto val1_1 = static_cast<TestItem*>(cache->Value(handle));
     ASSERT_NE(val1_1, nullptr);
     ASSERT_EQ(memcmp(val1_1->Buf(), str1.data(), str1.size()), 0);
     cache->Release(handle);
 
     // This Lookup should erase k1 from the secondary cache and insert
     // it into primary cache; then k3 is demoted.
+    // k2 and k3 are in secondary cache.
     handle = cache->Lookup("k1", &CompressedSecondaryCacheTest::helper_,
                            test_item_creator, Cache::Priority::LOW, true,
                            stats.get());
     ASSERT_NE(handle, nullptr);
     ASSERT_EQ(get_perf_context()->block_cache_standalone_handle_count, 1);
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 3);
     cache->Release(handle);
 
     // k2 is still in secondary cache.
@@ -421,6 +416,54 @@ class CompressedSecondaryCacheTest : public testing::Test {
     ASSERT_EQ(get_perf_context()->block_cache_standalone_handle_count, 2);
     cache->Release(handle);
 
+    // Testing SetCapacity().
+    ASSERT_OK(secondary_cache->SetCapacity(0));
+    handle = cache->Lookup("k3", &CompressedSecondaryCacheTest::helper_,
+                           test_item_creator, Cache::Priority::LOW, true,
+                           stats.get());
+    ASSERT_EQ(handle, nullptr);
+
+    ASSERT_OK(secondary_cache->SetCapacity(7000));
+    size_t capacity;
+    ASSERT_OK(secondary_cache->GetCapacity(capacity));
+    ASSERT_EQ(capacity, 7000);
+    auto item1_3 = new TestItem(str1.data(), str1.length());
+    // After this Insert, primary cache contains k1.
+    ASSERT_OK(cache->Insert(
+        "k1", item1_3, &CompressedSecondaryCacheTest::helper_, str2.length()));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 3);
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 4);
+
+    auto item2_3 = new TestItem(str2.data(), str2.length());
+    // After this Insert, primary cache contains k2 and secondary cache contains
+    // k1's dummy item.
+    ASSERT_OK(cache->Insert(
+        "k2", item2_3, &CompressedSecondaryCacheTest::helper_, str1.length()));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 4);
+
+    auto item1_4 = new TestItem(str1.data(), str1.length());
+    // After this Insert, primary cache contains k1 and secondary cache contains
+    // k1's dummy item and k2's dummy item.
+    ASSERT_OK(cache->Insert(
+        "k1", item1_4, &CompressedSecondaryCacheTest::helper_, str2.length()));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 5);
+
+    auto item2_4 = new TestItem(str2.data(), str2.length());
+    // After this Insert, primary cache contains k2 and secondary cache contains
+    // k1's real item and k2's dummy item.
+    ASSERT_OK(cache->Insert(
+        "k2", item2_4, &CompressedSecondaryCacheTest::helper_, str2.length()));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 5);
+    // This Lookup should just insert a dummy handle in the primary cache
+    // and the k1 is still in the secondary cache.
+    handle = cache->Lookup("k1", &CompressedSecondaryCacheTest::helper_,
+                           test_item_creator, Cache::Priority::LOW, true,
+                           stats.get());
+
+    ASSERT_NE(handle, nullptr);
+    cache->Release(handle);
+    ASSERT_EQ(get_perf_context()->block_cache_standalone_handle_count, 3);
+
     cache.reset();
     secondary_cache.reset();
   }
@@ -452,8 +495,7 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     Random rnd(301);
     std::string str1 = rnd.RandomString(1001);
-    auto item1 =
-        std::unique_ptr<TestItem>(new TestItem(str1.data(), str1.length()));
+    auto item1 = std::make_unique<TestItem>(str1.data(), str1.length());
     ASSERT_NOK(cache->Insert("k1", item1.get(), nullptr, str1.length()));
     ASSERT_OK(cache->Insert("k1", item1.get(),
                             &CompressedSecondaryCacheTest::helper_,
@@ -500,13 +542,13 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     Random rnd(301);
     std::string str1 = rnd.RandomString(1001);
-    TestItem* item1 = new TestItem(str1.data(), str1.length());
+    auto item1 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert("k1", item1,
                             &CompressedSecondaryCacheTest::helper_fail_,
                             str1.length()));
 
     std::string str2 = rnd.RandomString(1002);
-    TestItem* item2 = new TestItem(str2.data(), str2.length());
+    auto item2 = new TestItem(str2.data(), str2.length());
     // k1 should be demoted to the secondary cache.
     ASSERT_OK(cache->Insert("k2", item2,
                             &CompressedSecondaryCacheTest::helper_fail_,
@@ -559,12 +601,12 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     Random rnd(301);
     std::string str1 = rnd.RandomString(1001);
-    TestItem* item1 = new TestItem(str1.data(), str1.length());
+    auto item1 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert("k1", item1, &CompressedSecondaryCacheTest::helper_,
                             str1.length()));
 
     std::string str2 = rnd.RandomString(1002);
-    TestItem* item2 = new TestItem(str2.data(), str2.length());
+    auto item2 = new TestItem(str2.data(), str2.length());
     // k1 should be demoted to the secondary cache.
     ASSERT_OK(cache->Insert("k2", item2, &CompressedSecondaryCacheTest::helper_,
                             str2.length()));
@@ -617,13 +659,13 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     Random rnd(301);
     std::string str1 = rnd.RandomString(1001);
-    TestItem* item1_1 = new TestItem(str1.data(), str1.length());
+    auto item1_1 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert(
         "k1", item1_1, &CompressedSecondaryCacheTest::helper_, str1.length()));
 
     std::string str2 = rnd.RandomString(1002);
     std::string str2_clone{str2};
-    TestItem* item2 = new TestItem(str2.data(), str2.length());
+    auto item2 = new TestItem(str2.data(), str2.length());
     // After this Insert, primary cache contains k2 and secondary cache contains
     // k1's dummy item.
     ASSERT_OK(cache->Insert("k2", item2, &CompressedSecondaryCacheTest::helper_,
@@ -631,11 +673,11 @@ class CompressedSecondaryCacheTest : public testing::Test {
 
     // After this Insert, primary cache contains k1 and secondary cache contains
     // k1's dummy item and k2's dummy item.
-    TestItem* item1_2 = new TestItem(str1.data(), str1.length());
+    auto item1_2 = new TestItem(str1.data(), str1.length());
     ASSERT_OK(cache->Insert(
         "k1", item1_2, &CompressedSecondaryCacheTest::helper_, str1.length()));
 
-    TestItem* item2_2 = new TestItem(str2.data(), str2.length());
+    auto item2_2 = new TestItem(str2.data(), str2.length());
     // After this Insert, primary cache contains k2 and secondary cache contains
     // k1's item and k2's dummy item.
     ASSERT_OK(cache->Insert(
@@ -957,6 +999,7 @@ TEST_F(CompressedSecondaryCacheTest, SplictValueAndMergeChunksTest) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

cache/fast_lru_cache.cc

@@ -1,622 +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)
-    : CacheShard(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) {
-  // 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_;
-}
-
-size_t LRUCacheShard::GetOccupancyCount() const {
-  DMutexLock l(mutex_);
-  return table_.GetOccupancy();
-}
-
-size_t LRUCacheShard::GetTableAddressCount() const {
-  DMutexLock l(mutex_);
-  return table_.GetTableSize();
-}
-
-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

@@ -1,476 +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;
-  size_t GetOccupancyCount() const override;
-  size_t GetTableAddressCount() 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

@@ -22,6 +22,9 @@
 namespace ROCKSDB_NAMESPACE {
 namespace lru_cache {
 
+// A distinct pointer value for marking "dummy" cache entries
+void* const kDummyValueMarker = const_cast<char*>("kDummyValueMarker");
+
 LRUHandleTable::LRUHandleTable(int max_upper_hash_bits)
     : length_bits_(/* historical starting size*/ 4),
       list_(new LRUHandle* [size_t{1} << length_bits_] {}),
@@ -35,7 +38,7 @@ LRUHandleTable::~LRUHandleTable() {
           h->Free();
         }
       },
-      0, uint32_t{1} << length_bits_);
+      0, size_t{1} << length_bits_);
 }
 
 LRUHandle* LRUHandleTable::Lookup(const Slice& key, uint32_t hash) {
@@ -110,12 +113,13 @@ 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)
-    : CacheShard(metadata_charge_policy),
+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,
+                             SecondaryCache* secondary_cache)
+    : CacheShardBase(metadata_charge_policy),
       capacity_(0),
       high_pri_pool_usage_(0),
       low_pri_pool_usage_(0),
@@ -162,27 +166,27 @@ void LRUCacheShard::EraseUnRefEntries() {
 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) {
+    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(
@@ -361,7 +365,7 @@ 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;
@@ -382,7 +386,7 @@ 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.");
@@ -411,7 +415,7 @@ Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
         if (!e->HasRefs()) {
           e->Ref();
         }
-        *handle = reinterpret_cast<Cache::Handle*>(e);
+        *handle = e;
       }
     }
   }
@@ -425,16 +429,20 @@ void LRUCacheShard::Promote(LRUHandle* e) {
   SecondaryCacheResultHandle* secondary_handle = e->sec_handle;
 
   assert(secondary_handle->IsReady());
-  e->SetIncomplete(false);
-  e->SetInCache(false);
+  // 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;
 
   if (e->value) {
+    e->CalcTotalCharge(value_size, metadata_charge_policy_);
     Status s;
-    if (secondary_cache_ && secondary_cache_->SupportForceErase() &&
-        e->IsStandalone()) {
+    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;
@@ -464,15 +472,16 @@ void LRUCacheShard::Promote(LRUHandle* e) {
 
       // 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, /*value=*/nullptr, 0,
+      s = Insert(e->key(), e->hash, kDummyValueMarker, /*charge=*/0,
                  /*deleter=*/nullptr, /*helper=*/nullptr, /*handle=*/nullptr,
                  priority);
     } else {
       e->SetInCache(true);
-      e->SetIsStandalone(false);
-      Cache::Handle* handle = reinterpret_cast<Cache::Handle*>(e);
+      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
@@ -489,23 +498,19 @@ void LRUCacheShard::Promote(LRUHandle* e) {
       // 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);
-    e->SetIsStandalone(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,
+                                 const Cache::CreateCallback& create_cb,
+                                 Cache::Priority priority, bool wait,
+                                 Statistics* stats) {
   LRUHandle* e = nullptr;
   bool found_dummy_entry{false};
   {
@@ -513,21 +518,22 @@ Cache::Handle* LRUCacheShard::Lookup(
     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);
-      }
-      e->Ref();
-      e->SetHit();
-
-      // 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.
-      if (!e->value && secondary_cache_ &&
-          secondary_cache_->SupportForceErase()) {
+      if (e->value == kDummyValueMarker) {
+        // 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();
       }
     }
   }
@@ -541,30 +547,22 @@ Cache::Handle* LRUCacheShard::Lookup(
   // 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.
-  //
-  // Only support synchronous for now.
-  // TODO: Support asynchronous lookup in secondary cache
-  if ((!e || found_dummy_entry) && secondary_cache_ && helper &&
-      helper->saveto_cb) {
+  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);
-    // Release the dummy handle.
-    if (e) {
-      Release(reinterpret_cast<Cache::Handle*>(e), true /*erase_if_last_ref*/);
-    }
     bool is_in_sec_cache{false};
     std::unique_ptr<SecondaryCacheResultHandle> secondary_handle =
         secondary_cache_->Lookup(key, create_cb, 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->m_flags = 0;
+      e->im_flags = 0;
       e->SetSecondaryCacheCompatible(true);
       e->info_.helper = helper;
       e->key_length = key.size();
@@ -578,10 +576,8 @@ Cache::Handle* LRUCacheShard::Lookup(
       e->total_charge = 0;
       e->Ref();
       e->SetIsInSecondaryCache(is_in_sec_cache);
-
-      if (secondary_cache_->SupportForceErase() && !found_dummy_entry) {
-        e->SetIsStandalone(true);
-      }
+      e->SetIsStandalone(secondary_cache_->SupportForceErase() &&
+                         !found_dummy_entry);
 
       if (wait) {
         Promote(e);
@@ -599,24 +595,27 @@ Cache::Handle* LRUCacheShard::Lookup(
       } 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 {
-      e = nullptr;
+      // 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;
 }
@@ -635,12 +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();
@@ -658,12 +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;
     }
@@ -680,22 +678,25 @@ Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
                              size_t charge,
                              void (*deleter)(const Slice& key, void* value),
                              const Cache::CacheItemHelper* helper,
-                             Cache::Handle** handle, Cache::Priority priority) {
+                             LRUHandle** handle, Cache::Priority priority) {
   // 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;
+  e->m_flags = 0;
+  e->im_flags = 0;
   if (helper) {
+    // Use only one of the two parameters
+    assert(deleter == nullptr);
+    // value == nullptr is reserved for indicating failure for when secondary
+    // cache compatible
+    assert(value != nullptr);
     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->key_length = key.size();
@@ -736,9 +737,7 @@ void LRUCacheShard::Erase(const Slice& key, uint32_t hash) {
   }
 }
 
-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_);
@@ -769,7 +768,7 @@ size_t LRUCacheShard::GetTableAddressCount() const {
   return size_t{1} << table_.GetLengthBits();
 }
 
-std::string LRUCacheShard::GetPrintableOptions() const {
+void LRUCacheShard::AppendPrintableOptions(std::string& str) const {
   const int kBufferSize = 200;
   char buffer[kBufferSize];
   {
@@ -779,7 +778,7 @@ 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,
@@ -788,51 +787,30 @@ 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();
+  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;
-}
-
-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;
+  auto h = reinterpret_cast<const LRUHandle*>(handle);
+  assert(!h->IsPending() || h->value == nullptr);
+  assert(h->value != kDummyValueMarker);
+  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 {
@@ -844,32 +822,12 @@ Cache::DeleterFn LRUCache::GetDeleter(Handle* handle) const {
   }
 }
 
-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;
-  }
-}
-
 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) {
@@ -895,22 +853,17 @@ 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
@@ -930,7 +883,7 @@ 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

@@ -38,21 +38,14 @@ 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)
-// 4. The handle is never inserted into the LRUCache (both hash table and LRU
-//    list) and it doesn't experience the above three states.
-//    The entry can be freed when refs becomes 0.
-//    (refs >= 1 && in_cache == false && IS_STANDALONE == true)
-// All newly created LRUHandles are in state 1 or 4. 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;
@@ -60,7 +53,7 @@ struct LRUHandle {
     Info() {}
     ~Info() {}
     Cache::DeleterFn deleter;
-    const ShardedCache::CacheItemHelper* helper;
+    const Cache::CacheItemHelper* helper;
   } info_;
   // An entry is not added to the LRUHandleTable until the secondary cache
   // lookup is complete, so its safe to have this union.
@@ -77,45 +70,47 @@ 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),
-    // Whether this entry is not inserted into the cache (both hash table and
-    // LRU list).
-    IS_STANDALONE = (1 << 9),
+    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),
+    // Can this be inserted into the secondary cache.
+    IM_IS_SECONDARY_CACHE_COMPATIBLE = (1 << 2),
+    // Is the handle still being read from a lower tier.
+    IM_IS_PENDING = (1 << 3),
+    // Whether this handle is still in a lower tier
+    IM_IS_IN_SECONDARY_CACHE = (1 << 4),
+    // Marks result handles that should not be inserted into cache
+    IM_IS_STANDALONE = (1 << 5),
+  };
 
   // 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++; }
 
@@ -129,104 +124,95 @@ 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 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 {
-#ifdef __SANITIZE_THREAD__
-    return is_secondary_cache_compatible_for_tsan;
-#else
-    return flags & IS_SECONDARY_CACHE_COMPATIBLE;
-#endif  // __SANITIZE_THREAD__
+    return im_flags & IM_IS_SECONDARY_CACHE_COMPATIBLE;
   }
-  bool IsPending() const { return flags & IS_PENDING; }
-  bool IsInSecondaryCache() const { return flags & IS_IN_SECONDARY_CACHE; }
-  bool IsStandalone() const { return flags & IS_STANDALONE; }
+  bool IsPending() const { return im_flags & IM_IS_PENDING; }
+  bool IsInSecondaryCache() const {
+    return im_flags & IM_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;
+      im_flags |= IM_IS_SECONDARY_CACHE_COMPATIBLE;
     } else {
-      flags &= ~IS_SECONDARY_CACHE_COMPATIBLE;
+      im_flags &= ~IM_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;
+  void SetIsPending(bool pending) {
+    if (pending) {
+      im_flags |= IM_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 SetIsStandalone(bool is_standalone) {
     if (is_standalone) {
-      flags |= IS_STANDALONE;
+      im_flags |= IM_IS_STANDALONE;
     } else {
-      flags &= ~IS_STANDALONE;
+      im_flags &= ~IM_IS_STANDALONE;
     }
   }
 
   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()) {
@@ -241,7 +227,8 @@ struct LRUHandle {
         (*info_.helper->del_cb)(key(), value);
       }
     }
-    delete[] reinterpret_cast<char*>(this);
+
+    free(this);
   }
 
   inline size_t CalcuMetaCharge(
@@ -280,9 +267,6 @@ 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);
   ~LRUHandleTable();
 
@@ -291,8 +275,8 @@ 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;
@@ -331,23 +315,31 @@ class LRUHandleTable {
 };
 
 // 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, 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);
@@ -356,58 +348,49 @@ 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 {
+  inline Status Insert(const Slice& key, uint32_t hash, void* value,
+                       size_t charge, Cache::DeleterFn deleter,
+                       LRUHandle** handle, Cache::Priority priority) {
     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 {
+  inline Status Insert(const Slice& key, uint32_t hash, void* value,
+                       const Cache::CacheItemHelper* helper, size_t charge,
+                       LRUHandle** handle, Cache::Priority priority) {
     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 {
+  LRUHandle* Lookup(const Slice& key, uint32_t hash,
+                    const Cache::CacheItemHelper* helper,
+                    const Cache::CreateCallback& create_cb,
+                    Cache::Priority priority, bool wait, Statistics* stats);
+  inline LRUHandle* Lookup(const Slice& key, uint32_t hash) {
     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;
+  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;
-  virtual size_t GetOccupancyCount() const override;
-  virtual size_t GetTableAddressCount() const override;
+  size_t GetUsage() const;
+  size_t GetPinnedUsage() const;
+  size_t GetOccupancyCount() const;
+  size_t GetTableAddressCount() const;
 
-  virtual void ApplyToSomeEntries(
+  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;
+      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);
 
@@ -421,17 +404,19 @@ 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);
+                LRUHandle** 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
@@ -518,14 +503,15 @@ 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,
@@ -533,27 +519,21 @@ 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"; }
+  void* Value(Handle* handle) override;
+  size_t GetCharge(Handle* handle) const override;
+  DeleterFn GetDeleter(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/lru_cache_test.cc

@@ -10,7 +10,6 @@
 
 #include "cache/cache_key.h"
 #include "cache/clock_cache.h"
-#include "cache/fast_lru_cache.h"
 #include "db/db_test_util.h"
 #include "file/sst_file_manager_impl.h"
 #include "port/port.h"
@@ -67,7 +66,7 @@ class LRUCacheTest : public testing::Test {
   bool Lookup(const std::string& key) {
     auto handle = cache_->Lookup(key, 0 /*hash*/);
     if (handle) {
-      cache_->Release(handle);
+      cache_->Release(handle, true /*useful*/, false /*erase*/);
       return true;
     }
     return false;
@@ -364,152 +363,14 @@ TEST_F(LRUCacheTest, EntriesWithPriority) {
   ValidateLRUList({"x", "y", "g", "z", "d", "m"}, 2, 2, 2);
 }
 
-// TODO: FastLRUCache and ClockCache use the same tests. We can probably remove
-// them from FastLRUCache after ClockCache becomes productive, and we don't plan
-// to use or maintain FastLRUCache any more.
-namespace fast_lru_cache {
-
-// TODO(guido) Replicate LRU policy tests from LRUCache here.
-class FastLRUCacheTest : public testing::Test {
- public:
-  FastLRUCacheTest() {}
-  ~FastLRUCacheTest() override { DeleteCache(); }
-
-  void DeleteCache() {
-    if (cache_ != nullptr) {
-      cache_->~LRUCacheShard();
-      port::cacheline_aligned_free(cache_);
-      cache_ = nullptr;
-    }
-  }
-
-  void NewCache(size_t capacity) {
-    DeleteCache();
-    cache_ = reinterpret_cast<LRUCacheShard*>(
-        port::cacheline_aligned_alloc(sizeof(LRUCacheShard)));
-    new (cache_) LRUCacheShard(capacity, 1 /*estimated_value_size*/,
-                               false /*strict_capacity_limit*/,
-                               kDontChargeCacheMetadata);
-  }
-
-  Status Insert(const std::string& key) {
-    return cache_->Insert(key, 0 /*hash*/, nullptr /*value*/, 1 /*charge*/,
-                          nullptr /*deleter*/, nullptr /*handle*/,
-                          Cache::Priority::LOW);
-  }
-
-  Status Insert(char key, size_t len) { return Insert(std::string(len, key)); }
-
-  size_t CalcEstimatedHandleChargeWrapper(
-      size_t estimated_value_size,
-      CacheMetadataChargePolicy metadata_charge_policy) {
-    return LRUCacheShard::CalcEstimatedHandleCharge(estimated_value_size,
-                                                    metadata_charge_policy);
-  }
-
-  int CalcHashBitsWrapper(size_t capacity, size_t estimated_value_size,
-                          CacheMetadataChargePolicy metadata_charge_policy) {
-    return LRUCacheShard::CalcHashBits(capacity, estimated_value_size,
-                                       metadata_charge_policy);
-  }
-
-  // Maximum number of items that a shard can hold.
-  double CalcMaxOccupancy(size_t capacity, size_t estimated_value_size,
-                          CacheMetadataChargePolicy metadata_charge_policy) {
-    size_t handle_charge = LRUCacheShard::CalcEstimatedHandleCharge(
-        estimated_value_size, metadata_charge_policy);
-    return capacity / (kLoadFactor * handle_charge);
-  }
-  bool TableSizeIsAppropriate(int hash_bits, double max_occupancy) {
-    if (hash_bits == 0) {
-      return max_occupancy <= 1;
-    } else {
-      return (1 << hash_bits >= max_occupancy) &&
-             (1 << (hash_bits - 1) <= max_occupancy);
-    }
-  }
-
- private:
-  LRUCacheShard* cache_ = nullptr;
-};
-
-TEST_F(FastLRUCacheTest, ValidateKeySize) {
-  NewCache(3);
-  EXPECT_OK(Insert('a', 16));
-  EXPECT_NOK(Insert('b', 15));
-  EXPECT_OK(Insert('b', 16));
-  EXPECT_NOK(Insert('c', 17));
-  EXPECT_NOK(Insert('d', 1000));
-  EXPECT_NOK(Insert('e', 11));
-  EXPECT_NOK(Insert('f', 0));
-}
-
-TEST_F(FastLRUCacheTest, CalcHashBitsTest) {
-  size_t capacity;
-  size_t estimated_value_size;
-  double max_occupancy;
-  int hash_bits;
-  CacheMetadataChargePolicy metadata_charge_policy;
-  // Vary the cache capacity, fix the element charge.
-  for (int i = 0; i < 2048; i++) {
-    capacity = i;
-    estimated_value_size = 0;
-    metadata_charge_policy = kFullChargeCacheMetadata;
-    max_occupancy = CalcMaxOccupancy(capacity, estimated_value_size,
-                                     metadata_charge_policy);
-    hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                    metadata_charge_policy);
-    EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
-  }
-  // Fix the cache capacity, vary the element charge.
-  for (int i = 0; i < 1024; i++) {
-    capacity = 1024;
-    estimated_value_size = i;
-    metadata_charge_policy = kFullChargeCacheMetadata;
-    max_occupancy = CalcMaxOccupancy(capacity, estimated_value_size,
-                                     metadata_charge_policy);
-    hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                    metadata_charge_policy);
-    EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
-  }
-  // Zero-capacity cache, and only values have charge.
-  capacity = 0;
-  estimated_value_size = 1;
-  metadata_charge_policy = kDontChargeCacheMetadata;
-  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                  metadata_charge_policy);
-  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, 0 /* max_occupancy */));
-  // Zero-capacity cache, and only metadata has charge.
-  capacity = 0;
-  estimated_value_size = 0;
-  metadata_charge_policy = kFullChargeCacheMetadata;
-  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                  metadata_charge_policy);
-  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, 0 /* max_occupancy */));
-  // Small cache, large elements.
-  capacity = 1024;
-  estimated_value_size = 8192;
-  metadata_charge_policy = kFullChargeCacheMetadata;
-  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                  metadata_charge_policy);
-  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, 0 /* max_occupancy */));
-  // Large capacity.
-  capacity = 31924172;
-  estimated_value_size = 8192;
-  metadata_charge_policy = kFullChargeCacheMetadata;
-  max_occupancy =
-      CalcMaxOccupancy(capacity, estimated_value_size, metadata_charge_policy);
-  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
-                                  metadata_charge_policy);
-  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
-}
-
-}  // namespace fast_lru_cache
-
-namespace hyper_clock_cache {
+namespace clock_cache {
 
 class ClockCacheTest : public testing::Test {
  public:
+  using Shard = HyperClockCache::Shard;
+  using Table = HyperClockTable;
+  using HandleImpl = Shard::HandleImpl;
+
   ClockCacheTest() {}
   ~ClockCacheTest() override { DeleteShard(); }
 
@@ -523,28 +384,36 @@ class ClockCacheTest : public testing::Test {
 
   void NewShard(size_t capacity, bool strict_capacity_limit = true) {
     DeleteShard();
-    shard_ = reinterpret_cast<ClockCacheShard*>(
-        port::cacheline_aligned_alloc(sizeof(ClockCacheShard)));
-    new (shard_) ClockCacheShard(capacity, 1, strict_capacity_limit,
-                                 kDontChargeCacheMetadata);
+    shard_ =
+        reinterpret_cast<Shard*>(port::cacheline_aligned_alloc(sizeof(Shard)));
+
+    Table::Opts opts;
+    opts.estimated_value_size = 1;
+    new (shard_)
+        Shard(capacity, strict_capacity_limit, kDontChargeCacheMetadata, opts);
   }
 
-  Status Insert(const std::string& key,
+  Status Insert(const UniqueId64x2& hashed_key,
                 Cache::Priority priority = Cache::Priority::LOW) {
-    return shard_->Insert(key, 0 /*hash*/, nullptr /*value*/, 1 /*charge*/,
-                          nullptr /*deleter*/, nullptr /*handle*/, priority);
+    return shard_->Insert(TestKey(hashed_key), hashed_key, nullptr /*value*/,
+                          1 /*charge*/, nullptr /*deleter*/, nullptr /*handle*/,
+                          priority);
   }
 
   Status Insert(char key, Cache::Priority priority = Cache::Priority::LOW) {
-    return Insert(std::string(kCacheKeySize, key), priority);
+    return Insert(TestHashedKey(key), priority);
   }
 
   Status InsertWithLen(char key, size_t len) {
-    return Insert(std::string(len, key));
+    std::string skey(len, key);
+    return shard_->Insert(skey, TestHashedKey(key), nullptr /*value*/,
+                          1 /*charge*/, nullptr /*deleter*/, nullptr /*handle*/,
+                          Cache::Priority::LOW);
   }
 
-  bool Lookup(const std::string& key, bool useful = true) {
-    auto handle = shard_->Lookup(key, 0 /*hash*/);
+  bool Lookup(const Slice& key, const UniqueId64x2& hashed_key,
+              bool useful = true) {
+    auto handle = shard_->Lookup(key, hashed_key);
     if (handle) {
       shard_->Release(handle, useful, /*erase_if_last_ref=*/false);
       return true;
@@ -552,45 +421,30 @@ class ClockCacheTest : public testing::Test {
     return false;
   }
 
+  bool Lookup(const UniqueId64x2& hashed_key, bool useful = true) {
+    return Lookup(TestKey(hashed_key), hashed_key, useful);
+  }
+
   bool Lookup(char key, bool useful = true) {
-    return Lookup(std::string(kCacheKeySize, key), useful);
+    return Lookup(TestHashedKey(key), useful);
   }
 
-  void Erase(const std::string& key) { shard_->Erase(key, 0 /*hash*/); }
-
-#if 0  // FIXME
-  size_t CalcEstimatedHandleChargeWrapper(
-      size_t estimated_value_size,
-      CacheMetadataChargePolicy metadata_charge_policy) {
-    return ClockCacheShard::CalcEstimatedHandleCharge(estimated_value_size,
-                                                      metadata_charge_policy);
+  void Erase(char key) {
+    UniqueId64x2 hashed_key = TestHashedKey(key);
+    shard_->Erase(TestKey(hashed_key), hashed_key);
   }
 
-  int CalcHashBitsWrapper(size_t capacity, size_t estimated_value_size,
-                          CacheMetadataChargePolicy metadata_charge_policy) {
-    return ClockCacheShard::CalcHashBits(capacity, estimated_value_size,
-                                         metadata_charge_policy);
+  static inline Slice TestKey(const UniqueId64x2& hashed_key) {
+    return Slice(reinterpret_cast<const char*>(&hashed_key), 16U);
   }
 
-  // Maximum number of items that a shard can hold.
-  double CalcMaxOccupancy(size_t capacity, size_t estimated_value_size,
-                          CacheMetadataChargePolicy metadata_charge_policy) {
-    size_t handle_charge = ClockCacheShard::CalcEstimatedHandleCharge(
-        estimated_value_size, metadata_charge_policy);
-    return capacity / (kLoadFactor * handle_charge);
+  static inline UniqueId64x2 TestHashedKey(char key) {
+    // For testing hash near-collision behavior, put the variance in
+    // hashed_key in bits that are unlikely to be used as hash bits.
+    return {(static_cast<uint64_t>(key) << 56) + 1234U, 5678U};
   }
 
-  bool TableSizeIsAppropriate(int hash_bits, double max_occupancy) {
-    if (hash_bits == 0) {
-      return max_occupancy <= 1;
-    } else {
-      return (1 << hash_bits >= max_occupancy) &&
-             (1 << (hash_bits - 1) <= max_occupancy);
-    }
-  }
-#endif
-
-  ClockCacheShard* shard_ = nullptr;
+  Shard* shard_ = nullptr;
 };
 
 TEST_F(ClockCacheTest, Misc) {
@@ -607,14 +461,15 @@ TEST_F(ClockCacheTest, Misc) {
 
   // Some of this is motivated by code coverage
   std::string wrong_size_key(15, 'x');
-  EXPECT_FALSE(Lookup(wrong_size_key));
+  EXPECT_FALSE(Lookup(wrong_size_key, TestHashedKey('x')));
   EXPECT_FALSE(shard_->Ref(nullptr));
   EXPECT_FALSE(shard_->Release(nullptr));
-  shard_->Erase(wrong_size_key, /*hash*/ 42);  // no-op
+  shard_->Erase(wrong_size_key, TestHashedKey('x'));  // no-op
 }
 
 TEST_F(ClockCacheTest, Limits) {
-  NewShard(3, false /*strict_capacity_limit*/);
+  constexpr size_t kCapacity = 3;
+  NewShard(kCapacity, false /*strict_capacity_limit*/);
   for (bool strict_capacity_limit : {false, true, false}) {
     SCOPED_TRACE("strict_capacity_limit = " +
                  std::to_string(strict_capacity_limit));
@@ -622,11 +477,11 @@ TEST_F(ClockCacheTest, Limits) {
     // Also tests switching between strict limit and not
     shard_->SetStrictCapacityLimit(strict_capacity_limit);
 
-    std::string key(16, 'x');
+    UniqueId64x2 hkey = TestHashedKey('x');
 
     // Single entry charge beyond capacity
     {
-      Status s = shard_->Insert(key, 0 /*hash*/, nullptr /*value*/,
+      Status s = shard_->Insert(TestKey(hkey), hkey, nullptr /*value*/,
                                 5 /*charge*/, nullptr /*deleter*/,
                                 nullptr /*handle*/, Cache::Priority::LOW);
       if (strict_capacity_limit) {
@@ -638,9 +493,10 @@ TEST_F(ClockCacheTest, Limits) {
 
     // Single entry fills capacity
     {
-      Cache::Handle* h;
-      ASSERT_OK(shard_->Insert(key, 0 /*hash*/, nullptr /*value*/, 3 /*charge*/,
-                               nullptr /*deleter*/, &h, Cache::Priority::LOW));
+      HandleImpl* h;
+      ASSERT_OK(shard_->Insert(TestKey(hkey), hkey, nullptr /*value*/,
+                               3 /*charge*/, nullptr /*deleter*/, &h,
+                               Cache::Priority::LOW));
       // Try to insert more
       Status s = Insert('a');
       if (strict_capacity_limit) {
@@ -653,15 +509,17 @@ TEST_F(ClockCacheTest, Limits) {
       shard_->Release(h, false /*useful*/, false /*erase_if_last_ref*/);
     }
 
-    // Insert more than table size can handle (cleverly using zero-charge
-    // entries) to exceed occupancy limit.
+    // Insert more than table size can handle to exceed occupancy limit.
+    // (Cleverly using mostly zero-charge entries, but some non-zero to
+    // verify usage tracking on detached entries.)
     {
       size_t n = shard_->GetTableAddressCount() + 1;
-      std::unique_ptr<Cache::Handle* []> ha { new Cache::Handle* [n] {} };
+      std::unique_ptr<HandleImpl* []> ha { new HandleImpl* [n] {} };
       Status s;
       for (size_t i = 0; i < n && s.ok(); ++i) {
-        EncodeFixed64(&key[0], i);
-        s = shard_->Insert(key, 0 /*hash*/, nullptr /*value*/, 0 /*charge*/,
+        hkey[1] = i;
+        s = shard_->Insert(TestKey(hkey), hkey, nullptr /*value*/,
+                           (i + kCapacity < n) ? 0 : 1 /*charge*/,
                            nullptr /*deleter*/, &ha[i], Cache::Priority::LOW);
         if (i == 0) {
           EXPECT_OK(s);
@@ -807,12 +665,11 @@ void IncrementIntDeleter(const Slice& /*key*/, void* value) {
 // Testing calls to CorrectNearOverflow in Release
 TEST_F(ClockCacheTest, ClockCounterOverflowTest) {
   NewShard(6, /*strict_capacity_limit*/ false);
-  Cache::Handle* h;
+  HandleImpl* h;
   int deleted = 0;
-  std::string my_key(kCacheKeySize, 'x');
-  uint32_t my_hash = 42;
-  ASSERT_OK(shard_->Insert(my_key, my_hash, &deleted, 1, IncrementIntDeleter,
-                           &h, Cache::Priority::HIGH));
+  UniqueId64x2 hkey = TestHashedKey('x');
+  ASSERT_OK(shard_->Insert(TestKey(hkey), hkey, &deleted, 1,
+                           IncrementIntDeleter, &h, Cache::Priority::HIGH));
 
   // Some large number outstanding
   shard_->TEST_RefN(h, 123456789);
@@ -822,7 +679,7 @@ TEST_F(ClockCacheTest, ClockCounterOverflowTest) {
     shard_->TEST_ReleaseN(h, 1234567);
   }
   // Mark it invisible (to reach a different CorrectNearOverflow() in Release)
-  shard_->Erase(my_key, my_hash);
+  shard_->Erase(TestKey(hkey), hkey);
   // Simulate many more lookup/ref + release (one-by-one would be too
   // expensive for unit test)
   for (int i = 0; i < 10000; ++i) {
@@ -844,63 +701,65 @@ TEST_F(ClockCacheTest, ClockCounterOverflowTest) {
 TEST_F(ClockCacheTest, CollidingInsertEraseTest) {
   NewShard(6, /*strict_capacity_limit*/ false);
   int deleted = 0;
-  std::string key1(kCacheKeySize, 'x');
-  std::string key2(kCacheKeySize, 'y');
-  std::string key3(kCacheKeySize, 'z');
-  uint32_t my_hash = 42;
-  Cache::Handle* h1;
-  ASSERT_OK(shard_->Insert(key1, my_hash, &deleted, 1, IncrementIntDeleter, &h1,
+  UniqueId64x2 hkey1 = TestHashedKey('x');
+  Slice key1 = TestKey(hkey1);
+  UniqueId64x2 hkey2 = TestHashedKey('y');
+  Slice key2 = TestKey(hkey2);
+  UniqueId64x2 hkey3 = TestHashedKey('z');
+  Slice key3 = TestKey(hkey3);
+  HandleImpl* h1;
+  ASSERT_OK(shard_->Insert(key1, hkey1, &deleted, 1, IncrementIntDeleter, &h1,
                            Cache::Priority::HIGH));
-  Cache::Handle* h2;
-  ASSERT_OK(shard_->Insert(key2, my_hash, &deleted, 1, IncrementIntDeleter, &h2,
+  HandleImpl* h2;
+  ASSERT_OK(shard_->Insert(key2, hkey2, &deleted, 1, IncrementIntDeleter, &h2,
                            Cache::Priority::HIGH));
-  Cache::Handle* h3;
-  ASSERT_OK(shard_->Insert(key3, my_hash, &deleted, 1, IncrementIntDeleter, &h3,
+  HandleImpl* h3;
+  ASSERT_OK(shard_->Insert(key3, hkey3, &deleted, 1, IncrementIntDeleter, &h3,
                            Cache::Priority::HIGH));
 
   // Can repeatedly lookup+release despite the hash collision
-  Cache::Handle* tmp_h;
+  HandleImpl* tmp_h;
   for (bool erase_if_last_ref : {true, false}) {  // but not last ref
-    tmp_h = shard_->Lookup(key1, my_hash);
+    tmp_h = shard_->Lookup(key1, hkey1);
     ASSERT_EQ(h1, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
 
-    tmp_h = shard_->Lookup(key2, my_hash);
+    tmp_h = shard_->Lookup(key2, hkey2);
     ASSERT_EQ(h2, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
 
-    tmp_h = shard_->Lookup(key3, my_hash);
+    tmp_h = shard_->Lookup(key3, hkey3);
     ASSERT_EQ(h3, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
   }
 
   // Make h1 invisible
-  shard_->Erase(key1, my_hash);
+  shard_->Erase(key1, hkey1);
   // Redundant erase
-  shard_->Erase(key1, my_hash);
+  shard_->Erase(key1, hkey1);
 
   // All still alive
   ASSERT_EQ(deleted, 0);
 
   // Invisible to Lookup
-  tmp_h = shard_->Lookup(key1, my_hash);
+  tmp_h = shard_->Lookup(key1, hkey1);
   ASSERT_EQ(nullptr, tmp_h);
 
   // Can still find h2, h3
   for (bool erase_if_last_ref : {true, false}) {  // but not last ref
-    tmp_h = shard_->Lookup(key2, my_hash);
+    tmp_h = shard_->Lookup(key2, hkey2);
     ASSERT_EQ(h2, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
 
-    tmp_h = shard_->Lookup(key3, my_hash);
+    tmp_h = shard_->Lookup(key3, hkey3);
     ASSERT_EQ(h3, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
   }
 
   // Also Insert with invisible entry there
-  ASSERT_OK(shard_->Insert(key1, my_hash, &deleted, 1, IncrementIntDeleter,
+  ASSERT_OK(shard_->Insert(key1, hkey1, &deleted, 1, IncrementIntDeleter,
                            nullptr, Cache::Priority::HIGH));
-  tmp_h = shard_->Lookup(key1, my_hash);
+  tmp_h = shard_->Lookup(key1, hkey1);
   // Found but distinct handle
   ASSERT_NE(nullptr, tmp_h);
   ASSERT_NE(h1, tmp_h);
@@ -918,11 +777,11 @@ TEST_F(ClockCacheTest, CollidingInsertEraseTest) {
 
   // Can still find h2, h3
   for (bool erase_if_last_ref : {true, false}) {  // but not last ref
-    tmp_h = shard_->Lookup(key2, my_hash);
+    tmp_h = shard_->Lookup(key2, hkey2);
     ASSERT_EQ(h2, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
 
-    tmp_h = shard_->Lookup(key3, my_hash);
+    tmp_h = shard_->Lookup(key3, hkey3);
     ASSERT_EQ(h3, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
   }
@@ -934,7 +793,7 @@ TEST_F(ClockCacheTest, CollidingInsertEraseTest) {
   ASSERT_EQ(deleted, 0);
 
   // Can still find it
-  tmp_h = shard_->Lookup(key2, my_hash);
+  tmp_h = shard_->Lookup(key2, hkey2);
   ASSERT_EQ(h2, tmp_h);
 
   // Release last ref on h2, with erase
@@ -942,12 +801,12 @@ TEST_F(ClockCacheTest, CollidingInsertEraseTest) {
 
   // h2 deleted
   ASSERT_EQ(deleted--, 1);
-  tmp_h = shard_->Lookup(key2, my_hash);
+  tmp_h = shard_->Lookup(key2, hkey2);
   ASSERT_EQ(nullptr, tmp_h);
 
   // Can still find h3
   for (bool erase_if_last_ref : {true, false}) {  // but not last ref
-    tmp_h = shard_->Lookup(key3, my_hash);
+    tmp_h = shard_->Lookup(key3, hkey3);
     ASSERT_EQ(h3, tmp_h);
     ASSERT_FALSE(shard_->Release(tmp_h, erase_if_last_ref));
   }
@@ -959,11 +818,11 @@ TEST_F(ClockCacheTest, CollidingInsertEraseTest) {
   ASSERT_EQ(deleted, 0);
 
   // Explicit erase
-  shard_->Erase(key3, my_hash);
+  shard_->Erase(key3, hkey3);
 
   // h3 deleted
   ASSERT_EQ(deleted--, 1);
-  tmp_h = shard_->Lookup(key3, my_hash);
+  tmp_h = shard_->Lookup(key3, hkey3);
   ASSERT_EQ(nullptr, tmp_h);
 }
 
@@ -1007,7 +866,7 @@ TEST_F(ClockCacheTest, TableSizesTest) {
   }
 }
 
-}  // namespace hyper_clock_cache
+}  // namespace clock_cache
 
 class TestSecondaryCache : public SecondaryCache {
  public:
@@ -1273,14 +1132,19 @@ TEST_F(LRUCacheSecondaryCacheTest, BasicTest) {
                        nullptr /* memory_allocator */, kDefaultToAdaptiveMutex,
                        kDontChargeCacheMetadata);
   std::shared_ptr<TestSecondaryCache> secondary_cache =
-      std::make_shared<TestSecondaryCache>(2048);
+      std::make_shared<TestSecondaryCache>(4096);
   opts.secondary_cache = secondary_cache;
   std::shared_ptr<Cache> cache = NewLRUCache(opts);
   std::shared_ptr<Statistics> stats = CreateDBStatistics();
   CacheKey k1 = CacheKey::CreateUniqueForCacheLifetime(cache.get());
   CacheKey k2 = CacheKey::CreateUniqueForCacheLifetime(cache.get());
+  CacheKey k3 = CacheKey::CreateUniqueForCacheLifetime(cache.get());
 
   Random rnd(301);
+  // Start with warming k3
+  std::string str3 = rnd.RandomString(1021);
+  ASSERT_OK(secondary_cache->InsertSaved(k3.AsSlice(), str3));
+
   std::string str1 = rnd.RandomString(1020);
   TestItem* item1 = new TestItem(str1.data(), str1.length());
   ASSERT_OK(cache->Insert(k1.AsSlice(), item1,
@@ -1297,15 +1161,27 @@ TEST_F(LRUCacheSecondaryCacheTest, BasicTest) {
       cache->Lookup(k2.AsSlice(), &LRUCacheSecondaryCacheTest::helper_,
                     test_item_creator, Cache::Priority::LOW, true, stats.get());
   ASSERT_NE(handle, nullptr);
+  ASSERT_EQ(static_cast<TestItem*>(cache->Value(handle))->Size(), str2.size());
   cache->Release(handle);
+
   // This lookup should promote k1 and demote k2
   handle =
       cache->Lookup(k1.AsSlice(), &LRUCacheSecondaryCacheTest::helper_,
                     test_item_creator, Cache::Priority::LOW, true, stats.get());
   ASSERT_NE(handle, nullptr);
+  ASSERT_EQ(static_cast<TestItem*>(cache->Value(handle))->Size(), str1.size());
   cache->Release(handle);
-  ASSERT_EQ(secondary_cache->num_inserts(), 2u);
-  ASSERT_EQ(secondary_cache->num_lookups(), 1u);
+
+  // This lookup should promote k3 and demote k1
+  handle =
+      cache->Lookup(k3.AsSlice(), &LRUCacheSecondaryCacheTest::helper_,
+                    test_item_creator, Cache::Priority::LOW, true, stats.get());
+  ASSERT_NE(handle, nullptr);
+  ASSERT_EQ(static_cast<TestItem*>(cache->Value(handle))->Size(), str3.size());
+  cache->Release(handle);
+
+  ASSERT_EQ(secondary_cache->num_inserts(), 3u);
+  ASSERT_EQ(secondary_cache->num_lookups(), 2u);
   ASSERT_EQ(stats->getTickerCount(SECONDARY_CACHE_HITS),
             secondary_cache->num_lookups());
   PerfContext perf_ctx = *get_perf_context();
@@ -1371,9 +1247,11 @@ TEST_F(LRUCacheSecondaryCacheTest, SaveFailTest) {
   std::string str2 = rnd.RandomString(1020);
   TestItem* item2 = new TestItem(str2.data(), str2.length());
   // k1 should be demoted to NVM
+  ASSERT_EQ(secondary_cache->num_inserts(), 0u);
   ASSERT_OK(cache->Insert(k2.AsSlice(), item2,
                           &LRUCacheSecondaryCacheTest::helper_fail_,
                           str2.length()));
+  ASSERT_EQ(secondary_cache->num_inserts(), 1u);
 
   Cache::Handle* handle;
   handle =
@@ -1884,6 +1762,43 @@ TEST_F(DBSecondaryCacheTest, SecondaryCacheFailureTest) {
   Destroy(options);
 }
 
+TEST_F(DBSecondaryCacheTest, TestSecondaryWithCompressedCache) {
+  if (!Snappy_Supported()) {
+    ROCKSDB_GTEST_SKIP("Compressed cache test requires snappy support");
+    return;
+  }
+  LRUCacheOptions opts(2000 /* capacity */, 0 /* num_shard_bits */,
+                       false /* strict_capacity_limit */,
+                       0.5 /* high_pri_pool_ratio */,
+                       nullptr /* memory_allocator */, kDefaultToAdaptiveMutex,
+                       kDontChargeCacheMetadata);
+  std::shared_ptr<TestSecondaryCache> secondary_cache(
+      new TestSecondaryCache(2048 * 1024));
+  opts.secondary_cache = secondary_cache;
+  std::shared_ptr<Cache> cache = NewLRUCache(opts);
+  BlockBasedTableOptions table_options;
+  table_options.block_cache_compressed = cache;
+  table_options.no_block_cache = true;
+  table_options.block_size = 1234;
+  Options options = GetDefaultOptions();
+  options.compression = kSnappyCompression;
+  options.create_if_missing = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+  Random rnd(301);
+  const int N = 6;
+  for (int i = 0; i < N; i++) {
+    // Partly compressible
+    std::string p_v = rnd.RandomString(507) + std::string(500, ' ');
+    ASSERT_OK(Put(Key(i), p_v));
+  }
+  ASSERT_OK(Flush());
+  for (int i = 0; i < 2 * N; i++) {
+    std::string v = Get(Key(i % N));
+    ASSERT_EQ(1007, v.size());
+  }
+}
+
 TEST_F(LRUCacheSecondaryCacheTest, BasicWaitAllTest) {
   LRUCacheOptions opts(1024 /* capacity */, 2 /* num_shard_bits */,
                        false /* strict_capacity_limit */,
@@ -2703,6 +2618,7 @@ TEST_F(DBSecondaryCacheTest, TestSecondaryCacheOptionTwoDB) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

cache/secondary_cache.cc

@@ -0,0 +1,32 @@
+//  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 {
+
+size_t SliceSize(void* obj) { return static_cast<Slice*>(obj)->size(); }
+
+Status SliceSaveTo(void* from_obj, size_t from_offset, size_t length,
+                   void* out) {
+  const Slice& slice = *static_cast<Slice*>(from_obj);
+  std::memcpy(out, slice.data() + from_offset, length);
+  return Status::OK();
+}
+
+}  // namespace
+
+Status SecondaryCache::InsertSaved(const Slice& key, const Slice& saved) {
+  static Cache::CacheItemHelper helper{
+      &SliceSize, &SliceSaveTo, GetNoopDeleterForRole<CacheEntryRole::kMisc>()};
+  // 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);
@@ -210,7 +75,7 @@ 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;
 }
 
@@ -226,25 +91,10 @@ int GetDefaultCacheShardBits(size_t capacity, size_t min_shard_size) {
   return num_shard_bits;
 }
 
-int ShardedCache::GetNumShardBits() const { return BitsSetToOne(shard_mask_); }
-
-uint32_t ShardedCache::GetNumShards() const { return shard_mask_ + 1; }
-
-size_t ShardedCache::GetOccupancyCount() const {
-  size_t oc = 0;
-  uint32_t num_shards = GetNumShards();
-  for (uint32_t s = 0; s < num_shards; s++) {
-    oc += GetShard(s)->GetOccupancyCount();
-  }
-  return oc;
-}
-size_t ShardedCache::GetTableAddressCount() const {
-  size_t tac = 0;
-  uint32_t num_shards = GetNumShards();
-  for (uint32_t s = 0; s < num_shards; s++) {
-    tac += GetShard(s)->GetTableAddressCount();
-  }
-  return tac;
+int ShardedCacheBase::GetNumShardBits() const {
+  return BitsSetToOne(shard_mask_);
 }
 
+uint32_t ShardedCacheBase::GetNumShards() const { return shard_mask_ + 1; }
+
 }  // namespace ROCKSDB_NAMESPACE

cache/sharded_cache.h

@@ -10,122 +10,309 @@
 #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:
-  explicit CacheShard(CacheMetadataChargePolicy metadata_charge_policy)
+  explicit CacheShardBase(CacheMetadataChargePolicy metadata_charge_policy)
       : metadata_charge_policy_(metadata_charge_policy) {}
-  virtual ~CacheShard() = default;
 
   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;
-  virtual size_t GetOccupancyCount() const = 0;
-  virtual size_t GetTableAddressCount() 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, void* value, size_t charge,
+                DeleterFn deleter, HandleImpl** handle,
+                Cache::Priority priority) = 0;
+  Status Insert(const Slice& key, HashCref hash, void* value,
+                const Cache::CacheItemHelper* helper, size_t charge,
+                HandleImpl** handle, Cache::Priority priority) = 0;
+  HandleImpl* Lookup(const Slice& key, HashCref hash) = 0;
+  HandleImpl* Lookup(const Slice& key, HashCref hash,
+                        const Cache::CacheItemHelper* helper,
+                        const Cache::CreateCallback& create_cb,
+                        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(
+  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 ""; }
+      size_t average_entries_per_lock, size_t* state) = 0;
+  void EraseUnRefEntries() = 0;
+  */
 
  protected:
   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 size_t GetOccupancyCount() const override;
-  virtual size_t GetTableAddressCount() 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;
 
+  uint64_t NewId() override;
+
+  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_;
+
+  // Dynamic configuration parameters, guarded by config_mutex_
+  bool strict_capacity_limit_;
+  size_t capacity_;
+  mutable port::Mutex config_mutex_;
+};
+
+// 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, void* value, size_t charge, DeleterFn deleter,
+                Handle** handle, Priority priority) override {
+    HashVal hash = CacheShard::ComputeHash(key);
+    auto h_out = reinterpret_cast<HandleImpl**>(handle);
+    return GetShard(hash).Insert(key, hash, value, charge, deleter, h_out,
+                                 priority);
+  }
+  Status Insert(const Slice& key, void* value, const CacheItemHelper* helper,
+                size_t charge, Handle** handle = nullptr,
+                Priority priority = Priority::LOW) override {
+    if (!helper) {
+      return Status::InvalidArgument();
+    }
+    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, Statistics* /*stats*/) override {
+    HashVal hash = CacheShard::ComputeHash(key);
+    HandleImpl* result = GetShard(hash).Lookup(key, hash);
+    return reinterpret_cast<Handle*>(result);
+  }
+  Handle* Lookup(const Slice& key, const CacheItemHelper* helper,
+                 const CreateCallback& create_cb, Priority priority, bool wait,
+                 Statistics* stats = nullptr) override {
+    HashVal hash = CacheShard::ComputeHash(key);
+    HandleImpl* result = GetShard(hash).Lookup(key, hash, helper, create_cb,
+                                               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, void* value, size_t charge,
+                               DeleterFn deleter)>& 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 uint32_t Shard(uint32_t hash) { return hash & shard_mask_; }
+  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:
-  const uint32_t shard_mask_;
-  mutable port::Mutex capacity_mutex_;
-  size_t capacity_;
-  bool strict_capacity_limit_;
-  std::atomic<uint64_t> last_id_;
+  CacheShard* const shards_;
+  bool destroy_shards_in_dtor_;
 };
 
 // 512KB is traditional minimum shard size.

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 \

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() {
@@ -58,48 +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_, latest_seq,
-          /* allow_unprepared_value */ true, /* db_iter */ this);
-      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) {
-        assert(memtable_range_tombstone_iter_ != nullptr);
-        if (memtable_range_tombstone_iter_ != nullptr) {
-          SuperVersion* sv = cfd_->GetThreadLocalSuperVersion(db_impl_);
-          auto t = sv->mem->NewRangeTombstoneIterator(
-              read_options_, latest_seq, false /* immutable_memtable */);
-          delete *memtable_range_tombstone_iter_;
-          if (t == nullptr || t->empty()) {
-            *memtable_range_tombstone_iter_ = nullptr;
+        SuperVersion* sv = cfd_->GetThreadLocalSuperVersion(db_impl_);
+        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);
           }
-          cfd_->ReturnThreadLocalSuperVersion(sv);
         }
+        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"
@@ -107,6 +109,8 @@ 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;
 };
 

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_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_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_test.cc

@@ -1018,6 +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_test.cc

@@ -1304,12 +1304,20 @@ TEST_F(BlobSecondaryCacheTest, GetBlobsFromSecondaryCache) {
                                                blob_offsets[1]));
 
       // 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() can only
-      // get a dummy handle.
+      // 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->Lookup(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->Lookup(key1, statistics);
       ASSERT_NE(handle1, nullptr);
-      // handl1 is a dummy handle.
-      ASSERT_EQ(blob_cache->Value(handle1), nullptr);
+      ASSERT_NE(blob_cache->Value(handle1), nullptr);
       blob_cache->Release(handle1);
     }
   }
@@ -1610,6 +1618,7 @@ 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

@@ -779,7 +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);

db/builder.cc

@@ -90,7 +90,7 @@ 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) {
@@ -281,7 +281,8 @@ 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/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);
@@ -2839,53 +2833,57 @@ 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 +3007,7 @@ 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 +3019,8 @@ 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),
@@ -596,8 +609,8 @@ 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 +877,7 @@ 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 +918,7 @@ 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 +1011,8 @@ 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;
 
@@ -1144,7 +1158,8 @@ 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 +1219,7 @@ Compaction* ColumnFamilyData::CompactRange(
   if (result != nullptr) {
     result->SetInputVersion(current_);
   }
+  TEST_SYNC_POINT("ColumnFamilyData::CompactRange:Return");
   return result;
 }
 
@@ -1287,8 +1303,8 @@ 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_);
 }
@@ -1465,8 +1481,8 @@ 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(
@@ -1562,8 +1578,8 @@ 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
@@ -552,7 +549,7 @@ 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 +594,6 @@ 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_;
@@ -656,8 +651,7 @@ 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;

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),
@@ -259,7 +259,7 @@ Compaction::Compaction(
               ? mutable_cf_options()->blob_garbage_collection_age_cutoff
               : _blob_garbage_collection_age_cutoff),
       penultimate_level_(EvaluatePenultimateLevel(
-          immutable_options_, start_level_, output_level_)) {
+          vstorage, immutable_options_, start_level_, output_level_)) {
   MarkFilesBeingCompacted(true);
   if (is_manual_compaction_) {
     compaction_reason_ = CompactionReason::kManualCompaction;
@@ -268,6 +268,14 @@ 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 +322,67 @@ 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;
+    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::kFullRange;
+        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);
+
+  // If there's a case that the penultimate level output range is overlapping
+  // with the existing files, disable the penultimate level output by setting
+  // the range to empty. One example is the range delete could have overlap
+  // boundary with the next file. (which is actually a false overlap)
+  // TODO: Exclude such false overlap, so it won't disable the penultimate
+  //  output.
+  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() &&
+        OverlapPenultimateLevelOutputRange(file->smallest.user_key(),
+                                           file->largest.user_key())) {
+      // basically disable the penultimate range output. which should be rare
+      // or a false overlap caused by range del
+      penultimate_level_smallest_user_key_ = "";
+      penultimate_level_largest_user_key_ = "";
+      penultimate_output_range_type_ = PenultimateOutputRangeType::kDisabled;
+    }
+  }
 }
 
 Compaction::~Compaction() {
@@ -334,6 +400,8 @@ 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 +410,30 @@ 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 {
@@ -731,6 +808,7 @@ uint64_t Compaction::MinInputFileOldestAncesterTime(
 }
 
 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 +823,19 @@ 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_; }
@@ -363,11 +390,18 @@ 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 +418,6 @@ 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 +441,7 @@ 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 +500,11 @@ 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 +525,12 @@ 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);
@@ -164,7 +168,12 @@ 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);
@@ -821,8 +830,8 @@ 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 +858,8 @@ 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()) &&
@@ -873,8 +882,8 @@ void CompactionIterator::NextFromInput() {
       // object to minimize change to the existing flow.
       Status s = 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()) {
@@ -914,8 +923,17 @@ 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 +1097,7 @@ 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 +1105,13 @@ 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 +1173,7 @@ 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 +1219,8 @@ 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 +1235,8 @@ 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();
 
@@ -466,9 +472,12 @@ 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 +97,8 @@ const char* GetCompactionReasonString(CompactionReason compaction_reason) {
       return "ChangeTemperature";
     case CompactionReason::kForcedBlobGC:
       return "ForcedBlobGC";
+    case CompactionReason::kRoundRobinTtl:
+      return "RoundRobinTtl";
     case CompactionReason::kNumOfReasons:
       // fall through
     default:
@@ -105,6 +107,23 @@ 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,
@@ -263,11 +282,15 @@ 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 +316,27 @@ 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);
+      }
     }
   }
 }
@@ -674,11 +714,12 @@ 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(),
@@ -724,8 +765,8 @@ 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) {
@@ -1035,7 +1076,8 @@ 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.
@@ -1213,8 +1255,8 @@ 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
@@ -1237,6 +1279,10 @@ 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.
@@ -1455,7 +1501,7 @@ 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");
@@ -1661,7 +1707,9 @@ 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) {
@@ -1950,7 +1998,7 @@ 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();
@@ -1971,6 +2019,23 @@ 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_`.

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_job_test.cc

@@ -40,7 +40,7 @@ namespace ROCKSDB_NAMESPACE {
 namespace {
 
 void VerifyInitializationOfCompactionJobStats(
-      const CompactionJobStats& compaction_job_stats) {
+    const CompactionJobStats& compaction_job_stats) {
 #if !defined(IOS_CROSS_COMPILE)
   ASSERT_EQ(compaction_job_stats.elapsed_micros, 0U);
 
@@ -228,6 +228,9 @@ class CompactionJobTestBase : public testing::Test {
         test::CreateEnvFromSystem(ConfigOptions(), &base_env, &env_guard_));
     env_ = base_env;
     fs_ = env_->GetFileSystem();
+    // set default for the tests
+    mutable_cf_options_.target_file_size_base = 1024 * 1024;
+    mutable_cf_options_.max_compaction_bytes = 10 * 1024 * 1024;
   }
 
   void SetUp() override {
@@ -415,8 +418,9 @@ class CompactionJobTestBase : public testing::Test {
 
     auto cfd = versions_->GetColumnFamilySet()->GetDefault();
     if (table_type_ == TableTypeForTest::kMockTable) {
-      assert(expected_results.size() == 1);
-      mock_table_factory_->AssertLatestFile(expected_results[0]);
+      ASSERT_EQ(compaction_job_stats_.num_output_files,
+                expected_results.size());
+      mock_table_factory_->AssertLatestFiles(expected_results);
     } else {
       assert(table_type_ == TableTypeForTest::kBlockBasedTable);
     }
@@ -426,7 +430,8 @@ class CompactionJobTestBase : public testing::Test {
     ASSERT_EQ(expected_output_file_num, output_files.size());
 
     if (table_type_ == TableTypeForTest::kMockTable) {
-      assert(output_files.size() == 1);
+      assert(output_files.size() ==
+             static_cast<size_t>(expected_output_file_num));
       const FileMetaData* const output_file = output_files[0];
       ASSERT_EQ(output_file->oldest_blob_file_number,
                 expected_oldest_blob_file_numbers[0]);
@@ -494,8 +499,7 @@ class CompactionJobTestBase : public testing::Test {
 
         // This is how the key will look like once it's written in bottommost
         // file
-        InternalKey bottommost_internal_key(
-            key, 0, kTypeValue);
+        InternalKey bottommost_internal_key(key, 0, kTypeValue);
 
         if (corrupt_id(k)) {
           test::CorruptKeyType(&internal_key);
@@ -615,17 +619,27 @@ class CompactionJobTestBase : public testing::Test {
       CompactionInputFiles compaction_level;
       compaction_level.level = input_levels[i];
       compaction_level.files.insert(compaction_level.files.end(),
-          level_files.begin(), level_files.end());
+                                    level_files.begin(), level_files.end());
       compaction_input_files.push_back(compaction_level);
       num_input_files += level_files.size();
     }
 
+    std::vector<FileMetaData*> grandparents;
+    // it should actually be the next non-empty level
+    const int kGrandparentsLevel = output_level + 1;
+    if (kGrandparentsLevel < cf_options_.num_levels) {
+      grandparents =
+          cfd_->current()->storage_info()->LevelFiles(kGrandparentsLevel);
+    }
+
     Compaction compaction(
         cfd->current()->storage_info(), *cfd->ioptions(),
         *cfd->GetLatestMutableCFOptions(), mutable_db_options_,
-        compaction_input_files, output_level, 1024 * 1024, 10 * 1024 * 1024, 0,
-        kNoCompression, cfd->GetLatestMutableCFOptions()->compression_opts,
-        Temperature::kUnknown, max_subcompactions, {}, true);
+        compaction_input_files, output_level,
+        mutable_cf_options_.target_file_size_base,
+        mutable_cf_options_.max_compaction_bytes, 0, kNoCompression,
+        cfd->GetLatestMutableCFOptions()->compression_opts,
+        Temperature::kUnknown, max_subcompactions, grandparents, true);
     compaction.SetInputVersion(cfd->current());
 
     assert(db_options_.info_log);
@@ -1721,6 +1735,392 @@ TEST_F(CompactionJobTest, ResultSerialization) {
   }
 }
 
+class CompactionJobDynamicFileSizeTest
+    : public CompactionJobTestBase,
+      public ::testing::WithParamInterface<bool> {
+ public:
+  CompactionJobDynamicFileSizeTest()
+      : CompactionJobTestBase(
+            test::PerThreadDBPath("compaction_job_dynamic_file_size_test"),
+            BytewiseComparator(), [](uint64_t /*ts*/) { return ""; },
+            /*test_io_priority=*/false, TableTypeForTest::kMockTable) {}
+};
+
+TEST_P(CompactionJobDynamicFileSizeTest, CutForMaxCompactionBytes) {
+  // dynamic_file_size option should have no impact on cutting for max
+  // compaction bytes.
+  bool enable_dyanmic_file_size = GetParam();
+  cf_options_.level_compaction_dynamic_file_size = enable_dyanmic_file_size;
+
+  NewDB();
+  mutable_cf_options_.target_file_size_base = 80;
+  mutable_cf_options_.max_compaction_bytes = 21;
+
+  auto file1 = mock::MakeMockFile({
+      {KeyStr("c", 5U, kTypeValue), "val2"},
+      {KeyStr("n", 6U, kTypeValue), "val3"},
+  });
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({{KeyStr("h", 3U, kTypeValue), "val"},
+                                   {KeyStr("j", 4U, kTypeValue), "val"}});
+  AddMockFile(file2, 1);
+
+  // Create three L2 files, each size 10.
+  // max_compaction_bytes 21 means the compaction output in L1 will
+  // be cut to at least two files.
+  auto file3 = mock::MakeMockFile({{KeyStr("b", 1U, kTypeValue), "val"},
+                                   {KeyStr("c", 1U, kTypeValue), "val"},
+                                   {KeyStr("c1", 1U, kTypeValue), "val"},
+                                   {KeyStr("c2", 1U, kTypeValue), "val"},
+                                   {KeyStr("c3", 1U, kTypeValue), "val"},
+                                   {KeyStr("c4", 1U, kTypeValue), "val"},
+                                   {KeyStr("d", 1U, kTypeValue), "val"},
+                                   {KeyStr("e", 2U, kTypeValue), "val"}});
+  AddMockFile(file3, 2);
+
+  auto file4 = mock::MakeMockFile({{KeyStr("h", 1U, kTypeValue), "val"},
+                                   {KeyStr("i", 1U, kTypeValue), "val"},
+                                   {KeyStr("i1", 1U, kTypeValue), "val"},
+                                   {KeyStr("i2", 1U, kTypeValue), "val"},
+                                   {KeyStr("i3", 1U, kTypeValue), "val"},
+                                   {KeyStr("i4", 1U, kTypeValue), "val"},
+                                   {KeyStr("j", 1U, kTypeValue), "val"},
+                                   {KeyStr("k", 2U, kTypeValue), "val"}});
+  AddMockFile(file4, 2);
+
+  auto file5 = mock::MakeMockFile({{KeyStr("l", 1U, kTypeValue), "val"},
+                                   {KeyStr("m", 1U, kTypeValue), "val"},
+                                   {KeyStr("m1", 1U, kTypeValue), "val"},
+                                   {KeyStr("m2", 1U, kTypeValue), "val"},
+                                   {KeyStr("m3", 1U, kTypeValue), "val"},
+                                   {KeyStr("m4", 1U, kTypeValue), "val"},
+                                   {KeyStr("n", 1U, kTypeValue), "val"},
+                                   {KeyStr("o", 2U, kTypeValue), "val"}});
+  AddMockFile(file5, 2);
+
+  // The expected output should be:
+  //  L1:   [c,   h,  j]        [n]
+  //  L2: [b ... e] [h ... k] [l ... o]
+  // It's better to have "j" in the first file, because anyway it's overlapping
+  // with the second file on L2.
+  // (Note: before this PR, it was cut at "h" because it's using the internal
+  // comparator which think L1 "h" with seqno 3 is smaller than L2 "h" with
+  // seqno 1, but actually they're overlapped with the compaction picker).
+
+  auto expected_file1 =
+      mock::MakeMockFile({{KeyStr("c", 5U, kTypeValue), "val2"},
+                          {KeyStr("h", 3U, kTypeValue), "val"},
+                          {KeyStr("j", 4U, kTypeValue), "val"}});
+  auto expected_file2 =
+      mock::MakeMockFile({{KeyStr("n", 6U, kTypeValue), "val3"}});
+
+  SetLastSequence(6U);
+
+  const std::vector<int> input_levels = {0, 1};
+  auto lvl0_files = cfd_->current()->storage_info()->LevelFiles(0);
+  auto lvl1_files = cfd_->current()->storage_info()->LevelFiles(1);
+
+  RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                {expected_file1, expected_file2});
+}
+
+TEST_P(CompactionJobDynamicFileSizeTest, CutToSkipGrandparentFile) {
+  bool enable_dyanmic_file_size = GetParam();
+  cf_options_.level_compaction_dynamic_file_size = enable_dyanmic_file_size;
+
+  NewDB();
+  // Make sure the grandparent level file size (10) qualifies skipping.
+  // Currently, it has to be > 1/8 of target file size.
+  mutable_cf_options_.target_file_size_base = 70;
+
+  auto file1 = mock::MakeMockFile({
+      {KeyStr("a", 5U, kTypeValue), "val2"},
+      {KeyStr("z", 6U, kTypeValue), "val3"},
+  });
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({{KeyStr("c", 3U, kTypeValue), "val"},
+                                   {KeyStr("x", 4U, kTypeValue), "val"}});
+  AddMockFile(file2, 1);
+
+  auto file3 = mock::MakeMockFile({{KeyStr("b", 1U, kTypeValue), "val"},
+                                   {KeyStr("d", 2U, kTypeValue), "val"}});
+  AddMockFile(file3, 2);
+
+  auto file4 = mock::MakeMockFile({{KeyStr("h", 1U, kTypeValue), "val"},
+                                   {KeyStr("i", 2U, kTypeValue), "val"}});
+  AddMockFile(file4, 2);
+
+  auto file5 = mock::MakeMockFile({{KeyStr("v", 1U, kTypeValue), "val"},
+                                   {KeyStr("y", 2U, kTypeValue), "val"}});
+  AddMockFile(file5, 2);
+
+  auto expected_file1 =
+      mock::MakeMockFile({{KeyStr("a", 5U, kTypeValue), "val2"},
+                          {KeyStr("c", 3U, kTypeValue), "val"}});
+  auto expected_file2 =
+      mock::MakeMockFile({{KeyStr("x", 4U, kTypeValue), "val"},
+                          {KeyStr("z", 6U, kTypeValue), "val3"}});
+
+  auto expected_file_disable_dynamic_file_size =
+      mock::MakeMockFile({{KeyStr("a", 5U, kTypeValue), "val2"},
+                          {KeyStr("c", 3U, kTypeValue), "val"},
+                          {KeyStr("x", 4U, kTypeValue), "val"},
+                          {KeyStr("z", 6U, kTypeValue), "val3"}});
+
+  SetLastSequence(6U);
+  const std::vector<int> input_levels = {0, 1};
+  auto lvl0_files = cfd_->current()->storage_info()->LevelFiles(0);
+  auto lvl1_files = cfd_->current()->storage_info()->LevelFiles(1);
+  if (enable_dyanmic_file_size) {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file1, expected_file2});
+  } else {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file_disable_dynamic_file_size});
+  }
+}
+
+TEST_P(CompactionJobDynamicFileSizeTest, CutToAlignGrandparentBoundary) {
+  bool enable_dyanmic_file_size = GetParam();
+  cf_options_.level_compaction_dynamic_file_size = enable_dyanmic_file_size;
+  NewDB();
+
+  // MockTable has 1 byte per entry by default and each file is 10 bytes.
+  // When the file size is smaller than 100, it won't cut file earlier to align
+  // with its grandparent boundary.
+  const size_t kKeyValueSize = 10000;
+  mock_table_factory_->SetKeyValueSize(kKeyValueSize);
+
+  mutable_cf_options_.target_file_size_base = 10 * kKeyValueSize;
+
+  mock::KVVector file1;
+  char ch = 'd';
+  // Add value from d -> o
+  for (char i = 0; i < 12; i++) {
+    file1.emplace_back(KeyStr(std::string(1, ch + i), i + 10, kTypeValue),
+                       "val" + std::to_string(i));
+  }
+
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({{KeyStr("e", 3U, kTypeValue), "val"},
+                                   {KeyStr("s", 4U, kTypeValue), "val"}});
+  AddMockFile(file2, 1);
+
+  // the 1st grandparent file should be skipped
+  auto file3 = mock::MakeMockFile({{KeyStr("a", 1U, kTypeValue), "val"},
+                                   {KeyStr("b", 2U, kTypeValue), "val"}});
+  AddMockFile(file3, 2);
+
+  auto file4 = mock::MakeMockFile({{KeyStr("c", 1U, kTypeValue), "val"},
+                                   {KeyStr("e", 2U, kTypeValue), "val"}});
+  AddMockFile(file4, 2);
+
+  auto file5 = mock::MakeMockFile({{KeyStr("h", 1U, kTypeValue), "val"},
+                                   {KeyStr("j", 2U, kTypeValue), "val"}});
+  AddMockFile(file5, 2);
+
+  auto file6 = mock::MakeMockFile({{KeyStr("k", 1U, kTypeValue), "val"},
+                                   {KeyStr("n", 2U, kTypeValue), "val"}});
+  AddMockFile(file6, 2);
+
+  auto file7 = mock::MakeMockFile({{KeyStr("q", 1U, kTypeValue), "val"},
+                                   {KeyStr("t", 2U, kTypeValue), "val"}});
+  AddMockFile(file7, 2);
+
+  // The expected outputs are:
+  //  L1:         [d,e,f,g,h,i,j] [k,l,m,n,o,s]
+  //  L2: [a, b] [c,  e]   [h, j] [k, n]  [q, t]
+  // The first output cut earlier at "j", so it could be aligned with L2 files.
+  // If dynamic_file_size is not enabled, it will be cut based on the
+  // target_file_size
+  mock::KVVector expected_file1;
+  for (char i = 0; i < 7; i++) {
+    expected_file1.emplace_back(
+        KeyStr(std::string(1, ch + i), i + 10, kTypeValue),
+        "val" + std::to_string(i));
+  }
+
+  mock::KVVector expected_file2;
+  for (char i = 7; i < 12; i++) {
+    expected_file2.emplace_back(
+        KeyStr(std::string(1, ch + i), i + 10, kTypeValue),
+        "val" + std::to_string(i));
+  }
+  expected_file2.emplace_back(KeyStr("s", 4U, kTypeValue), "val");
+
+  mock::KVVector expected_file_disable_dynamic_file_size1;
+  for (char i = 0; i < 10; i++) {
+    expected_file_disable_dynamic_file_size1.emplace_back(
+        KeyStr(std::string(1, ch + i), i + 10, kTypeValue),
+        "val" + std::to_string(i));
+  }
+
+  mock::KVVector expected_file_disable_dynamic_file_size2;
+  for (char i = 10; i < 12; i++) {
+    expected_file_disable_dynamic_file_size2.emplace_back(
+        KeyStr(std::string(1, ch + i), i + 10, kTypeValue),
+        "val" + std::to_string(i));
+  }
+
+  expected_file_disable_dynamic_file_size2.emplace_back(
+      KeyStr("s", 4U, kTypeValue), "val");
+
+  SetLastSequence(22U);
+  const std::vector<int> input_levels = {0, 1};
+  auto lvl0_files = cfd_->current()->storage_info()->LevelFiles(0);
+  auto lvl1_files = cfd_->current()->storage_info()->LevelFiles(1);
+  if (enable_dyanmic_file_size) {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file1, expected_file2});
+  } else {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file_disable_dynamic_file_size1,
+                   expected_file_disable_dynamic_file_size2});
+  }
+}
+
+TEST_P(CompactionJobDynamicFileSizeTest, CutToAlignGrandparentBoundarySameKey) {
+  bool enable_dyanmic_file_size = GetParam();
+  cf_options_.level_compaction_dynamic_file_size = enable_dyanmic_file_size;
+  NewDB();
+
+  // MockTable has 1 byte per entry by default and each file is 10 bytes.
+  // When the file size is smaller than 100, it won't cut file earlier to align
+  // with its grandparent boundary.
+  const size_t kKeyValueSize = 10000;
+  mock_table_factory_->SetKeyValueSize(kKeyValueSize);
+
+  mutable_cf_options_.target_file_size_base = 10 * kKeyValueSize;
+
+  mock::KVVector file1;
+  for (int i = 0; i < 7; i++) {
+    file1.emplace_back(KeyStr("a", 100 - i, kTypeValue),
+                       "val" + std::to_string(100 - i));
+  }
+  file1.emplace_back(KeyStr("b", 90, kTypeValue), "valb");
+
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({{KeyStr("a", 93U, kTypeValue), "val93"},
+                                   {KeyStr("b", 90U, kTypeValue), "valb"}});
+  AddMockFile(file2, 1);
+
+  auto file3 = mock::MakeMockFile({{KeyStr("a", 89U, kTypeValue), "val"},
+                                   {KeyStr("a", 88U, kTypeValue), "val"}});
+  AddMockFile(file3, 2);
+
+  auto file4 = mock::MakeMockFile({{KeyStr("a", 87U, kTypeValue), "val"},
+                                   {KeyStr("a", 86U, kTypeValue), "val"}});
+  AddMockFile(file4, 2);
+
+  auto file5 = mock::MakeMockFile({{KeyStr("b", 85U, kTypeValue), "val"},
+                                   {KeyStr("b", 84U, kTypeValue), "val"}});
+  AddMockFile(file5, 2);
+
+  mock::KVVector expected_file1;
+  mock::KVVector expected_file_disable_dynamic_file_size;
+
+  for (int i = 0; i < 8; i++) {
+    expected_file1.emplace_back(KeyStr("a", 100 - i, kTypeValue),
+                                "val" + std::to_string(100 - i));
+    expected_file_disable_dynamic_file_size.emplace_back(
+        KeyStr("a", 100 - i, kTypeValue), "val" + std::to_string(100 - i));
+  }
+
+  // make sure `b` is cut in a separated file (so internally it's not using
+  // internal comparator, which will think the "b:90" (seqno 90) here is smaller
+  // than "b:85" on L2.)
+  auto expected_file2 =
+      mock::MakeMockFile({{KeyStr("b", 90U, kTypeValue), "valb"}});
+
+  expected_file_disable_dynamic_file_size.emplace_back(
+      KeyStr("b", 90U, kTypeValue), "valb");
+
+  SetLastSequence(122U);
+  const std::vector<int> input_levels = {0, 1};
+  auto lvl0_files = cfd_->current()->storage_info()->LevelFiles(0);
+  auto lvl1_files = cfd_->current()->storage_info()->LevelFiles(1);
+
+  // Just keep all the history
+  std::vector<SequenceNumber> snapshots;
+  for (int i = 80; i <= 100; i++) {
+    snapshots.emplace_back(i);
+  }
+  if (enable_dyanmic_file_size) {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file1, expected_file2}, snapshots);
+  } else {
+    RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                  {expected_file_disable_dynamic_file_size}, snapshots);
+  }
+}
+
+TEST_P(CompactionJobDynamicFileSizeTest, CutForMaxCompactionBytesSameKey) {
+  // dynamic_file_size option should have no impact on cutting for max
+  // compaction bytes.
+  bool enable_dyanmic_file_size = GetParam();
+  cf_options_.level_compaction_dynamic_file_size = enable_dyanmic_file_size;
+
+  NewDB();
+  mutable_cf_options_.target_file_size_base = 80;
+  mutable_cf_options_.max_compaction_bytes = 20;
+
+  auto file1 = mock::MakeMockFile({{KeyStr("a", 104U, kTypeValue), "val1"},
+                                   {KeyStr("b", 103U, kTypeValue), "val"}});
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({{KeyStr("a", 102U, kTypeValue), "val2"},
+                                   {KeyStr("c", 101U, kTypeValue), "val"}});
+  AddMockFile(file2, 1);
+
+  for (int i = 0; i < 10; i++) {
+    auto file =
+        mock::MakeMockFile({{KeyStr("a", 100 - (i * 2), kTypeValue), "val"},
+                            {KeyStr("a", 99 - (i * 2), kTypeValue), "val"}});
+    AddMockFile(file, 2);
+  }
+
+  for (int i = 0; i < 10; i++) {
+    auto file =
+        mock::MakeMockFile({{KeyStr("b", 80 - (i * 2), kTypeValue), "val"},
+                            {KeyStr("b", 79 - (i * 2), kTypeValue), "val"}});
+    AddMockFile(file, 2);
+  }
+
+  auto file5 = mock::MakeMockFile({{KeyStr("c", 60U, kTypeValue), "valc"},
+                                   {KeyStr("c", 59U, kTypeValue), "valc"}});
+
+  // "a" has 10 overlapped grandparent files (each size 10), which is far
+  // exceeded the `max_compaction_bytes`, but make sure 2 "a" are not separated,
+  // as splitting them won't help reducing the compaction size.
+  // also make sure "b" and "c" are cut separately.
+  mock::KVVector expected_file1 =
+      mock::MakeMockFile({{KeyStr("a", 104U, kTypeValue), "val1"},
+                          {KeyStr("a", 102U, kTypeValue), "val2"}});
+  mock::KVVector expected_file2 =
+      mock::MakeMockFile({{KeyStr("b", 103U, kTypeValue), "val"}});
+  mock::KVVector expected_file3 =
+      mock::MakeMockFile({{KeyStr("c", 101U, kTypeValue), "val"}});
+
+  SetLastSequence(122U);
+  const std::vector<int> input_levels = {0, 1};
+  auto lvl0_files = cfd_->current()->storage_info()->LevelFiles(0);
+  auto lvl1_files = cfd_->current()->storage_info()->LevelFiles(1);
+
+  // Just keep all the history
+  std::vector<SequenceNumber> snapshots;
+  for (int i = 80; i <= 105; i++) {
+    snapshots.emplace_back(i);
+  }
+  RunCompaction({lvl0_files, lvl1_files}, input_levels,
+                {expected_file1, expected_file2, expected_file3}, snapshots);
+}
+
+INSTANTIATE_TEST_CASE_P(CompactionJobDynamicFileSizeTest,
+                        CompactionJobDynamicFileSizeTest, testing::Bool());
 
 class CompactionJobTimestampTest : public CompactionJobTestBase {
  public:
@@ -2033,6 +2433,7 @@ TEST_F(CompactionJobIOPriorityTest, GetRateLimiterPriority) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   RegisterCustomObjects(argc, argv);
   return RUN_ALL_TESTS();

db/compaction/compaction_outputs.cc

@@ -76,14 +76,131 @@ IOStatus CompactionOutputs::WriterSyncClose(const Status& input_status,
   return io_s;
 }
 
+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());
 
+  // always update grandparent information like overlapped file number, size
+  // etc.
+  const Slice& internal_key = c_iter.key();
+  const uint64_t previous_overlapped_bytes = grandparent_overlapped_bytes_;
+  size_t num_grandparent_boundaries_crossed =
+      UpdateGrandparentBoundaryInfo(internal_key);
+
+  if (!HasBuilder()) {
+    return false;
+  }
+
   // If there's user defined partitioner, check that first
-  if (HasBuilder() && partitioner_ &&
-      partitioner_->ShouldPartition(
-          PartitionerRequest(last_key_for_partitioner_, c_iter.user_key(),
-                             current_output_file_size_)) == kRequired) {
+  if (partitioner_ && partitioner_->ShouldPartition(PartitionerRequest(
+                          last_key_for_partitioner_, c_iter.user_key(),
+                          current_output_file_size_)) == kRequired) {
     return true;
   }
 
@@ -92,12 +209,11 @@ bool CompactionOutputs::ShouldStopBefore(const CompactionIterator& c_iter) {
     return false;
   }
 
-  // reach the target file size
+  // reach the max file size
   if (current_output_file_size_ >= compaction_->max_output_file_size()) {
     return true;
   }
 
-  const Slice& internal_key = c_iter.key();
   const InternalKeyComparator* icmp =
       &compaction_->column_family_data()->internal_comparator();
 
@@ -111,32 +227,67 @@ bool CompactionOutputs::ShouldStopBefore(const CompactionIterator& c_iter) {
     }
   }
 
-  // Update grandparent information
-  const std::vector<FileMetaData*>& grandparents = compaction_->grandparents();
-  bool grandparant_file_switched = false;
-  // Scan to find the earliest grandparent file that contains key.
-  while (grandparent_index_ < grandparents.size() &&
-         icmp->Compare(internal_key,
-                       grandparents[grandparent_index_]->largest.Encode()) >
-             0) {
-    if (seen_key_) {
-      overlapped_bytes_ += grandparents[grandparent_index_]->fd.GetFileSize();
-      grandparant_file_switched = 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;
     }
-    assert(grandparent_index_ + 1 >= grandparents.size() ||
-           icmp->Compare(
-               grandparents[grandparent_index_]->largest.Encode(),
-               grandparents[grandparent_index_ + 1]->smallest.Encode()) <= 0);
-    grandparent_index_++;
-  }
-  seen_key_ = true;
 
-  if (grandparant_file_switched &&
-      overlapped_bytes_ + current_output_file_size_ >
-          compaction_->max_compaction_bytes()) {
-    // Too much overlap for current output; start new output
-    overlapped_bytes_ = 0;
-    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;
+    }
   }
 
   // check ttl file boundaries if there's any
@@ -189,6 +340,10 @@ Status CompactionOutputs::AddToOutput(
     if (!s.ok()) {
       return s;
     }
+    // reset grandparent information
+    grandparent_boundary_switched_num_ = 0;
+    grandparent_overlapped_bytes_ =
+        GetCurrentKeyGrandparentOverlappedBytes(key);
   }
 
   // Open output file if necessary
@@ -199,10 +354,9 @@ Status CompactionOutputs::AddToOutput(
     }
   }
 
-  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;
   }
@@ -232,10 +386,10 @@ 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();
@@ -249,7 +403,7 @@ Status CompactionOutputs::AddRangeDels(
   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
@@ -269,21 +423,22 @@ 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;
   }
@@ -292,8 +447,8 @@ 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
@@ -307,7 +462,8 @@ Status CompactionOutputs::AddRangeDels(
   for (; it->Valid(); it->Next()) {
     auto tombstone = it->Tombstone();
     if (upper_bound != nullptr) {
-      int cmp = ucmp->Compare(*upper_bound, tombstone.start_key_);
+      int cmp =
+          ucmp->CompareWithoutTimestamp(*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
@@ -320,7 +476,17 @@ Status CompactionOutputs::AddRangeDels(
       }
     }
 
-    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.
@@ -331,12 +497,13 @@ 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);
     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.
@@ -356,13 +523,23 @@ Status CompactionOutputs::AddRangeDels(
       // 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) {
+        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 {
+        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+      }
     }
     InternalKey largest_candidate = tombstone.SerializeEndKey();
     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.
@@ -376,9 +553,22 @@ 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;

db/compaction/compaction_outputs.h

@@ -168,11 +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);
+                      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 {
@@ -216,9 +221,16 @@ class CompactionOutputs {
     }
   }
 
-  uint64_t GetCurrentOutputFileSize() const {
-    return current_output_file_size_;
-  }
+  // 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.
@@ -311,12 +323,21 @@ class CompactionOutputs {
   // 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 overlapped_bytes_ = 0;
+  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

@@ -27,16 +27,6 @@
 
 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,
@@ -63,8 +53,6 @@ bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
     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 +61,10 @@ 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;
@@ -307,13 +294,12 @@ 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;
     }
   }
@@ -322,17 +308,24 @@ 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;
+      }
     }
   }
 
@@ -356,11 +349,25 @@ 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) {
@@ -507,8 +514,8 @@ 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;
@@ -527,13 +534,13 @@ 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);
@@ -554,9 +561,10 @@ 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;
       }
@@ -657,7 +665,10 @@ 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;
@@ -724,18 +735,18 @@ 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;
@@ -836,7 +847,10 @@ 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;
@@ -1121,7 +1135,8 @@ void CompactionPicker::RegisterCompaction(Compaction* c) {
   }
   assert(ioptions_.compaction_style != kCompactionStyleLevel ||
          c->output_level() == 0 ||
-         !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level()));
+         !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level(),
+                                          c->GetPenultimateLevel()));
   if (c->start_level() == 0 ||
       ioptions_.compaction_style == kCompactionStyleUniversal) {
     level0_compactions_in_progress_.insert(c);

db/compaction/compaction_picker.h

@@ -182,7 +182,8 @@ 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,

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 +212,58 @@ 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 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",
+                       cf_name.c_str(), f->fd.GetNumber(), tmp_fsize);
+      if (total_size <=
+          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+        break;
+      }
     }
   }
 
   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 +280,13 @@ 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);
 
@@ -327,8 +390,6 @@ 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);
-
   Compaction* c = nullptr;
   if (mutable_cf_options.ttl > 0) {
     c = PickTTLCompaction(cf_name, mutable_cf_options, mutable_db_options,

db/compaction/compaction_picker_level.cc

@@ -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
@@ -434,7 +455,9 @@ bool LevelCompactionBuilder::SetupOtherInputsIfNeeded() {
       // (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_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;
@@ -736,7 +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();

db/compaction/compaction_picker_test.cc

@@ -1340,6 +1340,7 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping4) {
   ioptions_.compaction_pri = kMinOverlappingRatio;
   mutable_cf_options_.max_bytes_for_level_base = 10000000;
   mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  mutable_cf_options_.ignore_max_compaction_bytes_for_input = false;
 
   // file 7 and 8 over lap with the same file, but file 8 is smaller so
   // it will be picked.
@@ -2358,6 +2359,7 @@ TEST_F(CompactionPickerTest, IsBottommostLevelTest) {
 TEST_F(CompactionPickerTest, MaxCompactionBytesHit) {
   mutable_cf_options_.max_bytes_for_level_base = 1000000u;
   mutable_cf_options_.max_compaction_bytes = 800000u;
+  mutable_cf_options_.ignore_max_compaction_bytes_for_input = false;
   ioptions_.level_compaction_dynamic_level_bytes = false;
   NewVersionStorage(6, kCompactionStyleLevel);
   // A compaction should be triggered and pick file 2 and 5.
@@ -2384,6 +2386,7 @@ TEST_F(CompactionPickerTest, MaxCompactionBytesHit) {
 TEST_F(CompactionPickerTest, MaxCompactionBytesNotHit) {
   mutable_cf_options_.max_bytes_for_level_base = 800000u;
   mutable_cf_options_.max_compaction_bytes = 1000000u;
+  mutable_cf_options_.ignore_max_compaction_bytes_for_input = false;
   ioptions_.level_compaction_dynamic_level_bytes = false;
   NewVersionStorage(6, kCompactionStyleLevel);
   // A compaction should be triggered and pick file 2 and 5.
@@ -3146,7 +3149,7 @@ TEST_F(CompactionPickerTest, UniversalMarkedL0Overlap2) {
   // should fail
   NewVersionStorage(1, kCompactionStyleUniversal);
 
-  // Mark file number 4 for compaction
+  // Mark file number 5 for compaction
   Add(0, 4U, "260", "300", 1 * kFileSize, 0, 260, 300);
   Add(0, 5U, "240", "290", 2 * kFileSize, 0, 201, 250, 0, true);
   Add(0, 3U, "301", "350", 4 * kFileSize, 0, 101, 150);
@@ -3521,9 +3524,11 @@ TEST_P(PerKeyPlacementCompactionPickerTest, OverlapWithNormalCompaction) {
   ASSERT_OK(level_compaction_picker.GetCompactionInputsFromFileNumbers(
       &input_files, &input_set, vstorage_.get(), comp_options));
 
-  ASSERT_EQ(
-      enable_per_key_placement_,
-      level_compaction_picker.FilesRangeOverlapWithCompaction(input_files, 6));
+  ASSERT_EQ(enable_per_key_placement_,
+            level_compaction_picker.FilesRangeOverlapWithCompaction(
+                input_files, 6,
+                Compaction::EvaluatePenultimateLevel(vstorage_.get(), ioptions_,
+                                                     0, 6)));
 }
 
 TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlap) {
@@ -3564,9 +3569,9 @@ TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlap) {
   ASSERT_OK(level_compaction_picker.GetCompactionInputsFromFileNumbers(
       &input_files, &input_set, vstorage_.get(), comp_options));
 
-  ASSERT_EQ(
-      enable_per_key_placement_,
-      level_compaction_picker.FilesRangeOverlapWithCompaction(input_files, 5));
+  ASSERT_EQ(enable_per_key_placement_,
+            level_compaction_picker.FilesRangeOverlapWithCompaction(
+                input_files, 5, Compaction::kInvalidLevel));
 }
 
 TEST_P(PerKeyPlacementCompactionPickerTest,
@@ -3609,7 +3614,9 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
 
   ASSERT_EQ(enable_per_key_placement_,
             universal_compaction_picker.FilesRangeOverlapWithCompaction(
-                input_files, 6));
+                input_files, 6,
+                Compaction::EvaluatePenultimateLevel(vstorage_.get(), ioptions_,
+                                                     0, 6)));
 }
 
 TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlapUniversal) {
@@ -3653,7 +3660,294 @@ TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlapUniversal) {
 
   ASSERT_EQ(enable_per_key_placement_,
             universal_compaction_picker.FilesRangeOverlapWithCompaction(
-                input_files, 5));
+                input_files, 5, Compaction::kInvalidLevel));
+}
+
+TEST_P(PerKeyPlacementCompactionPickerTest, PenultimateOverlapUniversal) {
+  // This test is make sure the Tiered compaction would lock whole range of
+  // both output level and penultimate level
+  if (enable_per_key_placement_) {
+    ioptions_.preclude_last_level_data_seconds = 10000;
+  }
+
+  int num_levels = ioptions_.num_levels;
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  NewVersionStorage(num_levels, kCompactionStyleUniversal);
+
+  // L4:   [200, 220]  [230, 250]       [360, 380]
+  // L5:
+  // L6: [101,                    351]
+  Add(4, 40U, "200", "220", 60000000U);
+  Add(4, 41U, "230", "250", 60000000U);
+  Add(4, 42U, "360", "380", 60000000U);
+  Add(6, 60U, "101", "351", 60000000U);
+  UpdateVersionStorageInfo();
+
+  // the existing compaction is the 1st L4 file + L6 file
+  // then compaction of the 2nd L4 file to L5 (penultimate level) is overlapped
+  // when the tiered compaction feature is on.
+  CompactionOptions comp_options;
+  std::unordered_set<uint64_t> input_set;
+  input_set.insert(40);
+  input_set.insert(60);
+  std::vector<CompactionInputFiles> input_files;
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(41);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_EQ(enable_per_key_placement_,
+            universal_compaction_picker.FilesRangeOverlapWithCompaction(
+                input_files, 5, Compaction::kInvalidLevel));
+
+  // compacting the 3rd L4 file is always safe:
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(42);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_FALSE(universal_compaction_picker.FilesRangeOverlapWithCompaction(
+      input_files, 5, Compaction::kInvalidLevel));
+}
+
+TEST_P(PerKeyPlacementCompactionPickerTest, LastLevelOnlyOverlapUniversal) {
+  if (enable_per_key_placement_) {
+    ioptions_.preclude_last_level_data_seconds = 10000;
+  }
+
+  int num_levels = ioptions_.num_levels;
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  NewVersionStorage(num_levels, kCompactionStyleUniversal);
+
+  // L4:   [200, 220]  [230, 250]       [360, 380]
+  // L5:
+  // L6: [101,                    351]
+  Add(4, 40U, "200", "220", 60000000U);
+  Add(4, 41U, "230", "250", 60000000U);
+  Add(4, 42U, "360", "380", 60000000U);
+  Add(6, 60U, "101", "351", 60000000U);
+  UpdateVersionStorageInfo();
+
+  CompactionOptions comp_options;
+  std::unordered_set<uint64_t> input_set;
+  input_set.insert(60);
+  std::vector<CompactionInputFiles> input_files;
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+
+  // cannot compact file 41 if the preclude_last_level feature is on, otherwise
+  // compact file 41 is okay.
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(41);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_EQ(enable_per_key_placement_,
+            universal_compaction_picker.FilesRangeOverlapWithCompaction(
+                input_files, 5, Compaction::kInvalidLevel));
+
+  // compacting the 3rd L4 file is always safe:
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(42);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_FALSE(universal_compaction_picker.FilesRangeOverlapWithCompaction(
+      input_files, 5, Compaction::kInvalidLevel));
+}
+
+TEST_P(PerKeyPlacementCompactionPickerTest,
+       LastLevelOnlyFailPenultimateUniversal) {
+  // This is to test last_level only compaction still unable to do the
+  // penultimate level compaction if there's already a file in the penultimate
+  // level.
+  // This should rarely happen in universal compaction, as the non-empty L5
+  // should be included in the compaction.
+  if (enable_per_key_placement_) {
+    ioptions_.preclude_last_level_data_seconds = 10000;
+  }
+
+  int num_levels = ioptions_.num_levels;
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  NewVersionStorage(num_levels, kCompactionStyleUniversal);
+
+  // L4:   [200, 220]
+  // L5:              [230, 250]
+  // L6: [101,                    351]
+  Add(4, 40U, "200", "220", 60000000U);
+  Add(5, 50U, "230", "250", 60000000U);
+  Add(6, 60U, "101", "351", 60000000U);
+  UpdateVersionStorageInfo();
+
+  CompactionOptions comp_options;
+  std::unordered_set<uint64_t> input_set;
+  input_set.insert(60);
+  std::vector<CompactionInputFiles> input_files;
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+
+  ASSERT_TRUE(comp1);
+  ASSERT_EQ(comp1->GetPenultimateLevel(), Compaction::kInvalidLevel);
+
+  // As comp1 cannot be output to the penultimate level, compacting file 40 to
+  // L5 is always safe.
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(40);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_FALSE(universal_compaction_picker.FilesRangeOverlapWithCompaction(
+      input_files, 5, Compaction::kInvalidLevel));
+
+  std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+  ASSERT_TRUE(comp2);
+  ASSERT_EQ(Compaction::kInvalidLevel, comp2->GetPenultimateLevel());
+}
+
+TEST_P(PerKeyPlacementCompactionPickerTest,
+       LastLevelOnlyConflictWithOngoingUniversal) {
+  // This is to test last_level only compaction still unable to do the
+  // penultimate level compaction if there's already an ongoing compaction to
+  // the penultimate level
+  if (enable_per_key_placement_) {
+    ioptions_.preclude_last_level_data_seconds = 10000;
+  }
+
+  int num_levels = ioptions_.num_levels;
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  NewVersionStorage(num_levels, kCompactionStyleUniversal);
+
+  // L4:   [200, 220]  [230, 250]       [360, 380]
+  // L5:
+  // L6: [101,                    351]
+  Add(4, 40U, "200", "220", 60000000U);
+  Add(4, 41U, "230", "250", 60000000U);
+  Add(4, 42U, "360", "380", 60000000U);
+  Add(6, 60U, "101", "351", 60000000U);
+  UpdateVersionStorageInfo();
+
+  // create an ongoing compaction to L5 (penultimate level)
+  CompactionOptions comp_options;
+  std::unordered_set<uint64_t> input_set;
+  input_set.insert(40);
+  std::vector<CompactionInputFiles> input_files;
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+
+  ASSERT_TRUE(comp1);
+  ASSERT_EQ(comp1->GetPenultimateLevel(), Compaction::kInvalidLevel);
+
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(60);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  ASSERT_EQ(enable_per_key_placement_,
+            universal_compaction_picker.FilesRangeOverlapWithCompaction(
+                input_files, 6,
+                Compaction::EvaluatePenultimateLevel(vstorage_.get(), ioptions_,
+                                                     6, 6)));
+
+  if (!enable_per_key_placement_) {
+    std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
+        comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+        mutable_db_options_, 0));
+    ASSERT_TRUE(comp2);
+    ASSERT_EQ(Compaction::kInvalidLevel, comp2->GetPenultimateLevel());
+  }
+}
+
+TEST_P(PerKeyPlacementCompactionPickerTest,
+       LastLevelOnlyNoConflictWithOngoingUniversal) {
+  // This is similar to `LastLevelOnlyConflictWithOngoingUniversal`, the only
+  // change is the ongoing compaction to L5 has no overlap with the last level
+  // compaction, so it's safe to move data from the last level to the
+  // penultimate level.
+  if (enable_per_key_placement_) {
+    ioptions_.preclude_last_level_data_seconds = 10000;
+  }
+
+  int num_levels = ioptions_.num_levels;
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  NewVersionStorage(num_levels, kCompactionStyleUniversal);
+
+  // L4:   [200, 220]  [230, 250]       [360, 380]
+  // L5:
+  // L6: [101,                    351]
+  Add(4, 40U, "200", "220", 60000000U);
+  Add(4, 41U, "230", "250", 60000000U);
+  Add(4, 42U, "360", "380", 60000000U);
+  Add(6, 60U, "101", "351", 60000000U);
+  UpdateVersionStorageInfo();
+
+  // create an ongoing compaction to L5 (penultimate level)
+  CompactionOptions comp_options;
+  std::unordered_set<uint64_t> input_set;
+  input_set.insert(42);
+  std::vector<CompactionInputFiles> input_files;
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+
+  ASSERT_TRUE(comp1);
+  ASSERT_EQ(comp1->GetPenultimateLevel(), Compaction::kInvalidLevel);
+
+  input_set.clear();
+  input_files.clear();
+  input_set.insert(60);
+  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
+      &input_files, &input_set, vstorage_.get(), comp_options));
+
+  // always safe to move data up
+  ASSERT_FALSE(universal_compaction_picker.FilesRangeOverlapWithCompaction(
+      input_files, 6,
+      Compaction::EvaluatePenultimateLevel(vstorage_.get(), ioptions_, 6, 6)));
+
+  // 2 compactions can be run in parallel
+  std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
+      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+      mutable_db_options_, 0));
+  ASSERT_TRUE(comp2);
+  if (enable_per_key_placement_) {
+    ASSERT_NE(Compaction::kInvalidLevel, comp2->GetPenultimateLevel());
+  } else {
+    ASSERT_EQ(Compaction::kInvalidLevel, comp2->GetPenultimateLevel());
+  }
 }
 
 INSTANTIATE_TEST_CASE_P(PerKeyPlacementCompactionPickerTest,
@@ -3664,6 +3958,7 @@ INSTANTIATE_TEST_CASE_P(PerKeyPlacementCompactionPickerTest,
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/compaction/compaction_picker_universal.cc

@@ -308,9 +308,10 @@ void UniversalCompactionBuilder::SortedRun::Dump(char* out_buf,
     if (file->fd.GetPathId() == 0 || !print_path) {
       snprintf(out_buf, out_buf_size, "file %" PRIu64, file->fd.GetNumber());
     } else {
-      snprintf(out_buf, out_buf_size, "file %" PRIu64
-                                      "(path "
-                                      "%" PRIu32 ")",
+      snprintf(out_buf, out_buf_size,
+               "file %" PRIu64
+               "(path "
+               "%" PRIu32 ")",
                file->fd.GetNumber(), file->fd.GetPathId());
     }
   } else {
@@ -743,6 +744,13 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSortedRuns(
     grandparents = vstorage_->LevelFiles(sorted_runs_[first_index_after].level);
   }
 
+  if (output_level != 0 &&
+      picker_->FilesRangeOverlapWithCompaction(
+          inputs, output_level,
+          Compaction::EvaluatePenultimateLevel(vstorage_, ioptions_,
+                                               start_level, output_level))) {
+    return nullptr;
+  }
   CompactionReason compaction_reason;
   if (max_number_of_files_to_compact == UINT_MAX) {
     compaction_reason = CompactionReason::kUniversalSizeRatio;
@@ -1081,6 +1089,24 @@ Compaction* UniversalCompactionBuilder::PickIncrementalForReduceSizeAmp(
   inputs.push_back(second_last_level_inputs);
   inputs.push_back(bottom_level_inputs);
 
+  int start_level = Compaction::kInvalidLevel;
+  for (const auto& in : inputs) {
+    if (!in.empty()) {
+      // inputs should already be sorted by level
+      start_level = in.level;
+      break;
+    }
+  }
+
+  // intra L0 compactions outputs could have overlap
+  if (output_level != 0 &&
+      picker_->FilesRangeOverlapWithCompaction(
+          inputs, output_level,
+          Compaction::EvaluatePenultimateLevel(vstorage_, ioptions_,
+                                               start_level, output_level))) {
+    return nullptr;
+  }
+
   // TODO support multi paths?
   uint32_t path_id = 0;
   return new Compaction(
@@ -1210,7 +1236,10 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
       if (!output_level_inputs.empty()) {
         inputs.push_back(output_level_inputs);
       }
-      if (picker_->FilesRangeOverlapWithCompaction(inputs, output_level)) {
+      if (picker_->FilesRangeOverlapWithCompaction(
+              inputs, output_level,
+              Compaction::EvaluatePenultimateLevel(
+                  vstorage_, ioptions_, start_level, output_level))) {
         return nullptr;
       }
 
@@ -1312,6 +1341,15 @@ Compaction* UniversalCompactionBuilder::PickCompactionWithSortedRunRange(
     output_level = sorted_runs_[end_index + 1].level - 1;
   }
 
+  // intra L0 compactions outputs could have overlap
+  if (output_level != 0 &&
+      picker_->FilesRangeOverlapWithCompaction(
+          inputs, output_level,
+          Compaction::EvaluatePenultimateLevel(vstorage_, ioptions_,
+                                               start_level, output_level))) {
+    return nullptr;
+  }
+
   // We never check size for
   // compaction_options_universal.compression_size_percent,
   // because we always compact all the files, so always compress.

db/compaction/compaction_service_test.cc

@@ -325,6 +325,7 @@ TEST_F(CompactionServiceTest, BasicCompactions) {
       });
   Reopen(options);
   ASSERT_GT(verify_passed, 0);
+  Close();
 }
 
 TEST_F(CompactionServiceTest, ManualCompaction) {

db/compaction/tiered_compaction_test.cc

@@ -10,7 +10,10 @@
 
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "rocksdb/iostats_context.h"
 #include "rocksdb/listener.h"
+#include "rocksdb/utilities/debug.h"
+#include "test_util/mock_time_env.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -414,38 +417,20 @@ TEST_P(TieredCompactionTest, RangeBasedTieredStorageUniversal) {
   // No data is moved from cold tier to hot tier because no input files from L5
   // or higher, it's not safe to move data to output_to_penultimate_level level.
   ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+  ASSERT_EQ("0,0,0,0,0,1", FilesPerLevel());
 
-  // Add 2 keys in higher level, but in separated files, the keys within that
-  // range should be moved up to the penultimate level
+  // Add 2 keys in higher level, but in separated files, all keys can be moved
+  // up if it's hot
   ASSERT_OK(Put(Key(0), "value" + std::to_string(0)));
   ASSERT_OK(Flush());
   ASSERT_OK(Put(Key(50), "value" + std::to_string(0)));
   ASSERT_OK(Flush());
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  ASSERT_EQ("0,0,0,0,0,1,1", FilesPerLevel());
-
-  // Add an SST with a key range cover all the data we want to move from the
-  // last level to the penultimate level
-  ASSERT_OK(Put(Key(0), "value" + std::to_string(0)));
-  ASSERT_OK(Put(Key(99), "value" + std::to_string(0)));
-  ASSERT_OK(Flush());
-
-  ResetAllStats(expect_stats, expect_pl_stats);
-
   ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
   ASSERT_EQ("0,0,0,0,0,1", FilesPerLevel());
+
   ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
   ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
 
-  last_stats.Add(kBasicCompStats);
-  last_stats.ResetCompactionReason(CompactionReason::kManualCompaction);
-  last_stats.bytes_read_output_level = kHasValue;
-  last_stats.num_input_files_in_output_level = kHasValue;
-  expect_pl_stats.Add(kBasicPerKeyPlacementCompStats);
-  expect_pl_stats.ResetCompactionReason(CompactionReason::kManualCompaction);
-  VerifyCompactionStats(expect_stats, expect_pl_stats);
-
   // change to only 1 key cold, to test compaction could stop even it matches
   // size amp compaction threshold
   {
@@ -1215,6 +1200,817 @@ TEST_P(TieredCompactionTest, RangeBasedTieredStorageLevel) {
 INSTANTIATE_TEST_CASE_P(TieredCompactionTest, TieredCompactionTest,
                         testing::Bool());
 
+class PrecludeLastLevelTest : public DBTestBase {
+ public:
+  PrecludeLastLevelTest()
+      : DBTestBase("preclude_last_level_test", /*env_do_fsync=*/false) {
+    mock_clock_ = std::make_shared<MockSystemClock>(env_->GetSystemClock());
+    mock_env_ = std::make_unique<CompositeEnvWrapper>(env_, mock_clock_);
+  }
+
+ protected:
+  std::unique_ptr<Env> mock_env_;
+  std::shared_ptr<MockSystemClock> mock_clock_;
+
+  void SetUp() override {
+    mock_clock_->InstallTimedWaitFixCallback();
+    SyncPoint::GetInstance()->SetCallBack(
+        "DBImpl::StartPeriodicTaskScheduler:Init", [&](void* arg) {
+          auto periodic_task_scheduler_ptr =
+              reinterpret_cast<PeriodicTaskScheduler*>(arg);
+          periodic_task_scheduler_ptr->TEST_OverrideTimer(mock_clock_.get());
+        });
+    mock_clock_->SetCurrentTime(0);
+  }
+};
+
+TEST_F(PrecludeLastLevelTest, MigrationFromPreserveTimeManualCompaction) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preserve_internal_time_seconds = 10000;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec)); });
+
+  int sst_num = 0;
+  // Write files that are overlap and enough to trigger compaction
+  for (; sst_num < kNumTrigger; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), "value"));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // all data is pushed to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  // enable preclude feature
+  options.preclude_last_level_data_seconds = 10000;
+  options.last_level_temperature = Temperature::kCold;
+  Reopen(options);
+
+  // all data is hot, even they're in the last level
+  ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  // Generate a sstable and trigger manual compaction
+  ASSERT_OK(Put(Key(10), "value"));
+  ASSERT_OK(Flush());
+
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // all data is moved up to the penultimate level
+  ASSERT_EQ("0,0,0,0,0,1", FilesPerLevel());
+  ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  // close explicitly, because the env is local variable which will be released
+  // first.
+  Close();
+}
+
+TEST_F(PrecludeLastLevelTest, MigrationFromPreserveTimeAutoCompaction) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preserve_internal_time_seconds = 10000;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec)); });
+
+  int sst_num = 0;
+  // Write files that are overlap and enough to trigger compaction
+  for (; sst_num < kNumTrigger; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), "value"));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // all data is pushed to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  // enable preclude feature
+  options.preclude_last_level_data_seconds = 10000;
+  options.last_level_temperature = Temperature::kCold;
+  // make sure it won't trigger Size Amp compaction, unlike normal Size Amp
+  // compaction which is typically a last level compaction, when tiered Storage
+  // ("preclude_last_level") is enabled, size amp won't include the last level.
+  // As the last level would be in cold tier and the size would not be a
+  // problem, which also avoid frequent hot to cold storage compaction.
+  options.compaction_options_universal.max_size_amplification_percent = 400;
+  Reopen(options);
+
+  // all data is hot, even they're in the last level
+  ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  // Write more data, but still all hot until the 10th SST, as:
+  // write a key every 10 seconds, 100 keys per SST, each SST takes 1000 seconds
+  // The preclude_last_level_data_seconds is 10k
+  Random rnd(301);
+  for (; sst_num < kNumTrigger * 2 - 1; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      // the value needs to be big enough to trigger full compaction
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), rnd.RandomString(100)));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+    ASSERT_OK(dbfull()->WaitForCompact(true));
+  }
+
+  // all data is moved up to the penultimate level
+  ASSERT_EQ("0,0,0,0,0,1", FilesPerLevel());
+  ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  // close explicitly, because the env is local variable which will be released
+  // first.
+  Close();
+}
+
+TEST_F(PrecludeLastLevelTest, MigrationFromPreserveTimePartial) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preserve_internal_time_seconds = 2000;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec)); });
+
+  int sst_num = 0;
+  // Write files that are overlap and enough to trigger compaction
+  for (; sst_num < kNumTrigger; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), "value"));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // all data is pushed to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  std::vector<KeyVersion> key_versions;
+  ASSERT_OK(GetAllKeyVersions(db_, Slice(), Slice(),
+                              std::numeric_limits<size_t>::max(),
+                              &key_versions));
+
+  // make sure there're more than 300 keys and first 100 keys are having seqno
+  // zeroed out, the last 100 key seqno not zeroed out
+  ASSERT_GT(key_versions.size(), 300);
+  for (int i = 0; i < 100; i++) {
+    ASSERT_EQ(key_versions[i].sequence, 0);
+  }
+  auto rit = key_versions.rbegin();
+  for (int i = 0; i < 100; i++) {
+    ASSERT_GT(rit->sequence, 0);
+    rit++;
+  }
+
+  // enable preclude feature
+  options.preclude_last_level_data_seconds = 2000;
+  options.last_level_temperature = Temperature::kCold;
+  Reopen(options);
+
+  // Generate a sstable and trigger manual compaction
+  ASSERT_OK(Put(Key(10), "value"));
+  ASSERT_OK(Flush());
+
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // some data are moved up, some are not
+  ASSERT_EQ("0,0,0,0,0,1,1", FilesPerLevel());
+  ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  Close();
+}
+
+TEST_F(PrecludeLastLevelTest, SmallPrecludeTime) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preclude_last_level_data_seconds = 60;
+  options.preserve_internal_time_seconds = 0;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  options.last_level_temperature = Temperature::kCold;
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+
+  dbfull()->TEST_WaitForPeridicTaskRun([&] {
+    mock_clock_->MockSleepForSeconds(static_cast<int>(rnd.Uniform(10) + 1));
+  });
+
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(Put(Key(i), rnd.RandomString(100)));
+    dbfull()->TEST_WaitForPeridicTaskRun([&] {
+      mock_clock_->MockSleepForSeconds(static_cast<int>(rnd.Uniform(2)));
+    });
+  }
+  ASSERT_OK(Flush());
+
+  TablePropertiesCollection tables_props;
+  ASSERT_OK(dbfull()->GetPropertiesOfAllTables(&tables_props));
+  ASSERT_EQ(tables_props.size(), 1);
+  ASSERT_FALSE(tables_props.begin()->second->seqno_to_time_mapping.empty());
+  SeqnoToTimeMapping tp_mapping;
+  ASSERT_OK(
+      tp_mapping.Add(tables_props.begin()->second->seqno_to_time_mapping));
+  ASSERT_OK(tp_mapping.Sort());
+  ASSERT_FALSE(tp_mapping.Empty());
+  auto seqs = tp_mapping.TEST_GetInternalMapping();
+  ASSERT_FALSE(seqs.empty());
+
+  // Wait more than preclude_last_level time, then make sure all the data is
+  // compacted to the last level even there's no write (no seqno -> time
+  // information was flushed to any SST).
+  mock_clock_->MockSleepForSeconds(100);
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+  ASSERT_EQ(GetSstSizeHelper(Temperature::kUnknown), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
+
+  Close();
+}
+
+TEST_F(PrecludeLastLevelTest, LastLevelOnlyCompactionPartial) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preserve_internal_time_seconds = 2000;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec)); });
+
+  int sst_num = 0;
+  // Write files that are overlap and enough to trigger compaction
+  for (; sst_num < kNumTrigger; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), "value"));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // all data is pushed to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  // enable preclude feature
+  options.preclude_last_level_data_seconds = 2000;
+  options.last_level_temperature = Temperature::kCold;
+  Reopen(options);
+
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // some data are moved up, some are not
+  ASSERT_EQ("0,0,0,0,0,1,1", FilesPerLevel());
+  ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+
+  std::vector<KeyVersion> key_versions;
+  ASSERT_OK(GetAllKeyVersions(db_, Slice(), Slice(),
+                              std::numeric_limits<size_t>::max(),
+                              &key_versions));
+
+  // make sure there're more than 300 keys and first 100 keys are having seqno
+  // zeroed out, the last 100 key seqno not zeroed out
+  ASSERT_GT(key_versions.size(), 300);
+  for (int i = 0; i < 100; i++) {
+    ASSERT_EQ(key_versions[i].sequence, 0);
+  }
+  auto rit = key_versions.rbegin();
+  for (int i = 0; i < 100; i++) {
+    ASSERT_GT(rit->sequence, 0);
+    rit++;
+  }
+
+  Close();
+}
+
+class PrecludeLastLevelTestWithParms
+    : public PrecludeLastLevelTest,
+      public testing::WithParamInterface<bool> {
+ public:
+  PrecludeLastLevelTestWithParms() : PrecludeLastLevelTest() {}
+};
+
+TEST_P(PrecludeLastLevelTestWithParms, LastLevelOnlyCompactionNoPreclude) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kNumKeys = 100;
+  const int kKeyPerSec = 10;
+
+  bool enable_preclude_last_level = GetParam();
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.preserve_internal_time_seconds = 2000;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec)); });
+
+  Random rnd(301);
+  int sst_num = 0;
+  // Write files that are overlap and enough to trigger compaction
+  for (; sst_num < kNumTrigger; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), rnd.RandomString(100)));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // all data is pushed to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  std::atomic_bool is_manual_compaction_running = false;
+  std::atomic_bool verified_compaction_order = false;
+
+  // Make sure the manual compaction is in progress and try to trigger a
+  // SizeRatio compaction by flushing 4 files to L0. The compaction will try to
+  // compact 4 files at L0 to L5 (the last empty level).
+  // If the preclude_last_feature is enabled, the auto triggered compaction
+  // cannot be picked. Otherwise, the auto triggered compaction can run in
+  // parallel with the last level compaction.
+  // L0: [a] [b] [c] [d]
+  // L5:     (locked if preclude_last_level is enabled)
+  // L6: [z] (locked: manual compaction in progress)
+  // TODO: in this case, L0 files should just be compacted to L4, so the 2
+  //  compactions won't be overlapped.
+  SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::ProcessKeyValueCompaction()::Processing", [&](void* arg) {
+        auto compaction = static_cast<Compaction*>(arg);
+        if (compaction->is_manual_compaction()) {
+          is_manual_compaction_running = true;
+          TEST_SYNC_POINT(
+              "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+              "ManualCompaction1");
+          TEST_SYNC_POINT(
+              "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+              "ManualCompaction2");
+          is_manual_compaction_running = false;
+        }
+      });
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "UniversalCompactionBuilder::PickCompaction:Return", [&](void* arg) {
+        auto compaction = static_cast<Compaction*>(arg);
+        if (enable_preclude_last_level && is_manual_compaction_running) {
+          ASSERT_TRUE(compaction == nullptr);
+          verified_compaction_order = true;
+        } else {
+          ASSERT_TRUE(compaction != nullptr);
+          verified_compaction_order = true;
+        }
+        if (!compaction || !compaction->is_manual_compaction()) {
+          TEST_SYNC_POINT(
+              "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+              "AutoCompactionPicked");
+        }
+      });
+
+  SyncPoint::GetInstance()->LoadDependency({
+      {"PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+       "ManualCompaction1",
+       "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:StartWrite"},
+      {"PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+       "AutoCompactionPicked",
+       "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:"
+       "ManualCompaction2"},
+  });
+
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  // only enable if the Parameter is true
+  if (enable_preclude_last_level) {
+    options.preclude_last_level_data_seconds = 2000;
+  }
+  options.max_background_jobs = 8;
+  options.last_level_temperature = Temperature::kCold;
+  Reopen(options);
+
+  auto manual_compaction_thread = port::Thread([this]() {
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    cro.exclusive_manual_compaction = false;
+    ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  });
+
+  TEST_SYNC_POINT(
+      "PrecludeLastLevelTest::LastLevelOnlyCompactionConflit:StartWrite");
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();
+
+  for (; sst_num < kNumTrigger * 2; sst_num++) {
+    for (int i = 0; i < kNumKeys; i++) {
+      // the value needs to be big enough to trigger full compaction
+      ASSERT_OK(Put(Key(sst_num * (kNumKeys - 1) + i), "value"));
+      dbfull()->TEST_WaitForPeridicTaskRun([&] {
+        mock_clock_->MockSleepForSeconds(static_cast<int>(kKeyPerSec));
+      });
+    }
+    ASSERT_OK(Flush());
+  }
+
+  manual_compaction_thread.join();
+
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  if (enable_preclude_last_level) {
+    ASSERT_NE("0,0,0,0,0,1,1", FilesPerLevel());
+  } else {
+    ASSERT_EQ("0,0,0,0,0,1,1", FilesPerLevel());
+  }
+  ASSERT_TRUE(verified_compaction_order);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  stop_token.reset();
+
+  Close();
+}
+
+INSTANTIATE_TEST_CASE_P(PrecludeLastLevelTestWithParms,
+                        PrecludeLastLevelTestWithParms, testing::Bool());
+
+// partition the SST into 3 ranges [0, 19] [20, 39] [40, ...]
+class ThreeRangesPartitioner : public SstPartitioner {
+ public:
+  const char* Name() const override { return "SingleKeySstPartitioner"; }
+
+  PartitionerResult ShouldPartition(
+      const PartitionerRequest& request) override {
+    if ((cmp->CompareWithoutTimestamp(*request.current_user_key,
+                                      DBTestBase::Key(20)) >= 0 &&
+         cmp->CompareWithoutTimestamp(*request.prev_user_key,
+                                      DBTestBase::Key(20)) < 0) ||
+        (cmp->CompareWithoutTimestamp(*request.current_user_key,
+                                      DBTestBase::Key(40)) >= 0 &&
+         cmp->CompareWithoutTimestamp(*request.prev_user_key,
+                                      DBTestBase::Key(40)) < 0)) {
+      return kRequired;
+    } else {
+      return kNotRequired;
+    }
+  }
+
+  bool CanDoTrivialMove(const Slice& /*smallest_user_key*/,
+                        const Slice& /*largest_user_key*/) override {
+    return false;
+  }
+
+  const Comparator* cmp = BytewiseComparator();
+};
+
+class ThreeRangesPartitionerFactory : public SstPartitionerFactory {
+ public:
+  static const char* kClassName() {
+    return "TombstoneTestSstPartitionerFactory";
+  }
+  const char* Name() const override { return kClassName(); }
+
+  std::unique_ptr<SstPartitioner> CreatePartitioner(
+      const SstPartitioner::Context& /* context */) const override {
+    return std::unique_ptr<SstPartitioner>(new ThreeRangesPartitioner());
+  }
+};
+
+TEST_F(PrecludeLastLevelTest, PartialPenultimateLevelCompaction) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.preserve_internal_time_seconds = 10000;
+  options.num_levels = kNumLevels;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(10)); });
+
+  Random rnd(301);
+
+  for (int i = 0; i < 300; i++) {
+    ASSERT_OK(Put(Key(i), rnd.RandomString(100)));
+    dbfull()->TEST_WaitForPeridicTaskRun(
+        [&] { mock_clock_->MockSleepForSeconds(kKeyPerSec); });
+  }
+  ASSERT_OK(Flush());
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // make sure all data is compacted to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  // Create 3 L5 files
+  auto factory = std::make_shared<ThreeRangesPartitionerFactory>();
+  options.sst_partitioner_factory = factory;
+
+  Reopen(options);
+
+  for (int i = 0; i < kNumTrigger - 1; i++) {
+    for (int j = 0; j < 100; j++) {
+      ASSERT_OK(Put(Key(i * 100 + j), rnd.RandomString(10)));
+    }
+    ASSERT_OK(Flush());
+  }
+
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  // L5: [0,19] [20,39] [40,299]
+  // L6: [0,                299]
+  ASSERT_EQ("0,0,0,0,0,3,1", FilesPerLevel());
+
+  // enable tiered storage feature
+  options.preclude_last_level_data_seconds = 10000;
+  options.last_level_temperature = Temperature::kCold;
+  options.statistics = CreateDBStatistics();
+  Reopen(options);
+
+  ColumnFamilyMetaData meta;
+  db_->GetColumnFamilyMetaData(&meta);
+  ASSERT_EQ(meta.levels[5].files.size(), 3);
+  ASSERT_EQ(meta.levels[6].files.size(), 1);
+  ASSERT_EQ(meta.levels[6].files[0].smallestkey, Key(0));
+  ASSERT_EQ(meta.levels[6].files[0].largestkey, Key(299));
+
+  std::string file_path = meta.levels[5].files[1].db_path;
+  std::vector<std::string> files;
+  // pick 3rd file @L5 + file@L6 for compaction
+  files.push_back(file_path + "/" + meta.levels[5].files[2].name);
+  files.push_back(file_path + "/" + meta.levels[6].files[0].name);
+  ASSERT_OK(db_->CompactFiles(CompactionOptions(), files, 6));
+
+  // The compaction only moved partial of the hot data to hot tier, range[0,39]
+  // is unsafe to move up, otherwise, they will be overlapped with the existing
+  // files@L5.
+  // The output should be:
+  //  L5: [0,19] [20,39] [40,299]    <-- Temperature::kUnknown
+  //  L6: [0,19] [20,39]             <-- Temperature::kCold
+  // L6 file is split because of the customized partitioner
+  ASSERT_EQ("0,0,0,0,0,3,2", FilesPerLevel());
+
+  // even all the data is hot, but not all data are moved to the hot tier
+  ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
+  ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
+
+  db_->GetColumnFamilyMetaData(&meta);
+  ASSERT_EQ(meta.levels[5].files.size(), 3);
+  ASSERT_EQ(meta.levels[6].files.size(), 2);
+  for (const auto& file : meta.levels[5].files) {
+    ASSERT_EQ(file.temperature, Temperature::kUnknown);
+  }
+  for (const auto& file : meta.levels[6].files) {
+    ASSERT_EQ(file.temperature, Temperature::kCold);
+  }
+  ASSERT_EQ(meta.levels[6].files[0].smallestkey, Key(0));
+  ASSERT_EQ(meta.levels[6].files[0].largestkey, Key(19));
+  ASSERT_EQ(meta.levels[6].files[1].smallestkey, Key(20));
+  ASSERT_EQ(meta.levels[6].files[1].largestkey, Key(39));
+
+  Close();
+}
+
+struct TestPropertiesCollector : public TablePropertiesCollector {
+  Status AddUserKey(const Slice& key, const Slice& /*value*/,
+                    EntryType /*type*/, SequenceNumber /*seq*/,
+                    uint64_t /*file_size*/) override {
+    if (cmp->Compare(key, DBTestBase::Key(100)) == 0) {
+      has_key_100 = true;
+    }
+    if (cmp->Compare(key, DBTestBase::Key(200)) == 0) {
+      has_key_200 = true;
+    }
+
+    return Status::OK();
+  }
+
+  const char* Name() const override { return "TestTablePropertiesCollector"; }
+
+  UserCollectedProperties GetReadableProperties() const override {
+    UserCollectedProperties ret;
+    return ret;
+  }
+
+  Status Finish(UserCollectedProperties* /*properties*/) override {
+    // The LSM tree would be like:
+    // L5: [0,19] [20,39] [40,299]
+    // L6: [0,                299]
+    // the 3rd file @L5 has both 100 and 200, which will be marked for
+    // compaction
+    // Also avoid marking flushed SST for compaction, which won't have both 100
+    // and 200
+    if (has_key_100 && has_key_200) {
+      need_compact_ = true;
+    } else {
+      need_compact_ = false;
+    }
+    has_key_100 = false;
+    has_key_200 = false;
+    return Status::OK();
+  }
+
+  bool NeedCompact() const override { return need_compact_; }
+
+  const Comparator* cmp = BytewiseComparator();
+
+ private:
+  bool has_key_100 = false;
+  bool has_key_200 = false;
+
+  bool need_compact_ = false;
+};
+
+class TestPropertiesCollectorFactory : public TablePropertiesCollectorFactory {
+ public:
+  TablePropertiesCollector* CreateTablePropertiesCollector(
+      TablePropertiesCollectorFactory::Context /*context*/) override {
+    return new TestPropertiesCollector;
+  }
+  const char* Name() const override { return "TestTablePropertiesCollector"; }
+};
+
+TEST_F(PrecludeLastLevelTest, PartialPenultimateLevelCompactionWithRangeDel) {
+  const int kNumTrigger = 4;
+  const int kNumLevels = 7;
+  const int kKeyPerSec = 10;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.env = mock_env_.get();
+  options.level0_file_num_compaction_trigger = kNumTrigger;
+  options.preserve_internal_time_seconds = 10000;
+  options.num_levels = kNumLevels;
+  // set a small max_compaction_bytes to avoid input level expansion
+  options.max_compaction_bytes = 30000;
+  options.ignore_max_compaction_bytes_for_input = false;
+  DestroyAndReopen(options);
+
+  // pass some time first, otherwise the first a few keys write time are going
+  // to be zero, and internally zero has special meaning: kUnknownSeqnoTime
+  dbfull()->TEST_WaitForPeridicTaskRun(
+      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(10)); });
+
+  Random rnd(301);
+
+  for (int i = 0; i < 300; i++) {
+    ASSERT_OK(Put(Key(i), rnd.RandomString(100)));
+    dbfull()->TEST_WaitForPeridicTaskRun(
+        [&] { mock_clock_->MockSleepForSeconds(kKeyPerSec); });
+  }
+  ASSERT_OK(Flush());
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // make sure all data is compacted to the last level
+  ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
+
+  // Create 3 L5 files
+  auto factory = std::make_shared<ThreeRangesPartitionerFactory>();
+  options.sst_partitioner_factory = factory;
+
+  // the user defined properties_collector will mark the 3rd file for compaction
+  auto collector_factory = std::make_shared<TestPropertiesCollectorFactory>();
+  options.table_properties_collector_factories.resize(1);
+  options.table_properties_collector_factories[0] = collector_factory;
+  // enable tiered storage feature
+  options.preclude_last_level_data_seconds = 10000;
+  options.last_level_temperature = Temperature::kCold;
+  Reopen(options);
+
+  for (int i = 0; i < kNumTrigger - 2; i++) {
+    for (int j = 0; j < 100; j++) {
+      ASSERT_OK(Put(Key(i * 100 + j), rnd.RandomString(10)));
+    }
+    ASSERT_OK(Flush());
+  }
+
+  // make sure there is one and only one compaction supports per-key placement
+  // but has the penultimate level output disabled.
+  std::atomic_int per_key_comp_num = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "UniversalCompactionBuilder::PickCompaction:Return", [&](void* arg) {
+        auto compaction = static_cast<Compaction*>(arg);
+        if (compaction->SupportsPerKeyPlacement()) {
+          ASSERT_EQ(compaction->GetPenultimateOutputRangeType(),
+                    Compaction::PenultimateOutputRangeType::kDisabled);
+          per_key_comp_num++;
+        }
+      });
+
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  for (int j = 0; j < 100; j++) {
+    ASSERT_OK(Put(Key(200 + j), rnd.RandomString(10)));
+  }
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                             Key(32), Key(40)));
+  ASSERT_OK(Flush());
+
+  // Before the per-key placement compaction, the LSM tress should be like:
+  // L5: [0,19] [20,40] [40,299]
+  // L6: [0,                299]
+  // The 2nd file @L5 has the largest key 40 because of range del
+
+  ASSERT_OK(dbfull()->WaitForCompact(true));
+
+  ASSERT_EQ(per_key_comp_num, 1);
+
+  // the compaction won't move any data to the penultimate level
+  ASSERT_EQ("0,0,0,0,0,2,3", FilesPerLevel());
+
+  Close();
+}
+
 #endif  // !defined(ROCKSDB_LITE)
 
 }  // namespace ROCKSDB_NAMESPACE

db/comparator_db_test.cc

@@ -17,7 +17,6 @@
 #include "util/string_util.h"
 #include "utilities/merge_operators.h"
 
-
 namespace ROCKSDB_NAMESPACE {
 namespace {
 
@@ -249,7 +248,7 @@ class TwoStrComparator : public Comparator {
 
   void FindShortSuccessor(std::string* /*key*/) const override {}
 };
-}  // namespace
+}  // anonymous namespace
 
 class ComparatorDBTest
     : public testing::Test,
@@ -470,7 +469,7 @@ void VerifySuccessor(const Slice& s, const Slice& t) {
   ASSERT_FALSE(rbc->IsSameLengthImmediateSuccessor(t, s));
 }
 
-}  // namespace
+}  // anonymous namespace
 
 TEST_P(ComparatorDBTest, IsSameLengthImmediateSuccessor) {
   {
@@ -673,6 +672,7 @@ TEST_P(ComparatorDBTest, SeparatorSuccessorRandomizeTest) {
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }

db/convenience.cc

@@ -26,8 +26,7 @@ Status DeleteFilesInRange(DB* db, ColumnFamilyHandle* column_family,
 }
 
 Status DeleteFilesInRanges(DB* db, ColumnFamilyHandle* column_family,
-                           const RangePtr* ranges, size_t n,
-                           bool include_end) {
+                           const RangePtr* ranges, size_t n, bool include_end) {
   return (static_cast_with_check<DBImpl>(db->GetRootDB()))
       ->DeleteFilesInRanges(column_family, ranges, n, include_end);
 }
@@ -47,9 +46,8 @@ Status VerifySstFileChecksum(const Options& options,
   InternalKeyComparator internal_comparator(options.comparator);
   ImmutableOptions ioptions(options);
 
-  Status s = ioptions.fs->NewRandomAccessFile(file_path,
-                                              FileOptions(env_options),
-                                              &file, nullptr);
+  Status s = ioptions.fs->NewRandomAccessFile(
+      file_path, FileOptions(env_options), &file, nullptr);
   if (s.ok()) {
     s = ioptions.fs->GetFileSize(file_path, IOOptions(), &file_size, nullptr);
   } else {

db/corruption_test.cc

@@ -65,7 +65,7 @@ class ErrorEnv : public EnvWrapper {
     return target()->NewWritableFile(fname, result, soptions);
   }
 };
-}  // namespace
+}  // anonymous namespace
 class CorruptionTest : public testing::Test {
  public:
   std::shared_ptr<Env> env_guard_;
@@ -138,9 +138,7 @@ class CorruptionTest : public testing::Test {
     return DB::Open(opt, dbname_, &db_);
   }
 
-  void Reopen(Options* options = nullptr) {
-    ASSERT_OK(TryReopen(options));
-  }
+  void Reopen(Options* options = nullptr) { ASSERT_OK(TryReopen(options)); }
 
   void RepairDB() {
     delete db_;
@@ -156,7 +154,7 @@ class CorruptionTest : public testing::Test {
         DBImpl* dbi = static_cast_with_check<DBImpl>(db_);
         ASSERT_OK(dbi->TEST_FlushMemTable());
       }
-      //if ((i % 100) == 0) fprintf(stderr, "@ %d of %d\n", i, n);
+      // if ((i % 100) == 0) fprintf(stderr, "@ %d of %d\n", i, n);
       Slice key = Key(i + start, &key_space);
       batch.Clear();
       ASSERT_OK(batch.Put(key, Value(i + start, &value_space)));
@@ -183,8 +181,7 @@ class CorruptionTest : public testing::Test {
       ASSERT_OK(iter->status());
       uint64_t key;
       Slice in(iter->key());
-      if (!ConsumeDecimalNumber(&in, &key) ||
-          !in.empty() ||
+      if (!ConsumeDecimalNumber(&in, &key) || !in.empty() ||
           key < next_expected) {
         bad_keys++;
         continue;
@@ -200,10 +197,11 @@ class CorruptionTest : public testing::Test {
     iter->status().PermitUncheckedError();
     delete iter;
 
-    fprintf(stderr,
-      "expected=%d..%d; got=%d; bad_keys=%d; bad_values=%d; missed=%llu\n",
-            min_expected, max_expected, correct, bad_keys, bad_values,
-            static_cast<unsigned long long>(missed));
+    fprintf(
+        stderr,
+        "expected=%d..%d; got=%d; bad_keys=%d; bad_values=%d; missed=%llu\n",
+        min_expected, max_expected, correct, bad_keys, bad_values,
+        static_cast<unsigned long long>(missed));
     ASSERT_LE(min_expected, correct);
     ASSERT_GE(max_expected, correct);
   }
@@ -217,8 +215,7 @@ class CorruptionTest : public testing::Test {
     std::string fname;
     int picked_number = -1;
     for (size_t i = 0; i < filenames.size(); i++) {
-      if (ParseFileName(filenames[i], &number, &type) &&
-          type == filetype &&
+      if (ParseFileName(filenames[i], &number, &type) && type == filetype &&
           static_cast<int>(number) > picked_number) {  // Pick latest file
         fname = dbname_ + "/" + filenames[i];
         picked_number = static_cast<int>(number);
@@ -244,7 +241,6 @@ class CorruptionTest : public testing::Test {
     FAIL() << "no file found at level";
   }
 
-
   int Property(const std::string& name) {
     std::string property;
     int result;

db/cuckoo_table_db_test.cc

@@ -77,9 +77,7 @@ class CuckooTableDBTest : public testing::Test {
     return db_->Put(WriteOptions(), k, v);
   }
 
-  Status Delete(const std::string& k) {
-    return db_->Delete(WriteOptions(), k);
-  }
+  Status Delete(const std::string& k) { return db_->Delete(WriteOptions(), k); }
 
   std::string Get(const std::string& k) {
     ReadOptions options;
@@ -313,23 +311,21 @@ TEST_F(CuckooTableDBTest, AdaptiveTable) {
   // Write some keys using plain table.
   std::shared_ptr<TableFactory> block_based_factory(
       NewBlockBasedTableFactory());
-  std::shared_ptr<TableFactory> plain_table_factory(
-      NewPlainTableFactory());
-  std::shared_ptr<TableFactory> cuckoo_table_factory(
-      NewCuckooTableFactory());
+  std::shared_ptr<TableFactory> plain_table_factory(NewPlainTableFactory());
+  std::shared_ptr<TableFactory> cuckoo_table_factory(NewCuckooTableFactory());
   options.create_if_missing = false;
-  options.table_factory.reset(NewAdaptiveTableFactory(
-    plain_table_factory, block_based_factory, plain_table_factory,
-    cuckoo_table_factory));
+  options.table_factory.reset(
+      NewAdaptiveTableFactory(plain_table_factory, block_based_factory,
+                              plain_table_factory, cuckoo_table_factory));
   Reopen(&options);
   ASSERT_OK(Put("key4", "v4"));
   ASSERT_OK(Put("key1", "v5"));
   ASSERT_OK(dbfull()->TEST_FlushMemTable());
 
   // Write some keys using block based table.
-  options.table_factory.reset(NewAdaptiveTableFactory(
-    block_based_factory, block_based_factory, plain_table_factory,
-    cuckoo_table_factory));
+  options.table_factory.reset(
+      NewAdaptiveTableFactory(block_based_factory, block_based_factory,
+                              plain_table_factory, cuckoo_table_factory));
   Reopen(&options);
   ASSERT_OK(Put("key5", "v6"));
   ASSERT_OK(Put("key2", "v7"));
@@ -345,6 +341,7 @@ TEST_F(CuckooTableDBTest, AdaptiveTable) {
 
 int main(int argc, char** argv) {
   if (ROCKSDB_NAMESPACE::port::kLittleEndian) {
+    ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
     ::testing::InitGoogleTest(&argc, argv);
     return RUN_ALL_TESTS();
   } else {

db/db_basic_test.cc

@@ -2158,11 +2158,11 @@ class DBMultiGetAsyncIOTest : public DBBasicTest,
       : DBBasicTest(), statistics_(ROCKSDB_NAMESPACE::CreateDBStatistics()) {
     BlockBasedTableOptions bbto;
     bbto.filter_policy.reset(NewBloomFilterPolicy(10));
-    Options options = CurrentOptions();
-    options.disable_auto_compactions = true;
-    options.statistics = statistics_;
-    options.table_factory.reset(NewBlockBasedTableFactory(bbto));
-    Reopen(options);
+    options_ = CurrentOptions();
+    options_.disable_auto_compactions = true;
+    options_.statistics = statistics_;
+    options_.table_factory.reset(NewBlockBasedTableFactory(bbto));
+    Reopen(options_);
     int num_keys = 0;
 
     // Put all keys in the bottommost level, and overwrite some keys
@@ -2227,8 +2227,12 @@ class DBMultiGetAsyncIOTest : public DBBasicTest,
 
   const std::shared_ptr<Statistics>& statistics() { return statistics_; }
 
+ protected:
+  void ReopenDB() { Reopen(options_); }
+
  private:
   std::shared_ptr<Statistics> statistics_;
+  Options options_;
 };
 
 TEST_P(DBMultiGetAsyncIOTest, GetFromL0) {
@@ -2305,6 +2309,69 @@ TEST_P(DBMultiGetAsyncIOTest, GetFromL1) {
   ASSERT_EQ(statistics()->getTickerCount(MULTIGET_COROUTINE_COUNT), 3);
 }
 
+TEST_P(DBMultiGetAsyncIOTest, GetFromL1Error) {
+  std::vector<std::string> key_strs;
+  std::vector<Slice> keys;
+  std::vector<PinnableSlice> values;
+  std::vector<Status> statuses;
+
+  key_strs.push_back(Key(33));
+  key_strs.push_back(Key(54));
+  key_strs.push_back(Key(102));
+  keys.push_back(key_strs[0]);
+  keys.push_back(key_strs[1]);
+  keys.push_back(key_strs[2]);
+  values.resize(keys.size());
+  statuses.resize(keys.size());
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "TableCache::GetTableReader:BeforeOpenFile", [&](void* status) {
+        static int count = 0;
+        count++;
+        // Fail the last table reader open, which is the 6th SST file
+        // since 3 overlapping L0 files + 3 L1 files containing the keys
+        if (count == 6) {
+          Status* s = static_cast<Status*>(status);
+          *s = Status::IOError();
+        }
+      });
+  // DB open will create table readers unless we reduce the table cache
+  // capacity.
+  // SanitizeOptions will set max_open_files to minimum of 20. Table cache
+  // is allocated with max_open_files - 10 as capacity. So override
+  // max_open_files to 11 so table cache capacity will become 1. This will
+  // prevent file open during DB open and force the file to be opened
+  // during MultiGet
+  SyncPoint::GetInstance()->SetCallBack(
+      "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+        int* max_open_files = (int*)arg;
+        *max_open_files = 11;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ReopenDB();
+
+  ReadOptions ro;
+  ro.async_io = true;
+  ro.optimize_multiget_for_io = GetParam();
+  dbfull()->MultiGet(ro, dbfull()->DefaultColumnFamily(), keys.size(),
+                     keys.data(), values.data(), statuses.data());
+  SyncPoint::GetInstance()->DisableProcessing();
+  ASSERT_EQ(values.size(), 3);
+  ASSERT_EQ(statuses[0], Status::OK());
+  ASSERT_EQ(statuses[1], Status::OK());
+  ASSERT_EQ(statuses[2], Status::IOError());
+
+  HistogramData multiget_io_batch_size;
+
+  statistics()->histogramData(MULTIGET_IO_BATCH_SIZE, &multiget_io_batch_size);
+
+  // A batch of 3 async IOs is expected, one for each overlapping file in L1
+  ASSERT_EQ(multiget_io_batch_size.count, 1);
+  ASSERT_EQ(multiget_io_batch_size.max, 2);
+  ASSERT_EQ(statistics()->getTickerCount(MULTIGET_COROUTINE_COUNT), 2);
+}
+
 TEST_P(DBMultiGetAsyncIOTest, LastKeyInFile) {
   std::vector<std::string> key_strs;
   std::vector<Slice> keys;
@@ -2805,7 +2872,7 @@ TEST_F(DBBasicTest, MultiGetIOBufferOverrun) {
   table_options.pin_l0_filter_and_index_blocks_in_cache = true;
   table_options.block_size = 16 * 1024;
   ASSERT_TRUE(table_options.block_size >
-            BlockBasedTable::kMultiGetReadStackBufSize);
+              BlockBasedTable::kMultiGetReadStackBufSize);
   options.table_factory.reset(NewBlockBasedTableFactory(table_options));
   Reopen(options);
 
@@ -2914,7 +2981,7 @@ class TableFileListener : public EventListener {
   InstrumentedMutex mutex_;
   std::unordered_map<std::string, std::vector<std::string>> cf_to_paths_;
 };
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(DBBasicTest, LastSstFileNotInManifest) {
   // If the last sst file is not tracked in MANIFEST,

db/db_block_cache_test.cc

@@ -13,7 +13,6 @@
 
 #include "cache/cache_entry_roles.h"
 #include "cache/cache_key.h"
-#include "cache/fast_lru_cache.h"
 #include "cache/lru_cache.h"
 #include "db/column_family.h"
 #include "db/db_impl/db_impl.h"
@@ -170,13 +169,16 @@ class DBBlockCacheTest : public DBTestBase {
 #ifndef ROCKSDB_LITE
   const std::array<size_t, kNumCacheEntryRoles> GetCacheEntryRoleCountsBg() {
     // Verify in cache entry role stats
-    ColumnFamilyHandleImpl* cfh =
-        static_cast<ColumnFamilyHandleImpl*>(dbfull()->DefaultColumnFamily());
-    InternalStats* internal_stats_ptr = cfh->cfd()->internal_stats();
-    InternalStats::CacheEntryRoleStats stats;
-    internal_stats_ptr->TEST_GetCacheEntryRoleStats(&stats,
-                                                    /*foreground=*/false);
-    return stats.entry_counts;
+    std::array<size_t, kNumCacheEntryRoles> cache_entry_role_counts;
+    std::map<std::string, std::string> values;
+    EXPECT_TRUE(db_->GetMapProperty(DB::Properties::kFastBlockCacheEntryStats,
+                                    &values));
+    for (size_t i = 0; i < kNumCacheEntryRoles; ++i) {
+      auto role = static_cast<CacheEntryRole>(i);
+      cache_entry_role_counts[i] =
+          ParseSizeT(values[BlockCacheEntryStatsMapKeys::EntryCount(role)]);
+    }
+    return cache_entry_role_counts;
   }
 #endif  // ROCKSDB_LITE
 };
@@ -411,7 +413,7 @@ class ReadOnlyCacheWrapper : public CacheWrapper {
   }
 };
 
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(DBBlockCacheTest, TestWithSameCompressed) {
   auto table_options = GetTableOptions();
@@ -941,10 +943,7 @@ TEST_F(DBBlockCacheTest, AddRedundantStats) {
             capacity,
             BlockBasedTableOptions().block_size /*estimated_value_size*/,
             num_shard_bits)
-            .MakeSharedCache(),
-        NewFastLRUCache(capacity, 1 /*estimated_value_size*/, num_shard_bits,
-                        false /*strict_capacity_limit*/,
-                        kDefaultCacheMetadataChargePolicy)}) {
+            .MakeSharedCache()}) {
     if (!base_cache) {
       // Skip clock cache when not supported
       continue;
@@ -1303,11 +1302,6 @@ TEST_F(DBBlockCacheTest, CacheEntryRoleStats) {
               capacity,
               BlockBasedTableOptions().block_size /*estimated_value_size*/)
               .MakeSharedCache()}) {
-      if (!cache) {
-        // Skip clock cache when not supported
-        continue;
-      }
-
       ++iterations_tested;
 
       Options options = CurrentOptions();
@@ -1500,23 +1494,38 @@ TEST_F(DBBlockCacheTest, CacheEntryRoleStats) {
       dbfull()->DumpStats();
       ASSERT_EQ(scan_count, 1);
 
-      env_->MockSleepForSeconds(10000);
+      env_->MockSleepForSeconds(60);
+      ASSERT_TRUE(db_->GetMapProperty(DB::Properties::kFastBlockCacheEntryStats,
+                                      &values));
+      ASSERT_EQ(scan_count, 1);
       ASSERT_TRUE(
           db_->GetMapProperty(DB::Properties::kBlockCacheEntryStats, &values));
       ASSERT_EQ(scan_count, 2);
 
       env_->MockSleepForSeconds(10000);
-      std::string value_str;
-      ASSERT_TRUE(
-          db_->GetProperty(DB::Properties::kBlockCacheEntryStats, &value_str));
+      ASSERT_TRUE(db_->GetMapProperty(DB::Properties::kFastBlockCacheEntryStats,
+                                      &values));
       ASSERT_EQ(scan_count, 3);
 
+      env_->MockSleepForSeconds(60);
+      std::string value_str;
+      ASSERT_TRUE(db_->GetProperty(DB::Properties::kFastBlockCacheEntryStats,
+                                   &value_str));
+      ASSERT_EQ(scan_count, 3);
+      ASSERT_TRUE(
+          db_->GetProperty(DB::Properties::kBlockCacheEntryStats, &value_str));
+      ASSERT_EQ(scan_count, 4);
+
       env_->MockSleepForSeconds(10000);
+      ASSERT_TRUE(db_->GetProperty(DB::Properties::kFastBlockCacheEntryStats,
+                                   &value_str));
+      ASSERT_EQ(scan_count, 5);
+
       ASSERT_TRUE(db_->GetProperty(DB::Properties::kCFStats, &value_str));
       // To match historical speed, querying this property no longer triggers
       // a scan, even if results are old. But periodic dump stats should keep
       // things reasonably updated.
-      ASSERT_EQ(scan_count, /*unchanged*/ 3);
+      ASSERT_EQ(scan_count, /*unchanged*/ 5);
 
       SyncPoint::GetInstance()->DisableProcessing();
       SyncPoint::GetInstance()->ClearAllCallBacks();
@@ -1525,6 +1534,124 @@ TEST_F(DBBlockCacheTest, CacheEntryRoleStats) {
   }
 }
 
+namespace {
+
+void DummyFillCache(Cache& cache, size_t entry_size,
+                    std::vector<CacheHandleGuard<void>>& handles) {
+  // fprintf(stderr, "Entry size: %zu\n", entry_size);
+  handles.clear();
+  cache.EraseUnRefEntries();
+  void* fake_value = &cache;
+  size_t capacity = cache.GetCapacity();
+  OffsetableCacheKey ck{"abc", "abc", 42};
+  for (size_t my_usage = 0; my_usage < capacity;) {
+    size_t charge = std::min(entry_size, capacity - my_usage);
+    Cache::Handle* handle;
+    Status st = cache.Insert(ck.WithOffset(my_usage).AsSlice(), fake_value,
+                             charge, /*deleter*/ nullptr, &handle);
+    ASSERT_OK(st);
+    handles.emplace_back(&cache, handle);
+    my_usage += charge;
+  }
+}
+
+class CountingLogger : public Logger {
+ public:
+  ~CountingLogger() override {}
+  using Logger::Logv;
+  void Logv(const InfoLogLevel log_level, const char* format,
+            va_list /*ap*/) override {
+    if (std::strstr(format, "HyperClockCache") == nullptr) {
+      // Not a match
+      return;
+    }
+    // static StderrLogger debug;
+    // debug.Logv(log_level, format, ap);
+    if (log_level == InfoLogLevel::INFO_LEVEL) {
+      ++info_count_;
+    } else if (log_level == InfoLogLevel::WARN_LEVEL) {
+      ++warn_count_;
+    } else if (log_level == InfoLogLevel::ERROR_LEVEL) {
+      ++error_count_;
+    }
+  }
+
+  std::array<int, 3> PopCounts() {
+    std::array<int, 3> rv{{info_count_, warn_count_, error_count_}};
+    info_count_ = warn_count_ = error_count_ = 0;
+    return rv;
+  }
+
+ private:
+  int info_count_{};
+  int warn_count_{};
+  int error_count_{};
+};
+
+}  // namespace
+
+TEST_F(DBBlockCacheTest, HyperClockCacheReportProblems) {
+  size_t capacity = 1024 * 1024;
+  size_t value_size_est = 8 * 1024;
+  HyperClockCacheOptions hcc_opts{capacity, value_size_est};
+  hcc_opts.num_shard_bits = 2;  // 4 shards
+  hcc_opts.metadata_charge_policy = kDontChargeCacheMetadata;
+  std::shared_ptr<Cache> cache = hcc_opts.MakeSharedCache();
+  std::shared_ptr<CountingLogger> logger = std::make_shared<CountingLogger>();
+
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  table_options.block_cache = cache;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  options.info_log = logger;
+  // Going to sample more directly
+  options.stats_dump_period_sec = 0;
+  Reopen(options);
+
+  std::vector<CacheHandleGuard<void>> handles;
+
+  // Clear anything from DB startup
+  logger->PopCounts();
+
+  // Fill cache based on expected size and check that when we
+  // don't report anything relevant in periodic stats dump
+  DummyFillCache(*cache, value_size_est, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 0, 0}}));
+
+  // Same, within reasonable bounds
+  DummyFillCache(*cache, value_size_est - value_size_est / 4, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 0, 0}}));
+
+  DummyFillCache(*cache, value_size_est + value_size_est / 3, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 0, 0}}));
+
+  // Estimate too high (value size too low) eventually reports ERROR
+  DummyFillCache(*cache, value_size_est / 2, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 1, 0}}));
+
+  DummyFillCache(*cache, value_size_est / 3, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 0, 1}}));
+
+  // Estimate too low (value size too high) starts with INFO
+  // and is only WARNING in the worst case
+  DummyFillCache(*cache, value_size_est * 2, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{1, 0, 0}}));
+
+  DummyFillCache(*cache, value_size_est * 3, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 1, 0}}));
+
+  DummyFillCache(*cache, value_size_est * 20, handles);
+  dbfull()->DumpStats();
+  EXPECT_EQ(logger->PopCounts(), (std::array<int, 3>{{0, 1, 0}}));
+}
+
 #endif  // ROCKSDB_LITE
 
 class DBBlockCacheKeyTest
@@ -1955,7 +2082,7 @@ struct CacheKeyDecoder {
                                    DownwardInvolution(decoded_session_counter));
   }
 };
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(CacheKeyTest, Encodings) {
   // This test primarily verifies this claim from cache_key.cc:

db/db_bloom_filter_test.cc

@@ -43,7 +43,7 @@ const std::string kStandard128Ribbon =
     test::Standard128RibbonFilterPolicy::kClassName();
 const std::string kAutoBloom = BloomFilterPolicy::kClassName();
 const std::string kAutoRibbon = RibbonFilterPolicy::kClassName();
-}  // namespace
+}  // anonymous namespace
 
 // DB tests related to bloom filter.
 
@@ -622,7 +622,7 @@ class AlwaysTrueFilterPolicy : public ReadOnlyBuiltinFilterPolicy {
   bool skip_;
 };
 
-}  // namespace
+}  // anonymous namespace
 
 TEST_P(DBBloomFilterTestWithParam, SkipFilterOnEssentiallyZeroBpk) {
   constexpr int maxKey = 10;
@@ -767,10 +767,10 @@ INSTANTIATE_TEST_CASE_P(
 
 INSTANTIATE_TEST_CASE_P(
     FormatLatest, DBBloomFilterTestWithParam,
-    ::testing::Values(
-        std::make_tuple(kAutoBloom, true, kLatestFormatVersion),
-        std::make_tuple(kAutoBloom, false, kLatestFormatVersion),
-        std::make_tuple(kAutoRibbon, false, kLatestFormatVersion)));
+    ::testing::Values(std::make_tuple(kAutoBloom, true, kLatestFormatVersion),
+                      std::make_tuple(kAutoBloom, false, kLatestFormatVersion),
+                      std::make_tuple(kAutoRibbon, false,
+                                      kLatestFormatVersion)));
 #endif  // !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 
 TEST_F(DBBloomFilterTest, BloomFilterRate) {
@@ -840,7 +840,7 @@ std::vector<CompatibilityConfig> kCompatibilityConfigs = {
      BlockBasedTableOptions().format_version},
     {kCompatibilityRibbonPolicy, true, BlockBasedTableOptions().format_version},
 };
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(DBBloomFilterTest, BloomFilterCompatibility) {
   Options options = CurrentOptions();
@@ -1678,7 +1678,7 @@ class TestingContextCustomFilterPolicy
  private:
   mutable std::string test_report_;
 };
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(DBBloomFilterTest, ContextCustomFilterPolicy) {
   auto policy = std::make_shared<TestingContextCustomFilterPolicy>(15, 8, 5);
@@ -1706,7 +1706,7 @@ TEST_F(DBBloomFilterTest, ContextCustomFilterPolicy) {
     ASSERT_OK(Put(1, Key(maxKey + 55555), Key(maxKey + 55555)));
     Flush(1);
     EXPECT_EQ(policy->DumpTestReport(),
-              fifo ? "cf=abe,s=kCompactionStyleFIFO,n=1,l=0,b=0,r=kFlush\n"
+              fifo ? "cf=abe,s=kCompactionStyleFIFO,n=7,l=0,b=0,r=kFlush\n"
                    : "cf=bob,s=kCompactionStyleLevel,n=7,l=0,b=0,r=kFlush\n");
 
     for (int i = maxKey / 2; i < maxKey; i++) {
@@ -1714,7 +1714,7 @@ TEST_F(DBBloomFilterTest, ContextCustomFilterPolicy) {
     }
     Flush(1);
     EXPECT_EQ(policy->DumpTestReport(),
-              fifo ? "cf=abe,s=kCompactionStyleFIFO,n=1,l=0,b=0,r=kFlush\n"
+              fifo ? "cf=abe,s=kCompactionStyleFIFO,n=7,l=0,b=0,r=kFlush\n"
                    : "cf=bob,s=kCompactionStyleLevel,n=7,l=0,b=0,r=kFlush\n");
 
     // Check that they can be found
@@ -1738,7 +1738,7 @@ TEST_F(DBBloomFilterTest, ContextCustomFilterPolicy) {
       EXPECT_LE(useful_count, maxKey * 2 * (fifo ? 0.9995 : 0.98));
     }
 
-    if (!fifo) {  // FIFO only has L0
+    if (!fifo) {  // FIFO doesn't fully support CompactRange
       // Full compaction
       ASSERT_OK(db_->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
                                   nullptr));
@@ -2186,16 +2186,14 @@ INSTANTIATE_TEST_CASE_P(DBBloomFilterTestVaryPrefixAndFormatVer,
                             std::make_tuple(false, 2),
                             std::make_tuple(false, 3),
                             std::make_tuple(false, 4),
-                            std::make_tuple(false, 5),
-                            std::make_tuple(true, 2),
-                            std::make_tuple(true, 3),
-                            std::make_tuple(true, 4),
+                            std::make_tuple(false, 5), std::make_tuple(true, 2),
+                            std::make_tuple(true, 3), std::make_tuple(true, 4),
                             std::make_tuple(true, 5)));
 
 #ifndef ROCKSDB_LITE
 namespace {
 static const std::string kPlainTable = "test_PlainTableBloom";
-}  // namespace
+}  // anonymous namespace
 
 class BloomStatsTestWithParam
     : public DBBloomFilterTest,
@@ -2408,7 +2406,7 @@ void PrefixScanInit(DBBloomFilterTest* dbtest) {
     dbtest->Flush();
   }
 }
-}  // namespace
+}  // anonymous namespace
 
 TEST_F(DBBloomFilterTest, PrefixScan) {
   while (ChangeFilterOptions()) {
@@ -2496,7 +2494,7 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
   for (int i = 0; i < numkeys; i += 2) {
     keys.push_back(i);
   }
-  RandomShuffle(std::begin(keys), std::end(keys));
+  RandomShuffle(std::begin(keys), std::end(keys), /*seed*/ 42);
   int num_inserted = 0;
   for (int key : keys) {
     ASSERT_OK(Put(1, Key(key), "val"));
@@ -3169,7 +3167,7 @@ std::pair<uint64_t, uint64_t> CheckedAndUseful(uint64_t checked,
                                                uint64_t useful) {
   return {checked, useful};
 }
-}  // namespace
+}  // anonymous namespace
 
 // This uses a prefix_extractor + comparator combination that violates
 // one of the old obsolete, unnecessary axioms of prefix extraction:
@@ -3377,7 +3375,7 @@ class NonIdempotentFixed4Transform : public SliceTransform {
 
   bool InDomain(const Slice& src) const override { return src.size() >= 5; }
 };
-}  // namespace
+}  // anonymous namespace
 
 // This uses a prefix_extractor + comparator combination that violates
 // two of the old obsolete, unnecessary axioms of prefix extraction:

db/db_compaction_test.cc

@@ -248,9 +248,8 @@ Options DeletionTriggerOptions(Options options) {
   return options;
 }
 
-bool HaveOverlappingKeyRanges(
-    const Comparator* c,
-    const SstFileMetaData& a, const SstFileMetaData& b) {
+bool HaveOverlappingKeyRanges(const Comparator* c, const SstFileMetaData& a,
+                              const SstFileMetaData& b) {
   if (c->CompareWithoutTimestamp(a.smallestkey, b.smallestkey) >= 0) {
     if (c->CompareWithoutTimestamp(a.smallestkey, b.largestkey) <= 0) {
       // b.smallestkey <= a.smallestkey <= b.largestkey
@@ -275,18 +274,15 @@ bool HaveOverlappingKeyRanges(
 // Identifies all files between level "min_level" and "max_level"
 // which has overlapping key range with "input_file_meta".
 void GetOverlappingFileNumbersForLevelCompaction(
-    const ColumnFamilyMetaData& cf_meta,
-    const Comparator* comparator,
-    int min_level, int max_level,
-    const SstFileMetaData* input_file_meta,
+    const ColumnFamilyMetaData& cf_meta, const Comparator* comparator,
+    int min_level, int max_level, const SstFileMetaData* input_file_meta,
     std::set<std::string>* overlapping_file_names) {
   std::set<const SstFileMetaData*> overlapping_files;
   overlapping_files.insert(input_file_meta);
   for (int m = min_level; m <= max_level; ++m) {
     for (auto& file : cf_meta.levels[m].files) {
       for (auto* included_file : overlapping_files) {
-        if (HaveOverlappingKeyRanges(
-                comparator, *included_file, file)) {
+        if (HaveOverlappingKeyRanges(comparator, *included_file, file)) {
           overlapping_files.insert(&file);
           overlapping_file_names->insert(file.name);
           break;
@@ -309,12 +305,9 @@ void VerifyCompactionResult(
 #endif
 }
 
-const SstFileMetaData* PickFileRandomly(
-    const ColumnFamilyMetaData& cf_meta,
-    Random* rand,
-    int* level = nullptr) {
-  auto file_id = rand->Uniform(static_cast<int>(
-      cf_meta.file_count)) + 1;
+const SstFileMetaData* PickFileRandomly(const ColumnFamilyMetaData& cf_meta,
+                                        Random* rand, int* level = nullptr) {
+  auto file_id = rand->Uniform(static_cast<int>(cf_meta.file_count)) + 1;
   for (auto& level_meta : cf_meta.levels) {
     if (file_id <= level_meta.files.size()) {
       if (level != nullptr) {
@@ -740,7 +733,6 @@ TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
   }
 }
 
-
 TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
   const int kNumKeysPerFile = 100;
 
@@ -883,7 +875,7 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
 
 TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {
   Options options = CurrentOptions();
-  options.write_buffer_size = 100000000;        // Large write buffer
+  options.write_buffer_size = 100000000;  // Large write buffer
   options.max_subcompactions = max_subcompactions_;
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -1069,7 +1061,7 @@ TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
   compact_opt.compression = kNoCompression;
   compact_opt.output_file_size_limit = 4096;
   const size_t key_len =
-    static_cast<size_t>(compact_opt.output_file_size_limit) / 5;
+      static_cast<size_t>(compact_opt.output_file_size_limit) / 5;
 
   DestroyAndReopen(options);
 
@@ -1247,14 +1239,8 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
   DestroyAndReopen(options);
   // non overlapping ranges
   std::vector<std::pair<int32_t, int32_t>> ranges = {
-    {100, 199},
-    {300, 399},
-    {0, 99},
-    {200, 299},
-    {600, 699},
-    {400, 499},
-    {500, 550},
-    {551, 599},
+      {100, 199}, {300, 399}, {0, 99},    {200, 299},
+      {600, 699}, {400, 499}, {500, 550}, {551, 599},
   };
   int32_t value_size = 10 * 1024;  // 10 KB
 
@@ -1297,14 +1283,15 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
   DestroyAndReopen(options);
   // Same ranges as above but overlapping
   ranges = {
-    {100, 199},
-    {300, 399},
-    {0, 99},
-    {200, 299},
-    {600, 699},
-    {400, 499},
-    {500, 560},  // this range overlap with the next one
-    {551, 599},
+      {100, 199},
+      {300, 399},
+      {0, 99},
+      {200, 299},
+      {600, 699},
+      {400, 499},
+      {500, 560},  // this range overlap with the next
+                   // one
+      {551, 599},
   };
   for (size_t i = 0; i < ranges.size(); i++) {
     for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
@@ -1907,7 +1894,7 @@ TEST_F(DBCompactionTest, DeleteFilesInRanges) {
   ASSERT_EQ("0,0,10", FilesPerLevel(0));
 
   // file [0 => 100), [200 => 300), ... [800, 900)
-  for (auto i = 0; i < 10; i+=2) {
+  for (auto i = 0; i < 10; i += 2) {
     for (auto j = 0; j < 100; j++) {
       auto k = i * 100 + j;
       ASSERT_OK(Put(Key(k), values[k]));
@@ -2349,14 +2336,14 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionCFPathUse) {
   cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_2", 4 * 1024 * 1024);
   cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_3", 1024 * 1024 * 1024);
   option_vector.emplace_back(DBOptions(options), cf_opt1);
-  CreateColumnFamilies({"one"},option_vector[1]);
+  CreateColumnFamilies({"one"}, option_vector[1]);
 
   // Configure CF2 specific paths.
   cf_opt2.cf_paths.emplace_back(dbname_ + "cf2", 500 * 1024);
   cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_2", 4 * 1024 * 1024);
   cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_3", 1024 * 1024 * 1024);
   option_vector.emplace_back(DBOptions(options), cf_opt2);
-  CreateColumnFamilies({"two"},option_vector[2]);
+  CreateColumnFamilies({"two"}, option_vector[2]);
 
   ReopenWithColumnFamilies({"default", "one", "two"}, option_vector);
 
@@ -2729,7 +2716,6 @@ TEST_P(DBCompactionTestWithParam, ManualCompaction) {
   }
 }
 
-
 TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {
   Options options = CurrentOptions();
   options.db_paths.emplace_back(dbname_ + "_2", 2 * 10485760);
@@ -2866,14 +2852,13 @@ TEST_P(DBCompactionTestWithParam, DISABLED_CompactFilesOnLevelCompaction) {
       auto file_meta = PickFileRandomly(cf_meta, &rnd, &level);
       compaction_input_file_names.push_back(file_meta->name);
       GetOverlappingFileNumbersForLevelCompaction(
-          cf_meta, options.comparator, level, output_level,
-          file_meta, &overlapping_file_names);
+          cf_meta, options.comparator, level, output_level, file_meta,
+          &overlapping_file_names);
     }
 
-    ASSERT_OK(dbfull()->CompactFiles(
-        CompactionOptions(), handles_[1],
-        compaction_input_file_names,
-        output_level));
+    ASSERT_OK(dbfull()->CompactFiles(CompactionOptions(), handles_[1],
+                                     compaction_input_file_names,
+                                     output_level));
 
     // Make sure all overlapping files do not exist after compaction
     dbfull()->GetColumnFamilyMetaData(handles_[1], &cf_meta);
@@ -2896,8 +2881,7 @@ TEST_P(DBCompactionTestWithParam, PartialCompactionFailure) {
   options.write_buffer_size = kKeysPerBuffer * kKvSize;
   options.max_write_buffer_number = 2;
   options.target_file_size_base =
-      options.write_buffer_size *
-      (options.max_write_buffer_number - 1);
+      options.write_buffer_size * (options.max_write_buffer_number - 1);
   options.level0_file_num_compaction_trigger = kNumL1Files;
   options.max_bytes_for_level_base =
       options.level0_file_num_compaction_trigger *
@@ -2917,10 +2901,9 @@ TEST_P(DBCompactionTestWithParam, PartialCompactionFailure) {
 
   DestroyAndReopen(options);
 
-  const int kNumInsertedKeys =
-      options.level0_file_num_compaction_trigger *
-      (options.max_write_buffer_number - 1) *
-      kKeysPerBuffer;
+  const int kNumInsertedKeys = options.level0_file_num_compaction_trigger *
+                               (options.max_write_buffer_number - 1) *
+                               kKeysPerBuffer;
 
   Random rnd(301);
   std::vector<std::string> keys;
@@ -3536,6 +3519,59 @@ TEST_P(DBCompactionTestWithParam, FullCompactionInBottomPriThreadPool) {
   Env::Default()->SetBackgroundThreads(0, Env::Priority::BOTTOM);
 }
 
+TEST_F(DBCompactionTest, CancelCompactionWaitingOnConflict) {
+  // This test verifies cancellation of a compaction waiting to be scheduled due
+  // to conflict with a running compaction.
+  //
+  // A `CompactRange()` in universal compacts all files, waiting for files to
+  // become available if they are locked for another compaction. This test
+  // triggers an automatic compaction that blocks a `CompactRange()`, and
+  // verifies that `DisableManualCompaction()` can successfully cancel the
+  // `CompactRange()` without waiting for the automatic compaction to finish.
+  const int kNumSortedRuns = 4;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.level0_file_num_compaction_trigger = kNumSortedRuns;
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+  Reopen(options);
+
+  test::SleepingBackgroundTask auto_compaction_sleeping_task;
+  // Block automatic compaction when it runs in the callback
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::Run():Start",
+      [&](void* /*arg*/) { auto_compaction_sleeping_task.DoSleep(); });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Fill overlapping files in L0 to trigger an automatic compaction
+  Random rnd(301);
+  for (int i = 0; i < kNumSortedRuns; ++i) {
+    int key_idx = 0;
+    GenerateNewFile(&rnd, &key_idx, true /* nowait */);
+  }
+  auto_compaction_sleeping_task.WaitUntilSleeping();
+
+  // Make sure the manual compaction has seen the conflict before being canceled
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
+      {{"ColumnFamilyData::CompactRange:Return",
+        "DBCompactionTest::CancelCompactionWaitingOnConflict:"
+        "PreDisableManualCompaction"}});
+  auto manual_compaction_thread = port::Thread([this]() {
+    ASSERT_TRUE(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr)
+                    .IsIncomplete());
+  });
+
+  // Cancel it. Thread should be joinable, i.e., manual compaction was unblocked
+  // despite finding a conflict with an automatic compaction that is still
+  // running
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CancelCompactionWaitingOnConflict:"
+      "PreDisableManualCompaction");
+  db_->DisableManualCompaction();
+  manual_compaction_thread.join();
+}
+
 TEST_F(DBCompactionTest, OptimizedDeletionObsoleting) {
   // Deletions can be dropped when compacted to non-last level if they fall
   // outside the lower-level files' key-ranges.
@@ -3618,9 +3654,8 @@ TEST_F(DBCompactionTest, CompactFilesPendingL0Bug) {
   ASSERT_EQ(kNumL0Files, cf_meta.levels[0].files.size());
   std::vector<std::string> input_filenames;
   input_filenames.push_back(cf_meta.levels[0].files.front().name);
-  ASSERT_OK(dbfull()
-                  ->CompactFiles(CompactionOptions(), input_filenames,
-                                 0 /* output_level */));
+  ASSERT_OK(dbfull()->CompactFiles(CompactionOptions(), input_filenames,
+                                   0 /* output_level */));
   TEST_SYNC_POINT("DBCompactionTest::CompactFilesPendingL0Bug:ManualCompacted");
   ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
 }
@@ -3824,6 +3859,242 @@ TEST_F(DBCompactionTest, NoCompactBottomLevelFilesWithDeletions) {
   }
 }
 
+TEST_F(DBCompactionTest, RoundRobinTtlCompactionNormal) {
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.level0_file_num_compaction_trigger = 20;
+  options.ttl = 24 * 60 * 60;  // 24 hours
+  options.compaction_pri = kRoundRobin;
+  env_->now_cpu_count_.store(0);
+  env_->SetMockSleep();
+  options.env = env_;
+
+  // add a small second for each wait time, to make sure the file is expired
+  int small_seconds = 1;
+
+  std::atomic_int ttl_compactions{0};
+  std::atomic_int round_robin_ttl_compactions{0};
+  std::atomic_int other_compactions{0};
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        Compaction* compaction = reinterpret_cast<Compaction*>(arg);
+        auto compaction_reason = compaction->compaction_reason();
+        if (compaction_reason == CompactionReason::kTtl) {
+          ttl_compactions++;
+        } else if (compaction_reason == CompactionReason::kRoundRobinTtl) {
+          round_robin_ttl_compactions++;
+        } else {
+          other_compactions++;
+        }
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  DestroyAndReopen(options);
+
+  // Setup the files from lower level to up level, each file is 1 hour's older
+  // than the next one.
+  // create 10 files on the last level (L6)
+  for (int i = 0; i < 10; i++) {
+    for (int j = 0; j < 100; j++) {
+      ASSERT_OK(Put(Key(i * 100 + j), "value" + std::to_string(i * 100 + j)));
+    }
+    ASSERT_OK(Flush());
+    env_->MockSleepForSeconds(60 * 60);  // generate 1 file per hour
+  }
+  MoveFilesToLevel(6);
+
+  // create 5 files on L5
+  for (int i = 0; i < 5; i++) {
+    for (int j = 0; j < 200; j++) {
+      ASSERT_OK(Put(Key(i * 200 + j), "value" + std::to_string(i * 200 + j)));
+    }
+    ASSERT_OK(Flush());
+    env_->MockSleepForSeconds(60 * 60);
+  }
+  MoveFilesToLevel(5);
+
+  // create 3 files on L4
+  for (int i = 0; i < 3; i++) {
+    for (int j = 0; j < 300; j++) {
+      ASSERT_OK(Put(Key(i * 300 + j), "value" + std::to_string(i * 300 + j)));
+    }
+    ASSERT_OK(Flush());
+    env_->MockSleepForSeconds(60 * 60);
+  }
+  MoveFilesToLevel(4);
+
+  // The LSM tree should be like:
+  // L4: [0,            299], [300,      599], [600,     899]
+  // L5: [0,  199]      [200,         399]...............[800, 999]
+  // L6: [0,99][100,199][200,299][300,399]...............[800,899][900,999]
+  ASSERT_EQ("0,0,0,0,3,5,10", FilesPerLevel());
+
+  // make sure the first L5 file is expired
+  env_->MockSleepForSeconds(16 * 60 * 60 + small_seconds++);
+
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(4), "value" + std::to_string(1)));
+  ASSERT_OK(Put(Key(5), "value" + std::to_string(1)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // verify there's a RoundRobin TTL compaction
+  ASSERT_EQ(1, round_robin_ttl_compactions);
+  round_robin_ttl_compactions = 0;
+
+  // expire 2 more files
+  env_->MockSleepForSeconds(2 * 60 * 60 + small_seconds++);
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(4), "value" + std::to_string(2)));
+  ASSERT_OK(Put(Key(5), "value" + std::to_string(2)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  ASSERT_EQ(2, round_robin_ttl_compactions);
+  round_robin_ttl_compactions = 0;
+
+  // expire 4 more files, 2 out of 3 files on L4 are expired
+  env_->MockSleepForSeconds(4 * 60 * 60 + small_seconds++);
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(6), "value" + std::to_string(3)));
+  ASSERT_OK(Put(Key(7), "value" + std::to_string(3)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  ASSERT_EQ(1, NumTableFilesAtLevel(4));
+  ASSERT_EQ(0, NumTableFilesAtLevel(5));
+
+  ASSERT_GT(round_robin_ttl_compactions, 0);
+  round_robin_ttl_compactions = 0;
+
+  // make the first L0 file expired, which triggers a normal TTL compaction
+  // instead of roundrobin TTL compaction, it will also include an extra file
+  // from L0 because of overlap
+  ASSERT_EQ(0, ttl_compactions);
+  env_->MockSleepForSeconds(19 * 60 * 60 + small_seconds++);
+
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(6), "value" + std::to_string(4)));
+  ASSERT_OK(Put(Key(7), "value" + std::to_string(4)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // L0 -> L1 compaction is normal TTL compaction, L1 -> next levels compactions
+  // are RoundRobin TTL compaction.
+  ASSERT_GT(ttl_compactions, 0);
+  ttl_compactions = 0;
+  ASSERT_GT(round_robin_ttl_compactions, 0);
+  round_robin_ttl_compactions = 0;
+
+  // All files are expired, so only the last level has data
+  env_->MockSleepForSeconds(24 * 60 * 60);
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(6), "value" + std::to_string(4)));
+  ASSERT_OK(Put(Key(7), "value" + std::to_string(4)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
+
+  ASSERT_GT(ttl_compactions, 0);
+  ttl_compactions = 0;
+  ASSERT_GT(round_robin_ttl_compactions, 0);
+  round_robin_ttl_compactions = 0;
+
+  ASSERT_EQ(0, other_compactions);
+}
+
+TEST_F(DBCompactionTest, RoundRobinTtlCompactionUnsortedTime) {
+  // This is to test the case that the RoundRobin compaction cursor not pointing
+  // to the oldest file, RoundRobin compaction should still compact the file
+  // after cursor until all expired files are compacted.
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.level0_file_num_compaction_trigger = 20;
+  options.ttl = 24 * 60 * 60;  // 24 hours
+  options.compaction_pri = kRoundRobin;
+  env_->now_cpu_count_.store(0);
+  env_->SetMockSleep();
+  options.env = env_;
+
+  std::atomic_int ttl_compactions{0};
+  std::atomic_int round_robin_ttl_compactions{0};
+  std::atomic_int other_compactions{0};
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        Compaction* compaction = reinterpret_cast<Compaction*>(arg);
+        auto compaction_reason = compaction->compaction_reason();
+        if (compaction_reason == CompactionReason::kTtl) {
+          ttl_compactions++;
+        } else if (compaction_reason == CompactionReason::kRoundRobinTtl) {
+          round_robin_ttl_compactions++;
+        } else {
+          other_compactions++;
+        }
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  DestroyAndReopen(options);
+
+  // create 10 files on the last level (L6)
+  for (int i = 0; i < 10; i++) {
+    for (int j = 0; j < 100; j++) {
+      ASSERT_OK(Put(Key(i * 100 + j), "value" + std::to_string(i * 100 + j)));
+    }
+    ASSERT_OK(Flush());
+    env_->MockSleepForSeconds(60 * 60);  // generate 1 file per hour
+  }
+  MoveFilesToLevel(6);
+
+  // create 5 files on L5
+  for (int i = 0; i < 5; i++) {
+    for (int j = 0; j < 200; j++) {
+      ASSERT_OK(Put(Key(i * 200 + j), "value" + std::to_string(i * 200 + j)));
+    }
+    ASSERT_OK(Flush());
+    env_->MockSleepForSeconds(60 * 60);  // 1 hour
+  }
+  MoveFilesToLevel(5);
+
+  // The LSM tree should be like:
+  // L5: [0,  199]      [200,         399] [400,599] [600,799] [800, 999]
+  // L6: [0,99][100,199][200,299][300,399]....................[800,899][900,999]
+  ASSERT_EQ("0,0,0,0,0,5,10", FilesPerLevel());
+
+  // point the compaction cursor to the 4th file on L5
+  VersionSet* const versions = dbfull()->GetVersionSet();
+  assert(versions);
+  ColumnFamilyData* const cfd = versions->GetColumnFamilySet()->GetDefault();
+  ASSERT_NE(cfd, nullptr);
+  Version* const current = cfd->current();
+  ASSERT_NE(current, nullptr);
+  VersionStorageInfo* storage_info = current->storage_info();
+  ASSERT_NE(storage_info, nullptr);
+  const InternalKey split_cursor = InternalKey(Key(600), 100000, kTypeValue);
+  storage_info->AddCursorForOneLevel(5, split_cursor);
+
+  // make the first file on L5 expired, there should be 3 TTL compactions:
+  // 4th one, 5th one, then 1st one.
+  env_->MockSleepForSeconds(19 * 60 * 60 + 1);
+  // trigger TTL compaction
+  ASSERT_OK(Put(Key(6), "value" + std::to_string(4)));
+  ASSERT_OK(Put(Key(7), "value" + std::to_string(4)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  ASSERT_EQ(2, NumTableFilesAtLevel(5));
+
+  ASSERT_EQ(3, round_robin_ttl_compactions);
+  ASSERT_EQ(0, ttl_compactions);
+  ASSERT_EQ(0, other_compactions);
+}
+
 TEST_F(DBCompactionTest, LevelCompactExpiredTtlFiles) {
   const int kNumKeysPerFile = 32;
   const int kNumLevelFiles = 2;
@@ -4267,9 +4538,9 @@ TEST_F(DBCompactionTest, LevelPeriodicAndTtlCompaction) {
   const int kValueSize = 100;
 
   Options options = CurrentOptions();
-  options.ttl = 10 * 60 * 60;  // 10 hours
+  options.ttl = 10 * 60 * 60;                          // 10 hours
   options.periodic_compaction_seconds = 48 * 60 * 60;  // 2 days
-  options.max_open_files = -1;   // needed for both periodic and ttl compactions
+  options.max_open_files = -1;  // needed for both periodic and ttl compactions
   env_->SetMockSleep();
   options.env = env_;
 
@@ -4691,7 +4962,7 @@ TEST_F(DBCompactionTest, CompactRangeSkipFlushAfterDelay) {
        {"DBImpl::FlushMemTable:StallWaitDone", "CompactionJob::Run():End"}});
   ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
 
-  //used for the delayable flushes
+  // used for the delayable flushes
   FlushOptions flush_opts;
   flush_opts.allow_write_stall = true;
   for (int i = 0; i < kNumL0FilesLimit - 1; ++i) {
@@ -4710,7 +4981,8 @@ TEST_F(DBCompactionTest, CompactRangeSkipFlushAfterDelay) {
   ASSERT_OK(Put(std::to_string(0), rnd.RandomString(1024)));
   ASSERT_OK(dbfull()->Flush(flush_opts));
   ASSERT_OK(Put(std::to_string(0), rnd.RandomString(1024)));
-  TEST_SYNC_POINT("DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush");
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush");
   manual_compaction_thread.join();
 
   // If CompactRange's flush was skipped, the final Put above will still be
@@ -5003,10 +5275,10 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
   }
 
   std::shared_ptr<ConcurrentTaskLimiter> unique_limiter(
-    NewConcurrentTaskLimiter("unique_limiter", -1));
+      NewConcurrentTaskLimiter("unique_limiter", -1));
 
-  const char* cf_names[] = {"default", "0", "1", "2", "3", "4", "5",
-    "6", "7", "8", "9", "a", "b", "c", "d", "e", "f" };
+  const char* cf_names[] = {"default", "0", "1", "2", "3", "4", "5", "6", "7",
+                            "8",       "9", "a", "b", "c", "d", "e", "f"};
   const unsigned int cf_count = sizeof cf_names / sizeof cf_names[0];
 
   std::unordered_map<std::string, CompactionLimiter*> cf_to_limiter;
@@ -5018,10 +5290,10 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
   options.level0_file_num_compaction_trigger = 4;
   options.level0_slowdown_writes_trigger = 64;
   options.level0_stop_writes_trigger = 64;
-  options.max_background_jobs = kMaxBackgroundThreads; // Enough threads
+  options.max_background_jobs = kMaxBackgroundThreads;  // Enough threads
   options.memtable_factory.reset(
       test::NewSpecialSkipListFactory(kNumKeysPerFile));
-  options.max_write_buffer_number = 10; // Enough memtables
+  options.max_write_buffer_number = 10;  // Enough memtables
   DestroyAndReopen(options);
 
   std::vector<Options> option_vector;
@@ -5049,9 +5321,8 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
     CreateColumnFamilies({cf_names[cf]}, option_vector[cf]);
   }
 
-  ReopenWithColumnFamilies(std::vector<std::string>(cf_names,
-                                                    cf_names + cf_count),
-                           option_vector);
+  ReopenWithColumnFamilies(
+      std::vector<std::string>(cf_names, cf_names + cf_count), option_vector);
 
   port::Mutex mutex;
 
@@ -5113,7 +5384,7 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
   // Enough L0 files to trigger compaction
   for (unsigned int cf = 0; cf < cf_count; cf++) {
     ASSERT_EQ(NumTableFilesAtLevel(0, cf),
-      options.level0_file_num_compaction_trigger);
+              options.level0_file_num_compaction_trigger);
   }
 
   // Create more files for one column family, which triggers speed up
@@ -5156,7 +5427,7 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
 
   // flush one more file to cf 1
   for (int i = 0; i < kNumKeysPerFile; i++) {
-      ASSERT_OK(Put(cf_test, Key(keyIndex++), ""));
+    ASSERT_OK(Put(cf_test, Key(keyIndex++), ""));
   }
   // put extra key to trigger flush
   ASSERT_OK(Put(cf_test, "", ""));
@@ -5191,9 +5462,7 @@ TEST_P(DBCompactionDirectIOTest, DirectIO) {
       });
   if (options.use_direct_io_for_flush_and_compaction) {
     SyncPoint::GetInstance()->SetCallBack(
-        "SanitizeOptions:direct_io", [&](void* /*arg*/) {
-          readahead = true;
-        });
+        "SanitizeOptions:direct_io", [&](void* /*arg*/) { readahead = true; });
   }
   SyncPoint::GetInstance()->EnableProcessing();
   CreateAndReopenWithCF({"pikachu"}, options);
@@ -7958,6 +8227,42 @@ TEST_F(DBCompactionTest, BottomPriCompactionCountsTowardConcurrencyLimit) {
   compact_range_thread.join();
 }
 
+TEST_F(DBCompactionTest, BottommostFileCompactionAllowIngestBehind) {
+  // allow_ingest_behind prevents seqnum zeroing, and could cause
+  // compaction loop with reason kBottommostFiles.
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.compaction_style = kCompactionStyleLevel;
+  options.allow_ingest_behind = true;
+  options.comparator = BytewiseComparator();
+  DestroyAndReopen(options);
+
+  WriteOptions write_opts;
+  ASSERT_OK(db_->Put(write_opts, "infinite", "compaction loop"));
+  ASSERT_OK(db_->Put(write_opts, "infinite", "loop"));
+
+  ASSERT_OK(Flush());
+  MoveFilesToLevel(1);
+  ASSERT_OK(db_->Put(write_opts, "bumpseqnum", ""));
+  ASSERT_OK(Flush());
+  auto snapshot = db_->GetSnapshot();
+  // Bump up oldest_snapshot_seqnum_ in VersionStorageInfo.
+  db_->ReleaseSnapshot(snapshot);
+  bool compacted = false;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* /* arg */) {
+        // There should not be a compaction.
+        compacted = true;
+      });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+  // Wait for compaction to be scheduled.
+  env_->SleepForMicroseconds(2000000);
+  ASSERT_FALSE(compacted);
+  // The following assert can be used to check for compaction loop:
+  // it used to wait forever before the fix.
+  // ASSERT_OK(dbfull()->TEST_WaitForCompact(true /* wait_unscheduled */));
+}
+
 #endif  // !defined(ROCKSDB_LITE)
 
 }  // namespace ROCKSDB_NAMESPACE
@@ -7968,8 +8273,8 @@ int main(int argc, char** argv) {
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 #else
-  (void) argc;
-  (void) argv;
+  (void)argc;
+  (void)argv;
   return 0;
 #endif
 }

db/db_dynamic_level_test.cc

@@ -500,8 +500,8 @@ int main(int argc, char** argv) {
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 #else
-  (void) argc;
-  (void) argv;
+  (void)argc;
+  (void)argv;
   return 0;
 #endif
 }

db/db_encryption_test.cc

@@ -43,7 +43,7 @@ TEST_F(DBEncryptionTest, CheckEncrypted) {
 
   Env* target = GetTargetEnv();
   int hits = 0;
-  for (auto it = fileNames.begin() ; it != fileNames.end(); ++it) {
+  for (auto it = fileNames.begin(); it != fileNames.end(); ++it) {
     if (*it == "LOCK") {
       continue;
     }
@@ -64,24 +64,24 @@ TEST_F(DBEncryptionTest, CheckEncrypted) {
     ASSERT_OK(status);
 
     if (data.ToString().find("foo567") != std::string::npos) {
-      hits++; 
-      //std::cout << "Hit in " << filePath << "\n";
+      hits++;
+      // std::cout << "Hit in " << filePath << "\n";
     }
     if (data.ToString().find("v1.fetdq") != std::string::npos) {
-      hits++; 
-      //std::cout << "Hit in " << filePath << "\n";
+      hits++;
+      // std::cout << "Hit in " << filePath << "\n";
     }
     if (data.ToString().find("bar123") != std::string::npos) {
-      hits++; 
-      //std::cout << "Hit in " << filePath << "\n";
+      hits++;
+      // std::cout << "Hit in " << filePath << "\n";
     }
     if (data.ToString().find("v2.dfgkjdfghsd") != std::string::npos) {
-      hits++; 
-      //std::cout << "Hit in " << filePath << "\n";
+      hits++;
+      // std::cout << "Hit in " << filePath << "\n";
     }
     if (data.ToString().find("dfgk") != std::string::npos) {
-      hits++; 
-      //std::cout << "Hit in " << filePath << "\n";
+      hits++;
+      // std::cout << "Hit in " << filePath << "\n";
     }
   }
   if (encrypted_env_) {
@@ -119,7 +119,7 @@ TEST_F(DBEncryptionTest, ReadEmptyFile) {
   ASSERT_TRUE(data.empty());
 }
 
-#endif // ROCKSDB_LITE
+#endif  // ROCKSDB_LITE
 
 }  // namespace ROCKSDB_NAMESPACE
 

db/db_filesnapshot.cc

@@ -65,8 +65,7 @@ Status DBImpl::FlushForGetLiveFiles() {
 }
 
 Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
-                            uint64_t* manifest_file_size,
-                            bool flush_memtable) {
+                            uint64_t* manifest_file_size, bool flush_memtable) {
   *manifest_file_size = 0;
 
   mutex_.Lock();

db/db_flush_test.cc

@@ -57,7 +57,7 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
   Reopen(options);
   FlushOptions no_wait;
   no_wait.wait = false;
-  no_wait.allow_write_stall=true;
+  no_wait.allow_write_stall = true;
 
   SyncPoint::GetInstance()->LoadDependency(
       {{"VersionSet::LogAndApply:WriteManifest",
@@ -746,6 +746,64 @@ class TestFlushListener : public EventListener {
 };
 #endif  // !ROCKSDB_LITE
 
+// RocksDB lite does not support GetLiveFiles()
+#ifndef ROCKSDB_LITE
+TEST_F(DBFlushTest, FixFlushReasonRaceFromConcurrentFlushes) {
+  Options options = CurrentOptions();
+  options.atomic_flush = true;
+  options.disable_auto_compactions = true;
+  CreateAndReopenWithCF({"cf1"}, options);
+
+  for (int idx = 0; idx < 1; ++idx) {
+    ASSERT_OK(Put(0, Key(idx), std::string(1, 'v')));
+    ASSERT_OK(Put(1, Key(idx), std::string(1, 'v')));
+  }
+
+  // To coerce a manual flush happenning in the middle of GetLiveFiles's flush,
+  // we need to pause background flush thread and enable it later.
+  std::shared_ptr<test::SleepingBackgroundTask> sleeping_task =
+      std::make_shared<test::SleepingBackgroundTask>();
+  env_->SetBackgroundThreads(1, Env::HIGH);
+  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                 sleeping_task.get(), Env::Priority::HIGH);
+  sleeping_task->WaitUntilSleeping();
+
+  // Coerce a manual flush happenning in the middle of GetLiveFiles's flush
+  bool get_live_files_paused_at_sync_point = false;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::AtomicFlushMemTables:AfterScheduleFlush", [&](void* /* arg */) {
+        if (get_live_files_paused_at_sync_point) {
+          // To prevent non-GetLiveFiles() flush from pausing at this sync point
+          return;
+        }
+        get_live_files_paused_at_sync_point = true;
+
+        FlushOptions fo;
+        fo.wait = false;
+        fo.allow_write_stall = true;
+        ASSERT_OK(dbfull()->Flush(fo));
+
+        // Resume background flush thread so GetLiveFiles() can finish
+        sleeping_task->WakeUp();
+        sleeping_task->WaitUntilDone();
+      });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<std::string> files;
+  uint64_t manifest_file_size;
+  // Before the fix, a race condition on default cf's flush reason due to
+  // concurrent GetLiveFiles's flush and manual flush will fail
+  // an internal assertion.
+  // After the fix, such race condition is fixed and there is no assertion
+  // failure.
+  ASSERT_OK(db_->GetLiveFiles(files, &manifest_file_size, /*flush*/ true));
+  ASSERT_TRUE(get_live_files_paused_at_sync_point);
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+}
+#endif  // !ROCKSDB_LITE
+
 TEST_F(DBFlushTest, MemPurgeBasic) {
   Options options = CurrentOptions();
 
@@ -1822,8 +1880,8 @@ TEST_F(DBFlushTest, ManualFlushFailsInReadOnlyMode) {
   ASSERT_NOK(dbfull()->TEST_WaitForFlushMemTable());
 #ifndef ROCKSDB_LITE
   uint64_t num_bg_errors;
-  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBackgroundErrors,
-                                  &num_bg_errors));
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBackgroundErrors, &num_bg_errors));
   ASSERT_GT(num_bg_errors, 0);
 #endif  // ROCKSDB_LITE
 
@@ -2440,7 +2498,9 @@ TEST_P(DBAtomicFlushTest, ManualFlushUnder2PC) {
   options.atomic_flush = GetParam();
   // 64MB so that memtable flush won't be trigger by the small writes.
   options.write_buffer_size = (static_cast<size_t>(64) << 20);
-
+  auto flush_listener = std::make_shared<FlushCounterListener>();
+  flush_listener->expected_flush_reason = FlushReason::kManualFlush;
+  options.listeners.push_back(flush_listener);
   // Destroy the DB to recreate as a TransactionDB.
   Close();
   Destroy(options, true);
@@ -2507,7 +2567,6 @@ TEST_P(DBAtomicFlushTest, ManualFlushUnder2PC) {
     auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
     ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
     ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
-    ASSERT_EQ(cfh->cfd()->GetFlushReason(), FlushReason::kManualFlush);
   }
 
   // The recovered min log number with prepared data should be non-zero.
@@ -2520,13 +2579,15 @@ TEST_P(DBAtomicFlushTest, ManualFlushUnder2PC) {
   ASSERT_TRUE(db_impl->allow_2pc());
   ASSERT_NE(db_impl->MinLogNumberToKeep(), 0);
 }
-#endif  // ROCKSDB_LITE
 
 TEST_P(DBAtomicFlushTest, ManualAtomicFlush) {
   Options options = CurrentOptions();
   options.create_if_missing = true;
   options.atomic_flush = GetParam();
   options.write_buffer_size = (static_cast<size_t>(64) << 20);
+  auto flush_listener = std::make_shared<FlushCounterListener>();
+  flush_listener->expected_flush_reason = FlushReason::kManualFlush;
+  options.listeners.push_back(flush_listener);
 
   CreateAndReopenWithCF({"pikachu", "eevee"}, options);
   size_t num_cfs = handles_.size();
@@ -2551,11 +2612,11 @@ TEST_P(DBAtomicFlushTest, ManualAtomicFlush) {
 
   for (size_t i = 0; i != num_cfs; ++i) {
     auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
-    ASSERT_EQ(cfh->cfd()->GetFlushReason(), FlushReason::kManualFlush);
     ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
     ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
   }
 }
+#endif  // ROCKSDB_LITE
 
 TEST_P(DBAtomicFlushTest, PrecomputeMinLogNumberToKeepNon2PC) {
   Options options = CurrentOptions();
@@ -3032,6 +3093,44 @@ TEST_P(DBAtomicFlushTest, BgThreadNoWaitAfterManifestError) {
   SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
+TEST_P(DBAtomicFlushTest, NoWaitWhenWritesStopped) {
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = GetParam();
+  options.max_write_buffer_number = 2;
+  options.memtable_factory.reset(test::NewSpecialSkipListFactory(1));
+
+  Reopen(options);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::DelayWrite:Start",
+        "DBAtomicFlushTest::NoWaitWhenWritesStopped:0"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(dbfull()->PauseBackgroundWork());
+  for (int i = 0; i < options.max_write_buffer_number; ++i) {
+    ASSERT_OK(Put("k" + std::to_string(i), "v" + std::to_string(i)));
+  }
+  std::thread stalled_writer([&]() { ASSERT_OK(Put("k", "v")); });
+
+  TEST_SYNC_POINT("DBAtomicFlushTest::NoWaitWhenWritesStopped:0");
+
+  {
+    FlushOptions flush_opts;
+    flush_opts.wait = false;
+    flush_opts.allow_write_stall = true;
+    ASSERT_TRUE(db_->Flush(flush_opts).IsTryAgain());
+  }
+
+  ASSERT_OK(dbfull()->ContinueBackgroundWork());
+  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+
+  stalled_writer.join();
+
+  SyncPoint::GetInstance()->DisableProcessing();
+}
+
 INSTANTIATE_TEST_CASE_P(DBFlushDirectIOTest, DBFlushDirectIOTest,
                         testing::Bool());
 

db/db_impl/compacted_db_impl.cc

@@ -26,16 +26,16 @@ CompactedDBImpl::CompactedDBImpl(const DBOptions& options,
       version_(nullptr),
       user_comparator_(nullptr) {}
 
-CompactedDBImpl::~CompactedDBImpl() {
-}
+CompactedDBImpl::~CompactedDBImpl() {}
 
 size_t CompactedDBImpl::FindFile(const Slice& key) {
   size_t right = files_.num_files - 1;
   auto cmp = [&](const FdWithKeyRange& f, const Slice& k) -> bool {
     return user_comparator_->Compare(ExtractUserKey(f.largest_key), k) < 0;
   };
-  return static_cast<size_t>(std::lower_bound(files_.files,
-                            files_.files + right, key, cmp) - files_.files);
+  return static_cast<size_t>(
+      std::lower_bound(files_.files, files_.files + right, key, cmp) -
+      files_.files);
 }
 
 Status CompactedDBImpl::Get(const ReadOptions& options, ColumnFamilyHandle*,
@@ -228,8 +228,8 @@ Status CompactedDBImpl::Init(const Options& options) {
   return Status::NotSupported("no file exists");
 }
 
-Status CompactedDBImpl::Open(const Options& options,
-                             const std::string& dbname, DB** dbptr) {
+Status CompactedDBImpl::Open(const Options& options, const std::string& dbname,
+                             DB** dbptr) {
   *dbptr = nullptr;
 
   if (options.max_open_files != -1) {

db/db_impl/compacted_db_impl.h

@@ -7,6 +7,7 @@
 #ifndef ROCKSDB_LITE
 #include <string>
 #include <vector>
+
 #include "db/db_impl/db_impl.h"
 
 namespace ROCKSDB_NAMESPACE {

db/db_impl/db_impl.cc

@@ -604,7 +604,7 @@ Status DBImpl::CloseHelper() {
 
   while (!flush_queue_.empty()) {
     const FlushRequest& flush_req = PopFirstFromFlushQueue();
-    for (const auto& iter : flush_req) {
+    for (const auto& iter : flush_req.cfd_to_max_mem_id_to_persist) {
       iter.first->UnrefAndTryDelete();
     }
   }
@@ -847,11 +847,13 @@ Status DBImpl::RegisterRecordSeqnoTimeWorker() {
     InstrumentedMutexLock l(&mutex_);
 
     for (auto cfd : *versions_->GetColumnFamilySet()) {
-      uint64_t preclude_last_option =
-          cfd->ioptions()->preclude_last_level_data_seconds;
-      if (!cfd->IsDropped() && preclude_last_option > 0) {
-        min_time_duration = std::min(preclude_last_option, min_time_duration);
-        max_time_duration = std::max(preclude_last_option, max_time_duration);
+      // preserve time is the max of 2 options.
+      uint64_t preserve_time_duration =
+          std::max(cfd->ioptions()->preserve_internal_time_seconds,
+                   cfd->ioptions()->preclude_last_level_data_seconds);
+      if (!cfd->IsDropped() && preserve_time_duration > 0) {
+        min_time_duration = std::min(preserve_time_duration, min_time_duration);
+        max_time_duration = std::max(preserve_time_duration, max_time_duration);
       }
     }
     if (min_time_duration == std::numeric_limits<uint64_t>::max()) {
@@ -863,8 +865,11 @@ Status DBImpl::RegisterRecordSeqnoTimeWorker() {
 
   uint64_t seqno_time_cadence = 0;
   if (min_time_duration != std::numeric_limits<uint64_t>::max()) {
+    // round up to 1 when the time_duration is smaller than
+    // kMaxSeqnoTimePairsPerCF
     seqno_time_cadence =
-        min_time_duration / SeqnoToTimeMapping::kMaxSeqnoTimePairsPerCF;
+        (min_time_duration + SeqnoToTimeMapping::kMaxSeqnoTimePairsPerCF - 1) /
+        SeqnoToTimeMapping::kMaxSeqnoTimePairsPerCF;
   }
 
   Status s;
@@ -1054,16 +1059,31 @@ void DBImpl::DumpStats() {
     return;
   }
 
+  // Also probe block cache(s) for problems, dump to info log
+  UnorderedSet<Cache*> probed_caches;
   TEST_SYNC_POINT("DBImpl::DumpStats:StartRunning");
   {
     InstrumentedMutexLock l(&mutex_);
     for (auto cfd : versions_->GetRefedColumnFamilySet()) {
-      if (cfd->initialized()) {
-        // Release DB mutex for gathering cache entry stats. Pass over all
-        // column families for this first so that other stats are dumped
-        // near-atomically.
-        InstrumentedMutexUnlock u(&mutex_);
-        cfd->internal_stats()->CollectCacheEntryStats(/*foreground=*/false);
+      if (!cfd->initialized()) {
+        continue;
+      }
+
+      // Release DB mutex for gathering cache entry stats. Pass over all
+      // column families for this first so that other stats are dumped
+      // near-atomically.
+      InstrumentedMutexUnlock u(&mutex_);
+      cfd->internal_stats()->CollectCacheEntryStats(/*foreground=*/false);
+
+      // Probe block cache for problems (if not already via another CF)
+      if (immutable_db_options_.info_log) {
+        auto* table_factory = cfd->ioptions()->table_factory.get();
+        assert(table_factory != nullptr);
+        Cache* cache =
+            table_factory->GetOptions<Cache>(TableFactory::kBlockCacheOpts());
+        if (cache && probed_caches.insert(cache).second) {
+          cache->ReportProblems(immutable_db_options_.info_log);
+        }
       }
     }
 
@@ -1074,18 +1094,7 @@ void DBImpl::DumpStats() {
     default_cf_internal_stats_->GetStringProperty(*property_info, *property,
                                                   &stats);
 
-    property = &DB::Properties::kCFStatsNoFileHistogram;
-    property_info = GetPropertyInfo(*property);
-    assert(property_info != nullptr);
-    assert(!property_info->need_out_of_mutex);
-    for (auto cfd : *versions_->GetColumnFamilySet()) {
-      if (cfd->initialized()) {
-        cfd->internal_stats()->GetStringProperty(*property_info, *property,
-                                                 &stats);
-      }
-    }
-
-    property = &DB::Properties::kCFFileHistogram;
+    property = &InternalStats::kPeriodicCFStats;
     property_info = GetPropertyInfo(*property);
     assert(property_info != nullptr);
     assert(!property_info->need_out_of_mutex);
@@ -1352,7 +1361,7 @@ Status DBImpl::SetDBOptions(
       file_options_for_compaction_ = fs_->OptimizeForCompactionTableWrite(
           file_options_for_compaction_, immutable_db_options_);
       versions_->ChangeFileOptions(mutable_db_options_);
-      //TODO(xiez): clarify why apply optimize for read to write options
+      // TODO(xiez): clarify why apply optimize for read to write options
       file_options_for_compaction_ = fs_->OptimizeForCompactionTableRead(
           file_options_for_compaction_, immutable_db_options_);
       file_options_for_compaction_.compaction_readahead_size =
@@ -1438,11 +1447,11 @@ Status DBImpl::FlushWAL(bool sync) {
       // future writes
       IOStatusCheck(io_s);
       // whether sync or not, we should abort the rest of function upon error
-      return std::move(io_s);
+      return static_cast<Status>(io_s);
     }
     if (!sync) {
       ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "FlushWAL sync=false");
-      return std::move(io_s);
+      return static_cast<Status>(io_s);
     }
   }
   if (!sync) {
@@ -1453,6 +1462,18 @@ Status DBImpl::FlushWAL(bool sync) {
   return SyncWAL();
 }
 
+bool DBImpl::WALBufferIsEmpty(bool lock) {
+  if (lock) {
+    log_write_mutex_.Lock();
+  }
+  log::Writer* cur_log_writer = logs_.back().writer;
+  auto res = cur_log_writer->BufferIsEmpty();
+  if (lock) {
+    log_write_mutex_.Unlock();
+  }
+  return res;
+}
+
 Status DBImpl::SyncWAL() {
   TEST_SYNC_POINT("DBImpl::SyncWAL:Begin");
   autovector<log::Writer*, 1> logs_to_sync;
@@ -1551,7 +1572,7 @@ Status DBImpl::ApplyWALToManifest(VersionEdit* synced_wals) {
 Status DBImpl::LockWAL() {
   log_write_mutex_.Lock();
   auto cur_log_writer = logs_.back().writer;
-  auto status = cur_log_writer->WriteBuffer();
+  IOStatus status = cur_log_writer->WriteBuffer();
   if (!status.ok()) {
     ROCKS_LOG_ERROR(immutable_db_options_.info_log, "WAL flush error %s",
                     status.ToString().c_str());
@@ -1559,7 +1580,7 @@ Status DBImpl::LockWAL() {
     // future writes
     WriteStatusCheck(status);
   }
-  return std::move(status);
+  return static_cast<Status>(status);
 }
 
 Status DBImpl::UnlockWAL() {
@@ -1802,32 +1823,33 @@ InternalIterator* DBImpl::NewInternalIterator(
   MergeIteratorBuilder merge_iter_builder(
       &cfd->internal_comparator(), arena,
       !read_options.total_order_seek &&
-          super_version->mutable_cf_options.prefix_extractor != nullptr);
-  // Collect iterator for mutable mem
-  merge_iter_builder.AddIterator(
-      super_version->mem->NewIterator(read_options, arena));
+          super_version->mutable_cf_options.prefix_extractor != nullptr,
+      read_options.iterate_upper_bound);
+  // Collect iterator for mutable memtable
+  auto mem_iter = super_version->mem->NewIterator(read_options, arena);
   Status s;
   if (!read_options.ignore_range_deletions) {
+    TruncatedRangeDelIterator* mem_tombstone_iter = nullptr;
     auto range_del_iter = super_version->mem->NewRangeTombstoneIterator(
         read_options, sequence, false /* immutable_memtable */);
     if (range_del_iter == nullptr || range_del_iter->empty()) {
       delete range_del_iter;
-      merge_iter_builder.AddRangeTombstoneIterator(nullptr);
     } else {
-      merge_iter_builder.AddRangeTombstoneIterator(
-          new TruncatedRangeDelIterator(
-              std::unique_ptr<FragmentedRangeTombstoneIterator>(range_del_iter),
-              &cfd->ioptions()->internal_comparator, nullptr /* smallest */,
-              nullptr /* largest */));
+      mem_tombstone_iter = new TruncatedRangeDelIterator(
+          std::unique_ptr<FragmentedRangeTombstoneIterator>(range_del_iter),
+          &cfd->ioptions()->internal_comparator, nullptr /* smallest */,
+          nullptr /* largest */);
     }
+    merge_iter_builder.AddPointAndTombstoneIterator(mem_iter,
+                                                    mem_tombstone_iter);
+  } else {
+    merge_iter_builder.AddIterator(mem_iter);
   }
+
   // Collect all needed child iterators for immutable memtables
   if (s.ok()) {
-    super_version->imm->AddIterators(read_options, &merge_iter_builder);
-    if (!read_options.ignore_range_deletions) {
-      s = super_version->imm->AddRangeTombstoneIterators(read_options, arena,
-                                                         merge_iter_builder);
-    }
+    super_version->imm->AddIterators(read_options, &merge_iter_builder,
+                                     !read_options.ignore_range_deletions);
   }
   TEST_SYNC_POINT_CALLBACK("DBImpl::NewInternalIterator:StatusCallback", &s);
   if (s.ok()) {
@@ -2340,8 +2362,8 @@ std::vector<Status> DBImpl::MultiGet(
     std::string* timestamp = timestamps ? &(*timestamps)[keys_read] : nullptr;
 
     LookupKey lkey(keys[keys_read], consistent_seqnum, read_options.timestamp);
-    auto cfh =
-        static_cast_with_check<ColumnFamilyHandleImpl>(column_family[keys_read]);
+    auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(
+        column_family[keys_read]);
     SequenceNumber max_covering_tombstone_seq = 0;
     auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
     assert(mgd_iter != multiget_cf_data.end());
@@ -3091,7 +3113,8 @@ Status DBImpl::CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
     }
   }  // InstrumentedMutexLock l(&mutex_)
 
-  if (cf_options.preclude_last_level_data_seconds > 0) {
+  if (cf_options.preserve_internal_time_seconds > 0 ||
+      cf_options.preclude_last_level_data_seconds > 0) {
     s = RegisterRecordSeqnoTimeWorker();
   }
   sv_context.Clean();
@@ -3182,7 +3205,8 @@ Status DBImpl::DropColumnFamilyImpl(ColumnFamilyHandle* column_family) {
     bg_cv_.SignalAll();
   }
 
-  if (cfd->ioptions()->preclude_last_level_data_seconds > 0) {
+  if (cfd->ioptions()->preserve_internal_time_seconds > 0 ||
+      cfd->ioptions()->preclude_last_level_data_seconds > 0) {
     s = RegisterRecordSeqnoTimeWorker();
   }
 
@@ -3662,14 +3686,16 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
     if (oldest_snapshot > bottommost_files_mark_threshold_) {
       CfdList cf_scheduled;
       for (auto* cfd : *versions_->GetColumnFamilySet()) {
-        cfd->current()->storage_info()->UpdateOldestSnapshot(oldest_snapshot);
-        if (!cfd->current()
-                 ->storage_info()
-                 ->BottommostFilesMarkedForCompaction()
-                 .empty()) {
-          SchedulePendingCompaction(cfd);
-          MaybeScheduleFlushOrCompaction();
-          cf_scheduled.push_back(cfd);
+        if (!cfd->ioptions()->allow_ingest_behind) {
+          cfd->current()->storage_info()->UpdateOldestSnapshot(oldest_snapshot);
+          if (!cfd->current()
+                   ->storage_info()
+                   ->BottommostFilesMarkedForCompaction()
+                   .empty()) {
+            SchedulePendingCompaction(cfd);
+            MaybeScheduleFlushOrCompaction();
+            cf_scheduled.push_back(cfd);
+          }
         }
       }
 
@@ -3678,7 +3704,8 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
       // mutex might be unlocked during the loop, making the result inaccurate.
       SequenceNumber new_bottommost_files_mark_threshold = kMaxSequenceNumber;
       for (auto* cfd : *versions_->GetColumnFamilySet()) {
-        if (CfdListContains(cf_scheduled, cfd)) {
+        if (CfdListContains(cf_scheduled, cfd) ||
+            cfd->ioptions()->allow_ingest_behind) {
           continue;
         }
         new_bottommost_files_mark_threshold = std::min(
@@ -3961,8 +3988,7 @@ SuperVersion* DBImpl::GetAndRefSuperVersion(uint32_t column_family_id) {
 void DBImpl::CleanupSuperVersion(SuperVersion* sv) {
   // Release SuperVersion
   if (sv->Unref()) {
-    bool defer_purge =
-            immutable_db_options().avoid_unnecessary_blocking_io;
+    bool defer_purge = immutable_db_options().avoid_unnecessary_blocking_io;
     {
       InstrumentedMutexLock l(&mutex_);
       sv->Cleanup();
@@ -4391,10 +4417,14 @@ Status DBImpl::CheckConsistency() {
       }
       files_by_directory[md.db_path].push_back(fname);
     }
+
+    IOOptions io_opts;
+    io_opts.do_not_recurse = true;
     for (const auto& dir_files : files_by_directory) {
       std::string directory = dir_files.first;
       std::vector<std::string> existing_files;
-      Status s = env_->GetChildren(directory, &existing_files);
+      Status s = fs_->GetChildren(directory, io_opts, &existing_files,
+                                  /*IODebugContext*=*/nullptr);
       if (!s.ok()) {
         corruption_messages +=
             "Can't list files in " + directory + ": " + s.ToString() + "\n";
@@ -4576,8 +4606,12 @@ Status DestroyDB(const std::string& dbname, const Options& options,
   // Reset the logger because it holds a handle to the
   // log file and prevents cleanup and directory removal
   soptions.info_log.reset();
+  IOOptions io_opts;
   // Ignore error in case directory does not exist
-  env->GetChildren(dbname, &filenames).PermitUncheckedError();
+  soptions.fs
+      ->GetChildren(dbname, io_opts, &filenames,
+                    /*IODebugContext*=*/nullptr)
+      .PermitUncheckedError();
 
   FileLock* lock;
   const std::string lockname = LockFileName(dbname);
@@ -4617,8 +4651,12 @@ Status DestroyDB(const std::string& dbname, const Options& options,
         paths.insert(cf_path.path);
       }
     }
+
     for (const auto& path : paths) {
-      if (env->GetChildren(path, &filenames).ok()) {
+      if (soptions.fs
+              ->GetChildren(path, io_opts, &filenames,
+                            /*IODebugContext*=*/nullptr)
+              .ok()) {
         for (const auto& fname : filenames) {
           if (ParseFileName(fname, &number, &type) &&
               (type == kTableFile ||
@@ -4640,7 +4678,11 @@ Status DestroyDB(const std::string& dbname, const Options& options,
     std::string archivedir = ArchivalDirectory(dbname);
     bool wal_dir_exists = false;
     if (!soptions.IsWalDirSameAsDBPath(dbname)) {
-      wal_dir_exists = env->GetChildren(soptions.wal_dir, &walDirFiles).ok();
+      wal_dir_exists =
+          soptions.fs
+              ->GetChildren(soptions.wal_dir, io_opts, &walDirFiles,
+                            /*IODebugContext*=*/nullptr)
+              .ok();
       archivedir = ArchivalDirectory(soptions.wal_dir);
     }
 
@@ -4648,7 +4690,10 @@ Status DestroyDB(const std::string& dbname, const Options& options,
     // processed and removed before those otherwise we have issues
     // removing them
     std::vector<std::string> archiveFiles;
-    if (env->GetChildren(archivedir, &archiveFiles).ok()) {
+    if (soptions.fs
+            ->GetChildren(archivedir, io_opts, &archiveFiles,
+                          /*IODebugContext*=*/nullptr)
+            .ok()) {
       // Delete archival files.
       for (const auto& file : archiveFiles) {
         if (ParseFileName(file, &number, &type) && type == kWalFile) {
@@ -4789,7 +4834,10 @@ Status DBImpl::DeleteObsoleteOptionsFiles() {
   // to the oldest.
   std::map<uint64_t, std::string> options_filenames;
   Status s;
-  s = GetEnv()->GetChildren(GetName(), &filenames);
+  IOOptions io_opts;
+  io_opts.do_not_recurse = true;
+  s = fs_->GetChildren(GetName(), io_opts, &filenames,
+                       /*IODebugContext*=*/nullptr);
   if (!s.ok()) {
     return s;
   }
@@ -5247,7 +5295,7 @@ Status DBImpl::IngestExternalFiles(
         mutex_.Unlock();
         status = AtomicFlushMemTables(cfds_to_flush, flush_opts,
                                       FlushReason::kExternalFileIngestion,
-                                      true /* writes_stopped */);
+                                      true /* entered_write_thread */);
         mutex_.Lock();
       } else {
         for (size_t i = 0; i != num_cfs; ++i) {
@@ -5258,7 +5306,7 @@ Status DBImpl::IngestExternalFiles(
                     ->cfd();
             status = FlushMemTable(cfd, flush_opts,
                                    FlushReason::kExternalFileIngestion,
-                                   true /* writes_stopped */);
+                                   true /* entered_write_thread */);
             mutex_.Lock();
             if (!status.ok()) {
               break;
@@ -5624,8 +5672,7 @@ Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
     }
   }
 
-  bool defer_purge =
-          immutable_db_options().avoid_unnecessary_blocking_io;
+  bool defer_purge = immutable_db_options().avoid_unnecessary_blocking_io;
   {
     InstrumentedMutexLock l(&mutex_);
     for (auto sv : sv_list) {
@@ -5738,8 +5785,24 @@ Status DBImpl::NewDefaultReplayer(
 Status DBImpl::StartBlockCacheTrace(
     const TraceOptions& trace_options,
     std::unique_ptr<TraceWriter>&& trace_writer) {
-  return block_cache_tracer_.StartTrace(immutable_db_options_.clock,
-                                        trace_options, std::move(trace_writer));
+  BlockCacheTraceOptions block_trace_opts;
+  block_trace_opts.sampling_frequency = trace_options.sampling_frequency;
+
+  BlockCacheTraceWriterOptions trace_writer_opt;
+  trace_writer_opt.max_trace_file_size = trace_options.max_trace_file_size;
+
+  std::unique_ptr<BlockCacheTraceWriter> block_cache_trace_writer =
+      NewBlockCacheTraceWriter(env_->GetSystemClock().get(), trace_writer_opt,
+                               std::move(trace_writer));
+
+  return block_cache_tracer_.StartTrace(block_trace_opts,
+                                        std::move(block_cache_trace_writer));
+}
+
+Status DBImpl::StartBlockCacheTrace(
+    const BlockCacheTraceOptions& trace_options,
+    std::unique_ptr<BlockCacheTraceWriter>&& trace_writer) {
+  return block_cache_tracer_.StartTrace(trace_options, std::move(trace_writer));
 }
 
 Status DBImpl::EndBlockCacheTrace() {

db/db_impl/db_impl.h

@@ -16,6 +16,7 @@
 #include <map>
 #include <set>
 #include <string>
+#include <unordered_map>
 #include <utility>
 #include <vector>
 
@@ -204,6 +205,9 @@ class DBImpl : public DB {
   using DB::Merge;
   Status Merge(const WriteOptions& options, ColumnFamilyHandle* column_family,
                const Slice& key, const Slice& value) override;
+  Status Merge(const WriteOptions& options, ColumnFamilyHandle* column_family,
+               const Slice& key, const Slice& ts, const Slice& value) override;
+
   using DB::Delete;
   Status Delete(const WriteOptions& options, ColumnFamilyHandle* column_family,
                 const Slice& key) override;
@@ -222,6 +226,9 @@ class DBImpl : public DB {
   Status DeleteRange(const WriteOptions& options,
                      ColumnFamilyHandle* column_family, const Slice& begin_key,
                      const Slice& end_key) override;
+  Status DeleteRange(const WriteOptions& options,
+                     ColumnFamilyHandle* column_family, const Slice& begin_key,
+                     const Slice& end_key, const Slice& ts) override;
 
   using DB::Write;
   virtual Status Write(const WriteOptions& options,
@@ -423,7 +430,7 @@ class DBImpl : public DB {
       const FlushOptions& options,
       const std::vector<ColumnFamilyHandle*>& column_families) override;
   virtual Status FlushWAL(bool sync) override;
-  bool TEST_WALBufferIsEmpty(bool lock = true);
+  bool WALBufferIsEmpty(bool lock = true);
   virtual Status SyncWAL() override;
   virtual Status LockWAL() override;
   virtual Status UnlockWAL() override;
@@ -565,9 +572,13 @@ class DBImpl : public DB {
 
   using DB::StartBlockCacheTrace;
   Status StartBlockCacheTrace(
-      const TraceOptions& options,
+      const TraceOptions& trace_options,
       std::unique_ptr<TraceWriter>&& trace_writer) override;
 
+  Status StartBlockCacheTrace(
+      const BlockCacheTraceOptions& options,
+      std::unique_ptr<BlockCacheTraceWriter>&& trace_writer) override;
+
   using DB::EndBlockCacheTrace;
   Status EndBlockCacheTrace() override;
 
@@ -1373,7 +1384,7 @@ class DBImpl : public DB {
 
   void NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
                           const MutableCFOptions& mutable_cf_options,
-                          int job_id);
+                          int job_id, FlushReason flush_reason);
 
   void NotifyOnFlushCompleted(
       ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
@@ -1665,12 +1676,17 @@ class DBImpl : public DB {
   // Argument required by background flush thread.
   struct BGFlushArg {
     BGFlushArg()
-        : cfd_(nullptr), max_memtable_id_(0), superversion_context_(nullptr) {}
+        : cfd_(nullptr),
+          max_memtable_id_(0),
+          superversion_context_(nullptr),
+          flush_reason_(FlushReason::kOthers) {}
     BGFlushArg(ColumnFamilyData* cfd, uint64_t max_memtable_id,
-               SuperVersionContext* superversion_context)
+               SuperVersionContext* superversion_context,
+               FlushReason flush_reason)
         : cfd_(cfd),
           max_memtable_id_(max_memtable_id),
-          superversion_context_(superversion_context) {}
+          superversion_context_(superversion_context),
+          flush_reason_(flush_reason) {}
 
     // Column family to flush.
     ColumnFamilyData* cfd_;
@@ -1681,6 +1697,7 @@ class DBImpl : public DB {
     // installs a new superversion for the column family. This operation
     // requires a SuperVersionContext object (currently embedded in JobContext).
     SuperVersionContext* superversion_context_;
+    FlushReason flush_reason_;
   };
 
   // Argument passed to flush thread.
@@ -1809,7 +1826,7 @@ class DBImpl : public DB {
   // installs a new super version for the column family.
   Status FlushMemTableToOutputFile(
       ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
-      bool* madeProgress, JobContext* job_context,
+      bool* madeProgress, JobContext* job_context, FlushReason flush_reason,
       SuperVersionContext* superversion_context,
       std::vector<SequenceNumber>& snapshot_seqs,
       SequenceNumber earliest_write_conflict_snapshot,
@@ -1877,12 +1894,13 @@ class DBImpl : public DB {
 
   // Force current memtable contents to be flushed.
   Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options,
-                       FlushReason flush_reason, bool writes_stopped = false);
+                       FlushReason flush_reason,
+                       bool entered_write_thread = false);
 
   Status AtomicFlushMemTables(
       const autovector<ColumnFamilyData*>& column_family_datas,
       const FlushOptions& options, FlushReason flush_reason,
-      bool writes_stopped = false);
+      bool entered_write_thread = false);
 
   // Wait until flushing this column family won't stall writes
   Status WaitUntilFlushWouldNotStallWrites(ColumnFamilyData* cfd,
@@ -2026,18 +2044,22 @@ class DBImpl : public DB {
 
   void MaybeScheduleFlushOrCompaction();
 
-  // A flush request specifies the column families to flush as well as the
-  // largest memtable id to persist for each column family. Once all the
-  // memtables whose IDs are smaller than or equal to this per-column-family
-  // specified value, this flush request is considered to have completed its
-  // work of flushing this column family. After completing the work for all
-  // column families in this request, this flush is considered complete.
-  using FlushRequest = std::vector<std::pair<ColumnFamilyData*, uint64_t>>;
+  struct FlushRequest {
+    FlushReason flush_reason;
+    // A map from column family to flush to largest memtable id to persist for
+    // each column family. Once all the memtables whose IDs are smaller than or
+    // equal to this per-column-family specified value, this flush request is
+    // considered to have completed its work of flushing this column family.
+    // After completing the work for all column families in this request, this
+    // flush is considered complete.
+    std::unordered_map<ColumnFamilyData*, uint64_t>
+        cfd_to_max_mem_id_to_persist;
+  };
 
   void GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
-                            FlushRequest* req);
+                            FlushReason flush_reason, FlushRequest* req);
 
-  void SchedulePendingFlush(const FlushRequest& req, FlushReason flush_reason);
+  void SchedulePendingFlush(const FlushRequest& req);
 
   void SchedulePendingCompaction(ColumnFamilyData* cfd);
   void SchedulePendingPurge(std::string fname, std::string dir_to_sync,

db/db_impl/db_impl_compaction_flush.cc

@@ -155,7 +155,7 @@ IOStatus DBImpl::SyncClosedLogs(JobContext* job_context,
 
 Status DBImpl::FlushMemTableToOutputFile(
     ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
-    bool* made_progress, JobContext* job_context,
+    bool* made_progress, JobContext* job_context, FlushReason flush_reason,
     SuperVersionContext* superversion_context,
     std::vector<SequenceNumber>& snapshot_seqs,
     SequenceNumber earliest_write_conflict_snapshot,
@@ -215,7 +215,8 @@ Status DBImpl::FlushMemTableToOutputFile(
       dbname_, cfd, immutable_db_options_, mutable_cf_options, max_memtable_id,
       file_options_for_compaction_, versions_.get(), &mutex_, &shutting_down_,
       snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
-      job_context, log_buffer, directories_.GetDbDir(), GetDataDir(cfd, 0U),
+      job_context, flush_reason, log_buffer, directories_.GetDbDir(),
+      GetDataDir(cfd, 0U),
       GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
       &event_logger_, mutable_cf_options.report_bg_io_stats,
       true /* sync_output_directory */, true /* write_manifest */, thread_pri,
@@ -260,7 +261,8 @@ Status DBImpl::FlushMemTableToOutputFile(
 
 #ifndef ROCKSDB_LITE
   // may temporarily unlock and lock the mutex.
-  NotifyOnFlushBegin(cfd, &file_meta, mutable_cf_options, job_context->job_id);
+  NotifyOnFlushBegin(cfd, &file_meta, mutable_cf_options, job_context->job_id,
+                     flush_reason);
 #endif  // ROCKSDB_LITE
 
   bool switched_to_mempurge = false;
@@ -390,8 +392,9 @@ Status DBImpl::FlushMemTablesToOutputFiles(
   MutableCFOptions mutable_cf_options_copy = *cfd->GetLatestMutableCFOptions();
   SuperVersionContext* superversion_context =
       bg_flush_arg.superversion_context_;
+  FlushReason flush_reason = bg_flush_arg.flush_reason_;
   Status s = FlushMemTableToOutputFile(
-      cfd, mutable_cf_options_copy, made_progress, job_context,
+      cfd, mutable_cf_options_copy, made_progress, job_context, flush_reason,
       superversion_context, snapshot_seqs, earliest_write_conflict_snapshot,
       snapshot_checker, log_buffer, thread_pri);
   return s;
@@ -420,7 +423,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
   for (const auto cfd : cfds) {
     assert(cfd->imm()->NumNotFlushed() != 0);
     assert(cfd->imm()->IsFlushPending());
-    assert(cfd->GetFlushReason() == cfds[0]->GetFlushReason());
+  }
+  for (const auto bg_flush_arg : bg_flush_args) {
+    assert(bg_flush_arg.flush_reason_ == bg_flush_args[0].flush_reason_);
   }
 #endif /* !NDEBUG */
 
@@ -459,13 +464,15 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     all_mutable_cf_options.emplace_back(*cfd->GetLatestMutableCFOptions());
     const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back();
     uint64_t max_memtable_id = bg_flush_args[i].max_memtable_id_;
+    FlushReason flush_reason = bg_flush_args[i].flush_reason_;
     jobs.emplace_back(new FlushJob(
         dbname_, cfd, immutable_db_options_, mutable_cf_options,
         max_memtable_id, file_options_for_compaction_, versions_.get(), &mutex_,
         &shutting_down_, snapshot_seqs, earliest_write_conflict_snapshot,
-        snapshot_checker, job_context, log_buffer, directories_.GetDbDir(),
-        data_dir, GetCompressionFlush(*cfd->ioptions(), mutable_cf_options),
-        stats_, &event_logger_, mutable_cf_options.report_bg_io_stats,
+        snapshot_checker, job_context, flush_reason, log_buffer,
+        directories_.GetDbDir(), data_dir,
+        GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
+        &event_logger_, mutable_cf_options.report_bg_io_stats,
         false /* sync_output_directory */, false /* write_manifest */,
         thread_pri, io_tracer_, seqno_time_mapping_, db_id_, db_session_id_,
         cfd->GetFullHistoryTsLow(), &blob_callback_));
@@ -483,8 +490,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
   for (int i = 0; i != num_cfs; ++i) {
     const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.at(i);
     // may temporarily unlock and lock the mutex.
+    FlushReason flush_reason = bg_flush_args[i].flush_reason_;
     NotifyOnFlushBegin(cfds[i], &file_meta[i], mutable_cf_options,
-                       job_context->job_id);
+                       job_context->job_id, flush_reason);
   }
 #endif /* !ROCKSDB_LITE */
 
@@ -642,8 +650,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
 
     bool resuming_from_bg_err =
         error_handler_.IsDBStopped() ||
-        (cfds[0]->GetFlushReason() == FlushReason::kErrorRecovery ||
-         cfds[0]->GetFlushReason() == FlushReason::kErrorRecoveryRetryFlush);
+        (bg_flush_args[0].flush_reason_ == FlushReason::kErrorRecovery ||
+         bg_flush_args[0].flush_reason_ ==
+             FlushReason::kErrorRecoveryRetryFlush);
     while ((!resuming_from_bg_err || error_handler_.GetRecoveryError().ok())) {
       std::pair<Status, bool> res = wait_to_install_func();
 
@@ -660,8 +669,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
 
       resuming_from_bg_err =
           error_handler_.IsDBStopped() ||
-          (cfds[0]->GetFlushReason() == FlushReason::kErrorRecovery ||
-           cfds[0]->GetFlushReason() == FlushReason::kErrorRecoveryRetryFlush);
+          (bg_flush_args[0].flush_reason_ == FlushReason::kErrorRecovery ||
+           bg_flush_args[0].flush_reason_ ==
+               FlushReason::kErrorRecoveryRetryFlush);
     }
 
     if (!resuming_from_bg_err) {
@@ -816,7 +826,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
 
 void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
                                 const MutableCFOptions& mutable_cf_options,
-                                int job_id) {
+                                int job_id, FlushReason flush_reason) {
 #ifndef ROCKSDB_LITE
   if (immutable_db_options_.listeners.size() == 0U) {
     return;
@@ -849,7 +859,7 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
     info.triggered_writes_stop = triggered_writes_stop;
     info.smallest_seqno = file_meta->fd.smallest_seqno;
     info.largest_seqno = file_meta->fd.largest_seqno;
-    info.flush_reason = cfd->GetFlushReason();
+    info.flush_reason = flush_reason;
     for (auto listener : immutable_db_options_.listeners) {
       listener->OnFlushBegin(this, info);
     }
@@ -862,6 +872,7 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
   (void)file_meta;
   (void)mutable_cf_options;
   (void)job_id;
+  (void)flush_reason;
 #endif  // ROCKSDB_LITE
 }
 
@@ -1055,10 +1066,10 @@ Status DBImpl::CompactRangeInternal(const CompactRangeOptions& options,
       SelectColumnFamiliesForAtomicFlush(&cfds);
       mutex_.Unlock();
       s = AtomicFlushMemTables(cfds, fo, FlushReason::kManualCompaction,
-                               false /* writes_stopped */);
+                               false /* entered_write_thread */);
     } else {
       s = FlushMemTable(cfd, fo, FlushReason::kManualCompaction,
-                        false /* writes_stopped*/);
+                        false /* entered_write_thread */);
     }
     if (!s.ok()) {
       LogFlush(immutable_db_options_.info_log);
@@ -1919,11 +1930,19 @@ Status DBImpl::RunManualCompaction(
                manual.begin, manual.end, &manual.manual_end, &manual_conflict,
                max_file_num_to_ignore, trim_ts)) == nullptr &&
           manual_conflict))) {
-      // exclusive manual compactions should not see a conflict during
-      // CompactRange
-      assert(!exclusive || !manual_conflict);
-      // Running either this or some other manual compaction
-      bg_cv_.Wait();
+      if (!scheduled) {
+        // There is a conflicting compaction
+        if (manual_compaction_paused_ > 0 || manual.canceled == true) {
+          // Stop waiting since it was canceled. Pretend the error came from
+          // compaction so the below cleanup/error handling code can process it.
+          manual.done = true;
+          manual.status =
+              Status::Incomplete(Status::SubCode::kManualCompactionPaused);
+        }
+      }
+      if (!manual.done) {
+        bg_cv_.Wait();
+      }
       if (manual_compaction_paused_ > 0 && scheduled && !unscheduled) {
         assert(thread_pool_priority != Env::Priority::TOTAL);
         // unschedule all manual compactions
@@ -2001,24 +2020,32 @@ Status DBImpl::RunManualCompaction(
 }
 
 void DBImpl::GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
-                                  FlushRequest* req) {
+                                  FlushReason flush_reason, FlushRequest* req) {
   assert(req != nullptr);
-  req->reserve(cfds.size());
+  req->flush_reason = flush_reason;
+  req->cfd_to_max_mem_id_to_persist.reserve(cfds.size());
   for (const auto cfd : cfds) {
     if (nullptr == cfd) {
       // cfd may be null, see DBImpl::ScheduleFlushes
       continue;
     }
     uint64_t max_memtable_id = cfd->imm()->GetLatestMemTableID();
-    req->emplace_back(cfd, max_memtable_id);
+    req->cfd_to_max_mem_id_to_persist.emplace(cfd, max_memtable_id);
   }
 }
 
 Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
                              const FlushOptions& flush_options,
-                             FlushReason flush_reason, bool writes_stopped) {
+                             FlushReason flush_reason,
+                             bool entered_write_thread) {
   // This method should not be called if atomic_flush is true.
   assert(!immutable_db_options_.atomic_flush);
+  if (!flush_options.wait && write_controller_.IsStopped()) {
+    std::ostringstream oss;
+    oss << "Writes have been stopped, thus unable to perform manual flush. "
+           "Please try again later after writes are resumed";
+    return Status::TryAgain(oss.str());
+  }
   Status s;
   if (!flush_options.allow_write_stall) {
     bool flush_needed = true;
@@ -2029,6 +2056,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
     }
   }
 
+  const bool needs_to_join_write_thread = !entered_write_thread;
   autovector<FlushRequest> flush_reqs;
   autovector<uint64_t> memtable_ids_to_wait;
   {
@@ -2037,7 +2065,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
 
     WriteThread::Writer w;
     WriteThread::Writer nonmem_w;
-    if (!writes_stopped) {
+    if (needs_to_join_write_thread) {
       write_thread_.EnterUnbatched(&w, &mutex_);
       if (two_write_queues_) {
         nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
@@ -2060,7 +2088,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
     if (s.ok()) {
       if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
           !cached_recoverable_state_empty_.load()) {
-        FlushRequest req{{cfd, flush_memtable_id}};
+        FlushRequest req{flush_reason, {{cfd, flush_memtable_id}}};
         flush_reqs.emplace_back(std::move(req));
         memtable_ids_to_wait.emplace_back(cfd->imm()->GetLatestMemTableID());
       }
@@ -2088,7 +2116,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
                            "to avoid holding old logs",
                            cfd->GetName().c_str());
             s = SwitchMemtable(cfd_stats, &context);
-            FlushRequest req{{cfd_stats, flush_memtable_id}};
+            FlushRequest req{flush_reason, {{cfd_stats, flush_memtable_id}}};
             flush_reqs.emplace_back(std::move(req));
             memtable_ids_to_wait.emplace_back(
                 cfd->imm()->GetLatestMemTableID());
@@ -2099,8 +2127,9 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
 
     if (s.ok() && !flush_reqs.empty()) {
       for (const auto& req : flush_reqs) {
-        assert(req.size() == 1);
-        ColumnFamilyData* loop_cfd = req[0].first;
+        assert(req.cfd_to_max_mem_id_to_persist.size() == 1);
+        ColumnFamilyData* loop_cfd =
+            req.cfd_to_max_mem_id_to_persist.begin()->first;
         loop_cfd->imm()->FlushRequested();
       }
       // If the caller wants to wait for this flush to complete, it indicates
@@ -2109,18 +2138,19 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
       // Therefore, we increase the cfd's ref count.
       if (flush_options.wait) {
         for (const auto& req : flush_reqs) {
-          assert(req.size() == 1);
-          ColumnFamilyData* loop_cfd = req[0].first;
+          assert(req.cfd_to_max_mem_id_to_persist.size() == 1);
+          ColumnFamilyData* loop_cfd =
+              req.cfd_to_max_mem_id_to_persist.begin()->first;
           loop_cfd->Ref();
         }
       }
       for (const auto& req : flush_reqs) {
-        SchedulePendingFlush(req, flush_reason);
+        SchedulePendingFlush(req);
       }
       MaybeScheduleFlushOrCompaction();
     }
 
-    if (!writes_stopped) {
+    if (needs_to_join_write_thread) {
       write_thread_.ExitUnbatched(&w);
       if (two_write_queues_) {
         nonmem_write_thread_.ExitUnbatched(&nonmem_w);
@@ -2134,8 +2164,8 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
     autovector<const uint64_t*> flush_memtable_ids;
     assert(flush_reqs.size() == memtable_ids_to_wait.size());
     for (size_t i = 0; i < flush_reqs.size(); ++i) {
-      assert(flush_reqs[i].size() == 1);
-      cfds.push_back(flush_reqs[i][0].first);
+      assert(flush_reqs[i].cfd_to_max_mem_id_to_persist.size() == 1);
+      cfds.push_back(flush_reqs[i].cfd_to_max_mem_id_to_persist.begin()->first);
       flush_memtable_ids.push_back(&(memtable_ids_to_wait[i]));
     }
     s = WaitForFlushMemTables(
@@ -2156,7 +2186,14 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
 Status DBImpl::AtomicFlushMemTables(
     const autovector<ColumnFamilyData*>& column_family_datas,
     const FlushOptions& flush_options, FlushReason flush_reason,
-    bool writes_stopped) {
+    bool entered_write_thread) {
+  assert(immutable_db_options_.atomic_flush);
+  if (!flush_options.wait && write_controller_.IsStopped()) {
+    std::ostringstream oss;
+    oss << "Writes have been stopped, thus unable to perform manual flush. "
+           "Please try again later after writes are resumed";
+    return Status::TryAgain(oss.str());
+  }
   Status s;
   if (!flush_options.allow_write_stall) {
     int num_cfs_to_flush = 0;
@@ -2173,6 +2210,7 @@ Status DBImpl::AtomicFlushMemTables(
       return s;
     }
   }
+  const bool needs_to_join_write_thread = !entered_write_thread;
   FlushRequest flush_req;
   autovector<ColumnFamilyData*> cfds;
   {
@@ -2181,7 +2219,7 @@ Status DBImpl::AtomicFlushMemTables(
 
     WriteThread::Writer w;
     WriteThread::Writer nonmem_w;
-    if (!writes_stopped) {
+    if (needs_to_join_write_thread) {
       write_thread_.EnterUnbatched(&w, &mutex_);
       if (two_write_queues_) {
         nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
@@ -2224,12 +2262,12 @@ Status DBImpl::AtomicFlushMemTables(
           cfd->Ref();
         }
       }
-      GenerateFlushRequest(cfds, &flush_req);
-      SchedulePendingFlush(flush_req, flush_reason);
+      GenerateFlushRequest(cfds, flush_reason, &flush_req);
+      SchedulePendingFlush(flush_req);
       MaybeScheduleFlushOrCompaction();
     }
 
-    if (!writes_stopped) {
+    if (needs_to_join_write_thread) {
       write_thread_.ExitUnbatched(&w);
       if (two_write_queues_) {
         nonmem_write_thread_.ExitUnbatched(&nonmem_w);
@@ -2240,7 +2278,7 @@ Status DBImpl::AtomicFlushMemTables(
   TEST_SYNC_POINT("DBImpl::AtomicFlushMemTables:BeforeWaitForBgFlush");
   if (s.ok() && flush_options.wait) {
     autovector<const uint64_t*> flush_memtable_ids;
-    for (auto& iter : flush_req) {
+    for (auto& iter : flush_req.cfd_to_max_mem_id_to_persist) {
       flush_memtable_ids.push_back(&(iter.second));
     }
     s = WaitForFlushMemTables(
@@ -2587,9 +2625,9 @@ DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
   FlushRequest flush_req = flush_queue_.front();
   flush_queue_.pop_front();
   if (!immutable_db_options_.atomic_flush) {
-    assert(flush_req.size() == 1);
+    assert(flush_req.cfd_to_max_mem_id_to_persist.size() == 1);
   }
-  for (const auto& elem : flush_req) {
+  for (const auto& elem : flush_req.cfd_to_max_mem_id_to_persist) {
     if (!immutable_db_options_.atomic_flush) {
       ColumnFamilyData* cfd = elem.first;
       assert(cfd);
@@ -2597,7 +2635,6 @@ DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
       cfd->set_queued_for_flush(false);
     }
   }
-  // TODO: need to unset flush reason?
   return flush_req;
 }
 
@@ -2627,31 +2664,29 @@ ColumnFamilyData* DBImpl::PickCompactionFromQueue(
   return cfd;
 }
 
-void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
-                                  FlushReason flush_reason) {
+void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req) {
   mutex_.AssertHeld();
-  if (flush_req.empty()) {
+  if (flush_req.cfd_to_max_mem_id_to_persist.empty()) {
     return;
   }
   if (!immutable_db_options_.atomic_flush) {
     // For the non-atomic flush case, we never schedule multiple column
     // families in the same flush request.
-    assert(flush_req.size() == 1);
-    ColumnFamilyData* cfd = flush_req[0].first;
+    assert(flush_req.cfd_to_max_mem_id_to_persist.size() == 1);
+    ColumnFamilyData* cfd =
+        flush_req.cfd_to_max_mem_id_to_persist.begin()->first;
     assert(cfd);
 
     if (!cfd->queued_for_flush() && cfd->imm()->IsFlushPending()) {
       cfd->Ref();
       cfd->set_queued_for_flush(true);
-      cfd->SetFlushReason(flush_reason);
       ++unscheduled_flushes_;
       flush_queue_.push_back(flush_req);
     }
   } else {
-    for (auto& iter : flush_req) {
+    for (auto& iter : flush_req.cfd_to_max_mem_id_to_persist) {
       ColumnFamilyData* cfd = iter.first;
       cfd->Ref();
-      cfd->SetFlushReason(flush_reason);
     }
     ++unscheduled_flushes_;
     flush_queue_.push_back(flush_req);
@@ -2783,10 +2818,12 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
   while (!flush_queue_.empty()) {
     // This cfd is already referenced
     const FlushRequest& flush_req = PopFirstFromFlushQueue();
+    FlushReason flush_reason = flush_req.flush_reason;
     superversion_contexts.clear();
-    superversion_contexts.reserve(flush_req.size());
+    superversion_contexts.reserve(
+        flush_req.cfd_to_max_mem_id_to_persist.size());
 
-    for (const auto& iter : flush_req) {
+    for (const auto& iter : flush_req.cfd_to_max_mem_id_to_persist) {
       ColumnFamilyData* cfd = iter.first;
       if (cfd->GetMempurgeUsed()) {
         // If imm() contains silent memtables (e.g.: because
@@ -2802,7 +2839,7 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
       }
       superversion_contexts.emplace_back(SuperVersionContext(true));
       bg_flush_args.emplace_back(cfd, iter.second,
-                                 &(superversion_contexts.back()));
+                                 &(superversion_contexts.back()), flush_reason);
     }
     if (!bg_flush_args.empty()) {
       break;
@@ -2826,9 +2863,14 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
     status = FlushMemTablesToOutputFiles(bg_flush_args, made_progress,
                                          job_context, log_buffer, thread_pri);
     TEST_SYNC_POINT("DBImpl::BackgroundFlush:BeforeFlush");
-    // All the CFDs in the FlushReq must have the same flush reason, so just
-    // grab the first one
-    *reason = bg_flush_args[0].cfd_->GetFlushReason();
+// All the CFD/bg_flush_arg in the FlushReq must have the same flush reason, so
+// just grab the first one
+#ifndef NDEBUG
+    for (const auto bg_flush_arg : bg_flush_args) {
+      assert(bg_flush_arg.flush_reason_ == bg_flush_args[0].flush_reason_);
+    }
+#endif /* !NDEBUG */
+    *reason = bg_flush_args[0].flush_reason_;
     for (auto& arg : bg_flush_args) {
       ColumnFamilyData* cfd = arg.cfd_;
       if (cfd->UnrefAndTryDelete()) {
@@ -3263,7 +3305,6 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
                              c->column_family_data());
     assert(c->num_input_files(1) == 0);
-    assert(c->level() == 0);
     assert(c->column_family_data()->ioptions()->compaction_style ==
            kCompactionStyleFIFO);
 
@@ -3756,9 +3797,11 @@ void DBImpl::InstallSuperVersionAndScheduleWork(
   // triggered soon anyway.
   bottommost_files_mark_threshold_ = kMaxSequenceNumber;
   for (auto* my_cfd : *versions_->GetColumnFamilySet()) {
-    bottommost_files_mark_threshold_ = std::min(
-        bottommost_files_mark_threshold_,
-        my_cfd->current()->storage_info()->bottommost_files_mark_threshold());
+    if (!my_cfd->ioptions()->allow_ingest_behind) {
+      bottommost_files_mark_threshold_ = std::min(
+          bottommost_files_mark_threshold_,
+          my_cfd->current()->storage_info()->bottommost_files_mark_threshold());
+    }
   }
 
   // Whenever we install new SuperVersion, we might need to issue new flushes or

db/db_impl/db_impl_debug.cc

@@ -31,18 +31,6 @@ Status DBImpl::TEST_SwitchWAL() {
   return s;
 }
 
-bool DBImpl::TEST_WALBufferIsEmpty(bool lock) {
-  if (lock) {
-    log_write_mutex_.Lock();
-  }
-  log::Writer* cur_log_writer = logs_.back().writer;
-  auto res = cur_log_writer->TEST_BufferIsEmpty();
-  if (lock) {
-    log_write_mutex_.Unlock();
-  }
-  return res;
-}
-
 uint64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
     ColumnFamilyHandle* column_family) {
   ColumnFamilyData* cfd;

db/db_impl/db_impl_files.cc

@@ -193,11 +193,14 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
       }
     }
 
+    IOOptions io_opts;
+    io_opts.do_not_recurse = true;
     for (auto& path : paths) {
       // set of all files in the directory. We'll exclude files that are still
       // alive in the subsequent processings.
       std::vector<std::string> files;
-      Status s = env_->GetChildren(path, &files);
+      Status s = immutable_db_options_.fs->GetChildren(
+          path, io_opts, &files, /*IODebugContext*=*/nullptr);
       s.PermitUncheckedError();  // TODO: What should we do on error?
       for (const std::string& file : files) {
         uint64_t number;
@@ -223,7 +226,9 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
     // Add log files in wal_dir
     if (!immutable_db_options_.IsWalDirSameAsDBPath(dbname_)) {
       std::vector<std::string> log_files;
-      Status s = env_->GetChildren(immutable_db_options_.wal_dir, &log_files);
+      Status s = immutable_db_options_.fs->GetChildren(
+          immutable_db_options_.wal_dir, io_opts, &log_files,
+          /*IODebugContext*=*/nullptr);
       s.PermitUncheckedError();  // TODO: What should we do on error?
       for (const std::string& log_file : log_files) {
         job_context->full_scan_candidate_files.emplace_back(
@@ -235,8 +240,9 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
     if (!immutable_db_options_.db_log_dir.empty() &&
         immutable_db_options_.db_log_dir != dbname_) {
       std::vector<std::string> info_log_files;
-      Status s =
-          env_->GetChildren(immutable_db_options_.db_log_dir, &info_log_files);
+      Status s = immutable_db_options_.fs->GetChildren(
+          immutable_db_options_.db_log_dir, io_opts, &info_log_files,
+          /*IODebugContext*=*/nullptr);
       s.PermitUncheckedError();  // TODO: What should we do on error?
       for (std::string& log_file : info_log_files) {
         job_context->full_scan_candidate_files.emplace_back(

db/db_impl/db_impl_open.cc

@@ -163,8 +163,11 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src,
     // DeleteScheduler::CleanupDirectory on the same dir later, it will be
     // safe
     std::vector<std::string> filenames;
+    IOOptions io_opts;
+    io_opts.do_not_recurse = true;
     auto wal_dir = immutable_db_options.GetWalDir();
-    Status s = result.env->GetChildren(wal_dir, &filenames);
+    Status s = immutable_db_options.fs->GetChildren(
+        wal_dir, io_opts, &filenames, /*IODebugContext*=*/nullptr);
     s.PermitUncheckedError();  //**TODO: What to do on error?
     for (std::string& filename : filenames) {
       if (filename.find(".log.trash", filename.length() -
@@ -267,7 +270,8 @@ Status DBImpl::ValidateOptions(const DBOptions& db_options) {
   if (db_options.unordered_write &&
       !db_options.allow_concurrent_memtable_write) {
     return Status::InvalidArgument(
-        "unordered_write is incompatible with !allow_concurrent_memtable_write");
+        "unordered_write is incompatible with "
+        "!allow_concurrent_memtable_write");
   }
 
   if (db_options.unordered_write && db_options.enable_pipelined_write) {
@@ -432,7 +436,10 @@ Status DBImpl::Recover(
       s = env_->FileExists(current_fname);
     } else {
       s = Status::NotFound();
-      Status io_s = env_->GetChildren(dbname_, &files_in_dbname);
+      IOOptions io_opts;
+      io_opts.do_not_recurse = true;
+      Status io_s = immutable_db_options_.fs->GetChildren(
+          dbname_, io_opts, &files_in_dbname, /*IODebugContext*=*/nullptr);
       if (!io_s.ok()) {
         s = io_s;
         files_in_dbname.clear();
@@ -498,7 +505,10 @@ Status DBImpl::Recover(
     }
   } else if (immutable_db_options_.best_efforts_recovery) {
     assert(files_in_dbname.empty());
-    Status s = env_->GetChildren(dbname_, &files_in_dbname);
+    IOOptions io_opts;
+    io_opts.do_not_recurse = true;
+    Status s = immutable_db_options_.fs->GetChildren(
+        dbname_, io_opts, &files_in_dbname, /*IODebugContext*=*/nullptr);
     if (s.IsNotFound()) {
       return Status::InvalidArgument(dbname_,
                                      "does not exist (open for read only)");
@@ -570,7 +580,10 @@ Status DBImpl::Recover(
     // produced by an older version of rocksdb.
     auto wal_dir = immutable_db_options_.GetWalDir();
     if (!immutable_db_options_.best_efforts_recovery) {
-      s = env_->GetChildren(wal_dir, &files_in_wal_dir);
+      IOOptions io_opts;
+      io_opts.do_not_recurse = true;
+      s = immutable_db_options_.fs->GetChildren(
+          wal_dir, io_opts, &files_in_wal_dir, /*IODebugContext*=*/nullptr);
     }
     if (s.IsNotFound()) {
       return Status::InvalidArgument("wal_dir not found", wal_dir);
@@ -678,7 +691,10 @@ Status DBImpl::Recover(
       } else if (normalized_dbname == normalized_wal_dir) {
         filenames = std::move(files_in_wal_dir);
       } else {
-        s = env_->GetChildren(GetName(), &filenames);
+        IOOptions io_opts;
+        io_opts.do_not_recurse = true;
+        s = immutable_db_options_.fs->GetChildren(
+            GetName(), io_opts, &filenames, /*IODebugContext*=*/nullptr);
       }
     }
     if (s.ok()) {
@@ -1043,9 +1059,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& wal_numbers,
     std::unique_ptr<SequentialFileReader> file_reader;
     {
       std::unique_ptr<FSSequentialFile> file;
-      status = fs_->NewSequentialFile(fname,
-                                      fs_->OptimizeForLogRead(file_options_),
-                                      &file, nullptr);
+      status = fs_->NewSequentialFile(
+          fname, fs_->OptimizeForLogRead(file_options_), &file, nullptr);
       if (!status.ok()) {
         MaybeIgnoreError(&status);
         if (!status.ok()) {
@@ -1929,18 +1944,6 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
 
   if (s.ok()) {
     for (auto cfd : *impl->versions_->GetColumnFamilySet()) {
-      if (cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
-        auto* vstorage = cfd->current()->storage_info();
-        for (int i = 1; i < vstorage->num_levels(); ++i) {
-          int num_files = vstorage->NumLevelFiles(i);
-          if (num_files > 0) {
-            s = Status::InvalidArgument(
-                "Not all files are at level 0. Cannot "
-                "open with FIFO compaction style.");
-            break;
-          }
-        }
-      }
       if (!cfd->mem()->IsSnapshotSupported()) {
         impl->is_snapshot_supported_ = false;
       }
@@ -2022,9 +2025,13 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
     // Remove duplicate paths.
     std::sort(paths.begin(), paths.end());
     paths.erase(std::unique(paths.begin(), paths.end()), paths.end());
+    IOOptions io_opts;
+    io_opts.do_not_recurse = true;
     for (auto& path : paths) {
       std::vector<std::string> existing_files;
-      impl->immutable_db_options_.env->GetChildren(path, &existing_files)
+      impl->immutable_db_options_.fs
+          ->GetChildren(path, io_opts, &existing_files,
+                        /*IODebugContext*=*/nullptr)
           .PermitUncheckedError();  //**TODO: What do to on error?
       for (auto& file_name : existing_files) {
         uint64_t file_number;
@@ -2060,10 +2067,16 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
     ROCKS_LOG_HEADER(impl->immutable_db_options_.info_log, "DB pointer %p",
                      impl);
     LogFlush(impl->immutable_db_options_.info_log);
-    assert(impl->TEST_WALBufferIsEmpty());
-    // If the assert above fails then we need to FlushWAL before returning
-    // control back to the user.
-    if (!persist_options_status.ok()) {
+    if (!impl->WALBufferIsEmpty()) {
+      s = impl->FlushWAL(false);
+      if (s.ok()) {
+        // Sync is needed otherwise WAL buffered data might get lost after a
+        // power reset.
+        log::Writer* log_writer = impl->logs_.back().writer;
+        s = log_writer->file()->Sync(impl->immutable_db_options_.use_fsync);
+      }
+    }
+    if (s.ok() && !persist_options_status.ok()) {
       s = Status::IOError(
           "DB::Open() failed --- Unable to persist Options file",
           persist_options_status.ToString());

db/db_impl/db_impl_readonly.h

@@ -9,6 +9,7 @@
 
 #include <string>
 #include <vector>
+
 #include "db/db_impl/db_impl.h"
 
 namespace ROCKSDB_NAMESPACE {

db/db_impl/db_impl_secondary.cc

@@ -100,7 +100,11 @@ Status DBImplSecondary::FindNewLogNumbers(std::vector<uint64_t>* logs) {
   assert(logs != nullptr);
   std::vector<std::string> filenames;
   Status s;
-  s = env_->GetChildren(immutable_db_options_.GetWalDir(), &filenames);
+  IOOptions io_opts;
+  io_opts.do_not_recurse = true;
+  s = immutable_db_options_.fs->GetChildren(immutable_db_options_.GetWalDir(),
+                                            io_opts, &filenames,
+                                            /*IODebugContext*=*/nullptr);
   if (s.IsNotFound()) {
     return Status::InvalidArgument("Failed to open wal_dir",
                                    immutable_db_options_.GetWalDir());
@@ -153,8 +157,7 @@ Status DBImplSecondary::MaybeInitLogReader(
     {
       std::unique_ptr<FSSequentialFile> file;
       Status status = fs_->NewSequentialFile(
-          fname, fs_->OptimizeForLogRead(file_options_), &file,
-          nullptr);
+          fname, fs_->OptimizeForLogRead(file_options_), &file, nullptr);
       if (!status.ok()) {
         *log_reader = nullptr;
         return status;
@@ -196,7 +199,7 @@ Status DBImplSecondary::RecoverLogFiles(
     assert(reader != nullptr);
   }
   for (auto log_number : log_numbers) {
-    auto it  = log_readers_.find(log_number);
+    auto it = log_readers_.find(log_number);
     assert(it != log_readers_.end());
     log::FragmentBufferedReader* reader = it->second->reader_;
     Status* wal_read_status = it->second->status_;

db/db_impl/db_impl_secondary.h

@@ -47,6 +47,7 @@ class LogReaderContainer {
     delete reporter_;
     delete status_;
   }
+
  private:
   struct LogReporter : public log::Reader::Reporter {
     Env* env;
@@ -247,7 +248,6 @@ class DBImplSecondary : public DBImpl {
   // method can take long time due to all the I/O and CPU costs.
   Status TryCatchUpWithPrimary() override;
 
-
   // Try to find log reader using log_number from log_readers_ map, initialize
   // if it doesn't exist
   Status MaybeInitLogReader(uint64_t log_number,

db/db_impl/db_impl_write.cc

@@ -62,6 +62,15 @@ Status DBImpl::Merge(const WriteOptions& o, ColumnFamilyHandle* column_family,
   }
 }
 
+Status DBImpl::Merge(const WriteOptions& o, ColumnFamilyHandle* column_family,
+                     const Slice& key, const Slice& ts, const Slice& val) {
+  const Status s = FailIfTsMismatchCf(column_family, ts, /*ts_for_read=*/false);
+  if (!s.ok()) {
+    return s;
+  }
+  return DB::Merge(o, column_family, key, ts, val);
+}
+
 Status DBImpl::Delete(const WriteOptions& write_options,
                       ColumnFamilyHandle* column_family, const Slice& key) {
   const Status s = FailIfCfHasTs(column_family);
@@ -111,6 +120,17 @@ Status DBImpl::DeleteRange(const WriteOptions& write_options,
   return DB::DeleteRange(write_options, column_family, begin_key, end_key);
 }
 
+Status DBImpl::DeleteRange(const WriteOptions& write_options,
+                           ColumnFamilyHandle* column_family,
+                           const Slice& begin_key, const Slice& end_key,
+                           const Slice& ts) {
+  const Status s = FailIfTsMismatchCf(column_family, ts, /*ts_for_read=*/false);
+  if (!s.ok()) {
+    return s;
+  }
+  return DB::DeleteRange(write_options, column_family, begin_key, end_key, ts);
+}
+
 void DBImpl::SetRecoverableStatePreReleaseCallback(
     PreReleaseCallback* callback) {
   recoverable_state_pre_release_callback_.reset(callback);
@@ -1624,14 +1644,14 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
       cfd->imm()->FlushRequested();
       if (!immutable_db_options_.atomic_flush) {
         FlushRequest flush_req;
-        GenerateFlushRequest({cfd}, &flush_req);
-        SchedulePendingFlush(flush_req, FlushReason::kWalFull);
+        GenerateFlushRequest({cfd}, FlushReason::kWalFull, &flush_req);
+        SchedulePendingFlush(flush_req);
       }
     }
     if (immutable_db_options_.atomic_flush) {
       FlushRequest flush_req;
-      GenerateFlushRequest(cfds, &flush_req);
-      SchedulePendingFlush(flush_req, FlushReason::kWalFull);
+      GenerateFlushRequest(cfds, FlushReason::kWalFull, &flush_req);
+      SchedulePendingFlush(flush_req);
     }
     MaybeScheduleFlushOrCompaction();
   }
@@ -1715,14 +1735,15 @@ Status DBImpl::HandleWriteBufferManagerFlush(WriteContext* write_context) {
       cfd->imm()->FlushRequested();
       if (!immutable_db_options_.atomic_flush) {
         FlushRequest flush_req;
-        GenerateFlushRequest({cfd}, &flush_req);
-        SchedulePendingFlush(flush_req, FlushReason::kWriteBufferManager);
+        GenerateFlushRequest({cfd}, FlushReason::kWriteBufferManager,
+                             &flush_req);
+        SchedulePendingFlush(flush_req);
       }
     }
     if (immutable_db_options_.atomic_flush) {
       FlushRequest flush_req;
-      GenerateFlushRequest(cfds, &flush_req);
-      SchedulePendingFlush(flush_req, FlushReason::kWriteBufferManager);
+      GenerateFlushRequest(cfds, FlushReason::kWriteBufferManager, &flush_req);
+      SchedulePendingFlush(flush_req);
     }
     MaybeScheduleFlushOrCompaction();
   }
@@ -1749,6 +1770,7 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
                  &time_delayed);
     uint64_t delay =
         write_controller_.GetDelay(immutable_db_options_.clock, num_bytes);
+    TEST_SYNC_POINT("DBImpl::DelayWrite:Start");
     if (delay > 0) {
       if (write_options.no_slowdown) {
         return Status::Incomplete("Write stall");
@@ -1758,8 +1780,8 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
       // Notify write_thread_ about the stall so it can setup a barrier and
       // fail any pending writers with no_slowdown
       write_thread_.BeginWriteStall();
-      TEST_SYNC_POINT("DBImpl::DelayWrite:BeginWriteStallDone");
       mutex_.Unlock();
+      TEST_SYNC_POINT("DBImpl::DelayWrite:BeginWriteStallDone");
       // We will delay the write until we have slept for `delay` microseconds
       // or we don't need a delay anymore. We check for cancellation every 1ms
       // (slightly longer because WriteController minimum delay is 1ms, in
@@ -1784,7 +1806,8 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
     // might wait here indefinitely as the background compaction may never
     // finish successfully, resulting in the stall condition lasting
     // indefinitely
-    while (error_handler_.GetBGError().ok() && write_controller_.IsStopped()) {
+    while (error_handler_.GetBGError().ok() && write_controller_.IsStopped() &&
+           !shutting_down_.load(std::memory_order_relaxed)) {
       if (write_options.no_slowdown) {
         return Status::Incomplete("Write stall");
       }
@@ -1810,9 +1833,13 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
   // proceed
   Status s;
   if (write_controller_.IsStopped()) {
-    // If writes are still stopped, it means we bailed due to a background
-    // error
-    s = Status::Incomplete(error_handler_.GetBGError().ToString());
+    if (!shutting_down_.load(std::memory_order_relaxed)) {
+      // If writes are still stopped and db not shutdown, it means we bailed
+      // due to a background error
+      s = Status::Incomplete(error_handler_.GetBGError().ToString());
+    } else {
+      s = Status::ShutdownInProgress("stalled writes");
+    }
   }
   if (error_handler_.IsDBStopped()) {
     s = error_handler_.GetBGError();
@@ -1972,13 +1999,13 @@ Status DBImpl::ScheduleFlushes(WriteContext* context) {
     if (immutable_db_options_.atomic_flush) {
       AssignAtomicFlushSeq(cfds);
       FlushRequest flush_req;
-      GenerateFlushRequest(cfds, &flush_req);
-      SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+      GenerateFlushRequest(cfds, FlushReason::kWriteBufferFull, &flush_req);
+      SchedulePendingFlush(flush_req);
     } else {
       for (auto* cfd : cfds) {
         FlushRequest flush_req;
-        GenerateFlushRequest({cfd}, &flush_req);
-        SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+        GenerateFlushRequest({cfd}, FlushReason::kWriteBufferFull, &flush_req);
+        SchedulePendingFlush(flush_req);
       }
     }
     MaybeScheduleFlushOrCompaction();
@@ -2361,6 +2388,24 @@ Status DB::DeleteRange(const WriteOptions& opt,
   return Write(opt, &batch);
 }
 
+Status DB::DeleteRange(const WriteOptions& opt,
+                       ColumnFamilyHandle* column_family,
+                       const Slice& begin_key, const Slice& end_key,
+                       const Slice& ts) {
+  ColumnFamilyHandle* default_cf = DefaultColumnFamily();
+  assert(default_cf);
+  const Comparator* const default_cf_ucmp = default_cf->GetComparator();
+  assert(default_cf_ucmp);
+  WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */,
+                   opt.protection_bytes_per_key,
+                   default_cf_ucmp->timestamp_size());
+  Status s = batch.DeleteRange(column_family, begin_key, end_key, ts);
+  if (!s.ok()) {
+    return s;
+  }
+  return Write(opt, &batch);
+}
+
 Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
                  const Slice& key, const Slice& value) {
   WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */,
@@ -2371,4 +2416,21 @@ Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
   }
   return Write(opt, &batch);
 }
+
+Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
+                 const Slice& key, const Slice& ts, const Slice& value) {
+  ColumnFamilyHandle* default_cf = DefaultColumnFamily();
+  assert(default_cf);
+  const Comparator* const default_cf_ucmp = default_cf->GetComparator();
+  assert(default_cf_ucmp);
+  WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */,
+                   opt.protection_bytes_per_key,
+                   default_cf_ucmp->timestamp_size());
+  Status s = batch.Merge(column_family, key, ts, value);
+  if (!s.ok()) {
+    return s;
+  }
+  return Write(opt, &batch);
+}
+
 }  // namespace ROCKSDB_NAMESPACE