Update history and bump patch number

Shared block cache in RocksJava
Summary: Changes to support sharing block cache using the Java API. Previously DB instances could share the block cache only when the same Options instance is passed to all the DB instances. But now, with this change, it is possible to explicitly create a cache and pass it to multiple options instances, to share the block cache. Implementing this for [Rocksandra](https://github.com/instagram/cassandra/tree/rocks_3.0), but this feature has been requested by many java api users over the years. Closes https://github.com/facebook/rocksdb/pull/3623 Differential Revision: D7305794 Pulled By: sagar0 fbshipit-source-id: 03e4e8ed7aeee6f88bada4a8365d4279ede2ad71
2026-11-07 03:22:34 -05:00 · 2018-03-23 11:39:50 -07:00 · 2018-03-23 11:33:26 -07:00 · 2018-03-23 11:15:55 -07:00 · 2018-03-16 14:09:20 -07:00 · 2018-03-16 14:08:01 -07:00
430 changed files with 28378 additions and 6413 deletions
@@ -52,6 +52,9 @@ install:
      brew install ccache;
      PATH=$PATH:/usr/local/opt/ccache/libexec;
    fi
+  - if [[ "${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

 before_script:
  # Increase the maximum number of open file descriptors, since some tests use
@@ -68,8 +71,8 @@ script:
  - 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 .. && make -j4 rocksdb; fi
-  - if [ "${JOB_NAME}" == 'cmake-mingw' ]; then mkdir build && cd build && cmake .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb; fi
+  - if [ "${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
 notifications:
    email:
      - leveldb@fb.com
@@ -14,7 +14,7 @@
 #        cd build
 # 3. Run cmake to generate project files for Windows, add more options to enable required third-party libraries.
 #    See thirdparty.inc for more information.
-#        sample command: cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
+#        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)
 #        msbuild rocksdb.sln
@@ -34,15 +34,25 @@

 cmake_minimum_required(VERSION 2.8.12)
 project(rocksdb)
+enable_language(CXX)
+enable_language(C)
+enable_language(ASM)

 if(POLICY CMP0042)
  cmake_policy(SET CMP0042 NEW)
 endif()

-list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/modules/")
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake/modules/")

 option(WITH_JEMALLOC "build with JeMalloc" OFF)
+option(WITH_SNAPPY "build with SNAPPY" OFF)
+option(WITH_LZ4 "build with lz4" OFF)
+option(WITH_ZLIB "build with zlib" OFF)
 if(MSVC)
+  # Defaults currently different for GFLAGS.
+  #  We will address find_package work a little later
+  option(WITH_GFLAGS "build with GFlags" OFF)
+  option(WITH_XPRESS "build with windows built in compression" OFF)
  include(${CMAKE_CURRENT_SOURCE_DIR}/thirdparty.inc)
 else()
  if(CMAKE_SYSTEM_NAME MATCHES "FreeBSD")
@@ -56,8 +66,18 @@ else()
      include_directories(${JEMALLOC_INCLUDE_DIR})
    endif()
  endif()
+  
+  # No config file for this
+  option(WITH_GFLAGS "build with GFlags" ON)
+  if(WITH_GFLAGS)
+    find_package(gflags)
+    if(gflags_FOUND)
+      add_definitions(-DGFLAGS=1)
+      include_directories(${gflags_INCLUDE_DIR})
+      list(APPEND THIRDPARTY_LIBS ${gflags_LIBRARIES})
+    endif()
+  endif()

-  option(WITH_SNAPPY "build with SNAPPY" OFF)
  if(WITH_SNAPPY)
    find_package(snappy REQUIRED)
    add_definitions(-DSNAPPY)
@@ -65,7 +85,6 @@ else()
    list(APPEND THIRDPARTY_LIBS ${SNAPPY_LIBRARIES})
  endif()

-  option(WITH_ZLIB "build with zlib" OFF)
  if(WITH_ZLIB)
    find_package(zlib REQUIRED)
    add_definitions(-DZLIB)
@@ -81,7 +100,6 @@ else()
    list(APPEND THIRDPARTY_LIBS ${BZIP2_LIBRARIES})
  endif()

-  option(WITH_LZ4 "build with lz4" OFF)
  if(WITH_LZ4)
    find_package(lz4 REQUIRED)
    add_definitions(-DLZ4)
@@ -114,20 +132,19 @@ endif()

 string(REGEX REPLACE "[^0-9a-f]+" "" GIT_SHA "${GIT_SHA}")

-if(NOT WIN32)
-  execute_process(COMMAND
-      "./build_tools/version.sh" "full"
-      WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
-      OUTPUT_VARIABLE ROCKSDB_VERSION
-  )
-  string(STRIP "${ROCKSDB_VERSION}" ROCKSDB_VERSION)
-  execute_process(COMMAND
-      "./build_tools/version.sh" "major"
-      WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
-      OUTPUT_VARIABLE ROCKSDB_VERSION_MAJOR
-  )
-  string(STRIP "${ROCKSDB_VERSION_MAJOR}" ROCKSDB_VERSION_MAJOR)
-endif()
+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)

 option(WITH_MD_LIBRARY "build with MD" ON)
 if(WIN32 AND MSVC)
@@ -163,31 +180,50 @@ else()
  endif()
 endif()

+include(CheckCCompilerFlag)
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc64le")
+  CHECK_C_COMPILER_FLAG("-maltivec" HAS_ALTIVEC)
+  if(HAS_ALTIVEC)
+    message(STATUS " HAS_ALTIVEC yes")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maltivec")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mcpu=power8")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcpu=power8")
+  endif(HAS_ALTIVEC)
+endif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc64le")
+
 option(PORTABLE "build a portable binary" OFF)
 option(FORCE_SSE42 "force building with SSE4.2, even when PORTABLE=ON" OFF)
 if(PORTABLE)
  # MSVC does not need a separate compiler flag to enable SSE4.2; if nmmintrin.h
  # is available, it is available by default.
  if(FORCE_SSE42 AND NOT MSVC)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.2")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.2 -mpclmul")
  endif()
 else()
  if(MSVC)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
  else()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
+    if(NOT HAVE_POWER8)
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
+    endif()
  endif()
 endif()

 include(CheckCXXSourceCompiles)
 if(NOT MSVC)
-  set(CMAKE_REQUIRED_FLAGS "-msse4.2")
+  set(CMAKE_REQUIRED_FLAGS "-msse4.2 -mpclmul")
 endif()
 CHECK_CXX_SOURCE_COMPILES("
 #include <cstdint>
 #include <nmmintrin.h>
+#include <wmmintrin.h>
 int main() {
  volatile uint32_t x = _mm_crc32_u32(0, 0);
+  const auto a = _mm_set_epi64x(0, 0);
+  const auto b = _mm_set_epi64x(0, 0);
+  const auto c = _mm_clmulepi64_si128(a, b, 0x00);
+  auto d = _mm_cvtsi128_si64(c);
 }
 " HAVE_SSE42)
 unset(CMAKE_REQUIRED_FLAGS)
@@ -249,6 +285,20 @@ if(WITH_UBSAN)
  endif()
 endif()

+find_package(NUMA)
+if(NUMA_FOUND)
+  add_definitions(-DNUMA)
+  include_directories(${NUMA_INCLUDE_DIR})
+  list(APPEND THIRDPARTY_LIBS ${NUMA_LIBRARIES})
+endif()
+
+find_package(TBB)
+if(TBB_FOUND)
+  add_definitions(-DTBB)
+  include_directories(${TBB_INCLUDE_DIR})
+  list(APPEND THIRDPARTY_LIBS ${TBB_LIBRARIES})
+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

@@ -316,11 +366,8 @@ if(NOT WIN32)
 endif()

 option(WITH_FALLOCATE "build with fallocate" ON)
-
 if(WITH_FALLOCATE)
-  set(CMAKE_REQUIRED_FLAGS ${CMAKE_C_FLAGS})
-  include(CheckCSourceCompiles)
-  CHECK_C_SOURCE_COMPILES("
+  CHECK_CXX_SOURCE_COMPILES("
 #include <fcntl.h>
 #include <linux/falloc.h>
 int main() {
@@ -333,12 +380,38 @@ int main() {
  endif()
 endif()

-include(CheckFunctionExists)
-CHECK_FUNCTION_EXISTS(malloc_usable_size HAVE_MALLOC_USABLE_SIZE)
+CHECK_CXX_SOURCE_COMPILES("
+#include <fcntl.h>
+int main() {
+  int fd = open(\"/dev/null\", 0);
+  sync_file_range(fd, 0, 1024, SYNC_FILE_RANGE_WRITE);
+}
+" HAVE_SYNC_FILE_RANGE_WRITE)
+if(HAVE_SYNC_FILE_RANGE_WRITE)
+  add_definitions(-DROCKSDB_RANGESYNC_PRESENT)
+endif()
+
+CHECK_CXX_SOURCE_COMPILES("
+#include <pthread.h>
+int main() {
+  (void) PTHREAD_MUTEX_ADAPTIVE_NP;
+}
+" HAVE_PTHREAD_MUTEX_ADAPTIVE_NP)
+if(HAVE_PTHREAD_MUTEX_ADAPTIVE_NP)
+  add_definitions(-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX)
+endif()
+
+include(CheckCXXSymbolExists)
+check_cxx_symbol_exists(malloc_usable_size malloc.h HAVE_MALLOC_USABLE_SIZE)
 if(HAVE_MALLOC_USABLE_SIZE)
  add_definitions(-DROCKSDB_MALLOC_USABLE_SIZE)
 endif()

+check_cxx_symbol_exists(sched_getcpu sched.h HAVE_SCHED_GETCPU)
+if(HAVE_SCHED_GETCPU)
+  add_definitions(-DROCKSDB_SCHED_GETCPU_PRESENT)
+endif()
+
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
 include_directories(SYSTEM ${PROJECT_SOURCE_DIR}/third-party/gtest-1.7.0/fused-src)
@@ -441,6 +514,7 @@ set(SOURCES
        table/block_based_table_factory.cc
        table/block_based_table_reader.cc
        table/block_builder.cc
+        table/block_fetcher.cc
        table/block_prefix_index.cc
        table/bloom_block.cc
        table/cuckoo_table_builder.cc
@@ -528,6 +602,7 @@ set(SOURCES
        utilities/leveldb_options/leveldb_options.cc
        utilities/lua/rocks_lua_compaction_filter.cc
        utilities/memory/memory_util.cc
+        utilities/merge_operators/bytesxor.cc
        utilities/merge_operators/max.cc
        utilities/merge_operators/put.cc
        utilities/merge_operators/string_append/stringappend.cc
@@ -564,10 +639,16 @@ if(HAVE_SSE42 AND NOT FORCE_SSE42)
  if(NOT MSVC)
 set_source_files_properties(
  util/crc32c.cc
-      PROPERTIES COMPILE_FLAGS "-msse4.2")
+      PROPERTIES COMPILE_FLAGS "-msse4.2 -mpclmul")
  endif()
 endif()

+if(HAVE_POWER8)
+  list(APPEND SOURCES
+    util/crc32c_ppc.c
+    util/crc32c_ppc_asm.S)
+endif(HAVE_POWER8)
+
 if(WIN32)
  list(APPEND SOURCES
    port/win/io_win.cc
@@ -662,7 +743,7 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
  set(package_config_destination ${CMAKE_INSTALL_LIBDIR}/cmake/rocksdb)

  configure_package_config_file(
-    ${CMAKE_SOURCE_DIR}/cmake/RocksDBConfig.cmake.in RocksDBConfig.cmake
+    ${CMAKE_CURRENT_LIST_DIR}/cmake/RocksDBConfig.cmake.in RocksDBConfig.cmake
    INSTALL_DESTINATION ${package_config_destination}
  )

@@ -811,7 +892,6 @@ if(WITH_TESTS)
        util/hash_test.cc
        util/heap_test.cc
        util/rate_limiter_test.cc
-        util/slice_test.cc
        util/slice_transform_test.cc
        util/timer_queue_test.cc
        util/thread_list_test.cc
@@ -893,17 +973,18 @@ if(WITH_TESTS)

  foreach(sourcefile ${TEST_EXES})
      get_filename_component(exename ${sourcefile} NAME_WE)
-      add_executable(${exename}${ARTIFACT_SUFFIX} ${sourcefile}
+      add_executable(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} ${sourcefile}
        $<TARGET_OBJECTS:testharness>)
-      set_target_properties(${exename}${ARTIFACT_SUFFIX}
+      set_target_properties(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX}
        PROPERTIES EXCLUDE_FROM_DEFAULT_BUILD_RELEASE 1
        EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
        EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
+        OUTPUT_NAME ${exename}${ARTIFACT_SUFFIX}
        )
-      target_link_libraries(${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} gtest ${LIBS})
+      target_link_libraries(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} gtest ${LIBS})
      if(NOT "${exename}" MATCHES "db_sanity_test")
        add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
-        add_dependencies(check ${exename}${ARTIFACT_SUFFIX})
+        add_dependencies(check ${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX})
      endif()
  endforeach(sourcefile ${TEST_EXES})

@@ -0,0 +1,3 @@
+# Code of Conduct
+
+Facebook has adopted a Code of Conduct that we expect project participants to adhere to. Please [read the full text](https://code.facebook.com/codeofconduct) so that you can understand what actions will and will not be tolerated.
@@ -1,5 +1,8 @@
 # Contributing to RocksDB

+## Code of Conduct
+The code of conduct is described in [`CODE_OF_CONDUCT.md`](CODE_OF_CONDUCT.md)
+
 ## Contributor License Agreement ("CLA")

 In order to accept your pull request, we need you to submit a CLA. You
@@ -1,8 +1,74 @@
 # Rocksdb Change Log
-## Unreleased
-### Public API Change
-### New Features
+## 5.12.2 (3/23/2018)
 ### Bug Fixes
+* Fsync after writing global seq number to the ingestion file in ExternalSstFileIngestionJob.
+
+### Java API Changes
+* Add `BlockBasedTableConfig.setBlockCache` to allow sharing a block cache across DB instances.
+
+## 5.12.1 (3/16/2018)
+### Public API Change
+* RocksDBOptionsParser::Parse()'s `ignore_unknown_options` argument will only be effective if the option file shows it is generated using a higher version of RocksDB than the current version.
+
+### New Features
+* Avoid unnecessarily flushing in `CompactRange()` when the range specified by the user does not overlap unflushed memtables.
+
+### Bug Fixes
+* Fix WAL corruption caused by race condition between user write thread and backup/checkpoint thread.
+
+## 5.12.0 (2/14/2018)
+### Public API Change
+* Iterator::SeekForPrev is now a pure virtual method. This is to prevent user who implement the Iterator interface fail to implement SeekForPrev by mistake.
+* Add `include_end` option to make the range end exclusive when `include_end == false` in `DeleteFilesInRange()`.
+* Add `CompactRangeOptions::allow_write_stall`, which makes `CompactRange` start working immediately, even if it causes user writes to stall. The default value is false, meaning we add delay to `CompactRange` calls until stalling can be avoided when possible. Note this delay is not present in previous RocksDB versions.
+* Creating checkpoint with empty directory now returns `Status::InvalidArgument`; previously, it returned `Status::IOError`.
+* Adds a BlockBasedTableOption to turn off index block compression.
+* Close() method now returns a status when closing a db.
+
+### New Features
+* Improve the performance of iterators doing long range scans by using readahead.
+* Add new function `DeleteFilesInRanges()` to delete files in multiple ranges at once for better performance.
+* FreeBSD build support for RocksDB and RocksJava.
+* Improved performance of long range scans with readahead.
+* Updated to and now continuously tested in Visual Studio 2017.
+
+### Bug Fixes
+* Fix `DisableFileDeletions()` followed by `GetSortedWalFiles()` to not return obsolete WAL files that `PurgeObsoleteFiles()` is going to delete.
+* Fix Handle error return from WriteBuffer() during WAL file close and DB close.
+* Fix advance reservation of arena block addresses.
+* Fix handling of empty string as checkpoint directory.
+
+## 5.11.0 (01/08/2018)
+### Public API Change
+* Add `autoTune` and `getBytesPerSecond()` to RocksJava RateLimiter
+
+### New Features
+* Add a new histogram stat called rocksdb.db.flush.micros for memtable flush.
+* Add "--use_txn" option to use transactional API in db_stress.
+* Disable onboard cache for compaction output in Windows platform.
+* Improve the performance of iterators doing long range scans by using readahead.
+
+### Bug Fixes
+* Fix a stack-use-after-scope bug in ForwardIterator.
+* Fix builds on platforms including Linux, Windows, and PowerPC.
+* Fix buffer overrun in backup engine for DBs with huge number of files.
+* Fix a mislabel bug for bottom-pri compaction threads.
+* Fix DB::Flush() keep waiting after flush finish under certain condition.
+
+## 5.10.0 (12/11/2017)
+### Public API Change
+* When running `make` with environment variable `USE_SSE` set and `PORTABLE` unset, will use all machine features available locally. Previously this combination only compiled SSE-related features.
+
+### New Features
+* Provide lifetime hints when writing files on Linux. This reduces hardware write-amp on storage devices supporting multiple streams.
+* Add a DB stat, `NUMBER_ITER_SKIP`, which returns how many internal keys were skipped during iterations (e.g., due to being tombstones or duplicate versions of a key).
+* Add PerfContext counters, `key_lock_wait_count` and `key_lock_wait_time`, which measure the number of times transactions wait on key locks and total amount of time waiting.
+
+### Bug Fixes
+* Fix IOError on WAL write doesn't propagate to write group follower
+* Make iterator invalid on merge error.
+* Fix performance issue in `IngestExternalFile()` affecting databases with large number of SST files.
+* Fix possible corruption to LSM structure when `DeleteFilesInRange()` deletes a subset of files spanned by a `DeleteRange()` marker.

 ## 5.9.0 (11/1/2017)
 ### Public API Change
@@ -11,11 +77,13 @@
 * API call `DB::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum)` was added, users who wish to preserve deletes are expected to periodically call this function to advance the cutoff seqnum (all deletes made before this seqnum can be dropped by DB). It's user responsibility to figure out how to advance the seqnum in the way so the tombstones are kept for the desired period of time, yet are eventually processed in time and don't eat up too much space.
 * `ReadOptions::iter_start_seqnum` was added; if set to something > 0 user will see 2 changes in iterators behavior 1) only keys written with sequence larger than this parameter would be returned and 2) the `Slice` returned by iter->key() now points to the the memory that keep User-oriented representation of the internal key, rather than user key. New struct `FullKey` was added to represent internal keys, along with a new helper function `ParseFullKey(const Slice& internal_key, FullKey* result);`.
 * Deprecate trash_dir param in NewSstFileManager, right now we will rename deleted files to <name>.trash instead of moving them to trash directory
+* Allow setting a custom trash/DB size ratio limit in the SstFileManager, after which files that are to be scheduled for deletion are deleted immediately, regardless of any delete ratelimit.
 * Return an error on write if write_options.sync = true and write_options.disableWAL = true to warn user of inconsistent options. Previously we will not write to WAL and not respecting the sync options in this case.

 ### New Features
+* CRC32C is now using the 3-way pipelined SSE algorithm `crc32c_3way` on supported platforms to improve performance. The system will choose to use this algorithm on supported platforms automatically whenever possible. If PCLMULQDQ is not supported it will fall back to the old Fast_CRC32 algorithm.
 * `DBOptions::writable_file_max_buffer_size` can now be changed dynamically.
-* `DBOptions::bytes_per_sync` and `DBOptions::wal_bytes_per_sync` can now be changed dynamically, `DBOptions::wal_bytes_per_sync` will flush all memtables and switch to a new WAL file.
+* `DBOptions::bytes_per_sync`, `DBOptions::compaction_readahead_size`, and `DBOptions::wal_bytes_per_sync` can now be changed dynamically, `DBOptions::wal_bytes_per_sync` will flush all memtables and switch to a new WAL file.
 * Support dynamic adjustment of rate limit according to demand for background I/O. It can be enabled by passing `true` to the `auto_tuned` parameter in `NewGenericRateLimiter()`. The value passed as `rate_bytes_per_sec` will still be respected as an upper-bound.
 * Support dynamically changing `ColumnFamilyOptions::compaction_options_fifo`.
 * Introduce `EventListener::OnStallConditionsChanged()` callback. Users can implement it to be notified when user writes are stalled, stopped, or resumed.
@@ -107,6 +107,41 @@ to build a portable binary, add `PORTABLE=1` before your make commands, like thi
            * run `brew tap homebrew/versions; brew install gcc48 --use-llvm` to install gcc 4.8 (or higher).
    * run `brew install rocksdb`

+* **FreeBSD** (11.01):
+
+    * You can either install RocksDB from the Ports system using `cd /usr/ports/databases/rocksdb && make install`, or you can follow the details below to install dependencies and compile from source code:
+
+    * Install the dependencies for RocksDB:
+
+        export BATCH=YES
+        cd /usr/ports/devel/gmake && make install
+        cd /usr/ports/devel/gflags && make install
+
+        cd /usr/ports/archivers/snappy && make install
+        cd /usr/ports/archivers/bzip2 && make install
+        cd /usr/ports/archivers/liblz4 && make install
+        cd /usr/ports/archivesrs/zstd && make install
+
+        cd /usr/ports/devel/git && make install
+
+
+    * Install the dependencies for RocksJava (optional):
+
+        export BATCH=yes
+        cd /usr/ports/java/openjdk7 && make install
+
+    * Build RocksDB from source:
+        cd ~
+        git clone https://github.com/facebook/rocksdb.git
+        cd rocksdb
+        gmake static_lib
+
+    * Build RocksJava from source (optional):
+        cd rocksdb
+        export JAVA_HOME=/usr/local/openjdk7
+        gmake rocksdbjava
+
+
 * **iOS**:
  * Run: `TARGET_OS=IOS make static_lib`. When building the project which uses rocksdb iOS library, make sure to define two important pre-processing macros: `ROCKSDB_LITE` and `IOS_CROSS_COMPILE`.

@@ -14,3 +14,5 @@ This is the list of all known third-party language bindings for RocksDB. If some
    * 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
@@ -232,9 +232,18 @@ endif
 # USAN doesn't work well with jemalloc. If we're compiling with USAN, we should use regular malloc.
 ifdef COMPILE_WITH_UBSAN
 	DISABLE_JEMALLOC=1
-	EXEC_LDFLAGS += -fsanitize=undefined
-	PLATFORM_CCFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
-	PLATFORM_CXXFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
+	# Suppress alignment warning because murmurhash relies on casting unaligned
+	# memory to integer. Fixing it may cause performance regression. 3-way crc32
+	# relies on it too, although it can be rewritten to eliminate with minimal
+	# performance regression.
+	EXEC_LDFLAGS += -fsanitize=undefined -fno-sanitize-recover=all
+	PLATFORM_CCFLAGS += -fsanitize=undefined -fno-sanitize-recover=all -DROCKSDB_UBSAN_RUN
+	PLATFORM_CXXFLAGS += -fsanitize=undefined -fno-sanitize-recover=all -DROCKSDB_UBSAN_RUN
+endif
+
+ifdef ROCKSDB_VALGRIND_RUN
+	PLATFORM_CCFLAGS += -DROCKSDB_VALGRIND_RUN
+	PLATFORM_CXXFLAGS += -DROCKSDB_VALGRIND_RUN
 endif

 ifndef DISABLE_JEMALLOC
@@ -300,6 +309,9 @@ LDFLAGS += $(LUA_LIB)

 endif

+ifeq ($(NO_THREEWAY_CRC32C), 1)
+	CXXFLAGS += -DNO_THREEWAY_CRC32C
+endif

 CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers
@@ -336,6 +348,8 @@ ifeq ($(HAVE_POWER8),1)
 LIB_CC_OBJECTS = $(LIB_SOURCES:.cc=.o)
 LIBOBJECTS += $(LIB_SOURCES_C:.c=.o)
 LIBOBJECTS += $(LIB_SOURCES_ASM:.S=.o)
+else
+LIB_CC_OBJECTS = $(LIB_SOURCES:.cc=.o)
 endif

 LIBOBJECTS += $(TOOL_LIB_SOURCES:.cc=.o)
@@ -494,12 +508,12 @@ TESTS = \
 	repair_test \
 	env_timed_test \
 	write_prepared_transaction_test \
-	slice_test \

 PARALLEL_TEST = \
 	backupable_db_test \
 	db_compaction_filter_test \
 	db_compaction_test \
+	db_merge_operator_test \
 	db_sst_test \
 	db_test \
 	db_universal_compaction_test \
@@ -511,8 +525,12 @@ PARALLEL_TEST = \
 	persistent_cache_test \
 	table_test \
 	transaction_test \
-	write_prepared_transaction_test
+	write_prepared_transaction_test \

+# options_settable_test doesn't pass with UBSAN as we use hack in the test
+ifdef COMPILE_WITH_UBSAN
+        TESTS := $(shell echo $(TESTS) | sed 's/\boptions_settable_test\b//g')
+endif
 SUBSET := $(TESTS)
 ifdef ROCKSDBTESTS_START
        SUBSET := $(shell echo $(SUBSET) | sed 's/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/')
@@ -871,7 +889,7 @@ ubsan_crash_test:
 	$(MAKE) clean

 valgrind_test:
-	DISABLE_JEMALLOC=1 $(MAKE) valgrind_check
+	ROCKSDB_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 $(MAKE) valgrind_check

 valgrind_check: $(TESTS)
 	$(MAKE) DRIVER="$(VALGRIND_VER) $(VALGRIND_OPTS)" gen_parallel_tests
@@ -1478,9 +1496,6 @@ range_del_aggregator_test: db/range_del_aggregator_test.o db/db_test_util.o $(LI
 blob_db_test: utilities/blob_db/blob_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-slice_test: util/slice_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1549,11 +1564,11 @@ BZIP2_DOWNLOAD_BASE ?= http://www.bzip.org
 SNAPPY_VER ?= 1.1.4
 SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
-LZ4_VER ?= 1.7.5
-LZ4_SHA256 ?= 0190cacd63022ccb86f44fa5041dc6c3804407ad61550ca21c382827319e7e7e
+LZ4_VER ?= 1.8.0
+LZ4_SHA256 ?= 2ca482ea7a9bb103603108b5a7510b7592b90158c151ff50a28f1ca8389fccf6
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.2.0
-ZSTD_SHA256 ?= 4a7e4593a3638276ca7f2a09dc4f38e674d8317bbea51626393ca73fc047cbfb
+ZSTD_VER ?= 1.3.3
+ZSTD_SHA256 ?= a77c47153ee7de02626c5b2a097005786b71688be61e9fb81806a011f90b297b
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive

 ifeq ($(PLATFORM), OS_MACOSX)
@@ -1567,7 +1582,7 @@ else
 endif
 endif
 ifeq ($(PLATFORM), OS_FREEBSD)
-	JAVA_INCLUDE += -I$(JAVA_HOME)/include/freebsd
+	JAVA_INCLUDE = -I$(JAVA_HOME)/include -I$(JAVA_HOME)/include/freebsd
 	ROCKSDBJNILIB = librocksdbjni-freebsd$(ARCH).so
 	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-freebsd$(ARCH).jar
 endif
@@ -1593,7 +1608,7 @@ libz.a:
 		exit 1; \
 	fi
 	tar xvzf zlib-$(ZLIB_VER).tar.gz
-	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC ${EXTRA_CFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --static && make
+	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC ${EXTRA_CFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --static && $(MAKE)
 	cp zlib-$(ZLIB_VER)/libz.a .

 libbz2.a:
@@ -1605,7 +1620,7 @@ libbz2.a:
 		exit 1; \
 	fi
 	tar xvzf bzip2-$(BZIP2_VER).tar.gz
-	cd bzip2-$(BZIP2_VER) && make CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64 ${EXTRA_CFLAGS}' AR='ar ${EXTRA_ARFLAGS}'
+	cd bzip2-$(BZIP2_VER) && $(MAKE) CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64 ${EXTRA_CFLAGS}' AR='ar ${EXTRA_ARFLAGS}'
 	cp bzip2-$(BZIP2_VER)/libbz2.a .

 libsnappy.a:
@@ -1618,7 +1633,7 @@ libsnappy.a:
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
 	cd snappy-$(SNAPPY_VER) && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --with-pic --enable-static --disable-shared
-	cd snappy-$(SNAPPY_VER) && make ${SNAPPY_MAKE_TARGET}
+	cd snappy-$(SNAPPY_VER) && $(MAKE) ${SNAPPY_MAKE_TARGET}
 	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .

 liblz4.a:
@@ -1631,7 +1646,7 @@ liblz4.a:
 		exit 1; \
 	fi
 	tar xvzf lz4-$(LZ4_VER).tar.gz
-	cd lz4-$(LZ4_VER)/lib && make CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
+	cd lz4-$(LZ4_VER)/lib && $(MAKE) CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
 	cp lz4-$(LZ4_VER)/lib/liblz4.a .

 libzstd.a:
@@ -1644,29 +1659,45 @@ libzstd.a:
 		exit 1; \
 	fi
 	tar xvzf zstd-$(ZSTD_VER).tar.gz
-	cd zstd-$(ZSTD_VER)/lib && make CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
+	cd zstd-$(ZSTD_VER)/lib && DESTDIR=. PREFIX= $(MAKE) CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' install
 	cp zstd-$(ZSTD_VER)/lib/libzstd.a .

 # A version of each $(LIBOBJECTS) compiled with -fPIC and a fixed set of static compression libraries
-java_static_libobjects = $(patsubst %,jls/%,$(LIBOBJECTS))
+java_static_libobjects = $(patsubst %,jls/%,$(LIB_CC_OBJECTS))
 CLEAN_FILES += jls
+java_static_all_libobjects = $(java_static_libobjects)

 ifneq ($(ROCKSDB_JAVA_NO_COMPRESSION), 1)
 JAVA_COMPRESSIONS = libz.a libbz2.a libsnappy.a liblz4.a libzstd.a
 endif

 JAVA_STATIC_FLAGS = -DZLIB -DBZIP2 -DSNAPPY -DLZ4 -DZSTD
-JAVA_STATIC_INCLUDES = -I./zlib-$(ZLIB_VER) -I./bzip2-$(BZIP2_VER) -I./snappy-$(SNAPPY_VER) -I./lz4-$(LZ4_VER)/lib -I./zstd-$(ZSTD_VER)/lib
+JAVA_STATIC_INCLUDES = -I./zlib-$(ZLIB_VER) -I./bzip2-$(BZIP2_VER) -I./snappy-$(SNAPPY_VER) -I./lz4-$(LZ4_VER)/lib -I./zstd-$(ZSTD_VER)/lib/include
+
+ifeq ($(HAVE_POWER8),1)
+JAVA_STATIC_C_LIBOBJECTS = $(patsubst %.c.o,jls/%.c.o,$(LIB_SOURCES_C:.c=.o))
+JAVA_STATIC_ASM_LIBOBJECTS = $(patsubst %.S.o,jls/%.S.o,$(LIB_SOURCES_ASM:.S=.o))
+
+java_static_ppc_libobjects = $(JAVA_STATIC_C_LIBOBJECTS) $(JAVA_STATIC_ASM_LIBOBJECTS)
+
+jls/util/crc32c_ppc.o: util/crc32c_ppc.c
+	$(AM_V_CC)$(CC) $(CFLAGS) $(JAVA_STATIC_FLAGS) $(JAVA_STATIC_INCLUDES) -c $< -o $@
+
+jls/util/crc32c_ppc_asm.o: util/crc32c_ppc_asm.S
+	$(AM_V_CC)$(CC) $(CFLAGS) $(JAVA_STATIC_FLAGS) $(JAVA_STATIC_INCLUDES) -c $< -o $@
+
+java_static_all_libobjects += $(java_static_ppc_libobjects)
+endif

 $(java_static_libobjects): jls/%.o: %.cc $(JAVA_COMPRESSIONS)
 	$(AM_V_CC)mkdir -p $(@D) && $(CXX) $(CXXFLAGS) $(JAVA_STATIC_FLAGS) $(JAVA_STATIC_INCLUDES) -fPIC -c $< -o $@ $(COVERAGEFLAGS)

-rocksdbjavastatic: $(java_static_libobjects)
+rocksdbjavastatic: $(java_static_all_libobjects)
 	cd java;$(MAKE) javalib;
 	rm -f ./java/target/$(ROCKSDBJNILIB)
 	$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC \
 	  -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) \
-	  $(java_static_libobjects) $(COVERAGEFLAGS) \
+	  $(java_static_all_libobjects) $(COVERAGEFLAGS) \
 	  $(JAVA_COMPRESSIONS) $(JAVA_STATIC_LDFLAGS)
 	cd java/target;strip $(STRIPFLAGS) $(ROCKSDBJNILIB)
 	cd java;jar -cf target/$(ROCKSDB_JAR) HISTORY*.md
@@ -1727,7 +1758,7 @@ JAVA_C_LIBOBJECTS = $(patsubst %.c.o,jl/%.c.o,$(JAVA_C_OBJECTS))
 JAVA_ASM_LIBOBJECTS = $(patsubst %.S.o,jl/%.S.o,$(JAVA_ASM_OBJECTS))
 endif

-java_libobjects = $(patsubst %,jl/%,$(LIBOBJECTS))
+java_libobjects = $(patsubst %,jl/%,$(LIB_CC_OBJECTS))
 CLEAN_FILES += jl
 java_all_libobjects = $(java_libobjects)

@@ -1,7 +1,8 @@
 ## RocksDB: A Persistent Key-Value Store for Flash and RAM Storage

-[![Build Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
-[![Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
+[![Linux/Mac Build Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
+[![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.
@@ -85,3 +85,6 @@ quasardb uses a heavily tuned RocksDB as its persistence layer.

 ## LzLabs
 LzLabs is using RocksDB as a storage engine in their multi-database distributed framework to store application configuration and user data.
+
+## ProfaneDB
+[ProfaneDB](https://profanedb.gitlab.io/) is a database for Protocol Buffers, and uses RocksDB for storage. It is accessible via gRPC, and the schema is defined using directly `.proto` files.
@@ -14,6 +14,11 @@ Vagrant.configure("2") do |config|
    box.vm.box = "chef/centos-6.5"
  end

+  config.vm.define "centos7" do |box|
+    box.vm.box = "centos/7"
+    box.vm.provision "shell", path: "build_tools/setup_centos7.sh"
+  end
+
  config.vm.define "FreeBSD10" do |box|
    box.vm.guest = :freebsd
    box.vm.box = "robin/freebsd-10"
@@ -1,9 +1,9 @@
 version: 1.0.{build}
-image: Visual Studio 2015
+image: Visual Studio 2017
 before_build:
 - md %APPVEYOR_BUILD_FOLDER%\build
 - cd %APPVEYOR_BUILD_FOLDER%\build
- cmake -G "Visual Studio 14 2015 Win64" -DOPTDBG=1 -DXPRESS=1 -DPORTABLE=1 ..
+- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 ..
 - cd ..
 build:
  project: build\rocksdb.sln
@@ -36,7 +36,7 @@ def parse_src_mk(repo_path):
 # get all .cc / .c files
 def get_cc_files(repo_path):
    cc_files = []
-    for root, dirnames, filenames in os.walk(repo_path):
+    for root, dirnames, filenames in os.walk(repo_path):  # noqa: B007 T25377293 Grandfathered in
        root = root[(len(repo_path) + 1):]
        if "java" in root:
            # Skip java
@@ -1,4 +1,5 @@
 #!/usr/bin/env bash
 # Create a tmp directory for the test to use
 TEST_DIR=$(mktemp -d /dev/shm/fbcode_rocksdb_XXXXXXX)
+# shellcheck disable=SC2068
 TEST_TMPDIR="$TEST_DIR" $@ && rm -rf "$TEST_DIR"
@@ -3,19 +3,17 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
 import targets_cfg
-import pprint

-# TODO(tec): replace this with PrettyPrinter
-def pretty_list(lst, indent=6):
+def pretty_list(lst, indent=8):
    if lst is None or len(lst) == 0:
        return ""

    if len(lst) == 1:
        return "\"%s\"" % lst[0]
-
+    
    separator = "\",\n%s\"" % (" " * indent)
-    res = separator.join(lst)
-    res = "\n" + (" " * indent) + "\"" + res + "\",\n" + (" " * (indent - 2))
+    res = separator.join(sorted(lst))
+    res = "\n" + (" " * indent) + "\"" + res + "\",\n" + (" " * (indent - 4))
    return res


@@ -27,7 +25,7 @@ class TARGETSBuilder:
        self.total_lib = 0
        self.total_bin = 0
        self.total_test = 0
-        self.tests_cfg = []
+        self.tests_cfg = ""

    def __del__(self):
        self.targets_file.close()
@@ -37,8 +35,8 @@ class TARGETSBuilder:
            headers = "AutoHeaders.RECURSIVE_GLOB"
        self.targets_file.write(targets_cfg.library_template % (
            name,
-            headers,
            pretty_list(srcs),
+            headers,
            pretty_list(deps)))
        self.total_lib = self.total_lib + 1

@@ -53,13 +51,13 @@ class TARGETSBuilder:
        exec_mode = "serial"
        if is_parallel:
            exec_mode = "parallel"
-        self.tests_cfg.append([test_name, str(src), str(exec_mode)])
+        self.tests_cfg += targets_cfg.test_cfg_template % (
+            test_name,
+            str(src),
+            str(exec_mode))

        self.total_test = self.total_test + 1

    def flush_tests(self):
-        self.targets_file.write(targets_cfg.unittests_template % (
-            pprint.PrettyPrinter().pformat(self.tests_cfg)
-        ))
-
-        self.tests_cfg = []
+        self.targets_file.write(targets_cfg.unittests_template % self.tests_cfg)
+        self.tests_cfg = ""
@@ -2,123 +2,141 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
-rocksdb_target_header = """
-import os
+rocksdb_target_header = """REPO_PATH = package_name() + "/"

-TARGETS_PATH = os.path.dirname(__file__)
-REPO_PATH = "rocksdb/src/"
 BUCK_BINS = "buck-out/gen/" + REPO_PATH
+
 TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
+
 rocksdb_compiler_flags = [
-  "-fno-builtin-memcmp",
-  "-DROCKSDB_PLATFORM_POSIX",
-  "-DROCKSDB_LIB_IO_POSIX",
-  "-DROCKSDB_FALLOCATE_PRESENT",
-  "-DROCKSDB_MALLOC_USABLE_SIZE",
-  "-DROCKSDB_RANGESYNC_PRESENT",
-  "-DROCKSDB_SCHED_GETCPU_PRESENT",
-  "-DROCKSDB_SUPPORT_THREAD_LOCAL",
-  "-DOS_LINUX",
-  # Flags to enable libs we include
-  "-DSNAPPY",
-  "-DZLIB",
-  "-DBZIP2",
-  "-DLZ4",
-  "-DZSTD",
-  "-DGFLAGS=gflags",
-  "-DNUMA",
-  "-DTBB",
-  # Needed to compile in fbcode
-  "-Wno-expansion-to-defined",
+    "-fno-builtin-memcmp",
+    "-DROCKSDB_PLATFORM_POSIX",
+    "-DROCKSDB_LIB_IO_POSIX",
+    "-DROCKSDB_FALLOCATE_PRESENT",
+    "-DROCKSDB_MALLOC_USABLE_SIZE",
+    "-DROCKSDB_RANGESYNC_PRESENT",
+    "-DROCKSDB_SCHED_GETCPU_PRESENT",
+    "-DROCKSDB_SUPPORT_THREAD_LOCAL",
+    "-DOS_LINUX",
+    # Flags to enable libs we include
+    "-DSNAPPY",
+    "-DZLIB",
+    "-DBZIP2",
+    "-DLZ4",
+    "-DZSTD",
+    "-DGFLAGS=gflags",
+    "-DNUMA",
+    "-DTBB",
+    # Needed to compile in fbcode
+    "-Wno-expansion-to-defined",
 ]

 rocksdb_external_deps = [
-  ('bzip2', None, 'bz2'),
-  ('snappy', None, "snappy"),
-  ('zlib', None, 'z'),
-  ('gflags', None, 'gflags'),
-  ('lz4', None, 'lz4'),
-  ('zstd', None),
-  ('tbb', None),
-  ("numa", None, "numa"),
-  ("googletest", None, "gtest"),
+    ("bzip2", None, "bz2"),
+    ("snappy", None, "snappy"),
+    ("zlib", None, "z"),
+    ("gflags", None, "gflags"),
+    ("lz4", None, "lz4"),
+    ("zstd", None),
+    ("tbb", None),
+    ("numa", None, "numa"),
+    ("googletest", None, "gtest"),
 ]

 rocksdb_preprocessor_flags = [
-  # Directories with files for #include
-  "-I" + REPO_PATH + "include/",
-  "-I" + REPO_PATH,
+    # Directories with files for #include
+    "-I" + REPO_PATH + "include/",
+    "-I" + REPO_PATH,
 ]

 rocksdb_arch_preprocessor_flags = {
-  "x86_64": ["-DHAVE_SSE42"],
+    "x86_64": ["-DHAVE_SSE42"],
 }
+
+build_mode = read_config("fbcode", "build_mode")
+
+is_opt_mode = build_mode.startswith("opt")
+
+# -DNDEBUG is added by default in opt mode in fbcode. But adding it twice
+# doesn't harm and avoid forgetting to add it.
+if is_opt_mode:
+    rocksdb_compiler_flags.append("-DNDEBUG")
 """


 library_template = """
 cpp_library(
    name = "%s",
-    headers = %s,
    srcs = [%s],
-    deps = [%s],
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    headers = %s,
    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
    compiler_flags = rocksdb_compiler_flags,
+    preprocessor_flags = rocksdb_preprocessor_flags,
+    deps = [%s],
    external_deps = rocksdb_external_deps,
 )
 """

 binary_template = """
 cpp_binary(
-  name = "%s",
-  srcs = [%s],
-  deps = [%s],
-  preprocessor_flags = rocksdb_preprocessor_flags,
-  arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-  compiler_flags = rocksdb_compiler_flags,
-  external_deps = rocksdb_external_deps,
+    name = "%s",
+    srcs = [%s],
+    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
+    compiler_flags = rocksdb_compiler_flags,
+    preprocessor_flags = rocksdb_preprocessor_flags,
+    deps = [%s],
+    external_deps = rocksdb_external_deps,
 )
 """

+test_cfg_template = """    [
+        "%s",
+        "%s",
+        "%s",
+    ],
+"""
+
 unittests_template = """
 # [test_name, test_src, test_type]
-ROCKS_TESTS = %s
-
+ROCKS_TESTS = [
+%s]

 # Generate a test rule for each entry in ROCKS_TESTS
-for test_cfg in ROCKS_TESTS:
-    test_name = test_cfg[0]
-    test_cc = test_cfg[1]
-    ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
-    test_bin = test_name + "_bin"
+# Do not build the tests in opt mode, since SyncPoint and other test code
+# will not be included.
+if not is_opt_mode:
+    for test_cfg in ROCKS_TESTS:
+        test_name = test_cfg[0]
+        test_cc = test_cfg[1]
+        ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
+        test_bin = test_name + "_bin"

-    cpp_binary (
-      name = test_bin,
-      srcs = [test_cc],
-      deps = [":rocksdb_test_lib"],
-      preprocessor_flags = rocksdb_preprocessor_flags,
-      arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-      compiler_flags = rocksdb_compiler_flags,
-      external_deps = rocksdb_external_deps,
-    )
+        cpp_binary (
+          name = test_bin,
+          srcs = [test_cc],
+          deps = [":rocksdb_test_lib"],
+          preprocessor_flags = rocksdb_preprocessor_flags,
+          arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
+          compiler_flags = rocksdb_compiler_flags,
+          external_deps = rocksdb_external_deps,
+        )

-    custom_unittest(
-      name = test_name,
-      type = ttype,
-      deps = [":" + test_bin],
-      command = [TEST_RUNNER, BUCK_BINS + test_bin]
-    )
+        custom_unittest(
+          name = test_name,
+          type = ttype,
+          deps = [":" + test_bin],
+          command = [TEST_RUNNER, BUCK_BINS + test_bin]
+        )

 custom_unittest(
    name = "make_rocksdbjavastatic",
-    type = "simple",
    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
+    type = "simple",
 )

 custom_unittest(
    name = "make_rocksdb_lite_release",
-    type = "simple",
    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
+    type = "simple",
 )
 """
@@ -141,6 +141,7 @@ case "$TARGET_OS" in
        ;;
    FreeBSD)
        PLATFORM=OS_FREEBSD
+        CXX=clang++
        COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_FREEBSD"
        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
        # PORT_FILES=port/freebsd/freebsd_specific.cc
@@ -210,119 +211,137 @@ EOF
        fi
    fi

-    # Test whether Snappy library is installed
-    # http://code.google.com/p/snappy/
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <snappy.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_SNAPPY; then
+        # Test whether Snappy library is installed
+        # http://code.google.com/p/snappy/
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <snappy.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DSNAPPY"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lsnappy"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -lsnappy"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DSNAPPY"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lsnappy"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lsnappy"
+        fi
    fi

-    # Test whether gflags library is installed
-    # http://gflags.github.io/gflags/
-    # check if the namespace is gflags
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
-      #include <gflags/gflags.h>
-      using namespace gflags;
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_GFLAGS; then
+        # Test whether gflags library is installed
+        # http://gflags.github.io/gflags/
+        # check if the namespace is gflags
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
+          #include <gflags/gflags.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=gflags"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
-    else
-      # check if namespace is google
-      $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
-        #include <gflags/gflags.h>
-        using namespace google;
-        int main() {}
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=1"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
+        else
+          # check if namespace is google
+          $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
+            #include <gflags/gflags.h>
+            using namespace google;
+            int main() {}
 EOF
-      if [ "$?" = 0 ]; then
-          COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=google"
-          PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
-      fi
+          if [ "$?" = 0 ]; then
+              COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=google"
+              PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
+          fi
+        fi
    fi

-    # Test whether zlib library is installed
-    $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <zlib.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_ZLIB; then
+        # Test whether zlib library is installed
+        $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <zlib.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DZLIB"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lz"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -lz"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DZLIB"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lz"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lz"
+        fi
    fi

-    # Test whether bzip library is installed
-    $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <bzlib.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_BZIP; then
+        # Test whether bzip library is installed
+        $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <bzlib.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DBZIP2"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbz2"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -lbz2"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DBZIP2"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbz2"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lbz2"
+        fi
    fi

-    # Test whether lz4 library is installed
-    $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <lz4.h>
-      #include <lz4hc.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_LZ4; then
+        # Test whether lz4 library is installed
+        $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <lz4.h>
+          #include <lz4hc.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DLZ4"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -llz4"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -llz4"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DLZ4"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -llz4"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -llz4"
+        fi
    fi

-    # Test whether zstd library is installed
-    $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <zstd.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_ZSTD; then
+        # Test whether zstd library is installed
+        $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <zstd.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DZSTD"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lzstd"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -lzstd"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DZSTD"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lzstd"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lzstd"
+        fi
    fi

-    # Test whether numa is available
-    $CXX $CFLAGS -x c++ - -o /dev/null -lnuma 2>/dev/null  <<EOF
-      #include <numa.h>
-      #include <numaif.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_NUMA; then
+        # Test whether numa is available
+        $CXX $CFLAGS -x c++ - -o /dev/null -lnuma 2>/dev/null  <<EOF
+          #include <numa.h>
+          #include <numaif.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DNUMA"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lnuma"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -lnuma"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DNUMA"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lnuma"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lnuma"
+        fi
    fi

-    # Test whether tbb is available
-    $CXX $CFLAGS $LDFLAGS -x c++ - -o /dev/null -ltbb 2>/dev/null  <<EOF
-      #include <tbb/tbb.h>
-      int main() {}
+    if ! test $ROCKSDB_DISABLE_TBB; then
+        # Test whether tbb is available
+        $CXX $CFLAGS $LDFLAGS -x c++ - -o /dev/null -ltbb 2>/dev/null  <<EOF
+          #include <tbb/tbb.h>
+          int main() {}
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DTBB"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltbb"
-        JAVA_LDFLAGS="$JAVA_LDFLAGS -ltbb"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DTBB"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltbb"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -ltbb"
+        fi
    fi

-    # Test whether jemalloc is available
-    if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null -ljemalloc \
-      2>/dev/null; then
-        # This will enable some preprocessor identifiers in the Makefile
-        JEMALLOC=1
-        # JEMALLOC can be enabled either using the flag (like here) or by
-        # providing direct link to the jemalloc library
-        WITH_JEMALLOC_FLAG=1
-    else
+    if ! test $ROCKSDB_DISABLE_JEMALLOC; then
+        # Test whether jemalloc is available
+        if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null -ljemalloc \
+          2>/dev/null; then
+            # This will enable some preprocessor identifiers in the Makefile
+            JEMALLOC=1
+            # JEMALLOC can be enabled either using the flag (like here) or by
+            # providing direct link to the jemalloc library
+            WITH_JEMALLOC_FLAG=1
+        fi
+    fi
+    if ! test $JEMALLOC && ! test $ROCKSDB_DISABLE_TCMALLOC; then
        # jemalloc is not available. Let's try tcmalloc
        if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null \
          -ltcmalloc 2>/dev/null; then
@@ -331,88 +350,100 @@ EOF
        fi
    fi

-    # Test whether malloc_usable_size is available
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <malloc.h>
-      int main() {
-        size_t res = malloc_usable_size(0);
-        return 0;
-      }
+    if ! test $ROCKSDB_DISABLE_MALLOC_USABLE_SIZE; then
+        # Test whether malloc_usable_size is available
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <malloc.h>
+          int main() {
+            size_t res = malloc_usable_size(0);
+            return 0;
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
+        fi
    fi

-    # Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <pthread.h>
-      int main() {
-        int x = PTHREAD_MUTEX_ADAPTIVE_NP;
-        return 0;
-      }
+    if ! test $ROCKSDB_DISABLE_PTHREAD_MUTEX_ADAPTIVE_NP; then
+        # Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <pthread.h>
+          int main() {
+            int x = PTHREAD_MUTEX_ADAPTIVE_NP;
+            return 0;
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_PTHREAD_ADAPTIVE_MUTEX"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_PTHREAD_ADAPTIVE_MUTEX"
+        fi
    fi

-    # Test whether backtrace is available
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <execinfo.h>>
-      int main() {
-        void* frames[1];
-        backtrace_symbols(frames, backtrace(frames, 1));
-        return 0;
-      }
-EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
-    else
-        # Test whether execinfo library is installed
-        $CXX $CFLAGS -lexecinfo -x c++ - -o /dev/null 2>/dev/null  <<EOF
-          #include <execinfo.h>
+    if ! test $ROCKSDB_DISABLE_BACKTRACE; then
+        # Test whether backtrace is available
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <execinfo.h>>
          int main() {
            void* frames[1];
            backtrace_symbols(frames, backtrace(frames, 1));
+            return 0;
          }
 EOF
        if [ "$?" = 0 ]; then
            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
-            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lexecinfo"
-            JAVA_LDFLAGS="$JAVA_LDFLAGS -lexecinfo"
+        else
+            # Test whether execinfo library is installed
+            $CXX $CFLAGS -lexecinfo -x c++ - -o /dev/null 2>/dev/null  <<EOF
+              #include <execinfo.h>
+              int main() {
+                void* frames[1];
+                backtrace_symbols(frames, backtrace(frames, 1));
+              }
+EOF
+            if [ "$?" = 0 ]; then
+                COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+                PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lexecinfo"
+                JAVA_LDFLAGS="$JAVA_LDFLAGS -lexecinfo"
+            fi
        fi
    fi

-    # Test if -pg is supported
-    $CXX $CFLAGS -pg -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      int main() {
-        return 0;
-      }
+    if ! test $ROCKSDB_DISABLE_PG; then
+        # Test if -pg is supported
+        $CXX $CFLAGS -pg -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          int main() {
+            return 0;
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        PROFILING_FLAGS=-pg
+        if [ "$?" = 0 ]; then
+            PROFILING_FLAGS=-pg
+        fi
    fi

-    # Test whether sync_file_range is supported for compatibility with an old glibc
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <fcntl.h>
-      int main() {
-        int fd = open("/dev/null", 0);
-        sync_file_range(fd, 0, 1024, SYNC_FILE_RANGE_WRITE);
-      }
+    if ! test $ROCKSDB_DISABLE_SYNC_FILE_RANGE; then
+        # Test whether sync_file_range is supported for compatibility with an old glibc
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <fcntl.h>
+          int main() {
+            int fd = open("/dev/null", 0);
+            sync_file_range(fd, 0, 1024, SYNC_FILE_RANGE_WRITE);
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_RANGESYNC_PRESENT"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_RANGESYNC_PRESENT"
+        fi
    fi

-    # Test whether sched_getcpu is supported
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <sched.h>
-      int main() {
-        int cpuid = sched_getcpu();
-      }
+    if ! test $ROCKSDB_DISABLE_SCHED_GETCPU; then
+        # Test whether sched_getcpu is supported
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <sched.h>
+          int main() {
+            int cpuid = sched_getcpu();
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SCHED_GETCPU_PRESENT"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SCHED_GETCPU_PRESENT"
+        fi
    fi
 fi

@@ -444,18 +475,23 @@ if test "$USE_HDFS"; then
  JAVA_LDFLAGS="$JAVA_LDFLAGS $HDFS_LDFLAGS"
 fi

-if test "$USE_SSE"; then
-  COMMON_FLAGS="$COMMON_FLAGS -msse4.2"
-elif test -z "$PORTABLE"; then
+if test -z "$PORTABLE"; then
  if test -n "`echo $TARGET_ARCHITECTURE | grep ^ppc64`"; then
    # Tune for this POWER processor, treating '+' models as base models
    POWER=`LD_SHOW_AUXV=1 /bin/true | grep AT_PLATFORM | grep -E -o power[0-9]+`
    COMMON_FLAGS="$COMMON_FLAGS -mcpu=$POWER -mtune=$POWER "
  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^s390x`"; then
    COMMON_FLAGS="$COMMON_FLAGS -march=z10 "
+  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
+    # TODO: Handle this with approprite options.
+    COMMON_FLAGS="$COMMON_FLAGS"
  elif [ "$TARGET_OS" != AIX ] && [ "$TARGET_OS" != SunOS ]; then
    COMMON_FLAGS="$COMMON_FLAGS -march=native "
+  elif test "$USE_SSE"; then
+    COMMON_FLAGS="$COMMON_FLAGS -msse4.2 -mpclmul"
  fi
+elif test "$USE_SSE"; then
+  COMMON_FLAGS="$COMMON_FLAGS -msse4.2 -mpclmul"
 fi

 $CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
@@ -469,6 +505,24 @@ if [ "$?" = 0 ]; then
  COMMON_FLAGS="$COMMON_FLAGS -DHAVE_SSE42"
 elif test "$USE_SSE"; then
  echo "warning: USE_SSE specified but compiler could not use SSE intrinsics, disabling"
+  exit 1
+fi
+
+$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
+  #include <cstdint>
+  #include <wmmintrin.h>
+  int main() {
+    const auto a = _mm_set_epi64x(0, 0);
+    const auto b = _mm_set_epi64x(0, 0);
+    const auto c = _mm_clmulepi64_si128(a, b, 0x00);
+    auto d = _mm_cvtsi128_si64(c);
+  }
+EOF
+if [ "$?" = 0 ]; then
+  COMMON_FLAGS="$COMMON_FLAGS -DHAVE_PCLMUL"
+elif test "$USE_SSE"; then
+  echo "warning: USE_SSE specified but compiler could not use PCLMUL intrinsics, disabling"
+  exit 1
 fi

 # iOS doesn't support thread-local storage, but this check would erroneously
@@ -13,10 +13,12 @@ error=0

 function log {
  DATE=`date +%Y-%m-%d:%H:%M:%S`
+  # shellcheck disable=SC2068
  echo $DATE $@
 }

 function log_err {
+  # shellcheck disable=SC2145
  log "ERROR: $@ Error code: $error."
 }

@@ -1,3 +1,4 @@
+# shellcheck disable=SC2148
 GCC_BASE=/mnt/gvfs/third-party2/gcc/8219ec1bcedf8ad9da05e121e193364de2cc4f61/5.x/centos6-native/c447969
 CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/64d8d58e3d84f8bde7a029763d4f5baf39d0d5b9/stable/centos6-native/6aaf4de
 LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/ba9be983c81de7299b59fe71950c664a84dcb5f8/5.x/gcc-5-glibc-2.23/339d858
@@ -1,3 +1,4 @@
+# shellcheck disable=SC2148
 GCC_BASE=/mnt/gvfs/third-party2/gcc/cf7d14c625ce30bae1a4661c2319c5a283e4dd22/4.8.1/centos6-native/cc6c9dc
 CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/8598c375b0e94e1448182eb3df034704144a838d/stable/centos6-native/3f16ddd
 LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/d6e0a7da6faba45f5e5b1638f9edd7afc2f34e7d/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc
@@ -83,6 +83,7 @@ CFLAGS+=" -DTBB"

 # use Intel SSE support for checksum calculations
 export USE_SSE=1
+export PORTABLE=1

 BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"
@@ -54,6 +54,7 @@ TBB_LIBS="$TBB_BASE/lib/libtbb.a"

 # use Intel SSE support for checksum calculations
 export USE_SSE=1
+export PORTABLE=1

 BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"
@@ -1,3 +1,4 @@
+# shellcheck disable=SC1113
 #/usr/bin/env bash

 set -e
@@ -28,12 +29,14 @@ function package() {
    if dpkg --get-selections | grep --quiet $1; then
      log "$1 is already installed. skipping."
    else
+      # shellcheck disable=SC2068
      apt-get install $@ -y
    fi
  elif [[ $OS = "centos" ]]; then
    if rpm -qa | grep --quiet $1; then
      log "$1 is already installed. skipping."
    else
+      # shellcheck disable=SC2068
      yum install $@ -y
    fi
  fi
@@ -52,6 +55,7 @@ function gem_install() {
  if gem list | grep --quiet $1; then
    log "$1 is already installed. skipping."
  else
+    # shellcheck disable=SC2068
    gem install $@
  fi
 }
@@ -125,4 +129,5 @@ function main() {
    include $LIB_DIR
 }

+# shellcheck disable=SC2068
 main $@
@@ -352,6 +352,22 @@ LITE_BUILD_COMMANDS="[
    }
 ]"

+#
+# Report RocksDB lite binary size to scuba
+REPORT_LITE_BINARY_SIZE_COMMANDS="[
+    {
+        'name':'Rocksdb Lite Binary Size',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Report RocksDB Lite binary size to scuba',
+                'shell':'tools/report_lite_binary_size.sh',
+                'user':'root',
+            },
+        ],
+]"
+
 #
 # RocksDB stress/crash test
 #
@@ -728,6 +744,9 @@ case $1 in
  lite)
    echo $LITE_BUILD_COMMANDS
    ;;
+  report_lite_binary_size)
+    echo $REPORT_LITE_BINARY_SIZE_COMMANDS
+    ;;
  stress_crash)
    echo $STRESS_CRASH_TEST_COMMANDS
    ;;
@@ -1,13 +1,16 @@
 # This script enables you running RocksDB tests by running
 # All the tests concurrently and utilizing all the cores
 Param(
-  [switch]$EnableJE = $false,  # Look for and use _je executable, append _je to listed exclusions
+  [switch]$EnableJE = $false,  # Look for and use test executable, append _je to listed exclusions
  [switch]$RunAll = $false,    # Will attempt discover all *_test[_je].exe binaries and run all
                               # of them as Google suites. I.e. It will run test cases concurrently
                               # except those mentioned as $Run, those will run as individual test cases
                               # And any execlued with $ExcludeExes or $ExcludeCases
                               # It will also not run any individual test cases
                               # excluded but $ExcludeCasese
+  [switch]$RunAllExe = $false, # Look for and use test exdcutables, append _je to exclusions automatically
+                               # It will attempt to run them in parallel w/o breaking them up on individual
+                               # test cases. Those listed with $ExcludeExes will be excluded
  [string]$SuiteRun = "",      # Split test suites in test cases and run in parallel, not compatible with $RunAll
  [string]$Run = "",           # Run specified executables in parallel but do not split to test cases
  [string]$ExcludeCases = "",  # Exclude test cases, expects a comma separated list, no spaces
@@ -39,13 +42,18 @@ $RunOnly.Add("compact_on_deletion_collector_test") | Out-Null
 $RunOnly.Add("merge_test") | Out-Null
 $RunOnly.Add("stringappend_test") | Out-Null # Apparently incorrectly written
 $RunOnly.Add("backupable_db_test") | Out-Null # Disabled
-
+$RunOnly.Add("timer_queue_test") | Out-Null # Not a gtest

 if($RunAll -and $SuiteRun -ne "") {
    Write-Error "$RunAll and $SuiteRun are not compatible"
    exit 1
 }

+if($RunAllExe -and $Run -ne "") {
+    Write-Error "$RunAllExe and $Run are not compatible"
+    exit 1
+}
+
 # If running under Appveyor assume that root
 [string]$Appveyor = $Env:APPVEYOR_BUILD_FOLDER
 if($Appveyor -ne "") {
@@ -131,12 +139,8 @@ function ExtractTestCases([string]$GTestExe, $HashTable) {

      # Leading whitespace is fine
      $l = $l -replace '^\s+',''
-      # but no whitespace any other place
-      if($l -match "\s+") {
-        continue
-      }
      # Trailing dot is a test group but no whitespace
-      elseif ( $l -match "\.$" ) {
+      if ($l -match "\.$" -and $l -notmatch "\s+") {
        $Group = $l
      }  else {
        # Otherwise it is a test name, remove leading space
@@ -223,13 +227,11 @@ $TestExes = [ordered]@{}
 if($Run -ne "") {

  $test_list = $Run -split ' '
-
  ForEach($t in $test_list) {

    if($EnableJE) {
      $t += "_je"
    }
-
    MakeAndAdd -token $t -HashTable $TestExes
  }

@@ -237,6 +239,38 @@ if($Run -ne "") {
     Write-Error "Failed to extract tests from $Run"
     exit 1
  }
+} elseif($RunAllExe) {
+  # Discover all the test binaries
+  if($EnableJE) {
+    $pattern = "*_test_je.exe"
+  } else {
+    $pattern = "*_test.exe"
+  }
+
+  $search_path = -join ($BinariesFolder, $pattern)
+  Write-Host "Binaries Search Path: $search_path"
+
+  $DiscoveredExe = @()
+  dir -Path $search_path | ForEach-Object {
+     $DiscoveredExe += ($_.Name)     
+  }
+
+  # Remove exclusions
+  ForEach($e in $DiscoveredExe) {
+    $e = $e -replace '.exe$', ''
+    $bare_name = $e -replace '_je$', ''
+
+    if($ExcludeExesSet.Contains($bare_name)) {
+      Write-Warning "Test $e is excluded"
+      continue
+    }
+    MakeAndAdd -token $e -HashTable $TestExes
+  }
+
+  if($TestExes.Count -lt 1) {
+     Write-Error "Failed to discover test executables"
+     exit 1
+  }
 }

 # Ordered by exe @{ Exe = @{ TestCase = LogName }}
@@ -245,9 +279,7 @@ $CasesToRun = [ordered]@{}
 if($SuiteRun -ne "") {
  $suite_list = $SuiteRun -split ' '
  ProcessSuites -ListOfSuites $suite_list -HashOfHashes $CasesToRun
-}
-
-if($RunAll) {
+} elseif ($RunAll) {
 # Discover all the test binaries
  if($EnableJE) {
    $pattern = "*_test_je.exe"
@@ -255,7 +287,6 @@ if($RunAll) {
    $pattern = "*_test.exe"
  }

-
  $search_path = -join ($BinariesFolder, $pattern)
  Write-Host "Binaries Search Path: $search_path"

@@ -287,8 +318,6 @@ if($RunAll) {
 }


-Write-Host "Attempting to start: $NumTestsToStart tests"
-
 # Invoke a test with a filter and redirect all output
 $InvokeTestCase = {
    param($exe, $test, $log);
@@ -365,6 +394,7 @@ function RunJobs($Suites, $TestCmds, [int]$ConcurrencyVal)
                 break
               }

+              Write-Host "Starting $exe_name"
              [string]$Exe =  -Join ($BinariesFolder, $exe_name)
              $job = Start-Job -Name $exe_name -ScriptBlock $InvokeTestAsync -ArgumentList @($Exe,$log_path)
              $JobToLog.Add($job, $log_path)
@@ -0,0 +1,43 @@
+#!/bin/bash
+set -e
+
+ROCKSDB_VERSION="5.10.3"
+ZSTD_VERSION="1.1.3"
+
+echo "This script configures CentOS with everything needed to build and run RocksDB"
+
+yum update -y && yum install epel-release -y
+
+yum install -y \
+  wget \
+  gcc-c++ \
+  snappy snappy-devel \
+  zlib zlib-devel \
+  bzip2 bzip2-devel \
+  lz4-devel \
+  libasan \
+  gflags
+
+mkdir -pv /usr/local/rocksdb-${ROCKSDB_VERSION}
+ln -sfT /usr/local/rocksdb-${ROCKSDB_VERSION} /usr/local/rocksdb
+
+wget -qO /tmp/zstd-${ZSTD_VERSION}.tar.gz https://github.com/facebook/zstd/archive/v${ZSTD_VERSION}.tar.gz
+wget -qO /tmp/rocksdb-${ROCKSDB_VERSION}.tar.gz https://github.com/facebook/rocksdb/archive/v${ROCKSDB_VERSION}.tar.gz
+
+cd /tmp
+
+tar xzvf zstd-${ZSTD_VERSION}.tar.gz
+tar xzvf rocksdb-${ROCKSDB_VERSION}.tar.gz -C /usr/local/
+
+echo "Installing ZSTD..."
+pushd zstd-${ZSTD_VERSION}
+make && make install
+popd
+
+echo "Compiling RocksDB..."
+cd /usr/local/rocksdb
+chown -R vagrant:vagrant /usr/local/rocksdb/
+sudo -u vagrant make static_lib
+cd examples/
+sudo -u vagrant make all
+sudo -u vagrant ./c_simple_example
@@ -17,16 +17,16 @@ int main() {
 #include <inttypes.h>
 #include <sys/types.h>
 #include <stdio.h>
-#include <gflags/gflags.h>

-#include "rocksdb/db.h"
-#include "rocksdb/cache.h"
-#include "rocksdb/env.h"
 #include "port/port.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/db.h"
+#include "rocksdb/env.h"
+#include "util/gflags_compat.h"
 #include "util/mutexlock.h"
 #include "util/random.h"

-using GFLAGS::ParseCommandLineFlags;
+using GFLAGS_NAMESPACE::ParseCommandLineFlags;

 static const uint32_t KB = 1024;

@@ -169,6 +169,11 @@ size_t LRUCacheShard::TEST_GetLRUSize() {
  return lru_size;
 }

+double LRUCacheShard::GetHighPriPoolRatio() {
+  MutexLock l(&mutex_);
+  return high_pri_pool_ratio_;
+}
+
 void LRUCacheShard::LRU_Remove(LRUHandle* e) {
  assert(e->next != nullptr);
  assert(e->prev != nullptr);
@@ -513,6 +518,20 @@ size_t LRUCache::TEST_GetLRUSize() {
  return lru_size_of_all_shards;
 }

+double LRUCache::GetHighPriPoolRatio() {
+  double result = 0.0;
+  if (num_shards_ > 0) {
+    result = shards_[0].GetHighPriPoolRatio();
+  }
+  return result;
+}
+
+std::shared_ptr<Cache> NewLRUCache(const LRUCacheOptions& cache_opts) {
+  return NewLRUCache(cache_opts.capacity, cache_opts.num_shard_bits,
+                     cache_opts.strict_capacity_limit,
+                     cache_opts.high_pri_pool_ratio);
+}
+
 std::shared_ptr<Cache> NewLRUCache(size_t capacity, int num_shard_bits,
                                   bool strict_capacity_limit,
                                   double high_pri_pool_ratio) {
@@ -57,7 +57,7 @@ struct LRUHandle {
  // Include the following flags:
  //   in_cache:    whether this entry is referenced by the hash table.
  //   is_high_pri: whether this entry is high priority entry.
-  //   in_high_pro_pool: whether this entry is in high-pri pool.
+  //   in_high_pri_pool: whether this entry is in high-pri pool.
  char flags;

  uint32_t hash;     // Hash of key(); used for fast sharding and comparisons
@@ -202,6 +202,9 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard : public CacheShard {
  //  not threadsafe
  size_t TEST_GetLRUSize();

+  //  Retrives high pri pool ratio
+  double GetHighPriPoolRatio();
+
  // Overloading to aligned it to cache line size
  void* operator new(size_t);

@@ -293,6 +296,8 @@ class LRUCache : public ShardedCache {

  //  Retrieves number of elements in LRU, for unit test purpose only
  size_t TEST_GetLRUSize();
+  //  Retrives high pri pool ratio
+  double GetHighPriPoolRatio();

 private:
  LRUCacheShard* shards_;
@@ -0,0 +1,21 @@
+# - Find NUMA
+# Find the NUMA library and includes
+#
+# NUMA_INCLUDE_DIR - where to find numa.h, etc.
+# NUMA_LIBRARIES - List of libraries when using NUMA.
+# NUMA_FOUND - True if NUMA found.
+
+find_path(NUMA_INCLUDE_DIR
+  NAMES numa.h numaif.h
+  HINTS ${NUMA_ROOT_DIR}/include)
+
+find_library(NUMA_LIBRARIES
+  NAMES numa
+  HINTS ${NUMA_ROOT_DIR}/lib)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(NUMA DEFAULT_MSG NUMA_LIBRARIES NUMA_INCLUDE_DIR)
+
+mark_as_advanced(
+  NUMA_LIBRARIES
+  NUMA_INCLUDE_DIR)
@@ -0,0 +1,21 @@
+# - Find TBB
+# Find the Thread Building Blocks library and includes
+#
+# TBB_INCLUDE_DIR - where to find tbb.h, etc.
+# TBB_LIBRARIES - List of libraries when using TBB.
+# TBB_FOUND - True if TBB found.
+
+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)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(TBB DEFAULT_MSG TBB_LIBRARIES TBB_INCLUDE_DIR)
+
+mark_as_advanced(
+TBB_LIBRARIES
+TBB_INCLUDE_DIR)
@@ -72,7 +72,7 @@ def display_file_coverage(per_file_coverage, total_coverage):
    header_template = \
        "%" + str(max_file_name_length) + "s\t%s\t%s"
    separator = "-" * (max_file_name_length + 10 + 20)
-    print header_template % ("Filename", "Coverage", "Lines")
+    print header_template % ("Filename", "Coverage", "Lines")  # noqa: E999 T25377293 Grandfathered in
    print separator

    # -- Print body
@@ -75,7 +75,7 @@ Status BuildTable(
    InternalStats* internal_stats, TableFileCreationReason reason,
    EventLogger* event_logger, int job_id, const Env::IOPriority io_priority,
    TableProperties* table_properties, int level, const uint64_t creation_time,
-    const uint64_t oldest_key_time) {
+    const uint64_t oldest_key_time, Env::WriteLifeTimeHint write_hint) {
  assert((column_family_id ==
          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
         column_family_name.empty());
@@ -117,6 +117,7 @@ Status BuildTable(
        return s;
      }
      file->SetIOPriority(io_priority);
+      file->SetWriteLifeTimeHint(write_hint);

      file_writer.reset(new WritableFileWriter(std::move(file), env_options,
                                               ioptions.statistics));
@@ -130,7 +131,8 @@ Status BuildTable(
    MergeHelper merge(env, internal_comparator.user_comparator(),
                      ioptions.merge_operator, nullptr, ioptions.info_log,
                      true /* internal key corruption is not ok */,
-                      snapshots.empty() ? 0 : snapshots.back());
+                      snapshots.empty() ? 0 : snapshots.back(),
+                      snapshot_checker);

    CompactionIterator c_iter(
        iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
@@ -79,6 +79,7 @@ extern Status BuildTable(
    EventLogger* event_logger = nullptr, int job_id = 0,
    const Env::IOPriority io_priority = Env::IO_HIGH,
    TableProperties* table_properties = nullptr, int level = -1,
-    const uint64_t creation_time = 0, const uint64_t oldest_key_time = 0);
+    const uint64_t creation_time = 0, const uint64_t oldest_key_time = 0,
+    Env::WriteLifeTimeHint write_hint = Env::WLTH_NOT_SET);

 }  // namespace rocksdb
@@ -117,7 +117,9 @@ struct rocksdb_flushoptions_t    { FlushOptions      rep; };
 struct rocksdb_fifo_compaction_options_t { CompactionOptionsFIFO rep; };
 struct rocksdb_readoptions_t {
   ReadOptions rep;
-   Slice upper_bound; // stack variable to set pointer to in ReadOptions
+   // stack variables to set pointers to in ReadOptions
+   Slice upper_bound;
+   Slice lower_bound;
 };
 struct rocksdb_writeoptions_t    { WriteOptions      rep; };
 struct rocksdb_options_t         { Options           rep; };
@@ -1824,6 +1826,26 @@ void rocksdb_block_based_options_set_block_restart_interval(
  options->rep.block_restart_interval = block_restart_interval;
 }

+void rocksdb_block_based_options_set_index_block_restart_interval(
+    rocksdb_block_based_table_options_t* options, int index_block_restart_interval) {
+  options->rep.index_block_restart_interval = index_block_restart_interval;
+}
+
+void rocksdb_block_based_options_set_metadata_block_size(
+    rocksdb_block_based_table_options_t* options, uint64_t metadata_block_size) {
+  options->rep.metadata_block_size = metadata_block_size;
+}
+
+void rocksdb_block_based_options_set_partition_filters(
+    rocksdb_block_based_table_options_t* options, unsigned char partition_filters) {
+  options->rep.partition_filters = partition_filters;
+}
+
+void rocksdb_block_based_options_set_use_delta_encoding(
+    rocksdb_block_based_table_options_t* options, unsigned char use_delta_encoding) {
+  options->rep.use_delta_encoding = use_delta_encoding;
+}
+
 void rocksdb_block_based_options_set_filter_policy(
    rocksdb_block_based_table_options_t* options,
    rocksdb_filterpolicy_t* filter_policy) {
@@ -1877,6 +1899,11 @@ void rocksdb_block_based_options_set_cache_index_and_filter_blocks(
  options->rep.cache_index_and_filter_blocks = v;
 }

+void rocksdb_block_based_options_set_cache_index_and_filter_blocks_with_high_priority(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.cache_index_and_filter_blocks_with_high_priority = v;
+}
+
 void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
    rocksdb_block_based_table_options_t* options, unsigned char v) {
  options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
@@ -1891,7 +1918,6 @@ void rocksdb_options_set_block_based_table_factory(
  }
 }

-
 rocksdb_cuckoo_table_options_t*
 rocksdb_cuckoo_options_create() {
  return new rocksdb_cuckoo_table_options_t;
@@ -1973,6 +1999,11 @@ void rocksdb_options_optimize_universal_style_compaction(
  opt->rep.OptimizeUniversalStyleCompaction(memtable_memory_budget);
 }

+void rocksdb_options_set_allow_ingest_behind(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.allow_ingest_behind = v;
+}
+
 void rocksdb_options_set_compaction_filter(
    rocksdb_options_t* opt,
    rocksdb_compactionfilter_t* filter) {
@@ -2728,6 +2759,18 @@ void rocksdb_readoptions_set_iterate_upper_bound(
  }
 }

+void rocksdb_readoptions_set_iterate_lower_bound(
+    rocksdb_readoptions_t *opt,
+    const char* key, size_t keylen) {
+  if (key == nullptr) {
+    opt->lower_bound = Slice();
+    opt->rep.iterate_lower_bound = nullptr;
+  } else {
+    opt->lower_bound = Slice(key, keylen);
+    opt->rep.iterate_lower_bound = &opt->lower_bound;
+  }
+}
+
 void rocksdb_readoptions_set_read_tier(
    rocksdb_readoptions_t* opt, int v) {
  opt->rep.read_tier = static_cast<rocksdb::ReadTier>(v);
@@ -2738,11 +2781,21 @@ void rocksdb_readoptions_set_tailing(
  opt->rep.tailing = v;
 }

+void rocksdb_readoptions_set_managed(
+    rocksdb_readoptions_t* opt, unsigned char v) {
+  opt->rep.managed = v;
+}
+
 void rocksdb_readoptions_set_readahead_size(
    rocksdb_readoptions_t* opt, size_t v) {
  opt->rep.readahead_size = v;
 }

+void rocksdb_readoptions_set_prefix_same_as_start(
+    rocksdb_readoptions_t* opt, unsigned char v) {
+  opt->rep.prefix_same_as_start = v;
+}
+
 void rocksdb_readoptions_set_pin_data(rocksdb_readoptions_t* opt,
                                      unsigned char v) {
  opt->rep.pin_data = v;
@@ -2753,6 +2806,22 @@ void rocksdb_readoptions_set_total_order_seek(rocksdb_readoptions_t* opt,
  opt->rep.total_order_seek = v;
 }

+void rocksdb_readoptions_set_max_skippable_internal_keys(
+    rocksdb_readoptions_t* opt,
+    uint64_t v) {
+  opt->rep.max_skippable_internal_keys = v;
+}
+
+void rocksdb_readoptions_set_background_purge_on_iterator_cleanup(
+    rocksdb_readoptions_t* opt, unsigned char v) {
+  opt->rep.background_purge_on_iterator_cleanup = v;
+}
+
+void rocksdb_readoptions_set_ignore_range_deletions(
+    rocksdb_readoptions_t* opt, unsigned char v) {
+  opt->rep.ignore_range_deletions = v;
+}
+
 rocksdb_writeoptions_t* rocksdb_writeoptions_create() {
  return new rocksdb_writeoptions_t;
 }
@@ -2770,6 +2839,24 @@ void rocksdb_writeoptions_disable_WAL(rocksdb_writeoptions_t* opt, int disable)
  opt->rep.disableWAL = disable;
 }

+void rocksdb_writeoptions_set_ignore_missing_column_families(
+    rocksdb_writeoptions_t* opt,
+    unsigned char v) {
+  opt->rep.ignore_missing_column_families = v;
+}
+
+void rocksdb_writeoptions_set_no_slowdown(
+    rocksdb_writeoptions_t* opt,
+    unsigned char v) {
+  opt->rep.no_slowdown = v;
+}
+
+void rocksdb_writeoptions_set_low_pri(
+    rocksdb_writeoptions_t* opt,
+    unsigned char v) {
+  opt->rep.low_pri = v;
+}
+
 rocksdb_compactoptions_t* rocksdb_compactoptions_create() {
  return new rocksdb_compactoptions_t;
 }
@@ -2962,6 +3049,12 @@ void rocksdb_ingestexternalfileoptions_set_allow_blocking_flush(
  opt->rep.allow_blocking_flush = allow_blocking_flush;
 }

+void rocksdb_ingestexternalfileoptions_set_ingest_behind(
+    rocksdb_ingestexternalfileoptions_t* opt,
+    unsigned char ingest_behind) {
+  opt->rep.ingest_behind = ingest_behind;
+}
+
 void rocksdb_ingestexternalfileoptions_destroy(
    rocksdb_ingestexternalfileoptions_t* opt) {
  delete opt;
@@ -31,6 +31,7 @@
 #include "monitoring/thread_status_util.h"
 #include "options/options_helper.h"
 #include "table/block_based_table_factory.h"
+#include "table/merging_iterator.h"
 #include "util/autovector.h"
 #include "util/compression.h"

@@ -123,6 +124,22 @@ Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
          " is not linked with the binary.");
    }
  }
+  if (cf_options.compression_opts.zstd_max_train_bytes > 0) {
+    if (!CompressionTypeSupported(CompressionType::kZSTD)) {
+      // Dictionary trainer is available since v0.6.1, but ZSTD was marked
+      // stable only since v0.8.0. For now we enable the feature in stable
+      // versions only.
+      return Status::InvalidArgument(
+          "zstd dictionary trainer cannot be used because " +
+          CompressionTypeToString(CompressionType::kZSTD) +
+          " is not linked with the binary.");
+    }
+    if (cf_options.compression_opts.max_dict_bytes == 0) {
+      return Status::InvalidArgument(
+          "The dictionary size limit (`CompressionOptions::max_dict_bytes`) "
+          "should be nonzero if we're using zstd's dictionary generator.");
+    }
+  }
  return Status::OK();
 }

@@ -328,12 +345,13 @@ void SuperVersionUnrefHandle(void* ptr) {
  // When latter happens, we are in ~ColumnFamilyData(), no get should happen as
  // well.
  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
-  if (sv->Unref()) {
-    sv->db_mutex->Lock();
-    sv->Cleanup();
-    sv->db_mutex->Unlock();
-    delete sv;
-  }
+  bool was_last_ref __attribute__((__unused__));
+  was_last_ref = sv->Unref();
+  // Thread-local SuperVersions can't outlive ColumnFamilyData::super_version_.
+  // This is important because we can't do SuperVersion cleanup here.
+  // That would require locking DB mutex, which would deadlock because
+  // SuperVersionUnrefHandle is called with locked ThreadLocalPtr mutex.
+  assert(!was_last_ref);
 }
 }  // anonymous namespace

@@ -365,11 +383,13 @@ ColumnFamilyData::ColumnFamilyData(
      next_(nullptr),
      prev_(nullptr),
      log_number_(0),
+      flush_reason_(FlushReason::kUnknown),
      column_family_set_(column_family_set),
      pending_flush_(false),
      pending_compaction_(false),
      prev_compaction_needed_bytes_(0),
-      allow_2pc_(db_options.allow_2pc) {
+      allow_2pc_(db_options.allow_2pc),
+      last_memtable_id_(0) {
  Ref();

  // Convert user defined table properties collector factories to internal ones.
@@ -452,7 +472,7 @@ ColumnFamilyData::~ColumnFamilyData() {
    local_sv_.reset();
    super_version_->db_mutex->Lock();

-    bool is_last_reference __attribute__((unused));
+    bool is_last_reference __attribute__((__unused__));
    is_last_reference = super_version_->Unref();
    assert(is_last_reference);
    super_version_->Cleanup();
@@ -463,7 +483,7 @@ ColumnFamilyData::~ColumnFamilyData() {
  if (dummy_versions_ != nullptr) {
    // List must be empty
    assert(dummy_versions_->TEST_Next() == dummy_versions_);
-    bool deleted __attribute__((unused));
+    bool deleted __attribute__((__unused__));
    deleted = dummy_versions_->Unref();
    assert(deleted);
  }
@@ -614,6 +634,41 @@ int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
 }
 }  // namespace

+std::pair<WriteStallCondition, ColumnFamilyData::WriteStallCause>
+ColumnFamilyData::GetWriteStallConditionAndCause(
+    int num_unflushed_memtables, int num_l0_files,
+    uint64_t num_compaction_needed_bytes,
+    const MutableCFOptions& mutable_cf_options) {
+  if (num_unflushed_memtables >= mutable_cf_options.max_write_buffer_number) {
+    return {WriteStallCondition::kStopped, WriteStallCause::kMemtableLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             num_l0_files >= mutable_cf_options.level0_stop_writes_trigger) {
+    return {WriteStallCondition::kStopped, WriteStallCause::kL0FileCountLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
+             num_compaction_needed_bytes >=
+                 mutable_cf_options.hard_pending_compaction_bytes_limit) {
+    return {WriteStallCondition::kStopped,
+            WriteStallCause::kPendingCompactionBytes};
+  } else if (mutable_cf_options.max_write_buffer_number > 3 &&
+             num_unflushed_memtables >=
+                 mutable_cf_options.max_write_buffer_number - 1) {
+    return {WriteStallCondition::kDelayed, WriteStallCause::kMemtableLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
+             num_l0_files >=
+                 mutable_cf_options.level0_slowdown_writes_trigger) {
+    return {WriteStallCondition::kDelayed, WriteStallCause::kL0FileCountLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
+             num_compaction_needed_bytes >=
+                 mutable_cf_options.soft_pending_compaction_bytes_limit) {
+    return {WriteStallCondition::kDelayed,
+            WriteStallCause::kPendingCompactionBytes};
+  }
+  return {WriteStallCondition::kNormal, WriteStallCause::kNone};
+}
+
 WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
      const MutableCFOptions& mutable_cf_options) {
  auto write_stall_condition = WriteStallCondition::kNormal;
@@ -623,25 +678,29 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
    uint64_t compaction_needed_bytes =
        vstorage->estimated_compaction_needed_bytes();

+    auto write_stall_condition_and_cause = GetWriteStallConditionAndCause(
+        imm()->NumNotFlushed(), vstorage->l0_delay_trigger_count(),
+        vstorage->estimated_compaction_needed_bytes(), mutable_cf_options);
+    write_stall_condition = write_stall_condition_and_cause.first;
+    auto write_stall_cause = write_stall_condition_and_cause.second;
+
    bool was_stopped = write_controller->IsStopped();
    bool needed_delay = write_controller->NeedsDelay();

-    if (imm()->NumNotFlushed() >= mutable_cf_options.max_write_buffer_number) {
+    if (write_stall_condition == WriteStallCondition::kStopped &&
+        write_stall_cause == WriteStallCause::kMemtableLimit) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stopping writes because we have %d immutable memtables "
          "(waiting for flush), max_write_buffer_number is set to %d",
          name_.c_str(), imm()->NumNotFlushed(),
          mutable_cf_options.max_write_buffer_number);
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               vstorage->l0_delay_trigger_count() >=
-                   mutable_cf_options.level0_stop_writes_trigger) {
+    } else if (write_stall_condition == WriteStallCondition::kStopped &&
+               write_stall_cause == WriteStallCause::kL0FileCountLimit) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::L0_FILE_COUNT_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
        internal_stats_->AddCFStats(
            InternalStats::LOCKED_L0_FILE_COUNT_LIMIT_STOPS, 1);
@@ -649,28 +708,23 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
      ROCKS_LOG_WARN(ioptions_.info_log,
                     "[%s] Stopping writes because we have %d level-0 files",
                     name_.c_str(), vstorage->l0_delay_trigger_count());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
-               compaction_needed_bytes >=
-                   mutable_cf_options.hard_pending_compaction_bytes_limit) {
+    } else if (write_stall_condition == WriteStallCondition::kStopped &&
+               write_stall_cause == WriteStallCause::kPendingCompactionBytes) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(
          InternalStats::PENDING_COMPACTION_BYTES_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stopping writes because of estimated pending compaction "
          "bytes %" PRIu64,
          name_.c_str(), compaction_needed_bytes);
-    } else if (mutable_cf_options.max_write_buffer_number > 3 &&
-               imm()->NumNotFlushed() >=
-                   mutable_cf_options.max_write_buffer_number - 1) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kMemtableLimit) {
      write_controller_token_ =
          SetupDelay(write_controller, compaction_needed_bytes,
                     prev_compaction_needed_bytes_, was_stopped,
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_LIMIT_SLOWDOWNS, 1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stalling writes because we have %d immutable memtables "
@@ -679,10 +733,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
          name_.c_str(), imm()->NumNotFlushed(),
          mutable_cf_options.max_write_buffer_number,
          write_controller->delayed_write_rate());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
-               vstorage->l0_delay_trigger_count() >=
-                   mutable_cf_options.level0_slowdown_writes_trigger) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kL0FileCountLimit) {
      // L0 is the last two files from stopping.
      bool near_stop = vstorage->l0_delay_trigger_count() >=
                       mutable_cf_options.level0_stop_writes_trigger - 2;
@@ -692,7 +744,6 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::L0_FILE_COUNT_LIMIT_SLOWDOWNS,
                                  1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
        internal_stats_->AddCFStats(
            InternalStats::LOCKED_L0_FILE_COUNT_LIMIT_SLOWDOWNS, 1);
@@ -702,10 +753,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     "rate %" PRIu64,
                     name_.c_str(), vstorage->l0_delay_trigger_count(),
                     write_controller->delayed_write_rate());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
-               vstorage->estimated_compaction_needed_bytes() >=
-                   mutable_cf_options.soft_pending_compaction_bytes_limit) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kPendingCompactionBytes) {
      // If the distance to hard limit is less than 1/4 of the gap between soft
      // and
      // hard bytes limit, we think it is near stop and speed up the slowdown.
@@ -723,7 +772,6 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(
          InternalStats::PENDING_COMPACTION_BYTES_LIMIT_SLOWDOWNS, 1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stalling writes because of estimated pending compaction "
@@ -731,6 +779,7 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
          name_.c_str(), vstorage->estimated_compaction_needed_bytes(),
          write_controller->delayed_write_rate());
    } else {
+      assert(write_stall_condition == WriteStallCondition::kNormal);
      if (vstorage->l0_delay_trigger_count() >=
          GetL0ThresholdSpeedupCompaction(
              mutable_cf_options.level0_file_num_compaction_trigger,
@@ -833,6 +882,68 @@ bool ColumnFamilyData::RangeOverlapWithCompaction(
      smallest_user_key, largest_user_key, level);
 }

+Status ColumnFamilyData::RangesOverlapWithMemtables(
+    const autovector<Range>& ranges, SuperVersion* super_version,
+    bool* overlap) {
+  assert(overlap != nullptr);
+  *overlap = false;
+  // Create an InternalIterator over all unflushed memtables
+  Arena arena;
+  ReadOptions read_opts;
+  read_opts.total_order_seek = true;
+  MergeIteratorBuilder merge_iter_builder(&internal_comparator_, &arena);
+  merge_iter_builder.AddIterator(
+      super_version->mem->NewIterator(read_opts, &arena));
+  super_version->imm->AddIterators(read_opts, &merge_iter_builder);
+  ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
+
+  std::vector<InternalIterator*> memtable_range_del_iters;
+  auto* active_range_del_iter =
+      super_version->mem->NewRangeTombstoneIterator(read_opts);
+  if (active_range_del_iter != nullptr) {
+    memtable_range_del_iters.push_back(active_range_del_iter);
+  }
+  super_version->imm->AddRangeTombstoneIterators(read_opts,
+                                                 &memtable_range_del_iters);
+  RangeDelAggregator range_del_agg(internal_comparator_, {} /* snapshots */,
+                                   false /* collapse_deletions */);
+  Status status;
+  {
+    std::unique_ptr<InternalIterator> memtable_range_del_iter(
+        NewMergingIterator(&internal_comparator_,
+                           memtable_range_del_iters.empty()
+                               ? nullptr
+                               : &memtable_range_del_iters[0],
+                           static_cast<int>(memtable_range_del_iters.size())));
+    status = range_del_agg.AddTombstones(std::move(memtable_range_del_iter));
+  }
+  for (size_t i = 0; i < ranges.size() && status.ok() && !*overlap; ++i) {
+    auto* vstorage = super_version->current->storage_info();
+    auto* ucmp = vstorage->InternalComparator()->user_comparator();
+    InternalKey range_start(ranges[i].start, kMaxSequenceNumber,
+                            kValueTypeForSeek);
+    memtable_iter->Seek(range_start.Encode());
+    status = memtable_iter->status();
+    ParsedInternalKey seek_result;
+    if (status.ok()) {
+      if (memtable_iter->Valid() &&
+          !ParseInternalKey(memtable_iter->key(), &seek_result)) {
+        status = Status::Corruption("DB have corrupted keys");
+      }
+    }
+    if (status.ok()) {
+      if (memtable_iter->Valid() &&
+          ucmp->Compare(seek_result.user_key, ranges[i].limit) <= 0) {
+        *overlap = true;
+      } else if (range_del_agg.IsRangeOverlapped(ranges[i].start,
+                                                 ranges[i].limit)) {
+        *overlap = true;
+      }
+    }
+  }
+  return status;
+}
+
 const int ColumnFamilyData::kCompactAllLevels = -1;
 const int ColumnFamilyData::kCompactToBaseLevel = -2;

@@ -851,8 +962,7 @@ Compaction* ColumnFamilyData::CompactRange(

 SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(
    InstrumentedMutex* db_mutex) {
-  SuperVersion* sv = nullptr;
-  sv = GetThreadLocalSuperVersion(db_mutex);
+  SuperVersion* sv = GetThreadLocalSuperVersion(db_mutex);
  sv->Ref();
  if (!ReturnThreadLocalSuperVersion(sv)) {
    // This Unref() corresponds to the Ref() in GetThreadLocalSuperVersion()
@@ -866,7 +976,6 @@ SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(

 SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
    InstrumentedMutex* db_mutex) {
-  SuperVersion* sv = nullptr;
  // The SuperVersion is cached in thread local storage to avoid acquiring
  // mutex when SuperVersion does not change since the last use. When a new
  // SuperVersion is installed, the compaction or flush thread cleans up
@@ -885,7 +994,7 @@ SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
  // should only keep kSVInUse before ReturnThreadLocalSuperVersion call
  // (if no Scrape happens).
  assert(ptr != SuperVersion::kSVInUse);
-  sv = static_cast<SuperVersion*>(ptr);
+  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
  if (sv == SuperVersion::kSVObsolete ||
      sv->version_number != super_version_number_.load()) {
    RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_ACQUIRES);
@@ -949,6 +1058,16 @@ void ColumnFamilyData::InstallSuperVersion(
      RecalculateWriteStallConditions(mutable_cf_options);

  if (old_superversion != nullptr) {
+    // Reset SuperVersions cached in thread local storage.
+    // This should be done before old_superversion->Unref(). That's to ensure
+    // that local_sv_ never holds the last reference to SuperVersion, since
+    // it has no means to safely do SuperVersion cleanup.
+    ResetThreadLocalSuperVersions();
+
+    if (old_superversion->mutable_cf_options.write_buffer_size !=
+        mutable_cf_options.write_buffer_size) {
+      mem_->UpdateWriteBufferSize(mutable_cf_options.write_buffer_size);
+    }
    if (old_superversion->write_stall_condition !=
        new_superversion->write_stall_condition) {
      sv_context->PushWriteStallNotification(
@@ -960,9 +1079,6 @@ void ColumnFamilyData::InstallSuperVersion(
      sv_context->superversions_to_free.push_back(old_superversion);
    }
  }
-
-  // Reset SuperVersions cached in thread local storage
-  ResetThreadLocalSuperVersions();
 }

 void ColumnFamilyData::ResetThreadLocalSuperVersions() {
@@ -974,10 +1090,12 @@ void ColumnFamilyData::ResetThreadLocalSuperVersions() {
      continue;
    }
    auto sv = static_cast<SuperVersion*>(ptr);
-    if (sv->Unref()) {
-      sv->Cleanup();
-      delete sv;
-    }
+    bool was_last_ref __attribute__((__unused__));
+    was_last_ref = sv->Unref();
+    // sv couldn't have been the last reference because
+    // ResetThreadLocalSuperVersions() is called before
+    // unref'ing super_version_.
+    assert(!was_last_ref);
  }
 }

@@ -995,6 +1113,24 @@ Status ColumnFamilyData::SetOptions(
 }
 #endif  // ROCKSDB_LITE

+// REQUIRES: DB mutex held
+Env::WriteLifeTimeHint ColumnFamilyData::CalculateSSTWriteHint(int level) {
+  if (initial_cf_options_.compaction_style != kCompactionStyleLevel) {
+    return Env::WLTH_NOT_SET;
+  }
+  if (level == 0) {
+    return Env::WLTH_MEDIUM;
+  }
+  int base_level = current_->storage_info()->base_level();
+
+  // L1: medium, L2: long, ...
+  if (level - base_level >= 2) {
+    return Env::WLTH_EXTREME;
+  }
+  return static_cast<Env::WriteLifeTimeHint>(level - base_level +
+                            static_cast<int>(Env::WLTH_MEDIUM));
+}
+
 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                 const ImmutableDBOptions* db_options,
                                 const EnvOptions& env_options,
@@ -1021,10 +1157,14 @@ ColumnFamilySet::~ColumnFamilySet() {
  while (column_family_data_.size() > 0) {
    // cfd destructor will delete itself from column_family_data_
    auto cfd = column_family_data_.begin()->second;
-    cfd->Unref();
+    bool last_ref __attribute__((__unused__));
+    last_ref = cfd->Unref();
+    assert(last_ref);
    delete cfd;
  }
-  dummy_cfd_->Unref();
+  bool dummy_last_ref __attribute__((__unused__));
+  dummy_last_ref = dummy_cfd_->Unref();
+  assert(dummy_last_ref);
  delete dummy_cfd_;
 }

@@ -30,6 +30,7 @@ namespace rocksdb {

 class Version;
 class VersionSet;
+class VersionStorageInfo;
 class MemTable;
 class MemTableListVersion;
 class CompactionPicker;
@@ -194,7 +195,7 @@ class ColumnFamilyData {
  // *) delete all memory associated with that column family
  // *) delete all the files associated with that column family
  void SetDropped();
-  bool IsDropped() const { return dropped_; }
+  bool IsDropped() const { return dropped_.load(std::memory_order_relaxed); }

  // thread-safe
  int NumberLevels() const { return ioptions_.num_levels; }
@@ -202,6 +203,10 @@ class ColumnFamilyData {
  void SetLogNumber(uint64_t log_number) { log_number_ = log_number; }
  uint64_t GetLogNumber() const { return log_number_; }

+  void SetFlushReason(FlushReason flush_reason) {
+    flush_reason_ = flush_reason;
+  }
+  FlushReason GetFlushReason() const { return flush_reason_; }
  // thread-safe
  const EnvOptions* soptions() const;
  const ImmutableCFOptions* ioptions() const { return &ioptions_; }
@@ -239,7 +244,11 @@ class ColumnFamilyData {
  void SetCurrent(Version* _current);
  uint64_t GetNumLiveVersions() const;  // REQUIRE: DB mutex held
  uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
-  void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }
+  void SetMemtable(MemTable* new_mem) {
+    uint64_t memtable_id = last_memtable_id_.fetch_add(1) + 1;
+    new_mem->SetID(memtable_id);
+    mem_ = new_mem;
+  }

  // calculate the oldest log needed for the durability of this column family
  uint64_t OldestLogToKeep();
@@ -265,6 +274,16 @@ class ColumnFamilyData {
                                  const Slice& largest_user_key,
                                  int level) const;

+  // Check if the passed ranges overlap with any unflushed memtables
+  // (immutable or mutable).
+  //
+  // @param super_version A referenced SuperVersion that will be held for the
+  //    duration of this function.
+  //
+  // Thread-safe
+  Status RangesOverlapWithMemtables(const autovector<Range>& ranges,
+                                    SuperVersion* super_version, bool* overlap);
+
  // A flag to tell a manual compaction is to compact all levels together
  // instead of a specific level.
  static const int kCompactAllLevels;
@@ -327,6 +346,17 @@ class ColumnFamilyData {
  bool pending_flush() { return pending_flush_; }
  bool pending_compaction() { return pending_compaction_; }

+  enum class WriteStallCause {
+    kNone,
+    kMemtableLimit,
+    kL0FileCountLimit,
+    kPendingCompactionBytes,
+  };
+  static std::pair<WriteStallCondition, WriteStallCause>
+  GetWriteStallConditionAndCause(int num_unflushed_memtables, int num_l0_files,
+                                 uint64_t num_compaction_needed_bytes,
+                                 const MutableCFOptions& mutable_cf_options);
+
  // Recalculate some small conditions, which are changed only during
  // compaction, adding new memtable and/or
  // recalculation of compaction score. These values are used in
@@ -339,6 +369,8 @@ class ColumnFamilyData {

  bool initialized() const { return initialized_.load(); }

+  Env::WriteLifeTimeHint CalculateSSTWriteHint(int level);
+
 private:
  friend class ColumnFamilySet;
  ColumnFamilyData(uint32_t id, const std::string& name,
@@ -356,7 +388,7 @@ class ColumnFamilyData {

  std::atomic<int> refs_;      // outstanding references to ColumnFamilyData
  std::atomic<bool> initialized_;
-  bool dropped_;               // true if client dropped it
+  std::atomic<bool> dropped_;  // true if client dropped it

  const InternalKeyComparator internal_comparator_;
  std::vector<std::unique_ptr<IntTblPropCollectorFactory>>
@@ -398,6 +430,8 @@ class ColumnFamilyData {
  // recovered from
  uint64_t log_number_;

+  std::atomic<FlushReason> flush_reason_;
+
  // An object that keeps all the compaction stats
  // and picks the next compaction
  std::unique_ptr<CompactionPicker> compaction_picker_;
@@ -417,6 +451,9 @@ class ColumnFamilyData {

  // if the database was opened with 2pc enabled
  bool allow_2pc_;
+
+  // Memtable id to track flush.
+  std::atomic<uint64_t> last_memtable_id_;
 };

 // ColumnFamilySet has interesting thread-safety requirements
@@ -1868,126 +1868,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  }
 }

-// This test checks for automatic getting a conflict if there is a
-// manual which has not yet been scheduled.
-// The manual compaction waits in NotScheduled
-// We generate more files and then trigger an automatic compaction
-// This will wait because there is an unscheduled manual compaction.
-// Once the conflict is hit, the manual compaction starts and ends
-// Then another automatic will start and end.
-TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
-  Open();
-  CreateColumnFamilies({"one"});
-  ColumnFamilyOptions default_cf, one;
-  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.max_background_compactions = 3;
-
-  default_cf.compaction_style = kCompactionStyleLevel;
-  default_cf.num_levels = 3;
-  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;
-  table_options.no_block_cache = true;
-  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
-
-  one.compaction_style = kCompactionStyleUniversal;
-
-  one.num_levels = 1;
-  // trigger compaction if there are >= 4 files
-  one.level0_file_num_compaction_trigger = 4;
-  one.write_buffer_size = 120000;
-
-  Reopen({default_cf, one});
-  // make sure all background compaction jobs can be scheduled
-  auto stop_token =
-      dbfull()->TEST_write_controler().GetCompactionPressureToken();
-
-  // SETUP column family "one" -- universal style
-  for (int i = 0; i < one.level0_file_num_compaction_trigger - 2; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-    AssertFilesPerLevel(ToString(i + 1), 1);
-  }
-  bool cf_1_1 = true;
-  bool cf_1_2 = true;
-  rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      {{"DBImpl::BackgroundCompaction()::Conflict",
-        "ColumnFamilyTest::ManualAutoCon:7"},
-       {"ColumnFamilyTest::ManualAutoCon:9",
-        "ColumnFamilyTest::ManualAutoCon:8"},
-       {"ColumnFamilyTest::ManualAutoCon:2",
-        "ColumnFamilyTest::ManualAutoCon:6"},
-       {"ColumnFamilyTest::ManualAutoCon:4",
-        "ColumnFamilyTest::ManualAutoCon:5"},
-       {"ColumnFamilyTest::ManualAutoCon:1",
-        "ColumnFamilyTest::ManualAutoCon:2"},
-       {"ColumnFamilyTest::ManualAutoCon:1",
-        "ColumnFamilyTest::ManualAutoCon:3"}});
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
-        if (cf_1_1) {
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:4");
-          cf_1_1 = false;
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:3");
-        } else if (cf_1_2) {
-          cf_1_2 = false;
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:2");
-        }
-      });
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::RunManualCompaction:NotScheduled", [&](void* arg) {
-        InstrumentedMutex* mutex = static_cast<InstrumentedMutex*>(arg);
-        mutex->Unlock();
-        TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:9");
-        TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:7");
-        mutex->Lock();
-      });
-
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  rocksdb::port::Thread threads([&] {
-    CompactRangeOptions compact_options;
-    compact_options.exclusive_manual_compaction = false;
-    ASSERT_OK(
-        db_->CompactRange(compact_options, handles_[1], nullptr, nullptr));
-    TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:6");
-  });
-
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:8");
-  WaitForFlush(1);
-
-  // Add more L0 files and force automatic compaction
-  for (int i = 0; i < one.level0_file_num_compaction_trigger; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-    AssertFilesPerLevel(ToString(one.level0_file_num_compaction_trigger + i),
-                        1);
-  }
-
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:5");
-  // Add more L0 files and force automatic compaction
-  for (int i = 0; i < one.level0_file_num_compaction_trigger; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-  }
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:1");
-
-  threads.join();
-  WaitForCompaction();
-  // VERIFY compaction "one"
-  ASSERT_LE(NumTableFilesAtLevel(0, 1), 3);
-
-  // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
-    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
-    key_iter++;
-  }
-}
-
 // In this test, we generate enough files to trigger automatic compactions.
 // The automatic compaction waits in NonTrivial:AfterRun
 // We generate more files and then trigger an automatic compaction
@@ -229,7 +229,7 @@ bool Compaction::IsTrivialMove() const {

  // Used in universal compaction, where trivial move can be done if the
  // input files are non overlapping
-  if ((immutable_cf_options_.compaction_options_universal.allow_trivial_move) &&
+  if ((mutable_cf_options_.compaction_options_universal.allow_trivial_move) &&
      (output_level_ != 0)) {
    return is_trivial_move_;
  }
@@ -405,20 +405,23 @@ void Compaction::Summary(char* output, int len) {
 uint64_t Compaction::OutputFilePreallocationSize() const {
  uint64_t preallocation_size = 0;

-  if (max_output_file_size_ != port::kMaxUint64 &&
-      (cfd_->ioptions()->compaction_style == kCompactionStyleLevel ||
-       output_level() > 0)) {
-    preallocation_size = max_output_file_size_;
-  } else {
-    for (const auto& level_files : inputs_) {
-      for (const auto& file : level_files.files) {
-        preallocation_size += file->fd.GetFileSize();
-      }
+  for (const auto& level_files : inputs_) {
+    for (const auto& file : level_files.files) {
+      preallocation_size += file->fd.GetFileSize();
    }
  }
+
+  if (max_output_file_size_ != port::kMaxUint64 &&
+      (immutable_cf_options_.compaction_style == kCompactionStyleLevel ||
+       output_level() > 0)) {
+    preallocation_size = std::min(max_output_file_size_, preallocation_size);
+  }
+
  // Over-estimate slightly so we don't end up just barely crossing
  // the threshold
-  return preallocation_size + (preallocation_size / 10);
+  // No point to prellocate more than 1GB.
+  return std::min(uint64_t{1073741824},
+                  preallocation_size + (preallocation_size / 10));
 }

 std::unique_ptr<CompactionFilter> Compaction::CreateCompactionFilter() const {
@@ -182,9 +182,13 @@ void CompactionIterator::Next() {

 void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
                                              Slice* skip_until) {
-  if (compaction_filter_ != nullptr && ikey_.type == kTypeValue &&
-      (visible_at_tip_ || ikey_.sequence > latest_snapshot_ ||
-       ignore_snapshots_)) {
+  if (compaction_filter_ != nullptr &&
+      (ikey_.type == kTypeValue || ikey_.type == kTypeBlobIndex) &&
+      (visible_at_tip_ || ignore_snapshots_ ||
+       ikey_.sequence > latest_snapshot_ ||
+       (snapshot_checker_ != nullptr &&
+        UNLIKELY(!snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                  latest_snapshot_))))) {
    // If the user has specified a compaction filter and the sequence
    // number is greater than any external snapshot, then invoke the
    // filter. If the return value of the compaction filter is true,
@@ -192,11 +196,13 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
    CompactionFilter::Decision filter;
    compaction_filter_value_.clear();
    compaction_filter_skip_until_.Clear();
+    CompactionFilter::ValueType value_type =
+        ikey_.type == kTypeValue ? CompactionFilter::ValueType::kValue
+                                 : CompactionFilter::ValueType::kBlobIndex;
    {
      StopWatchNano timer(env_, true);
      filter = compaction_filter_->FilterV2(
-          compaction_->level(), ikey_.user_key,
-          CompactionFilter::ValueType::kValue, value_,
+          compaction_->level(), ikey_.user_key, value_type, value_,
          &compaction_filter_value_, compaction_filter_skip_until_.rep());
      iter_stats_.total_filter_time +=
          env_ != nullptr ? timer.ElapsedNanos() : 0;
@@ -408,7 +414,10 @@ void CompactionIterator::NextFromInput() {
          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
        // Check whether the next key belongs to the same snapshot as the
        // SingleDelete.
-        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot) {
+        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
+            (snapshot_checker_ != nullptr &&
+             UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
+                                                       prev_snapshot)))) {
          if (next_ikey.type == kTypeSingleDeletion) {
            // We encountered two SingleDeletes in a row.  This could be due to
            // unexpected user input.
@@ -419,8 +428,12 @@ void CompactionIterator::NextFromInput() {
            // input_->Next().
            ++iter_stats_.num_record_drop_obsolete;
            ++iter_stats_.num_single_del_mismatch;
-          } else if ((ikey_.sequence <= earliest_write_conflict_snapshot_) ||
-                     has_outputted_key_) {
+          } else if (has_outputted_key_ ||
+                     (ikey_.sequence <= earliest_write_conflict_snapshot_ &&
+                      (snapshot_checker_ == nullptr ||
+                       LIKELY(snapshot_checker_->IsInSnapshot(
+                           ikey_.sequence,
+                           earliest_write_conflict_snapshot_))))) {
            // Found a matching value, we can drop the single delete and the
            // value.  It is safe to drop both records since we've already
            // outputted a key in this snapshot, or there is no earlier
@@ -468,6 +481,9 @@ void CompactionIterator::NextFromInput() {
        // comparison, so the value of has_current_user_key does not matter.
        has_current_user_key_ = false;
        if (compaction_ != nullptr && ikey_.sequence <= earliest_snapshot_ &&
+            (snapshot_checker_ == nullptr ||
+             LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                    earliest_snapshot_))) &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_)) {
          // Key doesn't exist outside of this range.
@@ -501,6 +517,9 @@ void CompactionIterator::NextFromInput() {
      input_->Next();
    } else if (compaction_ != nullptr && ikey_.type == kTypeDeletion &&
               ikey_.sequence <= earliest_snapshot_ &&
+               (snapshot_checker_ == nullptr ||
+                LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                       earliest_snapshot_))) &&
               ikeyNotNeededForIncrementalSnapshot() &&
               compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                          &level_ptrs_)) {
@@ -518,6 +537,11 @@ void CompactionIterator::NextFromInput() {
      //
      // Note:  Dropping this Delete will not affect TransactionDB
      // write-conflict checking since it is earlier than any snapshot.
+      //
+      // It seems that we can also drop deletion later than earliest snapshot
+      // given that:
+      // (1) The deletion is earlier than earliest_write_conflict_snapshot, and
+      // (2) No value exist earlier than the deletion.
      ++iter_stats_.num_record_drop_obsolete;
      if (!bottommost_level_) {
        ++iter_stats_.num_optimized_del_drop_obsolete;
@@ -598,9 +622,12 @@ void CompactionIterator::PrepareOutput() {
  // and the earliest snapshot is larger than this seqno
  // and the userkey differs from the last userkey in compaction
  // then we can squash the seqno to zero.
-
+  //
  // This is safe for TransactionDB write-conflict checking since transactions
  // only care about sequence number larger than any active snapshots.
+  //
+  // Can we do the same for levels above bottom level as long as
+  // KeyNotExistsBeyondOutputLevel() return true?
  if ((compaction_ != nullptr &&
      !compaction_->allow_ingest_behind()) &&
      ikeyNotNeededForIncrementalSnapshot() &&
@@ -9,8 +9,10 @@
 #include <vector>

 #include "port/port.h"
+#include "util/string_util.h"
 #include "util/testharness.h"
 #include "util/testutil.h"
+#include "utilities/merge_operators.h"

 namespace rocksdb {

@@ -39,9 +41,9 @@ class NoMergingMergeOp : public MergeOperator {
 // Always returns Decition::kRemove.
 class StallingFilter : public CompactionFilter {
 public:
-  virtual Decision FilterV2(int level, const Slice& key, ValueType t,
-                            const Slice& existing_value, std::string* new_value,
-                            std::string* skip_until) const override {
+  Decision FilterV2(int /*level*/, const Slice& key, ValueType /*type*/,
+                    const Slice& /*existing_value*/, std::string* /*new_value*/,
+                    std::string* /*skip_until*/) const override {
    int k = std::atoi(key.ToString().c_str());
    last_seen.store(k);
    while (k >= stall_at.load()) {
@@ -72,6 +74,18 @@ class StallingFilter : public CompactionFilter {
  mutable std::atomic<int> last_seen{0};
 };

+// Compaction filter that filter out all keys.
+class FilterAllKeysCompactionFilter : public CompactionFilter {
+ public:
+  Decision FilterV2(int /*level*/, const Slice& /*key*/, ValueType /*type*/,
+                    const Slice& /*existing_value*/, std::string* /*new_value*/,
+                    std::string* /*skip_until*/) const override {
+    return Decision::kRemove;
+  }
+
+  const char* Name() const override { return "AllKeysCompactionFilter"; }
+};
+
 class LoggingForwardVectorIterator : public InternalIterator {
 public:
  struct Action {
@@ -144,76 +158,146 @@ class FakeCompaction : public CompactionIterator::CompactionProxy {
 public:
  FakeCompaction() = default;

-  virtual int level(size_t compaction_input_level) const { return 0; }
+  virtual int level(size_t compaction_input_level) const override { return 0; }
  virtual bool KeyNotExistsBeyondOutputLevel(
-      const Slice& user_key, std::vector<size_t>* level_ptrs) const {
-    return key_not_exists_beyond_output_level;
+      const Slice& user_key, std::vector<size_t>* level_ptrs) const override {
+    return is_bottommost_level || key_not_exists_beyond_output_level;
  }
-  virtual bool bottommost_level() const { return false; }
-  virtual int number_levels() const { return 1; }
-  virtual Slice GetLargestUserKey() const {
+  virtual bool bottommost_level() const override { return is_bottommost_level; }
+  virtual int number_levels() const override { return 1; }
+  virtual Slice GetLargestUserKey() const override {
    return "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
  }
-  virtual bool allow_ingest_behind() const { return false; }
+  virtual bool allow_ingest_behind() const override { return false; }

-  virtual bool preserve_deletes() const {return false; }
+  virtual bool preserve_deletes() const override { return false; }

  bool key_not_exists_beyond_output_level = false;
+
+  bool is_bottommost_level = false;
 };

-class CompactionIteratorTest : public testing::Test {
+// A simplifed snapshot checker which assumes each snapshot has a global
+// last visible sequence.
+class TestSnapshotChecker : public SnapshotChecker {
+ public:
+  explicit TestSnapshotChecker(
+      SequenceNumber last_committed_sequence,
+      const std::unordered_map<SequenceNumber, SequenceNumber>& snapshots = {})
+      : last_committed_sequence_(last_committed_sequence),
+        snapshots_(snapshots) {}
+
+  bool IsInSnapshot(SequenceNumber seq,
+                    SequenceNumber snapshot_seq) const override {
+    if (snapshot_seq == kMaxSequenceNumber) {
+      return seq <= last_committed_sequence_;
+    }
+    assert(snapshots_.count(snapshot_seq) > 0);
+    return seq <= snapshots_.at(snapshot_seq);
+  }
+
+ private:
+  SequenceNumber last_committed_sequence_;
+  // A map of valid snapshot to last visible sequence to the snapshot.
+  std::unordered_map<SequenceNumber, SequenceNumber> snapshots_;
+};
+
+// Test param:
+//   bool: whether to pass snapshot_checker to compaction iterator.
+class CompactionIteratorTest : public testing::TestWithParam<bool> {
 public:
  CompactionIteratorTest()
      : cmp_(BytewiseComparator()), icmp_(cmp_), snapshots_({}) {}

-  void InitIterators(const std::vector<std::string>& ks,
-                     const std::vector<std::string>& vs,
-                     const std::vector<std::string>& range_del_ks,
-                     const std::vector<std::string>& range_del_vs,
-                     SequenceNumber last_sequence,
-                     MergeOperator* merge_op = nullptr,
-                     CompactionFilter* filter = nullptr) {
+  void InitIterators(
+      const std::vector<std::string>& ks, const std::vector<std::string>& vs,
+      const std::vector<std::string>& range_del_ks,
+      const std::vector<std::string>& range_del_vs,
+      SequenceNumber last_sequence,
+      SequenceNumber last_committed_sequence = kMaxSequenceNumber,
+      MergeOperator* merge_op = nullptr, CompactionFilter* filter = nullptr,
+      bool bottommost_level = false,
+      SequenceNumber earliest_write_conflict_snapshot = kMaxSequenceNumber) {
    std::unique_ptr<InternalIterator> range_del_iter(
        new test::VectorIterator(range_del_ks, range_del_vs));
    range_del_agg_.reset(new RangeDelAggregator(icmp_, snapshots_));
    ASSERT_OK(range_del_agg_->AddTombstones(std::move(range_del_iter)));

    std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
-    if (filter) {
+    if (filter || bottommost_level) {
      compaction_proxy_ = new FakeCompaction();
+      compaction_proxy_->is_bottommost_level = bottommost_level;
      compaction.reset(compaction_proxy_);
    }
-    // TODO(yiwu) add a mock snapshot checker and add test for it.
-    SnapshotChecker* snapshot_checker = nullptr;
+    bool use_snapshot_checker = UseSnapshotChecker() || GetParam();
+    if (use_snapshot_checker || last_committed_sequence < kMaxSequenceNumber) {
+      snapshot_checker_.reset(
+          new TestSnapshotChecker(last_committed_sequence, snapshot_map_));
+    }
+    merge_helper_.reset(
+        new MergeHelper(Env::Default(), cmp_, merge_op, filter, nullptr, false,
+                        0 /*latest_snapshot*/, snapshot_checker_.get(),
+                        0 /*level*/, nullptr /*statistics*/, &shutting_down_));

-    merge_helper_.reset(new MergeHelper(Env::Default(), cmp_, merge_op, filter,
-                                        nullptr, false, 0, 0, nullptr,
-                                        &shutting_down_));
    iter_.reset(new LoggingForwardVectorIterator(ks, vs));
    iter_->SeekToFirst();
    c_iter_.reset(new CompactionIterator(
        iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
-        kMaxSequenceNumber, snapshot_checker, Env::Default(), false,
-        range_del_agg_.get(), std::move(compaction), filter, nullptr,
-        &shutting_down_));
+        earliest_write_conflict_snapshot, snapshot_checker_.get(),
+        Env::Default(), false, range_del_agg_.get(), std::move(compaction),
+        filter, nullptr, &shutting_down_));
  }

-  void AddSnapshot(SequenceNumber snapshot) { snapshots_.push_back(snapshot); }
+  void AddSnapshot(SequenceNumber snapshot,
+                   SequenceNumber last_visible_seq = kMaxSequenceNumber) {
+    snapshots_.push_back(snapshot);
+    snapshot_map_[snapshot] = last_visible_seq;
+  }
+
+  virtual bool UseSnapshotChecker() const { return false; }
+
+  void RunTest(
+      const std::vector<std::string>& input_keys,
+      const std::vector<std::string>& input_values,
+      const std::vector<std::string>& expected_keys,
+      const std::vector<std::string>& expected_values,
+      SequenceNumber last_committed_seq = kMaxSequenceNumber,
+      MergeOperator* merge_operator = nullptr,
+      CompactionFilter* compaction_filter = nullptr,
+      bool bottommost_level = false,
+      SequenceNumber earliest_write_conflict_snapshot = kMaxSequenceNumber) {
+    InitIterators(input_keys, input_values, {}, {}, kMaxSequenceNumber,
+                  last_committed_seq, merge_operator, compaction_filter,
+                  bottommost_level, earliest_write_conflict_snapshot);
+    c_iter_->SeekToFirst();
+    for (size_t i = 0; i < expected_keys.size(); i++) {
+      std::string info = "i = " + ToString(i);
+      ASSERT_TRUE(c_iter_->Valid()) << info;
+      ASSERT_OK(c_iter_->status()) << info;
+      ASSERT_EQ(expected_keys[i], c_iter_->key().ToString()) << info;
+      ASSERT_EQ(expected_values[i], c_iter_->value().ToString()) << info;
+      c_iter_->Next();
+    }
+    ASSERT_FALSE(c_iter_->Valid());
+  }

  const Comparator* cmp_;
  const InternalKeyComparator icmp_;
  std::vector<SequenceNumber> snapshots_;
+  // A map of valid snapshot to last visible sequence to the snapshot.
+  std::unordered_map<SequenceNumber, SequenceNumber> snapshot_map_;
  std::unique_ptr<MergeHelper> merge_helper_;
  std::unique_ptr<LoggingForwardVectorIterator> iter_;
  std::unique_ptr<CompactionIterator> c_iter_;
  std::unique_ptr<RangeDelAggregator> range_del_agg_;
+  std::unique_ptr<SnapshotChecker> snapshot_checker_;
  std::atomic<bool> shutting_down_{false};
  FakeCompaction* compaction_proxy_;
 };

 // It is possible that the output of the compaction iterator is empty even if
 // the input is not.
-TEST_F(CompactionIteratorTest, EmptyResult) {
+TEST_P(CompactionIteratorTest, EmptyResult) {
  InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
                 test::KeyStr("a", 3, kTypeValue)},
                {"", "val"}, {}, {}, 5);
@@ -223,7 +307,7 @@ TEST_F(CompactionIteratorTest, EmptyResult) {

 // If there is a corruption after a single deletion, the corrupted key should
 // be preserved.
-TEST_F(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
+TEST_P(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
  InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
                 test::KeyStr("a", 3, kTypeValue, true),
                 test::KeyStr("b", 10, kTypeValue)},
@@ -242,7 +326,7 @@ TEST_F(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
  ASSERT_FALSE(c_iter_->Valid());
 }

-TEST_F(CompactionIteratorTest, SimpleRangeDeletion) {
+TEST_P(CompactionIteratorTest, SimpleRangeDeletion) {
  InitIterators({test::KeyStr("morning", 5, kTypeValue),
                 test::KeyStr("morning", 2, kTypeValue),
                 test::KeyStr("night", 3, kTypeValue)},
@@ -258,7 +342,7 @@ TEST_F(CompactionIteratorTest, SimpleRangeDeletion) {
  ASSERT_FALSE(c_iter_->Valid());
 }

-TEST_F(CompactionIteratorTest, RangeDeletionWithSnapshots) {
+TEST_P(CompactionIteratorTest, RangeDeletionWithSnapshots) {
  AddSnapshot(10);
  std::vector<std::string> ks1;
  ks1.push_back(test::KeyStr("ma", 28, kTypeRangeDeletion));
@@ -279,7 +363,7 @@ TEST_F(CompactionIteratorTest, RangeDeletionWithSnapshots) {
  ASSERT_FALSE(c_iter_->Valid());
 }

-TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
+TEST_P(CompactionIteratorTest, CompactionFilterSkipUntil) {
  class Filter : public CompactionFilter {
    virtual Decision FilterV2(int level, const Slice& key, ValueType t,
                              const Slice& existing_value,
@@ -354,7 +438,7 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
       test::KeyStr("j", 99, kTypeValue)},
      {"av50", "am45", "bv60", "bv40", "cv35", "dm70", "em71", "fm65", "fm30",
       "fv25", "gv90", "hv91", "im95", "jv99"},
-      {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+      {}, {}, kMaxSequenceNumber, kMaxSequenceNumber, &merge_op, &filter);

  // Compaction should output just "a", "e" and "h" keys.
  c_iter_->SeekToFirst();
@@ -389,13 +473,14 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
  ASSERT_EQ(expected_actions, iter_->log);
 }

-TEST_F(CompactionIteratorTest, ShuttingDownInFilter) {
+TEST_P(CompactionIteratorTest, ShuttingDownInFilter) {
  NoMergingMergeOp merge_op;
  StallingFilter filter;
  InitIterators(
      {test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeValue),
       test::KeyStr("3", 3, kTypeValue), test::KeyStr("4", 4, kTypeValue)},
-      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, kMaxSequenceNumber,
+      &merge_op, &filter);
  // Don't leave tombstones (kTypeDeletion) for filtered keys.
  compaction_proxy_->key_not_exists_beyond_output_level = true;

@@ -426,13 +511,14 @@ TEST_F(CompactionIteratorTest, ShuttingDownInFilter) {

 // Same as ShuttingDownInFilter, but shutdown happens during filter call for
 // a merge operand, not for a value.
-TEST_F(CompactionIteratorTest, ShuttingDownInMerge) {
+TEST_P(CompactionIteratorTest, ShuttingDownInMerge) {
  NoMergingMergeOp merge_op;
  StallingFilter filter;
  InitIterators(
      {test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeMerge),
       test::KeyStr("3", 3, kTypeMerge), test::KeyStr("4", 4, kTypeValue)},
-      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, kMaxSequenceNumber,
+      &merge_op, &filter);
  compaction_proxy_->key_not_exists_beyond_output_level = true;

  std::atomic<bool> seek_done{false};
@@ -460,7 +546,7 @@ TEST_F(CompactionIteratorTest, ShuttingDownInMerge) {
  EXPECT_EQ(2, filter.last_seen.load());
 }

-TEST_F(CompactionIteratorTest, SingleMergeOperand) {
+TEST_P(CompactionIteratorTest, SingleMergeOperand) {
  class Filter : public CompactionFilter {
    virtual Decision FilterV2(int level, const Slice& key, ValueType t,
                              const Slice& existing_value,
@@ -552,7 +638,7 @@ TEST_F(CompactionIteratorTest, SingleMergeOperand) {
       // c should invoke FullMerge due to kTypeValue at the beginning.
       test::KeyStr("c", 90, kTypeMerge), test::KeyStr("c", 80, kTypeValue)},
      {"av1", "bv2", "bv1", "cv2", "cv1"}, {}, {}, kMaxSequenceNumber,
-      &merge_op, &filter);
+      kMaxSequenceNumber, &merge_op, &filter);

  c_iter_->SeekToFirst();
  ASSERT_TRUE(c_iter_->Valid());
@@ -565,6 +651,298 @@ TEST_F(CompactionIteratorTest, SingleMergeOperand) {
  ASSERT_EQ("cv1cv2", c_iter_->value().ToString());
 }

+// In bottommost level, values earlier than earliest snapshot can be output
+// with sequence = 0.
+TEST_P(CompactionIteratorTest, ZeroOutSequenceAtBottomLevel) {
+  AddSnapshot(1);
+  RunTest({test::KeyStr("a", 1, kTypeValue), test::KeyStr("b", 2, kTypeValue)},
+          {"v1", "v2"},
+          {test::KeyStr("a", 0, kTypeValue), test::KeyStr("b", 2, kTypeValue)},
+          {"v1", "v2"}, kMaxSequenceNumber /*last_commited_seq*/,
+          nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
+          true /*bottommost_level*/);
+}
+
+// In bottommost level, deletions earlier than earliest snapshot can be removed
+// permanently.
+TEST_P(CompactionIteratorTest, RemoveDeletionAtBottomLevel) {
+  AddSnapshot(1);
+  RunTest({test::KeyStr("a", 1, kTypeDeletion),
+           test::KeyStr("b", 2, kTypeDeletion)},
+          {"", ""}, {test::KeyStr("b", 2, kTypeDeletion)}, {""},
+          kMaxSequenceNumber /*last_commited_seq*/, nullptr /*merge_operator*/,
+          nullptr /*compaction_filter*/, true /*bottommost_level*/);
+}
+
+// In bottommost level, single deletions earlier than earliest snapshot can be
+// removed permanently.
+TEST_P(CompactionIteratorTest, RemoveSingleDeletionAtBottomLevel) {
+  AddSnapshot(1);
+  RunTest({test::KeyStr("a", 1, kTypeSingleDeletion),
+           test::KeyStr("b", 2, kTypeSingleDeletion)},
+          {"", ""}, {test::KeyStr("b", 2, kTypeSingleDeletion)}, {""},
+          kMaxSequenceNumber /*last_commited_seq*/, nullptr /*merge_operator*/,
+          nullptr /*compaction_filter*/, true /*bottommost_level*/);
+}
+
+INSTANTIATE_TEST_CASE_P(CompactionIteratorTestInstance, CompactionIteratorTest,
+                        testing::Values(true, false));
+
+// Tests how CompactionIterator work together with SnapshotChecker.
+class CompactionIteratorWithSnapshotCheckerTest
+    : public CompactionIteratorTest {
+ public:
+  bool UseSnapshotChecker() const override { return true; }
+};
+
+// Uncommitted keys (keys with seq > last_committed_seq) should be output as-is
+// while committed version of these keys should get compacted as usual.
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       PreserveUncommittedKeys_Value) {
+  RunTest(
+      {test::KeyStr("foo", 3, kTypeValue), test::KeyStr("foo", 2, kTypeValue),
+       test::KeyStr("foo", 1, kTypeValue)},
+      {"v3", "v2", "v1"},
+      {test::KeyStr("foo", 3, kTypeValue), test::KeyStr("foo", 2, kTypeValue)},
+      {"v3", "v2"}, 2 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       PreserveUncommittedKeys_Deletion) {
+  RunTest({test::KeyStr("foo", 2, kTypeDeletion),
+           test::KeyStr("foo", 1, kTypeValue)},
+          {"", "v1"},
+          {test::KeyStr("foo", 2, kTypeDeletion),
+           test::KeyStr("foo", 1, kTypeValue)},
+          {"", "v1"}, 1 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       PreserveUncommittedKeys_Merge) {
+  auto merge_op = MergeOperators::CreateStringAppendOperator();
+  RunTest(
+      {test::KeyStr("foo", 3, kTypeMerge), test::KeyStr("foo", 2, kTypeMerge),
+       test::KeyStr("foo", 1, kTypeValue)},
+      {"v3", "v2", "v1"},
+      {test::KeyStr("foo", 3, kTypeMerge), test::KeyStr("foo", 2, kTypeValue)},
+      {"v3", "v1,v2"}, 2 /*last_committed_seq*/, merge_op.get());
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       PreserveUncommittedKeys_SingleDelete) {
+  RunTest({test::KeyStr("foo", 2, kTypeSingleDeletion),
+           test::KeyStr("foo", 1, kTypeValue)},
+          {"", "v1"},
+          {test::KeyStr("foo", 2, kTypeSingleDeletion),
+           test::KeyStr("foo", 1, kTypeValue)},
+          {"", "v1"}, 1 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       PreserveUncommittedKeys_BlobIndex) {
+  RunTest({test::KeyStr("foo", 3, kTypeBlobIndex),
+           test::KeyStr("foo", 2, kTypeBlobIndex),
+           test::KeyStr("foo", 1, kTypeBlobIndex)},
+          {"v3", "v2", "v1"},
+          {test::KeyStr("foo", 3, kTypeBlobIndex),
+           test::KeyStr("foo", 2, kTypeBlobIndex)},
+          {"v3", "v2"}, 2 /*last_committed_seq*/);
+}
+
+// Test compaction iterator dedup keys visible to the same snapshot.
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, DedupSameSnapshot_Value) {
+  AddSnapshot(2, 1);
+  RunTest(
+      {test::KeyStr("foo", 4, kTypeValue), test::KeyStr("foo", 3, kTypeValue),
+       test::KeyStr("foo", 2, kTypeValue), test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "v3", "v2", "v1"},
+      {test::KeyStr("foo", 4, kTypeValue), test::KeyStr("foo", 3, kTypeValue),
+       test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "v3", "v1"}, 3 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, DedupSameSnapshot_Deletion) {
+  AddSnapshot(2, 1);
+  RunTest(
+      {test::KeyStr("foo", 4, kTypeValue),
+       test::KeyStr("foo", 3, kTypeDeletion),
+       test::KeyStr("foo", 2, kTypeValue), test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "", "v2", "v1"},
+      {test::KeyStr("foo", 4, kTypeValue),
+       test::KeyStr("foo", 3, kTypeDeletion),
+       test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "", "v1"}, 3 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, DedupSameSnapshot_Merge) {
+  AddSnapshot(2, 1);
+  AddSnapshot(4, 3);
+  auto merge_op = MergeOperators::CreateStringAppendOperator();
+  RunTest(
+      {test::KeyStr("foo", 5, kTypeMerge), test::KeyStr("foo", 4, kTypeMerge),
+       test::KeyStr("foo", 3, kTypeMerge), test::KeyStr("foo", 2, kTypeMerge),
+       test::KeyStr("foo", 1, kTypeValue)},
+      {"v5", "v4", "v3", "v2", "v1"},
+      {test::KeyStr("foo", 5, kTypeMerge), test::KeyStr("foo", 4, kTypeMerge),
+       test::KeyStr("foo", 3, kTypeMerge), test::KeyStr("foo", 1, kTypeValue)},
+      {"v5", "v4", "v2,v3", "v1"}, 4 /*last_committed_seq*/, merge_op.get());
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       DedupSameSnapshot_SingleDeletion) {
+  AddSnapshot(2, 1);
+  RunTest(
+      {test::KeyStr("foo", 4, kTypeValue),
+       test::KeyStr("foo", 3, kTypeSingleDeletion),
+       test::KeyStr("foo", 2, kTypeValue), test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "", "v2", "v1"},
+      {test::KeyStr("foo", 4, kTypeValue), test::KeyStr("foo", 1, kTypeValue)},
+      {"v4", "v1"}, 3 /*last_committed_seq*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, DedupSameSnapshot_BlobIndex) {
+  AddSnapshot(2, 1);
+  RunTest({test::KeyStr("foo", 4, kTypeBlobIndex),
+           test::KeyStr("foo", 3, kTypeBlobIndex),
+           test::KeyStr("foo", 2, kTypeBlobIndex),
+           test::KeyStr("foo", 1, kTypeBlobIndex)},
+          {"v4", "v3", "v2", "v1"},
+          {test::KeyStr("foo", 4, kTypeBlobIndex),
+           test::KeyStr("foo", 3, kTypeBlobIndex),
+           test::KeyStr("foo", 1, kTypeBlobIndex)},
+          {"v4", "v3", "v1"}, 3 /*last_committed_seq*/);
+}
+
+// At bottom level, sequence numbers can be zero out, and deletions can be
+// removed, but only when they are visible to earliest snapshot.
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       NotZeroOutSequenceIfNotVisibleToEarliestSnapshot) {
+  AddSnapshot(2, 1);
+  RunTest({test::KeyStr("a", 1, kTypeValue), test::KeyStr("b", 2, kTypeValue),
+           test::KeyStr("c", 3, kTypeValue)},
+          {"v1", "v2", "v3"},
+          {test::KeyStr("a", 0, kTypeValue), test::KeyStr("b", 2, kTypeValue),
+           test::KeyStr("c", 3, kTypeValue)},
+          {"v1", "v2", "v3"}, kMaxSequenceNumber /*last_commited_seq*/,
+          nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
+          true /*bottommost_level*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       NotRemoveDeletionIfNotVisibleToEarliestSnapshot) {
+  AddSnapshot(2, 1);
+  RunTest(
+      {test::KeyStr("a", 1, kTypeDeletion), test::KeyStr("b", 2, kTypeDeletion),
+       test::KeyStr("c", 3, kTypeDeletion)},
+      {"", "", ""},
+      {test::KeyStr("b", 2, kTypeDeletion),
+       test::KeyStr("c", 3, kTypeDeletion)},
+      {"", ""}, kMaxSequenceNumber /*last_commited_seq*/,
+      nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
+      true /*bottommost_level*/);
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       NotRemoveSingleDeletionIfNotVisibleToEarliestSnapshot) {
+  AddSnapshot(2, 1);
+  RunTest({test::KeyStr("a", 1, kTypeSingleDeletion),
+           test::KeyStr("b", 2, kTypeSingleDeletion),
+           test::KeyStr("c", 3, kTypeSingleDeletion)},
+          {"", "", ""},
+          {test::KeyStr("b", 2, kTypeSingleDeletion),
+           test::KeyStr("c", 3, kTypeSingleDeletion)},
+          {"", ""}, kMaxSequenceNumber /*last_commited_seq*/,
+          nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
+          true /*bottommost_level*/);
+}
+
+// Single delete should not cancel out values that not visible to the
+// same set of snapshots
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       SingleDeleteAcrossSnapshotBoundary) {
+  AddSnapshot(2, 1);
+  RunTest({test::KeyStr("a", 2, kTypeSingleDeletion),
+           test::KeyStr("a", 1, kTypeValue)},
+          {"", "v1"},
+          {test::KeyStr("a", 2, kTypeSingleDeletion),
+           test::KeyStr("a", 1, kTypeValue)},
+          {"", "v1"}, 2 /*last_committed_seq*/);
+}
+
+// Single delete should be kept in case it is not visible to the
+// earliest write conflict snapshot. If a single delete is kept for this reason,
+// corresponding value can be trimmed to save space.
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       KeepSingleDeletionForWriteConflictChecking) {
+  AddSnapshot(2, 0);
+  RunTest({test::KeyStr("a", 2, kTypeSingleDeletion),
+           test::KeyStr("a", 1, kTypeValue)},
+          {"", "v1"},
+          {test::KeyStr("a", 2, kTypeSingleDeletion),
+           test::KeyStr("a", 1, kTypeValue)},
+          {"", ""}, 2 /*last_committed_seq*/, nullptr /*merge_operator*/,
+          nullptr /*compaction_filter*/, false /*bottommost_level*/,
+          2 /*earliest_write_conflict_snapshot*/);
+}
+
+// Compaction filter should keep uncommitted key as-is, and
+//   * Convert the latest velue to deletion, and/or
+//   * if latest value is a merge, apply filter to all suequent merges.
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, CompactionFilter_Value) {
+  std::unique_ptr<CompactionFilter> compaction_filter(
+      new FilterAllKeysCompactionFilter());
+  RunTest(
+      {test::KeyStr("a", 2, kTypeValue), test::KeyStr("a", 1, kTypeValue),
+       test::KeyStr("b", 3, kTypeValue), test::KeyStr("c", 1, kTypeValue)},
+      {"v2", "v1", "v3", "v4"},
+      {test::KeyStr("a", 2, kTypeValue), test::KeyStr("a", 1, kTypeDeletion),
+       test::KeyStr("b", 3, kTypeValue), test::KeyStr("c", 1, kTypeDeletion)},
+      {"v2", "", "v3", ""}, 1 /*last_committed_seq*/,
+      nullptr /*merge_operator*/, compaction_filter.get());
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, CompactionFilter_Deletion) {
+  std::unique_ptr<CompactionFilter> compaction_filter(
+      new FilterAllKeysCompactionFilter());
+  RunTest(
+      {test::KeyStr("a", 2, kTypeDeletion), test::KeyStr("a", 1, kTypeValue)},
+      {"", "v1"},
+      {test::KeyStr("a", 2, kTypeDeletion),
+       test::KeyStr("a", 1, kTypeDeletion)},
+      {"", ""}, 1 /*last_committed_seq*/, nullptr /*merge_operator*/,
+      compaction_filter.get());
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       CompactionFilter_PartialMerge) {
+  std::shared_ptr<MergeOperator> merge_op =
+      MergeOperators::CreateStringAppendOperator();
+  std::unique_ptr<CompactionFilter> compaction_filter(
+      new FilterAllKeysCompactionFilter());
+  RunTest({test::KeyStr("a", 3, kTypeMerge), test::KeyStr("a", 2, kTypeMerge),
+           test::KeyStr("a", 1, kTypeMerge)},
+          {"v3", "v2", "v1"}, {test::KeyStr("a", 3, kTypeMerge)}, {"v3"},
+          2 /*last_committed_seq*/, merge_op.get(), compaction_filter.get());
+}
+
+TEST_F(CompactionIteratorWithSnapshotCheckerTest, CompactionFilter_FullMerge) {
+  std::shared_ptr<MergeOperator> merge_op =
+      MergeOperators::CreateStringAppendOperator();
+  std::unique_ptr<CompactionFilter> compaction_filter(
+      new FilterAllKeysCompactionFilter());
+  RunTest(
+      {test::KeyStr("a", 3, kTypeMerge), test::KeyStr("a", 2, kTypeMerge),
+       test::KeyStr("a", 1, kTypeValue)},
+      {"v3", "v2", "v1"},
+      {test::KeyStr("a", 3, kTypeMerge), test::KeyStr("a", 1, kTypeDeletion)},
+      {"v3", ""}, 2 /*last_committed_seq*/, merge_op.get(),
+      compaction_filter.get());
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -265,8 +265,7 @@ CompactionJob::CompactionJob(
    int job_id, Compaction* compaction, const ImmutableDBOptions& db_options,
    const EnvOptions env_options, VersionSet* versions,
    const std::atomic<bool>* shutting_down,
-    const SequenceNumber preserve_deletes_seqnum,
-    LogBuffer* log_buffer,
+    const SequenceNumber preserve_deletes_seqnum, LogBuffer* log_buffer,
    Directory* db_directory, Directory* output_directory, Statistics* stats,
    InstrumentedMutex* db_mutex, Status* db_bg_error,
    std::vector<SequenceNumber> existing_snapshots,
@@ -282,6 +281,8 @@ CompactionJob::CompactionJob(
      db_options_(db_options),
      env_options_(env_options),
      env_(db_options.env),
+      env_optiosn_for_read_(
+          env_->OptimizeForCompactionTableRead(env_options, db_options_)),
      versions_(versions),
      shutting_down_(shutting_down),
      preserve_deletes_seqnum_(preserve_deletes_seqnum),
@@ -296,8 +297,10 @@ CompactionJob::CompactionJob(
      snapshot_checker_(snapshot_checker),
      table_cache_(std::move(table_cache)),
      event_logger_(event_logger),
+      bottommost_level_(false),
      paranoid_file_checks_(paranoid_file_checks),
-      measure_io_stats_(measure_io_stats) {
+      measure_io_stats_(measure_io_stats),
+      write_hint_(Env::WLTH_NOT_SET) {
  assert(log_buffer_ != nullptr);
  const auto* cfd = compact_->compaction->column_family_data();
  ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
@@ -368,6 +371,8 @@ void CompactionJob::Prepare() {
  assert(c->column_family_data()->current()->storage_info()
      ->NumLevelFiles(compact_->compaction->level()) > 0);

+  write_hint_ = c->column_family_data()->CalculateSSTWriteHint(
+    c->output_level());
  // Is this compaction producing files at the bottommost level?
  bottommost_level_ = c->bottommost_level();

@@ -620,7 +625,8 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
      "[%s] compacted to: %s, MB/sec: %.1f rd, %.1f wr, level %d, "
      "files in(%d, %d) out(%d) "
      "MB in(%.1f, %.1f) out(%.1f), read-write-amplify(%.1f) "
-      "write-amplify(%.1f) %s, records in: %d, records dropped: %d\n",
+      "write-amplify(%.1f) %s, records in: %" PRIu64
+      ", records dropped: %" PRIu64 " output_compression: %s\n",
      cfd->GetName().c_str(), vstorage->LevelSummary(&tmp), bytes_read_per_sec,
      bytes_written_per_sec, compact_->compaction->output_level(),
      stats.num_input_files_in_non_output_levels,
@@ -629,20 +635,23 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
      stats.bytes_read_output_level / 1048576.0,
      stats.bytes_written / 1048576.0, read_write_amp, write_amp,
      status.ToString().c_str(), stats.num_input_records,
-      stats.num_dropped_records);
+      stats.num_dropped_records,
+      CompressionTypeToString(compact_->compaction->output_compression())
+          .c_str());

  UpdateCompactionJobStats(stats);

  auto stream = event_logger_->LogToBuffer(log_buffer_);
-  stream << "job" << job_id_
-         << "event" << "compaction_finished"
+  stream << "job" << job_id_ << "event"
+         << "compaction_finished"
         << "compaction_time_micros" << compaction_stats_.micros
         << "output_level" << compact_->compaction->output_level()
         << "num_output_files" << compact_->NumOutputFiles()
-         << "total_output_size" << compact_->total_bytes
-         << "num_input_records" << compact_->num_input_records
-         << "num_output_records" << compact_->num_output_records
-         << "num_subcompactions" << compact_->sub_compact_states.size();
+         << "total_output_size" << compact_->total_bytes << "num_input_records"
+         << compact_->num_input_records << "num_output_records"
+         << compact_->num_output_records << "num_subcompactions"
+         << compact_->sub_compact_states.size() << "output_compression"
+         << CompressionTypeToString(compact_->compaction->output_compression());

  if (compaction_job_stats_ != nullptr) {
    stream << "num_single_delete_mismatches"
@@ -677,7 +686,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  std::unique_ptr<RangeDelAggregator> range_del_agg(
      new RangeDelAggregator(cfd->internal_comparator(), existing_snapshots_));
  std::unique_ptr<InternalIterator> input(versions_->MakeInputIterator(
-      sub_compact->compaction, range_del_agg.get()));
+      sub_compact->compaction, range_del_agg.get(), env_optiosn_for_read_));

  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_PROCESS_KV);
@@ -702,15 +711,18 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      sub_compact->compaction->mutable_cf_options();

  // To build compression dictionary, we sample the first output file, assuming
-  // it'll reach the maximum length, and then use the dictionary for compressing
-  // subsequent output files. The dictionary may be less than max_dict_bytes if
-  // the first output file's length is less than the maximum.
+  // it'll reach the maximum length. We optionally pass these samples through
+  // 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;
+  const size_t kSampleBytes =
+      kUseZstdTrainer ? cfd->ioptions()->compression_opts.zstd_max_train_bytes
+                      : cfd->ioptions()->compression_opts.max_dict_bytes;
  const int kSampleLenShift = 6;  // 2^6 = 64-byte samples
  std::set<size_t> sample_begin_offsets;
-  if (bottommost_level_ &&
-      cfd->ioptions()->compression_opts.max_dict_bytes > 0) {
-    const size_t kMaxSamples =
-        cfd->ioptions()->compression_opts.max_dict_bytes >> kSampleLenShift;
+  if (bottommost_level_ && kSampleBytes > 0) {
+    const size_t kMaxSamples = kSampleBytes >> kSampleLenShift;
    const size_t kOutFileLen = mutable_cf_options->MaxFileSizeForLevel(
        compact_->compaction->output_level());
    if (kOutFileLen != port::kMaxSizet) {
@@ -735,8 +747,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      compaction_filter, db_options_.info_log.get(),
      false /* internal key corruption is expected */,
      existing_snapshots_.empty() ? 0 : existing_snapshots_.back(),
-      compact_->compaction->level(), db_options_.statistics.get(),
-      shutting_down_);
+      snapshot_checker_, compact_->compaction->level(),
+      db_options_.statistics.get(), shutting_down_);

  TEST_SYNC_POINT("CompactionJob::Run():Inprogress");

@@ -780,11 +792,11 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  }
  const auto& c_iter_stats = c_iter->iter_stats();
  auto sample_begin_offset_iter = sample_begin_offsets.cbegin();
-  // data_begin_offset and compression_dict are only valid while generating
+  // data_begin_offset and dict_sample_data are only valid while generating
  // dictionary from the first output file.
  size_t data_begin_offset = 0;
-  std::string compression_dict;
-  compression_dict.reserve(cfd->ioptions()->compression_opts.max_dict_bytes);
+  std::string dict_sample_data;
+  dict_sample_data.reserve(kSampleBytes);

  while (status.ok() && !cfd->IsDropped() && c_iter->Valid()) {
    // Invariant: c_iter.status() is guaranteed to be OK if c_iter->Valid()
@@ -856,7 +868,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
            data_elmt_copy_len =
                data_end_offset - (data_begin_offset + data_elmt_copy_offset);
          }
-          compression_dict.append(&data_elmt.data()[data_elmt_copy_offset],
+          dict_sample_data.append(&data_elmt.data()[data_elmt_copy_offset],
                                  data_elmt_copy_len);
          if (sample_end_offset > data_end_offset) {
            // Didn't finish sample. Try to finish it with the next data_elmt.
@@ -911,9 +923,15 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      RecordDroppedKeys(range_del_out_stats,
                        &sub_compact->compaction_job_stats);
      if (sub_compact->outputs.size() == 1) {
-        // Use dictionary from first output file for compression of subsequent
-        // files.
-        sub_compact->compression_dict = std::move(compression_dict);
+        // Use samples from first output file to create dictionary for
+        // compression of subsequent files.
+        if (kUseZstdTrainer) {
+          sub_compact->compression_dict = ZSTD_TrainDictionary(
+              dict_sample_data, kSampleLenShift,
+              cfd->ioptions()->compression_opts.max_dict_bytes);
+        } else {
+          sub_compact->compression_dict = std::move(dict_sample_data);
+        }
      }
    }
  }
@@ -1296,6 +1314,7 @@ Status CompactionJob::OpenCompactionOutputFile(

  sub_compact->outputs.push_back(out);
  writable_file->SetIOPriority(Env::IO_LOW);
+  writable_file->SetWriteLifeTimeHint(write_hint_);
  writable_file->SetPreallocationBlockSize(static_cast<size_t>(
      sub_compact->compaction->OutputFilePreallocationSize()));
  sub_compact->outfile.reset(new WritableFileWriter(
@@ -1311,7 +1330,15 @@ Status CompactionJob::OpenCompactionOutputFile(
      sub_compact->compaction->MaxInputFileCreationTime();
  if (output_file_creation_time == 0) {
    int64_t _current_time = 0;
-    db_options_.env->GetCurrentTime(&_current_time);  // ignore error
+    auto status = db_options_.env->GetCurrentTime(&_current_time);
+    // Safe to proceed even if GetCurrentTime fails. So, log and proceed.
+    if (!status.ok()) {
+      ROCKS_LOG_WARN(
+          db_options_.info_log,
+          "Failed to get current time to populate creation_time property. "
+          "Status: %s",
+          status.ToString().c_str());
+    }
    output_file_creation_time = static_cast<uint64_t>(_current_time);
  }

@@ -134,6 +134,8 @@ class CompactionJob {
  const EnvOptions env_options_;

  Env* env_;
+  // env_option optimized for compaction table reads
+  EnvOptions env_optiosn_for_read_;
  VersionSet* versions_;
  const std::atomic<bool>* shutting_down_;
  const SequenceNumber preserve_deletes_seqnum_;
@@ -167,6 +169,7 @@ class CompactionJob {
  std::vector<Slice> boundaries_;
  // Stores the approx size of keys covered in the range of each subcompaction
  std::vector<uint64_t> sizes_;
+  Env::WriteLifeTimeHint write_hint_;
 };

 }  // namespace rocksdb
@@ -143,7 +143,8 @@ class CompactionJobTest : public testing::Test {
  }

  void SetLastSequence(const SequenceNumber sequence_number) {
-    versions_->SetLastToBeWrittenSequence(sequence_number + 1);
+    versions_->SetLastAllocatedSequence(sequence_number + 1);
+    versions_->SetLastPublishedSequence(sequence_number + 1);
    versions_->SetLastSequence(sequence_number + 1);
  }

@@ -89,7 +89,7 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
  // If bottommost_compression is set and we are compacting to the
  // bottommost level then we should use it.
  if (ioptions.bottommost_compression != kDisableCompressionOption &&
-      level > base_level && level >= (vstorage->num_non_empty_levels() - 1)) {
+      level >= (vstorage->num_non_empty_levels() - 1)) {
    return ioptions.bottommost_compression;
  }
  // If the user has specified a different compression level for each level,
@@ -175,6 +175,8 @@ TEST_F(CompactionPickerTest, Level1Trigger) {
 }

 TEST_F(CompactionPickerTest, Level1Trigger2) {
+  mutable_cf_options_.target_file_size_base = 10000000000;
+  mutable_cf_options_.RefreshDerivedOptions(ioptions_);
  NewVersionStorage(6, kCompactionStyleLevel);
  Add(1, 66U, "150", "200", 1000000001U);
  Add(1, 88U, "201", "300", 1000000000U);
@@ -191,13 +193,14 @@ TEST_F(CompactionPickerTest, Level1Trigger2) {
  ASSERT_EQ(66U, compaction->input(0, 0)->fd.GetNumber());
  ASSERT_EQ(6U, compaction->input(1, 0)->fd.GetNumber());
  ASSERT_EQ(7U, compaction->input(1, 1)->fd.GetNumber());
+  ASSERT_EQ(uint64_t{1073741824}, compaction->OutputFilePreallocationSize());
 }

 TEST_F(CompactionPickerTest, LevelMaxScore) {
  NewVersionStorage(6, kCompactionStyleLevel);
  mutable_cf_options_.target_file_size_base = 10000000;
-  mutable_cf_options_.target_file_size_multiplier = 10;
  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
+  mutable_cf_options_.RefreshDerivedOptions(ioptions_);
  Add(0, 1U, "150", "200", 1000000U);
  // Level 1 score 1.2
  Add(1, 66U, "150", "200", 6000000U);
@@ -218,6 +221,9 @@ TEST_F(CompactionPickerTest, LevelMaxScore) {
  ASSERT_TRUE(compaction.get() != nullptr);
  ASSERT_EQ(1U, compaction->num_input_files(0));
  ASSERT_EQ(7U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(mutable_cf_options_.target_file_size_base +
+                mutable_cf_options_.target_file_size_base / 10,
+            compaction->OutputFilePreallocationSize());
 }

 TEST_F(CompactionPickerTest, NeedsCompactionLevel) {
@@ -437,7 +443,7 @@ TEST_F(CompactionPickerTest, CompactionUniversalIngestBehindReservedLevel) {
 TEST_F(CompactionPickerTest, CannotTrivialMoveUniversal) {
  const uint64_t kFileSize = 100000;

-  ioptions_.compaction_options_universal.allow_trivial_move = true;
+  mutable_cf_options_.compaction_options_universal.allow_trivial_move = true;
  NewVersionStorage(1, kCompactionStyleUniversal);
  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
  // must return false when there's no files.
@@ -468,7 +474,7 @@ TEST_F(CompactionPickerTest, CannotTrivialMoveUniversal) {
 TEST_F(CompactionPickerTest, AllowsTrivialMoveUniversal) {
  const uint64_t kFileSize = 100000;

-  ioptions_.compaction_options_universal.allow_trivial_move = true;
+  mutable_cf_options_.compaction_options_universal.allow_trivial_move = true;
  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);

  NewVersionStorage(3, kCompactionStyleUniversal);
@@ -521,9 +527,10 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {
 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
  NewVersionStorage(6, kCompactionStyleLevel);
  ioptions_.compaction_pri = kMinOverlappingRatio;
-  mutable_cf_options_.target_file_size_base = 10000000;
+  mutable_cf_options_.target_file_size_base = 100000000000;
  mutable_cf_options_.target_file_size_multiplier = 10;
  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
+  mutable_cf_options_.RefreshDerivedOptions(ioptions_);

  Add(2, 6U, "150", "179", 50000000U);
  Add(2, 7U, "180", "220", 50000000U);
@@ -543,6 +550,8 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
  ASSERT_EQ(1U, compaction->num_input_files(0));
  // Pick file 8 because it overlaps with 0 files on level 3.
  ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
+  // Compaction input size * 1.1
+  ASSERT_GE(uint64_t{55000000}, compaction->OutputFilePreallocationSize());
 }

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
@@ -35,7 +35,7 @@ namespace {
 // and the index of the file in that level

 struct InputFileInfo {
-  InputFileInfo() : f(nullptr) {}
+  InputFileInfo() : f(nullptr), level(0), index(0) {}

  FileMetaData* f;
  size_t level;
@@ -204,7 +204,8 @@ void UniversalCompactionPicker::SortedRun::DumpSizeInfo(

 std::vector<UniversalCompactionPicker::SortedRun>
 UniversalCompactionPicker::CalculateSortedRuns(
-    const VersionStorageInfo& vstorage, const ImmutableCFOptions& ioptions) {
+    const VersionStorageInfo& vstorage, const ImmutableCFOptions& ioptions,
+    const MutableCFOptions& mutable_cf_options) {
  std::vector<UniversalCompactionPicker::SortedRun> ret;
  for (FileMetaData* f : vstorage.LevelFiles(0)) {
    ret.emplace_back(0, f, f->fd.GetFileSize(), f->compensated_file_size,
@@ -218,7 +219,8 @@ UniversalCompactionPicker::CalculateSortedRuns(
    for (FileMetaData* f : vstorage.LevelFiles(level)) {
      total_compensated_size += f->compensated_file_size;
      total_size += f->fd.GetFileSize();
-      if (ioptions.compaction_options_universal.allow_trivial_move == true) {
+      if (mutable_cf_options.compaction_options_universal.allow_trivial_move ==
+          true) {
        if (f->being_compacted) {
          being_compacted = f->being_compacted;
        }
@@ -227,7 +229,8 @@ UniversalCompactionPicker::CalculateSortedRuns(
        // non-zero level, all the files should share the same being_compacted
        // value.
        // This assumption is only valid when
-        // ioptions.compaction_options_universal.allow_trivial_move is false
+        // mutable_cf_options.compaction_options_universal.allow_trivial_move is
+        // false
        assert(is_first || f->being_compacted == being_compacted);
      }
      if (is_first) {
@@ -245,14 +248,13 @@ UniversalCompactionPicker::CalculateSortedRuns(

 // Universal style of compaction. Pick files that are contiguous in
 // time-range to compact.
-//
 Compaction* UniversalCompactionPicker::PickCompaction(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
  const int kLevel0 = 0;
  double score = vstorage->CompactionScore(kLevel0);
  std::vector<SortedRun> sorted_runs =
-      CalculateSortedRuns(*vstorage, ioptions_);
+      CalculateSortedRuns(*vstorage, ioptions_, mutable_cf_options);

  if (sorted_runs.size() == 0 ||
      sorted_runs.size() <
@@ -279,7 +281,8 @@ Compaction* UniversalCompactionPicker::PickCompaction(
  } else {
    // Size amplification is within limits. Try reducing read
    // amplification while maintaining file size ratios.
-    unsigned int ratio = ioptions_.compaction_options_universal.size_ratio;
+    unsigned int ratio =
+        mutable_cf_options.compaction_options_universal.size_ratio;

    if ((c = PickCompactionToReduceSortedRuns(
             cf_name, mutable_cf_options, vstorage, score, ratio, UINT_MAX,
@@ -327,7 +330,8 @@ Compaction* UniversalCompactionPicker::PickCompaction(
    return nullptr;
  }

-  if (ioptions_.compaction_options_universal.allow_trivial_move == true) {
+  if (mutable_cf_options.compaction_options_universal.allow_trivial_move ==
+      true) {
    c->set_is_trivial_move(IsInputFilesNonOverlapping(c));
  }

@@ -381,7 +385,8 @@ Compaction* UniversalCompactionPicker::PickCompaction(
 }

 uint32_t UniversalCompactionPicker::GetPathId(
-    const ImmutableCFOptions& ioptions, uint64_t file_size) {
+    const ImmutableCFOptions& ioptions,
+    const MutableCFOptions& mutable_cf_options, uint64_t file_size) {
  // Two conditions need to be satisfied:
  // (1) the target path needs to be able to hold the file's size
  // (2) Total size left in this and previous paths need to be not
@@ -398,8 +403,8 @@ uint32_t UniversalCompactionPicker::GetPathId(
  // that case. We need to improve it.
  uint64_t accumulated_size = 0;
  uint64_t future_size =
-      file_size * (100 - ioptions.compaction_options_universal.size_ratio) /
-      100;
+      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++) {
@@ -423,9 +428,9 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
    unsigned int max_number_of_files_to_compact,
    const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
  unsigned int min_merge_width =
-      ioptions_.compaction_options_universal.min_merge_width;
+      mutable_cf_options.compaction_options_universal.min_merge_width;
  unsigned int max_merge_width =
-      ioptions_.compaction_options_universal.max_merge_width;
+      mutable_cf_options.compaction_options_universal.max_merge_width;

  const SortedRun* sr = nullptr;
  bool done = false;
@@ -492,7 +497,7 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
      if (sz < static_cast<double>(succeeding_sr->size)) {
        break;
      }
-      if (ioptions_.compaction_options_universal.stop_style ==
+      if (mutable_cf_options.compaction_options_universal.stop_style ==
          kCompactionStopStyleSimilarSize) {
        // Similar-size stopping rule: also check the last picked file isn't
        // far larger than the next candidate file.
@@ -535,7 +540,7 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
  // size ratio of compression.
  bool enable_compression = true;
  int ratio_to_compress =
-      ioptions_.compaction_options_universal.compression_size_percent;
+      mutable_cf_options.compaction_options_universal.compression_size_percent;
  if (ratio_to_compress >= 0) {
    uint64_t total_size = 0;
    for (auto& sorted_run : sorted_runs) {
@@ -556,7 +561,8 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
  for (unsigned int i = 0; i < first_index_after; i++) {
    estimated_total_size += sorted_runs[i].size;
  }
-  uint32_t path_id = GetPathId(ioptions_, estimated_total_size);
+  uint32_t path_id =
+      GetPathId(ioptions_, mutable_cf_options, estimated_total_size);
  int start_level = sorted_runs[start_index].level;
  int output_level;
  if (first_index_after == sorted_runs.size()) {
@@ -597,9 +603,9 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(

  CompactionReason compaction_reason;
  if (max_number_of_files_to_compact == UINT_MAX) {
-    compaction_reason = CompactionReason::kUniversalSortedRunNum;
-  } else {
    compaction_reason = CompactionReason::kUniversalSizeRatio;
+  } else {
+    compaction_reason = CompactionReason::kUniversalSortedRunNum;
  }
  return new Compaction(
      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
@@ -621,8 +627,8 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
    VersionStorageInfo* vstorage, double score,
    const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
  // percentage flexibility while reducing size amplification
-  uint64_t ratio =
-      ioptions_.compaction_options_universal.max_size_amplification_percent;
+  uint64_t ratio = mutable_cf_options.compaction_options_universal
+                       .max_size_amplification_percent;

  unsigned int candidate_count = 0;
  uint64_t candidate_size = 0;
@@ -700,7 +706,8 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
  for (size_t loop = start_index; loop < sorted_runs.size(); loop++) {
    estimated_total_size += sorted_runs[loop].size;
  }
-  uint32_t path_id = GetPathId(ioptions_, estimated_total_size);
+  uint32_t path_id =
+      GetPathId(ioptions_, mutable_cf_options, estimated_total_size);
  int start_level = sorted_runs[start_index].level;

  std::vector<CompactionInputFiles> inputs(vstorage->num_levels());
@@ -80,11 +80,13 @@ class UniversalCompactionPicker : public CompactionPicker {
  bool IsInputFilesNonOverlapping(Compaction* c);

  static std::vector<SortedRun> CalculateSortedRuns(
-      const VersionStorageInfo& vstorage, const ImmutableCFOptions& ioptions);
+      const VersionStorageInfo& vstorage, const ImmutableCFOptions& ioptions,
+      const MutableCFOptions& mutable_cf_options);

  // Pick a path ID to place a newly generated file, with its estimated file
  // size.
  static uint32_t GetPathId(const ImmutableCFOptions& ioptions,
+                            const MutableCFOptions& mutable_cf_options,
                            uint64_t file_size);
 };
 }  // namespace rocksdb
@@ -19,9 +19,17 @@ void CancelAllBackgroundWork(DB* db, bool wait) {
 }

 Status DeleteFilesInRange(DB* db, ColumnFamilyHandle* column_family,
-                          const Slice* begin, const Slice* end) {
+                          const Slice* begin, const Slice* end,
+                          bool include_end) {
+  RangePtr range(begin, end);
+  return DeleteFilesInRanges(db, column_family, &range, 1, include_end);
+}
+
+Status DeleteFilesInRanges(DB* db, ColumnFamilyHandle* column_family,
+                           const RangePtr* ranges, size_t n,
+                           bool include_end) {
  return (static_cast_with_check<DBImpl, DB>(db->GetRootDB()))
-      ->DeleteFilesInRange(column_family, begin, end);
+      ->DeleteFilesInRanges(column_family, ranges, n, include_end);
 }

 Status VerifySstFileChecksum(const Options& options,
@@ -9,6 +9,7 @@
 #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
@@ -570,19 +571,19 @@ TEST_F(DBBasicTest, CompactBetweenSnapshots) {

 TEST_F(DBBasicTest, DBOpen_Options) {
  Options options = CurrentOptions();
-  std::string dbname = test::TmpDir(env_) + "/db_options_test";
-  ASSERT_OK(DestroyDB(dbname, options));
+  Close();
+  Destroy(options);

  // Does not exist, and create_if_missing == false: error
  DB* db = nullptr;
  options.create_if_missing = false;
-  Status s = DB::Open(options, dbname, &db);
+  Status s = DB::Open(options, dbname_, &db);
  ASSERT_TRUE(strstr(s.ToString().c_str(), "does not exist") != nullptr);
  ASSERT_TRUE(db == nullptr);

  // Does not exist, and create_if_missing == true: OK
  options.create_if_missing = true;
-  s = DB::Open(options, dbname, &db);
+  s = DB::Open(options, dbname_, &db);
  ASSERT_OK(s);
  ASSERT_TRUE(db != nullptr);

@@ -592,14 +593,14 @@ TEST_F(DBBasicTest, DBOpen_Options) {
  // Does exist, and error_if_exists == true: error
  options.create_if_missing = false;
  options.error_if_exists = true;
-  s = DB::Open(options, dbname, &db);
+  s = DB::Open(options, dbname_, &db);
  ASSERT_TRUE(strstr(s.ToString().c_str(), "exists") != nullptr);
  ASSERT_TRUE(db == nullptr);

  // Does exist, and error_if_exists == false: OK
  options.create_if_missing = true;
  options.error_if_exists = false;
-  s = DB::Open(options, dbname, &db);
+  s = DB::Open(options, dbname_, &db);
  ASSERT_OK(s);
  ASSERT_TRUE(db != nullptr);

@@ -847,6 +848,113 @@ TEST_F(DBBasicTest, MmapAndBufferOptions) {
 }
 #endif

+class TestEnv : public EnvWrapper {
+  public:
+    explicit TestEnv() : EnvWrapper(Env::Default()),
+                close_count(0) { }
+
+    class TestLogger : public Logger {
+      public:
+        using Logger::Logv;
+        TestLogger(TestEnv *env_ptr) : Logger() { env = env_ptr; }
+        ~TestLogger() {
+          if (!closed_) {
+            CloseHelper();
+          }
+        }
+        virtual void Logv(const char *format, va_list ap) override { };
+      protected:
+        virtual Status CloseImpl() override {
+          return CloseHelper();
+        }
+      private:
+        Status CloseHelper() {
+          env->CloseCountInc();;
+          return Status::IOError();
+        }
+        TestEnv *env;
+    };
+
+    void CloseCountInc() { close_count++; }
+
+    int GetCloseCount() { return close_count; }
+
+    virtual Status NewLogger(const std::string& fname,
+                             shared_ptr<Logger>* result) {
+      result->reset(new TestLogger(this));
+      return Status::OK();
+    }
+
+  private:
+    int close_count;
+};
+
+TEST_F(DBBasicTest, DBClose) {
+  Options options = GetDefaultOptions();
+  std::string dbname = test::TmpDir(env_) + "/db_close_test";
+  ASSERT_OK(DestroyDB(dbname, options));
+
+  DB* db = nullptr;
+  TestEnv *env = new TestEnv();
+  options.create_if_missing = true;
+  options.env = env;
+  Status s = DB::Open(options, dbname, &db);
+  ASSERT_OK(s);
+  ASSERT_TRUE(db != nullptr);
+
+  s = db->Close();
+  ASSERT_EQ(env->GetCloseCount(), 1);
+  ASSERT_EQ(s, Status::IOError());
+
+  delete db;
+  ASSERT_EQ(env->GetCloseCount(), 1);
+
+  // Do not call DB::Close() and ensure our logger Close() still gets called
+  s = DB::Open(options, dbname, &db);
+  ASSERT_OK(s);
+  ASSERT_TRUE(db != nullptr);
+  delete db;
+  ASSERT_EQ(env->GetCloseCount(), 2);
+
+  // Provide our own logger and ensure DB::Close() does not close it
+  options.info_log.reset(new TestEnv::TestLogger(env));
+  options.create_if_missing = false;
+  s = DB::Open(options, dbname, &db);
+  ASSERT_OK(s);
+  ASSERT_TRUE(db != nullptr);
+
+  s = db->Close();
+  ASSERT_EQ(s, Status::OK());
+  delete db;
+  ASSERT_EQ(env->GetCloseCount(), 2);
+  options.info_log.reset();
+  ASSERT_EQ(env->GetCloseCount(), 3);
+
+  delete options.env;
+}
+
+TEST_F(DBBasicTest, DBCloseFlushError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.manual_wal_flush = true;
+  options.write_buffer_size=100;
+  options.env = fault_injection_env.get();
+
+  Reopen(options);
+  ASSERT_OK(Put("key1", "value1"));
+  ASSERT_OK(Put("key2", "value2"));
+  ASSERT_OK(dbfull()->TEST_SwitchMemtable());
+  ASSERT_OK(Put("key3", "value3"));
+  fault_injection_env->SetFilesystemActive(false);
+  Status s = dbfull()->Close();
+  fault_injection_env->SetFilesystemActive(true);
+  ASSERT_NE(s, Status::OK());
+
+  Destroy(options);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -111,6 +111,31 @@ class DBBlockCacheTest : public DBTestBase {
  }
 };

+TEST_F(DBBlockCacheTest, IteratorBlockCacheUsage) {
+  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(0, 0, false);
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  ASSERT_EQ(0, cache->GetUsage());
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(0));
+  ASSERT_LT(0, cache->GetUsage());
+  delete iter;
+  iter = nullptr;
+  ASSERT_EQ(0, cache->GetUsage());
+}
+
 TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
  ReadOptions read_options;
  auto table_options = GetTableOptions();
@@ -1517,6 +1517,176 @@ TEST_F(DBCompactionTest, DeleteFileRange) {
  ASSERT_GT(old_num_files, new_num_files);
 }

+TEST_F(DBCompactionTest, DeleteFilesInRanges) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1024 * 1024;
+  options.max_bytes_for_level_multiplier = 2;
+  options.num_levels = 4;
+  options.max_background_compactions = 3;
+  options.disable_auto_compactions = true;
+
+  DestroyAndReopen(options);
+  int32_t value_size = 10 * 1024;  // 10 KB
+
+  Random rnd(301);
+  std::map<int32_t, std::string> values;
+
+  // file [0 => 100), [100 => 200), ... [900, 1000)
+  for (auto i = 0; i < 10; i++) {
+    for (auto j = 0; j < 100; j++) {
+      auto k = i * 100 + j;
+      values[k] = RandomString(&rnd, value_size);
+      ASSERT_OK(Put(Key(k), values[k]));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_EQ("10", FilesPerLevel(0));
+  CompactRangeOptions compact_options;
+  compact_options.change_level = true;
+  compact_options.target_level = 2;
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  ASSERT_EQ("0,0,10", FilesPerLevel(0));
+
+  // file [0 => 100), [200 => 300), ... [800, 900)
+  for (auto i = 0; i < 10; i+=2) {
+    for (auto j = 0; j < 100; j++) {
+      auto k = i * 100 + j;
+      ASSERT_OK(Put(Key(k), values[k]));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_EQ("5,0,10", FilesPerLevel(0));
+  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
+  ASSERT_EQ("0,5,10", FilesPerLevel(0));
+
+  // Delete files in range [0, 299] (inclusive)
+  {
+    auto begin_str1 = Key(0), end_str1 = Key(100);
+    auto begin_str2 = Key(100), end_str2 = Key(200);
+    auto begin_str3 = Key(200), end_str3 = Key(299);
+    Slice begin1(begin_str1), end1(end_str1);
+    Slice begin2(begin_str2), end2(end_str2);
+    Slice begin3(begin_str3), end3(end_str3);
+    std::vector<RangePtr> ranges;
+    ranges.push_back(RangePtr(&begin1, &end1));
+    ranges.push_back(RangePtr(&begin2, &end2));
+    ranges.push_back(RangePtr(&begin3, &end3));
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(),
+                                  ranges.data(), ranges.size()));
+    ASSERT_EQ("0,3,7", FilesPerLevel(0));
+
+    // Keys [0, 300) should not exist.
+    for (auto i = 0; i < 300; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+    for (auto i = 300; i < 1000; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+  }
+
+  // Delete files in range [600, 999) (exclusive)
+  {
+    auto begin_str1 = Key(600), end_str1 = Key(800);
+    auto begin_str2 = Key(700), end_str2 = Key(900);
+    auto begin_str3 = Key(800), end_str3 = Key(999);
+    Slice begin1(begin_str1), end1(end_str1);
+    Slice begin2(begin_str2), end2(end_str2);
+    Slice begin3(begin_str3), end3(end_str3);
+    std::vector<RangePtr> ranges;
+    ranges.push_back(RangePtr(&begin1, &end1));
+    ranges.push_back(RangePtr(&begin2, &end2));
+    ranges.push_back(RangePtr(&begin3, &end3));
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(),
+                                  ranges.data(), ranges.size(), false));
+    ASSERT_EQ("0,1,4", FilesPerLevel(0));
+
+    // Keys [600, 900) should not exist.
+    for (auto i = 600; i < 900; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+    for (auto i = 300; i < 600; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+    for (auto i = 900; i < 1000; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+  }
+
+  // Delete all files.
+  {
+    RangePtr range;
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(), &range, 1));
+    ASSERT_EQ("", FilesPerLevel(0));
+
+    for (auto i = 0; i < 1000; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+  }
+}
+
+TEST_F(DBCompactionTest, DeleteFileRangeFileEndpointsOverlapBug) {
+  // regression test for #2833: groups of files whose user-keys overlap at the
+  // endpoints could be split by `DeleteFilesInRange`. This caused old data to
+  // reappear, either because a new version of the key was removed, or a range
+  // deletion was partially dropped. It could also cause non-overlapping
+  // invariant to be violated if the files dropped by DeleteFilesInRange were
+  // a subset of files that a range deletion spans.
+  const int kNumL0Files = 2;
+  const int kValSize = 8 << 10;  // 8KB
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = kNumL0Files;
+  options.target_file_size_base = 1 << 10;  // 1KB
+  DestroyAndReopen(options);
+
+  // The snapshot prevents key 1 from having its old version dropped. The low
+  // `target_file_size_base` ensures two keys will be in each output file.
+  const Snapshot* snapshot = nullptr;
+  Random rnd(301);
+  // The value indicates which flush the key belonged to, which is enough
+  // for us to determine the keys' relative ages. After L0 flushes finish,
+  // files look like:
+  //
+  // File 0: 0 -> vals[0], 1 -> vals[0]
+  // File 1:               1 -> vals[1], 2 -> vals[1]
+  //
+  // Then L0->L1 compaction happens, which outputs keys as follows:
+  //
+  // File 0: 0 -> vals[0], 1 -> vals[1]
+  // File 1:               1 -> vals[0], 2 -> vals[1]
+  //
+  // DeleteFilesInRange shouldn't be allowed to drop just file 0, as that
+  // would cause `1 -> vals[0]` (an older key) to reappear.
+  std::string vals[kNumL0Files];
+  for (int i = 0; i < kNumL0Files; ++i) {
+    vals[i] = RandomString(&rnd, kValSize);
+    Put(Key(i), vals[i]);
+    Put(Key(i + 1), vals[i]);
+    Flush();
+    if (i == 0) {
+      snapshot = db_->GetSnapshot();
+    }
+  }
+  dbfull()->TEST_WaitForCompact();
+
+  // Verify `DeleteFilesInRange` can't drop only file 0 which would cause
+  // "1 -> vals[0]" to reappear.
+  std::string begin_str = Key(0), end_str = Key(1);
+  Slice begin = begin_str, end = end_str;
+  ASSERT_OK(DeleteFilesInRange(db_, db_->DefaultColumnFamily(), &begin, &end));
+  ASSERT_EQ(vals[1], Get(Key(1)));
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
@@ -2935,6 +3105,276 @@ TEST_F(DBCompactionTest, CompactBottomLevelFilesWithDeletions) {
  }
 }

+TEST_F(DBCompactionTest, CompactRangeDelayedByL0FileCount) {
+  // Verify that, when `CompactRangeOptions::allow_write_stall == false`, manual
+  // compaction only triggers flush after it's sure stall won't be triggered for
+  // L0 file count going too high.
+  const int kNumL0FilesTrigger = 4;
+  const int kNumL0FilesLimit = 8;
+  // i == 0: verifies normal case where stall is avoided by delay
+  // i == 1: verifies no delay in edge case where stall trigger is same as
+  //         compaction trigger, so stall can't be avoided
+  for (int i = 0; i < 2; ++i) {
+    Options options = CurrentOptions();
+    options.level0_slowdown_writes_trigger = kNumL0FilesLimit;
+    if (i == 0) {
+      options.level0_file_num_compaction_trigger = kNumL0FilesTrigger;
+    } else {
+      options.level0_file_num_compaction_trigger = kNumL0FilesLimit;
+    }
+    Reopen(options);
+
+    if (i == 0) {
+      // ensure the auto compaction doesn't finish until manual compaction has
+      // had a chance to be delayed.
+      rocksdb::SyncPoint::GetInstance()->LoadDependency(
+          {{"DBImpl::CompactRange:StallWait", "CompactionJob::Run():End"}});
+    } else {
+      // ensure the auto-compaction doesn't finish until manual compaction has
+      // continued without delay.
+      rocksdb::SyncPoint::GetInstance()->LoadDependency(
+          {{"DBImpl::CompactRange:StallWaitDone", "CompactionJob::Run():End"}});
+    }
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+    Random rnd(301);
+    for (int j = 0; j < kNumL0FilesLimit - 1; ++j) {
+      for (int k = 0; k < 2; ++k) {
+        ASSERT_OK(Put(Key(k), RandomString(&rnd, 1024)));
+      }
+      Flush();
+    }
+    auto manual_compaction_thread = port::Thread([this]() {
+      CompactRangeOptions cro;
+      cro.allow_write_stall = false;
+      db_->CompactRange(cro, nullptr, nullptr);
+    });
+
+    manual_compaction_thread.join();
+    dbfull()->TEST_WaitForCompact();
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_GT(NumTableFilesAtLevel(1), 0);
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  }
+}
+
+TEST_F(DBCompactionTest, CompactRangeDelayedByImmMemTableCount) {
+  // Verify that, when `CompactRangeOptions::allow_write_stall == false`, manual
+  // compaction only triggers flush after it's sure stall won't be triggered for
+  // immutable memtable count going too high.
+  const int kNumImmMemTableLimit = 8;
+  // i == 0: verifies normal case where stall is avoided by delay
+  // i == 1: verifies no delay in edge case where stall trigger is same as flush
+  //         trigger, so stall can't be avoided
+  for (int i = 0; i < 2; ++i) {
+    Options options = CurrentOptions();
+    options.disable_auto_compactions = true;
+    // the delay limit is one less than the stop limit. This test focuses on
+    // avoiding delay limit, but this option sets stop limit, so add one.
+    options.max_write_buffer_number = kNumImmMemTableLimit + 1;
+    if (i == 1) {
+      options.min_write_buffer_number_to_merge = kNumImmMemTableLimit;
+    }
+    Reopen(options);
+
+    if (i == 0) {
+      // ensure the flush doesn't finish until manual compaction has had a
+      // chance to be delayed.
+      rocksdb::SyncPoint::GetInstance()->LoadDependency(
+          {{"DBImpl::CompactRange:StallWait", "FlushJob::WriteLevel0Table"}});
+    } else {
+      // ensure the flush doesn't finish until manual compaction has continued
+      // without delay.
+      rocksdb::SyncPoint::GetInstance()->LoadDependency(
+          {{"DBImpl::CompactRange:StallWaitDone",
+            "FlushJob::WriteLevel0Table"}});
+    }
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+    Random rnd(301);
+    for (int j = 0; j < kNumImmMemTableLimit - 1; ++j) {
+      ASSERT_OK(Put(Key(0), RandomString(&rnd, 1024)));
+      FlushOptions flush_opts;
+      flush_opts.wait = false;
+      dbfull()->Flush(flush_opts);
+    }
+
+    auto manual_compaction_thread = port::Thread([this]() {
+      CompactRangeOptions cro;
+      cro.allow_write_stall = false;
+      db_->CompactRange(cro, nullptr, nullptr);
+    });
+
+    manual_compaction_thread.join();
+    dbfull()->TEST_WaitForFlushMemTable();
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_GT(NumTableFilesAtLevel(1), 0);
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  }
+}
+
+TEST_F(DBCompactionTest, CompactRangeShutdownWhileDelayed) {
+  // Verify that, when `CompactRangeOptions::allow_write_stall == false`, delay
+  // does not hang if CF is dropped or DB is closed
+  const int kNumL0FilesTrigger = 4;
+  const int kNumL0FilesLimit = 8;
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = kNumL0FilesTrigger;
+  options.level0_slowdown_writes_trigger = kNumL0FilesLimit;
+  // i == 0: DB::DropColumnFamily() on CompactRange's target CF unblocks it
+  // i == 1: DB::CancelAllBackgroundWork() unblocks CompactRange. This is to
+  //         simulate what happens during Close as we can't call Close (it
+  //         blocks on the auto-compaction, making a cycle).
+  for (int i = 0; i < 2; ++i) {
+    CreateAndReopenWithCF({"one"}, options);
+    // The calls to close CF/DB wait until the manual compaction stalls.
+    // The auto-compaction waits until the manual compaction finishes to ensure
+    // the signal comes from closing CF/DB, not from compaction making progress.
+    rocksdb::SyncPoint::GetInstance()->LoadDependency(
+        {{"DBImpl::CompactRange:StallWait",
+          "DBCompactionTest::CompactRangeShutdownWhileDelayed:PreShutdown"},
+         {"DBCompactionTest::CompactRangeShutdownWhileDelayed:PostManual",
+          "CompactionJob::Run():End"}});
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+    Random rnd(301);
+    for (int j = 0; j < kNumL0FilesLimit - 1; ++j) {
+      for (int k = 0; k < 2; ++k) {
+        ASSERT_OK(Put(1, Key(k), RandomString(&rnd, 1024)));
+      }
+      Flush(1);
+    }
+    auto manual_compaction_thread = port::Thread([this]() {
+      CompactRangeOptions cro;
+      cro.allow_write_stall = false;
+      ASSERT_TRUE(db_->CompactRange(cro, handles_[1], nullptr, nullptr)
+                      .IsShutdownInProgress());
+    });
+
+    TEST_SYNC_POINT(
+        "DBCompactionTest::CompactRangeShutdownWhileDelayed:PreShutdown");
+    if (i == 0) {
+      ASSERT_OK(db_->DropColumnFamily(handles_[1]));
+    } else {
+      dbfull()->CancelAllBackgroundWork(false /* wait */);
+    }
+    manual_compaction_thread.join();
+    TEST_SYNC_POINT(
+        "DBCompactionTest::CompactRangeShutdownWhileDelayed:PostManual");
+    dbfull()->TEST_WaitForCompact();
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  }
+}
+
+TEST_F(DBCompactionTest, CompactRangeSkipFlushAfterDelay) {
+  // Verify that, when `CompactRangeOptions::allow_write_stall == false`,
+  // CompactRange skips its flush if the delay is long enough that the memtables
+  // existing at the beginning of the call have already been flushed.
+  const int kNumL0FilesTrigger = 4;
+  const int kNumL0FilesLimit = 8;
+  Options options = CurrentOptions();
+  options.level0_slowdown_writes_trigger = kNumL0FilesLimit;
+  options.level0_file_num_compaction_trigger = kNumL0FilesTrigger;
+  Reopen(options);
+
+  Random rnd(301);
+  // The manual flush includes the memtable that was active when CompactRange
+  // began. So it unblocks CompactRange and precludes its flush. Throughout the
+  // test, stall conditions are upheld via high L0 file count.
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::CompactRange:StallWait",
+        "DBCompactionTest::CompactRangeSkipFlushAfterDelay:PreFlush"},
+       {"DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush",
+        "DBImpl::CompactRange:StallWaitDone"},
+       {"DBImpl::CompactRange:StallWaitDone", "CompactionJob::Run():End"}});
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  for (int i = 0; i < kNumL0FilesLimit - 1; ++i) {
+    for (int j = 0; j < 2; ++j) {
+      ASSERT_OK(Put(Key(j), RandomString(&rnd, 1024)));
+    }
+    Flush();
+  }
+  auto manual_compaction_thread = port::Thread([this]() {
+    CompactRangeOptions cro;
+    cro.allow_write_stall = false;
+    db_->CompactRange(cro, nullptr, nullptr);
+  });
+
+  TEST_SYNC_POINT("DBCompactionTest::CompactRangeSkipFlushAfterDelay:PreFlush");
+  Put(ToString(0), RandomString(&rnd, 1024));
+  Flush();
+  Put(ToString(0), RandomString(&rnd, 1024));
+  TEST_SYNC_POINT("DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush");
+  manual_compaction_thread.join();
+
+  // If CompactRange's flush was skipped, the final Put above will still be
+  // in the active memtable.
+  std::string num_keys_in_memtable;
+  db_->GetProperty(DB::Properties::kNumEntriesActiveMemTable, &num_keys_in_memtable);
+  ASSERT_EQ(ToString(1), num_keys_in_memtable);
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBCompactionTest, CompactRangeFlushOverlappingMemtable) {
+  // Verify memtable only gets flushed if it contains data overlapping the range
+  // provided to `CompactRange`. Tests all kinds of overlap/non-overlap.
+  const int kNumEndpointKeys = 5;
+  std::string keys[kNumEndpointKeys] = {"a", "b", "c", "d", "e"};
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  Reopen(options);
+
+  // One extra iteration for nullptr, which means left side of interval is
+  // unbounded.
+  for (int i = 0; i <= kNumEndpointKeys; ++i) {
+    Slice begin;
+    Slice* begin_ptr;
+    if (i == 0) {
+      begin_ptr = nullptr;
+    } else {
+      begin = keys[i - 1];
+      begin_ptr = &begin;
+    }
+    // Start at `i` so right endpoint comes after left endpoint. One extra
+    // iteration for nullptr, which means right side of interval is unbounded.
+    for (int j = std::max(0, i - 1); j <= kNumEndpointKeys; ++j) {
+      Slice end;
+      Slice* end_ptr;
+      if (j == kNumEndpointKeys) {
+        end_ptr = nullptr;
+      } else {
+        end = keys[j];
+        end_ptr = &end;
+      }
+      ASSERT_OK(Put("b", "val"));
+      ASSERT_OK(Put("d", "val"));
+      CompactRangeOptions compact_range_opts;
+      ASSERT_OK(db_->CompactRange(compact_range_opts, begin_ptr, end_ptr));
+
+      uint64_t get_prop_tmp, num_memtable_entries = 0;
+      ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kNumEntriesImmMemTables,
+                                      &get_prop_tmp));
+      num_memtable_entries += get_prop_tmp;
+      ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kNumEntriesActiveMemTable,
+                                      &get_prop_tmp));
+      num_memtable_entries += get_prop_tmp;
+      if (begin_ptr == nullptr || end_ptr == nullptr ||
+          (i <= 4 && j >= 1 && (begin != "c" || end != "c"))) {
+        // In this case `CompactRange`'s range overlapped in some way with the
+        // memtable's range, so flush should've happened. Then "b" and "d" won't
+        // be in the memtable.
+        ASSERT_EQ(0, num_memtable_entries);
+      } else {
+        ASSERT_EQ(2, num_memtable_entries);
+        // flush anyways to prepare for next iteration
+        db_->Flush(FlushOptions());
+      }
+    }
+  }
+}
+
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                        ::testing::Values(std::make_tuple(1, true),
                                          std::make_tuple(1, false),
@@ -57,6 +57,7 @@ Status DBImpl::EnableFileDeletions(bool force) {
      ROCKS_LOG_INFO(immutable_db_options_.info_log, "File Deletions Enabled");
      should_purge_files = true;
      FindObsoleteFiles(&job_context, true);
+      bg_cv_.SignalAll();
    } else {
      ROCKS_LOG_WARN(
          immutable_db_options_.info_log,
@@ -92,7 +93,7 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
      }
      cfd->Ref();
      mutex_.Unlock();
-      status = FlushMemTable(cfd, FlushOptions());
+      status = FlushMemTable(cfd, FlushOptions(), FlushReason::kGetLiveFiles);
      TEST_SYNC_POINT("DBImpl::GetLiveFiles:1");
      TEST_SYNC_POINT("DBImpl::GetLiveFiles:2");
      mutex_.Lock();
@@ -141,6 +142,18 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
 }

 Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
+  {
+    // If caller disabled deletions, this function should return files that are
+    // guaranteed not to be deleted until deletions are re-enabled. We need to
+    // wait for pending purges to finish since WalManager doesn't know which
+    // files are going to be purged. Additional purges won't be scheduled as
+    // long as deletions are disabled (so the below loop must terminate).
+    InstrumentedMutexLock l(&mutex_);
+    while (disable_delete_obsolete_files_ > 0 &&
+           pending_purge_obsolete_files_ > 0) {
+      bg_cv_.Wait();
+    }
+  }
  return wal_manager_.GetSortedWalFiles(files);
 }

@@ -126,6 +126,41 @@ TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
  ASSERT_EQ(1, num_compactions);
 }

+TEST_F(DBFlushTest, ManualFlushWithMinWriteBufferNumberToMerge) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 100;
+  options.max_write_buffer_number = 4;
+  options.min_write_buffer_number_to_merge = 3;
+  Reopen(options);
+
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::BGWorkFlush",
+        "DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge:1"},
+       {"DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge:2",
+        "FlushJob::WriteLevel0Table"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(Put("key1", "value1"));
+
+  port::Thread t([&]() {
+    // The call wait for flush to finish, i.e. with flush_options.wait = true.
+    ASSERT_OK(Flush());
+  });
+
+  // Wait for flush start.
+  TEST_SYNC_POINT("DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge:1");
+  // Insert a second memtable before the manual flush finish.
+  // At the end of the manual flush job, it will check if further flush
+  // is needed, but it will not trigger flush of the second memtable because
+  // min_write_buffer_number_to_merge is not reached.
+  ASSERT_OK(Put("key2", "value2"));
+  ASSERT_OK(dbfull()->TEST_SwitchMemtable());
+  TEST_SYNC_POINT("DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge:2");
+
+  // Manual flush should return, without waiting for flush indefinitely.
+  t.join();
+}
+
 TEST_P(DBFlushDirectIOTest, DirectIO) {
  Options options;
  options.create_if_missing = true;
@@ -150,6 +185,26 @@ TEST_P(DBFlushDirectIOTest, DirectIO) {
  delete options.env;
 }

+TEST_F(DBFlushTest, FlushError) {
+  Options options;
+  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
+      new FaultInjectionTestEnv(env_));
+  options.write_buffer_size = 100;
+  options.max_write_buffer_number = 4;
+  options.min_write_buffer_number_to_merge = 3;
+  options.disable_auto_compactions = true;
+  options.env = fault_injection_env.get();
+  Reopen(options);
+
+  ASSERT_OK(Put("key1", "value1"));
+  ASSERT_OK(Put("key2", "value2"));
+  fault_injection_env->SetFilesystemActive(false);
+  Status s = dbfull()->TEST_SwitchMemtable();
+  fault_injection_env->SetFilesystemActive(true);
+  Destroy(options);
+  ASSERT_NE(s, Status::OK());
+}
+
 INSTANTIATE_TEST_CASE_P(DBFlushDirectIOTest, DBFlushDirectIOTest,
                        testing::Bool());

@@ -106,7 +106,8 @@ CompressionType GetCompressionFlush(
  // optimization is used for leveled compaction. Otherwise the CPU and
  // latency overhead is not offset by saving much space.
  if (ioptions.compaction_style == kCompactionStyleUniversal) {
-    if (ioptions.compaction_options_universal.compression_size_percent < 0) {
+    if (mutable_cf_options.compaction_options_universal
+            .compression_size_percent < 0) {
      return mutable_cf_options.compression;
    } else {
      return kNoCompression;
@@ -122,13 +123,13 @@ CompressionType GetCompressionFlush(
 namespace {
 void DumpSupportInfo(Logger* logger) {
  ROCKS_LOG_HEADER(logger, "Compression algorithms supported:");
-  ROCKS_LOG_HEADER(logger, "\tSnappy supported: %d", Snappy_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZlib supported: %d", Zlib_Supported());
-  ROCKS_LOG_HEADER(logger, "\tBzip supported: %d", BZip2_Supported());
-  ROCKS_LOG_HEADER(logger, "\tLZ4 supported: %d", LZ4_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZSTDNotFinal supported: %d",
-                   ZSTDNotFinal_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZSTD supported: %d", ZSTD_Supported());
+  for (auto& compression : OptionsHelper::compression_type_string_map) {
+    if (compression.second != kNoCompression &&
+        compression.second != kDisableCompressionOption) {
+      ROCKS_LOG_HEADER(logger, "\t%s supported: %d", compression.first.c_str(),
+                       CompressionTypeSupported(compression.second));
+    }
+  }
  ROCKS_LOG_HEADER(logger, "Fast CRC32 supported: %s",
                   crc32c::IsFastCrc32Supported().c_str());
 }
@@ -136,9 +137,11 @@ void DumpSupportInfo(Logger* logger) {
 int64_t kDefaultLowPriThrottledRate = 2 * 1024 * 1024;
 } // namespace

-DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
+DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
+               const bool seq_per_batch)
    : env_(options.env),
      dbname_(dbname),
+      own_info_log_(options.info_log == nullptr),
      initial_db_options_(SanitizeOptions(dbname, options)),
      immutable_db_options_(initial_db_options_),
      mutable_db_options_(initial_db_options_),
@@ -175,6 +178,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
      num_running_flushes_(0),
      bg_purge_scheduled_(0),
      disable_delete_obsolete_files_(0),
+      pending_purge_obsolete_files_(0),
      delete_obsolete_files_last_run_(env_->NowMicros()),
      last_stats_dump_time_microsec_(0),
      next_job_id_(1),
@@ -185,19 +189,33 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
          env_options_, immutable_db_options_)),
      num_running_ingest_file_(0),
 #ifndef ROCKSDB_LITE
-      wal_manager_(immutable_db_options_, env_options_),
+      wal_manager_(immutable_db_options_, env_options_, seq_per_batch),
 #endif  // ROCKSDB_LITE
      event_logger_(immutable_db_options_.info_log.get()),
      bg_work_paused_(0),
      bg_compaction_paused_(0),
      refitting_level_(false),
      opened_successfully_(false),
-      concurrent_prepare_(options.concurrent_prepare),
+      two_write_queues_(options.two_write_queues),
      manual_wal_flush_(options.manual_wal_flush),
-      seq_per_batch_(options.seq_per_batch),
-      // TODO(myabandeh): revise this when we change options.seq_per_batch
-      use_custom_gc_(options.seq_per_batch),
-      preserve_deletes_(options.preserve_deletes) {
+      seq_per_batch_(seq_per_batch),
+      // 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
+      // enabled but seq_per_batch_ is disabled, last seq in memtable still
+      // indicates last published seq since wal-only writes that go to the 2nd
+      // queue do not consume a sequence number. Otherwise writes performed by
+      // the 2nd queue could change what is visible to the readers. In this
+      // cases, last_seq_same_as_publish_seq_==false, the 2nd queue maintains a
+      // separate variable to indicate the last published sequence.
+      last_seq_same_as_publish_seq_(
+          !(seq_per_batch && options.two_write_queues)),
+      // Since seq_per_batch_ is currently set only by WritePreparedTxn which
+      // requires a custom gc for compaction, we use that to set use_custom_gc_
+      // as well.
+      use_custom_gc_(seq_per_batch),
+      preserve_deletes_(options.preserve_deletes),
+      closed_(false) {
  env_->GetAbsolutePath(dbname, &db_absolute_path_);

  // Reserve ten files or so for other uses and give the rest to TableCache.
@@ -240,7 +258,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
      if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
        cfd->Ref();
        mutex_.Unlock();
-        FlushMemTable(cfd, FlushOptions());
+        FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
        mutex_.Lock();
        cfd->Unref();
      }
@@ -260,7 +278,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
  }
 }

-DBImpl::~DBImpl() {
+Status DBImpl::CloseHelper() {
  // CancelAllBackgroundWork called with false means we just set the shutdown
  // marker. After this we do a variant of the waiting and unschedule work
  // (to consider: moving all the waiting into CancelAllBackgroundWork(true))
@@ -269,6 +287,7 @@ DBImpl::~DBImpl() {
      env_->UnSchedule(this, Env::Priority::BOTTOM);
  int compactions_unscheduled = env_->UnSchedule(this, Env::Priority::LOW);
  int flushes_unscheduled = env_->UnSchedule(this, Env::Priority::HIGH);
+  Status ret;
  mutex_.Lock();
  bg_bottom_compaction_scheduled_ -= bottom_compactions_unscheduled;
  bg_compaction_scheduled_ -= compactions_unscheduled;
@@ -276,7 +295,8 @@ DBImpl::~DBImpl() {

  // Wait for background work to finish
  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
-         bg_flush_scheduled_ || bg_purge_scheduled_) {
+         bg_flush_scheduled_ || bg_purge_scheduled_ ||
+         pending_purge_obsolete_files_) {
    TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
    bg_cv_.Wait();
  }
@@ -330,7 +350,18 @@ DBImpl::~DBImpl() {
    delete l;
  }
  for (auto& log : logs_) {
-    log.ClearWriter();
+    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());
+      // Retain the first error
+      if (ret.ok()) {
+        ret = s;
+      }
+    }
  }
  logs_.clear();

@@ -363,6 +394,25 @@ DBImpl::~DBImpl() {

  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Shutdown complete");
  LogFlush(immutable_db_options_.info_log);
+
+  if (immutable_db_options_.info_log && own_info_log_) {
+    Status s = immutable_db_options_.info_log->Close();
+    if (ret.ok()) {
+      ret = s;
+    }
+  }
+  return ret;
+}
+
+Status DBImpl::CloseImpl() {
+  return CloseHelper();
+}
+
+DBImpl::~DBImpl() {
+  if (!closed_) {
+    closed_ = true;
+    CloseHelper();
+  }
 }

 void DBImpl::MaybeIgnoreError(Status* s) const {
@@ -508,6 +558,7 @@ Status DBImpl::SetOptions(ColumnFamilyHandle* column_family,

      persist_options_status = WriteOptionsFile(
          false /*need_mutex_lock*/, true /*need_enter_write_thread*/);
+      bg_cv_.SignalAll();
    }
  }
  sv_context.Clean();
@@ -574,13 +625,15 @@ Status DBImpl::SetDBOptions(
        new_options.bytes_per_sync = 1024 * 1024;
      }
      mutable_db_options_ = new_options;
-      env_options_for_compaction_ = EnvOptions(BuildDBOptions(
-                                                immutable_db_options_,
-                                                mutable_db_options_));
+      env_options_for_compaction_ = EnvOptions(
+          BuildDBOptions(immutable_db_options_, mutable_db_options_));
      env_options_for_compaction_ = env_->OptimizeForCompactionTableWrite(
-                                            env_options_for_compaction_,
-                                            immutable_db_options_);
+          env_options_for_compaction_, immutable_db_options_);
      versions_->ChangeEnvOptions(mutable_db_options_);
+      env_options_for_compaction_ = env_->OptimizeForCompactionTableRead(
+          env_options_for_compaction_, immutable_db_options_);
+      env_options_for_compaction_.compaction_readahead_size =
+          mutable_db_options_.compaction_readahead_size;
      write_thread_.EnterUnbatched(&w, &mutex_);
      if (total_log_size_ > GetMaxTotalWalSize() || wal_changed) {
        Status purge_wal_status = SwitchWAL(&write_context);
@@ -750,8 +803,8 @@ SequenceNumber DBImpl::GetLatestSequenceNumber() const {
  return versions_->LastSequence();
 }

-SequenceNumber DBImpl::IncAndFetchSequenceNumber() {
-  return versions_->FetchAddLastToBeWrittenSequence(1ull) + 1ull;
+void DBImpl::SetLastPublishedSequence(SequenceNumber seq) {
+  versions_->SetLastPublishedSequence(seq);
 }

 bool DBImpl::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) {
@@ -977,9 +1030,9 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
    // 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.
-    snapshot = concurrent_prepare_ && seq_per_batch_
-                   ? versions_->LastToBeWrittenSequence()
-                   : versions_->LastSequence();
+    snapshot = last_seq_same_as_publish_seq_
+                   ? versions_->LastSequence()
+                   : versions_->LastPublishedSequence();
  }
  TEST_SYNC_POINT("DBImpl::GetImpl:3");
  TEST_SYNC_POINT("DBImpl::GetImpl:4");
@@ -1070,9 +1123,9 @@ std::vector<Status> DBImpl::MultiGet(
    snapshot = reinterpret_cast<const SnapshotImpl*>(
        read_options.snapshot)->number_;
  } else {
-    snapshot = concurrent_prepare_ && seq_per_batch_
-                   ? versions_->LastToBeWrittenSequence()
-                   : versions_->LastSequence();
+    snapshot = last_seq_same_as_publish_seq_
+                   ? versions_->LastSequence()
+                   : versions_->LastPublishedSequence();
  }
  for (auto mgd_iter : multiget_cf_data) {
    mgd_iter.second->super_version =
@@ -1390,6 +1443,7 @@ Status DBImpl::DropColumnFamilyImpl(ColumnFamilyHandle* column_family) {
      }
      is_snapshot_supported_ = new_is_snapshot_supported;
    }
+    bg_cv_.SignalAll();
  }

  if (s.ok()) {
@@ -1478,8 +1532,9 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
        read_callback);
 #endif
  } else {
-    // Note: no need to consider the special case of concurrent_prepare_ &&
-    // seq_per_batch_ since NewIterator is overridden in WritePreparedTxnDB
+    // Note: no need to consider the special case of
+    // last_seq_same_as_publish_seq_==false since NewIterator is overridden in
+    // WritePreparedTxnDB
    auto snapshot = read_options.snapshot != nullptr
                        ? read_options.snapshot->GetSequenceNumber()
                        : versions_->LastSequence();
@@ -1492,7 +1547,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
                                            ColumnFamilyData* cfd,
                                            SequenceNumber snapshot,
                                            ReadCallback* read_callback,
-                                            bool allow_blob) {
+                                            bool allow_blob,
+                                            bool allow_refresh) {
  SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);

  // Try to generate a DB iterator tree in continuous memory area to be
@@ -1541,7 +1597,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
      env_, read_options, *cfd->ioptions(), snapshot,
      sv->mutable_cf_options.max_sequential_skip_in_iterations,
      sv->version_number, read_callback,
-      ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob);
+      ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob,
+      allow_refresh);

  InternalIterator* internal_iter =
      NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
@@ -1595,8 +1652,9 @@ Status DBImpl::NewIterators(
    }
 #endif
  } else {
-    // Note: no need to consider the special case of concurrent_prepare_ &&
-    // seq_per_batch_ since NewIterators is overridden in WritePreparedTxnDB
+    // Note: no need to consider the special case of
+    // last_seq_same_as_publish_seq_==false since NewIterators is overridden in
+    // WritePreparedTxnDB
    auto snapshot = read_options.snapshot != nullptr
                        ? read_options.snapshot->GetSequenceNumber()
                        : versions_->LastSequence();
@@ -1630,9 +1688,9 @@ const Snapshot* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
    delete s;
    return nullptr;
  }
-  auto snapshot_seq = concurrent_prepare_ && seq_per_batch_
-                          ? versions_->LastToBeWrittenSequence()
-                          : versions_->LastSequence();
+  auto snapshot_seq = last_seq_same_as_publish_seq_
+                          ? versions_->LastSequence()
+                          : versions_->LastPublishedSequence();
  return snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
 }

@@ -1643,9 +1701,9 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
    snapshots_.Delete(casted_s);
    uint64_t oldest_snapshot;
    if (snapshots_.empty()) {
-      oldest_snapshot = concurrent_prepare_ && seq_per_batch_
-                            ? versions_->LastToBeWrittenSequence()
-                            : versions_->LastSequence();
+      oldest_snapshot = last_seq_same_as_publish_seq_
+                            ? versions_->LastSequence()
+                            : versions_->LastPublishedSequence();
    } else {
      oldest_snapshot = snapshots_.oldest()->number_;
    }
@@ -1663,14 +1721,6 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
  delete casted_s;
 }

-bool DBImpl::HasActiveSnapshotLaterThanSN(SequenceNumber sn) {
-  InstrumentedMutexLock l(&mutex_);
-  if (snapshots_.empty()) {
-    return false;
-  }
-  return (snapshots_.newest()->GetSequenceNumber() >= sn);
-}
-
 #ifndef ROCKSDB_LITE
 Status DBImpl::GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
                                        TablePropertiesCollection* props) {
@@ -2034,8 +2084,7 @@ Status DBImpl::DeleteFile(std::string name) {
                      name.c_str());
      return Status::NotSupported("Delete only supported for archived logs");
    }
-    status =
-        env_->DeleteFile(immutable_db_options_.wal_dir + "/" + name.c_str());
+    status = wal_manager_.DeleteFile(name, number);
    if (!status.ok()) {
      ROCKS_LOG_ERROR(immutable_db_options_.info_log,
                      "DeleteFile %s failed -- %s.\n", name.c_str(),
@@ -2114,56 +2163,61 @@ Status DBImpl::DeleteFile(std::string name) {
  return status;
 }

-Status DBImpl::DeleteFilesInRange(ColumnFamilyHandle* column_family,
-                                  const Slice* begin, const Slice* end) {
+Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
+                                   const RangePtr* ranges, size_t n,
+                                   bool include_end) {
  Status status;
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  ColumnFamilyData* cfd = cfh->cfd();
  VersionEdit edit;
-  std::vector<FileMetaData*> deleted_files;
+  std::set<FileMetaData*> deleted_files;
  JobContext job_context(next_job_id_.fetch_add(1), true);
  {
    InstrumentedMutexLock l(&mutex_);
    Version* input_version = cfd->current();

    auto* vstorage = input_version->storage_info();
-    for (int i = 1; i < cfd->NumberLevels(); i++) {
-      if (vstorage->LevelFiles(i).empty() ||
-          !vstorage->OverlapInLevel(i, begin, end)) {
-        continue;
-      }
-      std::vector<FileMetaData*> level_files;
-      InternalKey begin_storage, end_storage, *begin_key, *end_key;
-      if (begin == nullptr) {
-        begin_key = nullptr;
-      } else {
-        begin_storage.SetMinPossibleForUserKey(*begin);
-        begin_key = &begin_storage;
-      }
-      if (end == nullptr) {
-        end_key = nullptr;
-      } else {
-        end_storage.SetMaxPossibleForUserKey(*end);
-        end_key = &end_storage;
-      }
+    for (size_t r = 0; r < n; r++) {
+      auto begin = ranges[r].start, end = ranges[r].limit;
+      for (int i = 1; i < cfd->NumberLevels(); i++) {
+        if (vstorage->LevelFiles(i).empty() ||
+            !vstorage->OverlapInLevel(i, begin, end)) {
+          continue;
+        }
+        std::vector<FileMetaData*> level_files;
+        InternalKey begin_storage, end_storage, *begin_key, *end_key;
+        if (begin == nullptr) {
+          begin_key = nullptr;
+        } else {
+          begin_storage.SetMinPossibleForUserKey(*begin);
+          begin_key = &begin_storage;
+        }
+        if (end == nullptr) {
+          end_key = nullptr;
+        } else {
+          end_storage.SetMaxPossibleForUserKey(*end);
+          end_key = &end_storage;
+        }

-      vstorage->GetOverlappingInputs(i, begin_key, end_key, &level_files, -1,
-                                     nullptr, false);
-      FileMetaData* level_file;
-      for (uint32_t j = 0; j < level_files.size(); j++) {
-        level_file = level_files[j];
-        if (((begin == nullptr) ||
-             (cfd->internal_comparator().user_comparator()->Compare(
-                  level_file->smallest.user_key(), *begin) >= 0)) &&
-            ((end == nullptr) ||
-             (cfd->internal_comparator().user_comparator()->Compare(
-                  level_file->largest.user_key(), *end) <= 0))) {
+        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];
          if (level_file->being_compacted) {
            continue;
          }
+          if (deleted_files.find(level_file) != deleted_files.end()) {
+            continue;
+          }
+          if (!include_end && end != nullptr &&
+              cfd->user_comparator()->Compare(level_file->largest.user_key(), *end) == 0) {
+            continue;
+          }
          edit.SetColumnFamily(cfd->GetID());
          edit.DeleteFile(i, level_file->fd.GetNumber());
-          deleted_files.push_back(level_file);
+          deleted_files.insert(level_file);
          level_file->being_compacted = true;
        }
      }
@@ -2314,7 +2368,15 @@ Status DB::DestroyColumnFamilyHandle(ColumnFamilyHandle* column_family) {
  return Status::OK();
 }

-DB::~DB() { }
+DB::~DB() {}
+
+Status DBImpl::Close() {
+  if (!closed_) {
+    closed_ = true;
+    return CloseImpl();
+  }
+  return Status::OK();
+}

 Status DB::ListColumnFamilies(const DBOptions& db_options,
                              const std::string& name,
@@ -2755,7 +2817,7 @@ Status DBImpl::IngestExternalFile(
    WriteThread::Writer w;
    write_thread_.EnterUnbatched(&w, &mutex_);
    WriteThread::Writer nonmem_w;
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
    }

@@ -2770,12 +2832,14 @@ Status DBImpl::IngestExternalFile(
    // Figure out if we need to flush the memtable first
    if (status.ok()) {
      bool need_flush = false;
-      status = ingestion_job.NeedsFlush(&need_flush);
+      status = ingestion_job.NeedsFlush(&need_flush, cfd->GetSuperVersion());
      TEST_SYNC_POINT_CALLBACK("DBImpl::IngestExternalFile:NeedFlush",
                               &need_flush);
      if (status.ok() && need_flush) {
        mutex_.Unlock();
-        status = FlushMemTable(cfd, FlushOptions(), true /* writes_stopped */);
+        status = FlushMemTable(cfd, FlushOptions(),
+                               FlushReason::kExternalFileIngestion,
+                               true /* writes_stopped */);
        mutex_.Lock();
      }
    }
@@ -2798,7 +2862,7 @@ Status DBImpl::IngestExternalFile(
    }

    // Resume writes to the DB
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      nonmem_write_thread_.ExitUnbatched(&nonmem_w);
    }
    write_thread_.ExitUnbatched(&w);
@@ -28,6 +28,7 @@
 #include "db/flush_scheduler.h"
 #include "db/internal_stats.h"
 #include "db/log_writer.h"
+#include "db/pre_release_callback.h"
 #include "db/read_callback.h"
 #include "db/snapshot_checker.h"
 #include "db/snapshot_impl.h"
@@ -68,7 +69,8 @@ struct MemTableInfo;

 class DBImpl : public DB {
 public:
-  DBImpl(const DBOptions& options, const std::string& dbname);
+  DBImpl(const DBOptions& options, const std::string& dbname,
+         const bool seq_per_batch = false);
  virtual ~DBImpl();

  // Implementations of the DB interface
@@ -146,7 +148,8 @@ class DBImpl : public DB {
                                      ColumnFamilyData* cfd,
                                      SequenceNumber snapshot,
                                      ReadCallback* read_callback,
-                                      bool allow_blob = false);
+                                      bool allow_blob = false,
+                                      bool allow_refresh = true);

  virtual const Snapshot* GetSnapshot() override;
  virtual void ReleaseSnapshot(const Snapshot* snapshot) override;
@@ -219,16 +222,14 @@ class DBImpl : public DB {
  virtual Status SyncWAL() override;

  virtual SequenceNumber GetLatestSequenceNumber() const override;
-  virtual SequenceNumber IncAndFetchSequenceNumber();
-  // Returns LastToBeWrittenSequence in concurrent_prepare_ && seq_per_batch_
-  // mode and LastSequence otherwise. This is useful when visiblility depends
-  // also on data written to the WAL but not to the memtable.
-  SequenceNumber TEST_GetLatestVisibleSequenceNumber() const;
+  virtual void SetLastPublishedSequence(SequenceNumber seq);
+  // Returns LastSequence in last_seq_same_as_publish_seq_
+  // mode and LastAllocatedSequence otherwise. This is useful when visiblility
+  // depends also on data written to the WAL but not to the memtable.
+  SequenceNumber TEST_GetLastVisibleSequence() const;

  virtual bool SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) override;

-  bool HasActiveSnapshotLaterThanSN(SequenceNumber sn);
-
 #ifndef ROCKSDB_LITE
  using DB::ResetStats;
  virtual Status ResetStats() override;
@@ -246,8 +247,9 @@ class DBImpl : public DB {
      const TransactionLogIterator::ReadOptions&
          read_options = TransactionLogIterator::ReadOptions()) override;
  virtual Status DeleteFile(std::string name) override;
-  Status DeleteFilesInRange(ColumnFamilyHandle* column_family,
-                            const Slice* begin, const Slice* end);
+  Status DeleteFilesInRanges(ColumnFamilyHandle* column_family,
+                             const RangePtr* ranges, size_t n,
+                             bool include_end = true);

  virtual void GetLiveFilesMetaData(
      std::vector<LiveFileMetaData>* metadata) override;
@@ -467,6 +469,8 @@ class DBImpl : public DB {
  // belong to live files are posibly 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.
+  // If FindObsoleteFiles() was run, we need to also run
+  // PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
  void PurgeObsoleteFiles(const JobContext& background_contet,
                          bool schedule_only = false);

@@ -604,10 +608,20 @@ class DBImpl : public DB {

  Status NewDB();

+  // This is to be used only by internal rocksdb classes.
+  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);
+
+  virtual Status Close() override;
+
 protected:
  Env* const env_;
  const std::string dbname_;
  unique_ptr<VersionSet> versions_;
+  // Flag to check whether we allocated and own the info log file
+  bool own_info_log_;
  const DBOptions initial_db_options_;
  const ImmutableDBOptions immutable_db_options_;
  MutableDBOptions mutable_db_options_;
@@ -653,10 +667,21 @@ 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.
  Status WriteImpl(const WriteOptions& options, WriteBatch* updates,
                   WriteCallback* callback = nullptr,
                   uint64_t* log_used = nullptr, uint64_t log_ref = 0,
-                   bool disable_memtable = false, uint64_t* seq_used = nullptr);
+                   bool disable_memtable = false, uint64_t* seq_used = nullptr,
+                   size_t batch_cnt = 0,
+                   PreReleaseCallback* pre_release_callback = nullptr);

  Status PipelinedWriteImpl(const WriteOptions& options, WriteBatch* updates,
                            WriteCallback* callback = nullptr,
@@ -664,10 +689,14 @@ class DBImpl : public DB {
                            bool disable_memtable = false,
                            uint64_t* seq_used = nullptr);

+  // 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 WriteImplWALOnly(const WriteOptions& options, WriteBatch* updates,
                          WriteCallback* callback = nullptr,
                          uint64_t* log_used = nullptr, uint64_t log_ref = 0,
-                          uint64_t* seq_used = nullptr);
+                          uint64_t* seq_used = nullptr, size_t batch_cnt = 0,
+                          PreReleaseCallback* pre_release_callback = nullptr);

  uint64_t FindMinLogContainingOutstandingPrep();
  uint64_t FindMinPrepLogReferencedByMemTable();
@@ -675,6 +704,9 @@ class DBImpl : public DB {
  // The writer must be the leader in write_thread_ and holding mutex_
  Status WriteRecoverableState();

+  // Actual implementation of Close()
+  Status CloseImpl();
+
 private:
  friend class DB;
  friend class InternalStats;
@@ -690,6 +722,7 @@ class DBImpl : public DB {
  friend class CompactedDBImpl;
 #ifndef NDEBUG
  friend class DBTest2_ReadCallbackTest_Test;
+  friend class WriteCallbackTest_WriteWithCallbackTest_Test;
  friend class XFTransactionWriteHandler;
  friend class DBBlobIndexTest;
 #endif
@@ -789,10 +822,14 @@ class DBImpl : public DB {

  // Force current memtable contents to be flushed.
  Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options,
-                       bool writes_stopped = false);
+                       FlushReason flush_reason, bool writes_stopped = false);

-  // Wait for memtable flushed
-  Status WaitForFlushMemTable(ColumnFamilyData* cfd);
+  // Wait for memtable flushed.
+  // If flush_memtable_id is non-null, wait until the memtable with the ID
+  // gets flush. Otherwise, wait until the column family don't have any
+  // memtable pending flush.
+  Status WaitForFlushMemTable(ColumnFamilyData* cfd,
+                              const uint64_t* flush_memtable_id = nullptr);

  // REQUIRES: mutex locked
  Status SwitchWAL(WriteContext* write_context);
@@ -847,7 +884,7 @@ class DBImpl : public DB {
  ColumnFamilyData* GetColumnFamilyDataByName(const std::string& cf_name);

  void MaybeScheduleFlushOrCompaction();
-  void SchedulePendingFlush(ColumnFamilyData* cfd);
+  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);
@@ -886,7 +923,7 @@ class DBImpl : public DB {
  // helper functions for adding and removing from flush & compaction queues
  void AddToCompactionQueue(ColumnFamilyData* cfd);
  ColumnFamilyData* PopFirstFromCompactionQueue();
-  void AddToFlushQueue(ColumnFamilyData* cfd);
+  void AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason);
  ColumnFamilyData* PopFirstFromFlushQueue();

  // helper function to call after some of the logs_ were synced
@@ -896,6 +933,8 @@ class DBImpl : public DB {

  uint64_t GetMaxTotalWalSize() const;

+  Status CloseHelper();
+
  // table_cache_ provides its own synchronization
  std::shared_ptr<Cache> table_cache_;

@@ -903,12 +942,12 @@ class DBImpl : public DB {
  FileLock* db_lock_;

  // In addition to mutex_, log_write_mutex_ protected writes to logs_ and
-  // logfile_number_. With concurrent_prepare it also protects alive_log_files_,
+  // logfile_number_. With two_write_queues it also protects alive_log_files_,
  // and log_empty_. Refer to the definition of each variable below for more
  // details.
  InstrumentedMutex log_write_mutex_;
  // State below is protected by mutex_
-  // With concurrent_prepare enabled, some of the variables that accessed during
+  // With two_write_queues enabled, some of the variables that accessed during
  // WriteToWAL need different synchronization: log_empty_, alive_log_files_,
  // logs_, logfile_number_. Refer to the definition of each variable below for
  // more description.
@@ -924,6 +963,10 @@ class DBImpl : public DB {
  // (i.e. whenever a flush is done, even if it didn't make any progress)
  // * whenever there is an error in background purge, flush or compaction
  // * whenever num_running_ingest_file_ goes to 0.
+  // * whenever pending_purge_obsolete_files_ goes to 0.
+  // * whenever disable_delete_obsolete_files_ goes to 0.
+  // * whenever SetOptions successfully updates options.
+  // * whenever a column family is dropped.
  InstrumentedCondVar bg_cv_;
  // Writes are protected by locking both mutex_ and log_write_mutex_, and reads
  // must be under either mutex_ or log_write_mutex_. Since after ::Open,
@@ -931,12 +974,12 @@ class DBImpl : public DB {
  // from the same write_thread_ without any locks.
  uint64_t logfile_number_;
  std::deque<uint64_t>
-      log_recycle_files;  // a list of log files that we can recycle
+      log_recycle_files_;  // a list of log files that we can recycle
  bool log_dir_synced_;
-  // Without concurrent_prepare, read and writes to log_empty_ are protected by
+  // Without two_write_queues, read and writes to log_empty_ are protected by
  // mutex_. Since it is currently updated/read only in write_thread_, it can be
  // accessed from the same write_thread_ without any locks. With
-  // concurrent_prepare writes, where it can be updated in different threads,
+  // two_write_queues writes, where it can be updated in different threads,
  // read and writes are protected by log_write_mutex_ instead. This is to avoid
  // expesnive mutex_ lock during WAL write, which update log_empty_.
  bool log_empty_;
@@ -961,9 +1004,11 @@ class DBImpl : public DB {
      writer = nullptr;
      return w;
    }
-    void ClearWriter() {
+    Status ClearWriter() {
+      Status s = writer->WriteBuffer();
      delete writer;
      writer = nullptr;
+      return s;
    }

    uint64_t number;
@@ -973,10 +1018,10 @@ class DBImpl : public DB {
    // true for some prefix of logs_
    bool getting_synced = false;
  };
-  // Without concurrent_prepare, read and writes to alive_log_files_ are
+  // Without two_write_queues, read and writes to alive_log_files_ are
  // protected by mutex_. However since back() is never popped, and push_back()
  // is done only from write_thread_, the same thread can access the item
-  // reffered by back() without mutex_. With concurrent_prepare_, writes
+  // reffered by back() without mutex_. With two_write_queues_, writes
  // are protected by locking both mutex_ and log_write_mutex_, and reads must
  // be under either mutex_ or log_write_mutex_.
  std::deque<LogFileNumberSize> alive_log_files_;
@@ -1001,7 +1046,7 @@ class DBImpl : public DB {
  // memtable on normal writes and hence improving the throughput. Each new
  // write of the state will replace the previous state entirely even if the
  // keys in the two consecuitive states do not overlap.
-  // It is protected by log_write_mutex_ when concurrent_prepare_ is enabled.
+  // It is protected by log_write_mutex_ when two_write_queues_ is enabled.
  // Otherwise only the heaad of write_thread_ can access it.
  WriteBatch cached_recoverable_state_;
  std::atomic<bool> cached_recoverable_state_empty_ = {true};
@@ -1190,6 +1235,10 @@ class DBImpl : public DB {
  // without any synchronization
  int disable_delete_obsolete_files_;

+  // Number of times FindObsoleteFiles has found deletable files and the
+  // corresponding call to PurgeObsoleteFiles has not yet finished.
+  int pending_purge_obsolete_files_;
+
  // last time when DeleteObsoleteFiles with full scan was executed. Originaly
  // initialized with startup time.
  uint64_t delete_obsolete_files_last_run_;
@@ -1314,12 +1363,27 @@ class DBImpl : public DB {
  bool MCOverlap(ManualCompactionState* m, ManualCompactionState* m1);

  size_t GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
+  Env::WriteLifeTimeHint CalculateWALWriteHint() {
+    return Env::WLTH_SHORT;
+  }

-  // When set, we use a seprate queue for writes that dont write to memtable. In
+  // When set, we use a separate queue for writes that dont write to memtable. In
  // 2PC these are the writes at Prepare phase.
-  const bool concurrent_prepare_;
+  const bool two_write_queues_;
  const bool manual_wal_flush_;
+  // Increase the sequence number after writing each batch, whether memtable is
+  // disabled for that or not. Otherwise the sequence number is increased after
+  // writing each key into memtable. This implies that when disable_memtable is
+  // set, the seq is not increased at all.
+  //
+  // Default: false
  const bool seq_per_batch_;
+  // LastSequence also indicates last published sequence visibile to the
+  // readers. Otherwise LastPublishedSequence should be used.
+  const bool last_seq_same_as_publish_seq_;
+  // It indicates that a customized gc algorithm must be used for
+  // flush/compaction and if it is not provided vis SnapshotChecker, we should
+  // disable gc to be safe.
  const bool use_custom_gc_;

  // Clients must periodically call SetPreserveDeletesSequenceNumber()
@@ -1328,6 +1392,9 @@ class DBImpl : public DB {
  // is set to false.
  std::atomic<SequenceNumber> preserve_deletes_seqnum_;
  const bool preserve_deletes_;
+
+  // Flag to check whether Close() has been called on this DB
+  bool closed_;
 };

 extern Options SanitizeOptions(const std::string& db,
@@ -23,6 +23,7 @@
 #include "util/sync_point.h"

 namespace rocksdb {
+
 Status DBImpl::SyncClosedLogs(JobContext* job_context) {
  TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
  mutex_.AssertHeld();
@@ -222,6 +223,7 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
    info.smallest_seqno = file_meta->smallest_seqno;
    info.largest_seqno = file_meta->largest_seqno;
    info.table_properties = prop;
+    info.flush_reason = cfd->GetFlushReason();
    for (auto listener : immutable_db_options_.listeners) {
      listener->OnFlushBegin(this, info);
    }
@@ -266,6 +268,7 @@ void DBImpl::NotifyOnFlushCompleted(ColumnFamilyData* cfd,
    info.smallest_seqno = file_meta->smallest_seqno;
    info.largest_seqno = file_meta->largest_seqno;
    info.table_properties = prop;
+    info.flush_reason = cfd->GetFlushReason();
    for (auto listener : immutable_db_options_.listeners) {
      listener->OnFlushCompleted(this, info);
    }
@@ -287,10 +290,75 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
  auto cfd = cfh->cfd();
  bool exclusive = options.exclusive_manual_compaction;

-  Status s = FlushMemTable(cfd, FlushOptions());
-  if (!s.ok()) {
-    LogFlush(immutable_db_options_.info_log);
-    return s;
+  bool flush_needed = true;
+  if (begin != nullptr && end != nullptr) {
+    // TODO(ajkr): We could also optimize away the flush in certain cases where
+    // one/both sides of the interval are unbounded. But it requires more
+    // changes to RangesOverlapWithMemtables.
+    Range range(*begin, *end);
+    SuperVersion* super_version = cfd->GetReferencedSuperVersion(&mutex_);
+    cfd->RangesOverlapWithMemtables({range}, super_version, &flush_needed);
+    CleanupSuperVersion(super_version);
+  }
+
+  if (!options.allow_write_stall && flush_needed) {
+    InstrumentedMutexLock l(&mutex_);
+    uint64_t orig_active_memtable_id = cfd->mem()->GetID();
+    WriteStallCondition write_stall_condition = WriteStallCondition::kNormal;
+    do {
+      if (write_stall_condition != WriteStallCondition::kNormal) {
+        TEST_SYNC_POINT("DBImpl::CompactRange:StallWait");
+        ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                       "[%s] CompactRange waiting on stall conditions to clear",
+                       cfd->GetName().c_str());
+        bg_cv_.Wait();
+      }
+      if (cfd->IsDropped() || shutting_down_.load(std::memory_order_acquire)) {
+        return Status::ShutdownInProgress();
+      }
+
+      uint64_t earliest_memtable_id =
+          std::min(cfd->mem()->GetID(), cfd->imm()->GetEarliestMemTableID());
+      if (earliest_memtable_id > orig_active_memtable_id) {
+        // We waited so long that the memtable we were originally waiting on was
+        // flushed.
+        flush_needed = false;
+        break;
+      }
+
+      const auto& mutable_cf_options = *cfd->GetLatestMutableCFOptions();
+      const auto* vstorage = cfd->current()->storage_info();
+
+      // Skip stalling check if we're below auto-flush and auto-compaction
+      // triggers. If it stalled in these conditions, that'd mean the stall
+      // triggers are so low that stalling is needed for any background work. In
+      // that case we shouldn't wait since background work won't be scheduled.
+      if (cfd->imm()->NumNotFlushed() <
+              cfd->ioptions()->min_write_buffer_number_to_merge &&
+          vstorage->l0_delay_trigger_count() <
+              mutable_cf_options.level0_file_num_compaction_trigger) {
+        break;
+      }
+
+      // check whether one extra immutable memtable or an extra L0 file would
+      // cause write stalling mode to be entered. It could still enter stall
+      // mode due to pending compaction bytes, but that's less common
+      write_stall_condition =
+          ColumnFamilyData::GetWriteStallConditionAndCause(
+              cfd->imm()->NumNotFlushed() + 1,
+              vstorage->l0_delay_trigger_count() + 1,
+              vstorage->estimated_compaction_needed_bytes(), mutable_cf_options)
+              .first;
+    } while (write_stall_condition != WriteStallCondition::kNormal);
+  }
+  TEST_SYNC_POINT("DBImpl::CompactRange:StallWaitDone");
+  Status s;
+  if (flush_needed) {
+    s = FlushMemTable(cfd, FlushOptions(), FlushReason::kManualCompaction);
+    if (!s.ok()) {
+      LogFlush(immutable_db_options_.info_log);
+      return s;
+    }
  }

  int max_level_with_files = 0;
@@ -441,7 +509,8 @@ Status DBImpl::CompactFiles(
  }  // release the mutex

  // delete unnecessary files if any, this is done outside the mutex
-  if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
+  if (job_context.HaveSomethingToClean() ||
+      job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
    // Have to flush the info logs before bg_compaction_scheduled_--
    // because if bg_flush_scheduled_ becomes 0 and the lock is
    // released, the deconstructor of DB can kick in and destroy all the
@@ -544,10 +613,10 @@ 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_, c->mutable_cf_options()->paranoid_file_checks,
+      directories_.GetDataDir(c->output_path_id()), stats_, &mutex_, &bg_error_,
+      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
                 // CompactFiles does not trigger OnCompactionCompleted(),
@@ -809,7 +878,14 @@ int DBImpl::Level0StopWriteTrigger(ColumnFamilyHandle* column_family) {
 Status DBImpl::Flush(const FlushOptions& flush_options,
                     ColumnFamilyHandle* column_family) {
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-  return FlushMemTable(cfh->cfd(), 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);
+  ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                 "[%s] Manual flush finished, status: %s\n",
+                 cfh->GetName().c_str(), s.ToString().c_str());
+  return s;
 }

 Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
@@ -942,8 +1018,9 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,

 Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
                             const FlushOptions& flush_options,
-                             bool writes_stopped) {
+                             FlushReason flush_reason, bool writes_stopped) {
  Status s;
+  uint64_t flush_memtable_id = 0;
  {
    WriteContext context;
    InstrumentedMutexLock guard_lock(&mutex_);
@@ -961,6 +1038,7 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,

    // SwitchMemtable() will release and reacquire mutex during execution
    s = SwitchMemtable(cfd, &context);
+    flush_memtable_id = cfd->imm()->GetLatestMemTableID();

    if (!writes_stopped) {
      write_thread_.ExitUnbatched(&w);
@@ -969,22 +1047,25 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
    cfd->imm()->FlushRequested();

    // schedule flush
-    SchedulePendingFlush(cfd);
+    SchedulePendingFlush(cfd, flush_reason);
    MaybeScheduleFlushOrCompaction();
  }

  if (s.ok() && flush_options.wait) {
    // Wait until the compaction completes
-    s = WaitForFlushMemTable(cfd);
+    s = WaitForFlushMemTable(cfd, &flush_memtable_id);
  }
  return s;
 }

-Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd) {
+Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd,
+                                    const uint64_t* flush_memtable_id) {
  Status s;
  // Wait until the compaction completes
  InstrumentedMutexLock l(&mutex_);
-  while (cfd->imm()->NumNotFlushed() > 0 && bg_error_.ok()) {
+  while (cfd->imm()->NumNotFlushed() > 0 && bg_error_.ok() &&
+         (flush_memtable_id == nullptr ||
+          cfd->imm()->GetEarliestMemTableID() <= *flush_memtable_id)) {
    if (shutting_down_.load(std::memory_order_acquire)) {
      return Status::ShutdownInProgress();
    }
@@ -1059,6 +1140,7 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
  if (HasExclusiveManualCompaction()) {
    // only manual compactions are allowed to run. don't schedule automatic
    // compactions
+    TEST_SYNC_POINT("DBImpl::MaybeScheduleFlushOrCompaction:Conflict");
    return;
  }

@@ -1121,11 +1203,12 @@ ColumnFamilyData* DBImpl::PopFirstFromCompactionQueue() {
  return cfd;
 }

-void DBImpl::AddToFlushQueue(ColumnFamilyData* cfd) {
+void DBImpl::AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason) {
  assert(!cfd->pending_flush());
  cfd->Ref();
  flush_queue_.push_back(cfd);
  cfd->set_pending_flush(true);
+  cfd->SetFlushReason(flush_reason);
 }

 ColumnFamilyData* DBImpl::PopFirstFromFlushQueue() {
@@ -1134,12 +1217,14 @@ ColumnFamilyData* DBImpl::PopFirstFromFlushQueue() {
  flush_queue_.pop_front();
  assert(cfd->pending_flush());
  cfd->set_pending_flush(false);
+  // TODO: need to unset flush reason?
  return cfd;
 }

-void DBImpl::SchedulePendingFlush(ColumnFamilyData* cfd) {
+void DBImpl::SchedulePendingFlush(ColumnFamilyData* cfd,
+                                  FlushReason flush_reason) {
  if (!cfd->pending_flush() && cfd->imm()->IsFlushPending()) {
-    AddToFlushQueue(cfd);
+    AddToFlushQueue(cfd, flush_reason);
    ++unscheduled_flushes_;
  }
 }
@@ -1303,8 +1388,10 @@ void DBImpl::BackgroundCallFlush() {
    // created. Thus, we force full scan in FindObsoleteFiles()
    FindObsoleteFiles(&job_context, !s.ok() && !s.IsShutdownInProgress());
    // delete unnecessary files if any, this is done outside the mutex
-    if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
+    if (job_context.HaveSomethingToClean() ||
+        job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
      mutex_.Unlock();
+      TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:FilesFound");
      // Have to flush the info logs before bg_flush_scheduled_--
      // because if bg_flush_scheduled_ becomes 0 and the lock is
      // released, the deconstructor of DB can kick in and destroy all the
@@ -1384,7 +1471,8 @@ void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
    FindObsoleteFiles(&job_context, !s.ok() && !s.IsShutdownInProgress());

    // delete unnecessary files if any, this is done outside the mutex
-    if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
+    if (job_context.HaveSomethingToClean() ||
+        job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
      mutex_.Unlock();
      // Have to flush the info logs before bg_compaction_scheduled_--
      // because if bg_flush_scheduled_ becomes 0 and the lock is
@@ -1503,7 +1591,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
               : m->manual_end->DebugString().c_str()));
    }
  } else if (!is_prepicked && !compaction_queue_.empty()) {
-    if (HaveManualCompaction(compaction_queue_.front())) {
+    if (HasExclusiveManualCompaction()) {
      // Can't compact right now, but try again later
      TEST_SYNC_POINT("DBImpl::BackgroundCompaction()::Conflict");

@@ -1679,7 +1767,7 @@ 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);
@@ -1913,7 +2001,7 @@ void DBImpl::InstallSuperVersionAndScheduleWork(

  // Whenever we install new SuperVersion, we might need to issue new flushes or
  // compactions.
-  SchedulePendingFlush(cfd);
+  SchedulePendingFlush(cfd, FlushReason::kSuperVersionChange);
  SchedulePendingCompaction(cfd);
  MaybeScheduleFlushOrCompaction();

@@ -103,7 +103,7 @@ Status DBImpl::TEST_FlushMemTable(bool wait, ColumnFamilyHandle* cfh) {
    auto cfhi = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh);
    cfd = cfhi->cfd();
  }
-  return FlushMemTable(cfd, fo);
+  return FlushMemTable(cfd, fo, FlushReason::kTest);
 }

 Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
@@ -209,11 +209,11 @@ int DBImpl::TEST_BGFlushesAllowed() const {
  return GetBGJobLimits().max_flushes;
 }

-SequenceNumber DBImpl::TEST_GetLatestVisibleSequenceNumber() const {
-  if (concurrent_prepare_ && seq_per_batch_) {
-    return versions_->LastToBeWrittenSequence();
-  } else {
+SequenceNumber DBImpl::TEST_GetLastVisibleSequence() const {
+  if (last_seq_same_as_publish_seq_) {
    return versions_->LastSequence();
+  } else {
+    return versions_->LastAllocatedSequence();
  }
 }

@@ -238,11 +238,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
    while (alive_log_files_.begin()->number < min_log_number) {
      auto& earliest = *alive_log_files_.begin();
      if (immutable_db_options_.recycle_log_file_num >
-          log_recycle_files.size()) {
+          log_recycle_files_.size()) {
        ROCKS_LOG_INFO(immutable_db_options_.info_log,
                       "adding log %" PRIu64 " to recycle list\n",
                       earliest.number);
-        log_recycle_files.push_back(earliest.number);
+        log_recycle_files_.push_back(earliest.number);
      } else {
        job_context->log_delete_files.push_back(earliest.number);
      }
@@ -252,11 +252,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
      }
      job_context->size_log_to_delete += earliest.size;
      total_log_size_ -= earliest.size;
-      if (concurrent_prepare_) {
+      if (two_write_queues_) {
        log_write_mutex_.Lock();
      }
      alive_log_files_.pop_front();
-      if (concurrent_prepare_) {
+      if (two_write_queues_) {
        log_write_mutex_.Unlock();
      }
      // Current log should always stay alive since it can't have
@@ -283,8 +283,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  // We're just cleaning up for DB::Write().
  assert(job_context->logs_to_free.empty());
  job_context->logs_to_free = logs_to_free_;
-  job_context->log_recycle_files.assign(log_recycle_files.begin(),
-                                        log_recycle_files.end());
+  job_context->log_recycle_files.assign(log_recycle_files_.begin(),
+                                        log_recycle_files_.end());
+  if (job_context->HaveSomethingToDelete()) {
+    ++pending_purge_obsolete_files_;
+  }
  logs_to_free_.clear();
 }

@@ -342,15 +345,12 @@ void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
 // 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) {
+  TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:Begin");
  // we'd better have sth to delete
  assert(state.HaveSomethingToDelete());

-  // this checks if FindObsoleteFiles() was run before. If not, don't do
-  // PurgeObsoleteFiles(). If FindObsoleteFiles() was run, we need to also
-  // run PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
-  if (state.manifest_file_number == 0) {
-    return;
-  }
+  // FindObsoleteFiles() should've populated this so nonzero
+  assert(state.manifest_file_number != 0);

  // Now, convert live list to an unordered map, WITHOUT mutex held;
  // set is slow.
@@ -533,6 +533,13 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
  wal_manager_.PurgeObsoleteWALFiles();
 #endif  // ROCKSDB_LITE
  LogFlush(immutable_db_options_.info_log);
+  InstrumentedMutexLock l(&mutex_);
+  --pending_purge_obsolete_files_;
+  assert(pending_purge_obsolete_files_ >= 0);
+  if (pending_purge_obsolete_files_ == 0) {
+    bg_cv_.SignalAll();
+  }
+  TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:End");
 }

 void DBImpl::DeleteObsoleteFiles() {
@@ -356,6 +356,29 @@ Status DBImpl::Recover(
      assert(s.IsIOError());
      return s;
    }
+    // Verify compatibility of env_options_ and filesystem
+    {
+      unique_ptr<RandomAccessFile> idfile;
+      EnvOptions customized_env(env_options_);
+      customized_env.use_direct_reads |=
+          immutable_db_options_.use_direct_io_for_flush_and_compaction;
+      s = env_->NewRandomAccessFile(IdentityFileName(dbname_), &idfile,
+                                    customized_env);
+      if (!s.ok()) {
+        const char* error_msg = s.ToString().c_str();
+        // Check if unsupported Direct I/O is the root cause
+        customized_env.use_direct_reads = false;
+        s = env_->NewRandomAccessFile(IdentityFileName(dbname_), &idfile,
+                                      customized_env);
+        if (s.ok()) {
+          return Status::InvalidArgument(
+              "Direct I/O is not supported by the specified DB.");
+        } else {
+          return Status::InvalidArgument(
+              "Found options incompatible with filesystem", error_msg);
+        }
+      }
+    }
  }

  Status s = versions_->Recover(column_families, read_only);
@@ -591,11 +614,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
        // consecutive, we continue recovery despite corruption. This could
        // happen when we open and write to a corrupted DB, where sequence id
        // will start from the last sequence id we recovered.
-        if (sequence == *next_sequence ||
-            // With seq_per_batch_, if previous run was with concurrent_prepare_
-            // then gap in the sequence numbers is expected by the commits
-            // without prepares.
-            (seq_per_batch_ && sequence >= *next_sequence)) {
+        if (sequence == *next_sequence) {
          stop_replay_for_corruption = false;
        }
        if (stop_replay_for_corruption) {
@@ -727,6 +746,12 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
    }

    if (!status.ok()) {
+      if (status.IsNotSupported()) {
+        // We should not treat NotSupported as corruption. It is rather a clear
+        // sign that we are processing a WAL that is produced by an incompatible
+        // version of the code.
+        return status;
+      }
      if (immutable_db_options_.wal_recovery_mode ==
          WALRecoveryMode::kSkipAnyCorruptedRecords) {
        // We should ignore all errors unconditionally
@@ -754,7 +779,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
    auto last_sequence = *next_sequence - 1;
    if ((*next_sequence != kMaxSequenceNumber) &&
        (versions_->LastSequence() <= last_sequence)) {
-      versions_->SetLastToBeWrittenSequence(last_sequence);
+      versions_->SetLastAllocatedSequence(last_sequence);
+      versions_->SetLastPublishedSequence(last_sequence);
      versions_->SetLastSequence(last_sequence);
    }
  }
@@ -845,13 +871,13 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
  if (data_seen && !flushed) {
    // Mark these as alive so they'll be considered for deletion later by
    // FindObsoleteFiles()
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      log_write_mutex_.Lock();
    }
    for (auto log_number : log_numbers) {
      alive_log_files_.push_back(LogFileNumberSize(log_number));
    }
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      log_write_mutex_.Unlock();
    }
  }
@@ -893,6 +919,7 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
    const uint64_t current_time = static_cast<uint64_t>(_current_time);

    {
+      auto write_hint = cfd->CalculateSSTWriteHint(0);
      mutex_.Unlock();

      SequenceNumber earliest_write_conflict_snapshot;
@@ -913,7 +940,7 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
          cfd->ioptions()->compression_opts, paranoid_file_checks,
          cfd->internal_stats(), TableFileCreationReason::kRecovery,
          &event_logger_, job_id, Env::IO_HIGH, nullptr /* table_properties */,
-          -1 /* level */, current_time);
+          -1 /* level */, current_time, write_hint);
      LogFlush(immutable_db_options_.info_log);
      ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
                      "[%s] [WriteLevel0TableForRecovery]"
@@ -966,6 +993,15 @@ 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;
+  return DBImpl::Open(db_options, dbname, column_families, handles, dbptr,
+                      !seq_per_batch);
+}
+
+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) {
  Status s = SanitizeOptionsByTable(db_options, column_families);
  if (!s.ok()) {
    return s;
@@ -985,7 +1021,7 @@ Status DB::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);
+  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch);
  s = impl->env_->CreateDirIfMissing(impl->immutable_db_options_.wal_dir);
  if (s.ok()) {
    for (auto db_path : impl->immutable_db_options_.db_paths) {
@@ -1007,6 +1043,7 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
    return s;
  }
  impl->mutex_.Lock();
+  auto write_hint = impl->CalculateWALWriteHint();
  // Handles create_if_missing, error_if_exists
  s = impl->Recover(column_families);
  if (s.ok()) {
@@ -1022,6 +1059,7 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
        LogFileName(impl->immutable_db_options_.wal_dir, new_log_number),
        &lfile, opt_env_options);
    if (s.ok()) {
+      lfile->SetWriteLifeTimeHint(write_hint);
      lfile->SetPreallocationBlockSize(
          impl->GetWalPreallocateBlockSize(max_write_buffer_size));
      {
@@ -1070,12 +1108,12 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
            cfd, &sv_context, *cfd->GetLatestMutableCFOptions());
      }
      sv_context.Clean();
-      if (impl->concurrent_prepare_) {
+      if (impl->two_write_queues_) {
        impl->log_write_mutex_.Lock();
      }
      impl->alive_log_files_.push_back(
          DBImpl::LogFileNumberSize(impl->logfile_number_));
-      if (impl->concurrent_prepare_) {
+      if (impl->two_write_queues_) {
        impl->log_write_mutex_.Unlock();
      }
      impl->DeleteObsoleteFiles();
@@ -57,21 +57,28 @@ Status DBImpl::WriteWithCallback(const WriteOptions& write_options,
 }
 #endif  // ROCKSDB_LITE

+// The main write queue. This is the only write queue that updates LastSequence.
+// When using one write queue, the same sequence also indicates the last
+// published sequence.
 Status DBImpl::WriteImpl(const WriteOptions& write_options,
                         WriteBatch* my_batch, WriteCallback* callback,
                         uint64_t* log_used, uint64_t log_ref,
-                         bool disable_memtable, uint64_t* seq_used) {
+                         bool disable_memtable, uint64_t* seq_used,
+                         size_t batch_cnt,
+                         PreReleaseCallback* pre_release_callback) {
+  assert(!seq_per_batch_ || batch_cnt != 0);
  if (my_batch == nullptr) {
    return Status::Corruption("Batch is nullptr!");
  }
  if (write_options.sync && write_options.disableWAL) {
    return Status::InvalidArgument("Sync writes has to enable WAL.");
  }
-  if (concurrent_prepare_ && immutable_db_options_.enable_pipelined_write) {
+  if (two_write_queues_ && immutable_db_options_.enable_pipelined_write) {
    return Status::NotSupported(
        "pipelined_writes is not compatible with concurrent prepares");
  }
  if (seq_per_batch_ && immutable_db_options_.enable_pipelined_write) {
+    // TODO(yiwu): update pipeline write with seq_per_batch and batch_cnt
    return Status::NotSupported(
        "pipelined_writes is not compatible with seq_per_batch");
  }
@@ -87,9 +94,9 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    }
  }

-  if (concurrent_prepare_ && disable_memtable) {
+  if (two_write_queues_ && disable_memtable) {
    return WriteImplWALOnly(write_options, my_batch, callback, log_used,
-                            log_ref, seq_used);
+                            log_ref, seq_used, batch_cnt, pre_release_callback);
  }

  if (immutable_db_options_.enable_pipelined_write) {
@@ -99,7 +106,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

  PERF_TIMER_GUARD(write_pre_and_post_process_time);
  WriteThread::Writer w(write_options, my_batch, callback, log_ref,
-                        disable_memtable);
+                        disable_memtable, batch_cnt, pre_release_callback);

  if (!write_options.disableWAL) {
    RecordTick(stats_, WRITE_WITH_WAL);
@@ -118,11 +125,22 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
      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_);
+          true /*concurrent_memtable_writes*/, seq_per_batch_, w.batch_cnt);
    }

    if (write_thread_.CompleteParallelMemTableWriter(&w)) {
      // we're responsible for exit batch group
+      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);
+          if (!ws.ok()) {
+            status = ws;
+            break;
+          }
+        }
+      }
+      // TODO(myabandeh): propagate status to write_group
      auto last_sequence = w.write_group->last_sequence;
      versions_->SetLastSequence(last_sequence);
      MemTableInsertStatusCheck(w.status);
@@ -154,7 +172,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  WriteThread::WriteGroup write_group;
  bool in_parallel_group = false;
  uint64_t last_sequence = kMaxSequenceNumber;
-  if (!concurrent_prepare_) {
+  if (!two_write_queues_) {
    last_sequence = versions_->LastSequence();
  }

@@ -162,7 +180,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

  bool need_log_sync = write_options.sync;
  bool need_log_dir_sync = need_log_sync && !log_dir_synced_;
-  if (!concurrent_prepare_ || !disable_memtable) {
+  if (!two_write_queues_ || !disable_memtable) {
    // 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
@@ -195,9 +213,11 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    bool parallel = immutable_db_options_.allow_concurrent_memtable_write &&
                    write_group.size > 1;
    size_t total_count = 0;
+    size_t valid_batches = 0;
    uint64_t total_byte_size = 0;
    for (auto* writer : write_group) {
      if (writer->CheckCallback(this)) {
+        valid_batches += writer->batch_cnt;
        if (writer->ShouldWriteToMemtable()) {
          total_count += WriteBatchInternal::Count(writer->batch);
          parallel = parallel && !writer->batch->HasMerge();
@@ -207,9 +227,15 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
            total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
      }
    }
-    size_t seq_inc = seq_per_batch_ ? write_group.size : total_count;
+    // Note about seq_per_batch_: either disableWAL is set for the entire write
+    // group or not. In either case we inc seq for each write batch with no
+    // failed callback. This means that there could be a batch with
+    // disalbe_memtable in between; although we do not write this batch to
+    // memtable it still consumes a seq. Otherwise, if !seq_per_batch_, we inc
+    // the seq per valid written key to mem.
+    size_t seq_inc = seq_per_batch_ ? valid_batches : total_count;

-    const bool concurrent_update = concurrent_prepare_;
+    const bool concurrent_update = two_write_queues_;
    // Update stats while we are an exclusive group leader, so we know
    // that nobody else can be writing to these particular stats.
    // We're optimistic, updating the stats before we successfully
@@ -237,7 +263,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

    PERF_TIMER_STOP(write_pre_and_post_process_time);

-    if (!concurrent_prepare_) {
+    if (!two_write_queues_) {
      if (status.ok() && !write_options.disableWAL) {
        PERF_TIMER_GUARD(write_wal_time);
        status = WriteToWAL(write_group, log_writer, log_used, need_log_sync,
@@ -246,13 +272,13 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    } else {
      if (status.ok() && !write_options.disableWAL) {
        PERF_TIMER_GUARD(write_wal_time);
-        // LastToBeWrittenSequence is increased inside WriteToWAL under
+        // LastAllocatedSequence is increased inside WriteToWAL under
        // wal_write_mutex_ to ensure ordered events in WAL
        status = ConcurrentWriteToWAL(write_group, log_used, &last_sequence,
                                      seq_inc);
      } else {
        // Otherwise we inc seq number for memtable writes
-        last_sequence = versions_->FetchAddLastToBeWrittenSequence(seq_inc);
+        last_sequence = versions_->FetchAddLastAllocatedSequence(seq_inc);
      }
    }
    assert(last_sequence != kMaxSequenceNumber);
@@ -263,16 +289,25 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
      PERF_TIMER_GUARD(write_memtable_time);

      if (!parallel) {
+        // w.sequence will be set inside InsertInto
        w.status = WriteBatchInternal::InsertInto(
            write_group, current_sequence, column_family_memtables_.get(),
            &flush_scheduler_, write_options.ignore_missing_column_families,
            0 /*recovery_log_number*/, this, parallel, seq_per_batch_);
      } else {
        SequenceNumber next_sequence = current_sequence;
+        // Note: the logic for advancing seq here must be consistent with the
+        // logic in WriteBatchInternal::InsertInto(write_group...) as well as
+        // with WriteBatchInternal::InsertInto(write_batch...) that is called on
+        // the merged batch during recovery from the WAL.
        for (auto* writer : write_group) {
+          if (writer->CallbackFailed()) {
+            continue;
+          }
          writer->sequence = next_sequence;
          if (seq_per_batch_) {
-            next_sequence++;
+            assert(writer->batch_cnt);
+            next_sequence += writer->batch_cnt;
          } else if (writer->ShouldWriteToMemtable()) {
            next_sequence += WriteBatchInternal::Count(writer->batch);
          }
@@ -292,7 +327,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
          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_);
+              this, true /*concurrent_memtable_writes*/, seq_per_batch_,
+              w.batch_cnt);
        }
      }
      if (seq_used != nullptr) {
@@ -310,9 +346,9 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    mutex_.Lock();
    MarkLogsSynced(logfile_number_, need_log_dir_sync, status);
    mutex_.Unlock();
-    // Requesting sync with concurrent_prepare_ is expected to be very rare. We
+    // Requesting sync with two_write_queues_ is expected to be very rare. We
    // hance provide a simple implementation that is not necessarily efficient.
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      if (manual_wal_flush_) {
        status = FlushWAL(true);
      } else {
@@ -329,10 +365,20 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  }
  if (should_exit_batch_group) {
    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);
+          if (!ws.ok()) {
+            status = ws;
+            break;
+          }
+        }
+      }
      versions_->SetLastSequence(last_sequence);
    }
    MemTableInsertStatusCheck(w.status);
-    write_thread_.ExitAsBatchGroupLeader(write_group, w.status);
+    write_thread_.ExitAsBatchGroupLeader(write_group, status);
  }

  if (status.ok()) {
@@ -468,17 +514,19 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
  return w.FinalStatus();
 }

+// The 2nd write queue. If enabled it will be used only for WAL-only writes.
+// This is the only queue that updates LastPublishedSequence which is only
+// applicable in a two-queue setting.
 Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
                                WriteBatch* my_batch, WriteCallback* callback,
                                uint64_t* log_used, uint64_t log_ref,
-                                uint64_t* seq_used) {
+                                uint64_t* seq_used, size_t batch_cnt,
+                                PreReleaseCallback* pre_release_callback) {
  Status status;
  PERF_TIMER_GUARD(write_pre_and_post_process_time);
  WriteThread::Writer w(write_options, my_batch, callback, log_ref,
-                        true /* disable_memtable */);
-  if (write_options.disableWAL) {
-    return status;
-  }
+                        true /* disable_memtable */, batch_cnt,
+                        pre_release_callback);
  RecordTick(stats_, WRITE_WITH_WAL);
  StopWatch write_sw(env_, immutable_db_options_.statistics.get(), DB_WRITE);

@@ -532,23 +580,39 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
  PERF_TIMER_STOP(write_pre_and_post_process_time);

  PERF_TIMER_GUARD(write_wal_time);
-  // LastToBeWrittenSequence is increased inside WriteToWAL under
+  // LastAllocatedSequence is increased inside WriteToWAL under
  // wal_write_mutex_ to ensure ordered events in WAL
-  size_t seq_inc = seq_per_batch_ ? write_group.size : 0 /*total_count*/;
-  status = ConcurrentWriteToWAL(write_group, log_used, &last_sequence, seq_inc);
+  size_t seq_inc = 0 /* total_count */;
+  if (seq_per_batch_) {
+    size_t total_batch_cnt = 0;
+    for (auto* writer : write_group) {
+      assert(writer->batch_cnt);
+      total_batch_cnt += writer->batch_cnt;
+    }
+    seq_inc = total_batch_cnt;
+  }
+  if (!write_options.disableWAL) {
+    status =
+        ConcurrentWriteToWAL(write_group, log_used, &last_sequence, seq_inc);
+  } else {
+    // Otherwise we inc seq number to do solely the seq allocation
+    last_sequence = versions_->FetchAddLastAllocatedSequence(seq_inc);
+  }
  auto curr_seq = last_sequence + 1;
  for (auto* writer : write_group) {
-    if (writer->CheckCallback(this)) {
-      writer->sequence = curr_seq;
+    if (writer->CallbackFailed()) {
+      continue;
    }
+    writer->sequence = curr_seq;
    if (seq_per_batch_) {
-      curr_seq++;
-    } else if (writer->CheckCallback(this)) {
-      curr_seq += WriteBatchInternal::Count(writer->batch);
+      assert(writer->batch_cnt);
+      curr_seq += writer->batch_cnt;
    }
+    // else seq advances only by memtable writes
  }
  if (status.ok() && write_options.sync) {
-    // Requesting sync with concurrent_prepare_ is expected to be very rare. We
+    assert(!write_options.disableWAL);
+    // Requesting sync with two_write_queues_ is expected to be very rare. We
    // hance provide a simple implementation that is not necessarily efficient.
    if (manual_wal_flush_) {
      status = FlushWAL(true);
@@ -561,7 +625,19 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
  if (!w.CallbackFailed()) {
    WriteCallbackStatusCheck(status);
  }
-  nonmem_write_thread_.ExitAsBatchGroupLeader(write_group, w.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);
+        if (!ws.ok()) {
+          status = ws;
+          break;
+        }
+      }
+    }
+  }
+  nonmem_write_thread_.ExitAsBatchGroupLeader(write_group, status);
  if (status.ok()) {
    status = w.FinalStatus();
  }
@@ -636,7 +712,7 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
  }

  if (UNLIKELY(status.ok() && !bg_error_.ok())) {
-    return bg_error_;
+    status = bg_error_;
  }

  if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
@@ -689,7 +765,8 @@ WriteBatch* DBImpl::MergeBatch(const WriteThread::WriteGroup& write_group,
  WriteBatch* merged_batch = nullptr;
  *write_with_wal = 0;
  auto* leader = write_group.leader;
-  if (write_group.size == 1 && leader->ShouldWriteToWAL() &&
+  assert(!leader->disable_wal);  // Same holds for all in the batch group
+  if (write_group.size == 1 && !leader->CallbackFailed() &&
      leader->batch->GetWalTerminationPoint().is_cleared()) {
    // we simply write the first WriteBatch to WAL if the group only
    // contains one batch, that batch should be written to the WAL,
@@ -705,7 +782,7 @@ WriteBatch* DBImpl::MergeBatch(const WriteThread::WriteGroup& write_group,
    // interface
    merged_batch = tmp_batch;
    for (auto writer : write_group) {
-      if (writer->ShouldWriteToWAL()) {
+      if (!writer->CallbackFailed()) {
        WriteBatchInternal::Append(merged_batch, writer->batch,
                                   /*WAL_only*/ true);
        if (WriteBatchInternal::IsLatestPersistentState(writer->batch)) {
@@ -719,7 +796,7 @@ WriteBatch* DBImpl::MergeBatch(const WriteThread::WriteGroup& write_group,
  return merged_batch;
 }

-// When concurrent_prepare_ is disabled, this function is called from the only
+// When two_write_queues_ is disabled, this function is called from the only
 // write thread. Otherwise this must be called holding log_write_mutex_.
 Status DBImpl::WriteToWAL(const WriteBatch& merged_batch,
                          log::Writer* log_writer, uint64_t* log_used,
@@ -745,6 +822,8 @@ Status DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
                          SequenceNumber sequence) {
  Status status;

+  assert(!write_group.leader->disable_wal);
+  // Same holds for all in the batch group
  size_t write_with_wal = 0;
  WriteBatch* to_be_cached_state = nullptr;
  WriteBatch* merged_batch = MergeBatch(write_group, &tmp_batch_,
@@ -812,6 +891,8 @@ Status DBImpl::ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
                                    size_t seq_inc) {
  Status status;

+  assert(!write_group.leader->disable_wal);
+  // Same holds for all in the batch group
  WriteBatch tmp_batch;
  size_t write_with_wal = 0;
  WriteBatch* to_be_cached_state = nullptr;
@@ -828,7 +909,7 @@ Status DBImpl::ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
      writer->log_used = logfile_number_;
    }
  }
-  *last_sequence = versions_->FetchAddLastToBeWrittenSequence(seq_inc);
+  *last_sequence = versions_->FetchAddLastAllocatedSequence(seq_inc);
  auto sequence = *last_sequence + 1;
  WriteBatchInternal::SetSequence(merged_batch, sequence);

@@ -858,18 +939,23 @@ Status DBImpl::WriteRecoverableState() {
  if (!cached_recoverable_state_empty_) {
    bool dont_care_bool;
    SequenceNumber next_seq;
-    if (concurrent_prepare_) {
+    if (two_write_queues_) {
      log_write_mutex_.Lock();
    }
    SequenceNumber seq = versions_->LastSequence();
-    WriteBatchInternal::SetSequence(&cached_recoverable_state_, ++seq);
+    WriteBatchInternal::SetSequence(&cached_recoverable_state_, seq + 1);
    auto status = WriteBatchInternal::InsertInto(
        &cached_recoverable_state_, column_family_memtables_.get(),
        &flush_scheduler_, true, 0 /*recovery_log_number*/, this,
        false /* concurrent_memtable_writes */, &next_seq, &dont_care_bool,
        seq_per_batch_);
-    versions_->SetLastSequence(--next_seq);
-    if (concurrent_prepare_) {
+    auto last_seq = next_seq - 1;
+    if (two_write_queues_) {
+      versions_->FetchAddLastAllocatedSequence(last_seq - seq);
+    }
+    versions_->SetLastSequence(last_seq);
+    versions_->SetLastPublishedSequence(last_seq);
+    if (two_write_queues_) {
      log_write_mutex_.Unlock();
    }
    if (status.ok()) {
@@ -933,7 +1019,7 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
        break;
      }
      cfd->imm()->FlushRequested();
-      SchedulePendingFlush(cfd);
+      SchedulePendingFlush(cfd, FlushReason::kWriteBufferManager);
    }
  }
  MaybeScheduleFlushOrCompaction();
@@ -979,7 +1065,7 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
    status = SwitchMemtable(cfd_picked, write_context);
    if (status.ok()) {
      cfd_picked->imm()->FlushRequested();
-      SchedulePendingFlush(cfd_picked);
+      SchedulePendingFlush(cfd_picked, FlushReason::kWriteBufferFull);
      MaybeScheduleFlushOrCompaction();
    }
  }
@@ -1109,7 +1195,7 @@ void DBImpl::NotifyOnMemTableSealed(ColumnFamilyData* cfd,
 Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  mutex_.AssertHeld();
  WriteThread::Writer nonmem_w;
-  if (concurrent_prepare_) {
+  if (two_write_queues_) {
    // SwitchMemtable is a rare event. To simply the reasoning, we make sure
    // that there is no concurrent thread writing to WAL.
    nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
@@ -1135,18 +1221,18 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  // Attempt to switch to a new memtable and trigger flush of old.
  // Do this without holding the dbmutex lock.
  assert(versions_->prev_log_number() == 0);
-  if (concurrent_prepare_) {
+  if (two_write_queues_) {
    log_write_mutex_.Lock();
  }
  bool creating_new_log = !log_empty_;
-  if (concurrent_prepare_) {
+  if (two_write_queues_) {
    log_write_mutex_.Unlock();
  }
  uint64_t recycle_log_number = 0;
  if (creating_new_log && immutable_db_options_.recycle_log_file_num &&
-      !log_recycle_files.empty()) {
-    recycle_log_number = log_recycle_files.front();
-    log_recycle_files.pop_front();
+      !log_recycle_files_.empty()) {
+    recycle_log_number = log_recycle_files_.front();
+    log_recycle_files_.pop_front();
  }
  uint64_t new_log_number =
      creating_new_log ? versions_->NewFileNumber() : logfile_number_;
@@ -1168,6 +1254,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
      BuildDBOptions(immutable_db_options_, mutable_db_options_);
  const auto preallocate_block_size =
    GetWalPreallocateBlockSize(mutable_cf_options.write_buffer_size);
+  auto write_hint = CalculateWALWriteHint();
  mutex_.Unlock();
  {
    if (creating_new_log) {
@@ -1193,6 +1280,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
        // use preallocate_block_size instead
        // of calling GetWalPreallocateBlockSize()
        lfile->SetPreallocationBlockSize(preallocate_block_size);
+        lfile->SetWriteLifeTimeHint(write_hint);
        unique_ptr<WritableFileWriter> file_writer(
            new WritableFileWriter(std::move(lfile), opt_env_opt));
        new_log = new log::Writer(
@@ -1219,17 +1307,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
                 ". Immutable memtables: %d.\n",
                 cfd->GetName().c_str(), new_log_number, num_imm_unflushed);
  mutex_.Lock();
-  if (!s.ok()) {
-    // how do we fail if we're not creating new log?
-    assert(creating_new_log);
-    assert(!new_mem);
-    assert(!new_log);
-    if (concurrent_prepare_) {
-      nonmem_write_thread_.ExitUnbatched(&nonmem_w);
-    }
-    return s;
-  }
-  if (creating_new_log) {
+  if (s.ok() && creating_new_log) {
    log_write_mutex_.Lock();
    logfile_number_ = new_log_number;
    assert(new_log != nullptr);
@@ -1238,12 +1316,31 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
    if (!logs_.empty()) {
      // Alway flush the buffer of the last log before switching to a new one
      log::Writer* cur_log_writer = logs_.back().writer;
-      cur_log_writer->WriteBuffer();
+      s = cur_log_writer->WriteBuffer();
+      if (!s.ok()) {
+        ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                       "[%s] Failed to switch from #%" PRIu64 " to #%" PRIu64
+                       "  WAL file -- %s\n",
+                       cfd->GetName().c_str(), cur_log_writer->get_log_number(),
+                       new_log_number);
+      }
    }
    logs_.emplace_back(logfile_number_, new_log);
    alive_log_files_.push_back(LogFileNumberSize(logfile_number_));
    log_write_mutex_.Unlock();
  }
+
+  if (!s.ok()) {
+    // how do we fail if we're not creating new log?
+    assert(creating_new_log);
+    assert(!new_mem);
+    assert(!new_log);
+    if (two_write_queues_) {
+      nonmem_write_thread_.ExitUnbatched(&nonmem_w);
+    }
+    return s;
+  }
+
  for (auto loop_cfd : *versions_->GetColumnFamilySet()) {
    // all this is just optimization to delete logs that
    // are no longer needed -- if CF is empty, that means it
@@ -1264,7 +1361,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  cfd->SetMemtable(new_mem);
  InstallSuperVersionAndScheduleWork(cfd, &context->superversion_context,
                                     mutable_cf_options);
-  if (concurrent_prepare_) {
+  if (two_write_queues_) {
    nonmem_write_thread_.ExitUnbatched(&nonmem_w);
  }
  return s;
@@ -1298,7 +1395,10 @@ Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,
  // 8 bytes are taken by header, 4 bytes for count, 1 byte for type,
  // and we allocate 11 extra bytes for key length, as well as value length.
  WriteBatch batch(key.size() + value.size() + 24);
-  batch.Put(column_family, key, value);
+  Status s = batch.Put(column_family, key, value);
+  if (!s.ok()) {
+    return s;
+  }
  return Write(opt, &batch);
 }

@@ -1327,7 +1427,10 @@ Status DB::DeleteRange(const WriteOptions& opt,
 Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
                 const Slice& key, const Slice& value) {
  WriteBatch batch;
-  batch.Merge(column_family, key, value);
+  Status s = batch.Merge(column_family, key, value);
+  if (!s.ok()) {
+    return s;
+  }
  return Write(opt, &batch);
 }
 }  // namespace rocksdb
@@ -76,6 +76,7 @@ class DBIter final: public Iterator {
      prev_count_ = 0;
      prev_found_count_ = 0;
      bytes_read_ = 0;
+      skip_count_ = 0;
    }

    void BumpGlobalStatistics(Statistics* global_statistics) {
@@ -84,6 +85,7 @@ class DBIter final: public Iterator {
      RecordTick(global_statistics, NUMBER_DB_PREV, prev_count_);
      RecordTick(global_statistics, NUMBER_DB_PREV_FOUND, prev_found_count_);
      RecordTick(global_statistics, ITER_BYTES_READ, bytes_read_);
+      RecordTick(global_statistics, NUMBER_ITER_SKIP, skip_count_);
      PERF_COUNTER_ADD(iter_read_bytes, bytes_read_);
      ResetCounters();
    }
@@ -98,6 +100,8 @@ class DBIter final: public Iterator {
    uint64_t prev_found_count_;
    // Map to Tickers::ITER_BYTES_READ
    uint64_t bytes_read_;
+    // Map to Tickers::NUMBER_ITER_SKIP
+    uint64_t skip_count_;
  };

  DBIter(Env* _env, const ReadOptions& read_options,
@@ -147,6 +151,7 @@ class DBIter final: public Iterator {
    // Compiler warning issue filed:
    // https://github.com/facebook/rocksdb/issues/3013
    RecordTick(statistics_, NO_ITERATORS, uint64_t(-1));
+    ResetInternalKeysSkippedCounter();
    local_stats_.BumpGlobalStatistics(statistics_);
    if (!arena_mode_) {
      delete iter_;
@@ -212,6 +217,9 @@ class DBIter final: public Iterator {
        *prop = "Iterator is not valid.";
      }
      return Status::OK();
+    } else if (prop_name == "rocksdb.iterator.internal-key") {
+      *prop = saved_key_.GetUserKey().ToString();
+      return Status::OK();
    }
    return Status::InvalidArgument("Undentified property.");
  }
@@ -267,6 +275,10 @@ class DBIter final: public Iterator {
  }

  inline void ResetInternalKeysSkippedCounter() {
+    local_stats_.skip_count_ += num_internal_keys_skipped_;
+    if (valid_) {
+      local_stats_.skip_count_--;
+    }
    num_internal_keys_skipped_ = 0;
  }

@@ -602,10 +614,12 @@ void DBIter::MergeValuesNewToOld() {
  // Start the merge process by pushing the first operand
  merge_context_.PushOperand(iter_->value(),
                             iter_->IsValuePinned() /* operand_pinned */);
+  TEST_SYNC_POINT("DBIter::MergeValuesNewToOld:PushedFirstOperand");

  ParsedInternalKey ikey;
  Status s;
  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
+    TEST_SYNC_POINT("DBIter::MergeValuesNewToOld:SteppedToNextOperand");
    if (!ParseKey(&ikey)) {
      // skip corrupted key
      continue;
@@ -631,6 +645,7 @@ void DBIter::MergeValuesNewToOld() {
          merge_operator_, ikey.user_key, &val, merge_context_.GetOperands(),
          &saved_value_, logger_, statistics_, env_, &pinned_value_, true);
      if (!s.ok()) {
+        valid_ = false;
        status_ = s;
      }
      // iter_ is positioned after put
@@ -668,6 +683,7 @@ void DBIter::MergeValuesNewToOld() {
                                  &saved_value_, logger_, statistics_, env_,
                                  &pinned_value_, true);
  if (!s.ok()) {
+    valid_ = false;
    status_ = s;
  }
 }
@@ -937,8 +953,10 @@ bool DBIter::FindValueForCurrentKey() {
      assert(false);
      break;
  }
-  valid_ = true;
-  if (!s.ok()) {
+  if (s.ok()) {
+    valid_ = true;
+  } else {
+    valid_ = false;
    status_ = s;
  }
  return true;
@@ -959,6 +977,17 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  // assume there is at least one parseable key for this user key
  ParsedInternalKey ikey;
  FindParseableKey(&ikey, kForward);
+  assert(iter_->Valid());
+  assert(user_comparator_->Equal(ikey.user_key, saved_key_.GetUserKey()));
+
+  // In case read_callback presents, the value we seek to may not be visible.
+  // Seek for the next value that's visible.
+  while (!IsVisible(ikey.sequence)) {
+    iter_->Next();
+    FindParseableKey(&ikey, kForward);
+    assert(iter_->Valid());
+    assert(user_comparator_->Equal(ikey.user_key, saved_key_.GetUserKey()));
+  }

  if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
      range_del_agg_.ShouldDelete(
@@ -1014,8 +1043,10 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
      iter_->Seek(last_key);
      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
    }
-    valid_ = true;
-    if (!s.ok()) {
+    if (s.ok()) {
+      valid_ = true;
+    } else {
+      valid_ = false;
      status_ = s;
    }
    return true;
@@ -1026,8 +1057,10 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
                                  &val, merge_context_.GetOperands(),
                                  &saved_value_, logger_, statistics_, env_,
                                  &pinned_value_, true);
-  valid_ = true;
-  if (!s.ok()) {
+  if (s.ok()) {
+    valid_ = true;
+  } else {
+    valid_ = false;
    status_ = s;
  }
  return true;
@@ -1125,6 +1158,13 @@ void DBIter::Seek(const Slice& target) {
  saved_key_.Clear();
  saved_key_.SetInternalKey(target, sequence_);

+  if (iterate_lower_bound_ != nullptr &&
+      user_comparator_->Compare(saved_key_.GetUserKey(),
+                                *iterate_lower_bound_) < 0) {
+    saved_key_.Clear();
+    saved_key_.SetInternalKey(*iterate_lower_bound_, sequence_);
+  }
+
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->Seek(saved_key_.GetInternalKey());
@@ -1143,6 +1183,7 @@ void DBIter::Seek(const Slice& target) {
    }
    if (statistics_ != nullptr) {
      if (valid_) {
+        // Decrement since we don't want to count this key as skipped
        RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
        RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
        PERF_COUNTER_ADD(iter_read_bytes, key().size() + value().size());
@@ -1167,6 +1208,13 @@ void DBIter::SeekForPrev(const Slice& target) {
  saved_key_.SetInternalKey(target, 0 /* sequence_number */,
                            kValueTypeForSeekForPrev);

+  if (iterate_upper_bound_ != nullptr &&
+      user_comparator_->Compare(saved_key_.GetUserKey(),
+                                *iterate_upper_bound_) >= 0) {
+    saved_key_.Clear();
+    saved_key_.SetInternalKey(*iterate_upper_bound_, kMaxSequenceNumber);
+  }
+
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekForPrev(saved_key_.GetInternalKey());
@@ -1269,7 +1317,8 @@ void DBIter::SeekToLast() {
    if (!Valid()) {
      return;
    } else if (user_comparator_->Equal(*iterate_upper_bound_, key())) {
-      Prev();
+      ReleaseTempPinnedData();
+      PrevInternal();
    }
  } else {
    PrevInternal();
@@ -1345,20 +1394,25 @@ void ArenaWrappedDBIter::Init(Env* env, const ReadOptions& read_options,
                              const SequenceNumber& sequence,
                              uint64_t max_sequential_skip_in_iteration,
                              uint64_t version_number,
-                              ReadCallback* read_callback, bool allow_blob) {
+                              ReadCallback* read_callback, bool allow_blob,
+                              bool allow_refresh) {
  auto mem = arena_.AllocateAligned(sizeof(DBIter));
  db_iter_ = new (mem)
      DBIter(env, read_options, cf_options, cf_options.user_comparator, nullptr,
             sequence, true, max_sequential_skip_in_iteration, read_callback,
             allow_blob);
  sv_number_ = version_number;
+  allow_refresh_ = allow_refresh;
 }

 Status ArenaWrappedDBIter::Refresh() {
-  if (cfd_ == nullptr || db_impl_ == nullptr) {
+  if (cfd_ == nullptr || db_impl_ == nullptr || !allow_refresh_) {
    return Status::NotSupported("Creating renew iterator is not allowed.");
  }
  assert(db_iter_ != nullptr);
+  // TODO(yiwu): For last_seq_same_as_publish_seq_==false, this is not the
+  // correct behavior. Will be corrected automatically when we take a snapshot
+  // here for the case of WritePreparedTxnDB.
  SequenceNumber latest_seq = db_impl_->GetLatestSequenceNumber();
  uint64_t cur_sv_number = cfd_->GetSuperVersionNumber();
  if (sv_number_ != cur_sv_number) {
@@ -1370,7 +1424,7 @@ Status ArenaWrappedDBIter::Refresh() {
    SuperVersion* sv = cfd_->GetReferencedSuperVersion(db_impl_->mutex());
    Init(env, read_options_, *(cfd_->ioptions()), latest_seq,
         sv->mutable_cf_options.max_sequential_skip_in_iterations,
-         cur_sv_number, read_callback_, allow_blob_);
+         cur_sv_number, read_callback_, allow_blob_, allow_refresh_);

    InternalIterator* internal_iter = db_impl_->NewInternalIterator(
        read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator());
@@ -1387,12 +1441,12 @@ ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    const ImmutableCFOptions& cf_options, const SequenceNumber& sequence,
    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
    ReadCallback* read_callback, DBImpl* db_impl, ColumnFamilyData* cfd,
-    bool allow_blob) {
+    bool allow_blob, bool allow_refresh) {
  ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
  iter->Init(env, read_options, cf_options, sequence,
             max_sequential_skip_in_iterations, version_number, read_callback,
-             allow_blob);
-  if (db_impl != nullptr && cfd != nullptr) {
+             allow_blob, allow_refresh);
+  if (db_impl != nullptr && cfd != nullptr && allow_refresh) {
    iter->StoreRefreshInfo(read_options, db_impl, cfd, read_callback,
                           allow_blob);
  }
@@ -73,7 +73,7 @@ class ArenaWrappedDBIter : public Iterator {
            const ImmutableCFOptions& cf_options,
            const SequenceNumber& sequence,
            uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
-            ReadCallback* read_callback, bool allow_blob);
+            ReadCallback* read_callback, bool allow_blob, bool allow_refresh);

  void StoreRefreshInfo(const ReadOptions& read_options, DBImpl* db_impl,
                        ColumnFamilyData* cfd, ReadCallback* read_callback,
@@ -94,6 +94,7 @@ class ArenaWrappedDBIter : public Iterator {
  ReadOptions read_options_;
  ReadCallback* read_callback_;
  bool allow_blob_ = false;
+  bool allow_refresh_ = true;
 };

 // Generate the arena wrapped iterator class.
@@ -104,6 +105,6 @@ extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    const ImmutableCFOptions& 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);
-
+    ColumnFamilyData* cfd = nullptr, bool allow_blob = false,
+    bool allow_refresh = true);
 }  // namespace rocksdb
@@ -356,7 +356,7 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    db_iter->SeekToLast();

    ASSERT_TRUE(db_iter->Valid());
-    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_key_skipped_count), 7);
+    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_key_skipped_count), 1);
    ASSERT_EQ(db_iter->key().ToString(), "b");

    SetPerfLevel(kDisable);
@@ -475,7 +475,7 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    db_iter->SeekToLast();

    ASSERT_TRUE(db_iter->Valid());
-    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_delete_skipped_count), 1);
+    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_delete_skipped_count), 0);
    ASSERT_EQ(db_iter->key().ToString(), "b");

    SetPerfLevel(kDisable);
@@ -2991,6 +2991,33 @@ TEST_F(DBIteratorTest, PrevLowerBound) {
  ASSERT_FALSE(db_iter->Valid());
 }

+TEST_F(DBIteratorTest, SeekLessLowerBound) {
+  const int kNumKeys = 3;
+  const int kLowerBound = 2;
+  TestIterator* internal_iter = new TestIterator(BytewiseComparator());
+  for (int j = 1; j <= kNumKeys; ++j) {
+    internal_iter->AddPut(std::to_string(j), "val");
+  }
+  internal_iter->Finish();
+
+  ReadOptions ro;
+  auto lower_bound_str = std::to_string(kLowerBound);
+  Slice lower_bound(lower_bound_str);
+  ro.iterate_lower_bound = &lower_bound;
+  Options options;
+  std::unique_ptr<Iterator> db_iter(NewDBIterator(
+      env_, ro, ImmutableCFOptions(options), BytewiseComparator(),
+      internal_iter, 10 /* sequence */,
+      options.max_sequential_skip_in_iterations, nullptr /* read_callback */));
+
+  auto before_lower_bound_str = std::to_string(kLowerBound - 1);
+  Slice before_lower_bound(lower_bound_str);
+
+  db_iter->Seek(before_lower_bound);
+  ASSERT_TRUE(db_iter->Valid());
+  ASSERT_EQ(lower_bound_str, db_iter->key().ToString());
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -28,13 +28,11 @@ class MockMemTableRep : public MemTableRep {
    return rep_->Allocate(len, buf);
  }

-  virtual void Insert(KeyHandle handle) override {
-    return rep_->Insert(handle);
-  }
+  virtual void Insert(KeyHandle handle) override { rep_->Insert(handle); }

  virtual void InsertWithHint(KeyHandle handle, void** hint) override {
    num_insert_with_hint_++;
-    ASSERT_NE(nullptr, hint);
+    EXPECT_NE(nullptr, hint);
    last_hint_in_ = *hint;
    rep_->InsertWithHint(handle, hint);
    last_hint_out_ = *hint;
@@ -129,6 +127,84 @@ class TestPrefixExtractor : public SliceTransform {
  }
 };

+// Test that ::Add properly returns false when inserting duplicate keys
+TEST_F(DBMemTableTest, DuplicateSeq) {
+  SequenceNumber seq = 123;
+  std::string value;
+  Status s;
+  MergeContext merge_context;
+  Options options;
+  InternalKeyComparator ikey_cmp(options.comparator);
+  RangeDelAggregator range_del_agg(ikey_cmp, {} /* snapshots */);
+
+  // Create a MemTable
+  InternalKeyComparator cmp(BytewiseComparator());
+  auto factory = std::make_shared<SkipListFactory>();
+  options.memtable_factory = factory;
+  ImmutableCFOptions ioptions(options);
+  WriteBufferManager wb(options.db_write_buffer_size);
+  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                               kMaxSequenceNumber, 0 /* column_family_id */);
+
+  // Write some keys and make sure it returns false on duplicates
+  bool res;
+  res = mem->Add(seq, kTypeValue, "key", "value2");
+  ASSERT_TRUE(res);
+  res = mem->Add(seq, kTypeValue, "key", "value2");
+  ASSERT_FALSE(res);
+  // Changing the type should still cause the duplicatae key
+  res = mem->Add(seq, kTypeMerge, "key", "value2");
+  ASSERT_FALSE(res);
+  // Changing the seq number will make the key fresh
+  res = mem->Add(seq + 1, kTypeMerge, "key", "value2");
+  ASSERT_TRUE(res);
+  // Test with different types for duplicate keys
+  res = mem->Add(seq, kTypeDeletion, "key", "");
+  ASSERT_FALSE(res);
+  res = mem->Add(seq, kTypeSingleDeletion, "key", "");
+  ASSERT_FALSE(res);
+
+  // Test the duplicate keys under stress
+  for (int i = 0; i < 10000; i++) {
+    bool insert_dup = i % 10 == 1;
+    if (!insert_dup) {
+      seq++;
+    }
+    res = mem->Add(seq, kTypeValue, "foo", "value" + ToString(seq));
+    if (insert_dup) {
+      ASSERT_FALSE(res);
+    } else {
+      ASSERT_TRUE(res);
+    }
+  }
+  delete mem;
+
+  // Test with InsertWithHint
+  options.memtable_insert_with_hint_prefix_extractor.reset(
+      new TestPrefixExtractor());  // which uses _ to extract the prefix
+  ioptions = ImmutableCFOptions(options);
+  mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                     kMaxSequenceNumber, 0 /* column_family_id */);
+  // Insert a duplicate key with _ in it
+  res = mem->Add(seq, kTypeValue, "key_1", "value");
+  ASSERT_TRUE(res);
+  res = mem->Add(seq, kTypeValue, "key_1", "value");
+  ASSERT_FALSE(res);
+  delete mem;
+
+  // Test when InsertConcurrently will be invoked
+  options.allow_concurrent_memtable_write = true;
+  ioptions = ImmutableCFOptions(options);
+  mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                     kMaxSequenceNumber, 0 /* column_family_id */);
+  MemTablePostProcessInfo post_process_info;
+  res = mem->Add(seq, kTypeValue, "key", "value", true, &post_process_info);
+  ASSERT_TRUE(res);
+  res = mem->Add(seq, kTypeValue, "key", "value", true, &post_process_info);
+  ASSERT_FALSE(res);
+  delete mem;
+}
+
 TEST_F(DBMemTableTest, InsertWithHint) {
  Options options;
  options.allow_concurrent_memtable_write = false;
@@ -14,10 +14,43 @@

 namespace rocksdb {

+class TestReadCallback : public ReadCallback {
+ public:
+  TestReadCallback(SnapshotChecker* snapshot_checker,
+                   SequenceNumber snapshot_seq)
+      : snapshot_checker_(snapshot_checker), snapshot_seq_(snapshot_seq) {}
+
+  bool IsCommitted(SequenceNumber seq) override {
+    return snapshot_checker_->IsInSnapshot(seq, snapshot_seq_);
+  }
+
+ private:
+  SnapshotChecker* snapshot_checker_;
+  SequenceNumber snapshot_seq_;
+};
+
 // Test merge operator functionality.
 class DBMergeOperatorTest : public DBTestBase {
 public:
  DBMergeOperatorTest() : DBTestBase("/db_merge_operator_test") {}
+
+  std::string GetWithReadCallback(SnapshotChecker* snapshot_checker,
+                                  const Slice& key,
+                                  const Snapshot* snapshot = nullptr) {
+    SequenceNumber seq = snapshot == nullptr ? db_->GetLatestSequenceNumber()
+                                             : snapshot->GetSequenceNumber();
+    TestReadCallback read_callback(snapshot_checker, seq);
+    ReadOptions read_opt;
+    read_opt.snapshot = snapshot;
+    PinnableSlice value;
+    Status s =
+        dbfull()->GetImpl(read_opt, db_->DefaultColumnFamily(), key, &value,
+                          nullptr /*value_found*/, &read_callback);
+    if (!s.ok()) {
+      return s.ToString();
+    }
+    return value.ToString();
+  }
 };

 TEST_F(DBMergeOperatorTest, LimitMergeOperands) {
@@ -133,16 +166,33 @@ TEST_F(DBMergeOperatorTest, MergeErrorOnIteration) {
  ASSERT_OK(Merge("k1", "v1"));
  ASSERT_OK(Merge("k1", "corrupted"));
  ASSERT_OK(Put("k2", "v2"));
-  VerifyDBFromMap({{"k1", ""}, {"k2", "v2"}}, nullptr, false,
-                  {{"k1", Status::Corruption()}});
+  auto* iter = db_->NewIterator(ReadOptions());
+  iter->Seek("k1");
+  ASSERT_FALSE(iter->Valid());
+  ASSERT_TRUE(iter->status().IsCorruption());
+  delete iter;
+  iter = db_->NewIterator(ReadOptions());
+  iter->Seek("k2");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  iter->Prev();
+  ASSERT_FALSE(iter->Valid());
+  ASSERT_TRUE(iter->status().IsCorruption());
+  delete iter;
  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v1"}, {"k2", "v2"}});

  DestroyAndReopen(options);
  ASSERT_OK(Merge("k1", "v1"));
  ASSERT_OK(Put("k2", "v2"));
  ASSERT_OK(Merge("k2", "corrupted"));
-  VerifyDBFromMap({{"k1", "v1"}, {"k2", ""}}, nullptr, false,
-                  {{"k2", Status::Corruption()}});
+  iter = db_->NewIterator(ReadOptions());
+  iter->Seek("k1");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  iter->Next();
+  ASSERT_FALSE(iter->Valid());
+  ASSERT_TRUE(iter->status().IsCorruption());
+  delete iter;
  VerifyDBInternal({{"k1", "v1"}, {"k2", "corrupted"}, {"k2", "v2"}});
 }

@@ -432,8 +482,152 @@ TEST_P(MergeOperatorPinningTest, TailingIterator) {
  writer_thread.join();
  reader_thread.join();
 }
+
+TEST_F(DBMergeOperatorTest, TailingIteratorMemtableUnrefedBySomeoneElse) {
+  Options options = CurrentOptions();
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  DestroyAndReopen(options);
+
+  // Overview of the test:
+  //  * There are two merge operands for the same key: one in an sst file,
+  //    another in a memtable.
+  //  * Seek a tailing iterator to this key.
+  //  * As part of the seek, the iterator will:
+  //      (a) first visit the operand in the memtable and tell ForwardIterator
+  //          to pin this operand, then
+  //      (b) move on to the operand in the sst file, then pass both operands
+  //          to merge operator.
+  //  * The memtable may get flushed and unreferenced by another thread between
+  //    (a) and (b). The test simulates it by flushing the memtable inside a
+  //    SyncPoint callback located between (a) and (b).
+  //  * In this case it's ForwardIterator's responsibility to keep the memtable
+  //    pinned until (b) is complete. There used to be a bug causing
+  //    ForwardIterator to not pin it in some circumstances. This test
+  //    reproduces it.
+
+  db_->Merge(WriteOptions(), "key", "sst");
+  db_->Flush(FlushOptions()); // Switch to SuperVersion A
+  db_->Merge(WriteOptions(), "key", "memtable");
+
+  // Pin SuperVersion A
+  std::unique_ptr<Iterator> someone_else(db_->NewIterator(ReadOptions()));
+
+  bool pushed_first_operand = false;
+  bool stepped_to_next_operand = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBIter::MergeValuesNewToOld:PushedFirstOperand", [&](void*) {
+        EXPECT_FALSE(pushed_first_operand);
+        pushed_first_operand = true;
+        db_->Flush(FlushOptions()); // Switch to SuperVersion B
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBIter::MergeValuesNewToOld:SteppedToNextOperand", [&](void*) {
+        EXPECT_FALSE(stepped_to_next_operand);
+        stepped_to_next_operand = true;
+        someone_else.reset(); // Unpin SuperVersion A
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  ReadOptions ro;
+  ro.tailing = true;
+  std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+  iter->Seek("key");
+
+  ASSERT_TRUE(iter->status().ok());
+  ASSERT_TRUE(iter->Valid());
+  EXPECT_EQ(std::string("sst,memtable"), iter->value().ToString());
+  EXPECT_TRUE(pushed_first_operand);
+  EXPECT_TRUE(stepped_to_next_operand);
+}
 #endif  // ROCKSDB_LITE

+TEST_F(DBMergeOperatorTest, SnapshotCheckerAndReadCallback) {
+  Options options = CurrentOptions();
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  DestroyAndReopen(options);
+
+  class TestSnapshotChecker : public SnapshotChecker {
+    bool IsInSnapshot(SequenceNumber seq,
+                      SequenceNumber snapshot_seq) const override {
+      switch (snapshot_seq) {
+        case 0:
+          return seq == 0;
+        case 1:
+          return seq <= 1;
+        case 2:
+          // seq = 2 not visible to snapshot with seq = 2
+          return seq <= 1;
+        case 3:
+          return seq <= 3;
+        case 4:
+          // seq = 4 not visible to snpahost with seq = 4
+          return seq <= 3;
+        default:
+          // seq >=4 is uncommitted
+          return seq <= 4;
+      };
+    }
+  };
+  TestSnapshotChecker* snapshot_checker = new TestSnapshotChecker();
+  dbfull()->SetSnapshotChecker(snapshot_checker);
+
+  std::string value;
+  ASSERT_OK(Merge("foo", "v1"));
+  ASSERT_EQ(1, db_->GetLatestSequenceNumber());
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo"));
+  ASSERT_OK(Merge("foo", "v2"));
+  ASSERT_EQ(2, db_->GetLatestSequenceNumber());
+  // v2 is not visible to latest snapshot, which has seq = 2.
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo"));
+  // Take a snapshot with seq = 2.
+  const Snapshot* snapshot1 = db_->GetSnapshot();
+  ASSERT_EQ(2, snapshot1->GetSequenceNumber());
+  // v2 is not visible to snapshot1, which has seq = 2
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo", snapshot1));
+
+  // Verify flush doesn't alter the result.
+  ASSERT_OK(Flush());
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo", snapshot1));
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo"));
+
+  ASSERT_OK(Merge("foo", "v3"));
+  ASSERT_EQ(3, db_->GetLatestSequenceNumber());
+  ASSERT_EQ("v1,v2,v3", GetWithReadCallback(snapshot_checker, "foo"));
+  ASSERT_OK(Merge("foo", "v4"));
+  ASSERT_EQ(4, db_->GetLatestSequenceNumber());
+  // v4 is not visible to latest snapshot, which has seq = 4.
+  ASSERT_EQ("v1,v2,v3", GetWithReadCallback(snapshot_checker, "foo"));
+  const Snapshot* snapshot2 = db_->GetSnapshot();
+  ASSERT_EQ(4, snapshot2->GetSequenceNumber());
+  // v4 is not visible to snapshot2, which has seq = 4.
+  ASSERT_EQ("v1,v2,v3",
+            GetWithReadCallback(snapshot_checker, "foo", snapshot2));
+
+  // Verify flush doesn't alter the result.
+  ASSERT_OK(Flush());
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo", snapshot1));
+  ASSERT_EQ("v1,v2,v3",
+            GetWithReadCallback(snapshot_checker, "foo", snapshot2));
+  ASSERT_EQ("v1,v2,v3", GetWithReadCallback(snapshot_checker, "foo"));
+
+  ASSERT_OK(Merge("foo", "v5"));
+  ASSERT_EQ(5, db_->GetLatestSequenceNumber());
+  // v5 is uncommitted
+  ASSERT_EQ("v1,v2,v3,v4", GetWithReadCallback(snapshot_checker, "foo"));
+
+  // full manual compaction.
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  // Verify compaction doesn't alter the result.
+  ASSERT_EQ("v1", GetWithReadCallback(snapshot_checker, "foo", snapshot1));
+  ASSERT_EQ("v1,v2,v3",
+            GetWithReadCallback(snapshot_checker, "foo", snapshot2));
+  ASSERT_EQ("v1,v2,v3,v4", GetWithReadCallback(snapshot_checker, "foo"));
+
+  db_->ReleaseSnapshot(snapshot1);
+  db_->ReleaseSnapshot(snapshot2);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -518,8 +518,13 @@ static void assert_candidate_files_empty(DBImpl* dbfull, const bool empty) {
  JobContext job_context(0);
  dbfull->FindObsoleteFiles(&job_context, false);
  ASSERT_EQ(empty, job_context.full_scan_candidate_files.empty());
-  job_context.Clean();
  dbfull->TEST_UnlockMutex();
+  if (job_context.HaveSomethingToDelete()) {
+    // fulfill the contract of FindObsoleteFiles by calling PurgeObsoleteFiles
+    // afterwards; otherwise the test may hang on shutdown
+    dbfull->PurgeObsoleteFiles(job_context);
+  }
+  job_context.Clean();
 }

 TEST_F(DBOptionsTest, DeleteObsoleteFilesPeriodChange) {
@@ -692,6 +697,32 @@ TEST_F(DBOptionsTest, SetFIFOCompactionOptions) {
  ASSERT_LE(NumTableFilesAtLevel(0), 5);
 }

+TEST_F(DBOptionsTest, CompactionReadaheadSizeChange) {
+  SpecialEnv env(env_);
+  Options options;
+  options.env = &env;
+
+  options.compaction_readahead_size = 0;
+  options.new_table_reader_for_compaction_inputs = true;
+  options.level0_file_num_compaction_trigger = 2;
+  const std::string kValue(1024, 'v');
+  Reopen(options);
+
+  ASSERT_EQ(0, dbfull()->GetDBOptions().compaction_readahead_size);
+  ASSERT_OK(dbfull()->SetDBOptions({{"compaction_readahead_size", "256"}}));
+  ASSERT_EQ(256, dbfull()->GetDBOptions().compaction_readahead_size);
+  for (int i = 0; i < 1024; i++) {
+    Put(Key(i), kValue);
+  }
+  Flush();
+  for (int i = 0; i < 1024 * 2; i++) {
+    Put(Key(i), kValue);
+  }
+  Flush();
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(256, env_->compaction_readahead_size_);
+  Close();
+}
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -330,11 +330,11 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  int64_t rate_bytes_per_sec = 1024 * 10;  // 10 Kbs / Sec
  Status s;
  options.sst_file_manager.reset(
-      NewSstFileManager(env_, nullptr, "", 0, false, &s));
+      NewSstFileManager(env_, nullptr, "", 0, false, &s, 0));
  ASSERT_OK(s);
  options.sst_file_manager->SetDeleteRateBytesPerSecond(rate_bytes_per_sec);
  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());
-  sfm->delete_scheduler()->TEST_SetMaxTrashDBRatio(1.1);
+  sfm->delete_scheduler()->SetMaxTrashDBRatio(1.1);

  ASSERT_OK(TryReopen(options));
  // Create 4 files in L0
@@ -376,6 +376,30 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

+TEST_F(DBSSTTest, OpenDBWithExistingTrash) {
+  Options options = CurrentOptions();
+
+  options.sst_file_manager.reset(
+      NewSstFileManager(env_, nullptr, "", 1024 * 1024 /* 1 MB/sec */));
+  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());
+
+  Destroy(last_options_);
+
+  // Add some trash files to the db directory so the DB can clean them up
+  env_->CreateDirIfMissing(dbname_);
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "001.sst.trash"));
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "002.sst.trash"));
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "003.sst.trash"));
+
+  // Reopen the DB and verify that it deletes existing trash files
+  ASSERT_OK(TryReopen(options));
+  sfm->WaitForEmptyTrash();
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "001.sst.trash"));
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "002.sst.trash"));
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "003.sst.trash"));
+}
+
+
 // Create a DB with 2 db_paths, and generate multiple files in the 2
 // db_paths using CompactRangeOptions, make sure that files that were
 // deleted from first db_path were deleted using DeleteScheduler and
@@ -396,10 +420,11 @@ TEST_F(DBSSTTest, DeleteSchedulerMultipleDBPaths) {
  int64_t rate_bytes_per_sec = 1024 * 1024;  // 1 Mb / Sec
  Status s;
  options.sst_file_manager.reset(
-      NewSstFileManager(env_, nullptr, "", rate_bytes_per_sec, false, &s));
+      NewSstFileManager(env_, nullptr, "", rate_bytes_per_sec, false, &s,
+                        /* max_trash_db_ratio= */ 1.1));
+
  ASSERT_OK(s);
  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());
-  sfm->delete_scheduler()->TEST_SetMaxTrashDBRatio(1.1);

  DestroyAndReopen(options);

@@ -459,7 +484,7 @@ TEST_F(DBSSTTest, DestroyDBWithRateLimitedDelete) {
  options.disable_auto_compactions = true;
  options.env = env_;
  options.sst_file_manager.reset(
-      NewSstFileManager(env_, nullptr, "", 0, false, &s));
+      NewSstFileManager(env_, nullptr, "", 0, false, &s, 0));
  ASSERT_OK(s);
  DestroyAndReopen(options);

@@ -477,7 +502,7 @@ TEST_F(DBSSTTest, DestroyDBWithRateLimitedDelete) {
  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());

  sfm->SetDeleteRateBytesPerSecond(1024 * 1024);
-  sfm->delete_scheduler()->TEST_SetMaxTrashDBRatio(1.1);
+  sfm->delete_scheduler()->SetMaxTrashDBRatio(1.1);
  ASSERT_OK(DestroyDB(dbname_, options));
  sfm->WaitForEmptyTrash();
  // We have deleted the 4 sst files in the delete_scheduler
@@ -480,6 +480,36 @@ TEST_F(DBTest, SingleDeletePutFlush) {
                         kSkipUniversalCompaction | kSkipMergePut));
 }

+// Disable because not all platform can run it.
+// It requires more than 9GB memory to run it, With single allocation
+// of more than 3GB.
+TEST_F(DBTest, DISABLED_SanitizeVeryVeryLargeValue) {
+  const size_t kValueSize = 4 * size_t{1024 * 1024 * 1024};  // 4GB value
+  std::string raw(kValueSize, 'v');
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.merge_operator = MergeOperators::CreatePutOperator();
+  options.write_buffer_size = 100000;  // Small write buffer
+  options.paranoid_checks = true;
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put("boo", "v1"));
+  ASSERT_TRUE(Put("foo", raw).IsInvalidArgument());
+  ASSERT_TRUE(Merge("foo", raw).IsInvalidArgument());
+
+  WriteBatch wb;
+  ASSERT_TRUE(wb.Put("foo", raw).IsInvalidArgument());
+  ASSERT_TRUE(wb.Merge("foo", raw).IsInvalidArgument());
+
+  Slice value_slice = raw;
+  Slice key_slice = "foo";
+  SliceParts sp_key(&key_slice, 1);
+  SliceParts sp_value(&value_slice, 1);
+
+  ASSERT_TRUE(wb.Put(sp_key, sp_value).IsInvalidArgument());
+  ASSERT_TRUE(wb.Merge(sp_key, sp_value).IsInvalidArgument());
+}
+
 // Disable because not all platform can run it.
 // It requires more than 9GB memory to run it, With single allocation
 // of more than 3GB.
@@ -503,7 +533,9 @@ TEST_F(DBTest, DISABLED_VeryLargeValue) {
  ASSERT_OK(Put(key2, raw));
  dbfull()->TEST_WaitForFlushMemTable();

+#ifndef ROCKSDB_LITE
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+#endif  // !ROCKSDB_LITE

  std::string value;
  Status s = db_->Get(ReadOptions(), key1, &value);
@@ -2191,6 +2223,8 @@ class ModelDB : public DB {
    batch.Put(cf, k, v);
    return Write(o, &batch);
  }
+  using DB::Close;
+  virtual Status Close() override { return Status::OK(); }
  using DB::Delete;
  virtual Status Delete(const WriteOptions& o, ColumnFamilyHandle* cf,
                        const Slice& key) override {
@@ -3354,11 +3388,23 @@ TEST_F(DBTest, DynamicMemtableOptions) {
      {"write_buffer_size", "131072"},
  }));

-  // The existing memtable is still 64KB in size, after it becomes immutable,
-  // the next memtable will be 128KB in size. Write 256KB total, we should
-  // have a 64KB L0 file, a 128KB L0 file, and a memtable with 64KB data
-  gen_l0_kb(256);
-  ASSERT_EQ(NumTableFilesAtLevel(0), 2);  // (A)
+  // The existing memtable inflated 64KB->128KB when we invoked SetOptions().
+  // Write 192KB, we should have a 128KB L0 file and a memtable with 64KB data.
+  gen_l0_kb(192);
+  ASSERT_EQ(NumTableFilesAtLevel(0), 1);  // (A)
+  ASSERT_LT(SizeAtLevel(0), k128KB + 2 * k5KB);
+  ASSERT_GT(SizeAtLevel(0), k128KB - 4 * k5KB);
+
+  // Decrease buffer size below current usage
+  ASSERT_OK(dbfull()->SetOptions({
+      {"write_buffer_size", "65536"},
+  }));
+  // The existing memtable became eligible for flush when we reduced its
+  // capacity to 64KB. Two keys need to be added to trigger flush: first causes
+  // memtable to be marked full, second schedules the flush. Then we should have
+  // a 128KB L0 file, a 64KB L0 file, and a memtable with just one key.
+  gen_l0_kb(2);
+  ASSERT_EQ(NumTableFilesAtLevel(0), 2);
  ASSERT_LT(SizeAtLevel(0), k128KB + k64KB + 2 * k5KB);
  ASSERT_GT(SizeAtLevel(0), k128KB + k64KB - 4 * k5KB);

@@ -3476,12 +3522,16 @@ TEST_F(DBTest, GetThreadStatus) {
    const int kTestCount = 3;
    const unsigned int kHighPriCounts[kTestCount] = {3, 2, 5};
    const unsigned int kLowPriCounts[kTestCount] = {10, 15, 3};
+    const unsigned int kBottomPriCounts[kTestCount] = {2, 1, 4};
    for (int test = 0; test < kTestCount; ++test) {
      // Change the number of threads in high / low priority pool.
      env_->SetBackgroundThreads(kHighPriCounts[test], Env::HIGH);
      env_->SetBackgroundThreads(kLowPriCounts[test], Env::LOW);
+      env_->SetBackgroundThreads(kBottomPriCounts[test], Env::BOTTOM);
      // Wait to ensure the all threads has been registered
      unsigned int thread_type_counts[ThreadStatus::NUM_THREAD_TYPES];
+      // TODO(ajkr): it'd be better if SetBackgroundThreads returned only after
+      // all threads have been registered.
      // Try up to 60 seconds.
      for (int num_try = 0; num_try < 60000; num_try++) {
        env_->SleepForMicroseconds(1000);
@@ -3496,20 +3546,21 @@ TEST_F(DBTest, GetThreadStatus) {
        if (thread_type_counts[ThreadStatus::HIGH_PRIORITY] ==
                kHighPriCounts[test] &&
            thread_type_counts[ThreadStatus::LOW_PRIORITY] ==
-                kLowPriCounts[test]) {
+                kLowPriCounts[test] &&
+            thread_type_counts[ThreadStatus::BOTTOM_PRIORITY] ==
+                kBottomPriCounts[test]) {
          break;
        }
      }
-      // Verify the total number of threades
-      ASSERT_EQ(thread_type_counts[ThreadStatus::HIGH_PRIORITY] +
-                    thread_type_counts[ThreadStatus::LOW_PRIORITY],
-                kHighPriCounts[test] + kLowPriCounts[test]);
      // Verify the number of high-priority threads
      ASSERT_EQ(thread_type_counts[ThreadStatus::HIGH_PRIORITY],
                kHighPriCounts[test]);
      // Verify the number of low-priority threads
      ASSERT_EQ(thread_type_counts[ThreadStatus::LOW_PRIORITY],
                kLowPriCounts[test]);
+      // Verify the number of bottom-priority threads
+      ASSERT_EQ(thread_type_counts[ThreadStatus::BOTTOM_PRIORITY],
+                kBottomPriCounts[test]);
    }
    if (i == 0) {
      // repeat the test with multiple column families
@@ -4363,6 +4414,74 @@ TEST_F(DBTest, DynamicFIFOCompactionOptions) {
  ASSERT_EQ(dbfull()->GetOptions().compaction_options_fifo.allow_compaction,
            true);
 }
+
+TEST_F(DBTest, DynamicUniversalCompactionOptions) {
+  Options options;
+  options.create_if_missing = true;
+  DestroyAndReopen(options);
+
+  // Initial defaults
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.size_ratio, 1);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.min_merge_width,
+            2);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.max_merge_width,
+            UINT_MAX);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.max_size_amplification_percent,
+            200);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.compression_size_percent,
+            -1);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.stop_style,
+            kCompactionStopStyleTotalSize);
+  ASSERT_EQ(
+      dbfull()->GetOptions().compaction_options_universal.allow_trivial_move,
+      false);
+
+  ASSERT_OK(dbfull()->SetOptions(
+      {{"compaction_options_universal", "{size_ratio=7;}"}}));
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.size_ratio, 7);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.min_merge_width,
+            2);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.max_merge_width,
+            UINT_MAX);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.max_size_amplification_percent,
+            200);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.compression_size_percent,
+            -1);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.stop_style,
+            kCompactionStopStyleTotalSize);
+  ASSERT_EQ(
+      dbfull()->GetOptions().compaction_options_universal.allow_trivial_move,
+      false);
+
+  ASSERT_OK(dbfull()->SetOptions(
+      {{"compaction_options_universal", "{min_merge_width=11;}"}}));
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.size_ratio, 7);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.min_merge_width,
+            11);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.max_merge_width,
+            UINT_MAX);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.max_size_amplification_percent,
+            200);
+  ASSERT_EQ(dbfull()
+                ->GetOptions()
+                .compaction_options_universal.compression_size_percent,
+            -1);
+  ASSERT_EQ(dbfull()->GetOptions().compaction_options_universal.stop_style,
+            kCompactionStopStyleTotalSize);
+  ASSERT_EQ(
+      dbfull()->GetOptions().compaction_options_universal.allow_trivial_move,
+      false);
+}
 #endif  // ROCKSDB_LITE

 TEST_F(DBTest, FileCreationRandomFailure) {
@@ -4969,55 +5088,84 @@ TEST_F(DBTest, PromoteL0Failure) {
  status = experimental::PromoteL0(db_, db_->DefaultColumnFamily());
  ASSERT_TRUE(status.IsInvalidArgument());
 }
-#endif  // ROCKSDB_LITE

 // Github issue #596
-TEST_F(DBTest, HugeNumberOfLevels) {
+TEST_F(DBTest, CompactRangeWithEmptyBottomLevel) {
+  const int kNumLevels = 2;
+  const int kNumL0Files = 2;
  Options options = CurrentOptions();
-  options.write_buffer_size = 2 * 1024 * 1024;         // 2MB
-  options.max_bytes_for_level_base = 2 * 1024 * 1024;  // 2MB
-  options.num_levels = 12;
-  options.max_background_compactions = 10;
-  options.max_bytes_for_level_multiplier = 2;
-  options.level_compaction_dynamic_level_bytes = true;
+  options.disable_auto_compactions = true;
+  options.num_levels = kNumLevels;
  DestroyAndReopen(options);

  Random rnd(301);
-  for (int i = 0; i < 300000; ++i) {
-    ASSERT_OK(Put(Key(i), RandomString(&rnd, 1024)));
+  for (int i = 0; i < kNumL0Files; ++i) {
+    ASSERT_OK(Put(Key(0), RandomString(&rnd, 1024)));
+    Flush();
  }
+  ASSERT_EQ(NumTableFilesAtLevel(0), kNumL0Files);
+  ASSERT_EQ(NumTableFilesAtLevel(1), 0);

  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_EQ(NumTableFilesAtLevel(0), 0);
+  ASSERT_EQ(NumTableFilesAtLevel(1), kNumL0Files);
 }
+#endif  // ROCKSDB_LITE

 TEST_F(DBTest, AutomaticConflictsWithManualCompaction) {
+  const int kNumL0Files = 50;
  Options options = CurrentOptions();
-  options.write_buffer_size = 2 * 1024 * 1024;         // 2MB
-  options.max_bytes_for_level_base = 2 * 1024 * 1024;  // 2MB
-  options.num_levels = 12;
+  options.level0_file_num_compaction_trigger = 4;
+  // never slowdown / stop
+  options.level0_slowdown_writes_trigger = 999999;
+  options.level0_stop_writes_trigger = 999999;
  options.max_background_compactions = 10;
-  options.max_bytes_for_level_multiplier = 2;
-  options.level_compaction_dynamic_level_bytes = true;
  DestroyAndReopen(options);

-  Random rnd(301);
-  for (int i = 0; i < 300000; ++i) {
-    ASSERT_OK(Put(Key(i), RandomString(&rnd, 1024)));
-  }
-
+  // schedule automatic compactions after the manual one starts, but before it
+  // finishes to ensure conflict.
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::BackgroundCompaction:Start",
+        "DBTest::AutomaticConflictsWithManualCompaction:PrePuts"},
+       {"DBTest::AutomaticConflictsWithManualCompaction:PostPuts",
+        "DBImpl::BackgroundCompaction:NonTrivial:AfterRun"}});
  std::atomic<int> callback_count(0);
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction()::Conflict",
+      "DBImpl::MaybeScheduleFlushOrCompaction:Conflict",
      [&](void* arg) { callback_count.fetch_add(1); });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  CompactRangeOptions croptions;
-  croptions.exclusive_manual_compaction = false;
-  ASSERT_OK(db_->CompactRange(croptions, nullptr, nullptr));
+
+  Random rnd(301);
+  for (int i = 0; i < 2; ++i) {
+    // put two keys to ensure no trivial move
+    for (int j = 0; j < 2; ++j) {
+      ASSERT_OK(Put(Key(j), RandomString(&rnd, 1024)));
+    }
+    ASSERT_OK(Flush());
+  }
+  std::thread manual_compaction_thread([this]() {
+    CompactRangeOptions croptions;
+    croptions.exclusive_manual_compaction = true;
+    ASSERT_OK(db_->CompactRange(croptions, nullptr, nullptr));
+  });
+
+  TEST_SYNC_POINT("DBTest::AutomaticConflictsWithManualCompaction:PrePuts");
+  for (int i = 0; i < kNumL0Files; ++i) {
+    // put two keys to ensure no trivial move
+    for (int j = 0; j < 2; ++j) {
+      ASSERT_OK(Put(Key(j), RandomString(&rnd, 1024)));
+    }
+    ASSERT_OK(Flush());
+  }
+  TEST_SYNC_POINT("DBTest::AutomaticConflictsWithManualCompaction:PostPuts");
+
  ASSERT_GE(callback_count.load(), 1);
-  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
-  for (int i = 0; i < 300000; ++i) {
+  for (int i = 0; i < 2; ++i) {
    ASSERT_NE("NOT_FOUND", Get(Key(i)));
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  manual_compaction_thread.join();
+  dbfull()->TEST_WaitForCompact();
 }

 // Github issue #595
@@ -5237,12 +5385,15 @@ class WriteStallListener : public EventListener {
 public:
  WriteStallListener() : condition_(WriteStallCondition::kNormal) {}
  void OnStallConditionsChanged(const WriteStallInfo& info) override {
+    MutexLock l(&mutex_);
    condition_ = info.condition.cur;
  }
  bool CheckCondition(WriteStallCondition expected) {
+    MutexLock l(&mutex_);
    return expected == condition_;
  }
 private:
+  port::Mutex mutex_;
  WriteStallCondition condition_;
 };

@@ -5571,6 +5722,50 @@ TEST_F(DBTest, PauseBackgroundWorkTest) {
  // now it's done
  ASSERT_TRUE(done.load());
 }
+
+// Keep spawning short-living threads that create an iterator and quit.
+// Meanwhile in another thread keep flushing memtables.
+// This used to cause a deadlock.
+TEST_F(DBTest, ThreadLocalPtrDeadlock) {
+  std::atomic<int> flushes_done{0};
+  std::atomic<int> threads_destroyed{0};
+  auto done = [&] {
+    return flushes_done.load() > 10;
+  };
+
+  std::thread flushing_thread([&] {
+    for (int i = 0; !done(); ++i) {
+      ASSERT_OK(db_->Put(WriteOptions(), Slice("hi"),
+                         Slice(std::to_string(i).c_str())));
+      ASSERT_OK(db_->Flush(FlushOptions()));
+      int cnt = ++flushes_done;
+      fprintf(stderr, "Flushed %d times\n", cnt);
+    }
+  });
+
+  std::vector<std::thread> thread_spawning_threads(10);
+  for (auto& t: thread_spawning_threads) {
+    t = std::thread([&] {
+      while (!done()) {
+        {
+          std::thread tmp_thread([&] {
+            auto it = db_->NewIterator(ReadOptions());
+            delete it;
+          });
+          tmp_thread.join();
+        }
+        ++threads_destroyed;
+      }
+    });
+  }
+
+  for (auto& t: thread_spawning_threads) {
+    t.join();
+  }
+  flushing_thread.join();
+  fprintf(stderr, "Done. Flushed %d times, destroyed %d threads\n",
+          flushes_done.load(), threads_destroyed.load());
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1027,6 +1027,7 @@ TEST_F(DBTest2, PresetCompressionDict) {
  const size_t kL0FileBytes = 128 << 10;
  const size_t kApproxPerBlockOverheadBytes = 50;
  const int kNumL0Files = 5;
+  const int kZstdTrainFactor = 16;

  Options options;
  options.env = CurrentOptions().env; // Make sure to use any custom env that the test is configured with.
@@ -1059,17 +1060,34 @@ TEST_F(DBTest2, PresetCompressionDict) {
  for (auto compression_type : compression_types) {
    options.compression = compression_type;
    size_t prev_out_bytes;
-    for (int i = 0; i < 2; ++i) {
+    for (int i = 0; i < 3; ++i) {
      // First iteration: compress without preset dictionary
      // Second iteration: compress with preset dictionary
-      // To make sure the compression dictionary was actually used, we verify
-      // the compressed size is smaller in the second iteration. Also in the
-      // second iteration, verify the data we get out is the same data we put
-      // in.
-      if (i) {
-        options.compression_opts.max_dict_bytes = kBlockSizeBytes;
-      } else {
-        options.compression_opts.max_dict_bytes = 0;
+      // Third iteration (zstd only): compress with zstd-trained dictionary
+      //
+      // To make sure the compression dictionary has the intended effect, we
+      // verify the compressed size is smaller in successive iterations. Also in
+      // the non-first iterations, verify the data we get out is the same data
+      // we put in.
+      switch (i) {
+        case 0:
+          options.compression_opts.max_dict_bytes = 0;
+          options.compression_opts.zstd_max_train_bytes = 0;
+          break;
+        case 1:
+          options.compression_opts.max_dict_bytes = kBlockSizeBytes;
+          options.compression_opts.zstd_max_train_bytes = 0;
+          break;
+        case 2:
+          if (compression_type != kZSTD) {
+            continue;
+          }
+          options.compression_opts.max_dict_bytes = kBlockSizeBytes;
+          options.compression_opts.zstd_max_train_bytes =
+              kZstdTrainFactor * kBlockSizeBytes;
+          break;
+        default:
+          assert(false);
      }

      options.statistics = rocksdb::CreateDBStatistics();
@@ -1134,7 +1152,7 @@ class CompactionCompressionListener : public EventListener {
    }

    if (db_options_->bottommost_compression != kDisableCompressionOption &&
-        ci.output_level == bottommost_level && ci.output_level >= 2) {
+        ci.output_level == bottommost_level) {
      ASSERT_EQ(ci.compression, db_options_->bottommost_compression);
    } else if (db_options_->compression_per_level.size() != 0) {
      ASSERT_EQ(ci.compression,
@@ -1802,11 +1820,17 @@ TEST_F(DBTest2, ReadAmpBitmap) {

 #ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
 TEST_F(DBTest2, ReadAmpBitmapLiveInCacheAfterDBClose) {
-  if (dbname_.find("dev/shm") != std::string::npos) {
-    // /dev/shm dont support getting a unique file id, this mean that
-    // running this test on /dev/shm will fail because lru_cache will load
-    // the blocks again regardless of them being already in the cache
-    return;
+  {
+    const int kIdBufLen = 100;
+    char id_buf[kIdBufLen];
+    std::unique_ptr<RandomAccessFile> file;
+    env_->NewRandomAccessFile(dbname_, &file, EnvOptions());
+    if (file->GetUniqueId(id_buf, kIdBufLen) == 0) {
+      // fs holding db directory doesn't support getting a unique file id,
+      // this means that running this test will fail because lru_cache will load
+      // the blocks again regardless of them being already in the cache
+      return;
+    }
  }
  uint32_t bytes_per_bit[2] = {1, 16};
  for (size_t k = 0; k < 2; k++) {
@@ -2412,6 +2436,60 @@ TEST_F(DBTest2, ReadCallbackTest) {
  }
 }

+#ifndef ROCKSDB_LITE
+
+TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
+  // Regression test for race condition where an obsolete file is returned to
+  // user as a "live file" but then deleted, all while file deletions are
+  // disabled.
+  //
+  // It happened like this:
+  //
+  // 1. [flush thread] Log file "x.log" found by FindObsoleteFiles
+  // 2. [user thread] DisableFileDeletions, GetSortedWalFiles are called and the
+  //    latter returned "x.log"
+  // 3. [flush thread] PurgeObsoleteFiles deleted "x.log"
+  // 4. [user thread] Reading "x.log" failed
+  //
+  // Unfortunately the only regression test I can come up with involves sleep.
+  // We cannot set SyncPoints to repro since, once the fix is applied, the
+  // SyncPoints would cause a deadlock as the repro's sequence of events is now
+  // prohibited.
+  //
+  // Instead, if we sleep for a second between Find and Purge, and ensure the
+  // read attempt happens after purge, then the sequence of events will almost
+  // certainly happen on the old code.
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"DBImpl::BackgroundCallFlush:FilesFound",
+       "DBTest2::LiveFilesOmitObsoleteFiles:FlushTriggered"},
+      {"DBImpl::PurgeObsoleteFiles:End",
+       "DBTest2::LiveFilesOmitObsoleteFiles:LiveFilesCaptured"},
+  });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::PurgeObsoleteFiles:Begin",
+      [&](void* arg) { env_->SleepForMicroseconds(1000000); });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put("key", "val");
+  FlushOptions flush_opts;
+  flush_opts.wait = false;
+  db_->Flush(flush_opts);
+  TEST_SYNC_POINT("DBTest2::LiveFilesOmitObsoleteFiles:FlushTriggered");
+
+  db_->DisableFileDeletions();
+  VectorLogPtr log_files;
+  db_->GetSortedWalFiles(log_files);
+  TEST_SYNC_POINT("DBTest2::LiveFilesOmitObsoleteFiles:LiveFilesCaptured");
+  for (const auto& log_file : log_files) {
+    ASSERT_OK(env_->FileExists(LogFileName(dbname_, log_file->LogNumber())));
+  }
+
+  db_->EnableFileDeletions();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -486,7 +486,7 @@ Options DBTestBase::GetOptions(
    }
    case kConcurrentWALWrites: {
      // This options optimize 2PC commit path
-      options.concurrent_prepare = true;
+      options.two_write_queues = true;
      options.manual_wal_flush = true;
      break;
    }
@@ -867,7 +867,6 @@ size_t DBTestBase::CountLiveFiles() {
  db_->GetLiveFilesMetaData(&metadata);
  return metadata.size();
 }
-#endif  // ROCKSDB_LITE

 int DBTestBase::NumTableFilesAtLevel(int level, int cf) {
  std::string property;
@@ -928,6 +927,7 @@ std::string DBTestBase::FilesPerLevel(int cf) {
  result.resize(last_non_zero_offset);
  return result;
 }
+#endif  // !ROCKSDB_LITE

 size_t DBTestBase::CountFiles() {
  std::vector<std::string> files;
@@ -997,6 +997,7 @@ void DBTestBase::MoveFilesToLevel(int level, int cf) {
  }
 }

+#ifndef ROCKSDB_LITE
 void DBTestBase::DumpFileCounts(const char* label) {
  fprintf(stderr, "---\n%s:\n", label);
  fprintf(stderr, "maxoverlap: %" PRIu64 "\n",
@@ -1008,6 +1009,7 @@ void DBTestBase::DumpFileCounts(const char* label) {
    }
  }
 }
+#endif  // !ROCKSDB_LITE

 std::string DBTestBase::DumpSSTableList() {
  std::string property;
@@ -1157,11 +1159,13 @@ UpdateStatus DBTestBase::updateInPlaceNoAction(char* prevValue,

 // Utility method to test InplaceUpdate
 void DBTestBase::validateNumberOfEntries(int numValues, int cf) {
-  ScopedArenaIterator iter;
  Arena arena;
  auto options = CurrentOptions();
  InternalKeyComparator icmp(options.comparator);
  RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
+  // This should be defined after range_del_agg so that it destructs the
+  // assigned iterator before it range_del_agg is already destructed.
+  ScopedArenaIterator iter;
  if (cf != 0) {
    iter.set(
        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[cf]));
@@ -137,8 +137,8 @@ class SpecialMemTableRep : public MemTableRep {
  // Insert key into the list.
  // REQUIRES: nothing that compares equal to key is currently in the list.
  virtual void Insert(KeyHandle handle) override {
-    memtable_->Insert(handle);
    num_entries_++;
+    memtable_->Insert(handle);
  }

  // Returns true iff an entry that compares equal to key is in the list.
@@ -445,6 +445,9 @@ class SpecialEnv : public EnvWrapper {
      r->reset(new CountingFile(std::move(*r), &random_read_counter_,
                                &random_read_bytes_counter_));
    }
+    if (s.ok() && soptions.compaction_readahead_size > 0) {
+      compaction_readahead_size_ = soptions.compaction_readahead_size;
+    }
    return s;
  }

@@ -570,6 +573,8 @@ class SpecialEnv : public EnvWrapper {
  bool no_slowdown_;

  std::atomic<bool> is_wal_sync_thread_safe_{true};
+
+  std::atomic<size_t> compaction_readahead_size_;
 };

 class MockTimeEnv : public EnvWrapper {
@@ -858,13 +863,13 @@ class DBTestBase : public testing::Test {
  size_t TotalLiveFiles(int cf = 0);

  size_t CountLiveFiles();
-#endif  // ROCKSDB_LITE

  int NumTableFilesAtLevel(int level, int cf = 0);

  double CompressionRatioAtLevel(int level, int cf = 0);

  int TotalTableFiles(int cf = 0, int levels = -1);
+#endif  // ROCKSDB_LITE

  // Return spread of files per level
  std::string FilesPerLevel(int cf = 0);
@@ -892,7 +897,9 @@ class DBTestBase : public testing::Test {

  void MoveFilesToLevel(int level, int cf = 0);

+#ifndef ROCKSDB_LITE
  void DumpFileCounts(const char* label);
+#endif  // ROCKSDB_LITE

  std::string DumpSSTableList();

@@ -374,6 +374,181 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSizeAmplification) {
  ASSERT_EQ(NumSortedRuns(1), 1);
 }

+TEST_P(DBTestUniversalCompaction, DynamicUniversalCompactionSizeAmplification) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.num_levels = 1;
+  options.write_buffer_size = 100 << 10;     // 100KB
+  options.target_file_size_base = 32 << 10;  // 32KB
+  options.level0_file_num_compaction_trigger = 3;
+  // Initial setup of compaction_options_universal will prevent universal
+  // compaction from happening
+  options.compaction_options_universal.size_ratio = 100;
+  options.compaction_options_universal.min_merge_width = 100;
+  DestroyAndReopen(options);
+
+  int total_picked_compactions = 0;
+  int total_size_amp_compactions = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "UniversalCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        if (arg) {
+          total_picked_compactions++;
+          Compaction* c = static_cast<Compaction*>(arg);
+          if (c->compaction_reason() ==
+              CompactionReason::kUniversalSizeAmplification) {
+            total_size_amp_compactions++;
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  MutableCFOptions mutable_cf_options;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  Random rnd(301);
+  int key_idx = 0;
+
+  //   Generate two files in Level 0. Both files are approx the same size.
+  for (int num = 0; num < options.level0_file_num_compaction_trigger - 1;
+       num++) {
+    // Write 110KB (11 values, each 10K)
+    for (int i = 0; i < 11; i++) {
+      ASSERT_OK(Put(1, Key(key_idx), RandomString(&rnd, 10000)));
+      key_idx++;
+    }
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    ASSERT_EQ(NumSortedRuns(1), num + 1);
+  }
+  ASSERT_EQ(NumSortedRuns(1), 2);
+
+  // Flush whatever is remaining in memtable. This is typically
+  // small, which should not trigger size ratio based compaction
+  // but could instead trigger size amplification if it's set
+  // to 110.
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  // Verify compaction did not happen
+  ASSERT_EQ(NumSortedRuns(1), 3);
+
+  // Trigger compaction if size amplification exceeds 110% without reopening DB
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.max_size_amplification_percent,
+            200);
+  ASSERT_OK(dbfull()->SetOptions(handles_[1],
+                                 {{"compaction_options_universal",
+                                   "{max_size_amplification_percent=110;}"}}));
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.max_size_amplification_percent,
+            110);
+  ASSERT_OK(dbfull()->TEST_GetLatestMutableCFOptions(handles_[1],
+                                                     &mutable_cf_options));
+  ASSERT_EQ(110, mutable_cf_options.compaction_options_universal
+                     .max_size_amplification_percent);
+
+  dbfull()->TEST_WaitForCompact();
+  // Verify that size amplification did happen
+  ASSERT_EQ(NumSortedRuns(1), 1);
+  ASSERT_EQ(total_picked_compactions, 1);
+  ASSERT_EQ(total_size_amp_compactions, 1);
+}
+
+TEST_P(DBTestUniversalCompaction, DynamicUniversalCompactionReadAmplification) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.num_levels = 1;
+  options.write_buffer_size = 100 << 10;     // 100KB
+  options.target_file_size_base = 32 << 10;  // 32KB
+  options.level0_file_num_compaction_trigger = 3;
+  // Initial setup of compaction_options_universal will prevent universal
+  // compaction from happening
+  options.compaction_options_universal.max_size_amplification_percent = 2000;
+  options.compaction_options_universal.size_ratio = 0;
+  options.compaction_options_universal.min_merge_width = 100;
+  DestroyAndReopen(options);
+
+  int total_picked_compactions = 0;
+  int total_size_ratio_compactions = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "UniversalCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        if (arg) {
+          total_picked_compactions++;
+          Compaction* c = static_cast<Compaction*>(arg);
+          if (c->compaction_reason() == CompactionReason::kUniversalSizeRatio) {
+            total_size_ratio_compactions++;
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  MutableCFOptions mutable_cf_options;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  Random rnd(301);
+  int key_idx = 0;
+
+  // Generate three files in Level 0. All files are approx the same size.
+  for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+    // Write 110KB (11 values, each 10K)
+    for (int i = 0; i < 11; i++) {
+      ASSERT_OK(Put(1, Key(key_idx), RandomString(&rnd, 10000)));
+      key_idx++;
+    }
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    ASSERT_EQ(NumSortedRuns(1), num + 1);
+  }
+  ASSERT_EQ(NumSortedRuns(1), options.level0_file_num_compaction_trigger);
+
+  // Flush whatever is remaining in memtable. This is typically small, about
+  // 30KB.
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  // Verify compaction did not happen
+  ASSERT_EQ(NumSortedRuns(1), options.level0_file_num_compaction_trigger + 1);
+  ASSERT_EQ(total_picked_compactions, 0);
+
+  ASSERT_OK(dbfull()->SetOptions(
+      handles_[1],
+      {{"compaction_options_universal",
+        "{min_merge_width=2;max_merge_width=2;size_ratio=100;}"}}));
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.min_merge_width,
+            2);
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.max_merge_width,
+            2);
+  ASSERT_EQ(
+      dbfull()->GetOptions(handles_[1]).compaction_options_universal.size_ratio,
+      100);
+
+  ASSERT_OK(dbfull()->TEST_GetLatestMutableCFOptions(handles_[1],
+                                                     &mutable_cf_options));
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.size_ratio, 100);
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.min_merge_width, 2);
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.max_merge_width, 2);
+
+  dbfull()->TEST_WaitForCompact();
+
+  // Files in L0 are approx: 0.3 (30KB), 1, 1, 1.
+  // On compaction: the files are below the size amp threshold, so we
+  // fallthrough to checking read amp conditions. The configured size ratio is
+  // not big enough to take 0.3 into consideration. So the next files 1 and 1
+  // are compacted together first as they satisfy size ratio condition and
+  // (min_merge_width, max_merge_width) condition, to give out a file size of 2.
+  // Next, the newly generated 2 and the last file 1 are compacted together. So
+  // at the end: #sortedRuns = 2, #picked_compactions = 2, and all the picked
+  // ones are size ratio based compactions.
+  ASSERT_EQ(NumSortedRuns(1), 2);
+  // If max_merge_width had not been changed dynamically above, and if it
+  // continued to be the default value of UINIT_MAX, total_picked_compactions
+  // would have been 1.
+  ASSERT_EQ(total_picked_compactions, 2);
+  ASSERT_EQ(total_size_ratio_compactions, 2);
+}
+
 TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
  const int kTestKeySize = 16;
  const int kTestValueSize = 984;
@@ -730,7 +730,8 @@ class RecoveryTestHelper {
        batch.Put(key, value);
        WriteBatchInternal::SetSequence(&batch, seq);
        current_log_writer->AddRecord(WriteBatchInternal::Contents(&batch));
-        versions->SetLastToBeWrittenSequence(seq);
+        versions->SetLastAllocatedSequence(seq);
+        versions->SetLastPublishedSequence(seq);
        versions->SetLastSequence(seq);
      }
    }
@@ -3,12 +3,18 @@
 //  COPYING file in the root directory) and Apache 2.0 License
 //  (found in the LICENSE.Apache file in the root directory).

+#include <atomic>
 #include <memory>
 #include <thread>
 #include <vector>
 #include "db/db_test_util.h"
 #include "db/write_batch_internal.h"
+#include "db/write_thread.h"
+#include "port/port.h"
 #include "port/stack_trace.h"
+#include "util/fault_injection_test_env.h"
+#include "util/string_util.h"
+#include "util/sync_point.h"

 namespace rocksdb {

@@ -17,7 +23,9 @@ class DBWriteTest : public DBTestBase, public testing::WithParamInterface<int> {
 public:
  DBWriteTest() : DBTestBase("/db_write_test") {}

-  void Open() { DBTestBase::Reopen(GetOptions(GetParam())); }
+  Options GetOptions() { return DBTestBase::GetOptions(GetParam()); }
+
+  void Open() { DBTestBase::Reopen(GetOptions()); }
 };

 // It is invalid to do sync write while disabling WAL.
@@ -31,50 +39,45 @@ TEST_P(DBWriteTest, SyncAndDisableWAL) {
  ASSERT_TRUE(dbfull()->Write(write_options, &batch).IsInvalidArgument());
 }

-// Sequence number should be return through input write batch.
-TEST_P(DBWriteTest, ReturnSeuqneceNumber) {
-  Random rnd(4422);
-  Open();
-  for (int i = 0; i < 100; i++) {
-    WriteBatch batch;
-    batch.Put("key" + ToString(i), test::RandomHumanReadableString(&rnd, 10));
-    ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
-    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(),
-              WriteBatchInternal::Sequence(&batch));
-  }
-}
-
-TEST_P(DBWriteTest, ReturnSeuqneceNumberMultiThreaded) {
-  constexpr size_t kThreads = 16;
-  constexpr size_t kNumKeys = 1000;
-  Open();
-  ASSERT_EQ(0, dbfull()->GetLatestSequenceNumber());
-  // Check each sequence is used once and only once.
-  std::vector<std::atomic_flag> flags(kNumKeys * kThreads + 1);
-  for (size_t i = 0; i < flags.size(); i++) {
-    flags[i].clear();
-  }
-  auto writer = [&](size_t id) {
-    Random rnd(4422 + static_cast<uint32_t>(id));
-    for (size_t k = 0; k < kNumKeys; k++) {
-      WriteBatch batch;
-      batch.Put("key" + ToString(id) + "-" + ToString(k),
-                test::RandomHumanReadableString(&rnd, 10));
-      ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
-      SequenceNumber sequence = WriteBatchInternal::Sequence(&batch);
-      ASSERT_GT(sequence, 0);
-      ASSERT_LE(sequence, kNumKeys * kThreads);
-      // The sequence isn't consumed by someone else.
-      ASSERT_FALSE(flags[sequence].test_and_set());
-    }
-  };
+TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
+  constexpr int kNumThreads = 5;
+  std::unique_ptr<FaultInjectionTestEnv> mock_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetOptions();
+  options.env = mock_env.get();
+  Reopen(options);
+  std::atomic<int> ready_count{0};
+  std::atomic<int> leader_count{0};
  std::vector<port::Thread> threads;
-  for (size_t i = 0; i < kThreads; i++) {
-    threads.emplace_back(writer, i);
+  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(
+      "WriteThread::JoinBatchGroup:Wait", [&](void* arg) {
+        ready_count++;
+        auto* w = reinterpret_cast<WriteThread::Writer*>(arg);
+        if (w->state == WriteThread::STATE_GROUP_LEADER) {
+          leader_count++;
+          while (ready_count < kNumThreads) {
+            // busy waiting
+          }
+        }
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  for (int i = 0; i < kNumThreads; i++) {
+    threads.push_back(port::Thread(
+        [&](int index) {
+          // All threads should fail.
+          ASSERT_FALSE(Put("key" + ToString(index), "value").ok());
+        },
+        i));
  }
-  for (size_t i = 0; i < kThreads; i++) {
+  for (int i = 0; i < kNumThreads; i++) {
    threads[i].join();
  }
+  ASSERT_EQ(1, leader_count);
+  // Close before mock_env destruct.
+  Close();
 }

 INSTANTIATE_TEST_CASE_P(DBWriteTestInstance, DBWriteTest,
@@ -125,6 +125,26 @@ int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
  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:
@@ -49,7 +49,12 @@ enum ValueType : unsigned char {
  kTypeRangeDeletion = 0xF,               // meta block
  kTypeColumnFamilyBlobIndex = 0x10,      // Blob DB only
  kTypeBlobIndex = 0x11,                  // Blob DB only
-  kMaxValue = 0x7F                        // Not used for storing records.
+  // When the prepared record is also persisted in db, we use a different
+  // record. This is to ensure that the WAL that is generated by a WritePolicy
+  // is not mistakenly read by another, which would result into data
+  // inconsistency.
+  kTypeBeginPersistedPrepareXID = 0x12,  // WAL only.
+  kMaxValue = 0x7F                       // Not used for storing records.
 };

 // Defined in dbformat.cc
@@ -82,7 +87,7 @@ struct ParsedInternalKey {

  ParsedInternalKey()
      : sequence(kMaxSequenceNumber)  // Make code analyzer happy
-  {}  // Intentionally left uninitialized (for speed)
+        {} // Intentionally left uninitialized (for speed)
  ParsedInternalKey(const Slice& u, const SequenceNumber& seq, ValueType t)
      : user_key(u), sequence(seq), type(t) { }
  std::string DebugString(bool hex = false) const;
@@ -157,6 +162,8 @@ class InternalKeyComparator

  virtual const char* Name() const override;
  virtual int Compare(const Slice& a, const Slice& b) const override;
+  // Same as Compare except that it excludes the value type from comparison
+  virtual int CompareKeySeq(const Slice& a, const Slice& b) const;
  virtual void FindShortestSeparator(std::string* start,
                                     const Slice& limit) const override;
  virtual void FindShortSuccessor(std::string* key) const override;
@@ -558,8 +558,10 @@ TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
 }

 TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
+  int kNumLevels = 7;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
+  options.num_levels = kNumLevels;
  Reopen(options);

  std::map<std::string, std::string> true_data;
@@ -567,43 +569,65 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
  // prevent range deletions from being dropped due to becoming obsolete.
  const Snapshot* snapshot = db_->GetSnapshot();

-  // range del [0, 50) in L0 file, [50, 100) in memtable
-  for (int i = 0; i < 2; i++) {
-    if (i == 1) {
+  // range del [0, 50) in L6 file, [50, 100) in L0 file, [100, 150) in memtable
+  for (int i = 0; i < 3; i++) {
+    if (i != 0) {
      db_->Flush(FlushOptions());
+      if (i == 1) {
+        MoveFilesToLevel(kNumLevels - 1);
+      }
    }
    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                               Key(50 * i), Key(50 * (i + 1))));
  }
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 1));

-  // overlaps with L0 file but not memtable, so flush is skipped
+  // overlaps with L0 file but not memtable, so flush is skipped and file is
+  // ingested into L0
  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {60, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
+      file_id++, &true_data));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
+
+  // overlaps with L6 file but not memtable or L0 file, so flush is skipped and
+  // file is ingested into L5
  ASSERT_OK(GenerateAndAddExternalFile(
      options, {10, 40}, {ValueType::kTypeValue, ValueType::kTypeValue},
      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 - 1));

-  // overlaps with memtable, so flush is triggered (thus file count increases by
-  // two at this step).
+  // 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.
  ASSERT_OK(GenerateAndAddExternalFile(
-      options, {50, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
+      options, {100, 140}, {ValueType::kTypeValue, ValueType::kTypeValue},
      file_id++, &true_data));
  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
  ASSERT_EQ(4, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));

-  // snapshot unneeded now that both range deletions are persisted
+  // snapshot unneeded now that all range deletions are persisted
  db_->ReleaseSnapshot(snapshot);

  // overlaps with nothing, so places at bottom level and skips incrementing
  // seqnum.
  ASSERT_OK(GenerateAndAddExternalFile(
-      options, {101, 125}, {ValueType::kTypeValue, ValueType::kTypeValue},
+      options, {151, 175}, {ValueType::kTypeValue, ValueType::kTypeValue},
      file_id++, &true_data));
  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
  ASSERT_EQ(4, NumTableFilesAtLevel(0));
-  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
+  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(2, NumTableFilesAtLevel(options.num_levels - 1));
 }

 #endif  // ROCKSDB_LITE
@@ -132,11 +132,15 @@ Status ExternalSstFileIngestionJob::Prepare(
  return status;
 }

-Status ExternalSstFileIngestionJob::NeedsFlush(bool* flush_needed) {
-  SuperVersion* super_version = cfd_->GetSuperVersion();
+Status ExternalSstFileIngestionJob::NeedsFlush(bool* flush_needed,
+                                               SuperVersion* super_version) {
+  autovector<Range> ranges;
+  for (const IngestedFileInfo& file_to_ingest : files_to_ingest_) {
+    ranges.emplace_back(file_to_ingest.smallest_user_key,
+                        file_to_ingest.largest_user_key);
+  }
  Status status =
-      IngestedFilesOverlapWithMemtables(super_version, flush_needed);
-
+      cfd_->RangesOverlapWithMemtables(ranges, super_version, flush_needed);
  if (status.ok() && *flush_needed &&
      !ingestion_options_.allow_blocking_flush) {
    status = Status::InvalidArgument("External file requires flush");
@@ -148,11 +152,12 @@ Status ExternalSstFileIngestionJob::NeedsFlush(bool* flush_needed) {
 // nonmem_write_thread_
 Status ExternalSstFileIngestionJob::Run() {
  Status status;
+  SuperVersion* super_version = cfd_->GetSuperVersion();
 #ifndef NDEBUG
  // We should never run the job with a memtable that is overlapping
  // with the files we are ingesting
  bool need_flush = false;
-  status = NeedsFlush(&need_flush);
+  status = NeedsFlush(&need_flush, super_version);
  assert(status.ok() && need_flush == false);
 #endif

@@ -164,10 +169,9 @@ Status ExternalSstFileIngestionJob::Run() {
    // if the dont overlap with any ranges since we have snapshots
    force_global_seqno = true;
  }
-  // It is safe to use this instead of LastToBeWrittenSequence since we are
+  // It is safe to use this instead of LastAllocatedSequence since we are
  // the only active writer, and hence they are equal
  const SequenceNumber last_seqno = versions_->LastSequence();
-  SuperVersion* super_version = cfd_->GetSuperVersion();
  edit_.SetColumnFamily(cfd_->GetID());
  // The levels that the files will be ingested into

@@ -199,7 +203,8 @@ Status ExternalSstFileIngestionJob::Run() {
  }

  if (consumed_seqno) {
-    versions_->SetLastToBeWrittenSequence(last_seqno + 1);
+    versions_->SetLastAllocatedSequence(last_seqno + 1);
+    versions_->SetLastPublishedSequence(last_seqno + 1);
    versions_->SetLastSequence(last_seqno + 1);
  }

@@ -374,46 +379,6 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
  return status;
 }

-Status ExternalSstFileIngestionJob::IngestedFilesOverlapWithMemtables(
-    SuperVersion* sv, bool* overlap) {
-  // Create an InternalIterator over all memtables
-  Arena arena;
-  ReadOptions ro;
-  ro.total_order_seek = true;
-  MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(), &arena);
-  merge_iter_builder.AddIterator(sv->mem->NewIterator(ro, &arena));
-  sv->imm->AddIterators(ro, &merge_iter_builder);
-  ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
-
-  std::vector<InternalIterator*> memtable_range_del_iters;
-  auto* active_range_del_iter = sv->mem->NewRangeTombstoneIterator(ro);
-  if (active_range_del_iter != nullptr) {
-    memtable_range_del_iters.push_back(active_range_del_iter);
-  }
-  sv->imm->AddRangeTombstoneIterators(ro, &memtable_range_del_iters);
-  std::unique_ptr<InternalIterator> memtable_range_del_iter(NewMergingIterator(
-      &cfd_->internal_comparator(),
-      memtable_range_del_iters.empty() ? nullptr : &memtable_range_del_iters[0],
-      static_cast<int>(memtable_range_del_iters.size())));
-
-  Status status;
-  *overlap = false;
-  for (IngestedFileInfo& f : files_to_ingest_) {
-    status =
-        IngestedFileOverlapWithIteratorRange(&f, memtable_iter.get(), overlap);
-    if (!status.ok() || *overlap == true) {
-      break;
-    }
-    status = IngestedFileOverlapWithRangeDeletions(
-        &f, memtable_range_del_iter.get(), overlap);
-    if (!status.ok() || *overlap == true) {
-      break;
-    }
-  }
-
-  return status;
-}
-
 Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
    SuperVersion* sv, bool force_global_seqno, CompactionStyle compaction_style,
    IngestedFileInfo* file_to_ingest, SequenceNumber* assigned_seqno) {
@@ -464,7 +429,10 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
                            return f1->largest_seqno < f2->largest_seqno;
                          }))
                ->largest_seqno;
-        if (level_largest_seqno != 0) {
+        // should only assign seqno to current level's largest seqno when
+        // the file fits
+        if (level_largest_seqno != 0 &&
+            IngestedFileFitInLevel(file_to_ingest, lvl)) {
          *assigned_seqno = level_largest_seqno;
        } else {
          continue;
@@ -541,6 +509,9 @@ Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
  std::string seqno_val;
  PutFixed64(&seqno_val, seqno);
  status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
+  if (status.ok()) {
+    status = rwfile->Fsync();
+  }
  if (status.ok()) {
    file_to_ingest->assigned_seqno = seqno;
  }
@@ -575,34 +546,6 @@ Status ExternalSstFileIngestionJob::IngestedFileOverlapWithIteratorRange(
  return iter->status();
 }

-Status ExternalSstFileIngestionJob::IngestedFileOverlapWithRangeDeletions(
-    const IngestedFileInfo* file_to_ingest, InternalIterator* range_del_iter,
-    bool* overlap) {
-  auto* vstorage = cfd_->current()->storage_info();
-  auto* ucmp = vstorage->InternalComparator()->user_comparator();
-
-  *overlap = false;
-  if (range_del_iter != nullptr) {
-    for (range_del_iter->SeekToFirst(); range_del_iter->Valid();
-         range_del_iter->Next()) {
-      ParsedInternalKey parsed_key;
-      if (!ParseInternalKey(range_del_iter->key(), &parsed_key)) {
-        return Status::Corruption("corrupted range deletion key: " +
-                                  range_del_iter->key().ToString());
-      }
-      RangeTombstone range_del(parsed_key, range_del_iter->value());
-      if (ucmp->Compare(range_del.start_key_,
-                        file_to_ingest->largest_user_key) <= 0 &&
-          ucmp->Compare(file_to_ingest->smallest_user_key,
-                        range_del.end_key_) <= 0) {
-        *overlap = true;
-        break;
-      }
-    }
-  }
-  return Status::OK();
-}
-
 bool ExternalSstFileIngestionJob::IngestedFileFitInLevel(
    const IngestedFileInfo* file_to_ingest, int level) {
  if (level == 0) {
@@ -639,23 +582,26 @@ Status ExternalSstFileIngestionJob::IngestedFileOverlapWithLevel(
  ro.total_order_seek = true;
  MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(),
                                          &arena);
+  // Files are opened lazily when the iterator needs them, thus range deletions
+  // are also added lazily to the aggregator. We need to check for range
+  // deletion overlap only in the case where there's no point-key overlap. Then,
+  // we've already opened the file with range containing the ingested file's
+  // begin key, and iterated through all files until the one containing the
+  // ingested file's end key. So any files maybe containing range deletions
+  // overlapping the ingested file must have been opened and had their range
+  // deletions added to the aggregator.
+  RangeDelAggregator range_del_agg(cfd_->internal_comparator(),
+                                   {} /* snapshots */,
+                                   false /* collapse_deletions */);
  sv->current->AddIteratorsForLevel(ro, env_options_, &merge_iter_builder, lvl,
-                                    nullptr /* range_del_agg */);
+                                    &range_del_agg);
  ScopedArenaIterator level_iter(merge_iter_builder.Finish());
-
-  std::vector<InternalIterator*> level_range_del_iters;
-  sv->current->AddRangeDelIteratorsForLevel(ro, env_options_, lvl,
-                                            &level_range_del_iters);
-  std::unique_ptr<InternalIterator> level_range_del_iter(NewMergingIterator(
-      &cfd_->internal_comparator(),
-      level_range_del_iters.empty() ? nullptr : &level_range_del_iters[0],
-      static_cast<int>(level_range_del_iters.size())));
-
  Status status = IngestedFileOverlapWithIteratorRange(
      file_to_ingest, level_iter.get(), overlap_with_level);
-  if (status.ok() && *overlap_with_level == false) {
-    status = IngestedFileOverlapWithRangeDeletions(
-        file_to_ingest, level_range_del_iter.get(), overlap_with_level);
+  if (status.ok() && *overlap_with_level == false &&
+      range_del_agg.IsRangeOverlapped(file_to_ingest->smallest_user_key,
+                                      file_to_ingest->largest_user_key)) {
+    *overlap_with_level = true;
  }
  return status;
 }
@@ -84,8 +84,12 @@ class ExternalSstFileIngestionJob {
  // 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
  // key range in the memtable.
-  // REQUIRES: Mutex held
-  Status NeedsFlush(bool* flush_needed);
+  //
+  // @param super_version A referenced SuperVersion that will be held for the
+  //    duration of this function.
+  //
+  // Thread-safe
+  Status NeedsFlush(bool* flush_needed, SuperVersion* super_version);

  // Will execute the ingestion job and prepare edit() to be applied.
  // REQUIRES: Mutex held
@@ -110,10 +114,6 @@ class ExternalSstFileIngestionJob {
  Status GetIngestedFileInfo(const std::string& external_file,
                             IngestedFileInfo* file_to_ingest);

-  // Check if the files we are ingesting overlap with any memtable.
-  // REQUIRES: Mutex held
-  Status IngestedFilesOverlapWithMemtables(SuperVersion* sv, bool* overlap);
-
  // Assign `file_to_ingest` the appropriate sequence number and  the lowest
  // possible level that it can be ingested to according to compaction_style.
  // REQUIRES: Mutex held
@@ -139,13 +139,6 @@ class ExternalSstFileIngestionJob {
      const IngestedFileInfo* file_to_ingest, InternalIterator* iter,
      bool* overlap);

-  // Check if `file_to_ingest` key range overlaps with any range deletions
-  // specified by `iter`.
-  // REQUIRES: Mutex held
-  Status IngestedFileOverlapWithRangeDeletions(
-      const IngestedFileInfo* file_to_ingest, InternalIterator* range_del_iter,
-      bool* overlap);
-
  // Check if `file_to_ingest` key range overlap with level
  // REQUIRES: Mutex held
  Status IngestedFileOverlapWithLevel(SuperVersion* sv,
@@ -388,9 +388,10 @@ class SstFileWriterCollector : public TablePropertiesCollector {
  const char* Name() const override { return name_.c_str(); }

  Status Finish(UserCollectedProperties* properties) override {
+    std::string count = std::to_string(count_);
    *properties = UserCollectedProperties{
        {prefix_ + "_SstFileWriterCollector", "YES"},
-        {prefix_ + "_Count", std::to_string(count_)},
+        {prefix_ + "_Count", count},
    };
    return Status::OK();
  }
@@ -1940,6 +1941,55 @@ TEST_F(ExternalSSTFileTest, IngestBehind) {
  size_t kcnt = 0;
  VerifyDBFromMap(true_data, &kcnt, false);
 }
+
+TEST_F(ExternalSSTFileTest, SkipBloomFilter) {
+  Options options = CurrentOptions();
+
+  BlockBasedTableOptions table_options;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10));
+  table_options.cache_index_and_filter_blocks = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+
+
+  // Create external SST file and include bloom filters
+  options.statistics = rocksdb::CreateDBStatistics();
+  DestroyAndReopen(options);
+  {
+    std::string file_path = sst_files_dir_ + "sst_with_bloom.sst";
+    SstFileWriter sst_file_writer(EnvOptions(), options);
+    ASSERT_OK(sst_file_writer.Open(file_path));
+    ASSERT_OK(sst_file_writer.Put("Key1", "Value1"));
+    ASSERT_OK(sst_file_writer.Finish());
+
+    ASSERT_OK(
+        db_->IngestExternalFile({file_path}, IngestExternalFileOptions()));
+
+    ASSERT_EQ(Get("Key1"), "Value1");
+    ASSERT_GE(
+        options.statistics->getTickerCount(Tickers::BLOCK_CACHE_FILTER_ADD), 1);
+  }
+
+  // Create external SST file but skip bloom filters
+  options.statistics = rocksdb::CreateDBStatistics();
+  DestroyAndReopen(options);
+  {
+    std::string file_path = sst_files_dir_ + "sst_with_no_bloom.sst";
+    SstFileWriter sst_file_writer(EnvOptions(), options, nullptr, true,
+                                  Env::IOPriority::IO_TOTAL,
+                                  true /* skip_filters */);
+    ASSERT_OK(sst_file_writer.Open(file_path));
+    ASSERT_OK(sst_file_writer.Put("Key1", "Value1"));
+    ASSERT_OK(sst_file_writer.Finish());
+
+    ASSERT_OK(
+        db_->IngestExternalFile({file_path}, IngestExternalFileOptions()));
+
+    ASSERT_EQ(Get("Key1"), "Value1");
+    ASSERT_EQ(
+        options.statistics->getTickerCount(Tickers::BLOCK_CACHE_FILTER_ADD), 0);
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -228,16 +228,9 @@ class FaultInjectionTest : public testing::Test,
    return Status::OK();
  }

-#ifdef ROCKSDB_UBSAN_RUN
-#if defined(__clang__)
-__attribute__((__no_sanitize__("shift"), no_sanitize("signed-integer-overflow")))
-#elif defined(__GNUC__)
-__attribute__((__no_sanitize_undefined__))
-#endif
-#endif
  // Return the ith key
  Slice Key(int i, std::string* storage) const {
-    int num = i;
+    unsigned long long num = i;
    if (!sequential_order_) {
      // random transfer
      const int m = 0x5bd1e995;
@@ -245,7 +238,7 @@ __attribute__((__no_sanitize_undefined__))
      num ^= num << 24;
    }
    char buf[100];
-    snprintf(buf, sizeof(buf), "%016d", num);
+    snprintf(buf, sizeof(buf), "%016d", static_cast<int>(num));
    storage->assign(buf, strlen(buf));
    return Slice(*storage);
  }
@@ -54,6 +54,32 @@

 namespace rocksdb {

+const char* GetFlushReasonString (FlushReason flush_reason) {
+  switch (flush_reason) {
+    case FlushReason::kUnknown:
+      return "Unknown";
+    case FlushReason::kGetLiveFiles:
+      return "Get Live Files";
+    case FlushReason::kShutDown:
+      return "Shut down";
+    case FlushReason::kExternalFileIngestion:
+      return "External File Ingestion";
+    case FlushReason::kManualCompaction:
+      return "Manual Compaction";
+    case FlushReason::kWriteBufferManager:
+      return "Write Buffer Manager";
+    case FlushReason::kWriteBufferFull:
+      return "Write Buffer Full";
+    case FlushReason::kTest:
+      return "Test";
+    case FlushReason::kSuperVersionChange:
+      return "SuperVersion Change";
+    default:
+      return "Invalid";
+  }
+}
+
+
 FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                   const ImmutableDBOptions& db_options,
                   const MutableCFOptions& mutable_cf_options,
@@ -209,6 +235,8 @@ Status FlushJob::Run(FileMetaData* file_meta) {
  auto stream = event_logger_->LogToBuffer(log_buffer_);
  stream << "job" << job_context_->job_id << "event"
         << "flush_finished";
+  stream << "output_compression"
+         << CompressionTypeToString(output_compression_);
  stream << "lsm_state";
  stream.StartArray();
  auto vstorage = cfd_->current()->storage_info();
@@ -246,6 +274,7 @@ Status FlushJob::WriteLevel0Table() {
  const uint64_t start_micros = db_options_.env->NowMicros();
  Status s;
  {
+    auto write_hint = cfd_->CalculateSSTWriteHint(0);
    db_mutex_->Unlock();
    if (log_buffer_) {
      log_buffer_->FlushBufferToLog();
@@ -275,12 +304,13 @@ Status FlushJob::WriteLevel0Table() {
      total_memory_usage += m->ApproximateMemoryUsage();
    }

-    event_logger_->Log() << "job" << job_context_->job_id << "event"
-                         << "flush_started"
-                         << "num_memtables" << mems_.size() << "num_entries"
-                         << total_num_entries << "num_deletes"
-                         << total_num_deletes << "memory_usage"
-                         << total_memory_usage;
+    event_logger_->Log()
+        << "job" << job_context_->job_id << "event"
+        << "flush_started"
+        << "num_memtables" << mems_.size() << "num_entries" << total_num_entries
+        << "num_deletes" << total_num_deletes << "memory_usage"
+        << total_memory_usage << "flush_reason"
+        << GetFlushReasonString(cfd_->GetFlushReason());

    {
      ScopedArenaIterator iter(
@@ -298,7 +328,15 @@ Status FlushJob::WriteLevel0Table() {
      TEST_SYNC_POINT_CALLBACK("FlushJob::WriteLevel0Table:output_compression",
                               &output_compression_);
      int64_t _current_time = 0;
-      db_options_.env->GetCurrentTime(&_current_time);  // ignore error
+      auto status = db_options_.env->GetCurrentTime(&_current_time);
+      // Safe to proceed even if GetCurrentTime fails. So, log and proceed.
+      if (!status.ok()) {
+        ROCKS_LOG_WARN(
+            db_options_.info_log,
+            "Failed to get current time to populate creation_time property. "
+            "Status: %s",
+            status.ToString().c_str());
+      }
      const uint64_t current_time = static_cast<uint64_t>(_current_time);

      uint64_t oldest_key_time =
@@ -315,7 +353,7 @@ Status FlushJob::WriteLevel0Table() {
          mutable_cf_options_.paranoid_file_checks, cfd_->internal_stats(),
          TableFileCreationReason::kFlush, event_logger_, job_context_->job_id,
          Env::IO_HIGH, &table_properties_, 0 /* level */, current_time,
-          oldest_key_time);
+          oldest_key_time, write_hint);
      LogFlush(db_options_.info_log);
    }
    ROCKS_LOG_INFO(db_options_.info_log,
@@ -353,6 +391,7 @@ Status FlushJob::WriteLevel0Table() {
  InternalStats::CompactionStats stats(1);
  stats.micros = db_options_.env->NowMicros() - start_micros;
  stats.bytes_written = meta_.fd.GetFileSize();
+  MeasureTime(stats_, FLUSH_TIME, stats.micros);
  cfd_->internal_stats()->AddCompactionStats(0 /* level */, stats);
  cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
                                     meta_.fd.GetFileSize());
@@ -40,6 +40,7 @@ class FlushJobTest : public testing::Test {
    EXPECT_OK(env_->CreateDirIfMissing(dbname_));
    db_options_.db_paths.emplace_back(dbname_,
                                      std::numeric_limits<uint64_t>::max());
+    db_options_.statistics = rocksdb::CreateDBStatistics();
    // TODO(icanadi) Remove this once we mock out VersionSet
    NewDB();
    std::vector<ColumnFamilyDescriptor> column_families;
@@ -138,16 +139,22 @@ TEST_F(FlushJobTest, NonEmpty) {

  EventLogger event_logger(db_options_.info_log.get());
  SnapshotChecker* snapshot_checker = nullptr;  // not relavant
-  FlushJob flush_job(
-      dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
-      *cfd->GetLatestMutableCFOptions(), env_options_, versions_.get(), &mutex_,
-      &shutting_down_, {}, kMaxSequenceNumber, snapshot_checker, &job_context,
-      nullptr, nullptr, nullptr, kNoCompression, nullptr, &event_logger, true);
+  FlushJob flush_job(dbname_, versions_->GetColumnFamilySet()->GetDefault(),
+                     db_options_, *cfd->GetLatestMutableCFOptions(),
+                     env_options_, versions_.get(), &mutex_, &shutting_down_,
+                     {}, kMaxSequenceNumber, snapshot_checker, &job_context,
+                     nullptr, nullptr, nullptr, kNoCompression,
+                     db_options_.statistics.get(), &event_logger, true);
+
+  HistogramData hist;
  FileMetaData fd;
  mutex_.Lock();
  flush_job.PickMemTable();
  ASSERT_OK(flush_job.Run(&fd));
  mutex_.Unlock();
+  db_options_.statistics->histogramData(FLUSH_TIME, &hist);
+  ASSERT_GT(hist.average, 0.0);
+
  ASSERT_EQ(ToString(0), fd.smallest.user_key().ToString());
  ASSERT_EQ("9999a",
            fd.largest.user_key().ToString());  // range tombstone end key
@@ -210,12 +217,15 @@ TEST_F(FlushJobTest, Snapshots) {
                     env_options_, versions_.get(), &mutex_, &shutting_down_,
                     snapshots, kMaxSequenceNumber, snapshot_checker,
                     &job_context, nullptr, nullptr, nullptr, kNoCompression,
-                     nullptr, &event_logger, true);
+                     db_options_.statistics.get(), &event_logger, true);
  mutex_.Lock();
  flush_job.PickMemTable();
  ASSERT_OK(flush_job.Run());
  mutex_.Unlock();
  mock_table_factory_->AssertSingleFile(inserted_keys);
+  HistogramData hist;
+  db_options_.statistics->histogramData(FLUSH_TIME, &hist);
+  ASSERT_GT(hist.average, 0.0);
  job_context.Clean();
 }

@@ -25,15 +25,15 @@
 namespace rocksdb {

 // Usage:
-//     LevelIterator iter;
+//     ForwardLevelIterator iter;
 //     iter.SetFileIndex(file_index);
 //     iter.Seek(target);
 //     iter.Next()
-class LevelIterator : public InternalIterator {
+class ForwardLevelIterator : public InternalIterator {
 public:
-  LevelIterator(const ColumnFamilyData* const cfd,
-                const ReadOptions& read_options,
-                const std::vector<FileMetaData*>& files)
+  ForwardLevelIterator(const ColumnFamilyData* const cfd,
+                       const ReadOptions& read_options,
+                       const std::vector<FileMetaData*>& files)
      : cfd_(cfd),
        read_options_(read_options),
        files_(files),
@@ -42,7 +42,7 @@ class LevelIterator : public InternalIterator {
        file_iter_(nullptr),
        pinned_iters_mgr_(nullptr) {}

-  ~LevelIterator() {
+  ~ForwardLevelIterator() {
    // Reset current pointer
    if (pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled()) {
      pinned_iters_mgr_->PinIterator(file_iter_);
@@ -84,11 +84,11 @@ class LevelIterator : public InternalIterator {
    }
  }
  void SeekToLast() override {
-    status_ = Status::NotSupported("LevelIterator::SeekToLast()");
+    status_ = Status::NotSupported("ForwardLevelIterator::SeekToLast()");
    valid_ = false;
  }
  void Prev() override {
-    status_ = Status::NotSupported("LevelIterator::Prev()");
+    status_ = Status::NotSupported("ForwardLevelIterator::Prev()");
    valid_ = false;
  }
  bool Valid() const override {
@@ -105,7 +105,7 @@ class LevelIterator : public InternalIterator {
    valid_ = file_iter_->Valid();
  }
  void SeekForPrev(const Slice& internal_key) override {
-    status_ = Status::NotSupported("LevelIterator::SeekForPrev()");
+    status_ = Status::NotSupported("ForwardLevelIterator::SeekForPrev()");
    valid_ = false;
  }
  void Next() override {
@@ -196,38 +196,60 @@ ForwardIterator::~ForwardIterator() {
  Cleanup(true);
 }

-namespace {
-// Used in PinnedIteratorsManager to release pinned SuperVersion
-static void ReleaseSuperVersionFunc(void* sv) {
-  delete reinterpret_cast<SuperVersion*>(sv);
-}
-}  // namespace
-
-void ForwardIterator::SVCleanup() {
-  if (sv_ != nullptr && sv_->Unref()) {
+void ForwardIterator::SVCleanup(DBImpl* db, SuperVersion* sv,
+                                bool background_purge_on_iterator_cleanup) {
+  if (sv->Unref()) {
    // Job id == 0 means that this is not our background process, but rather
    // user thread
    JobContext job_context(0);
-    db_->mutex_.Lock();
-    sv_->Cleanup();
-    db_->FindObsoleteFiles(&job_context, false, true);
-    if (read_options_.background_purge_on_iterator_cleanup) {
-      db_->ScheduleBgLogWriterClose(&job_context);
-    }
-    db_->mutex_.Unlock();
-    if (pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled()) {
-      pinned_iters_mgr_->PinPtr(sv_, &ReleaseSuperVersionFunc);
-    } else {
-      delete sv_;
+    db->mutex_.Lock();
+    sv->Cleanup();
+    db->FindObsoleteFiles(&job_context, false, true);
+    if (background_purge_on_iterator_cleanup) {
+      db->ScheduleBgLogWriterClose(&job_context);
    }
+    db->mutex_.Unlock();
+    delete sv;
    if (job_context.HaveSomethingToDelete()) {
-      db_->PurgeObsoleteFiles(
-          job_context, read_options_.background_purge_on_iterator_cleanup);
+      db->PurgeObsoleteFiles(job_context, background_purge_on_iterator_cleanup);
    }
    job_context.Clean();
  }
 }

+namespace {
+struct SVCleanupParams {
+  DBImpl* db;
+  SuperVersion* sv;
+  bool background_purge_on_iterator_cleanup;
+};
+}
+
+// Used in PinnedIteratorsManager to release pinned SuperVersion
+void ForwardIterator::DeferredSVCleanup(void* arg) {
+  auto d = reinterpret_cast<SVCleanupParams*>(arg);
+  ForwardIterator::SVCleanup(
+    d->db, d->sv, d->background_purge_on_iterator_cleanup);
+  delete d;
+}
+
+void ForwardIterator::SVCleanup() {
+  if (sv_ == nullptr) {
+    return;
+  }
+  if (pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled()) {
+    // pinned_iters_mgr_ tells us to make sure that all visited key-value slices
+    // are alive until pinned_iters_mgr_->ReleasePinnedData() is called.
+    // The slices may point into some memtables owned by sv_, so we need to keep
+    // sv_ referenced until pinned_iters_mgr_ unpins everything.
+    auto p = new SVCleanupParams{
+      db_, sv_, read_options_.background_purge_on_iterator_cleanup};
+    pinned_iters_mgr_->PinPtr(p, &ForwardIterator::DeferredSVCleanup);
+  } else {
+    SVCleanup(db_, sv_, read_options_.background_purge_on_iterator_cleanup);
+  }
+}
+
 void ForwardIterator::Cleanup(bool release_sv) {
  if (mutable_iter_ != nullptr) {
    DeleteIterator(mutable_iter_, true /* is_arena */);
@@ -361,14 +383,13 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,

      if (!l0_iters_[i]->status().ok()) {
        immutable_status_ = l0_iters_[i]->status();
-      } else if (l0_iters_[i]->Valid()) {
-        if (!IsOverUpperBound(l0_iters_[i]->key())) {
-          immutable_min_heap_.push(l0_iters_[i]);
-        } else {
-          has_iter_trimmed_for_upper_bound_ = true;
-          DeleteIterator(l0_iters_[i]);
-          l0_iters_[i] = nullptr;
-        }
+      } else if (l0_iters_[i]->Valid() &&
+                 !IsOverUpperBound(l0_iters_[i]->key())) {
+        immutable_min_heap_.push(l0_iters_[i]);
+      } else {
+        has_iter_trimmed_for_upper_bound_ = true;
+        DeleteIterator(l0_iters_[i]);
+        l0_iters_[i] = nullptr;
      }
    }

@@ -395,15 +416,14 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,

        if (!level_iters_[level - 1]->status().ok()) {
          immutable_status_ = level_iters_[level - 1]->status();
-        } else if (level_iters_[level - 1]->Valid()) {
-          if (!IsOverUpperBound(level_iters_[level - 1]->key())) {
-            immutable_min_heap_.push(level_iters_[level - 1]);
-          } else {
-            // Nothing in this level is interesting. Remove.
-            has_iter_trimmed_for_upper_bound_ = true;
-            DeleteIterator(level_iters_[level - 1]);
-            level_iters_[level - 1] = nullptr;
-          }
+        } else if (level_iters_[level - 1]->Valid() &&
+                   !IsOverUpperBound(level_iters_[level - 1]->key())) {
+          immutable_min_heap_.push(level_iters_[level - 1]);
+        } else {
+          // Nothing in this level is interesting. Remove.
+          has_iter_trimmed_for_upper_bound_ = true;
+          DeleteIterator(level_iters_[level - 1]);
+          level_iters_[level - 1] = nullptr;
        }
      }
    }
@@ -541,7 +561,7 @@ void ForwardIterator::UpdateChildrenPinnedItersMgr() {
  }

  // Set PinnedIteratorsManager for L1+ levels iterators.
-  for (LevelIterator* child_iter : level_iters_) {
+  for (ForwardLevelIterator* child_iter : level_iters_) {
    if (child_iter) {
      child_iter->SetPinnedItersMgr(pinned_iters_mgr_);
    }
@@ -566,7 +586,7 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
    sv_ = cfd_->GetReferencedSuperVersion(&(db_->mutex_));
  }
  RangeDelAggregator range_del_agg(
-      InternalKeyComparator(cfd_->internal_comparator()), {} /* snapshots */);
+      cfd_->internal_comparator(), {} /* snapshots */);
  mutable_iter_ = sv_->mem->NewIterator(read_options_, &arena_);
  sv_->imm->AddIterators(read_options_, &imm_iters_, &arena_);
  if (!read_options_.ignore_range_deletions) {
@@ -621,7 +641,7 @@ void ForwardIterator::RenewIterators() {
  mutable_iter_ = svnew->mem->NewIterator(read_options_, &arena_);
  svnew->imm->AddIterators(read_options_, &imm_iters_, &arena_);
  RangeDelAggregator range_del_agg(
-      InternalKeyComparator(cfd_->internal_comparator()), {} /* snapshots */);
+      cfd_->internal_comparator(), {} /* snapshots */);
  if (!read_options_.ignore_range_deletions) {
    std::unique_ptr<InternalIterator> range_del_iter(
        svnew->mem->NewRangeTombstoneIterator(read_options_));
@@ -703,7 +723,7 @@ void ForwardIterator::BuildLevelIterators(const VersionStorageInfo* vstorage) {
      }
    } else {
      level_iters_.push_back(
-          new LevelIterator(cfd_, read_options_, level_files));
+          new ForwardLevelIterator(cfd_, read_options_, level_files));
    }
  }
 }
@@ -23,7 +23,7 @@ class DBImpl;
 class Env;
 struct SuperVersion;
 class ColumnFamilyData;
-class LevelIterator;
+class ForwardLevelIterator;
 class VersionStorageInfo;
 struct FileMetaData;

@@ -85,7 +85,14 @@ class ForwardIterator : public InternalIterator {

 private:
  void Cleanup(bool release_sv);
+  // Unreference and, if needed, clean up the current SuperVersion. This is
+  // either done immediately or deferred until this iterator is unpinned by
+  // PinnedIteratorsManager.
  void SVCleanup();
+  static void SVCleanup(
+    DBImpl* db, SuperVersion* sv, bool background_purge_on_iterator_cleanup);
+  static void DeferredSVCleanup(void* arg);
+
  void RebuildIterators(bool refresh_sv);
  void RenewIterators();
  void BuildLevelIterators(const VersionStorageInfo* vstorage);
@@ -119,7 +126,7 @@ class ForwardIterator : public InternalIterator {
  InternalIterator* mutable_iter_;
  std::vector<InternalIterator*> imm_iters_;
  std::vector<InternalIterator*> l0_iters_;
-  std::vector<LevelIterator*> level_iters_;
+  std::vector<ForwardLevelIterator*> level_iters_;
  InternalIterator* current_;
  bool valid_;

@@ -17,7 +17,6 @@ int main() {
 // Block forward_iterator_bench under MAC and Windows
 int main() { return 0; }
 #else
-#include <gflags/gflags.h>
 #include <semaphore.h>
 #include <atomic>
 #include <bitset>
@@ -30,11 +29,12 @@ int main() { return 0; }
 #include <random>
 #include <thread>

+#include "port/port.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/db.h"
 #include "rocksdb/status.h"
 #include "rocksdb/table.h"
-#include "port/port.h"
+#include "util/gflags_compat.h"
 #include "util/testharness.h"

 const int MAX_SHARDS = 100000;
@@ -319,7 +319,7 @@ struct StatsThread {
 };

 int main(int argc, char** argv) {
-  GFLAGS::ParseCommandLineFlags(&argc, &argv, true);
+  GFLAGS_NAMESPACE::ParseCommandLineFlags(&argc, &argv, true);

  std::mt19937 rng{std::random_device()()};
  rocksdb::Status status;
@@ -81,8 +81,11 @@ struct SuperVersionContext {
 struct JobContext {
  inline bool HaveSomethingToDelete() const {
    return full_scan_candidate_files.size() || sst_delete_files.size() ||
-           log_delete_files.size() || manifest_delete_files.size() ||
-           memtables_to_free.size() > 0 || logs_to_free.size() > 0 ||
+           log_delete_files.size() || manifest_delete_files.size();
+  }
+
+  inline bool HaveSomethingToClean() const {
+    return memtables_to_free.size() > 0 || logs_to_free.size() > 0 ||
           superversion_context.HaveSomethingToDelete();
  }

@@ -528,7 +528,7 @@ TEST_F(EventListenerTest, CompactionReasonUniversal) {

  ASSERT_GT(listener->compaction_reasons_.size(), 0);
  for (auto compaction_reason : listener->compaction_reasons_) {
-    ASSERT_EQ(compaction_reason, CompactionReason::kUniversalSortedRunNum);
+    ASSERT_EQ(compaction_reason, CompactionReason::kUniversalSizeRatio);
  }

  options.level0_file_num_compaction_trigger = 8;
@@ -54,6 +54,7 @@ class Reader {
  // The Reader will start reading at the first record located at physical
  // position >= initial_offset within the file.
  Reader(std::shared_ptr<Logger> info_log,
+  // @lint-ignore TXT2 T25377293 Grandfathered in
 	 unique_ptr<SequentialFileReader>&& file,
         Reporter* reporter, bool checksum, uint64_t initial_offset,
         uint64_t log_num);
@@ -16,7 +16,12 @@
 namespace rocksdb {

 #ifdef ROCKSDB_JEMALLOC
+#ifdef __FreeBSD__
+#include <malloc_np.h>
+#define je_malloc_stats_print malloc_stats_print
+#else
 #include "jemalloc/jemalloc.h"
+#endif

 typedef struct {
  char* cur;
@@ -36,8 +41,8 @@ static void GetJemallocStatus(void* mstat_arg, const char* status) {
 }
 #endif  // ROCKSDB_JEMALLOC

-void DumpMallocStats(std::string* stats) {
 #ifdef ROCKSDB_JEMALLOC
+void DumpMallocStats(std::string* stats) {
  MallocStatus mstat;
  const unsigned int kMallocStatusLen = 1000000;
  std::unique_ptr<char[]> buf{new char[kMallocStatusLen + 1]};
@@ -45,8 +50,10 @@ void DumpMallocStats(std::string* stats) {
  mstat.end = buf.get() + kMallocStatusLen;
  je_malloc_stats_print(GetJemallocStatus, &mstat, "");
  stats->append(buf.get());
-#endif  // ROCKSDB_JEMALLOC
 }
+#else
+void DumpMallocStats(std::string*) {}
+#endif  // ROCKSDB_JEMALLOC

 }
 #endif  // !ROCKSDB_LITE
@@ -39,10 +39,10 @@

 namespace rocksdb {

-MemTableOptions::MemTableOptions(const ImmutableCFOptions& ioptions,
-                                 const MutableCFOptions& mutable_cf_options)
-    : write_buffer_size(mutable_cf_options.write_buffer_size),
-      arena_block_size(mutable_cf_options.arena_block_size),
+ImmutableMemTableOptions::ImmutableMemTableOptions(
+    const ImmutableCFOptions& ioptions,
+    const MutableCFOptions& mutable_cf_options)
+    : arena_block_size(mutable_cf_options.arena_block_size),
      memtable_prefix_bloom_bits(
          static_cast<uint32_t>(
              static_cast<double>(mutable_cf_options.write_buffer_size) *
@@ -83,6 +83,7 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
      data_size_(0),
      num_entries_(0),
      num_deletes_(0),
+      write_buffer_size_(mutable_cf_options.write_buffer_size),
      flush_in_progress_(false),
      flush_completed_(false),
      file_number_(0),
@@ -136,6 +137,7 @@ size_t MemTable::ApproximateMemoryUsage() {
 }

 bool MemTable::ShouldFlushNow() const {
+  size_t write_buffer_size = write_buffer_size_.load(std::memory_order_relaxed);
  // In a lot of times, we cannot allocate arena blocks that exactly matches the
  // buffer size. Thus we have to decide if we should over-allocate or
  // under-allocate.
@@ -153,16 +155,14 @@ bool MemTable::ShouldFlushNow() const {
  // if we can still allocate one more block without exceeding the
  // over-allocation ratio, then we should not flush.
  if (allocated_memory + kArenaBlockSize <
-      moptions_.write_buffer_size +
-      kArenaBlockSize * kAllowOverAllocationRatio) {
+      write_buffer_size + kArenaBlockSize * kAllowOverAllocationRatio) {
    return false;
  }

-  // if user keeps adding entries that exceeds moptions.write_buffer_size,
-  // we need to flush earlier even though we still have much available
-  // memory left.
-  if (allocated_memory > moptions_.write_buffer_size +
-      kArenaBlockSize * kAllowOverAllocationRatio) {
+  // if user keeps adding entries that exceeds write_buffer_size, we need to
+  // flush earlier even though we still have much available memory left.
+  if (allocated_memory >
+      write_buffer_size + kArenaBlockSize * kAllowOverAllocationRatio) {
    return true;
  }

@@ -225,7 +225,7 @@ int MemTable::KeyComparator::operator()(const char* prefix_len_key1,
  // Internal keys are encoded as length-prefixed strings.
  Slice k1 = GetLengthPrefixedSlice(prefix_len_key1);
  Slice k2 = GetLengthPrefixedSlice(prefix_len_key2);
-  return comparator.Compare(k1, k2);
+  return comparator.CompareKeySeq(k1, k2);
 }

 int MemTable::KeyComparator::operator()(const char* prefix_len_key,
@@ -233,7 +233,15 @@ int MemTable::KeyComparator::operator()(const char* prefix_len_key,
    const {
  // Internal keys are encoded as length-prefixed strings.
  Slice a = GetLengthPrefixedSlice(prefix_len_key);
-  return comparator.Compare(a, key);
+  return comparator.CompareKeySeq(a, key);
+}
+
+void MemTableRep::InsertConcurrently(KeyHandle handle) {
+#ifndef ROCKSDB_LITE
+  throw std::runtime_error("concurrent insert not supported");
+#else
+  abort();
+#endif
 }

 Slice MemTableRep::UserKey(const char* key) const {
@@ -265,7 +273,8 @@ class MemTableIterator : public InternalIterator {
        comparator_(mem.comparator_),
        valid_(false),
        arena_mode_(arena != nullptr),
-        value_pinned_(!mem.GetMemTableOptions()->inplace_update_support) {
+        value_pinned_(
+            !mem.GetImmutableMemTableOptions()->inplace_update_support) {
    if (use_range_del_table) {
      iter_ = mem.range_del_table_->GetIterator(arena);
    } else if (prefix_extractor_ != nullptr && !read_options.total_order_seek) {
@@ -280,8 +289,7 @@ class MemTableIterator : public InternalIterator {
 #ifndef NDEBUG
    // Assert that the MemTableIterator is never deleted while
    // Pinning is Enabled.
-    assert(!pinned_iters_mgr_ ||
-           (pinned_iters_mgr_ && !pinned_iters_mgr_->PinningEnabled()));
+    assert(!pinned_iters_mgr_ || !pinned_iters_mgr_->PinningEnabled());
 #endif
    if (arena_mode_) {
      iter_->~Iterator();
@@ -435,7 +443,7 @@ MemTable::MemTableStats MemTable::ApproximateStats(const Slice& start_ikey,
  return {entry_count * (data_size / n), entry_count};
 }

-void MemTable::Add(SequenceNumber s, ValueType type,
+bool MemTable::Add(SequenceNumber s, ValueType type,
                   const Slice& key, /* user key */
                   const Slice& value, bool allow_concurrent,
                   MemTablePostProcessInfo* post_process_info) {
@@ -470,9 +478,15 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    if (insert_with_hint_prefix_extractor_ != nullptr &&
        insert_with_hint_prefix_extractor_->InDomain(key_slice)) {
      Slice prefix = insert_with_hint_prefix_extractor_->Transform(key_slice);
-      table->InsertWithHint(handle, &insert_hints_[prefix]);
+      bool res = table->InsertKeyWithHint(handle, &insert_hints_[prefix]);
+      if (!res) {
+        return res;
+      }
    } else {
-      table->Insert(handle);
+      bool res = table->InsertKey(handle);
+      if (!res) {
+        return res;
+      }
    }

    // this is a bit ugly, but is the way to avoid locked instructions
@@ -505,7 +519,10 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    assert(post_process_info == nullptr);
    UpdateFlushState();
  } else {
-    table->InsertConcurrently(handle);
+    bool res = table->InsertKeyConcurrently(handle);
+    if (!res) {
+      return res;
+    }

    assert(post_process_info != nullptr);
    post_process_info->num_entries++;
@@ -535,6 +552,7 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    is_range_del_table_empty_ = false;
  }
  UpdateOldestKeyTime();
+  return true;
 }

 // Callback from MemTable::Get()
@@ -570,11 +588,12 @@ struct Saver {

 static bool SaveValue(void* arg, const char* entry) {
  Saver* s = reinterpret_cast<Saver*>(arg);
+  assert(s != nullptr);
  MergeContext* merge_context = s->merge_context;
  RangeDelAggregator* range_del_agg = s->range_del_agg;
  const MergeOperator* merge_operator = s->merge_operator;

-  assert(s != nullptr && merge_context != nullptr && range_del_agg != nullptr);
+  assert(merge_context != nullptr && range_del_agg != nullptr);

  // entry format is:
  //    klength  varint32
@@ -790,8 +809,9 @@ void MemTable::Update(SequenceNumber seq,
      // Correct user key
      const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
      ValueType type;
-      SequenceNumber unused;
-      UnPackSequenceAndType(tag, &unused, &type);
+      SequenceNumber existing_seq;
+      UnPackSequenceAndType(tag, &existing_seq, &type);
+      assert(existing_seq != seq);
      if (type == kTypeValue) {
        Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
        uint32_t prev_size = static_cast<uint32_t>(prev_value.size());
@@ -814,7 +834,10 @@ void MemTable::Update(SequenceNumber seq,
  }

  // key doesn't exist
-  Add(seq, kTypeValue, key, value);
+  bool add_res __attribute__((__unused__));
+  add_res = Add(seq, kTypeValue, key, value);
+  // We already checked unused != seq above. In that case, Add should not fail.
+  assert(add_res);
 }

 bool MemTable::UpdateCallback(SequenceNumber seq,
@@ -36,11 +36,9 @@ class MemTableIterator;
 class MergeContext;
 class InternalIterator;

-struct MemTableOptions {
-  explicit MemTableOptions(
-      const ImmutableCFOptions& ioptions,
-      const MutableCFOptions& mutable_cf_options);
-  size_t write_buffer_size;
+struct ImmutableMemTableOptions {
+  explicit ImmutableMemTableOptions(const ImmutableCFOptions& ioptions,
+                                    const MutableCFOptions& mutable_cf_options);
  size_t arena_block_size;
  uint32_t memtable_prefix_bloom_bits;
  size_t memtable_huge_page_size;
@@ -169,7 +167,10 @@ class MemTable {
  //
  // REQUIRES: if allow_concurrent = false, external synchronization to prevent
  // simultaneous operations on the same MemTable.
-  void Add(SequenceNumber seq, ValueType type, const Slice& key,
+  //
+  // Returns false if MemTableRepFactory::CanHandleDuplicatedKey() is true and
+  // the <key, seq> already exists.
+  bool Add(SequenceNumber seq, ValueType type, const Slice& key,
           const Slice& value, bool allow_concurrent = false,
           MemTablePostProcessInfo* post_process_info = nullptr);

@@ -262,6 +263,18 @@ class MemTable {
    return num_deletes_.load(std::memory_order_relaxed);
  }

+  // Dynamically change the memtable's capacity. If set below the current usage,
+  // the next key added will trigger a flush. Can only increase size when
+  // memtable prefix bloom is disabled, since we can't easily allocate more
+  // space.
+  void UpdateWriteBufferSize(size_t new_write_buffer_size) {
+    if (prefix_bloom_ == nullptr ||
+        new_write_buffer_size < write_buffer_size_) {
+      write_buffer_size_.store(new_write_buffer_size,
+                               std::memory_order_relaxed);
+    }
+  }
+
  // Returns the edits area that is needed for flushing the memtable
  VersionEdit* GetEdits() { return &edit_; }

@@ -350,12 +363,19 @@ class MemTable {
    return comparator_.comparator;
  }

-  const MemTableOptions* GetMemTableOptions() const { return &moptions_; }
+  const ImmutableMemTableOptions* GetImmutableMemTableOptions() const {
+    return &moptions_;
+  }

  uint64_t ApproximateOldestKeyTime() const {
    return oldest_key_time_.load(std::memory_order_relaxed);
  }

+  // REQUIRES: db_mutex held.
+  void SetID(uint64_t id) { id_ = id; }
+
+  uint64_t GetID() const { return id_; }
+
 private:
  enum FlushStateEnum { FLUSH_NOT_REQUESTED, FLUSH_REQUESTED, FLUSH_SCHEDULED };

@@ -364,7 +384,7 @@ class MemTable {
  friend class MemTableList;

  KeyComparator comparator_;
-  const MemTableOptions moptions_;
+  const ImmutableMemTableOptions moptions_;
  int refs_;
  const size_t kArenaBlockSize;
  AllocTracker mem_tracker_;
@@ -378,6 +398,9 @@ class MemTable {
  std::atomic<uint64_t> num_entries_;
  std::atomic<uint64_t> num_deletes_;

+  // Dynamically changeable memtable option
+  std::atomic<size_t> write_buffer_size_;
+
  // These are used to manage memtable flushes to storage
  bool flush_in_progress_; // started the flush
  bool flush_completed_;   // finished the flush
@@ -422,6 +445,9 @@ class MemTable {
  // Timestamp of oldest key
  std::atomic<uint64_t> oldest_key_time_;

+  // Memtable id to track flush.
+  uint64_t id_ = 0;
+
  // Returns a heuristic flush decision
  bool ShouldFlushNow() const;

@@ -5,11 +5,12 @@
 //
 #pragma once

-#include <string>
-#include <list>
-#include <vector>
-#include <set>
 #include <deque>
+#include <limits>
+#include <list>
+#include <set>
+#include <string>
+#include <vector>

 #include "db/dbformat.h"
 #include "db/memtable.h"
@@ -244,6 +245,22 @@ class MemTableList {

  uint64_t GetMinLogContainingPrepSection();

+  uint64_t GetEarliestMemTableID() const {
+    auto& memlist = current_->memlist_;
+    if (memlist.empty()) {
+      return std::numeric_limits<uint64_t>::max();
+    }
+    return memlist.back()->GetID();
+  }
+
+  uint64_t GetLatestMemTableID() const {
+    auto& memlist = current_->memlist_;
+    if (memlist.empty()) {
+      return 0;
+    }
+    return memlist.front()->GetID();
+  }
+
 private:
  // DB mutex held
  void InstallNewVersion();
@@ -5,12 +5,12 @@

 #include "db/merge_helper.h"

-#include <stdio.h>
 #include <string>

 #include "db/dbformat.h"
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/statistics.h"
+#include "port/likely.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/merge_operator.h"
@@ -22,7 +22,8 @@ MergeHelper::MergeHelper(Env* env, const Comparator* user_comparator,
                         const MergeOperator* user_merge_operator,
                         const CompactionFilter* compaction_filter,
                         Logger* logger, bool assert_valid_internal_key,
-                         SequenceNumber latest_snapshot, int level,
+                         SequenceNumber latest_snapshot,
+                         const SnapshotChecker* snapshot_checker, int level,
                         Statistics* stats,
                         const std::atomic<bool>* shutting_down)
    : env_(env),
@@ -34,6 +35,7 @@ MergeHelper::MergeHelper(Env* env, const Comparator* user_comparator,
      assert_valid_internal_key_(assert_valid_internal_key),
      allow_single_operand_(false),
      latest_snapshot_(latest_snapshot),
+      snapshot_checker_(snapshot_checker),
      level_(level),
      keys_(),
      filter_timer_(env_),
@@ -158,7 +160,10 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
      // hit a different user key, stop right here
      hit_the_next_user_key = true;
      break;
-    } else if (stop_before && ikey.sequence <= stop_before) {
+    } else if (stop_before > 0 && ikey.sequence <= stop_before &&
+               LIKELY(snapshot_checker_ == nullptr ||
+                      snapshot_checker_->IsInSnapshot(ikey.sequence,
+                                                      stop_before))) {
      // hit an entry that's visible by the previous snapshot, can't touch that
      break;
    }
@@ -13,6 +13,7 @@
 #include "db/dbformat.h"
 #include "db/merge_context.h"
 #include "db/range_del_aggregator.h"
+#include "db/snapshot_checker.h"
 #include "rocksdb/compaction_filter.h"
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
@@ -33,7 +34,8 @@ class MergeHelper {
              const MergeOperator* user_merge_operator,
              const CompactionFilter* compaction_filter, Logger* logger,
              bool assert_valid_internal_key, SequenceNumber latest_snapshot,
-              int level = 0, Statistics* stats = nullptr,
+              const SnapshotChecker* snapshot_checker = nullptr, int level = 0,
+              Statistics* stats = nullptr,
              const std::atomic<bool>* shutting_down = nullptr);

  // Wrapper around MergeOperator::FullMergeV2() that records perf statistics.
@@ -145,6 +147,7 @@ class MergeHelper {
  bool assert_valid_internal_key_; // enforce no internal key corruption?
  bool allow_single_operand_;
  SequenceNumber latest_snapshot_;
+  const SnapshotChecker* const snapshot_checker_;
  int level_;

  // the scratch area that holds the result of MergeUntil
@@ -0,0 +1,30 @@
+// Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#include "rocksdb/status.h"
+
+namespace rocksdb {
+
+class DB;
+
+class PreReleaseCallback {
+ public:
+  virtual ~PreReleaseCallback() {}
+
+  // Will be called while on the write thread after the write and before the
+  // release of the sequence number. This is useful if any operation needs to be
+  // done before the write gets visible to the readers, or if we want to reduce
+  // the overhead of locking by updating something sequentially while we are on
+  // the write thread. If the callback fails, this function returns a non-OK
+  // status, the sequence number will not be released, and same status will be
+  // propagated to all the writers in the write group.
+  // seq is the sequence number that is used for this write and will be
+  // released.
+  virtual Status Callback(SequenceNumber seq) = 0;
+};
+
+}  //  namespace rocksdb
@@ -17,7 +17,6 @@ int main() {
 #include <iostream>
 #include <vector>

-#include <gflags/gflags.h>
 #include "db/db_impl.h"
 #include "monitoring/histogram.h"
 #include "rocksdb/comparator.h"
@@ -27,14 +26,15 @@ int main() {
 #include "rocksdb/perf_context.h"
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/table.h"
+#include "util/coding.h"
+#include "util/gflags_compat.h"
 #include "util/random.h"
 #include "util/stop_watch.h"
 #include "util/string_util.h"
 #include "util/testharness.h"
 #include "utilities/merge_operators.h"
-#include "util/coding.h"

-using GFLAGS::ParseCommandLineFlags;
+using GFLAGS_NAMESPACE::ParseCommandLineFlags;

 DEFINE_bool(trigger_deadlock, false,
            "issue delete in range scan to trigger PrefixHashMap deadlock");
@@ -879,8 +879,6 @@ TEST_F(PrefixTest, PrefixSeekModePrev3) {
 int main(int argc, char** argv) {
  ::testing::InitGoogleTest(&argc, argv);
  ParseCommandLineFlags(&argc, &argv, true);
-  std::cout << kDbName << "\n";
-
  return RUN_ALL_TESTS();
 }