Skip CancelAllBackgroundWork if DBImpl is already closed (#6268)

Summary:
WritePreparedTxnDB calls CancelAllBackgroundWork in its destructor to avoid dangling references to it from background job's SnapshotChecker callback. However, if the DBImpl is already closed, the info log might be closed with it, which causes memory leak when CancelAllBackgroundWork tries to print to the info log. The patch fixes that by calling CancelAllBackgroundWork only if the db is not closed already.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6268

Differential Revision: D19303439

Pulled By: maysamyabandeh

fbshipit-source-id: 4228a6be7e78d43c90630347baa89b008200bd15

.travis.yml

@@ -6,25 +6,34 @@ os:
 compiler:
   - clang
   - gcc
-osx_image: xcode8.3
+osx_image: xcode9.4
 jdk:
   - openjdk7
 cache:
   - ccache
-  - apt
 
 addons:
   apt:
     sources:
       - ubuntu-toolchain-r-test
     packages:
-      - curl
-      - g++-8
-      - libbz2-dev
       - libgflags-dev
+      - libbz2-dev
+      - liblz4-dev
       - libsnappy-dev
-      - mingw-w64
+      - liblzma-dev  # xv
+      - libzstd-dev
       - zlib1g-dev
+  homebrew:
+    update: true
+    packages:
+      - ccache
+      - gflags
+      - lz4
+      - snappy
+      - xz
+      - zstd
+
 env:
   - TEST_GROUP=platform_dependent # 16-18 minutes
   - TEST_GROUP=1 # 33-35 minutes
@@ -60,17 +69,19 @@ matrix:
   - os : osx
     compiler: gcc
 
-# https://docs.travis-ci.com/user/caching/#ccache-cache
 install:
   - if [ "${TRAVIS_OS_NAME}" == osx ]; then
-      brew install ccache zstd lz4 snappy xz;
       PATH=$PATH:/usr/local/opt/ccache/libexec;
     fi
   - if [ "${JOB_NAME}" == cmake-gcc8 ]; then
+      sudo apt-get install -y g++-8;
       CC=gcc-8 && CXX=g++-8;
     fi
+  - if [ "${JOB_NAME}" == cmake-mingw ]; then
+      sudo apt-get install -y mingw-w64 ;
+    fi
   - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
-      mkdir cmake-dist && curl --silent --fail --show-error --location https://github.com/Kitware/CMake/releases/download/v3.14.5/cmake-3.14.5-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
+      mkdir cmake-dist && curl -sfSL https://github.com/Kitware/CMake/releases/download/v3.14.5/cmake-3.14.5-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
     fi
   - if [[ "${JOB_NAME}" == java_test ]]; then
       java -version && echo "JAVA_HOME=${JAVA_HOME}";
@@ -113,7 +124,7 @@ script:
       ;;
     cmake-mingw)
       sudo update-alternatives --set x86_64-w64-mingw32-g++ /usr/bin/x86_64-w64-mingw32-g++-posix;
-      mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb rocksdbjni
+      mkdir build && cd build && cmake -DJNI=1 -DWITH_GFLAGS=OFF .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb rocksdbjni
       ;;
     cmake*)
       mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_BUILD_TYPE=Release && make -j4 rocksdb rocksdbjni

CMakeLists.txt

@@ -45,6 +45,16 @@ if(POLICY CMP0042)
   cmake_policy(SET CMP0042 NEW)
 endif()
 
+if(NOT CMAKE_BUILD_TYPE)
+  if(EXISTS "${CMAKE_SOURCE_DIR}/.git")
+    set(default_build_type "Debug")
+  else()
+    set(default_build_type "RelWithDebInfo")
+  endif()
+  set(CMAKE_BUILD_TYPE "${default_build_type}" CACHE STRING
+    "Default BUILD_TYPE is ${default_build_type}" FORCE)
+endif()
+
 find_program(CCACHE_FOUND ccache)
 if(CCACHE_FOUND)
   set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ccache)
@@ -62,7 +72,7 @@ if (WITH_WINDOWS_UTF8_FILENAMES)
 endif()
 # third-party/folly is only validated to work on Linux and Windows for now.
 # So only turn it on there by default.
-if(CMAKE_SYSTEM_NAME MATCHES "Linux" OR CMAKE_SYSTEM_NAME MATCHES "Windows")
+if(CMAKE_SYSTEM_NAME MATCHES "Linux|Windows")
   if(MSVC AND MSVC_VERSION LESS 1910)
     # Folly does not compile with MSVC older than VS2017
     option(WITH_FOLLY_DISTRIBUTED_MUTEX "build with folly::DistributedMutex" OFF)
@@ -72,10 +82,12 @@ if(CMAKE_SYSTEM_NAME MATCHES "Linux" OR CMAKE_SYSTEM_NAME MATCHES "Windows")
 else()
   option(WITH_FOLLY_DISTRIBUTED_MUTEX "build with folly::DistributedMutex" OFF)
 endif()
+
+include(CMakeDependentOption)
+CMAKE_DEPENDENT_OPTION(WITH_GFLAGS "build with GFlags" ON
+  "NOT MSVC;NOT MINGW" OFF)
+
 if(MSVC)
-  # Defaults currently different for GFLAGS.
-  #  We will address find_package work a little later
-  option(WITH_GFLAGS "build with GFlags" OFF)
   option(WITH_XPRESS "build with windows built in compression" OFF)
   include(${CMAKE_CURRENT_SOURCE_DIR}/thirdparty.inc)
 else()
@@ -92,14 +104,11 @@ else()
   endif()
 
   # No config file for this
-  option(WITH_GFLAGS "build with GFlags" ON)
   if(WITH_GFLAGS)
-    find_package(gflags)
-    if(gflags_FOUND)
-      add_definitions(-DGFLAGS=1)
-      include_directories(${gflags_INCLUDE_DIR})
-      list(APPEND THIRDPARTY_LIBS ${gflags_LIBRARIES})
-    endif()
+    find_package(gflags REQUIRED)
+    add_definitions(-DGFLAGS=1)
+    include_directories(${gflags_INCLUDE_DIR})
+    list(APPEND THIRDPARTY_LIBS gflags::gflags)
   endif()
 
   if(WITH_SNAPPY)
@@ -550,6 +559,7 @@ set(SOURCES
         env/env_chroot.cc
         env/env_encryption.cc
         env/env_hdfs.cc
+        env/file_system.cc
         env/mock_env.cc
         file/delete_scheduler.cc
         file/file_prefetch_buffer.cc
@@ -636,7 +646,6 @@ set(SOURCES
         test_util/testutil.cc
         test_util/transaction_test_util.cc
         tools/block_cache_analyzer/block_cache_trace_analyzer.cc
-        tools/db_bench_tool.cc
         tools/dump/db_dump_tool.cc
         tools/ldb_cmd.cc
         tools/ldb_tool.cc
@@ -758,6 +767,7 @@ else()
   list(APPEND SOURCES
     port/port_posix.cc
     env/env_posix.cc
+    env/fs_posix.cc
     env/io_posix.cc)
 endif()
 
@@ -772,7 +782,8 @@ endif()
 
 set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
 set(ROCKSDB_SHARED_LIB rocksdb-shared${ARTIFACT_SUFFIX})
-set(ROCKSDB_IMPORT_LIB ${ROCKSDB_SHARED_LIB})
+
+option(ROCKSDB_BUILD_SHARED "Build shared versions of the RocksDB libraries" ON)
 
 option(WITH_LIBRADOS "Build with librados" OFF)
 if(WITH_LIBRADOS)
@@ -783,40 +794,44 @@ endif()
 
 if(WIN32)
   set(SYSTEM_LIBS ${SYSTEM_LIBS} shlwapi.lib rpcrt4.lib)
-  set(LIBS ${ROCKSDB_STATIC_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
 else()
   set(SYSTEM_LIBS ${CMAKE_THREAD_LIBS_INIT})
-  set(LIBS ${ROCKSDB_SHARED_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
-
-  add_library(${ROCKSDB_SHARED_LIB} SHARED ${SOURCES})
-  target_link_libraries(${ROCKSDB_SHARED_LIB}
-    ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
-  set_target_properties(${ROCKSDB_SHARED_LIB} PROPERTIES
-                        LINKER_LANGUAGE CXX
-                        VERSION ${rocksdb_VERSION}
-                        SOVERSION ${rocksdb_VERSION_MAJOR}
-                        CXX_STANDARD 11
-                        OUTPUT_NAME "rocksdb")
 endif()
 
 add_library(${ROCKSDB_STATIC_LIB} STATIC ${SOURCES})
 target_link_libraries(${ROCKSDB_STATIC_LIB}
   ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
 
-if(WIN32)
-  add_library(${ROCKSDB_IMPORT_LIB} SHARED ${SOURCES})
-  target_link_libraries(${ROCKSDB_IMPORT_LIB}
+if(ROCKSDB_BUILD_SHARED)
+  add_library(${ROCKSDB_SHARED_LIB} SHARED ${SOURCES})
+  target_link_libraries(${ROCKSDB_SHARED_LIB}
     ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
-  set_target_properties(${ROCKSDB_IMPORT_LIB} PROPERTIES
-    COMPILE_DEFINITIONS "ROCKSDB_DLL;ROCKSDB_LIBRARY_EXPORTS")
-  if(MSVC)
-    set_target_properties(${ROCKSDB_STATIC_LIB} PROPERTIES
-      COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_STATIC_LIB}.pdb")
-    set_target_properties(${ROCKSDB_IMPORT_LIB} PROPERTIES
-      COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_IMPORT_LIB}.pdb")
+
+  if(WIN32)
+    set_target_properties(${ROCKSDB_SHARED_LIB} PROPERTIES
+      COMPILE_DEFINITIONS "ROCKSDB_DLL;ROCKSDB_LIBRARY_EXPORTS")
+    if(MSVC)
+      set_target_properties(${ROCKSDB_STATIC_LIB} PROPERTIES
+        COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_STATIC_LIB}.pdb")
+      set_target_properties(${ROCKSDB_SHARED_LIB} PROPERTIES
+        COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_SHARED_LIB}.pdb")
+    endif()
+  else()
+    set_target_properties(${ROCKSDB_SHARED_LIB} PROPERTIES
+                          LINKER_LANGUAGE CXX
+                          VERSION ${rocksdb_VERSION}
+                          SOVERSION ${rocksdb_VERSION_MAJOR}
+                          CXX_STANDARD 11
+                          OUTPUT_NAME "rocksdb")
   endif()
 endif()
 
+if(ROCKSDB_BUILD_SHARED AND NOT WIN32)
+  set(ROCKSDB_LIB ${ROCKSDB_SHARED_LIB})
+else()
+  set(ROCKSDB_LIB ${ROCKSDB_STATIC_LIB})
+endif()
+
 option(WITH_JNI "build with JNI" OFF)
 if(WITH_JNI OR JNI)
   message(STATUS "JNI library is enabled")
@@ -863,15 +878,17 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
     INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
   )
 
-  install(
-    TARGETS ${ROCKSDB_SHARED_LIB}
-    EXPORT RocksDBTargets
-    COMPONENT runtime
-    ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}"
-    RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}"
-    LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
-    INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
-  )
+  if(ROCKSDB_BUILD_SHARED)
+    install(
+      TARGETS ${ROCKSDB_SHARED_LIB}
+      EXPORT RocksDBTargets
+      COMPONENT runtime
+      ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}"
+      RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}"
+      LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
+      INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
+    )
+  endif()
 
   install(
     EXPORT RocksDBTargets
@@ -889,11 +906,15 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
   )
 endif()
 
-option(WITH_TESTS "build with tests" ON)
-# For test libraries, utilities, and exes that are build iff WITH_TESTS=ON and
-# in Debug mode. Add test only code that is not #ifdefed for Release here.
-if(WITH_TESTS AND CMAKE_BUILD_TYPE STREQUAL "Debug")
-  add_subdirectory(third-party/gtest-1.8.1/fused-src/gtest)
+add_subdirectory(third-party/gtest-1.8.1/fused-src/gtest)
+add_library(testharness STATIC
+  test_util/testharness.cc)
+target_link_libraries(testharness gtest)
+
+# Tests are excluded from Release builds
+CMAKE_DEPENDENT_OPTION(WITH_TESTS "build with tests" ON
+  "CMAKE_BUILD_TYPE STREQUAL Debug" OFF)
+if(WITH_TESTS)
   set(TESTS
         cache/cache_test.cc
         cache/lru_cache_test.cc
@@ -1009,6 +1030,7 @@ if(WITH_TESTS AND CMAKE_BUILD_TYPE STREQUAL "Debug")
         util/filelock_test.cc
         util/hash_test.cc
         util/heap_test.cc
+        util/random_test.cc
         util/rate_limiter_test.cc
         util/repeatable_thread_test.cc
         util/slice_transform_test.cc
@@ -1058,7 +1080,7 @@ if(WITH_TESTS AND CMAKE_BUILD_TYPE STREQUAL "Debug")
   add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND})
   set(TESTUTILLIB testutillib${ARTIFACT_SUFFIX})
   add_library(${TESTUTILLIB} STATIC ${TESTUTIL_SOURCE})
-  target_link_libraries(${TESTUTILLIB} ${LIBS})
+  target_link_libraries(${TESTUTILLIB} ${ROCKSDB_LIB})
   if(MSVC)
     set_target_properties(${TESTUTILLIB} PROPERTIES COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/testutillib${ARTIFACT_SUFFIX}.pdb")
   endif()
@@ -1068,67 +1090,86 @@ if(WITH_TESTS AND CMAKE_BUILD_TYPE STREQUAL "Debug")
         EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
         )
 
-  set(TEST_EXES ${TESTS})
-  foreach(sourcefile ${TEST_EXES})
+  foreach(sourcefile ${TESTS})
       get_filename_component(exename ${sourcefile} NAME_WE)
-      add_executable(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} ${sourcefile}
-        $<TARGET_OBJECTS:testharness>)
+      add_executable(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} ${sourcefile})
       set_target_properties(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX}
         PROPERTIES EXCLUDE_FROM_DEFAULT_BUILD_RELEASE 1
         EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
         EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
         OUTPUT_NAME ${exename}${ARTIFACT_SUFFIX}
         )
-      target_link_libraries(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} gtest ${LIBS})
+      target_link_libraries(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} testharness gtest ${ROCKSDB_LIB})
       if(NOT "${exename}" MATCHES "db_sanity_test")
         add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
         add_dependencies(check ${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX})
       endif()
-  endforeach(sourcefile ${TEST_EXES})
+  endforeach(sourcefile ${TESTS})
 
-  # C executables must link to a shared object
-  set(C_TESTS db/c_test.c)
-  set(C_TEST_EXES ${C_TESTS})
+  if(WIN32)
+    # C executables must link to a shared object
+    if(ROCKSDB_BUILD_SHARED)
+      set(ROCKSDB_LIB_FOR_C ${ROCKSDB_SHARED_LIB})
+    else()
+      set(ROCKSDB_LIB_FOR_C OFF)
+    endif()
+  else()
+    set(ROCKSDB_LIB_FOR_C ${ROCKSDB_LIB})
+  endif()
 
-  foreach(sourcefile ${C_TEST_EXES})
-      string(REPLACE ".c" "" exename ${sourcefile})
-      string(REGEX REPLACE "^((.+)/)+" "" exename ${exename})
-      add_executable(${exename}${ARTIFACT_SUFFIX} ${sourcefile})
-      set_target_properties(${exename}${ARTIFACT_SUFFIX}
-        PROPERTIES EXCLUDE_FROM_DEFAULT_BUILD_RELEASE 1
-        EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
-        EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
-        )
-      target_link_libraries(${exename}${ARTIFACT_SUFFIX} ${ROCKSDB_IMPORT_LIB} testutillib${ARTIFACT_SUFFIX})
-      add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
-      add_dependencies(check ${exename}${ARTIFACT_SUFFIX})
-  endforeach(sourcefile ${C_TEST_EXES})
+  if(ROCKSDB_LIB_FOR_C)
+    set(C_TESTS db/c_test.c)
+    # C executables must link to a shared object
+    add_executable(c_test db/c_test.c)
+    target_link_libraries(c_test ${ROCKSDB_SHARED_LIB} testharness)
+    add_test(NAME c_test COMMAND c_test${ARTIFACT_SUFFIX})
+    add_dependencies(check c_test)
+  endif()
 endif()
 
 option(WITH_BENCHMARK_TOOLS "build with benchmarks" ON)
 if(WITH_BENCHMARK_TOOLS)
-  if(NOT TARGET gtest)
-    add_subdirectory(third-party/gtest-1.8.1/fused-src/gtest)
-  endif()
-  set(BENCHMARKS
-        cache/cache_bench.cc
-        memtable/memtablerep_bench.cc
-        db/range_del_aggregator_bench.cc
-        tools/db_bench.cc
-        table/table_reader_bench.cc
-        util/filter_bench.cc
-        utilities/persistent_cache/hash_table_bench.cc
-  )
-  add_library(testharness OBJECT test_util/testharness.cc)
-  foreach(sourcefile ${BENCHMARKS})
-    get_filename_component(exename ${sourcefile} NAME_WE)
-    add_executable(${exename}${ARTIFACT_SUFFIX} ${sourcefile}
-      $<TARGET_OBJECTS:testharness>)
-    target_link_libraries(${exename}${ARTIFACT_SUFFIX} gtest ${LIBS})
-  endforeach(sourcefile ${BENCHMARKS})
+  add_executable(db_bench
+    tools/db_bench.cc
+    tools/db_bench_tool.cc)
+  target_link_libraries(db_bench
+    ${ROCKSDB_LIB})
+
+  add_executable(cache_bench
+    cache/cache_bench.cc)
+  target_link_libraries(cache_bench
+    ${ROCKSDB_LIB})
+
+  add_executable(memtablerep_bench
+    memtable/memtablerep_bench.cc)
+  target_link_libraries(memtablerep_bench
+    ${ROCKSDB_LIB})
+
+  add_executable(range_del_aggregator_bench
+    db/range_del_aggregator_bench.cc)
+  target_link_libraries(range_del_aggregator_bench
+    ${ROCKSDB_LIB})
+
+  add_executable(table_reader_bench
+    table/table_reader_bench.cc)
+  target_link_libraries(table_reader_bench
+    ${ROCKSDB_LIB} testharness)
+
+  add_executable(filter_bench
+    util/filter_bench.cc)
+  target_link_libraries(filter_bench
+    ${ROCKSDB_LIB})
+
+  add_executable(hash_table_bench
+    utilities/persistent_cache/hash_table_bench.cc)
+  target_link_libraries(hash_table_bench
+    ${ROCKSDB_LIB})
 endif()
 
 option(WITH_TOOLS "build with tools" ON)
 if(WITH_TOOLS)
   add_subdirectory(tools)
+  add_subdirectory(db_stress_tool)
+  add_custom_target(tools
+    DEPENDS ${tool_deps})
 endif()

HISTORY.md

@@ -1,34 +1,32 @@
 # Rocksdb Change Log
 ## Unreleased
+### Public API Change
+* Added a rocksdb::FileSystem class in include/rocksdb/file_system.h to encapsulate file creation/read/write operations, and an option DBOptions::file_system to allow a user to pass in an instance of rocksdb::FileSystem. If its a non-null value, this will take precendence over DBOptions::env for file operations. A new API rocksdb::FileSystem::Default() returns a platform default object. The DBOptions::env option and Env::Default() API will continue to be used for threading and other OS related functions, and where DBOptions::file_system is not specified, for file operations. For storage developers who are accustomed to rocksdb::Env, the interface in rocksdb::FileSystem is new and will probably undergo some changes as more storage systems are ported to it from rocksdb::Env. As of now, no env other than Posix has been ported to the new interface.
+* A new rocksdb::NewSstFileManager() API that allows the caller to pass in separate Env and FileSystem objects.
+* Changed Java API for RocksDB.keyMayExist functions to use Holder<byte[]> instead of StringBuilder, so that retrieved values need not decode to Strings.
+* A new `OptimisticTransactionDBOptions` Option that allows users to configure occ validation policy. The default policy changes from kValidateSerial to kValidateParallel to reduce mutex contention.
 
-## 6.6.4 (1/31/2020)
 ### Bug Fixes
-* Fixed issue #6316 that can cause a corruption of the MANIFEST file in the middle when writing to it fails due to no disk space.
-
-## 6.6.3 (01/24/2020)
-### Bug Fixes
-* Fix a bug that can cause write threads to hang when a slowdown/stall happens and there is a mix of writers with WriteOptions::no_slowdown set/unset.
-
-## 6.6.2 (01/13/2020)
-### Bug Fixes
-* Fixed a bug where non-L0 compaction input files were not considered to compute the `creation_time` of new compaction outputs.
-
-## 6.6.1 (01/02/2020)
-### Bug Fixes
-* Fix a bug in WriteBatchWithIndex::MultiGetFromBatchAndDB, which is called by Transaction::MultiGet, that causes due to stale pointer access when the number of keys is > 32
+* Fix a bug that can cause unnecessary bg thread to be scheduled(#6104).
+* Fix a bug in which a snapshot read could be affected by a DeleteRange after the snapshot (#6062).
+* Fix crash caused by concurrent CF iterations and drops(#6147).
 * Fixed two performance issues related to memtable history trimming. First, a new SuperVersion is now created only if some memtables were actually trimmed. Second, trimming is only scheduled if there is at least one flushed memtable that is kept in memory for the purposes of transaction conflict checking.
+* Fix a bug in WriteBatchWithIndex::MultiGetFromBatchAndDB, which is called by Transaction::MultiGet, that causes due to stale pointer access when the number of keys is > 32
 * BlobDB no longer updates the SST to blob file mapping upon failed compactions.
-* Fix a bug in which a snapshot read through an iterator could be affected by a DeleteRange after the snapshot (#6062).
 * Fixed a bug where BlobDB was comparing the `ColumnFamilyHandle` pointers themselves instead of only the column family IDs when checking whether an API call uses the default column family or not.
-* Delete superversions in BackgroundCallPurge.
-* Fix use-after-free and double-deleting files in BackgroundCallPurge().
+* Fix a race condition for cfd->log_number_ between manifest switch and memtable switch (PR 6249) when number of column families is greater than 1.
+
+### New Features
+* It is now possible to enable periodic compactions for the base DB when using BlobDB.
+* BlobDB now garbage collects non-TTL blobs when `enable_garbage_collection` is set to `true` in `BlobDBOptions`. Garbage collection is performed during compaction: any valid blobs located in the oldest N files (where N is the number of non-TTL blob files multiplied by the value of `BlobDBOptions::garbage_collection_cutoff`) encountered during compaction get relocated to new blob files, and old blob files are dropped once they are no longer needed. Note: we recommend enabling periodic compactions for the base DB when using this feature to deal with the case when some old blob files are kept alive by SSTs that otherwise do not get picked for compaction.
+* `db_bench` now supports the `garbage_collection_cutoff` option for BlobDB.
 
 ## 6.6.0 (11/25/2019)
 ### Bug Fixes
-* Fix data corruption casued by output of intra-L0 compaction on ingested file not being placed in correct order in L0.
+* Fix data corruption caused by output of intra-L0 compaction on ingested file not being placed in correct order in L0.
 * Fix a data race between Version::GetColumnFamilyMetaData() and Compaction::MarkFilesBeingCompacted() for access to being_compacted (#6056). The current fix acquires the db mutex during Version::GetColumnFamilyMetaData(), which may cause regression.
 * Fix a bug in DBIter that is_blob_ state isn't updated when iterating backward using seek.
-* Fix a bug when format_version=3, partitioned fitlers, and prefix search are used in conjunction. The bug could result into Seek::(prefix) returning NotFound for an existing prefix.
+* Fix a bug when format_version=3, partitioned filters, and prefix search are used in conjunction. The bug could result into Seek::(prefix) returning NotFound for an existing prefix.
 * Revert the feature "Merging iterator to avoid child iterator reseek for some cases (#5286)" since it might cause strong results when reseek happens with a different iterator upper bound.
 * Fix a bug causing a crash during ingest external file when background compaction cause severe error (file not found).
 * Fix a bug when partitioned filters and prefix search are used in conjunction, ::SeekForPrev could return invalid for an existing prefix. ::SeekForPrev might be called by the user, or internally on ::Prev, or within ::Seek if the return value involves Delete or a Merge operand.
@@ -41,13 +39,13 @@
 * Universal compaction to support options.periodic_compaction_seconds. A full compaction will be triggered if any file is over the threshold.
 * `GetLiveFilesMetaData` and `GetColumnFamilyMetaData` now expose the file number of SST files as well as the oldest blob file referenced by each SST.
 * A batched MultiGet API (DB::MultiGet()) that supports retrieving keys from multiple column families.
-* Full and partitioned filters in the block-based table use an improved Bloom filter implementation, enabled with format_version 5 (or above) because previous releases cannot read this filter. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single (full) filter. For example, the new Bloom filter has the same false postive rate at 9.55 bits per key as the old one at 10 bits per key, and a lower false positive rate at 16 bits per key than the old one at 100 bits per key.
+* Full and partitioned filters in the block-based table use an improved Bloom filter implementation, enabled with format_version 5 (or above) because previous releases cannot read this filter. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single (full) filter. For example, the new Bloom filter has the same false positive rate at 9.55 bits per key as the old one at 10 bits per key, and a lower false positive rate at 16 bits per key than the old one at 100 bits per key.
 * Added AVX2 instructions to USE_SSE builds to accelerate the new Bloom filter and XXH3-based hash function on compatible x86_64 platforms (Haswell and later, ~2014).
 * Support options.ttl or options.periodic_compaction_seconds with options.max_open_files = -1. File's oldest ancester time and file creation time will be written to manifest. If it is availalbe, this information will be used instead of creation_time and file_creation_time in table properties.
 * Setting options.ttl for universal compaction now has the same meaning as setting periodic_compaction_seconds.
 * SstFileMetaData also returns file creation time and oldest ancester time.
 * The `sst_dump` command line tool `recompress` command now displays how many blocks were compressed and how many were not, in particular how many were not compressed because the compression ratio was not met (12.5% threshold for GoodCompressionRatio), as seen in the `number.block.not_compressed` counter stat since version 6.0.0.
-* The block cache usage is now takes into account the overhead of metadata per each entry. This results into more accurate managment of memory. A side-effect of this feature is that less items are fit into the block cache of the same size, which would result to higher cache miss rates. This can be remedied by increasing the block cache size or passing kDontChargeCacheMetadata to its constuctor to restore the old behavior.
+* The block cache usage is now takes into account the overhead of metadata per each entry. This results into more accurate management of memory. A side-effect of this feature is that less items are fit into the block cache of the same size, which would result to higher cache miss rates. This can be remedied by increasing the block cache size or passing kDontChargeCacheMetadata to its constuctor to restore the old behavior.
 * When using BlobDB, a mapping is maintained and persisted in the MANIFEST between each SST file and the oldest non-TTL blob file it references.
 * `db_bench` now supports and by default issues non-TTL Puts to BlobDB. TTL Puts can be enabled by specifying a non-zero value for the `blob_db_max_ttl_range` command line parameter explicitly.
 * `sst_dump` now supports printing BlobDB blob indexes in a human-readable format. This can be enabled by specifying the `decode_blob_index` flag on the command line.
@@ -70,7 +68,7 @@
 
 ### Default Option Changes
 * Changed the default value of periodic_compaction_seconds to `UINT64_MAX - 1` which allows RocksDB to auto-tune periodic compaction scheduling. When using the default value, periodic compactions are now auto-enabled if a compaction filter is used. A value of `0` will turn off the feature completely.
-* Changed the default value of ttl to `UINT64_MAX - 1` which allows RocksDB to auto-tune ttl value. When using the default value, TTL will be auto-enabled to 30 days, when the feature is supported. To revert the old behavior, you can explictly set it to 0.
+* Changed the default value of ttl to `UINT64_MAX - 1` which allows RocksDB to auto-tune ttl value. When using the default value, TTL will be auto-enabled to 30 days, when the feature is supported. To revert the old behavior, you can explicitly set it to 0.
 
 ### Performance Improvements
 * For 64-bit hashing, RocksDB is standardizing on a slightly modified preview version of XXH3. This function is now used for many non-persisted hashes, along with fastrange64() in place of the modulus operator, and some benchmarks show a slight improvement.
@@ -111,7 +109,7 @@
 
 ## 6.4.0 (7/30/2019)
 ### Default Option Change
-* LRUCacheOptions.high_pri_pool_ratio is set to 0.5 (previously 0.0) by default, which means that by default midpoint insertion is enabled. The same change is made for the default value of high_pri_pool_ratio argument in NewLRUCache(). When block cache is not explictly created, the small block cache created by BlockBasedTable will still has this option to be 0.0.
+* LRUCacheOptions.high_pri_pool_ratio is set to 0.5 (previously 0.0) by default, which means that by default midpoint insertion is enabled. The same change is made for the default value of high_pri_pool_ratio argument in NewLRUCache(). When block cache is not explicitly created, the small block cache created by BlockBasedTable will still has this option to be 0.0.
 * Change BlockBasedTableOptions.cache_index_and_filter_blocks_with_high_priority's default value from false to true.
 
 ### Public API Change
@@ -135,7 +133,7 @@
 * Support loading custom objects in unit tests. In the affected unit tests, RocksDB will create custom Env objects based on environment variable TEST_ENV_URI. Users need to make sure custom object types are properly registered. For example, a static library should expose a `RegisterCustomObjects` function. By linking the unit test binary with the static library, the unit test can execute this function.
 
 ### Performance Improvements
-* Reduce iterator key comparision for upper/lower bound check.
+* Reduce iterator key comparison for upper/lower bound check.
 * Improve performance of row_cache: make reads with newer snapshots than data in an SST file share the same cache key, except in some transaction cases.
 * The compression dictionary is no longer copied to a new object upon retrieval.
 
@@ -231,7 +229,7 @@
 * Introduce two more stats levels, kExceptHistogramOrTimers and kExceptTimers.
 * Added a feature to perform data-block sampling for compressibility, and report stats to user.
 * Add support for trace filtering.
-* Add DBOptions.avoid_unnecessary_blocking_io. If true, we avoid file deletion when destorying ColumnFamilyHandle and Iterator. Instead, a job is scheduled to delete the files in background.
+* Add DBOptions.avoid_unnecessary_blocking_io. If true, we avoid file deletion when destroying ColumnFamilyHandle and Iterator. Instead, a job is scheduled to delete the files in background.
 
 ### Public API Change
 * Remove bundled fbson library.
@@ -610,7 +608,7 @@ if set to something > 0 user will see 2 changes in iterators behavior 1) only ke
 
 ## 5.2.0 (02/08/2017)
 ### Public API Change
-* NewLRUCache() will determine number of shard bits automatically based on capacity, if the user doesn't pass one. This also impacts the default block cache when the user doesn't explict provide one.
+* NewLRUCache() will determine number of shard bits automatically based on capacity, if the user doesn't pass one. This also impacts the default block cache when the user doesn't explicit provide one.
 * Change the default of delayed slowdown value to 16MB/s and further increase the L0 stop condition to 36 files.
 * Options::use_direct_writes and Options::use_direct_reads are now ready to use.
 * (Experimental) Two-level indexing that partition the index and creates a 2nd level index on the partitions. The feature can be enabled by setting kTwoLevelIndexSearch as IndexType and configuring index_per_partition.
@@ -681,7 +679,7 @@ if set to something > 0 user will see 2 changes in iterators behavior 1) only ke
 
 ### New Features
 * A tool to migrate DB after options change. See include/rocksdb/utilities/option_change_migration.h.
-* Add ReadOptions.background_purge_on_iterator_cleanup. If true, we avoid file deletion when destorying iterators.
+* Add ReadOptions.background_purge_on_iterator_cleanup. If true, we avoid file deletion when destroying iterators.
 
 ## 4.10.0 (7/5/2016)
 ### Public API Change

Makefile

@@ -460,6 +460,7 @@ TESTS = \
 	env_test \
 	env_logger_test \
 	hash_test \
+	random_test \
 	thread_local_test \
 	rate_limiter_test \
 	perf_context_test \
@@ -743,7 +744,8 @@ endif  # PLATFORM_SHARED_EXT
 	release tags tags0 valgrind_check whitebox_crash_test format static_lib shared_lib all \
 	dbg rocksdbjavastatic rocksdbjava install install-static install-shared uninstall \
 	analyze tools tools_lib \
-	blackbox_crash_test_with_atomic_flush whitebox_crash_test_with_atomic_flush
+	blackbox_crash_test_with_atomic_flush whitebox_crash_test_with_atomic_flush  \
+	blackbox_crash_test_with_txn whitebox_crash_test_with_txn
 
 
 all: $(LIBRARY) $(BENCHMARKS) tools tools_lib test_libs $(TESTS)
@@ -937,6 +939,7 @@ check: all
 	fi
 	rm -rf $(TMPD)
 ifneq ($(PLATFORM), OS_AIX)
+	python tools/check_all_python.py
 ifeq ($(filter -DROCKSDB_LITE,$(OPT)),)
 	python tools/ldb_test.py
 	sh tools/rocksdb_dump_test.sh
@@ -955,6 +958,8 @@ crash_test: whitebox_crash_test blackbox_crash_test
 
 crash_test_with_atomic_flush: whitebox_crash_test_with_atomic_flush blackbox_crash_test_with_atomic_flush
 
+crash_test_with_txn: whitebox_crash_test_with_txn blackbox_crash_test_with_txn
+
 blackbox_crash_test: db_stress
 	python -u tools/db_crashtest.py --simple blackbox $(CRASH_TEST_EXT_ARGS)
 	python -u tools/db_crashtest.py blackbox $(CRASH_TEST_EXT_ARGS)
@@ -962,6 +967,9 @@ blackbox_crash_test: db_stress
 blackbox_crash_test_with_atomic_flush: db_stress
 	python -u tools/db_crashtest.py --cf_consistency blackbox $(CRASH_TEST_EXT_ARGS)
 
+blackbox_crash_test_with_txn: db_stress
+	python -u tools/db_crashtest.py --txn blackbox $(CRASH_TEST_EXT_ARGS)
+
 ifeq ($(CRASH_TEST_KILL_ODD),)
   CRASH_TEST_KILL_ODD=888887
 endif
@@ -976,6 +984,10 @@ whitebox_crash_test_with_atomic_flush: db_stress
 	python -u tools/db_crashtest.py --cf_consistency whitebox  --random_kill_odd \
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
 
+whitebox_crash_test_with_txn: db_stress
+	python -u tools/db_crashtest.py --txn whitebox --random_kill_odd \
+      $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
+
 asan_check:
 	$(MAKE) clean
 	COMPILE_WITH_ASAN=1 $(MAKE) check -j32
@@ -991,6 +1003,11 @@ asan_crash_test_with_atomic_flush:
 	COMPILE_WITH_ASAN=1 $(MAKE) crash_test_with_atomic_flush
 	$(MAKE) clean
 
+asan_crash_test_with_txn:
+	$(MAKE) clean
+	COMPILE_WITH_ASAN=1 $(MAKE) crash_test_with_txn
+	$(MAKE) clean
+
 ubsan_check:
 	$(MAKE) clean
 	COMPILE_WITH_UBSAN=1 $(MAKE) check -j32
@@ -1006,6 +1023,11 @@ ubsan_crash_test_with_atomic_flush:
 	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test_with_atomic_flush
 	$(MAKE) clean
 
+ubsan_crash_test_with_txn:
+	$(MAKE) clean
+	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test_with_txn
+	$(MAKE) clean
+
 valgrind_test:
 	ROCKSDB_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 $(MAKE) valgrind_check
 
@@ -1075,6 +1097,9 @@ parallel_check: $(TESTS)
 		TMPD=$(TMPD) J=$(J) db_test=0 parloop;
 
 analyze: clean
+	USE_CLANG=1 $(MAKE) analyze_incremental
+
+analyze_incremental:
 	$(CLANG_SCAN_BUILD) --use-analyzer=$(CLANG_ANALYZER) \
 		--use-c++=$(CXX) --use-cc=$(CC) --status-bugs \
 		-o $(CURDIR)/scan_build_report \
@@ -1103,21 +1128,16 @@ unity_test: db/db_test.o db/db_test_util.o $(TESTHARNESS) $(TOOLLIBOBJECTS) unit
 rocksdb.h rocksdb.cc: build_tools/amalgamate.py Makefile $(LIB_SOURCES) unity.cc
 	build_tools/amalgamate.py -I. -i./include unity.cc -x include/rocksdb/c.h -H rocksdb.h -o rocksdb.cc
 
-clean: clean-ext-libraries-all clean-rocks clean-rocksjava
+clean: clean-ext-libraries-all clean-rocks
 
-clean-not-downloaded: clean-ext-libraries-bin clean-rocks clean-not-downloaded-rocksjava
+clean-not-downloaded: clean-ext-libraries-bin clean-rocks
 
 clean-rocks:
 	rm -f $(BENCHMARKS) $(TOOLS) $(TESTS) $(PARALLEL_TEST) $(LIBRARY) $(SHARED)
 	rm -rf $(CLEAN_FILES) ios-x86 ios-arm scan_build_report
 	$(FIND) . -name "*.[oda]" -exec rm -f {} \;
 	$(FIND) . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
-
-clean-rocksjava:
-	cd java && $(MAKE) clean
-
-clean-not-downloaded-rocksjava:
-	cd java && $(MAKE) clean-not-downloaded
+	cd java; $(MAKE) clean
 
 clean-ext-libraries-all:
 	rm -rf bzip2* snappy* zlib* lz4* zstd*
@@ -1188,7 +1208,7 @@ memtablerep_bench: memtable/memtablerep_bench.o $(LIBOBJECTS) $(TESTUTIL)
 filter_bench: util/filter_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)
 
-db_stress: tools/db_stress.o $(STRESSTOOLOBJECTS)
+db_stress: db_stress_tool/db_stress.o $(STRESSTOOLOBJECTS)
 	$(AM_LINK)
 
 write_stress: tools/write_stress.o $(LIBOBJECTS) $(TESTUTIL)
@@ -1230,6 +1250,9 @@ coding_test: util/coding_test.o $(LIBOBJECTS) $(TESTHARNESS)
 hash_test: util/hash_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+random_test: util/random_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 option_change_migration_test: utilities/option_change_migration/option_change_migration_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1811,7 +1834,7 @@ endif
 libz.a:
 	-rm -rf zlib-$(ZLIB_VER)
 ifeq (,$(wildcard ./zlib-$(ZLIB_VER).tar.gz))
-	curl --fail --output zlib-$(ZLIB_VER).tar.gz --location ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+	curl --output zlib-$(ZLIB_VER).tar.gz -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
 endif
 	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
@@ -1825,7 +1848,7 @@ endif
 libbz2.a:
 	-rm -rf bzip2-$(BZIP2_VER)
 ifeq (,$(wildcard ./bzip2-$(BZIP2_VER).tar.gz))
-	curl --fail --output bzip2-$(BZIP2_VER).tar.gz --location ${CURL_SSL_OPTS} ${BZIP2_DOWNLOAD_BASE}/bzip2-$(BZIP2_VER).tar.gz
+	curl --output bzip2-$(BZIP2_VER).tar.gz -L ${CURL_SSL_OPTS} ${BZIP2_DOWNLOAD_BASE}/bzip2-$(BZIP2_VER).tar.gz
 endif
 	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
@@ -1839,7 +1862,7 @@ endif
 libsnappy.a:
 	-rm -rf snappy-$(SNAPPY_VER)
 ifeq (,$(wildcard ./snappy-$(SNAPPY_VER).tar.gz))
-	curl --fail --output snappy-$(SNAPPY_VER).tar.gz --location ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
+	curl --output snappy-$(SNAPPY_VER).tar.gz -L ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
 endif
 	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
@@ -1854,7 +1877,7 @@ endif
 liblz4.a:
 	-rm -rf lz4-$(LZ4_VER)
 ifeq (,$(wildcard ./lz4-$(LZ4_VER).tar.gz))
-	curl --fail --output lz4-$(LZ4_VER).tar.gz --location ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
+	curl --output lz4-$(LZ4_VER).tar.gz -L ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
 endif
 	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
@@ -1868,7 +1891,7 @@ endif
 libzstd.a:
 	-rm -rf zstd-$(ZSTD_VER)
 ifeq (,$(wildcard ./zstd-$(ZSTD_VER).tar.gz))
-	curl --fail --output zstd-$(ZSTD_VER).tar.gz --location ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
+	curl --output zstd-$(ZSTD_VER).tar.gz -L ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
 endif
 	ZSTD_SHA256_ACTUAL=`$(SHA256_CMD) zstd-$(ZSTD_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZSTD_SHA256)" != "$$ZSTD_SHA256_ACTUAL" ]; then \
@@ -1938,35 +1961,35 @@ rocksdbjavastaticreleasedocker: rocksdbjavastatic rocksdbjavastaticdockerx86 roc
 
 rocksdbjavastaticdockerx86:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x86-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos6_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_x86-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos6_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerx86_64:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x64-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_x64-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerppc64le:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_ppc64le-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos7_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_ppc64le-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos7_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerarm64v8:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_arm64v8-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_arm64v8-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerx86musl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x86-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_x86-musl-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerx86_64musl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x64-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_x64-musl-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerppc64lemusl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_ppc64le-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_ppc64le-musl-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticdockerarm64v8musl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_arm64v8-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
+	docker run --rm --name rocksdb_linux_arm64v8-musl-be --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) evolvedbinary/rocksjava:alpine3_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh
 
 rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral
 
@@ -2032,6 +2055,7 @@ jtest_run:
 
 jtest: rocksdbjava
 	cd java;$(MAKE) sample;$(MAKE) test;
+	python tools/check_all_python.py # TODO peterd: find a better place for this check in CI targets
 
 jdb_bench:
 	cd java;$(MAKE) db_bench;
@@ -2079,6 +2103,9 @@ endif
 .cc.o:
 	$(AM_V_CC)$(CXX) $(CXXFLAGS) -c $< -o $@ $(COVERAGEFLAGS)
 
+.cpp.o:
+	$(AM_V_CC)$(CXX) $(CXXFLAGS) -c $< -o $@ $(COVERAGEFLAGS)
+
 .c.o:
 	$(AM_V_CC)$(CC) $(CFLAGS) -c $< -o $@
 endif
@@ -2089,6 +2116,10 @@ endif
 all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(ANALYZER_LIB_SOURCES) $(STRESS_LIB_SOURCES)
 DEPFILES = $(all_sources:.cc=.cc.d)
 
+ifeq ($(USE_FOLLY_DISTRIBUTED_MUTEX),1)
+  DEPFILES += $(FOLLY_SOURCES:.cpp=.cpp.d)
+endif
+
 # Add proper dependency support so changing a .h file forces a .cc file to
 # rebuild.
 
@@ -2098,6 +2129,10 @@ DEPFILES = $(all_sources:.cc=.cc.d)
 	@$(CXX) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) \
 	  -MM -MT'$@' -MT'$(<:.cc=.o)' "$<" -o '$@'
 
+%.cpp.d: %.cpp
+	@$(CXX) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) \
+	  -MM -MT'$@' -MT'$(<:.cpp=.o)' "$<" -o '$@'
+
 ifeq ($(HAVE_POWER8),1)
 DEPFILES_C = $(LIB_SOURCES_C:.c=.c.d)
 DEPFILES_ASM = $(LIB_SOURCES_ASM:.S=.S.d)

TARGETS

@@ -26,6 +26,7 @@ ROCKSDB_EXTERNAL_DEPS = [
     ("lz4", None, "lz4"),
     ("zstd", None),
     ("tbb", None),
+    ("liburing", None, "uring"),
     ("googletest", None, "gtest"),
 ]
 
@@ -46,6 +47,7 @@ ROCKSDB_OS_PREPROCESSOR_FLAGS = [
             "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
             "-DROCKSDB_RANGESYNC_PRESENT",
             "-DROCKSDB_SCHED_GETCPU_PRESENT",
+            "-DROCKSDB_IOURING_PRESENT",
             "-DHAVE_SSE42",
             "-DNUMA",
         ],
@@ -70,6 +72,7 @@ ROCKSDB_PREPROCESSOR_FLAGS = [
     "-DZSTD_STATIC_LINKING_ONLY",
     "-DGFLAGS=gflags",
     "-DTBB",
+    "-DLIBURING",
 
     # Added missing flags from output of build_detect_platform
     "-DROCKSDB_BACKTRACE",
@@ -178,6 +181,8 @@ cpp_library(
         "env/env_encryption.cc",
         "env/env_hdfs.cc",
         "env/env_posix.cc",
+        "env/file_system.cc",
+        "env/fs_posix.cc",
         "env/io_posix.cc",
         "env/mock_env.cc",
         "file/delete_scheduler.cc",
@@ -399,9 +404,17 @@ cpp_library(
 cpp_library(
     name = "rocksdb_stress_lib",
     srcs = [
+        "db_stress_tool/batched_ops_stress.cc",
+        "db_stress_tool/cf_consistency_stress.cc",
+        "db_stress_tool/db_stress_common.cc",
+        "db_stress_tool/db_stress_driver.cc",
+        "db_stress_tool/db_stress_gflags.cc",
+        "db_stress_tool/db_stress_shared_state.cc",
+        "db_stress_tool/db_stress_test_base.cc",
+        "db_stress_tool/db_stress_tool.cc",
+        "db_stress_tool/no_batched_ops_stress.cc",
         "test_util/testutil.cc",
         "tools/block_cache_analyzer/block_cache_trace_analyzer.cc",
-        "tools/db_stress_tool.cc",
         "tools/trace_analyzer_tool.cc",
     ],
     auto_headers = AutoHeaders.RECURSIVE_GLOB,
@@ -1234,6 +1247,13 @@ ROCKS_TESTS = [
         [],
         [],
     ],
+    [
+        "random_test",
+        "util/random_test.cc",
+        "serial",
+        [],
+        [],
+    ],
     [
         "range_del_aggregator_test",
         "db/range_del_aggregator_test.cc",

appveyor.yml

@@ -29,7 +29,7 @@ install:
   - md %THIRDPARTY_HOME%
   - echo "Building Snappy dependency..."
   - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output snappy-1.1.7.zip --location https://github.com/google/snappy/archive/1.1.7.zip
+  - curl -fsSL -o snappy-1.1.7.zip https://github.com/google/snappy/archive/1.1.7.zip
   - unzip snappy-1.1.7.zip
   - cd snappy-1.1.7
   - mkdir build
@@ -39,7 +39,7 @@ install:
   - msbuild Snappy.sln /p:Configuration=Release /p:Platform=x64
   - echo "Building LZ4 dependency..."
   - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output lz4-1.8.3.zip --location https://github.com/lz4/lz4/archive/v1.8.3.zip
+  - curl -fsSL -o lz4-1.8.3.zip https://github.com/lz4/lz4/archive/v1.8.3.zip
   - unzip lz4-1.8.3.zip
   - cd lz4-1.8.3\visual\VS2010
   - ps: $CMD="$Env:DEV_ENV"; & $CMD lz4.sln /upgrade
@@ -47,7 +47,7 @@ install:
   - msbuild lz4.sln /p:Configuration=Release /p:Platform=x64
   - echo "Building ZStd dependency..."
   - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output zstd-1.4.0.zip --location https://github.com/facebook/zstd/archive/v1.4.0.zip
+  - curl -fsSL -o zstd-1.4.0.zip https://github.com/facebook/zstd/archive/v1.4.0.zip
   - unzip zstd-1.4.0.zip
   - cd zstd-1.4.0\build\VS2010
   - ps: $CMD="$Env:DEV_ENV"; & $CMD zstd.sln /upgrade

buckifier/targets_cfg.py

@@ -32,6 +32,7 @@ ROCKSDB_EXTERNAL_DEPS = [
     ("lz4", None, "lz4"),
     ("zstd", None),
     ("tbb", None),
+    ("liburing", None, "uring"),
     ("googletest", None, "gtest"),
 ]
 
@@ -52,6 +53,7 @@ ROCKSDB_OS_PREPROCESSOR_FLAGS = [
             "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
             "-DROCKSDB_RANGESYNC_PRESENT",
             "-DROCKSDB_SCHED_GETCPU_PRESENT",
+            "-DROCKSDB_IOURING_PRESENT",
             "-DHAVE_SSE42",
             "-DNUMA",
         ],
@@ -76,6 +78,7 @@ ROCKSDB_PREPROCESSOR_FLAGS = [
     "-DZSTD_STATIC_LINKING_ONLY",
     "-DGFLAGS=gflags",
     "-DTBB",
+    "-DLIBURING",
 
     # Added missing flags from output of build_detect_platform
     "-DROCKSDB_BACKTRACE",

build_tools/build_detect_platform

@@ -150,6 +150,19 @@ case "$TARGET_OS" in
             PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
         fi
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
+        # check for liburing
+        $CXX $CFLAGS -x c++ - -luring -o /dev/null 2>/dev/null  <<EOF
+          #include <liburing.h>
+          int main() {
+            struct io_uring ring;
+            io_uring_queue_init(1, &ring, 0);
+            return 0;
+          }
+EOF
+        if [ "$?" = 0 ]; then
+          PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -luring"
+          COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_IOURING_PRESENT"
+        fi
         if test -z "$USE_FOLLY_DISTRIBUTED_MUTEX"; then
           USE_FOLLY_DISTRIBUTED_MUTEX=1
         fi

build_tools/dependencies_platform007.sh

@@ -13,6 +13,7 @@ JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/c26f08f47ac35fc31da2633b7da92d6b86
 NUMA_BASE=/mnt/gvfs/third-party2/numa/3f3fb57a5ccc5fd21c66416c0b83e0aa76a05376/2.0.11/platform007/ca4da3d
 LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/40c73d874898b386a71847f1b99115d93822d11f/1.4/platform007/6f3e0a9
 TBB_BASE=/mnt/gvfs/third-party2/tbb/4ce8e8dba77cdbd81b75d6f0c32fd7a1b76a11ec/2018_U5/platform007/ca4da3d
+LIBURING_BASE=/mnt/gvfs/third-party2/liburing/79427253fd0d42677255aacfe6d13bfe63f752eb/20190828/platform007/ca4da3d
 KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/fb251ecd2f5ae16f8671f7014c246e52a748fe0b/fb/platform007/da39a3e
 BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/ab9f09bba370e7066cafd4eb59752db93f2e8312/2.29.1/platform007/15a3614
 VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d42d152a15636529b0861ec493927200ebebca8e/3.15.0/platform007/ca4da3d

build_tools/error_filter.py

@@ -133,13 +133,16 @@ _TEST_NAME_TO_PARSERS = {
     '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],

build_tools/fbcode_config_platform007.sh

@@ -86,6 +86,15 @@ else
 fi
 CFLAGS+=" -DTBB"
 
+# location of LIBURING
+LIBURING_INCLUDE=" -isystem $LIBURING_BASE/include/"
+if test -z $PIC_BUILD; then
+  LIBURING_LIBS="$LIBURING_BASE/lib/liburing.a"
+else
+  LIBURING_LIBS="$LIBURING_BASE/lib/liburing_pic.a"
+fi
+CFLAGS+=" -DLIBURING"
+
 test "$USE_SSE" || USE_SSE=1
 export USE_SSE
 test "$PORTABLE" || PORTABLE=1
@@ -94,7 +103,7 @@ export PORTABLE
 BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"
 
-DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE"
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE $LIBURING_INCLUDE"
 
 STDLIBS="-L $GCC_BASE/lib64"
 
@@ -135,10 +144,10 @@ else
 fi
 
 CFLAGS+=" $DEPS_INCLUDE"
-CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42 -DROCKSDB_IOURING_PRESENT"
 CXXFLAGS+=" $CFLAGS"
 
-EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
+EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS $LIBURING_LIBS"
 EXEC_LDFLAGS+=" -B$BINUTILS/gold"
 EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
@@ -148,7 +157,7 @@ EXEC_LDFLAGS+=" -ldl"
 
 PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
 
-EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS $LIBURING_LIBS"
 
 VALGRIND_VER="$VALGRIND_BASE/bin/"
 

build_tools/format-diff.sh

@@ -13,9 +13,14 @@ then
   echo "You didn't have clang-format-diff.py and/or clang-format available in your computer!"
   echo "You can download clang-format-diff.py by running: "
   echo "    curl --location http://goo.gl/iUW1u2 -o ${CLANG_FORMAT_DIFF}"
-  echo "You can download clang-format by running: "
+  echo "You can download clang-format by running:"
   echo "    brew install clang-format"
+  echo "  Or"
+  echo "    apt install clang-format"
+  echo "  This might work too:"
+  echo "    yum install git-clang-format"
   echo "Then, move both files (i.e. ${CLANG_FORMAT_DIFF} and clang-format) to some directory within PATH=${PATH}"
+  echo "and make sure ${CLANG_FORMAT_DIFF} is executable."
   exit 128
 fi
 

build_tools/regression_build_test.sh

@@ -378,7 +378,7 @@ function send_to_ods {
     echo >&2 "ERROR: Key $key doesn't have a value."
     return
   fi
-  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
+  curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
     --connect-timeout 60
 }
 

build_tools/rocksdb-lego-determinator

@@ -450,6 +450,36 @@ STRESS_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB stress/crash test with txn
+#
+STRESS_CRASH_TEST_WITH_TXN_COMMANDS="[
+    {
+        'name':'Rocksdb Stress and Crash Test with txn',
+        'oncall':'$ONCALL',
+        'executeLocal': 'true',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build and run RocksDB debug stress tests',
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 db_stress || $CONTRUN_NAME=db_stress $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            {
+                'name':'Build and run RocksDB debug crash tests with txn',
+                'timeout': 86400,
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 crash_test_with_txn || $CONTRUN_NAME=crash_test_with_txn $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            $UPLOAD_DB_DIR,
+        ],
+        $REPORT
+    }
+]"
+
 # RocksDB write stress test.
 # We run on disk device on purpose (i.e. no $SHM)
 # because we want to add some randomness to fsync commands
@@ -541,6 +571,30 @@ ASAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB crash testing with txn under address sanitizer
+#
+ASAN_CRASH_TEST_WITH_TXN_COMMANDS="[
+    {
+        'name':'Rocksdb crash test with txn under ASAN',
+        'oncall':'$ONCALL',
+        'executeLocal': 'true',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build and run RocksDB debug asan_crash_test_with_txn',
+                'timeout': 86400,
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 asan_crash_test_with_txn || $CONTRUN_NAME=asan_crash_test_with_txn $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            $UPLOAD_DB_DIR,
+        ],
+        $REPORT
+    }
+]"
+
 #
 # RocksDB test under undefined behavior sanitizer
 #
@@ -609,6 +663,30 @@ UBSAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB crash testing with txn under undefined behavior sanitizer
+#
+UBSAN_CRASH_TEST_WITH_TXN_COMMANDS="[
+    {
+        'name':'Rocksdb crash test with txn under UBSAN',
+        'oncall':'$ONCALL',
+        'executeLocal': 'true',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build and run RocksDB debug ubsan_crash_test_with_txn',
+                'timeout': 86400,
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH $CLANG make J=1 ubsan_crash_test_with_txn || $CONTRUN_NAME=ubsan_crash_test_with_txn $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            $UPLOAD_DB_DIR,
+        ],
+        $REPORT
+    }
+]"
+
 #
 # RocksDB unit test under valgrind
 #
@@ -702,6 +780,30 @@ TSAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB crash test with txn under TSAN
+#
+TSAN_CRASH_TEST_WITH_TXN_COMMANDS="[
+    {
+        'name':'Rocksdb Crash Test with txn under TSAN',
+        'oncall':'$ONCALL',
+        'executeLocal': 'true',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Compile and run',
+                'timeout': 86400,
+                'shell':'set -o pipefail && $SHM $DEBUG $TSAN $TSAN_CRASH CRASH_TEST_KILL_ODD=1887 make J=1 crash_test_with_txn || $CONTRUN_NAME=tsan_crash_test_with_txn $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            $UPLOAD_DB_DIR,
+        ],
+        $REPORT
+    }
+]"
+
 #
 # RocksDB format compatible
 #
@@ -778,7 +880,7 @@ run_regression()
 
   # parameters: $1 -- key, $2 -- value
   function send_size_to_ods {
-    curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
+    curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
       --connect-timeout 60
   }
 
@@ -889,6 +991,9 @@ case $1 in
   stress_crash_with_atomic_flush)
     echo $STRESS_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS
     ;;
+  stress_crash_with_txn)
+    echo $STRESS_CRASH_TEST_WITH_TXN_COMMANDS
+    ;;
   write_stress)
     echo $WRITE_STRESS_COMMANDS
     ;;
@@ -901,6 +1006,9 @@ case $1 in
   asan_crash_with_atomic_flush)
     echo $ASAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS
     ;;
+  asan_crash_with_txn)
+    echo $ASAN_CRASH_TEST_WITH_TXN_COMMANDS
+    ;;
   ubsan)
     echo $UBSAN_TEST_COMMANDS
     ;;
@@ -910,6 +1018,9 @@ case $1 in
   ubsan_crash_with_atomic_flush)
     echo $UBSAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS
     ;;
+  ubsan_crash_with_txn)
+    echo $UBSAN_CRASH_TEST_WITH_TXN_COMMANDS
+    ;;
   valgrind)
     echo $VALGRIND_TEST_COMMANDS
     ;;
@@ -922,6 +1033,9 @@ case $1 in
   tsan_crash_with_atomic_flush)
     echo $TSAN_CRASH_TEST_WITH_ATOMIC_FLUSH_COMMANDS
     ;;
+  tsan_crash_with_txn)
+    echo $TSAN_CRASH_TEST_WITH_TXN_COMMANDS
+    ;;
   format_compatible)
     echo $FORMAT_COMPATIBLE_COMMANDS
     ;;

build_tools/update_dependencies.sh

@@ -92,6 +92,7 @@ get_lib_base jemalloc   LATEST  platform007
 get_lib_base numa       LATEST  platform007
 get_lib_base libunwind  LATEST  platform007
 get_lib_base tbb        LATEST  platform007
+get_lib_base liburing   LATEST  platform007
 
 get_lib_base kernel-headers fb platform007
 get_lib_base binutils   LATEST centos7-native

cmake/modules/Findgflags.cmake

@@ -0,0 +1,29 @@
+# - Find gflags library
+# Find the gflags includes and library
+#
+# gflags_INCLUDE_DIR - where to find gflags.h.
+# gflags_LIBRARIES - List of libraries when using gflags.
+# gflags_FOUND - True if gflags found.
+
+find_path(GFLAGS_INCLUDE_DIR
+  NAMES gflags/gflags.h)
+
+find_library(GFLAGS_LIBRARIES
+  NAMES gflags)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(gflags
+  DEFAULT_MSG GFLAGS_LIBRARIES GFLAGS_INCLUDE_DIR)
+
+mark_as_advanced(
+  GFLAGS_LIBRARIES
+  GFLAGS_INCLUDE_DIR)
+
+if(gflags_FOUND AND NOT (TARGET gflags::gflags))
+  add_library(gflags::gflags UNKNOWN IMPORTED)
+  set_target_properties(gflags::gflags
+    PROPERTIES
+      IMPORTED_LOCATION ${GFLAGS_LIBRARIES}
+      INTERFACE_INCLUDE_DIRECTORIES ${GFLAGS_INCLUDE_DIR}
+      IMPORTED_LINK_INTERFACE_LANGUAGES "CXX")
+endif()

db/blob_index.h

@@ -111,7 +111,7 @@ class BlobIndex {
     return Status::OK();
   }
 
-  std::string DebugString(bool output_hex) {
+  std::string DebugString(bool output_hex) const {
     std::ostringstream oss;
 
     if (IsInlined()) {

db/builder.cc

@@ -66,8 +66,9 @@ TableBuilder* NewTableBuilder(
 }
 
 Status BuildTable(
-    const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
-    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    const std::string& dbname, Env* env, FileSystem* fs,
+    const ImmutableCFOptions& ioptions,
+    const MutableCFOptions& mutable_cf_options, const FileOptions& file_options,
     TableCache* table_cache, InternalIterator* iter,
     std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
         range_del_iters,
@@ -115,12 +116,12 @@ Status BuildTable(
     compression_opts_for_flush.max_dict_bytes = 0;
     compression_opts_for_flush.zstd_max_train_bytes = 0;
     {
-      std::unique_ptr<WritableFile> file;
+      std::unique_ptr<FSWritableFile> file;
 #ifndef NDEBUG
-      bool use_direct_writes = env_options.use_direct_writes;
+      bool use_direct_writes = file_options.use_direct_writes;
       TEST_SYNC_POINT_CALLBACK("BuildTable:create_file", &use_direct_writes);
 #endif  // !NDEBUG
-      s = NewWritableFile(env, fname, &file, env_options);
+      s = NewWritableFile(fs, fname, &file, file_options);
       if (!s.ok()) {
         EventHelpers::LogAndNotifyTableFileCreationFinished(
             event_logger, ioptions.listeners, dbname, column_family_name, fname,
@@ -131,7 +132,7 @@ Status BuildTable(
       file->SetWriteLifeTimeHint(write_hint);
 
       file_writer.reset(
-          new WritableFileWriter(std::move(file), fname, env_options, env,
+          new WritableFileWriter(std::move(file), fname, file_options, env,
                                  ioptions.statistics, ioptions.listeners));
       builder = NewTableBuilder(
           ioptions, mutable_cf_options, internal_comparator,
@@ -218,7 +219,7 @@ Status BuildTable(
       // we will regrad this verification as user reads since the goal is
       // to cache it here for further user reads
       std::unique_ptr<InternalIterator> it(table_cache->NewIterator(
-          ReadOptions(), env_options, internal_comparator, *meta,
+          ReadOptions(), file_options, internal_comparator, *meta,
           nullptr /* range_del_agg */,
           mutable_cf_options.prefix_extractor.get(), nullptr,
           (internal_stats == nullptr) ? nullptr
@@ -241,7 +242,7 @@ Status BuildTable(
   }
 
   if (!s.ok() || meta->fd.GetFileSize() == 0) {
-    env->DeleteFile(fname);
+    fs->DeleteFile(fname, IOOptions(), nullptr);
   }
 
   if (meta->fd.GetFileSize() == 0) {

db/builder.h

@@ -62,8 +62,9 @@ TableBuilder* NewTableBuilder(
 // @param column_family_name Name of the column family that is also identified
 //    by column_family_id, or empty string if unknown.
 extern Status BuildTable(
-    const std::string& dbname, Env* env, const ImmutableCFOptions& options,
-    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    const std::string& dbname, Env* env, FileSystem* fs,
+    const ImmutableCFOptions& options,
+    const MutableCFOptions& mutable_cf_options, const FileOptions& file_options,
     TableCache* table_cache, InternalIterator* iter,
     std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
         range_del_iters,

db/c.cc

@@ -852,6 +852,17 @@ void rocksdb_delete_cf(
         Slice(key, keylen)));
 }
 
+void rocksdb_delete_range_cf(rocksdb_t* db,
+                             const rocksdb_writeoptions_t* options,
+                             rocksdb_column_family_handle_t* column_family,
+                             const char* start_key, size_t start_key_len,
+                             const char* end_key, size_t end_key_len,
+                             char** errptr) {
+  SaveError(errptr, db->rep->DeleteRange(options->rep, column_family->rep,
+                                         Slice(start_key, start_key_len),
+                                         Slice(end_key, end_key_len)));
+}
+
 void rocksdb_merge(
     rocksdb_t* db,
     const rocksdb_writeoptions_t* options,
@@ -2032,6 +2043,17 @@ void rocksdb_block_based_options_set_index_type(
   options->rep.index_type = static_cast<BlockBasedTableOptions::IndexType>(v);
 }
 
+void rocksdb_block_based_options_set_data_block_index_type(
+    rocksdb_block_based_table_options_t* options, int v) {
+  options->rep.data_block_index_type =
+          static_cast<BlockBasedTableOptions::DataBlockIndexType>(v);
+}
+
+void rocksdb_block_based_options_set_data_block_hash_ratio(
+    rocksdb_block_based_table_options_t* options, double v) {
+  options->rep.data_block_hash_table_util_ratio = v;
+}
+
 void rocksdb_block_based_options_set_hash_index_allow_collision(
     rocksdb_block_based_table_options_t* options, unsigned char v) {
   options->rep.hash_index_allow_collision = v;
@@ -3037,7 +3059,16 @@ rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_format(int bits_per_ke
     bool KeyMayMatch(const Slice& key, const Slice& filter) const override {
       return rep_->KeyMayMatch(key, filter);
     }
-    static void DoNothing(void*) { }
+    // No need to override GetFilterBitsBuilder if this one is overridden
+    rocksdb::FilterBitsBuilder* GetBuilderWithContext(
+        const rocksdb::FilterBuildingContext& context) const override {
+      return rep_->GetBuilderWithContext(context);
+    }
+    rocksdb::FilterBitsReader* GetFilterBitsReader(
+        const Slice& contents) const override {
+      return rep_->GetFilterBitsReader(contents);
+    }
+    static void DoNothing(void*) {}
   };
   Wrapper* wrapper = new Wrapper;
   wrapper->rep_ = NewBloomFilterPolicy(bits_per_key, original_format);

db/c_test.c

@@ -487,8 +487,11 @@ int main(int argc, char** argv) {
   rocksdb_options_set_paranoid_checks(options, 1);
   rocksdb_options_set_max_open_files(options, 10);
   rocksdb_options_set_base_background_compactions(options, 1);
+
   table_options = rocksdb_block_based_options_create();
   rocksdb_block_based_options_set_block_cache(table_options, cache);
+  rocksdb_block_based_options_set_data_block_index_type(table_options, 1);
+  rocksdb_block_based_options_set_data_block_hash_ratio(table_options, 0.75);
   rocksdb_options_set_block_based_table_factory(options, table_options);
 
   rocksdb_options_set_compression(options, rocksdb_no_compression);
@@ -1048,17 +1051,20 @@ int main(int argc, char** argv) {
   }
 
   StartPhase("filter");
-  for (run = 0; run < 2; run++) {
-    // First run uses custom filter, second run uses bloom filter
+  for (run = 0; run <= 2; run++) {
+    // First run uses custom filter
+    // Second run uses old block-based bloom filter
+    // Third run uses full bloom filter
     CheckNoError(err);
     rocksdb_filterpolicy_t* policy;
     if (run == 0) {
-      policy = rocksdb_filterpolicy_create(
-          NULL, FilterDestroy, FilterCreate, FilterKeyMatch, NULL, FilterName);
+      policy = rocksdb_filterpolicy_create(NULL, FilterDestroy, FilterCreate,
+                                           FilterKeyMatch, NULL, FilterName);
+    } else if (run == 1) {
+      policy = rocksdb_filterpolicy_create_bloom(8);
     } else {
-      policy = rocksdb_filterpolicy_create_bloom(10);
+      policy = rocksdb_filterpolicy_create_bloom_full(8);
     }
-
     rocksdb_block_based_options_set_filter_policy(table_options, policy);
 
     // Create new database
@@ -1071,12 +1077,24 @@ int main(int argc, char** argv) {
     CheckNoError(err);
     rocksdb_put(db, woptions, "bar", 3, "barvalue", 8, &err);
     CheckNoError(err);
+
+    {
+      // Add enough keys to get just one reasonably populated Bloom filter
+      const int keys_to_add = 1500;
+      int i;
+      char keybuf[100];
+      for (i = 0; i < keys_to_add; i++) {
+        snprintf(keybuf, sizeof(keybuf), "yes%020d", i);
+        rocksdb_put(db, woptions, keybuf, strlen(keybuf), "val", 3, &err);
+        CheckNoError(err);
+      }
+    }
     rocksdb_compact_range(db, NULL, 0, NULL, 0);
 
     fake_filter_result = 1;
     CheckGet(db, roptions, "foo", "foovalue");
     CheckGet(db, roptions, "bar", "barvalue");
-    if (phase == 0) {
+    if (run == 0) {
       // Must not find value when custom filter returns false
       fake_filter_result = 0;
       CheckGet(db, roptions, "foo", NULL);
@@ -1086,6 +1104,43 @@ int main(int argc, char** argv) {
       CheckGet(db, roptions, "foo", "foovalue");
       CheckGet(db, roptions, "bar", "barvalue");
     }
+
+    {
+      // Query some keys not added to identify Bloom filter implementation
+      // from false positive queries, using perfcontext to detect Bloom
+      // filter behavior
+      rocksdb_perfcontext_t* perf = rocksdb_perfcontext_create();
+      rocksdb_perfcontext_reset(perf);
+
+      const int keys_to_query = 10000;
+      int i;
+      char keybuf[100];
+      for (i = 0; i < keys_to_query; i++) {
+        fake_filter_result = i % 2;
+        snprintf(keybuf, sizeof(keybuf), "no%020d", i);
+        CheckGet(db, roptions, keybuf, NULL);
+      }
+
+      const int hits =
+          (int)rocksdb_perfcontext_metric(perf, rocksdb_bloom_sst_hit_count);
+      if (run == 0) {
+        // Due to half true, half false with fake filter result
+        CheckCondition(hits == keys_to_query / 2);
+      } else if (run == 1) {
+        // Essentially a fingerprint of the block-based Bloom schema
+        CheckCondition(hits == 241);
+      } else {
+        // Essentially a fingerprint of the full Bloom schema(s),
+        // format_version < 5, which vary for three different CACHE_LINE_SIZEs
+        CheckCondition(hits == 224 || hits == 180 || hits == 125);
+      }
+      CheckCondition(
+          (keys_to_query - hits) ==
+          (int)rocksdb_perfcontext_metric(perf, rocksdb_bloom_sst_miss_count));
+
+      rocksdb_perfcontext_destroy(perf);
+    }
+
     // Reset the policy
     rocksdb_block_based_options_set_filter_policy(table_options, NULL);
     rocksdb_options_set_block_based_table_factory(options, table_options);
@@ -1189,6 +1244,15 @@ int main(int argc, char** argv) {
     rocksdb_put_cf(db, woptions, handles[1], "foo", 3, "hello", 5, &err);
     CheckNoError(err);
 
+    rocksdb_put_cf(db, woptions, handles[1], "foobar1", 7, "hello1", 6, &err);
+    CheckNoError(err);
+    rocksdb_put_cf(db, woptions, handles[1], "foobar2", 7, "hello2", 6, &err);
+    CheckNoError(err);
+    rocksdb_put_cf(db, woptions, handles[1], "foobar3", 7, "hello3", 6, &err);
+    CheckNoError(err);
+    rocksdb_put_cf(db, woptions, handles[1], "foobar4", 7, "hello4", 6, &err);
+    CheckNoError(err);
+
     rocksdb_flushoptions_t *flush_options = rocksdb_flushoptions_create();
     rocksdb_flushoptions_set_wait(flush_options, 1);
     rocksdb_flush_cf(db, flush_options, handles[1], &err);
@@ -1201,6 +1265,10 @@ int main(int argc, char** argv) {
     rocksdb_delete_cf(db, woptions, handles[1], "foo", 3, &err);
     CheckNoError(err);
 
+    rocksdb_delete_range_cf(db, woptions, handles[1], "foobar2", 7, "foobar4",
+                            7, &err);
+    CheckNoError(err);
+
     CheckGetCF(db, roptions, handles[1], "foo", NULL);
     CheckPinGetCF(db, roptions, handles[1], "foo", NULL);
 
@@ -1253,7 +1321,7 @@ int main(int argc, char** argv) {
     for (i = 0; rocksdb_iter_valid(iter) != 0; rocksdb_iter_next(iter)) {
       i++;
     }
-    CheckCondition(i == 1);
+    CheckCondition(i == 3);
     rocksdb_iter_get_error(iter, &err);
     CheckNoError(err);
     rocksdb_iter_destroy(iter);
@@ -1277,7 +1345,7 @@ int main(int argc, char** argv) {
     for (i = 0; rocksdb_iter_valid(iter) != 0; rocksdb_iter_next(iter)) {
       i++;
     }
-    CheckCondition(i == 1);
+    CheckCondition(i == 3);
     rocksdb_iter_get_error(iter, &err);
     CheckNoError(err);
     rocksdb_iter_destroy(iter);

db/column_family.cc

@@ -60,11 +60,8 @@ ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
     ColumnFamilyOptions initial_cf_options_copy = cfd_->initial_cf_options();
     JobContext job_context(0);
     mutex_->Lock();
-    if (cfd_->Unref()) {
-      bool dropped = cfd_->IsDropped();
-
-      delete cfd_;
-
+    bool dropped = cfd_->IsDropped();
+    if (cfd_->UnrefAndTryDelete()) {
       if (dropped) {
         db_->FindObsoleteFiles(&job_context, false, true);
       }
@@ -163,6 +160,11 @@ Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
     return Status::InvalidArgument(
         "Memtable doesn't concurrent writes (allow_concurrent_memtable_write)");
   }
+  if (cf_options.max_successive_merges != 0) {
+    return Status::InvalidArgument(
+        "max_successive_merges > 0 is incompatible "
+        "with concurrent writes (allow_concurrent_memtable_write)");
+  }
   return Status::OK();
 }
 
@@ -439,13 +441,18 @@ void SuperVersion::Cleanup() {
     to_delete.push_back(m);
   }
   current->Unref();
+  if (cfd->Unref()) {
+    delete cfd;
+  }
 }
 
-void SuperVersion::Init(MemTable* new_mem, MemTableListVersion* new_imm,
-                        Version* new_current) {
+void SuperVersion::Init(ColumnFamilyData* new_cfd, MemTable* new_mem,
+                        MemTableListVersion* new_imm, Version* new_current) {
+  cfd = new_cfd;
   mem = new_mem;
   imm = new_imm;
   current = new_current;
+  cfd->Ref();
   mem->Ref();
   imm->Ref();
   current->Ref();
@@ -473,7 +480,7 @@ ColumnFamilyData::ColumnFamilyData(
     uint32_t id, const std::string& name, Version* _dummy_versions,
     Cache* _table_cache, WriteBufferManager* write_buffer_manager,
     const ColumnFamilyOptions& cf_options, const ImmutableDBOptions& db_options,
-    const EnvOptions& env_options, ColumnFamilySet* column_family_set,
+    const FileOptions& file_options, ColumnFamilySet* column_family_set,
     BlockCacheTracer* const block_cache_tracer)
     : id_(id),
       name_(name),
@@ -515,7 +522,7 @@ ColumnFamilyData::ColumnFamilyData(
   if (_dummy_versions != nullptr) {
     internal_stats_.reset(
         new InternalStats(ioptions_.num_levels, db_options.env, this));
-    table_cache_.reset(new TableCache(ioptions_, env_options, _table_cache,
+    table_cache_.reset(new TableCache(ioptions_, file_options, _table_cache,
                                       block_cache_tracer));
     if (ioptions_.compaction_style == kCompactionStyleLevel) {
       compaction_picker_.reset(
@@ -581,21 +588,7 @@ ColumnFamilyData::~ColumnFamilyData() {
   // compaction_queue_ and we destroyed it
   assert(!queued_for_flush_);
   assert(!queued_for_compaction_);
-
-  if (super_version_ != nullptr) {
-    // Release SuperVersion reference kept in ThreadLocalPtr.
-    // This must be done outside of mutex_ since unref handler can lock mutex.
-    super_version_->db_mutex->Unlock();
-    local_sv_.reset();
-    super_version_->db_mutex->Lock();
-
-    bool is_last_reference __attribute__((__unused__));
-    is_last_reference = super_version_->Unref();
-    assert(is_last_reference);
-    super_version_->Cleanup();
-    delete super_version_;
-    super_version_ = nullptr;
-  }
+  assert(super_version_ == nullptr);
 
   if (dummy_versions_ != nullptr) {
     // List must be empty
@@ -615,6 +608,36 @@ ColumnFamilyData::~ColumnFamilyData() {
   }
 }
 
+bool ColumnFamilyData::UnrefAndTryDelete() {
+  int old_refs = refs_.fetch_sub(1);
+  assert(old_refs > 0);
+
+  if (old_refs == 1) {
+    assert(super_version_ == nullptr);
+    delete this;
+    return true;
+  }
+
+  if (old_refs == 2 && super_version_ != nullptr) {
+    // Only the super_version_ holds me
+    SuperVersion* sv = super_version_;
+    super_version_ = nullptr;
+    // Release SuperVersion reference kept in ThreadLocalPtr.
+    // This must be done outside of mutex_ since unref handler can lock mutex.
+    sv->db_mutex->Unlock();
+    local_sv_.reset();
+    sv->db_mutex->Lock();
+
+    if (sv->Unref()) {
+      // May delete this ColumnFamilyData after calling Cleanup()
+      sv->Cleanup();
+      delete sv;
+      return true;
+    }
+  }
+  return false;
+}
+
 void ColumnFamilyData::SetDropped() {
   // can't drop default CF
   assert(id_ != 0);
@@ -949,8 +972,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
   return write_stall_condition;
 }
 
-const EnvOptions* ColumnFamilyData::soptions() const {
-  return &(column_family_set_->env_options_);
+const FileOptions* ColumnFamilyData::soptions() const {
+  return &(column_family_set_->file_options_);
 }
 
 void ColumnFamilyData::SetCurrent(Version* current_version) {
@@ -1172,7 +1195,7 @@ void ColumnFamilyData::InstallSuperVersion(
   SuperVersion* new_superversion = sv_context->new_superversion.release();
   new_superversion->db_mutex = db_mutex;
   new_superversion->mutable_cf_options = mutable_cf_options;
-  new_superversion->Init(mem_, imm_.current(), current_);
+  new_superversion->Init(this, mem_, imm_.current(), current_);
   SuperVersion* old_superversion = super_version_;
   super_version_ = new_superversion;
   ++super_version_number_;
@@ -1288,6 +1311,10 @@ Env::WriteLifeTimeHint ColumnFamilyData::CalculateSSTWriteHint(int level) {
   // L1: medium, L2: long, ...
   if (level - base_level >= 2) {
     return Env::WLTH_EXTREME;
+  } else if (level < base_level) {
+    // There is no restriction which prevents level passed in to be smaller
+    // than base_level.
+    return Env::WLTH_MEDIUM;
   }
   return static_cast<Env::WriteLifeTimeHint>(level - base_level +
                             static_cast<int>(Env::WLTH_MEDIUM));
@@ -1320,7 +1347,7 @@ Directory* ColumnFamilyData::GetDataDir(size_t path_id) const {
 
 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                  const ImmutableDBOptions* db_options,
-                                 const EnvOptions& env_options,
+                                 const FileOptions& file_options,
                                  Cache* table_cache,
                                  WriteBufferManager* write_buffer_manager,
                                  WriteController* write_controller,
@@ -1328,11 +1355,11 @@ ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
     : max_column_family_(0),
       dummy_cfd_(new ColumnFamilyData(
           0, "", nullptr, nullptr, nullptr, ColumnFamilyOptions(), *db_options,
-          env_options, nullptr, block_cache_tracer)),
+          file_options, nullptr, block_cache_tracer)),
       default_cfd_cache_(nullptr),
       db_name_(dbname),
       db_options_(db_options),
-      env_options_(env_options),
+      file_options_(file_options),
       table_cache_(table_cache),
       write_buffer_manager_(write_buffer_manager),
       write_controller_(write_controller),
@@ -1347,14 +1374,12 @@ ColumnFamilySet::~ColumnFamilySet() {
     // cfd destructor will delete itself from column_family_data_
     auto cfd = column_family_data_.begin()->second;
     bool last_ref __attribute__((__unused__));
-    last_ref = cfd->Unref();
+    last_ref = cfd->UnrefAndTryDelete();
     assert(last_ref);
-    delete cfd;
   }
   bool dummy_last_ref __attribute__((__unused__));
-  dummy_last_ref = dummy_cfd_->Unref();
+  dummy_last_ref = dummy_cfd_->UnrefAndTryDelete();
   assert(dummy_last_ref);
-  delete dummy_cfd_;
 }
 
 ColumnFamilyData* ColumnFamilySet::GetDefault() const {
@@ -1404,7 +1429,7 @@ ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
   assert(column_families_.find(name) == column_families_.end());
   ColumnFamilyData* new_cfd = new ColumnFamilyData(
       id, name, dummy_versions, table_cache_, write_buffer_manager_, options,
-      *db_options_, env_options_, this, block_cache_tracer_);
+      *db_options_, file_options_, this, block_cache_tracer_);
   column_families_.insert({name, id});
   column_family_data_.insert({id, new_cfd});
   max_column_family_ = std::max(max_column_family_, id);

db/column_family.h

@@ -198,6 +198,7 @@ class ColumnFamilyHandleInternal : public ColumnFamilyHandleImpl {
 struct SuperVersion {
   // Accessing members of this class is not thread-safe and requires external
   // synchronization (ie db mutex held or on write thread).
+  ColumnFamilyData* cfd;
   MemTable* mem;
   MemTableListVersion* imm;
   Version* current;
@@ -221,8 +222,8 @@ struct SuperVersion {
   // that needs to be deleted in to_delete vector. Unrefing those
   // objects needs to be done in the mutex
   void Cleanup();
-  void Init(MemTable* new_mem, MemTableListVersion* new_imm,
-            Version* new_current);
+  void Init(ColumnFamilyData* new_cfd, MemTable* new_mem,
+            MemTableListVersion* new_imm, Version* new_current);
 
   // The value of dummy is not actually used. kSVInUse takes its address as a
   // mark in the thread local storage to indicate the SuperVersion is in use
@@ -288,6 +289,11 @@ class ColumnFamilyData {
     return old_refs == 1;
   }
 
+  // UnrefAndTryDelete() decreases the reference count and do free if needed,
+  // return true if this is freed else false, UnrefAndTryDelete() can only
+  // be called while holding a DB mutex, or during single-threaded recovery.
+  bool UnrefAndTryDelete();
+
   // SetDropped() can only be called under following conditions:
   // 1) Holding a DB mutex,
   // 2) from single-threaded write thread, AND
@@ -318,7 +324,7 @@ class ColumnFamilyData {
   }
   FlushReason GetFlushReason() const { return flush_reason_; }
   // thread-safe
-  const EnvOptions* soptions() const;
+  const FileOptions* soptions() const;
   const ImmutableCFOptions* ioptions() const { return &ioptions_; }
   // REQUIRES: DB mutex held
   // This returns the MutableCFOptions used by current SuperVersion
@@ -504,7 +510,7 @@ class ColumnFamilyData {
                    WriteBufferManager* write_buffer_manager,
                    const ColumnFamilyOptions& options,
                    const ImmutableDBOptions& db_options,
-                   const EnvOptions& env_options,
+                   const FileOptions& file_options,
                    ColumnFamilySet* column_family_set,
                    BlockCacheTracer* const block_cache_tracer);
 
@@ -632,7 +638,7 @@ class ColumnFamilySet {
 
   ColumnFamilySet(const std::string& dbname,
                   const ImmutableDBOptions* db_options,
-                  const EnvOptions& env_options, Cache* table_cache,
+                  const FileOptions& file_options, Cache* table_cache,
                   WriteBufferManager* write_buffer_manager,
                   WriteController* write_controller,
                   BlockCacheTracer* const block_cache_tracer);
@@ -690,7 +696,7 @@ class ColumnFamilySet {
 
   const std::string db_name_;
   const ImmutableDBOptions* const db_options_;
-  const EnvOptions env_options_;
+  const FileOptions file_options_;
   Cache* table_cache_;
   WriteBufferManager* write_buffer_manager_;
   WriteController* write_controller_;

db/column_family_test.cc

@@ -2365,6 +2365,45 @@ TEST_P(ColumnFamilyTest, ReadDroppedColumnFamily) {
   }
 }
 
+TEST_P(ColumnFamilyTest, LiveIteratorWithDroppedColumnFamily) {
+  db_options_.create_missing_column_families = true;
+  db_options_.max_open_files = 20;
+  // delete obsolete files always
+  db_options_.delete_obsolete_files_period_micros = 0;
+  Open({"default", "one", "two"});
+  ColumnFamilyOptions options;
+  options.level0_file_num_compaction_trigger = 100;
+  options.level0_slowdown_writes_trigger = 200;
+  options.level0_stop_writes_trigger = 200;
+  options.write_buffer_size = 100000;  // small write buffer size
+  Reopen({options, options, options});
+
+  // 1MB should create ~10 files for each CF
+  int kKeysNum = 10000;
+  PutRandomData(1, kKeysNum, 100);
+
+  {
+    std::unique_ptr<Iterator> iterator(
+        db_->NewIterator(ReadOptions(), handles_[1]));
+    iterator->SeekToFirst();
+
+    DropColumnFamilies({1});
+
+    // Make sure iterator created can still be used.
+    int count = 0;
+    for (; iterator->Valid(); iterator->Next()) {
+      ASSERT_OK(iterator->status());
+      ++count;
+    }
+    ASSERT_OK(iterator->status());
+    ASSERT_EQ(count, kKeysNum);
+  }
+
+  Reopen();
+  Close();
+  Destroy();
+}
+
 TEST_P(ColumnFamilyTest, FlushAndDropRaceCondition) {
   db_options_.create_missing_column_families = true;
   Open({"default", "one"});

db/compaction/compaction.cc

@@ -275,9 +275,7 @@ Compaction::~Compaction() {
     input_version_->Unref();
   }
   if (cfd_ != nullptr) {
-    if (cfd_->Unref()) {
-      delete cfd_;
-    }
+    cfd_->UnrefAndTryDelete();
   }
 }
 
@@ -547,13 +545,11 @@ bool Compaction::ShouldFormSubcompactions() const {
 
 uint64_t Compaction::MinInputFileOldestAncesterTime() const {
   uint64_t min_oldest_ancester_time = port::kMaxUint64;
-  for (const auto& level_files : inputs_) {
-    for (const auto& file : level_files.files) {
-      uint64_t oldest_ancester_time = file->TryGetOldestAncesterTime();
-      if (oldest_ancester_time != 0) {
-        min_oldest_ancester_time =
-            std::min(min_oldest_ancester_time, oldest_ancester_time);
-      }
+  for (const auto& file : inputs_[0].files) {
+    uint64_t oldest_ancester_time = file->TryGetOldestAncesterTime();
+    if (oldest_ancester_time != 0) {
+      min_oldest_ancester_time =
+          std::min(min_oldest_ancester_time, oldest_ancester_time);
     }
   }
   return min_oldest_ancester_time;

db/compaction/compaction_iterator.cc

@@ -639,28 +639,48 @@ void CompactionIterator::NextFromInput() {
 }
 
 void CompactionIterator::PrepareOutput() {
-  // Zeroing out the sequence number leads to better compression.
-  // If this is the bottommost level (no files in lower levels)
-  // and the earliest snapshot is larger than this seqno
-  // and the userkey differs from the last userkey in compaction
-  // then we can squash the seqno to zero.
-  //
-  // This is safe for TransactionDB write-conflict checking since transactions
-  // only care about sequence number larger than any active snapshots.
-  //
-  // Can we do the same for levels above bottom level as long as
-  // KeyNotExistsBeyondOutputLevel() return true?
-  if ((compaction_ != nullptr && !compaction_->allow_ingest_behind()) &&
-      ikeyNotNeededForIncrementalSnapshot() && bottommost_level_ && valid_ &&
-      IN_EARLIEST_SNAPSHOT(ikey_.sequence) && ikey_.type != kTypeMerge) {
-    assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
-    if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion) {
-      ROCKS_LOG_FATAL(info_log_,
-                      "Unexpected key type %d for seq-zero optimization",
-                      ikey_.type);
+  if (valid_) {
+    if (compaction_filter_ && ikey_.type == kTypeBlobIndex) {
+      const auto blob_decision = compaction_filter_->PrepareBlobOutput(
+          user_key(), value_, &compaction_filter_value_);
+
+      if (blob_decision == CompactionFilter::BlobDecision::kCorruption) {
+        status_ = Status::Corruption(
+            "Corrupted blob reference encountered during GC");
+        valid_ = false;
+      } else if (blob_decision == CompactionFilter::BlobDecision::kIOError) {
+        status_ = Status::IOError("Could not relocate blob during GC");
+        valid_ = false;
+      } else if (blob_decision ==
+                 CompactionFilter::BlobDecision::kChangeValue) {
+        value_ = compaction_filter_value_;
+      }
+    }
+
+    // Zeroing out the sequence number leads to better compression.
+    // If this is the bottommost level (no files in lower levels)
+    // and the earliest snapshot is larger than this seqno
+    // and the userkey differs from the last userkey in compaction
+    // then we can squash the seqno to zero.
+    //
+    // This is safe for TransactionDB write-conflict checking since transactions
+    // only care about sequence number larger than any active snapshots.
+    //
+    // Can we do the same for levels above bottom level as long as
+    // KeyNotExistsBeyondOutputLevel() return true?
+    if (valid_ && compaction_ != nullptr &&
+        !compaction_->allow_ingest_behind() &&
+        ikeyNotNeededForIncrementalSnapshot() && bottommost_level_ &&
+        IN_EARLIEST_SNAPSHOT(ikey_.sequence) && ikey_.type != kTypeMerge) {
+      assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
+      if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion) {
+        ROCKS_LOG_FATAL(info_log_,
+                        "Unexpected key type %d for seq-zero optimization",
+                        ikey_.type);
+      }
+      ikey_.sequence = 0;
+      current_key_.UpdateInternalKey(0, ikey_.type);
     }
-    ikey_.sequence = 0;
-    current_key_.UpdateInternalKey(0, ikey_.type);
   }
 }
 

db/compaction/compaction_job.cc

@@ -301,7 +301,7 @@ void CompactionJob::AggregateStatistics() {
 
 CompactionJob::CompactionJob(
     int job_id, Compaction* compaction, const ImmutableDBOptions& db_options,
-    const EnvOptions env_options, VersionSet* versions,
+    const FileOptions& file_options, VersionSet* versions,
     const std::atomic<bool>* shutting_down,
     const SequenceNumber preserve_deletes_seqnum, LogBuffer* log_buffer,
     Directory* db_directory, Directory* output_directory, Statistics* stats,
@@ -318,10 +318,11 @@ CompactionJob::CompactionJob(
       compaction_stats_(compaction->compaction_reason(), 1),
       dbname_(dbname),
       db_options_(db_options),
-      env_options_(env_options),
+      file_options_(file_options),
       env_(db_options.env),
-      env_options_for_read_(
-          env_->OptimizeForCompactionTableRead(env_options, db_options_)),
+      fs_(db_options.fs.get()),
+      file_options_for_read_(
+          fs_->OptimizeForCompactionTableRead(file_options, db_options_)),
       versions_(versions),
       shutting_down_(shutting_down),
       manual_compaction_paused_(manual_compaction_paused),
@@ -647,7 +648,7 @@ Status CompactionJob::Run() {
         // we will regard this verification as user reads since the goal is
         // to cache it here for further user reads
         InternalIterator* iter = cfd->table_cache()->NewIterator(
-            ReadOptions(), env_options_, cfd->internal_comparator(),
+            ReadOptions(), file_options_, cfd->internal_comparator(),
             *files_meta[file_idx], /*range_del_agg=*/nullptr, prefix_extractor,
             /*table_reader_ptr=*/nullptr,
             cfd->internal_stats()->GetFileReadHist(
@@ -836,7 +837,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   // Although the v2 aggregator is what the level iterator(s) know about,
   // the AddTombstones calls will be propagated down to the v1 aggregator.
   std::unique_ptr<InternalIterator> input(versions_->MakeInputIterator(
-      sub_compact->compaction, &range_del_agg, env_options_for_read_));
+      sub_compact->compaction, &range_del_agg, file_options_for_read_));
 
   AutoThreadOperationStageUpdater stage_updater(
       ThreadStatus::STAGE_COMPACTION_PROCESS_KV);
@@ -1457,13 +1458,13 @@ Status CompactionJob::OpenCompactionOutputFile(
       TableFileCreationReason::kCompaction);
 #endif  // !ROCKSDB_LITE
   // Make the output file
-  std::unique_ptr<WritableFile> writable_file;
+  std::unique_ptr<FSWritableFile> writable_file;
 #ifndef NDEBUG
-  bool syncpoint_arg = env_options_.use_direct_writes;
+  bool syncpoint_arg = file_options_.use_direct_writes;
   TEST_SYNC_POINT_CALLBACK("CompactionJob::OpenCompactionOutputFile",
                            &syncpoint_arg);
 #endif
-  Status s = NewWritableFile(env_, fname, &writable_file, env_options_);
+  Status s = NewWritableFile(fs_, fname, &writable_file, file_options_);
   if (!s.ok()) {
     ROCKS_LOG_ERROR(
         db_options_.info_log,
@@ -1506,14 +1507,14 @@ Status CompactionJob::OpenCompactionOutputFile(
     sub_compact->outputs.push_back(out);
   }
 
-  writable_file->SetIOPriority(Env::IO_LOW);
+  writable_file->SetIOPriority(Env::IOPriority::IO_LOW);
   writable_file->SetWriteLifeTimeHint(write_hint_);
   writable_file->SetPreallocationBlockSize(static_cast<size_t>(
       sub_compact->compaction->OutputFilePreallocationSize()));
   const auto& listeners =
       sub_compact->compaction->immutable_cf_options()->listeners;
   sub_compact->outfile.reset(
-      new WritableFileWriter(std::move(writable_file), fname, env_options_,
+      new WritableFileWriter(std::move(writable_file), fname, file_options_,
                              env_, db_options_.statistics.get(), listeners));
 
   // If the Column family flag is to only optimize filters for hits,

db/compaction/compaction_job.h

@@ -64,7 +64,7 @@ class CompactionJob {
  public:
   CompactionJob(int job_id, Compaction* compaction,
                 const ImmutableDBOptions& db_options,
-                const EnvOptions env_options, VersionSet* versions,
+                const FileOptions& file_options, VersionSet* versions,
                 const std::atomic<bool>* shutting_down,
                 const SequenceNumber preserve_deletes_seqnum,
                 LogBuffer* log_buffer, Directory* db_directory,
@@ -150,11 +150,12 @@ class CompactionJob {
   // DBImpl state
   const std::string& dbname_;
   const ImmutableDBOptions& db_options_;
-  const EnvOptions env_options_;
+  const FileOptions file_options_;
 
   Env* env_;
+  FileSystem* fs_;
   // env_option optimized for compaction table reads
-  EnvOptions env_options_for_read_;
+  FileOptions file_options_for_read_;
   VersionSet* versions_;
   const std::atomic<bool>* shutting_down_;
   const std::atomic<bool>* manual_compaction_paused_;

db/compaction/compaction_job_test.cc

@@ -72,6 +72,7 @@ class CompactionJobTest : public testing::Test {
  public:
   CompactionJobTest()
       : env_(Env::Default()),
+        fs_(std::make_shared<LegacyFileSystemWrapper>(env_)),
         dbname_(test::PerThreadDBPath("compaction_job_test")),
         db_options_(),
         mutable_cf_options_(cf_options_),
@@ -86,6 +87,8 @@ class CompactionJobTest : public testing::Test {
         mock_table_factory_(new mock::MockTableFactory()),
         error_handler_(nullptr, db_options_, &mutex_) {
     EXPECT_OK(env_->CreateDirIfMissing(dbname_));
+    db_options_.env = env_;
+    db_options_.fs = fs_;
     db_options_.db_paths.emplace_back(dbname_,
                                       std::numeric_limits<uint64_t>::max());
   }
@@ -267,8 +270,8 @@ class CompactionJobTest : public testing::Test {
     Status s = env_->NewWritableFile(
         manifest, &file, env_->OptimizeForManifestWrite(env_options_));
     ASSERT_OK(s);
-    std::unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), manifest, env_options_));
+    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+        NewLegacyWritableFileWrapper(std::move(file)), manifest, env_options_));
     {
       log::Writer log(std::move(file_writer), 0, false);
       std::string record;
@@ -360,6 +363,7 @@ class CompactionJobTest : public testing::Test {
   }
 
   Env* env_;
+  std::shared_ptr<FileSystem> fs_;
   std::string dbname_;
   EnvOptions env_options_;
   ImmutableDBOptions db_options_;

db/compaction/compaction_picker_universal.cc

@@ -247,7 +247,8 @@ bool UniversalCompactionBuilder::IsInputFilesNonOverlapping(Compaction* c) {
 
     next.f = nullptr;
 
-    if (curr.level != 0 && curr.index < c->num_input_files(curr.level) - 1) {
+    if (c->level(curr.level) != 0 &&
+        curr.index < c->num_input_files(curr.level) - 1) {
       next.f = c->input(curr.level, curr.index + 1);
       next.level = curr.level;
       next.index = curr.index + 1;

db/convenience.cc

@@ -41,14 +41,16 @@ Status VerifySstFileChecksum(const Options& options,
                              const EnvOptions& env_options,
                              const ReadOptions& read_options,
                              const std::string& file_path) {
-  std::unique_ptr<RandomAccessFile> file;
+  std::unique_ptr<FSRandomAccessFile> file;
   uint64_t file_size;
   InternalKeyComparator internal_comparator(options.comparator);
   ImmutableCFOptions ioptions(options);
 
-  Status s = ioptions.env->NewRandomAccessFile(file_path, &file, env_options);
+  Status s = ioptions.fs->NewRandomAccessFile(file_path,
+                                              FileOptions(env_options),
+                                              &file, nullptr);
   if (s.ok()) {
-    s = ioptions.env->GetFileSize(file_path, &file_size);
+    s = ioptions.fs->GetFileSize(file_path, IOOptions(), &file_size, nullptr);
   } else {
     return s;
   }

db/corruption_test.cc

@@ -20,6 +20,7 @@
 #include "db/db_test_util.h"
 #include "db/log_format.h"
 #include "db/version_set.h"
+#include "env/composite_env_wrapper.h"
 #include "file/filename.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/convenience.h"
@@ -189,6 +190,7 @@ class CorruptionTest : public testing::Test {
     ASSERT_TRUE(s.ok()) << s.ToString();
     Options options;
     EnvOptions env_options;
+    options.file_system.reset(new LegacyFileSystemWrapper(options.env));
     ASSERT_NOK(VerifySstFileChecksum(options, env_options, fname));
   }
 
@@ -539,7 +541,8 @@ TEST_F(CorruptionTest, RangeDeletionCorrupted) {
   std::unique_ptr<RandomAccessFile> file;
   ASSERT_OK(options_.env->NewRandomAccessFile(filename, &file, EnvOptions()));
   std::unique_ptr<RandomAccessFileReader> file_reader(
-      new RandomAccessFileReader(std::move(file), filename));
+      new RandomAccessFileReader(NewLegacyRandomAccessFileWrapper(file),
+                                 filename));
 
   uint64_t file_size;
   ASSERT_OK(options_.env->GetFileSize(filename, &file_size));

db/db_basic_test.cc

@@ -6,7 +6,6 @@
 // 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 <iostream>
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
@@ -525,6 +524,7 @@ TEST_F(DBBasicTest, Snapshot) {
     ASSERT_EQ(1U, GetNumSnapshots());
     uint64_t time_snap1 = GetTimeOldestSnapshots();
     ASSERT_GT(time_snap1, 0U);
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     Put(0, "foo", "0v2");
     Put(1, "foo", "1v2");
 
@@ -533,6 +533,7 @@ TEST_F(DBBasicTest, Snapshot) {
     const Snapshot* s2 = db_->GetSnapshot();
     ASSERT_EQ(2U, GetNumSnapshots());
     ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     Put(0, "foo", "0v3");
     Put(1, "foo", "1v3");
 
@@ -540,6 +541,7 @@ TEST_F(DBBasicTest, Snapshot) {
       ManagedSnapshot s3(db_);
       ASSERT_EQ(3U, GetNumSnapshots());
       ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+      ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
 
       Put(0, "foo", "0v4");
       Put(1, "foo", "1v4");
@@ -555,6 +557,7 @@ TEST_F(DBBasicTest, Snapshot) {
 
     ASSERT_EQ(2U, GetNumSnapshots());
     ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     ASSERT_EQ("0v1", Get(0, "foo", s1));
     ASSERT_EQ("1v1", Get(1, "foo", s1));
     ASSERT_EQ("0v2", Get(0, "foo", s2));
@@ -569,9 +572,11 @@ TEST_F(DBBasicTest, Snapshot) {
     ASSERT_EQ("1v4", Get(1, "foo"));
     ASSERT_EQ(1U, GetNumSnapshots());
     ASSERT_LT(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s2->GetSequenceNumber());
 
     db_->ReleaseSnapshot(s2);
     ASSERT_EQ(0U, GetNumSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), 0);
     ASSERT_EQ("0v4", Get(0, "foo"));
     ASSERT_EQ("1v4", Get(1, "foo"));
   } while (ChangeOptions());
@@ -1706,6 +1711,14 @@ class DBBasicTestWithParallelIO
       assert(Put(Key(i), values_[i]) == Status::OK());
     }
     Flush();
+
+    for (int i = 0; i < 100; ++i) {
+      // block cannot gain space by compression
+      uncompressable_values_.emplace_back(RandomString(&rnd, 256) + '\0');
+      std::string tmp_key = "a" + Key(i);
+      assert(Put(tmp_key, uncompressable_values_[i]) == Status::OK());
+    }
+    Flush();
   }
 
   bool CheckValue(int i, const std::string& value) {
@@ -1715,6 +1728,13 @@ class DBBasicTestWithParallelIO
     return false;
   }
 
+  bool CheckUncompressableValue(int i, const std::string& value) {
+    if (uncompressable_values_[i].compare(value) == 0) {
+      return true;
+    }
+    return false;
+  }
+
   int num_lookups() { return uncompressed_cache_->num_lookups(); }
   int num_found() { return uncompressed_cache_->num_found(); }
   int num_inserts() { return uncompressed_cache_->num_inserts(); }
@@ -1864,6 +1884,7 @@ class DBBasicTestWithParallelIO
   std::shared_ptr<MyBlockCache> uncompressed_cache_;
   bool compression_enabled_;
   std::vector<std::string> values_;
+  std::vector<std::string> uncompressable_values_;
   bool fill_cache_;
 };
 
@@ -1928,8 +1949,66 @@ TEST_P(DBBasicTestWithParallelIO, MultiGet) {
     ASSERT_OK(statuses[i]);
     ASSERT_TRUE(CheckValue(key_ints[i], values[i].ToString()));
   }
-  expected_reads += (read_from_cache ? 2 : 4);
+  if (compression_enabled() && !has_compressed_cache()) {
+    expected_reads += (read_from_cache ? 2 : 3);
+  } else {
+    expected_reads += (read_from_cache ? 2 : 4);
+  }
   ASSERT_EQ(env_->random_read_counter_.Read(), expected_reads);
+
+  keys.resize(10);
+  statuses.resize(10);
+  std::vector<int> key_uncmp{1, 2, 15, 16, 55, 81, 82, 83, 84, 85};
+  for (size_t i = 0; i < key_uncmp.size(); ++i) {
+    key_data[i] = "a" + Key(key_uncmp[i]);
+    keys[i] = Slice(key_data[i]);
+    statuses[i] = Status::OK();
+    values[i].Reset();
+  }
+  dbfull()->MultiGet(ro, dbfull()->DefaultColumnFamily(), keys.size(),
+                     keys.data(), values.data(), statuses.data(), true);
+  for (size_t i = 0; i < key_uncmp.size(); ++i) {
+    ASSERT_OK(statuses[i]);
+    ASSERT_TRUE(CheckUncompressableValue(key_uncmp[i], values[i].ToString()));
+  }
+  if (compression_enabled() && !has_compressed_cache()) {
+    expected_reads += (read_from_cache ? 3 : 3);
+  } else {
+    expected_reads += (read_from_cache ? 4 : 4);
+  }
+  ASSERT_EQ(env_->random_read_counter_.Read(), expected_reads);
+
+  keys.resize(5);
+  statuses.resize(5);
+  std::vector<int> key_tr{1, 2, 15, 16, 55};
+  for (size_t i = 0; i < key_tr.size(); ++i) {
+    key_data[i] = "a" + Key(key_tr[i]);
+    keys[i] = Slice(key_data[i]);
+    statuses[i] = Status::OK();
+    values[i].Reset();
+  }
+  dbfull()->MultiGet(ro, dbfull()->DefaultColumnFamily(), keys.size(),
+                     keys.data(), values.data(), statuses.data(), true);
+  for (size_t i = 0; i < key_tr.size(); ++i) {
+    ASSERT_OK(statuses[i]);
+    ASSERT_TRUE(CheckUncompressableValue(key_tr[i], values[i].ToString()));
+  }
+  if (compression_enabled() && !has_compressed_cache()) {
+    expected_reads += (read_from_cache ? 0 : 2);
+    ASSERT_EQ(env_->random_read_counter_.Read(), expected_reads);
+  } else {
+    if (has_uncompressed_cache()) {
+      expected_reads += (read_from_cache ? 0 : 3);
+      ASSERT_EQ(env_->random_read_counter_.Read(), expected_reads);
+    } else {
+      // A rare case, even we enable the block compression but some of data
+      // blocks are not compressed due to content. If user only enable the
+      // compressed cache, the uncompressed blocks will not tbe cached, and
+      // block reads will be triggered. The number of reads is related to
+      // the compression algorithm.
+      ASSERT_TRUE(env_->random_read_counter_.Read() >= expected_reads);
+    }
+  }
 }
 
 INSTANTIATE_TEST_CASE_P(
@@ -2045,48 +2124,81 @@ TEST_P(DBBasicTestWithTimestampWithParam, PutAndGet) {
       10 /*bits_per_key*/, false /*use_block_based_builder*/));
   bbto.whole_key_filtering = true;
   options.table_factory.reset(NewBlockBasedTableFactory(bbto));
-  DestroyAndReopen(options);
-  CreateAndReopenWithCF({"pikachu"}, options);
-  size_t num_cfs = handles_.size();
-  ASSERT_EQ(2, num_cfs);
-  std::vector<std::string> write_ts_strs(kNumTimestamps);
-  std::vector<std::string> read_ts_strs(kNumTimestamps);
-  std::vector<Slice> write_ts_list;
-  std::vector<Slice> read_ts_list;
 
-  for (size_t i = 0; i != kNumTimestamps; ++i) {
-    write_ts_list.emplace_back(EncodeTimestamp(i * 2, 0, &write_ts_strs[i]));
-    read_ts_list.emplace_back(EncodeTimestamp(1 + i * 2, 0, &read_ts_strs[i]));
-    const Slice& write_ts = write_ts_list.back();
-    WriteOptions wopts;
-    wopts.timestamp = &write_ts;
-    for (int cf = 0; cf != static_cast<int>(num_cfs); ++cf) {
-      for (size_t j = 0; j != (kNumKeysPerFile - 1) / kNumTimestamps; ++j) {
-        ASSERT_OK(Put(cf, "key" + std::to_string(j),
-                      "value_" + std::to_string(j) + "_" + std::to_string(i),
-                      wopts));
-      }
-      if (!memtable_only) {
-        ASSERT_OK(Flush(cf));
-      }
-    }
+  std::vector<CompressionType> compression_types;
+  compression_types.push_back(kNoCompression);
+  if (Zlib_Supported()) {
+    compression_types.push_back(kZlibCompression);
   }
-  const auto& verify_db_func = [&]() {
-    for (size_t i = 0; i != kNumTimestamps; ++i) {
-      ReadOptions ropts;
-      ropts.timestamp = &read_ts_list[i];
-      for (int cf = 0; cf != static_cast<int>(num_cfs); ++cf) {
-        ColumnFamilyHandle* cfh = handles_[cf];
-        for (size_t j = 0; j != (kNumKeysPerFile - 1) / kNumTimestamps; ++j) {
-          std::string value;
-          ASSERT_OK(db_->Get(ropts, cfh, "key" + std::to_string(j), &value));
-          ASSERT_EQ("value_" + std::to_string(j) + "_" + std::to_string(i),
-                    value);
+#if LZ4_VERSION_NUMBER >= 10400  // r124+
+  compression_types.push_back(kLZ4Compression);
+  compression_types.push_back(kLZ4HCCompression);
+#endif  // LZ4_VERSION_NUMBER >= 10400
+  if (ZSTD_Supported()) {
+    compression_types.push_back(kZSTD);
+  }
+
+  // Switch compression dictionary on/off to check key extraction
+  // correctness in kBuffered state
+  std::vector<uint32_t> max_dict_bytes_list = {0, 1 << 14};  // 0 or 16KB
+
+  for (auto compression_type : compression_types) {
+    for (uint32_t max_dict_bytes : max_dict_bytes_list) {
+      options.compression = compression_type;
+      options.compression_opts.max_dict_bytes = max_dict_bytes;
+      if (compression_type == kZSTD) {
+        options.compression_opts.zstd_max_train_bytes = max_dict_bytes;
+      }
+      options.target_file_size_base = 1 << 26;  // 64MB
+
+      DestroyAndReopen(options);
+      CreateAndReopenWithCF({"pikachu"}, options);
+      size_t num_cfs = handles_.size();
+      ASSERT_EQ(2, num_cfs);
+      std::vector<std::string> write_ts_strs(kNumTimestamps);
+      std::vector<std::string> read_ts_strs(kNumTimestamps);
+      std::vector<Slice> write_ts_list;
+      std::vector<Slice> read_ts_list;
+
+      for (size_t i = 0; i != kNumTimestamps; ++i) {
+        write_ts_list.emplace_back(
+            EncodeTimestamp(i * 2, 0, &write_ts_strs[i]));
+        read_ts_list.emplace_back(
+            EncodeTimestamp(1 + i * 2, 0, &read_ts_strs[i]));
+        const Slice& write_ts = write_ts_list.back();
+        WriteOptions wopts;
+        wopts.timestamp = &write_ts;
+        for (int cf = 0; cf != static_cast<int>(num_cfs); ++cf) {
+          for (size_t j = 0; j != (kNumKeysPerFile - 1) / kNumTimestamps; ++j) {
+            ASSERT_OK(Put(
+                cf, "key" + std::to_string(j),
+                "value_" + std::to_string(j) + "_" + std::to_string(i), wopts));
+          }
+          if (!memtable_only) {
+            ASSERT_OK(Flush(cf));
+          }
         }
       }
+      const auto& verify_db_func = [&]() {
+        for (size_t i = 0; i != kNumTimestamps; ++i) {
+          ReadOptions ropts;
+          ropts.timestamp = &read_ts_list[i];
+          for (int cf = 0; cf != static_cast<int>(num_cfs); ++cf) {
+            ColumnFamilyHandle* cfh = handles_[cf];
+            for (size_t j = 0; j != (kNumKeysPerFile - 1) / kNumTimestamps;
+                 ++j) {
+              std::string value;
+              ASSERT_OK(
+                  db_->Get(ropts, cfh, "key" + std::to_string(j), &value));
+              ASSERT_EQ("value_" + std::to_string(j) + "_" + std::to_string(i),
+                        value);
+            }
+          }
+        }
+      };
+      verify_db_func();
     }
-  };
-  verify_db_func();
+  }
 }
 
 INSTANTIATE_TEST_CASE_P(Timestamp, DBBasicTestWithTimestampWithParam,

db/db_block_cache_test.cc

@@ -10,6 +10,7 @@
 #include "cache/lru_cache.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "util/compression.h"
 
 namespace rocksdb {
 
@@ -685,16 +686,18 @@ TEST_F(DBBlockCacheTest, CacheCompressionDict) {
 
   // Try all the available libraries that support dictionary compression
   std::vector<CompressionType> compression_types;
-#ifdef ZLIB
-  compression_types.push_back(kZlibCompression);
-#endif  // ZLIB
-#if LZ4_VERSION_NUMBER >= 10400
-  compression_types.push_back(kLZ4Compression);
-  compression_types.push_back(kLZ4HCCompression);
-#endif  // LZ4_VERSION_NUMBER >= 10400
-#if ZSTD_VERSION_NUMBER >= 500
-  compression_types.push_back(kZSTD);
-#endif  // ZSTD_VERSION_NUMBER >= 500
+  if (Zlib_Supported()) {
+    compression_types.push_back(kZlibCompression);
+  }
+  if (LZ4_Supported()) {
+    compression_types.push_back(kLZ4Compression);
+    compression_types.push_back(kLZ4HCCompression);
+  }
+  if (ZSTD_Supported()) {
+    compression_types.push_back(kZSTD);
+  } else if (ZSTDNotFinal_Supported()) {
+    compression_types.push_back(kZSTDNotFinalCompression);
+  }
   Random rnd(301);
   for (auto compression_type : compression_types) {
     Options options = CurrentOptions();

db/db_compaction_test.cc

@@ -1503,6 +1503,37 @@ TEST_F(DBCompactionTest, DISABLED_ManualPartialFill) {
   }
 }
 
+TEST_F(DBCompactionTest, ManualCompactionWithUnorderedWrite) {
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::WriteImpl:UnorderedWriteAfterWriteWAL",
+        "DBCompactionTest::ManualCompactionWithUnorderedWrite:WaitWriteWAL"},
+       {"DBImpl::WaitForPendingWrites:BeforeBlock",
+        "DBImpl::WriteImpl:BeforeUnorderedWriteMemtable"}});
+
+  Options options = CurrentOptions();
+  options.unordered_write = true;
+  DestroyAndReopen(options);
+  Put("foo", "v1");
+  ASSERT_OK(Flush());
+
+  Put("bar", "v1");
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  port::Thread writer([&]() { Put("foo", "v2"); });
+
+  TEST_SYNC_POINT(
+      "DBCompactionTest::ManualCompactionWithUnorderedWrite:WaitWriteWAL");
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  writer.join();
+  ASSERT_EQ(Get("foo"), "v2");
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  Reopen(options);
+  ASSERT_EQ(Get("foo"), "v2");
+}
+
 TEST_F(DBCompactionTest, DeleteFileRange) {
   Options options = CurrentOptions();
   options.write_buffer_size = 10 * 1024 * 1024;
@@ -3576,15 +3607,6 @@ TEST_F(DBCompactionTest, LevelTtlCascadingCompactions) {
         ASSERT_OK(Put(Key(i), RandomString(&rnd, kValueSize)));
       }
       Flush();
-      // Get the first file's creation time. This will be the oldest file in the
-      // DB. Compactions inolving this file's descendents should keep getting
-      // this time.
-      std::vector<std::vector<FileMetaData>> level_to_files;
-      dbfull()->TEST_GetFilesMetaData(dbfull()->DefaultColumnFamily(),
-                                      &level_to_files);
-      uint64_t oldest_time = level_to_files[0][0].oldest_ancester_time;
-      // Add 1 hour and do another flush.
-      env_->addon_time_.fetch_add(1 * 60 * 60);
       for (int i = 101; i <= 200; ++i) {
         ASSERT_OK(Put(Key(i), RandomString(&rnd, kValueSize)));
       }
@@ -3592,13 +3614,11 @@ TEST_F(DBCompactionTest, LevelTtlCascadingCompactions) {
       MoveFilesToLevel(6);
       ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
 
-      env_->addon_time_.fetch_add(1 * 60 * 60);
       // Add two L4 files with key ranges: [1 .. 50], [51 .. 150].
       for (int i = 1; i <= 50; ++i) {
         ASSERT_OK(Put(Key(i), RandomString(&rnd, kValueSize)));
       }
       Flush();
-      env_->addon_time_.fetch_add(1 * 60 * 60);
       for (int i = 51; i <= 150; ++i) {
         ASSERT_OK(Put(Key(i), RandomString(&rnd, kValueSize)));
       }
@@ -3606,7 +3626,6 @@ TEST_F(DBCompactionTest, LevelTtlCascadingCompactions) {
       MoveFilesToLevel(4);
       ASSERT_EQ("0,0,0,0,2,0,2", FilesPerLevel());
 
-      env_->addon_time_.fetch_add(1 * 60 * 60);
       // Add one L1 file with key range: [26, 75].
       for (int i = 26; i <= 75; ++i) {
         ASSERT_OK(Put(Key(i), RandomString(&rnd, kValueSize)));
@@ -3648,10 +3667,6 @@ TEST_F(DBCompactionTest, LevelTtlCascadingCompactions) {
       ASSERT_EQ("1,0,0,0,0,0,1", FilesPerLevel());
       ASSERT_EQ(5, ttl_compactions);
 
-      dbfull()->TEST_GetFilesMetaData(dbfull()->DefaultColumnFamily(),
-                                      &level_to_files);
-      ASSERT_EQ(oldest_time, level_to_files[6][0].oldest_ancester_time);
-
       env_->addon_time_.fetch_add(25 * 60 * 60);
       ASSERT_OK(Put(Key(2), "1"));
       if (if_restart) {

db/db_filesnapshot.cc

@@ -102,7 +102,7 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
         TEST_SYNC_POINT("DBImpl::GetLiveFiles:1");
         TEST_SYNC_POINT("DBImpl::GetLiveFiles:2");
         mutex_.Lock();
-        cfd->Unref();
+        cfd->UnrefAndTryDelete();
         if (!status.ok()) {
           break;
         }

db/db_flush_test.cc

@@ -212,6 +212,30 @@ TEST_F(DBFlushTest, ManualFlushWithMinWriteBufferNumberToMerge) {
   t.join();
 }
 
+TEST_F(DBFlushTest, ScheduleOnlyOneBgThread) {
+  Options options = CurrentOptions();
+  Reopen(options);
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  int called = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MaybeScheduleFlushOrCompaction:AfterSchedule:0", [&](void* arg) {
+        ASSERT_NE(nullptr, arg);
+        auto unscheduled_flushes = *reinterpret_cast<int*>(arg);
+        ASSERT_EQ(0, unscheduled_flushes);
+        ++called;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(Put("a", "foo"));
+  FlushOptions flush_opts;
+  ASSERT_OK(dbfull()->Flush(flush_opts));
+  ASSERT_EQ(1, called);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
 TEST_P(DBFlushDirectIOTest, DirectIO) {
   Options options;
   options.create_if_missing = true;

db/db_impl/db_impl.cc

@@ -52,6 +52,7 @@
 #include "db/version_set.h"
 #include "db/write_batch_internal.h"
 #include "db/write_callback.h"
+#include "env/composite_env_wrapper.h"
 #include "file/file_util.h"
 #include "file/filename.h"
 #include "file/random_access_file_reader.h"
@@ -102,6 +103,7 @@
 #include "util/string_util.h"
 
 namespace rocksdb {
+
 const std::string kDefaultColumnFamilyName("default");
 const std::string kPersistentStatsColumnFamilyName(
     "___rocksdb_stats_history___");
@@ -141,16 +143,15 @@ void DumpSupportInfo(Logger* logger) {
   ROCKS_LOG_HEADER(logger, "Fast CRC32 supported: %s",
                    crc32c::IsFastCrc32Supported().c_str());
 }
-
-int64_t kDefaultLowPriThrottledRate = 2 * 1024 * 1024;
 }  // namespace
 
 DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
                const bool seq_per_batch, const bool batch_per_txn)
-    : env_(options.env),
-      dbname_(dbname),
+    : dbname_(dbname),
       own_info_log_(options.info_log == nullptr),
       initial_db_options_(SanitizeOptions(dbname, options)),
+      env_(initial_db_options_.env),
+      fs_(initial_db_options_.file_system),
       immutable_db_options_(initial_db_options_),
       mutable_db_options_(initial_db_options_),
       stats_(immutable_db_options_.statistics.get()),
@@ -158,9 +159,9 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
              immutable_db_options_.use_adaptive_mutex),
       default_cf_handle_(nullptr),
       max_total_in_memory_state_(0),
-      env_options_(BuildDBOptions(immutable_db_options_, mutable_db_options_)),
-      env_options_for_compaction_(env_->OptimizeForCompactionTableWrite(
-          env_options_, immutable_db_options_)),
+      file_options_(BuildDBOptions(immutable_db_options_, mutable_db_options_)),
+      file_options_for_compaction_(fs_->OptimizeForCompactionTableWrite(
+          file_options_, immutable_db_options_)),
       seq_per_batch_(seq_per_batch),
       batch_per_txn_(batch_per_txn),
       db_lock_(nullptr),
@@ -178,11 +179,6 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       write_thread_(immutable_db_options_),
       nonmem_write_thread_(immutable_db_options_),
       write_controller_(mutable_db_options_.delayed_write_rate),
-      // Use delayed_write_rate as a base line to determine the initial
-      // low pri write rate limit. It may be adjusted later.
-      low_pri_write_rate_limiter_(NewGenericRateLimiter(std::min(
-          static_cast<int64_t>(mutable_db_options_.delayed_write_rate / 8),
-          kDefaultLowPriThrottledRate))),
       last_batch_group_size_(0),
       unscheduled_flushes_(0),
       unscheduled_compactions_(0),
@@ -201,7 +197,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       unable_to_release_oldest_log_(false),
       num_running_ingest_file_(0),
 #ifndef ROCKSDB_LITE
-      wal_manager_(immutable_db_options_, env_options_, seq_per_batch),
+      wal_manager_(immutable_db_options_, file_options_, seq_per_batch),
 #endif  // ROCKSDB_LITE
       event_logger_(immutable_db_options_.info_log.get()),
       bg_work_paused_(0),
@@ -247,7 +243,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
   co.metadata_charge_policy = kDontChargeCacheMetadata;
   table_cache_ = NewLRUCache(co);
 
-  versions_.reset(new VersionSet(dbname_, &immutable_db_options_, env_options_,
+  versions_.reset(new VersionSet(dbname_, &immutable_db_options_, file_options_,
                                  table_cache_.get(), write_buffer_manager_,
                                  &write_controller_, &block_cache_tracer_));
   column_family_memtables_.reset(
@@ -337,7 +333,7 @@ Status DBImpl::ResumeImpl() {
         mutex_.Unlock();
         s = FlushMemTable(cfd, flush_opts, FlushReason::kErrorRecovery);
         mutex_.Lock();
-        cfd->Unref();
+        cfd->UnrefAndTryDelete();
         if (!s.ok()) {
           break;
         }
@@ -425,7 +421,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
           mutex_.Unlock();
           FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
           mutex_.Lock();
-          cfd->Unref();
+          cfd->UnrefAndTryDelete();
         }
       }
     }
@@ -482,17 +478,12 @@ Status DBImpl::CloseHelper() {
   while (!flush_queue_.empty()) {
     const FlushRequest& flush_req = PopFirstFromFlushQueue();
     for (const auto& iter : flush_req) {
-      ColumnFamilyData* cfd = iter.first;
-      if (cfd->Unref()) {
-        delete cfd;
-      }
+      iter.first->UnrefAndTryDelete();
     }
   }
   while (!compaction_queue_.empty()) {
     auto cfd = PopFirstFromCompactionQueue();
-    if (cfd->Unref()) {
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
 
   if (default_cf_handle_ != nullptr || persist_stats_cf_handle_ != nullptr) {
@@ -599,6 +590,7 @@ Status DBImpl::CloseHelper() {
       ret = s;
     }
   }
+
   if (ret.IsAborted()) {
     // Reserve IsAborted() error for those where users didn't release
     // certain resource and they can release them and come back and
@@ -1059,14 +1051,14 @@ Status DBImpl::SetDBOptions(
       wal_changed = mutable_db_options_.wal_bytes_per_sync !=
                     new_options.wal_bytes_per_sync;
       mutable_db_options_ = new_options;
-      env_options_for_compaction_ = EnvOptions(new_db_options);
-      env_options_for_compaction_ = env_->OptimizeForCompactionTableWrite(
-          env_options_for_compaction_, immutable_db_options_);
-      versions_->ChangeEnvOptions(mutable_db_options_);
+      file_options_for_compaction_ = FileOptions(new_db_options);
+      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
-      env_options_for_compaction_ = env_->OptimizeForCompactionTableRead(
-          env_options_for_compaction_, immutable_db_options_);
-      env_options_for_compaction_.compaction_readahead_size =
+      file_options_for_compaction_ = fs_->OptimizeForCompactionTableRead(
+          file_options_for_compaction_, immutable_db_options_);
+      file_options_for_compaction_.compaction_readahead_size =
           mutable_db_options_.compaction_readahead_size;
       WriteThread::Writer w;
       write_thread_.EnterUnbatched(&w, &mutex_);
@@ -1452,7 +1444,7 @@ InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
   if (s.ok()) {
     // Collect iterators for files in L0 - Ln
     if (read_options.read_tier != kMemtableTier) {
-      super_version->current->AddIterators(read_options, env_options_,
+      super_version->current->AddIterators(read_options, file_options_,
                                            &merge_iter_builder, range_del_agg);
     }
     internal_iter = merge_iter_builder.Finish();
@@ -2747,6 +2739,15 @@ const std::string& DBImpl::GetName() const { return dbname_; }
 
 Env* DBImpl::GetEnv() const { return env_; }
 
+FileSystem* DB::GetFileSystem() const {
+  static LegacyFileSystemWrapper fs_wrap(GetEnv());
+  return &fs_wrap;
+}
+
+FileSystem* DBImpl::GetFileSystem() const {
+  return immutable_db_options_.fs.get();
+}
+
 Options DBImpl::GetOptions(ColumnFamilyHandle* column_family) const {
   InstrumentedMutexLock l(&mutex_);
   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
@@ -3327,12 +3328,12 @@ Status DBImpl::GetDbIdentity(std::string& identity) const {
 
 Status DBImpl::GetDbIdentityFromIdentityFile(std::string* identity) const {
   std::string idfilename = IdentityFileName(dbname_);
-  const EnvOptions soptions;
+  const FileOptions soptions;
   std::unique_ptr<SequentialFileReader> id_file_reader;
   Status s;
   {
-    std::unique_ptr<SequentialFile> idfile;
-    s = env_->NewSequentialFile(idfilename, &idfile, soptions);
+    std::unique_ptr<FSSequentialFile> idfile;
+    s = fs_->NewSequentialFile(idfilename, soptions, &idfile, nullptr);
     if (!s.ok()) {
       return s;
     }
@@ -3341,7 +3342,7 @@ Status DBImpl::GetDbIdentityFromIdentityFile(std::string* identity) const {
   }
 
   uint64_t file_size;
-  s = env_->GetFileSize(idfilename, &file_size);
+  s = fs_->GetFileSize(idfilename, IOOptions(), &file_size, nullptr);
   if (!s.ok()) {
     return s;
   }
@@ -3415,7 +3416,12 @@ Status DBImpl::Close() {
 Status DB::ListColumnFamilies(const DBOptions& db_options,
                               const std::string& name,
                               std::vector<std::string>* column_families) {
-  return VersionSet::ListColumnFamilies(column_families, name, db_options.env);
+  FileSystem* fs = db_options.file_system.get();
+  LegacyFileSystemWrapper legacy_fs(db_options.env);
+  if (!fs) {
+    fs = &legacy_fs;
+  }
+  return VersionSet::ListColumnFamilies(column_families, name, fs);
 }
 
 Snapshot::~Snapshot() {}
@@ -3580,8 +3586,8 @@ Status DBImpl::WriteOptionsFile(bool need_mutex_lock,
 
   std::string file_name =
       TempOptionsFileName(GetName(), versions_->NewFileNumber());
-  Status s =
-      PersistRocksDBOptions(db_options, cf_names, cf_opts, file_name, GetEnv());
+  Status s = PersistRocksDBOptions(db_options, cf_names, cf_opts, file_name,
+                                   GetFileSystem());
 
   if (s.ok()) {
     s = RenameTempFileToOptionsFile(file_name);
@@ -3925,7 +3931,7 @@ Status DBImpl::IngestExternalFiles(
   for (const auto& arg : args) {
     auto* cfd = static_cast<ColumnFamilyHandleImpl*>(arg.column_family)->cfd();
     ingestion_jobs.emplace_back(
-        env_, versions_.get(), cfd, immutable_db_options_, env_options_,
+        env_, versions_.get(), cfd, immutable_db_options_, file_options_,
         &snapshots_, arg.options, &directories_, &event_logger_);
   }
   std::vector<std::pair<bool, Status>> exec_results;
@@ -3991,6 +3997,13 @@ Status DBImpl::IngestExternalFiles(
       nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
     }
 
+    // When unordered_write is enabled, the keys are writing to memtable in an
+    // unordered way. If the ingestion job checks memtable key range before the
+    // key landing in memtable, the ingestion job may skip the necessary
+    // memtable flush.
+    // So wait here to ensure there is no pending write to memtable.
+    WaitForPendingWrites();
+
     num_running_ingest_file_ += static_cast<int>(num_cfs);
     TEST_SYNC_POINT("DBImpl::IngestExternalFile:AfterIncIngestFileCounter");
 
@@ -4182,7 +4195,7 @@ Status DBImpl::CreateColumnFamilyWithImport(
   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(*handle);
   auto cfd = cfh->cfd();
   ImportColumnFamilyJob import_job(env_, versions_.get(), cfd,
-                                   immutable_db_options_, env_options_,
+                                   immutable_db_options_, file_options_,
                                    import_options, metadata.files);
 
   SuperVersionContext dummy_sv_ctx(/* create_superversion */ true);
@@ -4312,7 +4325,7 @@ Status DBImpl::VerifyChecksum(const ReadOptions& read_options) {
         const auto& fd = vstorage->LevelFilesBrief(i).files[j].fd;
         std::string fname = TableFileName(cfd->ioptions()->cf_paths,
                                           fd.GetNumber(), fd.GetPathId());
-        s = rocksdb::VerifySstFileChecksum(opts, env_options_, read_options,
+        s = rocksdb::VerifySstFileChecksum(opts, file_options_, read_options,
                                            fname);
       }
     }
@@ -4338,7 +4351,7 @@ Status DBImpl::VerifyChecksum(const ReadOptions& read_options) {
       SchedulePurge();
     }
     for (auto cfd : cfd_list) {
-      cfd->Unref();
+      cfd->UnrefAndTryDelete();
     }
   }
   return s;

db/db_impl/db_impl.h

@@ -306,6 +306,7 @@ class DBImpl : public DB {
       ColumnFamilyHandle* column_family) override;
   virtual const std::string& GetName() const override;
   virtual Env* GetEnv() const override;
+  virtual FileSystem* GetFileSystem() const override;
   using DB::GetOptions;
   virtual Options GetOptions(ColumnFamilyHandle* column_family) const override;
   using DB::GetDBOptions;
@@ -945,13 +946,14 @@ class DBImpl : public DB {
 #endif  // NDEBUG
 
  protected:
-  Env* const env_;
   const std::string dbname_;
   std::string db_id_;
   std::unique_ptr<VersionSet> versions_;
   // Flag to check whether we allocated and own the info log file
   bool own_info_log_;
   const DBOptions initial_db_options_;
+  Env* const env_;
+  std::shared_ptr<FileSystem> fs_;
   const ImmutableDBOptions immutable_db_options_;
   MutableDBOptions mutable_db_options_;
   Statistics* stats_;
@@ -979,10 +981,10 @@ class DBImpl : public DB {
   bool single_column_family_mode_;
 
   // The options to access storage files
-  const EnvOptions env_options_;
+  const FileOptions file_options_;
 
   // Additonal options for compaction and flush
-  EnvOptions env_options_for_compaction_;
+  FileOptions file_options_for_compaction_;
 
   std::unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
 
@@ -1420,6 +1422,7 @@ class DBImpl : public DB {
 
   inline void WaitForPendingWrites() {
     mutex_.AssertHeld();
+    TEST_SYNC_POINT("DBImpl::WaitForPendingWrites:BeforeBlock");
     // In case of pipelined write is enabled, wait for all pending memtable
     // writers.
     if (immutable_db_options_.enable_pipelined_write) {
@@ -1835,8 +1838,6 @@ class DBImpl : public DB {
 
   WriteController write_controller_;
 
-  std::unique_ptr<RateLimiter> low_pri_write_rate_limiter_;
-
   // Size of the last batch group. In slowdown mode, next write needs to
   // sleep if it uses up the quota.
   // Note: This is to protect memtable and compaction. If the batch only writes

db/db_impl/db_impl_compaction_flush.cc

@@ -148,7 +148,7 @@ Status DBImpl::FlushMemTableToOutputFile(
 
   FlushJob flush_job(
       dbname_, cfd, immutable_db_options_, mutable_cf_options,
-      nullptr /* memtable_id */, env_options_for_compaction_, versions_.get(),
+      nullptr /* 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),
@@ -332,7 +332,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     const uint64_t* max_memtable_id = &(bg_flush_args[i].max_memtable_id_);
     jobs.emplace_back(new FlushJob(
         dbname_, cfd, immutable_db_options_, mutable_cf_options,
-        max_memtable_id, env_options_for_compaction_, versions_.get(), &mutex_,
+        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),
@@ -968,7 +968,7 @@ Status DBImpl::CompactFilesImpl(
   CompactionJobStats compaction_job_stats;
   CompactionJob compaction_job(
       job_context->job_id, c.get(), immutable_db_options_,
-      env_options_for_compaction_, versions_.get(), &shutting_down_,
+      file_options_for_compaction_, versions_.get(), &shutting_down_,
       preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
       GetDataDir(c->column_family_data(), c->output_path_id()), stats_, &mutex_,
       &error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
@@ -1540,6 +1540,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
         nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
       }
     }
+    WaitForPendingWrites();
 
     if (!cfd->mem()->IsEmpty() || !cached_recoverable_state_empty_.load()) {
       s = SwitchMemtable(cfd, &context);
@@ -1617,19 +1618,16 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
     }
     s = WaitForFlushMemTables(cfds, flush_memtable_ids,
                               (flush_reason == FlushReason::kErrorRecovery));
+    InstrumentedMutexLock lock_guard(&mutex_);
     for (auto* tmp_cfd : cfds) {
-      if (tmp_cfd->Unref()) {
-        // Only one thread can reach here.
-        InstrumentedMutexLock lock_guard(&mutex_);
-        delete tmp_cfd;
-      }
+      tmp_cfd->UnrefAndTryDelete();
     }
   }
-  TEST_SYNC_POINT("FlushMemTableFinished");
+  TEST_SYNC_POINT("DBImpl::FlushMemTable:FlushMemTableFinished");
   return s;
 }
 
-// Flush all elments in 'column_family_datas'
+// Flush all elements in 'column_family_datas'
 // and atomically record the result to the MANIFEST.
 Status DBImpl::AtomicFlushMemTables(
     const autovector<ColumnFamilyData*>& column_family_datas,
@@ -1665,6 +1663,7 @@ Status DBImpl::AtomicFlushMemTables(
         nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
       }
     }
+    WaitForPendingWrites();
 
     for (auto cfd : column_family_datas) {
       if (cfd->IsDropped()) {
@@ -1681,7 +1680,7 @@ Status DBImpl::AtomicFlushMemTables(
       }
       cfd->Ref();
       s = SwitchMemtable(cfd, &context);
-      cfd->Unref();
+      cfd->UnrefAndTryDelete();
       if (!s.ok()) {
         break;
       }
@@ -1721,12 +1720,9 @@ Status DBImpl::AtomicFlushMemTables(
     }
     s = WaitForFlushMemTables(cfds, flush_memtable_ids,
                               (flush_reason == FlushReason::kErrorRecovery));
+    InstrumentedMutexLock lock_guard(&mutex_);
     for (auto* cfd : cfds) {
-      if (cfd->Unref()) {
-        // Only one thread can reach here.
-        InstrumentedMutexLock lock_guard(&mutex_);
-        delete cfd;
-      }
+      cfd->UnrefAndTryDelete();
     }
   }
   return s;
@@ -1918,6 +1914,10 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
     fta->thread_pri_ = Env::Priority::HIGH;
     env_->Schedule(&DBImpl::BGWorkFlush, fta, Env::Priority::HIGH, this,
                    &DBImpl::UnscheduleFlushCallback);
+    --unscheduled_flushes_;
+    TEST_SYNC_POINT_CALLBACK(
+        "DBImpl::MaybeScheduleFlushOrCompaction:AfterSchedule:0",
+        &unscheduled_flushes_);
   }
 
   // special case -- if high-pri (flush) thread pool is empty, then schedule
@@ -1932,6 +1932,7 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
       fta->thread_pri_ = Env::Priority::LOW;
       env_->Schedule(&DBImpl::BGWorkFlush, fta, Env::Priority::LOW, this,
                      &DBImpl::UnscheduleFlushCallback);
+      --unscheduled_flushes_;
     }
   }
 
@@ -2015,8 +2016,6 @@ ColumnFamilyData* DBImpl::PopFirstFromCompactionQueue() {
 DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
   assert(!flush_queue_.empty());
   FlushRequest flush_req = flush_queue_.front();
-  assert(unscheduled_flushes_ >= static_cast<int>(flush_req.size()));
-  unscheduled_flushes_ -= static_cast<int>(flush_req.size());
   flush_queue_.pop_front();
   // TODO: need to unset flush reason?
   return flush_req;
@@ -2058,7 +2057,7 @@ void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
     cfd->Ref();
     cfd->SetFlushReason(flush_reason);
   }
-  unscheduled_flushes_ += static_cast<int>(flush_req.size());
+  ++unscheduled_flushes_;
   flush_queue_.push_back(flush_req);
 }
 
@@ -2204,16 +2203,13 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
     *reason = bg_flush_args[0].cfd_->GetFlushReason();
     for (auto& arg : bg_flush_args) {
       ColumnFamilyData* cfd = arg.cfd_;
-      if (cfd->Unref()) {
-        delete cfd;
+      if (cfd->UnrefAndTryDelete()) {
         arg.cfd_ = nullptr;
       }
     }
   }
   for (auto cfd : column_families_not_to_flush) {
-    if (cfd->Unref()) {
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
   return status;
 }
@@ -2542,10 +2538,9 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     // reference).
     // This will all happen under a mutex so we don't have to be afraid of
     // somebody else deleting it.
-    if (cfd->Unref()) {
+    if (cfd->UnrefAndTryDelete()) {
       // This was the last reference of the column family, so no need to
       // compact.
-      delete cfd;
       return Status::OK();
     }
 
@@ -2752,7 +2747,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     assert(is_snapshot_supported_ || snapshots_.empty());
     CompactionJob compaction_job(
         job_context->job_id, c.get(), immutable_db_options_,
-        env_options_for_compaction_, versions_.get(), &shutting_down_,
+        file_options_for_compaction_, versions_.get(), &shutting_down_,
         preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
         GetDataDir(c->column_family_data(), c->output_path_id()), stats_,
         &mutex_, &error_handler_, snapshot_seqs,

db/db_impl/db_impl_open.cc

@@ -12,6 +12,7 @@
 
 #include "db/builder.h"
 #include "db/error_handler.h"
+#include "env/composite_env_wrapper.h"
 #include "file/read_write_util.h"
 #include "file/sst_file_manager_impl.h"
 #include "file/writable_file_writer.h"
@@ -34,6 +35,18 @@ Options SanitizeOptions(const std::string& dbname, const Options& src) {
 DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
   DBOptions result(src);
 
+  if (result.file_system == nullptr) {
+    if (result.env == Env::Default()) {
+      result.file_system = FileSystem::Default();
+    } else {
+      result.file_system.reset(new LegacyFileSystemWrapper(result.env));
+    }
+  } else {
+    if (result.env == nullptr) {
+      result.env = Env::Default();
+    }
+  }
+
   // result.max_open_files means an "infinite" open files.
   if (result.max_open_files != -1) {
     int max_max_open_files = port::GetMaxOpenFiles();
@@ -254,16 +267,16 @@ Status DBImpl::NewDB() {
   ROCKS_LOG_INFO(immutable_db_options_.info_log, "Creating manifest 1 \n");
   const std::string manifest = DescriptorFileName(dbname_, 1);
   {
-    std::unique_ptr<WritableFile> file;
-    EnvOptions env_options = env_->OptimizeForManifestWrite(env_options_);
-    s = NewWritableFile(env_, manifest, &file, env_options);
+    std::unique_ptr<FSWritableFile> file;
+    FileOptions file_options = fs_->OptimizeForManifestWrite(file_options_);
+    s = NewWritableFile(fs_.get(), manifest, &file, file_options);
     if (!s.ok()) {
       return s;
     }
     file->SetPreallocationBlockSize(
         immutable_db_options_.manifest_preallocation_size);
     std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
-        std::move(file), manifest, env_options, env_, nullptr /* stats */,
+        std::move(file), manifest, file_options, env_, nullptr /* stats */,
         immutable_db_options_.listeners));
     log::Writer log(std::move(file_writer), 0, false);
     std::string record;
@@ -277,7 +290,7 @@ Status DBImpl::NewDB() {
     // Make "CURRENT" file that points to the new manifest file.
     s = SetCurrentFile(env_, dbname_, 1, directories_.GetDbDir());
   } else {
-    env_->DeleteFile(manifest);
+    fs_->DeleteFile(manifest, IOOptions(), nullptr);
   }
   return s;
 }
@@ -350,7 +363,8 @@ Status DBImpl::Recover(
       return s;
     }
 
-    s = env_->FileExists(CurrentFileName(dbname_));
+    std::string current_fname = CurrentFileName(dbname_);
+    s = env_->FileExists(current_fname);
     if (s.IsNotFound()) {
       if (immutable_db_options_.create_if_missing) {
         // Has to be called only after Identity File creation is successful
@@ -363,7 +377,7 @@ Status DBImpl::Recover(
         }
       } else {
         return Status::InvalidArgument(
-            dbname_, "does not exist (create_if_missing is false)");
+            current_fname, "does not exist (create_if_missing is false)");
       }
     } else if (s.ok()) {
       if (immutable_db_options_.error_if_exists) {
@@ -375,20 +389,20 @@ Status DBImpl::Recover(
       assert(s.IsIOError());
       return s;
     }
-    // Verify compatibility of env_options_ and filesystem
+    // Verify compatibility of file_options_ and filesystem
     {
-      std::unique_ptr<RandomAccessFile> idfile;
-      EnvOptions customized_env(env_options_);
-      customized_env.use_direct_reads |=
+      std::unique_ptr<FSRandomAccessFile> idfile;
+      FileOptions customized_fs(file_options_);
+      customized_fs.use_direct_reads |=
           immutable_db_options_.use_direct_io_for_flush_and_compaction;
-      s = env_->NewRandomAccessFile(CurrentFileName(dbname_), &idfile,
-                                    customized_env);
+      s = fs_->NewRandomAccessFile(current_fname, customized_fs, &idfile,
+                                   nullptr);
       if (!s.ok()) {
         std::string error_str = s.ToString();
         // Check if unsupported Direct I/O is the root cause
-        customized_env.use_direct_reads = false;
-        s = env_->NewRandomAccessFile(CurrentFileName(dbname_), &idfile,
-                                      customized_env);
+        customized_fs.use_direct_reads = false;
+        s = fs_->NewRandomAccessFile(current_fname, customized_fs, &idfile,
+                                     nullptr);
         if (s.ok()) {
           return Status::InvalidArgument(
               "Direct I/O is not supported by the specified DB.");
@@ -408,7 +422,7 @@ Status DBImpl::Recover(
   // the very first time.
   if (db_id_.empty()) {
     // Check for the IDENTITY file and create it if not there.
-    s = env_->FileExists(IdentityFileName(dbname_));
+    s = fs_->FileExists(IdentityFileName(dbname_), IOOptions(), nullptr);
     // Typically Identity file is created in NewDB() and for some reason if
     // it is no longer available then at this point DB ID is not in Identity
     // file or Manifest.
@@ -748,9 +762,10 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
 
     std::unique_ptr<SequentialFileReader> file_reader;
     {
-      std::unique_ptr<SequentialFile> file;
-      status = env_->NewSequentialFile(fname, &file,
-                                       env_->OptimizeForLogRead(env_options_));
+      std::unique_ptr<FSSequentialFile> file;
+      status = fs_->NewSequentialFile(fname,
+                                      fs_->OptimizeForLogRead(file_options_),
+                                      &file, nullptr);
       if (!status.ok()) {
         MaybeIgnoreError(&status);
         if (!status.ok()) {
@@ -920,7 +935,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
         ColumnFamilyData* cfd;
 
         while ((cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
-          cfd->Unref();
+          cfd->UnrefAndTryDelete();
           // If this asserts, it means that InsertInto failed in
           // filtering updates to already-flushed column families
           assert(cfd->GetLogNumber() <= log_number);
@@ -1119,17 +1134,18 @@ Status DBImpl::RestoreAliveLogFiles(const std::vector<uint64_t>& log_numbers) {
     // preallocated space are not needed anymore. It is likely only the last
     // log has such preallocated space, so we only truncate for the last log.
     if (log_number == log_numbers.back()) {
-      std::unique_ptr<WritableFile> last_log;
-      Status truncate_status = env_->ReopenWritableFile(
-          fname, &last_log,
-          env_->OptimizeForLogWrite(
-              env_options_,
-              BuildDBOptions(immutable_db_options_, mutable_db_options_)));
+      std::unique_ptr<FSWritableFile> last_log;
+      Status truncate_status = fs_->ReopenWritableFile(
+          fname,
+          fs_->OptimizeForLogWrite(
+              file_options_,
+              BuildDBOptions(immutable_db_options_, mutable_db_options_)),
+          &last_log, nullptr);
       if (truncate_status.ok()) {
-        truncate_status = last_log->Truncate(log.size);
+        truncate_status = last_log->Truncate(log.size, IOOptions(), nullptr);
       }
       if (truncate_status.ok()) {
-        truncate_status = last_log->Close();
+        truncate_status = last_log->Close(IOOptions(), nullptr);
       }
       // Not a critical error if fail to truncate.
       if (!truncate_status.ok()) {
@@ -1196,8 +1212,8 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
         range_del_iters.emplace_back(range_del_iter);
       }
       s = BuildTable(
-          dbname_, env_, *cfd->ioptions(), mutable_cf_options,
-          env_options_for_compaction_, cfd->table_cache(), iter.get(),
+          dbname_, env_, fs_.get(), *cfd->ioptions(), mutable_cf_options,
+          file_options_for_compaction_, cfd->table_cache(), iter.get(),
           std::move(range_del_iters), &meta, cfd->internal_comparator(),
           cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
           snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
@@ -1280,12 +1296,12 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
 Status DBImpl::CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
                          size_t preallocate_block_size, log::Writer** new_log) {
   Status s;
-  std::unique_ptr<WritableFile> lfile;
+  std::unique_ptr<FSWritableFile> lfile;
 
   DBOptions db_options =
       BuildDBOptions(immutable_db_options_, mutable_db_options_);
-  EnvOptions opt_env_options =
-      env_->OptimizeForLogWrite(env_options_, db_options);
+  FileOptions opt_file_options =
+      fs_->OptimizeForLogWrite(file_options_, db_options);
   std::string log_fname =
       LogFileName(immutable_db_options_.wal_dir, log_file_num);
 
@@ -1295,10 +1311,10 @@ Status DBImpl::CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
                    recycle_log_number);
     std::string old_log_fname =
         LogFileName(immutable_db_options_.wal_dir, recycle_log_number);
-    s = env_->ReuseWritableFile(log_fname, old_log_fname, &lfile,
-                                opt_env_options);
+    s = fs_->ReuseWritableFile(log_fname, old_log_fname, opt_file_options,
+                               &lfile, /*dbg=*/nullptr);
   } else {
-    s = NewWritableFile(env_, log_fname, &lfile, opt_env_options);
+    s = NewWritableFile(fs_.get(), log_fname, &lfile, opt_file_options);
   }
 
   if (s.ok()) {
@@ -1307,7 +1323,7 @@ Status DBImpl::CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
 
     const auto& listeners = immutable_db_options_.listeners;
     std::unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(lfile), log_fname, opt_env_options,
+        new WritableFileWriter(std::move(lfile), log_fname, opt_file_options,
                                env_, nullptr /* stats */, listeners));
     *new_log = new log::Writer(std::move(file_writer), log_file_num,
                                immutable_db_options_.recycle_log_file_num > 0,

db/db_impl/db_impl_secondary.cc

@@ -144,9 +144,10 @@ Status DBImplSecondary::MaybeInitLogReader(
 
     std::unique_ptr<SequentialFileReader> file_reader;
     {
-      std::unique_ptr<SequentialFile> file;
-      Status status = env_->NewSequentialFile(
-          fname, &file, env_->OptimizeForLogRead(env_options_));
+      std::unique_ptr<FSSequentialFile> file;
+      Status status = fs_->NewSequentialFile(
+          fname, fs_->OptimizeForLogRead(file_options_), &file,
+          nullptr);
       if (!status.ok()) {
         *log_reader = nullptr;
         return status;
@@ -600,7 +601,7 @@ Status DB::OpenAsSecondary(
   handles->clear();
   DBImplSecondary* impl = new DBImplSecondary(tmp_opts, dbname);
   impl->versions_.reset(new ReactiveVersionSet(
-      dbname, &impl->immutable_db_options_, impl->env_options_,
+      dbname, &impl->immutable_db_options_, impl->file_options_,
       impl->table_cache_.get(), impl->write_buffer_manager_,
       &impl->write_controller_));
   impl->column_family_memtables_.reset(

db/db_impl/db_impl_write.cc

@@ -131,6 +131,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
                               log_used, log_ref, &seq, sub_batch_cnt,
                               pre_release_callback, kDoAssignOrder,
                               kDoPublishLastSeq, disable_memtable);
+    TEST_SYNC_POINT("DBImpl::WriteImpl:UnorderedWriteAfterWriteWAL");
     if (!status.ok()) {
       return status;
     }
@@ -138,6 +139,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
       *seq_used = seq;
     }
     if (!disable_memtable) {
+      TEST_SYNC_POINT("DBImpl::WriteImpl:BeforeUnorderedWriteMemtable");
       status = UnorderedWriteMemtable(write_options, my_batch, callback,
                                       log_ref, seq, sub_batch_cnt);
     }
@@ -1254,7 +1256,7 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
   for (const auto cfd : cfds) {
     cfd->Ref();
     status = SwitchMemtable(cfd, write_context);
-    cfd->Unref();
+    cfd->UnrefAndTryDelete();
     if (!status.ok()) {
       break;
     }
@@ -1333,7 +1335,7 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
     }
     cfd->Ref();
     status = SwitchMemtable(cfd, write_context);
-    cfd->Unref();
+    cfd->UnrefAndTryDelete();
     if (!status.ok()) {
       break;
     }
@@ -1525,8 +1527,7 @@ Status DBImpl::TrimMemtableHistory(WriteContext* context) {
       cfd->InstallSuperVersion(&context->superversion_context, &mutex_);
     }
 
-    if (cfd->Unref()) {
-      delete cfd;
+    if (cfd->UnrefAndTryDelete()) {
       cfd = nullptr;
     }
   }
@@ -1558,8 +1559,7 @@ Status DBImpl::ScheduleFlushes(WriteContext* context) {
     if (!cfd->mem()->IsEmpty()) {
       status = SwitchMemtable(cfd, context);
     }
-    if (cfd->Unref()) {
-      delete cfd;
+    if (cfd->UnrefAndTryDelete()) {
       cfd = nullptr;
     }
     if (!status.ok()) {

db/db_impl/db_secondary_test.cc

@@ -198,7 +198,7 @@ TEST_F(DBSecondaryTest, OpenAsSecondary) {
 namespace {
 class TraceFileEnv : public EnvWrapper {
  public:
-  explicit TraceFileEnv(Env* target) : EnvWrapper(target) {}
+  explicit TraceFileEnv(Env* _target) : EnvWrapper(_target) {}
   Status NewRandomAccessFile(const std::string& f,
                              std::unique_ptr<RandomAccessFile>* r,
                              const EnvOptions& env_options) override {

db/db_iterator_test.cc

@@ -804,6 +804,50 @@ TEST_P(DBIteratorTest, IteratorPinsRef) {
   } while (ChangeCompactOptions());
 }
 
+TEST_P(DBIteratorTest, IteratorDeleteAfterCfDelete) {
+  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+
+  Put(1, "foo", "delete-cf-then-delete-iter");
+  Put(1, "hello", "value2");
+
+  ColumnFamilyHandle* cf = handles_[1];
+  ReadOptions ro;
+
+  auto* iter = db_->NewIterator(ro, cf);
+  iter->SeekToFirst();
+  ASSERT_EQ(IterStatus(iter), "foo->delete-cf-then-delete-iter");
+
+  // delete CF handle
+  db_->DestroyColumnFamilyHandle(cf);
+  handles_.erase(std::begin(handles_) + 1);
+
+  // delete Iterator after CF handle is deleted
+  iter->Next();
+  ASSERT_EQ(IterStatus(iter), "hello->value2");
+  delete iter;
+}
+
+TEST_P(DBIteratorTest, IteratorDeleteAfterCfDrop) {
+  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+
+  Put(1, "foo", "drop-cf-then-delete-iter");
+
+  ReadOptions ro;
+  ColumnFamilyHandle* cf = handles_[1];
+
+  auto* iter = db_->NewIterator(ro, cf);
+  iter->SeekToFirst();
+  ASSERT_EQ(IterStatus(iter), "foo->drop-cf-then-delete-iter");
+
+  // drop and delete CF
+  db_->DropColumnFamily(cf);
+  db_->DestroyColumnFamilyHandle(cf);
+  handles_.erase(std::begin(handles_) + 1);
+
+  // delete Iterator after CF handle is dropped
+  delete iter;
+}
+
 // SetOptions not defined in ROCKSDB LITE
 #ifndef ROCKSDB_LITE
 TEST_P(DBIteratorTest, DBIteratorBoundTest) {
@@ -1297,19 +1341,19 @@ class DBIteratorTestForPinnedData : public DBIteratorTest {
 
       // Insert data to true_data map and to DB
       true_data[k] = v;
-      if (rnd.OneIn(static_cast<int>(100.0 / merge_percentage))) {
+      if (rnd.PercentTrue(merge_percentage)) {
         ASSERT_OK(db_->Merge(WriteOptions(), k, v));
       } else {
         ASSERT_OK(Put(k, v));
       }
 
       // Pick random keys to be used to test Seek()
-      if (rnd.OneIn(static_cast<int>(100.0 / seeks_percentage))) {
+      if (rnd.PercentTrue(seeks_percentage)) {
         random_keys.push_back(k);
       }
 
       // Delete some random keys
-      if (rnd.OneIn(static_cast<int>(100.0 / delete_percentage))) {
+      if (rnd.PercentTrue(delete_percentage)) {
         deleted_keys.push_back(k);
         true_data.erase(k);
         ASSERT_OK(Delete(k));
@@ -1867,7 +1911,7 @@ TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
   // make sure that we dont merge them while flushing but actually
   // merge them in the read path
   for (int i = 0; i < kNumKeys; i++) {
-    if (rnd.OneIn(static_cast<int>(100.0 / kNoMergeOpPercentage))) {
+    if (rnd.PercentTrue(kNoMergeOpPercentage)) {
       // Dont give merge operations for some keys
       continue;
     }
@@ -1883,7 +1927,7 @@ TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
   ASSERT_OK(Flush());
 
   for (int i = 0; i < kNumKeys; i++) {
-    if (rnd.OneIn(static_cast<int>(100.0 / kDeletePercentage))) {
+    if (rnd.PercentTrue(kDeletePercentage)) {
       gen_key = Key(i);
 
       ASSERT_OK(Delete(gen_key));

db/db_merge_operator_test.cc

@@ -150,6 +150,8 @@ TEST_F(DBMergeOperatorTest, MergeErrorOnWrite) {
   options.create_if_missing = true;
   options.merge_operator.reset(new TestPutOperator());
   options.max_successive_merges = 3;
+  // allow_concurrent_memtable_write is incompatible with max_successive_merges
+  options.allow_concurrent_memtable_write = false;
   options.env = env_;
   Reopen(options);
   ASSERT_OK(Merge("k1", "v1"));

db/db_test.cc

@@ -1705,6 +1705,7 @@ TEST_F(DBTest, Snapshot) {
     ASSERT_EQ(1U, GetNumSnapshots());
     uint64_t time_snap1 = GetTimeOldestSnapshots();
     ASSERT_GT(time_snap1, 0U);
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     Put(0, "foo", "0v2");
     Put(1, "foo", "1v2");
 
@@ -1713,6 +1714,7 @@ TEST_F(DBTest, Snapshot) {
     const Snapshot* s2 = db_->GetSnapshot();
     ASSERT_EQ(2U, GetNumSnapshots());
     ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     Put(0, "foo", "0v3");
     Put(1, "foo", "1v3");
 
@@ -1720,6 +1722,7 @@ TEST_F(DBTest, Snapshot) {
       ManagedSnapshot s3(db_);
       ASSERT_EQ(3U, GetNumSnapshots());
       ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+      ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
 
       Put(0, "foo", "0v4");
       Put(1, "foo", "1v4");
@@ -1735,6 +1738,7 @@ TEST_F(DBTest, Snapshot) {
 
     ASSERT_EQ(2U, GetNumSnapshots());
     ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s1->GetSequenceNumber());
     ASSERT_EQ("0v1", Get(0, "foo", s1));
     ASSERT_EQ("1v1", Get(1, "foo", s1));
     ASSERT_EQ("0v2", Get(0, "foo", s2));
@@ -1749,9 +1753,11 @@ TEST_F(DBTest, Snapshot) {
     ASSERT_EQ("1v4", Get(1, "foo"));
     ASSERT_EQ(1U, GetNumSnapshots());
     ASSERT_LT(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), s2->GetSequenceNumber());
 
     db_->ReleaseSnapshot(s2);
     ASSERT_EQ(0U, GetNumSnapshots());
+    ASSERT_EQ(GetSequenceOldestSnapshots(), 0);
     ASSERT_EQ("0v4", Get(0, "foo"));
     ASSERT_EQ("1v4", Get(1, "foo"));
   } while (ChangeOptions());

db/db_test2.cc

@@ -153,6 +153,8 @@ TEST_F(DBTest2, MaxSuccessiveMergesChangeWithDBRecovery) {
   options.create_if_missing = true;
   options.statistics = rocksdb::CreateDBStatistics();
   options.max_successive_merges = 3;
+  // allow_concurrent_memtable_write is incompatible with max_successive_merges
+  options.allow_concurrent_memtable_write = false;
   options.merge_operator = MergeOperators::CreatePutOperator();
   options.disable_auto_compactions = true;
   DestroyAndReopen(options);
@@ -4246,6 +4248,26 @@ TEST_F(DBTest2, BackgroundPurgeTest) {
   value = options.write_buffer_manager->memory_usage();
   ASSERT_EQ(base_value, value);
 }
+
+TEST_F(DBTest2, SwitchMemtableRaceWithNewManifest) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  options.max_manifest_file_size = 10;
+  options.create_if_missing = true;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  ASSERT_EQ(2, handles_.size());
+
+  ASSERT_OK(Put("foo", "value"));
+  const int kL0Files = options.level0_file_num_compaction_trigger;
+  for (int i = 0; i < kL0Files; ++i) {
+    ASSERT_OK(Put(/*cf=*/1, "a", std::to_string(i)));
+    ASSERT_OK(Flush(/*cf=*/1));
+  }
+
+  port::Thread thread([&]() { ASSERT_OK(Flush()); });
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  thread.join();
+}
 }  // namespace rocksdb
 
 #ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS

db/db_test_util.cc

@@ -867,6 +867,13 @@ uint64_t DBTestBase::GetTimeOldestSnapshots() {
   return int_num;
 }
 
+uint64_t DBTestBase::GetSequenceOldestSnapshots() {
+  uint64_t int_num;
+  EXPECT_TRUE(
+      dbfull()->GetIntProperty("rocksdb.oldest-snapshot-sequence", &int_num));
+  return int_num;
+}
+
 // Return a string that contains all key,value pairs in order,
 // formatted like "(k1->v1)(k2->v2)".
 std::string DBTestBase::Contents(int cf) {

db/db_test_util.h

@@ -860,6 +860,8 @@ class DBTestBase : public testing::Test {
 
   uint64_t GetTimeOldestSnapshots();
 
+  uint64_t GetSequenceOldestSnapshots();
+
   // Return a string that contains all key,value pairs in order,
   // formatted like "(k1->v1)(k2->v2)".
   std::string Contents(int cf = 0);

db/db_universal_compaction_test.cc

@@ -1753,13 +1753,14 @@ TEST_P(DBTestUniversalCompaction, RecalculateScoreAfterPicking) {
   // scheduling more; otherwise, other CFs/DBs may be delayed unnecessarily.
   const int kNumFilesTrigger = 8;
   Options options = CurrentOptions();
+  options.memtable_factory.reset(
+      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
   options.compaction_options_universal.max_merge_width = kNumFilesTrigger / 2;
   options.compaction_options_universal.max_size_amplification_percent =
       static_cast<unsigned int>(-1);
   options.compaction_style = kCompactionStyleUniversal;
   options.level0_file_num_compaction_trigger = kNumFilesTrigger;
   options.num_levels = num_levels_;
-  options.write_buffer_size = 100 << 10;  // 100KB
   Reopen(options);
 
   std::atomic<int> num_compactions_attempted(0);

db/db_wal_test.cc

@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "db/db_test_util.h"
+#include "env/composite_env_wrapper.h"
 #include "options/options_helper.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
@@ -902,8 +903,8 @@ class RecoveryTestHelper {
       std::string fname = LogFileName(test->dbname_, current_log_number);
       std::unique_ptr<WritableFile> file;
       ASSERT_OK(db_options.env->NewWritableFile(fname, &file, env_options));
-      std::unique_ptr<WritableFileWriter> file_writer(
-          new WritableFileWriter(std::move(file), fname, env_options));
+      std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+          NewLegacyWritableFileWrapper(std::move(file)), fname, env_options));
       current_log_writer.reset(
           new log::Writer(std::move(file_writer), current_log_number,
                           db_options.recycle_log_file_num > 0));

db/db_write_test.cc

@@ -39,97 +39,6 @@ TEST_P(DBWriteTest, SyncAndDisableWAL) {
   ASSERT_TRUE(dbfull()->Write(write_options, &batch).IsInvalidArgument());
 }
 
-TEST_P(DBWriteTest, WriteThreadHangOnWriteStall) {
-  Options options = GetOptions();
-  options.level0_stop_writes_trigger = options.level0_slowdown_writes_trigger = 4;
-  std::vector<port::Thread> threads;
-  std::atomic<int> thread_num(0);
-  port::Mutex mutex;
-  port::CondVar cv(&mutex);
-
-  Reopen(options);
-
-  std::function<void()> write_slowdown_func = [&]() {
-    int a = thread_num.fetch_add(1);
-    std::string key = "foo" + std::to_string(a);
-    WriteOptions wo;
-    wo.no_slowdown = false;
-    dbfull()->Put(wo, key, "bar");
-  };
-  std::function<void()> write_no_slowdown_func = [&]() {
-    int a = thread_num.fetch_add(1);
-    std::string key = "foo" + std::to_string(a);
-    WriteOptions wo;
-    wo.no_slowdown = true;
-    dbfull()->Put(wo, key, "bar");
-  };
-  std::function<void(void *)> unblock_main_thread_func = [&](void *) {
-    mutex.Lock();
-    cv.SignalAll();
-    mutex.Unlock();
-  };
-
-  // Create 3 L0 files and schedule 4th without waiting
-  Put("foo" + std::to_string(thread_num.fetch_add(1)), "bar");
-  Flush();
-  Put("foo" + std::to_string(thread_num.fetch_add(1)), "bar");
-  Flush();
-  Put("foo" + std::to_string(thread_num.fetch_add(1)), "bar");
-  Flush();
-  Put("foo" + std::to_string(thread_num.fetch_add(1)), "bar");
-
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-    "WriteThread::JoinBatchGroup:Start", unblock_main_thread_func);
-  rocksdb::SyncPoint::GetInstance()->LoadDependency(
-    {{"DBWriteTest::WriteThreadHangOnWriteStall:1",
-      "DBImpl::BackgroundCallFlush:start"},
-     {"DBWriteTest::WriteThreadHangOnWriteStall:2",
-      "DBImpl::WriteImpl:BeforeLeaderEnters"},
-     // Make compaction start wait for the write stall to be detected and
-     // implemented by a write group leader
-     {"DBWriteTest::WriteThreadHangOnWriteStall:3",
-      "BackgroundCallCompaction:0"}});
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-
-  // Schedule creation of 4th L0 file without waiting. This will seal the
-  // memtable and then wait for a sync point before writing the file. We need
-  // to do it this way because SwitchMemtable() needs to enter the
-  // write_thread
-  FlushOptions fopt;
-  fopt.wait = false;
-  dbfull()->Flush(fopt);
-
-  // Create a mix of slowdown/no_slowdown write threads
-  mutex.Lock();
-  // First leader
-  threads.emplace_back(write_slowdown_func);
-  cv.Wait();
-  // Second leader. Will stall writes
-  threads.emplace_back(write_slowdown_func);
-  cv.Wait();
-  threads.emplace_back(write_no_slowdown_func);
-  cv.Wait();
-  threads.emplace_back(write_slowdown_func);
-  cv.Wait();
-  threads.emplace_back(write_no_slowdown_func);
-  cv.Wait();
-  threads.emplace_back(write_slowdown_func);
-  cv.Wait();
-  mutex.Unlock();
-
-  TEST_SYNC_POINT("DBWriteTest::WriteThreadHangOnWriteStall:1");
-  dbfull()->TEST_WaitForFlushMemTable(nullptr);
-  // This would have triggered a write stall. Unblock the write group leader
-  TEST_SYNC_POINT("DBWriteTest::WriteThreadHangOnWriteStall:2");
-  // The leader is going to create missing newer links. When the leader finishes,
-  // the next leader is going to delay writes and fail writers with no_slowdown
-
-  TEST_SYNC_POINT("DBWriteTest::WriteThreadHangOnWriteStall:3");
-  for (auto& t : threads) {
-    t.join();
-  }
-}
-
 TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
   constexpr int kNumThreads = 5;
   std::unique_ptr<FaultInjectionTestEnv> mock_env(

db/error_handler.cc

@@ -166,6 +166,12 @@ Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reas
     return Status::OK();
   }
 
+  // Check if recovery is currently in progress. If it is, we will save this
+  // error so we can check it at the end to see if recovery succeeded or not
+  if (recovery_in_prog_ && recovery_error_.ok()) {
+    recovery_error_ = bg_err;
+  }
+
   bool paranoid = db_options_.paranoid_checks;
   Status::Severity sev = Status::Severity::kFatalError;
   Status new_bg_err;
@@ -198,15 +204,10 @@ Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reas
 
   new_bg_err = Status(bg_err, sev);
 
-  // Check if recovery is currently in progress. If it is, we will save this
-  // error so we can check it at the end to see if recovery succeeded or not
-  if (recovery_in_prog_ && recovery_error_.ok()) {
-    recovery_error_ = new_bg_err;
-  }
-
   bool auto_recovery = auto_recovery_;
   if (new_bg_err.severity() >= Status::Severity::kFatalError && auto_recovery) {
     auto_recovery = false;
+    ;
   }
 
   // Allow some error specific overrides

db/error_handler_test.cc

@@ -22,21 +22,6 @@ namespace rocksdb {
 class DBErrorHandlingTest : public DBTestBase {
  public:
   DBErrorHandlingTest() : DBTestBase("/db_error_handling_test") {}
-
-  std::string GetManifestNameFromLiveFiles() {
-    std::vector<std::string> live_files;
-    uint64_t manifest_size;
-
-    dbfull()->GetLiveFiles(live_files, &manifest_size, false);
-    for (auto& file : live_files) {
-      uint64_t num = 0;
-      FileType type;
-      if (ParseFileName(file, &num, &type) && type == kDescriptorFile) {
-        return file;
-      }
-    }
-    return "";
-  }
 };
 
 class DBErrorHandlingEnv : public EnvWrapper {
@@ -176,169 +161,6 @@ TEST_F(DBErrorHandlingTest, FLushWriteError) {
   Destroy(options);
 }
 
-TEST_F(DBErrorHandlingTest, ManifestWriteError) {
-  std::unique_ptr<FaultInjectionTestEnv> fault_env(
-      new FaultInjectionTestEnv(Env::Default()));
-  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
-  Options options = GetDefaultOptions();
-  options.create_if_missing = true;
-  options.env = fault_env.get();
-  options.listeners.emplace_back(listener);
-  Status s;
-  std::string old_manifest;
-  std::string new_manifest;
-
-  listener->EnableAutoRecovery(false);
-  DestroyAndReopen(options);
-  old_manifest = GetManifestNameFromLiveFiles();
-
-  Put(Key(0), "val");
-  Flush();
-  Put(Key(1), "val");
-  SyncPoint::GetInstance()->SetCallBack(
-      "VersionSet::LogAndApply:WriteManifest", [&](void *) {
-    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
-  });
-  SyncPoint::GetInstance()->EnableProcessing();
-  s = Flush();
-  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-  SyncPoint::GetInstance()->DisableProcessing();
-  fault_env->SetFilesystemActive(true);
-  s = dbfull()->Resume();
-  ASSERT_EQ(s, Status::OK());
-
-  new_manifest = GetManifestNameFromLiveFiles();
-  ASSERT_NE(new_manifest, old_manifest);
-
-  Reopen(options);
-  ASSERT_EQ("val", Get(Key(0)));
-  ASSERT_EQ("val", Get(Key(1)));
-  Close();
-}
-
-TEST_F(DBErrorHandlingTest, DoubleManifestWriteError) {
-  std::unique_ptr<FaultInjectionTestEnv> fault_env(
-      new FaultInjectionTestEnv(Env::Default()));
-  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
-  Options options = GetDefaultOptions();
-  options.create_if_missing = true;
-  options.env = fault_env.get();
-  options.listeners.emplace_back(listener);
-  Status s;
-  std::string old_manifest;
-  std::string new_manifest;
-
-  listener->EnableAutoRecovery(false);
-  DestroyAndReopen(options);
-  old_manifest = GetManifestNameFromLiveFiles();
-
-  Put(Key(0), "val");
-  Flush();
-  Put(Key(1), "val");
-  SyncPoint::GetInstance()->SetCallBack(
-      "VersionSet::LogAndApply:WriteManifest", [&](void *) {
-    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
-  });
-  SyncPoint::GetInstance()->EnableProcessing();
-  s = Flush();
-  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
-  fault_env->SetFilesystemActive(true);
-
-  // This Resume() will attempt to create a new manifest file and fail again
-  s = dbfull()->Resume();
-  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
-  fault_env->SetFilesystemActive(true);
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-  SyncPoint::GetInstance()->DisableProcessing();
-
-  // A successful Resume() will create a new manifest file
-  s = dbfull()->Resume();
-  ASSERT_EQ(s, Status::OK());
-
-  new_manifest = GetManifestNameFromLiveFiles();
-  ASSERT_NE(new_manifest, old_manifest);
-
-  Reopen(options);
-  ASSERT_EQ("val", Get(Key(0)));
-  ASSERT_EQ("val", Get(Key(1)));
-  Close();
-}
-
-TEST_F(DBErrorHandlingTest, CompactionManifestWriteError) {
-  std::unique_ptr<FaultInjectionTestEnv> fault_env(
-      new FaultInjectionTestEnv(Env::Default()));
-  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
-  Options options = GetDefaultOptions();
-  options.create_if_missing = true;
-  options.level0_file_num_compaction_trigger = 2;
-  options.listeners.emplace_back(listener);
-  options.env = fault_env.get();
-  Status s;
-  std::string old_manifest;
-  std::string new_manifest;
-  std::atomic<bool> fail_manifest(false);
-  DestroyAndReopen(options);
-  old_manifest = GetManifestNameFromLiveFiles();
-
-  Put(Key(0), "val");
-  Put(Key(2), "val");
-  s = Flush();
-  ASSERT_EQ(s, Status::OK());
-
-  rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      // Wait for flush of 2nd L0 file before starting compaction
-      {{"FlushMemTableFinished",
-        "BackgroundCallCompaction:0"},
-      // Wait for compaction to detect manifest write error
-       {"BackgroundCallCompaction:1",
-        "CompactionManifestWriteError:0"},
-      // Make compaction thread wait for error to be cleared
-       {"CompactionManifestWriteError:1",
-        "DBImpl::BackgroundCallCompaction:FoundObsoleteFiles"},
-      // Wait for DB instance to clear bg_error before calling
-      // TEST_WaitForCompact
-       {"SstFileManagerImpl::ClearError",
-        "CompactionManifestWriteError:2"}});
-  // trigger manifest write failure in compaction thread
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "BackgroundCallCompaction:0", [&](void *) {
-      fail_manifest.store(true);
-      });
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "VersionSet::LogAndApply:WriteManifest", [&](void *) {
-      if (fail_manifest.load()) {
-        fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
-      }
-      });
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-
-  Put(Key(1), "val");
-  // This Flush will trigger a compaction, which will fail when appending to
-  // the manifest
-  s = Flush();
-  ASSERT_EQ(s, Status::OK());
-
-  TEST_SYNC_POINT("CompactionManifestWriteError:0");
-  // Clear all errors so when the compaction is retried, it will succeed
-  fault_env->SetFilesystemActive(true);
-  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
-  TEST_SYNC_POINT("CompactionManifestWriteError:1");
-  TEST_SYNC_POINT("CompactionManifestWriteError:2");
-
-  s = dbfull()->TEST_WaitForCompact();
-  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
-  ASSERT_EQ(s, Status::OK());
-
-  new_manifest = GetManifestNameFromLiveFiles();
-  ASSERT_NE(new_manifest, old_manifest);
-  Reopen(options);
-  ASSERT_EQ("val", Get(Key(0)));
-  ASSERT_EQ("val", Get(Key(1)));
-  ASSERT_EQ("val", Get(Key(2)));
-  Close();
-}
-
 TEST_F(DBErrorHandlingTest, CompactionWriteError) {
   std::unique_ptr<FaultInjectionTestEnv> fault_env(
       new FaultInjectionTestEnv(Env::Default()));
@@ -361,7 +183,8 @@ TEST_F(DBErrorHandlingTest, CompactionWriteError) {
       );
   listener->EnableAutoRecovery(false);
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+      {{"DBImpl::FlushMemTable:FlushMemTableFinished",
+        "BackgroundCallCompaction:0"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "BackgroundCallCompaction:0", [&](void *) {
       fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
@@ -397,7 +220,8 @@ TEST_F(DBErrorHandlingTest, CorruptionError) {
   ASSERT_EQ(s, Status::OK());
 
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+      {{"DBImpl::FlushMemTable:FlushMemTableFinished",
+        "BackgroundCallCompaction:0"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "BackgroundCallCompaction:0", [&](void *) {
       fault_env->SetFilesystemActive(false, Status::Corruption("Corruption"));

db/external_sst_file_ingestion_job.cc

@@ -101,14 +101,15 @@ Status ExternalSstFileIngestionJob::Prepare(
         TableFileName(cfd_->ioptions()->cf_paths, f.fd.GetNumber(),
                       f.fd.GetPathId());
     if (ingestion_options_.move_files) {
-      status = env_->LinkFile(path_outside_db, path_inside_db);
+      status =
+          fs_->LinkFile(path_outside_db, path_inside_db, IOOptions(), nullptr);
       if (status.ok()) {
         // It is unsafe to assume application had sync the file and file
         // directory before ingest the file. For integrity of RocksDB we need
         // to sync the file.
-        std::unique_ptr<WritableFile> file_to_sync;
-        status = env_->ReopenWritableFile(path_inside_db, &file_to_sync,
-                                          env_options_);
+        std::unique_ptr<FSWritableFile> file_to_sync;
+        status = fs_->ReopenWritableFile(path_inside_db, env_options_,
+                                         &file_to_sync, nullptr);
         if (status.ok()) {
           TEST_SYNC_POINT(
               "ExternalSstFileIngestionJob::BeforeSyncIngestedFile");
@@ -133,7 +134,7 @@ Status ExternalSstFileIngestionJob::Prepare(
       TEST_SYNC_POINT_CALLBACK("ExternalSstFileIngestionJob::Prepare:CopyFile",
                                nullptr);
       // CopyFile also sync the new file.
-      status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+      status = CopyFile(fs_, path_outside_db, path_inside_db, 0,
                         db_options_.use_fsync);
     }
     TEST_SYNC_POINT("ExternalSstFileIngestionJob::Prepare:FileAdded");
@@ -360,17 +361,19 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
   file_to_ingest->external_file_path = external_file;
 
   // Get external file size
-  Status status = env_->GetFileSize(external_file, &file_to_ingest->file_size);
+  Status status = fs_->GetFileSize(external_file, IOOptions(),
+                                   &file_to_ingest->file_size, nullptr);
   if (!status.ok()) {
     return status;
   }
 
   // Create TableReader for external file
   std::unique_ptr<TableReader> table_reader;
-  std::unique_ptr<RandomAccessFile> sst_file;
+  std::unique_ptr<FSRandomAccessFile> sst_file;
   std::unique_ptr<RandomAccessFileReader> sst_file_reader;
 
-  status = env_->NewRandomAccessFile(external_file, &sst_file, env_options_);
+  status = fs_->NewRandomAccessFile(external_file, env_options_,
+                                    &sst_file, nullptr);
   if (!status.ok()) {
     return status;
   }
@@ -651,13 +654,15 @@ Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
     // Determine if we can write global_seqno to a given offset of file.
     // If the file system does not support random write, then we should not.
     // Otherwise we should.
-    std::unique_ptr<RandomRWFile> rwfile;
-    Status status = env_->NewRandomRWFile(file_to_ingest->internal_file_path,
-                                          &rwfile, env_options_);
+    std::unique_ptr<FSRandomRWFile> rwfile;
+    Status status =
+        fs_->NewRandomRWFile(file_to_ingest->internal_file_path, env_options_,
+                             &rwfile, nullptr);
     if (status.ok()) {
       std::string seqno_val;
       PutFixed64(&seqno_val, seqno);
-      status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
+      status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val,
+                             IOOptions(), nullptr);
       if (status.ok()) {
         TEST_SYNC_POINT("ExternalSstFileIngestionJob::BeforeSyncGlobalSeqno");
         status = SyncIngestedFile(rwfile.get());
@@ -715,9 +720,9 @@ template <typename TWritableFile>
 Status ExternalSstFileIngestionJob::SyncIngestedFile(TWritableFile* file) {
   assert(file != nullptr);
   if (db_options_.use_fsync) {
-    return file->Fsync();
+    return file->Fsync(IOOptions(), nullptr);
   } else {
-    return file->Sync();
+    return file->Sync(IOOptions(), nullptr);
   }
 }
 

db/external_sst_file_ingestion_job.h

@@ -74,6 +74,7 @@ class ExternalSstFileIngestionJob {
       const IngestExternalFileOptions& ingestion_options,
       Directories* directories, EventLogger* event_logger)
       : env_(env),
+        fs_(db_options.fs.get()),
         versions_(versions),
         cfd_(cfd),
         db_options_(db_options),
@@ -158,6 +159,7 @@ class ExternalSstFileIngestionJob {
   Status SyncIngestedFile(TWritableFile* file);
 
   Env* env_;
+  FileSystem* fs_;
   VersionSet* versions_;
   ColumnFamilyData* cfd_;
   const ImmutableDBOptions& db_options_;

db/external_sst_file_test.cc

@@ -1700,6 +1700,35 @@ TEST_F(ExternalSSTFileTest, SstFileWriterNonSharedKeys) {
   ASSERT_OK(DeprecatedAddFile({file_path}));
 }
 
+TEST_F(ExternalSSTFileTest, WithUnorderedWrite) {
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::WriteImpl:UnorderedWriteAfterWriteWAL",
+        "ExternalSSTFileTest::WithUnorderedWrite:WaitWriteWAL"},
+       {"DBImpl::WaitForPendingWrites:BeforeBlock",
+        "DBImpl::WriteImpl:BeforeUnorderedWriteMemtable"}});
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::IngestExternalFile:NeedFlush", [&](void* need_flush) {
+        ASSERT_TRUE(*reinterpret_cast<bool*>(need_flush));
+      });
+
+  Options options = CurrentOptions();
+  options.unordered_write = true;
+  DestroyAndReopen(options);
+  Put("foo", "v1");
+  SyncPoint::GetInstance()->EnableProcessing();
+  port::Thread writer([&]() { Put("bar", "v2"); });
+
+  TEST_SYNC_POINT("ExternalSSTFileTest::WithUnorderedWrite:WaitWriteWAL");
+  ASSERT_OK(GenerateAndAddExternalFile(options, {{"bar", "v3"}}, -1,
+                                       true /* allow_global_seqno */));
+  ASSERT_EQ(Get("bar"), "v3");
+
+  writer.join();
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
 TEST_P(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
   Options options = CurrentOptions();
   options.IncreaseParallelism(20);

db/flush_job.cc

@@ -86,7 +86,7 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                    const ImmutableDBOptions& db_options,
                    const MutableCFOptions& mutable_cf_options,
                    const uint64_t* max_memtable_id,
-                   const EnvOptions& env_options, VersionSet* versions,
+                   const FileOptions& file_options, VersionSet* versions,
                    InstrumentedMutex* db_mutex,
                    std::atomic<bool>* shutting_down,
                    std::vector<SequenceNumber> existing_snapshots,
@@ -103,7 +103,7 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
       db_options_(db_options),
       mutable_cf_options_(mutable_cf_options),
       max_memtable_id_(max_memtable_id),
-      env_options_(env_options),
+      file_options_(file_options),
       versions_(versions),
       db_mutex_(db_mutex),
       shutting_down_(shutting_down),
@@ -249,7 +249,8 @@ Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
   }
   RecordFlushIOStats();
 
-  auto stream = event_logger_->LogToBuffer(log_buffer_);
+  // When measure_io_stats_ is true, the default 512 bytes is not enough.
+  auto stream = event_logger_->LogToBuffer(log_buffer_, 1024);
   stream << "job" << job_context_->job_id << "event"
          << "flush_finished";
   stream << "output_compression"
@@ -371,8 +372,8 @@ Status FlushJob::WriteLevel0Table() {
       meta_.file_creation_time = current_time;
 
       s = BuildTable(
-          dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
-          env_options_, cfd_->table_cache(), iter.get(),
+          dbname_, db_options_.env, db_options_.fs.get(), *cfd_->ioptions(),
+          mutable_cf_options_, file_options_, cfd_->table_cache(), iter.get(),
           std::move(range_del_iters), &meta_, cfd_->internal_comparator(),
           cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
           cfd_->GetName(), existing_snapshots_,

db/flush_job.h

@@ -61,7 +61,7 @@ class FlushJob {
   FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
            const ImmutableDBOptions& db_options,
            const MutableCFOptions& mutable_cf_options,
-           const uint64_t* max_memtable_id, const EnvOptions& env_options,
+           const uint64_t* max_memtable_id, const FileOptions& file_options,
            VersionSet* versions, InstrumentedMutex* db_mutex,
            std::atomic<bool>* shutting_down,
            std::vector<SequenceNumber> existing_snapshots,
@@ -108,7 +108,7 @@ class FlushJob {
   // equal to *max_memtable_id_ will be selected for flush. If null, then all
   // memtables in the column family will be selected.
   const uint64_t* max_memtable_id_;
-  const EnvOptions env_options_;
+  const FileOptions file_options_;
   VersionSet* versions_;
   InstrumentedMutex* db_mutex_;
   std::atomic<bool>* shutting_down_;

db/flush_job_test.cc

@@ -30,16 +30,13 @@ class FlushJobTest : public testing::Test {
  public:
   FlushJobTest()
       : env_(Env::Default()),
+        fs_(std::make_shared<LegacyFileSystemWrapper>(env_)),
         dbname_(test::PerThreadDBPath("flush_job_test")),
         options_(),
         db_options_(options_),
         column_family_names_({kDefaultColumnFamilyName, "foo", "bar"}),
         table_cache_(NewLRUCache(50000, 16)),
         write_buffer_manager_(db_options_.db_write_buffer_size),
-        versions_(new VersionSet(dbname_, &db_options_, env_options_,
-                                 table_cache_.get(), &write_buffer_manager_,
-                                 &write_controller_,
-                                 /*block_cache_tracer=*/nullptr)),
         shutting_down_(false),
         mock_table_factory_(new mock::MockTableFactory()) {
     EXPECT_OK(env_->CreateDirIfMissing(dbname_));
@@ -54,6 +51,12 @@ class FlushJobTest : public testing::Test {
       column_families.emplace_back(cf_name, cf_options_);
     }
 
+    db_options_.env = env_;
+    db_options_.fs = fs_;
+    versions_.reset(new VersionSet(dbname_, &db_options_, env_options_,
+                                   table_cache_.get(), &write_buffer_manager_,
+                                   &write_controller_,
+                                   /*block_cache_tracer=*/nullptr));
     EXPECT_OK(versions_->Recover(column_families, false));
   }
 
@@ -88,8 +91,8 @@ class FlushJobTest : public testing::Test {
     Status s = env_->NewWritableFile(
         manifest, &file, env_->OptimizeForManifestWrite(env_options_));
     ASSERT_OK(s);
-    std::unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), manifest, EnvOptions()));
+    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+        NewLegacyWritableFileWrapper(std::move(file)), manifest, EnvOptions()));
     {
       log::Writer log(std::move(file_writer), 0, false);
       std::string record;
@@ -109,6 +112,7 @@ class FlushJobTest : public testing::Test {
   }
 
   Env* env_;
+  std::shared_ptr<FileSystem> fs_;
   std::string dbname_;
   EnvOptions env_options_;
   Options options_;

db/flush_scheduler.cc

@@ -60,9 +60,7 @@ ColumnFamilyData* FlushScheduler::TakeNextColumnFamily() {
     }
 
     // no longer relevant, retry
-    if (cfd->Unref()) {
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
 }
 
@@ -80,9 +78,7 @@ bool FlushScheduler::Empty() {
 void FlushScheduler::Clear() {
   ColumnFamilyData* cfd;
   while ((cfd = TakeNextColumnFamily()) != nullptr) {
-    if (cfd->Unref()) {
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
   assert(head_.load(std::memory_order_relaxed) == nullptr);
 }

db/import_column_family_job.cc

@@ -92,14 +92,15 @@ Status ImportColumnFamilyJob::Prepare(uint64_t next_file_number,
         cfd_->ioptions()->cf_paths, f.fd.GetNumber(), f.fd.GetPathId());
 
     if (hardlink_files) {
-      status = env_->LinkFile(path_outside_db, path_inside_db);
+      status =
+          fs_->LinkFile(path_outside_db, path_inside_db, IOOptions(), nullptr);
       if (status.IsNotSupported()) {
         // Original file is on a different FS, use copy instead of hard linking
         hardlink_files = false;
       }
     }
     if (!hardlink_files) {
-      status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+      status = CopyFile(fs_, path_outside_db, path_inside_db, 0,
                         db_options_.use_fsync);
     }
     if (!status.ok()) {
@@ -115,7 +116,8 @@ Status ImportColumnFamilyJob::Prepare(uint64_t next_file_number,
       if (f.internal_file_path.empty()) {
         break;
       }
-      const auto s = env_->DeleteFile(f.internal_file_path);
+      const auto s =
+          fs_->DeleteFile(f.internal_file_path, IOOptions(), nullptr);
       if (!s.ok()) {
         ROCKS_LOG_WARN(db_options_.info_log,
                        "AddFile() clean up for file %s failed : %s",
@@ -168,7 +170,8 @@ void ImportColumnFamilyJob::Cleanup(const Status& status) {
   if (!status.ok()) {
     // We failed to add files to the database remove all the files we copied.
     for (const auto& f : files_to_import_) {
-      const auto s = env_->DeleteFile(f.internal_file_path);
+      const auto s =
+          fs_->DeleteFile(f.internal_file_path, IOOptions(), nullptr);
       if (!s.ok()) {
         ROCKS_LOG_WARN(db_options_.info_log,
                        "AddFile() clean up for file %s failed : %s",
@@ -178,7 +181,8 @@ void ImportColumnFamilyJob::Cleanup(const Status& status) {
   } else if (status.ok() && import_options_.move_files) {
     // The files were moved and added successfully, remove original file links
     for (IngestedFileInfo& f : files_to_import_) {
-      const auto s = env_->DeleteFile(f.external_file_path);
+      const auto s =
+          fs_->DeleteFile(f.external_file_path, IOOptions(), nullptr);
       if (!s.ok()) {
         ROCKS_LOG_WARN(
             db_options_.info_log,
@@ -196,17 +200,19 @@ Status ImportColumnFamilyJob::GetIngestedFileInfo(
   file_to_import->external_file_path = external_file;
 
   // Get external file size
-  auto status = env_->GetFileSize(external_file, &file_to_import->file_size);
+  Status status = fs_->GetFileSize(external_file, IOOptions(),
+                                   &file_to_import->file_size, nullptr);
   if (!status.ok()) {
     return status;
   }
 
   // Create TableReader for external file
   std::unique_ptr<TableReader> table_reader;
-  std::unique_ptr<RandomAccessFile> sst_file;
+  std::unique_ptr<FSRandomAccessFile> sst_file;
   std::unique_ptr<RandomAccessFileReader> sst_file_reader;
 
-  status = env_->NewRandomAccessFile(external_file, &sst_file, env_options_);
+  status = fs_->NewRandomAccessFile(external_file, env_options_,
+                                    &sst_file, nullptr);
   if (!status.ok()) {
     return status;
   }

db/import_column_family_job.h

@@ -29,6 +29,7 @@ class ImportColumnFamilyJob {
         versions_(versions),
         cfd_(cfd),
         db_options_(db_options),
+        fs_(db_options_.fs.get()),
         env_options_(env_options),
         import_options_(import_options),
         metadata_(metadata) {}
@@ -60,6 +61,7 @@ class ImportColumnFamilyJob {
   VersionSet* versions_;
   ColumnFamilyData* cfd_;
   const ImmutableDBOptions& db_options_;
+  FileSystem* fs_;
   const EnvOptions& env_options_;
   autovector<IngestedFileInfo> files_to_import_;
   VersionEdit edit_;

db/internal_stats.cc

@@ -231,6 +231,7 @@ static const std::string is_file_deletions_enabled =
     "is-file-deletions-enabled";
 static const std::string num_snapshots = "num-snapshots";
 static const std::string oldest_snapshot_time = "oldest-snapshot-time";
+static const std::string oldest_snapshot_sequence = "oldest-snapshot-sequence";
 static const std::string num_live_versions = "num-live-versions";
 static const std::string current_version_number =
     "current-super-version-number";
@@ -309,6 +310,8 @@ const std::string DB::Properties::kNumSnapshots =
     rocksdb_prefix + num_snapshots;
 const std::string DB::Properties::kOldestSnapshotTime =
     rocksdb_prefix + oldest_snapshot_time;
+const std::string DB::Properties::kOldestSnapshotSequence =
+    rocksdb_prefix + oldest_snapshot_sequence;
 const std::string DB::Properties::kNumLiveVersions =
     rocksdb_prefix + num_live_versions;
 const std::string DB::Properties::kCurrentSuperVersionNumber =
@@ -427,6 +430,9 @@ const std::unordered_map<std::string, DBPropertyInfo>
         {DB::Properties::kOldestSnapshotTime,
          {false, nullptr, &InternalStats::HandleOldestSnapshotTime, nullptr,
           nullptr}},
+        {DB::Properties::kOldestSnapshotSequence,
+         {false, nullptr, &InternalStats::HandleOldestSnapshotSequence, nullptr,
+          nullptr}},
         {DB::Properties::kNumLiveVersions,
          {false, nullptr, &InternalStats::HandleNumLiveVersions, nullptr,
           nullptr}},
@@ -772,6 +778,12 @@ bool InternalStats::HandleOldestSnapshotTime(uint64_t* value, DBImpl* db,
   return true;
 }
 
+bool InternalStats::HandleOldestSnapshotSequence(uint64_t* value, DBImpl* db,
+                                                 Version* /*version*/) {
+  *value = static_cast<uint64_t>(db->snapshots().GetOldestSnapshotSequence());
+  return true;
+}
+
 bool InternalStats::HandleNumLiveVersions(uint64_t* value, DBImpl* /*db*/,
                                           Version* /*version*/) {
   *value = cfd_->GetNumLiveVersions();

db/internal_stats.h

@@ -553,6 +553,8 @@ class InternalStats {
   bool HandleEstimateNumKeys(uint64_t* value, DBImpl* db, Version* version);
   bool HandleNumSnapshots(uint64_t* value, DBImpl* db, Version* version);
   bool HandleOldestSnapshotTime(uint64_t* value, DBImpl* db, Version* version);
+  bool HandleOldestSnapshotSequence(uint64_t* value, DBImpl* db,
+                                    Version* version);
   bool HandleNumLiveVersions(uint64_t* value, DBImpl* db, Version* version);
   bool HandleCurrentSuperVersionNumber(uint64_t* value, DBImpl* db,
                                        Version* version);

db/log_test.cc

@@ -9,6 +9,7 @@
 
 #include "db/log_reader.h"
 #include "db/log_writer.h"
+#include "env/composite_env_wrapper.h"
 #include "file/sequence_file_reader.h"
 #include "file/writable_file_writer.h"
 #include "rocksdb/env.h"
@@ -120,6 +121,13 @@ class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
     }
   };
 
+  inline StringSource* GetStringSourceFromLegacyReader(
+      SequentialFileReader* reader) {
+    LegacySequentialFileWrapper* file =
+        static_cast<LegacySequentialFileWrapper*>(reader->file());
+    return static_cast<StringSource*>(file->target());
+  }
+
   class ReportCollector : public Reader::Reporter {
    public:
     size_t dropped_bytes_;
@@ -133,21 +141,19 @@ class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
   };
 
   std::string& dest_contents() {
-    auto dest =
-      dynamic_cast<test::StringSink*>(writer_.file()->writable_file());
+    auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
     assert(dest);
     return dest->contents_;
   }
 
   const std::string& dest_contents() const {
-    auto dest =
-      dynamic_cast<const test::StringSink*>(writer_.file()->writable_file());
+    auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
     assert(dest);
     return dest->contents_;
   }
 
   void reset_source_contents() {
-    auto src = dynamic_cast<StringSource*>(reader_->file()->file());
+    auto src = GetStringSourceFromLegacyReader(reader_->file());
     assert(src);
     src->contents_ = dest_contents();
   }
@@ -214,8 +220,7 @@ class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
   }
 
   void ShrinkSize(int bytes) {
-    auto dest =
-      dynamic_cast<test::StringSink*>(writer_.file()->writable_file());
+    auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
     assert(dest);
     dest->Drop(bytes);
   }
@@ -230,7 +235,7 @@ class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
   }
 
   void ForceError(size_t position = 0) {
-    auto src = dynamic_cast<StringSource*>(reader_->file()->file());
+    auto src = GetStringSourceFromLegacyReader(reader_->file());
     src->force_error_ = true;
     src->force_error_position_ = position;
   }
@@ -244,13 +249,13 @@ class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
   }
 
   void ForceEOF(size_t position = 0) {
-    auto src = dynamic_cast<StringSource*>(reader_->file()->file());
+    auto src = GetStringSourceFromLegacyReader(reader_->file());
     src->force_eof_ = true;
     src->force_eof_position_ = position;
   }
 
   void UnmarkEOF() {
-    auto src = dynamic_cast<StringSource*>(reader_->file()->file());
+    auto src = GetStringSourceFromLegacyReader(reader_->file());
     src->returned_partial_ = false;
     reader_->UnmarkEOF();
   }
@@ -752,8 +757,9 @@ class RetriableLogTest : public ::testing::TestWithParam<int> {
       s = env_->NewWritableFile(log_file_, &writable_file, env_options_);
     }
     if (s.ok()) {
-      writer_.reset(new WritableFileWriter(std::move(writable_file), log_file_,
-                                           env_options_));
+      writer_.reset(new WritableFileWriter(
+          NewLegacyWritableFileWrapper(std::move(writable_file)), log_file_,
+          env_options_));
       assert(writer_ != nullptr);
     }
     std::unique_ptr<SequentialFile> seq_file;
@@ -761,7 +767,8 @@ class RetriableLogTest : public ::testing::TestWithParam<int> {
       s = env_->NewSequentialFile(log_file_, &seq_file, env_options_);
     }
     if (s.ok()) {
-      reader_.reset(new SequentialFileReader(std::move(seq_file), log_file_));
+      reader_.reset(new SequentialFileReader(
+          NewLegacySequentialFileWrapper(seq_file), log_file_));
       assert(reader_ != nullptr);
       log_reader_.reset(new FragmentBufferedReader(
           nullptr, std::move(reader_), &report_, true /* checksum */,
@@ -772,8 +779,7 @@ class RetriableLogTest : public ::testing::TestWithParam<int> {
   }
 
   std::string contents() {
-    auto file =
-        dynamic_cast<test::StringSink*>(log_writer_->file()->writable_file());
+    auto file = test::GetStringSinkFromLegacyWriter(log_writer_->file());
     assert(file != nullptr);
     return file->contents_;
   }

db/memtable_list_test.cc

@@ -92,6 +92,7 @@ class MemTableListTest : public testing::Test {
     CreateDB();
     // Create a mock VersionSet
     DBOptions db_options;
+    db_options.file_system = FileSystem::Default();
     ImmutableDBOptions immutable_db_options(db_options);
     EnvOptions env_options;
     std::shared_ptr<Cache> table_cache(NewLRUCache(50000, 16));
@@ -138,6 +139,8 @@ class MemTableListTest : public testing::Test {
     CreateDB();
     // Create a mock VersionSet
     DBOptions db_options;
+    db_options.file_system.reset(new LegacyFileSystemWrapper(db_options.env));
+
     ImmutableDBOptions immutable_db_options(db_options);
     EnvOptions env_options;
     std::shared_ptr<Cache> table_cache(NewLRUCache(50000, 16));

db/merge_test.cc

@@ -78,6 +78,9 @@ std::shared_ptr<DB> OpenDb(const std::string& dbname, const bool ttl = false,
   options.create_if_missing = true;
   options.merge_operator = std::make_shared<CountMergeOperator>();
   options.max_successive_merges = max_successive_merges;
+  if (max_successive_merges > 0) {
+    options.allow_concurrent_memtable_write = false;
+  }
   Status s;
   DestroyDB(dbname, Options());
 // DBWithTTL is not supported in ROCKSDB_LITE

db/repair.cc

@@ -71,6 +71,7 @@
 #include "db/table_cache.h"
 #include "db/version_edit.h"
 #include "db/write_batch_internal.h"
+#include "env/composite_env_wrapper.h"
 #include "file/filename.h"
 #include "file/writable_file_writer.h"
 #include "options/cf_options.h"
@@ -348,8 +349,8 @@ class Repairer {
     if (!status.ok()) {
       return status;
     }
-    std::unique_ptr<SequentialFileReader> lfile_reader(
-        new SequentialFileReader(std::move(lfile), logname));
+    std::unique_ptr<SequentialFileReader> lfile_reader(new SequentialFileReader(
+        NewLegacySequentialFileWrapper(lfile), logname));
 
     // Create the log reader.
     LogReporter reporter;
@@ -421,17 +422,19 @@ class Repairer {
       if (range_del_iter != nullptr) {
         range_del_iters.emplace_back(range_del_iter);
       }
+
+      LegacyFileSystemWrapper fs(env_);
       status = BuildTable(
-          dbname_, env_, *cfd->ioptions(), *cfd->GetLatestMutableCFOptions(),
-          env_options_, table_cache_, iter.get(), std::move(range_del_iters),
-          &meta, cfd->internal_comparator(),
-          cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
-          {}, kMaxSequenceNumber, snapshot_checker, kNoCompression,
-          0 /* sample_for_compression */, CompressionOptions(), false,
-          nullptr /* internal_stats */, TableFileCreationReason::kRecovery,
-          nullptr /* event_logger */, 0 /* job_id */, Env::IO_HIGH,
-          nullptr /* table_properties */, -1 /* level */, current_time,
-          write_hint);
+          dbname_, env_, &fs, *cfd->ioptions(),
+          *cfd->GetLatestMutableCFOptions(), env_options_, table_cache_,
+          iter.get(), std::move(range_del_iters), &meta,
+          cfd->internal_comparator(), cfd->int_tbl_prop_collector_factories(),
+          cfd->GetID(), cfd->GetName(), {}, kMaxSequenceNumber,
+          snapshot_checker, kNoCompression, 0 /* sample_for_compression */,
+          CompressionOptions(), false, nullptr /* internal_stats */,
+          TableFileCreationReason::kRecovery, nullptr /* event_logger */,
+          0 /* job_id */, Env::IO_HIGH, nullptr /* table_properties */,
+          -1 /* level */, current_time, write_hint);
       ROCKS_LOG_INFO(db_options_.info_log,
                      "Log #%" PRIu64 ": %d ops saved to Table #%" PRIu64 " %s",
                      log, counter, meta.fd.GetNumber(),
@@ -667,8 +670,14 @@ Status RepairDB(const std::string& dbname, const DBOptions& db_options,
 }
 
 Status RepairDB(const std::string& dbname, const Options& options) {
-  DBOptions db_options(options);
-  ColumnFamilyOptions cf_options(options);
+  Options opts(options);
+  if (opts.file_system == nullptr) {
+    opts.file_system.reset(new LegacyFileSystemWrapper(opts.env));
+    ;
+  }
+
+  DBOptions db_options(opts);
+  ColumnFamilyOptions cf_options(opts);
   Repairer repairer(dbname, db_options,
                     {}, cf_options /* default_cf_opts */,
                     cf_options /* unknown_cf_opts */,

db/repair_test.cc

@@ -74,7 +74,9 @@ TEST_F(RepairTest, CorruptManifest) {
 
   Close();
   ASSERT_OK(env_->FileExists(manifest_path));
-  CreateFile(env_, manifest_path, "blah", false /* use_fsync */);
+
+  LegacyFileSystemWrapper fs(env_);
+  CreateFile(&fs, manifest_path, "blah", false /* use_fsync */);
   ASSERT_OK(RepairDB(dbname_, CurrentOptions()));
   Reopen(CurrentOptions());
 
@@ -153,7 +155,9 @@ TEST_F(RepairTest, CorruptSst) {
   Flush();
   auto sst_path = GetFirstSstPath();
   ASSERT_FALSE(sst_path.empty());
-  CreateFile(env_, sst_path, "blah", false /* use_fsync */);
+
+  LegacyFileSystemWrapper fs(env_);
+  CreateFile(&fs, sst_path, "blah", false /* use_fsync */);
 
   Close();
   ASSERT_OK(RepairDB(dbname_, CurrentOptions()));

db/snapshot_impl.h

@@ -148,6 +148,14 @@ class SnapshotList {
     }
   }
 
+  int64_t GetOldestSnapshotSequence() const {
+    if (empty()) {
+      return 0;
+    } else {
+      return oldest()->GetSequenceNumber();
+    }
+  }
+
   uint64_t count() const { return count_; }
 
  private:

db/table_cache.cc

@@ -63,10 +63,10 @@ void AppendVarint64(IterKey* key, uint64_t v) {
 }  // namespace
 
 TableCache::TableCache(const ImmutableCFOptions& ioptions,
-                       const EnvOptions& env_options, Cache* const cache,
+                       const FileOptions& file_options, Cache* const cache,
                        BlockCacheTracer* const block_cache_tracer)
     : ioptions_(ioptions),
-      env_options_(env_options),
+      file_options_(file_options),
       cache_(cache),
       immortal_tables_(false),
       block_cache_tracer_(block_cache_tracer) {
@@ -89,7 +89,7 @@ void TableCache::ReleaseHandle(Cache::Handle* handle) {
 }
 
 Status TableCache::GetTableReader(
-    const EnvOptions& env_options,
+    const FileOptions& file_options,
     const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
     bool sequential_mode, bool record_read_stats, HistogramImpl* file_read_hist,
     std::unique_ptr<TableReader>* table_reader,
@@ -97,13 +97,14 @@ Status TableCache::GetTableReader(
     bool prefetch_index_and_filter_in_cache) {
   std::string fname =
       TableFileName(ioptions_.cf_paths, fd.GetNumber(), fd.GetPathId());
-  std::unique_ptr<RandomAccessFile> file;
-  Status s = ioptions_.env->NewRandomAccessFile(fname, &file, env_options);
+  std::unique_ptr<FSRandomAccessFile> file;
+  Status s = ioptions_.fs->NewRandomAccessFile(fname, file_options, &file,
+                                               nullptr);
 
   RecordTick(ioptions_.statistics, NO_FILE_OPENS);
   if (s.ok()) {
     if (!sequential_mode && ioptions_.advise_random_on_open) {
-      file->Hint(RandomAccessFile::RANDOM);
+      file->Hint(FSRandomAccessFile::kRandom);
     }
     StopWatch sw(ioptions_.env, ioptions_.statistics, TABLE_OPEN_IO_MICROS);
     std::unique_ptr<RandomAccessFileReader> file_reader(
@@ -112,7 +113,7 @@ Status TableCache::GetTableReader(
             record_read_stats ? ioptions_.statistics : nullptr, SST_READ_MICROS,
             file_read_hist, ioptions_.rate_limiter, ioptions_.listeners));
     s = ioptions_.table_factory->NewTableReader(
-        TableReaderOptions(ioptions_, prefix_extractor, env_options,
+        TableReaderOptions(ioptions_, prefix_extractor, file_options,
                            internal_comparator, skip_filters, immortal_tables_,
                            level, fd.largest_seqno, block_cache_tracer_),
         std::move(file_reader), fd.GetFileSize(), table_reader,
@@ -129,7 +130,7 @@ void TableCache::EraseHandle(const FileDescriptor& fd, Cache::Handle* handle) {
   cache_->Erase(key);
 }
 
-Status TableCache::FindTable(const EnvOptions& env_options,
+Status TableCache::FindTable(const FileOptions& file_options,
                              const InternalKeyComparator& internal_comparator,
                              const FileDescriptor& fd, Cache::Handle** handle,
                              const SliceTransform* prefix_extractor,
@@ -150,7 +151,7 @@ Status TableCache::FindTable(const EnvOptions& env_options,
       return Status::Incomplete("Table not found in table_cache, no_io is set");
     }
     std::unique_ptr<TableReader> table_reader;
-    s = GetTableReader(env_options, internal_comparator, fd,
+    s = GetTableReader(file_options, internal_comparator, fd,
                        false /* sequential mode */, record_read_stats,
                        file_read_hist, &table_reader, prefix_extractor,
                        skip_filters, level, prefetch_index_and_filter_in_cache);
@@ -172,7 +173,7 @@ Status TableCache::FindTable(const EnvOptions& env_options,
 }
 
 InternalIterator* TableCache::NewIterator(
-    const ReadOptions& options, const EnvOptions& env_options,
+    const ReadOptions& options, const FileOptions& file_options,
     const InternalKeyComparator& icomparator, const FileMetaData& file_meta,
     RangeDelAggregator* range_del_agg, const SliceTransform* prefix_extractor,
     TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
@@ -191,7 +192,7 @@ InternalIterator* TableCache::NewIterator(
   auto& fd = file_meta.fd;
   table_reader = fd.table_reader;
   if (table_reader == nullptr) {
-    s = FindTable(env_options, icomparator, fd, &handle, prefix_extractor,
+    s = FindTable(file_options, icomparator, fd, &handle, prefix_extractor,
                   options.read_tier == kBlockCacheTier /* no_io */,
                   !for_compaction /* record_read_stats */, file_read_hist,
                   skip_filters, level);
@@ -206,8 +207,8 @@ InternalIterator* TableCache::NewIterator(
       result = NewEmptyInternalIterator<Slice>(arena);
     } else {
       result = table_reader->NewIterator(options, prefix_extractor, arena,
-                                         skip_filters, caller,
-                                         env_options.compaction_readahead_size);
+                                   skip_filters, caller,
+                                   file_options.compaction_readahead_size);
     }
     if (handle != nullptr) {
       result->RegisterCleanup(&UnrefEntry, cache_, handle);
@@ -264,7 +265,7 @@ Status TableCache::GetRangeTombstoneIterator(
   TableReader* t = fd.table_reader;
   Cache::Handle* handle = nullptr;
   if (t == nullptr) {
-    s = FindTable(env_options_, internal_comparator, fd, &handle);
+    s = FindTable(file_options_, internal_comparator, fd, &handle);
     if (s.ok()) {
       t = GetTableReaderFromHandle(handle);
     }
@@ -381,7 +382,7 @@ Status TableCache::Get(const ReadOptions& options,
   if (!done && s.ok()) {
     if (t == nullptr) {
       s = FindTable(
-          env_options_, internal_comparator, fd, &handle, prefix_extractor,
+          file_options_, internal_comparator, fd, &handle, prefix_extractor,
           options.read_tier == kBlockCacheTier /* no_io */,
           true /* record_read_stats */, file_read_hist, skip_filters, level);
       if (s.ok()) {
@@ -481,7 +482,7 @@ Status TableCache::MultiGet(const ReadOptions& options,
   if (s.ok() && !table_range.empty()) {
     if (t == nullptr) {
       s = FindTable(
-          env_options_, internal_comparator, fd, &handle, prefix_extractor,
+          file_options_, internal_comparator, fd, &handle, prefix_extractor,
           options.read_tier == kBlockCacheTier /* no_io */,
           true /* record_read_stats */, file_read_hist, skip_filters, level);
       if (s.ok()) {
@@ -551,7 +552,7 @@ Status TableCache::MultiGet(const ReadOptions& options,
 }
 
 Status TableCache::GetTableProperties(
-    const EnvOptions& env_options,
+    const FileOptions& file_options,
     const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
     std::shared_ptr<const TableProperties>* properties,
     const SliceTransform* prefix_extractor, bool no_io) {
@@ -565,7 +566,7 @@ Status TableCache::GetTableProperties(
   }
 
   Cache::Handle* table_handle = nullptr;
-  s = FindTable(env_options, internal_comparator, fd, &table_handle,
+  s = FindTable(file_options, internal_comparator, fd, &table_handle,
                 prefix_extractor, no_io);
   if (!s.ok()) {
     return s;
@@ -578,7 +579,7 @@ Status TableCache::GetTableProperties(
 }
 
 size_t TableCache::GetMemoryUsageByTableReader(
-    const EnvOptions& env_options,
+    const FileOptions& file_options,
     const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
     const SliceTransform* prefix_extractor) {
   Status s;
@@ -589,7 +590,7 @@ size_t TableCache::GetMemoryUsageByTableReader(
   }
 
   Cache::Handle* table_handle = nullptr;
-  s = FindTable(env_options, internal_comparator, fd, &table_handle,
+  s = FindTable(file_options, internal_comparator, fd, &table_handle,
                 prefix_extractor, true);
   if (!s.ok()) {
     return 0;
@@ -614,7 +615,7 @@ uint64_t TableCache::ApproximateOffsetOf(
   Cache::Handle* table_handle = nullptr;
   if (table_reader == nullptr) {
     const bool for_compaction = (caller == TableReaderCaller::kCompaction);
-    Status s = FindTable(env_options_, internal_comparator, fd, &table_handle,
+    Status s = FindTable(file_options_, internal_comparator, fd, &table_handle,
                          prefix_extractor, false /* no_io */,
                          !for_compaction /* record_read_stats */);
     if (s.ok()) {
@@ -641,7 +642,7 @@ uint64_t TableCache::ApproximateSize(
   Cache::Handle* table_handle = nullptr;
   if (table_reader == nullptr) {
     const bool for_compaction = (caller == TableReaderCaller::kCompaction);
-    Status s = FindTable(env_options_, internal_comparator, fd, &table_handle,
+    Status s = FindTable(file_options_, internal_comparator, fd, &table_handle,
                          prefix_extractor, false /* no_io */,
                          !for_compaction /* record_read_stats */);
     if (s.ok()) {

db/table_cache.h

@@ -49,7 +49,7 @@ class HistogramImpl;
 class TableCache {
  public:
   TableCache(const ImmutableCFOptions& ioptions,
-             const EnvOptions& storage_options, Cache* cache,
+             const FileOptions& storage_options, Cache* cache,
              BlockCacheTracer* const block_cache_tracer);
   ~TableCache();
 
@@ -67,7 +67,7 @@ class TableCache {
   // @param skip_filters Disables loading/accessing the filter block
   // @param level The level this table is at, -1 for "not set / don't know"
   InternalIterator* NewIterator(
-      const ReadOptions& options, const EnvOptions& toptions,
+      const ReadOptions& options, const FileOptions& toptions,
       const InternalKeyComparator& internal_comparator,
       const FileMetaData& file_meta, RangeDelAggregator* range_del_agg,
       const SliceTransform* prefix_extractor, TableReader** table_reader_ptr,
@@ -128,7 +128,7 @@ class TableCache {
   // Find table reader
   // @param skip_filters Disables loading/accessing the filter block
   // @param level == -1 means not specified
-  Status FindTable(const EnvOptions& toptions,
+  Status FindTable(const FileOptions& toptions,
                    const InternalKeyComparator& internal_comparator,
                    const FileDescriptor& file_fd, Cache::Handle**,
                    const SliceTransform* prefix_extractor = nullptr,
@@ -146,7 +146,7 @@ class TableCache {
   // @returns: `properties` will be reset on success. Please note that we will
   //            return Status::Incomplete() if table is not present in cache and
   //            we set `no_io` to be true.
-  Status GetTableProperties(const EnvOptions& toptions,
+  Status GetTableProperties(const FileOptions& toptions,
                             const InternalKeyComparator& internal_comparator,
                             const FileDescriptor& file_meta,
                             std::shared_ptr<const TableProperties>* properties,
@@ -156,7 +156,7 @@ class TableCache {
   // Return total memory usage of the table reader of the file.
   // 0 if table reader of the file is not loaded.
   size_t GetMemoryUsageByTableReader(
-      const EnvOptions& toptions,
+      const FileOptions& toptions,
       const InternalKeyComparator& internal_comparator,
       const FileDescriptor& fd,
       const SliceTransform* prefix_extractor = nullptr);
@@ -193,7 +193,7 @@ class TableCache {
 
  private:
   // Build a table reader
-  Status GetTableReader(const EnvOptions& env_options,
+  Status GetTableReader(const FileOptions& file_options,
                         const InternalKeyComparator& internal_comparator,
                         const FileDescriptor& fd, bool sequential_mode,
                         bool record_read_stats, HistogramImpl* file_read_hist,
@@ -216,7 +216,7 @@ class TableCache {
                        size_t prefix_size, GetContext* get_context);
 
   const ImmutableCFOptions& ioptions_;
-  const EnvOptions& env_options_;
+  const FileOptions& file_options_;
   Cache* const cache_;
   std::string row_cache_id_;
   bool immortal_tables_;

db/table_properties_collector_test.cc

@@ -12,6 +12,7 @@
 #include "db/db_impl/db_impl.h"
 #include "db/dbformat.h"
 #include "db/table_properties_collector.h"
+#include "env/composite_env_wrapper.h"
 #include "file/sequence_file_reader.h"
 #include "file/writable_file_writer.h"
 #include "options/cf_options.h"
@@ -48,7 +49,8 @@ void MakeBuilder(const Options& options, const ImmutableCFOptions& ioptions,
                  std::unique_ptr<TableBuilder>* builder) {
   std::unique_ptr<WritableFile> wf(new test::StringSink);
   writable->reset(
-      new WritableFileWriter(std::move(wf), "" /* don't care */, EnvOptions()));
+      new WritableFileWriter(NewLegacyWritableFileWrapper(std::move(wf)),
+                             "" /* don't care */, EnvOptions()));
   int unknown_level = -1;
   builder->reset(NewTableBuilder(
       ioptions, moptions, internal_comparator, int_tbl_prop_collector_factories,
@@ -282,8 +284,9 @@ void TestCustomizedTablePropertiesCollector(
   writer->Flush();
 
   // -- Step 2: Read properties
-  test::StringSink* fwf =
-      static_cast<test::StringSink*>(writer->writable_file());
+  LegacyWritableFileWrapper* file =
+      static_cast<LegacyWritableFileWrapper*>(writer->writable_file());
+  test::StringSink* fwf = static_cast<test::StringSink*>(file->target());
   std::unique_ptr<RandomAccessFileReader> fake_file_reader(
       test::GetRandomAccessFileReader(
           new test::StringSource(fwf->contents())));
@@ -422,8 +425,9 @@ void TestInternalKeyPropertiesCollector(
     ASSERT_OK(builder->Finish());
     writable->Flush();
 
-    test::StringSink* fwf =
-        static_cast<test::StringSink*>(writable->writable_file());
+    LegacyWritableFileWrapper* file =
+        static_cast<LegacyWritableFileWrapper*>(writable->writable_file());
+    test::StringSink* fwf = static_cast<test::StringSink*>(file->target());
     std::unique_ptr<RandomAccessFileReader> reader(
         test::GetRandomAccessFileReader(
             new test::StringSource(fwf->contents())));

db/transaction_log_impl.cc

@@ -42,22 +42,23 @@ TransactionLogIteratorImpl::TransactionLogIteratorImpl(
 Status TransactionLogIteratorImpl::OpenLogFile(
     const LogFile* log_file,
     std::unique_ptr<SequentialFileReader>* file_reader) {
-  Env* env = options_->env;
-  std::unique_ptr<SequentialFile> file;
+  FileSystem* fs = options_->fs.get();
+  std::unique_ptr<FSSequentialFile> file;
   std::string fname;
   Status s;
-  EnvOptions optimized_env_options = env->OptimizeForLogRead(soptions_);
+  EnvOptions optimized_env_options = fs->OptimizeForLogRead(soptions_);
   if (log_file->Type() == kArchivedLogFile) {
     fname = ArchivedLogFileName(dir_, log_file->LogNumber());
-    s = env->NewSequentialFile(fname, &file, optimized_env_options);
+    s = fs->NewSequentialFile(fname, optimized_env_options, &file, nullptr);
   } else {
     fname = LogFileName(dir_, log_file->LogNumber());
-    s = env->NewSequentialFile(fname, &file, optimized_env_options);
+    s = fs->NewSequentialFile(fname, optimized_env_options, &file, nullptr);
     if (!s.ok()) {
       //  If cannot open file in DB directory.
       //  Try the archive dir, as it could have moved in the meanwhile.
       fname = ArchivedLogFileName(dir_, log_file->LogNumber());
-      s = env->NewSequentialFile(fname, &file, optimized_env_options);
+      s = fs->NewSequentialFile(fname, optimized_env_options,
+                                &file, nullptr);
     }
   }
   if (s.ok()) {

db/trim_history_scheduler.cc

@@ -34,10 +34,7 @@ ColumnFamilyData* TrimHistoryScheduler::TakeNextColumnFamily() {
       // success
       return cfd;
     }
-    if (cfd->Unref()) {
-      // no longer relevant, retry
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
 }
 
@@ -49,9 +46,7 @@ bool TrimHistoryScheduler::Empty() {
 void TrimHistoryScheduler::Clear() {
   ColumnFamilyData* cfd;
   while ((cfd = TakeNextColumnFamily()) != nullptr) {
-    if (cfd->Unref()) {
-      delete cfd;
-    }
+    cfd->UnrefAndTryDelete();
   }
   assert(Empty());
 }

db/version_builder.cc

@@ -83,7 +83,7 @@ class VersionBuilder::Rep {
     std::unordered_map<uint64_t, FileMetaData*> added_files;
   };
 
-  const EnvOptions& env_options_;
+  const FileOptions& file_options_;
   Logger* info_log_;
   TableCache* table_cache_;
   VersionStorageInfo* base_vstorage_;
@@ -101,9 +101,10 @@ class VersionBuilder::Rep {
   FileComparator level_nonzero_cmp_;
 
  public:
-  Rep(const EnvOptions& env_options, Logger* info_log, TableCache* table_cache,
+  Rep(const FileOptions& file_options, Logger* info_log,
+      TableCache* table_cache,
       VersionStorageInfo* base_vstorage)
-      : env_options_(env_options),
+      : file_options_(file_options),
         info_log_(info_log),
         table_cache_(table_cache),
         base_vstorage_(base_vstorage),
@@ -461,7 +462,7 @@ class VersionBuilder::Rep {
         auto* file_meta = files_meta[file_idx].first;
         int level = files_meta[file_idx].second;
         statuses[file_idx] = table_cache_->FindTable(
-            env_options_, *(base_vstorage_->InternalComparator()),
+            file_options_, *(base_vstorage_->InternalComparator()),
             file_meta->fd, &file_meta->table_reader_handle, prefix_extractor,
             false /*no_io */, true /* record_read_stats */,
             internal_stats->GetFileReadHist(level), false, level,
@@ -500,11 +501,11 @@ class VersionBuilder::Rep {
   }
 };
 
-VersionBuilder::VersionBuilder(const EnvOptions& env_options,
+VersionBuilder::VersionBuilder(const FileOptions& file_options,
                                TableCache* table_cache,
                                VersionStorageInfo* base_vstorage,
                                Logger* info_log)
-    : rep_(new Rep(env_options, info_log, table_cache, base_vstorage)) {}
+    : rep_(new Rep(file_options, info_log, table_cache, base_vstorage)) {}
 
 VersionBuilder::~VersionBuilder() { delete rep_; }
 

db/version_builder.h

@@ -8,7 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 //
 #pragma once
-#include "rocksdb/env.h"
+#include "rocksdb/file_system.h"
 #include "rocksdb/slice_transform.h"
 
 namespace rocksdb {
@@ -24,7 +24,7 @@ class InternalStats;
 // Versions that contain full copies of the intermediate state.
 class VersionBuilder {
  public:
-  VersionBuilder(const EnvOptions& env_options, TableCache* table_cache,
+  VersionBuilder(const FileOptions& file_options, TableCache* table_cache,
                  VersionStorageInfo* base_vstorage, Logger* info_log = nullptr);
   ~VersionBuilder();
   Status CheckConsistency(VersionStorageInfo* vstorage);

db/version_set.cc

@@ -853,7 +853,7 @@ namespace {
 class LevelIterator final : public InternalIterator {
  public:
   LevelIterator(TableCache* table_cache, const ReadOptions& read_options,
-                const EnvOptions& env_options,
+                const FileOptions& file_options,
                 const InternalKeyComparator& icomparator,
                 const LevelFilesBrief* flevel,
                 const SliceTransform* prefix_extractor, bool should_sample,
@@ -863,7 +863,7 @@ class LevelIterator final : public InternalIterator {
                     compaction_boundaries = nullptr)
       : table_cache_(table_cache),
         read_options_(read_options),
-        env_options_(env_options),
+        file_options_(file_options),
         icomparator_(icomparator),
         user_comparator_(icomparator.user_comparator()),
         flevel_(flevel),
@@ -975,7 +975,7 @@ class LevelIterator final : public InternalIterator {
     }
     CheckMayBeOutOfLowerBound();
     return table_cache_->NewIterator(
-        read_options_, env_options_, icomparator_, *file_meta.file_metadata,
+        read_options_, file_options_, icomparator_, *file_meta.file_metadata,
         range_del_agg_, prefix_extractor_,
         nullptr /* don't need reference to table */, file_read_hist_, caller_,
         /*arena=*/nullptr, skip_filters_, level_, smallest_compaction_key,
@@ -998,7 +998,7 @@ class LevelIterator final : public InternalIterator {
 
   TableCache* table_cache_;
   const ReadOptions read_options_;
-  const EnvOptions& env_options_;
+  const FileOptions& file_options_;
   const InternalKeyComparator& icomparator_;
   const UserComparatorWrapper user_comparator_;
   const LevelFilesBrief* flevel_;
@@ -1206,7 +1206,7 @@ class BaseReferencedVersionBuilder {
  public:
   explicit BaseReferencedVersionBuilder(ColumnFamilyData* cfd)
       : version_builder_(new VersionBuilder(
-            cfd->current()->version_set()->env_options(), cfd->table_cache(),
+            cfd->current()->version_set()->file_options(), cfd->table_cache(),
             cfd->current()->storage_info(), cfd->ioptions()->info_log)),
         version_(cfd->current()) {
     version_->Ref();
@@ -1227,7 +1227,7 @@ Status Version::GetTableProperties(std::shared_ptr<const TableProperties>* tp,
   auto table_cache = cfd_->table_cache();
   auto ioptions = cfd_->ioptions();
   Status s = table_cache->GetTableProperties(
-      env_options_, cfd_->internal_comparator(), file_meta->fd, tp,
+      file_options_, cfd_->internal_comparator(), file_meta->fd, tp,
       mutable_cf_options_.prefix_extractor.get(), true /* no io */);
   if (s.ok()) {
     return s;
@@ -1241,7 +1241,7 @@ Status Version::GetTableProperties(std::shared_ptr<const TableProperties>* tp,
 
   // 2. Table is not present in table cache, we'll read the table properties
   // directly from the properties block in the file.
-  std::unique_ptr<RandomAccessFile> file;
+  std::unique_ptr<FSRandomAccessFile> file;
   std::string file_name;
   if (fname != nullptr) {
     file_name = *fname;
@@ -1250,7 +1250,8 @@ Status Version::GetTableProperties(std::shared_ptr<const TableProperties>* tp,
       TableFileName(ioptions->cf_paths, file_meta->fd.GetNumber(),
                     file_meta->fd.GetPathId());
   }
-  s = ioptions->env->NewRandomAccessFile(file_name, &file, env_options_);
+  s = ioptions->fs->NewRandomAccessFile(file_name, file_options_, &file,
+                                        nullptr);
   if (!s.ok()) {
     return s;
   }
@@ -1420,7 +1421,7 @@ size_t Version::GetMemoryUsageByTableReaders() {
   for (auto& file_level : storage_info_.level_files_brief_) {
     for (size_t i = 0; i < file_level.num_files; i++) {
       total_usage += cfd_->table_cache()->GetMemoryUsageByTableReader(
-          env_options_, cfd_->internal_comparator(), file_level.files[i].fd,
+          file_options_, cfd_->internal_comparator(), file_level.files[i].fd,
           mutable_cf_options_.prefix_extractor.get());
     }
   }
@@ -1547,7 +1548,7 @@ double VersionStorageInfo::GetEstimatedCompressionRatioAtLevel(
 }
 
 void Version::AddIterators(const ReadOptions& read_options,
-                           const EnvOptions& soptions,
+                           const FileOptions& soptions,
                            MergeIteratorBuilder* merge_iter_builder,
                            RangeDelAggregator* range_del_agg) {
   assert(storage_info_.finalized_);
@@ -1559,7 +1560,7 @@ void Version::AddIterators(const ReadOptions& read_options,
 }
 
 void Version::AddIteratorsForLevel(const ReadOptions& read_options,
-                                   const EnvOptions& soptions,
+                                   const FileOptions& soptions,
                                    MergeIteratorBuilder* merge_iter_builder,
                                    int level,
                                    RangeDelAggregator* range_del_agg) {
@@ -1614,7 +1615,7 @@ void Version::AddIteratorsForLevel(const ReadOptions& read_options,
 }
 
 Status Version::OverlapWithLevelIterator(const ReadOptions& read_options,
-                                         const EnvOptions& env_options,
+                                         const FileOptions& file_options,
                                          const Slice& smallest_user_key,
                                          const Slice& largest_user_key,
                                          int level, bool* overlap) {
@@ -1638,7 +1639,7 @@ Status Version::OverlapWithLevelIterator(const ReadOptions& read_options,
         continue;
       }
       ScopedArenaIterator iter(cfd_->table_cache()->NewIterator(
-          read_options, env_options, cfd_->internal_comparator(),
+          read_options, file_options, cfd_->internal_comparator(),
           *file->file_metadata, &range_del_agg,
           mutable_cf_options_.prefix_extractor.get(), nullptr,
           cfd_->internal_stats()->GetFileReadHist(0),
@@ -1655,7 +1656,7 @@ Status Version::OverlapWithLevelIterator(const ReadOptions& read_options,
   } else if (storage_info_.LevelFilesBrief(level).num_files > 0) {
     auto mem = arena.AllocateAligned(sizeof(LevelIterator));
     ScopedArenaIterator iter(new (mem) LevelIterator(
-        cfd_->table_cache(), read_options, env_options,
+        cfd_->table_cache(), read_options, file_options,
         cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level),
         mutable_cf_options_.prefix_extractor.get(), should_sample_file_read(),
         cfd_->internal_stats()->GetFileReadHist(level),
@@ -1719,7 +1720,7 @@ VersionStorageInfo::VersionStorageInfo(
 }
 
 Version::Version(ColumnFamilyData* column_family_data, VersionSet* vset,
-                 const EnvOptions& env_opt,
+                 const FileOptions& file_opt,
                  const MutableCFOptions mutable_cf_options,
                  uint64_t version_number)
     : env_(vset->env_),
@@ -1744,7 +1745,7 @@ Version::Version(ColumnFamilyData* column_family_data, VersionSet* vset,
       next_(this),
       prev_(this),
       refs_(0),
-      env_options_(env_opt),
+      file_options_(file_opt),
       mutable_cf_options_(mutable_cf_options),
       version_number_(version_number) {}
 
@@ -3509,7 +3510,7 @@ void AtomicGroupReadBuffer::Clear() {
 
 VersionSet::VersionSet(const std::string& dbname,
                        const ImmutableDBOptions* _db_options,
-                       const EnvOptions& storage_options, Cache* table_cache,
+                       const FileOptions& storage_options, Cache* table_cache,
                        WriteBufferManager* write_buffer_manager,
                        WriteController* write_controller,
                        BlockCacheTracer* const block_cache_tracer)
@@ -3517,6 +3518,7 @@ VersionSet::VersionSet(const std::string& dbname,
           dbname, _db_options, storage_options, table_cache,
           write_buffer_manager, write_controller, block_cache_tracer)),
       env_(_db_options->env),
+      fs_(_db_options->fs.get()),
       dbname_(dbname),
       db_options_(_db_options),
       next_file_number_(2),
@@ -3529,7 +3531,7 @@ VersionSet::VersionSet(const std::string& dbname,
       prev_log_number_(0),
       current_version_number_(0),
       manifest_file_size_(0),
-      env_options_(storage_options),
+      file_options_(storage_options),
       block_cache_tracer_(block_cache_tracer) {}
 
 VersionSet::~VersionSet() {
@@ -3655,7 +3657,7 @@ Status VersionSet::ProcessManifestWrites(
         }
       }
       if (version == nullptr) {
-        version = new Version(last_writer->cfd, this, env_options_,
+        version = new Version(last_writer->cfd, this, file_options_,
                               last_writer->mutable_cf_options,
                               current_version_number_++);
         versions.push_back(version);
@@ -3751,6 +3753,10 @@ Status VersionSet::ProcessManifestWrites(
     pending_manifest_file_number_ = manifest_file_number_;
   }
 
+  // Local cached copy of state variable(s). WriteCurrentStateToManifest()
+  // reads its content after releasing db mutex to avoid race with
+  // SwitchMemtable().
+  std::unordered_map<uint32_t, MutableCFState> curr_state;
   if (new_descriptor_log) {
     // if we are writing out new snapshot make sure to persist max column
     // family.
@@ -3758,10 +3764,14 @@ Status VersionSet::ProcessManifestWrites(
       first_writer.edit_list.front()->SetMaxColumnFamily(
           column_family_set_->GetMaxColumnFamily());
     }
+    for (const auto* cfd : *column_family_set_) {
+      assert(curr_state.find(cfd->GetID()) == curr_state.end());
+      curr_state[cfd->GetID()] = {cfd->GetLogNumber()};
+    }
   }
 
   {
-    EnvOptions opt_env_opts = env_->OptimizeForManifestWrite(env_options_);
+    FileOptions opt_file_opts = fs_->OptimizeForManifestWrite(file_options_);
     mu->Unlock();
 
     TEST_SYNC_POINT("VersionSet::LogAndApply:WriteManifest");
@@ -3788,19 +3798,19 @@ Status VersionSet::ProcessManifestWrites(
                      pending_manifest_file_number_);
       std::string descriptor_fname =
           DescriptorFileName(dbname_, pending_manifest_file_number_);
-      std::unique_ptr<WritableFile> descriptor_file;
-      s = NewWritableFile(env_, descriptor_fname, &descriptor_file,
-                          opt_env_opts);
+      std::unique_ptr<FSWritableFile> descriptor_file;
+      s = NewWritableFile(fs_, descriptor_fname, &descriptor_file,
+                          opt_file_opts);
       if (s.ok()) {
         descriptor_file->SetPreallocationBlockSize(
             db_options_->manifest_preallocation_size);
 
         std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
-            std::move(descriptor_file), descriptor_fname, opt_env_opts, env_,
+            std::move(descriptor_file), descriptor_fname, opt_file_opts, env_,
             nullptr, db_options_->listeners));
         descriptor_log_.reset(
             new log::Writer(std::move(file_writer), 0, false));
-        s = WriteCurrentStateToManifest(descriptor_log_.get());
+        s = WriteCurrentStateToManifest(curr_state, descriptor_log_.get());
       }
     }
 
@@ -3888,9 +3898,7 @@ Status VersionSet::ProcessManifestWrites(
     } else if (first_writer.edit_list.front()->is_column_family_drop_) {
       assert(batch_edits.size() == 1);
       first_writer.cfd->SetDropped();
-      if (first_writer.cfd->Unref()) {
-        delete first_writer.cfd;
-      }
+      first_writer.cfd->UnrefAndTryDelete();
     } else {
       // Each version in versions corresponds to a column family.
       // For each column family, update its log number indicating that logs
@@ -3942,15 +3950,12 @@ Status VersionSet::ProcessManifestWrites(
     for (auto v : versions) {
       delete v;
     }
-    // If manifest append failed for whatever reason, the file could be
-    // corrupted. So we need to force the next version update to start a
-    // new manifest file.
-    descriptor_log_.reset();
     if (new_descriptor_log) {
       ROCKS_LOG_INFO(db_options_->info_log,
                      "Deleting manifest %" PRIu64 " current manifest %" PRIu64
                      "\n",
                      manifest_file_number_, pending_manifest_file_number_);
+      descriptor_log_.reset();
       env_->DeleteFile(
           DescriptorFileName(dbname_, pending_manifest_file_number_));
     }
@@ -4171,8 +4176,7 @@ Status VersionSet::ApplyOneVersionEditToBuilder(
       builders.erase(builder);
       cfd = column_family_set_->GetColumnFamily(edit.column_family_);
       assert(cfd != nullptr);
-      if (cfd->Unref()) {
-        delete cfd;
+      if (cfd->UnrefAndTryDelete()) {
         cfd = nullptr;
       } else {
         // who else can have reference to cfd!?
@@ -4257,15 +4261,16 @@ Status VersionSet::ExtractInfoFromVersionEdit(
   return Status::OK();
 }
 
-Status VersionSet::GetCurrentManifestPath(const std::string& dbname, Env* env,
+Status VersionSet::GetCurrentManifestPath(const std::string& dbname,
+                                          FileSystem* fs,
                                           std::string* manifest_path,
                                           uint64_t* manifest_file_number) {
-  assert(env != nullptr);
+  assert(fs != nullptr);
   assert(manifest_path != nullptr);
   assert(manifest_file_number != nullptr);
 
   std::string fname;
-  Status s = ReadFileToString(env, CurrentFileName(dbname), &fname);
+  Status s = ReadFileToString(fs, CurrentFileName(dbname), &fname);
   if (!s.ok()) {
     return s;
   }
@@ -4367,7 +4372,7 @@ Status VersionSet::Recover(
 
   // Read "CURRENT" file, which contains a pointer to the current manifest file
   std::string manifest_path;
-  Status s = GetCurrentManifestPath(dbname_, env_, &manifest_path,
+  Status s = GetCurrentManifestPath(dbname_, fs_, &manifest_path,
                                     &manifest_file_number_);
   if (!s.ok()) {
     return s;
@@ -4378,9 +4383,10 @@ Status VersionSet::Recover(
 
   std::unique_ptr<SequentialFileReader> manifest_file_reader;
   {
-    std::unique_ptr<SequentialFile> manifest_file;
-    s = env_->NewSequentialFile(manifest_path, &manifest_file,
-                                env_->OptimizeForManifestRead(env_options_));
+    std::unique_ptr<FSSequentialFile> manifest_file;
+    s = fs_->NewSequentialFile(manifest_path,
+                               fs_->OptimizeForManifestRead(file_options_),
+                               &manifest_file, nullptr);
     if (!s.ok()) {
       return s;
     }
@@ -4389,7 +4395,8 @@ Status VersionSet::Recover(
                                  db_options_->log_readahead_size));
   }
   uint64_t current_manifest_file_size;
-  s = env_->GetFileSize(manifest_path, &current_manifest_file_size);
+  s = fs_->GetFileSize(manifest_path, IOOptions(), &current_manifest_file_size,
+                       nullptr);
   if (!s.ok()) {
     return s;
   }
@@ -4496,7 +4503,7 @@ Status VersionSet::Recover(
           true /* is_initial_load */,
           cfd->GetLatestMutableCFOptions()->prefix_extractor.get());
 
-      Version* v = new Version(cfd, this, env_options_,
+      Version* v = new Version(cfd, this, file_options_,
                                *cfd->GetLatestMutableCFOptions(),
                                current_version_number_++);
       builder->SaveTo(v->storage_info());
@@ -4541,23 +4548,24 @@ Status VersionSet::Recover(
 }
 
 Status VersionSet::ListColumnFamilies(std::vector<std::string>* column_families,
-                                      const std::string& dbname, Env* env) {
+                                      const std::string& dbname,
+                                      FileSystem* fs) {
   // these are just for performance reasons, not correcntes,
   // so we're fine using the defaults
-  EnvOptions soptions;
+  FileOptions soptions;
   // Read "CURRENT" file, which contains a pointer to the current manifest file
   std::string manifest_path;
   uint64_t manifest_file_number;
-  Status s = GetCurrentManifestPath(dbname, env, &manifest_path,
-                                    &manifest_file_number);
+  Status s =
+      GetCurrentManifestPath(dbname, fs, &manifest_path, &manifest_file_number);
   if (!s.ok()) {
     return s;
   }
 
   std::unique_ptr<SequentialFileReader> file_reader;
   {
-    std::unique_ptr<SequentialFile> file;
-    s = env->NewSequentialFile(manifest_path, &file, soptions);
+    std::unique_ptr<FSSequentialFile> file;
+    s = fs->NewSequentialFile(manifest_path, soptions, &file, nullptr);
     if (!s.ok()) {
       return s;
   }
@@ -4611,7 +4619,7 @@ Status VersionSet::ListColumnFamilies(std::vector<std::string>* column_families,
 #ifndef ROCKSDB_LITE
 Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
                                         const Options* options,
-                                        const EnvOptions& env_options,
+                                        const FileOptions& file_options,
                                         int new_levels) {
   if (new_levels <= 1) {
     return Status::InvalidArgument(
@@ -4624,7 +4632,7 @@ Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
                                         options->table_cache_numshardbits));
   WriteController wc(options->delayed_write_rate);
   WriteBufferManager wb(options->db_write_buffer_size);
-  VersionSet versions(dbname, &db_options, env_options, tc.get(), &wb, &wc,
+  VersionSet versions(dbname, &db_options, file_options, tc.get(), &wb, &wc,
                       /*block_cache_tracer=*/nullptr);
   Status status;
 
@@ -4700,9 +4708,11 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
   std::unique_ptr<SequentialFileReader> file_reader;
   Status s;
   {
-    std::unique_ptr<SequentialFile> file;
-    s = options.env->NewSequentialFile(
-        dscname, &file, env_->OptimizeForManifestRead(env_options_));
+    std::unique_ptr<FSSequentialFile> file;
+    s = options.file_system->NewSequentialFile(
+        dscname,
+        options.file_system->OptimizeForManifestRead(file_options_), &file,
+        nullptr);
     if (!s.ok()) {
       return s;
     }
@@ -4784,8 +4794,7 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
         comparators.erase(edit.column_family_);
         cfd = column_family_set_->GetColumnFamily(edit.column_family_);
         assert(cfd != nullptr);
-        cfd->Unref();
-        delete cfd;
+        cfd->UnrefAndTryDelete();
         cfd = nullptr;
       } else {
         if (!cf_in_builders) {
@@ -4863,7 +4872,7 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
       assert(builders_iter != builders.end());
       auto builder = builders_iter->second->version_builder();
 
-      Version* v = new Version(cfd, this, env_options_,
+      Version* v = new Version(cfd, this, file_options_,
                                *cfd->GetLatestMutableCFOptions(),
                                current_version_number_++);
       builder->SaveTo(v->storage_info());
@@ -4917,7 +4926,9 @@ void VersionSet::MarkMinLogNumberToKeep2PC(uint64_t number) {
   }
 }
 
-Status VersionSet::WriteCurrentStateToManifest(log::Writer* log) {
+Status VersionSet::WriteCurrentStateToManifest(
+    const std::unordered_map<uint32_t, MutableCFState>& curr_state,
+    log::Writer* log) {
   // TODO: Break up into multiple records to reduce memory usage on recovery?
 
   // WARNING: This method doesn't hold a mutex!!
@@ -4983,7 +4994,10 @@ Status VersionSet::WriteCurrentStateToManifest(log::Writer* log) {
                        f->oldest_ancester_time, f->file_creation_time);
         }
       }
-      edit.SetLogNumber(cfd->GetLogNumber());
+      const auto iter = curr_state.find(cfd->GetID());
+      assert(iter != curr_state.end());
+      uint64_t log_number = iter->second.log_number;
+      edit.SetLogNumber(log_number);
       std::string record;
       if (!edit.EncodeTo(&record)) {
         return Status::Corruption(
@@ -5240,7 +5254,7 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
 
 InternalIterator* VersionSet::MakeInputIterator(
     const Compaction* c, RangeDelAggregator* range_del_agg,
-    const EnvOptions& env_options_compactions) {
+    const FileOptions& file_options_compactions) {
   auto cfd = c->column_family_data();
   ReadOptions read_options;
   read_options.verify_checksums = true;
@@ -5265,7 +5279,8 @@ InternalIterator* VersionSet::MakeInputIterator(
         const LevelFilesBrief* flevel = c->input_levels(which);
         for (size_t i = 0; i < flevel->num_files; i++) {
           list[num++] = cfd->table_cache()->NewIterator(
-              read_options, env_options_compactions, cfd->internal_comparator(),
+              read_options, file_options_compactions,
+              cfd->internal_comparator(),
               *flevel->files[i].file_metadata, range_del_agg,
               c->mutable_cf_options()->prefix_extractor.get(),
               /*table_reader_ptr=*/nullptr,
@@ -5278,7 +5293,7 @@ InternalIterator* VersionSet::MakeInputIterator(
       } else {
         // Create concatenating iterator for the files from this level
         list[num++] = new LevelIterator(
-            cfd->table_cache(), read_options, env_options_compactions,
+            cfd->table_cache(), read_options, file_options_compactions,
             cfd->internal_comparator(), c->input_levels(which),
             c->mutable_cf_options()->prefix_extractor.get(),
             /*should_sample=*/false,
@@ -5431,7 +5446,7 @@ ColumnFamilyData* VersionSet::CreateColumnFamily(
 
   MutableCFOptions dummy_cf_options;
   Version* dummy_versions =
-      new Version(nullptr, this, env_options_, dummy_cf_options);
+      new Version(nullptr, this, file_options_, dummy_cf_options);
   // Ref() dummy version once so that later we can call Unref() to delete it
   // by avoiding calling "delete" explicitly (~Version is private)
   dummy_versions->Ref();
@@ -5439,7 +5454,7 @@ ColumnFamilyData* VersionSet::CreateColumnFamily(
       edit->column_family_name_, edit->column_family_, dummy_versions,
       cf_options);
 
-  Version* v = new Version(new_cfd, this, env_options_,
+  Version* v = new Version(new_cfd, this, file_options_,
                            *new_cfd->GetLatestMutableCFOptions(),
                            current_version_number_++);
 
@@ -5483,11 +5498,11 @@ uint64_t VersionSet::GetTotalSstFilesSize(Version* dummy_versions) {
 
 ReactiveVersionSet::ReactiveVersionSet(const std::string& dbname,
                                        const ImmutableDBOptions* _db_options,
-                                       const EnvOptions& _env_options,
+                                       const FileOptions& _file_options,
                                        Cache* table_cache,
                                        WriteBufferManager* write_buffer_manager,
                                        WriteController* write_controller)
-    : VersionSet(dbname, _db_options, _env_options, table_cache,
+    : VersionSet(dbname, _db_options, _file_options, table_cache,
                  write_buffer_manager, write_controller,
                  /*block_cache_tracer=*/nullptr),
       number_of_edits_to_skip_(0) {}
@@ -5618,7 +5633,7 @@ Status ReactiveVersionSet::Recover(
       assert(builders_iter != builders.end());
       auto* builder = builders_iter->second->version_builder();
 
-      Version* v = new Version(cfd, this, env_options_,
+      Version* v = new Version(cfd, this, file_options_,
                                *cfd->GetLatestMutableCFOptions(),
                                current_version_number_++);
       builder->SaveTo(v->storage_info());
@@ -5811,8 +5826,7 @@ Status ReactiveVersionSet::ApplyOneVersionEditToBuilder(
     // secondary instance. (Is it possible that the ref count for cfd is 0 but
     // the ref count for its versions is higher than 0?)
     cfd->SetDropped();
-    if (cfd->Unref()) {
-      delete cfd;
+    if (cfd->UnrefAndTryDelete()) {
       cfd = nullptr;
     }
     active_version_builders_.erase(builder_iter);
@@ -5839,7 +5853,7 @@ Status ReactiveVersionSet::ApplyOneVersionEditToBuilder(
         &s);
 
     if (s.ok()) {
-      auto version = new Version(cfd, this, env_options_,
+      auto version = new Version(cfd, this, file_options_,
                                  *cfd->GetLatestMutableCFOptions(),
                                  current_version_number_++);
       builder->SaveTo(version->storage_info());
@@ -5882,9 +5896,9 @@ Status ReactiveVersionSet::MaybeSwitchManifest(
   Status s;
   do {
     std::string manifest_path;
-    s = GetCurrentManifestPath(dbname_, env_, &manifest_path,
+    s = GetCurrentManifestPath(dbname_, fs_, &manifest_path,
                                &manifest_file_number_);
-    std::unique_ptr<SequentialFile> manifest_file;
+    std::unique_ptr<FSSequentialFile> manifest_file;
     if (s.ok()) {
       if (nullptr == manifest_reader->get() ||
           manifest_reader->get()->file()->file_name() != manifest_path) {
@@ -5894,9 +5908,9 @@ Status ReactiveVersionSet::MaybeSwitchManifest(
         TEST_SYNC_POINT(
             "ReactiveVersionSet::MaybeSwitchManifest:"
             "AfterGetCurrentManifestPath:1");
-        s = env_->NewSequentialFile(
-            manifest_path, &manifest_file,
-            env_->OptimizeForManifestRead(env_options_));
+        s = fs_->NewSequentialFile(manifest_path,
+                                   env_->OptimizeForManifestRead(file_options_),
+                                   &manifest_file, nullptr);
       } else {
         // No need to switch manifest.
         break;

db/version_set.h

@@ -559,15 +559,15 @@ class Version {
   // Append to *iters a sequence of iterators that will
   // yield the contents of this Version when merged together.
   // REQUIRES: This version has been saved (see VersionSet::SaveTo)
-  void AddIterators(const ReadOptions&, const EnvOptions& soptions,
+  void AddIterators(const ReadOptions&, const FileOptions& soptions,
                     MergeIteratorBuilder* merger_iter_builder,
                     RangeDelAggregator* range_del_agg);
 
-  void AddIteratorsForLevel(const ReadOptions&, const EnvOptions& soptions,
+  void AddIteratorsForLevel(const ReadOptions&, const FileOptions& soptions,
                             MergeIteratorBuilder* merger_iter_builder,
                             int level, RangeDelAggregator* range_del_agg);
 
-  Status OverlapWithLevelIterator(const ReadOptions&, const EnvOptions&,
+  Status OverlapWithLevelIterator(const ReadOptions&, const FileOptions&,
                                   const Slice& smallest_user_key,
                                   const Slice& largest_user_key,
                                   int level, bool* overlap);
@@ -684,6 +684,7 @@ class Version {
 
  private:
   Env* env_;
+  FileSystem* fs_;
   friend class ReactiveVersionSet;
   friend class VersionSet;
 
@@ -729,14 +730,14 @@ class Version {
   Version* next_;               // Next version in linked list
   Version* prev_;               // Previous version in linked list
   int refs_;                    // Number of live refs to this version
-  const EnvOptions env_options_;
+  const FileOptions file_options_;
   const MutableCFOptions mutable_cf_options_;
 
   // A version number that uniquely represents this version. This is
   // used for debugging and logging purposes only.
   uint64_t version_number_;
 
-  Version(ColumnFamilyData* cfd, VersionSet* vset, const EnvOptions& env_opt,
+  Version(ColumnFamilyData* cfd, VersionSet* vset, const FileOptions& file_opt,
           MutableCFOptions mutable_cf_options, uint64_t version_number = 0);
 
   ~Version();
@@ -801,7 +802,7 @@ class AtomicGroupReadBuffer {
 class VersionSet {
  public:
   VersionSet(const std::string& dbname, const ImmutableDBOptions* db_options,
-             const EnvOptions& env_options, Cache* table_cache,
+             const FileOptions& file_options, Cache* table_cache,
              WriteBufferManager* write_buffer_manager,
              WriteController* write_controller,
              BlockCacheTracer* const block_cache_tracer);
@@ -863,7 +864,8 @@ class VersionSet {
       bool new_descriptor_log = false,
       const ColumnFamilyOptions* new_cf_options = nullptr);
 
-  static Status GetCurrentManifestPath(const std::string& dbname, Env* env,
+  static Status GetCurrentManifestPath(const std::string& dbname,
+                                       FileSystem* fs,
                                        std::string* manifest_filename,
                                        uint64_t* manifest_file_number);
 
@@ -876,7 +878,7 @@ class VersionSet {
   // Reads a manifest file and returns a list of column families in
   // column_families.
   static Status ListColumnFamilies(std::vector<std::string>* column_families,
-                                   const std::string& dbname, Env* env);
+                                   const std::string& dbname, FileSystem* fs);
 
 #ifndef ROCKSDB_LITE
   // Try to reduce the number of levels. This call is valid when
@@ -890,7 +892,7 @@ class VersionSet {
   // among [4-6] contains files.
   static Status ReduceNumberOfLevels(const std::string& dbname,
                                      const Options* options,
-                                     const EnvOptions& env_options,
+                                     const FileOptions& file_options,
                                      int new_levels);
 
   // printf contents (for debugging)
@@ -1004,7 +1006,7 @@ class VersionSet {
   // The caller should delete the iterator when no longer needed.
   InternalIterator* MakeInputIterator(
       const Compaction* c, RangeDelAggregator* range_del_agg,
-      const EnvOptions& env_options_compactions);
+      const FileOptions& file_options_compactions);
 
   // Add all files listed in any live version to *live.
   void AddLiveFiles(std::vector<FileDescriptor>* live_list);
@@ -1037,9 +1039,9 @@ class VersionSet {
                         uint64_t min_pending_output);
 
   ColumnFamilySet* GetColumnFamilySet() { return column_family_set_.get(); }
-  const EnvOptions& env_options() { return env_options_; }
-  void ChangeEnvOptions(const MutableDBOptions& new_options) {
-    env_options_.writable_file_max_buffer_size =
+  const FileOptions& file_options() { return file_options_; }
+  void ChangeFileOptions(const MutableDBOptions& new_options) {
+    file_options_.writable_file_max_buffer_size =
         new_options.writable_file_max_buffer_size;
   }
 
@@ -1073,8 +1075,14 @@ class VersionSet {
                            const Slice& start, const Slice& end,
                            TableReaderCaller caller);
 
+  struct MutableCFState {
+    uint64_t log_number;
+  };
+
   // Save current contents to *log
-  Status WriteCurrentStateToManifest(log::Writer* log);
+  Status WriteCurrentStateToManifest(
+      const std::unordered_map<uint32_t, MutableCFState>& curr_state,
+      log::Writer* log);
 
   void AppendVersion(ColumnFamilyData* column_family_data, Version* v);
 
@@ -1106,6 +1114,7 @@ class VersionSet {
   std::unique_ptr<ColumnFamilySet> column_family_set_;
 
   Env* const env_;
+  FileSystem* const fs_;
   const std::string dbname_;
   std::string db_id_;
   const ImmutableDBOptions* const db_options_;
@@ -1150,7 +1159,7 @@ class VersionSet {
   std::vector<std::string> obsolete_manifests_;
 
   // env options for all reads and writes except compactions
-  EnvOptions env_options_;
+  FileOptions file_options_;
 
   BlockCacheTracer* const block_cache_tracer_;
 
@@ -1174,7 +1183,7 @@ class ReactiveVersionSet : public VersionSet {
  public:
   ReactiveVersionSet(const std::string& dbname,
                      const ImmutableDBOptions* _db_options,
-                     const EnvOptions& _env_options, Cache* table_cache,
+                     const FileOptions& _file_options, Cache* table_cache,
                      WriteBufferManager* write_buffer_manager,
                      WriteController* write_controller);
 

db/version_set_test.cc

@@ -610,21 +610,25 @@ class VersionSetTestBase {
 
   VersionSetTestBase()
       : env_(Env::Default()),
+        fs_(std::make_shared<LegacyFileSystemWrapper>(env_)),
         dbname_(test::PerThreadDBPath("version_set_test")),
         db_options_(),
         mutable_cf_options_(cf_options_),
         table_cache_(NewLRUCache(50000, 16)),
         write_buffer_manager_(db_options_.db_write_buffer_size),
-        versions_(new VersionSet(dbname_, &db_options_, env_options_,
-                                 table_cache_.get(), &write_buffer_manager_,
-                                 &write_controller_,
-                                 /*block_cache_tracer=*/nullptr)),
-        reactive_versions_(std::make_shared<ReactiveVersionSet>(
-            dbname_, &db_options_, env_options_, table_cache_.get(),
-            &write_buffer_manager_, &write_controller_)),
         shutting_down_(false),
         mock_table_factory_(std::make_shared<mock::MockTableFactory>()) {
     EXPECT_OK(env_->CreateDirIfMissing(dbname_));
+
+    db_options_.env = env_;
+    db_options_.fs = fs_;
+    versions_.reset(new VersionSet(dbname_, &db_options_, env_options_,
+                                   table_cache_.get(), &write_buffer_manager_,
+                                   &write_controller_,
+                                   /*block_cache_tracer=*/nullptr)),
+        reactive_versions_ = std::make_shared<ReactiveVersionSet>(
+            dbname_, &db_options_, env_options_, table_cache_.get(),
+            &write_buffer_manager_, &write_controller_);
     db_options_.db_paths.emplace_back(dbname_,
                                       std::numeric_limits<uint64_t>::max());
   }
@@ -669,8 +673,8 @@ class VersionSetTestBase {
     Status s = env_->NewWritableFile(
         manifest, &file, env_->OptimizeForManifestWrite(env_options_));
     ASSERT_OK(s);
-    std::unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), manifest, env_options_));
+    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+        NewLegacyWritableFileWrapper(std::move(file)), manifest, env_options_));
     {
       log_writer->reset(new log::Writer(std::move(file_writer), 0, false));
       std::string record;
@@ -709,6 +713,7 @@ class VersionSetTestBase {
   }
 
   Env* env_;
+  std::shared_ptr<FileSystem> fs_;
   const std::string dbname_;
   EnvOptions env_options_;
   ImmutableDBOptions db_options_;

db/wal_manager.cc

@@ -120,7 +120,7 @@ Status WalManager::GetUpdatesSince(
     return s;
   }
   iter->reset(new TransactionLogIteratorImpl(
-      db_options_.wal_dir, &db_options_, read_options, env_options_, seq,
+      db_options_.wal_dir, &db_options_, read_options, file_options_, seq,
       std::move(wal_files), version_set, seq_per_batch_));
   return (*iter)->status();
 }
@@ -464,9 +464,10 @@ Status WalManager::ReadFirstLine(const std::string& fname,
     }
   };
 
-  std::unique_ptr<SequentialFile> file;
-  Status status = env_->NewSequentialFile(
-      fname, &file, env_->OptimizeForLogRead(env_options_));
+  std::unique_ptr<FSSequentialFile> file;
+  Status status = fs_->NewSequentialFile(fname,
+                                         fs_->OptimizeForLogRead(file_options_),
+                                         &file, nullptr);
   std::unique_ptr<SequentialFileReader> file_reader(
       new SequentialFileReader(std::move(file), fname));
 

db/wal_manager.h

@@ -36,10 +36,11 @@ namespace rocksdb {
 class WalManager {
  public:
   WalManager(const ImmutableDBOptions& db_options,
-             const EnvOptions& env_options, const bool seq_per_batch = false)
+             const FileOptions& file_options, const bool seq_per_batch = false)
       : db_options_(db_options),
-        env_options_(env_options),
+        file_options_(file_options),
         env_(db_options.env),
+        fs_(db_options.fs.get()),
         purge_wal_files_last_run_(0),
         seq_per_batch_(seq_per_batch),
         wal_in_db_path_(IsWalDirSameAsDBPath(&db_options)) {}
@@ -88,8 +89,9 @@ class WalManager {
 
   // ------- state from DBImpl ------
   const ImmutableDBOptions& db_options_;
-  const EnvOptions& env_options_;
+  const FileOptions file_options_;
   Env* env_;
+  FileSystem* fs_;
 
   // ------- WalManager state -------
   // cache for ReadFirstRecord() calls

db/wal_manager_test.cc

@@ -47,6 +47,8 @@ class WalManagerTest : public testing::Test {
                                       std::numeric_limits<uint64_t>::max());
     db_options_.wal_dir = dbname_;
     db_options_.env = env_.get();
+    fs_.reset(new LegacyFileSystemWrapper(env_.get()));
+    db_options_.fs = fs_;
 
     versions_.reset(new VersionSet(dbname_, &db_options_, env_options_,
                                    table_cache_.get(), &write_buffer_manager_,
@@ -79,8 +81,8 @@ class WalManagerTest : public testing::Test {
     std::string fname = ArchivedLogFileName(dbname_, current_log_number_);
     std::unique_ptr<WritableFile> file;
     ASSERT_OK(env_->NewWritableFile(fname, &file, env_options_));
-    std::unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), fname, env_options_));
+    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+        NewLegacyWritableFileWrapper(std::move(file)), fname, env_options_));
     current_log_writer_.reset(new log::Writer(std::move(file_writer), 0, false));
   }
 
@@ -111,6 +113,7 @@ class WalManagerTest : public testing::Test {
   WriteBufferManager write_buffer_manager_;
   std::unique_ptr<VersionSet> versions_;
   std::unique_ptr<WalManager> wal_manager_;
+  std::shared_ptr<LegacyFileSystemWrapper> fs_;
 
   std::unique_ptr<log::Writer> current_log_writer_;
   uint64_t current_log_number_;
@@ -130,8 +133,8 @@ TEST_F(WalManagerTest, ReadFirstRecordCache) {
       wal_manager_->TEST_ReadFirstRecord(kAliveLogFile, 1 /* number */, &s));
   ASSERT_EQ(s, 0U);
 
-  std::unique_ptr<WritableFileWriter> file_writer(
-      new WritableFileWriter(std::move(file), path, EnvOptions()));
+  std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+      NewLegacyWritableFileWrapper(std::move(file)), path, EnvOptions()));
   log::Writer writer(std::move(file_writer), 1,
                      db_options_.recycle_log_file_num > 0);
   WriteBatch batch;

db/write_thread.cc

@@ -344,9 +344,6 @@ void WriteThread::BeginWriteStall() {
       prev->link_older = w->link_older;
       w->status = Status::Incomplete("Write stall");
       SetState(w, STATE_COMPLETED);
-      if (prev->link_older) {
-        prev->link_older->link_newer = prev;
-      }
       w = prev->link_older;
     } else {
       prev = w;
@@ -358,11 +355,7 @@ void WriteThread::BeginWriteStall() {
 void WriteThread::EndWriteStall() {
   MutexLock lock(&stall_mu_);
 
-  // Unlink write_stall_dummy_ from the write queue. This will unblock
-  // pending write threads to enqueue themselves
   assert(newest_writer_.load(std::memory_order_relaxed) == &write_stall_dummy_);
-  assert(write_stall_dummy_.link_older != nullptr);
-  write_stall_dummy_.link_older->link_newer = write_stall_dummy_.link_newer;
   newest_writer_.exchange(write_stall_dummy_.link_older);
 
   // Wake up writers

db_stress_tool/CMakeLists.txt

@@ -0,0 +1,14 @@
+add_executable(db_stress${ARTIFACT_SUFFIX}
+  db_stress.cc
+  db_stress_tool.cc
+  batched_ops_stress.cc
+  cf_consistency_stress.cc
+  db_stress_common.cc
+  db_stress_driver.cc
+  db_stress_test_base.cc
+  db_stress_shared_state.cc
+  db_stress_gflags.cc
+  db_stress_tool.cc
+  no_batched_ops_stress.cc)
+target_link_libraries(db_stress${ARTIFACT_SUFFIX} ${ROCKSDB_LIB})
+list(APPEND tool_deps db_stress)

db_stress_tool/batched_ops_stress.cc

@@ -0,0 +1,341 @@
+//  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.
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_common.h"
+
+namespace rocksdb {
+class BatchedOpsStressTest : public StressTest {
+ public:
+  BatchedOpsStressTest() {}
+  virtual ~BatchedOpsStressTest() {}
+
+  // Given a key K and value V, this puts ("0"+K, "0"+V), ("1"+K, "1"+V), ...
+  // ("9"+K, "9"+V) in DB atomically i.e in a single batch.
+  // Also refer BatchedOpsStressTest::TestGet
+  Status TestPut(ThreadState* thread, WriteOptions& write_opts,
+                 const ReadOptions& /* read_opts */,
+                 const std::vector<int>& rand_column_families,
+                 const std::vector<int64_t>& rand_keys, char (&value)[100],
+                 std::unique_ptr<MutexLock>& /* lock */) override {
+    uint32_t value_base =
+        thread->rand.Next() % thread->shared->UNKNOWN_SENTINEL;
+    size_t sz = GenerateValue(value_base, value, sizeof(value));
+    Slice v(value, sz);
+    std::string keys[10] = {"9", "8", "7", "6", "5", "4", "3", "2", "1", "0"};
+    std::string values[10] = {"9", "8", "7", "6", "5", "4", "3", "2", "1", "0"};
+    Slice value_slices[10];
+    WriteBatch batch;
+    Status s;
+    auto cfh = column_families_[rand_column_families[0]];
+    std::string key_str = Key(rand_keys[0]);
+    for (int i = 0; i < 10; i++) {
+      keys[i] += key_str;
+      values[i] += v.ToString();
+      value_slices[i] = values[i];
+      if (FLAGS_use_merge) {
+        batch.Merge(cfh, keys[i], value_slices[i]);
+      } else {
+        batch.Put(cfh, keys[i], value_slices[i]);
+      }
+    }
+
+    s = db_->Write(write_opts, &batch);
+    if (!s.ok()) {
+      fprintf(stderr, "multiput error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    } else {
+      // we did 10 writes each of size sz + 1
+      thread->stats.AddBytesForWrites(10, (sz + 1) * 10);
+    }
+
+    return s;
+  }
+
+  // Given a key K, this deletes ("0"+K), ("1"+K),... ("9"+K)
+  // in DB atomically i.e in a single batch. Also refer MultiGet.
+  Status TestDelete(ThreadState* thread, WriteOptions& writeoptions,
+                    const std::vector<int>& rand_column_families,
+                    const std::vector<int64_t>& rand_keys,
+                    std::unique_ptr<MutexLock>& /* lock */) override {
+    std::string keys[10] = {"9", "7", "5", "3", "1", "8", "6", "4", "2", "0"};
+
+    WriteBatch batch;
+    Status s;
+    auto cfh = column_families_[rand_column_families[0]];
+    std::string key_str = Key(rand_keys[0]);
+    for (int i = 0; i < 10; i++) {
+      keys[i] += key_str;
+      batch.Delete(cfh, keys[i]);
+    }
+
+    s = db_->Write(writeoptions, &batch);
+    if (!s.ok()) {
+      fprintf(stderr, "multidelete error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    } else {
+      thread->stats.AddDeletes(10);
+    }
+
+    return s;
+  }
+
+  Status TestDeleteRange(ThreadState* /* thread */,
+                         WriteOptions& /* write_opts */,
+                         const std::vector<int>& /* rand_column_families */,
+                         const std::vector<int64_t>& /* rand_keys */,
+                         std::unique_ptr<MutexLock>& /* lock */) override {
+    assert(false);
+    return Status::NotSupported(
+        "BatchedOpsStressTest does not support "
+        "TestDeleteRange");
+  }
+
+  void TestIngestExternalFile(
+      ThreadState* /* thread */,
+      const std::vector<int>& /* rand_column_families */,
+      const std::vector<int64_t>& /* rand_keys */,
+      std::unique_ptr<MutexLock>& /* lock */) override {
+    assert(false);
+    fprintf(stderr,
+            "BatchedOpsStressTest does not support "
+            "TestIngestExternalFile\n");
+    std::terminate();
+  }
+
+  // Given a key K, this gets values for "0"+K, "1"+K,..."9"+K
+  // in the same snapshot, and verifies that all the values are of the form
+  // "0"+V, "1"+V,..."9"+V.
+  // ASSUMES that BatchedOpsStressTest::TestPut was used to put (K, V) into
+  // the DB.
+  Status TestGet(ThreadState* thread, const ReadOptions& readoptions,
+                 const std::vector<int>& rand_column_families,
+                 const std::vector<int64_t>& rand_keys) override {
+    std::string keys[10] = {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"};
+    Slice key_slices[10];
+    std::string values[10];
+    ReadOptions readoptionscopy = readoptions;
+    readoptionscopy.snapshot = db_->GetSnapshot();
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    auto cfh = column_families_[rand_column_families[0]];
+    std::string from_db;
+    Status s;
+    for (int i = 0; i < 10; i++) {
+      keys[i] += key.ToString();
+      key_slices[i] = keys[i];
+      s = db_->Get(readoptionscopy, cfh, key_slices[i], &from_db);
+      if (!s.ok() && !s.IsNotFound()) {
+        fprintf(stderr, "get error: %s\n", s.ToString().c_str());
+        values[i] = "";
+        thread->stats.AddErrors(1);
+        // we continue after error rather than exiting so that we can
+        // find more errors if any
+      } else if (s.IsNotFound()) {
+        values[i] = "";
+        thread->stats.AddGets(1, 0);
+      } else {
+        values[i] = from_db;
+
+        char expected_prefix = (keys[i])[0];
+        char actual_prefix = (values[i])[0];
+        if (actual_prefix != expected_prefix) {
+          fprintf(stderr, "error expected prefix = %c actual = %c\n",
+                  expected_prefix, actual_prefix);
+        }
+        (values[i])[0] = ' ';  // blank out the differing character
+        thread->stats.AddGets(1, 1);
+      }
+    }
+    db_->ReleaseSnapshot(readoptionscopy.snapshot);
+
+    // Now that we retrieved all values, check that they all match
+    for (int i = 1; i < 10; i++) {
+      if (values[i] != values[0]) {
+        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
+                key.ToString(true).c_str(), StringToHex(values[0]).c_str(),
+                StringToHex(values[i]).c_str());
+        // we continue after error rather than exiting so that we can
+        // find more errors if any
+      }
+    }
+
+    return s;
+  }
+
+  std::vector<Status> TestMultiGet(
+      ThreadState* thread, const ReadOptions& readoptions,
+      const std::vector<int>& rand_column_families,
+      const std::vector<int64_t>& rand_keys) override {
+    size_t num_keys = rand_keys.size();
+    std::vector<Status> ret_status(num_keys);
+    std::array<std::string, 10> keys = {{"0", "1", "2", "3", "4",
+                                         "5", "6", "7", "8", "9"}};
+    size_t num_prefixes = keys.size();
+    for (size_t rand_key = 0; rand_key < num_keys; ++rand_key) {
+      std::vector<Slice> key_slices;
+      std::vector<PinnableSlice> values(num_prefixes);
+      std::vector<Status> statuses(num_prefixes);
+      ReadOptions readoptionscopy = readoptions;
+      readoptionscopy.snapshot = db_->GetSnapshot();
+      std::vector<std::string> key_str;
+      key_str.reserve(num_prefixes);
+      key_slices.reserve(num_prefixes);
+      std::string from_db;
+      ColumnFamilyHandle* cfh = column_families_[rand_column_families[0]];
+
+      for (size_t key = 0; key < num_prefixes; ++key) {
+        key_str.emplace_back(keys[key] + Key(rand_keys[rand_key]));
+        key_slices.emplace_back(key_str.back());
+      }
+      db_->MultiGet(readoptionscopy, cfh, num_prefixes, key_slices.data(),
+                    values.data(), statuses.data());
+      for (size_t i = 0; i < num_prefixes; i++) {
+        Status s = statuses[i];
+        if (!s.ok() && !s.IsNotFound()) {
+          fprintf(stderr, "get error: %s\n", s.ToString().c_str());
+          thread->stats.AddErrors(1);
+          ret_status[rand_key] = s;
+          // we continue after error rather than exiting so that we can
+          // find more errors if any
+        } else if (s.IsNotFound()) {
+          thread->stats.AddGets(1, 0);
+          ret_status[rand_key] = s;
+        } else {
+          char expected_prefix = (keys[i])[0];
+          char actual_prefix = (values[i])[0];
+          if (actual_prefix != expected_prefix) {
+            fprintf(stderr, "error expected prefix = %c actual = %c\n",
+                    expected_prefix, actual_prefix);
+          }
+          std::string str;
+          str.assign(values[i].data(), values[i].size());
+          values[i].Reset();
+          str[0] = ' ';  // blank out the differing character
+          values[i].PinSelf(str);
+          thread->stats.AddGets(1, 1);
+        }
+      }
+      db_->ReleaseSnapshot(readoptionscopy.snapshot);
+
+      // Now that we retrieved all values, check that they all match
+      for (size_t i = 1; i < num_prefixes; i++) {
+        if (values[i] != values[0]) {
+          fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
+                  key_str[i].c_str(), StringToHex(values[0].ToString()).c_str(),
+                  StringToHex(values[i].ToString()).c_str());
+          // we continue after error rather than exiting so that we can
+          // find more errors if any
+        }
+      }
+    }
+
+    return ret_status;
+  }
+
+  // Given a key, this does prefix scans for "0"+P, "1"+P,..."9"+P
+  // in the same snapshot where P is the first FLAGS_prefix_size - 1 bytes
+  // of the key. Each of these 10 scans returns a series of values;
+  // each series should be the same length, and it is verified for each
+  // index i that all the i'th values are of the form "0"+V, "1"+V,..."9"+V.
+  // ASSUMES that MultiPut was used to put (K, V)
+  Status TestPrefixScan(ThreadState* thread, const ReadOptions& readoptions,
+                        const std::vector<int>& rand_column_families,
+                        const std::vector<int64_t>& rand_keys) override {
+    size_t prefix_to_use =
+        (FLAGS_prefix_size < 0) ? 7 : static_cast<size_t>(FLAGS_prefix_size);
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    auto cfh = column_families_[rand_column_families[0]];
+    std::string prefixes[10] = {"0", "1", "2", "3", "4",
+                                "5", "6", "7", "8", "9"};
+    Slice prefix_slices[10];
+    ReadOptions readoptionscopy[10];
+    const Snapshot* snapshot = db_->GetSnapshot();
+    Iterator* iters[10];
+    std::string upper_bounds[10];
+    Slice ub_slices[10];
+    Status s = Status::OK();
+    for (int i = 0; i < 10; i++) {
+      prefixes[i] += key.ToString();
+      prefixes[i].resize(prefix_to_use);
+      prefix_slices[i] = Slice(prefixes[i]);
+      readoptionscopy[i] = readoptions;
+      readoptionscopy[i].snapshot = snapshot;
+      if (thread->rand.OneIn(2) &&
+          GetNextPrefix(prefix_slices[i], &(upper_bounds[i]))) {
+        // For half of the time, set the upper bound to the next prefix
+        ub_slices[i] = Slice(upper_bounds[i]);
+        readoptionscopy[i].iterate_upper_bound = &(ub_slices[i]);
+      }
+      iters[i] = db_->NewIterator(readoptionscopy[i], cfh);
+      iters[i]->Seek(prefix_slices[i]);
+    }
+
+    long count = 0;
+    while (iters[0]->Valid() && iters[0]->key().starts_with(prefix_slices[0])) {
+      count++;
+      std::string values[10];
+      // get list of all values for this iteration
+      for (int i = 0; i < 10; i++) {
+        // no iterator should finish before the first one
+        assert(iters[i]->Valid() &&
+               iters[i]->key().starts_with(prefix_slices[i]));
+        values[i] = iters[i]->value().ToString();
+
+        char expected_first = (prefixes[i])[0];
+        char actual_first = (values[i])[0];
+
+        if (actual_first != expected_first) {
+          fprintf(stderr, "error expected first = %c actual = %c\n",
+                  expected_first, actual_first);
+        }
+        (values[i])[0] = ' ';  // blank out the differing character
+      }
+      // make sure all values are equivalent
+      for (int i = 0; i < 10; i++) {
+        if (values[i] != values[0]) {
+          fprintf(stderr,
+                  "error : %d, inconsistent values for prefix %s: %s, %s\n", i,
+                  prefixes[i].c_str(), StringToHex(values[0]).c_str(),
+                  StringToHex(values[i]).c_str());
+          // we continue after error rather than exiting so that we can
+          // find more errors if any
+        }
+        iters[i]->Next();
+      }
+    }
+
+    // cleanup iterators and snapshot
+    for (int i = 0; i < 10; i++) {
+      // if the first iterator finished, they should have all finished
+      assert(!iters[i]->Valid() ||
+             !iters[i]->key().starts_with(prefix_slices[i]));
+      assert(iters[i]->status().ok());
+      delete iters[i];
+    }
+    db_->ReleaseSnapshot(snapshot);
+
+    if (s.ok()) {
+      thread->stats.AddPrefixes(1, count);
+    } else {
+      fprintf(stderr, "TestPrefixScan error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    }
+
+    return s;
+  }
+
+  void VerifyDb(ThreadState* /* thread */) const override {}
+};
+
+StressTest* CreateBatchedOpsStressTest() { return new BatchedOpsStressTest(); }
+
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/cf_consistency_stress.cc

@@ -0,0 +1,581 @@
+//  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.
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_common.h"
+
+namespace rocksdb {
+class CfConsistencyStressTest : public StressTest {
+ public:
+  CfConsistencyStressTest() : batch_id_(0) {}
+
+  ~CfConsistencyStressTest() override {}
+
+  Status TestPut(ThreadState* thread, WriteOptions& write_opts,
+                 const ReadOptions& /* read_opts */,
+                 const std::vector<int>& rand_column_families,
+                 const std::vector<int64_t>& rand_keys, char (&value)[100],
+                 std::unique_ptr<MutexLock>& /* lock */) override {
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    uint64_t value_base = batch_id_.fetch_add(1);
+    size_t sz =
+        GenerateValue(static_cast<uint32_t>(value_base), value, sizeof(value));
+    Slice v(value, sz);
+    WriteBatch batch;
+    for (auto cf : rand_column_families) {
+      ColumnFamilyHandle* cfh = column_families_[cf];
+      if (FLAGS_use_merge) {
+        batch.Merge(cfh, key, v);
+      } else { /* !FLAGS_use_merge */
+        batch.Put(cfh, key, v);
+      }
+    }
+    Status s = db_->Write(write_opts, &batch);
+    if (!s.ok()) {
+      fprintf(stderr, "multi put or merge error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    } else {
+      auto num = static_cast<long>(rand_column_families.size());
+      thread->stats.AddBytesForWrites(num, (sz + 1) * num);
+    }
+
+    return s;
+  }
+
+  Status TestDelete(ThreadState* thread, WriteOptions& write_opts,
+                    const std::vector<int>& rand_column_families,
+                    const std::vector<int64_t>& rand_keys,
+                    std::unique_ptr<MutexLock>& /* lock */) override {
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    WriteBatch batch;
+    for (auto cf : rand_column_families) {
+      ColumnFamilyHandle* cfh = column_families_[cf];
+      batch.Delete(cfh, key);
+    }
+    Status s = db_->Write(write_opts, &batch);
+    if (!s.ok()) {
+      fprintf(stderr, "multidel error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    } else {
+      thread->stats.AddDeletes(static_cast<long>(rand_column_families.size()));
+    }
+    return s;
+  }
+
+  Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts,
+                         const std::vector<int>& rand_column_families,
+                         const std::vector<int64_t>& rand_keys,
+                         std::unique_ptr<MutexLock>& /* lock */) override {
+    int64_t rand_key = rand_keys[0];
+    auto shared = thread->shared;
+    int64_t max_key = shared->GetMaxKey();
+    if (rand_key > max_key - FLAGS_range_deletion_width) {
+      rand_key =
+          thread->rand.Next() % (max_key - FLAGS_range_deletion_width + 1);
+    }
+    std::string key_str = Key(rand_key);
+    Slice key = key_str;
+    std::string end_key_str = Key(rand_key + FLAGS_range_deletion_width);
+    Slice end_key = end_key_str;
+    WriteBatch batch;
+    for (auto cf : rand_column_families) {
+      ColumnFamilyHandle* cfh = column_families_[rand_column_families[cf]];
+      batch.DeleteRange(cfh, key, end_key);
+    }
+    Status s = db_->Write(write_opts, &batch);
+    if (!s.ok()) {
+      fprintf(stderr, "multi del range error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    } else {
+      thread->stats.AddRangeDeletions(
+          static_cast<long>(rand_column_families.size()));
+    }
+    return s;
+  }
+
+  void TestIngestExternalFile(
+      ThreadState* /* thread */,
+      const std::vector<int>& /* rand_column_families */,
+      const std::vector<int64_t>& /* rand_keys */,
+      std::unique_ptr<MutexLock>& /* lock */) override {
+    assert(false);
+    fprintf(stderr,
+            "CfConsistencyStressTest does not support TestIngestExternalFile "
+            "because it's not possible to verify the result\n");
+    std::terminate();
+  }
+
+  Status TestGet(ThreadState* thread, const ReadOptions& readoptions,
+                 const std::vector<int>& rand_column_families,
+                 const std::vector<int64_t>& rand_keys) override {
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    Status s;
+    bool is_consistent = true;
+
+    if (thread->rand.OneIn(2)) {
+      // 1/2 chance, does a random read from random CF
+      auto cfh =
+          column_families_[rand_column_families[thread->rand.Next() %
+                                                rand_column_families.size()]];
+      std::string from_db;
+      s = db_->Get(readoptions, cfh, key, &from_db);
+    } else {
+      // 1/2 chance, comparing one key is the same across all CFs
+      const Snapshot* snapshot = db_->GetSnapshot();
+      ReadOptions readoptionscopy = readoptions;
+      readoptionscopy.snapshot = snapshot;
+
+      std::string value0;
+      s = db_->Get(readoptionscopy, column_families_[rand_column_families[0]],
+                   key, &value0);
+      if (s.ok() || s.IsNotFound()) {
+        bool found = s.ok();
+        for (size_t i = 1; i < rand_column_families.size(); i++) {
+          std::string value1;
+          s = db_->Get(readoptionscopy,
+                       column_families_[rand_column_families[i]], key, &value1);
+          if (!s.ok() && !s.IsNotFound()) {
+            break;
+          }
+          if (!found && s.ok()) {
+            fprintf(stderr, "Get() return different results with key %s\n",
+                    Slice(key_str).ToString(true).c_str());
+            fprintf(stderr, "CF %s is not found\n",
+                    column_family_names_[0].c_str());
+            fprintf(stderr, "CF %s returns value %s\n",
+                    column_family_names_[i].c_str(),
+                    Slice(value1).ToString(true).c_str());
+            is_consistent = false;
+          } else if (found && s.IsNotFound()) {
+            fprintf(stderr, "Get() return different results with key %s\n",
+                    Slice(key_str).ToString(true).c_str());
+            fprintf(stderr, "CF %s returns value %s\n",
+                    column_family_names_[0].c_str(),
+                    Slice(value0).ToString(true).c_str());
+            fprintf(stderr, "CF %s is not found\n",
+                    column_family_names_[i].c_str());
+            is_consistent = false;
+          } else if (s.ok() && value0 != value1) {
+            fprintf(stderr, "Get() return different results with key %s\n",
+                    Slice(key_str).ToString(true).c_str());
+            fprintf(stderr, "CF %s returns value %s\n",
+                    column_family_names_[0].c_str(),
+                    Slice(value0).ToString(true).c_str());
+            fprintf(stderr, "CF %s returns value %s\n",
+                    column_family_names_[i].c_str(),
+                    Slice(value1).ToString(true).c_str());
+            is_consistent = false;
+          }
+          if (!is_consistent) {
+            break;
+          }
+        }
+      }
+
+      db_->ReleaseSnapshot(snapshot);
+    }
+    if (!is_consistent) {
+      fprintf(stderr, "TestGet error: is_consistent is false\n");
+      thread->stats.AddErrors(1);
+      // Fail fast to preserve the DB state.
+      thread->shared->SetVerificationFailure();
+    } else if (s.ok()) {
+      thread->stats.AddGets(1, 1);
+    } else if (s.IsNotFound()) {
+      thread->stats.AddGets(1, 0);
+    } else {
+      fprintf(stderr, "TestGet error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    }
+    return s;
+  }
+
+  std::vector<Status> TestMultiGet(
+      ThreadState* thread, const ReadOptions& read_opts,
+      const std::vector<int>& rand_column_families,
+      const std::vector<int64_t>& rand_keys) override {
+    size_t num_keys = rand_keys.size();
+    std::vector<std::string> key_str;
+    std::vector<Slice> keys;
+    keys.reserve(num_keys);
+    key_str.reserve(num_keys);
+    std::vector<PinnableSlice> values(num_keys);
+    std::vector<Status> statuses(num_keys);
+    ColumnFamilyHandle* cfh = column_families_[rand_column_families[0]];
+
+    for (size_t i = 0; i < num_keys; ++i) {
+      key_str.emplace_back(Key(rand_keys[i]));
+      keys.emplace_back(key_str.back());
+    }
+    db_->MultiGet(read_opts, cfh, num_keys, keys.data(), values.data(),
+                  statuses.data());
+    for (auto s : statuses) {
+      if (s.ok()) {
+        // found case
+        thread->stats.AddGets(1, 1);
+      } else if (s.IsNotFound()) {
+        // not found case
+        thread->stats.AddGets(1, 0);
+      } else {
+        // errors case
+        fprintf(stderr, "MultiGet error: %s\n", s.ToString().c_str());
+        thread->stats.AddErrors(1);
+      }
+    }
+    return statuses;
+  }
+
+  Status TestPrefixScan(ThreadState* thread, const ReadOptions& readoptions,
+                        const std::vector<int>& rand_column_families,
+                        const std::vector<int64_t>& rand_keys) override {
+    size_t prefix_to_use =
+        (FLAGS_prefix_size < 0) ? 7 : static_cast<size_t>(FLAGS_prefix_size);
+
+    std::string key_str = Key(rand_keys[0]);
+    Slice key = key_str;
+    Slice prefix = Slice(key.data(), prefix_to_use);
+
+    std::string upper_bound;
+    Slice ub_slice;
+    ReadOptions ro_copy = readoptions;
+    if (thread->rand.OneIn(2) && GetNextPrefix(prefix, &upper_bound)) {
+      ub_slice = Slice(upper_bound);
+      ro_copy.iterate_upper_bound = &ub_slice;
+    }
+    auto cfh =
+        column_families_[rand_column_families[thread->rand.Next() %
+                                              rand_column_families.size()]];
+    Iterator* iter = db_->NewIterator(ro_copy, cfh);
+    long count = 0;
+    for (iter->Seek(prefix); iter->Valid() && iter->key().starts_with(prefix);
+         iter->Next()) {
+      ++count;
+    }
+    assert(prefix_to_use == 0 ||
+           count <= (static_cast<long>(1) << ((8 - prefix_to_use) * 8)));
+    Status s = iter->status();
+    if (s.ok()) {
+      thread->stats.AddPrefixes(1, count);
+    } else {
+      fprintf(stderr, "TestPrefixScan error: %s\n", s.ToString().c_str());
+      thread->stats.AddErrors(1);
+    }
+    delete iter;
+    return s;
+  }
+
+  ColumnFamilyHandle* GetControlCfh(ThreadState* thread,
+                                    int /*column_family_id*/
+                                    ) override {
+    // All column families should contain the same data. Randomly pick one.
+    return column_families_[thread->rand.Next() % column_families_.size()];
+  }
+
+#ifdef ROCKSDB_LITE
+  Status TestCheckpoint(ThreadState* /* thread */,
+                        const std::vector<int>& /* rand_column_families */,
+                        const std::vector<int64_t>& /* rand_keys */) override {
+    assert(false);
+    fprintf(stderr,
+            "RocksDB lite does not support "
+            "TestCheckpoint\n");
+    std::terminate();
+  }
+#else
+  Status TestCheckpoint(ThreadState* thread,
+                        const std::vector<int>& /* rand_column_families */,
+                        const std::vector<int64_t>& /* rand_keys */) override {
+    std::string checkpoint_dir =
+        FLAGS_db + "/.checkpoint" + ToString(thread->tid);
+
+    // We need to clear DB including manifest files, so make a copy
+    Options opt_copy = options_;
+    opt_copy.env = db_stress_env->target();
+    DestroyDB(checkpoint_dir, opt_copy);
+
+    Checkpoint* checkpoint = nullptr;
+    Status s = Checkpoint::Create(db_, &checkpoint);
+    if (s.ok()) {
+      s = checkpoint->CreateCheckpoint(checkpoint_dir);
+    }
+    std::vector<ColumnFamilyHandle*> cf_handles;
+    DB* checkpoint_db = nullptr;
+    if (s.ok()) {
+      delete checkpoint;
+      checkpoint = nullptr;
+      Options options(options_);
+      options.listeners.clear();
+      std::vector<ColumnFamilyDescriptor> cf_descs;
+      // TODO(ajkr): `column_family_names_` is not safe to access here when
+      // `clear_column_family_one_in != 0`. But we can't easily switch to
+      // `ListColumnFamilies` to get names because it won't necessarily give
+      // the same order as `column_family_names_`.
+      if (FLAGS_clear_column_family_one_in == 0) {
+        for (const auto& name : column_family_names_) {
+          cf_descs.emplace_back(name, ColumnFamilyOptions(options));
+        }
+        s = DB::OpenForReadOnly(DBOptions(options), checkpoint_dir, cf_descs,
+                                &cf_handles, &checkpoint_db);
+      }
+    }
+    if (checkpoint_db != nullptr) {
+      for (auto cfh : cf_handles) {
+        delete cfh;
+      }
+      cf_handles.clear();
+      delete checkpoint_db;
+      checkpoint_db = nullptr;
+    }
+    DestroyDB(checkpoint_dir, opt_copy);
+    if (!s.ok()) {
+      fprintf(stderr, "A checkpoint operation failed with: %s\n",
+              s.ToString().c_str());
+    }
+    return s;
+  }
+#endif  // !ROCKSDB_LITE
+
+  void VerifyDb(ThreadState* thread) const override {
+    ReadOptions options(FLAGS_verify_checksum, true);
+    // We must set total_order_seek to true because we are doing a SeekToFirst
+    // on a column family whose memtables may support (by default) prefix-based
+    // iterator. In this case, NewIterator with options.total_order_seek being
+    // false returns a prefix-based iterator. Calling SeekToFirst using this
+    // iterator causes the iterator to become invalid. That means we cannot
+    // iterate the memtable using this iterator any more, although the memtable
+    // contains the most up-to-date key-values.
+    options.total_order_seek = true;
+    const auto ss_deleter = [this](const Snapshot* ss) {
+      db_->ReleaseSnapshot(ss);
+    };
+    std::unique_ptr<const Snapshot, decltype(ss_deleter)> snapshot_guard(
+        db_->GetSnapshot(), ss_deleter);
+    options.snapshot = snapshot_guard.get();
+    assert(thread != nullptr);
+    auto shared = thread->shared;
+    std::vector<std::unique_ptr<Iterator>> iters(column_families_.size());
+    for (size_t i = 0; i != column_families_.size(); ++i) {
+      iters[i].reset(db_->NewIterator(options, column_families_[i]));
+    }
+    for (auto& iter : iters) {
+      iter->SeekToFirst();
+    }
+    size_t num = column_families_.size();
+    assert(num == iters.size());
+    std::vector<Status> statuses(num, Status::OK());
+    do {
+      if (shared->HasVerificationFailedYet()) {
+        break;
+      }
+      size_t valid_cnt = 0;
+      size_t idx = 0;
+      for (auto& iter : iters) {
+        if (iter->Valid()) {
+          ++valid_cnt;
+        } else {
+          statuses[idx] = iter->status();
+        }
+        ++idx;
+      }
+      if (valid_cnt == 0) {
+        Status status;
+        for (size_t i = 0; i != num; ++i) {
+          const auto& s = statuses[i];
+          if (!s.ok()) {
+            status = s;
+            fprintf(stderr, "Iterator on cf %s has error: %s\n",
+                    column_families_[i]->GetName().c_str(),
+                    s.ToString().c_str());
+            shared->SetVerificationFailure();
+          }
+        }
+        break;
+      } else if (valid_cnt != iters.size()) {
+        shared->SetVerificationFailure();
+        for (size_t i = 0; i != num; ++i) {
+          if (!iters[i]->Valid()) {
+            if (statuses[i].ok()) {
+              fprintf(stderr, "Finished scanning cf %s\n",
+                      column_families_[i]->GetName().c_str());
+            } else {
+              fprintf(stderr, "Iterator on cf %s has error: %s\n",
+                      column_families_[i]->GetName().c_str(),
+                      statuses[i].ToString().c_str());
+            }
+          } else {
+            fprintf(stderr, "cf %s has remaining data to scan\n",
+                    column_families_[i]->GetName().c_str());
+          }
+        }
+        break;
+      }
+      if (shared->HasVerificationFailedYet()) {
+        break;
+      }
+      // If the program reaches here, then all column families' iterators are
+      // still valid.
+      if (shared->PrintingVerificationResults()) {
+        continue;
+      }
+      Slice key;
+      Slice value;
+      int num_mismatched_cfs = 0;
+      for (size_t i = 0; i != num; ++i) {
+        if (i == 0) {
+          key = iters[i]->key();
+          value = iters[i]->value();
+        } else {
+          int cmp = key.compare(iters[i]->key());
+          if (cmp != 0) {
+            ++num_mismatched_cfs;
+            if (1 == num_mismatched_cfs) {
+              fprintf(stderr, "Verification failed\n");
+              fprintf(stderr, "Latest Sequence Number: %" PRIu64 "\n",
+                      db_->GetLatestSequenceNumber());
+              fprintf(stderr, "[%s] %s => %s\n",
+                      column_families_[0]->GetName().c_str(),
+                      key.ToString(true /* hex */).c_str(),
+                      value.ToString(true /* hex */).c_str());
+            }
+            fprintf(stderr, "[%s] %s => %s\n",
+                    column_families_[i]->GetName().c_str(),
+                    iters[i]->key().ToString(true /* hex */).c_str(),
+                    iters[i]->value().ToString(true /* hex */).c_str());
+#ifndef ROCKSDB_LITE
+            Slice begin_key;
+            Slice end_key;
+            if (cmp < 0) {
+              begin_key = key;
+              end_key = iters[i]->key();
+            } else {
+              begin_key = iters[i]->key();
+              end_key = key;
+            }
+            std::vector<KeyVersion> versions;
+            const size_t kMaxNumIKeys = 8;
+            const auto print_key_versions = [&](ColumnFamilyHandle* cfh) {
+              Status s = GetAllKeyVersions(db_, cfh, begin_key, end_key,
+                                           kMaxNumIKeys, &versions);
+              if (!s.ok()) {
+                fprintf(stderr, "%s\n", s.ToString().c_str());
+                return;
+              }
+              assert(nullptr != cfh);
+              fprintf(stderr,
+                      "Internal keys in CF '%s', [%s, %s] (max %" ROCKSDB_PRIszt
+                      ")\n",
+                      cfh->GetName().c_str(),
+                      begin_key.ToString(true /* hex */).c_str(),
+                      end_key.ToString(true /* hex */).c_str(), kMaxNumIKeys);
+              for (const KeyVersion& kv : versions) {
+                fprintf(stderr, "  key %s seq %" PRIu64 " type %d\n",
+                        Slice(kv.user_key).ToString(true).c_str(), kv.sequence,
+                        kv.type);
+              }
+            };
+            if (1 == num_mismatched_cfs) {
+              print_key_versions(column_families_[0]);
+            }
+            print_key_versions(column_families_[i]);
+#endif  // ROCKSDB_LITE
+            shared->SetVerificationFailure();
+          }
+        }
+      }
+      shared->FinishPrintingVerificationResults();
+      for (auto& iter : iters) {
+        iter->Next();
+      }
+    } while (true);
+  }
+
+#ifndef ROCKSDB_LITE
+  void ContinuouslyVerifyDb(ThreadState* thread) const override {
+    assert(thread);
+    Status status;
+
+    DB* db_ptr = cmp_db_ ? cmp_db_ : db_;
+    const auto& cfhs = cmp_db_ ? cmp_cfhs_ : column_families_;
+    const auto ss_deleter = [&](const Snapshot* ss) {
+      db_ptr->ReleaseSnapshot(ss);
+    };
+    std::unique_ptr<const Snapshot, decltype(ss_deleter)> snapshot_guard(
+        db_ptr->GetSnapshot(), ss_deleter);
+    if (cmp_db_) {
+      status = cmp_db_->TryCatchUpWithPrimary();
+    }
+    SharedState* shared = thread->shared;
+    assert(shared);
+    if (!status.ok()) {
+      shared->SetShouldStopTest();
+      return;
+    }
+    assert(cmp_db_ || snapshot_guard.get());
+    const auto checksum_column_family = [](Iterator* iter,
+                                           uint32_t* checksum) -> Status {
+      assert(nullptr != checksum);
+      uint32_t ret = 0;
+      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+        ret = crc32c::Extend(ret, iter->key().data(), iter->key().size());
+        ret = crc32c::Extend(ret, iter->value().data(), iter->value().size());
+      }
+      *checksum = ret;
+      return iter->status();
+    };
+    ReadOptions ropts;
+    ropts.total_order_seek = true;
+    ropts.snapshot = snapshot_guard.get();
+    uint32_t crc = 0;
+    {
+      // Compute crc for all key-values of default column family.
+      std::unique_ptr<Iterator> it(db_ptr->NewIterator(ropts));
+      status = checksum_column_family(it.get(), &crc);
+    }
+    uint32_t tmp_crc = 0;
+    if (status.ok()) {
+      for (ColumnFamilyHandle* cfh : cfhs) {
+        if (cfh == db_ptr->DefaultColumnFamily()) {
+          continue;
+        }
+        std::unique_ptr<Iterator> it(db_ptr->NewIterator(ropts, cfh));
+        status = checksum_column_family(it.get(), &tmp_crc);
+        if (!status.ok() || tmp_crc != crc) {
+          break;
+        }
+      }
+    }
+    if (!status.ok() || tmp_crc != crc) {
+      shared->SetShouldStopTest();
+    }
+  }
+#endif  // !ROCKSDB_LITE
+
+  std::vector<int> GenerateColumnFamilies(
+      const int /* num_column_families */,
+      int /* rand_column_family */) const override {
+    std::vector<int> ret;
+    int num = static_cast<int>(column_families_.size());
+    int k = 0;
+    std::generate_n(back_inserter(ret), num, [&k]() -> int { return k++; });
+    return ret;
+  }
+
+ private:
+  std::atomic<int64_t> batch_id_;
+};
+
+StressTest* CreateCfConsistencyStressTest() {
+  return new CfConsistencyStressTest();
+}
+
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_common.cc

@@ -0,0 +1,224 @@
+//  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.
+//
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_common.h"
+#include <cmath>
+
+rocksdb::DbStressEnvWrapper* db_stress_env = nullptr;
+enum rocksdb::CompressionType compression_type_e = rocksdb::kSnappyCompression;
+enum rocksdb::CompressionType bottommost_compression_type_e =
+    rocksdb::kSnappyCompression;
+enum rocksdb::ChecksumType checksum_type_e = rocksdb::kCRC32c;
+enum RepFactory FLAGS_rep_factory = kSkipList;
+std::vector<double> sum_probs(100001);
+int64_t zipf_sum_size = 100000;
+
+namespace rocksdb {
+
+// Zipfian distribution is generated based on a pre-calculated array.
+// It should be used before start the stress test.
+// First, the probability distribution function (PDF) of this Zipfian follows
+// power low. P(x) = 1/(x^alpha).
+// So we calculate the PDF when x is from 0 to zipf_sum_size in first for loop
+// and add the PDF value togetger as c. So we get the total probability in c.
+// Next, we calculate inverse CDF of Zipfian and store the value of each in
+// an array (sum_probs). The rank is from 0 to zipf_sum_size. For example, for
+// integer k, its Zipfian CDF value is sum_probs[k].
+// Third, when we need to get an integer whose probability follows Zipfian
+// distribution, we use a rand_seed [0,1] which follows uniform distribution
+// as a seed and search it in the sum_probs via binary search. When we find
+// the closest sum_probs[i] of rand_seed, i is the integer that in
+// [0, zipf_sum_size] following Zipfian distribution with parameter alpha.
+// Finally, we can scale i to [0, max_key] scale.
+// In order to avoid that hot keys are close to each other and skew towards 0,
+// we use Rando64 to shuffle it.
+void InitializeHotKeyGenerator(double alpha) {
+  double c = 0;
+  for (int64_t i = 1; i <= zipf_sum_size; i++) {
+    c = c + (1.0 / std::pow(static_cast<double>(i), alpha));
+  }
+  c = 1.0 / c;
+
+  sum_probs[0] = 0;
+  for (int64_t i = 1; i <= zipf_sum_size; i++) {
+    sum_probs[i] =
+        sum_probs[i - 1] + c / std::pow(static_cast<double>(i), alpha);
+  }
+}
+
+// Generate one key that follows the Zipfian distribution. The skewness
+// is decided by the parameter alpha. Input is the rand_seed [0,1] and
+// the max of the key to be generated. If we directly return tmp_zipf_seed,
+// the closer to 0, the higher probability will be. To randomly distribute
+// the hot keys in [0, max_key], we use Random64 to shuffle it.
+int64_t GetOneHotKeyID(double rand_seed, int64_t max_key) {
+  int64_t low = 1, mid, high = zipf_sum_size, zipf = 0;
+  while (low <= high) {
+    mid = std::floor((low + high) / 2);
+    if (sum_probs[mid] >= rand_seed && sum_probs[mid - 1] < rand_seed) {
+      zipf = mid;
+      break;
+    } else if (sum_probs[mid] >= rand_seed) {
+      high = mid - 1;
+    } else {
+      low = mid + 1;
+    }
+  }
+  int64_t tmp_zipf_seed = static_cast<int64_t>(
+      std::floor(zipf * max_key / (static_cast<double>(zipf_sum_size))));
+  Random64 rand_local(tmp_zipf_seed);
+  return rand_local.Next() % max_key;
+}
+
+void PoolSizeChangeThread(void* v) {
+  assert(FLAGS_compaction_thread_pool_adjust_interval > 0);
+  ThreadState* thread = reinterpret_cast<ThreadState*>(v);
+  SharedState* shared = thread->shared;
+
+  while (true) {
+    {
+      MutexLock l(shared->GetMutex());
+      if (shared->ShouldStopBgThread()) {
+        shared->IncBgThreadsFinished();
+        if (shared->BgThreadsFinished()) {
+          shared->GetCondVar()->SignalAll();
+        }
+        return;
+      }
+    }
+
+    auto thread_pool_size_base = FLAGS_max_background_compactions;
+    auto thread_pool_size_var = FLAGS_compaction_thread_pool_variations;
+    int new_thread_pool_size =
+        thread_pool_size_base - thread_pool_size_var +
+        thread->rand.Next() % (thread_pool_size_var * 2 + 1);
+    if (new_thread_pool_size < 1) {
+      new_thread_pool_size = 1;
+    }
+    db_stress_env->SetBackgroundThreads(new_thread_pool_size,
+                                        rocksdb::Env::Priority::LOW);
+    // Sleep up to 3 seconds
+    db_stress_env->SleepForMicroseconds(
+        thread->rand.Next() % FLAGS_compaction_thread_pool_adjust_interval *
+            1000 +
+        1);
+  }
+}
+
+void DbVerificationThread(void* v) {
+  assert(FLAGS_continuous_verification_interval > 0);
+  auto* thread = reinterpret_cast<ThreadState*>(v);
+  SharedState* shared = thread->shared;
+  StressTest* stress_test = shared->GetStressTest();
+  assert(stress_test != nullptr);
+  while (true) {
+    {
+      MutexLock l(shared->GetMutex());
+      if (shared->ShouldStopBgThread()) {
+        shared->IncBgThreadsFinished();
+        if (shared->BgThreadsFinished()) {
+          shared->GetCondVar()->SignalAll();
+        }
+        return;
+      }
+    }
+    if (!shared->HasVerificationFailedYet()) {
+      stress_test->ContinuouslyVerifyDb(thread);
+    }
+    db_stress_env->SleepForMicroseconds(
+        thread->rand.Next() % FLAGS_continuous_verification_interval * 1000 +
+        1);
+  }
+}
+
+void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz) {
+  if (!FLAGS_verbose) {
+    return;
+  }
+  std::string tmp;
+  tmp.reserve(sz * 2 + 16);
+  char buf[4];
+  for (size_t i = 0; i < sz; i++) {
+    snprintf(buf, 4, "%X", value[i]);
+    tmp.append(buf);
+  }
+  fprintf(stdout, "[CF %d] %" PRIi64 " == > (%" ROCKSDB_PRIszt ") %s\n", cf,
+          key, sz, tmp.c_str());
+}
+
+// Note that if hot_key_alpha != 0, it generates the key based on Zipfian
+// distribution. Keys are randomly scattered to [0, FLAGS_max_key]. It does
+// not ensure the order of the keys being generated and the keys does not have
+// the active range which is related to FLAGS_active_width.
+int64_t GenerateOneKey(ThreadState* thread, uint64_t iteration) {
+  const double completed_ratio =
+      static_cast<double>(iteration) / FLAGS_ops_per_thread;
+  const int64_t base_key = static_cast<int64_t>(
+      completed_ratio * (FLAGS_max_key - FLAGS_active_width));
+  int64_t rand_seed = base_key + thread->rand.Next() % FLAGS_active_width;
+  int64_t cur_key = rand_seed;
+  if (FLAGS_hot_key_alpha != 0) {
+    // If set the Zipfian distribution Alpha to non 0, use Zipfian
+    double float_rand =
+        (static_cast<double>(thread->rand.Next() % FLAGS_max_key)) /
+        FLAGS_max_key;
+    cur_key = GetOneHotKeyID(float_rand, FLAGS_max_key);
+  }
+  return cur_key;
+}
+
+// Note that if hot_key_alpha != 0, it generates the key based on Zipfian
+// distribution. Keys being generated are in random order.
+// If user want to generate keys based on uniform distribution, user needs to
+// set hot_key_alpha == 0. It will generate the random keys in increasing
+// order in the key array (ensure key[i] >= key[i+1]) and constrained in a
+// range related to FLAGS_active_width.
+std::vector<int64_t> GenerateNKeys(ThreadState* thread, int num_keys,
+                                   uint64_t iteration) {
+  const double completed_ratio =
+      static_cast<double>(iteration) / FLAGS_ops_per_thread;
+  const int64_t base_key = static_cast<int64_t>(
+      completed_ratio * (FLAGS_max_key - FLAGS_active_width));
+  std::vector<int64_t> keys;
+  keys.reserve(num_keys);
+  int64_t next_key = base_key + thread->rand.Next() % FLAGS_active_width;
+  keys.push_back(next_key);
+  for (int i = 1; i < num_keys; ++i) {
+    // Generate the key follows zipfian distribution
+    if (FLAGS_hot_key_alpha != 0) {
+      double float_rand =
+          (static_cast<double>(thread->rand.Next() % FLAGS_max_key)) /
+          FLAGS_max_key;
+      next_key = GetOneHotKeyID(float_rand, FLAGS_max_key);
+    } else {
+      // This may result in some duplicate keys
+      next_key = next_key + thread->rand.Next() %
+                                (FLAGS_active_width - (next_key - base_key));
+    }
+    keys.push_back(next_key);
+  }
+  return keys;
+}
+
+size_t GenerateValue(uint32_t rand, char* v, size_t max_sz) {
+  size_t value_sz =
+      ((rand % kRandomValueMaxFactor) + 1) * FLAGS_value_size_mult;
+  assert(value_sz <= max_sz && value_sz >= sizeof(uint32_t));
+  (void)max_sz;
+  *((uint32_t*)v) = rand;
+  for (size_t i = sizeof(uint32_t); i < value_sz; i++) {
+    v[i] = (char)(rand ^ i);
+  }
+  v[value_sz] = '\0';
+  return value_sz;  // the size of the value set.
+}
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_common.h

@@ -0,0 +1,415 @@
+//  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.
+//
+// The test uses an array to compare against values written to the database.
+// Keys written to the array are in 1:1 correspondence to the actual values in
+// the database according to the formula in the function GenerateValue.
+
+// Space is reserved in the array from 0 to FLAGS_max_key and values are
+// randomly written/deleted/read from those positions. During verification we
+// compare all the positions in the array. To shorten/elongate the running
+// time, you could change the settings: FLAGS_max_key, FLAGS_ops_per_thread,
+// (sometimes also FLAGS_threads).
+//
+// NOTE that if FLAGS_test_batches_snapshots is set, the test will have
+// different behavior. See comment of the flag for details.
+
+#ifdef GFLAGS
+#pragma once
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <algorithm>
+#include <array>
+#include <chrono>
+#include <cinttypes>
+#include <exception>
+#include <queue>
+#include <thread>
+
+#include "db/db_impl/db_impl.h"
+#include "db/version_set.h"
+#include "db_stress_tool/db_stress_env_wrapper.h"
+#include "db_stress_tool/db_stress_listener.h"
+#include "db_stress_tool/db_stress_shared_state.h"
+#include "db_stress_tool/db_stress_test_base.h"
+#include "hdfs/env_hdfs.h"
+#include "logging/logging.h"
+#include "monitoring/histogram.h"
+#include "options/options_helper.h"
+#include "port/port.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/env.h"
+#include "rocksdb/slice.h"
+#include "rocksdb/slice_transform.h"
+#include "rocksdb/statistics.h"
+#include "rocksdb/utilities/backupable_db.h"
+#include "rocksdb/utilities/checkpoint.h"
+#include "rocksdb/utilities/db_ttl.h"
+#include "rocksdb/utilities/debug.h"
+#include "rocksdb/utilities/options_util.h"
+#include "rocksdb/utilities/transaction.h"
+#include "rocksdb/utilities/transaction_db.h"
+#include "rocksdb/write_batch.h"
+#include "util/coding.h"
+#include "util/compression.h"
+#include "util/crc32c.h"
+#include "util/gflags_compat.h"
+#include "util/mutexlock.h"
+#include "util/random.h"
+#include "util/string_util.h"
+#include "utilities/blob_db/blob_db.h"
+// SyncPoint is not supported in Released Windows Mode.
+#if !(defined NDEBUG) || !defined(OS_WIN)
+#include "test_util/sync_point.h"
+#endif  // !(defined NDEBUG) || !defined(OS_WIN)
+#include "test_util/testutil.h"
+
+#include "utilities/merge_operators.h"
+
+using GFLAGS_NAMESPACE::ParseCommandLineFlags;
+using GFLAGS_NAMESPACE::RegisterFlagValidator;
+using GFLAGS_NAMESPACE::SetUsageMessage;
+
+DECLARE_uint64(seed);
+DECLARE_bool(read_only);
+DECLARE_int64(max_key);
+DECLARE_double(hot_key_alpha);
+DECLARE_int32(column_families);
+DECLARE_string(options_file);
+DECLARE_int64(active_width);
+DECLARE_bool(test_batches_snapshots);
+DECLARE_bool(atomic_flush);
+DECLARE_bool(test_cf_consistency);
+DECLARE_int32(threads);
+DECLARE_int32(ttl);
+DECLARE_int32(value_size_mult);
+DECLARE_int32(compaction_readahead_size);
+DECLARE_bool(enable_pipelined_write);
+DECLARE_bool(verify_before_write);
+DECLARE_bool(histogram);
+DECLARE_bool(destroy_db_initially);
+DECLARE_bool(verbose);
+DECLARE_bool(progress_reports);
+DECLARE_uint64(db_write_buffer_size);
+DECLARE_int32(write_buffer_size);
+DECLARE_int32(max_write_buffer_number);
+DECLARE_int32(min_write_buffer_number_to_merge);
+DECLARE_int32(max_write_buffer_number_to_maintain);
+DECLARE_int64(max_write_buffer_size_to_maintain);
+DECLARE_double(memtable_prefix_bloom_size_ratio);
+DECLARE_bool(memtable_whole_key_filtering);
+DECLARE_int32(open_files);
+DECLARE_int64(compressed_cache_size);
+DECLARE_int32(compaction_style);
+DECLARE_int32(level0_file_num_compaction_trigger);
+DECLARE_int32(level0_slowdown_writes_trigger);
+DECLARE_int32(level0_stop_writes_trigger);
+DECLARE_int32(block_size);
+DECLARE_int32(format_version);
+DECLARE_int32(index_block_restart_interval);
+DECLARE_int32(max_background_compactions);
+DECLARE_int32(num_bottom_pri_threads);
+DECLARE_int32(compaction_thread_pool_adjust_interval);
+DECLARE_int32(compaction_thread_pool_variations);
+DECLARE_int32(max_background_flushes);
+DECLARE_int32(universal_size_ratio);
+DECLARE_int32(universal_min_merge_width);
+DECLARE_int32(universal_max_merge_width);
+DECLARE_int32(universal_max_size_amplification_percent);
+DECLARE_int32(clear_column_family_one_in);
+DECLARE_int32(get_live_files_and_wal_files_one_in);
+DECLARE_int32(set_options_one_in);
+DECLARE_int32(set_in_place_one_in);
+DECLARE_int64(cache_size);
+DECLARE_bool(cache_index_and_filter_blocks);
+DECLARE_bool(use_clock_cache);
+DECLARE_uint64(subcompactions);
+DECLARE_uint64(periodic_compaction_seconds);
+DECLARE_uint64(compaction_ttl);
+DECLARE_bool(allow_concurrent_memtable_write);
+DECLARE_bool(enable_write_thread_adaptive_yield);
+DECLARE_int32(reopen);
+DECLARE_double(bloom_bits);
+DECLARE_bool(use_block_based_filter);
+DECLARE_bool(partition_filters);
+DECLARE_int32(index_type);
+DECLARE_string(db);
+DECLARE_string(secondaries_base);
+DECLARE_bool(test_secondary);
+DECLARE_string(expected_values_path);
+DECLARE_bool(verify_checksum);
+DECLARE_bool(mmap_read);
+DECLARE_bool(mmap_write);
+DECLARE_bool(use_direct_reads);
+DECLARE_bool(use_direct_io_for_flush_and_compaction);
+DECLARE_bool(statistics);
+DECLARE_bool(sync);
+DECLARE_bool(use_fsync);
+DECLARE_int32(kill_random_test);
+DECLARE_string(kill_prefix_blacklist);
+DECLARE_bool(disable_wal);
+DECLARE_uint64(recycle_log_file_num);
+DECLARE_int64(target_file_size_base);
+DECLARE_int32(target_file_size_multiplier);
+DECLARE_uint64(max_bytes_for_level_base);
+DECLARE_double(max_bytes_for_level_multiplier);
+DECLARE_int32(range_deletion_width);
+DECLARE_uint64(rate_limiter_bytes_per_sec);
+DECLARE_bool(rate_limit_bg_reads);
+DECLARE_bool(use_txn);
+DECLARE_uint64(txn_write_policy);
+DECLARE_bool(unordered_write);
+DECLARE_int32(backup_one_in);
+DECLARE_int32(checkpoint_one_in);
+DECLARE_int32(ingest_external_file_one_in);
+DECLARE_int32(ingest_external_file_width);
+DECLARE_int32(compact_files_one_in);
+DECLARE_int32(compact_range_one_in);
+DECLARE_int32(flush_one_in);
+DECLARE_int32(pause_background_one_in);
+DECLARE_int32(compact_range_width);
+DECLARE_int32(acquire_snapshot_one_in);
+DECLARE_bool(compare_full_db_state_snapshot);
+DECLARE_uint64(snapshot_hold_ops);
+DECLARE_bool(long_running_snapshots);
+DECLARE_bool(use_multiget);
+DECLARE_int32(readpercent);
+DECLARE_int32(prefixpercent);
+DECLARE_int32(writepercent);
+DECLARE_int32(delpercent);
+DECLARE_int32(delrangepercent);
+DECLARE_int32(nooverwritepercent);
+DECLARE_int32(iterpercent);
+DECLARE_uint64(num_iterations);
+DECLARE_string(compression_type);
+DECLARE_string(bottommost_compression_type);
+DECLARE_int32(compression_max_dict_bytes);
+DECLARE_int32(compression_zstd_max_train_bytes);
+DECLARE_string(checksum_type);
+DECLARE_string(hdfs);
+DECLARE_string(env_uri);
+DECLARE_uint64(ops_per_thread);
+DECLARE_uint64(log2_keys_per_lock);
+DECLARE_uint64(max_manifest_file_size);
+DECLARE_bool(in_place_update);
+DECLARE_int32(secondary_catch_up_one_in);
+DECLARE_string(memtablerep);
+DECLARE_int32(prefix_size);
+DECLARE_bool(use_merge);
+DECLARE_bool(use_full_merge_v1);
+DECLARE_int32(sync_wal_one_in);
+DECLARE_bool(avoid_unnecessary_blocking_io);
+DECLARE_bool(write_dbid_to_manifest);
+DECLARE_uint64(max_write_batch_group_size_bytes);
+DECLARE_bool(level_compaction_dynamic_level_bytes);
+DECLARE_int32(verify_checksum_one_in);
+DECLARE_int32(verify_db_one_in);
+DECLARE_int32(continuous_verification_interval);
+
+#ifndef ROCKSDB_LITE
+DECLARE_bool(use_blob_db);
+DECLARE_uint64(blob_db_min_blob_size);
+DECLARE_uint64(blob_db_bytes_per_sync);
+DECLARE_uint64(blob_db_file_size);
+DECLARE_bool(blob_db_enable_gc);
+DECLARE_double(blob_db_gc_cutoff);
+#endif  // !ROCKSDB_LITE
+DECLARE_int32(approximate_size_one_in);
+
+const long KB = 1024;
+const int kRandomValueMaxFactor = 3;
+const int kValueMaxLen = 100;
+
+// wrapped posix or hdfs environment
+extern rocksdb::DbStressEnvWrapper* db_stress_env;
+
+extern enum rocksdb::CompressionType compression_type_e;
+extern enum rocksdb::CompressionType bottommost_compression_type_e;
+extern enum rocksdb::ChecksumType checksum_type_e;
+
+enum RepFactory { kSkipList, kHashSkipList, kVectorRep };
+
+inline enum RepFactory StringToRepFactory(const char* ctype) {
+  assert(ctype);
+
+  if (!strcasecmp(ctype, "skip_list"))
+    return kSkipList;
+  else if (!strcasecmp(ctype, "prefix_hash"))
+    return kHashSkipList;
+  else if (!strcasecmp(ctype, "vector"))
+    return kVectorRep;
+
+  fprintf(stdout, "Cannot parse memreptable %s\n", ctype);
+  return kSkipList;
+}
+
+extern enum RepFactory FLAGS_rep_factory;
+
+namespace rocksdb {
+inline enum rocksdb::CompressionType StringToCompressionType(
+    const char* ctype) {
+  assert(ctype);
+
+  rocksdb::CompressionType ret_compression_type;
+
+  if (!strcasecmp(ctype, "disable")) {
+    ret_compression_type = rocksdb::kDisableCompressionOption;
+  } else if (!strcasecmp(ctype, "none")) {
+    ret_compression_type = rocksdb::kNoCompression;
+  } else if (!strcasecmp(ctype, "snappy")) {
+    ret_compression_type = rocksdb::kSnappyCompression;
+  } else if (!strcasecmp(ctype, "zlib")) {
+    ret_compression_type = rocksdb::kZlibCompression;
+  } else if (!strcasecmp(ctype, "bzip2")) {
+    ret_compression_type = rocksdb::kBZip2Compression;
+  } else if (!strcasecmp(ctype, "lz4")) {
+    ret_compression_type = rocksdb::kLZ4Compression;
+  } else if (!strcasecmp(ctype, "lz4hc")) {
+    ret_compression_type = rocksdb::kLZ4HCCompression;
+  } else if (!strcasecmp(ctype, "xpress")) {
+    ret_compression_type = rocksdb::kXpressCompression;
+  } else if (!strcasecmp(ctype, "zstd")) {
+    ret_compression_type = rocksdb::kZSTD;
+  } else {
+    fprintf(stderr, "Cannot parse compression type '%s'\n", ctype);
+    ret_compression_type = rocksdb::kSnappyCompression;  // default value
+  }
+  if (ret_compression_type != rocksdb::kDisableCompressionOption &&
+      !CompressionTypeSupported(ret_compression_type)) {
+    // Use no compression will be more portable but considering this is
+    // only a stress test and snappy is widely available. Use snappy here.
+    ret_compression_type = rocksdb::kSnappyCompression;
+  }
+  return ret_compression_type;
+}
+
+inline enum rocksdb::ChecksumType StringToChecksumType(const char* ctype) {
+  assert(ctype);
+  auto iter = rocksdb::checksum_type_string_map.find(ctype);
+  if (iter != rocksdb::checksum_type_string_map.end()) {
+    return iter->second;
+  }
+  fprintf(stderr, "Cannot parse checksum type '%s'\n", ctype);
+  return rocksdb::kCRC32c;
+}
+
+inline std::string ChecksumTypeToString(rocksdb::ChecksumType ctype) {
+  auto iter = std::find_if(
+      rocksdb::checksum_type_string_map.begin(),
+      rocksdb::checksum_type_string_map.end(),
+      [&](const std::pair<std::string, rocksdb::ChecksumType>&
+              name_and_enum_val) { return name_and_enum_val.second == ctype; });
+  assert(iter != rocksdb::checksum_type_string_map.end());
+  return iter->first;
+}
+
+inline std::vector<std::string> SplitString(std::string src) {
+  std::vector<std::string> ret;
+  if (src.empty()) {
+    return ret;
+  }
+  size_t pos = 0;
+  size_t pos_comma;
+  while ((pos_comma = src.find(',', pos)) != std::string::npos) {
+    ret.push_back(src.substr(pos, pos_comma - pos));
+    pos = pos_comma + 1;
+  }
+  ret.push_back(src.substr(pos, src.length()));
+  return ret;
+}
+
+#ifdef _MSC_VER
+#pragma warning(push)
+// truncation of constant value on static_cast
+#pragma warning(disable : 4309)
+#endif
+inline bool GetNextPrefix(const rocksdb::Slice& src, std::string* v) {
+  std::string ret = src.ToString();
+  for (int i = static_cast<int>(ret.size()) - 1; i >= 0; i--) {
+    if (ret[i] != static_cast<char>(255)) {
+      ret[i] = ret[i] + 1;
+      break;
+    } else if (i != 0) {
+      ret[i] = 0;
+    } else {
+      // all FF. No next prefix
+      return false;
+    }
+  }
+  *v = ret;
+  return true;
+}
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+// convert long to a big-endian slice key
+extern inline std::string Key(int64_t val) {
+  std::string little_endian_key;
+  std::string big_endian_key;
+  PutFixed64(&little_endian_key, val);
+  assert(little_endian_key.size() == sizeof(val));
+  big_endian_key.resize(sizeof(val));
+  for (size_t i = 0; i < sizeof(val); ++i) {
+    big_endian_key[i] = little_endian_key[sizeof(val) - 1 - i];
+  }
+  return big_endian_key;
+}
+
+extern inline bool GetIntVal(std::string big_endian_key, uint64_t* key_p) {
+  unsigned int size_key = sizeof(*key_p);
+  assert(big_endian_key.size() == size_key);
+  std::string little_endian_key;
+  little_endian_key.resize(size_key);
+  for (size_t i = 0; i < size_key; ++i) {
+    little_endian_key[i] = big_endian_key[size_key - 1 - i];
+  }
+  Slice little_endian_slice = Slice(little_endian_key);
+  return GetFixed64(&little_endian_slice, key_p);
+}
+
+extern inline std::string StringToHex(const std::string& str) {
+  std::string result = "0x";
+  result.append(Slice(str).ToString(true));
+  return result;
+}
+
+// Unified output format for double parameters
+extern inline std::string FormatDoubleParam(double param) {
+  return std::to_string(param);
+}
+
+// Make sure that double parameter is a value we can reproduce by
+// re-inputting the value printed.
+extern inline void SanitizeDoubleParam(double* param) {
+  *param = std::atof(FormatDoubleParam(*param).c_str());
+}
+
+extern void PoolSizeChangeThread(void* v);
+
+extern void DbVerificationThread(void* v);
+
+extern void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz);
+
+extern int64_t GenerateOneKey(ThreadState* thread, uint64_t iteration);
+
+extern std::vector<int64_t> GenerateNKeys(ThreadState* thread, int num_keys,
+                                          uint64_t iteration);
+
+extern size_t GenerateValue(uint32_t rand, char* v, size_t max_sz);
+
+extern StressTest* CreateCfConsistencyStressTest();
+extern StressTest* CreateBatchedOpsStressTest();
+extern StressTest* CreateNonBatchedOpsStressTest();
+extern void InitializeHotKeyGenerator(double alpha);
+extern int64_t GetOneHotKeyID(double rand_seed, int64_t max_key);
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_driver.cc

@@ -0,0 +1,163 @@
+//  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.
+//
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_common.h"
+
+namespace rocksdb {
+void ThreadBody(void* v) {
+  ThreadState* thread = reinterpret_cast<ThreadState*>(v);
+  SharedState* shared = thread->shared;
+
+  if (shared->ShouldVerifyAtBeginning()) {
+    thread->shared->GetStressTest()->VerifyDb(thread);
+  }
+  {
+    MutexLock l(shared->GetMutex());
+    shared->IncInitialized();
+    if (shared->AllInitialized()) {
+      shared->GetCondVar()->SignalAll();
+    }
+    while (!shared->Started()) {
+      shared->GetCondVar()->Wait();
+    }
+  }
+  thread->shared->GetStressTest()->OperateDb(thread);
+
+  {
+    MutexLock l(shared->GetMutex());
+    shared->IncOperated();
+    if (shared->AllOperated()) {
+      shared->GetCondVar()->SignalAll();
+    }
+    while (!shared->VerifyStarted()) {
+      shared->GetCondVar()->Wait();
+    }
+  }
+
+  thread->shared->GetStressTest()->VerifyDb(thread);
+
+  {
+    MutexLock l(shared->GetMutex());
+    shared->IncDone();
+    if (shared->AllDone()) {
+      shared->GetCondVar()->SignalAll();
+    }
+  }
+}
+
+bool RunStressTest(StressTest* stress) {
+  stress->InitDb();
+
+  SharedState shared(db_stress_env, stress);
+  if (FLAGS_read_only) {
+    stress->InitReadonlyDb(&shared);
+  }
+
+  uint32_t n = shared.GetNumThreads();
+
+  uint64_t now = db_stress_env->NowMicros();
+  fprintf(stdout, "%s Initializing worker threads\n",
+          db_stress_env->TimeToString(now / 1000000).c_str());
+  std::vector<ThreadState*> threads(n);
+  for (uint32_t i = 0; i < n; i++) {
+    threads[i] = new ThreadState(i, &shared);
+    db_stress_env->StartThread(ThreadBody, threads[i]);
+  }
+  ThreadState bg_thread(0, &shared);
+  if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
+    db_stress_env->StartThread(PoolSizeChangeThread, &bg_thread);
+  }
+  ThreadState continuous_verification_thread(0, &shared);
+  if (FLAGS_continuous_verification_interval > 0) {
+    db_stress_env->StartThread(DbVerificationThread,
+                               &continuous_verification_thread);
+  }
+
+  // Each thread goes through the following states:
+  // initializing -> wait for others to init -> read/populate/depopulate
+  // wait for others to operate -> verify -> done
+
+  {
+    MutexLock l(shared.GetMutex());
+    while (!shared.AllInitialized()) {
+      shared.GetCondVar()->Wait();
+    }
+    if (shared.ShouldVerifyAtBeginning()) {
+      if (shared.HasVerificationFailedYet()) {
+        fprintf(stderr, "Crash-recovery verification failed :(\n");
+      } else {
+        fprintf(stdout, "Crash-recovery verification passed :)\n");
+      }
+    }
+
+    now = db_stress_env->NowMicros();
+    fprintf(stdout, "%s Starting database operations\n",
+            db_stress_env->TimeToString(now / 1000000).c_str());
+
+    shared.SetStart();
+    shared.GetCondVar()->SignalAll();
+    while (!shared.AllOperated()) {
+      shared.GetCondVar()->Wait();
+    }
+
+    now = db_stress_env->NowMicros();
+    if (FLAGS_test_batches_snapshots) {
+      fprintf(stdout, "%s Limited verification already done during gets\n",
+              db_stress_env->TimeToString((uint64_t)now / 1000000).c_str());
+    } else {
+      fprintf(stdout, "%s Starting verification\n",
+              db_stress_env->TimeToString((uint64_t)now / 1000000).c_str());
+    }
+
+    shared.SetStartVerify();
+    shared.GetCondVar()->SignalAll();
+    while (!shared.AllDone()) {
+      shared.GetCondVar()->Wait();
+    }
+  }
+
+  for (unsigned int i = 1; i < n; i++) {
+    threads[0]->stats.Merge(threads[i]->stats);
+  }
+  threads[0]->stats.Report("Stress Test");
+
+  for (unsigned int i = 0; i < n; i++) {
+    delete threads[i];
+    threads[i] = nullptr;
+  }
+  now = db_stress_env->NowMicros();
+  if (!FLAGS_test_batches_snapshots && !shared.HasVerificationFailedYet()) {
+    fprintf(stdout, "%s Verification successful\n",
+            db_stress_env->TimeToString(now / 1000000).c_str());
+  }
+  stress->PrintStatistics();
+
+  if (FLAGS_compaction_thread_pool_adjust_interval > 0 ||
+      FLAGS_continuous_verification_interval > 0) {
+    MutexLock l(shared.GetMutex());
+    shared.SetShouldStopBgThread();
+    while (!shared.BgThreadsFinished()) {
+      shared.GetCondVar()->Wait();
+    }
+  }
+
+  if (!stress->VerifySecondaries()) {
+    return false;
+  }
+
+  if (shared.HasVerificationFailedYet()) {
+    fprintf(stderr, "Verification failed :(\n");
+    return false;
+  }
+  return true;
+}
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_driver.h

@@ -0,0 +1,17 @@
+//  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.
+
+#ifdef GFLAGS
+#pragma once
+#include "db_stress_tool/db_stress_test_base.h"
+namespace rocksdb {
+extern void ThreadBody(void* /*thread_state*/);
+extern bool RunStressTest(StressTest*);
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_env_wrapper.h

@@ -0,0 +1,36 @@
+//  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.
+
+#ifdef GFLAGS
+#pragma once
+#include "db_stress_tool/db_stress_common.h"
+
+namespace rocksdb {
+class DbStressEnvWrapper : public EnvWrapper {
+ public:
+  explicit DbStressEnvWrapper(Env* t) : EnvWrapper(t) {}
+
+  Status DeleteFile(const std::string& f) override {
+    // We determine whether it is a manifest file by searching a strong,
+    // so that there will be false positive if the directory path contains the
+    // keyword but it is unlikely.
+    // Checkpoint directory needs to be exempted.
+    if (!if_preserve_all_manifests ||
+        f.find("MANIFEST-") == std::string::npos ||
+        f.find("checkpoint") != std::string::npos) {
+      return target()->DeleteFile(f);
+    }
+    return Status::OK();
+  }
+
+  // If true, all manifest files will not be delted in DeleteFile().
+  bool if_preserve_all_manifests = true;
+};
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_gflags.cc

@@ -0,0 +1,632 @@
+//  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.
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_common.h"
+
+static bool ValidateUint32Range(const char* flagname, uint64_t value) {
+  if (value > std::numeric_limits<uint32_t>::max()) {
+    fprintf(stderr, "Invalid value for --%s: %lu, overflow\n", flagname,
+            (unsigned long)value);
+    return false;
+  }
+  return true;
+}
+
+DEFINE_uint64(seed, 2341234, "Seed for PRNG");
+static const bool FLAGS_seed_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_seed, &ValidateUint32Range);
+
+DEFINE_bool(read_only, false, "True if open DB in read-only mode during tests");
+
+DEFINE_int64(max_key, 1 * KB * KB,
+             "Max number of key/values to place in database");
+
+DEFINE_int32(column_families, 10, "Number of column families");
+
+DEFINE_double(
+    hot_key_alpha, 0,
+    "Use Zipfian distribution to generate the key "
+    "distribution. If it is not specified, write path will use random "
+    "distribution to generate the keys. The parameter is [0, double_max]). "
+    "However, the larger alpha is, the more shewed will be. If alpha is "
+    "larger than 2, it is likely that only 1 key will be accessed. The "
+    "Recommended value is [0.8-1.5]. The distribution is also related to "
+    "max_key and total iterations of generating the hot key. ");
+
+DEFINE_string(
+    options_file, "",
+    "The path to a RocksDB options file.  If specified, then db_stress will "
+    "run with the RocksDB options in the default column family of the "
+    "specified options file. Note that, when an options file is provided, "
+    "db_stress will ignore the flag values for all options that may be passed "
+    "via options file.");
+
+DEFINE_int64(
+    active_width, 0,
+    "Number of keys in active span of the key-range at any given time. The "
+    "span begins with its left endpoint at key 0, gradually moves rightwards, "
+    "and ends with its right endpoint at max_key. If set to 0, active_width "
+    "will be sanitized to be equal to max_key.");
+
+// TODO(noetzli) Add support for single deletes
+DEFINE_bool(test_batches_snapshots, false,
+            "If set, the test uses MultiGet(), MultiPut() and MultiDelete()"
+            " which read/write/delete multiple keys in a batch. In this mode,"
+            " we do not verify db content by comparing the content with the "
+            "pre-allocated array. Instead, we do partial verification inside"
+            " MultiGet() by checking various values in a batch. Benefit of"
+            " this mode:\n"
+            "\t(a) No need to acquire mutexes during writes (less cache "
+            "flushes in multi-core leading to speed up)\n"
+            "\t(b) No long validation at the end (more speed up)\n"
+            "\t(c) Test snapshot and atomicity of batch writes");
+
+DEFINE_bool(atomic_flush, false,
+            "If set, enables atomic flush in the options.\n");
+
+DEFINE_bool(test_cf_consistency, false,
+            "If set, runs the stress test dedicated to verifying writes to "
+            "multiple column families are consistent. Setting this implies "
+            "`atomic_flush=true` is set true if `disable_wal=false`.\n");
+
+DEFINE_int32(threads, 32, "Number of concurrent threads to run.");
+
+DEFINE_int32(ttl, -1,
+             "Opens the db with this ttl value if this is not -1. "
+             "Carefully specify a large value such that verifications on "
+             "deleted values don't fail");
+
+DEFINE_int32(value_size_mult, 8,
+             "Size of value will be this number times rand_int(1,3) bytes");
+
+DEFINE_int32(compaction_readahead_size, 0, "Compaction readahead size");
+
+DEFINE_bool(enable_pipelined_write, false, "Pipeline WAL/memtable writes");
+
+DEFINE_bool(verify_before_write, false, "Verify before write");
+
+DEFINE_bool(histogram, false, "Print histogram of operation timings");
+
+DEFINE_bool(destroy_db_initially, true,
+            "Destroys the database dir before start if this is true");
+
+DEFINE_bool(verbose, false, "Verbose");
+
+DEFINE_bool(progress_reports, true,
+            "If true, db_stress will report number of finished operations");
+
+DEFINE_uint64(db_write_buffer_size, rocksdb::Options().db_write_buffer_size,
+              "Number of bytes to buffer in all memtables before compacting");
+
+DEFINE_int32(write_buffer_size,
+             static_cast<int32_t>(rocksdb::Options().write_buffer_size),
+             "Number of bytes to buffer in memtable before compacting");
+
+DEFINE_int32(max_write_buffer_number,
+             rocksdb::Options().max_write_buffer_number,
+             "The number of in-memory memtables. "
+             "Each memtable is of size FLAGS_write_buffer_size.");
+
+DEFINE_int32(min_write_buffer_number_to_merge,
+             rocksdb::Options().min_write_buffer_number_to_merge,
+             "The minimum number of write buffers that will be merged together "
+             "before writing to storage. This is cheap because it is an "
+             "in-memory merge. If this feature is not enabled, then all these "
+             "write buffers are flushed to L0 as separate files and this "
+             "increases read amplification because a get request has to check "
+             "in all of these files. Also, an in-memory merge may result in "
+             "writing less data to storage if there are duplicate records in"
+             " each of these individual write buffers.");
+
+DEFINE_int32(max_write_buffer_number_to_maintain,
+             rocksdb::Options().max_write_buffer_number_to_maintain,
+             "The total maximum number of write buffers to maintain in memory "
+             "including copies of buffers that have already been flushed. "
+             "Unlike max_write_buffer_number, this parameter does not affect "
+             "flushing. This controls the minimum amount of write history "
+             "that will be available in memory for conflict checking when "
+             "Transactions are used. If this value is too low, some "
+             "transactions may fail at commit time due to not being able to "
+             "determine whether there were any write conflicts. Setting this "
+             "value to 0 will cause write buffers to be freed immediately "
+             "after they are flushed.  If this value is set to -1, "
+             "'max_write_buffer_number' will be used.");
+
+DEFINE_int64(max_write_buffer_size_to_maintain,
+             rocksdb::Options().max_write_buffer_size_to_maintain,
+             "The total maximum size of write buffers to maintain in memory "
+             "including copies of buffers that have already been flushed. "
+             "Unlike max_write_buffer_number, this parameter does not affect "
+             "flushing. This controls the minimum amount of write history "
+             "that will be available in memory for conflict checking when "
+             "Transactions are used. If this value is too low, some "
+             "transactions may fail at commit time due to not being able to "
+             "determine whether there were any write conflicts. Setting this "
+             "value to 0 will cause write buffers to be freed immediately "
+             "after they are flushed.  If this value is set to -1, "
+             "'max_write_buffer_number' will be used.");
+
+DEFINE_double(memtable_prefix_bloom_size_ratio,
+              rocksdb::Options().memtable_prefix_bloom_size_ratio,
+              "creates prefix blooms for memtables, each with size "
+              "`write_buffer_size * memtable_prefix_bloom_size_ratio`.");
+
+DEFINE_bool(memtable_whole_key_filtering,
+            rocksdb::Options().memtable_whole_key_filtering,
+            "Enable whole key filtering in memtables.");
+
+DEFINE_int32(open_files, rocksdb::Options().max_open_files,
+             "Maximum number of files to keep open at the same time "
+             "(use default if == 0)");
+
+DEFINE_int64(compressed_cache_size, -1,
+             "Number of bytes to use as a cache of compressed data."
+             " Negative means use default settings.");
+
+DEFINE_int32(compaction_style, rocksdb::Options().compaction_style, "");
+
+DEFINE_int32(level0_file_num_compaction_trigger,
+             rocksdb::Options().level0_file_num_compaction_trigger,
+             "Level0 compaction start trigger");
+
+DEFINE_int32(level0_slowdown_writes_trigger,
+             rocksdb::Options().level0_slowdown_writes_trigger,
+             "Number of files in level-0 that will slow down writes");
+
+DEFINE_int32(level0_stop_writes_trigger,
+             rocksdb::Options().level0_stop_writes_trigger,
+             "Number of files in level-0 that will trigger put stop.");
+
+DEFINE_int32(block_size,
+             static_cast<int32_t>(rocksdb::BlockBasedTableOptions().block_size),
+             "Number of bytes in a block.");
+
+DEFINE_int32(
+    format_version,
+    static_cast<int32_t>(rocksdb::BlockBasedTableOptions().format_version),
+    "Format version of SST files.");
+
+DEFINE_int32(index_block_restart_interval,
+             rocksdb::BlockBasedTableOptions().index_block_restart_interval,
+             "Number of keys between restart points "
+             "for delta encoding of keys in index block.");
+
+DEFINE_int32(max_background_compactions,
+             rocksdb::Options().max_background_compactions,
+             "The maximum number of concurrent background compactions "
+             "that can occur in parallel.");
+
+DEFINE_int32(num_bottom_pri_threads, 0,
+             "The number of threads in the bottom-priority thread pool (used "
+             "by universal compaction only).");
+
+DEFINE_int32(compaction_thread_pool_adjust_interval, 0,
+             "The interval (in milliseconds) to adjust compaction thread pool "
+             "size. Don't change it periodically if the value is 0.");
+
+DEFINE_int32(compaction_thread_pool_variations, 2,
+             "Range of background thread pool size variations when adjusted "
+             "periodically.");
+
+DEFINE_int32(max_background_flushes, rocksdb::Options().max_background_flushes,
+             "The maximum number of concurrent background flushes "
+             "that can occur in parallel.");
+
+DEFINE_int32(universal_size_ratio, 0,
+             "The ratio of file sizes that trigger"
+             " compaction in universal style");
+
+DEFINE_int32(universal_min_merge_width, 0,
+             "The minimum number of files to "
+             "compact in universal style compaction");
+
+DEFINE_int32(universal_max_merge_width, 0,
+             "The max number of files to compact"
+             " in universal style compaction");
+
+DEFINE_int32(universal_max_size_amplification_percent, 0,
+             "The max size amplification for universal style compaction");
+
+DEFINE_int32(clear_column_family_one_in, 1000000,
+             "With a chance of 1/N, delete a column family and then recreate "
+             "it again. If N == 0, never drop/create column families. "
+             "When test_batches_snapshots is true, this flag has no effect");
+
+DEFINE_int32(get_live_files_and_wal_files_one_in, 1000000,
+             "With a chance of 1/N, call GetLiveFiles, GetSortedWalFiles "
+             "and GetCurrentWalFile to verify if it returns correctly. If "
+             "N == 0, never call the three interfaces.");
+
+DEFINE_int32(set_options_one_in, 0,
+             "With a chance of 1/N, change some random options");
+
+DEFINE_int32(set_in_place_one_in, 0,
+             "With a chance of 1/N, toggle in place support option");
+
+DEFINE_int64(cache_size, 2LL * KB * KB * KB,
+             "Number of bytes to use as a cache of uncompressed data.");
+
+DEFINE_bool(cache_index_and_filter_blocks, false,
+            "True if indexes/filters should be cached in block cache.");
+
+DEFINE_bool(use_clock_cache, false,
+            "Replace default LRU block cache with clock cache.");
+
+DEFINE_uint64(subcompactions, 1,
+              "Maximum number of subcompactions to divide L0-L1 compactions "
+              "into.");
+
+DEFINE_uint64(periodic_compaction_seconds, 1000,
+              "Files older than this value will be picked up for compaction.");
+
+DEFINE_uint64(compaction_ttl, 1000,
+              "Files older than TTL will be compacted to the next level.");
+
+DEFINE_bool(allow_concurrent_memtable_write, false,
+            "Allow multi-writers to update mem tables in parallel.");
+
+DEFINE_bool(enable_write_thread_adaptive_yield, true,
+            "Use a yielding spin loop for brief writer thread waits.");
+
+#ifndef ROCKSDB_LITE
+// BlobDB Options
+DEFINE_bool(use_blob_db, false, "Use BlobDB.");
+
+DEFINE_uint64(blob_db_min_blob_size,
+              rocksdb::blob_db::BlobDBOptions().min_blob_size,
+              "Smallest blob to store in a file. Blobs smaller than this "
+              "will be inlined with the key in the LSM tree.");
+
+DEFINE_uint64(blob_db_bytes_per_sync,
+              rocksdb::blob_db::BlobDBOptions().bytes_per_sync,
+              "Sync blob files once per every N bytes written.");
+
+DEFINE_uint64(blob_db_file_size,
+              rocksdb::blob_db::BlobDBOptions().blob_file_size,
+              "Target size of each blob file.");
+
+DEFINE_bool(blob_db_enable_gc,
+            rocksdb::blob_db::BlobDBOptions().enable_garbage_collection,
+            "Enable BlobDB garbage collection.");
+
+DEFINE_double(blob_db_gc_cutoff,
+              rocksdb::blob_db::BlobDBOptions().garbage_collection_cutoff,
+              "Cutoff ratio for BlobDB garbage collection.");
+#endif  // !ROCKSDB_LITE
+
+static const bool FLAGS_subcompactions_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_subcompactions, &ValidateUint32Range);
+
+static bool ValidateInt32Positive(const char* flagname, int32_t value) {
+  if (value < 0) {
+    fprintf(stderr, "Invalid value for --%s: %d, must be >=0\n", flagname,
+            value);
+    return false;
+  }
+  return true;
+}
+DEFINE_int32(reopen, 10, "Number of times database reopens");
+static const bool FLAGS_reopen_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_reopen, &ValidateInt32Positive);
+
+DEFINE_double(bloom_bits, 10,
+              "Bloom filter bits per key. "
+              "Negative means use default settings.");
+
+DEFINE_bool(use_block_based_filter, false,
+            "use block based filter"
+            "instead of full filter for block based table");
+
+DEFINE_bool(partition_filters, false,
+            "use partitioned filters "
+            "for block-based table");
+
+DEFINE_int32(
+    index_type,
+    static_cast<int32_t>(rocksdb::BlockBasedTableOptions::kBinarySearch),
+    "Type of block-based table index (see `enum IndexType` in table.h)");
+
+DEFINE_string(db, "", "Use the db with the following name.");
+
+DEFINE_string(secondaries_base, "",
+              "Use this path as the base path for secondary instances.");
+
+DEFINE_bool(test_secondary, false, "Test secondary instance.");
+
+DEFINE_string(
+    expected_values_path, "",
+    "File where the array of expected uint32_t values will be stored. If "
+    "provided and non-empty, the DB state will be verified against these "
+    "values after recovery. --max_key and --column_family must be kept the "
+    "same across invocations of this program that use the same "
+    "--expected_values_path.");
+
+DEFINE_bool(verify_checksum, false,
+            "Verify checksum for every block read from storage");
+
+DEFINE_bool(mmap_read, rocksdb::Options().allow_mmap_reads,
+            "Allow reads to occur via mmap-ing files");
+
+DEFINE_bool(mmap_write, rocksdb::Options().allow_mmap_writes,
+            "Allow writes to occur via mmap-ing files");
+
+DEFINE_bool(use_direct_reads, rocksdb::Options().use_direct_reads,
+            "Use O_DIRECT for reading data");
+
+DEFINE_bool(use_direct_io_for_flush_and_compaction,
+            rocksdb::Options().use_direct_io_for_flush_and_compaction,
+            "Use O_DIRECT for writing data");
+
+DEFINE_bool(statistics, false, "Create database statistics");
+
+DEFINE_bool(sync, false, "Sync all writes to disk");
+
+DEFINE_bool(use_fsync, false, "If true, issue fsync instead of fdatasync");
+
+DEFINE_int32(kill_random_test, 0,
+             "If non-zero, kill at various points in source code with "
+             "probability 1/this");
+static const bool FLAGS_kill_random_test_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_kill_random_test, &ValidateInt32Positive);
+extern int rocksdb_kill_odds;
+
+DEFINE_string(kill_prefix_blacklist, "",
+              "If non-empty, kill points with prefix in the list given will be"
+              " skipped. Items are comma-separated.");
+extern std::vector<std::string> rocksdb_kill_prefix_blacklist;
+
+DEFINE_bool(disable_wal, false, "If true, do not write WAL for write.");
+
+DEFINE_uint64(recycle_log_file_num, rocksdb::Options().recycle_log_file_num,
+              "Number of old WAL files to keep around for later recycling");
+
+DEFINE_int64(target_file_size_base, rocksdb::Options().target_file_size_base,
+             "Target level-1 file size for compaction");
+
+DEFINE_int32(target_file_size_multiplier, 1,
+             "A multiplier to compute target level-N file size (N >= 2)");
+
+DEFINE_uint64(max_bytes_for_level_base,
+              rocksdb::Options().max_bytes_for_level_base,
+              "Max bytes for level-1");
+
+DEFINE_double(max_bytes_for_level_multiplier, 2,
+              "A multiplier to compute max bytes for level-N (N >= 2)");
+
+DEFINE_int32(range_deletion_width, 10,
+             "The width of the range deletion intervals.");
+
+DEFINE_uint64(rate_limiter_bytes_per_sec, 0, "Set options.rate_limiter value.");
+
+DEFINE_bool(rate_limit_bg_reads, false,
+            "Use options.rate_limiter on compaction reads");
+
+DEFINE_bool(use_txn, false,
+            "Use TransactionDB. Currently the default write policy is "
+            "TxnDBWritePolicy::WRITE_PREPARED");
+
+DEFINE_uint64(txn_write_policy, 0,
+              "The transaction write policy. Default is "
+              "TxnDBWritePolicy::WRITE_COMMITTED. Note that this should not be "
+              "changed accross crashes.");
+
+DEFINE_bool(unordered_write, false,
+            "Turn on the unordered_write feature. This options is currently "
+            "tested only in combination with use_txn=true and "
+            "txn_write_policy=TxnDBWritePolicy::WRITE_PREPARED.");
+
+DEFINE_int32(backup_one_in, 0,
+             "If non-zero, then CreateNewBackup() will be called once for "
+             "every N operations on average.  0 indicates CreateNewBackup() "
+             "is disabled.");
+
+DEFINE_int32(checkpoint_one_in, 0,
+             "If non-zero, then CreateCheckpoint() will be called once for "
+             "every N operations on average.  0 indicates CreateCheckpoint() "
+             "is disabled.");
+
+DEFINE_int32(ingest_external_file_one_in, 0,
+             "If non-zero, then IngestExternalFile() will be called once for "
+             "every N operations on average.  0 indicates IngestExternalFile() "
+             "is disabled.");
+
+DEFINE_int32(ingest_external_file_width, 1000,
+             "The width of the ingested external files.");
+
+DEFINE_int32(compact_files_one_in, 0,
+             "If non-zero, then CompactFiles() will be called once for every N "
+             "operations on average.  0 indicates CompactFiles() is disabled.");
+
+DEFINE_int32(compact_range_one_in, 0,
+             "If non-zero, then CompactRange() will be called once for every N "
+             "operations on average.  0 indicates CompactRange() is disabled.");
+
+DEFINE_int32(flush_one_in, 0,
+             "If non-zero, then Flush() will be called once for every N ops "
+             "on average.  0 indicates calls to Flush() are disabled.");
+
+DEFINE_int32(pause_background_one_in, 0,
+             "If non-zero, then PauseBackgroundWork()+Continue will be called "
+             "once for every N ops on average.  0 disables.");
+
+DEFINE_int32(compact_range_width, 10000,
+             "The width of the ranges passed to CompactRange().");
+
+DEFINE_int32(acquire_snapshot_one_in, 0,
+             "If non-zero, then acquires a snapshot once every N operations on "
+             "average.");
+
+DEFINE_bool(compare_full_db_state_snapshot, false,
+            "If set we compare state of entire db (in one of the threads) with"
+            "each snapshot.");
+
+DEFINE_uint64(snapshot_hold_ops, 0,
+              "If non-zero, then releases snapshots N operations after they're "
+              "acquired.");
+
+DEFINE_bool(long_running_snapshots, false,
+            "If set, hold on some some snapshots for much longer time.");
+
+DEFINE_bool(use_multiget, false,
+            "If set, use the batched MultiGet API for reads");
+
+static bool ValidateInt32Percent(const char* flagname, int32_t value) {
+  if (value < 0 || value > 100) {
+    fprintf(stderr, "Invalid value for --%s: %d, 0<= pct <=100 \n", flagname,
+            value);
+    return false;
+  }
+  return true;
+}
+
+DEFINE_int32(readpercent, 10,
+             "Ratio of reads to total workload (expressed as a percentage)");
+static const bool FLAGS_readpercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_readpercent, &ValidateInt32Percent);
+
+DEFINE_int32(prefixpercent, 20,
+             "Ratio of prefix iterators to total workload (expressed as a"
+             " percentage)");
+static const bool FLAGS_prefixpercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_prefixpercent, &ValidateInt32Percent);
+
+DEFINE_int32(writepercent, 45,
+             "Ratio of writes to total workload (expressed as a percentage)");
+static const bool FLAGS_writepercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_writepercent, &ValidateInt32Percent);
+
+DEFINE_int32(delpercent, 15,
+             "Ratio of deletes to total workload (expressed as a percentage)");
+static const bool FLAGS_delpercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_delpercent, &ValidateInt32Percent);
+
+DEFINE_int32(delrangepercent, 0,
+             "Ratio of range deletions to total workload (expressed as a "
+             "percentage). Cannot be used with test_batches_snapshots");
+static const bool FLAGS_delrangepercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_delrangepercent, &ValidateInt32Percent);
+
+DEFINE_int32(nooverwritepercent, 60,
+             "Ratio of keys without overwrite to total workload (expressed as "
+             " a percentage)");
+static const bool FLAGS_nooverwritepercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_nooverwritepercent, &ValidateInt32Percent);
+
+DEFINE_int32(iterpercent, 10,
+             "Ratio of iterations to total workload"
+             " (expressed as a percentage)");
+static const bool FLAGS_iterpercent_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_iterpercent, &ValidateInt32Percent);
+
+DEFINE_uint64(num_iterations, 10, "Number of iterations per MultiIterate run");
+static const bool FLAGS_num_iterations_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_num_iterations, &ValidateUint32Range);
+
+DEFINE_string(compression_type, "snappy",
+              "Algorithm to use to compress the database");
+
+DEFINE_int32(compression_max_dict_bytes, 0,
+             "Maximum size of dictionary used to prime the compression "
+             "library.");
+
+DEFINE_int32(compression_zstd_max_train_bytes, 0,
+             "Maximum size of training data passed to zstd's dictionary "
+             "trainer.");
+
+DEFINE_string(bottommost_compression_type, "disable",
+              "Algorithm to use to compress bottommost level of the database. "
+              "\"disable\" means disabling the feature");
+
+DEFINE_string(checksum_type, "kCRC32c", "Algorithm to use to checksum blocks");
+
+DEFINE_string(hdfs, "", "Name of hdfs environment");
+
+DEFINE_string(env_uri, "",
+              "URI for env lookup. Mutually exclusive with --hdfs");
+
+DEFINE_uint64(ops_per_thread, 1200000, "Number of operations per thread.");
+static const bool FLAGS_ops_per_thread_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_ops_per_thread, &ValidateUint32Range);
+
+DEFINE_uint64(log2_keys_per_lock, 2, "Log2 of number of keys per lock");
+static const bool FLAGS_log2_keys_per_lock_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_log2_keys_per_lock, &ValidateUint32Range);
+
+DEFINE_uint64(max_manifest_file_size, 16384, "Maximum size of a MANIFEST file");
+
+DEFINE_bool(in_place_update, false, "On true, does inplace update in memtable");
+
+DEFINE_int32(secondary_catch_up_one_in, 0,
+             "If non-zero, the secondaries attemp to catch up with the primary "
+             "once for every N operations on average. 0 indicates the "
+             "secondaries do not try to catch up after open.");
+
+DEFINE_string(memtablerep, "skip_list", "");
+
+inline static bool ValidatePrefixSize(const char* flagname, int32_t value) {
+  if (value < -1 || value > 8) {
+    fprintf(stderr, "Invalid value for --%s: %d. -1 <= PrefixSize <= 8\n",
+            flagname, value);
+    return false;
+  }
+  return true;
+}
+DEFINE_int32(prefix_size, 7,
+             "Control the prefix size for HashSkipListRep. "
+             "-1 is disabled.");
+static const bool FLAGS_prefix_size_dummy __attribute__((__unused__)) =
+    RegisterFlagValidator(&FLAGS_prefix_size, &ValidatePrefixSize);
+
+DEFINE_bool(use_merge, false,
+            "On true, replaces all writes with a Merge "
+            "that behaves like a Put");
+
+DEFINE_bool(use_full_merge_v1, false,
+            "On true, use a merge operator that implement the deprecated "
+            "version of FullMerge");
+
+DEFINE_int32(sync_wal_one_in, 0,
+             "If non-zero, then SyncWAL() will be called once for every N ops "
+             "on average. 0 indicates that calls to SyncWAL() are disabled.");
+
+DEFINE_bool(avoid_unnecessary_blocking_io,
+            rocksdb::Options().avoid_unnecessary_blocking_io,
+            "If true, some expensive cleaning up operations will be moved from "
+            "user reads to high-pri background threads.");
+
+DEFINE_bool(write_dbid_to_manifest, rocksdb::Options().write_dbid_to_manifest,
+            "Write DB_ID to manifest");
+
+DEFINE_uint64(max_write_batch_group_size_bytes,
+              rocksdb::Options().max_write_batch_group_size_bytes,
+              "Max write batch group size");
+
+DEFINE_bool(level_compaction_dynamic_level_bytes,
+            rocksdb::Options().level_compaction_dynamic_level_bytes,
+            "Use dynamic level");
+
+DEFINE_int32(verify_checksum_one_in, 0,
+             "If non-zero, then DB::VerifyChecksum() will be called to do"
+             " checksum verification of all the files in the database once for"
+             " every N ops on average. 0 indicates that calls to"
+             " VerifyChecksum() are disabled.");
+DEFINE_int32(verify_db_one_in, 0,
+             "If non-zero, call VerifyDb() once for every N ops. 0 indicates "
+             "that VerifyDb() will not be called in OperateDb(). Note that "
+             "enabling this can slow down tests.");
+
+DEFINE_int32(continuous_verification_interval, 1000,
+             "While test is running, verify db every N milliseconds. 0 "
+             "disables continuous verification.");
+
+DEFINE_int32(approximate_size_one_in, 64,
+             "If non-zero, DB::GetApproximateSizes() will be called against"
+             " random key ranges.");
+#endif  // GFLAGS

db_stress_tool/db_stress_listener.h

@@ -0,0 +1,222 @@
+//  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).
+
+#ifdef GFLAGS
+#pragma once
+
+#include "rocksdb/listener.h"
+#include "util/gflags_compat.h"
+
+DECLARE_int32(compact_files_one_in);
+
+namespace rocksdb {
+class DbStressListener : public EventListener {
+ public:
+  DbStressListener(const std::string& db_name,
+                   const std::vector<DbPath>& db_paths,
+                   const std::vector<ColumnFamilyDescriptor>& column_families)
+      : db_name_(db_name),
+        db_paths_(db_paths),
+        column_families_(column_families),
+        num_pending_file_creations_(0) {}
+#ifndef ROCKSDB_LITE
+  ~DbStressListener() override { assert(num_pending_file_creations_ == 0); }
+  void OnFlushCompleted(DB* /*db*/, const FlushJobInfo& info) override {
+    assert(IsValidColumnFamilyName(info.cf_name));
+    VerifyFilePath(info.file_path);
+    // pretending doing some work here
+    RandomSleep();
+  }
+
+  void OnFlushBegin(DB* /*db*/,
+                    const FlushJobInfo& /*flush_job_info*/) override {
+    RandomSleep();
+  }
+
+  void OnTableFileDeleted(const TableFileDeletionInfo& /*info*/) override {
+    RandomSleep();
+  }
+
+  void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& /*ci*/) override {
+    RandomSleep();
+  }
+
+  void OnCompactionCompleted(DB* /*db*/, const CompactionJobInfo& ci) override {
+    assert(IsValidColumnFamilyName(ci.cf_name));
+    assert(ci.input_files.size() + ci.output_files.size() > 0U);
+    for (const auto& file_path : ci.input_files) {
+      VerifyFilePath(file_path);
+    }
+    for (const auto& file_path : ci.output_files) {
+      VerifyFilePath(file_path);
+    }
+    // pretending doing some work here
+    RandomSleep();
+  }
+
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& /*info*/) override {
+    ++num_pending_file_creations_;
+  }
+
+  void OnTableFileCreated(const TableFileCreationInfo& info) override {
+    assert(info.db_name == db_name_);
+    assert(IsValidColumnFamilyName(info.cf_name));
+    if (info.file_size) {
+      VerifyFilePath(info.file_path);
+    }
+    assert(info.job_id > 0 || FLAGS_compact_files_one_in > 0);
+    if (info.status.ok() && info.file_size > 0) {
+      assert(info.table_properties.data_size > 0 ||
+             info.table_properties.num_range_deletions > 0);
+      assert(info.table_properties.raw_key_size > 0);
+      assert(info.table_properties.num_entries > 0);
+    }
+    --num_pending_file_creations_;
+  }
+
+  void OnMemTableSealed(const MemTableInfo& /*info*/) override {
+    RandomSleep();
+  }
+
+  void OnColumnFamilyHandleDeletionStarted(
+      ColumnFamilyHandle* /*handle*/) override {
+    RandomSleep();
+  }
+
+  void OnExternalFileIngested(
+      DB* /*db*/, const ExternalFileIngestionInfo& /*info*/) override {
+    RandomSleep();
+  }
+
+  void OnBackgroundError(BackgroundErrorReason /* reason */,
+                         Status* /* bg_error */) override {
+    RandomSleep();
+  }
+
+  void OnStallConditionsChanged(const WriteStallInfo& /*info*/) override {
+    RandomSleep();
+  }
+
+  void OnFileReadFinish(const FileOperationInfo& info) override {
+    // Even empty callback is valuable because sometimes some locks are
+    // released in order to make the callback.
+
+    // Sleep carefully here as it is a frequent operation and we don't want
+    // to slow down the tests. We always sleep when the read is large.
+    // When read is small, sleep in a small chance.
+    size_t length_read = info.length;
+    if (length_read >= 1000000 || Random::GetTLSInstance()->OneIn(1000)) {
+      RandomSleep();
+    }
+  }
+
+  void OnFileWriteFinish(const FileOperationInfo& info) override {
+    // Even empty callback is valuable because sometimes some locks are
+    // released in order to make the callback.
+
+    // Sleep carefully here as it is a frequent operation and we don't want
+    // to slow down the tests. When the write is large, always sleep.
+    // Otherwise, sleep in a relatively small chance.
+    size_t length_write = info.length;
+    if (length_write >= 1000000 || Random::GetTLSInstance()->OneIn(64)) {
+      RandomSleep();
+    }
+  }
+
+  bool ShouldBeNotifiedOnFileIO() override {
+    RandomSleep();
+    return static_cast<bool>(Random::GetTLSInstance()->OneIn(1));
+  }
+
+  void OnErrorRecoveryBegin(BackgroundErrorReason /* reason */,
+                            Status /* bg_error */,
+                            bool* /* auto_recovery */) override {
+    RandomSleep();
+  }
+
+  void OnErrorRecoveryCompleted(Status /* old_bg_error */) override {
+    RandomSleep();
+  }
+
+ protected:
+  bool IsValidColumnFamilyName(const std::string& cf_name) const {
+    if (cf_name == kDefaultColumnFamilyName) {
+      return true;
+    }
+    // The column family names in the stress tests are numbers.
+    for (size_t i = 0; i < cf_name.size(); ++i) {
+      if (cf_name[i] < '0' || cf_name[i] > '9') {
+        return false;
+      }
+    }
+    return true;
+  }
+
+  void VerifyFileDir(const std::string& file_dir) {
+#ifndef NDEBUG
+    if (db_name_ == file_dir) {
+      return;
+    }
+    for (const auto& db_path : db_paths_) {
+      if (db_path.path == file_dir) {
+        return;
+      }
+    }
+    for (auto& cf : column_families_) {
+      for (const auto& cf_path : cf.options.cf_paths) {
+        if (cf_path.path == file_dir) {
+          return;
+        }
+      }
+    }
+    assert(false);
+#else
+    (void)file_dir;
+#endif  // !NDEBUG
+  }
+
+  void VerifyFileName(const std::string& file_name) {
+#ifndef NDEBUG
+    uint64_t file_number;
+    FileType file_type;
+    bool result = ParseFileName(file_name, &file_number, &file_type);
+    assert(result);
+    assert(file_type == kTableFile);
+#else
+    (void)file_name;
+#endif  // !NDEBUG
+  }
+
+  void VerifyFilePath(const std::string& file_path) {
+#ifndef NDEBUG
+    size_t pos = file_path.find_last_of("/");
+    if (pos == std::string::npos) {
+      VerifyFileName(file_path);
+    } else {
+      if (pos > 0) {
+        VerifyFileDir(file_path.substr(0, pos));
+      }
+      VerifyFileName(file_path.substr(pos));
+    }
+#else
+    (void)file_path;
+#endif  // !NDEBUG
+  }
+
+  void RandomSleep() {
+    std::this_thread::sleep_for(
+        std::chrono::microseconds(Random::GetTLSInstance()->Uniform(5000)));
+  }
+#endif  // !ROCKSDB_LITE
+
+ private:
+  std::string db_name_;
+  std::vector<DbPath> db_paths_;
+  std::vector<ColumnFamilyDescriptor> column_families_;
+  std::atomic<int> num_pending_file_creations_;
+};
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_shared_state.cc

@@ -0,0 +1,18 @@
+//  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.
+//
+
+#ifdef GFLAGS
+#include "db_stress_tool/db_stress_shared_state.h"
+
+namespace rocksdb {
+const uint32_t SharedState::UNKNOWN_SENTINEL = 0xfffffffe;
+const uint32_t SharedState::DELETION_SENTINEL = 0xffffffff;
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_shared_state.h

@@ -0,0 +1,390 @@
+//  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
+
+#ifdef GFLAGS
+#pragma once
+
+#include "db_stress_tool/db_stress_stat.h"
+#include "util/gflags_compat.h"
+
+DECLARE_uint64(seed);
+DECLARE_int64(max_key);
+DECLARE_uint64(log2_keys_per_lock);
+DECLARE_int32(threads);
+DECLARE_int32(column_families);
+DECLARE_int32(nooverwritepercent);
+DECLARE_string(expected_values_path);
+DECLARE_int32(clear_column_family_one_in);
+DECLARE_bool(test_batches_snapshots);
+DECLARE_int32(compaction_thread_pool_adjust_interval);
+DECLARE_int32(continuous_verification_interval);
+
+namespace rocksdb {
+class StressTest;
+
+// State shared by all concurrent executions of the same benchmark.
+class SharedState {
+ public:
+  // indicates a key may have any value (or not be present) as an operation on
+  // it is incomplete.
+  static const uint32_t UNKNOWN_SENTINEL;
+  // indicates a key should definitely be deleted
+  static const uint32_t DELETION_SENTINEL;
+
+  SharedState(Env* env, StressTest* stress_test)
+      : cv_(&mu_),
+        seed_(static_cast<uint32_t>(FLAGS_seed)),
+        max_key_(FLAGS_max_key),
+        log2_keys_per_lock_(static_cast<uint32_t>(FLAGS_log2_keys_per_lock)),
+        num_threads_(FLAGS_threads),
+        num_initialized_(0),
+        num_populated_(0),
+        vote_reopen_(0),
+        num_done_(0),
+        start_(false),
+        start_verify_(false),
+        num_bg_threads_(0),
+        should_stop_bg_thread_(false),
+        bg_thread_finished_(0),
+        stress_test_(stress_test),
+        verification_failure_(false),
+        should_stop_test_(false),
+        no_overwrite_ids_(FLAGS_column_families),
+        values_(nullptr),
+        printing_verification_results_(false) {
+    // Pick random keys in each column family that will not experience
+    // overwrite
+
+    fprintf(stdout, "Choosing random keys with no overwrite\n");
+    Random64 rnd(seed_);
+    // Start with the identity permutation. Subsequent iterations of
+    // for loop below will start with perm of previous for loop
+    int64_t* permutation = new int64_t[max_key_];
+    for (int64_t i = 0; i < max_key_; i++) {
+      permutation[i] = i;
+    }
+    // Now do the Knuth shuffle
+    int64_t num_no_overwrite_keys = (max_key_ * FLAGS_nooverwritepercent) / 100;
+    // Only need to figure out first num_no_overwrite_keys of permutation
+    no_overwrite_ids_.reserve(num_no_overwrite_keys);
+    for (int64_t i = 0; i < num_no_overwrite_keys; i++) {
+      int64_t rand_index = i + rnd.Next() % (max_key_ - i);
+      // Swap i and rand_index;
+      int64_t temp = permutation[i];
+      permutation[i] = permutation[rand_index];
+      permutation[rand_index] = temp;
+      // Fill no_overwrite_ids_ with the first num_no_overwrite_keys of
+      // permutation
+      no_overwrite_ids_.insert(permutation[i]);
+    }
+    delete[] permutation;
+
+    size_t expected_values_size =
+        sizeof(std::atomic<uint32_t>) * FLAGS_column_families * max_key_;
+    bool values_init_needed = false;
+    Status status;
+    if (!FLAGS_expected_values_path.empty()) {
+      if (!std::atomic<uint32_t>{}.is_lock_free()) {
+        status = Status::InvalidArgument(
+            "Cannot use --expected_values_path on platforms without lock-free "
+            "std::atomic<uint32_t>");
+      }
+      if (status.ok() && FLAGS_clear_column_family_one_in > 0) {
+        status = Status::InvalidArgument(
+            "Cannot use --expected_values_path on when "
+            "--clear_column_family_one_in is greater than zero.");
+      }
+      uint64_t size = 0;
+      if (status.ok()) {
+        status = env->GetFileSize(FLAGS_expected_values_path, &size);
+      }
+      std::unique_ptr<WritableFile> wfile;
+      if (status.ok() && size == 0) {
+        const EnvOptions soptions;
+        status =
+            env->NewWritableFile(FLAGS_expected_values_path, &wfile, soptions);
+      }
+      if (status.ok() && size == 0) {
+        std::string buf(expected_values_size, '\0');
+        status = wfile->Append(buf);
+        values_init_needed = true;
+      }
+      if (status.ok()) {
+        status = env->NewMemoryMappedFileBuffer(FLAGS_expected_values_path,
+                                                &expected_mmap_buffer_);
+      }
+      if (status.ok()) {
+        assert(expected_mmap_buffer_->GetLen() == expected_values_size);
+        values_ = static_cast<std::atomic<uint32_t>*>(
+            expected_mmap_buffer_->GetBase());
+        assert(values_ != nullptr);
+      } else {
+        fprintf(stderr, "Failed opening shared file '%s' with error: %s\n",
+                FLAGS_expected_values_path.c_str(), status.ToString().c_str());
+        assert(values_ == nullptr);
+      }
+    }
+    if (values_ == nullptr) {
+      values_allocation_.reset(
+          new std::atomic<uint32_t>[FLAGS_column_families * max_key_]);
+      values_ = &values_allocation_[0];
+      values_init_needed = true;
+    }
+    assert(values_ != nullptr);
+    if (values_init_needed) {
+      for (int i = 0; i < FLAGS_column_families; ++i) {
+        for (int j = 0; j < max_key_; ++j) {
+          Delete(i, j, false /* pending */);
+        }
+      }
+    }
+
+    if (FLAGS_test_batches_snapshots) {
+      fprintf(stdout, "No lock creation because test_batches_snapshots set\n");
+      return;
+    }
+
+    long num_locks = static_cast<long>(max_key_ >> log2_keys_per_lock_);
+    if (max_key_ & ((1 << log2_keys_per_lock_) - 1)) {
+      num_locks++;
+    }
+    fprintf(stdout, "Creating %ld locks\n", num_locks * FLAGS_column_families);
+    key_locks_.resize(FLAGS_column_families);
+
+    for (int i = 0; i < FLAGS_column_families; ++i) {
+      key_locks_[i].resize(num_locks);
+      for (auto& ptr : key_locks_[i]) {
+        ptr.reset(new port::Mutex);
+      }
+    }
+    if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
+      ++num_bg_threads_;
+      fprintf(stdout, "Starting compaction_thread_pool_adjust_thread\n");
+    }
+    if (FLAGS_continuous_verification_interval > 0) {
+      ++num_bg_threads_;
+      fprintf(stdout, "Starting continuous_verification_thread\n");
+    }
+  }
+
+  ~SharedState() {}
+
+  port::Mutex* GetMutex() { return &mu_; }
+
+  port::CondVar* GetCondVar() { return &cv_; }
+
+  StressTest* GetStressTest() const { return stress_test_; }
+
+  int64_t GetMaxKey() const { return max_key_; }
+
+  uint32_t GetNumThreads() const { return num_threads_; }
+
+  void IncInitialized() { num_initialized_++; }
+
+  void IncOperated() { num_populated_++; }
+
+  void IncDone() { num_done_++; }
+
+  void IncVotedReopen() { vote_reopen_ = (vote_reopen_ + 1) % num_threads_; }
+
+  bool AllInitialized() const { return num_initialized_ >= num_threads_; }
+
+  bool AllOperated() const { return num_populated_ >= num_threads_; }
+
+  bool AllDone() const { return num_done_ >= num_threads_; }
+
+  bool AllVotedReopen() { return (vote_reopen_ == 0); }
+
+  void SetStart() { start_ = true; }
+
+  void SetStartVerify() { start_verify_ = true; }
+
+  bool Started() const { return start_; }
+
+  bool VerifyStarted() const { return start_verify_; }
+
+  void SetVerificationFailure() { verification_failure_.store(true); }
+
+  bool HasVerificationFailedYet() const { return verification_failure_.load(); }
+
+  void SetShouldStopTest() { should_stop_test_.store(true); }
+
+  bool ShouldStopTest() const { return should_stop_test_.load(); }
+
+  port::Mutex* GetMutexForKey(int cf, int64_t key) {
+    return key_locks_[cf][key >> log2_keys_per_lock_].get();
+  }
+
+  void LockColumnFamily(int cf) {
+    for (auto& mutex : key_locks_[cf]) {
+      mutex->Lock();
+    }
+  }
+
+  void UnlockColumnFamily(int cf) {
+    for (auto& mutex : key_locks_[cf]) {
+      mutex->Unlock();
+    }
+  }
+
+  std::atomic<uint32_t>& Value(int cf, int64_t key) const {
+    return values_[cf * max_key_ + key];
+  }
+
+  void ClearColumnFamily(int cf) {
+    std::fill(&Value(cf, 0 /* key */), &Value(cf + 1, 0 /* key */),
+              DELETION_SENTINEL);
+  }
+
+  // @param pending True if the update may have started but is not yet
+  //    guaranteed finished. This is useful for crash-recovery testing when the
+  //    process may crash before updating the expected values array.
+  void Put(int cf, int64_t key, uint32_t value_base, bool pending) {
+    if (!pending) {
+      // prevent expected-value update from reordering before Write
+      std::atomic_thread_fence(std::memory_order_release);
+    }
+    Value(cf, key).store(pending ? UNKNOWN_SENTINEL : value_base,
+                         std::memory_order_relaxed);
+    if (pending) {
+      // prevent Write from reordering before expected-value update
+      std::atomic_thread_fence(std::memory_order_release);
+    }
+  }
+
+  uint32_t Get(int cf, int64_t key) const { return Value(cf, key); }
+
+  // @param pending See comment above Put()
+  // Returns true if the key was not yet deleted.
+  bool Delete(int cf, int64_t key, bool pending) {
+    if (Value(cf, key) == DELETION_SENTINEL) {
+      return false;
+    }
+    Put(cf, key, DELETION_SENTINEL, pending);
+    return true;
+  }
+
+  // @param pending See comment above Put()
+  // Returns true if the key was not yet deleted.
+  bool SingleDelete(int cf, int64_t key, bool pending) {
+    return Delete(cf, key, pending);
+  }
+
+  // @param pending See comment above Put()
+  // Returns number of keys deleted by the call.
+  int DeleteRange(int cf, int64_t begin_key, int64_t end_key, bool pending) {
+    int covered = 0;
+    for (int64_t key = begin_key; key < end_key; ++key) {
+      if (Delete(cf, key, pending)) {
+        ++covered;
+      }
+    }
+    return covered;
+  }
+
+  bool AllowsOverwrite(int64_t key) {
+    return no_overwrite_ids_.find(key) == no_overwrite_ids_.end();
+  }
+
+  bool Exists(int cf, int64_t key) {
+    // UNKNOWN_SENTINEL counts as exists. That assures a key for which overwrite
+    // is disallowed can't be accidentally added a second time, in which case
+    // SingleDelete wouldn't be able to properly delete the key. It does allow
+    // the case where a SingleDelete might be added which covers nothing, but
+    // that's not a correctness issue.
+    uint32_t expected_value = Value(cf, key).load();
+    return expected_value != DELETION_SENTINEL;
+  }
+
+  uint32_t GetSeed() const { return seed_; }
+
+  void SetShouldStopBgThread() { should_stop_bg_thread_ = true; }
+
+  bool ShouldStopBgThread() { return should_stop_bg_thread_; }
+
+  void IncBgThreadsFinished() { ++bg_thread_finished_; }
+
+  bool BgThreadsFinished() const {
+    return bg_thread_finished_ == num_bg_threads_;
+  }
+
+  bool ShouldVerifyAtBeginning() const {
+    return expected_mmap_buffer_.get() != nullptr;
+  }
+
+  bool PrintingVerificationResults() {
+    bool tmp = false;
+    return !printing_verification_results_.compare_exchange_strong(
+        tmp, true, std::memory_order_relaxed);
+  }
+
+  void FinishPrintingVerificationResults() {
+    printing_verification_results_.store(false, std::memory_order_relaxed);
+  }
+
+ private:
+  port::Mutex mu_;
+  port::CondVar cv_;
+  const uint32_t seed_;
+  const int64_t max_key_;
+  const uint32_t log2_keys_per_lock_;
+  const int num_threads_;
+  long num_initialized_;
+  long num_populated_;
+  long vote_reopen_;
+  long num_done_;
+  bool start_;
+  bool start_verify_;
+  int num_bg_threads_;
+  bool should_stop_bg_thread_;
+  int bg_thread_finished_;
+  StressTest* stress_test_;
+  std::atomic<bool> verification_failure_;
+  std::atomic<bool> should_stop_test_;
+
+  // Keys that should not be overwritten
+  std::unordered_set<size_t> no_overwrite_ids_;
+
+  std::atomic<uint32_t>* values_;
+  std::unique_ptr<std::atomic<uint32_t>[]> values_allocation_;
+  // Has to make it owned by a smart ptr as port::Mutex is not copyable
+  // and storing it in the container may require copying depending on the impl.
+  std::vector<std::vector<std::unique_ptr<port::Mutex>>> key_locks_;
+  std::unique_ptr<MemoryMappedFileBuffer> expected_mmap_buffer_;
+  std::atomic<bool> printing_verification_results_;
+};
+
+// Per-thread state for concurrent executions of the same benchmark.
+struct ThreadState {
+  uint32_t tid;  // 0..n-1
+  Random rand;   // Has different seeds for different threads
+  SharedState* shared;
+  Stats stats;
+  struct SnapshotState {
+    const Snapshot* snapshot;
+    // The cf from which we did a Get at this snapshot
+    int cf_at;
+    // The name of the cf at the time that we did a read
+    std::string cf_at_name;
+    // The key with which we did a Get at this snapshot
+    std::string key;
+    // The status of the Get
+    Status status;
+    // The value of the Get
+    std::string value;
+    // optional state of all keys in the db
+    std::vector<bool>* key_vec;
+  };
+  std::queue<std::pair<uint64_t, SnapshotState>> snapshot_queue;
+
+  ThreadState(uint32_t index, SharedState* _shared)
+      : tid(index), rand(1000 + index + _shared->GetSeed()), shared(_shared) {}
+};
+}  // namespace rocksdb
+#endif  // GFLAGS

db_stress_tool/db_stress_stat.h

@@ -0,0 +1,213 @@
+//  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).
+
+#pragma once
+#include <cinttypes>
+#include <memory>
+#include <queue>
+#include <unordered_set>
+
+#include "monitoring/histogram.h"
+#include "port/port.h"
+#include "rocksdb/env.h"
+#include "rocksdb/snapshot.h"
+#include "rocksdb/statistics.h"
+#include "util/gflags_compat.h"
+#include "util/random.h"
+
+DECLARE_bool(histogram);
+DECLARE_bool(progress_reports);
+
+namespace rocksdb {
+// Database statistics
+static std::shared_ptr<rocksdb::Statistics> dbstats;
+static std::shared_ptr<rocksdb::Statistics> dbstats_secondaries;
+
+class Stats {
+ private:
+  uint64_t start_;
+  uint64_t finish_;
+  double seconds_;
+  long done_;
+  long gets_;
+  long prefixes_;
+  long writes_;
+  long deletes_;
+  size_t single_deletes_;
+  long iterator_size_sums_;
+  long founds_;
+  long iterations_;
+  long range_deletions_;
+  long covered_by_range_deletions_;
+  long errors_;
+  long num_compact_files_succeed_;
+  long num_compact_files_failed_;
+  int next_report_;
+  size_t bytes_;
+  uint64_t last_op_finish_;
+  HistogramImpl hist_;
+
+ public:
+  Stats() {}
+
+  void Start() {
+    next_report_ = 100;
+    hist_.Clear();
+    done_ = 0;
+    gets_ = 0;
+    prefixes_ = 0;
+    writes_ = 0;
+    deletes_ = 0;
+    single_deletes_ = 0;
+    iterator_size_sums_ = 0;
+    founds_ = 0;
+    iterations_ = 0;
+    range_deletions_ = 0;
+    covered_by_range_deletions_ = 0;
+    errors_ = 0;
+    bytes_ = 0;
+    seconds_ = 0;
+    num_compact_files_succeed_ = 0;
+    num_compact_files_failed_ = 0;
+    start_ = Env::Default()->NowMicros();
+    last_op_finish_ = start_;
+    finish_ = start_;
+  }
+
+  void Merge(const Stats& other) {
+    hist_.Merge(other.hist_);
+    done_ += other.done_;
+    gets_ += other.gets_;
+    prefixes_ += other.prefixes_;
+    writes_ += other.writes_;
+    deletes_ += other.deletes_;
+    single_deletes_ += other.single_deletes_;
+    iterator_size_sums_ += other.iterator_size_sums_;
+    founds_ += other.founds_;
+    iterations_ += other.iterations_;
+    range_deletions_ += other.range_deletions_;
+    covered_by_range_deletions_ = other.covered_by_range_deletions_;
+    errors_ += other.errors_;
+    bytes_ += other.bytes_;
+    seconds_ += other.seconds_;
+    num_compact_files_succeed_ += other.num_compact_files_succeed_;
+    num_compact_files_failed_ += other.num_compact_files_failed_;
+    if (other.start_ < start_) start_ = other.start_;
+    if (other.finish_ > finish_) finish_ = other.finish_;
+  }
+
+  void Stop() {
+    finish_ = Env::Default()->NowMicros();
+    seconds_ = (finish_ - start_) * 1e-6;
+  }
+
+  void FinishedSingleOp() {
+    if (FLAGS_histogram) {
+      auto now = Env::Default()->NowMicros();
+      auto micros = now - last_op_finish_;
+      hist_.Add(micros);
+      if (micros > 20000) {
+        fprintf(stdout, "long op: %" PRIu64 " micros%30s\r", micros, "");
+      }
+      last_op_finish_ = now;
+    }
+
+    done_++;
+    if (FLAGS_progress_reports) {
+      if (done_ >= next_report_) {
+        if (next_report_ < 1000)
+          next_report_ += 100;
+        else if (next_report_ < 5000)
+          next_report_ += 500;
+        else if (next_report_ < 10000)
+          next_report_ += 1000;
+        else if (next_report_ < 50000)
+          next_report_ += 5000;
+        else if (next_report_ < 100000)
+          next_report_ += 10000;
+        else if (next_report_ < 500000)
+          next_report_ += 50000;
+        else
+          next_report_ += 100000;
+        fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
+      }
+    }
+  }
+
+  void AddBytesForWrites(long nwrites, size_t nbytes) {
+    writes_ += nwrites;
+    bytes_ += nbytes;
+  }
+
+  void AddGets(long ngets, long nfounds) {
+    founds_ += nfounds;
+    gets_ += ngets;
+  }
+
+  void AddPrefixes(long nprefixes, long count) {
+    prefixes_ += nprefixes;
+    iterator_size_sums_ += count;
+  }
+
+  void AddIterations(long n) { iterations_ += n; }
+
+  void AddDeletes(long n) { deletes_ += n; }
+
+  void AddSingleDeletes(size_t n) { single_deletes_ += n; }
+
+  void AddRangeDeletions(long n) { range_deletions_ += n; }
+
+  void AddCoveredByRangeDeletions(long n) { covered_by_range_deletions_ += n; }
+
+  void AddErrors(long n) { errors_ += n; }
+
+  void AddNumCompactFilesSucceed(long n) { num_compact_files_succeed_ += n; }
+
+  void AddNumCompactFilesFailed(long n) { num_compact_files_failed_ += n; }
+
+  void Report(const char* name) {
+    std::string extra;
+    if (bytes_ < 1 || done_ < 1) {
+      fprintf(stderr, "No writes or ops?\n");
+      return;
+    }
+
+    double elapsed = (finish_ - start_) * 1e-6;
+    double bytes_mb = bytes_ / 1048576.0;
+    double rate = bytes_mb / elapsed;
+    double throughput = (double)done_ / elapsed;
+
+    fprintf(stdout, "%-12s: ", name);
+    fprintf(stdout, "%.3f micros/op %ld ops/sec\n", seconds_ * 1e6 / done_,
+            (long)throughput);
+    fprintf(stdout, "%-12s: Wrote %.2f MB (%.2f MB/sec) (%ld%% of %ld ops)\n",
+            "", bytes_mb, rate, (100 * writes_) / done_, done_);
+    fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_);
+    fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_);
+    fprintf(stdout, "%-12s: Single deleted %" ROCKSDB_PRIszt " times\n", "",
+            single_deletes_);
+    fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "", gets_,
+            founds_);
+    fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_);
+    fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "",
+            iterator_size_sums_);
+    fprintf(stdout, "%-12s: Iterated %ld times\n", "", iterations_);
+    fprintf(stdout, "%-12s: Deleted %ld key-ranges\n", "", range_deletions_);
+    fprintf(stdout, "%-12s: Range deletions covered %ld keys\n", "",
+            covered_by_range_deletions_);
+
+    fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_);
+    fprintf(stdout, "%-12s: %ld CompactFiles() succeed\n", "",
+            num_compact_files_succeed_);
+    fprintf(stdout, "%-12s: %ld CompactFiles() did not succeed\n", "",
+            num_compact_files_failed_);
+
+    if (FLAGS_histogram) {
+      fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str());
+    }
+    fflush(stdout);
+  }
+};
+}  // namespace rocksdb