Separate some IndexBlockIter logic from BlockIter (#4136)

Summary:
Some logic only related to IndexBlockIter is separated from BlockIter to IndexBlockIter. This is done by writing an exclusive Seek() and SeekForPrev() for DataBlockIter, and all metadata block iter and tombstone block iter now use data block iter. Dealing with the BinarySeek() sharing problem by passing in the comparator to use.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4136

Reviewed By: maysamyabandeh

Differential Revision: D8859673

Pulled By: siying

fbshipit-source-id: 703e5e6824b82b7cbf4721f3594b94127797ca9e

.lgtm.yml

@@ -0,0 +1,4 @@
+extraction:
+  cpp:
+    index:
+      build_command: make static_lib

.travis.yml

@@ -15,8 +15,17 @@ cache:
   - apt
 
 addons:
-   apt:
-      packages: ['zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'libgflags-dev', 'mingw-w64']
+  apt:
+    sources:
+      - ubuntu-toolchain-r-test
+    packages:
+      - curl
+      - g++-8
+      - libbz2-dev
+      - libgflags-dev
+      - libsnappy-dev
+      - mingw-w64
+      - zlib1g-dev
 env:
   - TEST_GROUP=platform_dependent # 16-18 minutes
   - TEST_GROUP=1 # 33-35 minutes
@@ -30,6 +39,7 @@ env:
   # Build examples
   - JOB_NAME=examples # 5-7 minutes
   - JOB_NAME=cmake # 3-5 minutes
+  - JOB_NAME=cmake-gcc8 # 3-5 minutes
   - JOB_NAME=cmake-mingw # 3 minutes
 
 matrix:
@@ -42,6 +52,8 @@ matrix:
     env: TEST_GROUP=3
   - os: osx
     env: TEST_GROUP=4
+  - os: osx
+    env: JOB_NAME=cmake-gcc8
   - os : osx
     env: JOB_NAME=cmake-mingw
   - os : linux
@@ -52,9 +64,12 @@ matrix:
 # https://docs.travis-ci.com/user/caching/#ccache-cache
 install:
   - if [ "${TRAVIS_OS_NAME}" == osx ]; then
-      brew install ccache;
+      brew install ccache zstd lz4 snappy xz;
       PATH=$PATH:/usr/local/opt/ccache/libexec;
     fi
+  - if [ "${JOB_NAME}" == cmake-gcc8 ]; then
+      CC=gcc-8 && CXX=g++-8;
+    fi
   - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
       mkdir cmake-dist && curl -sfSL https://cmake.org/files/v3.8/cmake-3.8.1-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
     fi
@@ -67,16 +82,40 @@ before_script:
 script:
   - ${CXX} --version
   - if [ `command -v ccache` ]; then ccache -C; fi
-  - if [ "${TEST_GROUP}" == 'platform_dependent' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=db_block_cache_test make -j4 all_but_some_tests check_some; fi
-  - if [ "${TEST_GROUP}" == '1' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=full_filter_block_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '2' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=full_filter_block_test ROCKSDBTESTS_END=write_batch_with_index_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '3' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_batch_with_index_test ROCKSDBTESTS_END=write_prepared_transaction_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '4' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_prepared_transaction_test make -j4 check_some; fi
-  - if [ "${JOB_NAME}" == 'java_test' ]; then OPT=-DTRAVIS V=1 make clean jclean && make rocksdbjava jtest; fi
-  - if [ "${JOB_NAME}" == 'lite_build' ]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib tools; fi
-  - if [ "${JOB_NAME}" == 'examples' ]; then OPT=-DTRAVIS V=1 make -j4 static_lib; cd examples; make -j4; fi
-  - if [ "${JOB_NAME}" == 'cmake' ]; then mkdir build && cd build && cmake -DJNI=1 .. && make -j4 rocksdb rocksdbjni; fi
-  - if [ "${JOB_NAME}" == 'cmake-mingw' ]; then 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; fi
+  - case $TEST_GROUP in
+    platform_dependent)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=db_block_cache_test make -j4 all_but_some_tests check_some
+      ;;
+    1)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=full_filter_block_test make -j4 check_some
+      ;;
+    2)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=full_filter_block_test ROCKSDBTESTS_END=write_batch_with_index_test make -j4 check_some
+      ;;
+    3)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_batch_with_index_test ROCKSDBTESTS_END=write_prepared_transaction_test make -j4 check_some
+      ;;
+    4)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_prepared_transaction_test make -j4 check_some
+      ;;
+    esac
+  - case $JOB_NAME in
+    java_test)
+      OPT=-DTRAVIS V=1 make clean jclean && make rocksdbjava jtest
+      ;;
+    lite_build)
+      OPT='-DTRAVIS -DROCKSDB_LITE' V=1 make -j4 static_lib tools
+      ;;
+    examples)
+      OPT=-DTRAVIS V=1 make -j4 static_lib && cd examples && make -j4
+      ;;
+    cmake-mingw)
+      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
+      ;;
+    cmake*)
+      mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_BUILD_TYPE=Release && make -j4 rocksdb rocksdbjni
+      ;;
+    esac
 notifications:
     email:
       - leveldb@fb.com

CMakeLists.txt

@@ -16,7 +16,7 @@
 #    See thirdparty.inc for more information.
 #        sample command: cmake -G "Visual Studio 15 Win64" -DWITH_GFLAGS=1 -DWITH_SNAPPY=1 -DWITH_JEMALLOC=1 -DWITH_JNI=1 ..
 # 4. Then build the project in debug mode (you may want to add /m[:<N>] flag to run msbuild in <N> parallel threads
-#                                          or simply /m ot use all avail cores)
+#                                          or simply /m to use all avail cores)
 #        msbuild rocksdb.sln
 #
 #        rocksdb.sln build features exclusions of test only code in Release. If you build ALL_BUILD then everything
@@ -64,9 +64,10 @@ else()
       find_package(JeMalloc REQUIRED)
       add_definitions(-DROCKSDB_JEMALLOC -DJEMALLOC_NO_DEMANGLE)
       include_directories(${JEMALLOC_INCLUDE_DIR})
+      list(APPEND THIRDPARTY_LIBS ${JEMALLOC_LIBRARIES})
     endif()
   endif()
-  
+
   # No config file for this
   option(WITH_GFLAGS "build with GFlags" ON)
   if(WITH_GFLAGS)
@@ -132,19 +133,17 @@ endif()
 
 string(REGEX REPLACE "[^0-9a-f]+" "" GIT_SHA "${GIT_SHA}")
 
-set(SH_CMD "sh")
-execute_process(COMMAND
-  ${SH_CMD} -c "build_tools/version.sh full"
-  WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
-  OUTPUT_VARIABLE ROCKSDB_VERSION
-)
-string(STRIP "${ROCKSDB_VERSION}" ROCKSDB_VERSION)
-execute_process(COMMAND
-  ${SH_CMD} -c "build_tools/version.sh major"
-  WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
-  OUTPUT_VARIABLE ROCKSDB_VERSION_MAJOR
-)
-string(STRIP "${ROCKSDB_VERSION_MAJOR}" ROCKSDB_VERSION_MAJOR)
+
+# Read rocksdb version from version.h header file.
+file(READ include/rocksdb/version.h version_header_file)
+string(REGEX MATCH "#define ROCKSDB_MAJOR ([0-9]+)" _ ${version_header_file})
+set(ROCKSDB_VERSION_MAJOR ${CMAKE_MATCH_1})
+string(REGEX MATCH "#define ROCKSDB_MINOR ([0-9]+)" _ ${version_header_file})
+set(ROCKSDB_VERSION_MINOR ${CMAKE_MATCH_1})
+string(REGEX MATCH "#define ROCKSDB_PATCH ([0-9]+)" _ ${version_header_file})
+set(ROCKSDB_VERSION_PATCH ${CMAKE_MATCH_1})
+set(ROCKSDB_VERSION ${ROCKSDB_VERSION_MAJOR}.${ROCKSDB_VERSION_MINOR}.${ROCKSDB_VERSION_PATCH})
+
 
 option(WITH_MD_LIBRARY "build with MD" ON)
 if(WIN32 AND MSVC)
@@ -171,7 +170,7 @@ else()
   endif()
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
   if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -fno-omit-frame-pointer")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
     include(CheckCXXCompilerFlag)
     CHECK_CXX_COMPILER_FLAG("-momit-leaf-frame-pointer" HAVE_OMIT_LEAF_FRAME_POINTER)
     if(HAVE_OMIT_LEAF_FRAME_POINTER)
@@ -229,6 +228,7 @@ int main() {
 unset(CMAKE_REQUIRED_FLAGS)
 if(HAVE_SSE42)
   add_definitions(-DHAVE_SSE42)
+  add_definitions(-DHAVE_PCLMUL)
 elseif(FORCE_SSE42)
   message(FATAL_ERROR "FORCE_SSE42=ON but unable to compile with SSE4.2 enabled")
 endif()
@@ -285,20 +285,28 @@ if(WITH_UBSAN)
   endif()
 endif()
 
-find_package(NUMA)
-if(NUMA_FOUND)
+option(WITH_NUMA "build with NUMA policy support" OFF)
+if(WITH_NUMA)
+  find_package(NUMA REQUIRED)
   add_definitions(-DNUMA)
   include_directories(${NUMA_INCLUDE_DIR})
   list(APPEND THIRDPARTY_LIBS ${NUMA_LIBRARIES})
 endif()
 
-find_package(TBB)
-if(TBB_FOUND)
+option(WITH_TBB "build with Threading Building Blocks (TBB)" OFF)
+if(WITH_TBB)
+  find_package(TBB REQUIRED)
   add_definitions(-DTBB)
   include_directories(${TBB_INCLUDE_DIR})
   list(APPEND THIRDPARTY_LIBS ${TBB_LIBRARIES})
 endif()
 
+# Stall notifications eat some performance from inserts
+option(DISABLE_STALL_NOTIF "Build with stall notifications" OFF)
+if(DISABLE_STALL_NOTIF)
+  add_definitions(-DROCKSDB_DISABLE_STALL_NOTIFICATION)
+endif()
+
 # Used to run CI build and tests so we can run faster
 set(OPTIMIZE_DEBUG_DEFAULT 0)        # Debug build is unoptimized by default use -DOPTDBG=1 to optimize
 
@@ -308,6 +316,32 @@ else()
    set(OPTIMIZE_DEBUG ${OPTIMIZE_DEBUG_DEFAULT})
 endif()
 
+if(DEFINED USE_RTTI)
+  if(USE_RTTI)
+    message(STATUS "Enabling RTTI")
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -DROCKSDB_USE_RTTI")
+  else()
+    if(MSVC)
+      message(STATUS "Disabling RTTI in Release builds. Always on in Debug.")
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
+      set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /GR-")
+    else()
+      message(STATUS "Disabling RTTI in Release builds")
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-rtti")
+      set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+    endif()
+  endif()
+else()
+  message(STATUS "Enabling RTTI in Debug builds only (default)")
+  set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
+  if(MSVC)
+     set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /GR-")
+  else()
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+  endif()
+endif()
+
 if(MSVC)
   if((${OPTIMIZE_DEBUG} EQUAL 1))
     message(STATUS "Debug optimization is enabled")
@@ -447,6 +481,7 @@ set(SOURCES
         db/db_info_dumper.cc
         db/db_iter.cc
         db/dbformat.cc
+        db/error_handler.cc
         db/event_helpers.cc
         db/experimental.cc
         db/external_sst_file_ingestion_job.cc
@@ -455,6 +490,7 @@ set(SOURCES
         db/flush_scheduler.cc
         db/forward_iterator.cc
         db/internal_stats.cc
+        db/logs_with_prep_tracker.cc
         db/log_reader.cc
         db/log_writer.cc
         db/malloc_stats.cc
@@ -549,6 +585,7 @@ set(SOURCES
         util/coding.cc
         util/compaction_job_stats_impl.cc
         util/comparator.cc
+        util/compression_context_cache.cc
         util/concurrent_arena.cc
         util/crc32c.cc
         util/delete_scheduler.cc
@@ -569,6 +606,7 @@ set(SOURCES
         util/status_message.cc
         util/string_util.cc
         util/sync_point.cc
+        util/sync_point_impl.cc
         util/testutil.cc
         util/thread_local.cc
         util/threadpool_imp.cc
@@ -631,6 +669,8 @@ set(SOURCES
         utilities/transactions/transaction_util.cc
         utilities/transactions/write_prepared_txn.cc
         utilities/transactions/write_prepared_txn_db.cc
+        utilities/transactions/write_unprepared_txn.cc
+        utilities/transactions/write_unprepared_txn_db.cc
         utilities/ttl/db_ttl_impl.cc
         utilities/write_batch_with_index/write_batch_with_index.cc
         utilities/write_batch_with_index/write_batch_with_index_internal.cc
@@ -657,8 +697,12 @@ if(WIN32)
     port/win/env_default.cc
     port/win/port_win.cc
     port/win/win_logger.cc
-    port/win/win_thread.cc
+    port/win/win_thread.cc)
+
+if(WITH_XPRESS)
+  list(APPEND SOURCES
     port/win/xpress_win.cc)
+endif()
 
 if(WITH_JEMALLOC)
   list(APPEND SOURCES
@@ -675,6 +719,14 @@ endif()
 set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
 set(ROCKSDB_SHARED_LIB rocksdb-shared${ARTIFACT_SUFFIX})
 set(ROCKSDB_IMPORT_LIB ${ROCKSDB_SHARED_LIB})
+
+option(WITH_LIBRADOS "Build with librados" OFF)
+if(WITH_LIBRADOS)
+  list(APPEND SOURCES
+    utilities/env_librados.cc)
+  list(APPEND THIRDPARTY_LIBS rados)
+endif()
+
 if(WIN32)
   set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)
   set(LIBS ${ROCKSDB_STATIC_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
@@ -693,13 +745,6 @@ else()
                         OUTPUT_NAME "rocksdb")
 endif()
 
-option(WITH_LIBRADOS "Build with librados" OFF)
-if(WITH_LIBRADOS)
-  list(APPEND SOURCES
-    utilities/env_librados.cc)
-  list(APPEND THIRDPARTY_LIBS rados)
-endif()
-
 add_library(${ROCKSDB_STATIC_LIB} STATIC ${SOURCES})
 target_link_libraries(${ROCKSDB_STATIC_LIB}
   ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
@@ -814,6 +859,7 @@ if(WITH_TESTS)
         db/db_inplace_update_test.cc
         db/db_io_failure_test.cc
         db/db_iter_test.cc
+        db/db_iter_stress_test.cc
         db/db_iterator_test.cc
         db/db_log_iter_test.cc
         db/db_memtable_test.cc
@@ -832,6 +878,8 @@ if(WITH_TESTS)
         db/db_write_test.cc
         db/dbformat_test.cc
         db/deletefile_test.cc
+        db/error_handler_test.cc
+        db/obsolete_files_test.cc
         db/external_sst_file_basic_test.cc
         db/external_sst_file_test.cc
         db/fault_injection_test.cc
@@ -924,6 +972,7 @@ if(WITH_TESTS)
         utilities/transactions/optimistic_transaction_test.cc
         utilities/transactions/transaction_test.cc
         utilities/transactions/write_prepared_transaction_test.cc
+        utilities/transactions/write_unprepared_transaction_test.cc
         utilities/ttl/ttl_test.cc
         utilities/write_batch_with_index/write_batch_with_index_test.cc
   )

HISTORY.md

@@ -1,24 +1,56 @@
 # Rocksdb Change Log
-## 5.13.4 (6/12/2018)
-### Bug Fixes
-* Fix regression bug of Prev() with ReadOptions.iterate_upper_bound.
-
-## 5.13.3 (6/6/2018)
-### Bug Fixes
-* Fix assertion when reading bloom filter of SST files containing range deletions but no data
-
-## 5.13.2 (5/25/2018)
+## Unreleased
 ### Public API Change
+* For users of `Statistics` objects created via `CreateDBStatistics()`, the format of the string returned by its `ToString()` method has changed.
+* With LRUCache, when high_pri_pool_ratio > 0, midpoint insertion strategy will be enabled to put low-pri items to the tail of low-pri list (the midpoint) when they first inserted into the cache. This is to make cache entries never get hit age out faster, improving cache efficiency when large background scan presents.
+* For bottommost_compression, a compatible CompressionOptions is added via `bottommost_compression_opts`. To keep backward compatible, a new boolean `enabled` is added to CompressionOptions. For compression_opts, it will be always used no matter what value of `enabled` is. For bottommost_compression_opts, it will only be used when user set `enabled=true`, otherwise, compression_opts will be used for bottommost_compression as default.
+* The "rocksdb.num.entries" table property no longer counts range deletion tombstones as entries.
+
+### New Features
+* Changes the format of index blocks by storing the key in their raw form rather than converting them to InternalKey. This saves 8 bytes per index key. The feature is backward compatbile but not forward compatible. It is disabled by default unless format_version 3 or above is used.
+* Add a new table property, "rocksdb.num.range-deletions", which counts the number of range deletion tombstones in the table.
+* Improve the performance of iterators doing long range scans by using readahead, when using direct IO.
+* pin_top_level_index_and_filter (default true) in BlockBasedTableOptions can be used in combination with cache_index_and_filter_blocks to prefetch and pin the top-level index of partitioned index and filter blocks in cache. It has no impact when cache_index_and_filter_blocks is false.
+* Avoid memcpy when reading mmap files with OpenReadOnly and max_open_files==-1
+
+### Bug Fixes
+* fix deadlock with enable_pipelined_write=true and max_successive_merges > 0
+* Fix corruption in non-iterator reads when mmap is used for file reads
+* Fix bug with prefix search in partition filters where a shared prefix would be ignored from the later partitions. The bug could report an eixstent key as missing. The bug could be triggered if prefix_extractor is set and partition filters is enabled.
+* Change default value of `bytes_max_delete_chunk` to 0 in NewSstFileManager() as it doesn't work well with checkpoints.
+
+## 5.14.0 (5/16/2018)
+### Public API Change
+* Add a BlockBasedTableOption to align uncompressed data blocks on the smaller of block size or page size boundary, to reduce flash reads by avoiding reads spanning 4K pages.
+* The background thread naming convention changed (on supporting platforms) to "rocksdb:<thread pool priority><thread number>", e.g., "rocksdb:low0".
+* Add a new ticker stat rocksdb.number.multiget.keys.found to count number of keys successfully read in MultiGet calls
+* Touch-up to write-related counters in PerfContext. New counters added: write_scheduling_flushes_compactions_time, write_thread_wait_nanos. Counters whose behavior was fixed or modified: write_memtable_time, write_pre_and_post_process_time, write_delay_time.
+* Posix Env's NewRandomRWFile() will fail if the file doesn't exist.
+* Now, `DBOptions::use_direct_io_for_flush_and_compaction` only applies to background writes, and `DBOptions::use_direct_reads` applies to both user reads and background reads. This conforms with Linux's `open(2)` manpage, which advises against simultaneously reading a file in buffered and direct modes, due to possibly undefined behavior and degraded performance.
+* Iterator::Valid() always returns false if !status().ok(). So, now when doing a Seek() followed by some Next()s, there's no need to check status() after every operation.
+* Iterator::Seek()/SeekForPrev()/SeekToFirst()/SeekToLast() always resets status().
 * Introduced `CompressionOptions::kDefaultCompressionLevel`, which is a generic way to tell RocksDB to use the compression library's default level. It is now the default value for `CompressionOptions::level`. Previously the level defaulted to -1, which gave poor compression ratios in ZSTD.
 
-### Bug Fixes
-* Fix segfault caused by object premature destruction (PR #3898)
-* Fix an issue with unnecessary capture in lambda expressions (PR #3904)
-
-## 5.13.1 (4/30/2018)
 ### New Features
+* Introduce TTL for level compaction so that all files older than ttl go through the compaction process to get rid of old data.
+* TransactionDBOptions::write_policy can be configured to enable WritePrepared 2PC transactions. Read more about them in the wiki.
+* Add DB properties "rocksdb.block-cache-capacity", "rocksdb.block-cache-usage", "rocksdb.block-cache-pinned-usage" to show block cache usage.
 * Add `Env::LowerThreadPoolCPUPriority(Priority)` method, which lowers the CPU priority of background (esp. compaction) threads to minimize interference with foreground tasks.
-* Eliminate use of temporary directories in BackupEngine to improve reliability on distributed file systems.
+* Fsync parent directory after deleting a file in delete scheduler.
+* In level-based compaction, if bottom-pri thread pool was setup via `Env::SetBackgroundThreads()`, compactions to the bottom level will be delegated to that thread pool.
+* `prefix_extractor` has been moved from ImmutableCFOptions to MutableCFOptions, meaning it can be dynamically changed without a DB restart.
+
+### Bug Fixes
+* Fsync after writing global seq number to the ingestion file in ExternalSstFileIngestionJob.
+* Fix WAL corruption caused by race condition between user write thread and FlushWAL when two_write_queue is not set.
+* Fix `BackupableDBOptions::max_valid_backups_to_open` to not delete backup files when refcount cannot be accurately determined.
+* Fix memory leak when pin_l0_filter_and_index_blocks_in_cache is used with partitioned filters
+* Disable rollback of merge operands in WritePrepared transactions to work around an issue in MyRocks. It can be enabled back by setting TransactionDBOptions::rollback_merge_operands to true.
+* Fix wrong results by ReverseBytewiseComparator::FindShortSuccessor()
+
+### Java API Changes
+* Add `BlockBasedTableConfig.setBlockCache` to allow sharing a block cache across DB instances.
+* Added SstFileManager to the Java API to allow managing SST files across DB instances.
 
 ## 5.13.0 (3/20/2018)
 ### Public API Change
@@ -30,15 +62,11 @@
 * Avoid unnecessarily flushing in `CompactRange()` when the range specified by the user does not overlap unflushed memtables.
 * If `ColumnFamilyOptions::max_subcompactions` is set greater than one, we now parallelize large manual level-based compactions.
 * Add "rocksdb.live-sst-files-size" DB property to return total bytes of all SST files belong to the latest LSM tree.
+* NewSstFileManager to add an argument bytes_max_delete_chunk with default 64MB. With this argument, a file larger than 64MB will be ftruncated multiple times based on this size.
 
 ### Bug Fixes
 * Fix a leak in prepared_section_completed_ where the zeroed entries would not removed from the map.
 * Fix WAL corruption caused by race condition between user write thread and backup/checkpoint thread.
-* Fsync after writing global seq number to the ingestion file in ExternalSstFileIngestionJob.
-* Fix memory leak when pin_l0_filter_and_index_blocks_in_cache is used with partitioned filters
-
-### Java API Changes
-* Add `BlockBasedTableConfig.setBlockCache` to allow sharing a block cache across DB instances.
 
 ## 5.12.0 (2/14/2018)
 ### Public API Change

LANGUAGE-BINDINGS.md

@@ -10,9 +10,11 @@ This is the list of all known third-party language bindings for RocksDB. If some
 * PHP - https://github.com/Photonios/rocksdb-php
 * C# - https://github.com/warrenfalk/rocksdb-sharp
 * Rust
+    * https://github.com/pingcap/rust-rocksdb (used in production fork of https://github.com/spacejam/rust-rocksdb)
     * https://github.com/spacejam/rust-rocksdb
     * https://github.com/bh1xuw/rust-rocks
 * D programming language - https://github.com/b1naryth1ef/rocksdb
 * Erlang - https://gitlab.com/barrel-db/erlang-rocksdb
 * Elixir - https://github.com/urbint/rox
 * Nim - https://github.com/status-im/nim-rocksdb
+* Swift and Objective-C (iOS/OSX) - https://github.com/iabudiab/ObjectiveRocks 

Makefile

@@ -76,13 +76,19 @@ ifeq ($(MAKECMDGOALS),install)
 endif
 
 ifeq ($(MAKECMDGOALS),rocksdbjavastatic)
-	DEBUG_LEVEL=0
+	ifneq ($(DEBUG_LEVEL),2)
+		DEBUG_LEVEL=0
+	endif
 endif
 
 ifeq ($(MAKECMDGOALS),rocksdbjavastaticrelease)
 	DEBUG_LEVEL=0
 endif
 
+ifeq ($(MAKECMDGOALS),rocksdbjavastaticreleasedocker)
+        DEBUG_LEVEL=0
+endif
+
 ifeq ($(MAKECMDGOALS),rocksdbjavastaticpublish)
 	DEBUG_LEVEL=0
 endif
@@ -96,6 +102,10 @@ OPT += -momit-leaf-frame-pointer
 endif
 endif
 
+ifeq (,$(shell $(CXX) -fsyntax-only -faligned-new -xc++ /dev/null 2>&1))
+CXXFLAGS += -faligned-new -DHAVE_ALIGNED_NEW
+endif
+
 ifeq (,$(shell $(CXX) -fsyntax-only -maltivec -xc /dev/null 2>&1))
 CXXFLAGS += -DHAS_ALTIVEC
 CFLAGS += -DHAS_ALTIVEC
@@ -222,6 +232,9 @@ ifdef COMPILE_WITH_TSAN
 	PROFILING_FLAGS =
 	# LUA is not supported under TSAN
 	LUA_PATH =
+	# Limit keys for crash test under TSAN to avoid error:
+	# "ThreadSanitizer: DenseSlabAllocator overflow. Dying."
+	CRASH_TEST_EXT_ARGS += --max_key=1000000
 endif
 
 # AIX doesn't work with -pg
@@ -276,7 +289,7 @@ endif
 default: all
 
 WARNING_FLAGS = -W -Wextra -Wall -Wsign-compare -Wshadow \
-  -Wno-unused-parameter
+  -Wunused-parameter
 
 ifeq ($(PLATFORM), OS_OPENBSD)
 	WARNING_FLAGS += -Wno-unused-lambda-capture
@@ -397,6 +410,7 @@ TESTS = \
 	db_blob_index_test \
 	db_bloom_filter_test \
 	db_iter_test \
+	db_iter_stress_test \
 	db_log_iter_test \
 	db_compaction_filter_test \
 	db_compaction_test \
@@ -415,6 +429,7 @@ TESTS = \
 	db_table_properties_test \
 	db_statistics_test \
 	db_write_test \
+	error_handler_test \
 	autovector_test \
 	blob_db_test \
 	cleanable_test \
@@ -474,6 +489,7 @@ TESTS = \
 	write_batch_with_index_test \
 	write_controller_test\
 	deletefile_test \
+	obsolete_files_test \
 	table_test \
 	geodb_test \
 	delete_scheduler_test \
@@ -511,6 +527,7 @@ TESTS = \
 	repair_test \
 	env_timed_test \
 	write_prepared_transaction_test \
+	write_unprepared_transaction_test \
 	db_universal_compaction_test \
 
 PARALLEL_TEST = \
@@ -530,6 +547,7 @@ PARALLEL_TEST = \
 	table_test \
 	transaction_test \
 	write_prepared_transaction_test \
+	write_unprepared_transaction_test \
 
 # options_settable_test doesn't pass with UBSAN as we use hack in the test
 ifdef COMPILE_WITH_UBSAN
@@ -642,7 +660,7 @@ ifeq ($(HAVE_POWER8),1)
 shared_all_libobjects = $(shared_libobjects) $(shared-ppc-objects)
 endif
 $(SHARED4): $(shared_all_libobjects)
-	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED3) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(shared_libobjects) $(LDFLAGS) -o $@
+	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED3) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(shared_all_libobjects) $(LDFLAGS) -o $@
 
 endif  # PLATFORM_SHARED_EXT
 
@@ -795,7 +813,7 @@ check_0:
 	  | grep -E '$(tests-regexp)'					\
 	  | build_tools/gnu_parallel -j$(J) --plain --joblog=LOG $$eta --gnu '{} >& t/log-{/}'
 
-valgrind-blacklist-regexp = InlineSkipTest.ConcurrentInsert|TransactionTest.DeadlockStress|DBCompactionTest.SuggestCompactRangeNoTwoLevel0Compactions|BackupableDBTest.RateLimiting|DBTest.CloseSpeedup|DBTest.ThreadStatusFlush|DBTest.RateLimitingTest|DBTest.EncodeDecompressedBlockSizeTest|FaultInjectionTest.UninstalledCompaction|HarnessTest.Randomized|ExternalSSTFileTest.CompactDuringAddFileRandom|ExternalSSTFileTest.IngestFileWithGlobalSeqnoRandomized
+valgrind-blacklist-regexp = InlineSkipTest.ConcurrentInsert|TransactionStressTest.DeadlockStress|DBCompactionTest.SuggestCompactRangeNoTwoLevel0Compactions|BackupableDBTest.RateLimiting|DBTest.CloseSpeedup|DBTest.ThreadStatusFlush|DBTest.RateLimitingTest|DBTest.EncodeDecompressedBlockSizeTest|FaultInjectionTest.UninstalledCompaction|HarnessTest.Randomized|ExternalSSTFileTest.CompactDuringAddFileRandom|ExternalSSTFileTest.IngestFileWithGlobalSeqnoRandomized|MySQLStyleTransactionTest.TransactionStressTest
 
 .PHONY: valgrind_check_0
 valgrind_check_0:
@@ -1177,6 +1195,9 @@ db_statistics_test: db/db_statistics_test.o db/db_test_util.o $(LIBOBJECTS) $(TE
 db_write_test: db/db_write_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+error_handler_test: db/error_handler_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 external_sst_file_basic_test: db/external_sst_file_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1189,6 +1210,9 @@ db_tailing_iter_test: db/db_tailing_iter_test.o db/db_test_util.o $(LIBOBJECTS)
 db_iter_test: db/db_iter_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+db_iter_stress_test: db/db_iter_stress_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_universal_compaction_test: db/db_universal_compaction_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1380,6 +1404,9 @@ options_file_test: db/options_file_test.o $(LIBOBJECTS) $(TESTHARNESS)
 deletefile_test: db/deletefile_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+obsolete_files_test: db/obsolete_files_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 geodb_test: utilities/geodb/geodb_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1461,6 +1488,9 @@ transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TEST
 write_prepared_transaction_test: utilities/transactions/write_prepared_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+write_unprepared_transaction_test: utilities/transactions/write_unprepared_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 	$(AM_LINK)
 
@@ -1715,7 +1745,9 @@ rocksdbjavastatic: $(java_static_all_libobjects)
 	  -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) \
 	  $(java_static_all_libobjects) $(COVERAGEFLAGS) \
 	  $(JAVA_COMPRESSIONS) $(JAVA_STATIC_LDFLAGS)
-	cd java/target;strip $(STRIPFLAGS) $(ROCKSDBJNILIB)
+	cd java/target;if [ "$(DEBUG_LEVEL)" == "0" ]; then \
+		strip $(STRIPFLAGS) $(ROCKSDBJNILIB); \
+	fi
 	cd java;jar -cf target/$(ROCKSDB_JAR) HISTORY*.md
 	cd java/target;jar -uf $(ROCKSDB_JAR) $(ROCKSDBJNILIB)
 	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR) org/rocksdb/*.class org/rocksdb/util/*.class

README.md

@@ -4,7 +4,6 @@
 [![Windows Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
 [![PPC64le Build Status](http://140.211.168.68:8080/buildStatus/icon?job=Rocksdb)](http://140.211.168.68:8080/job/Rocksdb)
 
-
 RocksDB is developed and maintained by Facebook Database Engineering Team.
 It is built on earlier work on LevelDB by Sanjay Ghemawat (sanjay@google.com)
 and Jeff Dean (jeff@google.com)

TARGETS

@@ -25,6 +25,10 @@ rocksdb_compiler_flags = [
     "-DTBB",
     # Needed to compile in fbcode
     "-Wno-expansion-to-defined",
+    # Added missing flags from output of build_detect_platform
+    "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
+    "-DROCKSDB_BACKTRACE",
+    "-Wnarrowing",
 ]
 
 rocksdb_external_deps = [
@@ -89,6 +93,7 @@ cpp_library(
         "db/db_info_dumper.cc",
         "db/db_iter.cc",
         "db/dbformat.cc",
+        "db/error_handler.cc",
         "db/event_helpers.cc",
         "db/experimental.cc",
         "db/external_sst_file_ingestion_job.cc",
@@ -99,6 +104,7 @@ cpp_library(
         "db/internal_stats.cc",
         "db/log_reader.cc",
         "db/log_writer.cc",
+        "db/logs_with_prep_tracker.cc",
         "db/malloc_stats.cc",
         "db/managed_iterator.cc",
         "db/memtable.cc",
@@ -195,6 +201,7 @@ cpp_library(
         "util/coding.cc",
         "util/compaction_job_stats_impl.cc",
         "util/comparator.cc",
+        "util/compression_context_cache.cc",
         "util/concurrent_arena.cc",
         "util/crc32c.cc",
         "util/delete_scheduler.cc",
@@ -215,6 +222,7 @@ cpp_library(
         "util/status_message.cc",
         "util/string_util.cc",
         "util/sync_point.cc",
+        "util/sync_point_impl.cc",
         "util/thread_local.cc",
         "util/threadpool_imp.cc",
         "util/transaction_test_util.cc",
@@ -274,6 +282,8 @@ cpp_library(
         "utilities/transactions/transaction_util.cc",
         "utilities/transactions/write_prepared_txn.cc",
         "utilities/transactions/write_prepared_txn_db.cc",
+        "utilities/transactions/write_unprepared_txn.cc",
+        "utilities/transactions/write_unprepared_txn_db.cc",
         "utilities/ttl/db_ttl_impl.cc",
         "utilities/write_batch_with_index/write_batch_with_index.cc",
         "utilities/write_batch_with_index/write_batch_with_index_internal.cc",
@@ -549,6 +559,11 @@ ROCKS_TESTS = [
         "db/db_io_failure_test.cc",
         "serial",
     ],
+    [
+        "db_iter_stress_test",
+        "db/db_iter_stress_test.cc",
+        "serial",
+    ],
     [
         "db_iter_test",
         "db/db_iter_test.cc",
@@ -674,6 +689,11 @@ ROCKS_TESTS = [
         "utilities/env_timed_test.cc",
         "serial",
     ],
+    [
+        "error_handler_test",
+        "db/error_handler_test.cc",
+        "serial",
+    ],
     [
         "event_logger_test",
         "util/event_logger_test.cc",
@@ -824,6 +844,11 @@ ROCKS_TESTS = [
         "utilities/object_registry_test.cc",
         "serial",
     ],
+    [
+        "obsolete_files_test",
+        "db/obsolete_files_test.cc",
+        "serial",
+    ],
     [
         "optimistic_transaction_test",
         "utilities/transactions/optimistic_transaction_test.cc",
@@ -1024,6 +1049,11 @@ ROCKS_TESTS = [
         "utilities/transactions/write_prepared_transaction_test.cc",
         "parallel",
     ],
+    [
+        "write_unprepared_transaction_test",
+        "utilities/transactions/write_unprepared_transaction_test.cc",
+        "parallel",
+    ],
 ]
 
 # Generate a test rule for each entry in ROCKS_TESTS
@@ -1053,14 +1083,3 @@ if not is_opt_mode:
           command = [TEST_RUNNER, BUCK_BINS + test_bin]
         )
 
-custom_unittest(
-    name = "make_rocksdbjavastatic",
-    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
-    type = "simple",
-)
-
-custom_unittest(
-    name = "make_rocksdb_lite_release",
-    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
-    type = "simple",
-)

USERS.md

@@ -26,7 +26,7 @@ Learn more about those use cases in a Tech Talk by Ankit Gupta and Naveen Somasu
 Yahoo is using RocksDB as a storage engine for their biggest distributed data store Sherpa. Learn more about it here: http://yahooeng.tumblr.com/post/120730204806/sherpa-scales-new-heights
 
 ## CockroachDB
-CockroachDB is an open-source geo-replicated transactional database (still in development). They are using RocksDB as their storage engine. Check out their github: https://github.com/cockroachdb/cockroach
+CockroachDB is an open-source geo-replicated transactional database. They are using RocksDB as their storage engine. Check out their github: https://github.com/cockroachdb/cockroach
 
 ## DNANexus
 DNANexus is using RocksDB to speed up processing of genomics data.

WINDOWS_PORT.md

@@ -45,7 +45,7 @@ We plan to use this port for our business purposes here at Bing and this provide
 * All posix specific headers were replaced to port/port.h which worked well
 * Replaced `dirent.h` for `port/dirent.h` (very few places) with the implementation of the relevant interfaces within `rocksdb::port` namespace
 * Replaced `sys/time.h` to `port/sys_time.h` (few places) implemented equivalents within `rocksdb::port`
-* `printf %z` specification is not supported on Windows. To imitate existing standards we came up with a string macro `ROCKSDB_PRIszt` which expands to `%z` on posix systems and to Iu on windows.
+* `printf %z` specification is not supported on Windows. To imitate existing standards we came up with a string macro `ROCKSDB_PRIszt` which expands to `zu` on posix systems and to `Iu` on windows.
 * in class member initialization were moved to a __ctors in some cases
 * `constexpr` is not supported. We had to replace `std::numeric_limits<>::max/min()` to its C macros for constants. Sometimes we had to make class members `static const` and place a definition within a .cc file.
 * `constexpr` for functions was replaced to a template specialization (1 place)

buckifier/buckify_rocksdb.py

@@ -3,14 +3,11 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
 from targets_builder import TARGETSBuilder
-from optparse import OptionParser
 import os
 import fnmatch
 import sys
-import tempfile
 
 from util import ColorString
-import util
 
 # tests to export as libraries for inclusion in other projects
 _EXPORTED_TEST_LIBS = ["env_basic_test"]

buckifier/targets_cfg.py

@@ -29,6 +29,10 @@ rocksdb_compiler_flags = [
     "-DTBB",
     # Needed to compile in fbcode
     "-Wno-expansion-to-defined",
+    # Added missing flags from output of build_detect_platform
+    "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
+    "-DROCKSDB_BACKTRACE",
+    "-Wnarrowing",
 ]
 
 rocksdb_external_deps = [
@@ -131,15 +135,4 @@ if not is_opt_mode:
           command = [TEST_RUNNER, BUCK_BINS + test_bin]
         )
 
-custom_unittest(
-    name = "make_rocksdbjavastatic",
-    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
-    type = "simple",
-)
-
-custom_unittest(
-    name = "make_rocksdb_lite_release",
-    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
-    type = "simple",
-)
 """

build_tools/build_detect_platform

@@ -1,4 +1,4 @@
-#!/bin/sh
+#!/usr/bin/env bash
 #
 # Detects OS we're compiling on and outputs a file specified by the first
 # argument, which in turn gets read while processing Makefile.
@@ -66,11 +66,23 @@ rm -f "$OUTPUT"
 touch "$OUTPUT"
 
 if test -z "$CC"; then
-   CC=cc
+    if [ -x "$(command -v cc)" ]; then
+        CC=cc
+    elif [ -x "$(command -v clang)" ]; then
+        CC=clang
+    else
+        CC=cc
+    fi
 fi
 
 if test -z "$CXX"; then
-    CXX=g++
+    if [ -x "$(command -v g++)" ]; then
+        CXX=g++
+    elif [ -x "$(command -v clang++)" ]; then
+        CXX=clang++
+    else
+        CXX=g++
+    fi
 fi
 
 # Detect OS
@@ -87,7 +99,7 @@ if test -z "$CLANG_SCAN_BUILD"; then
 fi
 
 if test -z "$CLANG_ANALYZER"; then
-    CLANG_ANALYZER=$(which clang++ 2> /dev/null)
+    CLANG_ANALYZER=$(command -v clang++ 2> /dev/null)
 fi
 
 if test -z "$FIND"; then
@@ -132,6 +144,8 @@ case "$TARGET_OS" in
         COMMON_FLAGS="$COMMON_FLAGS -DOS_LINUX"
         if [ -z "$USE_CLANG" ]; then
             COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
+        else
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
         fi
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
         # PORT_FILES=port/linux/linux_specific.cc
@@ -184,6 +198,8 @@ case "$TARGET_OS" in
         COMMON_FLAGS="$COMMON_FLAGS -DCYGWIN"
         if [ -z "$USE_CLANG" ]; then
             COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
+        else
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
         fi
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
         # PORT_FILES=port/linux/linux_specific.cc
@@ -498,9 +514,9 @@ if test -z "$PORTABLE"; then
   elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
     # TODO: Handle this with approprite options.
     COMMON_FLAGS="$COMMON_FLAGS"
-  elif [ "$TARGET_OS" == IOS ]; then
+  elif [ "$TARGET_OS" == "IOS" ]; then
     COMMON_FLAGS="$COMMON_FLAGS"
-  elif [ "$TARGET_OS" != AIX ] && [ "$TARGET_OS" != SunOS ]; then
+  elif [ "$TARGET_OS" != "AIX" ] && [ "$TARGET_OS" != "SunOS" ]; then
     COMMON_FLAGS="$COMMON_FLAGS -march=native "
   elif test "$USE_SSE"; then
     COMMON_FLAGS="$COMMON_FLAGS -msse4.2 -mpclmul"

build_tools/version.sh

@@ -1,4 +1,4 @@
-#!/bin/sh
+#!/usr/bin/env bash
 if [ "$#" = "0" ]; then
   echo "Usage: $0 major|minor|patch|full"
   exit 1

cache/cache_bench.cc

@@ -52,7 +52,7 @@ namespace rocksdb {
 
 class CacheBench;
 namespace {
-void deleter(const Slice& key, void* value) {
+void deleter(const Slice& /*key*/, void* value) {
     delete reinterpret_cast<char *>(value);
 }
 

cache/clock_cache.cc

@@ -13,8 +13,8 @@
 
 namespace rocksdb {
 
-std::shared_ptr<Cache> NewClockCache(size_t capacity, int num_shard_bits,
-                                     bool strict_capacity_limit) {
+std::shared_ptr<Cache> NewClockCache(size_t /*capacity*/, int /*num_shard_bits*/,
+                                     bool /*strict_capacity_limit*/) {
   // Clock cache not supported.
   return nullptr;
 }
@@ -367,7 +367,9 @@ ClockCacheShard::~ClockCacheShard() {
   for (auto& handle : list_) {
     uint32_t flags = handle.flags.load(std::memory_order_relaxed);
     if (InCache(flags) || CountRefs(flags) > 0) {
-      (*handle.deleter)(handle.key, handle.value);
+      if (handle.deleter != nullptr) {
+        (*handle.deleter)(handle.key, handle.value);
+      }
       delete[] handle.key.data();
     }
   }

cache/lru_cache.cc

@@ -99,14 +99,20 @@ void LRUHandleTable::Resize() {
   length_ = new_length;
 }
 
-LRUCacheShard::LRUCacheShard()
-    : capacity_(0), high_pri_pool_usage_(0), strict_capacity_limit_(false),
-      high_pri_pool_ratio_(0), high_pri_pool_capacity_(0), usage_(0),
+LRUCacheShard::LRUCacheShard(size_t capacity, bool strict_capacity_limit,
+                             double high_pri_pool_ratio)
+    : capacity_(0),
+      high_pri_pool_usage_(0),
+      strict_capacity_limit_(strict_capacity_limit),
+      high_pri_pool_ratio_(high_pri_pool_ratio),
+      high_pri_pool_capacity_(0),
+      usage_(0),
       lru_usage_(0) {
   // Make empty circular linked list
   lru_.next = &lru_;
   lru_.prev = &lru_;
   lru_low_pri_ = &lru_;
+  SetCapacity(capacity);
 }
 
 LRUCacheShard::~LRUCacheShard() {}
@@ -193,7 +199,7 @@ void LRUCacheShard::LRU_Remove(LRUHandle* e) {
 void LRUCacheShard::LRU_Insert(LRUHandle* e) {
   assert(e->next == nullptr);
   assert(e->prev == nullptr);
-  if (high_pri_pool_ratio_ > 0 && e->IsHighPri()) {
+  if (high_pri_pool_ratio_ > 0 && (e->IsHighPri() || e->HasHit())) {
     // Inset "e" to head of LRU list.
     e->next = &lru_;
     e->prev = lru_.prev;
@@ -240,22 +246,6 @@ void LRUCacheShard::EvictFromLRU(size_t charge,
   }
 }
 
-void* LRUCacheShard::operator new(size_t size) {
-  return port::cacheline_aligned_alloc(size);
-}
-
-void* LRUCacheShard::operator new[](size_t size) {
-  return port::cacheline_aligned_alloc(size);
-}
-
-void LRUCacheShard::operator delete(void *memblock) {
-  port::cacheline_aligned_free(memblock);
-}
-
-void LRUCacheShard::operator delete[](void* memblock) {
-  port::cacheline_aligned_free(memblock);
-}
-
 void LRUCacheShard::SetCapacity(size_t capacity) {
   autovector<LRUHandle*> last_reference_list;
   {
@@ -285,6 +275,7 @@ Cache::Handle* LRUCacheShard::Lookup(const Slice& key, uint32_t hash) {
       LRU_Remove(e);
     }
     e->refs++;
+    e->SetHit();
   }
   return reinterpret_cast<Cache::Handle*>(e);
 }
@@ -473,15 +464,24 @@ LRUCache::LRUCache(size_t capacity, int num_shard_bits,
                    bool strict_capacity_limit, double high_pri_pool_ratio)
     : ShardedCache(capacity, num_shard_bits, strict_capacity_limit) {
   num_shards_ = 1 << num_shard_bits;
-  shards_ = new LRUCacheShard[num_shards_];
-  SetCapacity(capacity);
-  SetStrictCapacityLimit(strict_capacity_limit);
+  shards_ = reinterpret_cast<LRUCacheShard*>(
+      port::cacheline_aligned_alloc(sizeof(LRUCacheShard) * num_shards_));
+  size_t per_shard = (capacity + (num_shards_ - 1)) / num_shards_;
   for (int i = 0; i < num_shards_; i++) {
-    shards_[i].SetHighPriorityPoolRatio(high_pri_pool_ratio);
+    new (&shards_[i])
+        LRUCacheShard(per_shard, strict_capacity_limit, high_pri_pool_ratio);
   }
 }
 
-LRUCache::~LRUCache() { delete[] shards_; }
+LRUCache::~LRUCache() {
+  if (shards_ != nullptr) {
+    assert(num_shards_ > 0);
+    for (int i = 0; i < num_shards_; i++) {
+      shards_[i].~LRUCacheShard();
+    }
+    port::cacheline_aligned_free(shards_);
+  }
+}
 
 CacheShard* LRUCache::GetShard(int shard) {
   return reinterpret_cast<CacheShard*>(&shards_[shard]);
@@ -505,9 +505,17 @@ uint32_t LRUCache::GetHash(Handle* handle) const {
 
 void LRUCache::DisownData() {
 // Do not drop data if compile with ASAN to suppress leak warning.
+#if defined(__clang__)
+#if !defined(__has_feature) || !__has_feature(address_sanitizer)
+  shards_ = nullptr;
+  num_shards_ = 0;
+#endif
+#else  // __clang__
 #ifndef __SANITIZE_ADDRESS__
   shards_ = nullptr;
+  num_shards_ = 0;
 #endif  // !__SANITIZE_ADDRESS__
+#endif  // __clang__
 }
 
 size_t LRUCache::TEST_GetLRUSize() {

cache/lru_cache.h

@@ -77,6 +77,7 @@ struct LRUHandle {
   bool InCache() { return flags & 1; }
   bool IsHighPri() { return flags & 2; }
   bool InHighPriPool() { return flags & 4; }
+  bool HasHit() { return flags & 8; }
 
   void SetInCache(bool in_cache) {
     if (in_cache) {
@@ -102,6 +103,8 @@ struct LRUHandle {
     }
   }
 
+  void SetHit() { flags |= 8; }
+
   void Free() {
     assert((refs == 1 && InCache()) || (refs == 0 && !InCache()));
     if (deleter) {
@@ -156,7 +159,8 @@ class LRUHandleTable {
 // A single shard of sharded cache.
 class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard : public CacheShard {
  public:
-  LRUCacheShard();
+  LRUCacheShard(size_t capacity, bool strict_capacity_limit,
+                double high_pri_pool_ratio);
   virtual ~LRUCacheShard();
 
   // Separate from constructor so caller can easily make an array of LRUCache
@@ -205,15 +209,6 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard : public CacheShard {
   //  Retrives high pri pool ratio
   double GetHighPriPoolRatio();
 
-  // Overloading to aligned it to cache line size
-  void* operator new(size_t);
-
-  void* operator new[](size_t);
-
-  void operator delete(void *);
-
-  void operator delete[](void*);
-
  private:
   void LRU_Remove(LRUHandle* e);
   void LRU_Insert(LRUHandle* e);
@@ -300,7 +295,7 @@ class LRUCache : public ShardedCache {
   double GetHighPriPoolRatio();
 
  private:
-  LRUCacheShard* shards_;
+  LRUCacheShard* shards_ = nullptr;
   int num_shards_ = 0;
 };
 

cache/lru_cache_test.cc

@@ -7,6 +7,7 @@
 
 #include <string>
 #include <vector>
+#include "port/port.h"
 #include "util/testharness.h"
 
 namespace rocksdb {
@@ -14,22 +15,22 @@ namespace rocksdb {
 class LRUCacheTest : public testing::Test {
  public:
   LRUCacheTest() {}
-  ~LRUCacheTest() {}
+  ~LRUCacheTest() { DeleteCache(); }
+
+  void DeleteCache() {
+    if (cache_ != nullptr) {
+      cache_->~LRUCacheShard();
+      port::cacheline_aligned_free(cache_);
+      cache_ = nullptr;
+    }
+  }
 
   void NewCache(size_t capacity, double high_pri_pool_ratio = 0.0) {
-    cache_.reset(
-#if defined(_MSC_VER)
-#pragma warning(push)
-#pragma warning(disable: 4316) // We've validated the alignment with the new operators
-#endif
-      new LRUCacheShard()
-#if defined(_MSC_VER)
-#pragma warning(pop)
-#endif
-    );
-    cache_->SetCapacity(capacity);
-    cache_->SetStrictCapacityLimit(false);
-    cache_->SetHighPriorityPoolRatio(high_pri_pool_ratio);
+    DeleteCache();
+    cache_ = reinterpret_cast<LRUCacheShard*>(
+        port::cacheline_aligned_alloc(sizeof(LRUCacheShard)));
+    new (cache_) LRUCacheShard(capacity, false /*strict_capcity_limit*/,
+                               high_pri_pool_ratio);
   }
 
   void Insert(const std::string& key,
@@ -85,7 +86,7 @@ class LRUCacheTest : public testing::Test {
   }
 
  private:
-  std::unique_ptr<LRUCacheShard> cache_;
+  LRUCacheShard* cache_ = nullptr;
 };
 
 TEST_F(LRUCacheTest, BasicLRU) {
@@ -114,7 +115,30 @@ TEST_F(LRUCacheTest, BasicLRU) {
   ValidateLRUList({"e", "z", "d", "u", "v"});
 }
 
-TEST_F(LRUCacheTest, MidPointInsertion) {
+TEST_F(LRUCacheTest, MidpointInsertion) {
+  // Allocate 2 cache entries to high-pri pool.
+  NewCache(5, 0.45);
+
+  Insert("a", Cache::Priority::LOW);
+  Insert("b", Cache::Priority::LOW);
+  Insert("c", Cache::Priority::LOW);
+  Insert("x", Cache::Priority::HIGH);
+  Insert("y", Cache::Priority::HIGH);
+  ValidateLRUList({"a", "b", "c", "x", "y"}, 2);
+
+  // Low-pri entries inserted to the tail of low-pri list (the midpoint).
+  // After lookup, it will move to the tail of the full list.
+  Insert("d", Cache::Priority::LOW);
+  ValidateLRUList({"b", "c", "d", "x", "y"}, 2);
+  ASSERT_TRUE(Lookup("d"));
+  ValidateLRUList({"b", "c", "x", "y", "d"}, 2);
+
+  // High-pri entries will be inserted to the tail of full list.
+  Insert("z", Cache::Priority::HIGH);
+  ValidateLRUList({"c", "x", "y", "d", "z"}, 2);
+}
+
+TEST_F(LRUCacheTest, EntriesWithPriority) {
   // Allocate 2 cache entries to high-pri pool.
   NewCache(5, 0.45);
 
@@ -140,15 +164,15 @@ TEST_F(LRUCacheTest, MidPointInsertion) {
   Insert("a", Cache::Priority::LOW);
   ValidateLRUList({"v", "X", "a", "Y", "Z"}, 2);
 
-  // Low-pri entries will be inserted to head of low-pri pool after lookup.
+  // Low-pri entries will be inserted to head of high-pri pool after lookup.
   ASSERT_TRUE(Lookup("v"));
-  ValidateLRUList({"X", "a", "v", "Y", "Z"}, 2);
+  ValidateLRUList({"X", "a", "Y", "Z", "v"}, 2);
 
   // High-pri entries will be inserted to the head of the list after lookup.
   ASSERT_TRUE(Lookup("X"));
-  ValidateLRUList({"a", "v", "Y", "Z", "X"}, 2);
+  ValidateLRUList({"a", "Y", "Z", "v", "X"}, 2);
   ASSERT_TRUE(Lookup("Z"));
-  ValidateLRUList({"a", "v", "Y", "X", "Z"}, 2);
+  ValidateLRUList({"a", "Y", "v", "X", "Z"}, 2);
 
   Erase("Y");
   ValidateLRUList({"a", "v", "X", "Z"}, 2);
@@ -161,7 +185,7 @@ TEST_F(LRUCacheTest, MidPointInsertion) {
   Insert("g", Cache::Priority::LOW);
   ValidateLRUList({"d", "e", "f", "g", "Z"}, 1);
   ASSERT_TRUE(Lookup("d"));
-  ValidateLRUList({"e", "f", "g", "d", "Z"}, 1);
+  ValidateLRUList({"e", "f", "g", "Z", "d"}, 2);
 }
 
 }  // namespace rocksdb

cmake/modules/FindTBB.cmake

@@ -5,13 +5,17 @@
 # TBB_LIBRARIES - List of libraries when using TBB.
 # TBB_FOUND - True if TBB found.
 
+if(NOT DEFINED TBB_ROOT_DIR)
+  set(TBB_ROOT_DIR "$ENV{TBBROOT}")
+endif()
+
 find_path(TBB_INCLUDE_DIR
 NAMES tbb/tbb.h
 HINTS ${TBB_ROOT_DIR}/include)
 
 find_library(TBB_LIBRARIES
 NAMES tbb
-HINTS ${TBB_ROOT_DIR}/lib)
+HINTS ${TBB_ROOT_DIR}/lib ENV LIBRARY_PATH)
 
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(TBB DEFAULT_MSG TBB_LIBRARIES TBB_INCLUDE_DIR)

coverage/parse_gcov_output.py

@@ -1,4 +1,3 @@
-import optparse
 import re
 import sys
 

db/builder.cc

@@ -28,6 +28,7 @@
 #include "rocksdb/options.h"
 #include "rocksdb/table.h"
 #include "table/block_based_table_builder.h"
+#include "table/format.h"
 #include "table/internal_iterator.h"
 #include "util/file_reader_writer.h"
 #include "util/filename.h"
@@ -39,7 +40,7 @@ namespace rocksdb {
 class TableFactory;
 
 TableBuilder* NewTableBuilder(
-    const ImmutableCFOptions& ioptions,
+    const ImmutableCFOptions& ioptions, const MutableCFOptions& moptions,
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
@@ -52,19 +53,20 @@ TableBuilder* NewTableBuilder(
           TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
          column_family_name.empty());
   return ioptions.table_factory->NewTableBuilder(
-      TableBuilderOptions(
-          ioptions, internal_comparator, int_tbl_prop_collector_factories,
-          compression_type, compression_opts, compression_dict, skip_filters,
-          column_family_name, level, creation_time, oldest_key_time),
+      TableBuilderOptions(ioptions, moptions, internal_comparator,
+                          int_tbl_prop_collector_factories, compression_type,
+                          compression_opts, compression_dict, skip_filters,
+                          column_family_name, level, creation_time,
+                          oldest_key_time),
       column_family_id, file);
 }
 
 Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
-    const MutableCFOptions& /*mutable_cf_options*/,
-    const EnvOptions& env_options, TableCache* table_cache,
-    InternalIterator* iter, std::unique_ptr<InternalIterator> range_del_iter,
-    FileMetaData* meta, const InternalKeyComparator& internal_comparator,
+    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    TableCache* table_cache, InternalIterator* iter,
+    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
+    const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
     uint32_t column_family_id, const std::string& column_family_name,
@@ -92,7 +94,7 @@ Status BuildTable(
     return s;
   }
 
-  std::string fname = TableFileName(ioptions.db_paths, meta->fd.GetNumber(),
+  std::string fname = TableFileName(ioptions.cf_paths, meta->fd.GetNumber(),
                                     meta->fd.GetPathId());
 #ifndef ROCKSDB_LITE
   EventHelpers::NotifyTableFileCreationStarted(
@@ -100,7 +102,7 @@ Status BuildTable(
 #endif  // !ROCKSDB_LITE
   TableProperties tp;
 
-  if (iter->Valid() || range_del_agg->ShouldAddTombstones()) {
+  if (iter->Valid() || !range_del_agg->IsEmpty()) {
     TableBuilder* builder;
     unique_ptr<WritableFileWriter> file_writer;
     {
@@ -122,10 +124,11 @@ Status BuildTable(
       file_writer.reset(new WritableFileWriter(std::move(file), env_options,
                                                ioptions.statistics));
       builder = NewTableBuilder(
-          ioptions, internal_comparator, int_tbl_prop_collector_factories,
-          column_family_id, column_family_name, file_writer.get(), compression,
-          compression_opts, level, nullptr /* compression_dict */,
-          false /* skip_filters */, creation_time, oldest_key_time);
+          ioptions, mutable_cf_options, internal_comparator,
+          int_tbl_prop_collector_factories, column_family_id,
+          column_family_name, file_writer.get(), compression, compression_opts,
+          level, nullptr /* compression_dict */, false /* skip_filters */,
+          creation_time, oldest_key_time);
     }
 
     MergeHelper merge(env, internal_comparator.user_comparator(),
@@ -137,6 +140,7 @@ Status BuildTable(
     CompactionIterator c_iter(
         iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
         &snapshots, earliest_write_conflict_snapshot, snapshot_checker, env,
+        ShouldReportDetailedTime(env, ioptions.statistics),
         true /* internal key corruption is not ok */, range_del_agg.get());
     c_iter.SeekToFirst();
     for (; c_iter.Valid(); c_iter.Next()) {
@@ -152,12 +156,18 @@ Status BuildTable(
             ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
       }
     }
-    // nullptr for table_{min,max} so all range tombstones will be flushed
-    range_del_agg->AddToBuilder(builder, nullptr /* lower_bound */,
-                                nullptr /* upper_bound */, meta);
+
+    for (auto it = range_del_agg->NewIterator(); it->Valid(); it->Next()) {
+      auto tombstone = it->Tombstone();
+      auto kv = tombstone.Serialize();
+      builder->Add(kv.first.Encode(), kv.second);
+      meta->UpdateBoundariesForRange(kv.first, tombstone.SerializeEndKey(),
+                                     tombstone.seq_, internal_comparator);
+    }
 
     // Finish and check for builder errors
-    bool empty = builder->NumEntries() == 0;
+    tp = builder->GetTableProperties();
+    bool empty = builder->NumEntries() == 0 && tp.num_range_deletions == 0;
     s = c_iter.status();
     if (!s.ok() || empty) {
       builder->Abandon();
@@ -170,7 +180,7 @@ Status BuildTable(
       meta->fd.file_size = file_size;
       meta->marked_for_compaction = builder->NeedCompact();
       assert(meta->fd.GetFileSize() > 0);
-      tp = builder->GetTableProperties();
+      tp = builder->GetTableProperties(); // refresh now that builder is finished
       if (table_properties) {
         *table_properties = tp;
       }
@@ -194,8 +204,9 @@ 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->fd,
-          nullptr /* range_del_agg */, nullptr,
+          ReadOptions(), env_options, internal_comparator, *meta,
+          nullptr /* range_del_agg */,
+          mutable_cf_options.prefix_extractor.get(), nullptr,
           (internal_stats == nullptr) ? nullptr
                                       : internal_stats->GetFileReadHist(0),
           false /* for_compaction */, nullptr /* arena */,
@@ -219,9 +230,11 @@ Status BuildTable(
   }
 
   // Output to event logger and fire events.
-  EventHelpers::LogAndNotifyTableFileCreationFinished(
-      event_logger, ioptions.listeners, dbname, column_family_name, fname,
-      job_id, meta->fd, tp, reason, s);
+  if (!s.ok() || meta->fd.GetFileSize() > 0) {
+    EventHelpers::LogAndNotifyTableFileCreationFinished(
+        event_logger, ioptions.listeners, dbname, column_family_name, fname,
+        job_id, meta->fd, tp, reason, s);
+  }
 
   return s;
 }

db/builder.h

@@ -43,7 +43,7 @@ class InternalIterator;
 // @param compression_dict Data for presetting the compression library's
 //    dictionary, or nullptr.
 TableBuilder* NewTableBuilder(
-    const ImmutableCFOptions& options,
+    const ImmutableCFOptions& options, const MutableCFOptions& moptions,
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,

db/c.cc

@@ -38,6 +38,7 @@
 #include "rocksdb/utilities/transaction_db.h"
 #include "rocksdb/utilities/write_batch_with_index.h"
 #include "rocksdb/write_batch.h"
+#include "rocksdb/perf_context.h"
 #include "utilities/merge_operators.h"
 
 using rocksdb::BytewiseComparator;
@@ -89,8 +90,10 @@ using rocksdb::LiveFileMetaData;
 using rocksdb::BackupEngine;
 using rocksdb::BackupableDBOptions;
 using rocksdb::BackupInfo;
+using rocksdb::BackupID;
 using rocksdb::RestoreOptions;
 using rocksdb::CompactRangeOptions;
+using rocksdb::BottommostLevelCompaction;
 using rocksdb::RateLimiter;
 using rocksdb::NewGenericRateLimiter;
 using rocksdb::PinnableSlice;
@@ -101,6 +104,10 @@ using rocksdb::OptimisticTransactionDB;
 using rocksdb::OptimisticTransactionOptions;
 using rocksdb::Transaction;
 using rocksdb::Checkpoint;
+using rocksdb::TransactionLogIterator;
+using rocksdb::BatchResult;
+using rocksdb::PerfLevel;
+using rocksdb::PerfContext;
 
 using std::shared_ptr;
 
@@ -132,6 +139,8 @@ struct rocksdb_cuckoo_table_options_t  { CuckooTableOptions rep; };
 struct rocksdb_seqfile_t         { SequentialFile*   rep; };
 struct rocksdb_randomfile_t      { RandomAccessFile* rep; };
 struct rocksdb_writablefile_t    { WritableFile*     rep; };
+struct rocksdb_wal_iterator_t { TransactionLogIterator* rep; };
+struct rocksdb_wal_readoptions_t { TransactionLogIterator::ReadOptions rep; };
 struct rocksdb_filelock_t        { FileLock*         rep; };
 struct rocksdb_logger_t          { shared_ptr<Logger>  rep; };
 struct rocksdb_cache_t           { shared_ptr<Cache>   rep; };
@@ -140,7 +149,8 @@ struct rocksdb_column_family_handle_t  { ColumnFamilyHandle* rep; };
 struct rocksdb_envoptions_t      { EnvOptions        rep; };
 struct rocksdb_ingestexternalfileoptions_t  { IngestExternalFileOptions rep; };
 struct rocksdb_sstfilewriter_t   { SstFileWriter*    rep; };
-struct rocksdb_ratelimiter_t     { RateLimiter*      rep; };
+struct rocksdb_ratelimiter_t     { shared_ptr<RateLimiter>      rep; };
+struct rocksdb_perfcontext_t     { PerfContext*      rep; };
 struct rocksdb_pinnableslice_t {
   PinnableSlice rep;
 };
@@ -523,10 +533,18 @@ rocksdb_backup_engine_t* rocksdb_backup_engine_open(
 }
 
 void rocksdb_backup_engine_create_new_backup(rocksdb_backup_engine_t* be,
-                                             rocksdb_t* db, char** errptr) {
+                                             rocksdb_t* db,
+                                             char** errptr) {
   SaveError(errptr, be->rep->CreateNewBackup(db->rep));
 }
 
+void rocksdb_backup_engine_create_new_backup_flush(rocksdb_backup_engine_t* be,
+                                                   rocksdb_t* db,
+                                                   unsigned char flush_before_backup,
+                                                   char** errptr) {
+  SaveError(errptr, be->rep->CreateNewBackup(db->rep, flush_before_backup));
+}
+
 void rocksdb_backup_engine_purge_old_backups(rocksdb_backup_engine_t* be,
                                              uint32_t num_backups_to_keep,
                                              char** errptr) {
@@ -546,6 +564,12 @@ void rocksdb_restore_options_set_keep_log_files(rocksdb_restore_options_t* opt,
   opt->rep.keep_log_files = v;
 }
 
+
+void rocksdb_backup_engine_verify_backup(rocksdb_backup_engine_t* be,
+    uint32_t backup_id, char** errptr) {
+  SaveError(errptr, be->rep->VerifyBackup(static_cast<BackupID>(backup_id)));
+}
+
 void rocksdb_backup_engine_restore_db_from_latest_backup(
     rocksdb_backup_engine_t* be, const char* db_dir, const char* wal_dir,
     const rocksdb_restore_options_t* restore_options, char** errptr) {
@@ -926,6 +950,54 @@ rocksdb_iterator_t* rocksdb_create_iterator(
   return result;
 }
 
+rocksdb_wal_iterator_t* rocksdb_get_updates_since(
+        rocksdb_t* db, uint64_t seq_number,
+        const rocksdb_wal_readoptions_t* options,
+        char** errptr) {
+  std::unique_ptr<TransactionLogIterator> iter;
+  TransactionLogIterator::ReadOptions ro;
+  if (options!=nullptr) {
+      ro = options->rep;
+  }
+  if (SaveError(errptr, db->rep->GetUpdatesSince(seq_number, &iter, ro))) {
+    return nullptr;
+  }
+  rocksdb_wal_iterator_t* result = new rocksdb_wal_iterator_t;
+  result->rep = iter.release();
+  return result;
+}
+
+void rocksdb_wal_iter_next(rocksdb_wal_iterator_t* iter) {
+    iter->rep->Next();
+}
+
+unsigned char rocksdb_wal_iter_valid(const rocksdb_wal_iterator_t* iter) {
+    return iter->rep->Valid();
+}
+
+void rocksdb_wal_iter_status (const rocksdb_wal_iterator_t* iter, char** errptr) {
+    SaveError(errptr, iter->rep->status());
+}
+
+void rocksdb_wal_iter_destroy (const rocksdb_wal_iterator_t* iter) {
+  delete iter->rep;
+  delete iter;
+}
+
+rocksdb_writebatch_t* rocksdb_wal_iter_get_batch (const rocksdb_wal_iterator_t* iter, uint64_t* seq) {
+  rocksdb_writebatch_t* result = rocksdb_writebatch_create();
+  BatchResult wal_batch = iter->rep->GetBatch();
+  result->rep = * wal_batch.writeBatchPtr.release();
+  if (seq != nullptr) {
+    *seq = wal_batch.sequence;
+  }
+  return result;
+}
+
+uint64_t rocksdb_get_latest_sequence_number (rocksdb_t *db) {
+    return db->rep->GetLatestSequenceNumber();
+}
+
 rocksdb_iterator_t* rocksdb_create_iterator_cf(
     rocksdb_t* db,
     const rocksdb_readoptions_t* options,
@@ -1403,23 +1475,24 @@ void rocksdb_writebatch_put_log_data(
   b->rep.PutLogData(Slice(blob, len));
 }
 
+class H : public WriteBatch::Handler {
+ public:
+  void* state_;
+  void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
+  void (*deleted_)(void*, const char* k, size_t klen);
+  virtual void Put(const Slice& key, const Slice& value) override {
+    (*put_)(state_, key.data(), key.size(), value.data(), value.size());
+  }
+  virtual void Delete(const Slice& key) override {
+    (*deleted_)(state_, key.data(), key.size());
+  }
+};
+
 void rocksdb_writebatch_iterate(
     rocksdb_writebatch_t* b,
     void* state,
     void (*put)(void*, const char* k, size_t klen, const char* v, size_t vlen),
     void (*deleted)(void*, const char* k, size_t klen)) {
-  class H : public WriteBatch::Handler {
-   public:
-    void* state_;
-    void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
-    void (*deleted_)(void*, const char* k, size_t klen);
-    virtual void Put(const Slice& key, const Slice& value) override {
-      (*put_)(state_, key.data(), key.size(), value.data(), value.size());
-    }
-    virtual void Delete(const Slice& key) override {
-      (*deleted_)(state_, key.data(), key.size());
-    }
-  };
   H handler;
   handler.state_ = state;
   handler.put_ = put;
@@ -1664,18 +1737,6 @@ void rocksdb_writebatch_wi_iterate(
     void* state,
     void (*put)(void*, const char* k, size_t klen, const char* v, size_t vlen),
     void (*deleted)(void*, const char* k, size_t klen)) {
-  class H : public WriteBatch::Handler {
-   public:
-    void* state_;
-    void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
-    void (*deleted_)(void*, const char* k, size_t klen);
-    virtual void Put(const Slice& key, const Slice& value) override {
-      (*put_)(state_, key.data(), key.size(), value.data(), value.size());
-    }
-    virtual void Delete(const Slice& key) override {
-      (*deleted_)(state_, key.data(), key.size());
-    }
-  };
   H handler;
   handler.state_ = state;
   handler.put_ = put;
@@ -1924,6 +1985,11 @@ void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
   options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
 }
 
+void rocksdb_block_based_options_set_pin_top_level_index_and_filter(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.pin_top_level_index_and_filter = v;
+}
+
 void rocksdb_options_set_block_based_table_factory(
     rocksdb_options_t *opt,
     rocksdb_block_based_table_options_t* table_options) {
@@ -2202,6 +2268,18 @@ void rocksdb_options_set_compression_per_level(rocksdb_options_t* opt,
   }
 }
 
+void rocksdb_options_set_bottommost_compression_options(rocksdb_options_t* opt,
+                                                        int w_bits, int level,
+                                                        int strategy,
+                                                        int max_dict_bytes,
+                                                        bool enabled) {
+  opt->rep.bottommost_compression_opts.window_bits = w_bits;
+  opt->rep.bottommost_compression_opts.level = level;
+  opt->rep.bottommost_compression_opts.strategy = strategy;
+  opt->rep.bottommost_compression_opts.max_dict_bytes = max_dict_bytes;
+  opt->rep.bottommost_compression_opts.enabled = enabled;
+}
+
 void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
                                              int level, int strategy,
                                              int max_dict_bytes) {
@@ -2531,8 +2609,9 @@ char *rocksdb_options_statistics_get_string(rocksdb_options_t *opt) {
 }
 
 void rocksdb_options_set_ratelimiter(rocksdb_options_t *opt, rocksdb_ratelimiter_t *limiter) {
-  opt->rep.rate_limiter.reset(limiter->rep);
-  limiter->rep = nullptr;
+  if (limiter) {
+    opt->rep.rate_limiter = limiter->rep;
+  }
 }
 
 rocksdb_ratelimiter_t* rocksdb_ratelimiter_create(
@@ -2540,18 +2619,186 @@ rocksdb_ratelimiter_t* rocksdb_ratelimiter_create(
     int64_t refill_period_us,
     int32_t fairness) {
   rocksdb_ratelimiter_t* rate_limiter = new rocksdb_ratelimiter_t;
-  rate_limiter->rep = NewGenericRateLimiter(rate_bytes_per_sec,
-                                            refill_period_us, fairness);
+  rate_limiter->rep.reset(
+               NewGenericRateLimiter(rate_bytes_per_sec,
+                                     refill_period_us, fairness));
   return rate_limiter;
 }
 
 void rocksdb_ratelimiter_destroy(rocksdb_ratelimiter_t *limiter) {
-  if (limiter->rep) {
-    delete limiter->rep;
-  }
   delete limiter;
 }
 
+void rocksdb_set_perf_level(int v) {
+  PerfLevel level = static_cast<PerfLevel>(v);
+  SetPerfLevel(level);
+}
+
+rocksdb_perfcontext_t* rocksdb_perfcontext_create() {
+  rocksdb_perfcontext_t* context = new rocksdb_perfcontext_t;
+  context->rep = rocksdb::get_perf_context();
+  return context;
+}
+
+void rocksdb_perfcontext_reset(rocksdb_perfcontext_t* context) {
+  context->rep->Reset();
+}
+
+char* rocksdb_perfcontext_report(rocksdb_perfcontext_t* context,
+    unsigned char exclude_zero_counters) {
+  return strdup(context->rep->ToString(exclude_zero_counters).c_str());
+}
+
+uint64_t rocksdb_perfcontext_metric(rocksdb_perfcontext_t* context,
+    int metric) {
+  PerfContext* rep = context->rep;
+  switch (metric) {
+    case rocksdb_user_key_comparison_count:
+      return rep->user_key_comparison_count;
+    case rocksdb_block_cache_hit_count:
+      return rep->block_cache_hit_count;
+    case rocksdb_block_read_count:
+      return rep->block_read_count;
+    case rocksdb_block_read_byte:
+      return rep->block_read_byte;
+    case rocksdb_block_read_time:
+      return rep->block_read_time;
+    case rocksdb_block_checksum_time:
+      return rep->block_checksum_time;
+    case rocksdb_block_decompress_time:
+      return rep->block_decompress_time;
+    case rocksdb_get_read_bytes:
+      return rep->get_read_bytes;
+    case rocksdb_multiget_read_bytes:
+      return rep->multiget_read_bytes;
+    case rocksdb_iter_read_bytes:
+      return rep->iter_read_bytes;
+    case rocksdb_internal_key_skipped_count:
+      return rep->internal_key_skipped_count;
+    case rocksdb_internal_delete_skipped_count:
+      return rep->internal_delete_skipped_count;
+    case rocksdb_internal_recent_skipped_count:
+      return rep->internal_recent_skipped_count;
+    case rocksdb_internal_merge_count:
+      return rep->internal_merge_count;
+    case rocksdb_get_snapshot_time:
+      return rep->get_snapshot_time;
+    case rocksdb_get_from_memtable_time:
+      return rep->get_from_memtable_time;
+    case rocksdb_get_from_memtable_count:
+      return rep->get_from_memtable_count;
+    case rocksdb_get_post_process_time:
+      return rep->get_post_process_time;
+    case rocksdb_get_from_output_files_time:
+      return rep->get_from_output_files_time;
+    case rocksdb_seek_on_memtable_time:
+      return rep->seek_on_memtable_time;
+    case rocksdb_seek_on_memtable_count:
+      return rep->seek_on_memtable_count;
+    case rocksdb_next_on_memtable_count:
+      return rep->next_on_memtable_count;
+    case rocksdb_prev_on_memtable_count:
+      return rep->prev_on_memtable_count;
+    case rocksdb_seek_child_seek_time:
+      return rep->seek_child_seek_time;
+    case rocksdb_seek_child_seek_count:
+      return rep->seek_child_seek_count;
+    case rocksdb_seek_min_heap_time:
+      return rep->seek_min_heap_time;
+    case rocksdb_seek_max_heap_time:
+      return rep->seek_max_heap_time;
+    case rocksdb_seek_internal_seek_time:
+      return rep->seek_internal_seek_time;
+    case rocksdb_find_next_user_entry_time:
+      return rep->find_next_user_entry_time;
+    case rocksdb_write_wal_time:
+      return rep->write_wal_time;
+    case rocksdb_write_memtable_time:
+      return rep->write_memtable_time;
+    case rocksdb_write_delay_time:
+      return rep->write_delay_time;
+    case rocksdb_write_pre_and_post_process_time:
+      return rep->write_pre_and_post_process_time;
+    case rocksdb_db_mutex_lock_nanos:
+      return rep->db_mutex_lock_nanos;
+    case rocksdb_db_condition_wait_nanos:
+      return rep->db_condition_wait_nanos;
+    case rocksdb_merge_operator_time_nanos:
+      return rep->merge_operator_time_nanos;
+    case rocksdb_read_index_block_nanos:
+      return rep->read_index_block_nanos;
+    case rocksdb_read_filter_block_nanos:
+      return rep->read_filter_block_nanos;
+    case rocksdb_new_table_block_iter_nanos:
+      return rep->new_table_block_iter_nanos;
+    case rocksdb_new_table_iterator_nanos:
+      return rep->new_table_iterator_nanos;
+    case rocksdb_block_seek_nanos:
+      return rep->block_seek_nanos;
+    case rocksdb_find_table_nanos:
+      return rep->find_table_nanos;
+    case rocksdb_bloom_memtable_hit_count:
+      return rep->bloom_memtable_hit_count;
+    case rocksdb_bloom_memtable_miss_count:
+      return rep->bloom_memtable_miss_count;
+    case rocksdb_bloom_sst_hit_count:
+      return rep->bloom_sst_hit_count;
+    case rocksdb_bloom_sst_miss_count:
+      return rep->bloom_sst_miss_count;
+    case rocksdb_key_lock_wait_time:
+      return rep->key_lock_wait_time;
+    case rocksdb_key_lock_wait_count:
+      return rep->key_lock_wait_count;
+    case rocksdb_env_new_sequential_file_nanos:
+      return rep->env_new_sequential_file_nanos;
+    case rocksdb_env_new_random_access_file_nanos:
+      return rep->env_new_random_access_file_nanos;
+    case rocksdb_env_new_writable_file_nanos:
+      return rep->env_new_writable_file_nanos;
+    case rocksdb_env_reuse_writable_file_nanos:
+      return rep->env_reuse_writable_file_nanos;
+    case rocksdb_env_new_random_rw_file_nanos:
+      return rep->env_new_random_rw_file_nanos;
+    case rocksdb_env_new_directory_nanos:
+      return rep->env_new_directory_nanos;
+    case rocksdb_env_file_exists_nanos:
+      return rep->env_file_exists_nanos;
+    case rocksdb_env_get_children_nanos:
+      return rep->env_get_children_nanos;
+    case rocksdb_env_get_children_file_attributes_nanos:
+      return rep->env_get_children_file_attributes_nanos;
+    case rocksdb_env_delete_file_nanos:
+      return rep->env_delete_file_nanos;
+    case rocksdb_env_create_dir_nanos:
+      return rep->env_create_dir_nanos;
+    case rocksdb_env_create_dir_if_missing_nanos:
+      return rep->env_create_dir_if_missing_nanos;
+    case rocksdb_env_delete_dir_nanos:
+      return rep->env_delete_dir_nanos;
+    case rocksdb_env_get_file_size_nanos:
+      return rep->env_get_file_size_nanos;
+    case rocksdb_env_get_file_modification_time_nanos:
+      return rep->env_get_file_modification_time_nanos;
+    case rocksdb_env_rename_file_nanos:
+      return rep->env_rename_file_nanos;
+    case rocksdb_env_link_file_nanos:
+      return rep->env_link_file_nanos;
+    case rocksdb_env_lock_file_nanos:
+      return rep->env_lock_file_nanos;
+    case rocksdb_env_unlock_file_nanos:
+      return rep->env_unlock_file_nanos;
+    case rocksdb_env_new_logger_nanos:
+      return rep->env_new_logger_nanos;
+    default:
+      break;
+  }
+  return 0;
+}
+
+void rocksdb_perfcontext_destroy(rocksdb_perfcontext_t* context) {
+  delete context;
+}
+
 /*
 TODO:
 DB::OpenForReadOnly
@@ -2880,6 +3127,11 @@ void rocksdb_compactoptions_destroy(rocksdb_compactoptions_t* opt) {
   delete opt;
 }
 
+void rocksdb_compactoptions_set_bottommost_level_compaction(
+    rocksdb_compactoptions_t* opt, unsigned char v) {
+  opt->rep.bottommost_level_compaction = static_cast<BottommostLevelCompaction>(v);
+}
+
 void rocksdb_compactoptions_set_exclusive_manual_compaction(
     rocksdb_compactoptions_t* opt, unsigned char v) {
   opt->rep.exclusive_manual_compaction = v;
@@ -3029,6 +3281,11 @@ void rocksdb_sstfilewriter_finish(rocksdb_sstfilewriter_t* writer,
   SaveError(errptr, writer->rep->Finish(nullptr));
 }
 
+void rocksdb_sstfilewriter_file_size(rocksdb_sstfilewriter_t* writer,
+                                  uint64_t* file_size) {
+  *file_size = writer->rep->FileSize();
+}
+
 void rocksdb_sstfilewriter_destroy(rocksdb_sstfilewriter_t* writer) {
   delete writer->rep;
   delete writer;
@@ -3124,20 +3381,21 @@ void rocksdb_slicetransform_destroy(rocksdb_slicetransform_t* st) {
   delete st;
 }
 
+struct Wrapper : public rocksdb_slicetransform_t {
+  const SliceTransform* rep_;
+  ~Wrapper() { delete rep_; }
+  const char* Name() const override { return rep_->Name(); }
+  Slice Transform(const Slice& src) const override {
+    return rep_->Transform(src);
+  }
+  bool InDomain(const Slice& src) const override {
+    return rep_->InDomain(src);
+  }
+  bool InRange(const Slice& src) const override { return rep_->InRange(src); }
+  static void DoNothing(void*) { }
+};
+
 rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t prefixLen) {
-  struct Wrapper : public rocksdb_slicetransform_t {
-    const SliceTransform* rep_;
-    ~Wrapper() { delete rep_; }
-    const char* Name() const override { return rep_->Name(); }
-    Slice Transform(const Slice& src) const override {
-      return rep_->Transform(src);
-    }
-    bool InDomain(const Slice& src) const override {
-      return rep_->InDomain(src);
-    }
-    bool InRange(const Slice& src) const override { return rep_->InRange(src); }
-    static void DoNothing(void*) { }
-  };
   Wrapper* wrapper = new Wrapper;
   wrapper->rep_ = rocksdb::NewFixedPrefixTransform(prefixLen);
   wrapper->state_ = nullptr;
@@ -3146,19 +3404,6 @@ rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t pref
 }
 
 rocksdb_slicetransform_t* rocksdb_slicetransform_create_noop() {
-  struct Wrapper : public rocksdb_slicetransform_t {
-    const SliceTransform* rep_;
-    ~Wrapper() { delete rep_; }
-    const char* Name() const override { return rep_->Name(); }
-    Slice Transform(const Slice& src) const override {
-      return rep_->Transform(src);
-    }
-    bool InDomain(const Slice& src) const override {
-      return rep_->InDomain(src);
-    }
-    bool InRange(const Slice& src) const override { return rep_->InRange(src); }
-    static void DoNothing(void*) { }
-  };
   Wrapper* wrapper = new Wrapper;
   wrapper->rep_ = rocksdb::NewNoopTransform();
   wrapper->state_ = nullptr;

db/c_test.c

@@ -192,10 +192,11 @@ static void CheckDel(void* ptr, const char* k, size_t klen) {
   (*state)++;
 }
 
-static void CmpDestroy(void* arg) { }
+static void CmpDestroy(void* arg) { (void)arg; }
 
 static int CmpCompare(void* arg, const char* a, size_t alen,
                       const char* b, size_t blen) {
+  (void)arg;
   size_t n = (alen < blen) ? alen : blen;
   int r = memcmp(a, b, n);
   if (r == 0) {
@@ -206,13 +207,15 @@ static int CmpCompare(void* arg, const char* a, size_t alen,
 }
 
 static const char* CmpName(void* arg) {
+  (void)arg;
   return "foo";
 }
 
 // Custom filter policy
 static unsigned char fake_filter_result = 1;
-static void FilterDestroy(void* arg) { }
+static void FilterDestroy(void* arg) { (void)arg; }
 static const char* FilterName(void* arg) {
+  (void)arg;
   return "TestFilter";
 }
 static char* FilterCreate(
@@ -220,6 +223,10 @@ static char* FilterCreate(
     const char* const* key_array, const size_t* key_length_array,
     int num_keys,
     size_t* filter_length) {
+  (void)arg;
+  (void)key_array;
+  (void)key_length_array;
+  (void)num_keys;
   *filter_length = 4;
   char* result = malloc(4);
   memcpy(result, "fake", 4);
@@ -229,20 +236,30 @@ static unsigned char FilterKeyMatch(
     void* arg,
     const char* key, size_t length,
     const char* filter, size_t filter_length) {
+  (void)arg;
+  (void)key;
+  (void)length;
   CheckCondition(filter_length == 4);
   CheckCondition(memcmp(filter, "fake", 4) == 0);
   return fake_filter_result;
 }
 
 // Custom compaction filter
-static void CFilterDestroy(void* arg) {}
-static const char* CFilterName(void* arg) { return "foo"; }
+static void CFilterDestroy(void* arg) { (void)arg; }
+static const char* CFilterName(void* arg) {
+  (void)arg;
+  return "foo";
+}
 static unsigned char CFilterFilter(void* arg, int level, const char* key,
                                    size_t key_length,
                                    const char* existing_value,
                                    size_t value_length, char** new_value,
                                    size_t* new_value_length,
                                    unsigned char* value_changed) {
+  (void)arg;
+  (void)level;
+  (void)existing_value;
+  (void)value_length;
   if (key_length == 3) {
     if (memcmp(key, "bar", key_length) == 0) {
       return 1;
@@ -256,10 +273,15 @@ static unsigned char CFilterFilter(void* arg, int level, const char* key,
   return 0;
 }
 
-static void CFilterFactoryDestroy(void* arg) {}
-static const char* CFilterFactoryName(void* arg) { return "foo"; }
+static void CFilterFactoryDestroy(void* arg) { (void)arg; }
+static const char* CFilterFactoryName(void* arg) {
+  (void)arg;
+  return "foo";
+}
 static rocksdb_compactionfilter_t* CFilterCreate(
     void* arg, rocksdb_compactionfiltercontext_t* context) {
+  (void)arg;
+  (void)context;
   return rocksdb_compactionfilter_create(NULL, CFilterDestroy, CFilterFilter,
                                          CFilterName);
 }
@@ -290,8 +312,9 @@ static rocksdb_t* CheckCompaction(rocksdb_t* db, rocksdb_options_t* options,
 }
 
 // Custom merge operator
-static void MergeOperatorDestroy(void* arg) { }
+static void MergeOperatorDestroy(void* arg) { (void)arg; }
 static const char* MergeOperatorName(void* arg) {
+  (void)arg;
   return "TestMergeOperator";
 }
 static char* MergeOperatorFullMerge(
@@ -301,6 +324,14 @@ static char* MergeOperatorFullMerge(
     const char* const* operands_list, const size_t* operands_list_length,
     int num_operands,
     unsigned char* success, size_t* new_value_length) {
+  (void)arg;
+  (void)key;
+  (void)key_length;
+  (void)existing_value;
+  (void)existing_value_length;
+  (void)operands_list;
+  (void)operands_list_length;
+  (void)num_operands;
   *new_value_length = 4;
   *success = 1;
   char* result = malloc(4);
@@ -313,6 +344,12 @@ static char* MergeOperatorPartialMerge(
     const char* const* operands_list, const size_t* operands_list_length,
     int num_operands,
     unsigned char* success, size_t* new_value_length) {
+  (void)arg;
+  (void)key;
+  (void)key_length;
+  (void)operands_list;
+  (void)operands_list_length;
+  (void)num_operands;
   *new_value_length = 4;
   *success = 1;
   char* result = malloc(4);
@@ -377,6 +414,8 @@ static void CheckTxnDBGetCF(rocksdb_transactiondb_t* txn_db,
 }
 
 int main(int argc, char** argv) {
+  (void)argc;
+  (void)argv;
   rocksdb_t* db;
   rocksdb_comparator_t* cmp;
   rocksdb_cache_t* cache;

db/column_family.cc

@@ -34,6 +34,7 @@
 #include "table/merging_iterator.h"
 #include "util/autovector.h"
 #include "util/compression.h"
+#include "util/sst_file_manager_impl.h"
 
 namespace rocksdb {
 
@@ -54,6 +55,9 @@ ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
 #endif  // ROCKSDB_LITE
     // Job id == 0 means that this is not our background process, but rather
     // user thread
+    // Need to hold some shared pointers owned by the initial_cf_options
+    // before final cleaning up finishes.
+    ColumnFamilyOptions initial_cf_options_copy = cfd_->initial_cf_options();
     JobContext job_context(0);
     mutex_->Lock();
     if (cfd_->Unref()) {
@@ -81,6 +85,7 @@ Status ColumnFamilyHandleImpl::GetDescriptor(ColumnFamilyDescriptor* desc) {
   *desc = ColumnFamilyDescriptor(cfd()->GetName(), cfd()->GetLatestCFOptions());
   return Status::OK();
 #else
+  (void)desc;
   return Status::NotSupported();
 #endif  // !ROCKSDB_LITE
 }
@@ -156,6 +161,28 @@ Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
   return Status::OK();
 }
 
+Status CheckCFPathsSupported(const DBOptions& db_options,
+                             const ColumnFamilyOptions& cf_options) {
+  // More than one cf_paths are supported only in universal
+  // and level compaction styles. This function also checks the case
+  // in which cf_paths is not specified, which results in db_paths
+  // being used.
+  if ((cf_options.compaction_style != kCompactionStyleUniversal) &&
+      (cf_options.compaction_style != kCompactionStyleLevel)) {
+    if (cf_options.cf_paths.size() > 1) {
+      return Status::NotSupported(
+          "More than one CF paths are only supported in "
+          "universal and level compaction styles. ");
+    } else if (cf_options.cf_paths.empty() &&
+               db_options.db_paths.size() > 1) {
+      return Status::NotSupported(
+          "More than one DB paths are only supported in "
+          "universal and level compaction styles. ");
+    }
+  }
+  return Status::OK();
+}
+
 ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                     const ColumnFamilyOptions& src) {
   ColumnFamilyOptions result = src;
@@ -274,9 +301,24 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
         result.hard_pending_compaction_bytes_limit;
   }
 
+#ifndef ROCKSDB_LITE
+  // When the DB is stopped, it's possible that there are some .trash files that
+  // were not deleted yet, when we open the DB we will find these .trash files
+  // and schedule them to be deleted (or delete immediately if SstFileManager
+  // was not used)
+  auto sfm = static_cast<SstFileManagerImpl*>(db_options.sst_file_manager.get());
+  for (size_t i = 0; i < result.cf_paths.size(); i++) {
+    DeleteScheduler::CleanupDirectory(db_options.env, sfm, result.cf_paths[i].path);
+  }
+#endif
+
+  if (result.cf_paths.empty()) {
+    result.cf_paths = db_options.db_paths;
+  }
+
   if (result.level_compaction_dynamic_level_bytes) {
     if (result.compaction_style != kCompactionStyleLevel ||
-        db_options.db_paths.size() > 1U) {
+        result.cf_paths.size() > 1U) {
       // 1. level_compaction_dynamic_level_bytes only makes sense for
       //    level-based compaction.
       // 2. we don't yet know how to make both of this feature and multiple
@@ -383,10 +425,10 @@ ColumnFamilyData::ColumnFamilyData(
       next_(nullptr),
       prev_(nullptr),
       log_number_(0),
-      flush_reason_(FlushReason::kUnknown),
+      flush_reason_(FlushReason::kOthers),
       column_family_set_(column_family_set),
-      pending_flush_(false),
-      pending_compaction_(false),
+      queued_for_flush_(false),
+      queued_for_compaction_(false),
       prev_compaction_needed_bytes_(0),
       allow_2pc_(db_options.allow_2pc),
       last_memtable_id_(0) {
@@ -462,8 +504,8 @@ ColumnFamilyData::~ColumnFamilyData() {
 
   // It would be wrong if this ColumnFamilyData is in flush_queue_ or
   // compaction_queue_ and we destroyed it
-  assert(!pending_flush_);
-  assert(!pending_compaction_);
+  assert(!queued_for_flush_);
+  assert(!queued_for_compaction_);
 
   if (super_version_ != nullptr) {
     // Release SuperVersion reference kept in ThreadLocalPtr.
@@ -516,7 +558,9 @@ uint64_t ColumnFamilyData::OldestLogToKeep() {
   auto current_log = GetLogNumber();
 
   if (allow_2pc_) {
-    auto imm_prep_log = imm()->GetMinLogContainingPrepSection();
+    autovector<MemTable*> empty_list;
+    auto imm_prep_log =
+        imm()->PrecomputeMinLogContainingPrepSection(empty_list);
     auto mem_prep_log = mem()->GetMinLogContainingPrepSection();
 
     if (imm_prep_log > 0 && imm_prep_log < current_log) {
@@ -953,11 +997,13 @@ const int ColumnFamilyData::kCompactToBaseLevel = -2;
 
 Compaction* ColumnFamilyData::CompactRange(
     const MutableCFOptions& mutable_cf_options, int input_level,
-    int output_level, uint32_t output_path_id, const InternalKey* begin,
-    const InternalKey* end, InternalKey** compaction_end, bool* conflict) {
+    int output_level, uint32_t output_path_id, uint32_t max_subcompactions,
+    const InternalKey* begin, const InternalKey* end,
+    InternalKey** compaction_end, bool* conflict) {
   auto* result = compaction_picker_->CompactRange(
       GetName(), mutable_cf_options, current_->storage_info(), input_level,
-      output_level, output_path_id, begin, end, compaction_end, conflict);
+      output_level, output_path_id, max_subcompactions, begin, end,
+      compaction_end, conflict);
   if (result != nullptr) {
     result->SetInputVersion(current_);
   }
@@ -1107,8 +1153,9 @@ void ColumnFamilyData::ResetThreadLocalSuperVersions() {
 Status ColumnFamilyData::SetOptions(
       const std::unordered_map<std::string, std::string>& options_map) {
   MutableCFOptions new_mutable_cf_options;
-  Status s = GetMutableOptionsFromStrings(mutable_cf_options_, options_map,
-                                          &new_mutable_cf_options);
+  Status s =
+      GetMutableOptionsFromStrings(mutable_cf_options_, options_map,
+                                   ioptions_.info_log, &new_mutable_cf_options);
   if (s.ok()) {
     mutable_cf_options_ = new_mutable_cf_options;
     mutable_cf_options_.RefreshDerivedOptions(ioptions_);
@@ -1135,6 +1182,31 @@ Env::WriteLifeTimeHint ColumnFamilyData::CalculateSSTWriteHint(int level) {
                             static_cast<int>(Env::WLTH_MEDIUM));
 }
 
+Status ColumnFamilyData::AddDirectories() {
+  Status s;
+  assert(data_dirs_.empty());
+  for (auto& p : ioptions_.cf_paths) {
+    std::unique_ptr<Directory> path_directory;
+    s = DBImpl::CreateAndNewDirectory(ioptions_.env, p.path, &path_directory);
+    if (!s.ok()) {
+      return s;
+    }
+    assert(path_directory != nullptr);
+    data_dirs_.emplace_back(path_directory.release());
+  }
+  assert(data_dirs_.size() == ioptions_.cf_paths.size());
+  return s;
+}
+
+Directory* ColumnFamilyData::GetDataDir(size_t path_id) const {
+  if (data_dirs_.empty()) {
+    return nullptr;
+  }
+
+  assert(path_id < data_dirs_.size());
+  return data_dirs_[path_id].get();
+}
+
 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                  const ImmutableDBOptions* db_options,
                                  const EnvOptions& env_options,

db/column_family.h

@@ -139,6 +139,9 @@ extern Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options);
 extern Status CheckConcurrentWritesSupported(
     const ColumnFamilyOptions& cf_options);
 
+extern Status CheckCFPathsSupported(const DBOptions& db_options,
+                                    const ColumnFamilyOptions& cf_options);
+
 extern ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                            const ColumnFamilyOptions& src);
 // Wrap user defined table proproties collector factories `from cf_options`
@@ -293,9 +296,9 @@ class ColumnFamilyData {
   // REQUIRES: DB mutex held
   Compaction* CompactRange(const MutableCFOptions& mutable_cf_options,
                            int input_level, int output_level,
-                           uint32_t output_path_id, const InternalKey* begin,
-                           const InternalKey* end, InternalKey** compaction_end,
-                           bool* manual_conflict);
+                           uint32_t output_path_id, uint32_t max_subcompactions,
+                           const InternalKey* begin, const InternalKey* end,
+                           InternalKey** compaction_end, bool* manual_conflict);
 
   CompactionPicker* compaction_picker() { return compaction_picker_.get(); }
   // thread-safe
@@ -342,10 +345,10 @@ class ColumnFamilyData {
   void ResetThreadLocalSuperVersions();
 
   // Protected by DB mutex
-  void set_pending_flush(bool value) { pending_flush_ = value; }
-  void set_pending_compaction(bool value) { pending_compaction_ = value; }
-  bool pending_flush() { return pending_flush_; }
-  bool pending_compaction() { return pending_compaction_; }
+  void set_queued_for_flush(bool value) { queued_for_flush_ = value; }
+  void set_queued_for_compaction(bool value) { queued_for_compaction_ = value; }
+  bool queued_for_flush() { return queued_for_flush_; }
+  bool queued_for_compaction() { return queued_for_compaction_; }
 
   enum class WriteStallCause {
     kNone,
@@ -370,8 +373,16 @@ class ColumnFamilyData {
 
   bool initialized() const { return initialized_.load(); }
 
+  const ColumnFamilyOptions& initial_cf_options() {
+    return initial_cf_options_;
+  }
+
   Env::WriteLifeTimeHint CalculateSSTWriteHint(int level);
 
+  Status AddDirectories();
+
+  Directory* GetDataDir(size_t path_id) const;
+
  private:
   friend class ColumnFamilySet;
   ColumnFamilyData(uint32_t id, const std::string& name,
@@ -442,11 +453,11 @@ class ColumnFamilyData {
   std::unique_ptr<WriteControllerToken> write_controller_token_;
 
   // If true --> this ColumnFamily is currently present in DBImpl::flush_queue_
-  bool pending_flush_;
+  bool queued_for_flush_;
 
   // If true --> this ColumnFamily is currently present in
   // DBImpl::compaction_queue_
-  bool pending_compaction_;
+  bool queued_for_compaction_;
 
   uint64_t prev_compaction_needed_bytes_;
 
@@ -455,6 +466,9 @@ class ColumnFamilyData {
 
   // Memtable id to track flush.
   std::atomic<uint64_t> last_memtable_id_;
+
+  // Directories corresponding to cf_paths.
+  std::vector<std::unique_ptr<Directory>> data_dirs_;
 };
 
 // ColumnFamilySet has interesting thread-safety requirements

db/column_family_test.cc

@@ -57,24 +57,36 @@ class EnvCounter : public EnvWrapper {
   std::atomic<int> num_new_writable_file_;
 };
 
-class ColumnFamilyTest : public testing::Test {
+class ColumnFamilyTestBase : public testing::Test {
  public:
-  ColumnFamilyTest() : rnd_(139) {
+  ColumnFamilyTestBase(uint32_t format) : rnd_(139), format_(format) {
     env_ = new EnvCounter(Env::Default());
-    dbname_ = test::TmpDir() + "/column_family_test";
+    dbname_ = test::PerThreadDBPath("column_family_test");
     db_options_.create_if_missing = true;
     db_options_.fail_if_options_file_error = true;
     db_options_.env = env_;
     DestroyDB(dbname_, Options(db_options_, column_family_options_));
   }
 
-  ~ColumnFamilyTest() {
+  virtual ~ColumnFamilyTestBase() {
+    std::vector<ColumnFamilyDescriptor> column_families;
+    for (auto h : handles_) {
+      ColumnFamilyDescriptor cfdescriptor;
+      h->GetDescriptor(&cfdescriptor);
+      column_families.push_back(cfdescriptor);
+    }
     Close();
     rocksdb::SyncPoint::GetInstance()->DisableProcessing();
-    Destroy();
+    Destroy(column_families);
     delete env_;
   }
 
+  BlockBasedTableOptions GetBlockBasedTableOptions() {
+    BlockBasedTableOptions options;
+    options.format_version = format_;
+    return options;
+  }
+
   // Return the value to associate with the specified key
   Slice Value(int k, std::string* storage) {
     if (k == 0) {
@@ -236,9 +248,11 @@ class ColumnFamilyTest : public testing::Test {
 #endif  // !ROCKSDB_LITE
   }
 
-  void Destroy() {
+  void Destroy(const std::vector<ColumnFamilyDescriptor>& column_families =
+                  std::vector<ColumnFamilyDescriptor>()) {
     Close();
-    ASSERT_OK(DestroyDB(dbname_, Options(db_options_, column_family_options_)));
+    ASSERT_OK(DestroyDB(dbname_, Options(db_options_, column_family_options_),
+                        column_families));
   }
 
   void CreateColumnFamilies(
@@ -291,6 +305,9 @@ class ColumnFamilyTest : public testing::Test {
   }
 
   void PutRandomData(int cf, int num, int key_value_size, bool save = false) {
+    if (cf >= static_cast<int>(keys_.size())) {
+      keys_.resize(cf + 1);
+    }
     for (int i = 0; i < num; ++i) {
       // 10 bytes for key, rest is value
       if (!save) {
@@ -298,7 +315,7 @@ class ColumnFamilyTest : public testing::Test {
                       RandomString(&rnd_, key_value_size - 10)));
       } else {
         std::string key = test::RandomKey(&rnd_, 11);
-        keys_.insert(key);
+        keys_[cf].insert(key);
         ASSERT_OK(Put(cf, key, RandomString(&rnd_, key_value_size - 10)));
       }
     }
@@ -490,18 +507,37 @@ class ColumnFamilyTest : public testing::Test {
     ASSERT_OK(destfile->Close());
   }
 
+  int GetSstFileCount(std::string path) {
+    std::vector<std::string> files;
+    DBTestBase::GetSstFiles(env_, path, &files);
+    return static_cast<int>(files.size());
+  }
+
   std::vector<ColumnFamilyHandle*> handles_;
   std::vector<std::string> names_;
-  std::set<std::string> keys_;
+  std::vector<std::set<std::string>> keys_;
   ColumnFamilyOptions column_family_options_;
   DBOptions db_options_;
   std::string dbname_;
   DB* db_ = nullptr;
   EnvCounter* env_;
   Random rnd_;
+  uint32_t format_;
 };
 
-TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
+class ColumnFamilyTest
+    : public ColumnFamilyTestBase,
+      virtual public ::testing::WithParamInterface<uint32_t> {
+ public:
+  ColumnFamilyTest() : ColumnFamilyTestBase(GetParam()) {}
+};
+
+INSTANTIATE_TEST_CASE_P(FormatDef, ColumnFamilyTest,
+                        testing::Values(test::kDefaultFormatVersion));
+INSTANTIATE_TEST_CASE_P(FormatLatest, ColumnFamilyTest,
+                        testing::Values(test::kLatestFormatVersion));
+
+TEST_P(ColumnFamilyTest, DontReuseColumnFamilyID) {
   for (int iter = 0; iter < 3; ++iter) {
     Open();
     CreateColumnFamilies({"one", "two", "three"});
@@ -520,7 +556,8 @@ TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
       Reopen();
     }
     CreateColumnFamilies({"three2"});
-    // ID 3 that was used for dropped column family "three" should not be reused
+    // ID 3 that was used for dropped column family "three" should not be
+    // reused
     auto cfh3 = reinterpret_cast<ColumnFamilyHandleImpl*>(handles_[3]);
     ASSERT_EQ(4U, cfh3->GetID());
     Close();
@@ -529,7 +566,7 @@ TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
 }
 
 #ifndef ROCKSDB_LITE
-TEST_F(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
+TEST_P(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
   Open();
 
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
@@ -552,10 +589,13 @@ TEST_F(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
 }
 #endif  // !ROCKSDB_LITE
 
-class FlushEmptyCFTestWithParam : public ColumnFamilyTest,
-                                  public testing::WithParamInterface<bool> {
+class FlushEmptyCFTestWithParam
+    : public ColumnFamilyTestBase,
+      virtual public testing::WithParamInterface<std::tuple<uint32_t, bool>> {
  public:
-  FlushEmptyCFTestWithParam() { allow_2pc_ = GetParam(); }
+  FlushEmptyCFTestWithParam()
+      : ColumnFamilyTestBase(std::get<0>(GetParam())),
+        allow_2pc_(std::get<1>(GetParam())) {}
 
   // Required if inheriting from testing::WithParamInterface<>
   static void SetUpTestCase() {}
@@ -680,10 +720,16 @@ TEST_P(FlushEmptyCFTestWithParam, FlushEmptyCFTest2) {
   db_options_.env = env_;
 }
 
-INSTANTIATE_TEST_CASE_P(FlushEmptyCFTestWithParam, FlushEmptyCFTestWithParam,
-                        ::testing::Bool());
+INSTANTIATE_TEST_CASE_P(
+    FormatDef, FlushEmptyCFTestWithParam,
+    testing::Values(std::make_tuple(test::kDefaultFormatVersion, true),
+                    std::make_tuple(test::kDefaultFormatVersion, false)));
+INSTANTIATE_TEST_CASE_P(
+    FormatLatest, FlushEmptyCFTestWithParam,
+    testing::Values(std::make_tuple(test::kLatestFormatVersion, true),
+                    std::make_tuple(test::kLatestFormatVersion, false)));
 
-TEST_F(ColumnFamilyTest, AddDrop) {
+TEST_P(ColumnFamilyTest, AddDrop) {
   Open();
   CreateColumnFamilies({"one", "two", "three"});
   ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
@@ -709,7 +755,7 @@ TEST_F(ColumnFamilyTest, AddDrop) {
               std::vector<std::string>({"default", "four", "three"}));
 }
 
-TEST_F(ColumnFamilyTest, BulkAddDrop) {
+TEST_P(ColumnFamilyTest, BulkAddDrop) {
   constexpr int kNumCF = 1000;
   ColumnFamilyOptions cf_options;
   WriteOptions write_options;
@@ -747,7 +793,7 @@ TEST_F(ColumnFamilyTest, BulkAddDrop) {
   ASSERT_TRUE(families == std::vector<std::string>({"default"}));
 }
 
-TEST_F(ColumnFamilyTest, DropTest) {
+TEST_P(ColumnFamilyTest, DropTest) {
   // first iteration - dont reopen DB before dropping
   // second iteration - reopen DB before dropping
   for (int iter = 0; iter < 2; ++iter) {
@@ -771,7 +817,7 @@ TEST_F(ColumnFamilyTest, DropTest) {
   }
 }
 
-TEST_F(ColumnFamilyTest, WriteBatchFailure) {
+TEST_P(ColumnFamilyTest, WriteBatchFailure) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
   WriteBatch batch;
@@ -789,7 +835,7 @@ TEST_F(ColumnFamilyTest, WriteBatchFailure) {
   Close();
 }
 
-TEST_F(ColumnFamilyTest, ReadWrite) {
+TEST_P(ColumnFamilyTest, ReadWrite) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
   ASSERT_OK(Put(0, "foo", "v1"));
@@ -813,7 +859,7 @@ TEST_F(ColumnFamilyTest, ReadWrite) {
   Close();
 }
 
-TEST_F(ColumnFamilyTest, IgnoreRecoveredLog) {
+TEST_P(ColumnFamilyTest, IgnoreRecoveredLog) {
   std::string backup_logs = dbname_ + "/backup_logs";
 
   // delete old files in backup_logs directory
@@ -889,7 +935,7 @@ TEST_F(ColumnFamilyTest, IgnoreRecoveredLog) {
 }
 
 #ifndef ROCKSDB_LITE  // TEST functions used are not supported
-TEST_F(ColumnFamilyTest, FlushTest) {
+TEST_P(ColumnFamilyTest, FlushTest) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
   ASSERT_OK(Put(0, "foo", "v1"));
@@ -937,7 +983,7 @@ TEST_F(ColumnFamilyTest, FlushTest) {
 }
 
 // Makes sure that obsolete log files get deleted
-TEST_F(ColumnFamilyTest, LogDeletionTest) {
+TEST_P(ColumnFamilyTest, LogDeletionTest) {
   db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
   column_family_options_.arena_block_size = 4 * 1024;
   column_family_options_.write_buffer_size = 128000;  // 128KB
@@ -1005,7 +1051,7 @@ TEST_F(ColumnFamilyTest, LogDeletionTest) {
 }
 #endif  // !ROCKSDB_LITE
 
-TEST_F(ColumnFamilyTest, CrashAfterFlush) {
+TEST_P(ColumnFamilyTest, CrashAfterFlush) {
   std::unique_ptr<FaultInjectionTestEnv> fault_env(
       new FaultInjectionTestEnv(env_));
   db_options_.env = fault_env.get();
@@ -1037,7 +1083,7 @@ TEST_F(ColumnFamilyTest, CrashAfterFlush) {
   db_options_.env = env_;
 }
 
-TEST_F(ColumnFamilyTest, OpenNonexistentColumnFamily) {
+TEST_P(ColumnFamilyTest, OpenNonexistentColumnFamily) {
   ASSERT_OK(TryOpen({"default"}));
   Close();
   ASSERT_TRUE(TryOpen({"default", "dne"}).IsInvalidArgument());
@@ -1045,7 +1091,7 @@ TEST_F(ColumnFamilyTest, OpenNonexistentColumnFamily) {
 
 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 // Makes sure that obsolete log files get deleted
-TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
+TEST_P(ColumnFamilyTest, DifferentWriteBufferSizes) {
   // disable flushing stale column families
   db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
   Open();
@@ -1151,7 +1197,7 @@ TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  // Cuckoo is not supported in lite
-TEST_F(ColumnFamilyTest, MemtableNotSupportSnapshot) {
+TEST_P(ColumnFamilyTest, MemtableNotSupportSnapshot) {
   db_options_.allow_concurrent_memtable_write = false;
   Open();
   auto* s1 = dbfull()->GetSnapshot();
@@ -1189,7 +1235,7 @@ static TestComparator third_comparator;
 static TestComparator fourth_comparator;
 
 // Test that we can retrieve the comparator from a created CF
-TEST_F(ColumnFamilyTest, GetComparator) {
+TEST_P(ColumnFamilyTest, GetComparator) {
   Open();
   // Add a column family with no comparator specified
   CreateColumnFamilies({"first"});
@@ -1208,7 +1254,7 @@ TEST_F(ColumnFamilyTest, GetComparator) {
   Close();
 }
 
-TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
+TEST_P(ColumnFamilyTest, DifferentMergeOperators) {
   Open();
   CreateColumnFamilies({"first", "second"});
   ColumnFamilyOptions default_cf, first, second;
@@ -1239,7 +1285,7 @@ TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
 }
 
 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
-TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
+TEST_P(ColumnFamilyTest, DifferentCompactionStyles) {
   Open();
   CreateColumnFamilies({"one", "two"});
   ColumnFamilyOptions default_cf, one, two;
@@ -1251,7 +1297,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = static_cast<uint64_t>(1) << 60;
 
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1311,7 +1357,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
 #ifndef ROCKSDB_LITE
 // Sync points not supported in RocksDB Lite
 
-TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
+TEST_P(ColumnFamilyTest, MultipleManualCompactions) {
   Open();
   CreateColumnFamilies({"one", "two"});
   ColumnFamilyOptions default_cf, one, two;
@@ -1323,7 +1369,7 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1354,7 +1400,7 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
        {"ColumnFamilyTest::MultiManual:2", "ColumnFamilyTest::MultiManual:5"},
        {"ColumnFamilyTest::MultiManual:2", "ColumnFamilyTest::MultiManual:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::MultiManual:4");
           cf_1_1 = false;
@@ -1400,15 +1446,17 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
   CompactAll(2);
   AssertFilesPerLevel("0,1", 2);
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
   Close();
 }
 
-TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
+TEST_P(ColumnFamilyTest, AutomaticAndManualCompactions) {
   Open();
   CreateColumnFamilies({"one", "two"});
   ColumnFamilyOptions default_cf, one, two;
@@ -1420,7 +1468,8 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1447,7 +1496,7 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
        {"ColumnFamilyTest::AutoManual:2", "ColumnFamilyTest::AutoManual:5"},
        {"ColumnFamilyTest::AutoManual:2", "ColumnFamilyTest::AutoManual:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           cf_1_1 = false;
           TEST_SYNC_POINT("ColumnFamilyTest::AutoManual:4");
@@ -1494,14 +1543,16 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
   CompactAll(2);
   AssertFilesPerLevel("0,1", 2);
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
-TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
+TEST_P(ColumnFamilyTest, ManualAndAutomaticCompactions) {
   Open();
   CreateColumnFamilies({"one", "two"});
   ColumnFamilyOptions default_cf, one, two;
@@ -1513,7 +1564,8 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1548,7 +1600,7 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
        {"ColumnFamilyTest::ManualAuto:5", "ColumnFamilyTest::ManualAuto:2"},
        {"ColumnFamilyTest::ManualAuto:2", "ColumnFamilyTest::ManualAuto:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::ManualAuto:4");
           cf_1_1 = false;
@@ -1590,14 +1642,16 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
   CompactAll(2);
   AssertFilesPerLevel("0,1", 2);
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
-TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
+TEST_P(ColumnFamilyTest, SameCFManualManualCompactions) {
   Open();
   CreateColumnFamilies({"one"});
   ColumnFamilyOptions default_cf, one;
@@ -1609,7 +1663,8 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1641,7 +1696,7 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
        {"ColumnFamilyTest::ManualManual:1",
         "ColumnFamilyTest::ManualManual:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::ManualManual:4");
           cf_1_1 = false;
@@ -1689,14 +1744,16 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
   ASSERT_LE(NumTableFilesAtLevel(0, 1), 2);
 
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
-TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
+TEST_P(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
   Open();
   CreateColumnFamilies({"one"});
   ColumnFamilyOptions default_cf, one;
@@ -1708,7 +1765,8 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1739,7 +1797,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
        {"ColumnFamilyTest::ManualAuto:1", "ColumnFamilyTest::ManualAuto:2"},
        {"ColumnFamilyTest::ManualAuto:1", "ColumnFamilyTest::ManualAuto:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::ManualAuto:4");
           cf_1_1 = false;
@@ -1779,14 +1837,16 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
   ASSERT_LE(NumTableFilesAtLevel(0, 1), 2);
 
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
-TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
+TEST_P(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
   Open();
   CreateColumnFamilies({"one"});
   ColumnFamilyOptions default_cf, one;
@@ -1798,7 +1858,8 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1831,7 +1892,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
         "ColumnFamilyTest::ManualAuto:3"},
        {"ColumnFamilyTest::ManualAuto:1", "ColumnFamilyTest::ManualAuto:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::ManualAuto:4");
           cf_1_1 = false;
@@ -1869,11 +1930,13 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
   AssertFilesPerLevel("0,1", 1);
 
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 
 // In this test, we generate enough files to trigger automatic compactions.
@@ -1882,7 +1945,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
 // This will wait because the automatic compaction has files it needs.
 // Once the conflict is hit, the automatic compaction starts and ends
 // Then the manual will run and end.
-TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
+TEST_P(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
   Open();
   CreateColumnFamilies({"one"});
   ColumnFamilyOptions default_cf, one;
@@ -1894,7 +1957,8 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
   default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
   table_options.no_block_cache = true;
   default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
@@ -1918,7 +1982,7 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
        {"CompactionPicker::CompactRange:Conflict",
         "ColumnFamilyTest::AutoManual:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (cf_1_1) {
           TEST_SYNC_POINT("ColumnFamilyTest::AutoManual:4");
           cf_1_1 = false;
@@ -1958,11 +2022,13 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
   // VERIFY compaction "one"
   AssertFilesPerLevel("1", 1);
   // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
+  std::set<std::string>::iterator key_iter = keys_[1].begin();
+  while (key_iter != keys_[1].end()) {
     ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
     key_iter++;
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 #endif  // !ROCKSDB_LITE
 
@@ -1979,7 +2045,7 @@ std::string IterStatus(Iterator* iter) {
 }
 }  // anonymous namespace
 
-TEST_F(ColumnFamilyTest, NewIteratorsTest) {
+TEST_P(ColumnFamilyTest, NewIteratorsTest) {
   // iter == 0 -- no tailing
   // iter == 2 -- tailing
   for (int iter = 0; iter < 2; ++iter) {
@@ -2026,7 +2092,7 @@ TEST_F(ColumnFamilyTest, NewIteratorsTest) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  // ReadOnlyDB is not supported
-TEST_F(ColumnFamilyTest, ReadOnlyDBTest) {
+TEST_P(ColumnFamilyTest, ReadOnlyDBTest) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
   ASSERT_OK(Put(0, "a", "b"));
@@ -2078,7 +2144,7 @@ TEST_F(ColumnFamilyTest, ReadOnlyDBTest) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  //  WaitForFlush() is not supported in lite
-TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
+TEST_P(ColumnFamilyTest, DontRollEmptyLogs) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
 
@@ -2102,7 +2168,7 @@ TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  //  WaitForCompaction() is not supported in lite
-TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
+TEST_P(ColumnFamilyTest, FlushStaleColumnFamilies) {
   Open();
   CreateColumnFamilies({"one", "two"});
   ColumnFamilyOptions default_cf, one, two;
@@ -2137,7 +2203,7 @@ TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
 }
 #endif  // !ROCKSDB_LITE
 
-TEST_F(ColumnFamilyTest, CreateMissingColumnFamilies) {
+TEST_P(ColumnFamilyTest, CreateMissingColumnFamilies) {
   Status s = TryOpen({"one", "two"});
   ASSERT_TRUE(!s.ok());
   db_options_.create_missing_column_families = true;
@@ -2146,7 +2212,7 @@ TEST_F(ColumnFamilyTest, CreateMissingColumnFamilies) {
   Close();
 }
 
-TEST_F(ColumnFamilyTest, SanitizeOptions) {
+TEST_P(ColumnFamilyTest, SanitizeOptions) {
   DBOptions db_options;
   for (int s = kCompactionStyleLevel; s <= kCompactionStyleUniversal; ++s) {
     for (int l = 0; l <= 2; l++) {
@@ -2195,7 +2261,7 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
   }
 }
 
-TEST_F(ColumnFamilyTest, ReadDroppedColumnFamily) {
+TEST_P(ColumnFamilyTest, ReadDroppedColumnFamily) {
   // iter 0 -- drop CF, don't reopen
   // iter 1 -- delete CF, reopen
   for (int iter = 0; iter < 2; ++iter) {
@@ -2267,7 +2333,7 @@ TEST_F(ColumnFamilyTest, ReadDroppedColumnFamily) {
   }
 }
 
-TEST_F(ColumnFamilyTest, FlushAndDropRaceCondition) {
+TEST_P(ColumnFamilyTest, FlushAndDropRaceCondition) {
   db_options_.create_missing_column_families = true;
   Open({"default", "one"});
   ColumnFamilyOptions options;
@@ -2333,12 +2399,13 @@ TEST_F(ColumnFamilyTest, FlushAndDropRaceCondition) {
 // skipped as persisting options is not supported in ROCKSDB_LITE
 namespace {
 std::atomic<int> test_stage(0);
+std::atomic<bool> ordered_by_writethread(false);
 const int kMainThreadStartPersistingOptionsFile = 1;
 const int kChildThreadFinishDroppingColumnFamily = 2;
-const int kChildThreadWaitingMainThreadPersistOptions = 3;
 void DropSingleColumnFamily(ColumnFamilyTest* cf_test, int cf_id,
                             std::vector<Comparator*>* comparators) {
-  while (test_stage < kMainThreadStartPersistingOptionsFile) {
+  while (test_stage < kMainThreadStartPersistingOptionsFile &&
+         !ordered_by_writethread) {
     Env::Default()->SleepForMicroseconds(100);
   }
   cf_test->DropColumnFamilies({cf_id});
@@ -2350,7 +2417,7 @@ void DropSingleColumnFamily(ColumnFamilyTest* cf_test, int cf_id,
 }
 }  // namespace
 
-TEST_F(ColumnFamilyTest, CreateAndDropRace) {
+TEST_P(ColumnFamilyTest, CreateAndDropRace) {
   const int kCfCount = 5;
   std::vector<ColumnFamilyOptions> cf_opts;
   std::vector<Comparator*> comparators;
@@ -2365,28 +2432,25 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
   auto main_thread_id = std::this_thread::get_id();
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack("PersistRocksDBOptions:start",
-                                                 [&](void* arg) {
+                                                 [&](void* /*arg*/) {
     auto current_thread_id = std::this_thread::get_id();
     // If it's the main thread hitting this sync-point, then it
     // will be blocked until some other thread update the test_stage.
     if (main_thread_id == current_thread_id) {
       test_stage = kMainThreadStartPersistingOptionsFile;
-      while (test_stage < kChildThreadFinishDroppingColumnFamily) {
+      while (test_stage < kChildThreadFinishDroppingColumnFamily &&
+             !ordered_by_writethread) {
         Env::Default()->SleepForMicroseconds(100);
       }
     }
   });
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "WriteThread::EnterUnbatched:Wait", [&](void* arg) {
+      "WriteThread::EnterUnbatched:Wait", [&](void* /*arg*/) {
         // This means a thread doing DropColumnFamily() is waiting for
         // other thread to finish persisting options.
         // In such case, we update the test_stage to unblock the main thread.
-        test_stage = kChildThreadWaitingMainThreadPersistOptions;
-
-        // Note that based on the test setting, this must not be the
-        // main thread.
-        ASSERT_NE(main_thread_id, std::this_thread::get_id());
+        ordered_by_writethread = true;
       });
 
   // Create a database with four column families
@@ -2397,7 +2461,8 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
 
   // Start a thread that will drop the first column family
   // and its comparator
-  rocksdb::port::Thread drop_cf_thread(DropSingleColumnFamily, this, 1, &comparators);
+  rocksdb::port::Thread drop_cf_thread(DropSingleColumnFamily, this, 1,
+                                       &comparators);
 
   DropColumnFamilies({2});
 
@@ -2409,10 +2474,13 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
       delete comparator;
     }
   }
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 #endif  // !ROCKSDB_LITE
 
-TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
+TEST_P(ColumnFamilyTest, WriteStallSingleColumnFamily) {
   const uint64_t kBaseRate = 800000u;
   db_options_.delayed_write_rate = kBaseRate;
   db_options_.max_background_compactions = 6;
@@ -2594,7 +2662,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 }
 
-TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
+TEST_P(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
   db_options_.max_background_compactions = 6;
   Open({"default"});
   ColumnFamilyData* cfd =
@@ -2658,7 +2726,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }
 
-TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
+TEST_P(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
   const uint64_t kBaseRate = 810000u;
   db_options_.delayed_write_rate = kBaseRate;
   Open();
@@ -2733,7 +2801,7 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 }
 
-TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
+TEST_P(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
   db_options_.max_background_compactions = 6;
   column_family_options_.soft_pending_compaction_bytes_limit = 200;
   column_family_options_.hard_pending_compaction_bytes_limit = 2000;
@@ -2798,7 +2866,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }
 
-TEST_F(ColumnFamilyTest, CreateAndDestoryOptions) {
+TEST_P(ColumnFamilyTest, CreateAndDestoryOptions) {
   std::unique_ptr<ColumnFamilyOptions> cfo(new ColumnFamilyOptions());
   ColumnFamilyHandle* cfh;
   Open();
@@ -2810,7 +2878,7 @@ TEST_F(ColumnFamilyTest, CreateAndDestoryOptions) {
   ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
 }
 
-TEST_F(ColumnFamilyTest, CreateDropAndDestroy) {
+TEST_P(ColumnFamilyTest, CreateDropAndDestroy) {
   ColumnFamilyHandle* cfh;
   Open();
   ASSERT_OK(db_->CreateColumnFamily(ColumnFamilyOptions(), "yoyo", &cfh));
@@ -2820,7 +2888,8 @@ TEST_F(ColumnFamilyTest, CreateDropAndDestroy) {
   ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
 }
 
-TEST_F(ColumnFamilyTest, CreateDropAndDestroyWithoutFileDeletion) {
+#ifndef ROCKSDB_LITE
+TEST_P(ColumnFamilyTest, CreateDropAndDestroyWithoutFileDeletion) {
   ColumnFamilyHandle* cfh;
   Open();
   ASSERT_OK(db_->CreateColumnFamily(ColumnFamilyOptions(), "yoyo", &cfh));
@@ -2831,8 +2900,7 @@ TEST_F(ColumnFamilyTest, CreateDropAndDestroyWithoutFileDeletion) {
   ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
 }
 
-#ifndef ROCKSDB_LITE
-TEST_F(ColumnFamilyTest, FlushCloseWALFiles) {
+TEST_P(ColumnFamilyTest, FlushCloseWALFiles) {
   SpecialEnv env(Env::Default());
   db_options_.env = &env;
   db_options_.max_background_flushes = 1;
@@ -2874,7 +2942,7 @@ TEST_F(ColumnFamilyTest, FlushCloseWALFiles) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
-TEST_F(ColumnFamilyTest, IteratorCloseWALFile1) {
+TEST_P(ColumnFamilyTest, IteratorCloseWALFile1) {
   SpecialEnv env(Env::Default());
   db_options_.env = &env;
   db_options_.max_background_flushes = 1;
@@ -2919,7 +2987,7 @@ TEST_F(ColumnFamilyTest, IteratorCloseWALFile1) {
   Close();
 }
 
-TEST_F(ColumnFamilyTest, IteratorCloseWALFile2) {
+TEST_P(ColumnFamilyTest, IteratorCloseWALFile2) {
   SpecialEnv env(Env::Default());
   // Allow both of flush and purge job to schedule.
   env.SetBackgroundThreads(2, Env::HIGH);
@@ -2976,7 +3044,7 @@ TEST_F(ColumnFamilyTest, IteratorCloseWALFile2) {
 #endif  // !ROCKSDB_LITE
 
 #ifndef ROCKSDB_LITE  // TEST functions are not supported in lite
-TEST_F(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
+TEST_P(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
   SpecialEnv env(Env::Default());
   // Allow both of flush and purge job to schedule.
   env.SetBackgroundThreads(2, Env::HIGH);
@@ -3053,7 +3121,7 @@ TEST_F(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
 // Disable on windows because SyncWAL requires env->IsSyncThreadSafe()
 // to return true which is not so in unbuffered mode.
 #ifndef OS_WIN
-TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
+TEST_P(ColumnFamilyTest, LogSyncConflictFlush) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
 
@@ -3088,7 +3156,7 @@ TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
 // test is being used to ensure a roll of wal files.
 // Basic idea is to test that WAL truncation is being detected and not
 // ignored
-TEST_F(ColumnFamilyTest, DISABLED_LogTruncationTest) {
+TEST_P(ColumnFamilyTest, DISABLED_LogTruncationTest) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
 
@@ -3157,6 +3225,77 @@ TEST_F(ColumnFamilyTest, DISABLED_LogTruncationTest) {
   // cleanup
   env_->DeleteDir(backup_logs);
 }
+
+TEST_P(ColumnFamilyTest, DefaultCfPathsTest) {
+  Open();
+  // Leave cf_paths for one column families to be empty.
+  // Files should be generated according to db_paths for that
+  // column family.
+  ColumnFamilyOptions cf_opt1, cf_opt2;
+  cf_opt1.cf_paths.emplace_back(dbname_ + "_one_1",
+                                std::numeric_limits<uint64_t>::max());
+  CreateColumnFamilies({"one", "two"}, {cf_opt1, cf_opt2});
+  Reopen({ColumnFamilyOptions(), cf_opt1, cf_opt2});
+
+  // Fill Column family 1.
+  PutRandomData(1, 100, 100);
+  Flush(1);
+
+  ASSERT_EQ(1, GetSstFileCount(cf_opt1.cf_paths[0].path));
+  ASSERT_EQ(0, GetSstFileCount(dbname_));
+
+  // Fill column family 2
+  PutRandomData(2, 100, 100);
+  Flush(2);
+
+  // SST from Column family 2 should be generated in
+  // db_paths which is dbname_ in this case.
+  ASSERT_EQ(1, GetSstFileCount(dbname_));
+}
+
+TEST_P(ColumnFamilyTest, MultipleCFPathsTest) {
+  Open();
+  // Configure Column family specific paths.
+  ColumnFamilyOptions cf_opt1, cf_opt2;
+  cf_opt1.cf_paths.emplace_back(dbname_ + "_one_1",
+                                std::numeric_limits<uint64_t>::max());
+  cf_opt2.cf_paths.emplace_back(dbname_ + "_two_1",
+                                std::numeric_limits<uint64_t>::max());
+  CreateColumnFamilies({"one", "two"}, {cf_opt1, cf_opt2});
+  Reopen({ColumnFamilyOptions(), cf_opt1, cf_opt2});
+
+  PutRandomData(1, 100, 100, true /* save */);
+  Flush(1);
+
+  // Check that files are generated in appropriate paths.
+  ASSERT_EQ(1, GetSstFileCount(cf_opt1.cf_paths[0].path));
+  ASSERT_EQ(0, GetSstFileCount(dbname_));
+
+  PutRandomData(2, 100, 100, true /* save */);
+  Flush(2);
+
+  ASSERT_EQ(1, GetSstFileCount(cf_opt2.cf_paths[0].path));
+  ASSERT_EQ(0, GetSstFileCount(dbname_));
+
+  // Re-open and verify the keys.
+  Reopen({ColumnFamilyOptions(), cf_opt1, cf_opt2});
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  for (int cf = 1; cf != 3; ++cf) {
+    ReadOptions read_options;
+    read_options.readahead_size = 0;
+    auto it = dbi->NewIterator(read_options, handles_[cf]);
+    for (it->SeekToFirst(); it->Valid(); it->Next()) {
+      Slice key(it->key());
+      ASSERT_NE(keys_[cf].end(), keys_[cf].find(key.ToString()));
+    }
+    delete it;
+
+    for (const auto& key : keys_[cf]) {
+      ASSERT_NE("NOT_FOUND", Get(cf, key));
+    }
+  }
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/compact_files_test.cc

@@ -24,7 +24,7 @@ class CompactFilesTest : public testing::Test {
  public:
   CompactFilesTest() {
     env_ = Env::Default();
-    db_name_ = test::TmpDir(env_) + "/compact_files_test";
+    db_name_ = test::PerThreadDBPath("compact_files_test");
   }
 
   std::string db_name_;

db/compacted_db_impl.cc

@@ -49,8 +49,8 @@ Status CompactedDBImpl::Get(const ReadOptions& options, ColumnFamilyHandle*,
                          GetContext::kNotFound, key, value, nullptr, nullptr,
                          nullptr, nullptr);
   LookupKey lkey(key, kMaxSequenceNumber);
-  files_.files[FindFile(key)].fd.table_reader->Get(
-      options, lkey.internal_key(), &get_context);
+  files_.files[FindFile(key)].fd.table_reader->Get(options, lkey.internal_key(),
+                                                   &get_context, nullptr);
   if (get_context.State() == GetContext::kFound) {
     return Status::OK();
   }
@@ -82,7 +82,7 @@ std::vector<Status> CompactedDBImpl::MultiGet(const ReadOptions& options,
                              GetContext::kNotFound, keys[idx], &pinnable_val,
                              nullptr, nullptr, nullptr, nullptr);
       LookupKey lkey(keys[idx], kMaxSequenceNumber);
-      r->Get(options, lkey.internal_key(), &get_context);
+      r->Get(options, lkey.internal_key(), &get_context, nullptr);
       value.assign(pinnable_val.data(), pinnable_val.size());
       if (get_context.State() == GetContext::kFound) {
         statuses[idx] = Status::OK();

db/compaction.cc

@@ -134,6 +134,8 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
                        int _output_level, uint64_t _target_file_size,
                        uint64_t _max_compaction_bytes, uint32_t _output_path_id,
                        CompressionType _compression,
+                       CompressionOptions _compression_opts,
+                       uint32_t _max_subcompactions,
                        std::vector<FileMetaData*> _grandparents,
                        bool _manual_compaction, double _score,
                        bool _deletion_compaction,
@@ -143,6 +145,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
       output_level_(_output_level),
       max_output_file_size_(_target_file_size),
       max_compaction_bytes_(_max_compaction_bytes),
+      max_subcompactions_(_max_subcompactions),
       immutable_cf_options_(_immutable_cf_options),
       mutable_cf_options_(_mutable_cf_options),
       input_version_(nullptr),
@@ -150,6 +153,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
       cfd_(nullptr),
       output_path_id_(_output_path_id),
       output_compression_(_compression),
+      output_compression_opts_(_compression_opts),
       deletion_compaction_(_deletion_compaction),
       inputs_(std::move(_inputs)),
       grandparents_(std::move(_grandparents)),
@@ -163,6 +167,9 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
   if (is_manual_compaction_) {
     compaction_reason_ = CompactionReason::kManualCompaction;
   }
+  if (max_subcompactions_ == 0) {
+    max_subcompactions_ = immutable_cf_options_.max_subcompactions;
+  }
 
 #ifndef NDEBUG
   for (size_t i = 1; i < inputs_.size(); ++i) {
@@ -442,7 +449,7 @@ bool Compaction::IsOutputLevelEmpty() const {
 }
 
 bool Compaction::ShouldFormSubcompactions() const {
-  if (immutable_cf_options_.max_subcompactions <= 1 || cfd_ == nullptr) {
+  if (max_subcompactions_ <= 1 || cfd_ == nullptr) {
     return false;
   }
   if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel) {
@@ -468,4 +475,8 @@ uint64_t Compaction::MaxInputFileCreationTime() const {
   return max_creation_time;
 }
 
+int Compaction::GetInputBaseLevel() const {
+  return input_vstorage_->base_level();
+}
+
 }  // namespace rocksdb

db/compaction.h

@@ -40,6 +40,7 @@ class Compaction {
              std::vector<CompactionInputFiles> inputs, int output_level,
              uint64_t target_file_size, uint64_t max_compaction_bytes,
              uint32_t output_path_id, CompressionType compression,
+             CompressionOptions compression_opts, uint32_t max_subcompactions,
              std::vector<FileMetaData*> grandparents,
              bool manual_compaction = false, double score = -1,
              bool deletion_compaction = false,
@@ -118,6 +119,11 @@ class Compaction {
   // What compression for output
   CompressionType output_compression() const { return output_compression_; }
 
+  // What compression options for output
+  CompressionOptions output_compression_opts() const {
+    return output_compression_opts_;
+  }
+
   // Whether need to write output file to second DB path.
   uint32_t output_path_id() const { return output_path_id_; }
 
@@ -233,6 +239,8 @@ class Compaction {
 
   Slice GetLargestUserKey() const { return largest_user_key_; }
 
+  int GetInputBaseLevel() const;
+
   CompactionReason compaction_reason() { return compaction_reason_; }
 
   const std::vector<FileMetaData*>& grandparents() const {
@@ -241,6 +249,8 @@ class Compaction {
 
   uint64_t max_compaction_bytes() const { return max_compaction_bytes_; }
 
+  uint32_t max_subcompactions() const { return max_subcompactions_; }
+
   uint64_t MaxInputFileCreationTime() const;
 
  private:
@@ -267,6 +277,7 @@ class Compaction {
   const int output_level_;  // levels to which output files are stored
   uint64_t max_output_file_size_;
   uint64_t max_compaction_bytes_;
+  uint32_t max_subcompactions_;
   const ImmutableCFOptions immutable_cf_options_;
   const MutableCFOptions mutable_cf_options_;
   Version* input_version_;
@@ -277,6 +288,7 @@ class Compaction {
 
   const uint32_t output_path_id_;
   CompressionType output_compression_;
+  CompressionOptions output_compression_opts_;
   // If true, then the comaction can be done by simply deleting input files.
   const bool deletion_compaction_;
 

db/compaction_iterator.cc

@@ -17,14 +17,15 @@ CompactionIterator::CompactionIterator(
     SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, Env* env,
-    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    bool report_detailed_time, bool expect_valid_internal_key,
+    RangeDelAggregator* range_del_agg,
     const Compaction* compaction, const CompactionFilter* compaction_filter,
     const std::atomic<bool>* shutting_down,
     const SequenceNumber preserve_deletes_seqnum)
     : CompactionIterator(
           input, cmp, merge_helper, last_sequence, snapshots,
           earliest_write_conflict_snapshot, snapshot_checker, env,
-          expect_valid_internal_key, range_del_agg,
+          report_detailed_time, expect_valid_internal_key, range_del_agg,
           std::unique_ptr<CompactionProxy>(
               compaction ? new CompactionProxy(compaction) : nullptr),
           compaction_filter, shutting_down, preserve_deletes_seqnum) {}
@@ -34,7 +35,8 @@ CompactionIterator::CompactionIterator(
     SequenceNumber /*last_sequence*/, std::vector<SequenceNumber>* snapshots,
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, Env* env,
-    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    bool report_detailed_time, bool expect_valid_internal_key,
+    RangeDelAggregator* range_del_agg,
     std::unique_ptr<CompactionProxy> compaction,
     const CompactionFilter* compaction_filter,
     const std::atomic<bool>* shutting_down,
@@ -46,6 +48,7 @@ CompactionIterator::CompactionIterator(
       earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
       snapshot_checker_(snapshot_checker),
       env_(env),
+      report_detailed_time_(report_detailed_time),
       expect_valid_internal_key_(expect_valid_internal_key),
       range_del_agg_(range_del_agg),
       compaction_(std::move(compaction)),
@@ -171,12 +174,12 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
     // to get sequence number.
     Slice& filter_key = ikey_.type == kTypeValue ? ikey_.user_key : key_;
     {
-      StopWatchNano timer(env_, true);
+      StopWatchNano timer(env_, report_detailed_time_);
       filter = compaction_filter_->FilterV2(
           compaction_->level(), filter_key, value_type, value_,
           &compaction_filter_value_, compaction_filter_skip_until_.rep());
       iter_stats_.total_filter_time +=
-          env_ != nullptr ? timer.ElapsedNanos() : 0;
+          env_ != nullptr && report_detailed_time_ ? timer.ElapsedNanos() : 0;
     }
 
     if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil &&
@@ -551,7 +554,7 @@ void CompactionIterator::NextFromInput() {
       // 1. new user key -OR-
       // 2. different snapshot stripe
       bool should_delete = range_del_agg_->ShouldDelete(
-          key_, RangeDelAggregator::RangePositioningMode::kForwardTraversal);
+          key_, RangeDelPositioningMode::kForwardTraversal);
       if (should_delete) {
         ++iter_stats_.num_record_drop_hidden;
         ++iter_stats_.num_record_drop_range_del;

db/compaction_iterator.h

@@ -63,7 +63,7 @@ class CompactionIterator {
                      std::vector<SequenceNumber>* snapshots,
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* snapshot_checker, Env* env,
-                     bool expect_valid_internal_key,
+                     bool report_detailed_time, bool expect_valid_internal_key,
                      RangeDelAggregator* range_del_agg,
                      const Compaction* compaction = nullptr,
                      const CompactionFilter* compaction_filter = nullptr,
@@ -76,7 +76,7 @@ class CompactionIterator {
                      std::vector<SequenceNumber>* snapshots,
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* snapshot_checker, Env* env,
-                     bool expect_valid_internal_key,
+                     bool report_detailed_time, bool expect_valid_internal_key,
                      RangeDelAggregator* range_del_agg,
                      std::unique_ptr<CompactionProxy> compaction,
                      const CompactionFilter* compaction_filter = nullptr,
@@ -139,6 +139,7 @@ class CompactionIterator {
   const SequenceNumber earliest_write_conflict_snapshot_;
   const SnapshotChecker* const snapshot_checker_;
   Env* env_;
+  bool report_detailed_time_;
   bool expect_valid_internal_key_;
   RangeDelAggregator* range_del_agg_;
   std::unique_ptr<CompactionProxy> compaction_;

db/compaction_iterator_test.cc

@@ -247,8 +247,9 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
     c_iter_.reset(new CompactionIterator(
         iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
         earliest_write_conflict_snapshot, snapshot_checker_.get(),
-        Env::Default(), false, range_del_agg_.get(), std::move(compaction),
-        filter, &shutting_down_));
+        Env::Default(), false /* report_detailed_time */,
+        false, range_del_agg_.get(), std::move(compaction), filter,
+        &shutting_down_));
   }
 
   void AddSnapshot(SequenceNumber snapshot,

db/compaction_job.cc

@@ -25,8 +25,10 @@
 #include <vector>
 
 #include "db/builder.h"
+#include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/dbformat.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "db/log_reader.h"
 #include "db/log_writer.h"
@@ -62,6 +64,46 @@
 
 namespace rocksdb {
 
+const char* GetCompactionReasonString(CompactionReason compaction_reason) {
+  switch (compaction_reason) {
+    case CompactionReason::kUnknown:
+      return "Unknown";
+    case CompactionReason::kLevelL0FilesNum:
+      return "LevelL0FilesNum";
+    case CompactionReason::kLevelMaxLevelSize:
+      return "LevelMaxLevelSize";
+    case CompactionReason::kUniversalSizeAmplification:
+      return "UniversalSizeAmplification";
+    case CompactionReason::kUniversalSizeRatio:
+      return "UniversalSizeRatio";
+    case CompactionReason::kUniversalSortedRunNum:
+      return "UniversalSortedRunNum";
+    case CompactionReason::kFIFOMaxSize:
+      return "FIFOMaxSize";
+    case CompactionReason::kFIFOReduceNumFiles:
+      return "FIFOReduceNumFiles";
+    case CompactionReason::kFIFOTtl:
+      return "FIFOTtl";
+    case CompactionReason::kManualCompaction:
+      return "ManualCompaction";
+    case CompactionReason::kFilesMarkedForCompaction:
+      return "FilesMarkedForCompaction";
+    case CompactionReason::kBottommostFiles:
+      return "BottommostFiles";
+    case CompactionReason::kTtl:
+      return "Ttl";
+    case CompactionReason::kFlush:
+      return "Flush";
+    case CompactionReason::kExternalSstIngestion:
+      return "ExternalSstIngestion";
+    case CompactionReason::kNumOfReasons:
+      // fall through
+    default:
+      assert(false);
+      return "Invalid";
+  }
+}
+
 // Maintains state for each sub-compaction
 struct CompactionJob::SubcompactionState {
   const Compaction* compaction;
@@ -267,7 +309,7 @@ CompactionJob::CompactionJob(
     const std::atomic<bool>* shutting_down,
     const SequenceNumber preserve_deletes_seqnum, LogBuffer* log_buffer,
     Directory* db_directory, Directory* output_directory, Statistics* stats,
-    InstrumentedMutex* db_mutex, Status* db_bg_error,
+    InstrumentedMutex* db_mutex, ErrorHandler* db_error_handler,
     std::vector<SequenceNumber> existing_snapshots,
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, std::shared_ptr<Cache> table_cache,
@@ -276,7 +318,7 @@ CompactionJob::CompactionJob(
     : job_id_(job_id),
       compact_(new CompactionState(compaction)),
       compaction_job_stats_(compaction_job_stats),
-      compaction_stats_(1),
+      compaction_stats_(compaction->compaction_reason(), 1),
       dbname_(dbname),
       db_options_(db_options),
       env_options_(env_options),
@@ -291,7 +333,7 @@ CompactionJob::CompactionJob(
       output_directory_(output_directory),
       stats_(stats),
       db_mutex_(db_mutex),
-      db_bg_error_(db_bg_error),
+      db_error_handler_(db_error_handler),
       existing_snapshots_(std::move(existing_snapshots)),
       earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
       snapshot_checker_(snapshot_checker),
@@ -314,15 +356,13 @@ CompactionJob::~CompactionJob() {
   ThreadStatusUtil::ResetThreadStatus();
 }
 
-void CompactionJob::ReportStartedCompaction(
-    Compaction* compaction) {
+void CompactionJob::ReportStartedCompaction(Compaction* compaction) {
   const auto* cfd = compact_->compaction->column_family_data();
   ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
                                     db_options_.enable_thread_tracking);
 
-  ThreadStatusUtil::SetThreadOperationProperty(
-      ThreadStatus::COMPACTION_JOB_ID,
-      job_id_);
+  ThreadStatusUtil::SetThreadOperationProperty(ThreadStatus::COMPACTION_JOB_ID,
+                                               job_id_);
 
   ThreadStatusUtil::SetThreadOperationProperty(
       ThreadStatus::COMPACTION_INPUT_OUTPUT_LEVEL,
@@ -352,8 +392,7 @@ void CompactionJob::ReportStartedCompaction(
 
   // Set the thread operation after operation properties
   // to ensure GetThreadList() can always show them all together.
-  ThreadStatusUtil::SetThreadOperation(
-      ThreadStatus::OP_COMPACTION);
+  ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_COMPACTION);
 
   if (compaction_job_stats_) {
     compaction_job_stats_->is_manual_compaction =
@@ -368,11 +407,11 @@ void CompactionJob::Prepare() {
   // Generate file_levels_ for compaction berfore making Iterator
   auto* c = compact_->compaction;
   assert(c->column_family_data() != nullptr);
-  assert(c->column_family_data()->current()->storage_info()
-      ->NumLevelFiles(compact_->compaction->level()) > 0);
+  assert(c->column_family_data()->current()->storage_info()->NumLevelFiles(
+             compact_->compaction->level()) > 0);
 
-  write_hint_ = c->column_family_data()->CalculateSSTWriteHint(
-    c->output_level());
+  write_hint_ =
+      c->column_family_data()->CalculateSSTWriteHint(c->output_level());
   // Is this compaction producing files at the bottommost level?
   bottommost_level_ = c->bottommost_level();
 
@@ -455,14 +494,18 @@ void CompactionJob::GenSubcompactionBoundaries() {
   }
 
   std::sort(bounds.begin(), bounds.end(),
-    [cfd_comparator] (const Slice& a, const Slice& b) -> bool {
-      return cfd_comparator->Compare(ExtractUserKey(a), ExtractUserKey(b)) < 0;
-    });
+            [cfd_comparator](const Slice& a, const Slice& b) -> bool {
+              return cfd_comparator->Compare(ExtractUserKey(a),
+                                             ExtractUserKey(b)) < 0;
+            });
   // Remove duplicated entries from bounds
-  bounds.erase(std::unique(bounds.begin(), bounds.end(),
-    [cfd_comparator] (const Slice& a, const Slice& b) -> bool {
-      return cfd_comparator->Compare(ExtractUserKey(a), ExtractUserKey(b)) == 0;
-    }), bounds.end());
+  bounds.erase(
+      std::unique(bounds.begin(), bounds.end(),
+                  [cfd_comparator](const Slice& a, const Slice& b) -> bool {
+                    return cfd_comparator->Compare(ExtractUserKey(a),
+                                                   ExtractUserKey(b)) == 0;
+                  }),
+      bounds.end());
 
   // Combine consecutive pairs of boundaries into ranges with an approximate
   // size of data covered by keys in that range
@@ -485,12 +528,15 @@ void CompactionJob::GenSubcompactionBoundaries() {
 
   // Group the ranges into subcompactions
   const double min_file_fill_percent = 4.0 / 5;
-  uint64_t max_output_files = static_cast<uint64_t>(
-      std::ceil(sum / min_file_fill_percent /
-                c->mutable_cf_options()->MaxFileSizeForLevel(out_lvl)));
+  int base_level = v->storage_info()->base_level();
+  uint64_t max_output_files = static_cast<uint64_t>(std::ceil(
+      sum / min_file_fill_percent /
+      MaxFileSizeForLevel(*(c->mutable_cf_options()), out_lvl,
+          c->immutable_cf_options()->compaction_style, base_level,
+          c->immutable_cf_options()->level_compaction_dynamic_level_bytes)));
   uint64_t subcompactions =
       std::min({static_cast<uint64_t>(ranges.size()),
-                static_cast<uint64_t>(db_options_.max_subcompactions),
+                static_cast<uint64_t>(c->max_subcompactions()),
                 max_output_files});
 
   if (subcompactions > 1) {
@@ -563,11 +609,75 @@ Status CompactionJob::Run() {
     }
   }
 
+  if (status.ok()) {
+    thread_pool.clear();
+    std::vector<const FileMetaData*> files_meta;
+    for (const auto& state : compact_->sub_compact_states) {
+      for (const auto& output : state.outputs) {
+        files_meta.emplace_back(&output.meta);
+      }
+    }
+    ColumnFamilyData* cfd = compact_->compaction->column_family_data();
+    auto prefix_extractor =
+        compact_->compaction->mutable_cf_options()->prefix_extractor.get();
+    std::atomic<size_t> next_file_meta_idx(0);
+    auto verify_table = [&](Status& output_status) {
+      while (true) {
+        size_t file_idx = next_file_meta_idx.fetch_add(1);
+        if (file_idx >= files_meta.size()) {
+          break;
+        }
+        // Verify that the table is usable
+        // We set for_compaction to false and don't OptimizeForCompactionTableRead
+        // here because this is a special case after we finish the table building
+        // No matter whether use_direct_io_for_flush_and_compaction is true,
+        // we will regard this verification as user reads since the goal is
+        // to cache it here for further user reads
+        InternalIterator* iter = cfd->table_cache()->NewIterator(
+            ReadOptions(), env_options_, cfd->internal_comparator(),
+            *files_meta[file_idx], nullptr /* range_del_agg */,
+            prefix_extractor, nullptr,
+            cfd->internal_stats()->GetFileReadHist(
+                compact_->compaction->output_level()),
+            false, nullptr /* arena */, false /* skip_filters */,
+            compact_->compaction->output_level());
+        auto s = iter->status();
+
+        if (s.ok() && paranoid_file_checks_) {
+          for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {}
+          s = iter->status();
+        }
+
+        delete iter;
+
+        if (!s.ok()) {
+          output_status = s;
+          break;
+        }
+      }
+    };
+    for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
+      thread_pool.emplace_back(verify_table,
+                               std::ref(compact_->sub_compact_states[i].status));
+    }
+    verify_table(compact_->sub_compact_states[0].status);
+    for (auto& thread : thread_pool) {
+      thread.join();
+    }
+    for (const auto& state : compact_->sub_compact_states) {
+      if (!state.status.ok()) {
+        status = state.status;
+        break;
+      }
+    }
+  }
+
   TablePropertiesCollection tp;
   for (const auto& state : compact_->sub_compact_states) {
     for (const auto& output : state.outputs) {
-      auto fn = TableFileName(db_options_.db_paths, output.meta.fd.GetNumber(),
-                              output.meta.fd.GetPathId());
+      auto fn =
+          TableFileName(state.compaction->immutable_cf_options()->cf_paths,
+                        output.meta.fd.GetNumber(), output.meta.fd.GetPathId());
       tp[fn] = output.table_properties;
     }
   }
@@ -715,23 +825,29 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   // zstd's dictionary trainer, or just use them directly. Then, the dictionary
   // is used for compressing subsequent output files in the same subcompaction.
   const bool kUseZstdTrainer =
-      cfd->ioptions()->compression_opts.zstd_max_train_bytes > 0;
+      sub_compact->compaction->output_compression_opts().zstd_max_train_bytes >
+      0;
   const size_t kSampleBytes =
-      kUseZstdTrainer ? cfd->ioptions()->compression_opts.zstd_max_train_bytes
-                      : cfd->ioptions()->compression_opts.max_dict_bytes;
+      kUseZstdTrainer
+          ? sub_compact->compaction->output_compression_opts()
+                .zstd_max_train_bytes
+          : sub_compact->compaction->output_compression_opts().max_dict_bytes;
   const int kSampleLenShift = 6;  // 2^6 = 64-byte samples
   std::set<size_t> sample_begin_offsets;
   if (bottommost_level_ && kSampleBytes > 0) {
     const size_t kMaxSamples = kSampleBytes >> kSampleLenShift;
-    const size_t kOutFileLen = static_cast<size_t>(
-      mutable_cf_options->MaxFileSizeForLevel(
-        compact_->compaction->output_level()));
+    const size_t kOutFileLen =
+        static_cast<size_t>(MaxFileSizeForLevel(*mutable_cf_options,
+            compact_->compaction->output_level(),
+            cfd->ioptions()->compaction_style,
+            compact_->compaction->GetInputBaseLevel(),
+            cfd->ioptions()->level_compaction_dynamic_level_bytes));
     if (kOutFileLen != port::kMaxSizet) {
       const size_t kOutFileNumSamples = kOutFileLen >> kSampleLenShift;
       Random64 generator{versions_->NewFileNumber()};
       for (size_t i = 0; i < kMaxSamples; ++i) {
         sample_begin_offsets.insert(
-          static_cast<size_t>(generator.Uniform(kOutFileNumSamples))
+            static_cast<size_t>(generator.Uniform(kOutFileNumSamples))
             << kSampleLenShift);
       }
     }
@@ -768,18 +884,17 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   sub_compact->c_iter.reset(new CompactionIterator(
       input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
       &existing_snapshots_, earliest_write_conflict_snapshot_,
-      snapshot_checker_, env_, false, range_del_agg.get(),
-      sub_compact->compaction, compaction_filter, shutting_down_,
-      preserve_deletes_seqnum_));
+      snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_), false,
+      range_del_agg.get(), sub_compact->compaction, compaction_filter,
+      shutting_down_, preserve_deletes_seqnum_));
   auto c_iter = sub_compact->c_iter.get();
   c_iter->SeekToFirst();
-  if (c_iter->Valid() &&
-      sub_compact->compaction->output_level() != 0) {
+  if (c_iter->Valid() && sub_compact->compaction->output_level() != 0) {
     // ShouldStopBefore() maintains state based on keys processed so far. The
     // compaction loop always calls it on the "next" key, thus won't tell it the
     // first key. So we do that here.
-    sub_compact->ShouldStopBefore(
-      c_iter->key(), sub_compact->current_output_file_size);
+    sub_compact->ShouldStopBefore(c_iter->key(),
+                                  sub_compact->current_output_file_size);
   }
   const auto& c_iter_stats = c_iter->iter_stats();
   auto sample_begin_offset_iter = sample_begin_offsets.cbegin();
@@ -893,8 +1008,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
     c_iter->Next();
     if (!output_file_ended && c_iter->Valid() &&
         sub_compact->compaction->output_level() != 0 &&
-        sub_compact->ShouldStopBefore(
-          c_iter->key(), sub_compact->current_output_file_size) &&
+        sub_compact->ShouldStopBefore(c_iter->key(),
+                                      sub_compact->current_output_file_size) &&
         sub_compact->builder != nullptr) {
       // (2) this key belongs to the next file. For historical reasons, the
       // iterator status after advancing will be given to
@@ -919,7 +1034,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
         if (kUseZstdTrainer) {
           sub_compact->compression_dict = ZSTD_TrainDictionary(
               dict_sample_data, kSampleLenShift,
-              cfd->ioptions()->compression_opts.max_dict_bytes);
+              sub_compact->compaction->output_compression_opts()
+                  .max_dict_bytes);
         } else {
           sub_compact->compression_dict = std::move(dict_sample_data);
         }
@@ -946,8 +1062,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   RecordDroppedKeys(c_iter_stats, &sub_compact->compaction_job_stats);
   RecordCompactionIOStats();
 
-  if (status.ok() && (shutting_down_->load(std::memory_order_relaxed) ||
-                      cfd->IsDropped())) {
+  if (status.ok() &&
+      (shutting_down_->load(std::memory_order_relaxed) || cfd->IsDropped())) {
     status = Status::ShutdownInProgress(
         "Database shutdown or Column family drop during compaction");
   }
@@ -959,8 +1075,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   if (status.ok() && sub_compact->builder == nullptr &&
-      sub_compact->outputs.size() == 0 &&
-      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
+      sub_compact->outputs.size() == 0) {
     // handle subcompaction containing only range deletions
     status = OpenCompactionOutputFile(sub_compact);
   }
@@ -1048,9 +1163,13 @@ Status CompactionJob::FinishCompactionOutputFile(
   uint64_t output_number = sub_compact->current_output()->meta.fd.GetNumber();
   assert(output_number != 0);
 
+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  const Comparator* ucmp = cfd->user_comparator();
+
   // Check for iterator errors
   Status s = input_status;
   auto meta = &sub_compact->current_output()->meta;
+  assert(meta != nullptr);
   if (s.ok()) {
     Slice lower_bound_guard, upper_bound_guard;
     std::string smallest_user_key;
@@ -1079,9 +1198,71 @@ Status CompactionJob::FinishCompactionOutputFile(
       // subcompaction ends.
       upper_bound = sub_compact->end;
     }
-    range_del_agg->AddToBuilder(sub_compact->builder.get(), lower_bound,
-                                upper_bound, meta, range_del_out_stats,
-                                bottommost_level_);
+    auto earliest_snapshot = kMaxSequenceNumber;
+    if (existing_snapshots_.size() > 0) {
+      earliest_snapshot = existing_snapshots_[0];
+    }
+    auto it = range_del_agg->NewIterator();
+    if (lower_bound != nullptr) {
+      it->Seek(*lower_bound);
+    }
+    for (; it->Valid(); it->Next()) {
+      auto tombstone = it->Tombstone();
+      if (upper_bound != nullptr &&
+          ucmp->Compare(*upper_bound, tombstone.start_key_) <= 0) {
+        // Tombstones starting at upper_bound or later only need to be included
+        // in the next table. Break because subsequent tombstones will start
+        // even later.
+        break;
+      }
+
+      if (bottommost_level_ && tombstone.seq_ <= earliest_snapshot) {
+        // TODO(andrewkr): tombstones that span multiple output files are
+        // counted for each compaction output file, so lots of double counting.
+        range_del_out_stats->num_range_del_drop_obsolete++;
+        range_del_out_stats->num_record_drop_obsolete++;
+        continue;
+      }
+
+      auto kv = tombstone.Serialize();
+      sub_compact->builder->Add(kv.first.Encode(), kv.second);
+      InternalKey smallest_candidate = std::move(kv.first);
+      if (lower_bound != nullptr &&
+          ucmp->Compare(smallest_candidate.user_key(), *lower_bound) <= 0) {
+        // Pretend the smallest key has the same user key as lower_bound
+        // (the max key in the previous table or subcompaction) in order for
+        // files to appear key-space partitioned.
+        //
+        // Choose lowest seqnum so this file's smallest internal key comes
+        // after the previous file's/subcompaction's largest. The fake seqnum
+        // is OK because the read path's file-picking code only considers user
+        // key.
+        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+      }
+      InternalKey largest_candidate = tombstone.SerializeEndKey();
+      if (upper_bound != nullptr &&
+          ucmp->Compare(*upper_bound, largest_candidate.user_key()) <= 0) {
+        // Pretend the largest key has the same user key as upper_bound (the
+        // min key in the following table or subcompaction) in order for files
+        // to appear key-space partitioned.
+        //
+        // Choose highest seqnum so this file's largest internal key comes
+        // before the next file's/subcompaction's smallest. The fake seqnum is
+        // OK because the read path's file-picking code only considers the user
+        // key portion.
+        //
+        // Note Seek() also creates InternalKey with (user_key,
+        // kMaxSequenceNumber), but with kTypeDeletion (0x7) instead of
+        // kTypeRangeDeletion (0xF), so the range tombstone comes before the
+        // Seek() key in InternalKey's ordering. So Seek() will look in the
+        // next file for the user key.
+        largest_candidate =
+            InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      }
+      meta->UpdateBoundariesForRange(smallest_candidate, largest_candidate,
+                                     tombstone.seq_,
+                                     cfd->internal_comparator());
+    }
     meta->marked_for_compaction = sub_compact->builder->NeedCompact();
   }
   const uint64_t current_entries = sub_compact->builder->NumEntries();
@@ -1107,12 +1288,18 @@ Status CompactionJob::FinishCompactionOutputFile(
   }
   sub_compact->outfile.reset();
 
-  if (s.ok() && current_entries == 0) {
+  TableProperties tp;
+  if (s.ok()) {
+    tp = sub_compact->builder->GetTableProperties();
+  }
+
+  if (s.ok() && current_entries == 0 && tp.num_range_deletions == 0) {
     // If there is nothing to output, no necessary to generate a sst file.
     // This happens when the output level is bottom level, at the same time
     // the sub_compact output nothing.
-    std::string fname = TableFileName(
-        db_options_.db_paths, meta->fd.GetNumber(), meta->fd.GetPathId());
+    std::string fname =
+        TableFileName(sub_compact->compaction->immutable_cf_options()->cf_paths,
+                      meta->fd.GetNumber(), meta->fd.GetPathId());
     env_->DeleteFile(fname);
 
     // Also need to remove the file from outputs, or it will be added to the
@@ -1124,49 +1311,23 @@ Status CompactionJob::FinishCompactionOutputFile(
     return s;
   }
 
-  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
-  TableProperties tp;
-  if (s.ok() && current_entries > 0) {
-    // Verify that the table is usable
-    // We set for_compaction to false and don't OptimizeForCompactionTableRead
-    // here because this is a special case after we finish the table building
-    // No matter whether use_direct_io_for_flush_and_compaction is true,
-    // we will regrad 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(), meta->fd,
-        nullptr /* range_del_agg */, nullptr,
-        cfd->internal_stats()->GetFileReadHist(
-            compact_->compaction->output_level()),
-        false, nullptr /* arena */, false /* skip_filters */,
-        compact_->compaction->output_level());
-    s = iter->status();
-
-    if (s.ok() && paranoid_file_checks_) {
-      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {}
-      s = iter->status();
-    }
-
-    delete iter;
-
+  if (s.ok() && (current_entries > 0 || tp.num_range_deletions > 0)) {
     // Output to event logger and fire events.
-    if (s.ok()) {
-      tp = sub_compact->builder->GetTableProperties();
-      sub_compact->current_output()->table_properties =
-          std::make_shared<TableProperties>(tp);
-      ROCKS_LOG_INFO(db_options_.info_log,
-                     "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
-                     " keys, %" PRIu64 " bytes%s",
-                     cfd->GetName().c_str(), job_id_, output_number,
-                     current_entries, current_bytes,
-                     meta->marked_for_compaction ? " (need compaction)" : "");
-    }
+    sub_compact->current_output()->table_properties =
+        std::make_shared<TableProperties>(tp);
+    ROCKS_LOG_INFO(db_options_.info_log,
+                   "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
+                   " keys, %" PRIu64 " bytes%s",
+                   cfd->GetName().c_str(), job_id_, output_number,
+                   current_entries, current_bytes,
+                   meta->marked_for_compaction ? " (need compaction)" : "");
   }
   std::string fname;
   FileDescriptor output_fd;
   if (meta != nullptr) {
-    fname = TableFileName(db_options_.db_paths, meta->fd.GetNumber(),
-                          meta->fd.GetPathId());
+    fname =
+        TableFileName(sub_compact->compaction->immutable_cf_options()->cf_paths,
+                      meta->fd.GetNumber(), meta->fd.GetPathId());
     output_fd = meta->fd;
   } else {
     fname = "(nil)";
@@ -1180,27 +1341,19 @@ Status CompactionJob::FinishCompactionOutputFile(
   auto sfm =
       static_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
   if (sfm && meta != nullptr && meta->fd.GetPathId() == 0) {
-    auto fn = TableFileName(cfd->ioptions()->db_paths, meta->fd.GetNumber(),
-                            meta->fd.GetPathId());
+    auto fn =
+        TableFileName(sub_compact->compaction->immutable_cf_options()->cf_paths,
+                      meta->fd.GetNumber(), meta->fd.GetPathId());
     sfm->OnAddFile(fn);
     if (sfm->IsMaxAllowedSpaceReached()) {
       // TODO(ajkr): should we return OK() if max space was reached by the final
       // compaction output file (similarly to how flush works when full)?
-      s = Status::IOError("Max allowed space was reached");
+      s = Status::SpaceLimit("Max allowed space was reached");
       TEST_SYNC_POINT(
           "CompactionJob::FinishCompactionOutputFile:"
           "MaxAllowedSpaceReached");
       InstrumentedMutexLock l(db_mutex_);
-      if (db_bg_error_->ok()) {
-        Status new_bg_error = s;
-        // may temporarily unlock and lock the mutex.
-        EventHelpers::NotifyOnBackgroundError(
-            cfd->ioptions()->listeners, BackgroundErrorReason::kCompaction,
-            &new_bg_error, db_mutex_);
-        if (!new_bg_error.ok()) {
-          *db_bg_error_ = new_bg_error;
-        }
-      }
+      db_error_handler_->SetBGError(s, BackgroundErrorReason::kCompaction);
     }
   }
 #endif
@@ -1236,7 +1389,7 @@ Status CompactionJob::InstallCompactionResults(
         compaction->InputLevelSummary(&inputs_summary), compact_->total_bytes);
   }
 
-  // Add compaction outputs
+  // Add compaction inputs
   compaction->AddInputDeletions(compact_->compaction->edit());
 
   for (const auto& sub_compact : compact_->sub_compact_states) {
@@ -1266,8 +1419,9 @@ Status CompactionJob::OpenCompactionOutputFile(
   assert(sub_compact->builder == nullptr);
   // no need to lock because VersionSet::next_file_number_ is atomic
   uint64_t file_number = versions_->NewFileNumber();
-  std::string fname = TableFileName(db_options_.db_paths, file_number,
-                                    sub_compact->compaction->output_path_id());
+  std::string fname =
+      TableFileName(sub_compact->compaction->immutable_cf_options()->cf_paths,
+                    file_number, sub_compact->compaction->output_path_id());
   // Fire events.
   ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
 #ifndef ROCKSDB_LITE
@@ -1334,10 +1488,11 @@ Status CompactionJob::OpenCompactionOutputFile(
   }
 
   sub_compact->builder.reset(NewTableBuilder(
-      *cfd->ioptions(), cfd->internal_comparator(),
-      cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
-      sub_compact->outfile.get(), sub_compact->compaction->output_compression(),
-      cfd->ioptions()->compression_opts,
+      *cfd->ioptions(), *(sub_compact->compaction->mutable_cf_options()),
+      cfd->internal_comparator(), cfd->int_tbl_prop_collector_factories(),
+      cfd->GetID(), cfd->GetName(), sub_compact->outfile.get(),
+      sub_compact->compaction->output_compression(),
+      sub_compact->compaction->output_compression_opts(),
       sub_compact->compaction->output_level(), &sub_compact->compression_dict,
       skip_filters, output_file_creation_time));
   LogFlush(db_options_.info_log);
@@ -1369,8 +1524,7 @@ void CompactionJob::CleanupCompaction() {
 
 #ifndef ROCKSDB_LITE
 namespace {
-void CopyPrefix(
-    const Slice& src, size_t prefix_length, std::string* dst) {
+void CopyPrefix(const Slice& src, size_t prefix_length, std::string* dst) {
   assert(prefix_length > 0);
   size_t length = src.size() > prefix_length ? prefix_length : src.size();
   dst->assign(src.data(), length);
@@ -1389,13 +1543,11 @@ void CompactionJob::UpdateCompactionStats() {
     if (compaction->level(input_level) != compaction->output_level()) {
       UpdateCompactionInputStatsHelper(
           &compaction_stats_.num_input_files_in_non_output_levels,
-          &compaction_stats_.bytes_read_non_output_levels,
-          input_level);
+          &compaction_stats_.bytes_read_non_output_levels, input_level);
     } else {
       UpdateCompactionInputStatsHelper(
           &compaction_stats_.num_input_files_in_output_level,
-          &compaction_stats_.bytes_read_output_level,
-          input_level);
+          &compaction_stats_.bytes_read_output_level, input_level);
     }
   }
 
@@ -1418,8 +1570,9 @@ void CompactionJob::UpdateCompactionStats() {
   }
 }
 
-void CompactionJob::UpdateCompactionInputStatsHelper(
-    int* num_files, uint64_t* bytes_read, int input_level) {
+void CompactionJob::UpdateCompactionInputStatsHelper(int* num_files,
+                                                     uint64_t* bytes_read,
+                                                     int input_level) {
   const Compaction* compaction = compact_->compaction;
   auto num_input_files = compaction->num_input_files(input_level);
   *num_files += static_cast<int>(num_input_files);
@@ -1440,10 +1593,8 @@ void CompactionJob::UpdateCompactionJobStats(
 
     // input information
     compaction_job_stats_->total_input_bytes =
-        stats.bytes_read_non_output_levels +
-        stats.bytes_read_output_level;
-    compaction_job_stats_->num_input_records =
-        compact_->num_input_records;
+        stats.bytes_read_non_output_levels + stats.bytes_read_output_level;
+    compaction_job_stats_->num_input_records = compact_->num_input_records;
     compaction_job_stats_->num_input_files =
         stats.num_input_files_in_non_output_levels +
         stats.num_input_files_in_output_level;
@@ -1452,21 +1603,20 @@ void CompactionJob::UpdateCompactionJobStats(
 
     // output information
     compaction_job_stats_->total_output_bytes = stats.bytes_written;
-    compaction_job_stats_->num_output_records =
-        compact_->num_output_records;
+    compaction_job_stats_->num_output_records = compact_->num_output_records;
     compaction_job_stats_->num_output_files = stats.num_output_files;
 
     if (compact_->NumOutputFiles() > 0U) {
-      CopyPrefix(
-          compact_->SmallestUserKey(),
-          CompactionJobStats::kMaxPrefixLength,
-          &compaction_job_stats_->smallest_output_key_prefix);
-      CopyPrefix(
-          compact_->LargestUserKey(),
-          CompactionJobStats::kMaxPrefixLength,
-          &compaction_job_stats_->largest_output_key_prefix);
+      CopyPrefix(compact_->SmallestUserKey(),
+                 CompactionJobStats::kMaxPrefixLength,
+                 &compaction_job_stats_->smallest_output_key_prefix);
+      CopyPrefix(compact_->LargestUserKey(),
+                 CompactionJobStats::kMaxPrefixLength,
+                 &compaction_job_stats_->largest_output_key_prefix);
     }
   }
+#else
+  (void)stats;
 #endif  // !ROCKSDB_LITE
 }
 
@@ -1489,7 +1639,9 @@ void CompactionJob::LogCompaction() {
     // build event logger report
     auto stream = event_logger_->Log();
     stream << "job" << job_id_ << "event"
-           << "compaction_started";
+           << "compaction_started"
+           << "compaction_reason"
+           << GetCompactionReasonString(compaction->compaction_reason());
     for (size_t i = 0; i < compaction->num_input_levels(); ++i) {
       stream << ("files_L" + ToString(compaction->level(i)));
       stream.StartArray();

db/compaction_job.h

@@ -30,6 +30,7 @@
 #include "db/write_controller.h"
 #include "db/write_thread.h"
 #include "options/db_options.h"
+#include "options/cf_options.h"
 #include "port/port.h"
 #include "rocksdb/compaction_filter.h"
 #include "rocksdb/compaction_job_stats.h"
@@ -46,6 +47,7 @@
 namespace rocksdb {
 
 class Arena;
+class ErrorHandler;
 class MemTable;
 class SnapshotChecker;
 class TableCache;
@@ -63,7 +65,7 @@ class CompactionJob {
                 LogBuffer* log_buffer,
                 Directory* db_directory, Directory* output_directory,
                 Statistics* stats, InstrumentedMutex* db_mutex,
-                Status* db_bg_error,
+                ErrorHandler* db_error_handler,
                 std::vector<SequenceNumber> existing_snapshots,
                 SequenceNumber earliest_write_conflict_snapshot,
                 const SnapshotChecker* snapshot_checker,
@@ -144,7 +146,7 @@ class CompactionJob {
   Directory* output_directory_;
   Statistics* stats_;
   InstrumentedMutex* db_mutex_;
-  Status* db_bg_error_;
+  ErrorHandler* db_error_handler_;
   // If there were two snapshots with seq numbers s1 and
   // s2 and s1 < s2, and if we find two instances of a key k1 then lies
   // entirely within s1 and s2, then the earlier version of k1 can be safely

db/compaction_job_stats_test.cc

@@ -98,7 +98,7 @@ class CompactionJobStatsTest : public testing::Test,
   CompactionJobStatsTest() : env_(Env::Default()) {
     env_->SetBackgroundThreads(1, Env::LOW);
     env_->SetBackgroundThreads(1, Env::HIGH);
-    dbname_ = test::TmpDir(env_) + "/compaction_job_stats_test";
+    dbname_ = test::PerThreadDBPath("compaction_job_stats_test");
     alternative_wal_dir_ = dbname_ + "/wal";
     Options options;
     options.create_if_missing = true;
@@ -806,7 +806,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
     stats_checker->set_verify_next_comp_io_stats(true);
     std::atomic<bool> first_prepare_write(true);
     rocksdb::SyncPoint::GetInstance()->SetCallBack(
-        "WritableFileWriter::Append:BeforePrepareWrite", [&](void* arg) {
+        "WritableFileWriter::Append:BeforePrepareWrite", [&](void* /*arg*/) {
           if (first_prepare_write.load()) {
             options.env->SleepForMicroseconds(3);
             first_prepare_write.store(false);
@@ -815,7 +815,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
 
     std::atomic<bool> first_flush(true);
     rocksdb::SyncPoint::GetInstance()->SetCallBack(
-        "WritableFileWriter::Flush:BeforeAppend", [&](void* arg) {
+        "WritableFileWriter::Flush:BeforeAppend", [&](void* /*arg*/) {
           if (first_flush.load()) {
             options.env->SleepForMicroseconds(3);
             first_flush.store(false);
@@ -824,7 +824,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
 
     std::atomic<bool> first_sync(true);
     rocksdb::SyncPoint::GetInstance()->SetCallBack(
-        "WritableFileWriter::SyncInternal:0", [&](void* arg) {
+        "WritableFileWriter::SyncInternal:0", [&](void* /*arg*/) {
           if (first_sync.load()) {
             options.env->SleepForMicroseconds(3);
             first_sync.store(false);
@@ -833,7 +833,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
 
     std::atomic<bool> first_range_sync(true);
     rocksdb::SyncPoint::GetInstance()->SetCallBack(
-        "WritableFileWriter::RangeSync:0", [&](void* arg) {
+        "WritableFileWriter::RangeSync:0", [&](void* /*arg*/) {
           if (first_range_sync.load()) {
             options.env->SleepForMicroseconds(3);
             first_range_sync.store(false);

db/compaction_job_test.cc

@@ -12,6 +12,7 @@
 
 #include "db/column_family.h"
 #include "db/compaction_job.h"
+#include "db/error_handler.h"
 #include "db/version_set.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/db.h"
@@ -67,7 +68,7 @@ class CompactionJobTest : public testing::Test {
  public:
   CompactionJobTest()
       : env_(Env::Default()),
-        dbname_(test::TmpDir() + "/compaction_job_test"),
+        dbname_(test::PerThreadDBPath("compaction_job_test")),
         db_options_(),
         mutable_cf_options_(cf_options_),
         table_cache_(NewLRUCache(50000, 16)),
@@ -77,7 +78,8 @@ class CompactionJobTest : public testing::Test {
                                  &write_controller_)),
         shutting_down_(false),
         preserve_deletes_seqnum_(0),
-        mock_table_factory_(new mock::MockTableFactory()) {
+        mock_table_factory_(new mock::MockTableFactory()),
+        error_handler_(db_options_, &mutex_) {
     EXPECT_OK(env_->CreateDirIfMissing(dbname_));
     db_options_.db_paths.emplace_back(dbname_,
                                       std::numeric_limits<uint64_t>::max());
@@ -246,7 +248,8 @@ class CompactionJobTest : public testing::Test {
     Compaction compaction(cfd->current()->storage_info(), *cfd->ioptions(),
                           *cfd->GetLatestMutableCFOptions(),
                           compaction_input_files, 1, 1024 * 1024,
-                          10 * 1024 * 1024, 0, kNoCompression, {}, true);
+                          10 * 1024 * 1024, 0, kNoCompression,
+                          cfd->ioptions()->compression_opts, 0, {}, true);
     compaction.SetInputVersion(cfd->current());
 
     LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
@@ -254,13 +257,12 @@ class CompactionJobTest : public testing::Test {
     EventLogger event_logger(db_options_.info_log.get());
     // TODO(yiwu) add a mock snapshot checker and add test for it.
     SnapshotChecker* snapshot_checker = nullptr;
-    CompactionJob compaction_job(0, &compaction, db_options_, env_options_,
-                                 versions_.get(), &shutting_down_,
-                                 preserve_deletes_seqnum_, &log_buffer,
-                                 nullptr, nullptr, nullptr, &mutex_, &bg_error_,
-                                 snapshots, earliest_write_conflict_snapshot,
-                                 snapshot_checker, table_cache_, &event_logger,
-                                 false, false, dbname_, &compaction_job_stats_);
+    CompactionJob compaction_job(
+        0, &compaction, db_options_, env_options_, versions_.get(),
+        &shutting_down_, preserve_deletes_seqnum_, &log_buffer, nullptr,
+        nullptr, nullptr, &mutex_, &error_handler_, snapshots,
+        earliest_write_conflict_snapshot, snapshot_checker, table_cache_,
+        &event_logger, false, false, dbname_, &compaction_job_stats_);
     VerifyInitializationOfCompactionJobStats(compaction_job_stats_);
 
     compaction_job.Prepare();
@@ -302,7 +304,7 @@ class CompactionJobTest : public testing::Test {
   ColumnFamilyData* cfd_;
   std::unique_ptr<CompactionFilter> compaction_filter_;
   std::shared_ptr<MergeOperator> merge_op_;
-  Status bg_error_;
+  ErrorHandler error_handler_;
 };
 
 TEST_F(CompactionJobTest, Simple) {

db/compaction_picker.cc

@@ -59,7 +59,7 @@ bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
   }
 
   if (span_len >= min_files_to_compact &&
-      new_compact_bytes_per_del_file < max_compact_bytes_per_del_file) {
+      compact_bytes_per_del_file < max_compact_bytes_per_del_file) {
     assert(comp_inputs != nullptr);
     comp_inputs->level = 0;
     for (size_t i = 0; i < span_len; ++i) {
@@ -110,6 +110,24 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
   }
 }
 
+CompressionOptions GetCompressionOptions(const ImmutableCFOptions& ioptions,
+                                         const VersionStorageInfo* vstorage,
+                                         int level,
+                                         const bool enable_compression) {
+  if (!enable_compression) {
+    return ioptions.compression_opts;
+  }
+  // If bottommost_compression is set and we are compacting to the
+  // bottommost level then we should use the specified compression options
+  // for the bottmomost_compression.
+  if (ioptions.bottommost_compression != kDisableCompressionOption &&
+      level >= (vstorage->num_non_empty_levels() - 1) &&
+      ioptions.bottommost_compression_opts.enabled) {
+    return ioptions.bottommost_compression_opts;
+  }
+  return ioptions.compression_opts;
+}
+
 CompactionPicker::CompactionPicker(const ImmutableCFOptions& ioptions,
                                    const InternalKeyComparator* icmp)
     : ioptions_(ioptions), icmp_(icmp) {}
@@ -301,7 +319,9 @@ Compaction* CompactionPicker::CompactFiles(
       new Compaction(vstorage, ioptions_, mutable_cf_options, input_files,
                      output_level, compact_options.output_file_size_limit,
                      mutable_cf_options.max_compaction_bytes, output_path_id,
-                     compact_options.compression, /* grandparents */ {}, true);
+                     compact_options.compression, ioptions_.compression_opts,
+                     compact_options.max_subcompactions,
+                     /* grandparents */ {}, true);
   RegisterCompaction(c);
   return c;
 }
@@ -364,7 +384,7 @@ bool CompactionPicker::IsRangeInCompaction(VersionStorageInfo* vstorage,
   assert(level < NumberLevels());
 
   vstorage->GetOverlappingInputs(level, smallest, largest, &inputs,
-                                 *level_index, level_index);
+                                 level_index ? *level_index : 0, level_index);
   return AreFilesInCompaction(inputs);
 }
 
@@ -475,7 +495,7 @@ bool CompactionPicker::SetupOtherInputs(
       ROCKS_LOG_INFO(ioptions_.info_log,
                      "[%s] Expanding@%d %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt
                      "(%" PRIu64 "+%" PRIu64 " bytes) to %" ROCKSDB_PRIszt
-                     "+%" ROCKSDB_PRIszt " (%" PRIu64 "+%" PRIu64 "bytes)\n",
+                     "+%" ROCKSDB_PRIszt " (%" PRIu64 "+%" PRIu64 " bytes)\n",
                      cf_name.c_str(), input_level, inputs->size(),
                      output_level_inputs->size(), inputs_size,
                      output_level_inputs_size, expanded_inputs.size(),
@@ -504,7 +524,8 @@ void CompactionPicker::GetGrandparents(
 Compaction* CompactionPicker::CompactRange(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     VersionStorageInfo* vstorage, int input_level, int output_level,
-    uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
+    uint32_t output_path_id, uint32_t max_subcompactions,
+    const InternalKey* begin, const InternalKey* end,
     InternalKey** compaction_end, bool* manual_conflict) {
   // CompactionPickerFIFO has its own implementation of compact range
   assert(ioptions_.compaction_style != kCompactionStyleFIFO);
@@ -566,11 +587,14 @@ Compaction* CompactionPicker::CompactRange(
 
     Compaction* c = new Compaction(
         vstorage, ioptions_, mutable_cf_options, std::move(inputs),
-        output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
+        output_level,
+        MaxFileSizeForLevel(mutable_cf_options, output_level,
+                            ioptions_.compaction_style),
         /* max_compaction_bytes */ LLONG_MAX, output_path_id,
         GetCompressionType(ioptions_, vstorage, mutable_cf_options,
                            output_level, 1),
-        /* grandparents */ {}, /* is manual */ true);
+        GetCompressionOptions(ioptions_, vstorage, output_level),
+        max_subcompactions, /* grandparents */ {}, /* is manual */ true);
     RegisterCompaction(c);
     return c;
   }
@@ -616,7 +640,7 @@ Compaction* CompactionPicker::CompactRange(
       }
     }
   }
-  assert(output_path_id < static_cast<uint32_t>(ioptions_.db_paths.size()));
+  assert(output_path_id < static_cast<uint32_t>(ioptions_.cf_paths.size()));
 
   if (ExpandInputsToCleanCut(cf_name, vstorage, &inputs) == false) {
     // manual compaction is now multi-threaded, so it can
@@ -673,11 +697,16 @@ Compaction* CompactionPicker::CompactRange(
   GetGrandparents(vstorage, inputs, output_level_inputs, &grandparents);
   Compaction* compaction = new Compaction(
       vstorage, ioptions_, mutable_cf_options, std::move(compaction_inputs),
-      output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
+      output_level,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          ioptions_.compaction_style, vstorage->base_level(),
+                          ioptions_.level_compaction_dynamic_level_bytes),
       mutable_cf_options.max_compaction_bytes, output_path_id,
       GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
                          vstorage->base_level()),
-      std::move(grandparents), /* is manual compaction */ true);
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      max_subcompactions, std::move(grandparents),
+      /* is manual compaction */ true);
 
   TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
   RegisterCompaction(compaction);
@@ -845,6 +874,11 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
       }
     }
   }
+  if (RangeOverlapWithCompaction(smallestkey, largestkey, output_level)) {
+    return Status::Aborted(
+        "A running compaction is writing to the same output level in an "
+        "overlapping key range");
+  }
   return Status::OK();
 }
 
@@ -939,8 +973,83 @@ void CompactionPicker::UnregisterCompaction(Compaction* c) {
   compactions_in_progress_.erase(c);
 }
 
+void CompactionPicker::PickFilesMarkedForCompaction(
+    const std::string& cf_name, VersionStorageInfo* vstorage, int* start_level,
+    int* output_level, CompactionInputFiles* start_level_inputs) {
+  if (vstorage->FilesMarkedForCompaction().empty()) {
+    return;
+  }
+
+  auto continuation = [&, cf_name](std::pair<int, FileMetaData*> level_file) {
+    // If it's being compacted it has nothing to do here.
+    // If this assert() fails that means that some function marked some
+    // files as being_compacted, but didn't call ComputeCompactionScore()
+    assert(!level_file.second->being_compacted);
+    *start_level = level_file.first;
+    *output_level =
+        (*start_level == 0) ? vstorage->base_level() : *start_level + 1;
+
+    if (*start_level == 0 && !level0_compactions_in_progress()->empty()) {
+      return false;
+    }
+
+    start_level_inputs->files = {level_file.second};
+    start_level_inputs->level = *start_level;
+    return ExpandInputsToCleanCut(cf_name, vstorage, start_level_inputs);
+  };
+
+  // take a chance on a random file first
+  Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage));
+  size_t random_file_index = static_cast<size_t>(rnd.Uniform(
+      static_cast<uint64_t>(vstorage->FilesMarkedForCompaction().size())));
+
+  if (continuation(vstorage->FilesMarkedForCompaction()[random_file_index])) {
+    // found the compaction!
+    return;
+  }
+
+  for (auto& level_file : vstorage->FilesMarkedForCompaction()) {
+    if (continuation(level_file)) {
+      // found the compaction!
+      return;
+    }
+  }
+  start_level_inputs->files.clear();
+}
+
+bool CompactionPicker::GetOverlappingL0Files(
+    VersionStorageInfo* vstorage, CompactionInputFiles* start_level_inputs,
+    int output_level, int* parent_index) {
+  // Two level 0 compaction won't run at the same time, so don't need to worry
+  // about files on level 0 being compacted.
+  assert(level0_compactions_in_progress()->empty());
+  InternalKey smallest, largest;
+  GetRange(*start_level_inputs, &smallest, &largest);
+  // Note that the next call will discard the file we placed in
+  // c->inputs_[0] earlier and replace it with an overlapping set
+  // which will include the picked file.
+  start_level_inputs->files.clear();
+  vstorage->GetOverlappingInputs(0, &smallest, &largest,
+                                 &(start_level_inputs->files));
+
+  // If we include more L0 files in the same compaction run it can
+  // cause the 'smallest' and 'largest' key to get extended to a
+  // larger range. So, re-invoke GetRange to get the new key range
+  GetRange(*start_level_inputs, &smallest, &largest);
+  if (IsRangeInCompaction(vstorage, &smallest, &largest, output_level,
+                          parent_index)) {
+    return false;
+  }
+  assert(!start_level_inputs->files.empty());
+
+  return true;
+}
+
 bool LevelCompactionPicker::NeedsCompaction(
     const VersionStorageInfo* vstorage) const {
+  if (!vstorage->ExpiredTtlFiles().empty()) {
+    return true;
+  }
   if (!vstorage->BottommostFilesMarkedForCompaction().empty()) {
     return true;
   }
@@ -1007,8 +1116,7 @@ class LevelCompactionBuilder {
   // otherwise, returns false.
   bool PickIntraL0Compaction();
 
-  // If there is any file marked for compaction, put put it into inputs.
-  void PickFilesMarkedForCompaction();
+  void PickExpiredTtlFiles();
 
   const std::string& cf_name_;
   VersionStorageInfo* vstorage_;
@@ -1036,8 +1144,8 @@ class LevelCompactionBuilder {
   static const int kMinFilesForIntraL0Compaction = 4;
 };
 
-void LevelCompactionBuilder::PickFilesMarkedForCompaction() {
-  if (vstorage_->FilesMarkedForCompaction().empty()) {
+void LevelCompactionBuilder::PickExpiredTtlFiles() {
+  if (vstorage_->ExpiredTtlFiles().empty()) {
     return;
   }
 
@@ -1050,8 +1158,9 @@ void LevelCompactionBuilder::PickFilesMarkedForCompaction() {
     output_level_ =
         (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1;
 
-    if (start_level_ == 0 &&
-        !compaction_picker_->level0_compactions_in_progress()->empty()) {
+    if ((start_level_ == vstorage_->num_non_empty_levels() - 1) ||
+        (start_level_ == 0 &&
+         !compaction_picker_->level0_compactions_in_progress()->empty())) {
       return false;
     }
 
@@ -1061,22 +1170,13 @@ void LevelCompactionBuilder::PickFilesMarkedForCompaction() {
                                                       &start_level_inputs_);
   };
 
-  // take a chance on a random file first
-  Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage_));
-  size_t random_file_index = static_cast<size_t>(rnd.Uniform(
-      static_cast<uint64_t>(vstorage_->FilesMarkedForCompaction().size())));
-
-  if (continuation(vstorage_->FilesMarkedForCompaction()[random_file_index])) {
-    // found the compaction!
-    return;
-  }
-
-  for (auto& level_file : vstorage_->FilesMarkedForCompaction()) {
+  for (auto& level_file : vstorage_->ExpiredTtlFiles()) {
     if (continuation(level_file)) {
       // found the compaction!
       return;
     }
   }
+
   start_level_inputs_.files.clear();
 }
 
@@ -1131,61 +1231,51 @@ void LevelCompactionBuilder::SetupInitialFiles() {
   // if we didn't find a compaction, check if there are any files marked for
   // compaction
   if (start_level_inputs_.empty()) {
-    is_manual_ = true;
     parent_index_ = base_index_ = -1;
-    PickFilesMarkedForCompaction();
-    if (start_level_inputs_.empty()) {
-      size_t i;
-      for (i = 0; i < vstorage_->BottommostFilesMarkedForCompaction().size();
-           ++i) {
-        auto& level_and_file =
-            vstorage_->BottommostFilesMarkedForCompaction()[i];
-        assert(!level_and_file.second->being_compacted);
-        start_level_inputs_.level = output_level_ = start_level_ =
-            level_and_file.first;
-        start_level_inputs_.files = {level_and_file.second};
-        if (compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
-                                                       &start_level_inputs_)) {
-          break;
-        }
-      }
-      if (i == vstorage_->BottommostFilesMarkedForCompaction().size()) {
-        start_level_inputs_.clear();
-      } else {
-        assert(!start_level_inputs_.empty());
-        compaction_reason_ = CompactionReason::kBottommostFiles;
-      }
-    } else {
+
+    // PickFilesMarkedForCompaction();
+    compaction_picker_->PickFilesMarkedForCompaction(
+        cf_name_, vstorage_, &start_level_, &output_level_, &start_level_inputs_);
+    if (!start_level_inputs_.empty()) {
+      is_manual_ = true;
       compaction_reason_ = CompactionReason::kFilesMarkedForCompaction;
+      return;
+    }
+
+    size_t i;
+    for (i = 0; i < vstorage_->BottommostFilesMarkedForCompaction().size();
+         ++i) {
+      auto& level_and_file = vstorage_->BottommostFilesMarkedForCompaction()[i];
+      assert(!level_and_file.second->being_compacted);
+      start_level_inputs_.level = output_level_ = start_level_ =
+          level_and_file.first;
+      start_level_inputs_.files = {level_and_file.second};
+      if (compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
+                                                     &start_level_inputs_)) {
+        break;
+      }
+    }
+    if (i == vstorage_->BottommostFilesMarkedForCompaction().size()) {
+      start_level_inputs_.clear();
+    } else {
+      assert(!start_level_inputs_.empty());
+      compaction_reason_ = CompactionReason::kBottommostFiles;
+      return;
+    }
+
+    assert(start_level_inputs_.empty());
+    PickExpiredTtlFiles();
+    if (!start_level_inputs_.empty()) {
+      compaction_reason_ = CompactionReason::kTtl;
     }
   }
 }
 
 bool LevelCompactionBuilder::SetupOtherL0FilesIfNeeded() {
   if (start_level_ == 0 && output_level_ != 0) {
-    // Two level 0 compaction won't run at the same time, so don't need to worry
-    // about files on level 0 being compacted.
-    assert(compaction_picker_->level0_compactions_in_progress()->empty());
-    InternalKey smallest, largest;
-    compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest);
-    // Note that the next call will discard the file we placed in
-    // c->inputs_[0] earlier and replace it with an overlapping set
-    // which will include the picked file.
-    start_level_inputs_.files.clear();
-    vstorage_->GetOverlappingInputs(0, &smallest, &largest,
-                                    &start_level_inputs_.files);
-
-    // If we include more L0 files in the same compaction run it can
-    // cause the 'smallest' and 'largest' key to get extended to a
-    // larger range. So, re-invoke GetRange to get the new key range
-    compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest);
-    if (compaction_picker_->IsRangeInCompaction(
-            vstorage_, &smallest, &largest, output_level_, &parent_index_)) {
-      return false;
-    }
+    return compaction_picker_->GetOverlappingL0Files(
+        vstorage_, &start_level_inputs_, output_level_, &parent_index_);
   }
-  assert(!start_level_inputs_.files.empty());
-
   return true;
 }
 
@@ -1258,13 +1348,17 @@ Compaction* LevelCompactionBuilder::PickCompaction() {
 Compaction* LevelCompactionBuilder::GetCompaction() {
   auto c = new Compaction(
       vstorage_, ioptions_, mutable_cf_options_, std::move(compaction_inputs_),
-      output_level_, mutable_cf_options_.MaxFileSizeForLevel(output_level_),
+      output_level_,
+      MaxFileSizeForLevel(mutable_cf_options_, output_level_,
+                          ioptions_.compaction_style, vstorage_->base_level(),
+                          ioptions_.level_compaction_dynamic_level_bytes),
       mutable_cf_options_.max_compaction_bytes,
       GetPathId(ioptions_, mutable_cf_options_, output_level_),
       GetCompressionType(ioptions_, vstorage_, mutable_cf_options_,
                          output_level_, vstorage_->base_level()),
-      std::move(grandparents_), is_manual_, start_level_score_,
-      false /* deletion_compaction */, compaction_reason_);
+      GetCompressionOptions(ioptions_, vstorage_, output_level_),
+      /* max_subcompactions */ 0, std::move(grandparents_), is_manual_,
+      start_level_score_, false /* deletion_compaction */, compaction_reason_);
 
   // If it's level 0 compaction, make sure we don't execute any other level 0
   // compactions in parallel
@@ -1287,10 +1381,10 @@ uint32_t LevelCompactionBuilder::GetPathId(
     const ImmutableCFOptions& ioptions,
     const MutableCFOptions& mutable_cf_options, int level) {
   uint32_t p = 0;
-  assert(!ioptions.db_paths.empty());
+  assert(!ioptions.cf_paths.empty());
 
   // size remaining in the most recent path
-  uint64_t current_path_size = ioptions.db_paths[0].target_size;
+  uint64_t current_path_size = ioptions.cf_paths[0].target_size;
 
   uint64_t level_size;
   int cur_level = 0;
@@ -1300,7 +1394,7 @@ uint32_t LevelCompactionBuilder::GetPathId(
   level_size = mutable_cf_options.max_bytes_for_level_base;
 
   // Last path is the fallback
-  while (p < ioptions.db_paths.size() - 1) {
+  while (p < ioptions.cf_paths.size() - 1) {
     if (level_size <= current_path_size) {
       if (cur_level == level) {
         // Does desired level fit in this path?
@@ -1327,7 +1421,7 @@ uint32_t LevelCompactionBuilder::GetPathId(
       }
     }
     p++;
-    current_path_size = ioptions.db_paths[p].target_size;
+    current_path_size = ioptions.cf_paths[p].target_size;
   }
   return p;
 }
@@ -1512,7 +1606,8 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(
 
   Compaction* c = new Compaction(
       vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
-      kNoCompression, {}, /* is manual */ false, vstorage->CompactionScore(0),
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
       /* is deletion compaction */ true, CompactionReason::kFIFOTtl);
   return c;
 }
@@ -1531,16 +1626,27 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
     if (mutable_cf_options.compaction_options_fifo.allow_compaction &&
         level_files.size() > 0) {
       CompactionInputFiles comp_inputs;
+      // try to prevent same files from being compacted multiple times, which
+      // could produce large files that may never TTL-expire. Achieve this by
+      // disallowing compactions with files larger than memtable (inflate its
+      // size by 10% to account for uncompressed L0 files that may have size
+      // slightly greater than memtable size limit).
+      size_t max_compact_bytes_per_del_file =
+          static_cast<size_t>(MultiplyCheckOverflow(
+              static_cast<uint64_t>(mutable_cf_options.write_buffer_size),
+              1.1));
       if (FindIntraL0Compaction(
               level_files,
               mutable_cf_options
-                  .level0_file_num_compaction_trigger /* min_files_to_compact */,
-              mutable_cf_options.write_buffer_size, &comp_inputs)) {
+                  .level0_file_num_compaction_trigger /* min_files_to_compact */
+              ,
+              max_compact_bytes_per_del_file, &comp_inputs)) {
         Compaction* c = new Compaction(
             vstorage, ioptions_, mutable_cf_options, {comp_inputs}, 0,
             16 * 1024 * 1024 /* output file size limit */,
             0 /* max compaction bytes, not applicable */,
-            0 /* output path ID */, mutable_cf_options.compression, {},
+            0 /* output path ID */, mutable_cf_options.compression,
+            ioptions_.compression_opts, 0 /* max_subcompactions */, {},
             /* is manual */ false, vstorage->CompactionScore(0),
             /* is deletion compaction */ false,
             CompactionReason::kFIFOReduceNumFiles);
@@ -1588,7 +1694,8 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
 
   Compaction* c = new Compaction(
       vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
-      kNoCompression, {}, /* is manual */ false, vstorage->CompactionScore(0),
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
       /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
   return c;
 }
@@ -1612,9 +1719,13 @@ Compaction* FIFOCompactionPicker::PickCompaction(
 Compaction* FIFOCompactionPicker::CompactRange(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     VersionStorageInfo* vstorage, int input_level, int output_level,
-    uint32_t /*output_path_id*/, const InternalKey* /*begin*/,
-    const InternalKey* /*end*/, InternalKey** compaction_end,
-    bool* /*manual_conflict*/) {
+    uint32_t /*output_path_id*/, uint32_t /*max_subcompactions*/,
+    const InternalKey* /*begin*/, const InternalKey* /*end*/,
+    InternalKey** compaction_end, bool* /*manual_conflict*/) {
+#ifdef NDEBUG
+  (void)input_level;
+  (void)output_level;
+#endif
   assert(input_level == 0);
   assert(output_level == 0);
   *compaction_end = nullptr;

db/compaction_picker.h

@@ -58,7 +58,8 @@ class CompactionPicker {
   virtual Compaction* CompactRange(
       const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
       VersionStorageInfo* vstorage, int input_level, int output_level,
-      uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
+      uint32_t output_path_id, uint32_t max_subcompactions,
+      const InternalKey* begin, const InternalKey* end,
       InternalKey** compaction_end, bool* manual_conflict);
 
   // The maximum allowed output level.  Default value is NumberLevels() - 1.
@@ -174,6 +175,15 @@ class CompactionPicker {
                        const CompactionInputFiles& output_level_inputs,
                        std::vector<FileMetaData*>* grandparents);
 
+  void PickFilesMarkedForCompaction(const std::string& cf_name,
+                                    VersionStorageInfo* vstorage,
+                                    int* start_level, int* output_level,
+                                    CompactionInputFiles* start_level_inputs);
+
+  bool GetOverlappingL0Files(VersionStorageInfo* vstorage,
+                             CompactionInputFiles* start_level_inputs,
+                             int output_level, int* parent_index);
+
   // Register this compaction in the set of running compactions
   void RegisterCompaction(Compaction* c);
 
@@ -238,7 +248,8 @@ class FIFOCompactionPicker : public CompactionPicker {
   virtual Compaction* CompactRange(
       const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
       VersionStorageInfo* vstorage, int input_level, int output_level,
-      uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
+      uint32_t output_path_id, uint32_t max_subcompactions,
+      const InternalKey* begin, const InternalKey* end,
       InternalKey** compaction_end, bool* manual_conflict) override;
 
   // The maximum allowed output level.  Always returns 0.
@@ -280,6 +291,7 @@ class NullCompactionPicker : public CompactionPicker {
                            VersionStorageInfo* /*vstorage*/,
                            int /*input_level*/, int /*output_level*/,
                            uint32_t /*output_path_id*/,
+                           uint32_t /*max_subcompactions*/,
                            const InternalKey* /*begin*/,
                            const InternalKey* /*end*/,
                            InternalKey** /*compaction_end*/,
@@ -301,4 +313,9 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
                                    int level, int base_level,
                                    const bool enable_compression = true);
 
+CompressionOptions GetCompressionOptions(const ImmutableCFOptions& ioptions,
+                                         const VersionStorageInfo* vstorage,
+                                         int level,
+                                         const bool enable_compression = true);
+
 }  // namespace rocksdb

db/compaction_picker_test.cc

@@ -59,7 +59,7 @@ class CompactionPickerTest : public testing::Test {
         vstorage_(nullptr) {
     fifo_options_.max_table_files_size = 1;
     mutable_cf_options_.RefreshDerivedOptions(ioptions_);
-    ioptions_.db_paths.emplace_back("dummy",
+    ioptions_.cf_paths.emplace_back("dummy",
                                     std::numeric_limits<uint64_t>::max());
   }
 
@@ -391,10 +391,10 @@ TEST_F(CompactionPickerTest, NeedsCompactionUniversal) {
   NewVersionStorage(1, kCompactionStyleUniversal);
   UniversalCompactionPicker universal_compaction_picker(
       ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
   // must return false when there's no files.
   ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
             false);
-  UpdateVersionStorageInfo();
 
   // verify the trigger given different number of L0 files.
   for (int i = 1;
@@ -415,6 +415,7 @@ TEST_F(CompactionPickerTest, CompactionUniversalIngestBehindReservedLevel) {
   ioptions_.allow_ingest_behind = true;
   ioptions_.num_levels = 3;
   UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
   // must return false when there's no files.
   ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
             false);
@@ -448,6 +449,7 @@ TEST_F(CompactionPickerTest, CannotTrivialMoveUniversal) {
   mutable_cf_options_.compaction_options_universal.allow_trivial_move = true;
   NewVersionStorage(1, kCompactionStyleUniversal);
   UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
   // must return false when there's no files.
   ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
             false);

db/compaction_picker_universal.cc

@@ -162,7 +162,13 @@ bool UniversalCompactionPicker::IsInputFilesNonOverlapping(Compaction* c) {
 bool UniversalCompactionPicker::NeedsCompaction(
     const VersionStorageInfo* vstorage) const {
   const int kLevel0 = 0;
-  return vstorage->CompactionScore(kLevel0) >= 1;
+  if (vstorage->CompactionScore(kLevel0) >= 1) {
+    return true;
+  }
+  if (!vstorage->FilesMarkedForCompaction().empty()) {
+    return true;
+  }
+  return false;
 }
 
 void UniversalCompactionPicker::SortedRun::Dump(char* out_buf,
@@ -257,8 +263,9 @@ Compaction* UniversalCompactionPicker::PickCompaction(
       CalculateSortedRuns(*vstorage, ioptions_, mutable_cf_options);
 
   if (sorted_runs.size() == 0 ||
-      sorted_runs.size() <
-          (unsigned int)mutable_cf_options.level0_file_num_compaction_trigger) {
+      (vstorage->FilesMarkedForCompaction().empty() &&
+       sorted_runs.size() < (unsigned int)mutable_cf_options
+                                .level0_file_num_compaction_trigger)) {
     ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: nothing to do\n",
                      cf_name.c_str());
     TEST_SYNC_POINT_CALLBACK("UniversalCompactionPicker::PickCompaction:Return",
@@ -272,58 +279,73 @@ Compaction* UniversalCompactionPicker::PickCompaction(
       cf_name.c_str(), sorted_runs.size(), vstorage->LevelSummary(&tmp));
 
   // Check for size amplification first.
-  Compaction* c;
-  if ((c = PickCompactionToReduceSizeAmp(cf_name, mutable_cf_options, vstorage,
-                                         score, sorted_runs, log_buffer)) !=
-      nullptr) {
-    ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: compacting for size amp\n",
-                     cf_name.c_str());
-  } else {
-    // Size amplification is within limits. Try reducing read
-    // amplification while maintaining file size ratios.
-    unsigned int ratio =
-        mutable_cf_options.compaction_options_universal.size_ratio;
-
-    if ((c = PickCompactionToReduceSortedRuns(
-             cf_name, mutable_cf_options, vstorage, score, ratio, UINT_MAX,
-             sorted_runs, log_buffer)) != nullptr) {
-      ROCKS_LOG_BUFFER(log_buffer,
-                       "[%s] Universal: compacting for size ratio\n",
+  Compaction* c = nullptr;
+  if (sorted_runs.size() >=
+      static_cast<size_t>(
+          mutable_cf_options.level0_file_num_compaction_trigger)) {
+    if ((c = PickCompactionToReduceSizeAmp(cf_name, mutable_cf_options,
+                                           vstorage, score, sorted_runs,
+                                           log_buffer)) != nullptr) {
+      ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: compacting for size amp\n",
                        cf_name.c_str());
     } else {
-      // Size amplification and file size ratios are within configured limits.
-      // If max read amplification is exceeding configured limits, then force
-      // compaction without looking at filesize ratios and try to reduce
-      // the number of files to fewer than level0_file_num_compaction_trigger.
-      // This is guaranteed by NeedsCompaction()
-      assert(sorted_runs.size() >=
-             static_cast<size_t>(
-                 mutable_cf_options.level0_file_num_compaction_trigger));
-      // Get the total number of sorted runs that are not being compacted
-      int num_sr_not_compacted = 0;
-      for (size_t i = 0; i < sorted_runs.size(); i++) {
-        if (sorted_runs[i].being_compacted == false) {
-          num_sr_not_compacted++;
-        }
-      }
+      // Size amplification is within limits. Try reducing read
+      // amplification while maintaining file size ratios.
+      unsigned int ratio =
+          mutable_cf_options.compaction_options_universal.size_ratio;
 
-      // The number of sorted runs that are not being compacted is greater than
-      // the maximum allowed number of sorted runs
-      if (num_sr_not_compacted >
-          mutable_cf_options.level0_file_num_compaction_trigger) {
-        unsigned int num_files =
-            num_sr_not_compacted -
-            mutable_cf_options.level0_file_num_compaction_trigger + 1;
-        if ((c = PickCompactionToReduceSortedRuns(
-                 cf_name, mutable_cf_options, vstorage, score, UINT_MAX,
-                 num_files, sorted_runs, log_buffer)) != nullptr) {
-          ROCKS_LOG_BUFFER(log_buffer,
-                           "[%s] Universal: compacting for file num -- %u\n",
-                           cf_name.c_str(), num_files);
+      if ((c = PickCompactionToReduceSortedRuns(
+               cf_name, mutable_cf_options, vstorage, score, ratio, UINT_MAX,
+               sorted_runs, log_buffer)) != nullptr) {
+        ROCKS_LOG_BUFFER(log_buffer,
+                         "[%s] Universal: compacting for size ratio\n",
+                         cf_name.c_str());
+      } else {
+        // Size amplification and file size ratios are within configured limits.
+        // If max read amplification is exceeding configured limits, then force
+        // compaction without looking at filesize ratios and try to reduce
+        // the number of files to fewer than level0_file_num_compaction_trigger.
+        // This is guaranteed by NeedsCompaction()
+        assert(sorted_runs.size() >=
+               static_cast<size_t>(
+                   mutable_cf_options.level0_file_num_compaction_trigger));
+        // Get the total number of sorted runs that are not being compacted
+        int num_sr_not_compacted = 0;
+        for (size_t i = 0; i < sorted_runs.size(); i++) {
+          if (sorted_runs[i].being_compacted == false) {
+            num_sr_not_compacted++;
+          }
+        }
+
+        // The number of sorted runs that are not being compacted is greater
+        // than the maximum allowed number of sorted runs
+        if (num_sr_not_compacted >
+            mutable_cf_options.level0_file_num_compaction_trigger) {
+          unsigned int num_files =
+              num_sr_not_compacted -
+              mutable_cf_options.level0_file_num_compaction_trigger + 1;
+          if ((c = PickCompactionToReduceSortedRuns(
+                   cf_name, mutable_cf_options, vstorage, score, UINT_MAX,
+                   num_files, sorted_runs, log_buffer)) != nullptr) {
+            ROCKS_LOG_BUFFER(log_buffer,
+                             "[%s] Universal: compacting for file num -- %u\n",
+                             cf_name.c_str(), num_files);
+          }
         }
       }
     }
   }
+
+  if (c == nullptr) {
+    if ((c = PickDeleteTriggeredCompaction(cf_name, mutable_cf_options,
+                                           vstorage, score, sorted_runs,
+                                           log_buffer)) != nullptr) {
+      ROCKS_LOG_BUFFER(log_buffer,
+                       "[%s] Universal: delete triggered compaction\n",
+                       cf_name.c_str());
+    }
+  }
+
   if (c == nullptr) {
     TEST_SYNC_POINT_CALLBACK("UniversalCompactionPicker::PickCompaction:Return",
                              nullptr);
@@ -406,9 +428,9 @@ uint32_t UniversalCompactionPicker::GetPathId(
       file_size *
       (100 - mutable_cf_options.compaction_options_universal.size_ratio) / 100;
   uint32_t p = 0;
-  assert(!ioptions.db_paths.empty());
-  for (; p < ioptions.db_paths.size() - 1; p++) {
-    uint64_t target_size = ioptions.db_paths[p].target_size;
+  assert(!ioptions.cf_paths.empty());
+  for (; p < ioptions.cf_paths.size() - 1; p++) {
+    uint64_t target_size = ioptions.cf_paths[p].target_size;
     if (target_size > file_size &&
         accumulated_size + (target_size - file_size) > future_size) {
       return p;
@@ -609,11 +631,15 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
   }
   return new Compaction(
       vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
-      mutable_cf_options.MaxFileSizeForLevel(output_level), LLONG_MAX, path_id,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          kCompactionStyleUniversal),
+      LLONG_MAX, path_id,
       GetCompressionType(ioptions_, vstorage, mutable_cf_options, start_level,
                          1, enable_compression),
-      /* grandparents */ {}, /* is manual */ false, score,
-      false /* deletion_compaction */, compaction_reason);
+      GetCompressionOptions(ioptions_, vstorage, start_level,
+                            enable_compression),
+      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
+      score, false /* deletion_compaction */, compaction_reason);
 }
 
 // Look at overall size amplification. If size amplification
@@ -635,6 +661,10 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
   size_t start_index = 0;
   const SortedRun* sr = nullptr;
 
+  if (sorted_runs.back().being_compacted) {
+    return nullptr;
+  }
+
   // Skip files that are already being compacted
   for (size_t loop = 0; loop < sorted_runs.size() - 1; loop++) {
     sr = &sorted_runs[loop];
@@ -741,15 +771,137 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
   }
 
   return new Compaction(
-      vstorage, ioptions_, mutable_cf_options, std::move(inputs),
-      output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          kCompactionStyleUniversal),
       /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, vstorage, mutable_cf_options,
-                         output_level, 1),
-      /* grandparents */ {}, /* is manual */ false, score,
-      false /* deletion_compaction */,
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         1),
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
+      score, false /* deletion_compaction */,
       CompactionReason::kUniversalSizeAmplification);
 }
+
+// Pick files marked for compaction. Typically, files are marked by
+// CompactOnDeleteCollector due to the presence of tombstones.
+Compaction* UniversalCompactionPicker::PickDeleteTriggeredCompaction(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, double score,
+    const std::vector<SortedRun>& /*sorted_runs*/, LogBuffer* /*log_buffer*/) {
+  CompactionInputFiles start_level_inputs;
+  int output_level;
+  std::vector<CompactionInputFiles> inputs;
+
+  if (vstorage->num_levels() == 1) {
+    // This is single level universal. Since we're basically trying to reclaim
+    // space by processing files marked for compaction due to high tombstone
+    // density, let's do the same thing as compaction to reduce size amp which
+    // has the same goals.
+    bool compact = false;
+
+    start_level_inputs.level = 0;
+    start_level_inputs.files.clear();
+    output_level = 0;
+    for (FileMetaData* f : vstorage->LevelFiles(0)) {
+      if (f->marked_for_compaction) {
+        compact = true;
+      }
+      if (compact) {
+        start_level_inputs.files.push_back(f);
+      }
+    }
+    if (start_level_inputs.size() <= 1) {
+      // If only the last file in L0 is marked for compaction, ignore it
+      return nullptr;
+    }
+    inputs.push_back(start_level_inputs);
+  } else {
+    int start_level;
+
+    // For multi-level universal, the strategy is to make this look more like
+    // leveled. We pick one of the files marked for compaction and compact with
+    // overlapping files in the adjacent level.
+    PickFilesMarkedForCompaction(cf_name, vstorage, &start_level, &output_level,
+                                 &start_level_inputs);
+    if (start_level_inputs.empty()) {
+      return nullptr;
+    }
+
+    // Pick the first non-empty level after the start_level
+    for (output_level = start_level + 1; output_level < vstorage->num_levels();
+         output_level++) {
+      if (vstorage->NumLevelFiles(output_level) != 0) {
+        break;
+      }
+    }
+
+    // If all higher levels are empty, pick the highest level as output level
+    if (output_level == vstorage->num_levels()) {
+      if (start_level == 0) {
+        output_level = vstorage->num_levels() - 1;
+      } else {
+        // If start level is non-zero and all higher levels are empty, this
+        // compaction will translate into a trivial move. Since the idea is
+        // to reclaim space and trivial move doesn't help with that, we
+        // skip compaction in this case and return nullptr
+        return nullptr;
+      }
+    }
+    if (ioptions_.allow_ingest_behind &&
+        output_level == vstorage->num_levels() - 1) {
+      assert(output_level > 1);
+      output_level--;
+    }
+
+    if (output_level != 0) {
+      if (start_level == 0) {
+        if (!GetOverlappingL0Files(vstorage, &start_level_inputs, output_level,
+                                   nullptr)) {
+          return nullptr;
+        }
+      }
+
+      CompactionInputFiles output_level_inputs;
+      int parent_index = -1;
+
+      output_level_inputs.level = output_level;
+      if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage,
+                            &start_level_inputs, &output_level_inputs,
+                            &parent_index, -1)) {
+        return nullptr;
+      }
+      inputs.push_back(start_level_inputs);
+      if (!output_level_inputs.empty()) {
+        inputs.push_back(output_level_inputs);
+      }
+      if (FilesRangeOverlapWithCompaction(inputs, output_level)) {
+        return nullptr;
+      }
+    } else {
+      inputs.push_back(start_level_inputs);
+    }
+  }
+
+  uint64_t estimated_total_size = 0;
+  // Use size of the output level as estimated file size
+  for (FileMetaData* f : vstorage->LevelFiles(output_level)) {
+    estimated_total_size += f->fd.GetFileSize();
+  }
+  uint32_t path_id =
+      GetPathId(ioptions_, mutable_cf_options, estimated_total_size);
+  return new Compaction(
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          kCompactionStyleUniversal),
+      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         1),
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ true,
+      score, false /* deletion_compaction */,
+      CompactionReason::kFilesMarkedForCompaction);
+}
 }  // namespace rocksdb
 
 #endif  // !ROCKSDB_LITE

db/compaction_picker_universal.h

@@ -73,6 +73,11 @@ class UniversalCompactionPicker : public CompactionPicker {
       VersionStorageInfo* vstorage, double score,
       const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer);
 
+  Compaction* PickDeleteTriggeredCompaction(
+      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+      VersionStorageInfo* vstorage, double score,
+      const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer);
+
   // Used in universal compaction when the enabled_trivial_move
   // option is set. Checks whether there are any overlapping files
   // in the input. Returns true if the input files are non

db/comparator_db_test.cc

@@ -2,6 +2,7 @@
 //  This source code is licensed under both the GPLv2 (found in the
 //  COPYING file in the root directory) and Apache 2.0 License
 //  (found in the LICENSE.Apache file in the root directory).
+#include <array>
 #include <map>
 #include <string>
 
@@ -250,7 +251,9 @@ class TwoStrComparator : public Comparator {
 };
 }  // namespace
 
-class ComparatorDBTest : public testing::Test {
+class ComparatorDBTest
+    : public testing::Test,
+      virtual public ::testing::WithParamInterface<uint32_t> {
  private:
   std::string dbname_;
   Env* env_;
@@ -261,7 +264,11 @@ class ComparatorDBTest : public testing::Test {
  public:
   ComparatorDBTest() : env_(Env::Default()), db_(nullptr) {
     comparator = BytewiseComparator();
-    dbname_ = test::TmpDir() + "/comparator_db_test";
+    dbname_ = test::PerThreadDBPath("comparator_db_test");
+    BlockBasedTableOptions toptions;
+    toptions.format_version = GetParam();
+    last_options_.table_factory.reset(
+        rocksdb::NewBlockBasedTableFactory(toptions));
     EXPECT_OK(DestroyDB(dbname_, last_options_));
   }
 
@@ -273,8 +280,12 @@ class ComparatorDBTest : public testing::Test {
 
   DB* GetDB() { return db_; }
 
-  void SetOwnedComparator(const Comparator* cmp) {
-    comparator_guard.reset(cmp);
+  void SetOwnedComparator(const Comparator* cmp, bool owner = true) {
+    if (owner) {
+      comparator_guard.reset(cmp);
+    } else {
+      comparator_guard.reset();
+    }
     comparator = cmp;
     last_options_.comparator = cmp;
   }
@@ -303,7 +314,12 @@ class ComparatorDBTest : public testing::Test {
   }
 };
 
-TEST_F(ComparatorDBTest, Bytewise) {
+INSTANTIATE_TEST_CASE_P(FormatDef, ComparatorDBTest,
+                        testing::Values(test::kDefaultFormatVersion));
+INSTANTIATE_TEST_CASE_P(FormatLatest, ComparatorDBTest,
+                        testing::Values(test::kLatestFormatVersion));
+
+TEST_P(ComparatorDBTest, Bytewise) {
   for (int rand_seed = 301; rand_seed < 306; rand_seed++) {
     DestroyAndReopen();
     Random rnd(rand_seed);
@@ -313,7 +329,7 @@ TEST_F(ComparatorDBTest, Bytewise) {
   }
 }
 
-TEST_F(ComparatorDBTest, SimpleSuffixReverseComparator) {
+TEST_P(ComparatorDBTest, SimpleSuffixReverseComparator) {
   SetOwnedComparator(new test::SimpleSuffixReverseComparator());
 
   for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -339,8 +355,8 @@ TEST_F(ComparatorDBTest, SimpleSuffixReverseComparator) {
   }
 }
 
-TEST_F(ComparatorDBTest, Uint64Comparator) {
-  SetOwnedComparator(test::Uint64Comparator());
+TEST_P(ComparatorDBTest, Uint64Comparator) {
+  SetOwnedComparator(test::Uint64Comparator(), false /* owner */);
 
   for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
     Options* opt = GetOptions();
@@ -363,7 +379,7 @@ TEST_F(ComparatorDBTest, Uint64Comparator) {
   }
 }
 
-TEST_F(ComparatorDBTest, DoubleComparator) {
+TEST_P(ComparatorDBTest, DoubleComparator) {
   SetOwnedComparator(new DoubleComparator());
 
   for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -388,7 +404,7 @@ TEST_F(ComparatorDBTest, DoubleComparator) {
   }
 }
 
-TEST_F(ComparatorDBTest, HashComparator) {
+TEST_P(ComparatorDBTest, HashComparator) {
   SetOwnedComparator(new HashComparator());
 
   for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -407,7 +423,7 @@ TEST_F(ComparatorDBTest, HashComparator) {
   }
 }
 
-TEST_F(ComparatorDBTest, TwoStrComparator) {
+TEST_P(ComparatorDBTest, TwoStrComparator) {
   SetOwnedComparator(new TwoStrComparator());
 
   for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -433,6 +449,209 @@ TEST_F(ComparatorDBTest, TwoStrComparator) {
   }
 }
 
+TEST_P(ComparatorDBTest, IsSameLengthImmediateSuccessor) {
+  {
+    // different length
+    Slice s("abcxy");
+    Slice t("abcxyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxyz");
+    Slice t("abcxy");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    // not last byte different
+    Slice s("abc1xyz");
+    Slice t("abc2xyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    // same string
+    Slice s("abcxyz");
+    Slice t("abcxyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxy");
+    Slice t("abcxz");
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxz");
+    Slice t("abcxy");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xac";
+    const char t_array[] = "\x50\x8a\xad";
+    Slice s(s_array);
+    Slice t(t_array);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff";
+    const char t_array[] = "\x50\x8b\x00";
+    Slice s(s_array, 3);
+    Slice t(t_array, 3);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff\xff";
+    const char t_array[] = "\x50\x8b\x00\x00";
+    Slice s(s_array, 4);
+    Slice t(t_array, 4);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff\xff";
+    const char t_array[] = "\x50\x8b\x00\x01";
+    Slice s(s_array, 4);
+    Slice t(t_array, 4);
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+}
+
+TEST_P(ComparatorDBTest, FindShortestSeparator) {
+  std::string s1 = "abc1xyz";
+  std::string s2 = "abc3xy";
+
+  BytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_EQ("abc2", s1);
+
+  s1 = "abc5xyztt";
+
+  ReverseBytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_EQ("abc5", s1);
+
+  s1 = "abc3";
+  s2 = "abc2xy";
+  ReverseBytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_EQ("abc3", s1);
+
+  s1 = "abc3xyz";
+  s2 = "abc2xy";
+  ReverseBytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_EQ("abc3", s1);
+
+  s1 = "abc3xyz";
+  s2 = "abc2";
+  ReverseBytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_EQ("abc3", s1);
+
+  std::string old_s1 = s1 = "abc2xy";
+  s2 = "abc2";
+  ReverseBytewiseComparator()->FindShortestSeparator(&s1, s2);
+  ASSERT_TRUE(old_s1 >= s1);
+  ASSERT_TRUE(s1 > s2);
+}
+
+TEST_P(ComparatorDBTest, SeparatorSuccessorRandomizeTest) {
+  // Char list for boundary cases.
+  std::array<unsigned char, 6> char_list{{0, 1, 2, 253, 254, 255}};
+  Random rnd(301);
+
+  for (int attempts = 0; attempts < 1000; attempts++) {
+    uint32_t size1 = rnd.Skewed(4);
+    uint32_t size2;
+
+    if (rnd.OneIn(2)) {
+      // size2 to be random size
+      size2 = rnd.Skewed(4);
+    } else {
+      // size1 is within [-2, +2] of size1
+      int diff = static_cast<int>(rnd.Uniform(5)) - 2;
+      int tmp_size2 = static_cast<int>(size1) + diff;
+      if (tmp_size2 < 0) {
+        tmp_size2 = 0;
+      }
+      size2 = static_cast<uint32_t>(tmp_size2);
+    }
+
+    std::string s1;
+    std::string s2;
+    for (uint32_t i = 0; i < size1; i++) {
+      if (rnd.OneIn(2)) {
+        // Use random byte
+        s1 += static_cast<char>(rnd.Uniform(256));
+      } else {
+        // Use one byte in char_list
+        char c = static_cast<char>(char_list[rnd.Uniform(sizeof(char_list))]);
+        s1 += c;
+      }
+    }
+
+    // First set s2 to be the same as s1, and then modify s2.
+    s2 = s1;
+    s2.resize(size2);
+    // We start from the back of the string
+    if (size2 > 0) {
+      uint32_t pos = size2 - 1;
+      do {
+        if (pos >= size1 || rnd.OneIn(4)) {
+          // For 1/4 chance, use random byte
+          s2[pos] = static_cast<char>(rnd.Uniform(256));
+        } else if (rnd.OneIn(4)) {
+          // In 1/4 chance, stop here.
+          break;
+        } else {
+          // Create a char within [-2, +2] of the matching char of s1.
+          int diff = static_cast<int>(rnd.Uniform(5)) - 2;
+          // char may be signed or unsigned based on platform.
+          int s1_char = static_cast<int>(static_cast<unsigned char>(s1[pos]));
+          int s2_char = s1_char + diff;
+          if (s2_char < 0) {
+            s2_char = 0;
+          }
+          if (s2_char > 255) {
+            s2_char = 255;
+          }
+          s2[pos] = static_cast<char>(s2_char);
+        }
+      } while (pos-- != 0);
+    }
+
+    // Test separators
+    for (int rev = 0; rev < 2; rev++) {
+      if (rev == 1) {
+        // switch s1 and s2
+        std::string t = s1;
+        s1 = s2;
+        s2 = t;
+      }
+      std::string separator = s1;
+      BytewiseComparator()->FindShortestSeparator(&separator, s2);
+      std::string rev_separator = s1;
+      ReverseBytewiseComparator()->FindShortestSeparator(&rev_separator, s2);
+
+      if (s1 == s2) {
+        ASSERT_EQ(s1, separator);
+        ASSERT_EQ(s2, rev_separator);
+      } else if (s1 < s2) {
+        ASSERT_TRUE(s1 <= separator);
+        ASSERT_TRUE(s2 > separator);
+        ASSERT_LE(separator.size(), std::max(s1.size(), s2.size()));
+        ASSERT_EQ(s1, rev_separator);
+      } else {
+        ASSERT_TRUE(s1 >= rev_separator);
+        ASSERT_TRUE(s2 < rev_separator);
+        ASSERT_LE(rev_separator.size(), std::max(s1.size(), s2.size()));
+        ASSERT_EQ(s1, separator);
+      }
+    }
+
+    // Test successors
+    std::string succ = s1;
+    BytewiseComparator()->FindShortSuccessor(&succ);
+    ASSERT_TRUE(succ >= s1);
+
+    succ = s1;
+    ReverseBytewiseComparator()->FindShortSuccessor(&succ);
+    ASSERT_TRUE(succ <= s1);
+  }
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/convenience.cc

@@ -49,9 +49,11 @@ Status VerifySstFileChecksum(const Options& options,
   unique_ptr<TableReader> table_reader;
   std::unique_ptr<RandomAccessFileReader> file_reader(
       new RandomAccessFileReader(std::move(file), file_path));
+  const bool kImmortal = true;
   s = ioptions.table_factory->NewTableReader(
-      TableReaderOptions(ioptions, env_options, internal_comparator,
-                         false /* skip_filters */, -1 /* level */),
+      TableReaderOptions(ioptions, options.prefix_extractor.get(), env_options,
+                         internal_comparator, false /* skip_filters */,
+                         !kImmortal, -1 /* level */),
       std::move(file_reader), file_size, &table_reader,
       false /* prefetch_index_and_filter_in_cache */);
   if (!s.ok()) {

db/corruption_test.cc

@@ -48,7 +48,7 @@ class CorruptionTest : public testing::Test {
     tiny_cache_ = NewLRUCache(100, 4);
     options_.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
     options_.env = &env_;
-    dbname_ = test::TmpDir() + "/corruption_test";
+    dbname_ = test::PerThreadDBPath("corruption_test");
     DestroyDB(dbname_, options_);
 
     db_ = nullptr;
@@ -333,9 +333,9 @@ TEST_F(CorruptionTest, TableFileIndexData) {
   Corrupt(kTableFile, -2000, 500);
   Reopen();
   dbi = reinterpret_cast<DBImpl*>(db_);
-  // one full file should be readable, since only one was corrupted
+  // one full file may be readable, since only one was corrupted
   // the other file should be fully non-readable, since index was corrupted
-  Check(5000, 5000);
+  Check(0, 5000);
   ASSERT_NOK(dbi->VerifyChecksum());
 }
 

db/cuckoo_table_db_test.cc

@@ -25,7 +25,7 @@ class CuckooTableDBTest : public testing::Test {
 
  public:
   CuckooTableDBTest() : env_(Env::Default()) {
-    dbname_ = test::TmpDir() + "/cuckoo_table_db_test";
+    dbname_ = test::PerThreadDBPath("cuckoo_table_db_test");
     EXPECT_OK(DestroyDB(dbname_, Options()));
     db_ = nullptr;
     Reopen();

db/db_basic_test.cc

@@ -892,7 +892,7 @@ class TestEnv : public EnvWrapper {
 
 TEST_F(DBBasicTest, DBClose) {
   Options options = GetDefaultOptions();
-  std::string dbname = test::TmpDir(env_) + "/db_close_test";
+  std::string dbname = test::PerThreadDBPath("db_close_test");
   ASSERT_OK(DestroyDB(dbname, options));
 
   DB* db = nullptr;

db/db_block_cache_test.cc

@@ -305,6 +305,41 @@ TEST_F(DBBlockCacheTest, IndexAndFilterBlocksOfNewTableAddedToCache) {
             TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
 }
 
+// With fill_cache = false, fills up the cache, then iterates over the entire
+// db, verify dummy entries inserted in `BlockBasedTable::NewDataBlockIterator`
+// does not cause heap-use-after-free errors in COMPILE_WITH_ASAN=1 runs
+TEST_F(DBBlockCacheTest, FillCacheAndIterateDB) {
+  ReadOptions read_options;
+  read_options.fill_cache = false;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(10, 0, true);
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_OK(Put("key1", "val1"));
+  ASSERT_OK(Put("key2", "val2"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("key3", "val3"));
+  ASSERT_OK(Put("key4", "val4"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("key5", "val5"));
+  ASSERT_OK(Put("key6", "val6"));
+  ASSERT_OK(Flush());
+
+  Iterator* iter = nullptr;
+
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(0));
+  while (iter->Valid()) {
+    iter->Next();
+  }
+  delete iter;
+  iter = nullptr;
+}
+
 TEST_F(DBBlockCacheTest, IndexAndFilterBlocksStats) {
   Options options = CurrentOptions();
   options.create_if_missing = true;
@@ -355,7 +390,10 @@ class MockCache : public LRUCache {
   static uint32_t high_pri_insert_count;
   static uint32_t low_pri_insert_count;
 
-  MockCache() : LRUCache(1 << 25, 0, false, 0.0) {}
+  MockCache()
+      : LRUCache((size_t)1 << 25 /*capacity*/, 0 /*num_shard_bits*/,
+                 false /*strict_capacity_limit*/, 0.0 /*high_pri_pool_ratio*/) {
+  }
 
   virtual Status Insert(const Slice& key, void* value, size_t charge,
                         void (*deleter)(const Slice& key, void* value),

db/db_bloom_filter_test.cc

@@ -39,6 +39,26 @@ class DBBloomFilterTestWithParam
   }
 };
 
+class SliceTransformLimitedDomainGeneric : public SliceTransform {
+  const char* Name() const override {
+    return "SliceTransformLimitedDomainGeneric";
+  }
+
+  Slice Transform(const Slice& src) const override {
+    return Slice(src.data(), 5);
+  }
+
+  bool InDomain(const Slice& src) const override {
+    // prefix will be x????
+    return src.size() >= 5;
+  }
+
+  bool InRange(const Slice& dst) const override {
+    // prefix will be x????
+    return dst.size() == 5;
+  }
+};
+
 // KeyMayExist can lead to a few false positives, but not false negatives.
 // To make test deterministic, use a much larger number of bits per key-20 than
 // bits in the key, so that false positives are eliminated
@@ -117,6 +137,53 @@ TEST_P(DBBloomFilterTestWithParam, KeyMayExist) {
       ChangeOptions(kSkipPlainTable | kSkipHashIndex | kSkipFIFOCompaction));
 }
 
+TEST_F(DBBloomFilterTest, GetFilterByPrefixBloomCustomPrefixExtractor) {
+  for (bool partition_filters : {true, false}) {
+    Options options = last_options_;
+    options.prefix_extractor =
+        std::make_shared<SliceTransformLimitedDomainGeneric>();
+    options.statistics = rocksdb::CreateDBStatistics();
+    BlockBasedTableOptions bbto;
+    bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
+    if (partition_filters) {
+      bbto.partition_filters = true;
+      bbto.index_type = BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
+    }
+    bbto.whole_key_filtering = false;
+    options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+    DestroyAndReopen(options);
+
+    WriteOptions wo;
+    ReadOptions ro;
+    FlushOptions fo;
+    fo.wait = true;
+    std::string value;
+
+    ASSERT_OK(dbfull()->Put(wo, "barbarbar", "foo"));
+    ASSERT_OK(dbfull()->Put(wo, "barbarbar2", "foo2"));
+    ASSERT_OK(dbfull()->Put(wo, "foofoofoo", "bar"));
+
+    dbfull()->Flush(fo);
+
+    ASSERT_EQ("foo", Get("barbarbar"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ("foo2", Get("barbarbar2"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ("NOT_FOUND", Get("barbarbar3"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+
+    ASSERT_EQ("NOT_FOUND", Get("barfoofoo"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 1);
+
+    ASSERT_EQ("NOT_FOUND", Get("foobarbar"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+
+    ro.total_order_seek = true;
+    ASSERT_TRUE(db_->Get(ro, "foobarbar", &value).IsNotFound());
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+  }
+}
+
 TEST_F(DBBloomFilterTest, GetFilterByPrefixBloom) {
   for (bool partition_filters : {true, false}) {
     Options options = last_options_;
@@ -1057,10 +1124,10 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   CompactRangeOptions compact_options;
@@ -1097,6 +1164,413 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
             TestGetTickerCount(options, BLOCK_CACHE_ADD));
 }
 
+int CountIter(std::unique_ptr<Iterator>& iter, const Slice& key) {
+  int count = 0;
+  for (iter->Seek(key); iter->Valid() && iter->status() == Status::OK();
+       iter->Next()) {
+    count++;
+  }
+  return count;
+}
+
+// use iterate_upper_bound to hint compatiability of existing bloom filters.
+// The BF is considered compatible if 1) upper bound and seek key transform
+// into the same string, or 2) the transformed seek key is of the same length
+// as the upper bound and two keys are adjacent according to the comparator.
+TEST_F(DBBloomFilterTest, DynamicBloomFilterUpperBound) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewCappedPrefixTransform(4));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    ASSERT_OK(Put("abcdxxx0", "val1"));
+    ASSERT_OK(Put("abcdxxx1", "val2"));
+    ASSERT_OK(Put("abcdxxx2", "val3"));
+    ASSERT_OK(Put("abcdxxx3", "val4"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // prefix_extractor has not changed, BF will always be read
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcd0000"), 4);
+    }
+    {
+      Slice upper_bound("abcdzzzz");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcd0000"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:5"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.FixedPrefix.5"));
+    {
+      // BF changed, [abcdxx00, abce) is a valid bound, will trigger BF read
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx00"), 4);
+      // should check bloom filter since upper bound meets requirement
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [abcdxx01, abcey) is not valid bound since upper bound is too long for
+      // the BF in SST (capped:4)
+      Slice upper_bound("abcey");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx01"), 4);
+      // should skip bloom filter since upper bound is too long
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [abcdxx02, abcdy) is a valid bound since the prefix is the same
+      Slice upper_bound("abcdy");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx02"), 4);
+      // should check bloom filter since upper bound matches transformed seek
+      // key
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [aaaaaaaa, abce) is not a valid bound since 1) they don't share the
+      // same prefix, 2) the prefixes are not consecutive
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "aaaaaaaa"), 0);
+      // should skip bloom filter since mismatch is found
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:3"}}));
+    {
+      // [abc, abd) is not a valid bound since the upper bound is too short
+      // for BF (capped:4)
+      Slice upper_bound("abd");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abc"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:4"}}));
+    {
+      // set back to capped:4 and verify BF is always read
+      Slice upper_bound("abd");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abc"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                3 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+    }
+    iteration++;
+  }
+}
+
+// Create multiple SST files each with a different prefix_extractor config,
+// verify iterators can read all SST files using the latest config.
+TEST_F(DBBloomFilterTest, DynamicBloomFilterMultipleSST) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    table_options.cache_index_and_filter_blocks = true;
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    Slice upper_bound("foz90000");
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+
+    // first SST with fixed:1 BF
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_OK(Put("foq1", "bar1"));
+    ASSERT_OK(Put("fpa", "0"));
+    dbfull()->Flush(FlushOptions());
+    std::unique_ptr<Iterator> iter_old(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter_old, "foo"), 4);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 1);
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    read_options.iterate_upper_bound = &upper_bound;
+    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter, "foo"), 2);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              1 + iteration);
+    ASSERT_EQ(CountIter(iter, "gpk"), 0);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              1 + iteration);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+
+    // second SST with capped:3 BF
+    ASSERT_OK(Put("foo3", "bar3"));
+    ASSERT_OK(Put("foo4", "bar4"));
+    ASSERT_OK(Put("foq5", "bar5"));
+    ASSERT_OK(Put("fpb", "1"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // BF is cappped:3 now
+      std::unique_ptr<Iterator> iter_tmp(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_tmp, "foo"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+      ASSERT_EQ(CountIter(iter_tmp, "gpk"), 0);
+      // both counters are incremented because BF is "not changed" for 1 of the
+      // 2 SST files, so filter is checked once and found no match.
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                3 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+    }
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:2"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.FixedPrefix.2"));
+    // third SST with fixed:2 BF
+    ASSERT_OK(Put("foo6", "bar6"));
+    ASSERT_OK(Put("foo7", "bar7"));
+    ASSERT_OK(Put("foq8", "bar8"));
+    ASSERT_OK(Put("fpc", "2"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // BF is fixed:2 now
+      std::unique_ptr<Iterator> iter_tmp(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_tmp, "foo"), 9);
+      // the first and last BF are checked
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                4 + iteration * 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+      ASSERT_EQ(CountIter(iter_tmp, "gpk"), 0);
+      // only last BF is checked and not found
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                5 + iteration * 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 2);
+    }
+
+    // iter_old can only see the first SST, so checked plus 1
+    ASSERT_EQ(CountIter(iter_old, "foo"), 4);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              6 + iteration * 3);
+    // iter was created after the first setoptions call so only full filter
+    // will check the filter
+    ASSERT_EQ(CountIter(iter, "foo"), 2);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              6 + iteration * 4);
+
+    {
+      // keys in all three SSTs are visible to iterator
+      // The range of [foo, foz90000] is compatible with (fixed:1) and (fixed:2)
+      // so +2 for checked counter
+      std::unique_ptr<Iterator> iter_all(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_all, "foo"), 9);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                7 + iteration * 5);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 2);
+      ASSERT_EQ(CountIter(iter_all, "gpk"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                8 + iteration * 5);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    {
+      std::unique_ptr<Iterator> iter_all(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_all, "foo"), 6);
+      // all three SST are checked because the current options has the same as
+      // the remaining SST (capped:3)
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                9 + iteration * 7);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+      ASSERT_EQ(CountIter(iter_all, "gpk"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                10 + iteration * 7);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 4);
+    }
+    // TODO(Zhongyi): Maybe also need to add Get calls to test point look up?
+    iteration++;
+  }
+}
+
+// Create a new column family in a running DB, change prefix_extractor
+// dynamically, verify the iterator created on the new column family behaves
+// as expected
+TEST_F(DBBloomFilterTest, DynamicBloomFilterNewColumnFamily) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    CreateAndReopenWithCF({"pikachu" + std::to_string(iteration)}, options);
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+    // create a new CF and set prefix_extractor dynamically
+    options.prefix_extractor.reset(NewCappedPrefixTransform(3));
+    CreateColumnFamilies({"ramen_dojo_" + std::to_string(iteration)}, options);
+    ASSERT_EQ(0,
+              strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
+                     "rocksdb.CappedPrefix.3"));
+    ASSERT_OK(Put(2, "foo3", "bar3"));
+    ASSERT_OK(Put(2, "foo4", "bar4"));
+    ASSERT_OK(Put(2, "foo5", "bar5"));
+    ASSERT_OK(Put(2, "foq6", "bar6"));
+    ASSERT_OK(Put(2, "fpq7", "bar7"));
+    dbfull()->Flush(FlushOptions());
+    {
+      std::unique_ptr<Iterator> iter(
+          db_->NewIterator(read_options, handles_[2]));
+      ASSERT_EQ(CountIter(iter, "foo"), 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(
+        dbfull()->SetOptions(handles_[2], {{"prefix_extractor", "fixed:2"}}));
+    ASSERT_EQ(0,
+              strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
+                     "rocksdb.FixedPrefix.2"));
+    {
+      std::unique_ptr<Iterator> iter(
+          db_->NewIterator(read_options, handles_[2]));
+      ASSERT_EQ(CountIter(iter, "foo"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[2]));
+    dbfull()->DestroyColumnFamilyHandle(handles_[2]);
+    handles_[2] = nullptr;
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+    dbfull()->DestroyColumnFamilyHandle(handles_[1]);
+    handles_[1] = nullptr;
+    iteration++;
+  }
+}
+
+// Verify it's possible to change prefix_extractor at runtime and iterators
+// behaves as expected
+TEST_F(DBBloomFilterTest, DynamicBloomFilterOptions) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_OK(Put("foo1", "bar1"));
+    ASSERT_OK(Put("fpa", "0"));
+    dbfull()->Flush(FlushOptions());
+    ASSERT_OK(Put("foo3", "bar3"));
+    ASSERT_OK(Put("foo4", "bar4"));
+    ASSERT_OK(Put("foo5", "bar5"));
+    ASSERT_OK(Put("fpb", "1"));
+    dbfull()->Flush(FlushOptions());
+    ASSERT_OK(Put("foo6", "bar6"));
+    ASSERT_OK(Put("foo7", "bar7"));
+    ASSERT_OK(Put("foo8", "bar8"));
+    ASSERT_OK(Put("fpc", "2"));
+    dbfull()->Flush(FlushOptions());
+
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+    {
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "foo"), 12);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    std::unique_ptr<Iterator> iter_old(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter_old, "foo"), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 6);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    {
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      // "fp*" should be skipped
+      ASSERT_EQ(CountIter(iter, "foo"), 9);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 6);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+
+    // iterator created before should not be affected and see all keys
+    ASSERT_EQ(CountIter(iter_old, "foo"), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 9);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    ASSERT_EQ(CountIter(iter_old, "abc"), 0);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+    iteration++;
+  }
+}
+
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

db/db_compaction_filter_test.cc

@@ -45,6 +45,7 @@ class DBTestCompactionFilterWithCompactParam
   }
 };
 
+#ifndef ROCKSDB_VALGRIND_RUN
 INSTANTIATE_TEST_CASE_P(
     DBTestCompactionFilterWithCompactOption,
     DBTestCompactionFilterWithCompactParam,
@@ -53,6 +54,12 @@ INSTANTIATE_TEST_CASE_P(
                       DBTestBase::OptionConfig::kUniversalCompactionMultiLevel,
                       DBTestBase::OptionConfig::kLevelSubcompactions,
                       DBTestBase::OptionConfig::kUniversalSubcompactions));
+#else
+// Run fewer cases in valgrind
+INSTANTIATE_TEST_CASE_P(DBTestCompactionFilterWithCompactOption,
+                        DBTestCompactionFilterWithCompactParam,
+                        ::testing::Values(DBTestBase::OptionConfig::kDefault));
+#endif  // ROCKSDB_VALGRIND_RUN
 
 class KeepFilter : public CompactionFilter {
  public:

db/db_compaction_test.cc

@@ -74,6 +74,48 @@ class FlushedFileCollector : public EventListener {
   std::mutex mutex_;
 };
 
+class CompactionStatsCollector : public EventListener {
+public:
+  CompactionStatsCollector()
+      : compaction_completed_(static_cast<int>(CompactionReason::kNumOfReasons)) {
+    for (auto& v : compaction_completed_) {
+      v.store(0);
+    }
+  }
+
+  ~CompactionStatsCollector() {}
+
+  virtual void OnCompactionCompleted(DB* /* db */,
+      const CompactionJobInfo& info) override {
+    int k = static_cast<int>(info.compaction_reason);
+    int num_of_reasons = static_cast<int>(CompactionReason::kNumOfReasons);
+    assert(k >= 0 && k < num_of_reasons);
+    compaction_completed_[k]++;
+  }
+
+  virtual void OnExternalFileIngested(DB* /* db */,
+      const ExternalFileIngestionInfo& /* info */) override {
+    int k = static_cast<int>(CompactionReason::kExternalSstIngestion);
+    compaction_completed_[k]++;
+  }
+
+  virtual void OnFlushCompleted(DB* /* db */,
+      const FlushJobInfo& /* info */) override {
+    int k = static_cast<int>(CompactionReason::kFlush);
+    compaction_completed_[k]++;
+  }
+
+  int NumberOfCompactions(CompactionReason reason) const {
+    int num_of_reasons = static_cast<int>(CompactionReason::kNumOfReasons);
+    int k = static_cast<int>(reason);
+    assert(k >= 0 && k < num_of_reasons);
+    return compaction_completed_.at(k).load();
+  }
+
+private:
+  std::vector<std::atomic<int>> compaction_completed_;
+};
+
 static const int kCDTValueSize = 1000;
 static const int kCDTKeysPerBuffer = 4;
 static const int kCDTNumLevels = 8;
@@ -154,6 +196,40 @@ void VerifyCompactionResult(
 #endif
 }
 
+/*
+ * Verifies compaction stats of cfd are valid.
+ *
+ * For each level of cfd, its compaction stats are valid if
+ * 1) sum(stat.counts) == stat.count, and
+ * 2) stat.counts[i] == collector.NumberOfCompactions(i)
+ */
+void VerifyCompactionStats(ColumnFamilyData& cfd,
+    const CompactionStatsCollector& collector) {
+#ifndef NDEBUG
+  InternalStats* internal_stats_ptr = cfd.internal_stats();
+  ASSERT_TRUE(internal_stats_ptr != nullptr);
+  const std::vector<InternalStats::CompactionStats>& comp_stats =
+      internal_stats_ptr->TEST_GetCompactionStats();
+  const int num_of_reasons = static_cast<int>(CompactionReason::kNumOfReasons);
+  std::vector<int> counts(num_of_reasons, 0);
+  // Count the number of compactions caused by each CompactionReason across
+  // all levels.
+  for (const auto& stat : comp_stats) {
+    int sum = 0;
+    for (int i = 0; i < num_of_reasons; i++) {
+      counts[i] += stat.counts[i];
+      sum += stat.counts[i];
+    }
+    ASSERT_EQ(sum, stat.count);
+  }
+  // Verify InternalStats bookkeeping matches that of CompactionStatsCollector,
+  // assuming that all compactions complete.
+  for (int i = 0; i < num_of_reasons; i++) {
+    ASSERT_EQ(collector.NumberOfCompactions(static_cast<CompactionReason>(i)), counts[i]);
+  }
+#endif /* NDEBUG */
+}
+
 const SstFileMetaData* PickFileRandomly(
     const ColumnFamilyMetaData& cf_meta,
     Random* rand,
@@ -175,6 +251,7 @@ const SstFileMetaData* PickFileRandomly(
 }
 }  // anonymous namespace
 
+#ifndef ROCKSDB_VALGRIND_RUN
 // All the TEST_P tests run once with sub_compactions disabled (i.e.
 // options.max_subcompactions = 1) and once with it enabled
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
@@ -217,6 +294,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
     ASSERT_GT(db_size[0] / 3, db_size[1]);
   }
 }
+#endif  // ROCKSDB_VALGRIND_RUN
 
 TEST_P(DBCompactionTestWithParam, CompactionsPreserveDeletes) {
   //  For each options type we test following
@@ -360,7 +438,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
       });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "TableCache::GetTableReader:0",
-      [&](void* arg) { num_new_table_reader++; });
+      [&](void* /*arg*/) { num_new_table_reader++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   for (int k = 0; k < options.level0_file_num_compaction_trigger; ++k) {
@@ -916,7 +994,7 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
   int32_t trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -973,10 +1051,10 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -1072,10 +1150,10 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveTargetLevel) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -1131,10 +1209,10 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   bool first = true;
   // Purpose of dependencies:
   // 4 -> 1: ensure the order of two non-trivial compactions
@@ -1145,7 +1223,7 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
        {"DBCompaction::ManualPartial:5", "DBCompaction::ManualPartial:2"},
        {"DBCompaction::ManualPartial:5", "DBCompaction::ManualPartial:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (first) {
           first = false;
           TEST_SYNC_POINT("DBCompaction::ManualPartial:4");
@@ -1276,17 +1354,17 @@ TEST_F(DBCompactionTest, DISABLED_ManualPartialFill) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   bool first = true;
   bool second = true;
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
       {{"DBCompaction::PartialFill:4", "DBCompaction::PartialFill:1"},
        {"DBCompaction::PartialFill:2", "DBCompaction::PartialFill:3"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* /*arg*/) {
         if (first) {
           TEST_SYNC_POINT("DBCompaction::PartialFill:4");
           first = false;
@@ -1692,10 +1770,10 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -1983,6 +2061,125 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionPathUse) {
   Destroy(options);
 }
 
+TEST_P(DBCompactionTestWithParam, LevelCompactionCFPathUse) {
+  Options options = CurrentOptions();
+  options.db_paths.emplace_back(dbname_, 500 * 1024);
+  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
+  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
+  options.memtable_factory.reset(
+    new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+  options.compaction_style = kCompactionStyleLevel;
+  options.write_buffer_size = 110 << 10;  // 110KB
+  options.arena_block_size = 4 << 10;
+  options.level0_file_num_compaction_trigger = 2;
+  options.num_levels = 4;
+  options.max_bytes_for_level_base = 400 * 1024;
+  options.max_subcompactions = max_subcompactions_;
+
+  std::vector<Options> option_vector;
+  option_vector.emplace_back(options);
+  ColumnFamilyOptions cf_opt1(options), cf_opt2(options);
+  // Configure CF1 specific paths.
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1", 500 * 1024);
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_2", 4 * 1024 * 1024);
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_3", 1024 * 1024 * 1024);
+  option_vector.emplace_back(DBOptions(options), cf_opt1);
+  CreateColumnFamilies({"one"},option_vector[1]);
+
+  // Configura CF2 specific paths.
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2", 500 * 1024);
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_2", 4 * 1024 * 1024);
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_3", 1024 * 1024 * 1024);
+  option_vector.emplace_back(DBOptions(options), cf_opt2);
+  CreateColumnFamilies({"two"},option_vector[2]);
+
+  ReopenWithColumnFamilies({"default", "one", "two"}, option_vector);
+
+  Random rnd(301);
+  int key_idx = 0;
+  int key_idx1 = 0;
+  int key_idx2 = 0;
+
+  auto generate_file = [&]() {
+    GenerateNewFile(0, &rnd, &key_idx);
+    GenerateNewFile(1, &rnd, &key_idx1);
+    GenerateNewFile(2, &rnd, &key_idx2);
+  };
+
+  auto check_sstfilecount = [&](int path_id, int expected) {
+    ASSERT_EQ(expected, GetSstFileCount(options.db_paths[path_id].path));
+    ASSERT_EQ(expected, GetSstFileCount(cf_opt1.cf_paths[path_id].path));
+    ASSERT_EQ(expected, GetSstFileCount(cf_opt2.cf_paths[path_id].path));
+  };
+
+  auto check_filesperlevel = [&](const std::string& expected) {
+    ASSERT_EQ(expected, FilesPerLevel(0));
+    ASSERT_EQ(expected, FilesPerLevel(1));
+    ASSERT_EQ(expected, FilesPerLevel(2));
+  };
+
+  auto check_getvalues = [&]() {
+    for (int i = 0; i < key_idx; i++) {
+      auto v = Get(0, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+
+    for (int i = 0; i < key_idx1; i++) {
+      auto v = Get(1, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+
+    for (int i = 0; i < key_idx2; i++) {
+      auto v = Get(2, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+  };
+
+  // Check that default column family uses db_paths.
+  // And Column family "one" uses cf_paths.
+
+  // First three 110KB files are not going to second path.
+  // After that, (100K, 200K)
+  for (int num = 0; num < 3; num++) {
+    generate_file();
+  }
+
+  // Another 110KB triggers a compaction to 400K file to fill up first path
+  generate_file();
+  check_sstfilecount(1, 3);
+
+  // (1, 4)
+  generate_file();
+  check_filesperlevel("1,4");
+  check_sstfilecount(1, 4);
+  check_sstfilecount(0, 1);
+
+  // (1, 4, 1)
+  generate_file();
+  check_filesperlevel("1,4,1");
+  check_sstfilecount(2, 1);
+  check_sstfilecount(1, 4);
+  check_sstfilecount(0, 1);
+
+  // (1, 4, 2)
+  generate_file();
+  check_filesperlevel("1,4,2");
+  check_sstfilecount(2, 2);
+  check_sstfilecount(1, 4);
+  check_sstfilecount(0, 1);
+
+  check_getvalues();
+
+  ReopenWithColumnFamilies({"default", "one", "two"}, option_vector);
+
+  check_getvalues();
+
+  Destroy(options, true);
+}
+
 TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
   Random rnd(301);
   int max_key_level_insert = 200;
@@ -2573,16 +2770,16 @@ TEST_P(DBCompactionTestWithParam, CompressLevelCompaction) {
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "Compaction::InputCompressionMatchesOutput:Matches",
-      [&](void* arg) { matches++; });
+      [&](void* /*arg*/) { matches++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "Compaction::InputCompressionMatchesOutput:DidntMatch",
-      [&](void* arg) { didnt_match++; });
+      [&](void* /*arg*/) { didnt_match++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial++; });
+      [&](void* /*arg*/) { non_trivial++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Reopen(options);
@@ -2744,10 +2941,10 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -2933,6 +3130,48 @@ TEST_P(DBCompactionTestWithParam, IntraL0CompactionDoesNotObsoleteDeletions) {
   ASSERT_TRUE(db_->Get(roptions, Key(0), &result).IsNotFound());
 }
 
+TEST_P(DBCompactionTestWithParam, FullCompactionInBottomPriThreadPool) {
+  const int kNumFilesTrigger = 3;
+  Env::Default()->SetBackgroundThreads(1, Env::Priority::BOTTOM);
+  for (bool use_universal_compaction : {false, true}) {
+    Options options = CurrentOptions();
+    if (use_universal_compaction) {
+      options.compaction_style = kCompactionStyleUniversal;
+    } else {
+      options.compaction_style = kCompactionStyleLevel;
+      options.level_compaction_dynamic_level_bytes = true;
+    }
+    options.num_levels = 4;
+    options.write_buffer_size = 100 << 10;     // 100KB
+    options.target_file_size_base = 32 << 10;  // 32KB
+    options.level0_file_num_compaction_trigger = kNumFilesTrigger;
+    // Trigger compaction if size amplification exceeds 110%
+    options.compaction_options_universal.max_size_amplification_percent = 110;
+    DestroyAndReopen(options);
+
+    int num_bottom_pri_compactions = 0;
+    SyncPoint::GetInstance()->SetCallBack(
+        "DBImpl::BGWorkBottomCompaction",
+        [&](void* /*arg*/) { ++num_bottom_pri_compactions; });
+    SyncPoint::GetInstance()->EnableProcessing();
+
+    Random rnd(301);
+    for (int num = 0; num < kNumFilesTrigger; num++) {
+      ASSERT_EQ(NumSortedRuns(), num);
+      int key_idx = 0;
+      GenerateNewFile(&rnd, &key_idx);
+    }
+    dbfull()->TEST_WaitForCompact();
+
+    ASSERT_EQ(1, num_bottom_pri_compactions);
+
+    // Verify that size amplification did occur
+    ASSERT_EQ(NumSortedRuns(), 1);
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  }
+  Env::Default()->SetBackgroundThreads(0, Env::Priority::BOTTOM);
+}
+
 TEST_F(DBCompactionTest, OptimizedDeletionObsoleting) {
   // Deletions can be dropped when compacted to non-last level if they fall
   // outside the lower-level files' key-ranges.
@@ -3091,6 +3330,13 @@ TEST_F(DBCompactionTest, CompactBottomLevelFilesWithDeletions) {
   // compactions should be triggered, which reduce the size of each bottom-level
   // file without changing file count.
   db_->ReleaseSnapshot(snapshot);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        Compaction* compaction = reinterpret_cast<Compaction*>(arg);
+        ASSERT_TRUE(compaction->compaction_reason() ==
+                    CompactionReason::kBottommostFiles);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
   dbfull()->TEST_WaitForCompact();
   db_->GetLiveFilesMetaData(&post_release_metadata);
   ASSERT_EQ(pre_release_metadata.size(), post_release_metadata.size());
@@ -3104,6 +3350,66 @@ TEST_F(DBCompactionTest, CompactBottomLevelFilesWithDeletions) {
     // deletion markers/deleted keys.
     ASSERT_LT(post_file.size, pre_file.size);
   }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBCompactionTest, LevelCompactExpiredTtlFiles) {
+  const int kNumKeysPerFile = 32;
+  const int kNumLevelFiles = 2;
+  const int kValueSize = 1024;
+
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.ttl = 24 * 60 * 60;  // 24 hours
+  options.max_open_files = -1;
+  env_->time_elapse_only_sleep_ = false;
+  options.env = env_;
+
+  env_->addon_time_.store(0);
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+  for (int i = 0; i < kNumLevelFiles; ++i) {
+    for (int j = 0; j < kNumKeysPerFile; ++j) {
+      ASSERT_OK(
+          Put(Key(i * kNumKeysPerFile + j), RandomString(&rnd, kValueSize)));
+    }
+    Flush();
+  }
+  Flush();
+  dbfull()->TEST_WaitForCompact();
+  MoveFilesToLevel(3);
+  ASSERT_EQ("0,0,0,2", FilesPerLevel());
+
+  for (int i = 0; i < kNumLevelFiles; ++i) {
+    for (int j = 0; j < kNumKeysPerFile; ++j) {
+      // Overwrite previous keys with smaller, but predictable, values.
+      ASSERT_OK(Delete(Key(i * kNumKeysPerFile + j)));
+    }
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ("2,0,0,2", FilesPerLevel());
+  MoveFilesToLevel(1);
+  ASSERT_EQ("0,2,0,2", FilesPerLevel());
+
+  env_->addon_time_.fetch_add(36 * 60 * 60);  // 36 hours
+  ASSERT_EQ("0,2,0,2", FilesPerLevel());
+
+  // Just do a siimple write + flush so that the Ttl expired files get
+  // compacted.
+  ASSERT_OK(Put("a", "1"));
+  Flush();
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        Compaction* compaction = reinterpret_cast<Compaction*>(arg);
+        ASSERT_TRUE(compaction->compaction_reason() == CompactionReason::kTtl);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  dbfull()->TEST_WaitForCompact();
+  // All non-L0 files are deleted, as they contained only deleted data.
+  ASSERT_EQ("1", FilesPerLevel());
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
 
 TEST_F(DBCompactionTest, CompactRangeDelayedByL0FileCount) {
@@ -3376,6 +3682,90 @@ TEST_F(DBCompactionTest, CompactRangeFlushOverlappingMemtable) {
   }
 }
 
+TEST_F(DBCompactionTest, CompactionStatsTest) {
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = 2;
+  CompactionStatsCollector* collector = new CompactionStatsCollector();
+  options.listeners.emplace_back(collector);
+  DestroyAndReopen(options);
+
+  for (int i = 0; i < 32; i++) {
+    for (int j = 0; j < 5000; j++) {
+      Put(std::to_string(j), std::string(1, 'A'));
+    }
+    ASSERT_OK(Flush());
+    ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+  }
+  dbfull()->TEST_WaitForCompact();
+  ColumnFamilyHandleImpl* cfh =
+      static_cast<ColumnFamilyHandleImpl*>(dbfull()->DefaultColumnFamily());
+  ColumnFamilyData* cfd = cfh->cfd();
+
+  VerifyCompactionStats(*cfd, *collector);
+}
+
+TEST_F(DBCompactionTest, CompactFilesOutputRangeConflict) {
+  // LSM setup:
+  // L1:      [ba bz]
+  // L2: [a b]       [c d]
+  // L3: [a b]       [c d]
+  //
+  // Thread 1:                        Thread 2:
+  // Begin compacting all L2->L3
+  //                                  Compact [ba bz] L1->L3
+  // End compacting all L2->L3
+  //
+  // The compaction operation in thread 2 should be disallowed because the range
+  // overlaps with the compaction in thread 1, which also covers that range in
+  // L3.
+  Options options = CurrentOptions();
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+  Reopen(options);
+
+  for (int level = 3; level >= 2; --level) {
+    ASSERT_OK(Put("a", "val"));
+    ASSERT_OK(Put("b", "val"));
+    ASSERT_OK(Flush());
+    ASSERT_OK(Put("c", "val"));
+    ASSERT_OK(Put("d", "val"));
+    ASSERT_OK(Flush());
+    MoveFilesToLevel(level);
+  }
+  ASSERT_OK(Put("ba", "val"));
+  ASSERT_OK(Put("bz", "val"));
+  ASSERT_OK(Flush());
+  MoveFilesToLevel(1);
+
+  SyncPoint::GetInstance()->LoadDependency({
+      {"CompactFilesImpl:0",
+       "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2Begin"},
+      {"DBCompactionTest::CompactFilesOutputRangeConflict:Thread2End",
+       "CompactFilesImpl:1"},
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  auto bg_thread = port::Thread([&]() {
+    // Thread 1
+    std::vector<std::string> filenames = collector->GetFlushedFiles();
+    filenames.pop_back();
+    ASSERT_OK(db_->CompactFiles(CompactionOptions(), filenames,
+                                3 /* output_level */));
+  });
+
+  // Thread 2
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2Begin");
+  std::string filename = collector->GetFlushedFiles().back();
+  ASSERT_FALSE(
+      db_->CompactFiles(CompactionOptions(), {filename}, 3 /* output_level */)
+          .ok());
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2End");
+
+  bg_thread.join();
+}
+
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                         ::testing::Values(std::make_tuple(1, true),
                                           std::make_tuple(1, false),
@@ -3395,7 +3785,7 @@ TEST_P(DBCompactionDirectIOTest, DirectIO) {
       "TableCache::NewIterator:for_compaction", [&](void* arg) {
         bool* use_direct_reads = static_cast<bool*>(arg);
         ASSERT_EQ(*use_direct_reads,
-                  options.use_direct_io_for_flush_and_compaction);
+                  options.use_direct_reads);
       });
   SyncPoint::GetInstance()->SetCallBack(
       "CompactionJob::OpenCompactionOutputFile", [&](void* arg) {
@@ -3405,7 +3795,7 @@ TEST_P(DBCompactionDirectIOTest, DirectIO) {
       });
   if (options.use_direct_io_for_flush_and_compaction) {
     SyncPoint::GetInstance()->SetCallBack(
-        "SanitizeOptions:direct_io", [&](void* arg) {
+        "SanitizeOptions:direct_io", [&](void* /*arg*/) {
           readahead = true;
         });
   }
@@ -3414,7 +3804,7 @@ TEST_P(DBCompactionDirectIOTest, DirectIO) {
   MakeTables(3, "p", "q", 1);
   ASSERT_EQ("1,1,1", FilesPerLevel(1));
   Compact(1, "p1", "p9");
-  ASSERT_FALSE(readahead ^ options.use_direct_io_for_flush_and_compaction);
+  ASSERT_EQ(readahead, options.use_direct_reads);
   ASSERT_EQ("0,0,1", FilesPerLevel(1));
   Destroy(options);
   delete options.env;

db/db_dynamic_level_test.cc

@@ -194,7 +194,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   // Hold compaction jobs to make sure
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "CompactionJob::Run():Start",
-      [&](void* arg) { env_->SleepForMicroseconds(100000); });
+      [&](void* /*arg*/) { env_->SleepForMicroseconds(100000); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
@@ -378,7 +378,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBaseInc) {
   int non_trivial = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial++; });
+      [&](void* /*arg*/) { non_trivial++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Random rnd(301);

db/db_filesnapshot.cc

@@ -74,7 +74,7 @@ Status DBImpl::EnableFileDeletions(bool force) {
 }
 
 int DBImpl::IsFileDeletionsEnabled() const {
-  return disable_delete_obsolete_files_;
+  return !disable_delete_obsolete_files_;
 }
 
 Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,

db/db_flush_test.cc

@@ -72,19 +72,23 @@ TEST_F(DBFlushTest, SyncFail) {
   auto* cfd =
       reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
           ->cfd();
-  int refs_before = cfd->current()->TEST_refs();
   FlushOptions flush_options;
   flush_options.wait = false;
   ASSERT_OK(dbfull()->Flush(flush_options));
+  // Flush installs a new super-version. Get the ref count after that.
+  auto current_before = cfd->current();
+  int refs_before = cfd->current()->TEST_refs();
   fault_injection_env->SetFilesystemActive(false);
   TEST_SYNC_POINT("DBFlushTest::SyncFail:1");
   TEST_SYNC_POINT("DBFlushTest::SyncFail:2");
   fault_injection_env->SetFilesystemActive(true);
+  // Now the background job will do the flush; wait for it.
   dbfull()->TEST_WaitForFlushMemTable();
 #ifndef ROCKSDB_LITE
   ASSERT_EQ("", FilesPerLevel());  // flush failed.
 #endif                             // ROCKSDB_LITE
-  // Flush job should release ref count to current version.
+  // Backgroun flush job should release ref count to current version.
+  ASSERT_EQ(current_before, cfd->current());
   ASSERT_EQ(refs_before, cfd->current()->TEST_refs());
   Destroy(options);
 }
@@ -101,7 +105,7 @@ TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
   std::thread::id tid;
   int num_flushes = 0, num_compactions = 0;
   SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BGWorkFlush", [&](void* arg) {
+      "DBImpl::BGWorkFlush", [&](void* /*arg*/) {
         if (tid == std::thread::id()) {
           tid = std::this_thread::get_id();
         } else {
@@ -110,7 +114,7 @@ TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
         ++num_flushes;
       });
   SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BGWorkCompaction", [&](void* arg) {
+      "DBImpl::BGWorkCompaction", [&](void* /*arg*/) {
         ASSERT_EQ(tid, std::this_thread::get_id());
         ++num_compactions;
       });

db/db_impl.cc

@@ -32,6 +32,7 @@
 #include "db/db_info_dumper.h"
 #include "db/db_iter.h"
 #include "db/dbformat.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "db/external_sst_file_ingestion_job.h"
 #include "db/flush_job.h"
@@ -97,7 +98,7 @@
 
 namespace rocksdb {
 const std::string kDefaultColumnFamilyName("default");
-void DumpRocksDBBuildVersion(Logger * log);
+void DumpRocksDBBuildVersion(Logger* log);
 
 CompressionType GetCompressionFlush(
     const ImmutableCFOptions& ioptions,
@@ -135,10 +136,10 @@ void DumpSupportInfo(Logger* logger) {
 }
 
 int64_t kDefaultLowPriThrottledRate = 2 * 1024 * 1024;
-} // namespace
+}  // namespace
 
 DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
-               const bool seq_per_batch)
+               const bool seq_per_batch, const bool batch_per_txn)
     : env_(options.env),
       dbname_(dbname),
       own_info_log_(options.info_log == nullptr),
@@ -183,7 +184,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       last_stats_dump_time_microsec_(0),
       next_job_id_(1),
       has_unpersisted_data_(false),
-      unable_to_flush_oldest_log_(false),
+      unable_to_release_oldest_log_(false),
       env_options_(BuildDBOptions(immutable_db_options_, mutable_db_options_)),
       env_options_for_compaction_(env_->OptimizeForCompactionTableWrite(
           env_options_, immutable_db_options_)),
@@ -199,6 +200,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       two_write_queues_(options.two_write_queues),
       manual_wal_flush_(options.manual_wal_flush),
       seq_per_batch_(seq_per_batch),
+      batch_per_txn_(batch_per_txn),
       // last_sequencee_ is always maintained by the main queue that also writes
       // to the memtable. When two_write_queues_ is disabled last seq in
       // memtable is the same as last seq published to the readers. When it is
@@ -215,7 +217,11 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       // as well.
       use_custom_gc_(seq_per_batch),
       preserve_deletes_(options.preserve_deletes),
-      closed_(false) {
+      closed_(false),
+      error_handler_(immutable_db_options_, &mutex_) {
+  // !batch_per_trx_ implies seq_per_batch_ because it is only unset for
+  // WriteUnprepared, which should use seq_per_batch_.
+  assert(batch_per_txn_ || seq_per_batch_);
   env_->GetAbsolutePath(dbname, &db_absolute_path_);
 
   // Reserve ten files or so for other uses and give the rest to TableCache.
@@ -244,6 +250,43 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
   preserve_deletes_seqnum_.store(0);
 }
 
+Status DBImpl::Resume() {
+  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Resuming DB");
+
+  InstrumentedMutexLock db_mutex(&mutex_);
+
+  if (!error_handler_.IsDBStopped() && !error_handler_.IsBGWorkStopped()) {
+    // Nothing to do
+    return Status::OK();
+  }
+
+  Status s = error_handler_.GetBGError();
+  if (s.severity() > Status::Severity::kHardError) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log,
+        "DB resume requested but failed due to Fatal/Unrecoverable error");
+    return s;
+  }
+
+  JobContext job_context(0);
+  FindObsoleteFiles(&job_context, true);
+  error_handler_.ClearBGError();
+  mutex_.Unlock();
+
+  job_context.manifest_file_number = 1;
+  if (job_context.HaveSomethingToDelete()) {
+    PurgeObsoleteFiles(job_context);
+  }
+  job_context.Clean();
+
+  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Successfully resumed DB");
+  mutex_.Lock();
+  MaybeScheduleFlushOrCompaction();
+
+  // No need to check BGError again. If something happened, event listener would be
+  // notified and the operation causing it would have failed
+  return Status::OK();
+}
+
 // Will lock the mutex_,  will wait for completion if wait is true
 void DBImpl::CancelAllBackgroundWork(bool wait) {
   InstrumentedMutexLock l(&mutex_);
@@ -300,6 +343,8 @@ Status DBImpl::CloseHelper() {
     TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
     bg_cv_.Wait();
   }
+  TEST_SYNC_POINT_CALLBACK("DBImpl::CloseHelper:PendingPurgeFinished",
+                           &files_grabbed_for_purge_);
   EraseThreadStatusDbInfo();
   flush_scheduler_.Clear();
 
@@ -353,10 +398,11 @@ Status DBImpl::CloseHelper() {
     uint64_t log_number = log.writer->get_log_number();
     Status s = log.ClearWriter();
     if (!s.ok()) {
-      ROCKS_LOG_WARN(immutable_db_options_.info_log,
-                     "Unable to Sync WAL file %s with error -- %s",
-                     LogFileName(immutable_db_options_.wal_dir, log_number).c_str(),
-                     s.ToString().c_str());
+      ROCKS_LOG_WARN(
+          immutable_db_options_.info_log,
+          "Unable to Sync WAL file %s with error -- %s",
+          LogFileName(immutable_db_options_.wal_dir, log_number).c_str(),
+          s.ToString().c_str());
       // Retain the first error
       if (ret.ok()) {
         ret = s;
@@ -512,7 +558,16 @@ void DBImpl::ScheduleBgLogWriterClose(JobContext* job_context) {
   }
 }
 
-Directory* DBImpl::Directories::GetDataDir(size_t path_id) {
+Directory* DBImpl::GetDataDir(ColumnFamilyData* cfd, size_t path_id) const {
+  assert(cfd);
+  Directory* ret_dir = cfd->GetDataDir(path_id);
+  if (ret_dir == nullptr) {
+    return directories_.GetDataDir(path_id);
+  }
+  return ret_dir;
+}
+
+Directory* DBImpl::Directories::GetDataDir(size_t path_id) const {
   assert(path_id < data_dirs_.size());
   Directory* ret_dir = data_dirs_[path_id].get();
   if (ret_dir == nullptr) {
@@ -522,9 +577,12 @@ Directory* DBImpl::Directories::GetDataDir(size_t path_id) {
   return ret_dir;
 }
 
-Status DBImpl::SetOptions(ColumnFamilyHandle* column_family,
+Status DBImpl::SetOptions(
+    ColumnFamilyHandle* column_family,
     const std::unordered_map<std::string, std::string>& options_map) {
 #ifdef ROCKSDB_LITE
+  (void)column_family;
+  (void)options_map;
   return Status::NotSupported("Not supported in ROCKSDB LITE");
 #else
   auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family)->cfd();
@@ -561,9 +619,9 @@ Status DBImpl::SetOptions(ColumnFamilyHandle* column_family,
   }
   sv_context.Clean();
 
-  ROCKS_LOG_INFO(immutable_db_options_.info_log,
-                 "SetOptions() on column family [%s], inputs:",
-                 cfd->GetName().c_str());
+  ROCKS_LOG_INFO(
+      immutable_db_options_.info_log,
+      "SetOptions() on column family [%s], inputs:", cfd->GetName().c_str());
   for (const auto& o : options_map) {
     ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s: %s\n", o.first.c_str(),
                    o.second.c_str());
@@ -587,6 +645,7 @@ Status DBImpl::SetOptions(ColumnFamilyHandle* column_family,
 Status DBImpl::SetDBOptions(
     const std::unordered_map<std::string, std::string>& options_map) {
 #ifdef ROCKSDB_LITE
+  (void)options_map;
   return Status::NotSupported("Not supported in ROCKSDB LITE");
 #else
   if (options_map.empty()) {
@@ -613,7 +672,8 @@ Status DBImpl::SetDBOptions(
         MaybeScheduleFlushOrCompaction();
       }
 
-      write_controller_.set_max_delayed_write_rate(new_options.delayed_write_rate);
+      write_controller_.set_max_delayed_write_rate(
+          new_options.delayed_write_rate);
       table_cache_.get()->SetCapacity(new_options.max_open_files == -1
                                           ? TableCache::kInfiniteCapacity
                                           : new_options.max_open_files - 10);
@@ -693,7 +753,7 @@ int DBImpl::FindMinimumEmptyLevelFitting(
 }
 
 Status DBImpl::FlushWAL(bool sync) {
-  {
+  if (manual_wal_flush_) {
     // We need to lock log_write_mutex_ since logs_ might change concurrently
     InstrumentedMutexLock wl(&log_write_mutex_);
     log::Writer* cur_log_writer = logs_.back().writer;
@@ -701,12 +761,20 @@ Status DBImpl::FlushWAL(bool sync) {
     if (!s.ok()) {
       ROCKS_LOG_ERROR(immutable_db_options_.info_log, "WAL flush error %s",
                       s.ToString().c_str());
+      // In case there is a fs error we should set it globally to prevent the
+      // future writes
+      WriteStatusCheck(s);
+      // whether sync or not, we should abort the rest of function upon error
+      return s;
     }
     if (!sync) {
       ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "FlushWAL sync=false");
       return s;
     }
   }
+  if (!sync) {
+    return Status::OK();
+  }
   // sync = true
   ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "FlushWAL sync=true");
   return SyncWAL();
@@ -774,12 +842,10 @@ Status DBImpl::SyncWAL() {
   return status;
 }
 
-void DBImpl::MarkLogsSynced(
-    uint64_t up_to, bool synced_dir, const Status& status) {
+void DBImpl::MarkLogsSynced(uint64_t up_to, bool synced_dir,
+                            const Status& status) {
   mutex_.AssertHeld();
-  if (synced_dir &&
-      logfile_number_ == up_to &&
-      status.ok()) {
+  if (synced_dir && logfile_number_ == up_to && status.ok()) {
     log_dir_synced_ = true;
   }
   for (auto it = logs_.begin(); it != logs_.end() && it->number <= up_to;) {
@@ -787,6 +853,8 @@ void DBImpl::MarkLogsSynced(
     assert(log.getting_synced);
     if (status.ok() && logs_.size() > 1) {
       logs_to_free_.push_back(log.ReleaseWriter());
+      // To modify logs_ both mutex_ and log_write_mutex_ must be held
+      InstrumentedMutexLock l(&log_write_mutex_);
       it = logs_.erase(it);
     } else {
       log.getting_synced = false;
@@ -853,14 +921,14 @@ void DBImpl::BackgroundCallPurge() {
     if (!purge_queue_.empty()) {
       auto purge_file = purge_queue_.begin();
       auto fname = purge_file->fname;
+      auto dir_to_sync = purge_file->dir_to_sync;
       auto type = purge_file->type;
       auto number = purge_file->number;
-      auto path_id = purge_file->path_id;
       auto job_id = purge_file->job_id;
       purge_queue_.pop_front();
 
       mutex_.Unlock();
-      DeleteObsoleteFileImpl(job_id, fname, type, number, path_id);
+      DeleteObsoleteFileImpl(job_id, fname, dir_to_sync, type, number);
       mutex_.Lock();
     } else {
       assert(!logs_to_free_queue_.empty());
@@ -944,7 +1012,7 @@ InternalIterator* DBImpl::NewInternalIterator(
   MergeIteratorBuilder merge_iter_builder(
       &cfd->internal_comparator(), arena,
       !read_options.total_order_seek &&
-          cfd->ioptions()->prefix_extractor != nullptr);
+          super_version->mutable_cf_options.prefix_extractor != nullptr);
   // Collect iterator for mutable mem
   merge_iter_builder.AddIterator(
       super_version->mem->NewIterator(read_options, arena));
@@ -1012,12 +1080,19 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
 
   SequenceNumber snapshot;
   if (read_options.snapshot != nullptr) {
-    // Note: In WritePrepared txns this is not necessary but not harmful either.
-    // Because prep_seq > snapshot => commit_seq > snapshot so if a snapshot is
-    // specified we should be fine with skipping seq numbers that are greater
-    // than that.
-    snapshot = reinterpret_cast<const SnapshotImpl*>(
-        read_options.snapshot)->number_;
+    // Note: In WritePrepared txns this is not necessary but not harmful
+    // either.  Because prep_seq > snapshot => commit_seq > snapshot so if
+    // a snapshot is specified we should be fine with skipping seq numbers
+    // that are greater than that.
+    //
+    // In WriteUnprepared, we cannot set snapshot in the lookup key because we
+    // may skip uncommitted data that should be visible to the transaction for
+    // reading own writes.
+    snapshot =
+        reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
+    if (callback) {
+      snapshot = std::max(snapshot, callback->MaxUnpreparedSequenceNumber());
+    }
   } else {
     // Since we get and reference the super version before getting
     // the snapshot number, without a mutex protection, it is possible
@@ -1025,10 +1100,10 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
     // data for this snapshot is available. But it will contain all
     // the data available in the super version we have, which is also
     // a valid snapshot to read from.
-    // We shouldn't get snapshot before finding and referencing the
-    // super versipon because a flush happening in between may compact
-    // away data for the snapshot, but the snapshot is earlier than the
-    // data overwriting it, so users may see wrong results.
+    // We shouldn't get snapshot before finding and referencing the super
+    // version because a flush happening in between may compact away data for
+    // the snapshot, but the snapshot is earlier than the data overwriting it,
+    // so users may see wrong results.
     snapshot = last_seq_same_as_publish_seq_
                    ? versions_->LastSequence()
                    : versions_->LastPublishedSequence();
@@ -1095,7 +1170,6 @@ std::vector<Status> DBImpl::MultiGet(
     const ReadOptions& read_options,
     const std::vector<ColumnFamilyHandle*>& column_family,
     const std::vector<Slice>& keys, std::vector<std::string>* values) {
-
   StopWatch sw(env_, stats_, DB_MULTIGET);
   PERF_TIMER_GUARD(get_snapshot_time);
 
@@ -1119,8 +1193,8 @@ std::vector<Status> DBImpl::MultiGet(
 
   mutex_.Lock();
   if (read_options.snapshot != nullptr) {
-    snapshot = reinterpret_cast<const SnapshotImpl*>(
-        read_options.snapshot)->number_;
+    snapshot =
+        reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
   } else {
     snapshot = last_seq_same_as_publish_seq_
                    ? versions_->LastSequence()
@@ -1148,6 +1222,7 @@ std::vector<Status> DBImpl::MultiGet(
   // First look in the memtable, then in the immutable memtable (if any).
   // s is both in/out. When in, s could either be OK or MergeInProgress.
   // merge_operands will contain the sequence of merges in the latter case.
+  size_t num_found = 0;
   for (size_t i = 0; i < num_keys; ++i) {
     merge_context.Clear();
     Status& s = stat_list[i];
@@ -1169,11 +1244,11 @@ std::vector<Status> DBImpl::MultiGet(
       if (super_version->mem->Get(lkey, value, &s, &merge_context,
                                   &range_del_agg, read_options)) {
         done = true;
-        // TODO(?): RecordTick(stats_, MEMTABLE_HIT)?
+        RecordTick(stats_, MEMTABLE_HIT);
       } else if (super_version->imm->Get(lkey, value, &s, &merge_context,
                                          &range_del_agg, read_options)) {
         done = true;
-        // TODO(?): RecordTick(stats_, MEMTABLE_HIT)?
+        RecordTick(stats_, MEMTABLE_HIT);
       }
     }
     if (!done) {
@@ -1182,11 +1257,12 @@ std::vector<Status> DBImpl::MultiGet(
       super_version->current->Get(read_options, lkey, &pinnable_val, &s,
                                   &merge_context, &range_del_agg);
       value->assign(pinnable_val.data(), pinnable_val.size());
-      // TODO(?): RecordTick(stats_, MEMTABLE_MISS)?
+      RecordTick(stats_, MEMTABLE_MISS);
     }
 
     if (s.ok()) {
       bytes_read += value->size();
+      num_found++;
     }
   }
 
@@ -1214,6 +1290,7 @@ std::vector<Status> DBImpl::MultiGet(
 
   RecordTick(stats_, NUMBER_MULTIGET_CALLS);
   RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
+  RecordTick(stats_, NUMBER_MULTIGET_KEYS_FOUND, num_found);
   RecordTick(stats_, NUMBER_MULTIGET_BYTES_READ, bytes_read);
   MeasureTime(stats_, BYTES_PER_MULTIGET, bytes_read);
   PERF_COUNTER_ADD(multiget_read_bytes, bytes_read);
@@ -1301,6 +1378,17 @@ Status DBImpl::CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
   if (s.ok() && immutable_db_options_.allow_concurrent_memtable_write) {
     s = CheckConcurrentWritesSupported(cf_options);
   }
+  if (s.ok()) {
+    s = CheckCFPathsSupported(initial_db_options_, cf_options);
+  }
+  if (s.ok()) {
+    for (auto& cf_path : cf_options.cf_paths) {
+      s = env_->CreateDirIfMissing(cf_path.path);
+      if (!s.ok()) {
+        break;
+      }
+    }
+  }
   if (!s.ok()) {
     return s;
   }
@@ -1332,13 +1420,19 @@ Status DBImpl::CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
                                  &cf_options);
       write_thread_.ExitUnbatched(&w);
     }
+    if (s.ok()) {
+      auto* cfd =
+          versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name);
+      assert(cfd != nullptr);
+      s = cfd->AddDirectories();
+    }
     if (s.ok()) {
       single_column_family_mode_ = false;
       auto* cfd =
           versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name);
       assert(cfd != nullptr);
-      InstallSuperVersionAndScheduleWork(
-              cfd, &sv_context, *cfd->GetLatestMutableCFOptions());
+      InstallSuperVersionAndScheduleWork(cfd, &sv_context,
+                                         *cfd->GetLatestMutableCFOptions());
 
       if (!cfd->mem()->IsSnapshotSupported()) {
         is_snapshot_supported_ = false;
@@ -1420,8 +1514,8 @@ Status DBImpl::DropColumnFamilyImpl(ColumnFamilyHandle* column_family) {
       // we drop column family from a single write thread
       WriteThread::Writer w;
       write_thread_.EnterUnbatched(&w, &mutex_);
-      s = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
-                                 &edit, &mutex_);
+      s = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(), &edit,
+                                 &mutex_);
       write_thread_.ExitUnbatched(&w);
     }
     if (s.ok()) {
@@ -1471,7 +1565,7 @@ bool DBImpl::KeyMayExist(const ReadOptions& read_options,
     *value_found = true;
   }
   ReadOptions roptions = read_options;
-  roptions.read_tier = kBlockCacheTier; // read from block cache only
+  roptions.read_tier = kBlockCacheTier;  // read from block cache only
   PinnableSlice pinnable_val;
   auto s = GetImpl(roptions, column_family, key, &pinnable_val, value_found);
   value->assign(pinnable_val.data(), pinnable_val.size());
@@ -1494,17 +1588,17 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
   if (immutable_db_options_.preserve_deletes &&
       read_options.iter_start_seqnum > 0 &&
       read_options.iter_start_seqnum < preserve_deletes_seqnum_.load()) {
-        return NewErrorIterator(Status::InvalidArgument(
-          "Iterator requested internal keys which are too old and are not"
-          " guaranteed to be preserved, try larger iter_start_seqnum opt."));
-      }
+    return NewErrorIterator(Status::InvalidArgument(
+        "Iterator requested internal keys which are too old and are not"
+        " guaranteed to be preserved, try larger iter_start_seqnum opt."));
+  }
   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
   auto cfd = cfh->cfd();
   ReadCallback* read_callback = nullptr;  // No read callback provided.
   if (read_options.managed) {
 #ifdef ROCKSDB_LITE
     // not supported in lite version
-    result =  NewErrorIterator(Status::InvalidArgument(
+    result = NewErrorIterator(Status::InvalidArgument(
         "Managed Iterators not supported in RocksDBLite."));
 #else
     if ((read_options.tailing) || (read_options.snapshot != nullptr) ||
@@ -1513,7 +1607,7 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
     } else {
       // Managed iter not supported
       result = NewErrorIterator(Status::InvalidArgument(
-        "Managed Iterators not supported without snapshots."));
+          "Managed Iterators not supported without snapshots."));
     }
 #endif
   } else if (read_options.tailing) {
@@ -1525,8 +1619,8 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
     SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
     auto iter = new ForwardIterator(this, read_options, cfd, sv);
     result = NewDBIterator(
-        env_, read_options, *cfd->ioptions(), cfd->user_comparator(), iter,
-        kMaxSequenceNumber,
+        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
+        cfd->user_comparator(), iter, kMaxSequenceNumber,
         sv->mutable_cf_options.max_sequential_skip_in_iterations,
         read_callback);
 #endif
@@ -1593,7 +1687,7 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
   // likely that any iterator pointer is close to the iterator it points to so
   // that they are likely to be in the same cache line and/or page.
   ArenaWrappedDBIter* db_iter = NewArenaWrappedDbIterator(
-      env_, read_options, *cfd->ioptions(), snapshot,
+      env_, read_options, *cfd->ioptions(), sv->mutable_cf_options, snapshot,
       sv->mutable_cf_options.max_sequential_skip_in_iterations,
       sv->version_number, read_callback,
       ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob,
@@ -1644,8 +1738,8 @@ Status DBImpl::NewIterators(
       SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
       auto iter = new ForwardIterator(this, read_options, cfd, sv);
       iterators->push_back(NewDBIterator(
-          env_, read_options, *cfd->ioptions(), cfd->user_comparator(), iter,
-          kMaxSequenceNumber,
+          env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
+          cfd->user_comparator(), iter, kMaxSequenceNumber,
           sv->mutable_cf_options.max_sequential_skip_in_iterations,
           read_callback));
     }
@@ -1658,8 +1752,8 @@ Status DBImpl::NewIterators(
                         ? read_options.snapshot->GetSequenceNumber()
                         : versions_->LastSequence();
     for (size_t i = 0; i < column_families.size(); ++i) {
-      auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
-          column_families[i])->cfd();
+      auto* cfd =
+          reinterpret_cast<ColumnFamilyHandleImpl*>(column_families[i])->cfd();
       iterators->push_back(
           NewIteratorImpl(read_options, cfd, snapshot, read_callback));
     }
@@ -1676,7 +1770,7 @@ const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
 }
 #endif  // ROCKSDB_LITE
 
-const Snapshot* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
+SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
   int64_t unix_time = 0;
   env_->GetCurrentTime(&unix_time);  // Ignore error
   SnapshotImpl* s = new SnapshotImpl;
@@ -1766,13 +1860,9 @@ Status DBImpl::GetPropertiesOfTablesInRange(ColumnFamilyHandle* column_family,
 
 #endif  // ROCKSDB_LITE
 
-const std::string& DBImpl::GetName() const {
-  return dbname_;
-}
+const std::string& DBImpl::GetName() const { return dbname_; }
 
-Env* DBImpl::GetEnv() const {
-  return env_;
-}
+Env* DBImpl::GetEnv() const { return env_; }
 
 Options DBImpl::GetOptions(ColumnFamilyHandle* column_family) const {
   InstrumentedMutexLock l(&mutex_);
@@ -1805,6 +1895,13 @@ bool DBImpl::GetProperty(ColumnFamilyHandle* column_family,
     InstrumentedMutexLock l(&mutex_);
     return cfd->internal_stats()->GetStringProperty(*property_info, property,
                                                     value);
+  } else if (property_info->handle_string_dbimpl) {
+    std::string tmp_value;
+    bool ret_value = (this->*(property_info->handle_string_dbimpl))(&tmp_value);
+    if (ret_value) {
+      *value = tmp_value;
+    }
+    return ret_value;
   }
   // Shouldn't reach here since exactly one of handle_string and handle_int
   // should be non-nullptr.
@@ -1871,6 +1968,16 @@ bool DBImpl::GetIntPropertyInternal(ColumnFamilyData* cfd,
   }
 }
 
+bool DBImpl::GetPropertyHandleOptionsStatistics(std::string* value) {
+  assert(value != nullptr);
+  Statistics* statistics = immutable_db_options_.statistics.get();
+  if (!statistics) {
+    return false;
+  }
+  *value = statistics->ToString();
+  return true;
+}
+
 #ifndef ROCKSDB_LITE
 Status DBImpl::ResetStats() {
   InstrumentedMutexLock l(&mutex_);
@@ -2145,9 +2252,9 @@ Status DBImpl::DeleteFile(std::string name) {
     status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                     &edit, &mutex_, directories_.GetDbDir());
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_context,
-          *cfd->GetLatestMutableCFOptions());
+      InstallSuperVersionAndScheduleWork(cfd, &job_context.superversion_context,
+                                         *cfd->GetLatestMutableCFOptions(),
+                                         FlushReason::kDeleteFiles);
     }
     FindObsoleteFiles(&job_context, false);
   }  // lock released here
@@ -2198,9 +2305,9 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
           end_key = &end_storage;
         }
 
-        vstorage->GetCleanInputsWithinInterval(i, begin_key, end_key,
-                                               &level_files, -1 /* hint_index */,
-                                               nullptr /* file_index */);
+        vstorage->GetCleanInputsWithinInterval(
+            i, begin_key, end_key, &level_files, -1 /* hint_index */,
+            nullptr /* file_index */);
         FileMetaData* level_file;
         for (uint32_t j = 0; j < level_files.size(); j++) {
           level_file = level_files[j];
@@ -2211,7 +2318,8 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
             continue;
           }
           if (!include_end && end != nullptr &&
-              cfd->user_comparator()->Compare(level_file->largest.user_key(), *end) == 0) {
+              cfd->user_comparator()->Compare(level_file->largest.user_key(),
+                                              *end) == 0) {
             continue;
           }
           edit.SetColumnFamily(cfd->GetID());
@@ -2229,9 +2337,9 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
     status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                     &edit, &mutex_, directories_.GetDbDir());
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_context,
-          *cfd->GetLatestMutableCFOptions());
+      InstallSuperVersionAndScheduleWork(cfd, &job_context.superversion_context,
+                                         *cfd->GetLatestMutableCFOptions(),
+                                         FlushReason::kDeleteFiles);
     }
     for (auto* deleted_file : deleted_files) {
       deleted_file->being_compacted = false;
@@ -2255,9 +2363,8 @@ void DBImpl::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
   versions_->GetLiveFilesMetaData(metadata);
 }
 
-void DBImpl::GetColumnFamilyMetaData(
-    ColumnFamilyHandle* column_family,
-    ColumnFamilyMetaData* cf_meta) {
+void DBImpl::GetColumnFamilyMetaData(ColumnFamilyHandle* column_family,
+                                     ColumnFamilyMetaData* cf_meta) {
   assert(column_family);
   auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family)->cfd();
   auto* sv = GetAndRefSuperVersion(cfd);
@@ -2311,7 +2418,8 @@ Status DBImpl::GetDbIdentity(std::string& identity) const {
     if (!s.ok()) {
       return s;
     }
-    id_file_reader.reset(new SequentialFileReader(std::move(idfile)));
+    id_file_reader.reset(
+        new SequentialFileReader(std::move(idfile), idfilename));
   }
 
   uint64_t file_size;
@@ -2383,10 +2491,10 @@ Status DB::ListColumnFamilies(const DBOptions& db_options,
   return VersionSet::ListColumnFamilies(column_families, name, db_options.env);
 }
 
-Snapshot::~Snapshot() {
-}
+Snapshot::~Snapshot() {}
 
-Status DestroyDB(const std::string& dbname, const Options& options) {
+Status DestroyDB(const std::string& dbname, const Options& options,
+                 const std::vector<ColumnFamilyDescriptor>& column_families) {
   ImmutableDBOptions soptions(SanitizeOptions(dbname, options));
   Env* env = soptions.env;
   std::vector<std::string> filenames;
@@ -2404,15 +2512,15 @@ Status DestroyDB(const std::string& dbname, const Options& options) {
     uint64_t number;
     FileType type;
     InfoLogPrefix info_log_prefix(!soptions.db_log_dir.empty(), dbname);
-    for (size_t i = 0; i < filenames.size(); i++) {
-      if (ParseFileName(filenames[i], &number, info_log_prefix.prefix, &type) &&
-          type != kDBLockFile) {  // Lock file will be deleted at end
+    for (const auto& fname : filenames) {
+      if (ParseFileName(fname, &number, info_log_prefix.prefix, &type) &&
+        type != kDBLockFile) {  // Lock file will be deleted at end
         Status del;
-        std::string path_to_delete = dbname + "/" + filenames[i];
+        std::string path_to_delete = dbname + "/" + fname;
         if (type == kMetaDatabase) {
           del = DestroyDB(path_to_delete, options);
         } else if (type == kTableFile) {
-          del = DeleteSSTFile(&soptions, path_to_delete, 0);
+          del = DeleteSSTFile(&soptions, path_to_delete, dbname);
         } else {
           del = env->DeleteFile(path_to_delete);
         }
@@ -2422,59 +2530,82 @@ Status DestroyDB(const std::string& dbname, const Options& options) {
       }
     }
 
-    for (size_t path_id = 0; path_id < options.db_paths.size(); path_id++) {
-      const auto& db_path = options.db_paths[path_id];
-      env->GetChildren(db_path.path, &filenames);
-      for (size_t i = 0; i < filenames.size(); i++) {
-        if (ParseFileName(filenames[i], &number, &type) &&
+    std::vector<std::string> paths;
+
+    for (const auto& path : options.db_paths) {
+      paths.emplace_back(path.path);
+    }
+    for (const auto& cf : column_families) {
+      for (const auto& path : cf.options.cf_paths) {
+        paths.emplace_back(path.path);
+      }
+    }
+
+    // Remove duplicate paths.
+    // Note that we compare only the actual paths but not path ids.
+    // This reason is that same path can appear at different path_ids
+    // for different column families.
+    std::sort(paths.begin(), paths.end());
+    paths.erase(std::unique(paths.begin(), paths.end()), paths.end());
+
+    for (const auto& path : paths) {
+      if (env->GetChildren(path, &filenames).ok()) {
+        for (const auto& fname : filenames) {
+          if (ParseFileName(fname, &number, &type) &&
             type == kTableFile) {  // Lock file will be deleted at end
-          std::string table_path = db_path.path + "/" + filenames[i];
-          Status del = DeleteSSTFile(&soptions, table_path,
-                                     static_cast<uint32_t>(path_id));
-          if (result.ok() && !del.ok()) {
-            result = del;
+            std::string table_path = path + "/" + fname;
+            Status del = DeleteSSTFile(&soptions, table_path, dbname);
+            if (result.ok() && !del.ok()) {
+              result = del;
+            }
           }
         }
+        env->DeleteDir(path);
       }
     }
 
     std::vector<std::string> walDirFiles;
     std::string archivedir = ArchivalDirectory(dbname);
+    bool wal_dir_exists = false;
     if (dbname != soptions.wal_dir) {
-      env->GetChildren(soptions.wal_dir, &walDirFiles);
+      wal_dir_exists = env->GetChildren(soptions.wal_dir, &walDirFiles).ok();
       archivedir = ArchivalDirectory(soptions.wal_dir);
     }
 
-    // Delete log files in the WAL dir
-    for (const auto& file : walDirFiles) {
-      if (ParseFileName(file, &number, &type) && type == kLogFile) {
-        Status del = env->DeleteFile(LogFileName(soptions.wal_dir, number));
-        if (result.ok() && !del.ok()) {
-          result = del;
-        }
-      }
-    }
-
+    // Archive dir may be inside wal dir or dbname and should be
+    // processed and removed before those otherwise we have issues
+    // removing them
     std::vector<std::string> archiveFiles;
-    env->GetChildren(archivedir, &archiveFiles);
-    // Delete archival files.
-    for (size_t i = 0; i < archiveFiles.size(); ++i) {
-      if (ParseFileName(archiveFiles[i], &number, &type) &&
+    if (env->GetChildren(archivedir, &archiveFiles).ok()) {
+      // Delete archival files.
+      for (const auto& file : archiveFiles) {
+        if (ParseFileName(file, &number, &type) &&
           type == kLogFile) {
-        Status del = env->DeleteFile(archivedir + "/" + archiveFiles[i]);
-        if (result.ok() && !del.ok()) {
-          result = del;
+          Status del = env->DeleteFile(archivedir + "/" + file);
+          if (result.ok() && !del.ok()) {
+            result = del;
+          }
         }
       }
+      env->DeleteDir(archivedir);
     }
 
-    // ignore case where no archival directory is present
-    env->DeleteDir(archivedir);
+    // Delete log files in the WAL dir
+    if (wal_dir_exists) {
+      for (const auto& file : walDirFiles) {
+        if (ParseFileName(file, &number, &type) && type == kLogFile) {
+          Status del = env->DeleteFile(LogFileName(soptions.wal_dir, number));
+          if (result.ok() && !del.ok()) {
+            result = del;
+          }
+        }
+      }
+      env->DeleteDir(soptions.wal_dir);
+    }
 
     env->UnlockFile(lock);  // Ignore error since state is already gone
     env->DeleteFile(lockname);
     env->DeleteDir(dbname);  // Ignore error in case dir contains other files
-    env->DeleteDir(soptions.wal_dir);
   }
   return result;
 }
@@ -2536,6 +2667,9 @@ Status DBImpl::WriteOptionsFile(bool need_mutex_lock,
                              s.ToString().c_str());
     }
   }
+#else
+  (void)need_mutex_lock;
+  (void)need_enter_write_thread;
 #endif  // !ROCKSDB_LITE
   return Status::OK();
 }
@@ -2601,25 +2735,26 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
   // Retry if the file name happen to conflict with an existing one.
   s = GetEnv()->RenameFile(file_name, options_file_name);
 
-  DeleteObsoleteOptionsFiles();
+  if (0 == disable_delete_obsolete_files_) {
+    DeleteObsoleteOptionsFiles();
+  }
   return s;
 #else
+  (void)file_name;
   return Status::OK();
 #endif  // !ROCKSDB_LITE
 }
 
 #ifdef ROCKSDB_USING_THREAD_STATUS
 
-void DBImpl::NewThreadStatusCfInfo(
-    ColumnFamilyData* cfd) const {
+void DBImpl::NewThreadStatusCfInfo(ColumnFamilyData* cfd) const {
   if (immutable_db_options_.enable_thread_tracking) {
     ThreadStatusUtil::NewColumnFamilyInfo(this, cfd, cfd->GetName(),
                                           cfd->ioptions()->env);
   }
 }
 
-void DBImpl::EraseThreadStatusCfInfo(
-    ColumnFamilyData* cfd) const {
+void DBImpl::EraseThreadStatusCfInfo(ColumnFamilyData* cfd) const {
   if (immutable_db_options_.enable_thread_tracking) {
     ThreadStatusUtil::EraseColumnFamilyInfo(cfd);
   }
@@ -2632,21 +2767,16 @@ void DBImpl::EraseThreadStatusDbInfo() const {
 }
 
 #else
-void DBImpl::NewThreadStatusCfInfo(
-    ColumnFamilyData* cfd) const {
-}
+void DBImpl::NewThreadStatusCfInfo(ColumnFamilyData* /*cfd*/) const {}
 
-void DBImpl::EraseThreadStatusCfInfo(
-    ColumnFamilyData* cfd) const {
-}
+void DBImpl::EraseThreadStatusCfInfo(ColumnFamilyData* /*cfd*/) const {}
 
-void DBImpl::EraseThreadStatusDbInfo() const {
-}
+void DBImpl::EraseThreadStatusDbInfo() const {}
 #endif  // ROCKSDB_USING_THREAD_STATUS
 
 //
 // A global method that can dump out the build version
-void DumpRocksDBBuildVersion(Logger * log) {
+void DumpRocksDBBuildVersion(Logger* log) {
 #if !defined(IOS_CROSS_COMPILE)
   // if we compile with Xcode, we don't run build_detect_version, so we don't
   // generate util/build_version.cc
@@ -2783,7 +2913,7 @@ Status DBImpl::IngestExternalFile(
   if (ingestion_options.ingest_behind) {
     if (!immutable_db_options_.allow_ingest_behind) {
       return Status::InvalidArgument(
-        "Can't ingest_behind file in DB with allow_ingest_behind=false");
+          "Can't ingest_behind file in DB with allow_ingest_behind=false");
     }
   }
 
@@ -2794,16 +2924,18 @@ Status DBImpl::IngestExternalFile(
   std::list<uint64_t>::iterator pending_output_elem;
   {
     InstrumentedMutexLock l(&mutex_);
-    if (!bg_error_.ok()) {
+    if (error_handler_.IsDBStopped()) {
       // Don't ingest files when there is a bg_error
-      return bg_error_;
+      return error_handler_.GetBGError();
     }
 
     // Make sure that bg cleanup wont delete the files that we are ingesting
     pending_output_elem = CaptureCurrentFileNumberInPendingOutputs();
   }
 
-  status = ingestion_job.Prepare(external_files);
+  SuperVersion* super_version = cfd->GetReferencedSuperVersion(&mutex_);
+  status = ingestion_job.Prepare(external_files, super_version);
+  CleanupSuperVersion(super_version);
   if (!status.ok()) {
     return status;
   }
@@ -2859,8 +2991,8 @@ Status DBImpl::IngestExternalFile(
                                  &mutex_, directories_.GetDbDir());
     }
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(cfd, &sv_context,
-                                         *mutable_cf_options);
+      InstallSuperVersionAndScheduleWork(cfd, &sv_context, *mutable_cf_options,
+                                         FlushReason::kExternalFileIngestion);
     }
 
     // Resume writes to the DB
@@ -2916,11 +3048,12 @@ Status DBImpl::VerifyChecksum() {
   }
   for (auto& sv : sv_list) {
     VersionStorageInfo* vstorage = sv->current->storage_info();
+    ColumnFamilyData* cfd = sv->current->cfd();
     for (int i = 0; i < vstorage->num_non_empty_levels() && s.ok(); i++) {
       for (size_t j = 0; j < vstorage->LevelFilesBrief(i).num_files && s.ok();
            j++) {
         const auto& fd = vstorage->LevelFilesBrief(i).files[j].fd;
-        std::string fname = TableFileName(immutable_db_options_.db_paths,
+        std::string fname = TableFileName(cfd->ioptions()->cf_paths,
                                           fd.GetNumber(), fd.GetPathId());
         s = rocksdb::VerifySstFileChecksum(options, env_options, fname);
       }
@@ -2938,7 +3071,7 @@ Status DBImpl::VerifyChecksum() {
       }
     }
     for (auto cfd : cfd_list) {
-        cfd->Unref();
+      cfd->Unref();
     }
   }
   return s;
@@ -2974,5 +3107,4 @@ void DBImpl::WaitForIngestFile() {
 }
 
 #endif  // ROCKSDB_LITE
-
 }  // namespace rocksdb

db/db_impl.h

@@ -23,10 +23,13 @@
 #include "db/compaction_job.h"
 #include "db/dbformat.h"
 #include "db/external_sst_file_ingestion_job.h"
+#include "db/error_handler.h"
+#include "db/event_helpers.h"
 #include "db/flush_job.h"
 #include "db/flush_scheduler.h"
 #include "db/internal_stats.h"
 #include "db/log_writer.h"
+#include "db/logs_with_prep_tracker.h"
 #include "db/pre_release_callback.h"
 #include "db/read_callback.h"
 #include "db/snapshot_checker.h"
@@ -69,9 +72,12 @@ struct MemTableInfo;
 class DBImpl : public DB {
  public:
   DBImpl(const DBOptions& options, const std::string& dbname,
-         const bool seq_per_batch = false);
+         const bool seq_per_batch = false, const bool batch_per_txn = true);
   virtual ~DBImpl();
 
+  using DB::Resume;
+  virtual Status Resume() override;
+
   // Implementations of the DB interface
   using DB::Put;
   virtual Status Put(const WriteOptions& options,
@@ -220,9 +226,12 @@ class DBImpl : public DB {
   virtual Status Flush(const FlushOptions& options,
                        ColumnFamilyHandle* column_family) override;
   virtual Status FlushWAL(bool sync) override;
+  bool TEST_WALBufferIsEmpty();
   virtual Status SyncWAL() override;
 
   virtual SequenceNumber GetLatestSequenceNumber() const override;
+  // REQUIRES: joined the main write queue if two_write_queues is disabled, and
+  // the second write queue otherwise.
   virtual void SetLastPublishedSequence(SequenceNumber seq);
   // Returns LastSequence in last_seq_same_as_publish_seq_
   // mode and LastAllocatedSequence otherwise. This is useful when visiblility
@@ -340,6 +349,7 @@ class DBImpl : public DB {
 
   Status RunManualCompaction(ColumnFamilyData* cfd, int input_level,
                              int output_level, uint32_t output_path_id,
+                             uint32_t max_subcompactions,
                              const Slice* begin, const Slice* end,
                              bool exclusive,
                              bool disallow_trivial_move = false);
@@ -351,6 +361,10 @@ class DBImpl : public DB {
       Arena* arena, RangeDelAggregator* range_del_agg,
       ColumnFamilyHandle* column_family = nullptr);
 
+  LogsWithPrepTracker* logs_with_prep_tracker() {
+    return &logs_with_prep_tracker_;
+  }
+
 #ifndef NDEBUG
   // Extra methods (for testing) that are not in the public DB interface
   // Implemented in db_impl_debug.cc
@@ -362,9 +376,7 @@ class DBImpl : public DB {
 
   void TEST_SwitchWAL();
 
-  bool TEST_UnableToFlushOldestLog() {
-    return unable_to_flush_oldest_log_;
-  }
+  bool TEST_UnableToReleaseOldestLog() { return unable_to_release_oldest_log_; }
 
   bool TEST_IsLogGettingFlushed() {
     return alive_log_files_.begin()->getting_flushed;
@@ -476,7 +488,7 @@ class DBImpl : public DB {
   // It is not necessary to hold the mutex when invoking this method.
   // If FindObsoleteFiles() was run, we need to also run
   // PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
-  void PurgeObsoleteFiles(const JobContext& background_contet,
+  void PurgeObsoleteFiles(JobContext& background_contet,
                           bool schedule_only = false);
 
   void SchedulePurge();
@@ -549,25 +561,50 @@ class DBImpl : public DB {
   // these will then be passed to TransactionDB so that
   // locks can be reacquired before writing can resume.
   struct RecoveredTransaction {
-    uint64_t log_number_;
     std::string name_;
-    WriteBatch* batch_;
-    // The seq number of the first key in the batch
-    SequenceNumber seq_;
-    // Number of sub-batched. A new sub-batch is created if we txn attempts to
-    // inserts a duplicate key,seq to memtable. This is currently used in
-    // WritePrparedTxn
-    size_t batch_cnt_;
+    bool unprepared_;
+
+    struct BatchInfo {
+      uint64_t log_number_;
+      // TODO(lth): For unprepared, the memory usage here can be big for
+      // unprepared transactions. This is only useful for rollbacks, and we
+      // can in theory just keep keyset for that.
+      WriteBatch* batch_;
+      // Number of sub-batches. A new sub-batch is created if txn attempts to
+      // insert a duplicate key,seq to memtable. This is currently used in
+      // WritePreparedTxn/WriteUnpreparedTxn.
+      size_t batch_cnt_;
+    };
+
+    // This maps the seq of the first key in the batch to BatchInfo, which
+    // contains WriteBatch and other information relevant to the batch.
+    //
+    // For WriteUnprepared, batches_ can have size greater than 1, but for
+    // other write policies, it must be of size 1.
+    std::map<SequenceNumber, BatchInfo> batches_;
+
     explicit RecoveredTransaction(const uint64_t log, const std::string& name,
                                   WriteBatch* batch, SequenceNumber seq,
-                                  size_t batch_cnt)
-        : log_number_(log),
-          name_(name),
-          batch_(batch),
-          seq_(seq),
-          batch_cnt_(batch_cnt) {}
+                                  size_t batch_cnt, bool unprepared)
+        : name_(name), unprepared_(unprepared) {
+      batches_[seq] = {log, batch, batch_cnt};
+    }
 
-    ~RecoveredTransaction() { delete batch_; }
+    ~RecoveredTransaction() {
+      for (auto& it : batches_) {
+        delete it.second.batch_;
+      }
+    }
+
+    void AddBatch(SequenceNumber seq, uint64_t log_number, WriteBatch* batch,
+                  size_t batch_cnt, bool unprepared) {
+      assert(batches_.count(seq) == 0);
+      batches_[seq] = {log_number, batch, batch_cnt};
+      // Prior state must be unprepared, since the prepare batch must be the
+      // last batch.
+      assert(unprepared_);
+      unprepared_ = unprepared;
+    }
   };
 
   bool allow_2pc() const { return immutable_db_options_.allow_2pc; }
@@ -588,10 +625,20 @@ class DBImpl : public DB {
 
   void InsertRecoveredTransaction(const uint64_t log, const std::string& name,
                                   WriteBatch* batch, SequenceNumber seq,
-                                  size_t batch_cnt) {
-    recovered_transactions_[name] =
-        new RecoveredTransaction(log, name, batch, seq, batch_cnt);
-    MarkLogAsContainingPrepSection(log);
+                                  size_t batch_cnt, bool unprepared_batch) {
+    // For WriteUnpreparedTxn, InsertRecoveredTransaction is called multiple
+    // times for every unprepared batch encountered during recovery.
+    //
+    // If the transaction is prepared, then the last call to
+    // InsertRecoveredTransaction will have unprepared_batch = false.
+    auto rtxn = recovered_transactions_.find(name);
+    if (rtxn == recovered_transactions_.end()) {
+      recovered_transactions_[name] = new RecoveredTransaction(
+          log, name, batch, seq, batch_cnt, unprepared_batch);
+    } else {
+      rtxn->second->AddBatch(seq, log, batch, batch_cnt, unprepared_batch);
+    }
+    logs_with_prep_tracker_.MarkLogAsContainingPrepSection(log);
   }
 
   void DeleteRecoveredTransaction(const std::string& name) {
@@ -599,7 +646,10 @@ class DBImpl : public DB {
     assert(it != recovered_transactions_.end());
     auto* trx = it->second;
     recovered_transactions_.erase(it);
-    MarkLogAsHavingPrepSectionFlushed(trx->log_number_);
+    for (const auto& info : trx->batches_) {
+      logs_with_prep_tracker_.MarkLogAsHavingPrepSectionFlushed(
+          info.second.log_number_);
+    }
     delete trx;
   }
 
@@ -611,14 +661,15 @@ class DBImpl : public DB {
     recovered_transactions_.clear();
   }
 
-  void MarkLogAsHavingPrepSectionFlushed(uint64_t log);
-  void MarkLogAsContainingPrepSection(uint64_t log);
   void AddToLogsToFreeQueue(log::Writer* log_writer) {
     logs_to_free_queue_.push_back(log_writer);
   }
 
   void SetSnapshotChecker(SnapshotChecker* snapshot_checker);
 
+  // Not thread-safe.
+  void SetRecoverableStatePreReleaseCallback(PreReleaseCallback* callback);
+
   InstrumentedMutex* mutex() { return &mutex_; }
 
   Status NewDB();
@@ -627,10 +678,13 @@ class DBImpl : public DB {
   static Status Open(const DBOptions& db_options, const std::string& name,
                      const std::vector<ColumnFamilyDescriptor>& column_families,
                      std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
-                     const bool seq_per_batch);
+                     const bool seq_per_batch, const bool batch_per_txn);
 
   virtual Status Close() override;
 
+  static Status CreateAndNewDirectory(Env* env, const std::string& dirname,
+                                      std::unique_ptr<Directory>* directory);
+
  protected:
   Env* const env_;
   const std::string dbname_;
@@ -683,14 +737,20 @@ class DBImpl : public DB {
   void EraseThreadStatusDbInfo() const;
 
   // If disable_memtable is set the application logic must guarantee that the
-  // batch will still be skipped from memtable during the recovery. In
-  // WriteCommitted it is guarnateed since disable_memtable is used for prepare
-  // batch which will be written to memtable later during the commit, and in
-  // WritePrepared it is guaranteed since it will be used only for WAL markers
-  // which will never be written to memtable.
-  // batch_cnt is expected to be non-zero in seq_per_batch mode and indicates
-  // the number of sub-patches. A sub-patch is a subset of the write batch that
-  // does not have duplicate keys.
+  // batch will still be skipped from memtable during the recovery. An excption
+  // to this is seq_per_batch_ mode, in which since each batch already takes one
+  // seq, it is ok for the batch to write to memtable during recovery as long as
+  // it only takes one sequence number: i.e., no duplicate keys.
+  // In WriteCommitted it is guarnateed since disable_memtable is used for
+  // prepare batch which will be written to memtable later during the commit,
+  // and in WritePrepared it is guaranteed since it will be used only for WAL
+  // markers which will never be written to memtable. If the commit marker is
+  // accompanied with CommitTimeWriteBatch that is not written to memtable as
+  // long as it has no duplicate keys, it does not violate the one-seq-per-batch
+  // policy.
+  // batch_cnt is expected to be non-zero in seq_per_batch mode and
+  // indicates the number of sub-patches. A sub-patch is a subset of the write
+  // batch that does not have duplicate keys.
   Status WriteImpl(const WriteOptions& options, WriteBatch* updates,
                    WriteCallback* callback = nullptr,
                    uint64_t* log_used = nullptr, uint64_t log_ref = 0,
@@ -713,8 +773,6 @@ class DBImpl : public DB {
                           uint64_t* seq_used = nullptr, size_t batch_cnt = 0,
                           PreReleaseCallback* pre_release_callback = nullptr);
 
-  uint64_t FindMinLogContainingOutstandingPrep();
-  uint64_t FindMinPrepLogReferencedByMemTable();
   // write cached_recoverable_state_ to memtable if it is not empty
   // The writer must be the leader in write_thread_ and holding mutex_
   Status WriteRecoverableState();
@@ -726,20 +784,24 @@ class DBImpl : public DB {
   friend class DB;
   friend class InternalStats;
   friend class PessimisticTransaction;
+  friend class TransactionBaseImpl;
   friend class WriteCommittedTxn;
   friend class WritePreparedTxn;
   friend class WritePreparedTxnDB;
   friend class WriteBatchWithIndex;
+  friend class WriteUnpreparedTxnDB;
 #ifndef ROCKSDB_LITE
   friend class ForwardIterator;
 #endif
   friend struct SuperVersion;
   friend class CompactedDBImpl;
+  friend class DBTest_ConcurrentFlushWAL_Test;
 #ifndef NDEBUG
   friend class DBTest2_ReadCallbackTest_Test;
   friend class WriteCallbackTest_WriteWithCallbackTest_Test;
   friend class XFTransactionWriteHandler;
   friend class DBBlobIndexTest;
+  friend class WriteUnpreparedTransactionTest_RecoveryRollbackUnprepared_Test;
 #endif
   struct CompactionState;
 
@@ -782,7 +844,8 @@ class DBImpl : public DB {
   void DeleteObsoleteFiles();
   // Delete obsolete files and log status and information of file deletion
   void DeleteObsoleteFileImpl(int job_id, const std::string& fname,
-                              FileType type, uint64_t number, uint32_t path_id);
+                              const std::string& path_to_sync, FileType type,
+                              uint64_t number);
 
   // Background process needs to call
   //     auto x = CaptureCurrentFileNumberInPendingOutputs()
@@ -833,7 +896,8 @@ class DBImpl : public DB {
 
   Status ScheduleFlushes(WriteContext* context);
 
-  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context);
+  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
+                        FlushReason flush_reason = FlushReason::kOthers);
 
   // Force current memtable contents to be flushed.
   Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options,
@@ -873,7 +937,7 @@ class DBImpl : public DB {
                               size_t seq_inc);
 
   // Used by WriteImpl to update bg_error_ if paranoid check is enabled.
-  void WriteCallbackStatusCheck(const Status& status);
+  void WriteStatusCheck(const Status& status);
 
   // Used by WriteImpl to update bg_error_ in case of memtable insert error.
   void MemTableInsertStatusCheck(const Status& memtable_insert_status);
@@ -902,8 +966,8 @@ class DBImpl : public DB {
   void MaybeScheduleFlushOrCompaction();
   void SchedulePendingFlush(ColumnFamilyData* cfd, FlushReason flush_reason);
   void SchedulePendingCompaction(ColumnFamilyData* cfd);
-  void SchedulePendingPurge(std::string fname, FileType type, uint64_t number,
-                            uint32_t path_id, int job_id);
+  void SchedulePendingPurge(std::string fname, std::string dir_to_sync,
+                            FileType type, uint64_t number, int job_id);
   static void BGWorkCompaction(void* arg);
   // Runs a pre-chosen universal compaction involving bottom level in a
   // separate, bottom-pri thread pool.
@@ -921,6 +985,9 @@ class DBImpl : public DB {
   Status BackgroundFlush(bool* madeProgress, JobContext* job_context,
                          LogBuffer* log_buffer);
 
+  bool EnoughRoomForCompaction(const std::vector<CompactionInputFiles>& inputs,
+                               bool* sfm_bookkeeping, LogBuffer* log_buffer);
+
   void PrintStatistics();
 
   // dump rocksdb.stats to LOG
@@ -945,10 +1012,12 @@ class DBImpl : public DB {
   // helper function to call after some of the logs_ were synced
   void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);
 
-  const Snapshot* GetSnapshotImpl(bool is_write_conflict_boundary);
+  SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary);
 
   uint64_t GetMaxTotalWalSize() const;
 
+  Directory* GetDataDir(ColumnFamilyData* cfd, size_t path_id) const;
+
   Status CloseHelper();
 
   // table_cache_ provides its own synchronization
@@ -1086,7 +1155,7 @@ class DBImpl : public DB {
                           const std::string& wal_dir,
                           const std::vector<DbPath>& data_paths);
 
-    Directory* GetDataDir(size_t path_id);
+    Directory* GetDataDir(size_t path_id) const;
 
     Directory* GetWalDir() {
       if (wal_dir_) {
@@ -1101,9 +1170,6 @@ class DBImpl : public DB {
     std::unique_ptr<Directory> db_dir_;
     std::vector<std::unique_ptr<Directory>> data_dirs_;
     std::unique_ptr<Directory> wal_dir_;
-
-    Status CreateAndNewDirectory(Env* env, const std::string& dirname,
-                                 std::unique_ptr<Directory>* directory) const;
   };
 
   Directories directories_;
@@ -1145,13 +1211,13 @@ class DBImpl : public DB {
   // purge_queue_
   struct PurgeFileInfo {
     std::string fname;
+    std::string dir_to_sync;
     FileType type;
     uint64_t number;
-    uint32_t path_id;
     int job_id;
-    PurgeFileInfo(std::string fn, FileType t, uint64_t num, uint32_t pid,
+    PurgeFileInfo(std::string fn, std::string d, FileType t, uint64_t num,
                   int jid)
-        : fname(fn), type(t), number(num), path_id(pid), job_id(jid) {}
+        : fname(fn), dir_to_sync(d), type(t), number(num), job_id(jid) {}
   };
 
   // flush_queue_ and compaction_queue_ hold column families that we need to
@@ -1182,6 +1248,10 @@ class DBImpl : public DB {
   // A queue to store filenames of the files to be purged
   std::deque<PurgeFileInfo> purge_queue_;
 
+  // A vector to store the file numbers that have been assigned to certain
+  // JobContext. Current implementation tracks ssts only.
+  std::vector<uint64_t> files_grabbed_for_purge_;
+
   // A queue to store log writers to close
   std::deque<log::Writer*> logs_to_free_queue_;
   int unscheduled_flushes_;
@@ -1240,9 +1310,6 @@ class DBImpl : public DB {
     PrepickedCompaction* prepicked_compaction;
   };
 
-  // Have we encountered a background error in paranoid mode?
-  Status bg_error_;
-
   // shall we disable deletion of obsolete files
   // if 0 the deletion is enabled.
   // if non-zero, files will not be getting deleted
@@ -1277,7 +1344,7 @@ class DBImpl : public DB {
   // We must attempt to free the dependent memtables again
   // at a later time after the transaction in the oldest
   // log is fully commited.
-  bool unable_to_flush_oldest_log_;
+  bool unable_to_release_oldest_log_;
 
   static const int KEEP_LOG_FILE_NUM = 1000;
   // MSVC version 1800 still does not have constexpr for ::max()
@@ -1314,38 +1381,16 @@ class DBImpl : public DB {
   // Indicate DB was opened successfully
   bool opened_successfully_;
 
-  // REQUIRES: logs_with_prep_mutex_ held
-  //
-  // sorted list of log numbers still containing prepared data.
-  // this is used by FindObsoleteFiles to determine which
-  // flushed logs we must keep around because they still
-  // contain prepared data which has not been committed or rolled back
-  struct LogCnt {
-    uint64_t log;  // the log number
-    uint64_t cnt;  // number of prepared sections in the log
-  };
-  std::vector<LogCnt> logs_with_prep_;
-  std::mutex logs_with_prep_mutex_;
-
-  // REQUIRES: prepared_section_completed_mutex_ held
-  //
-  // to be used in conjunction with logs_with_prep_.
-  // once a transaction with data in log L is committed or rolled back
-  // rather than updating logs_with_prep_ directly we keep track of that
-  // in prepared_section_completed_ which maps LOG -> instance_count. This helps
-  // avoiding contention between a commit thread and the prepare threads.
-  //
-  // when trying to determine the minimum log still active we first
-  // consult logs_with_prep_. while that root value maps to
-  // an equal value in prepared_section_completed_ we erase the log from
-  // both logs_with_prep_ and prepared_section_completed_.
-  std::unordered_map<uint64_t, uint64_t> prepared_section_completed_;
-  std::mutex prepared_section_completed_mutex_;
+  LogsWithPrepTracker logs_with_prep_tracker_;
 
   // Callback for compaction to check if a key is visible to a snapshot.
   // REQUIRES: mutex held
   std::unique_ptr<SnapshotChecker> snapshot_checker_;
 
+  // Callback for when the cached_recoverable_state_ is written to memtable
+  // Only to be set during initialization
+  std::unique_ptr<PreReleaseCallback> recoverable_state_pre_release_callback_;
+
   // No copying allowed
   DBImpl(const DBImpl&);
   void operator=(const DBImpl&);
@@ -1359,7 +1404,8 @@ class DBImpl : public DB {
   // state needs flush or compaction.
   void InstallSuperVersionAndScheduleWork(
       ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-      const MutableCFOptions& mutable_cf_options);
+      const MutableCFOptions& mutable_cf_options,
+      FlushReason flush_reason = FlushReason::kOthers);
 
 #ifndef ROCKSDB_LITE
   using DB::GetPropertiesOfAllTables;
@@ -1375,6 +1421,7 @@ class DBImpl : public DB {
   bool GetIntPropertyInternal(ColumnFamilyData* cfd,
                               const DBPropertyInfo& property_info,
                               bool is_locked, uint64_t* value);
+  bool GetPropertyHandleOptionsStatistics(std::string* value);
 
   bool HasPendingManualCompaction();
   bool HasExclusiveManualCompaction();
@@ -1384,6 +1431,9 @@ class DBImpl : public DB {
   bool HaveManualCompaction(ColumnFamilyData* cfd);
   bool MCOverlap(ManualCompactionState* m, ManualCompactionState* m1);
 
+  bool ShouldPurge(uint64_t file_number) const;
+  void MarkAsGrabbedForPurge(uint64_t file_number);
+
   size_t GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
   Env::WriteLifeTimeHint CalculateWALWriteHint() {
     return Env::WLTH_SHORT;
@@ -1400,6 +1450,13 @@ class DBImpl : public DB {
   //
   // Default: false
   const bool seq_per_batch_;
+  // This determines during recovery whether we expect one writebatch per
+  // recovered transaction, or potentially multiple writebatches per
+  // transaction. For WriteUnprepared, this is set to false, since multiple
+  // batches can exist per transaction.
+  //
+  // Default: true
+  const bool batch_per_txn_;
   // LastSequence also indicates last published sequence visibile to the
   // readers. Otherwise LastPublishedSequence should be used.
   const bool last_seq_same_as_publish_seq_;
@@ -1417,6 +1474,8 @@ class DBImpl : public DB {
 
   // Flag to check whether Close() has been called on this DB
   bool closed_;
+
+  ErrorHandler error_handler_;
 };
 
 extern Options SanitizeOptions(const std::string& db,
@@ -1428,6 +1487,25 @@ extern CompressionType GetCompressionFlush(
     const ImmutableCFOptions& ioptions,
     const MutableCFOptions& mutable_cf_options);
 
+// Return the earliest log file to keep after the memtable flush is
+// finalized.
+// `cfd_to_flush` is the column family whose memtable (specified in
+// `memtables_to_flush`) will be flushed and thus will not depend on any WAL
+// file.
+// The function is only applicable to 2pc mode.
+extern uint64_t PrecomputeMinLogNumberToKeep(
+    VersionSet* vset, const ColumnFamilyData& cfd_to_flush,
+    autovector<VersionEdit*> edit_list,
+    const autovector<MemTable*>& memtables_to_flush,
+    LogsWithPrepTracker* prep_tracker);
+
+// `cfd_to_flush` is the column family whose memtable will be flushed and thus
+// will not depend on any WAL file. nullptr means no memtable is being flushed.
+// The function is only applicable to 2pc mode.
+extern uint64_t FindMinPrepLogReferencedByMemTable(
+    VersionSet* vset, const ColumnFamilyData* cfd_to_flush,
+    const autovector<MemTable*>& memtables_to_flush);
+
 // Fix user-supplied options to be reasonable
 template <class T, class V>
 static void ClipToRange(T* ptr, V minvalue, V maxvalue) {

db/db_impl_compaction_flush.cc

@@ -14,6 +14,7 @@
 #include <inttypes.h>
 
 #include "db/builder.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
@@ -24,6 +25,35 @@
 
 namespace rocksdb {
 
+bool DBImpl::EnoughRoomForCompaction(
+    const std::vector<CompactionInputFiles>& inputs,
+    bool* sfm_reserved_compact_space, LogBuffer* log_buffer) {
+  // Check if we have enough room to do the compaction
+  bool enough_room = true;
+#ifndef ROCKSDB_LITE
+  auto sfm = static_cast<SstFileManagerImpl*>(
+      immutable_db_options_.sst_file_manager.get());
+  if (sfm) {
+    enough_room = sfm->EnoughRoomForCompaction(inputs);
+    if (enough_room) {
+      *sfm_reserved_compact_space = true;
+    }
+  }
+#else
+  (void)inputs;
+  (void)sfm_reserved_compact_space;
+#endif  // ROCKSDB_LITE
+  if (!enough_room) {
+    // Just in case tests want to change the value of enough_room
+    TEST_SYNC_POINT_CALLBACK(
+        "DBImpl::BackgroundCompaction():CancelledCompaction", &enough_room);
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "Cancelled compaction because not enough room");
+    RecordTick(stats_, COMPACTION_CANCELLED, 1);
+  }
+  return enough_room;
+}
+
 Status DBImpl::SyncClosedLogs(JobContext* job_context) {
   TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
   mutex_.AssertHeld();
@@ -50,6 +80,9 @@ Status DBImpl::SyncClosedLogs(JobContext* job_context) {
                      "[JOB %d] Syncing log #%" PRIu64, job_context->job_id,
                      log->get_log_number());
       s = log->file()->Sync(immutable_db_options_.use_fsync);
+      if (!s.ok()) {
+        break;
+      }
     }
     if (s.ok()) {
       s = directories_.GetWalDir()->Fsync();
@@ -61,14 +94,7 @@ Status DBImpl::SyncClosedLogs(JobContext* job_context) {
     // "number < current_log_number".
     MarkLogsSynced(current_log_number - 1, true, s);
     if (!s.ok()) {
-      Status new_bg_error = s;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kFlush,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
-      }
+      error_handler_.SetBGError(s, BackgroundErrorReason::kFlush);
       TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Failed");
       return s;
     }
@@ -95,8 +121,7 @@ Status DBImpl::FlushMemTableToOutputFile(
       dbname_, cfd, immutable_db_options_, mutable_cf_options,
       env_options_for_compaction_, versions_.get(), &mutex_, &shutting_down_,
       snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
-      job_context, log_buffer, directories_.GetDbDir(),
-      directories_.GetDataDir(0U),
+      job_context, log_buffer, directories_.GetDbDir(), GetDataDir(cfd, 0U),
       GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
       &event_logger_, mutable_cf_options.report_bg_io_stats);
 
@@ -129,7 +154,7 @@ Status DBImpl::FlushMemTableToOutputFile(
   // and EventListener callback will be called when the db_mutex
   // is unlocked by the current thread.
   if (s.ok()) {
-    s = flush_job.Run(&file_meta);
+    s = flush_job.Run(&logs_with_prep_tracker_, &file_meta);
   } else {
     flush_job.Cancel();
   }
@@ -146,18 +171,9 @@ Status DBImpl::FlushMemTableToOutputFile(
                      cfd->current()->storage_info()->LevelSummary(&tmp));
   }
 
-  if (!s.ok() && !s.IsShutdownInProgress() &&
-      immutable_db_options_.paranoid_checks && bg_error_.ok()) {
+  if (!s.ok() && !s.IsShutdownInProgress()) {
     Status new_bg_error = s;
-    // may temporarily unlock and lock the mutex.
-    EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                          BackgroundErrorReason::kFlush,
-                                          &new_bg_error, &mutex_);
-    if (!new_bg_error.ok()) {
-      // if a bad error happened (not ShutdownInProgress), paranoid_checks is
-      // true, and the error isn't handled by callback, mark DB read-only
-      bg_error_ = new_bg_error;
-    }
+    error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
   }
   if (s.ok()) {
 #ifndef ROCKSDB_LITE
@@ -169,20 +185,14 @@ Status DBImpl::FlushMemTableToOutputFile(
     if (sfm) {
       // Notify sst_file_manager that a new file was added
       std::string file_path = MakeTableFileName(
-          immutable_db_options_.db_paths[0].path, file_meta.fd.GetNumber());
+          cfd->ioptions()->cf_paths[0].path, file_meta.fd.GetNumber());
       sfm->OnAddFile(file_path);
-      if (sfm->IsMaxAllowedSpaceReached() && bg_error_.ok()) {
-        Status new_bg_error = Status::IOError("Max allowed space was reached");
+      if (sfm->IsMaxAllowedSpaceReached()) {
+        Status new_bg_error = Status::SpaceLimit("Max allowed space was reached");
         TEST_SYNC_POINT_CALLBACK(
             "DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
             &new_bg_error);
-        // may temporarily unlock and lock the mutex.
-        EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                              BackgroundErrorReason::kFlush,
-                                              &new_bg_error, &mutex_);
-        if (!new_bg_error.ok()) {
-          bg_error_ = new_bg_error;
-        }
+        error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
       }
     }
 #endif  // ROCKSDB_LITE
@@ -214,7 +224,7 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
     info.cf_name = cfd->GetName();
     // TODO(yhchiang): make db_paths dynamic in case flush does not
     //                 go to L0 in the future.
-    info.file_path = MakeTableFileName(immutable_db_options_.db_paths[0].path,
+    info.file_path = MakeTableFileName(cfd->ioptions()->cf_paths[0].path,
                                        file_meta->fd.GetNumber());
     info.thread_id = env_->GetThreadID();
     info.job_id = job_id;
@@ -231,6 +241,12 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
   mutex_.Lock();
 // no need to signal bg_cv_ as it will be signaled at the end of the
 // flush process.
+#else
+  (void)cfd;
+  (void)file_meta;
+  (void)mutable_cf_options;
+  (void)job_id;
+  (void)prop;
 #endif  // ROCKSDB_LITE
 }
 
@@ -259,7 +275,7 @@ void DBImpl::NotifyOnFlushCompleted(ColumnFamilyData* cfd,
     info.cf_name = cfd->GetName();
     // TODO(yhchiang): make db_paths dynamic in case flush does not
     //                 go to L0 in the future.
-    info.file_path = MakeTableFileName(immutable_db_options_.db_paths[0].path,
+    info.file_path = MakeTableFileName(cfd->ioptions()->cf_paths[0].path,
                                        file_meta->fd.GetNumber());
     info.thread_id = env_->GetThreadID();
     info.job_id = job_id;
@@ -276,18 +292,25 @@ void DBImpl::NotifyOnFlushCompleted(ColumnFamilyData* cfd,
   mutex_.Lock();
   // no need to signal bg_cv_ as it will be signaled at the end of the
   // flush process.
+#else
+  (void)cfd;
+  (void)file_meta;
+  (void)mutable_cf_options;
+  (void)job_id;
+  (void)prop;
 #endif  // ROCKSDB_LITE
 }
 
 Status DBImpl::CompactRange(const CompactRangeOptions& options,
                             ColumnFamilyHandle* column_family,
                             const Slice* begin, const Slice* end) {
-  if (options.target_path_id >= immutable_db_options_.db_paths.size()) {
+  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  auto cfd = cfh->cfd();
+
+  if (options.target_path_id >= cfd->ioptions()->cf_paths.size()) {
     return Status::InvalidArgument("Invalid target path ID");
   }
 
-  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-  auto cfd = cfh->cfd();
   bool exclusive = options.exclusive_manual_compaction;
 
   bool flush_needed = true;
@@ -384,7 +407,7 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
     }
     s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
                             final_output_level, options.target_path_id,
-                            begin, end, exclusive);
+                            options.max_subcompactions, begin, end, exclusive);
   } else {
     for (int level = 0; level <= max_level_with_files; level++) {
       int output_level;
@@ -418,7 +441,7 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
         }
       }
       s = RunManualCompaction(cfd, level, output_level, options.target_path_id,
-                              begin, end, exclusive);
+                              options.max_subcompactions, begin, end, exclusive);
       if (!s.ok()) {
         break;
       }
@@ -457,14 +480,19 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
   return s;
 }
 
-Status DBImpl::CompactFiles(
-    const CompactionOptions& compact_options,
-    ColumnFamilyHandle* column_family,
-    const std::vector<std::string>& input_file_names,
-    const int output_level, const int output_path_id,
-    std::vector<std::string>* const output_file_names) {
+Status DBImpl::CompactFiles(const CompactionOptions& compact_options,
+                            ColumnFamilyHandle* column_family,
+                            const std::vector<std::string>& input_file_names,
+                            const int output_level, const int output_path_id,
+                            std::vector<std::string>* const output_file_names) {
 #ifdef ROCKSDB_LITE
-    // not supported in lite version
+  (void)compact_options;
+  (void)column_family;
+  (void)input_file_names;
+  (void)output_level;
+  (void)output_path_id;
+  (void)output_file_names;
+  // not supported in lite version
   return Status::NotSupported("Not supported in ROCKSDB LITE");
 #else
   if (column_family == nullptr) {
@@ -488,9 +516,9 @@ Status DBImpl::CompactFiles(
     // IngestExternalFile() calls to finish.
     WaitForIngestFile();
 
-    s = CompactFilesImpl(compact_options, cfd, sv->current,
-                         input_file_names, output_file_names, output_level,
-                         output_path_id, &job_context, &log_buffer);
+    s = CompactFilesImpl(compact_options, cfd, sv->current, input_file_names,
+                         output_file_names, output_level, output_path_id,
+                         &job_context, &log_buffer);
   }
   if (sv->Unref()) {
     mutex_.Lock();
@@ -533,9 +561,8 @@ Status DBImpl::CompactFiles(
 Status DBImpl::CompactFilesImpl(
     const CompactionOptions& compact_options, ColumnFamilyData* cfd,
     Version* version, const std::vector<std::string>& input_file_names,
-    std::vector<std::string>* const output_file_names,
-    const int output_level, int output_path_id, JobContext* job_context,
-    LogBuffer* log_buffer) {
+    std::vector<std::string>* const output_file_names, const int output_level,
+    int output_path_id, JobContext* job_context, LogBuffer* log_buffer) {
   mutex_.AssertHeld();
 
   if (shutting_down_.load(std::memory_order_acquire)) {
@@ -553,7 +580,7 @@ Status DBImpl::CompactFilesImpl(
   version->GetColumnFamilyMetaData(&cf_meta);
 
   if (output_path_id < 0) {
-    if (immutable_db_options_.db_paths.size() == 1U) {
+    if (cfd->ioptions()->cf_paths.size() == 1U) {
       output_path_id = 0;
     } else {
       return Status::NotSupported(
@@ -582,6 +609,16 @@ Status DBImpl::CompactFilesImpl(
           "files are already being compacted");
     }
   }
+  bool sfm_reserved_compact_space = false;
+  // First check if we have enough room to do the compaction
+  bool enough_room = EnoughRoomForCompaction(
+      input_files, &sfm_reserved_compact_space, log_buffer);
+
+  if (!enough_room) {
+    // m's vars will get set properly at the end of this function,
+    // as long as status == CompactionTooLarge
+    return Status::CompactionTooLarge();
+  }
 
   // At this point, CompactFiles will be run.
   bg_compaction_scheduled_++;
@@ -615,9 +652,9 @@ Status DBImpl::CompactFilesImpl(
       job_context->job_id, c.get(), immutable_db_options_,
       env_options_for_compaction_, versions_.get(), &shutting_down_,
       preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
-      directories_.GetDataDir(c->output_path_id()), stats_, &mutex_, &bg_error_,
-      snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
-      table_cache_, &event_logger_,
+      GetDataDir(c->column_family_data(), c->output_path_id()), stats_, &mutex_,
+      &error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
+      snapshot_checker, table_cache_, &event_logger_,
       c->mutable_cf_options()->paranoid_file_checks,
       c->mutable_cf_options()->report_bg_io_stats, dbname_,
       nullptr);  // Here we pass a nullptr for CompactionJobStats because
@@ -655,9 +692,17 @@ Status DBImpl::CompactFilesImpl(
   if (status.ok()) {
     InstallSuperVersionAndScheduleWork(
         c->column_family_data(), &job_context->superversion_context,
-       *c->mutable_cf_options());
+        *c->mutable_cf_options(), FlushReason::kManualCompaction);
   }
   c->ReleaseCompactionFiles(s);
+#ifndef ROCKSDB_LITE
+  // Need to make sure SstFileManager does its bookkeeping
+  auto sfm = static_cast<SstFileManagerImpl*>(
+      immutable_db_options_.sst_file_manager.get());
+  if (sfm && sfm_reserved_compact_space) {
+    sfm->OnCompactionCompletion(c.get());
+  }
+#endif  // ROCKSDB_LITE
 
   ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);
 
@@ -670,23 +715,15 @@ Status DBImpl::CompactFilesImpl(
                    "[%s] [JOB %d] Compaction error: %s",
                    c->column_family_data()->GetName().c_str(),
                    job_context->job_id, status.ToString().c_str());
-    if (immutable_db_options_.paranoid_checks && bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kCompaction,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
-      }
-    }
+    error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
   }
 
   if (output_file_names != nullptr) {
     for (const auto newf : c->edit()->GetNewFiles()) {
-      (*output_file_names).push_back(TableFileName(
-              immutable_db_options_.db_paths, newf.second.fd.GetNumber(),
-              newf.second.fd.GetPathId()) );
+      (*output_file_names)
+          .push_back(TableFileName(c->immutable_cf_options()->cf_paths,
+                                   newf.second.fd.GetNumber(),
+                                   newf.second.fd.GetPathId()));
     }
   }
 
@@ -696,6 +733,7 @@ Status DBImpl::CompactFilesImpl(
   if (bg_compaction_scheduled_ == 0) {
     bg_cv_.SignalAll();
   }
+  TEST_SYNC_POINT("CompactFilesImpl:End");
 
   return status;
 }
@@ -730,9 +768,8 @@ Status DBImpl::ContinueBackgroundWork() {
 }
 
 void DBImpl::NotifyOnCompactionCompleted(
-    ColumnFamilyData* cfd, Compaction *c, const Status &st,
-    const CompactionJobStats& compaction_job_stats,
-    const int job_id) {
+    ColumnFamilyData* cfd, Compaction* c, const Status& st,
+    const CompactionJobStats& compaction_job_stats, const int job_id) {
 #ifndef ROCKSDB_LITE
   if (immutable_db_options_.listeners.size() == 0U) {
     return;
@@ -760,7 +797,7 @@ void DBImpl::NotifyOnCompactionCompleted(
     info.compression = c->output_compression();
     for (size_t i = 0; i < c->num_input_levels(); ++i) {
       for (const auto fmd : *c->inputs(i)) {
-        auto fn = TableFileName(immutable_db_options_.db_paths,
+        auto fn = TableFileName(c->immutable_cf_options()->cf_paths,
                                 fmd->fd.GetNumber(), fmd->fd.GetPathId());
         info.input_files.push_back(fn);
         if (info.table_properties.count(fn) == 0) {
@@ -773,9 +810,9 @@ void DBImpl::NotifyOnCompactionCompleted(
       }
     }
     for (const auto newf : c->edit()->GetNewFiles()) {
-      info.output_files.push_back(TableFileName(immutable_db_options_.db_paths,
-                                                newf.second.fd.GetNumber(),
-                                                newf.second.fd.GetPathId()));
+      info.output_files.push_back(TableFileName(
+          c->immutable_cf_options()->cf_paths, newf.second.fd.GetNumber(),
+          newf.second.fd.GetPathId()));
     }
     for (auto listener : immutable_db_options_.listeners) {
       listener->OnCompactionCompleted(this, info);
@@ -785,6 +822,12 @@ void DBImpl::NotifyOnCompactionCompleted(
   current->Unref();
   // no need to signal bg_cv_ as it will be signaled at the end of the
   // flush process.
+#else
+  (void)cfd;
+  (void)c;
+  (void)st;
+  (void)compaction_job_stats;
+  (void)job_id;
 #endif  // ROCKSDB_LITE
 }
 
@@ -881,8 +924,9 @@ int DBImpl::MaxMemCompactionLevel(ColumnFamilyHandle* /*column_family*/) {
 int DBImpl::Level0StopWriteTrigger(ColumnFamilyHandle* column_family) {
   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
   InstrumentedMutexLock l(&mutex_);
-  return cfh->cfd()->GetSuperVersion()->
-      mutable_cf_options.level0_stop_writes_trigger;
+  return cfh->cfd()
+      ->GetSuperVersion()
+      ->mutable_cf_options.level0_stop_writes_trigger;
 }
 
 Status DBImpl::Flush(const FlushOptions& flush_options,
@@ -891,7 +935,7 @@ Status DBImpl::Flush(const FlushOptions& flush_options,
   ROCKS_LOG_INFO(immutable_db_options_.info_log, "[%s] Manual flush start.",
                  cfh->GetName().c_str());
   Status s =
-      FlushMemTable(cfh->cfd(), flush_options, FlushReason::kManualCompaction);
+      FlushMemTable(cfh->cfd(), flush_options, FlushReason::kManualFlush);
   ROCKS_LOG_INFO(immutable_db_options_.info_log,
                  "[%s] Manual flush finished, status: %s\n",
                  cfh->GetName().c_str(), s.ToString().c_str());
@@ -900,6 +944,7 @@ Status DBImpl::Flush(const FlushOptions& flush_options,
 
 Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
                                    int output_level, uint32_t output_path_id,
+                                   uint32_t max_subcompactions,
                                    const Slice* begin, const Slice* end,
                                    bool exclusive, bool disallow_trivial_move) {
   assert(input_level == ColumnFamilyData::kCompactAllLevels ||
@@ -985,11 +1030,12 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
     if (ShouldntRunManualCompaction(&manual) || (manual.in_progress == true) ||
         scheduled ||
         (((manual.manual_end = &manual.tmp_storage1) != nullptr) &&
-             ((compaction = manual.cfd->CompactRange(
-                  *manual.cfd->GetLatestMutableCFOptions(), manual.input_level,
-                  manual.output_level, manual.output_path_id, manual.begin,
-                  manual.end, &manual.manual_end, &manual_conflict)) == nullptr &&
-         manual_conflict))) {
+         ((compaction = manual.cfd->CompactRange(
+               *manual.cfd->GetLatestMutableCFOptions(), manual.input_level,
+               manual.output_level, manual.output_path_id, max_subcompactions,
+               manual.begin, manual.end, &manual.manual_end,
+               &manual_conflict)) == nullptr &&
+          manual_conflict))) {
       // exclusive manual compactions should not see a conflict during
       // CompactRange
       assert(!exclusive || !manual_conflict);
@@ -1065,6 +1111,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
     // Wait until the compaction completes
     s = WaitForFlushMemTable(cfd, &flush_memtable_id);
   }
+  TEST_SYNC_POINT("FlushMemTableFinished");
   return s;
 }
 
@@ -1073,7 +1120,7 @@ Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd,
   Status s;
   // Wait until the compaction completes
   InstrumentedMutexLock l(&mutex_);
-  while (cfd->imm()->NumNotFlushed() > 0 && bg_error_.ok() &&
+  while (cfd->imm()->NumNotFlushed() > 0 && !error_handler_.IsDBStopped() &&
          (flush_memtable_id == nullptr ||
           cfd->imm()->GetEarliestMemTableID() <= *flush_memtable_id)) {
     if (shutting_down_.load(std::memory_order_acquire)) {
@@ -1087,8 +1134,8 @@ Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd,
     }
     bg_cv_.Wait();
   }
-  if (!bg_error_.ok()) {
-    s = bg_error_;
+  if (error_handler_.IsDBStopped()) {
+    s = error_handler_.GetBGError();
   }
   return s;
 }
@@ -1122,7 +1169,7 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
   }
   auto bg_job_limits = GetBGJobLimits();
   bool is_flush_pool_empty =
-    env_->GetBackgroundThreads(Env::Priority::HIGH) == 0;
+      env_->GetBackgroundThreads(Env::Priority::HIGH) == 0;
   while (!is_flush_pool_empty && unscheduled_flushes_ > 0 &&
          bg_flush_scheduled_ < bg_job_limits.max_flushes) {
     unscheduled_flushes_--;
@@ -1198,26 +1245,26 @@ DBImpl::BGJobLimits DBImpl::GetBGJobLimits(int max_background_flushes,
 }
 
 void DBImpl::AddToCompactionQueue(ColumnFamilyData* cfd) {
-  assert(!cfd->pending_compaction());
+  assert(!cfd->queued_for_compaction());
   cfd->Ref();
   compaction_queue_.push_back(cfd);
-  cfd->set_pending_compaction(true);
+  cfd->set_queued_for_compaction(true);
 }
 
 ColumnFamilyData* DBImpl::PopFirstFromCompactionQueue() {
   assert(!compaction_queue_.empty());
   auto cfd = *compaction_queue_.begin();
   compaction_queue_.pop_front();
-  assert(cfd->pending_compaction());
-  cfd->set_pending_compaction(false);
+  assert(cfd->queued_for_compaction());
+  cfd->set_queued_for_compaction(false);
   return cfd;
 }
 
 void DBImpl::AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason) {
-  assert(!cfd->pending_flush());
+  assert(!cfd->queued_for_flush());
   cfd->Ref();
   flush_queue_.push_back(cfd);
-  cfd->set_pending_flush(true);
+  cfd->set_queued_for_flush(true);
   cfd->SetFlushReason(flush_reason);
 }
 
@@ -1225,32 +1272,31 @@ ColumnFamilyData* DBImpl::PopFirstFromFlushQueue() {
   assert(!flush_queue_.empty());
   auto cfd = *flush_queue_.begin();
   flush_queue_.pop_front();
-  assert(cfd->pending_flush());
-  cfd->set_pending_flush(false);
+  assert(cfd->queued_for_flush());
+  cfd->set_queued_for_flush(false);
   // TODO: need to unset flush reason?
   return cfd;
 }
 
 void DBImpl::SchedulePendingFlush(ColumnFamilyData* cfd,
                                   FlushReason flush_reason) {
-  if (!cfd->pending_flush() && cfd->imm()->IsFlushPending()) {
+  if (!cfd->queued_for_flush() && cfd->imm()->IsFlushPending()) {
     AddToFlushQueue(cfd, flush_reason);
     ++unscheduled_flushes_;
   }
 }
 
 void DBImpl::SchedulePendingCompaction(ColumnFamilyData* cfd) {
-  if (!cfd->pending_compaction() && cfd->NeedsCompaction()) {
+  if (!cfd->queued_for_compaction() && cfd->NeedsCompaction()) {
     AddToCompactionQueue(cfd);
     ++unscheduled_compactions_;
   }
 }
 
-void DBImpl::SchedulePendingPurge(std::string fname, FileType type,
-                                  uint64_t number, uint32_t path_id,
-                                  int job_id) {
+void DBImpl::SchedulePendingPurge(std::string fname, std::string dir_to_sync,
+                                  FileType type, uint64_t number, int job_id) {
   mutex_.AssertHeld();
-  PurgeFileInfo file_info(fname, type, number, path_id, job_id);
+  PurgeFileInfo file_info(fname, dir_to_sync, type, number, job_id);
   purge_queue_.push_back(std::move(file_info));
 }
 
@@ -1308,9 +1354,13 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
                                LogBuffer* log_buffer) {
   mutex_.AssertHeld();
 
-  Status status = bg_error_;
-  if (status.ok() && shutting_down_.load(std::memory_order_acquire)) {
-    status = Status::ShutdownInProgress();
+  Status status;
+  if (!error_handler_.IsBGWorkStopped()) {
+    if (shutting_down_.load(std::memory_order_acquire)) {
+      status = Status::ShutdownInProgress();
+    }
+  } else {
+    status = error_handler_.GetBGError();
   }
 
   if (!status.ok()) {
@@ -1379,7 +1429,7 @@ void DBImpl::BackgroundCallFlush() {
       // chew up resources for failed flushes for the duration of
       // the problem.
       uint64_t error_cnt =
-        default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
+          default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
       bg_cv_.SignalAll();  // In case a waiter can proceed despite the error
       mutex_.Unlock();
       ROCKS_LOG_ERROR(immutable_db_options_.info_log,
@@ -1479,6 +1529,7 @@ void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
     // have created (they might not be all recorded in job_context in case of a
     // failure). Thus, we force full scan in FindObsoleteFiles()
     FindObsoleteFiles(&job_context, !s.ok() && !s.IsShutdownInProgress());
+    TEST_SYNC_POINT("DBImpl::BackgroundCallCompaction:FoundObsoleteFiles");
 
     // delete unnecessary files if any, this is done outside the mutex
     if (job_context.HaveSomethingToClean() ||
@@ -1492,6 +1543,7 @@ void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
       log_buffer.FlushBufferToLog();
       if (job_context.HaveSomethingToDelete()) {
         PurgeObsoleteFiles(job_context);
+        TEST_SYNC_POINT("DBImpl::BackgroundCallCompaction:PurgedObsoleteFiles");
       }
       job_context.Clean();
       mutex_.Lock();
@@ -1554,9 +1606,13 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
       is_manual && manual_compaction->disallow_trivial_move;
 
   CompactionJobStats compaction_job_stats;
-  Status status = bg_error_;
-  if (status.ok() && shutting_down_.load(std::memory_order_acquire)) {
-    status = Status::ShutdownInProgress();
+  Status status;
+  if (!error_handler_.IsBGWorkStopped()) {
+    if (shutting_down_.load(std::memory_order_acquire)) {
+      status = Status::ShutdownInProgress();
+    }
+  } else {
+    status = error_handler_.GetBGError();
   }
 
   if (!status.ok()) {
@@ -1576,9 +1632,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
 
   // InternalKey manual_end_storage;
   // InternalKey* manual_end = &manual_end_storage;
-#ifndef ROCKSDB_LITE
-  bool sfm_bookkeeping = false;
-#endif  // ROCKSDB_LITE
+  bool sfm_reserved_compact_space = false;
   if (is_manual) {
     ManualCompactionState* m = manual_compaction;
     assert(m->in_progress);
@@ -1592,16 +1646,29 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                        (m->begin ? m->begin->DebugString().c_str() : "(begin)"),
                        (m->end ? m->end->DebugString().c_str() : "(end)"));
     } else {
-      ROCKS_LOG_BUFFER(
-          log_buffer,
-          "[%s] Manual compaction from level-%d to level-%d from %s .. "
-          "%s; will stop at %s\n",
-          m->cfd->GetName().c_str(), m->input_level, c->output_level(),
-          (m->begin ? m->begin->DebugString().c_str() : "(begin)"),
-          (m->end ? m->end->DebugString().c_str() : "(end)"),
-          ((m->done || m->manual_end == nullptr)
-               ? "(end)"
-               : m->manual_end->DebugString().c_str()));
+      // First check if we have enough room to do the compaction
+      bool enough_room = EnoughRoomForCompaction(
+          *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
+
+      if (!enough_room) {
+        // Then don't do the compaction
+        c->ReleaseCompactionFiles(status);
+        c.reset();
+        // m's vars will get set properly at the end of this function,
+        // as long as status == CompactionTooLarge
+        status = Status::CompactionTooLarge();
+      } else {
+        ROCKS_LOG_BUFFER(
+            log_buffer,
+            "[%s] Manual compaction from level-%d to level-%d from %s .. "
+            "%s; will stop at %s\n",
+            m->cfd->GetName().c_str(), m->input_level, c->output_level(),
+            (m->begin ? m->begin->DebugString().c_str() : "(begin)"),
+            (m->end ? m->end->DebugString().c_str() : "(end)"),
+            ((m->done || m->manual_end == nullptr)
+                 ? "(end)"
+                 : m->manual_end->DebugString().c_str()));
+      }
     }
   } else if (!is_prepicked && !compaction_queue_.empty()) {
     if (HasExclusiveManualCompaction()) {
@@ -1642,24 +1709,10 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
       c.reset(cfd->PickCompaction(*mutable_cf_options, log_buffer));
       TEST_SYNC_POINT("DBImpl::BackgroundCompaction():AfterPickCompaction");
 
-      bool enough_room = true;
       if (c != nullptr) {
-#ifndef ROCKSDB_LITE
-        auto sfm = static_cast<SstFileManagerImpl*>(
-            immutable_db_options_.sst_file_manager.get());
-        if (sfm) {
-          enough_room = sfm->EnoughRoomForCompaction(c.get());
-          if (enough_room) {
-            sfm_bookkeeping = true;
-          }
-        }
-#endif  // ROCKSDB_LITE
-        if (!enough_room) {
-          // Just in case tests want to change the value of enough_room
-          TEST_SYNC_POINT_CALLBACK(
-              "DBImpl::BackgroundCompaction():CancelledCompaction",
-              &enough_room);
-        }
+        bool enough_room = EnoughRoomForCompaction(
+            *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
+
         if (!enough_room) {
           // Then don't do the compaction
           c->ReleaseCompactionFiles(status);
@@ -1668,9 +1721,6 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
               ->storage_info()
               ->ComputeCompactionScore(*(c->immutable_cf_options()),
                                        *(c->mutable_cf_options()));
-
-          ROCKS_LOG_BUFFER(log_buffer,
-                           "Cancelled compaction because not enough room");
           AddToCompactionQueue(cfd);
           ++unscheduled_compactions_;
 
@@ -1678,7 +1728,6 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
           // Don't need to sleep here, because BackgroundCallCompaction
           // will sleep if !s.ok()
           status = Status::CompactionTooLarge();
-          RecordTick(stats_, COMPACTION_CANCELLED, 1);
         } else {
           // update statistics
           MeasureTime(stats_, NUM_FILES_IN_SINGLE_COMPACTION,
@@ -1727,7 +1776,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                                     &mutex_, directories_.GetDbDir());
     InstallSuperVersionAndScheduleWork(
         c->column_family_data(), &job_context->superversion_context,
-        *c->mutable_cf_options());
+        *c->mutable_cf_options(), FlushReason::kAutoCompaction);
     ROCKS_LOG_BUFFER(log_buffer, "[%s] Deleted %d files\n",
                      c->column_family_data()->GetName().c_str(),
                      c->num_input_files(0));
@@ -1758,11 +1807,11 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                            f->largest, f->smallest_seqno, f->largest_seqno,
                            f->marked_for_compaction);
 
-        ROCKS_LOG_BUFFER(log_buffer, "[%s] Moving #%" PRIu64
-                                     " to level-%d %" PRIu64 " bytes\n",
-                         c->column_family_data()->GetName().c_str(),
-                         f->fd.GetNumber(), c->output_level(),
-                         f->fd.GetFileSize());
+        ROCKS_LOG_BUFFER(
+            log_buffer,
+            "[%s] Moving #%" PRIu64 " to level-%d %" PRIu64 " bytes\n",
+            c->column_family_data()->GetName().c_str(), f->fd.GetNumber(),
+            c->output_level(), f->fd.GetFileSize());
         ++moved_files;
         moved_bytes += f->fd.GetFileSize();
       }
@@ -1774,7 +1823,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     // Use latest MutableCFOptions
     InstallSuperVersionAndScheduleWork(
         c->column_family_data(), &job_context->superversion_context,
-        *c->mutable_cf_options());
+        *c->mutable_cf_options(), FlushReason::kAutoCompaction);
 
     VersionStorageInfo::LevelSummaryStorage tmp;
     c->column_family_data()->internal_stats()->IncBytesMoved(c->output_level(),
@@ -1796,9 +1845,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
 
     // Clear Instrument
     ThreadStatusUtil::ResetThreadStatus();
-  } else if (c->column_family_data()->ioptions()->compaction_style ==
-                 kCompactionStyleUniversal &&
-             !is_prepicked && c->output_level() > 0 &&
+  } else if (!is_prepicked && c->output_level() > 0 &&
              c->output_level() ==
                  c->column_family_data()
                      ->current()
@@ -1806,9 +1853,9 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                      ->MaxOutputLevel(
                          immutable_db_options_.allow_ingest_behind) &&
              env_->GetBackgroundThreads(Env::Priority::BOTTOM) > 0) {
-    // Forward universal compactions involving last level to the bottom pool
-    // if it exists, such that long-running compactions can't block short-
-    // lived ones, like L0->L0s.
+    // Forward compactions involving last level to the bottom pool if it exists,
+    // such that compactions unlikely to contribute to write stalls can be
+    // delayed or deprioritized.
     TEST_SYNC_POINT("DBImpl::BackgroundCompaction:ForwardToBottomPriPool");
     CompactionArg* ca = new CompactionArg;
     ca->db = this;
@@ -1819,7 +1866,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     env_->Schedule(&DBImpl::BGWorkBottomCompaction, ca, Env::Priority::BOTTOM,
                    this, &DBImpl::UnscheduleCallback);
   } else {
-    int output_level  __attribute__((__unused__));
+    int output_level __attribute__((__unused__));
     output_level = c->output_level();
     TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:NonTrivial",
                              &output_level);
@@ -1836,8 +1883,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
         job_context->job_id, c.get(), immutable_db_options_,
         env_options_for_compaction_, versions_.get(), &shutting_down_,
         preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
-        directories_.GetDataDir(c->output_path_id()), stats_, &mutex_,
-        &bg_error_, snapshot_seqs, earliest_write_conflict_snapshot,
+        GetDataDir(c->column_family_data(), c->output_path_id()), stats_,
+        &mutex_, &error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
         snapshot_checker, table_cache_, &event_logger_,
         c->mutable_cf_options()->paranoid_file_checks,
         c->mutable_cf_options()->report_bg_io_stats, dbname_,
@@ -1853,7 +1900,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     if (status.ok()) {
       InstallSuperVersionAndScheduleWork(
           c->column_family_data(), &job_context->superversion_context,
-          *c->mutable_cf_options());
+          *c->mutable_cf_options(), FlushReason::kAutoCompaction);
     }
     *made_progress = true;
   }
@@ -1865,14 +1912,13 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     // Need to make sure SstFileManager does its bookkeeping
     auto sfm = static_cast<SstFileManagerImpl*>(
         immutable_db_options_.sst_file_manager.get());
-    if (sfm && sfm_bookkeeping) {
+    if (sfm && sfm_reserved_compact_space) {
       sfm->OnCompactionCompletion(c.get());
     }
 #endif  // ROCKSDB_LITE
 
-    NotifyOnCompactionCompleted(
-        c->column_family_data(), c.get(), status,
-        compaction_job_stats, job_context->job_id);
+    NotifyOnCompactionCompleted(c->column_family_data(), c.get(), status,
+                                compaction_job_stats, job_context->job_id);
   }
   // this will unref its input_version and column_family_data
   c.reset();
@@ -1884,16 +1930,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
   } else {
     ROCKS_LOG_WARN(immutable_db_options_.info_log, "Compaction error: %s",
                    status.ToString().c_str());
-    if (immutable_db_options_.paranoid_checks && bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kCompaction,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
-      }
-    }
+    error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
   }
 
   if (is_manual) {
@@ -1930,7 +1967,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
       m->begin = &m->tmp_storage;
       m->incomplete = true;
     }
-    m->in_progress = false; // not being processed anymore
+    m->in_progress = false;  // not being processed anymore
   }
   TEST_SYNC_POINT("DBImpl::BackgroundCompaction:Finish");
   return status;
@@ -2030,8 +2067,7 @@ bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {
 }
 
 // SuperVersionContext gets created and destructed outside of the lock --
-// we
-// use this convinently to:
+// we use this conveniently to:
 // * malloc one SuperVersion() outside of the lock -- new_superversion
 // * delete SuperVersion()s outside of the lock -- superversions_to_free
 //
@@ -2044,7 +2080,7 @@ bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {
 
 void DBImpl::InstallSuperVersionAndScheduleWork(
     ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-    const MutableCFOptions& mutable_cf_options) {
+    const MutableCFOptions& mutable_cf_options, FlushReason flush_reason) {
   mutex_.AssertHeld();
 
   // Update max_total_in_memory_state_
@@ -2058,20 +2094,49 @@ void DBImpl::InstallSuperVersionAndScheduleWork(
   if (sv_context->new_superversion == nullptr) {
     sv_context->NewSuperVersion();
   }
-  cfd->InstallSuperVersion(sv_context, &mutex_,
-                           mutable_cf_options);
+  cfd->InstallSuperVersion(sv_context, &mutex_, mutable_cf_options);
 
   // Whenever we install new SuperVersion, we might need to issue new flushes or
   // compactions.
-  SchedulePendingFlush(cfd, FlushReason::kSuperVersionChange);
+  SchedulePendingFlush(cfd, flush_reason);
   SchedulePendingCompaction(cfd);
   MaybeScheduleFlushOrCompaction();
 
   // Update max_total_in_memory_state_
-  max_total_in_memory_state_ =
-      max_total_in_memory_state_ - old_memtable_size +
-      mutable_cf_options.write_buffer_size *
-      mutable_cf_options.max_write_buffer_number;
+  max_total_in_memory_state_ = max_total_in_memory_state_ - old_memtable_size +
+                               mutable_cf_options.write_buffer_size *
+                                   mutable_cf_options.max_write_buffer_number;
+}
+
+// ShouldPurge is called by FindObsoleteFiles when doing a full scan,
+// and db mutex (mutex_) should already be held. This function performs a
+// linear scan of an vector (files_grabbed_for_purge_) in search of a
+// certain element. We expect FindObsoleteFiles with full scan to occur once
+// every 10 hours by default, and the size of the vector is small.
+// Therefore, the cost is affordable even if the mutex is held.
+// Actually, the current implementation of FindObsoleteFiles with
+// full_scan=true can issue I/O requests to obtain list of files in
+// directories, e.g. env_->getChildren while holding db mutex.
+// In the future, if we want to reduce the cost of search, we may try to keep
+// the vector sorted.
+bool DBImpl::ShouldPurge(uint64_t file_number) const {
+  for (auto fn : files_grabbed_for_purge_) {
+    if (file_number == fn) {
+      return false;
+    }
+  }
+  for (const auto& purge_file_info : purge_queue_) {
+    if (purge_file_info.number == file_number) {
+      return false;
+    }
+  }
+  return true;
+}
+
+// MarkAsGrabbedForPurge is called by FindObsoleteFiles, and db mutex
+// (mutex_) should already be held.
+void DBImpl::MarkAsGrabbedForPurge(uint64_t file_number) {
+  files_grabbed_for_purge_.emplace_back(file_number);
 }
 
 void DBImpl::SetSnapshotChecker(SnapshotChecker* snapshot_checker) {

db/db_impl_debug.cc

@@ -10,6 +10,7 @@
 #ifndef NDEBUG
 
 #include "db/db_impl.h"
+#include "db/error_handler.h"
 #include "monitoring/thread_status_updater.h"
 
 namespace rocksdb {
@@ -25,6 +26,12 @@ void DBImpl::TEST_SwitchWAL() {
   SwitchWAL(&write_context);
 }
 
+bool DBImpl::TEST_WALBufferIsEmpty() {
+  InstrumentedMutexLock wl(&log_write_mutex_);
+  log::Writer* cur_log_writer = logs_.back().writer;
+  return cur_log_writer->TEST_BufferIsEmpty();
+}
+
 int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
     ColumnFamilyHandle* column_family) {
   ColumnFamilyData* cfd;
@@ -80,7 +87,7 @@ Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
        cfd->ioptions()->compaction_style == kCompactionStyleFIFO)
           ? level
           : level + 1;
-  return RunManualCompaction(cfd, level, output_level, 0, begin, end, true,
+  return RunManualCompaction(cfd, level, output_level, 0, 0, begin, end, true,
                              disallow_trivial_move);
 }
 
@@ -128,10 +135,10 @@ Status DBImpl::TEST_WaitForCompact(bool wait_unscheduled) {
   while ((bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
           bg_flush_scheduled_ ||
           (wait_unscheduled && unscheduled_compactions_)) &&
-         bg_error_.ok()) {
+         !error_handler_.IsDBStopped()) {
     bg_cv_.Wait();
   }
-  return bg_error_;
+  return error_handler_.GetBGError();
 }
 
 void DBImpl::TEST_LockMutex() {
@@ -184,17 +191,21 @@ Status DBImpl::TEST_GetAllImmutableCFOptions(
 }
 
 uint64_t DBImpl::TEST_FindMinLogContainingOutstandingPrep() {
-  return FindMinLogContainingOutstandingPrep();
+  return logs_with_prep_tracker_.FindMinLogContainingOutstandingPrep();
 }
 
 size_t DBImpl::TEST_PreparedSectionCompletedSize() {
-  return prepared_section_completed_.size();
+  return logs_with_prep_tracker_.TEST_PreparedSectionCompletedSize();
 }
 
-size_t DBImpl::TEST_LogsWithPrepSize() { return logs_with_prep_.size(); }
+size_t DBImpl::TEST_LogsWithPrepSize() {
+  return logs_with_prep_tracker_.TEST_LogsWithPrepSize();
+}
 
 uint64_t DBImpl::TEST_FindMinPrepLogReferencedByMemTable() {
-  return FindMinPrepLogReferencedByMemTable();
+  autovector<MemTable*> empty_list;
+  return FindMinPrepLogReferencedByMemTable(versions_.get(), nullptr,
+                                            empty_list);
 }
 
 Status DBImpl::TEST_GetLatestMutableCFOptions(

db/db_impl_files.cc

@@ -12,125 +12,20 @@
 #define __STDC_FORMAT_MACROS
 #endif
 #include <inttypes.h>
+#include <set>
+#include <unordered_set>
 #include "db/event_helpers.h"
+#include "db/memtable_list.h"
 #include "util/file_util.h"
 #include "util/sst_file_manager_impl.h"
 
-
 namespace rocksdb {
-uint64_t DBImpl::FindMinPrepLogReferencedByMemTable() {
-  if (!allow_2pc()) {
-    return 0;
-  }
-
-  uint64_t min_log = 0;
-
-  // we must look through the memtables for two phase transactions
-  // that have been committed but not yet flushed
-  for (auto loop_cfd : *versions_->GetColumnFamilySet()) {
-    if (loop_cfd->IsDropped()) {
-      continue;
-    }
-
-    auto log = loop_cfd->imm()->GetMinLogContainingPrepSection();
-
-    if (log > 0 && (min_log == 0 || log < min_log)) {
-      min_log = log;
-    }
-
-    log = loop_cfd->mem()->GetMinLogContainingPrepSection();
-
-    if (log > 0 && (min_log == 0 || log < min_log)) {
-      min_log = log;
-    }
-  }
-
-  return min_log;
-}
-
-void DBImpl::MarkLogAsHavingPrepSectionFlushed(uint64_t log) {
-  assert(log != 0);
-  std::lock_guard<std::mutex> lock(prepared_section_completed_mutex_);
-  auto it = prepared_section_completed_.find(log);
-  if (UNLIKELY(it == prepared_section_completed_.end())) {
-    prepared_section_completed_[log] = 1;
-  } else {
-    it->second += 1;
-  }
-}
-
-void DBImpl::MarkLogAsContainingPrepSection(uint64_t log) {
-  assert(log != 0);
-  std::lock_guard<std::mutex> lock(logs_with_prep_mutex_);
-
-  auto rit = logs_with_prep_.rbegin();
-  bool updated = false;
-  // Most probably the last log is the one that is being marked for
-  // having a prepare section; so search from the end.
-  for (; rit != logs_with_prep_.rend() && rit->log >= log; ++rit) {
-    if (rit->log == log) {
-      rit->cnt++;
-      updated = true;
-      break;
-    }
-  }
-  if (!updated) {
-    // We are either at the start, or at a position with rit->log < log
-    logs_with_prep_.insert(rit.base(), {log, 1});
-  }
-}
-
-uint64_t DBImpl::FindMinLogContainingOutstandingPrep() {
-  std::lock_guard<std::mutex> lock(logs_with_prep_mutex_);
-  auto it = logs_with_prep_.begin();
-  // start with the smallest log
-  for (; it != logs_with_prep_.end();) {
-    auto min_log = it->log;
-    {
-      std::lock_guard<std::mutex> lock2(prepared_section_completed_mutex_);
-      auto completed_it = prepared_section_completed_.find(min_log);
-      if (completed_it == prepared_section_completed_.end() ||
-          completed_it->second < it->cnt) {
-        return min_log;
-      }
-      assert(completed_it != prepared_section_completed_.end() &&
-             completed_it->second == it->cnt);
-      prepared_section_completed_.erase(completed_it);
-    }
-    // erase from beginning in vector is not efficient but this function is not
-    // on the fast path.
-    it = logs_with_prep_.erase(it);
-  }
-  // no such log found
-  return 0;
-}
-
 uint64_t DBImpl::MinLogNumberToKeep() {
-  uint64_t log_number = versions_->MinLogNumber();
-
   if (allow_2pc()) {
-    // if are 2pc we must consider logs containing prepared
-    // sections of outstanding transactions.
-    //
-    // We must check min logs with outstanding prep before we check
-    // logs references by memtables because a log referenced by the
-    // first data structure could transition to the second under us.
-    //
-    // TODO(horuff): iterating over all column families under db mutex.
-    // should find more optimal solution
-    auto min_log_in_prep_heap = FindMinLogContainingOutstandingPrep();
-
-    if (min_log_in_prep_heap != 0 && min_log_in_prep_heap < log_number) {
-      log_number = min_log_in_prep_heap;
-    }
-
-    auto min_log_refed_by_mem = FindMinPrepLogReferencedByMemTable();
-
-    if (min_log_refed_by_mem != 0 && min_log_refed_by_mem < log_number) {
-      log_number = min_log_refed_by_mem;
-    }
+    return versions_->min_log_number_to_keep_2pc();
+  } else {
+    return versions_->MinLogNumberWithUnflushedData();
   }
-  return log_number;
 }
 
 // * Returns the list of live files in 'sst_live'
@@ -187,27 +82,69 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
                               &job_context->manifest_delete_files,
                               job_context->min_pending_output);
 
+  // Mark the elements in job_context->sst_delete_files as grabbedForPurge
+  // so that other threads calling FindObsoleteFiles with full_scan=true
+  // will not add these files to candidate list for purge.
+  for (const auto& sst_to_del : job_context->sst_delete_files) {
+    MarkAsGrabbedForPurge(sst_to_del.metadata->fd.GetNumber());
+  }
+
   // store the current filenum, lognum, etc
   job_context->manifest_file_number = versions_->manifest_file_number();
   job_context->pending_manifest_file_number =
       versions_->pending_manifest_file_number();
   job_context->log_number = MinLogNumberToKeep();
-
   job_context->prev_log_number = versions_->prev_log_number();
 
   versions_->AddLiveFiles(&job_context->sst_live);
   if (doing_the_full_scan) {
+    InfoLogPrefix info_log_prefix(!immutable_db_options_.db_log_dir.empty(),
+        dbname_);
+    std::set<std::string> paths;
     for (size_t path_id = 0; path_id < immutable_db_options_.db_paths.size();
          path_id++) {
+      paths.insert(immutable_db_options_.db_paths[path_id].path);
+    }
+
+    // Note that if cf_paths is not specified in the ColumnFamilyOptions
+    // of a particular column family, we use db_paths as the cf_paths
+    // setting. Hence, there can be multiple duplicates of files from db_paths
+    // in the following code. The duplicate are removed while identifying
+    // unique files in PurgeObsoleteFiles.
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      for (size_t path_id = 0; path_id < cfd->ioptions()->cf_paths.size();
+           path_id++) {
+        auto& path = cfd->ioptions()->cf_paths[path_id].path;
+
+        if (paths.find(path) == paths.end()) {
+          paths.insert(path);
+        }
+      }
+    }
+
+    for (auto& path : paths) {
       // set of all files in the directory. We'll exclude files that are still
       // alive in the subsequent processings.
       std::vector<std::string> files;
-      env_->GetChildren(immutable_db_options_.db_paths[path_id].path,
-                        &files);  // Ignore errors
-      for (std::string file : files) {
+      env_->GetChildren(path, &files);  // Ignore errors
+      for (const std::string& file : files) {
+        uint64_t number;
+        FileType type;
+        // 1. If we cannot parse the file name, we skip;
+        // 2. If the file with file_number equals number has already been
+        // grabbed for purge by another compaction job, or it has already been
+        // schedule for purge, we also skip it if we
+        // are doing full scan in order to avoid double deletion of the same
+        // file under race conditions. See
+        // https://github.com/facebook/rocksdb/issues/3573
+        if (!ParseFileName(file, &number, info_log_prefix.prefix, &type) ||
+            !ShouldPurge(number)) {
+          continue;
+        }
+
         // TODO(icanadi) clean up this mess to avoid having one-off "/" prefixes
         job_context->full_scan_candidate_files.emplace_back(
-            "/" + file, static_cast<uint32_t>(path_id));
+            "/" + file, path);
       }
     }
 
@@ -216,8 +153,9 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
       std::vector<std::string> log_files;
       env_->GetChildren(immutable_db_options_.wal_dir,
                         &log_files);  // Ignore errors
-      for (std::string log_file : log_files) {
-        job_context->full_scan_candidate_files.emplace_back(log_file, 0);
+      for (const std::string& log_file : log_files) {
+        job_context->full_scan_candidate_files.emplace_back(log_file,
+            immutable_db_options_.wal_dir);
       }
     }
     // Add info log files in db_log_dir
@@ -226,8 +164,9 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
       std::vector<std::string> info_log_files;
       // Ignore errors
       env_->GetChildren(immutable_db_options_.db_log_dir, &info_log_files);
-      for (std::string log_file : info_log_files) {
-        job_context->full_scan_candidate_files.emplace_back(log_file, 0);
+      for (std::string& log_file : info_log_files) {
+        job_context->full_scan_candidate_files.emplace_back(log_file,
+            immutable_db_options_.db_log_dir);
       }
     }
   }
@@ -302,22 +241,24 @@ bool CompareCandidateFile(const JobContext::CandidateFileInfo& first,
   } else if (first.file_name < second.file_name) {
     return false;
   } else {
-    return (first.path_id > second.path_id);
+    return (first.file_path > second.file_path);
   }
 }
 };  // namespace
 
 // Delete obsolete files and log status and information of file deletion
 void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
-                                    FileType type, uint64_t number,
-                                    uint32_t path_id) {
+                                    const std::string& path_to_sync,
+                                    FileType type, uint64_t number) {
   Status file_deletion_status;
   if (type == kTableFile) {
     file_deletion_status =
-        DeleteSSTFile(&immutable_db_options_, fname, path_id);
+        DeleteSSTFile(&immutable_db_options_, fname, path_to_sync);
   } else {
     file_deletion_status = env_->DeleteFile(fname);
   }
+  TEST_SYNC_POINT_CALLBACK("DBImpl::DeleteObsoleteFileImpl:AfterDeletion",
+      &file_deletion_status);
   if (file_deletion_status.ok()) {
     ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
                     "[JOB %d] Delete %s type=%d #%" PRIu64 " -- %s\n", job_id,
@@ -347,7 +288,7 @@ void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
 // belong to live files are possibly removed. Also, removes all the
 // files in sst_delete_files and log_delete_files.
 // It is not necessary to hold the mutex when invoking this method.
-void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
+void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
   TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:Begin");
   // we'd better have sth to delete
   assert(state.HaveSomethingToDelete());
@@ -370,23 +311,23 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
       state.log_delete_files.size() + state.manifest_delete_files.size());
   // We may ignore the dbname when generating the file names.
   const char* kDumbDbName = "";
-  for (auto file : state.sst_delete_files) {
+  for (auto& file : state.sst_delete_files) {
     candidate_files.emplace_back(
-        MakeTableFileName(kDumbDbName, file->fd.GetNumber()),
-        file->fd.GetPathId());
-    if (file->table_reader_handle) {
-      table_cache_->Release(file->table_reader_handle);
+        MakeTableFileName(kDumbDbName, file.metadata->fd.GetNumber()), file.path);
+    if (file.metadata->table_reader_handle) {
+      table_cache_->Release(file.metadata->table_reader_handle);
     }
-    delete file;
+    file.DeleteMetadata();
   }
 
   for (auto file_num : state.log_delete_files) {
     if (file_num > 0) {
-      candidate_files.emplace_back(LogFileName(kDumbDbName, file_num), 0);
+      candidate_files.emplace_back(LogFileName(kDumbDbName, file_num),
+          immutable_db_options_.wal_dir);
     }
   }
   for (const auto& filename : state.manifest_delete_files) {
-    candidate_files.emplace_back(filename, 0);
+    candidate_files.emplace_back(filename, dbname_);
   }
 
   // dedup state.candidate_files so we don't try to delete the same
@@ -409,9 +350,32 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
   std::vector<std::string> old_info_log_files;
   InfoLogPrefix info_log_prefix(!immutable_db_options_.db_log_dir.empty(),
                                 dbname_);
+
+  // File numbers of most recent two OPTIONS file in candidate_files (found in
+  // previos FindObsoleteFiles(full_scan=true))
+  // At this point, there must not be any duplicate file numbers in
+  // candidate_files.
+  uint64_t optsfile_num1 = std::numeric_limits<uint64_t>::min();
+  uint64_t optsfile_num2 = std::numeric_limits<uint64_t>::min();
   for (const auto& candidate_file : candidate_files) {
-    std::string to_delete = candidate_file.file_name;
-    uint32_t path_id = candidate_file.path_id;
+    const std::string& fname = candidate_file.file_name;
+    uint64_t number;
+    FileType type;
+    if (!ParseFileName(fname, &number, info_log_prefix.prefix, &type) ||
+        type != kOptionsFile) {
+      continue;
+    }
+    if (number > optsfile_num1) {
+      optsfile_num2 = optsfile_num1;
+      optsfile_num1 = number;
+    } else if (number > optsfile_num2) {
+      optsfile_num2 = number;
+    }
+  }
+
+  std::unordered_set<uint64_t> files_to_del;
+  for (const auto& candidate_file : candidate_files) {
+    const std::string& to_delete = candidate_file.file_name;
     uint64_t number;
     FileType type;
     // Ignore file if we cannot recognize it.
@@ -437,6 +401,9 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
         // DontDeletePendingOutputs fail
         keep = (sst_live_map.find(number) != sst_live_map.end()) ||
                number >= state.min_pending_output;
+        if (!keep) {
+          files_to_del.insert(number);
+        }
         break;
       case kTempFile:
         // Any temp files that are currently being written to must
@@ -457,11 +424,19 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
           old_info_log_files.push_back(to_delete);
         }
         break;
+      case kOptionsFile:
+        keep = (number >= optsfile_num2);
+        TEST_SYNC_POINT_CALLBACK(
+            "DBImpl::PurgeObsoleteFiles:CheckOptionsFiles:1",
+            reinterpret_cast<void*>(&number));
+        TEST_SYNC_POINT_CALLBACK(
+            "DBImpl::PurgeObsoleteFiles:CheckOptionsFiles:2",
+            reinterpret_cast<void*>(&keep));
+        break;
       case kCurrentFile:
       case kDBLockFile:
       case kIdentityFile:
       case kMetaDatabase:
-      case kOptionsFile:
       case kBlobFile:
         keep = true;
         break;
@@ -472,13 +447,16 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
     }
 
     std::string fname;
+    std::string dir_to_sync;
     if (type == kTableFile) {
       // evict from cache
       TableCache::Evict(table_cache_.get(), number);
-      fname = TableFileName(immutable_db_options_.db_paths, number, path_id);
+      fname = MakeTableFileName(candidate_file.file_path, number);
+      dir_to_sync = candidate_file.file_path;
     } else {
-      fname = ((type == kLogFile) ? immutable_db_options_.wal_dir : dbname_) +
-              "/" + to_delete;
+      dir_to_sync =
+          (type == kLogFile) ? immutable_db_options_.wal_dir : dbname_;
+      fname = dir_to_sync + "/" + to_delete;
     }
 
 #ifndef ROCKSDB_LITE
@@ -492,12 +470,25 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
     Status file_deletion_status;
     if (schedule_only) {
       InstrumentedMutexLock guard_lock(&mutex_);
-      SchedulePendingPurge(fname, type, number, path_id, state.job_id);
+      SchedulePendingPurge(fname, dir_to_sync, type, number, state.job_id);
     } else {
-      DeleteObsoleteFileImpl(state.job_id, fname, type, number, path_id);
+      DeleteObsoleteFileImpl(state.job_id, fname, dir_to_sync, type, number);
     }
   }
 
+  {
+    // After purging obsolete files, remove them from files_grabbed_for_purge_.
+    // Use a temporary vector to perform bulk deletion via swap.
+    InstrumentedMutexLock guard_lock(&mutex_);
+    std::vector<uint64_t> tmp;
+    for (auto fn : files_grabbed_for_purge_) {
+      if (files_to_del.count(fn) == 0) {
+        tmp.emplace_back(fn);
+      }
+    }
+    files_grabbed_for_purge_.swap(tmp);
+  }
+
   // Delete old info log files.
   size_t old_info_log_file_count = old_info_log_files.size();
   if (old_info_log_file_count != 0 &&
@@ -557,4 +548,95 @@ void DBImpl::DeleteObsoleteFiles() {
   job_context.Clean();
   mutex_.Lock();
 }
+
+uint64_t FindMinPrepLogReferencedByMemTable(
+    VersionSet* vset, const ColumnFamilyData* cfd_to_flush,
+    const autovector<MemTable*>& memtables_to_flush) {
+  uint64_t min_log = 0;
+
+  // we must look through the memtables for two phase transactions
+  // that have been committed but not yet flushed
+  for (auto loop_cfd : *vset->GetColumnFamilySet()) {
+    if (loop_cfd->IsDropped() || loop_cfd == cfd_to_flush) {
+      continue;
+    }
+
+    auto log = loop_cfd->imm()->PrecomputeMinLogContainingPrepSection(
+        memtables_to_flush);
+
+    if (log > 0 && (min_log == 0 || log < min_log)) {
+      min_log = log;
+    }
+
+    log = loop_cfd->mem()->GetMinLogContainingPrepSection();
+
+    if (log > 0 && (min_log == 0 || log < min_log)) {
+      min_log = log;
+    }
+  }
+
+  return min_log;
+}
+
+uint64_t PrecomputeMinLogNumberToKeep(
+    VersionSet* vset, const ColumnFamilyData& cfd_to_flush,
+    autovector<VersionEdit*> edit_list,
+    const autovector<MemTable*>& memtables_to_flush,
+    LogsWithPrepTracker* prep_tracker) {
+  assert(vset != nullptr);
+  assert(prep_tracker != nullptr);
+  // Calculate updated min_log_number_to_keep
+  // Since the function should only be called in 2pc mode, log number in
+  // the version edit should be sufficient.
+
+  // Precompute the min log number containing unflushed data for the column
+  // family being flushed (`cfd_to_flush`).
+  uint64_t cf_min_log_number_to_keep = 0;
+  for (auto& e : edit_list) {
+    if (e->has_log_number()) {
+      cf_min_log_number_to_keep =
+          std::max(cf_min_log_number_to_keep, e->log_number());
+    }
+  }
+  if (cf_min_log_number_to_keep == 0) {
+    // No version edit contains information on log number. The log number
+    // for this column family should stay the same as it is.
+    cf_min_log_number_to_keep = cfd_to_flush.GetLogNumber();
+  }
+
+  // Get min log number containing unflushed data for other column families.
+  uint64_t min_log_number_to_keep =
+      vset->PreComputeMinLogNumberWithUnflushedData(&cfd_to_flush);
+  if (cf_min_log_number_to_keep != 0) {
+    min_log_number_to_keep =
+        std::min(cf_min_log_number_to_keep, min_log_number_to_keep);
+  }
+
+  // if are 2pc we must consider logs containing prepared
+  // sections of outstanding transactions.
+  //
+  // We must check min logs with outstanding prep before we check
+  // logs references by memtables because a log referenced by the
+  // first data structure could transition to the second under us.
+  //
+  // TODO: iterating over all column families under db mutex.
+  // should find more optimal solution
+  auto min_log_in_prep_heap =
+      prep_tracker->FindMinLogContainingOutstandingPrep();
+
+  if (min_log_in_prep_heap != 0 &&
+      min_log_in_prep_heap < min_log_number_to_keep) {
+    min_log_number_to_keep = min_log_in_prep_heap;
+  }
+
+  uint64_t min_log_refed_by_mem = FindMinPrepLogReferencedByMemTable(
+      vset, &cfd_to_flush, memtables_to_flush);
+
+  if (min_log_refed_by_mem != 0 &&
+      min_log_refed_by_mem < min_log_number_to_keep) {
+    min_log_number_to_keep = min_log_refed_by_mem;
+  }
+  return min_log_number_to_keep;
+}
+
 }  // namespace rocksdb

db/db_impl_open.cc

@@ -14,6 +14,7 @@
 #include <inttypes.h>
 
 #include "db/builder.h"
+#include "db/error_handler.h"
 #include "options/options_helper.h"
 #include "rocksdb/wal_filter.h"
 #include "table/block_based_table_factory.h"
@@ -106,14 +107,14 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
   }
 
-  if (result.use_direct_io_for_flush_and_compaction &&
+  if (result.use_direct_reads &&
       result.compaction_readahead_size == 0) {
     TEST_SYNC_POINT_CALLBACK("SanitizeOptions:direct_io", nullptr);
     result.compaction_readahead_size = 1024 * 1024 * 2;
   }
 
   if (result.compaction_readahead_size > 0 ||
-      result.use_direct_io_for_flush_and_compaction) {
+      result.use_direct_reads) {
     result.new_table_reader_for_compaction_inputs = true;
   }
 
@@ -163,28 +164,23 @@ static Status ValidateOptions(
     if (s.ok() && db_options.allow_concurrent_memtable_write) {
       s = CheckConcurrentWritesSupported(cfd.options);
     }
+    if (s.ok()) {
+      s = CheckCFPathsSupported(db_options, cfd.options);
+    }
     if (!s.ok()) {
       return s;
     }
-    if (db_options.db_paths.size() > 1) {
-      if ((cfd.options.compaction_style != kCompactionStyleUniversal) &&
-          (cfd.options.compaction_style != kCompactionStyleLevel)) {
-        return Status::NotSupported(
-            "More than one DB paths are only supported in "
-            "universal and level compaction styles. ");
-      }
-    }
-    if (cfd.options.compaction_options_fifo.ttl > 0) {
+
+    if (cfd.options.ttl > 0 || cfd.options.compaction_options_fifo.ttl > 0) {
       if (db_options.max_open_files != -1) {
         return Status::NotSupported(
-            "FIFO Compaction with TTL is only supported when files are always "
+            "TTL is only supported when files are always "
             "kept open (set max_open_files = -1). ");
       }
       if (cfd.options.table_factory->Name() !=
           BlockBasedTableFactory().Name()) {
         return Status::NotSupported(
-            "FIFO Compaction with TTL is only supported in "
-            "Block-Based Table format. ");
+            "TTL is only supported in Block-Based Table format. ");
       }
     }
   }
@@ -254,9 +250,9 @@ Status DBImpl::NewDB() {
   return s;
 }
 
-Status DBImpl::Directories::CreateAndNewDirectory(
+Status DBImpl::CreateAndNewDirectory(
     Env* env, const std::string& dirname,
-    std::unique_ptr<Directory>* directory) const {
+    std::unique_ptr<Directory>* directory) {
   // We call CreateDirIfMissing() as the directory may already exist (if we
   // are reopening a DB), when this happens we don't want creating the
   // directory to cause an error. However, we need to check if creating the
@@ -274,12 +270,12 @@ Status DBImpl::Directories::CreateAndNewDirectory(
 Status DBImpl::Directories::SetDirectories(
     Env* env, const std::string& dbname, const std::string& wal_dir,
     const std::vector<DbPath>& data_paths) {
-  Status s = CreateAndNewDirectory(env, dbname, &db_dir_);
+  Status s = DBImpl::CreateAndNewDirectory(env, dbname, &db_dir_);
   if (!s.ok()) {
     return s;
   }
   if (!wal_dir.empty() && dbname != wal_dir) {
-    s = CreateAndNewDirectory(env, wal_dir, &wal_dir_);
+    s = DBImpl::CreateAndNewDirectory(env, wal_dir, &wal_dir_);
     if (!s.ok()) {
       return s;
     }
@@ -292,7 +288,7 @@ Status DBImpl::Directories::SetDirectories(
       data_dirs_.emplace_back(nullptr);
     } else {
       std::unique_ptr<Directory> path_directory;
-      s = CreateAndNewDirectory(env, db_path, &path_directory);
+      s = DBImpl::CreateAndNewDirectory(env, db_path, &path_directory);
       if (!s.ok()) {
         return s;
       }
@@ -385,6 +381,14 @@ Status DBImpl::Recover(
   if (immutable_db_options_.paranoid_checks && s.ok()) {
     s = CheckConsistency();
   }
+  if (s.ok() && !read_only) {
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      s = cfd->AddDirectories();
+      if (!s.ok()) {
+        return s;
+      }
+    }
+  }
   if (s.ok()) {
     SequenceNumber next_sequence(kMaxSequenceNumber);
     default_cf_handle_ = new ColumnFamilyHandleImpl(
@@ -529,6 +533,13 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
   bool flushed = false;
   uint64_t corrupted_log_number = kMaxSequenceNumber;
   for (auto log_number : log_numbers) {
+    if (log_number < versions_->min_log_number_to_keep_2pc()) {
+      ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                     "Skipping log #%" PRIu64
+                     " since it is older than min log to keep #%" PRIu64,
+                     log_number, versions_->min_log_number_to_keep_2pc());
+      continue;
+    }
     // The previous incarnation may not have written any MANIFEST
     // records after allocating this log number.  So we manually
     // update the file number allocation counter in VersionSet.
@@ -567,7 +578,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
           continue;
         }
       }
-      file_reader.reset(new SequentialFileReader(std::move(file)));
+      file_reader.reset(new SequentialFileReader(std::move(file), fname));
     }
 
     // Create the log reader.
@@ -709,7 +720,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
       status = WriteBatchInternal::InsertInto(
           &batch, column_family_memtables_.get(), &flush_scheduler_, true,
           log_number, this, false /* concurrent_memtable_writes */,
-          next_sequence, &has_valid_writes, seq_per_batch_);
+          next_sequence, &has_valid_writes, seq_per_batch_, batch_per_txn_);
       MaybeIgnoreError(&status);
       if (!status.ok()) {
         // We are treating this as a failure while reading since we read valid
@@ -962,7 +973,7 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
                   meta.marked_for_compaction);
   }
 
-  InternalStats::CompactionStats stats(1);
+  InternalStats::CompactionStats stats(CompactionReason::kFlush, 1);
   stats.micros = env_->NowMicros() - start_micros;
   stats.bytes_written = meta.fd.GetFileSize();
   stats.num_output_files = 1;
@@ -993,15 +1004,16 @@ Status DB::Open(const Options& options, const std::string& dbname, DB** dbptr) {
 Status DB::Open(const DBOptions& db_options, const std::string& dbname,
                 const std::vector<ColumnFamilyDescriptor>& column_families,
                 std::vector<ColumnFamilyHandle*>* handles, DB** dbptr) {
-  const bool seq_per_batch = true;
+  const bool kSeqPerBatch = true;
+  const bool kBatchPerTxn = true;
   return DBImpl::Open(db_options, dbname, column_families, handles, dbptr,
-                      !seq_per_batch);
+                      !kSeqPerBatch, kBatchPerTxn);
 }
 
 Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
                     const std::vector<ColumnFamilyDescriptor>& column_families,
                     std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
-                    const bool seq_per_batch) {
+                    const bool seq_per_batch, const bool batch_per_txn) {
   Status s = SanitizeOptionsByTable(db_options, column_families);
   if (!s.ok()) {
     return s;
@@ -1021,11 +1033,20 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
         std::max(max_write_buffer_size, cf.options.write_buffer_size);
   }
 
-  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch);
+  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch, batch_per_txn);
   s = impl->env_->CreateDirIfMissing(impl->immutable_db_options_.wal_dir);
   if (s.ok()) {
-    for (auto db_path : impl->immutable_db_options_.db_paths) {
-      s = impl->env_->CreateDirIfMissing(db_path.path);
+    std::vector<std::string> paths;
+    for (auto& db_path : impl->immutable_db_options_.db_paths) {
+      paths.emplace_back(db_path.path);
+    }
+    for (auto& cf : column_families) {
+      for (auto& cf_path : cf.options.cf_paths) {
+        paths.emplace_back(cf_path.path);
+      }
+    }
+    for (auto& path : paths) {
+      s = impl->env_->CreateDirIfMissing(path);
       if (!s.ok()) {
         break;
       }
@@ -1071,7 +1092,8 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
             new_log_number,
             new log::Writer(
                 std::move(file_writer), new_log_number,
-                impl->immutable_db_options_.recycle_log_file_num > 0));
+                impl->immutable_db_options_.recycle_log_file_num > 0,
+                impl->immutable_db_options_.manual_wal_flush));
       }
 
       // set column family handles
@@ -1168,17 +1190,28 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
       impl->immutable_db_options_.sst_file_manager.get());
   if (s.ok() && sfm) {
     // Notify SstFileManager about all sst files that already exist in
-    // db_paths[0] when the DB is opened.
-    auto& db_path = impl->immutable_db_options_.db_paths[0];
-    std::vector<std::string> existing_files;
-    impl->immutable_db_options_.env->GetChildren(db_path.path, &existing_files);
-    for (auto& file_name : existing_files) {
-      uint64_t file_number;
-      FileType file_type;
-      std::string file_path = db_path.path + "/" + file_name;
-      if (ParseFileName(file_name, &file_number, &file_type) &&
-          file_type == kTableFile) {
-        sfm->OnAddFile(file_path);
+    // db_paths[0] and cf_paths[0] when the DB is opened.
+    std::vector<std::string> paths;
+    paths.emplace_back(impl->immutable_db_options_.db_paths[0].path);
+    for (auto& cf : column_families) {
+      if (!cf.options.cf_paths.empty()) {
+        paths.emplace_back(cf.options.cf_paths[0].path);
+      }
+    }
+    // Remove duplicate paths.
+    std::sort(paths.begin(), paths.end());
+    paths.erase(std::unique(paths.begin(), paths.end()), paths.end());
+    for (auto& path : paths) {
+      std::vector<std::string> existing_files;
+      impl->immutable_db_options_.env->GetChildren(path, &existing_files);
+      for (auto& file_name : existing_files) {
+        uint64_t file_number;
+        FileType file_type;
+        std::string file_path = path + "/" + file_name;
+        if (ParseFileName(file_name, &file_number, &file_type) &&
+            file_type == kTableFile) {
+          sfm->OnAddFile(file_path);
+        }
       }
     }
   }
@@ -1187,6 +1220,9 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
   if (s.ok()) {
     ROCKS_LOG_INFO(impl->immutable_db_options_.info_log, "DB pointer %p", impl);
     LogFlush(impl->immutable_db_options_.info_log);
+    assert(impl->TEST_WALBufferIsEmpty());
+    // If the assert above fails then we need to FlushWAL before returning
+    // control back to the user.
     if (!persist_options_status.ok()) {
       s = Status::IOError(
           "DB::Open() failed --- Unable to persist Options file",

db/db_impl_readonly.cc

@@ -24,8 +24,7 @@ DBImplReadOnly::DBImplReadOnly(const DBOptions& db_options,
   LogFlush(immutable_db_options_.info_log);
 }
 
-DBImplReadOnly::~DBImplReadOnly() {
-}
+DBImplReadOnly::~DBImplReadOnly() {}
 
 // Implementations of the DB interface
 Status DBImplReadOnly::Get(const ReadOptions& read_options,
@@ -59,7 +58,7 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
   SequenceNumber latest_snapshot = versions_->LastSequence();
   ReadCallback* read_callback = nullptr;  // No read callback provided.
   auto db_iter = NewArenaWrappedDbIterator(
-      env_, read_options, *cfd->ioptions(),
+      env_, read_options, *cfd->ioptions(), super_version->mutable_cf_options,
       (read_options.snapshot != nullptr
            ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
                  ->number_
@@ -89,7 +88,7 @@ Status DBImplReadOnly::NewIterators(
     auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
     auto* sv = cfd->GetSuperVersion()->Ref();
     auto* db_iter = NewArenaWrappedDbIterator(
-        env_, read_options, *cfd->ioptions(),
+        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
         (read_options.snapshot != nullptr
              ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
                    ->number_
@@ -183,20 +182,21 @@ Status DB::OpenForReadOnly(
   return s;
 }
 
-#else  // !ROCKSDB_LITE
+#else   // !ROCKSDB_LITE
 
-Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
-                           DB** dbptr, bool error_if_log_file_exist) {
+Status DB::OpenForReadOnly(const Options& /*options*/,
+                           const std::string& /*dbname*/, DB** /*dbptr*/,
+                           bool /*error_if_log_file_exist*/) {
   return Status::NotSupported("Not supported in ROCKSDB_LITE.");
 }
 
 Status DB::OpenForReadOnly(
-    const DBOptions& db_options, const std::string& dbname,
-    const std::vector<ColumnFamilyDescriptor>& column_families,
-    std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
-    bool error_if_log_file_exist) {
+    const DBOptions& /*db_options*/, const std::string& /*dbname*/,
+    const std::vector<ColumnFamilyDescriptor>& /*column_families*/,
+    std::vector<ColumnFamilyHandle*>* /*handles*/, DB** /*dbptr*/,
+    bool /*error_if_log_file_exist*/) {
   return Status::NotSupported("Not supported in ROCKSDB_LITE.");
 }
 #endif  // !ROCKSDB_LITE
 
-}   // namespace rocksdb
+}  // namespace rocksdb

db/db_impl_write.cc

@@ -12,6 +12,7 @@
 #define __STDC_FORMAT_MACROS
 #endif
 #include <inttypes.h>
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "monitoring/perf_context_imp.h"
 #include "options/options_helper.h"
@@ -45,6 +46,11 @@ Status DBImpl::SingleDelete(const WriteOptions& write_options,
   return DB::SingleDelete(write_options, column_family, key);
 }
 
+void DBImpl::SetRecoverableStatePreReleaseCallback(
+    PreReleaseCallback* callback) {
+  recoverable_state_pre_release_callback_.reset(callback);
+}
+
 Status DBImpl::Write(const WriteOptions& write_options, WriteBatch* my_batch) {
   return WriteImpl(write_options, my_batch, nullptr, nullptr);
 }
@@ -117,15 +123,19 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
   write_thread_.JoinBatchGroup(&w);
   if (w.state == WriteThread::STATE_PARALLEL_MEMTABLE_WRITER) {
     // we are a non-leader in a parallel group
-    PERF_TIMER_GUARD(write_memtable_time);
 
     if (w.ShouldWriteToMemtable()) {
+      PERF_TIMER_STOP(write_pre_and_post_process_time);
+      PERF_TIMER_GUARD(write_memtable_time);
+
       ColumnFamilyMemTablesImpl column_family_memtables(
           versions_->GetColumnFamilySet());
       w.status = WriteBatchInternal::InsertInto(
           &w, w.sequence, &column_family_memtables, &flush_scheduler_,
           write_options.ignore_missing_column_families, 0 /*log_number*/, this,
           true /*concurrent_memtable_writes*/, seq_per_batch_, w.batch_cnt);
+
+      PERF_TIMER_START(write_pre_and_post_process_time);
     }
 
     if (write_thread_.CompleteParallelMemTableWriter(&w)) {
@@ -133,7 +143,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
       for (auto* writer : *(w.write_group)) {
         if (!writer->CallbackFailed() && writer->pre_release_callback) {
           assert(writer->sequence != kMaxSequenceNumber);
-          Status ws = writer->pre_release_callback->Callback(writer->sequence);
+          Status ws = writer->pre_release_callback->Callback(writer->sequence,
+                                                             disable_memtable);
           if (!ws.ok()) {
             status = ws;
             break;
@@ -184,7 +195,13 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     // With concurrent writes we do preprocess only in the write thread that
     // also does write to memtable to avoid sync issue on shared data structure
     // with the other thread
+
+    // PreprocessWrite does its own perf timing.
+    PERF_TIMER_STOP(write_pre_and_post_process_time);
+
     status = PreprocessWrite(write_options, &need_log_sync, &write_context);
+
+    PERF_TIMER_START(write_pre_and_post_process_time);
   }
   log::Writer* log_writer = logs_.back().writer;
 
@@ -195,6 +212,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
   // and protects against concurrent loggers and concurrent writes
   // into memtables
 
+  TEST_SYNC_POINT("DBImpl::WriteImpl:BeforeLeaderEnters");
   last_batch_group_size_ =
       write_thread_.EnterAsBatchGroupLeader(&w, &write_group);
 
@@ -339,7 +357,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
   PERF_TIMER_START(write_pre_and_post_process_time);
 
   if (!w.CallbackFailed()) {
-    WriteCallbackStatusCheck(status);
+    WriteStatusCheck(status);
   }
 
   if (need_log_sync) {
@@ -347,7 +365,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     MarkLogsSynced(logfile_number_, need_log_dir_sync, status);
     mutex_.Unlock();
     // Requesting sync with two_write_queues_ is expected to be very rare. We
-    // hance provide a simple implementation that is not necessarily efficient.
+    // hence provide a simple implementation that is not necessarily efficient.
     if (two_write_queues_) {
       if (manual_wal_flush_) {
         status = FlushWAL(true);
@@ -368,7 +386,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
       for (auto* writer : write_group) {
         if (!writer->CallbackFailed() && writer->pre_release_callback) {
           assert(writer->sequence != kMaxSequenceNumber);
-          Status ws = writer->pre_release_callback->Callback(writer->sequence);
+          Status ws = writer->pre_release_callback->Callback(writer->sequence,
+                                                             disable_memtable);
           if (!ws.ok()) {
             status = ws;
             break;
@@ -407,7 +426,10 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
     mutex_.Lock();
     bool need_log_sync = !write_options.disableWAL && write_options.sync;
     bool need_log_dir_sync = need_log_sync && !log_dir_synced_;
+    // PreprocessWrite does its own perf timing.
+    PERF_TIMER_STOP(write_pre_and_post_process_time);
     w.status = PreprocessWrite(write_options, &need_log_sync, &write_context);
+    PERF_TIMER_START(write_pre_and_post_process_time);
     log::Writer* log_writer = logs_.back().writer;
     mutex_.Unlock();
 
@@ -462,7 +484,7 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
     }
 
     if (!w.CallbackFailed()) {
-      WriteCallbackStatusCheck(w.status);
+      WriteStatusCheck(w.status);
     }
 
     if (need_log_sync) {
@@ -623,13 +645,15 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
   PERF_TIMER_START(write_pre_and_post_process_time);
 
   if (!w.CallbackFailed()) {
-    WriteCallbackStatusCheck(status);
+    WriteStatusCheck(status);
   }
   if (status.ok()) {
     for (auto* writer : write_group) {
       if (!writer->CallbackFailed() && writer->pre_release_callback) {
         assert(writer->sequence != kMaxSequenceNumber);
-        Status ws = writer->pre_release_callback->Callback(writer->sequence);
+        const bool DISABLE_MEMTABLE = true;
+        Status ws = writer->pre_release_callback->Callback(writer->sequence,
+                                                           DISABLE_MEMTABLE);
         if (!ws.ok()) {
           status = ws;
           break;
@@ -647,22 +671,13 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
   return status;
 }
 
-void DBImpl::WriteCallbackStatusCheck(const Status& status) {
+void DBImpl::WriteStatusCheck(const Status& status) {
   // Is setting bg_error_ enough here?  This will at least stop
   // compaction and fail any further writes.
   if (immutable_db_options_.paranoid_checks && !status.ok() &&
       !status.IsBusy() && !status.IsIncomplete()) {
     mutex_.Lock();
-    if (bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kWriteCallback,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;  // stop compaction & fail any further writes
-      }
-    }
+    error_handler_.SetBGError(status, BackgroundErrorReason::kWriteCallback);
     mutex_.Unlock();
   }
 }
@@ -675,15 +690,8 @@ void DBImpl::MemTableInsertStatusCheck(const Status& status) {
   // ignore_missing_column_families.
   if (!status.ok()) {
     mutex_.Lock();
-    assert(bg_error_.ok());
-    Status new_bg_error = status;
-    // may temporarily unlock and lock the mutex.
-    EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                          BackgroundErrorReason::kMemTable,
-                                          &new_bg_error, &mutex_);
-    if (!new_bg_error.ok()) {
-      bg_error_ = new_bg_error;  // stop compaction & fail any further writes
-    }
+    assert(!error_handler_.IsBGWorkStopped());
+    error_handler_.SetBGError(status, BackgroundErrorReason::kMemTable);
     mutex_.Unlock();
   }
 }
@@ -695,6 +703,8 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
   assert(write_context != nullptr && need_log_sync != nullptr);
   Status status;
 
+  PERF_TIMER_GUARD(write_scheduling_flushes_compactions_time);
+
   assert(!single_column_family_mode_ ||
          versions_->GetColumnFamilySet()->NumberOfColumnFamilies() == 1);
   if (UNLIKELY(status.ok() && !single_column_family_mode_ &&
@@ -711,22 +721,27 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
     status = HandleWriteBufferFull(write_context);
   }
 
-  if (UNLIKELY(status.ok() && !bg_error_.ok())) {
-    status = bg_error_;
+  if (UNLIKELY(status.ok())) {
+    status = error_handler_.GetBGError();
   }
 
   if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
     status = ScheduleFlushes(write_context);
   }
 
+  PERF_TIMER_STOP(write_scheduling_flushes_compactions_time);
+  PERF_TIMER_GUARD(write_pre_and_post_process_time);
+
   if (UNLIKELY(status.ok() && (write_controller_.IsStopped() ||
                                write_controller_.NeedsDelay()))) {
+    PERF_TIMER_STOP(write_pre_and_post_process_time);
     PERF_TIMER_GUARD(write_delay_time);
     // We don't know size of curent batch so that we always use the size
     // for previous one. It might create a fairness issue that expiration
     // might happen for smaller writes but larger writes can go through.
     // Can optimize it if it is an issue.
     status = DelayWrite(last_batch_group_size_, write_options);
+    PERF_TIMER_START(write_pre_and_post_process_time);
   }
 
   if (status.ok() && *need_log_sync) {
@@ -804,7 +819,21 @@ Status DBImpl::WriteToWAL(const WriteBatch& merged_batch,
   assert(log_size != nullptr);
   Slice log_entry = WriteBatchInternal::Contents(&merged_batch);
   *log_size = log_entry.size();
+  // When two_write_queues_ WriteToWAL has to be protected from concurretn calls
+  // from the two queues anyway and log_write_mutex_ is already held. Otherwise
+  // if manual_wal_flush_ is enabled we need to protect log_writer->AddRecord
+  // from possible concurrent calls via the FlushWAL by the application.
+  const bool needs_locking = manual_wal_flush_ && !two_write_queues_;
+  // Due to performance cocerns of missed branch prediction penalize the new
+  // manual_wal_flush_ feature (by UNLIKELY) instead of the more common case
+  // when we do not need any locking.
+  if (UNLIKELY(needs_locking)) {
+    log_write_mutex_.Lock();
+  }
   Status status = log_writer->AddRecord(log_entry);
+  if (UNLIKELY(needs_locking)) {
+    log_write_mutex_.Unlock();
+  }
   if (log_used != nullptr) {
     *log_used = logfile_number_;
   }
@@ -942,7 +971,12 @@ Status DBImpl::WriteRecoverableState() {
     if (two_write_queues_) {
       log_write_mutex_.Lock();
     }
-    SequenceNumber seq = versions_->LastSequence();
+    SequenceNumber seq;
+    if (two_write_queues_) {
+      seq = versions_->FetchAddLastAllocatedSequence(0);
+    } else {
+      seq = versions_->LastSequence();
+    }
     WriteBatchInternal::SetSequence(&cached_recoverable_state_, seq + 1);
     auto status = WriteBatchInternal::InsertInto(
         &cached_recoverable_state_, column_family_memtables_.get(),
@@ -952,12 +986,20 @@ Status DBImpl::WriteRecoverableState() {
     auto last_seq = next_seq - 1;
     if (two_write_queues_) {
       versions_->FetchAddLastAllocatedSequence(last_seq - seq);
+      versions_->SetLastPublishedSequence(last_seq);
     }
     versions_->SetLastSequence(last_seq);
-    versions_->SetLastPublishedSequence(last_seq);
     if (two_write_queues_) {
       log_write_mutex_.Unlock();
     }
+    if (status.ok() && recoverable_state_pre_release_callback_) {
+      const bool DISABLE_MEMTABLE = true;
+      for (uint64_t sub_batch_seq = seq + 1;
+           sub_batch_seq < next_seq && status.ok(); sub_batch_seq++) {
+        status = recoverable_state_pre_release_callback_->Callback(
+            sub_batch_seq, !DISABLE_MEMTABLE);
+      }
+    }
     if (status.ok()) {
       cached_recoverable_state_.Clear();
       cached_recoverable_state_empty_ = true;
@@ -977,28 +1019,34 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
   }
 
   auto oldest_alive_log = alive_log_files_.begin()->number;
-  auto oldest_log_with_uncommited_prep = FindMinLogContainingOutstandingPrep();
+  bool flush_wont_release_oldest_log = false;
+  if (allow_2pc()) {
+    auto oldest_log_with_uncommited_prep =
+        logs_with_prep_tracker_.FindMinLogContainingOutstandingPrep();
 
-  if (allow_2pc() &&
-      oldest_log_with_uncommited_prep > 0 &&
-      oldest_log_with_uncommited_prep <= oldest_alive_log) {
-    if (unable_to_flush_oldest_log_) {
+    assert(oldest_log_with_uncommited_prep == 0 ||
+           oldest_log_with_uncommited_prep >= oldest_alive_log);
+    if (oldest_log_with_uncommited_prep > 0 &&
+        oldest_log_with_uncommited_prep == oldest_alive_log) {
+      if (unable_to_release_oldest_log_) {
         // we already attempted to flush all column families dependent on
-        // the oldest alive log but the log still contained uncommited transactions.
-        // the oldest alive log STILL contains uncommited transaction so there
-        // is still nothing that we can do.
+        // the oldest alive log but the log still contained uncommited
+        // transactions so there is still nothing that we can do.
         return status;
-    } else {
-      ROCKS_LOG_WARN(
-          immutable_db_options_.info_log,
-          "Unable to release oldest log due to uncommited transaction");
-      unable_to_flush_oldest_log_ = true;
+      } else {
+        ROCKS_LOG_WARN(
+            immutable_db_options_.info_log,
+            "Unable to release oldest log due to uncommited transaction");
+        unable_to_release_oldest_log_ = true;
+        flush_wont_release_oldest_log = true;
+      }
     }
-  } else {
+  }
+  if (!flush_wont_release_oldest_log) {
     // we only mark this log as getting flushed if we have successfully
     // flushed all data in this log. If this log contains outstanding prepared
     // transactions then we cannot flush this log until those transactions are commited.
-    unable_to_flush_oldest_log_ = false;
+    unable_to_release_oldest_log_ = false;
     alive_log_files_.begin()->getting_flushed = true;
   }
 
@@ -1062,7 +1110,8 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
     }
   }
   if (cfd_picked != nullptr) {
-    status = SwitchMemtable(cfd_picked, write_context);
+    status = SwitchMemtable(cfd_picked, write_context,
+                            FlushReason::kWriteBufferFull);
     if (status.ok()) {
       cfd_picked->imm()->FlushRequested();
       SchedulePendingFlush(cfd_picked, FlushReason::kWriteBufferFull);
@@ -1112,7 +1161,7 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
       mutex_.Lock();
     }
 
-    while (bg_error_.ok() && write_controller_.IsStopped()) {
+    while (!error_handler_.IsDBStopped() && write_controller_.IsStopped()) {
       if (write_options.no_slowdown) {
         return Status::Incomplete();
       }
@@ -1128,7 +1177,7 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
     RecordTick(stats_, STALL_MICROS, time_delayed);
   }
 
-  return bg_error_;
+  return error_handler_.GetBGError();
 }
 
 Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
@@ -1152,6 +1201,7 @@ Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
       // is that in case the write is heavy, low pri writes may never have
       // a chance to run. Now we guarantee we are still slowly making
       // progress.
+      PERF_TIMER_GUARD(write_delay_time);
       write_controller_.low_pri_rate_limiter()->Request(
           my_batch->GetDataSize(), Env::IO_HIGH, nullptr /* stats */,
           RateLimiter::OpType::kWrite);
@@ -1163,7 +1213,7 @@ Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
 Status DBImpl::ScheduleFlushes(WriteContext* context) {
   ColumnFamilyData* cfd;
   while ((cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
-    auto status = SwitchMemtable(cfd, context);
+    auto status = SwitchMemtable(cfd, context, FlushReason::kWriteBufferFull);
     if (cfd->Unref()) {
       delete cfd;
     }
@@ -1192,7 +1242,8 @@ void DBImpl::NotifyOnMemTableSealed(ColumnFamilyData* /*cfd*/,
 
 // REQUIRES: mutex_ is held
 // REQUIRES: this thread is currently at the front of the writer queue
-Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
+Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
+                              FlushReason flush_reason) {
   mutex_.AssertHeld();
   WriteThread::Writer nonmem_w;
   if (two_write_queues_) {
@@ -1215,7 +1266,11 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
   // In case of pipelined write is enabled, wait for all pending memtable
   // writers.
   if (immutable_db_options_.enable_pipelined_write) {
+    // Memtable writers may call DB::Get in case max_successive_merges > 0,
+    // which may lock mutex. Unlocking mutex here to avoid deadlock.
+    mutex_.Unlock();
     write_thread_.WaitForMemTableWriters();
+    mutex_.Lock();
   }
 
   // Attempt to switch to a new memtable and trigger flush of old.
@@ -1360,7 +1415,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
   new_mem->Ref();
   cfd->SetMemtable(new_mem);
   InstallSuperVersionAndScheduleWork(cfd, &context->superversion_context,
-                                     mutable_cf_options);
+                                     mutable_cf_options, flush_reason);
   if (two_write_queues_) {
     nonmem_write_thread_.ExitUnbatched(&nonmem_w);
   }

db/db_iter.h

@@ -30,6 +30,7 @@ class InternalIterator;
 // into appropriate user keys.
 extern Iterator* NewDBIterator(Env* env, const ReadOptions& read_options,
                                const ImmutableCFOptions& cf_options,
+                               const MutableCFOptions& mutable_cf_options,
                                const Comparator* user_key_comparator,
                                InternalIterator* internal_iter,
                                const SequenceNumber& sequence,
@@ -71,6 +72,7 @@ class ArenaWrappedDBIter : public Iterator {
 
   void Init(Env* env, const ReadOptions& read_options,
             const ImmutableCFOptions& cf_options,
+            const MutableCFOptions& mutable_cf_options,
             const SequenceNumber& sequence,
             uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
             ReadCallback* read_callback, bool allow_blob, bool allow_refresh);
@@ -102,7 +104,8 @@ class ArenaWrappedDBIter : public Iterator {
 // be supported.
 extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
     Env* env, const ReadOptions& read_options,
-    const ImmutableCFOptions& cf_options, const SequenceNumber& sequence,
+    const ImmutableCFOptions& cf_options,
+    const MutableCFOptions& mutable_cf_options, const SequenceNumber& sequence,
     uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
     ReadCallback* read_callback, DBImpl* db_impl = nullptr,
     ColumnFamilyData* cfd = nullptr, bool allow_blob = false,

db/db_iter_stress_test.cc

@@ -0,0 +1,654 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "db/db_iter.h"
+#include "db/dbformat.h"
+#include "rocksdb/comparator.h"
+#include "rocksdb/options.h"
+#include "rocksdb/slice.h"
+#include "util/random.h"
+#include "util/string_util.h"
+#include "util/testharness.h"
+#include "utilities/merge_operators.h"
+
+#ifdef GFLAGS
+
+#include "util/gflags_compat.h"
+
+using GFLAGS_NAMESPACE::ParseCommandLineFlags;
+
+DEFINE_bool(verbose, false,
+            "Print huge, detailed trace. Intended for debugging failures.");
+
+#else
+
+void ParseCommandLineFlags(int*, char***, bool) {}
+bool FLAGS_verbose = false;
+
+#endif
+
+namespace rocksdb {
+
+class DBIteratorStressTest : public testing::Test {
+ public:
+  Env* env_;
+
+  DBIteratorStressTest() : env_(Env::Default()) {}
+};
+
+namespace {
+
+struct Entry {
+  std::string key;
+  ValueType type;  // kTypeValue, kTypeDeletion, kTypeMerge
+  uint64_t sequence;
+  std::string ikey;  // internal key, made from `key`, `sequence` and `type`
+  std::string value;
+  // If false, we'll pretend that this entry doesn't exist.
+  bool visible = true;
+
+  bool operator<(const Entry& e) const {
+    if (key != e.key) return key < e.key;
+    return std::tie(sequence, type) > std::tie(e.sequence, e.type);
+  }
+};
+
+struct Data {
+  std::vector<Entry> entries;
+
+  // Indices in `entries` with `visible` = false.
+  std::vector<size_t> hidden;
+  // Keys of entries whose `visible` changed since the last seek of iterators.
+  std::set<std::string> recently_touched_keys;
+};
+
+struct StressTestIterator : public InternalIterator {
+  Data* data;
+  Random64* rnd;
+  InternalKeyComparator cmp;
+
+  // Each operation will return error with this probability...
+  double error_probability = 0;
+  // ... and add/remove entries with this probability.
+  double mutation_probability = 0;
+  // The probability of adding vs removing entries will be chosen so that the
+  // amount of removed entries stays somewhat close to this number.
+  double target_hidden_fraction = 0;
+  // If true, print all mutations to stdout for debugging.
+  bool trace = false;
+
+  int iter = -1;
+  Status status_;
+
+  StressTestIterator(Data* _data, Random64* _rnd, const Comparator* _cmp)
+      : data(_data), rnd(_rnd), cmp(_cmp) {}
+
+  bool Valid() const override {
+    if (iter >= 0 && iter < (int)data->entries.size()) {
+      assert(status_.ok());
+      return true;
+    }
+    return false;
+  }
+
+  Status status() const override { return status_; }
+
+  bool MaybeFail() {
+    if (rnd->Next() >=
+        std::numeric_limits<uint64_t>::max() * error_probability) {
+      return false;
+    }
+    if (rnd->Next() % 2) {
+      status_ = Status::Incomplete("test");
+    } else {
+      status_ = Status::IOError("test");
+    }
+    if (trace) {
+      std::cout << "injecting " << status_.ToString() << std::endl;
+    }
+    iter = -1;
+    return true;
+  }
+
+  void MaybeMutate() {
+    if (rnd->Next() >=
+        std::numeric_limits<uint64_t>::max() * mutation_probability) {
+      return;
+    }
+    do {
+      // If too many entries are hidden, hide less, otherwise hide more.
+      double hide_probability =
+          data->hidden.size() > data->entries.size() * target_hidden_fraction
+              ? 1. / 3
+              : 2. / 3;
+      if (data->hidden.empty()) {
+        hide_probability = 1;
+      }
+      bool do_hide =
+          rnd->Next() < std::numeric_limits<uint64_t>::max() * hide_probability;
+      if (do_hide) {
+        // Hide a random entry.
+        size_t idx = rnd->Next() % data->entries.size();
+        Entry& e = data->entries[idx];
+        if (e.visible) {
+          if (trace) {
+            std::cout << "hiding idx " << idx << std::endl;
+          }
+          e.visible = false;
+          data->hidden.push_back(idx);
+          data->recently_touched_keys.insert(e.key);
+        } else {
+          // Already hidden. Let's go unhide something instead, just because
+          // it's easy and it doesn't really matter what we do.
+          do_hide = false;
+        }
+      }
+      if (!do_hide) {
+        // Unhide a random entry.
+        size_t hi = rnd->Next() % data->hidden.size();
+        size_t idx = data->hidden[hi];
+        if (trace) {
+          std::cout << "unhiding idx " << idx << std::endl;
+        }
+        Entry& e = data->entries[idx];
+        assert(!e.visible);
+        e.visible = true;
+        data->hidden[hi] = data->hidden.back();
+        data->hidden.pop_back();
+        data->recently_touched_keys.insert(e.key);
+      }
+    } while (rnd->Next() % 3 != 0);  // do 3 mutations on average
+  }
+
+  void SkipForward() {
+    while (iter < (int)data->entries.size() && !data->entries[iter].visible) {
+      ++iter;
+    }
+  }
+  void SkipBackward() {
+    while (iter >= 0 && !data->entries[iter].visible) {
+      --iter;
+    }
+  }
+
+  void SeekToFirst() override {
+    if (MaybeFail()) return;
+    MaybeMutate();
+
+    status_ = Status::OK();
+    iter = 0;
+    SkipForward();
+  }
+  void SeekToLast() override {
+    if (MaybeFail()) return;
+    MaybeMutate();
+
+    status_ = Status::OK();
+    iter = (int)data->entries.size() - 1;
+    SkipBackward();
+  }
+
+  void Seek(const Slice& target) override {
+    if (MaybeFail()) return;
+    MaybeMutate();
+
+    status_ = Status::OK();
+    // Binary search.
+    auto it = std::partition_point(
+        data->entries.begin(), data->entries.end(),
+        [&](const Entry& e) { return cmp.Compare(e.ikey, target) < 0; });
+    iter = (int)(it - data->entries.begin());
+    SkipForward();
+  }
+  void SeekForPrev(const Slice& target) override {
+    if (MaybeFail()) return;
+    MaybeMutate();
+
+    status_ = Status::OK();
+    // Binary search.
+    auto it = std::partition_point(
+        data->entries.begin(), data->entries.end(),
+        [&](const Entry& e) { return cmp.Compare(e.ikey, target) <= 0; });
+    iter = (int)(it - data->entries.begin());
+    --iter;
+    SkipBackward();
+  }
+
+  void Next() override {
+    assert(Valid());
+    if (MaybeFail()) return;
+    MaybeMutate();
+    ++iter;
+    SkipForward();
+  }
+  void Prev() override {
+    assert(Valid());
+    if (MaybeFail()) return;
+    MaybeMutate();
+    --iter;
+    SkipBackward();
+  }
+
+  Slice key() const override {
+    assert(Valid());
+    return data->entries[iter].ikey;
+  }
+  Slice value() const override {
+    assert(Valid());
+    return data->entries[iter].value;
+  }
+
+  bool IsKeyPinned() const override { return true; }
+  bool IsValuePinned() const override { return true; }
+};
+
+// A small reimplementation of DBIter, supporting only some of the features,
+// and doing everything in O(log n).
+// Skips all keys that are in recently_touched_keys.
+struct ReferenceIterator {
+  Data* data;
+  uint64_t sequence;  // ignore entries with sequence number below this
+
+  bool valid = false;
+  std::string key;
+  std::string value;
+
+  ReferenceIterator(Data* _data, uint64_t _sequence)
+      : data(_data), sequence(_sequence) {}
+
+  bool Valid() const { return valid; }
+
+  // Finds the first entry with key
+  // greater/less/greater-or-equal/less-or-equal than `key`, depending on
+  // arguments: if `skip`, inequality is strict; if `forward`, it's
+  // greater/greater-or-equal, otherwise less/less-or-equal.
+  // Sets `key` to the result.
+  // If no such key exists, returns false. Doesn't check `visible`.
+  bool FindNextKey(bool skip, bool forward) {
+    valid = false;
+    auto it = std::partition_point(data->entries.begin(), data->entries.end(),
+                                   [&](const Entry& e) {
+                                     if (forward != skip) {
+                                       return e.key < key;
+                                     } else {
+                                       return e.key <= key;
+                                     }
+                                   });
+    if (forward) {
+      if (it != data->entries.end()) {
+        key = it->key;
+        return true;
+      }
+    } else {
+      if (it != data->entries.begin()) {
+        --it;
+        key = it->key;
+        return true;
+      }
+    }
+    return false;
+  }
+
+  bool FindValueForCurrentKey() {
+    if (data->recently_touched_keys.count(key)) {
+      return false;
+    }
+
+    // Find the first entry for the key. The caller promises that it exists.
+    auto it = std::partition_point(data->entries.begin(), data->entries.end(),
+                                   [&](const Entry& e) {
+                                     if (e.key != key) {
+                                       return e.key < key;
+                                     }
+                                     return e.sequence > sequence;
+                                   });
+
+    // Find the first visible entry.
+    for (;; ++it) {
+      if (it == data->entries.end()) {
+        return false;
+      }
+      Entry& e = *it;
+      if (e.key != key) {
+        return false;
+      }
+      assert(e.sequence <= sequence);
+      if (!e.visible) continue;
+      if (e.type == kTypeDeletion) {
+        return false;
+      }
+      if (e.type == kTypeValue) {
+        value = e.value;
+        valid = true;
+        return true;
+      }
+      assert(e.type == kTypeMerge);
+      break;
+    }
+
+    // Collect merge operands.
+    std::vector<Slice> operands;
+    for (; it != data->entries.end(); ++it) {
+      Entry& e = *it;
+      if (e.key != key) {
+        break;
+      }
+      assert(e.sequence <= sequence);
+      if (!e.visible) continue;
+      if (e.type == kTypeDeletion) {
+        break;
+      }
+      operands.push_back(e.value);
+      if (e.type == kTypeValue) {
+        break;
+      }
+    }
+
+    // Do a merge.
+    value = operands.back().ToString();
+    for (int i = (int)operands.size() - 2; i >= 0; --i) {
+      value.append(",");
+      value.append(operands[i].data(), operands[i].size());
+    }
+
+    valid = true;
+    return true;
+  }
+
+  // Start at `key` and move until we encounter a valid value.
+  // `forward` defines the direction of movement.
+  // If `skip` is true, we're looking for key not equal to `key`.
+  void DoTheThing(bool skip, bool forward) {
+    while (FindNextKey(skip, forward) && !FindValueForCurrentKey()) {
+      skip = true;
+    }
+  }
+
+  void Seek(const Slice& target) {
+    key = target.ToString();
+    DoTheThing(false, true);
+  }
+  void SeekForPrev(const Slice& target) {
+    key = target.ToString();
+    DoTheThing(false, false);
+  }
+  void SeekToFirst() { Seek(""); }
+  void SeekToLast() {
+    key = data->entries.back().key;
+    DoTheThing(false, false);
+  }
+  void Next() {
+    assert(Valid());
+    DoTheThing(true, true);
+  }
+  void Prev() {
+    assert(Valid());
+    DoTheThing(true, false);
+  }
+};
+
+}  // namespace
+
+// Use an internal iterator that sometimes returns errors and sometimes
+// adds/removes entries on the fly. Do random operations on a DBIter and
+// check results.
+// TODO: can be improved for more coverage:
+//   * Override IsKeyPinned() and IsValuePinned() to actually use
+//     PinnedIteratorManager and check that there's no use-after free.
+//   * Try different combinations of prefix_extractor, total_order_seek,
+//     prefix_same_as_start, iterate_lower_bound, iterate_upper_bound.
+TEST_F(DBIteratorStressTest, StressTest) {
+  // We use a deterministic RNG, and everything happens in a single thread.
+  Random64 rnd(826909345792864532ll);
+
+  auto gen_key = [&](int max_key) {
+    assert(max_key > 0);
+    int len = 0;
+    int a = max_key;
+    while (a) {
+      a /= 10;
+      ++len;
+    }
+    std::string s = ToString(rnd.Next() % static_cast<uint64_t>(max_key));
+    s.insert(0, len - (int)s.size(), '0');
+    return s;
+  };
+
+  Options options;
+  options.merge_operator = MergeOperators::CreateFromStringId("stringappend");
+  ReadOptions ropt;
+
+  size_t num_matching = 0;
+  size_t num_at_end = 0;
+  size_t num_not_ok = 0;
+  size_t num_recently_removed = 0;
+
+  // Number of iterations for each combination of parameters
+  // (there are ~250 of those).
+  // Tweak this to change the test run time.
+  // As of the time of writing, the test takes ~4 seconds for value of 5000.
+  const int num_iterations = 5000;
+  // Enable this to print all the operations for debugging.
+  bool trace = FLAGS_verbose;
+
+  for (int num_entries : {5, 10, 100}) {
+    for (double key_space : {0.1, 1.0, 3.0}) {
+      for (ValueType prevalent_entry_type :
+           {kTypeValue, kTypeDeletion, kTypeMerge}) {
+        for (double error_probability : {0.01, 0.1}) {
+          for (double mutation_probability : {0.01, 0.5}) {
+            for (double target_hidden_fraction : {0.1, 0.5}) {
+              std::string trace_str =
+                  "entries: " + ToString(num_entries) +
+                  ", key_space: " + ToString(key_space) +
+                  ", error_probability: " + ToString(error_probability) +
+                  ", mutation_probability: " + ToString(mutation_probability) +
+                  ", target_hidden_fraction: " +
+                  ToString(target_hidden_fraction);
+              SCOPED_TRACE(trace_str);
+              if (trace) {
+                std::cout << trace_str << std::endl;
+              }
+
+              // Generate data.
+              Data data;
+              int max_key = (int)(num_entries * key_space) + 1;
+              for (int i = 0; i < num_entries; ++i) {
+                Entry e;
+                e.key = gen_key(max_key);
+                if (rnd.Next() % 10 != 0) {
+                  e.type = prevalent_entry_type;
+                } else {
+                  const ValueType types[] = {kTypeValue, kTypeDeletion,
+                                             kTypeMerge};
+                  e.type =
+                      types[rnd.Next() % (sizeof(types) / sizeof(types[0]))];
+                }
+                e.sequence = i;
+                e.value = "v" + ToString(i);
+                ParsedInternalKey internal_key(e.key, e.sequence, e.type);
+                AppendInternalKey(&e.ikey, internal_key);
+
+                data.entries.push_back(e);
+              }
+              std::sort(data.entries.begin(), data.entries.end());
+              if (trace) {
+                std::cout << "entries:";
+                for (size_t i = 0; i < data.entries.size(); ++i) {
+                  Entry& e = data.entries[i];
+                  std::cout
+                      << "\n  idx " << i << ": \"" << e.key << "\": \""
+                      << e.value << "\" seq: " << e.sequence << " type: "
+                      << (e.type == kTypeValue
+                              ? "val"
+                              : e.type == kTypeDeletion ? "del" : "merge");
+                }
+                std::cout << std::endl;
+              }
+
+              std::unique_ptr<Iterator> db_iter;
+              std::unique_ptr<ReferenceIterator> ref_iter;
+              for (int iteration = 0; iteration < num_iterations; ++iteration) {
+                SCOPED_TRACE(iteration);
+                // Create a new iterator every ~30 operations.
+                if (db_iter == nullptr || rnd.Next() % 30 == 0) {
+                  uint64_t sequence = rnd.Next() % (data.entries.size() + 2);
+                  ref_iter.reset(new ReferenceIterator(&data, sequence));
+                  if (trace) {
+                    std::cout << "new iterator, seq: " << sequence << std::endl;
+                  }
+
+                  auto internal_iter =
+                      new StressTestIterator(&data, &rnd, BytewiseComparator());
+                  internal_iter->error_probability = error_probability;
+                  internal_iter->mutation_probability = mutation_probability;
+                  internal_iter->target_hidden_fraction =
+                      target_hidden_fraction;
+                  internal_iter->trace = trace;
+                  db_iter.reset(NewDBIterator(
+                      env_, ropt, ImmutableCFOptions(options),
+                      MutableCFOptions(options), BytewiseComparator(),
+                      internal_iter, sequence,
+                      options.max_sequential_skip_in_iterations,
+                      nullptr /*read_callback*/));
+                }
+
+                // Do a random operation. It's important to do it on ref_it
+                // later than on db_iter to make sure ref_it sees the correct
+                // recently_touched_keys.
+                std::string old_key;
+                bool forward = rnd.Next() % 2 > 0;
+                // Do Next()/Prev() ~90% of the time.
+                bool seek = !ref_iter->Valid() || rnd.Next() % 10 == 0;
+                if (trace) {
+                  std::cout << iteration << ": ";
+                }
+
+                if (!seek) {
+                  assert(db_iter->Valid());
+                  old_key = ref_iter->key;
+                  if (trace) {
+                    std::cout << (forward ? "Next" : "Prev") << std::endl;
+                  }
+
+                  if (forward) {
+                    db_iter->Next();
+                    ref_iter->Next();
+                  } else {
+                    db_iter->Prev();
+                    ref_iter->Prev();
+                  }
+                } else {
+                  data.recently_touched_keys.clear();
+                  // Do SeekToFirst less often than Seek.
+                  if (rnd.Next() % 4 == 0) {
+                    if (trace) {
+                      std::cout << (forward ? "SeekToFirst" : "SeekToLast")
+                                << std::endl;
+                    }
+
+                    if (forward) {
+                      old_key = "";
+                      db_iter->SeekToFirst();
+                      ref_iter->SeekToFirst();
+                    } else {
+                      old_key = data.entries.back().key;
+                      db_iter->SeekToLast();
+                      ref_iter->SeekToLast();
+                    }
+                  } else {
+                    old_key = gen_key(max_key);
+                    if (trace) {
+                      std::cout << (forward ? "Seek" : "SeekForPrev") << " \""
+                                << old_key << '"' << std::endl;
+                    }
+                    if (forward) {
+                      db_iter->Seek(old_key);
+                      ref_iter->Seek(old_key);
+                    } else {
+                      db_iter->SeekForPrev(old_key);
+                      ref_iter->SeekForPrev(old_key);
+                    }
+                  }
+                }
+
+                // Check the result.
+                if (db_iter->Valid()) {
+                  ASSERT_TRUE(db_iter->status().ok());
+                  if (data.recently_touched_keys.count(
+                          db_iter->key().ToString())) {
+                    // Ended on a key that may have been mutated during the
+                    // operation. Reference iterator skips such keys, so we
+                    // can't check the exact result.
+
+                    // Check that the key moved in the right direction.
+                    if (forward) {
+                      if (seek)
+                        ASSERT_GE(db_iter->key().ToString(), old_key);
+                      else
+                        ASSERT_GT(db_iter->key().ToString(), old_key);
+                    } else {
+                      if (seek)
+                        ASSERT_LE(db_iter->key().ToString(), old_key);
+                      else
+                        ASSERT_LT(db_iter->key().ToString(), old_key);
+                    }
+
+                    if (ref_iter->Valid()) {
+                      // Check that DBIter didn't miss any non-mutated key.
+                      if (forward) {
+                        ASSERT_LT(db_iter->key().ToString(), ref_iter->key);
+                      } else {
+                        ASSERT_GT(db_iter->key().ToString(), ref_iter->key);
+                      }
+                    }
+                    // Tell the next iteration of the loop to reseek the
+                    // iterators.
+                    ref_iter->valid = false;
+
+                    ++num_recently_removed;
+                  } else {
+                    ASSERT_TRUE(ref_iter->Valid());
+                    ASSERT_EQ(ref_iter->key, db_iter->key().ToString());
+                    ASSERT_EQ(ref_iter->value, db_iter->value());
+                    ++num_matching;
+                  }
+                } else if (db_iter->status().ok()) {
+                  ASSERT_FALSE(ref_iter->Valid());
+                  ++num_at_end;
+                } else {
+                  // Non-ok status. Nothing to check here.
+                  // Tell the next iteration of the loop to reseek the
+                  // iterators.
+                  ref_iter->valid = false;
+                  ++num_not_ok;
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+
+  // Check that all cases were hit many times.
+  EXPECT_GT(num_matching, 10000);
+  EXPECT_GT(num_at_end, 10000);
+  EXPECT_GT(num_not_ok, 10000);
+  EXPECT_GT(num_recently_removed, 10000);
+
+  std::cout << "stats:\n  exact matches: " << num_matching
+            << "\n  end reached: " << num_at_end
+            << "\n  non-ok status: " << num_not_ok
+            << "\n  mutated on the fly: " << num_recently_removed << std::endl;
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  ParseCommandLineFlags(&argc, &argv, true);
+  return RUN_ALL_TESTS();
+}

db/db_iterator_test.cc

@@ -20,7 +20,7 @@ namespace rocksdb {
 
 // A dumb ReadCallback which saying every key is committed.
 class DummyReadCallback : public ReadCallback {
-  bool IsCommitted(SequenceNumber /*seq*/) { return true; }
+  bool IsVisible(SequenceNumber /*seq*/) override { return true; }
 };
 
 // Test param:
@@ -862,6 +862,8 @@ TEST_P(DBIteratorTest, IteratorPinsRef) {
   } while (ChangeCompactOptions());
 }
 
+// SetOptions not defined in ROCKSDB LITE
+#ifndef ROCKSDB_LITE
 TEST_P(DBIteratorTest, DBIteratorBoundTest) {
   Options options = CurrentOptions();
   options.env = env_;
@@ -946,9 +948,7 @@ TEST_P(DBIteratorTest, DBIteratorBoundTest) {
   }
 
   // prefix is the first letter of the key
-  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
-
-  DestroyAndReopen(options);
+  ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:1"}}));
   ASSERT_OK(Put("a", "0"));
   ASSERT_OK(Put("foo", "bar"));
   ASSERT_OK(Put("foo1", "bar1"));
@@ -1036,6 +1036,61 @@ TEST_P(DBIteratorTest, DBIteratorBoundTest) {
   }
 }
 
+TEST_P(DBIteratorTest, DBIteratorBoundMultiSeek) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.prefix_extractor = nullptr;
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("a", "0"));
+  ASSERT_OK(Put("z", "0"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("foo1", "bar1"));
+  ASSERT_OK(Put("foo2", "bar2"));
+  ASSERT_OK(Put("foo3", "bar3"));
+  ASSERT_OK(Put("foo4", "bar4"));
+
+  {
+    std::string up_str = "foo5";
+    Slice up(up_str);
+    ReadOptions ro;
+    ro.iterate_upper_bound = &up;
+    std::unique_ptr<Iterator> iter(NewIterator(ro));
+
+    iter->Seek("foo1");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo1")), 0);
+
+    uint64_t prev_block_cache_hit =
+        TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    uint64_t prev_block_cache_miss =
+        TestGetTickerCount(options, BLOCK_CACHE_MISS);
+
+    ASSERT_GT(prev_block_cache_hit + prev_block_cache_miss, 0);
+
+    iter->Seek("foo4");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo4")), 0);
+    ASSERT_EQ(prev_block_cache_hit,
+              TestGetTickerCount(options, BLOCK_CACHE_HIT));
+    ASSERT_EQ(prev_block_cache_miss,
+              TestGetTickerCount(options, BLOCK_CACHE_MISS));
+
+    iter->Seek("foo2");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo2")), 0);
+    iter->Next();
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo3")), 0);
+    ASSERT_EQ(prev_block_cache_hit,
+              TestGetTickerCount(options, BLOCK_CACHE_HIT));
+    ASSERT_EQ(prev_block_cache_miss,
+              TestGetTickerCount(options, BLOCK_CACHE_MISS));
+  }
+}
+#endif
+
 TEST_P(DBIteratorTest, DBIteratorBoundOptimizationTest) {
   int upper_bound_hits = 0;
   Options options = CurrentOptions();
@@ -1482,6 +1537,26 @@ TEST_P(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
   delete iter;
 }
 
+class SliceTransformLimitedDomainGeneric : public SliceTransform {
+  const char* Name() const override {
+    return "SliceTransformLimitedDomainGeneric";
+  }
+
+  Slice Transform(const Slice& src) const override {
+    return Slice(src.data(), 1);
+  }
+
+  bool InDomain(const Slice& src) const override {
+    // prefix will be x????
+    return src.size() >= 1;
+  }
+
+  bool InRange(const Slice& dst) const override {
+    // prefix will be x????
+    return dst.size() == 1;
+  }
+};
+
 TEST_P(DBIteratorTest, IterSeekForPrevCrossingFiles) {
   Options options = CurrentOptions();
   options.prefix_extractor.reset(NewFixedPrefixTransform(1));
@@ -1536,6 +1611,61 @@ TEST_P(DBIteratorTest, IterSeekForPrevCrossingFiles) {
   }
 }
 
+TEST_P(DBIteratorTest, IterSeekForPrevCrossingFilesCustomPrefixExtractor) {
+  Options options = CurrentOptions();
+  options.prefix_extractor =
+      std::make_shared<SliceTransformLimitedDomainGeneric>();
+  options.disable_auto_compactions = true;
+  // Enable prefix bloom for SST files
+  BlockBasedTableOptions table_options;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put("a1", "va1"));
+  ASSERT_OK(Put("a2", "va2"));
+  ASSERT_OK(Put("a3", "va3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b1", "vb1"));
+  ASSERT_OK(Put("b2", "vb2"));
+  ASSERT_OK(Put("b3", "vb3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b4", "vb4"));
+  ASSERT_OK(Put("d1", "vd1"));
+  ASSERT_OK(Put("d2", "vd2"));
+  ASSERT_OK(Put("d4", "vd4"));
+  ASSERT_OK(Flush());
+
+  MoveFilesToLevel(1);
+  {
+    ReadOptions ro;
+    Iterator* iter = NewIterator(ro);
+
+    iter->SeekForPrev("a4");
+    ASSERT_EQ(iter->key().ToString(), "a3");
+    ASSERT_EQ(iter->value().ToString(), "va3");
+
+    iter->SeekForPrev("c2");
+    ASSERT_EQ(iter->key().ToString(), "b3");
+    iter->SeekForPrev("d3");
+    ASSERT_EQ(iter->key().ToString(), "d2");
+    iter->SeekForPrev("b5");
+    ASSERT_EQ(iter->key().ToString(), "b4");
+    delete iter;
+  }
+
+  {
+    ReadOptions ro;
+    ro.prefix_same_as_start = true;
+    Iterator* iter = NewIterator(ro);
+    iter->SeekForPrev("c2");
+    ASSERT_TRUE(!iter->Valid());
+    delete iter;
+  }
+}
+
 TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocks) {
   Options options = CurrentOptions();
   BlockBasedTableOptions table_options;
@@ -2310,6 +2440,46 @@ TEST_P(DBIteratorTest, SeekAfterHittingManyInternalKeys) {
   ASSERT_EQ(iter2->value().ToString(), "val_6");
 }
 
+// Reproduces a former bug where iterator would skip some records when DBIter
+// re-seeks subiterator with Incomplete status.
+TEST_P(DBIteratorTest, NonBlockingIterationBugRepro) {
+  Options options = CurrentOptions();
+  BlockBasedTableOptions table_options;
+  // Make sure the sst file has more than one block.
+  table_options.flush_block_policy_factory =
+      std::make_shared<FlushBlockEveryKeyPolicyFactory>();
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  // Two records in sst file, each in its own block.
+  Put("b", "");
+  Put("d", "");
+  Flush();
+
+  // Create a nonblocking iterator before writing to memtable.
+  ReadOptions ropt;
+  ropt.read_tier = kBlockCacheTier;
+  unique_ptr<Iterator> iter(NewIterator(ropt));
+
+  // Overwrite a key in memtable many times to hit
+  // max_sequential_skip_in_iterations (which is 8 by default).
+  for (int i = 0; i < 20; ++i) {
+    Put("c", "");
+  }
+
+  // Load the second block in sst file into the block cache.
+  {
+    unique_ptr<Iterator> iter2(NewIterator(ReadOptions()));
+    iter2->Seek("d");
+  }
+
+  // Finally seek the nonblocking iterator.
+  iter->Seek("a");
+  // With the bug, the status used to be OK, and the iterator used to point to
+  // "d".
+  EXPECT_TRUE(iter->status().IsIncomplete());
+}
+
 INSTANTIATE_TEST_CASE_P(DBIteratorTestInstance, DBIteratorTest,
                         testing::Values(true, false));
 
@@ -2322,7 +2492,7 @@ TEST_F(DBIteratorWithReadCallbackTest, ReadCallback) {
     explicit TestReadCallback(SequenceNumber last_visible_seq)
         : last_visible_seq_(last_visible_seq) {}
 
-    bool IsCommitted(SequenceNumber seq) override {
+    bool IsVisible(SequenceNumber seq) override {
       return seq <= last_visible_seq_;
     }
 

db/db_merge_operator_test.cc

@@ -20,7 +20,7 @@ class TestReadCallback : public ReadCallback {
                    SequenceNumber snapshot_seq)
       : snapshot_checker_(snapshot_checker), snapshot_seq_(snapshot_seq) {}
 
-  bool IsCommitted(SequenceNumber seq) override {
+  bool IsVisible(SequenceNumber seq) override {
     return snapshot_checker_->IsInSnapshot(seq, snapshot_seq_);
   }
 

db/db_options_test.cc

@@ -134,7 +134,7 @@ TEST_F(DBOptionsTest, SetBytesPerSync) {
   const std::string kValue(kValueSize, 'v');
   ASSERT_EQ(options.bytes_per_sync, dbfull()->GetDBOptions().bytes_per_sync);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "WritableFileWriter::RangeSync:0", [&](void* arg) {
+      "WritableFileWriter::RangeSync:0", [&](void* /*arg*/) {
         counter++;
       });
 
@@ -183,7 +183,7 @@ TEST_F(DBOptionsTest, SetWalBytesPerSync) {
   int counter = 0;
   int low_bytes_per_sync = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "WritableFileWriter::RangeSync:0", [&](void* arg) {
+      "WritableFileWriter::RangeSync:0", [&](void* /*arg*/) {
         counter++;
       });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
@@ -242,11 +242,12 @@ TEST_F(DBOptionsTest, WritableFileMaxBufferSize) {
   ASSERT_EQ(unmatch_cnt, 0);
   ASSERT_GE(match_cnt, 11);
 
-  buffer_size = 512 * 1024;
-  match_cnt = 0;
-  unmatch_cnt = 0;
   ASSERT_OK(
       dbfull()->SetDBOptions({{"writable_file_max_buffer_size", "524288"}}));
+  buffer_size = 512 * 1024;
+  match_cnt = 0;
+  unmatch_cnt = 0;  // SetDBOptions() will create a WriteableFileWriter
+
   ASSERT_EQ(buffer_size,
             dbfull()->GetDBOptions().writable_file_max_buffer_size);
   i = 0;

db/db_properties_test.cc

@@ -69,27 +69,27 @@ TEST_F(DBPropertiesTest, Empty) {
     ASSERT_OK(db_->DisableFileDeletions());
     ASSERT_TRUE(
         dbfull()->GetProperty("rocksdb.is-file-deletions-enabled", &num));
-    ASSERT_EQ("1", num);
+    ASSERT_EQ("0", num);
 
     ASSERT_OK(db_->DisableFileDeletions());
     ASSERT_TRUE(
         dbfull()->GetProperty("rocksdb.is-file-deletions-enabled", &num));
-    ASSERT_EQ("2", num);
+    ASSERT_EQ("0", num);
 
     ASSERT_OK(db_->DisableFileDeletions());
     ASSERT_TRUE(
         dbfull()->GetProperty("rocksdb.is-file-deletions-enabled", &num));
-    ASSERT_EQ("3", num);
+    ASSERT_EQ("0", num);
 
     ASSERT_OK(db_->EnableFileDeletions(false));
     ASSERT_TRUE(
         dbfull()->GetProperty("rocksdb.is-file-deletions-enabled", &num));
-    ASSERT_EQ("2", num);
+    ASSERT_EQ("0", num);
 
     ASSERT_OK(db_->EnableFileDeletions());
     ASSERT_TRUE(
         dbfull()->GetProperty("rocksdb.is-file-deletions-enabled", &num));
-    ASSERT_EQ("0", num);
+    ASSERT_EQ("1", num);
   } while (ChangeOptions());
 }
 
@@ -170,6 +170,7 @@ void ResetTableProperties(TableProperties* tp) {
   tp->raw_value_size = 0;
   tp->num_data_blocks = 0;
   tp->num_entries = 0;
+  tp->num_range_deletions = 0;
 }
 
 void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
@@ -177,17 +178,19 @@ void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
   std::replace(tp_string.begin(), tp_string.end(), ';', ' ');
   std::replace(tp_string.begin(), tp_string.end(), '=', ' ');
   ResetTableProperties(tp);
-
   sscanf(tp_string.c_str(),
-         "# data blocks %" SCNu64 " # entries %" SCNu64 " raw key size %" SCNu64
+         "# data blocks %" SCNu64 " # entries %" SCNu64
+         " # range deletions %" SCNu64
+         " raw key size %" SCNu64
          " raw average key size %lf "
          " raw value size %" SCNu64
          " raw average value size %lf "
-         " data block size %" SCNu64 " index block size %" SCNu64
-         " filter block size %" SCNu64,
-         &tp->num_data_blocks, &tp->num_entries, &tp->raw_key_size,
-         &dummy_double, &tp->raw_value_size, &dummy_double, &tp->data_size,
-         &tp->index_size, &tp->filter_size);
+         " data block size %" SCNu64 " index block size (user-key? %" SCNu64
+         ") %" SCNu64 " filter block size %" SCNu64,
+         &tp->num_data_blocks, &tp->num_entries, &tp->num_range_deletions,
+         &tp->raw_key_size, &dummy_double, &tp->raw_value_size, &dummy_double,
+         &tp->data_size, &tp->index_key_is_user_key, &tp->index_size,
+         &tp->filter_size);
 }
 
 void VerifySimilar(uint64_t a, uint64_t b, double bias) {
@@ -218,19 +221,25 @@ void VerifyTableProperties(const TableProperties& base_tp,
   ASSERT_EQ(base_tp.raw_key_size, new_tp.raw_key_size);
   ASSERT_EQ(base_tp.raw_value_size, new_tp.raw_value_size);
   ASSERT_EQ(base_tp.num_entries, new_tp.num_entries);
+  ASSERT_EQ(base_tp.num_range_deletions, new_tp.num_range_deletions);
 }
 
 void GetExpectedTableProperties(TableProperties* expected_tp,
                                 const int kKeySize, const int kValueSize,
-                                const int kKeysPerTable, const int kTableCount,
+                                const int kKeysPerTable,
+                                const int kRangeDeletionsPerTable,
+                                const int kTableCount,
                                 const int kBloomBitsPerKey,
-                                const size_t kBlockSize) {
+                                const size_t kBlockSize,
+                                const bool index_key_is_user_key) {
   const int kKeyCount = kTableCount * kKeysPerTable;
+  const int kRangeDeletionCount = kTableCount * kRangeDeletionsPerTable;
   const int kAvgSuccessorSize = kKeySize / 5;
   const int kEncodingSavePerKey = kKeySize / 4;
-  expected_tp->raw_key_size = kKeyCount * (kKeySize + 8);
-  expected_tp->raw_value_size = kKeyCount * kValueSize;
+  expected_tp->raw_key_size = (kKeyCount + kRangeDeletionCount) * (kKeySize + 8);
+  expected_tp->raw_value_size = (kKeyCount + kRangeDeletionCount) * kValueSize;
   expected_tp->num_entries = kKeyCount;
+  expected_tp->num_range_deletions = kRangeDeletionCount;
   expected_tp->num_data_blocks =
       kTableCount *
       (kKeysPerTable * (kKeySize - kEncodingSavePerKey + kValueSize)) /
@@ -238,7 +247,8 @@ void GetExpectedTableProperties(TableProperties* expected_tp,
   expected_tp->data_size =
       kTableCount * (kKeysPerTable * (kKeySize + 8 + kValueSize));
   expected_tp->index_size =
-      expected_tp->num_data_blocks * (kAvgSuccessorSize + 8);
+      expected_tp->num_data_blocks *
+      (kAvgSuccessorSize + (index_key_is_user_key ? 0 : 8));
   expected_tp->filter_size =
       kTableCount * (kKeysPerTable * kBloomBitsPerKey / 8);
 }
@@ -251,8 +261,11 @@ TEST_F(DBPropertiesTest, ValidatePropertyInfo) {
     ASSERT_TRUE(ppt_name_and_info.first.empty() ||
                 !isdigit(ppt_name_and_info.first.back()));
 
-    ASSERT_TRUE((ppt_name_and_info.second.handle_string == nullptr) !=
-                (ppt_name_and_info.second.handle_int == nullptr));
+    int count = 0;
+    count += (ppt_name_and_info.second.handle_string == nullptr) ? 0 : 1;
+    count += (ppt_name_and_info.second.handle_int == nullptr) ? 0 : 1;
+    count += (ppt_name_and_info.second.handle_string_dbimpl == nullptr) ? 0 : 1;
+    ASSERT_TRUE(count == 1);
   }
 }
 
@@ -287,6 +300,7 @@ TEST_F(DBPropertiesTest, ValidateSampleNumber) {
 
 TEST_F(DBPropertiesTest, AggregatedTableProperties) {
   for (int kTableCount = 40; kTableCount <= 100; kTableCount += 30) {
+    const int kRangeDeletionsPerTable = 5;
     const int kKeysPerTable = 100;
     const int kKeySize = 80;
     const int kValueSize = 200;
@@ -305,24 +319,35 @@ TEST_F(DBPropertiesTest, AggregatedTableProperties) {
 
     DestroyAndReopen(options);
 
+    // Hold open a snapshot to prevent range tombstones from being compacted
+    // away.
+    ManagedSnapshot snapshot(db_);
+
     Random rnd(5632);
     for (int table = 1; table <= kTableCount; ++table) {
       for (int i = 0; i < kKeysPerTable; ++i) {
         db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
                  RandomString(&rnd, kValueSize));
       }
+      for (int i = 0; i < kRangeDeletionsPerTable; i++) {
+        std::string start = RandomString(&rnd, kKeySize);
+        std::string end = start;
+        end.resize(kValueSize);
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), start, end);
+      }
       db_->Flush(FlushOptions());
     }
     std::string property;
     db_->GetProperty(DB::Properties::kAggregatedTableProperties, &property);
+    TableProperties output_tp;
+    ParseTablePropertiesString(property, &output_tp);
+    bool index_key_is_user_key = output_tp.index_key_is_user_key > 0;
 
     TableProperties expected_tp;
     GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                               kKeysPerTable, kTableCount, kBloomBitsPerKey,
-                               table_options.block_size);
-
-    TableProperties output_tp;
-    ParseTablePropertiesString(property, &output_tp);
+                               kKeysPerTable, kRangeDeletionsPerTable,
+                               kTableCount, kBloomBitsPerKey,
+                               table_options.block_size, index_key_is_user_key);
 
     VerifyTableProperties(expected_tp, output_tp);
   }
@@ -371,6 +396,13 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
   for (int key = 0; key < key_index; key++) {
     Get(Key(key));
   }
+
+  // Test for getting immutable_db_options_.statistics
+  ASSERT_TRUE(dbfull()->GetProperty(dbfull()->DefaultColumnFamily(),
+                                    "rocksdb.options-statistics", &prop));
+  ASSERT_NE(std::string::npos, prop.find("rocksdb.block.cache.miss"));
+  ASSERT_EQ(std::string::npos, prop.find("rocksdb.db.f.micros"));
+
   ASSERT_TRUE(dbfull()->GetProperty(dbfull()->DefaultColumnFamily(),
                                     "rocksdb.cf-file-histogram", &prop));
   ASSERT_NE(std::string::npos, prop.find("** Level 0 read latency histogram"));
@@ -437,6 +469,7 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
 
 TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
   const int kTableCount = 100;
+  const int kRangeDeletionsPerTable = 2;
   const int kKeysPerTable = 10;
   const int kKeySize = 50;
   const int kValueSize = 400;
@@ -462,6 +495,9 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
 
   DestroyAndReopen(options);
 
+  // Hold open a snapshot to prevent range tombstones from being compacted away.
+  ManagedSnapshot snapshot(db_);
+
   std::string level_tp_strings[kMaxLevel];
   std::string tp_string;
   TableProperties level_tps[kMaxLevel];
@@ -471,6 +507,12 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
       db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
                RandomString(&rnd, kValueSize));
     }
+    for (int i = 0; i < kRangeDeletionsPerTable; i++) {
+      std::string start = RandomString(&rnd, kKeySize);
+      std::string end = start;
+      end.resize(kValueSize);
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), start, end);
+    }
     db_->Flush(FlushOptions());
     db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
     ResetTableProperties(&sum_tp);
@@ -486,9 +528,11 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
       sum_tp.raw_value_size += level_tps[level].raw_value_size;
       sum_tp.num_data_blocks += level_tps[level].num_data_blocks;
       sum_tp.num_entries += level_tps[level].num_entries;
+      sum_tp.num_range_deletions += level_tps[level].num_range_deletions;
     }
     db_->GetProperty(DB::Properties::kAggregatedTableProperties, &tp_string);
     ParseTablePropertiesString(tp_string, &tp);
+    bool index_key_is_user_key = tp.index_key_is_user_key > 0;
     ASSERT_EQ(sum_tp.data_size, tp.data_size);
     ASSERT_EQ(sum_tp.index_size, tp.index_size);
     ASSERT_EQ(sum_tp.filter_size, tp.filter_size);
@@ -496,13 +540,15 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
     ASSERT_EQ(sum_tp.raw_value_size, tp.raw_value_size);
     ASSERT_EQ(sum_tp.num_data_blocks, tp.num_data_blocks);
     ASSERT_EQ(sum_tp.num_entries, tp.num_entries);
+    ASSERT_EQ(sum_tp.num_range_deletions, tp.num_range_deletions);
     if (table > 3) {
-      GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                                 kKeysPerTable, table, kBloomBitsPerKey,
-                                 table_options.block_size);
+      GetExpectedTableProperties(
+          &expected_tp, kKeySize, kValueSize, kKeysPerTable,
+          kRangeDeletionsPerTable, table, kBloomBitsPerKey,
+          table_options.block_size, index_key_is_user_key);
       // Gives larger bias here as index block size, filter block size,
       // and data block size become much harder to estimate in this test.
-      VerifyTableProperties(tp, expected_tp, 0.5, 0.4, 0.4, 0.25);
+      VerifyTableProperties(expected_tp, tp, 0.5, 0.4, 0.4, 0.25);
     }
   }
 }
@@ -1436,6 +1482,107 @@ TEST_F(DBPropertiesTest, SstFilesSize) {
   ASSERT_TRUE(listener->callback_triggered);
 }
 
+TEST_F(DBPropertiesTest, BlockCacheProperties) {
+  Options options;
+  uint64_t value;
+
+  // Block cache properties are not available for tables other than
+  // block-based table.
+  options.table_factory.reset(NewPlainTableFactory());
+  Reopen(options);
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_FALSE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+
+  options.table_factory.reset(NewCuckooTableFactory());
+  Reopen(options);
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_FALSE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+
+  // Block cache properties are not available if block cache is not used.
+  BlockBasedTableOptions table_options;
+  table_options.no_block_cache = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_FALSE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_FALSE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+
+  // Test with empty block cache.
+  constexpr size_t kCapacity = 100;
+  auto block_cache = NewLRUCache(kCapacity, 0 /*num_shard_bits*/);
+  table_options.block_cache = block_cache;
+  table_options.no_block_cache = false;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_EQ(kCapacity, value);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_EQ(0, value);
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+  ASSERT_EQ(0, value);
+
+  // Insert unpinned item to the cache and check size.
+  constexpr size_t kSize1 = 50;
+  block_cache->Insert("item1", nullptr /*value*/, kSize1, nullptr /*deleter*/);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_EQ(kCapacity, value);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_EQ(kSize1, value);
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+  ASSERT_EQ(0, value);
+
+  // Insert pinned item to the cache and check size.
+  constexpr size_t kSize2 = 30;
+  Cache::Handle* item2 = nullptr;
+  block_cache->Insert("item2", nullptr /*value*/, kSize2, nullptr /*deleter*/,
+                      &item2);
+  ASSERT_NE(nullptr, item2);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_EQ(kCapacity, value);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  ASSERT_EQ(kSize1 + kSize2, value);
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+  ASSERT_EQ(kSize2, value);
+
+  // Insert another pinned item to make the cache over-sized.
+  constexpr size_t kSize3 = 80;
+  Cache::Handle* item3 = nullptr;
+  block_cache->Insert("item3", nullptr /*value*/, kSize3, nullptr /*deleter*/,
+                      &item3);
+  ASSERT_NE(nullptr, item2);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_EQ(kCapacity, value);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  // Item 1 is evicted.
+  ASSERT_EQ(kSize2 + kSize3, value);
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+  ASSERT_EQ(kSize2 + kSize3, value);
+
+  // Check size after release.
+  block_cache->Release(item2);
+  block_cache->Release(item3);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheCapacity, &value));
+  ASSERT_EQ(kCapacity, value);
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBlockCacheUsage, &value));
+  // item2 will be evicted, while item3 remain in cache after release.
+  ASSERT_EQ(kSize3, value);
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kBlockCachePinnedUsage, &value));
+  ASSERT_EQ(0, value);
+}
+
 #endif  // ROCKSDB_LITE
 }  // namespace rocksdb
 

db/db_range_del_test.cc

@@ -27,47 +27,53 @@ class DBRangeDelTest : public DBTestBase {
 // ROCKSDB_LITE
 #ifndef ROCKSDB_LITE
 TEST_F(DBRangeDelTest, NonBlockBasedTableNotSupported) {
-  if (!IsMemoryMappedAccessSupported()) {
-    return;
+  // TODO: figure out why MmapReads trips the iterator pinning assertion in
+  // RangeDelAggregator. Ideally it would be supported; otherwise it should at
+  // least be explicitly unsupported.
+  for (auto config : {kPlainTableAllBytesPrefix, /* kWalDirAndMmapReads */}) {
+    option_config_ = config;
+    DestroyAndReopen(CurrentOptions());
+    ASSERT_TRUE(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                                 "dr1", "dr1")
+                    .IsNotSupported());
   }
-  Options opts = CurrentOptions();
-  opts.table_factory.reset(new PlainTableFactory());
-  opts.prefix_extractor.reset(NewNoopTransform());
-  opts.allow_mmap_reads = true;
-  opts.max_sequential_skip_in_iterations = 999999;
-  Reopen(opts);
-
-  ASSERT_TRUE(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1", "dr1")
-          .IsNotSupported());
 }
 
 TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
-  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
-                             "dr2"));
-  ASSERT_OK(db_->Flush(FlushOptions()));
-  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                               "dr1", "dr2"));
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }
 
 TEST_F(DBRangeDelTest, CompactionOutputHasOnlyRangeTombstone) {
-  Options opts = CurrentOptions();
-  opts.disable_auto_compactions = true;
-  opts.statistics = CreateDBStatistics();
-  Reopen(opts);
+  do {
+    Options opts = CurrentOptions();
+    opts.disable_auto_compactions = true;
+    opts.statistics = CreateDBStatistics();
+    DestroyAndReopen(opts);
 
-  // snapshot protects range tombstone from dropping due to becoming obsolete.
-  const Snapshot* snapshot = db_->GetSnapshot();
-  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
-  db_->Flush(FlushOptions());
+    // snapshot protects range tombstone from dropping due to becoming obsolete.
+    const Snapshot* snapshot = db_->GetSnapshot();
+    db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
+    db_->Flush(FlushOptions());
 
-  ASSERT_EQ(1, NumTableFilesAtLevel(0));
-  ASSERT_EQ(0, NumTableFilesAtLevel(1));
-  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
-                              true /* disallow_trivial_move */);
-  ASSERT_EQ(0, NumTableFilesAtLevel(0));
-  ASSERT_EQ(1, NumTableFilesAtLevel(1));
-  ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
-  db_->ReleaseSnapshot(snapshot);
+    ASSERT_EQ(1, NumTableFilesAtLevel(0));
+    ASSERT_EQ(0, NumTableFilesAtLevel(1));
+    dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                                true /* disallow_trivial_move */);
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_EQ(1, NumTableFilesAtLevel(1));
+    ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
+    db_->ReleaseSnapshot(snapshot);
+    // Skip cuckoo memtables, which do not support snapshots. Skip non-leveled
+    // compactions as the above assertions about the number of files in a level
+    // do not hold true.
+  } while (ChangeOptions(kRangeDelSkipConfigs | kSkipHashCuckoo |
+                         kSkipUniversalCompaction | kSkipFIFOCompaction));
 }
 
 TEST_F(DBRangeDelTest, CompactionOutputFilesExactlyFilled) {
@@ -433,8 +439,8 @@ TEST_F(DBRangeDelTest, ValidUniversalSubcompactionBoundaries) {
       reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
           ->cfd(),
       1 /* input_level */, 2 /* output_level */, 0 /* output_path_id */,
-      nullptr /* begin */, nullptr /* end */, true /* exclusive */,
-      true /* disallow_trivial_move */));
+      0 /* max_subcompactions */, nullptr /* begin */, nullptr /* end */,
+      true /* exclusive */, true /* disallow_trivial_move */));
 }
 #endif  // ROCKSDB_LITE
 
@@ -496,12 +502,12 @@ TEST_F(DBRangeDelTest, ObsoleteTombstoneCleanup) {
   Reopen(opts);
 
   db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
-                   "dr1");  // obsolete after compaction
+                   "dr10");  // obsolete after compaction
   db_->Put(WriteOptions(), "key", "val");
   db_->Flush(FlushOptions());
   const Snapshot* snapshot = db_->GetSnapshot();
   db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr2",
-                   "dr2");  // protected by snapshot
+                   "dr20");  // protected by snapshot
   db_->Put(WriteOptions(), "key", "val");
   db_->Flush(FlushOptions());
 
@@ -590,48 +596,57 @@ TEST_F(DBRangeDelTest, TableEvictedDuringScan) {
 }
 
 TEST_F(DBRangeDelTest, GetCoveredKeyFromMutableMemtable) {
-  db_->Put(WriteOptions(), "key", "val");
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    db_->Put(WriteOptions(), "key", "val");
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
 
-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }
 
 TEST_F(DBRangeDelTest, GetCoveredKeyFromImmutableMemtable) {
-  Options opts = CurrentOptions();
-  opts.max_write_buffer_number = 3;
-  opts.min_write_buffer_number_to_merge = 2;
-  // SpecialSkipListFactory lets us specify maximum number of elements the
-  // memtable can hold. It switches the active memtable to immutable (flush is
-  // prevented by the above options) upon inserting an element that would
-  // overflow the memtable.
-  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
-  Reopen(opts);
+  do {
+    Options opts = CurrentOptions();
+    opts.max_write_buffer_number = 3;
+    opts.min_write_buffer_number_to_merge = 2;
+    // SpecialSkipListFactory lets us specify maximum number of elements the
+    // memtable can hold. It switches the active memtable to immutable (flush is
+    // prevented by the above options) upon inserting an element that would
+    // overflow the memtable.
+    opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+    DestroyAndReopen(opts);
 
-  db_->Put(WriteOptions(), "key", "val");
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
-  db_->Put(WriteOptions(), "blah", "val");
+    db_->Put(WriteOptions(), "key", "val");
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+    db_->Put(WriteOptions(), "blah", "val");
 
-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }
 
 TEST_F(DBRangeDelTest, GetCoveredKeyFromSst) {
-  db_->Put(WriteOptions(), "key", "val");
-  // snapshot prevents key from being deleted during flush
-  const Snapshot* snapshot = db_->GetSnapshot();
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
-  ASSERT_OK(db_->Flush(FlushOptions()));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    db_->Put(WriteOptions(), "key", "val");
+    // snapshot prevents key from being deleted during flush
+    const Snapshot* snapshot = db_->GetSnapshot();
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+    ASSERT_OK(db_->Flush(FlushOptions()));
 
-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
-  db_->ReleaseSnapshot(snapshot);
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    db_->ReleaseSnapshot(snapshot);
+    // Cuckoo memtables do not support snapshots.
+  } while (ChangeOptions(kRangeDelSkipConfigs | kSkipHashCuckoo));
 }
 
 TEST_F(DBRangeDelTest, GetCoveredMergeOperandFromMemtable) {
@@ -895,11 +910,14 @@ TEST_F(DBRangeDelTest, MemtableBloomFilter) {
 }
 
 TEST_F(DBRangeDelTest, CompactionTreatsSplitInputLevelDeletionAtomically) {
-  // make sure compaction treats files containing a split range deletion in the
-  // input level as an atomic unit. I.e., compacting any input-level file(s)
-  // containing a portion of the range deletion causes all other input-level
-  // files containing portions of that same range deletion to be included in the
-  // compaction.
+  // This test originally verified that compaction treated files containing a
+  // split range deletion in the input level as an atomic unit. I.e.,
+  // compacting any input-level file(s) containing a portion of the range
+  // deletion causes all other input-level files containing portions of that
+  // same range deletion to be included in the compaction. Range deletion
+  // tombstones are now truncated to sstable boundaries which removed the need
+  // for that behavior (which could lead to excessively large
+  // compactions).
   const int kNumFilesPerLevel = 4, kValueBytes = 4 << 10;
   Options options = CurrentOptions();
   options.compression = kNoCompression;
@@ -946,22 +964,111 @@ TEST_F(DBRangeDelTest, CompactionTreatsSplitInputLevelDeletionAtomically) {
     if (i == 0) {
       ASSERT_OK(db_->CompactFiles(
           CompactionOptions(), {meta.levels[1].files[0].name}, 2 /* level */));
+      ASSERT_EQ(0, NumTableFilesAtLevel(1));
     } else if (i == 1) {
       auto begin_str = Key(0), end_str = Key(1);
       Slice begin = begin_str, end = end_str;
       ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &begin, &end));
+      ASSERT_EQ(3, NumTableFilesAtLevel(1));
     } else if (i == 2) {
       ASSERT_OK(db_->SetOptions(db_->DefaultColumnFamily(),
                                 {{"max_bytes_for_level_base", "10000"}}));
       dbfull()->TEST_WaitForCompact();
+      ASSERT_EQ(1, NumTableFilesAtLevel(1));
     }
-    ASSERT_EQ(0, NumTableFilesAtLevel(1));
     ASSERT_GT(NumTableFilesAtLevel(2), 0);
 
     db_->ReleaseSnapshot(snapshot);
   }
 }
 
+TEST_F(DBRangeDelTest, RangeTombstoneEndKeyAsSstableUpperBound) {
+  // Test the handling of the range-tombstone end-key as the
+  // upper-bound for an sstable.
+
+  const int kNumFilesPerLevel = 2, kValueBytes = 4 << 10;
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.level0_file_num_compaction_trigger = kNumFilesPerLevel;
+  options.memtable_factory.reset(
+      new SpecialSkipListFactory(2 /* num_entries_flush */));
+  options.target_file_size_base = kValueBytes;
+  options.disable_auto_compactions = true;
+
+  DestroyAndReopen(options);
+
+  // Create an initial sstable at L2:
+  //   [key000000#1,1, key000000#1,1]
+  ASSERT_OK(Put(Key(0), ""));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  MoveFilesToLevel(2);
+  ASSERT_EQ(1, NumTableFilesAtLevel(2));
+
+  // A snapshot protects the range tombstone from dropping due to
+  // becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                   Key(0), Key(2 * kNumFilesPerLevel));
+
+  // Create 2 additional sstables in L0. Note that the first sstable
+  // contains the range tombstone.
+  //   [key000000#3,1, key000004#72057594037927935,15]
+  //   [key000001#5,1, key000002#6,1]
+  Random rnd(301);
+  std::string value = RandomString(&rnd, kValueBytes);
+  for (int j = 0; j < kNumFilesPerLevel; ++j) {
+    // Give files overlapping key-ranges to prevent a trivial move when we
+    // compact from L0 to L1.
+    ASSERT_OK(Put(Key(j), value));
+    ASSERT_OK(Put(Key(2 * kNumFilesPerLevel - 1 - j), value));
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    ASSERT_EQ(j + 1, NumTableFilesAtLevel(0));
+  }
+  // Compact the 2 L0 sstables to L1, resulting in the following LSM. There
+  // are 2 sstables generated in L1 due to the target_file_size_base setting.
+  //   L1:
+  //     [key000000#3,1, key000002#72057594037927935,15]
+  //     [key000002#6,1, key000004#72057594037927935,15]
+  //   L2:
+  //     [key000000#1,1, key000000#1,1]
+  MoveFilesToLevel(1);
+  ASSERT_EQ(2, NumTableFilesAtLevel(1));
+
+  {
+    // Compact the second sstable in L1:
+    //   L1:
+    //     [key000000#3,1, key000002#72057594037927935,15]
+    //   L2:
+    //     [key000000#1,1, key000000#1,1]
+    //     [key000002#6,1, key000004#72057594037927935,15]
+    auto begin_str = Key(3);
+    const rocksdb::Slice begin = begin_str;
+    dbfull()->TEST_CompactRange(1, &begin, nullptr);
+    ASSERT_EQ(1, NumTableFilesAtLevel(1));
+    ASSERT_EQ(2, NumTableFilesAtLevel(2));
+  }
+
+  {
+    // Compact the first sstable in L1. This should be copacetic, but
+    // was previously resulting in overlapping sstables in L2 due to
+    // mishandling of the range tombstone end-key when used as the
+    // largest key for an sstable. The resulting LSM structure should
+    // be:
+    //
+    //   L2:
+    //     [key000000#1,1, key000001#72057594037927935,15]
+    //     [key000001#5,1, key000002#72057594037927935,15]
+    //     [key000002#6,1, key000004#72057594037927935,15]
+    auto begin_str = Key(0);
+    const rocksdb::Slice begin = begin_str;
+    dbfull()->TEST_CompactRange(1, &begin, &begin);
+    ASSERT_EQ(0, NumTableFilesAtLevel(1));
+    ASSERT_EQ(3, NumTableFilesAtLevel(2));
+  }
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_F(DBRangeDelTest, UnorderedTombstones) {
   // Regression test for #2752. Range delete tombstones between
   // different snapshot stripes are not stored in order, so the first

db/db_sst_test.cc

@@ -20,6 +20,37 @@ class DBSSTTest : public DBTestBase {
   DBSSTTest() : DBTestBase("/db_sst_test") {}
 };
 
+#ifndef ROCKSDB_LITE
+// A class which remembers the name of each flushed file.
+class FlushedFileCollector : public EventListener {
+ public:
+  FlushedFileCollector() {}
+  ~FlushedFileCollector() {}
+
+  virtual void OnFlushCompleted(DB* /*db*/, const FlushJobInfo& info) override {
+    std::lock_guard<std::mutex> lock(mutex_);
+    flushed_files_.push_back(info.file_path);
+  }
+
+  std::vector<std::string> GetFlushedFiles() {
+    std::lock_guard<std::mutex> lock(mutex_);
+    std::vector<std::string> result;
+    for (auto fname : flushed_files_) {
+      result.push_back(fname);
+    }
+    return result;
+  }
+  void ClearFlushedFiles() {
+    std::lock_guard<std::mutex> lock(mutex_);
+    flushed_files_.clear();
+  }
+
+ private:
+  std::vector<std::string> flushed_files_;
+  std::mutex mutex_;
+};
+#endif  // ROCKSDB_LITE
+
 TEST_F(DBSSTTest, DontDeletePendingOutputs) {
   Options options;
   options.env = env_;
@@ -72,7 +103,7 @@ TEST_F(DBSSTTest, SSTsWithLdbSuffixHandling) {
   ASSERT_GT(num_files, 0);
 
   std::vector<std::string> filenames;
-  GetSstFiles(dbname_, &filenames);
+  GetSstFiles(env_, dbname_, &filenames);
   int num_ldb_files = 0;
   for (size_t i = 0; i < filenames.size(); ++i) {
     if (i & 1) {
@@ -231,11 +262,11 @@ TEST_F(DBSSTTest, DBWithSstFileManager) {
   int files_deleted = 0;
   int files_moved = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "SstFileManagerImpl::OnAddFile", [&](void* arg) { files_added++; });
+      "SstFileManagerImpl::OnAddFile", [&](void* /*arg*/) { files_added++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "SstFileManagerImpl::OnDeleteFile", [&](void* arg) { files_deleted++; });
+      "SstFileManagerImpl::OnDeleteFile", [&](void* /*arg*/) { files_deleted++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "SstFileManagerImpl::OnMoveFile", [&](void* arg) { files_moved++; });
+      "SstFileManagerImpl::OnMoveFile", [&](void* /*arg*/) { files_moved++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -408,12 +439,12 @@ TEST_F(DBSSTTest, DeleteSchedulerMultipleDBPaths) {
   int bg_delete_file = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DeleteScheduler::DeleteTrashFile:DeleteFile",
-      [&](void* arg) { bg_delete_file++; });
+      [&](void* /*arg*/) { bg_delete_file++; });
   // The deletion scheduler sometimes skips marking file as trash according to
   // a heuristic. In that case the deletion will go through the below SyncPoint.
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DeleteScheduler::DeleteFile",
-      [&](void* arg) { bg_delete_file++; });
+      [&](void* /*arg*/) { bg_delete_file++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Options options = CurrentOptions();
@@ -483,7 +514,7 @@ TEST_F(DBSSTTest, DestroyDBWithRateLimitedDelete) {
   int bg_delete_file = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DeleteScheduler::DeleteTrashFile:DeleteFile",
-      [&](void* arg) { bg_delete_file++; });
+      [&](void* /*arg*/) { bg_delete_file++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Status s;
@@ -557,12 +588,13 @@ TEST_F(DBSSTTest, CancellingCompactionsWorks) {
 
   int completed_compactions = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction():CancelledCompaction", [&](void* arg) {
+      "DBImpl::BackgroundCompaction():CancelledCompaction", [&](void* /*arg*/) {
         sfm->SetMaxAllowedSpaceUsage(0);
+        ASSERT_EQ(sfm->GetCompactionsReservedSize(), 0);
       });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial:AfterRun",
-      [&](void* arg) { completed_compactions++; });
+      [&](void* /*arg*/) { completed_compactions++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Random rnd(301);
@@ -584,6 +616,8 @@ TEST_F(DBSSTTest, CancellingCompactionsWorks) {
   ASSERT_OK(Flush());
   dbfull()->TEST_WaitForCompact(true);
 
+  // Because we set a callback in CancelledCompaction, we actually
+  // let the compaction run
   ASSERT_GT(completed_compactions, 0);
   ASSERT_EQ(sfm->GetCompactionsReservedSize(), 0);
   // Make sure the stat is bumped
@@ -591,6 +625,78 @@ TEST_F(DBSSTTest, CancellingCompactionsWorks) {
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
 
+TEST_F(DBSSTTest, CancellingManualCompactionsWorks) {
+  std::shared_ptr<SstFileManager> sst_file_manager(NewSstFileManager(env_));
+  auto sfm = static_cast<SstFileManagerImpl*>(sst_file_manager.get());
+
+  Options options = CurrentOptions();
+  options.sst_file_manager = sst_file_manager;
+  options.statistics = CreateDBStatistics();
+
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+
+  // Generate a file containing 10 keys.
+  for (int i = 0; i < 10; i++) {
+    ASSERT_OK(Put(Key(i), RandomString(&rnd, 50)));
+  }
+  ASSERT_OK(Flush());
+  uint64_t total_file_size = 0;
+  auto files_in_db = GetAllSSTFiles(&total_file_size);
+  // Set the maximum allowed space usage to the current total size
+  sfm->SetMaxAllowedSpaceUsage(2 * total_file_size + 1);
+
+  // Generate another file to trigger compaction.
+  for (int i = 0; i < 10; i++) {
+    ASSERT_OK(Put(Key(i), RandomString(&rnd, 50)));
+  }
+  ASSERT_OK(Flush());
+
+  // OK, now trigger a manual compaction
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  // Wait for manual compaction to get scheduled and finish
+  dbfull()->TEST_WaitForCompact(true);
+
+  ASSERT_EQ(sfm->GetCompactionsReservedSize(), 0);
+  // Make sure the stat is bumped
+  ASSERT_EQ(dbfull()->immutable_db_options().statistics.get()->getTickerCount(
+                COMPACTION_CANCELLED),
+            1);
+
+  // Now make sure CompactFiles also gets cancelled
+  auto l0_files = collector->GetFlushedFiles();
+  dbfull()->CompactFiles(rocksdb::CompactionOptions(), l0_files, 0);
+
+  // Wait for manual compaction to get scheduled and finish
+  dbfull()->TEST_WaitForCompact(true);
+
+  ASSERT_EQ(dbfull()->immutable_db_options().statistics.get()->getTickerCount(
+                COMPACTION_CANCELLED),
+            2);
+  ASSERT_EQ(sfm->GetCompactionsReservedSize(), 0);
+
+  // Now let the flush through and make sure GetCompactionsReservedSize
+  // returns to normal
+  sfm->SetMaxAllowedSpaceUsage(0);
+  int completed_compactions = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "CompactFilesImpl:End", [&](void* /*arg*/) { completed_compactions++; });
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  dbfull()->CompactFiles(rocksdb::CompactionOptions(), l0_files, 0);
+  dbfull()->TEST_WaitForCompact(true);
+
+  ASSERT_EQ(sfm->GetCompactionsReservedSize(), 0);
+  ASSERT_GT(completed_compactions, 0);
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
 TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
   // This test will set a maximum allowed space for the DB, then it will
   // keep filling the DB until the limit is reached and bg_error_ is set.
@@ -627,7 +733,7 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "CompactionJob::FinishCompactionOutputFile:MaxAllowedSpaceReached",
-      [&](void* arg) {
+      [&](void* /*arg*/) {
         bg_error_set = true;
         GetAllSSTFiles(&total_sst_files_size);
         reached_max_space_on_compaction++;

db/db_tailing_iter_test.cc

@@ -157,10 +157,10 @@ TEST_F(DBTestTailingIterator, TailingIteratorTrimSeekToNext) {
       });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "ForwardIterator::RenewIterators:Null",
-      [&](void* arg) { file_iters_renewed_null = true; });
+      [&](void* /*arg*/) { file_iters_renewed_null = true; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "ForwardIterator::RenewIterators:Copy",
-      [&](void* arg) { file_iters_renewed_copy = true; });
+      [&](void* /*arg*/) { file_iters_renewed_copy = true; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
   const int num_records = 1000;
   for (int i = 1; i < num_records; ++i) {
@@ -415,7 +415,7 @@ TEST_F(DBTestTailingIterator, TailingIteratorUpperBound) {
   int immutable_seeks = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "ForwardIterator::SeekInternal:Immutable",
-      [&](void* arg) { ++immutable_seeks; });
+      [&](void* /*arg*/) { ++immutable_seeks; });
 
   // Seek to 13. This should not require any immutable seeks.
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();

db/db_test.cc

@@ -234,11 +234,11 @@ TEST_F(DBTest, SkipDelay) {
       std::atomic<int> sleep_count(0);
       rocksdb::SyncPoint::GetInstance()->SetCallBack(
           "DBImpl::DelayWrite:Sleep",
-          [&](void* arg) { sleep_count.fetch_add(1); });
+          [&](void* /*arg*/) { sleep_count.fetch_add(1); });
       std::atomic<int> wait_count(0);
       rocksdb::SyncPoint::GetInstance()->SetCallBack(
           "DBImpl::DelayWrite:Wait",
-          [&](void* arg) { wait_count.fetch_add(1); });
+          [&](void* /*arg*/) { wait_count.fetch_add(1); });
       rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
       WriteOptions wo;
@@ -1682,7 +1682,7 @@ TEST_F(DBTest, CustomComparator) {
 
 TEST_F(DBTest, DBOpen_Options) {
   Options options = CurrentOptions();
-  std::string dbname = test::TmpDir(env_) + "/db_options_test";
+  std::string dbname = test::PerThreadDBPath("db_options_test");
   ASSERT_OK(DestroyDB(dbname, options));
 
   // Does not exist, and create_if_missing == false: error
@@ -1740,7 +1740,7 @@ TEST_F(DBTest, DBOpen_Change_NumLevels) {
 }
 
 TEST_F(DBTest, DestroyDBMetaDatabase) {
-  std::string dbname = test::TmpDir(env_) + "/db_meta";
+  std::string dbname = test::PerThreadDBPath("db_meta");
   ASSERT_OK(env_->CreateDirIfMissing(dbname));
   std::string metadbname = MetaDatabaseName(dbname, 0);
   ASSERT_OK(env_->CreateDirIfMissing(metadbname));
@@ -2153,10 +2153,14 @@ TEST_F(DBTest, GroupCommitTest) {
   do {
     Options options = CurrentOptions();
     options.env = env_;
-    env_->log_write_slowdown_.store(100);
     options.statistics = rocksdb::CreateDBStatistics();
     Reopen(options);
 
+    rocksdb::SyncPoint::GetInstance()->LoadDependency(
+        {{"WriteThread::JoinBatchGroup:BeganWaiting",
+          "DBImpl::WriteImpl:BeforeLeaderEnters"}});
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
     // Start threads
     GCThread thread[kGCNumThreads];
     for (int id = 0; id < kGCNumThreads; id++) {
@@ -2165,13 +2169,7 @@ TEST_F(DBTest, GroupCommitTest) {
       thread[id].done = false;
       env_->StartThread(GCThreadBody, &thread[id]);
     }
-
-    for (int id = 0; id < kGCNumThreads; id++) {
-      while (thread[id].done == false) {
-        env_->SleepForMicroseconds(100000);
-      }
-    }
-    env_->log_write_slowdown_.store(0);
+    env_->WaitForJoin();
 
     ASSERT_GT(TestGetTickerCount(options, WRITE_DONE_BY_OTHER), 0);
 
@@ -2563,6 +2561,7 @@ class ModelDB : public DB {
   std::string name_ = "";
 };
 
+#ifndef ROCKSDB_VALGRIND_RUN
 static std::string RandomKey(Random* rnd, int minimum = 0) {
   int len;
   do {
@@ -2719,6 +2718,7 @@ TEST_P(DBTestRandomized, Randomized) {
   if (model_snap != nullptr) model.ReleaseSnapshot(model_snap);
   if (db_snap != nullptr) db_->ReleaseSnapshot(db_snap);
 }
+#endif  // ROCKSDB_VALGRIND_RUN
 
 TEST_F(DBTest, BlockBasedTablePrefixIndexTest) {
   // create a DB with block prefix index
@@ -3348,6 +3348,56 @@ TEST_F(DBTest, WriteSingleThreadEntry) {
   }
 }
 
+TEST_F(DBTest, ConcurrentFlushWAL) {
+  const size_t cnt = 100;
+  Options options;
+  WriteOptions wopt;
+  ReadOptions ropt;
+  for (bool two_write_queues : {false, true}) {
+    for (bool manual_wal_flush : {false, true}) {
+      options.two_write_queues = two_write_queues;
+      options.manual_wal_flush = manual_wal_flush;
+      options.create_if_missing = true;
+      DestroyAndReopen(options);
+      std::vector<port::Thread> threads;
+      threads.emplace_back([&] {
+        for (size_t i = 0; i < cnt; i++) {
+          auto istr = ToString(i);
+          db_->Put(wopt, db_->DefaultColumnFamily(), "a" + istr, "b" + istr);
+        }
+      });
+      if (two_write_queues) {
+        threads.emplace_back([&] {
+          for (size_t i = cnt; i < 2 * cnt; i++) {
+            auto istr = ToString(i);
+            WriteBatch batch;
+            batch.Put("a" + istr, "b" + istr);
+            dbfull()->WriteImpl(wopt, &batch, nullptr, nullptr, 0, true);
+          }
+        });
+      }
+      threads.emplace_back([&] {
+        for (size_t i = 0; i < cnt * 100; i++) {  // FlushWAL is faster than Put
+          db_->FlushWAL(false);
+        }
+      });
+      for (auto& t : threads) {
+        t.join();
+      }
+      options.create_if_missing = false;
+      // Recover from the wal and make sure that it is not corrupted
+      Reopen(options);
+      for (size_t i = 0; i < cnt; i++) {
+        PinnableSlice pval;
+        auto istr = ToString(i);
+        ASSERT_OK(
+            db_->Get(ropt, db_->DefaultColumnFamily(), "a" + istr, &pval));
+        ASSERT_TRUE(pval == ("b" + istr));
+      }
+    }
+  }
+}
+
 #ifndef ROCKSDB_LITE
 TEST_F(DBTest, DynamicMemtableOptions) {
   const uint64_t k64KB = 1 << 16;
@@ -3441,7 +3491,7 @@ TEST_F(DBTest, DynamicMemtableOptions) {
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::DelayWrite:Wait",
-      [&](void* arg) { sleeping_task_low.WakeUp(); });
+      [&](void* /*arg*/) { sleeping_task_low.WakeUp(); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   while (!sleeping_task_low.WokenUp() && count < 256) {
@@ -4557,6 +4607,7 @@ TEST_F(DBTest, FileCreationRandomFailure) {
 }
 
 #ifndef ROCKSDB_LITE
+
 TEST_F(DBTest, DynamicMiscOptions) {
   // Test max_sequential_skip_in_iterations
   Options options;
@@ -4874,6 +4925,7 @@ TEST_P(DBTestWithParam, FilterCompactionTimeTest) {
   options.disable_auto_compactions = true;
   options.create_if_missing = true;
   options.statistics = rocksdb::CreateDBStatistics();
+  options.statistics->stats_level_ = kExceptTimeForMutex;
   options.max_subcompactions = max_subcompactions_;
   DestroyAndReopen(options);
 
@@ -5144,7 +5196,7 @@ TEST_F(DBTest, AutomaticConflictsWithManualCompaction) {
   std::atomic<int> callback_count(0);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::MaybeScheduleFlushOrCompaction:Conflict",
-      [&](void* arg) { callback_count.fetch_add(1); });
+      [&](void* /*arg*/) { callback_count.fetch_add(1); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Random rnd(301);
@@ -5367,7 +5419,7 @@ TEST_F(DBTest, HardLimit) {
 
   std::atomic<int> callback_count(0);
   rocksdb::SyncPoint::GetInstance()->SetCallBack("DBImpl::DelayWrite:Wait",
-                                                 [&](void* arg) {
+                                                 [&](void* /*arg*/) {
                                                    callback_count.fetch_add(1);
                                                    sleeping_task_low.WakeUp();
                                                  });
@@ -5392,21 +5444,20 @@ TEST_F(DBTest, HardLimit) {
   sleeping_task_low.WaitUntilDone();
 }
 
-#ifndef ROCKSDB_LITE
+#if !defined(ROCKSDB_LITE) && !defined(ROCKSDB_DISABLE_STALL_NOTIFICATION)
 class WriteStallListener : public EventListener {
  public:
-  WriteStallListener() : cond_(&mutex_), 
-    condition_(WriteStallCondition::kNormal),
-    expected_(WriteStallCondition::kNormal),
-    expected_set_(false) 
-  {}
+  WriteStallListener()
+      : cond_(&mutex_),
+        condition_(WriteStallCondition::kNormal),
+        expected_(WriteStallCondition::kNormal),
+        expected_set_(false) {}
   void OnStallConditionsChanged(const WriteStallInfo& info) override {
     MutexLock l(&mutex_);
     condition_ = info.condition.cur;
-    if (expected_set_ && 
-      condition_ == expected_) {
-        cond_.Signal();
-        expected_set_ = false;
+    if (expected_set_ && condition_ == expected_) {
+      cond_.Signal();
+      expected_set_ = false;
     }
   }
   bool CheckCondition(WriteStallCondition expected) {
@@ -5504,7 +5555,7 @@ TEST_F(DBTest, SoftLimit) {
 
   // Only allow one compactin going through.
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "BackgroundCallCompaction:0", [&](void* arg) {
+      "BackgroundCallCompaction:0", [&](void* /*arg*/) {
         // Schedule a sleeping task.
         sleeping_task_low.Reset();
         env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
@@ -5614,7 +5665,7 @@ TEST_F(DBTest, LastWriteBufferDelay) {
   sleeping_task.WakeUp();
   sleeping_task.WaitUntilDone();
 }
-#endif  // ROCKSDB_LITE
+#endif  // !defined(ROCKSDB_LITE) && !defined(ROCKSDB_DISABLE_STALL_NOTIFICATION)
 
 TEST_F(DBTest, FailWhenCompressionNotSupportedTest) {
   CompressionType compressions[] = {kZlibCompression, kBZip2Compression,

db/db_test2.cc

@@ -354,7 +354,7 @@ INSTANTIATE_TEST_CASE_P(DBTestSharedWriteBufferAcrossCFs,
                                           std::make_tuple(false, true)));
 
 TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
-  std::string dbname2 = test::TmpDir(env_) + "/db_shared_wb_db2";
+  std::string dbname2 = test::PerThreadDBPath("db_shared_wb_db2");
   Options options = CurrentOptions();
   options.arena_block_size = 4096;
   // Avoid undeterministic value by malloc_usable_size();
@@ -1112,7 +1112,7 @@ TEST_F(DBTest2, PresetCompressionDict) {
 
       size_t out_bytes = 0;
       std::vector<std::string> files;
-      GetSstFiles(dbname_, &files);
+      GetSstFiles(env_, dbname_, &files);
       for (const auto& file : files) {
         uint64_t curr_bytes;
         env_->GetFileSize(dbname_ + "/" + file, &curr_bytes);
@@ -1691,7 +1691,7 @@ TEST_F(DBTest2, SyncPointMarker) {
   std::atomic<int> sync_point_called(0);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBTest2::MarkedPoint",
-      [&](void* arg) { sync_point_called.fetch_add(1); });
+      [&](void* /*arg*/) { sync_point_called.fetch_add(1); });
 
   // The first dependency enforces Marker can be loaded before MarkedPoint.
   // The second checks that thread 1's MarkedPoint should be disabled here.
@@ -1978,7 +1978,7 @@ TEST_F(DBTest2, AutomaticCompactionOverlapManualCompaction) {
   // can fit in L2, these 2 files will be moved to L2 and overlap with
   // the running compaction and break the LSM consistency.
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "CompactionJob::Run():Start", [&](void* arg) {
+      "CompactionJob::Run():Start", [&](void* /*arg*/) {
         ASSERT_OK(
             dbfull()->SetOptions({{"level0_file_num_compaction_trigger", "2"},
                                   {"max_bytes_for_level_base", "1"}}));
@@ -2044,7 +2044,7 @@ TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {
   // the running compaction and break the LSM consistency.
   std::atomic<bool> flag(false);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "CompactionJob::Run():Start", [&](void* arg) {
+      "CompactionJob::Run():Start", [&](void* /*arg*/) {
         if (flag.exchange(true)) {
           // We want to make sure to call this callback only once
           return;
@@ -2296,7 +2296,8 @@ TEST_F(DBTest2, RateLimitedCompactionReads) {
                              kBytesPerKey) /* rate_bytes_per_sec */,
         10 * 1000 /* refill_period_us */, 10 /* fairness */,
         RateLimiter::Mode::kReadsOnly));
-    options.use_direct_io_for_flush_and_compaction = use_direct_io;
+    options.use_direct_reads = options.use_direct_io_for_flush_and_compaction =
+        use_direct_io;
     BlockBasedTableOptions bbto;
     bbto.block_size = 16384;
     bbto.no_block_cache = true;
@@ -2318,7 +2319,7 @@ TEST_F(DBTest2, RateLimitedCompactionReads) {
     // chose 1MB as the upper bound on the total bytes read.
     size_t rate_limited_bytes =
         options.rate_limiter->GetTotalBytesThrough(Env::IO_LOW);
-    // Include the explict prefetch of the footer in direct I/O case.
+    // Include the explicit prefetch of the footer in direct I/O case.
     size_t direct_io_extra = use_direct_io ? 512 * 1024 : 0;
     ASSERT_GE(rate_limited_bytes,
               static_cast<size_t>(kNumKeysPerFile * kBytesPerKey * kNumL0Files +
@@ -2421,7 +2422,7 @@ TEST_F(DBTest2, ReadCallbackTest) {
   class TestReadCallback : public ReadCallback {
    public:
     explicit TestReadCallback(SequenceNumber snapshot) : snapshot_(snapshot) {}
-    virtual bool IsCommitted(SequenceNumber seq) override {
+    virtual bool IsVisible(SequenceNumber seq) override {
       return seq <= snapshot_;
     }
 
@@ -2478,7 +2479,7 @@ TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
   });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::PurgeObsoleteFiles:Begin",
-      [&](void* arg) { env_->SleepForMicroseconds(1000000); });
+      [&](void* /*arg*/) { env_->SleepForMicroseconds(1000000); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   Put("key", "val");
@@ -2501,6 +2502,51 @@ TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
 
 #endif  // ROCKSDB_LITE
 
+TEST_F(DBTest2, PinnableSliceAndMmapReads) {
+  Options options = CurrentOptions();
+  options.allow_mmap_reads = true;
+  options.max_open_files = 100;
+  options.compression = kNoCompression;
+  Reopen(options);
+
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+
+  PinnableSlice pinned_value;
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  // It is not safe to pin mmap files as they might disappear by compaction
+  ASSERT_FALSE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+  dbfull()->TEST_CompactRange(0 /* level */, nullptr /* begin */,
+                              nullptr /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
+
+  // Ensure pinned_value doesn't rely on memory munmap'd by the above
+  // compaction. It crashes if it does.
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+#ifndef ROCKSDB_LITE
+  pinned_value.Reset();
+  // Unsafe to pin mmap files when they could be kicked out of table cache
+  Close();
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  ASSERT_FALSE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+  pinned_value.Reset();
+  // In read-only mode with infinite capacity on table cache it should pin the
+  // value and avoid the memcpy
+  Close();
+  options.max_open_files = -1;
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  ASSERT_TRUE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+#endif
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/db_test_util.cc

@@ -63,7 +63,7 @@ DBTestBase::DBTestBase(const std::string path)
       option_config_(kDefault) {
   env_->SetBackgroundThreads(1, Env::LOW);
   env_->SetBackgroundThreads(1, Env::HIGH);
-  dbname_ = test::TmpDir(env_) + path;
+  dbname_ = test::PerThreadDBPath(env_, path);
   alternative_wal_dir_ = dbname_ + "/wal";
   alternative_db_log_dir_ = dbname_ + "/db_log_dir";
   auto options = CurrentOptions();
@@ -118,7 +118,8 @@ bool DBTestBase::ShouldSkipOptions(int option_config, int skip_mask) {
 
     if ((skip_mask & kSkipUniversalCompaction) &&
         (option_config == kUniversalCompaction ||
-         option_config == kUniversalCompactionMultiLevel)) {
+         option_config == kUniversalCompactionMultiLevel ||
+         option_config == kUniversalSubcompactions)) {
       return true;
     }
     if ((skip_mask & kSkipMergePut) && option_config == kMergePut) {
@@ -346,6 +347,26 @@ Options DBTestBase::GetOptions(
           NewHashCuckooRepFactory(options.write_buffer_size));
       options.allow_concurrent_memtable_write = false;
       break;
+      case kDirectIO: {
+        options.use_direct_reads = true;
+        options.use_direct_io_for_flush_and_compaction = true;
+        options.compaction_readahead_size = 2 * 1024 * 1024;
+  #if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
+      !defined(OS_AIX) && !defined(OS_OPENBSD)
+        rocksdb::SyncPoint::GetInstance()->SetCallBack(
+            "NewWritableFile:O_DIRECT", [&](void* arg) {
+              int* val = static_cast<int*>(arg);
+              *val &= ~O_DIRECT;
+            });
+        rocksdb::SyncPoint::GetInstance()->SetCallBack(
+            "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
+              int* val = static_cast<int*>(arg);
+              *val &= ~O_DIRECT;
+            });
+        rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  #endif
+        break;
+      }
 #endif  // ROCKSDB_LITE
     case kMergePut:
       options.merge_operator = MergeOperators::CreatePutOperator();
@@ -428,6 +449,17 @@ Options DBTestBase::GetOptions(
       options.prefix_extractor.reset(NewNoopTransform());
       break;
     }
+    case kBlockBasedTableWithPartitionedIndexFormat3: {
+      table_options.format_version = 3;
+      // Format 3 changes the binary index format. Since partitioned index is a
+      // super-set of simple indexes, we are also using kTwoLevelIndexSearch to
+      // test this format.
+      table_options.index_type = BlockBasedTableOptions::kTwoLevelIndexSearch;
+      // The top-level index in partition filters are also affected by format 3.
+      table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
+      table_options.partition_filters = true;
+      break;
+    }
     case kBlockBasedTableWithIndexRestartInterval: {
       table_options.index_block_restart_interval = 8;
       break;
@@ -460,26 +492,6 @@ Options DBTestBase::GetOptions(
       options.enable_write_thread_adaptive_yield = true;
       break;
     }
-    case kDirectIO: {
-      options.use_direct_reads = true;
-      options.use_direct_io_for_flush_and_compaction = true;
-      options.compaction_readahead_size = 2 * 1024 * 1024;
-#if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
-    !defined(OS_AIX) && !defined(OS_OPENBSD)
-      rocksdb::SyncPoint::GetInstance()->SetCallBack(
-          "NewWritableFile:O_DIRECT", [&](void* arg) {
-            int* val = static_cast<int*>(arg);
-            *val &= ~O_DIRECT;
-          });
-      rocksdb::SyncPoint::GetInstance()->SetCallBack(
-          "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
-            int* val = static_cast<int*>(arg);
-            *val &= ~O_DIRECT;
-          });
-      rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-#endif
-      break;
-    }
     case kPipelinedWrite: {
       options.enable_pipelined_write = true;
       break;
@@ -575,9 +587,17 @@ void DBTestBase::DestroyAndReopen(const Options& options) {
   ASSERT_OK(TryReopen(options));
 }
 
-void DBTestBase::Destroy(const Options& options) {
+void DBTestBase::Destroy(const Options& options, bool delete_cf_paths) {
+  std::vector<ColumnFamilyDescriptor> column_families;
+  if (delete_cf_paths) {
+    for (size_t i = 0; i < handles_.size(); ++i) {
+      ColumnFamilyDescriptor cfdescriptor;
+      handles_[i]->GetDescriptor(&cfdescriptor);
+      column_families.push_back(cfdescriptor);
+    }
+  }
   Close();
-  ASSERT_OK(DestroyDB(dbname_, options));
+  ASSERT_OK(DestroyDB(dbname_, options, column_families));
 }
 
 Status DBTestBase::ReadOnlyReopen(const Options& options) {
@@ -1017,9 +1037,9 @@ std::string DBTestBase::DumpSSTableList() {
   return property;
 }
 
-void DBTestBase::GetSstFiles(std::string path,
+void DBTestBase::GetSstFiles(Env* env, std::string path,
                              std::vector<std::string>* files) {
-  env_->GetChildren(path, files);
+  env->GetChildren(path, files);
 
   files->erase(
       std::remove_if(files->begin(), files->end(), [](std::string name) {
@@ -1031,7 +1051,7 @@ void DBTestBase::GetSstFiles(std::string path,
 
 int DBTestBase::GetSstFileCount(std::string path) {
   std::vector<std::string> files;
-  GetSstFiles(path, &files);
+  DBTestBase::GetSstFiles(env_, path, &files);
   return static_cast<int>(files.size());
 }
 

db/db_test_util.h

@@ -451,8 +451,9 @@ class SpecialEnv : public EnvWrapper {
     return s;
   }
 
-  Status NewSequentialFile(const std::string& f, unique_ptr<SequentialFile>* r,
-                           const EnvOptions& soptions) override {
+  virtual Status NewSequentialFile(const std::string& f,
+                                   unique_ptr<SequentialFile>* r,
+                                   const EnvOptions& soptions) override {
     class CountingFile : public SequentialFile {
      public:
       CountingFile(unique_ptr<SequentialFile>&& target,
@@ -605,7 +606,7 @@ class MockTimeEnv : public EnvWrapper {
   }
 
  private:
-  uint64_t current_time_ = 0;
+  std::atomic<uint64_t> current_time_{0};
 };
 
 #ifndef ROCKSDB_LITE
@@ -693,13 +694,15 @@ class DBTestBase : public testing::Test {
     kConcurrentSkipList = 29,
     kPipelinedWrite = 30,
     kConcurrentWALWrites = 31,
-    kEnd = 32,
-    kDirectIO = 33,
-    kLevelSubcompactions = 34,
-    kUniversalSubcompactions = 35,
-    kBlockBasedTableWithIndexRestartInterval = 36,
-    kBlockBasedTableWithPartitionedIndex = 37,
-    kPartitionedFilterWithNewTableReaderForCompactions = 38,
+    kDirectIO,
+    kLevelSubcompactions,
+    kBlockBasedTableWithIndexRestartInterval,
+    kBlockBasedTableWithPartitionedIndex,
+    kBlockBasedTableWithPartitionedIndexFormat3,
+    kPartitionedFilterWithNewTableReaderForCompactions,
+    kUniversalSubcompactions,
+    // This must be the last line
+    kEnd,
   };
 
  public:
@@ -730,6 +733,13 @@ class DBTestBase : public testing::Test {
     kSkipMmapReads = 256,
   };
 
+  const int kRangeDelSkipConfigs =
+      // Plain tables do not support range deletions.
+      kSkipPlainTable |
+      // MmapReads disables the iterator pinning that RangeDelAggregator
+      // requires.
+      kSkipMmapReads;
+
   explicit DBTestBase(const std::string path);
 
   ~DBTestBase();
@@ -803,7 +813,7 @@ class DBTestBase : public testing::Test {
 
   void DestroyAndReopen(const Options& options);
 
-  void Destroy(const Options& options);
+  void Destroy(const Options& options, bool delete_cf_paths = false);
 
   Status ReadOnlyReopen(const Options& options);
 
@@ -903,7 +913,8 @@ class DBTestBase : public testing::Test {
 
   std::string DumpSSTableList();
 
-  void GetSstFiles(std::string path, std::vector<std::string>* files);
+  static void GetSstFiles(Env* env, std::string path,
+                          std::vector<std::string>* files);
 
   int GetSstFileCount(std::string path);
 

db/db_universal_compaction_test.cc

@@ -10,6 +10,7 @@
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #if !defined(ROCKSDB_LITE)
+#include "rocksdb/utilities/table_properties_collectors.h"
 #include "util/sync_point.h"
 
 namespace rocksdb {
@@ -40,6 +41,12 @@ class DBTestUniversalCompaction : public DBTestUniversalCompactionBase {
       DBTestUniversalCompactionBase("/db_universal_compaction_test") {}
 };
 
+class DBTestUniversalDeleteTrigCompaction : public DBTestBase {
+ public:
+  DBTestUniversalDeleteTrigCompaction()
+      : DBTestBase("/db_universal_compaction_test") {}
+};
+
 namespace {
 void VerifyCompactionResult(
     const ColumnFamilyMetaData& cf_meta,
@@ -655,7 +662,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTargetLevel) {
   ASSERT_EQ("0,0,0,0,1", FilesPerLevel(0));
 }
 
-
+#ifndef ROCKSDB_VALGRIND_RUN
 class DBTestUniversalCompactionMultiLevels
     : public DBTestUniversalCompactionBase {
  public:
@@ -691,13 +698,14 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionMultiLevels) {
     ASSERT_EQ(Get(1, Key(i % num_keys)), Key(i));
   }
 }
+
 // Tests universal compaction with trivial move enabled
 TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
   int32_t trivial_move = 0;
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial", [&](void* arg) {
         non_trivial_move++;
@@ -769,7 +777,7 @@ TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
   std::atomic<int> num_compactions_running(0);
   std::atomic<bool> has_parallel(false);
   rocksdb::SyncPoint::GetInstance()->SetCallBack("CompactionJob::Run():Start",
-                                                 [&](void* arg) {
+                                                 [&](void* /*arg*/) {
     if (num_compactions_running.fetch_add(1) > 0) {
       has_parallel.store(true);
       return;
@@ -784,7 +792,7 @@ TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
   });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "CompactionJob::Run():End",
-      [&](void* arg) { num_compactions_running.fetch_add(-1); });
+      [&](void* /*arg*/) { num_compactions_running.fetch_add(-1); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   options = CurrentOptions(options);
@@ -933,6 +941,7 @@ INSTANTIATE_TEST_CASE_P(DBTestUniversalCompactionParallel,
                         DBTestUniversalCompactionParallel,
                         ::testing::Combine(::testing::Values(1, 10),
                                            ::testing::Values(false)));
+#endif  // ROCKSDB_VALGRIND_RUN
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
   Options options = CurrentOptions();
@@ -1148,13 +1157,14 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio2) {
   ASSERT_LT(TotalSize(), 120000U * 12 * 0.8 + 120000 * 2);
 }
 
+#ifndef ROCKSDB_VALGRIND_RUN
 // Test that checks trivial move in universal compaction
 TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
   int32_t trivial_move = 0;
   int32_t non_trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial", [&](void* arg) {
         non_trivial_move++;
@@ -1200,7 +1210,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
   int32_t trivial_move = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:TrivialMove",
-      [&](void* arg) { trivial_move++; });
+      [&](void* /*arg*/) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial", [&](void* arg) {
         ASSERT_TRUE(arg != nullptr);
@@ -1240,6 +1250,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
 
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
+#endif  // ROCKSDB_VALGRIND_RUN
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
   Options options = CurrentOptions();
@@ -1343,6 +1354,146 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
   Destroy(options);
 }
 
+TEST_P(DBTestUniversalCompaction, UniversalCompactionCFPathUse) {
+  Options options = CurrentOptions();
+  options.db_paths.emplace_back(dbname_, 300 * 1024);
+  options.db_paths.emplace_back(dbname_ + "_2", 300 * 1024);
+  options.db_paths.emplace_back(dbname_ + "_3", 500 * 1024);
+  options.db_paths.emplace_back(dbname_ + "_4", 1024 * 1024 * 1024);
+  options.memtable_factory.reset(
+      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+  options.compaction_style = kCompactionStyleUniversal;
+  options.compaction_options_universal.size_ratio = 10;
+  options.write_buffer_size = 111 << 10;  // 114KB
+  options.arena_block_size = 4 << 10;
+  options.level0_file_num_compaction_trigger = 2;
+  options.num_levels = 1;
+
+  std::vector<Options> option_vector;
+  option_vector.emplace_back(options);
+  ColumnFamilyOptions cf_opt1(options), cf_opt2(options);
+  // Configure CF1 specific paths.
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1", 300 * 1024);
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_2", 300 * 1024);
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_3", 500 * 1024);
+  cf_opt1.cf_paths.emplace_back(dbname_ + "cf1_4", 1024 * 1024 * 1024);
+  option_vector.emplace_back(DBOptions(options), cf_opt1);
+  CreateColumnFamilies({"one"},option_vector[1]);
+
+  // Configura CF2 specific paths.
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2", 300 * 1024);
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_2", 300 * 1024);
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_3", 500 * 1024);
+  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_4", 1024 * 1024 * 1024);
+  option_vector.emplace_back(DBOptions(options), cf_opt2);
+  CreateColumnFamilies({"two"},option_vector[2]);
+
+  ReopenWithColumnFamilies({"default", "one", "two"}, option_vector);
+
+  Random rnd(301);
+  int key_idx = 0;
+  int key_idx1 = 0;
+  int key_idx2 = 0;
+
+  auto generate_file = [&]() {
+    GenerateNewFile(0, &rnd, &key_idx);
+    GenerateNewFile(1, &rnd, &key_idx1);
+    GenerateNewFile(2, &rnd, &key_idx2);
+  };
+
+  auto check_sstfilecount = [&](int path_id, int expected) {
+    ASSERT_EQ(expected, GetSstFileCount(options.db_paths[path_id].path));
+    ASSERT_EQ(expected, GetSstFileCount(cf_opt1.cf_paths[path_id].path));
+    ASSERT_EQ(expected, GetSstFileCount(cf_opt2.cf_paths[path_id].path));
+  };
+
+  auto check_getvalues = [&]() {
+    for (int i = 0; i < key_idx; i++) {
+      auto v = Get(0, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+
+    for (int i = 0; i < key_idx1; i++) {
+      auto v = Get(1, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+
+    for (int i = 0; i < key_idx2; i++) {
+      auto v = Get(2, Key(i));
+      ASSERT_NE(v, "NOT_FOUND");
+      ASSERT_TRUE(v.size() == 1 || v.size() == 990);
+    }
+  };
+
+  // First three 110KB files are not going to second path.
+  // After that, (100K, 200K)
+  for (int num = 0; num < 3; num++) {
+    generate_file();
+  }
+
+  // Another 110KB triggers a compaction to 400K file to second path
+  generate_file();
+  check_sstfilecount(2, 1);
+
+  // (1, 4)
+  generate_file();
+  check_sstfilecount(2, 1);
+  check_sstfilecount(0, 1);
+
+  // (1,1,4) -> (2, 4)
+  generate_file();
+  check_sstfilecount(2, 1);
+  check_sstfilecount(1, 1);
+  check_sstfilecount(0, 0);
+
+  // (1, 2, 4) -> (3, 4)
+  generate_file();
+  check_sstfilecount(2, 1);
+  check_sstfilecount(1, 1);
+  check_sstfilecount(0, 0);
+
+  // (1, 3, 4) -> (8)
+  generate_file();
+  check_sstfilecount(3, 1);
+
+  // (1, 8)
+  generate_file();
+  check_sstfilecount(3, 1);
+  check_sstfilecount(0, 1);
+
+  // (1, 1, 8) -> (2, 8)
+  generate_file();
+  check_sstfilecount(3, 1);
+  check_sstfilecount(1, 1);
+
+  // (1, 2, 8) -> (3, 8)
+  generate_file();
+  check_sstfilecount(3, 1);
+  check_sstfilecount(1, 1);
+  check_sstfilecount(0, 0);
+
+  // (1, 3, 8) -> (4, 8)
+  generate_file();
+  check_sstfilecount(2, 1);
+  check_sstfilecount(3, 1);
+
+  // (1, 4, 8) -> (5, 8)
+  generate_file();
+  check_sstfilecount(3, 1);
+  check_sstfilecount(2, 1);
+  check_sstfilecount(0, 0);
+
+  check_getvalues();
+
+  ReopenWithColumnFamilies({"default", "one", "two"}, option_vector);
+
+  check_getvalues();
+
+  Destroy(options, true);
+}
+
 TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
   std::function<void(int)> verify_func = [&](int num_keys_in_db) {
     std::string keys_in_db;
@@ -1540,50 +1691,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSecondPathRatio) {
   Destroy(options);
 }
 
-TEST_P(DBTestUniversalCompaction, FullCompactionInBottomPriThreadPool) {
-  const int kNumFilesTrigger = 3;
-  Env::Default()->SetBackgroundThreads(1, Env::Priority::BOTTOM);
-  for (bool allow_ingest_behind : {false, true}) {
-    Options options = CurrentOptions();
-    options.allow_ingest_behind = allow_ingest_behind;
-    options.compaction_style = kCompactionStyleUniversal;
-    options.num_levels = num_levels_;
-    options.write_buffer_size = 100 << 10;     // 100KB
-    options.target_file_size_base = 32 << 10;  // 32KB
-    options.level0_file_num_compaction_trigger = kNumFilesTrigger;
-    // Trigger compaction if size amplification exceeds 110%
-    options.compaction_options_universal.max_size_amplification_percent = 110;
-    DestroyAndReopen(options);
-
-    int num_bottom_pri_compactions = 0;
-    SyncPoint::GetInstance()->SetCallBack(
-        "DBImpl::BGWorkBottomCompaction",
-        [&](void* arg) { ++num_bottom_pri_compactions; });
-    SyncPoint::GetInstance()->EnableProcessing();
-
-    Random rnd(301);
-    for (int num = 0; num < kNumFilesTrigger; num++) {
-      ASSERT_EQ(NumSortedRuns(), num);
-      int key_idx = 0;
-      GenerateNewFile(&rnd, &key_idx);
-    }
-    dbfull()->TEST_WaitForCompact();
-
-    if (allow_ingest_behind || num_levels_ > 1) {
-      // allow_ingest_behind increases number of levels while sanitizing.
-      ASSERT_EQ(1, num_bottom_pri_compactions);
-    } else {
-      // for single-level universal, everything's bottom level so nothing should
-      // be executed in bottom-pri thread pool.
-      ASSERT_EQ(0, num_bottom_pri_compactions);
-    }
-    // Verify that size amplification did occur
-    ASSERT_EQ(NumSortedRuns(), 1);
-    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
-  }
-  Env::Default()->SetBackgroundThreads(0, Env::Priority::BOTTOM);
-}
-
 TEST_P(DBTestUniversalCompaction, ConcurrentBottomPriLowPriCompactions) {
   if (num_levels_ == 1) {
     // for single-level universal, everything's bottom level so nothing should
@@ -1656,7 +1763,7 @@ TEST_P(DBTestUniversalCompaction, RecalculateScoreAfterPicking) {
 
   std::atomic<int> num_compactions_attempted(0);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:Start", [&](void* arg) {
+      "DBImpl::BackgroundCompaction:Start", [&](void* /*arg*/) {
         ++num_compactions_attempted;
       });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
@@ -1674,6 +1781,63 @@ TEST_P(DBTestUniversalCompaction, RecalculateScoreAfterPicking) {
   ASSERT_EQ(NumSortedRuns(), 5);
 }
 
+TEST_P(DBTestUniversalCompaction, FinalSortedRunCompactFilesConflict) {
+  // Regression test for conflict between:
+  // (1) Running CompactFiles including file in the final sorted run; and
+  // (2) Picking universal size-amp-triggered compaction, which always includes
+  //     the final sorted run.
+  if (exclusive_manual_compaction_) {
+    return;
+  }
+
+  Options opts = CurrentOptions();
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.compaction_options_universal.max_size_amplification_percent = 50;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compression = kNoCompression;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.max_background_compactions = 2;
+  opts.num_levels = num_levels_;
+  Reopen(opts);
+
+  // make sure compaction jobs can be parallelized
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();
+
+  Put("key", "val");
+  Flush();
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  ASSERT_EQ(NumTableFilesAtLevel(num_levels_ - 1), 1);
+  ColumnFamilyMetaData cf_meta;
+  ColumnFamilyHandle* default_cfh = db_->DefaultColumnFamily();
+  dbfull()->GetColumnFamilyMetaData(default_cfh, &cf_meta);
+  ASSERT_EQ(1, cf_meta.levels[num_levels_ - 1].files.size());
+  std::string first_sst_filename =
+      cf_meta.levels[num_levels_ - 1].files[0].name;
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"CompactFilesImpl:0",
+        "DBTestUniversalCompaction:FinalSortedRunCompactFilesConflict:0"},
+       {"DBImpl::BackgroundCompaction():AfterPickCompaction",
+        "CompactFilesImpl:1"}});
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::thread compact_files_thread([&]() {
+    ASSERT_OK(dbfull()->CompactFiles(CompactionOptions(), default_cfh,
+        {first_sst_filename}, num_levels_ - 1));
+  });
+
+  TEST_SYNC_POINT(
+      "DBTestUniversalCompaction:FinalSortedRunCompactFilesConflict:0");
+  for (int i = 0; i < 2; ++i) {
+    Put("key", "val");
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+
+  compact_files_thread.join();
+}
+
 INSTANTIATE_TEST_CASE_P(UniversalCompactionNumLevels, DBTestUniversalCompaction,
                         ::testing::Combine(::testing::Values(1, 3, 5),
                                            ::testing::Bool()));
@@ -1749,6 +1913,241 @@ INSTANTIATE_TEST_CASE_P(DBTestUniversalManualCompactionOutputPathId,
                         ::testing::Combine(::testing::Values(1, 8),
                                            ::testing::Bool()));
 
+TEST_F(DBTestUniversalDeleteTrigCompaction, BasicL0toL1) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  //  MoveFilesToLevel(6);
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, SingleLevel) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.num_levels = 1;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, MultipleLevels) {
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 4;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 500; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 500; i < 1000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 1000; i < 1500; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 1500; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+
+  for (i = 1999; i < 2333; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2333; i < 2666; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2666; i < 2999; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+  ASSERT_GT(NumTableFilesAtLevel(5), 0);
+
+  for (i = 1900; i < 2100; ++i) {
+    Delete(Key(i));
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(1));
+  ASSERT_EQ(0, NumTableFilesAtLevel(2));
+  ASSERT_EQ(0, NumTableFilesAtLevel(3));
+  ASSERT_EQ(0, NumTableFilesAtLevel(4));
+  ASSERT_EQ(0, NumTableFilesAtLevel(5));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, OverlappingL0) {
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 5;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2000; i < 3000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 3500; i < 4000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2900; i < 3100; ++i) {
+    Delete(Key(i));
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, IngestBehind) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.allow_ingest_behind = true;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  //  MoveFilesToLevel(6);
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(6));
+  ASSERT_GT(NumTableFilesAtLevel(5), 0);
+}
+
 }  // namespace rocksdb
 
 #endif  // !defined(ROCKSDB_LITE)

db/db_wal_test.cc

@@ -20,6 +20,106 @@ class DBWALTest : public DBTestBase {
   DBWALTest() : DBTestBase("/db_wal_test") {}
 };
 
+// A SpecialEnv enriched to give more insight about deleted files
+class EnrichedSpecialEnv : public SpecialEnv {
+ public:
+  explicit EnrichedSpecialEnv(Env* base) : SpecialEnv(base) {}
+  Status NewSequentialFile(const std::string& f, unique_ptr<SequentialFile>* r,
+                           const EnvOptions& soptions) override {
+    InstrumentedMutexLock l(&env_mutex_);
+    if (f == skipped_wal) {
+      deleted_wal_reopened = true;
+      if (IsWAL(f) && largetest_deleted_wal.size() != 0 &&
+          f.compare(largetest_deleted_wal) <= 0) {
+        gap_in_wals = true;
+      }
+    }
+    return SpecialEnv::NewSequentialFile(f, r, soptions);
+  }
+  Status DeleteFile(const std::string& fname) override {
+    if (IsWAL(fname)) {
+      deleted_wal_cnt++;
+      InstrumentedMutexLock l(&env_mutex_);
+      // If this is the first WAL, remember its name and skip deleting it. We
+      // remember its name partly because the application might attempt to
+      // delete the file again.
+      if (skipped_wal.size() != 0 && skipped_wal != fname) {
+        if (largetest_deleted_wal.size() == 0 ||
+            largetest_deleted_wal.compare(fname) < 0) {
+          largetest_deleted_wal = fname;
+        }
+      } else {
+        skipped_wal = fname;
+        return Status::OK();
+      }
+    }
+    return SpecialEnv::DeleteFile(fname);
+  }
+  bool IsWAL(const std::string& fname) {
+    // printf("iswal %s\n", fname.c_str());
+    return fname.compare(fname.size() - 3, 3, "log") == 0;
+  }
+
+  InstrumentedMutex env_mutex_;
+  // the wal whose actual delete was skipped by the env
+  std::string skipped_wal = "";
+  // the largest WAL that was requested to be deleted
+  std::string largetest_deleted_wal = "";
+  // number of WALs that were successfully deleted
+  std::atomic<size_t> deleted_wal_cnt = {0};
+  // the WAL whose delete from fs was skipped is reopened during recovery
+  std::atomic<bool> deleted_wal_reopened = {false};
+  // whether a gap in the WALs was detected during recovery
+  std::atomic<bool> gap_in_wals = {false};
+};
+
+class DBWALTestWithEnrichedEnv : public DBTestBase {
+ public:
+  DBWALTestWithEnrichedEnv() : DBTestBase("/db_wal_test") {
+    enriched_env_ = new EnrichedSpecialEnv(env_->target());
+    auto options = CurrentOptions();
+    options.env = enriched_env_;
+    Reopen(options);
+    delete env_;
+    // to be deleted by the parent class
+    env_ = enriched_env_;
+  }
+
+ protected:
+  EnrichedSpecialEnv* enriched_env_;
+};
+
+// Test that the recovery would successfully avoid the gaps between the logs.
+// One known scenario that could cause this is that the application issue the
+// WAL deletion out of order. For the sake of simplicity in the test, here we
+// create the gap by manipulating the env to skip deletion of the first WAL but
+// not the ones after it.
+TEST_F(DBWALTestWithEnrichedEnv, SkipDeletedWALs) {
+  auto options = last_options_;
+  // To cause frequent WAL deletion
+  options.write_buffer_size = 128;
+  Reopen(options);
+
+  WriteOptions writeOpt = WriteOptions();
+  for (int i = 0; i < 128 * 5; i++) {
+    ASSERT_OK(dbfull()->Put(writeOpt, "foo", "v1"));
+  }
+  FlushOptions fo;
+  fo.wait = true;
+  ASSERT_OK(db_->Flush(fo));
+
+  // some wals are deleted
+  ASSERT_NE(0, enriched_env_->deleted_wal_cnt);
+  // but not the first one
+  ASSERT_NE(0, enriched_env_->skipped_wal.size());
+
+  // Test that the WAL that was not deleted will be skipped during recovery
+  options = last_options_;
+  Reopen(options);
+  ASSERT_FALSE(enriched_env_->deleted_wal_reopened);
+  ASSERT_FALSE(enriched_env_->gap_in_wals);
+}
+
 TEST_F(DBWALTest, WAL) {
   do {
     CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
@@ -891,7 +991,7 @@ TEST_F(DBWALTest, kPointInTimeRecoveryCFConsistency) {
 
   // Record the offset at this point
   Env* env = options.env;
-  int wal_file_id = RecoveryTestHelper::kWALFileOffset + 1;
+  uint64_t wal_file_id = dbfull()->TEST_LogfileNumber();
   std::string fname = LogFileName(dbname_, wal_file_id);
   uint64_t offset_to_corrupt;
   ASSERT_OK(env->GetFileSize(fname, &offset_to_corrupt));

db/db_write_test.cc

@@ -50,6 +50,7 @@ TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
   std::atomic<int> leader_count{0};
   std::vector<port::Thread> threads;
   mock_env->SetFilesystemActive(false);
+
   // Wait until all threads linked to write threads, to make sure
   // all threads join the same batch group.
   SyncPoint::GetInstance()->SetCallBack(
@@ -68,7 +69,19 @@ TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
     threads.push_back(port::Thread(
         [&](int index) {
           // All threads should fail.
-          ASSERT_FALSE(Put("key" + ToString(index), "value").ok());
+          auto res = Put("key" + ToString(index), "value");
+          if (options.manual_wal_flush) {
+            ASSERT_TRUE(res.ok());
+            // we should see fs error when we do the flush
+
+            // TSAN reports a false alarm for lock-order-inversion but Open and
+            // FlushWAL are not run concurrently. Disabling this until TSAN is
+            // fixed.
+            // res = dbfull()->FlushWAL(false);
+            // ASSERT_FALSE(res.ok());
+          } else {
+            ASSERT_FALSE(res.ok());
+          }
         },
         i));
   }
@@ -80,6 +93,53 @@ TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
   Close();
 }
 
+TEST_P(DBWriteTest, ManualWalFlushInEffect) {
+  Options options = GetOptions();
+  Reopen(options);
+  // try the 1st WAL created during open
+  ASSERT_TRUE(Put("key" + ToString(0), "value").ok());
+  ASSERT_TRUE(options.manual_wal_flush != dbfull()->TEST_WALBufferIsEmpty());
+  ASSERT_TRUE(dbfull()->FlushWAL(false).ok());
+  ASSERT_TRUE(dbfull()->TEST_WALBufferIsEmpty());
+  // try the 2nd wal created during SwitchWAL
+  dbfull()->TEST_SwitchWAL();
+  ASSERT_TRUE(Put("key" + ToString(0), "value").ok());
+  ASSERT_TRUE(options.manual_wal_flush != dbfull()->TEST_WALBufferIsEmpty());
+  ASSERT_TRUE(dbfull()->FlushWAL(false).ok());
+  ASSERT_TRUE(dbfull()->TEST_WALBufferIsEmpty());
+}
+
+TEST_P(DBWriteTest, IOErrorOnWALWriteTriggersReadOnlyMode) {
+  std::unique_ptr<FaultInjectionTestEnv> mock_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetOptions();
+  options.env = mock_env.get();
+  Reopen(options);
+  for (int i = 0; i < 2; i++) {
+    // Forcibly fail WAL write for the first Put only. Subsequent Puts should
+    // fail due to read-only mode
+    mock_env->SetFilesystemActive(i != 0);
+    auto res = Put("key" + ToString(i), "value");
+    // TSAN reports a false alarm for lock-order-inversion but Open and
+    // FlushWAL are not run concurrently. Disabling this until TSAN is
+    // fixed.
+    /*
+    if (options.manual_wal_flush && i == 0) {
+      // even with manual_wal_flush the 2nd Put should return error because of
+      // the read-only mode
+      ASSERT_TRUE(res.ok());
+      // we should see fs error when we do the flush
+      res = dbfull()->FlushWAL(false);
+    }
+    */
+    if (!options.manual_wal_flush) {
+      ASSERT_FALSE(res.ok());
+    }
+  }
+  // Close before mock_env destruct.
+  Close();
+}
+
 INSTANTIATE_TEST_CASE_P(DBWriteTestInstance, DBWriteTest,
                         testing::Values(DBTestBase::kDefault,
                                         DBTestBase::kConcurrentWALWrites,

db/dbformat.cc

@@ -46,6 +46,8 @@ EntryType GetEntryType(ValueType value_type) {
       return kEntrySingleDelete;
     case kTypeMerge:
       return kEntryMerge;
+    case kTypeRangeDeletion:
+      return kEntryRangeDeletion;
     default:
       return kEntryOther;
   }
@@ -106,45 +108,6 @@ const char* InternalKeyComparator::Name() const {
   return name_.c_str();
 }
 
-int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
-  // Order by:
-  //    increasing user key (according to user-supplied comparator)
-  //    decreasing sequence number
-  //    decreasing type (though sequence# should be enough to disambiguate)
-  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
-  PERF_COUNTER_ADD(user_key_comparison_count, 1);
-  if (r == 0) {
-    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8);
-    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8);
-    if (anum > bnum) {
-      r = -1;
-    } else if (anum < bnum) {
-      r = +1;
-    }
-  }
-  return r;
-}
-
-int InternalKeyComparator::CompareKeySeq(const Slice& akey,
-                                         const Slice& bkey) const {
-  // Order by:
-  //    increasing user key (according to user-supplied comparator)
-  //    decreasing sequence number
-  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
-  PERF_COUNTER_ADD(user_key_comparison_count, 1);
-  if (r == 0) {
-    // Shift the number to exclude the last byte which contains the value type
-    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8) >> 8;
-    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8) >> 8;
-    if (anum > bnum) {
-      r = -1;
-    } else if (anum < bnum) {
-      r = +1;
-    }
-  }
-  return r;
-}
-
 int InternalKeyComparator::Compare(const ParsedInternalKey& a,
                                    const ParsedInternalKey& b) const {
   // Order by:

db/dbformat.h

@@ -11,6 +11,7 @@
 #include <stdio.h>
 #include <string>
 #include <utility>
+#include "monitoring/perf_context_imp.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/filter_policy.h"
@@ -54,7 +55,11 @@ enum ValueType : unsigned char {
   // is not mistakenly read by another, which would result into data
   // inconsistency.
   kTypeBeginPersistedPrepareXID = 0x12,  // WAL only.
-  kMaxValue = 0x7F                       // Not used for storing records.
+  // Similar to kTypeBeginPersistedPrepareXID, this is to ensure that WAL
+  // generated by WriteUnprepared write policy is not mistakenly read by
+  // another.
+  kTypeBeginUnprepareXID = 0x13,  // WAL only.
+  kMaxValue = 0x7F                // Not used for storing records.
 };
 
 // Defined in dbformat.cc
@@ -135,10 +140,14 @@ inline Slice ExtractUserKey(const Slice& internal_key) {
   return Slice(internal_key.data(), internal_key.size() - 8);
 }
 
-inline ValueType ExtractValueType(const Slice& internal_key) {
+inline uint64_t ExtractInternalKeyFooter(const Slice& internal_key) {
   assert(internal_key.size() >= 8);
   const size_t n = internal_key.size();
-  uint64_t num = DecodeFixed64(internal_key.data() + n - 8);
+  return DecodeFixed64(internal_key.data() + n - 8);
+}
+
+inline ValueType ExtractValueType(const Slice& internal_key) {
+  uint64_t num = ExtractInternalKeyFooter(internal_key);
   unsigned char c = num & 0xff;
   return static_cast<ValueType>(c);
 }
@@ -601,10 +610,58 @@ struct RangeTombstone {
     return InternalKey(start_key_, seq_, kTypeRangeDeletion);
   }
 
+  // The tombstone end-key is exclusive, so we generate an internal-key here
+  // which has a similar property. Using kMaxSequenceNumber guarantees that
+  // the returned internal-key will compare less than any other internal-key
+  // with the same user-key. This in turn guarantees that the serialized
+  // end-key for a tombstone such as [a-b] will compare less than the key "b".
+  //
   // be careful to use SerializeEndKey(), allocates new memory
   InternalKey SerializeEndKey() const {
-    return InternalKey(end_key_, seq_, kTypeRangeDeletion);
+    return InternalKey(end_key_, kMaxSequenceNumber, kTypeRangeDeletion);
   }
 };
 
+inline
+int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
+  // Order by:
+  //    increasing user key (according to user-supplied comparator)
+  //    decreasing sequence number
+  //    decreasing type (though sequence# should be enough to disambiguate)
+  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
+  PERF_COUNTER_ADD(user_key_comparison_count, 1);
+  if (r == 0) {
+    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8);
+    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8);
+    if (anum > bnum) {
+      r = -1;
+    } else if (anum < bnum) {
+      r = +1;
+    }
+  }
+  return r;
+}
+
+inline
+int InternalKeyComparator::CompareKeySeq(const Slice& akey,
+                                         const Slice& bkey) const {
+  // Order by:
+  //    increasing user key (according to user-supplied comparator)
+  //    decreasing sequence number
+  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
+  PERF_COUNTER_ADD(user_key_comparison_count, 1);
+  if (r == 0) {
+    // Shift the number to exclude the last byte which contains the value type
+    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8) >> 8;
+    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8) >> 8;
+    if (anum > bnum) {
+      r = -1;
+    } else if (anum < bnum) {
+      r = +1;
+    }
+  }
+  return r;
+}
+
+
 }  // namespace rocksdb

db/dbformat_test.cc

@@ -192,6 +192,13 @@ TEST_F(FormatTest, UpdateInternalKey) {
   ASSERT_EQ(new_val_type, decoded.type);
 }
 
+TEST_F(FormatTest, RangeTombstoneSerializeEndKey) {
+  RangeTombstone t("a", "b", 2);
+  InternalKey k("b", 3, kTypeValue);
+  const InternalKeyComparator cmp(BytewiseComparator());
+  ASSERT_LT(cmp.Compare(t.SerializeEndKey(), k), 0);
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/deletefile_test.cc

@@ -45,7 +45,7 @@ class DeleteFileTest : public testing::Test {
     options_.max_bytes_for_level_base = 1024*1024*1000;
     options_.WAL_ttl_seconds = 300; // Used to test log files
     options_.WAL_size_limit_MB = 1024; // Used to test log files
-    dbname_ = test::TmpDir() + "/deletefile_test";
+    dbname_ = test::PerThreadDBPath("deletefile_test");
     options_.wal_dir = dbname_ + "/wal_files";
 
     // clean up all the files that might have been there before

db/error_handler.cc

@@ -0,0 +1,170 @@
+//  Copyright (c) 2018-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+#include "db/error_handler.h"
+#include "db/event_helpers.h"
+
+namespace rocksdb {
+
+// Maps to help decide the severity of an error based on the
+// BackgroundErrorReason, Code, SubCode and whether db_options.paranoid_checks
+// is set or not. There are 3 maps, going from most specific to least specific
+// (i.e from all 4 fields in a tuple to only the BackgroundErrorReason and
+// paranoid_checks). The less specific map serves as a catch all in case we miss
+// a specific error code or subcode.
+std::map<std::tuple<BackgroundErrorReason, Status::Code, Status::SubCode, bool>,
+         Status::Severity>
+    ErrorSeverityMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         true),
+         Status::Severity::kSoftError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kSpaceLimit,
+                         true),
+         Status::Severity::kHardError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kNoSpace, true),
+         Status::Severity::kSoftError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kNoSpace, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kSpaceLimit, true),
+         Status::Severity::kHardError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         false),
+         Status::Severity::kFatalError},
+};
+
+std::map<std::tuple<BackgroundErrorReason, Status::Code, bool>, Status::Severity>
+    DefaultErrorSeverityMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+};
+
+std::map<std::tuple<BackgroundErrorReason, bool>, Status::Severity>
+    DefaultReasonMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction, false),
+          Status::Severity::kNoError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, false),
+          Status::Severity::kNoError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback, false),
+          Status::Severity::kFatalError},
+        // Errors during Memtable update
+        {std::make_tuple(BackgroundErrorReason::kMemTable, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kMemTable, false),
+          Status::Severity::kFatalError},
+};
+
+Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reason) {
+  db_mutex_->AssertHeld();
+
+  if (bg_err.ok()) {
+    return Status::OK();
+  }
+
+  bool paranoid = db_options_.paranoid_checks;
+  Status::Severity sev = Status::Severity::kFatalError;
+  Status new_bg_err;
+  bool found = false;
+
+  {
+    auto entry = ErrorSeverityMap.find(std::make_tuple(reason, bg_err.code(),
+          bg_err.subcode(), paranoid));
+    if (entry != ErrorSeverityMap.end()) {
+      sev = entry->second;
+      found = true;
+    }
+  }
+
+  if (!found) {
+    auto entry = DefaultErrorSeverityMap.find(std::make_tuple(reason,
+          bg_err.code(), paranoid));
+    if (entry != DefaultErrorSeverityMap.end()) {
+      sev = entry->second;
+      found = true;
+    }
+  }
+
+  if (!found) {
+    auto entry = DefaultReasonMap.find(std::make_tuple(reason, paranoid));
+    if (entry != DefaultReasonMap.end()) {
+      sev = entry->second;
+    }
+  }
+
+  new_bg_err = Status(bg_err, sev);
+  if (!new_bg_err.ok()) {
+    Status s = new_bg_err;
+    EventHelpers::NotifyOnBackgroundError(db_options_.listeners, reason, &s, db_mutex_);
+    if (!s.ok() && (s.severity() > bg_error_.severity())) {
+      bg_error_ = s;
+    }
+  }
+
+  return bg_error_;
+}
+
+}

db/error_handler.h

@@ -0,0 +1,52 @@
+//  Copyright (c) 2018-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 "monitoring/instrumented_mutex.h"
+#include "options/db_options.h"
+#include "rocksdb/listener.h"
+#include "rocksdb/status.h"
+
+namespace rocksdb {
+
+class ErrorHandler {
+  public:
+    ErrorHandler(const ImmutableDBOptions& db_options,
+        InstrumentedMutex* db_mutex)
+      : db_options_(db_options),
+        bg_error_(Status::OK()),
+        db_mutex_(db_mutex)
+      {}
+    ~ErrorHandler() {}
+
+    Status::Severity GetErrorSeverity(BackgroundErrorReason reason,
+        Status::Code code, Status::SubCode subcode);
+
+    Status SetBGError(const Status& bg_err, BackgroundErrorReason reason);
+
+    Status GetBGError()
+    {
+      return bg_error_;
+    }
+
+    void ClearBGError() {
+      bg_error_ = Status::OK();
+    }
+
+    bool IsDBStopped() {
+      return !bg_error_.ok();
+    }
+
+    bool IsBGWorkStopped() {
+      return !bg_error_.ok();
+    }
+
+  private:
+    const ImmutableDBOptions& db_options_;
+    Status bg_error_;
+    InstrumentedMutex* db_mutex_;
+};
+
+}

db/error_handler_test.cc

@@ -0,0 +1,138 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+#include "rocksdb/perf_context.h"
+#include "util/fault_injection_test_env.h"
+#if !defined(ROCKSDB_LITE)
+#include "util/sync_point.h"
+#endif
+
+namespace rocksdb {
+
+class DBErrorHandlingTest : public DBTestBase {
+ public:
+  DBErrorHandlingTest() : DBTestBase("/db_error_handling_test") {}
+};
+
+class DBErrorHandlingEnv : public EnvWrapper {
+  public:
+    DBErrorHandlingEnv() : EnvWrapper(Env::Default()),
+      trig_no_space(false), trig_io_error(false) {}
+
+    void SetTrigNoSpace() {trig_no_space = true;}
+    void SetTrigIoError() {trig_io_error = true;}
+  private:
+    bool trig_no_space;
+    bool trig_io_error;
+};
+
+TEST_F(DBErrorHandlingTest, FLushWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  Status s;
+  DestroyAndReopen(options);
+
+  Put(Key(0), "va;");
+  SyncPoint::GetInstance()->SetCallBack(
+      "FlushJob::Start", [&](void *) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kSoftError);
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_EQ(s, Status::OK());
+
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, CompactionWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.level0_file_num_compaction_trigger = 2;
+  options.env = fault_env.get();
+  Status s;
+  DestroyAndReopen(options);
+
+  Put(Key(0), "va;");
+  Put(Key(2), "va;");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BackgroundCallCompaction:0", [&](void *) {
+      fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put(Key(1), "val");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  s = dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kSoftError);
+
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_EQ(s, Status::OK());
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, CorruptionError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.level0_file_num_compaction_trigger = 2;
+  options.env = fault_env.get();
+  Status s;
+  DestroyAndReopen(options);
+
+  Put(Key(0), "va;");
+  Put(Key(2), "va;");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BackgroundCallCompaction:0", [&](void *) {
+      fault_env->SetFilesystemActive(false, Status::Corruption("Corruption"));
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put(Key(1), "val");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  s = dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kUnrecoverableError);
+
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_NE(s, Status::OK());
+  Destroy(options);
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}

db/event_helpers.cc

@@ -8,7 +8,7 @@
 namespace rocksdb {
 
 namespace {
-template<class T>
+template <class T>
 inline T SafeDivide(T a, T b) {
   return b == 0 ? 0 : a / b;
 }
@@ -17,7 +17,8 @@ inline T SafeDivide(T a, T b) {
 void EventHelpers::AppendCurrentTime(JSONWriter* jwriter) {
   *jwriter << "time_micros"
            << std::chrono::duration_cast<std::chrono::microseconds>(
-                  std::chrono::system_clock::now().time_since_epoch()).count();
+                  std::chrono::system_clock::now().time_since_epoch())
+                  .count();
 }
 
 #ifndef ROCKSDB_LITE
@@ -52,6 +53,11 @@ void EventHelpers::NotifyOnBackgroundError(
     listener->OnBackgroundError(reason, bg_error);
   }
   db_mutex->Lock();
+#else
+  (void)listeners;
+  (void)reason;
+  (void)bg_error;
+  (void)db_mutex;
 #endif  // ROCKSDB_LITE
 }
 
@@ -117,20 +123,25 @@ void EventHelpers::LogAndNotifyTableFileCreationFinished(
   for (auto& listener : listeners) {
     listener->OnTableFileCreated(info);
   }
+#else
+  (void)listeners;
+  (void)db_name;
+  (void)cf_name;
+  (void)file_path;
+  (void)reason;
 #endif  // !ROCKSDB_LITE
 }
 
 void EventHelpers::LogAndNotifyTableFileDeletion(
-    EventLogger* event_logger, int job_id,
-    uint64_t file_number, const std::string& file_path,
-    const Status& status, const std::string& dbname,
+    EventLogger* event_logger, int job_id, uint64_t file_number,
+    const std::string& file_path, const Status& status,
+    const std::string& dbname,
     const std::vector<std::shared_ptr<EventListener>>& listeners) {
-
   JSONWriter jwriter;
   AppendCurrentTime(&jwriter);
 
-  jwriter << "job" << job_id
-          << "event" << "table_file_deletion"
+  jwriter << "job" << job_id << "event"
+          << "table_file_deletion"
           << "file_number" << file_number;
   if (!status.ok()) {
     jwriter << "status" << status.ToString();
@@ -149,6 +160,10 @@ void EventHelpers::LogAndNotifyTableFileDeletion(
   for (auto& listener : listeners) {
     listener->OnTableFileDeleted(info);
   }
+#else
+  (void)file_path;
+  (void)dbname;
+  (void)listeners;
 #endif  // !ROCKSDB_LITE
 }
 

db/experimental.cc

@@ -30,12 +30,13 @@ Status PromoteL0(DB* db, ColumnFamilyHandle* column_family, int target_level) {
 
 #else  // ROCKSDB_LITE
 
-Status SuggestCompactRange(DB* db, ColumnFamilyHandle* column_family,
-                           const Slice* begin, const Slice* end) {
+Status SuggestCompactRange(DB* /*db*/, ColumnFamilyHandle* /*column_family*/,
+                           const Slice* /*begin*/, const Slice* /*end*/) {
   return Status::NotSupported("Not supported in RocksDB LITE");
 }
 
-Status PromoteL0(DB* db, ColumnFamilyHandle* column_family, int target_level) {
+Status PromoteL0(DB* /*db*/, ColumnFamilyHandle* /*column_family*/,
+                 int /*target_level*/) {
   return Status::NotSupported("Not supported in RocksDB LITE");
 }
 

db/external_sst_file_basic_test.cc

@@ -39,7 +39,8 @@ class ExternalSSTFileBasicTest : public DBTestBase {
 
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<int> keys,
-      const std::vector<ValueType>& value_types, int file_id,
+      const std::vector<ValueType>& value_types,
+      std::vector<std::pair<int, int>> range_deletions, int file_id,
       std::map<std::string, std::string>* true_data) {
     assert(value_types.size() == 1 || keys.size() == value_types.size());
     std::string file_path = sst_files_dir_ + ToString(file_id);
@@ -49,6 +50,29 @@ class ExternalSSTFileBasicTest : public DBTestBase {
     if (!s.ok()) {
       return s;
     }
+    for (size_t i = 0; i < range_deletions.size(); i++) {
+      // Account for the effect of range deletions on true_data before
+      // all point operators, even though sst_file_writer.DeleteRange
+      // must be called before other sst_file_writer methods. This is
+      // because point writes take precedence over range deletions
+      // in the same ingested sst.
+      std::string start_key = Key(range_deletions[i].first);
+      std::string end_key = Key(range_deletions[i].second);
+      s = sst_file_writer.DeleteRange(start_key, end_key);
+      if (!s.ok()) {
+        sst_file_writer.Finish();
+        return s;
+      }
+      auto start_key_it = true_data->find(start_key);
+      if (start_key_it == true_data->end()) {
+        start_key_it = true_data->upper_bound(start_key);
+      }
+      auto end_key_it = true_data->find(end_key);
+      if (end_key_it == true_data->end()) {
+        end_key_it = true_data->upper_bound(end_key);
+      }
+      true_data->erase(start_key_it, end_key_it);
+    }
     for (size_t i = 0; i < keys.size(); i++) {
       std::string key = Key(keys[i]);
       std::string value = Key(keys[i]) + ToString(file_id);
@@ -86,6 +110,14 @@ class ExternalSSTFileBasicTest : public DBTestBase {
     return s;
   }
 
+  Status GenerateAndAddExternalFile(
+      const Options options, std::vector<int> keys,
+      const std::vector<ValueType>& value_types, int file_id,
+      std::map<std::string, std::string>* true_data) {
+    return GenerateAndAddExternalFile(options, keys, value_types, {}, file_id,
+                                      true_data);
+  }
+
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<int> keys, const ValueType value_type,
       int file_id, std::map<std::string, std::string>* true_data) {
@@ -126,9 +158,14 @@ TEST_F(ExternalSSTFileBasicTest, Basic) {
   ASSERT_EQ(file1_info.num_entries, 100);
   ASSERT_EQ(file1_info.smallest_key, Key(0));
   ASSERT_EQ(file1_info.largest_key, Key(99));
+  ASSERT_EQ(file1_info.num_range_del_entries, 0);
+  ASSERT_EQ(file1_info.smallest_range_del_key, "");
+  ASSERT_EQ(file1_info.largest_range_del_key, "");
   // sst_file_writer already finished, cannot add this value
   s = sst_file_writer.Put(Key(100), "bad_val");
   ASSERT_FALSE(s.ok()) << s.ToString();
+  s = sst_file_writer.DeleteRange(Key(100), Key(200));
+  ASSERT_FALSE(s.ok()) << s.ToString();
 
   DestroyAndReopen(options);
   // Add file using file path
@@ -189,6 +226,7 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
   ASSERT_EQ(file3_info.num_entries, 15);
   ASSERT_EQ(file3_info.smallest_key, Key(110));
   ASSERT_EQ(file3_info.largest_key, Key(124));
+
   s = DeprecatedAddFile({file1}, true /* move file */);
   ASSERT_TRUE(s.ok()) << s.ToString();
   ASSERT_EQ(Status::NotFound(), env_->FileExists(file1));
@@ -197,8 +235,8 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
   ASSERT_TRUE(s.ok()) << s.ToString();
   ASSERT_OK(env_->FileExists(file2));
 
-  // This file have overlapping values with the existing data
-  s = DeprecatedAddFile({file2}, true /* move file */);
+  // This file has overlapping values with the existing data
+  s = DeprecatedAddFile({file3}, true /* move file */);
   ASSERT_FALSE(s.ok()) << s.ToString();
   ASSERT_OK(env_->FileExists(file3));
 
@@ -218,34 +256,33 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
     ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
                                          ValueType::kTypeValue, file_id++,
                                          &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
                                          ValueType::kTypeValue, file_id++,
-
                                          &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 4, 6}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {11, 15, 19}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {120, 130}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 130}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
     // Write some keys through normal write path
@@ -257,17 +294,17 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {40, 41, 42}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {20, 30, 40}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
     const Snapshot* snapshot = db_->GetSnapshot();
@@ -314,37 +351,54 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
     ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
                                          ValueType::kTypeValue, file_id++,
                                          &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
                                          ValueType::kTypeValue, file_id++,
-
                                          &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 4, 6}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(options, {11, 15, 19},
                                          ValueType::kTypeDeletion, file_id++,
                                          &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {120, 130}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 130}, ValueType::kTypeDeletion, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
+    ASSERT_OK(GenerateAndAddExternalFile(options, {120},
+                                         {ValueType::kTypeValue}, {{120, 135}},
+                                         file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{110, 120}},
+                                         file_id++, &true_data));
+    // The range deletion ends on a key, but it doesn't actually delete
+    // this key because the largest key in the range is exclusive. Still,
+    // it counts as an overlap so a new seqno will be assigned.
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{100, 109}},
+                                         file_id++, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
     // Write some keys through normal write path
     for (int i = 0; i < 50; i++) {
       ASSERT_OK(Put(Key(i), "memtable"));
@@ -354,18 +408,18 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {40, 41, 42}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(options, {20, 30, 40},
                                          ValueType::kTypeDeletion, file_id++,
                                          &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
     const Snapshot* snapshot = db_->GetSnapshot();
@@ -414,7 +468,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
          ValueType::kTypeMerge, ValueType::kTypeValue, ValueType::kTypeMerge},
         file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
@@ -422,35 +476,58 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
          ValueType::kTypeMerge},
         file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 4, 6}, {ValueType::kTypeDeletion, ValueType::kTypeValue,
                              ValueType::kTypeMerge},
         file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {11, 15, 19}, {ValueType::kTypeDeletion, ValueType::kTypeMerge,
                                 ValueType::kTypeValue},
         file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {120, 130}, {ValueType::kTypeValue, ValueType::kTypeMerge},
         file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 130}, {ValueType::kTypeMerge, ValueType::kTypeDeletion},
         file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {150, 151, 152},
+        {ValueType::kTypeValue, ValueType::kTypeMerge,
+         ValueType::kTypeDeletion},
+        {{150, 160}, {180, 190}}, file_id++, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {150, 151, 152},
+        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
+        {{200, 250}}, file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {300, 301, 302},
+        {ValueType::kTypeValue, ValueType::kTypeMerge,
+         ValueType::kTypeDeletion},
+        {{1, 2}, {152, 154}}, file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
     // Write some keys through normal write path
     for (int i = 0; i < 50; i++) {
       ASSERT_OK(Put(Key(i), "memtable"));
@@ -462,14 +539,15 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {60, 61, 62},
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
         file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {40, 41, 42}, {ValueType::kTypeValue, ValueType::kTypeDeletion,
-                                ValueType::kTypeDeletion},
+        options, {40, 41, 42},
+        {ValueType::kTypeValue, ValueType::kTypeDeletion,
+         ValueType::kTypeDeletion},
         file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
@@ -477,7 +555,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
          ValueType::kTypeDeletion},
         file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
     const Snapshot* snapshot = db_->GetSnapshot();
@@ -589,7 +667,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {60, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      {{65, 70}, {70, 85}}, file_id++, &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
   ASSERT_EQ(2, NumTableFilesAtLevel(0));
   ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
@@ -605,6 +683,16 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
   ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
 
+  // overlaps with L5 file but not memtable or L0 file, so flush is skipped and
+  // file is ingested into L4
+  ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{5, 15}}, file_id++,
+                                       &true_data));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
+
   // ingested file overlaps with memtable, so flush is triggered before the file
   // is ingested such that the ingested data is considered newest. So L0 file
   // count increases by two.
@@ -623,7 +711,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   // seqnum.
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {151, 175}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      {{160, 200}}, file_id++, &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
   ASSERT_EQ(4, NumTableFilesAtLevel(0));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));

db/external_sst_file_ingestion_job.cc

@@ -29,13 +29,13 @@
 namespace rocksdb {
 
 Status ExternalSstFileIngestionJob::Prepare(
-    const std::vector<std::string>& external_files_paths) {
+    const std::vector<std::string>& external_files_paths, SuperVersion* sv) {
   Status status;
 
   // Read the information of files we are ingesting
   for (const std::string& file_path : external_files_paths) {
     IngestedFileInfo file_to_ingest;
-    status = GetIngestedFileInfo(file_path, &file_to_ingest);
+    status = GetIngestedFileInfo(file_path, &file_to_ingest, sv);
     if (!status.ok()) {
       return status;
     }
@@ -78,7 +78,7 @@ Status ExternalSstFileIngestionJob::Prepare(
   }
 
   for (IngestedFileInfo& f : files_to_ingest_) {
-    if (f.num_entries == 0) {
+    if (f.num_entries == 0 && f.num_range_deletions == 0) {
       return Status::InvalidArgument("File contain no entries");
     }
 
@@ -94,7 +94,8 @@ Status ExternalSstFileIngestionJob::Prepare(
 
     const std::string path_outside_db = f.external_file_path;
     const std::string path_inside_db =
-        TableFileName(db_options_.db_paths, f.fd.GetNumber(), f.fd.GetPathId());
+        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);
@@ -102,12 +103,16 @@ Status ExternalSstFileIngestionJob::Prepare(
         // Original file is on a different FS, use copy instead of hard linking
         status = CopyFile(env_, path_outside_db, path_inside_db, 0,
                           db_options_.use_fsync);
+        f.copy_file = true;
+      } else {
+        f.copy_file = false;
       }
     } else {
       status = CopyFile(env_, path_outside_db, path_inside_db, 0,
                         db_options_.use_fsync);
+      f.copy_file = true;
     }
-    TEST_SYNC_POINT("DBImpl::AddFile:FileCopied");
+    TEST_SYNC_POINT("ExternalSstFileIngestionJob::Prepare:FileAdded");
     if (!status.ok()) {
       break;
     }
@@ -117,7 +122,7 @@ Status ExternalSstFileIngestionJob::Prepare(
   if (!status.ok()) {
     // We failed, remove all files that we copied into the db
     for (IngestedFileInfo& f : files_to_ingest_) {
-      if (f.internal_file_path == "") {
+      if (f.internal_file_path.empty()) {
         break;
       }
       Status s = env_->DeleteFile(f.internal_file_path);
@@ -217,9 +222,17 @@ void ExternalSstFileIngestionJob::UpdateStats() {
   uint64_t total_l0_files = 0;
   uint64_t total_time = env_->NowMicros() - job_start_time_;
   for (IngestedFileInfo& f : files_to_ingest_) {
-    InternalStats::CompactionStats stats(1);
+    InternalStats::CompactionStats stats(CompactionReason::kExternalSstIngestion, 1);
     stats.micros = total_time;
-    stats.bytes_written = f.fd.GetFileSize();
+    // If actual copy occured for this file, then we need to count the file
+    // size as the actual bytes written. If the file was linked, then we ignore
+    // the bytes written for file metadata.
+    // TODO (yanqin) maybe account for file metadata bytes for exact accuracy?
+    if (f.copy_file) {
+      stats.bytes_written = f.fd.GetFileSize();
+    } else {
+      stats.bytes_moved = f.fd.GetFileSize();
+    }
     stats.num_output_files = 1;
     cfd_->internal_stats()->AddCompactionStats(f.picked_level, stats);
     cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_INGESTED_ADD_FILE,
@@ -271,7 +284,8 @@ void ExternalSstFileIngestionJob::Cleanup(const Status& status) {
 }
 
 Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
-    const std::string& external_file, IngestedFileInfo* file_to_ingest) {
+    const std::string& external_file, IngestedFileInfo* file_to_ingest,
+    SuperVersion* sv) {
   file_to_ingest->external_file_path = external_file;
 
   // Get external file size
@@ -293,8 +307,9 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
                                                    external_file));
 
   status = cfd_->ioptions()->table_factory->NewTableReader(
-      TableReaderOptions(*cfd_->ioptions(), env_options_,
-                         cfd_->internal_comparator()),
+      TableReaderOptions(*cfd_->ioptions(),
+                         sv->mutable_cf_options.prefix_extractor.get(),
+                         env_options_, cfd_->internal_comparator()),
       std::move(sst_file_reader), file_to_ingest->file_size, &table_reader);
   if (!status.ok()) {
     return status;
@@ -321,12 +336,14 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
 
     // Set the global sequence number
     file_to_ingest->original_seqno = DecodeFixed64(seqno_iter->second.c_str());
-    file_to_ingest->global_seqno_offset = props->properties_offsets.at(
+    auto offsets_iter = props->properties_offsets.find(
         ExternalSstFilePropertyNames::kGlobalSeqno);
-
-    if (file_to_ingest->global_seqno_offset == 0) {
+    if (offsets_iter == props->properties_offsets.end() ||
+        offsets_iter->second == 0) {
+      file_to_ingest->global_seqno_offset = 0;
       return Status::Corruption("Was not able to find file global seqno field");
     }
+    file_to_ingest->global_seqno_offset = offsets_iter->second;
   } else if (file_to_ingest->version == 1) {
     // SST file V1 should not have global seqno field
     assert(seqno_iter == uprops.end());
@@ -341,6 +358,7 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
   }
   // Get number of entries in table
   file_to_ingest->num_entries = props->num_entries;
+  file_to_ingest->num_range_deletions = props->num_range_deletions;
 
   ParsedInternalKey key;
   ReadOptions ro;
@@ -350,27 +368,57 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
   // We need to disable fill_cache so that we read from the file without
   // updating the block cache.
   ro.fill_cache = false;
-  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(ro));
+  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(
+      ro, sv->mutable_cf_options.prefix_extractor.get()));
+  std::unique_ptr<InternalIterator> range_del_iter(
+      table_reader->NewRangeTombstoneIterator(ro));
 
-  // Get first (smallest) key from file
+  // Get first (smallest) and last (largest) key from file.
+  bool bounds_set = false;
   iter->SeekToFirst();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("external file have corrupted keys");
-  }
-  if (key.sequence != 0) {
-    return Status::Corruption("external file have non zero sequence number");
-  }
-  file_to_ingest->smallest_user_key = key.user_key.ToString();
+  if (iter->Valid()) {
+    if (!ParseInternalKey(iter->key(), &key)) {
+      return Status::Corruption("external file have corrupted keys");
+    }
+    if (key.sequence != 0) {
+      return Status::Corruption("external file have non zero sequence number");
+    }
+    file_to_ingest->smallest_user_key = key.user_key.ToString();
 
-  // Get last (largest) key from file
-  iter->SeekToLast();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("external file have corrupted keys");
+    iter->SeekToLast();
+    if (!ParseInternalKey(iter->key(), &key)) {
+      return Status::Corruption("external file have corrupted keys");
+    }
+    if (key.sequence != 0) {
+      return Status::Corruption("external file have non zero sequence number");
+    }
+    file_to_ingest->largest_user_key = key.user_key.ToString();
+
+    bounds_set = true;
   }
-  if (key.sequence != 0) {
-    return Status::Corruption("external file have non zero sequence number");
+
+  // We may need to adjust these key bounds, depending on whether any range
+  // deletion tombstones extend past them.
+  const Comparator* ucmp = cfd_->internal_comparator().user_comparator();
+  if (range_del_iter != nullptr) {
+    for (range_del_iter->SeekToFirst(); range_del_iter->Valid();
+         range_del_iter->Next()) {
+      if (!ParseInternalKey(range_del_iter->key(), &key)) {
+        return Status::Corruption("external file have corrupted keys");
+      }
+      RangeTombstone tombstone(key, range_del_iter->value());
+
+      if (!bounds_set || ucmp->Compare(tombstone.start_key_,
+                                       file_to_ingest->smallest_user_key) < 0) {
+        file_to_ingest->smallest_user_key = tombstone.start_key_.ToString();
+      }
+      if (!bounds_set || ucmp->Compare(tombstone.end_key_,
+                                       file_to_ingest->largest_user_key) > 0) {
+        file_to_ingest->largest_user_key = tombstone.end_key_.ToString();
+      }
+      bounds_set = true;
+    }
   }
-  file_to_ingest->largest_user_key = key.user_key.ToString();
 
   file_to_ingest->cf_id = static_cast<uint32_t>(props->column_family_id);
 

db/external_sst_file_ingestion_job.h

@@ -36,6 +36,8 @@ struct IngestedFileInfo {
   uint64_t file_size;
   // total number of keys in external file
   uint64_t num_entries;
+  // total number of range deletions in external file
+  uint64_t num_range_deletions;
   // Id of column family this file shoule be ingested into
   uint32_t cf_id;
   // TableProperties read from external file
@@ -46,11 +48,18 @@ struct IngestedFileInfo {
   // FileDescriptor for the file inside the DB
   FileDescriptor fd;
   // file path that we picked for file inside the DB
-  std::string internal_file_path = "";
+  std::string internal_file_path;
   // Global sequence number that we picked for the file inside the DB
   SequenceNumber assigned_seqno = 0;
   // Level inside the DB we picked for the external file.
   int picked_level = 0;
+  // Whether to copy or link the external sst file. copy_file will be set to
+  // false if ingestion_options.move_files is true and underlying FS
+  // supports link operation. Need to provide a default value to make the
+  // undefined-behavior sanity check of llvm happy. Since
+  // ingestion_options.move_files is false by default, thus copy_file is true
+  // by default.
+  bool copy_file = true;
 
   InternalKey smallest_internal_key() const {
     return InternalKey(smallest_user_key, assigned_seqno,
@@ -79,7 +88,8 @@ class ExternalSstFileIngestionJob {
         job_start_time_(env_->NowMicros()) {}
 
   // Prepare the job by copying external files into the DB.
-  Status Prepare(const std::vector<std::string>& external_files_paths);
+  Status Prepare(const std::vector<std::string>& external_files_paths,
+                 SuperVersion* sv);
 
   // Check if we need to flush the memtable before running the ingestion job
   // This will be true if the files we are ingesting are overlapping with any
@@ -112,7 +122,8 @@ class ExternalSstFileIngestionJob {
   // Open the external file and populate `file_to_ingest` with all the
   // external information we need to ingest this file.
   Status GetIngestedFileInfo(const std::string& external_file,
-                             IngestedFileInfo* file_to_ingest);
+                             IngestedFileInfo* file_to_ingest,
+                             SuperVersion* sv);
 
   // Assign `file_to_ingest` the appropriate sequence number and  the lowest
   // possible level that it can be ingested to according to compaction_style.

db/external_sst_file_test.cc

@@ -222,6 +222,9 @@ TEST_F(ExternalSSTFileTest, Basic) {
     ASSERT_EQ(file1_info.num_entries, 100);
     ASSERT_EQ(file1_info.smallest_key, Key(0));
     ASSERT_EQ(file1_info.largest_key, Key(99));
+    ASSERT_EQ(file1_info.num_range_del_entries, 0);
+    ASSERT_EQ(file1_info.smallest_range_del_key, "");
+    ASSERT_EQ(file1_info.largest_range_del_key, "");
     // sst_file_writer already finished, cannot add this value
     s = sst_file_writer.Put(Key(100), "bad_val");
     ASSERT_FALSE(s.ok()) << s.ToString();
@@ -290,6 +293,58 @@ TEST_F(ExternalSSTFileTest, Basic) {
     ASSERT_EQ(file5_info.smallest_key, Key(400));
     ASSERT_EQ(file5_info.largest_key, Key(499));
 
+    // file6.sst (delete 400 => 500)
+    std::string file6 = sst_files_dir_ + "file6.sst";
+    ASSERT_OK(sst_file_writer.Open(file6));
+    sst_file_writer.DeleteRange(Key(400), Key(500));
+    ExternalSstFileInfo file6_info;
+    s = sst_file_writer.Finish(&file6_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file6_info.file_path, file6);
+    ASSERT_EQ(file6_info.num_entries, 0);
+    ASSERT_EQ(file6_info.smallest_key, "");
+    ASSERT_EQ(file6_info.largest_key, "");
+    ASSERT_EQ(file6_info.num_range_del_entries, 1);
+    ASSERT_EQ(file6_info.smallest_range_del_key, Key(400));
+    ASSERT_EQ(file6_info.largest_range_del_key, Key(500));
+
+    // file7.sst (delete 500 => 570, put 520 => 599 divisible by 2)
+    std::string file7 = sst_files_dir_ + "file7.sst";
+    ASSERT_OK(sst_file_writer.Open(file7));
+    sst_file_writer.DeleteRange(Key(500), Key(550));
+    for (int k = 520; k < 560; k += 2) {
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
+    }
+    sst_file_writer.DeleteRange(Key(525), Key(575));
+    for (int k = 560; k < 600; k += 2) {
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
+    }
+    ExternalSstFileInfo file7_info;
+    s = sst_file_writer.Finish(&file7_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file7_info.file_path, file7);
+    ASSERT_EQ(file7_info.num_entries, 40);
+    ASSERT_EQ(file7_info.smallest_key, Key(520));
+    ASSERT_EQ(file7_info.largest_key, Key(598));
+    ASSERT_EQ(file7_info.num_range_del_entries, 2);
+    ASSERT_EQ(file7_info.smallest_range_del_key, Key(500));
+    ASSERT_EQ(file7_info.largest_range_del_key, Key(575));
+
+    // file8.sst (delete 600 => 700)
+    std::string file8 = sst_files_dir_ + "file8.sst";
+    ASSERT_OK(sst_file_writer.Open(file8));
+    sst_file_writer.DeleteRange(Key(600), Key(700));
+    ExternalSstFileInfo file8_info;
+    s = sst_file_writer.Finish(&file8_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file8_info.file_path, file8);
+    ASSERT_EQ(file8_info.num_entries, 0);
+    ASSERT_EQ(file8_info.smallest_key, "");
+    ASSERT_EQ(file8_info.largest_key, "");
+    ASSERT_EQ(file8_info.num_range_del_entries, 1);
+    ASSERT_EQ(file8_info.smallest_range_del_key, Key(600));
+    ASSERT_EQ(file8_info.largest_range_del_key, Key(700));
+
     // Cannot create an empty sst file
     std::string file_empty = sst_files_dir_ + "file_empty.sst";
     ExternalSstFileInfo file_empty_info;
@@ -336,6 +391,16 @@ TEST_F(ExternalSSTFileTest, Basic) {
     // Key range of file5 (400 => 499) dont overlap with any keys in DB
     ASSERT_OK(DeprecatedAddFile({file5}));
 
+    // This file has overlapping values with the existing data
+    s = DeprecatedAddFile({file6});
+    ASSERT_FALSE(s.ok()) << s.ToString();
+
+    // Key range of file7 (500 => 598) dont overlap with any keys in DB
+    ASSERT_OK(DeprecatedAddFile({file7}));
+
+    // Key range of file7 (600 => 700) dont overlap with any keys in DB
+    ASSERT_OK(DeprecatedAddFile({file8}));
+
     // Make sure values are correct before and after flush/compaction
     for (int i = 0; i < 2; i++) {
       for (int k = 0; k < 200; k++) {
@@ -349,6 +414,13 @@ TEST_F(ExternalSSTFileTest, Basic) {
         std::string value = Key(k) + "_val";
         ASSERT_EQ(Get(Key(k)), value);
       }
+      for (int k = 500; k < 600; k++) {
+        std::string value = Key(k) + "_val";
+        if (k < 520 || k % 2 == 1) {
+          value = "NOT_FOUND";
+        }
+        ASSERT_EQ(Get(Key(k)), value);
+      }
       ASSERT_OK(Flush());
       ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
     }
@@ -377,7 +449,8 @@ TEST_F(ExternalSSTFileTest, Basic) {
       ASSERT_EQ(Get(Key(k)), value);
     }
     DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction |
+                         kRangeDelSkipConfigs));
 }
 class SstFileWriterCollector : public TablePropertiesCollector {
  public:
@@ -518,17 +591,57 @@ TEST_F(ExternalSSTFileTest, AddList) {
     ASSERT_EQ(file5_info.smallest_key, Key(200));
     ASSERT_EQ(file5_info.largest_key, Key(299));
 
+    // file6.sst (delete 0 => 100)
+    std::string file6 = sst_files_dir_ + "file6.sst";
+    ASSERT_OK(sst_file_writer.Open(file6));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(0), Key(75)));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(25), Key(100)));
+    ExternalSstFileInfo file6_info;
+    s = sst_file_writer.Finish(&file6_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file6_info.file_path, file6);
+    ASSERT_EQ(file6_info.num_entries, 0);
+    ASSERT_EQ(file6_info.smallest_key, "");
+    ASSERT_EQ(file6_info.largest_key, "");
+    ASSERT_EQ(file6_info.num_range_del_entries, 2);
+    ASSERT_EQ(file6_info.smallest_range_del_key, Key(0));
+    ASSERT_EQ(file6_info.largest_range_del_key, Key(100));
+
+    // file7.sst (delete 100 => 200)
+    std::string file7 = sst_files_dir_ + "file7.sst";
+    ASSERT_OK(sst_file_writer.Open(file7));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(100), Key(200)));
+    ExternalSstFileInfo file7_info;
+    s = sst_file_writer.Finish(&file7_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file7_info.file_path, file7);
+    ASSERT_EQ(file7_info.num_entries, 0);
+    ASSERT_EQ(file7_info.smallest_key, "");
+    ASSERT_EQ(file7_info.largest_key, "");
+    ASSERT_EQ(file7_info.num_range_del_entries, 1);
+    ASSERT_EQ(file7_info.smallest_range_del_key, Key(100));
+    ASSERT_EQ(file7_info.largest_range_del_key, Key(200));
+
     // list 1 has internal key range conflict
     std::vector<std::string> file_list0({file1, file2});
     std::vector<std::string> file_list1({file3, file2, file1});
     std::vector<std::string> file_list2({file5});
     std::vector<std::string> file_list3({file3, file4});
+    std::vector<std::string> file_list4({file5, file7});
+    std::vector<std::string> file_list5({file6, file7});
 
     DestroyAndReopen(options);
 
-    // This list of files have key ranges are overlapping with each other
+    // These lists of files have key ranges that overlap with each other
     s = DeprecatedAddFile(file_list1);
     ASSERT_FALSE(s.ok()) << s.ToString();
+    // Both of the following overlap on the end key of a range deletion
+    // tombstone. This is a limitation because these tombstones have exclusive
+    // end keys that should not count as overlapping with other keys.
+    s = DeprecatedAddFile(file_list4);
+    ASSERT_FALSE(s.ok()) << s.ToString();
+    s = DeprecatedAddFile(file_list5);
+    ASSERT_FALSE(s.ok()) << s.ToString();
 
     // Add files using file path list
     s = DeprecatedAddFile(file_list0);
@@ -619,7 +732,8 @@ TEST_F(ExternalSSTFileTest, AddList) {
       ASSERT_EQ(Get(Key(k)), value);
     }
     DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction |
+                         kRangeDelSkipConfigs));
 }
 
 TEST_F(ExternalSSTFileTest, AddListAtomicity) {
@@ -689,7 +803,7 @@ TEST_F(ExternalSSTFileTest, PurgeObsoleteFilesBug) {
   DestroyAndReopen(options);
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::AddFile:FileCopied", [&](void* arg) {
+      "ExternalSstFileIngestionJob::Prepare:FileAdded", [&](void* /* arg */) {
         ASSERT_OK(Put("aaa", "bbb"));
         ASSERT_OK(Flush());
         ASSERT_OK(Put("aaa", "xxx"));
@@ -897,11 +1011,11 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {
 
 TEST_F(ExternalSSTFileTest, OverlappingRanges) {
   Random rnd(301);
-  int picked_level = 0;
+  SequenceNumber assigned_seqno = 0;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-    "ExternalSstFileIngestionJob::Run", [&picked_level](void* arg) {
+    "ExternalSstFileIngestionJob::Run", [&assigned_seqno](void* arg) {
       ASSERT_TRUE(arg != nullptr);
-      picked_level = *(static_cast<int*>(arg));
+      assigned_seqno = *(static_cast<SequenceNumber*>(arg));
     });
   bool need_flush = false;
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -925,7 +1039,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
 
     printf("Option config = %d\n", option_config_);
     std::vector<std::pair<int, int>> key_ranges;
-    for (int i = 0; i < 500; i++) {
+    for (int i = 0; i < 100; i++) {
       int range_start = rnd.Uniform(20000);
       int keys_per_range = 10 + rnd.Uniform(41);
 
@@ -971,7 +1085,8 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
         s = DeprecatedAddFile({file_name});
         auto it = true_data.lower_bound(Key(range_start));
         if (option_config_ != kUniversalCompaction &&
-            option_config_ != kUniversalCompactionMultiLevel) {
+            option_config_ != kUniversalCompactionMultiLevel &&
+            option_config_ != kUniversalSubcompactions) {
           if (it != true_data.end() && it->first <= Key(range_end)) {
             // This range overlap with data already exist in DB
             ASSERT_NOK(s);
@@ -982,7 +1097,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
           }
         } else {
           if ((it != true_data.end() && it->first <= Key(range_end)) ||
-              need_flush || picked_level > 0 || overlap_with_db) {
+              need_flush || assigned_seqno > 0 || overlap_with_db) {
             // This range overlap with data already exist in DB
             ASSERT_NOK(s);
             failed_add_file++;
@@ -1128,7 +1243,7 @@ TEST_F(ExternalSSTFileTest, PickedLevelBug) {
   std::atomic<bool> bg_compact_started(false);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:Start",
-      [&](void* arg) { bg_compact_started.store(true); });
+      [&](void* /*arg*/) { bg_compact_started.store(true); });
 
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
@@ -1194,8 +1309,9 @@ TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
     ASSERT_OK(GenerateAndAddExternalFile(options, file_keys, range_id));
   };
 
+  const int num_of_ranges = 1000;
   std::vector<port::Thread> threads;
-  while (range_id < 5000) {
+  while (range_id < num_of_ranges) {
     int range_start = range_id * 10;
     int range_end = range_start + 10;
 
@@ -1220,7 +1336,7 @@ TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
     range_id++;
   }
 
-  for (int rid = 0; rid < 5000; rid++) {
+  for (int rid = 0; rid < num_of_ranges; rid++) {
     int range_start = rid * 10;
     int range_end = range_start + 10;
 
@@ -1409,12 +1525,16 @@ TEST_F(ExternalSSTFileTest, AddFileTrivialMoveBug) {
   ASSERT_OK(GenerateAndAddExternalFile(options, {22, 23}, 6));  // L2
 
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "CompactionJob::Run():Start", [&](void* arg) {
+      "CompactionJob::Run():Start", [&](void* /*arg*/) {
         // fit in L3 but will overlap with compaction so will be added
         // to L2 but a compaction will trivially move it to L3
         // and break LSM consistency
-        ASSERT_OK(dbfull()->SetOptions({{"max_bytes_for_level_base", "1"}}));
-        ASSERT_OK(GenerateAndAddExternalFile(options, {15, 16}, 7));
+        static std::atomic<bool> called = {false};
+        if (!called) {
+          called = true;
+          ASSERT_OK(dbfull()->SetOptions({{"max_bytes_for_level_base", "1"}}));
+          ASSERT_OK(GenerateAndAddExternalFile(options, {15, 16}, 7));
+        }
       });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
@@ -1471,7 +1591,7 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
 
     Random rnd(301);
     std::map<std::string, std::string> true_data;
-    for (int i = 0; i < 2000; i++) {
+    for (int i = 0; i < 500; i++) {
       std::vector<std::pair<std::string, std::string>> random_data;
       for (int j = 0; j < 100; j++) {
         std::string k;
@@ -1797,6 +1917,56 @@ TEST_F(ExternalSSTFileTest, FileWithCFInfo) {
   ASSERT_OK(db_->IngestExternalFile(handles_[2], {unknown_sst}, ifo));
 }
 
+/*
+ * Test and verify the functionality of ingestion_options.move_files.
+ */
+TEST_F(ExternalSSTFileTest, LinkExternalSst) {
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  DestroyAndReopen(options);
+  const int kNumKeys = 10000;
+
+  std::string file_path = sst_files_dir_ + "file1.sst";
+  // Create SstFileWriter for default column family
+  SstFileWriter sst_file_writer(EnvOptions(), options);
+  ASSERT_OK(sst_file_writer.Open(file_path));
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(sst_file_writer.Put(Key(i), Key(i) + "_value"));
+  }
+  ASSERT_OK(sst_file_writer.Finish());
+  uint64_t file_size = 0;
+  ASSERT_OK(env_->GetFileSize(file_path, &file_size));
+
+  IngestExternalFileOptions ifo;
+  ifo.move_files = true;
+  ASSERT_OK(db_->IngestExternalFile({file_path}, ifo));
+
+  ColumnFamilyHandleImpl* cfh =
+      static_cast<ColumnFamilyHandleImpl*>(dbfull()->DefaultColumnFamily());
+  ColumnFamilyData* cfd = cfh->cfd();
+  const InternalStats* internal_stats_ptr = cfd->internal_stats();
+  const std::vector<InternalStats::CompactionStats>& comp_stats =
+      internal_stats_ptr->TEST_GetCompactionStats();
+  uint64_t bytes_copied = 0;
+  uint64_t bytes_moved = 0;
+  for (const auto& stats : comp_stats) {
+    bytes_copied += stats.bytes_written;
+    bytes_moved += stats.bytes_moved;
+  }
+  // If bytes_moved > 0, it means external sst resides on the same FS
+  // supporting hard link operation. Therefore,
+  // 0 bytes should be copied, and the bytes_moved == file_size.
+  // Otherwise, FS does not support hard link, or external sst file resides on
+  // a different file system, then the bytes_copied should be equal to
+  // file_size.
+  if (bytes_moved > 0) {
+    ASSERT_EQ(0, bytes_copied);
+    ASSERT_EQ(file_size, bytes_moved);
+  } else {
+    ASSERT_EQ(file_size, bytes_copied);
+  }
+}
+
 class TestIngestExternalFileListener : public EventListener {
  public:
   void OnExternalFileIngested(DB* /*db*/,

db/fault_injection_test.cc

@@ -34,19 +34,22 @@ static const int kValueSize = 1000;
 static const int kMaxNumValues = 2000;
 static const size_t kNumIterations = 3;
 
-class FaultInjectionTest : public testing::Test,
-                           public testing::WithParamInterface<bool> {
+enum FaultInjectionOptionConfig {
+  kDefault,
+  kDifferentDataDir,
+  kWalDir,
+  kSyncWal,
+  kWalDirSyncWal,
+  kMultiLevels,
+  kEnd,
+};
+class FaultInjectionTest
+    : public testing::Test,
+      public testing::WithParamInterface<std::tuple<
+          bool, FaultInjectionOptionConfig, FaultInjectionOptionConfig>> {
  protected:
-  enum OptionConfig {
-    kDefault,
-    kDifferentDataDir,
-    kWalDir,
-    kSyncWal,
-    kWalDirSyncWal,
-    kMultiLevels,
-    kEnd,
-  };
   int option_config_;
+  int non_inclusive_end_range_;  // kEnd or equivalent to that
   // When need to make sure data is persistent, sync WAL
   bool sync_use_wal_;
   // When need to make sure data is persistent, call DB::CompactRange()
@@ -72,13 +75,13 @@ class FaultInjectionTest : public testing::Test,
   DB* db_;
 
   FaultInjectionTest()
-      : option_config_(kDefault),
+      : option_config_(std::get<1>(GetParam())),
+        non_inclusive_end_range_(std::get<2>(GetParam())),
         sync_use_wal_(false),
         sync_use_compact_(true),
         base_env_(nullptr),
         env_(nullptr),
-        db_(nullptr) {
-  }
+        db_(nullptr) {}
 
   ~FaultInjectionTest() {
     rocksdb::SyncPoint::GetInstance()->DisableProcessing();
@@ -87,7 +90,7 @@ class FaultInjectionTest : public testing::Test,
 
   bool ChangeOptions() {
     option_config_++;
-    if (option_config_ >= kEnd) {
+    if (option_config_ >= non_inclusive_end_range_) {
       return false;
     } else {
       if (option_config_ == kMultiLevels) {
@@ -104,18 +107,18 @@ class FaultInjectionTest : public testing::Test,
     Options options;
     switch (option_config_) {
       case kWalDir:
-        options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
+        options.wal_dir = test::PerThreadDBPath(env_, "fault_test_wal");
         break;
       case kDifferentDataDir:
-        options.db_paths.emplace_back(test::TmpDir(env_) + "/fault_test_data",
-                                      1000000U);
+        options.db_paths.emplace_back(
+            test::PerThreadDBPath(env_, "fault_test_data"), 1000000U);
         break;
       case kSyncWal:
         sync_use_wal_ = true;
         sync_use_compact_ = false;
         break;
       case kWalDirSyncWal:
-        options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
+        options.wal_dir = test::PerThreadDBPath(env_, "/fault_test_wal");
         sync_use_wal_ = true;
         sync_use_compact_ = false;
         break;
@@ -155,7 +158,7 @@ class FaultInjectionTest : public testing::Test,
     table_options.block_cache = tiny_cache_;
     options_.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
-    dbname_ = test::TmpDir() + "/fault_test";
+    dbname_ = test::PerThreadDBPath("fault_test");
 
     EXPECT_OK(DestroyDB(dbname_, options_));
 
@@ -166,7 +169,7 @@ class FaultInjectionTest : public testing::Test,
   }
 
   void SetUp() override {
-    sequential_order_ = GetParam();
+    sequential_order_ = std::get<0>(GetParam());
     ASSERT_OK(NewDB());
   }
 
@@ -343,7 +346,9 @@ class FaultInjectionTest : public testing::Test,
   }
 };
 
-TEST_P(FaultInjectionTest, FaultTest) {
+class FaultInjectionTestSplitted : public FaultInjectionTest {};
+
+TEST_P(FaultInjectionTestSplitted, FaultTest) {
   do {
     Random rnd(301);
 
@@ -456,10 +461,10 @@ TEST_P(FaultInjectionTest, UninstalledCompaction) {
 
   std::atomic<bool> opened(false);
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::Open:Opened", [&](void* arg) { opened.store(true); });
+      "DBImpl::Open:Opened", [&](void* /*arg*/) { opened.store(true); });
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BGWorkCompaction",
-      [&](void* arg) { ASSERT_TRUE(opened.load()); });
+      [&](void* /*arg*/) { ASSERT_TRUE(opened.load()); });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
   ASSERT_OK(OpenDB());
   ASSERT_OK(Verify(0, kNumKeys, FaultInjectionTest::kValExpectFound));
@@ -530,7 +535,17 @@ TEST_P(FaultInjectionTest, WriteBatchWalTerminationTest) {
   ASSERT_EQ(db_->Get(ro, "boys", &val), Status::NotFound());
 }
 
-INSTANTIATE_TEST_CASE_P(FaultTest, FaultInjectionTest, ::testing::Bool());
+INSTANTIATE_TEST_CASE_P(
+    FaultTest, FaultInjectionTest,
+    ::testing::Values(std::make_tuple(false, kDefault, kEnd),
+                      std::make_tuple(true, kDefault, kEnd)));
+
+INSTANTIATE_TEST_CASE_P(
+    FaultTest, FaultInjectionTestSplitted,
+    ::testing::Values(std::make_tuple(false, kDefault, kSyncWal),
+                      std::make_tuple(true, kDefault, kSyncWal),
+                      std::make_tuple(false, kSyncWal, kEnd),
+                      std::make_tuple(true, kSyncWal, kEnd)));
 
 }  // namespace rocksdb
 

db/flush_job.cc

@@ -56,8 +56,8 @@ namespace rocksdb {
 
 const char* GetFlushReasonString (FlushReason flush_reason) {
   switch (flush_reason) {
-    case FlushReason::kUnknown:
-      return "Unknown";
+    case FlushReason::kOthers:
+      return "Other Reasons";
     case FlushReason::kGetLiveFiles:
       return "Get Live Files";
     case FlushReason::kShutDown:
@@ -72,8 +72,12 @@ const char* GetFlushReasonString (FlushReason flush_reason) {
       return "Write Buffer Full";
     case FlushReason::kTest:
       return "Test";
-    case FlushReason::kSuperVersionChange:
-      return "SuperVersion Change";
+    case FlushReason::kDeleteFiles:
+      return "Delete Files";
+    case FlushReason::kAutoCompaction:
+      return "Auto Compaction";
+    case FlushReason::kManualFlush:
+      return "Manual Flush";
     default:
       return "Invalid";
   }
@@ -181,7 +185,9 @@ void FlushJob::PickMemTable() {
   base_->Ref();  // it is likely that we do not need this reference
 }
 
-Status FlushJob::Run(FileMetaData* file_meta) {
+Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
+                     FileMetaData* file_meta) {
+  TEST_SYNC_POINT("FlushJob::Start");
   db_mutex_->AssertHeld();
   assert(pick_memtable_called);
   AutoThreadOperationStageUpdater stage_run(
@@ -222,7 +228,7 @@ Status FlushJob::Run(FileMetaData* file_meta) {
     TEST_SYNC_POINT("FlushJob::InstallResults");
     // Replace immutable memtable with the generated Table
     s = cfd_->imm()->InstallMemtableFlushResults(
-        cfd_, mutable_cf_options_, mems_, versions_, db_mutex_,
+        cfd_, mutable_cf_options_, mems_, prep_tracker, versions_, db_mutex_,
         meta_.fd.GetNumber(), &job_context_->memtables_to_free, db_directory_,
         log_buffer_);
   }
@@ -388,7 +394,7 @@ Status FlushJob::WriteLevel0Table() {
   }
 
   // Note that here we treat flush as level 0 compaction in internal stats
-  InternalStats::CompactionStats stats(1);
+  InternalStats::CompactionStats stats(CompactionReason::kFlush, 1);
   stats.micros = db_options_.env->NowMicros() - start_micros;
   stats.bytes_written = meta_.fd.GetFileSize();
   MeasureTime(stats_, FLUSH_TIME, stats.micros);

db/flush_job.h

@@ -22,6 +22,7 @@
 #include "db/internal_stats.h"
 #include "db/job_context.h"
 #include "db/log_writer.h"
+#include "db/logs_with_prep_tracker.h"
 #include "db/memtable_list.h"
 #include "db/snapshot_impl.h"
 #include "db/version_edit.h"
@@ -42,6 +43,7 @@
 
 namespace rocksdb {
 
+class DBImpl;
 class MemTable;
 class SnapshotChecker;
 class TableCache;
@@ -71,7 +73,8 @@ class FlushJob {
   // Require db_mutex held.
   // Once PickMemTable() is called, either Run() or Cancel() has to be called.
   void PickMemTable();
-  Status Run(FileMetaData* file_meta = nullptr);
+  Status Run(LogsWithPrepTracker* prep_tracker = nullptr,
+             FileMetaData* file_meta = nullptr);
   void Cancel();
   TableProperties GetTableProperties() const { return table_properties_; }
 

db/flush_job_test.cc

@@ -27,7 +27,7 @@ class FlushJobTest : public testing::Test {
  public:
   FlushJobTest()
       : env_(Env::Default()),
-        dbname_(test::TmpDir() + "/flush_job_test"),
+        dbname_(test::PerThreadDBPath("flush_job_test")),
         options_(),
         db_options_(options_),
         table_cache_(NewLRUCache(50000, 16)),
@@ -150,7 +150,7 @@ TEST_F(FlushJobTest, NonEmpty) {
   FileMetaData fd;
   mutex_.Lock();
   flush_job.PickMemTable();
-  ASSERT_OK(flush_job.Run(&fd));
+  ASSERT_OK(flush_job.Run(nullptr, &fd));
   mutex_.Unlock();
   db_options_.statistics->histogramData(FLUSH_TIME, &hist);
   ASSERT_GT(hist.average, 0.0);

db/forward_iterator.cc

@@ -27,20 +27,22 @@ namespace rocksdb {
 // Usage:
 //     ForwardLevelIterator iter;
 //     iter.SetFileIndex(file_index);
-//     iter.Seek(target);
+//     iter.Seek(target); // or iter.SeekToFirst();
 //     iter.Next()
 class ForwardLevelIterator : public InternalIterator {
  public:
   ForwardLevelIterator(const ColumnFamilyData* const cfd,
                        const ReadOptions& read_options,
-                       const std::vector<FileMetaData*>& files)
+                       const std::vector<FileMetaData*>& files,
+                       const SliceTransform* prefix_extractor)
       : cfd_(cfd),
         read_options_(read_options),
         files_(files),
         valid_(false),
         file_index_(std::numeric_limits<uint32_t>::max()),
         file_iter_(nullptr),
-        pinned_iters_mgr_(nullptr) {}
+        pinned_iters_mgr_(nullptr),
+        prefix_extractor_(prefix_extractor) {}
 
   ~ForwardLevelIterator() {
     // Reset current pointer
@@ -53,11 +55,11 @@ class ForwardLevelIterator : public InternalIterator {
 
   void SetFileIndex(uint32_t file_index) {
     assert(file_index < files_.size());
+    status_ = Status::OK();
     if (file_index != file_index_) {
       file_index_ = file_index;
       Reset();
     }
-    valid_ = false;
   }
   void Reset() {
     assert(file_index_ < files_.size());
@@ -73,14 +75,14 @@ class ForwardLevelIterator : public InternalIterator {
         cfd_->internal_comparator(), {} /* snapshots */);
     file_iter_ = cfd_->table_cache()->NewIterator(
         read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
-        files_[file_index_]->fd,
+        *files_[file_index_],
         read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
-        nullptr /* table_reader_ptr */, nullptr, false);
+        prefix_extractor_, nullptr /* table_reader_ptr */, nullptr, false);
     file_iter_->SetPinnedItersMgr(pinned_iters_mgr_);
+    valid_ = false;
     if (!range_del_agg.IsEmpty()) {
       status_ = Status::NotSupported(
           "Range tombstones unsupported with ForwardIterator");
-      valid_ = false;
     }
   }
   void SeekToLast() override {
@@ -95,12 +97,27 @@ class ForwardLevelIterator : public InternalIterator {
     return valid_;
   }
   void SeekToFirst() override {
-    SetFileIndex(0);
+    assert(file_iter_ != nullptr);
+    if (!status_.ok()) {
+      assert(!valid_);
+      return;
+    }
     file_iter_->SeekToFirst();
     valid_ = file_iter_->Valid();
   }
   void Seek(const Slice& internal_key) override {
     assert(file_iter_ != nullptr);
+
+    // This deviates from the usual convention for InternalIterator::Seek() in
+    // that it doesn't discard pre-existing error status. That's because this
+    // Seek() is only supposed to be called immediately after SetFileIndex()
+    // (which discards pre-existing error status), and SetFileIndex() may set
+    // an error status, which we shouldn't discard.
+    if (!status_.ok()) {
+      assert(!valid_);
+      return;
+    }
+
     file_iter_->Seek(internal_key);
     valid_ = file_iter_->Valid();
   }
@@ -112,8 +129,12 @@ class ForwardLevelIterator : public InternalIterator {
     assert(valid_);
     file_iter_->Next();
     for (;;) {
-      if (file_iter_->status().IsIncomplete() || file_iter_->Valid()) {
-        valid_ = !file_iter_->status().IsIncomplete();
+      valid_ = file_iter_->Valid();
+      if (!file_iter_->status().ok()) {
+        assert(!valid_);
+        return;
+      }
+      if (valid_) {
         return;
       }
       if (file_index_ + 1 >= files_.size()) {
@@ -121,6 +142,10 @@ class ForwardLevelIterator : public InternalIterator {
         return;
       }
       SetFileIndex(file_index_ + 1);
+      if (!status_.ok()) {
+        assert(!valid_);
+        return;
+      }
       file_iter_->SeekToFirst();
     }
   }
@@ -135,7 +160,7 @@ class ForwardLevelIterator : public InternalIterator {
   Status status() const override {
     if (!status_.ok()) {
       return status_;
-    } else if (file_iter_ && !file_iter_->status().ok()) {
+    } else if (file_iter_) {
       return file_iter_->status();
     }
     return Status::OK();
@@ -165,6 +190,7 @@ class ForwardLevelIterator : public InternalIterator {
   Status status_;
   InternalIterator* file_iter_;
   PinnedIteratorsManager* pinned_iters_mgr_;
+  const SliceTransform* prefix_extractor_;
 };
 
 ForwardIterator::ForwardIterator(DBImpl* db, const ReadOptions& read_options,
@@ -173,7 +199,7 @@ ForwardIterator::ForwardIterator(DBImpl* db, const ReadOptions& read_options,
     : db_(db),
       read_options_(read_options),
       cfd_(cfd),
-      prefix_extractor_(cfd->ioptions()->prefix_extractor),
+      prefix_extractor_(current_sv->mutable_cf_options.prefix_extractor.get()),
       user_comparator_(cfd->user_comparator()),
       immutable_min_heap_(MinIterComparator(&cfd_->internal_comparator())),
       sv_(current_sv),
@@ -299,9 +325,6 @@ bool ForwardIterator::IsOverUpperBound(const Slice& internal_key) const {
 }
 
 void ForwardIterator::Seek(const Slice& internal_key) {
-  if (IsOverUpperBound(internal_key)) {
-    valid_ = false;
-  }
   if (sv_ == nullptr) {
     RebuildIterators(true);
   } else if (sv_->version_number != cfd_->GetSuperVersionNumber()) {
@@ -605,13 +628,16 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
     if ((read_options_.iterate_upper_bound != nullptr) &&
         cfd_->internal_comparator().user_comparator()->Compare(
             l0->smallest.user_key(), *read_options_.iterate_upper_bound) > 0) {
-      has_iter_trimmed_for_upper_bound_ = true;
+      // No need to set has_iter_trimmed_for_upper_bound_: this ForwardIterator
+      // will never be interested in files with smallest key above
+      // iterate_upper_bound, since iterate_upper_bound can't be changed.
       l0_iters_.push_back(nullptr);
       continue;
     }
     l0_iters_.push_back(cfd_->table_cache()->NewIterator(
-        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), l0->fd,
-        read_options_.ignore_range_deletions ? nullptr : &range_del_agg));
+        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), *l0,
+        read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
+        sv_->mutable_cf_options.prefix_extractor.get()));
   }
   BuildLevelIterators(vstorage);
   current_ = nullptr;
@@ -680,8 +706,9 @@ void ForwardIterator::RenewIterators() {
     }
     l0_iters_new.push_back(cfd_->table_cache()->NewIterator(
         read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files_new[inew]->fd,
-        read_options_.ignore_range_deletions ? nullptr : &range_del_agg));
+        *l0_files_new[inew],
+        read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
+        svnew->mutable_cf_options.prefix_extractor.get()));
   }
 
   for (auto* f : l0_iters_) {
@@ -722,8 +749,9 @@ void ForwardIterator::BuildLevelIterators(const VersionStorageInfo* vstorage) {
         has_iter_trimmed_for_upper_bound_ = true;
       }
     } else {
-      level_iters_.push_back(
-          new ForwardLevelIterator(cfd_, read_options_, level_files));
+      level_iters_.push_back(new ForwardLevelIterator(
+          cfd_, read_options_, level_files,
+          sv_->mutable_cf_options.prefix_extractor.get()));
     }
   }
 }
@@ -738,7 +766,8 @@ void ForwardIterator::ResetIncompleteIterators() {
     DeleteIterator(l0_iters_[i]);
     l0_iters_[i] = cfd_->table_cache()->NewIterator(
         read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files[i]->fd, nullptr /* range_del_agg */);
+        *l0_files[i], nullptr /* range_del_agg */,
+        sv_->mutable_cf_options.prefix_extractor.get());
     l0_iters_[i]->SetPinnedItersMgr(pinned_iters_mgr_);
   }
 
@@ -773,7 +802,7 @@ void ForwardIterator::UpdateCurrent() {
       current_ = mutable_iter_;
     }
   }
-  valid_ = (current_ != nullptr);
+  valid_ = current_ != nullptr && immutable_status_.ok();
   if (!status_.ok()) {
     status_ = Status::OK();
   }