db_bench_tool: fix buffer size

Summary: Found by gcc warning: x86_64-pc-linux-gnu-g++ --version x86_64-pc-linux-gnu-g++ (GCC) 7.1.1 20170710 tools/db_bench_tool.cc: In member function 'void rocksdb::Benchmark::RandomWithVerify(rocksdb::ThreadState*)': tools/db_bench_tool.cc:4430:8: error: '%lu' directive output may be truncated writing between 1 and 19 bytes into a region of size between 0 and 66 [-Werror=format-truncation=] void RandomWithVerify(ThreadState* thread) { ^~~~~~~~~~~~~~~~ tools/db_bench_tool.cc:4430:8: note: directive argument in the range [0, 9223372036854775807] tools/db_bench_tool.cc:4492:13: note: 'snprintf' output between 37 and 128 bytes into a destination of size 100 snprintf(msg, sizeof(msg), ~~~~~~~~^~~~~~~~~~~~~~~~~~ "( get:%" PRIu64 " put:%" PRIu64 " del:%" PRIu64 " total:%" \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PRIu64 " found:%" PRIu64 ")", ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ gets_done, puts_done, deletes_done, readwrites_, found); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ cc1plus: all warnings being treated as errors Makefile:1707: recipe for target 'tools/db_bench_tool.o' failed Closes https://github.com/facebook/rocksdb/pull/2558 Differential Revision: D5398703 Pulled By: siying fbshipit-source-id: 6ffa552bbd8b59cfc2c36289f86ff9b9acca8ca6
table/block.h: change memset
2026-11-06 11:02:36 -05:00 · 2019-10-30 14:12:09 -07:00 · 2019-10-30 14:06:58 -07:00 · 2019-10-30 14:06:58 -07:00 · 2019-10-30 14:06:58 -07:00 · 2019-10-30 14:06:58 -07:00
354 changed files with 12038 additions and 5764 deletions
@@ -31,6 +31,7 @@ build/
 ldb
 manifest_dump
 sst_dump
+blob_dump
 column_aware_encoding_exp
 util/build_version.cc
 build_tools/VALGRIND_LOGS/
@@ -41,6 +42,7 @@ package/
 .phutil_module_cache
 unity.a
 tags
+etags
 rocksdb_dump
 rocksdb_undump
 db_test2
@@ -59,7 +61,7 @@ rocksdb.h
 unity.cc
 java/crossbuild/.vagrant
 .vagrant/
-java/**.asc
+java/**/*.asc
 java/javadoc

 scan_build_report/
@@ -7,6 +7,7 @@ os:
 compiler:
  - clang
  - gcc
+osx_image: xcode8.3
 jdk:
  - oraclejdk7
 cache:
@@ -15,47 +16,55 @@ cache:

 addons:
   apt:
-      sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
-      packages: ['clang-3.6' , 'g++-6', 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'libgflags-dev']
+      packages: ['zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'libgflags-dev', 'mingw-w64']
 env:
-  - TEST_GROUP=platform_dependent1
-  - TEST_GROUP=platform_dependent2
-  - TEST_GROUP=platform_dependent3
-  - TEST_GROUP=1
-  - TEST_GROUP=2
+  - TEST_GROUP=platform_dependent # 16-18 minutes
+  - TEST_GROUP=1 # 33-35 minutes
+  - TEST_GROUP=2 # 30-32 minutes
  # Run java tests
-  - JOB_NAME=java_test
+  - JOB_NAME=java_test # 4-11 minutes
  # Build ROCKSDB_LITE
-  - JOB_NAME=lite_build
+  - JOB_NAME=lite_build # 3-4 minutes
  # Build examples
-  - JOB_NAME=examples
+  - JOB_NAME=examples # 5-7 minutes
+  - JOB_NAME=cmake # 3-5 minutes
+  - JOB_NAME=cmake-mingw # 3 minutes

 matrix:
  exclude:
-  - os: osx
-    compiler: gcc
  - os: osx
    env: TEST_GROUP=1
  - os: osx
    env: TEST_GROUP=2
+  - os : osx
+    env: JOB_NAME=cmake-mingw
+  - os : linux
+    compiler: clang
+  - os : osx
+    compiler: gcc
+
+# https://docs.travis-ci.com/user/caching/#ccache-cache
+install:
+  - if [ "${TRAVIS_OS_NAME}" == osx ]; then
+      brew install ccache;
+      PATH=$PATH:/usr/local/opt/ccache/libexec;
+    fi

 before_script:
-  - if [[ "${TRAVIS_OS_NAME}" == 'linux' && "${CXX}" == 'clang++' ]]; then CXX=clang++-3.6; fi
-  # test one linux g++ build with g++-6
-  - if [[ "${TRAVIS_OS_NAME}" == 'linux' && "${CXX}" == 'g++' && "${JOB_NAME}" == 'unittests' ]]; then CXX=g++-6; fi
-  # Limit the maximum number of open file descriptors to 8192
-  - ulimit -n 8192 || true
+  # Increase the maximum number of open file descriptors, since some tests use
+  # more FDs than the default limit.
+  - ulimit -n 8192

 script:
  - ${CXX} --version
-  - if [[ "${TEST_GROUP}" == 'platform_dependent1' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=external_sst_file_basic_test make -j4 check_some; fi
-  - if [[ "${TEST_GROUP}" == 'platform_dependent2' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=external_sst_file_basic_test ROCKSDBTESTS_END=checkpoint_test make -j4 check_some; fi
-  - if [[ "${TEST_GROUP}" == 'platform_dependent3' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=checkpoint_test ROCKSDBTESTS_END=db_block_cache_test make -j4 check_some; fi
-  - if [[ "${TEST_GROUP}" == '1' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=comparator_db_test make -j4 check_some; fi
-  - if [[ "${TEST_GROUP}" == '2' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=comparator_db_test make -j4 check_some; fi
-  - if [[ "${JOB_NAME}" == 'java_test' ]]; then OPT=-DTRAVIS V=1 make clean jclean rocksdbjava jtest; fi
-  - if [[ "${JOB_NAME}" == 'lite_build' ]]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib; fi
-  - if [[ "${JOB_NAME}" == 'examples' ]]; then OPT=-DTRAVIS V=1 make -j4 static_lib; cd examples; make -j4; fi
+  - if [ "${TEST_GROUP}" == 'platform_dependent' ]; then OPT=-DTRAVIS V=1 make -j4 all; OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=db_block_cache_test make -j4 check_some; fi
+  - if [ "${TEST_GROUP}" == '1' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=comparator_db_test make -j4 check_some; fi
+  - if [ "${TEST_GROUP}" == '2' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=comparator_db_test make -j4 check_some; fi
+  - if [ "${JOB_NAME}" == 'java_test' ]; then OPT=-DTRAVIS V=1 make clean jclean && make rocksdbjava jtest; fi
+  - if [ "${JOB_NAME}" == 'lite_build' ]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib; 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 [ "${JOB_NAME}" == 'cmake-mingw' ]; then mkdir build && cd build && cmake .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb; fi
 notifications:
    email:
      - leveldb@fb.com
@@ -55,7 +55,7 @@ else()
      add_definitions(-DROCKSDB_JEMALLOC -DJEMALLOC_NO_DEMANGLE)
      include_directories(${JEMALLOC_INCLUDE_DIR})
    endif()
-  endif() 
+  endif()
  option(WITH_SNAPPY "build with SNAPPY" OFF)
  if(WITH_SNAPPY)
    find_package(snappy REQUIRED)
@@ -66,8 +66,8 @@ else()
 endif()

 if(WIN32)
-  execute_process(COMMAND powershell -Command "Get-Date -format MM_dd_yyyy" OUTPUT_VARIABLE DATE)
-  execute_process(COMMAND powershell -Command "Get-Date -format HH:mm:ss" OUTPUT_VARIABLE TIME)
+  execute_process(COMMAND powershell -noprofile -Command "Get-Date -format MM_dd_yyyy" OUTPUT_VARIABLE DATE)
+  execute_process(COMMAND powershell -noprofile -Command "Get-Date -format HH:mm:ss" OUTPUT_VARIABLE TIME)
  string(REGEX REPLACE "(..)_(..)_..(..).*" "\\1/\\2/\\3" DATE "${DATE}")
  string(REGEX REPLACE "(..):(.....).*" " \\1:\\2" TIME "${TIME}")
  set(GIT_DATE_TIME "${DATE} ${TIME}")
@@ -106,17 +106,58 @@ if(NOT WIN32)
  string(STRIP "${ROCKSDB_VERSION_MAJOR}" ROCKSDB_VERSION_MAJOR)
 endif()

-if(WIN32)
-  option(WITH_AVX2 "build with AVX2" ON)
-  if(WITH_AVX2)
-    if(MSVC)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
-    else()
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx2")
-    endif()
+option(WITH_MD_LIBRARY "build with MD" ON)
+if(WIN32 AND MSVC)
+  if(WITH_MD_LIBRARY)
+    set(RUNTIME_LIBRARY "MD")
+  else()
+    set(RUNTIME_LIBRARY "MT")
  endif()
 endif()

+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 "${CMAKE_CXX_FLAGS} -msse4.2")
+  endif()
+else()
+  if(MSVC)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
+  else()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
+  endif()
+endif()
+
+set(CMAKE_REQUIRED_FLAGS ${CMAKE_CXX_FLAGS})
+include(CheckCXXSourceCompiles)
+CHECK_CXX_SOURCE_COMPILES("
+#include <cstdint>
+#include <nmmintrin.h>
+int main() {
+  volatile uint32_t x = _mm_crc32_u32(0, 0);
+}
+" HAVE_SSE42)
+if(HAVE_SSE42)
+  add_definitions(-DHAVE_SSE42)
+elseif(FORCE_SSE42)
+  message(FATAL_ERROR "FORCE_SSE42=ON but unable to compile with SSE4.2 enabled")
+endif()
+
+CHECK_CXX_SOURCE_COMPILES("
+#if defined(_MSC_VER) && !defined(__thread)
+#define __thread __declspec(thread)
+#endif
+int main() {
+  static __thread int tls;
+}
+" HAVE_THREAD_LOCAL)
+if(HAVE_THREAD_LOCAL)
+  add_definitions(-DROCKSDB_SUPPORT_THREAD_LOCAL)
+endif()
+
 set(BUILD_VERSION_CC ${CMAKE_BINARY_DIR}/build_version.cc)
 configure_file(util/build_version.cc.in ${BUILD_VERSION_CC} @ONLY)
 add_library(build_version OBJECT ${BUILD_VERSION_CC})
@@ -133,7 +174,8 @@ else()
  endif()
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
+    add_definitions(-DNDEBUG)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -fno-omit-frame-pointer")
    include(CheckCXXCompilerFlag)
    CHECK_CXX_COMPILER_FLAG("-momit-leaf-frame-pointer" HAVE_OMIT_LEAF_FRAME_POINTER)
    if(HAVE_OMIT_LEAF_FRAME_POINTER)
@@ -153,7 +195,6 @@ endif()

 option(WITH_ASAN "build with ASAN" OFF)
 if(WITH_ASAN)
-  add_definitions(-DROCKSDB_TSAN_RUN)
  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=address")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=address")
  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=address")
@@ -165,7 +206,6 @@ endif()
 option(WITH_TSAN "build with TSAN" OFF)
 if(WITH_TSAN)
  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=thread -pie")
-  add_definitions(-DROCKSDB_TSAN_RUN)
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=thread -fPIC")
  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=thread -fPIC")
  if(WITH_JEMALLOC)
@@ -175,6 +215,7 @@ endif()

 option(WITH_UBSAN "build with UBSAN" OFF)
 if(WITH_UBSAN)
+  add_definitions(-DROCKSDB_UBSAN_RUN)
  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=undefined")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=undefined")
  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=undefined")
@@ -195,11 +236,11 @@ endif()
 if(MSVC)
  if((${OPTIMIZE_DEBUG} EQUAL 1))
    message(STATUS "Debug optimization is enabled")
-    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt /MDd")
+    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt /${RUNTIME_LIBRARY}d")
  else()
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm /MDd")
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm /${RUNTIME_LIBRARY}d")
  endif()
-  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /MD")
+  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /${RUNTIME_LIBRARY}")

  set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /DEBUG")
  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /DEBUG")
@@ -317,6 +358,7 @@ set(SOURCES
        db/internal_stats.cc
        db/log_reader.cc
        db/log_writer.cc
+        db/malloc_stats.cc
        db/managed_iterator.cc
        db/memtable.cc
        db/memtable_list.cc
@@ -339,13 +381,14 @@ set(SOURCES
        env/env.cc
        env/env_chroot.cc
        env/env_hdfs.cc
-        env/memenv.cc
+        env/mock_env.cc
+        memtable/alloc_tracker.cc
        memtable/hash_cuckoo_rep.cc
        memtable/hash_linklist_rep.cc
        memtable/hash_skiplist_rep.cc
-        memtable/memtable_allocator.cc
        memtable/skiplistrep.cc
        memtable/vectorrep.cc
+        memtable/write_buffer_manager.cc
        monitoring/histogram.cc
        monitoring/histogram_windowing.cc
        monitoring/instrumented_mutex.cc
@@ -434,16 +477,18 @@ set(SOURCES
        utilities/blob_db/blob_db.cc
        utilities/blob_db/blob_db_impl.cc
        utilities/blob_db/blob_db_options_impl.cc
+        utilities/blob_db/blob_dump_tool.cc
        utilities/blob_db/blob_file.cc
        utilities/blob_db/blob_log_reader.cc
        utilities/blob_db/blob_log_writer.cc
        utilities/blob_db/blob_log_format.cc
-        utilities/checkpoint/checkpoint.cc
+        utilities/checkpoint/checkpoint_impl.cc
        utilities/col_buf_decoder.cc
        utilities/col_buf_encoder.cc
        utilities/column_aware_encoding_util.cc
        utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc
        utilities/date_tiered/date_tiered_db_impl.cc
+        utilities/debug.cc
        utilities/document/document_db.cc
        utilities/document/json_document.cc
        utilities/document/json_document_builder.cc
@@ -455,6 +500,8 @@ set(SOURCES
        utilities/memory/memory_util.cc
        utilities/merge_operators/max.cc
        utilities/merge_operators/put.cc
+        utilities/merge_operators/cassandra/format.cc
+        utilities/merge_operators/cassandra/merge_operator.cc
        utilities/merge_operators/string_append/stringappend.cc
        utilities/merge_operators/string_append/stringappend2.cc
        utilities/merge_operators/uint64add.cc
@@ -498,16 +545,14 @@ else()
    env/io_posix.cc)
 endif()

+set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
+set(ROCKSDB_SHARED_LIB rocksdb-shared${ARTIFACT_SUFFIX})
+set(ROCKSDB_IMPORT_LIB ${ROCKSDB_SHARED_LIB})
 if(WIN32)
  set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)
-  set(ROCKSDB_STATIC_LIB rocksdblib${ARTIFACT_SUFFIX})
-  set(ROCKSDB_IMPORT_LIB rocksdb${ARTIFACT_SUFFIX})
  set(LIBS ${ROCKSDB_STATIC_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
 else()
-  set(SYSTEM_LIBS ${CMAKE_THREAD_LIBS_INIT} rt)
-  set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
-  set(ROCKSDB_SHARED_LIB rocksdb-shared)
-  set(ROCKSDB_IMPORT_LIB ${ROCKSDB_SHARED_LIB})
+  set(SYSTEM_LIBS ${CMAKE_THREAD_LIBS_INIT})
  set(LIBS ${ROCKSDB_SHARED_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})

  add_library(${ROCKSDB_SHARED_LIB} SHARED ${SOURCES})
@@ -591,6 +636,7 @@ set(TESTS
        db/db_test2.cc
        db/db_universal_compaction_test.cc
        db/db_wal_test.cc
+        db/db_write_test.cc
        db/dbformat_test.cc
        db/deletefile_test.cc
        db/external_sst_file_basic_test.cc
@@ -623,6 +669,7 @@ set(TESTS
        env/mock_env_test.cc
        memtable/inlineskiplist_test.cc
        memtable/skiplist_test.cc
+        memtable/write_buffer_manager_test.cc
        monitoring/histogram_test.cc
        monitoring/iostats_context_test.cc
        monitoring/statistics_test.cc
@@ -649,6 +696,7 @@ set(TESTS
        util/event_logger_test.cc
        util/file_reader_writer_test.cc
        util/filelock_test.cc
+        util/hash_test.cc
        util/heap_test.cc
        util/rate_limiter_test.cc
        util/slice_transform_test.cc
@@ -656,6 +704,7 @@ set(TESTS
        util/thread_list_test.cc
        util/thread_local_test.cc
        utilities/backupable/backupable_db_test.cc
+        utilities/blob_db/blob_db_test.cc
        utilities/checkpoint/checkpoint_test.cc
        utilities/column_aware_encoding_test.cc
        utilities/date_tiered/date_tiered_test.cc
@@ -664,6 +713,10 @@ set(TESTS
        utilities/geodb/geodb_test.cc
        utilities/lua/rocks_lua_test.cc
        utilities/memory/memory_test.cc
+        utilities/merge_operators/cassandra/cassandra_merge_test.cc
+        utilities/merge_operators/cassandra/cassandra_format_test.cc
+        utilities/merge_operators/cassandra/cassandra_row_merge_test.cc
+        utilities/merge_operators/cassandra/cassandra_serialize_test.cc
        utilities/merge_operators/string_append/stringappend_test.cc
        utilities/object_registry_test.cc
        utilities/option_change_migration/option_change_migration_test.cc
@@ -701,10 +754,10 @@ endforeach(sourcefile ${BENCHMARKS})
 # and linked to tests. Add test only code that is not #ifdefed for Release here.
 set(TESTUTIL_SOURCE
    db/db_test_util.cc
-    env/mock_env.cc
    monitoring/thread_status_updater_debug.cc
    table/mock_table.cc
    util/fault_injection_test_env.cc
+    utilities/merge_operators/cassandra/test_utils.cc
 )
 # test utilities are only build in debug
 enable_testing()
@@ -1,5 +1,8 @@
 # RocksDB default options change log
 ## Unreleased
+* delayed_write_rate takes the rate given by rate_limiter if not specified.
+
+## 5.2
 * Change the default of delayed slowdown value to 16MB/s and further increase the L0 stop condition to 36 files.

 ## 5.0 (11/17/2016)
@@ -1,64 +1,50 @@
 # Rocksdb Change Log
-## 5.4.12 (08/21/2017)
-* Statistics::Reset() to reset user stats.
-
-## 5.4.11 (08/17/2017)
-### Bug Fixes
-* Fix a bug in PosixWritableFile::Close()
-
-## 5.4.10 (08/12/2017)
+## 5.6.2 (08/12/2017)
 ### Bug Fixes
 * Fix incorrect dropping of deletions during intra-L0 compaction.

-## 5.4.9 (07/20/2017)
-* Change license to Apache + GPLv2
+## 5.6.1 (07/25/2017)
+### Bug Fixes
+* Fix lite build.

-## 5.4.8 (07/13/2017)
-No change from 5.4.7.
+## 5.6.0 (06/06/2017)
+### Public API Change
+* Scheduling flushes and compactions in the same thread pool is no longer supported by setting `max_background_flushes=0`. Instead, users can achieve this by configuring their high-pri thread pool to have zero threads.
+* Replace `Options::max_background_flushes`, `Options::max_background_compactions`, and `Options::base_background_compactions` all with `Options::max_background_jobs`, which automatically decides how many threads to allocate towards flush/compaction.
+* options.delayed_write_rate by default take the value of options.rate_limiter rate.
+* Replace global variable `IOStatsContext iostats_context` with `IOStatsContext* get_iostats_context()`; replace global variable `PerfContext perf_context` with `PerfContext* get_perf_context()`.
+
+### New Features
+* Change ticker/histogram statistics implementations to use core-local storage. This improves aggregation speed compared to our previous thread-local approach, particularly for applications with many threads.
+* Users can pass a cache object to write buffer manager, so that they can cap memory usage for memtable and block cache using one single limit.
+* Flush will be triggered when 7/8 of the limit introduced by write_buffer_manager or db_write_buffer_size is triggered, so that the hard threshold is hard to hit.
+* Introduce WriteOptions.low_pri. If it is true, low priority writes will be throttled if the compaction is behind.
+* `DB::IngestExternalFile()` now supports ingesting files into a database containing range deletions.

-## 5.4.7 (06/28/2017)
 ### Bug Fixes
 * Shouldn't ignore return value of fsync() in flush.

-## 5.4.6 (05/30/2017)
-### Bug Fixes
-* Fix db_bench with lite build.
-
-## 5.4.5 (05/19/2017)
-### Bug Fixes
-* Facility for cross-building RocksJava using Docker.
-* Build and link with ZStd when creating the static RocksJava build.
-* Fix snprintf's usage to be cross-platform.
-* Fix build error with blob DB.
-* Fix readamp test type inconsistency.
-
-## 5.4.4 (05/11/2017)
+## 5.5.0 (05/17/2017)
 ### New Features
+* FIFO compaction to support Intra L0 compaction too with CompactionOptionsFIFO.allow_compaction=true.
+* DB::ResetStats() to reset internal stats.
+* Statistics::Reset() to reset user stats.
+* ldb add option --try_load_options, which will open DB with its own option file.
+* Introduce WriteBatch::PopSavePoint to pop the most recent save point explicitly.
+* Support dynamically change `max_open_files` option via SetDBOptions()
 * Added DB::CreateColumnFamilie() and DB::DropColumnFamilies() to bulk create/drop column families.
+* Add debugging function `GetAllKeyVersions` to see internal versions of a range of keys.
+* Support file ingestion with universal compaction style
+* Support file ingestion behind with option `allow_ingest_behind`
+* New option enable_pipelined_write which may improve write throughput in case writing from multiple threads and WAL enabled.

 ### Bug Fixes
-* Fix a manual / auto compaction data race.
-* Fix CentOS 5 cross-building of RocksJava.
-
-## 5.4.3 (05/10/2017)
-### Bug Fixes
-* Fix memory alignment with logical sector size.
-* Fix alignment in ReadaheadRandomAccessFile.
-* Fix bias with read amplification stats (READ_AMP_ESTIMATE_USEFUL_BYTES and READ_AMP_TOTAL_READ_BYTES).
-
-## 5.4.2 (05/08/2017)
-### Bug Fixes
-* Fix huge fallocate calls fail and make XFS unhappy.
-
-## 5.4.1 (05/02/2017)
-### Bug Fxies
-* Fix WriteBatchWithIndex address use after scope error.
-* Fix WritableFile buffer size in direct IO.
-* Add prefetch to PosixRandomAccessFile in buffered io.
-* Fix PinnableSlice access invalid address when row cache is enabled.
+* Fix the bug that Direct I/O uses direct reads for non-SST file

 ## 5.4.0 (04/11/2017)
 ### Public API Change
+* random_access_max_buffer_size no longer has any effect
+* Removed Env::EnableReadAhead(), Env::ShouldForwardRawRequest()
 * Support dynamically change `stats_dump_period_sec` option via SetDBOptions().
 * Added ReadOptions::max_skippable_internal_keys to set a threshold to fail a request as incomplete when too many keys are being skipped when using iterators.
 * DB::Get in place of std::string accepts PinnableSlice, which avoids the extra memcpy of value to std::string in most of cases.
@@ -71,7 +57,6 @@ No change from 5.4.7.
 * Memtable flush can be avoided during checkpoint creation if total log file size is smaller than a threshold specified by the user.
 * Introduce level-based L0->L0 compactions to reduce file count, so write delays are incurred less often.
 * (Experimental) Partitioning filters which creates an index on the partitions. The feature can be enabled by setting partition_filters when using kFullFilter. Currently the feature also requires two-level indexing to be enabled. Number of partitions is the same as the number of partitions for indexes, which is controlled by metadata_block_size.
-* DB::ResetStats() to reset internal stats.

 ## 5.3.0 (03/08/2017)
 ### Public API Change
@@ -298,7 +283,7 @@ No change from 5.4.7.
 * Added a new way to report QPS from db_bench (check out --report_file and --report_interval_seconds)
 * Added a cache for individual rows. See DBOptions::row_cache for more info.
 * Several new features on EventListener (see include/rocksdb/listener.h):
- - OnCompationCompleted() now returns per-compaciton job statistics, defined in include/rocksdb/compaction_job_stats.h.
+ - OnCompationCompleted() now returns per-compaction job statistics, defined in include/rocksdb/compaction_job_stats.h.
 - Added OnTableFileCreated() and OnTableFileDeleted().
 * Add compaction_options_universal.enable_trivial_move to true, to allow trivial move while performing universal compaction. Trivial move will happen only when all the input files are non overlapping.

@@ -21,9 +21,11 @@ depend on gflags. You will need to have gflags installed to run `make all`. This
 use binaries compiled by `make all` in production.

 * By default the binary we produce is optimized for the platform you're compiling on
-(-march=native or the equivalent). If you want to build a portable binary, add 'PORTABLE=1' before
-your make commands, like this: `PORTABLE=1 make static_lib`. If you want to build a binary that
-makes use of SSE4, add 'USE_SSE=1' before your make commands, like this: `USE_SSE=1 make static_lib`.
+(`-march=native` or the equivalent). SSE4.2 will thus be enabled automatically if your
+CPU supports it. To print a warning if your CPU does not support SSE4.2, build with
+`USE_SSE=1 make static_lib` or, if using CMake, `cmake -DFORCE_SSE42=ON`. If you want
+to build a portable binary, add `PORTABLE=1` before your make commands, like this:
+`PORTABLE=1 make static_lib`.

 ## Dependencies

@@ -63,26 +65,25 @@ makes use of SSE4, add 'USE_SSE=1' before your make commands, like this: `USE_SS
              git checkout v2.0
              ./configure && make && sudo make install

-      **Notice**: Once installed, please add the include path for gflags to your CPATH env var and the
-      lib path to LIBRARY_PATH. If installed with default settings, the lib will be /usr/local/lib
-      and the include path will be /usr/local/include.
+      **Notice**: Once installed, please add the include path for gflags to your `CPATH` environment variable and the
+      lib path to `LIBRARY_PATH`. If installed with default settings, the include path will be `/usr/local/include`
+      and the lib path will be `/usr/local/lib`.

    * Install snappy:

-              wget https://github.com/google/snappy/releases/download/1.1.4/snappy-1.1.4.tar.gz
-              tar -xzvf snappy-1.1.4.tar.gz
-              cd snappy-1.1.4
-              ./configure && make && sudo make install
+              sudo yum install snappy snappy-devel

    * Install zlib:

-              sudo yum install zlib
-              sudo yum install zlib-devel
+              sudo yum install zlib zlib-devel

    * Install bzip2:

-              sudo yum install bzip2
-              sudo yum install bzip2-devel
+              sudo yum install bzip2 bzip2-devel
+
+    * Install ASAN (optional for debugging):
+
+              sudo yum install libasan

    * Install zstandard:

@@ -106,3 +107,30 @@ makes use of SSE4, add 'USE_SSE=1' before your make commands, like this: `USE_SS
 * **Windows**:
  * For building with MS Visual Studio 13 you will need Update 4 installed.
  * Read and follow the instructions at CMakeLists.txt
+
+* **AIX 6.1**
+    * Install AIX Toolbox rpms with gcc
+    * Use these environment variables:
+  
+             export PORTABLE=1
+             export CC=gcc
+             export AR="ar -X64"
+             export EXTRA_ARFLAGS=-X64
+             export EXTRA_CFLAGS=-maix64
+             export EXTRA_CXXFLAGS=-maix64
+             export PLATFORM_LDFLAGS="-static-libstdc++ -static-libgcc"
+             export LIBPATH=/opt/freeware/lib
+             export JAVA_HOME=/usr/java8_64
+             export PATH=/opt/freeware/bin:$PATH
+  
+* **Solaris Sparc**
+    * Install GCC 4.8.2 and higher.
+    * Use these environment variables:
+
+             export CC=gcc
+             export EXTRA_CFLAGS=-m64
+             export EXTRA_CXXFLAGS=-m64
+             export EXTRA_LDFLAGS=-m64
+             export PORTABLE=1
+             export PLATFORM_LDFLAGS="-static-libstdc++ -static-libgcc"
+
@@ -14,7 +14,8 @@ CFLAGS += ${EXTRA_CFLAGS}
 CXXFLAGS += ${EXTRA_CXXFLAGS}
 LDFLAGS += $(EXTRA_LDFLAGS)
 MACHINE ?= $(shell uname -m)
-ARFLAGS = rs
+ARFLAGS = ${EXTRA_ARFLAGS} rs
+STRIPFLAGS = -S -x

 # Transform parallel LOG output into something more readable.
 perl_command = perl -n \
@@ -154,7 +155,9 @@ missing_make_config_paths := $(shell				\
 $(foreach path, $(missing_make_config_paths), \
 	$(warning Warning: $(path) dont exist))

-ifneq ($(PLATFORM), IOS)
+ifeq ($(PLATFORM), OS_AIX)
+# no debug info
+else ifneq ($(PLATFORM), IOS)
 CFLAGS += -g
 CXXFLAGS += -g
 else
@@ -162,6 +165,11 @@ else
 OPT += -DNDEBUG
 endif

+ifeq ($(PLATFORM), OS_AIX)
+ARFLAGS = -X64 rs
+STRIPFLAGS = -X64 -x
+endif
+
 ifeq ($(PLATFORM), OS_SOLARIS)
 	PLATFORM_CXXFLAGS += -D _GLIBCXX_USE_C99
 endif
@@ -184,9 +192,9 @@ endif
 # TSAN doesn't work well with jemalloc. If we're compiling with TSAN, we should use regular malloc.
 ifdef COMPILE_WITH_TSAN
 	DISABLE_JEMALLOC=1
-	EXEC_LDFLAGS += -fsanitize=thread -pie
-	PLATFORM_CCFLAGS += -fsanitize=thread -fPIC -DROCKSDB_TSAN_RUN
-	PLATFORM_CXXFLAGS += -fsanitize=thread -fPIC -DROCKSDB_TSAN_RUN
+	EXEC_LDFLAGS += -fsanitize=thread
+	PLATFORM_CCFLAGS += -fsanitize=thread -fPIC
+	PLATFORM_CXXFLAGS += -fsanitize=thread -fPIC
        # Turn off -pg when enabling TSAN testing, because that induces
        # a link failure.  TODO: find the root cause
 	PROFILING_FLAGS =
@@ -194,12 +202,17 @@ ifdef COMPILE_WITH_TSAN
 	LUA_PATH =
 endif

+# AIX doesn't work with -pg
+ifeq ($(PLATFORM), OS_AIX)
+	PROFILING_FLAGS =
+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
-	PLATFORM_CXXFLAGS += -fsanitize=undefined
+	PLATFORM_CCFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
+	PLATFORM_CXXFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
 endif

 ifndef DISABLE_JEMALLOC
@@ -215,8 +228,14 @@ endif
 export GTEST_THROW_ON_FAILURE=1
 export GTEST_HAS_EXCEPTIONS=1
 GTEST_DIR = ./third-party/gtest-1.7.0/fused-src
-PLATFORM_CCFLAGS += -isystem $(GTEST_DIR)
-PLATFORM_CXXFLAGS += -isystem $(GTEST_DIR)
+# AIX: pre-defined system headers are surrounded by an extern "C" block
+ifeq ($(PLATFORM), OS_AIX)
+	PLATFORM_CCFLAGS += -I$(GTEST_DIR)
+	PLATFORM_CXXFLAGS += -I$(GTEST_DIR)
+else
+	PLATFORM_CCFLAGS += -isystem $(GTEST_DIR)
+	PLATFORM_CXXFLAGS += -isystem $(GTEST_DIR)
+endif

 # This (the first rule) must depend on "all".
 default: all
@@ -224,6 +243,11 @@ default: all
 WARNING_FLAGS = -W -Wextra -Wall -Wsign-compare -Wshadow \
  -Wno-unused-parameter

+ifndef DISABLE_WARNING_AS_ERROR
+	WARNING_FLAGS += -Werror
+endif
+
+
 ifdef LUA_PATH

 ifndef LUA_INCLUDE
@@ -256,6 +280,9 @@ CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverl

 LDFLAGS += $(PLATFORM_LDFLAGS)

+# If NO_UPDATE_BUILD_VERSION is set we don't update util/build_version.cc, but
+# the file needs to already exist or else the build will fail
+ifndef NO_UPDATE_BUILD_VERSION
 date := $(shell date +%F)
 ifdef FORCE_GIT_SHA
 	git_sha := $(FORCE_GIT_SHA)
@@ -269,7 +296,6 @@ gen_build_version = sed -e s/@@GIT_SHA@@/$(git_sha)/ -e s/@@GIT_DATE_TIME@@/$(da
 # as a regular source file as part of the compilation process.
 # One can run "strings executable_filename | grep _build_" to find
 # the version of the source that we used to build the executable file.
-CLEAN_FILES += util/build_version.cc:
 FORCE:
 util/build_version.cc: FORCE
 	$(AM_V_GEN)rm -f $@-t
@@ -277,6 +303,7 @@ util/build_version.cc: FORCE
 	$(AM_V_at)if test -f $@; then					\
 	  cmp -s $@-t $@ && rm -f $@-t || mv -f $@-t $@;		\
 	else mv -f $@-t $@; fi
+endif

 LIBOBJECTS = $(LIB_SOURCES:.cc=.o)
 LIBOBJECTS += $(TOOL_LIB_SOURCES:.cc=.o)
@@ -308,6 +335,7 @@ TESTS = \
 	inlineskiplist_test \
 	env_basic_test \
 	env_test \
+	hash_test \
 	thread_local_test \
 	rate_limiter_test \
 	perf_context_test \
@@ -335,7 +363,9 @@ TESTS = \
 	db_properties_test \
 	db_table_properties_test \
 	db_statistics_test \
+	db_write_test \
 	autovector_test \
+	blob_db_test \
 	cleanable_test \
 	column_family_test \
 	table_properties_collector_test \
@@ -371,7 +401,12 @@ TESTS = \
 	external_sst_file_test \
 	prefix_test \
 	skiplist_test \
+	write_buffer_manager_test \
 	stringappend_test \
+	cassandra_format_test \
+	cassandra_merge_test \
+	cassandra_row_merge_test \
+	cassandra_serialize_test \
 	ttl_test \
 	date_tiered_test \
 	backupable_db_test \
@@ -458,7 +493,8 @@ TOOLS = \
 	ldb \
 	db_repl_stress \
 	rocksdb_dump \
-	rocksdb_undump
+	rocksdb_undump \
+	blob_dump \

 TEST_LIBS = \
 	librocksdb_env_basic_test.a
@@ -519,9 +555,14 @@ $(SHARED3): $(SHARED4)
 	ln -fs $(SHARED4) $(SHARED3)
 endif

-$(SHARED4):
-	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED3) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(LIB_SOURCES) $(TOOL_LIB_SOURCES) \
-		$(LDFLAGS) -o $@
+shared_libobjects = $(patsubst %,shared-objects/%,$(LIBOBJECTS))
+CLEAN_FILES += shared-objects
+
+$(shared_libobjects): shared-objects/%.o: %.cc
+	$(AM_V_CC)mkdir -p $(@D) && $(CXX) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) -c $< -o $@
+
+$(SHARED4): $(shared_libobjects)
+	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED3) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(shared_libobjects) $(LDFLAGS) -o $@

 endif  # PLATFORM_SHARED_EXT

@@ -562,7 +603,8 @@ ifneq (,$(filter check parallel_check,$(MAKECMDGOALS)),)
 # and create a randomly-named rocksdb.XXXX directory therein.
 # We'll use that directory in the "make check" rules.
 ifeq ($(TMPD),)
-TMPD := $(shell f=/dev/shm; test -k $$f || f=/tmp;			\
+TMPDIR := $(shell echo $${TMPDIR:-/tmp})
+TMPD := $(shell f=/dev/shm; test -k $$f || f=$(TMPDIR);     \
  perl -le 'use File::Temp "tempdir";'					\
    -e 'print tempdir("'$$f'/rocksdb.XXXX", CLEANUP => 0)')
 endif
@@ -717,10 +759,12 @@ check: all
 	      echo "===== Running $$t"; ./$$t || exit 1; done;          \
 	fi
 	rm -rf $(TMPD)
+ifneq ($(PLATFORM), OS_AIX)
 ifeq ($(filter -DROCKSDB_LITE,$(OPT)),)
 	python tools/ldb_test.py
 	sh tools/rocksdb_dump_test.sh
 endif
+endif

 # TODO add ldb_tests
 check_some: $(SUBSET)
@@ -866,12 +910,13 @@ clean:
 	rm -rf $(CLEAN_FILES) ios-x86 ios-arm scan_build_report
 	find . -name "*.[oda]" -exec rm -f {} \;
 	find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
-	rm -rf bzip2* snappy* zlib* lz4*
+	rm -rf bzip2* snappy* zlib* lz4* zstd*
 	cd java; $(MAKE) clean

 tags:
 	ctags * -R
 	cscope -b `find . -name '*.cc'` `find . -name '*.h'` `find . -name '*.c'`
+	ctags -e -R -o etags *

 format:
 	build_tools/format-diff.sh
@@ -945,12 +990,27 @@ cache_test: cache/cache_test.o $(LIBOBJECTS) $(TESTHARNESS)
 coding_test: util/coding_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+hash_test: util/hash_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 option_change_migration_test: utilities/option_change_migration/option_change_migration_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 stringappend_test: utilities/merge_operators/string_append/stringappend_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+cassandra_format_test: utilities/merge_operators/cassandra/cassandra_format_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+cassandra_merge_test: utilities/merge_operators/cassandra/cassandra_merge_test.o utilities/merge_operators/cassandra/test_utils.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+cassandra_row_merge_test: utilities/merge_operators/cassandra/cassandra_row_merge_test.o utilities/merge_operators/cassandra/test_utils.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+cassandra_serialize_test: utilities/merge_operators/cassandra/cassandra_serialize_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 redis_test: utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1026,6 +1086,9 @@ db_sst_test: db/db_sst_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 db_statistics_test: db/db_statistics_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_write_test: db/db_write_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 external_sst_file_basic_test: db/external_sst_file_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1181,6 +1244,9 @@ inlineskiplist_test: memtable/inlineskiplist_test.o $(LIBOBJECTS) $(TESTHARNESS)
 skiplist_test: memtable/skiplist_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+write_buffer_manager_test: memtable/write_buffer_manager_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 version_edit_test: db/version_edit_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1307,6 +1373,9 @@ transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TEST
 sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 	$(AM_LINK)

+blob_dump: tools/blob_dump.o $(LIBOBJECTS)
+	$(AM_LINK)
+
 column_aware_encoding_exp: utilities/column_aware_encoding_exp.o $(EXPOBJECTS)
 	$(AM_LINK)

@@ -1337,6 +1406,9 @@ lua_test: utilities/lua/rocks_lua_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTH
 range_del_aggregator_test: db/range_del_aggregator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+blob_db_test: utilities/blob_db/blob_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1396,14 +1468,18 @@ ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PA
 ROCKSDB_SOURCES_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar
 SHA256_CMD = sha256sum

-ZLIB_VER = 1.2.11
-ZLIB_SHA256 = c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
-BZIP2_VER = 1.0.6
-BZIP2_SHA256 = a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-SNAPPY_VER = 1.1.4
-SNAPPY_SHA256 = 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
-LZ4_VER = 1.7.5
-LZ4_SHA256 = 0190cacd63022ccb86f44fa5041dc6c3804407ad61550ca21c382827319e7e7e
+ZLIB_VER ?= 1.2.11
+ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
+ZLIB_DOWNLOAD_BASE ?= http://zlib.net
+BZIP2_VER ?= 1.0.6
+BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
+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_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
 ZSTD_VER ?= 1.2.0
 ZSTD_SHA256 ?= 4a7e4593a3638276ca7f2a09dc4f38e674d8317bbea51626393ca73fc047cbfb
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
@@ -1429,47 +1505,53 @@ ifeq ($(PLATFORM), OS_SOLARIS)
 	JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/solaris
 	SHA256_CMD = digest -a sha256
 endif
+ifeq ($(PLATFORM), OS_AIX)
+	JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/aix
+	ROCKSDBJNILIB = librocksdbjni-aix.so
+	EXTRACT_SOURCES = gunzip < TAR_GZ | tar xvf -
+	SNAPPY_MAKE_TARGET = libsnappy.la
+endif

 libz.a:
 	-rm -rf zlib-$(ZLIB_VER)
-	curl -O -L http://zlib.net/zlib-$(ZLIB_VER).tar.gz
+	curl -O -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
 	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
 		echo zlib-$(ZLIB_VER).tar.gz checksum mismatch, expected=\"$(ZLIB_SHA256)\" actual=\"$$ZLIB_SHA256_ACTUAL\"; \
 		exit 1; \
 	fi
 	tar xvzf zlib-$(ZLIB_VER).tar.gz
-	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC' ./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:
 	-rm -rf bzip2-$(BZIP2_VER)
-	curl -O -L http://www.bzip.org/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
+	curl -O -L ${BZIP2_DOWNLOAD_BASE}/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
 	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
 		echo bzip2-$(BZIP2_VER).tar.gz checksum mismatch, expected=\"$(BZIP2_SHA256)\" actual=\"$$BZIP2_SHA256_ACTUAL\"; \
 		exit 1; \
 	fi
 	tar xvzf bzip2-$(BZIP2_VER).tar.gz
-	cd bzip2-$(BZIP2_VER) && make CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64'
+	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:
 	-rm -rf snappy-$(SNAPPY_VER)
-	curl -O -L https://github.com/google/snappy/releases/download/$(SNAPPY_VER)/snappy-$(SNAPPY_VER).tar.gz
+	curl -O -L ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER)/snappy-$(SNAPPY_VER).tar.gz
 	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
 		echo snappy-$(SNAPPY_VER).tar.gz checksum mismatch, expected=\"$(SNAPPY_SHA256)\" actual=\"$$SNAPPY_SHA256_ACTUAL\"; \
 		exit 1; \
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
-	cd snappy-$(SNAPPY_VER) && ./configure --with-pic --enable-static
-	cd snappy-$(SNAPPY_VER) && make
+	cd snappy-$(SNAPPY_VER) && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --with-pic --enable-static --disable-shared
+	cd snappy-$(SNAPPY_VER) && make ${SNAPPY_MAKE_TARGET}
 	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .

 liblz4.a:
 	-rm -rf lz4-$(LZ4_VER)
-	curl -O -L https://github.com/lz4/lz4/archive/v$(LZ4_VER).tar.gz
+	curl -O -L ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
 	mv v$(LZ4_VER).tar.gz lz4-$(LZ4_VER).tar.gz
 	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
@@ -1477,7 +1559,7 @@ liblz4.a:
 		exit 1; \
 	fi
 	tar xvzf lz4-$(LZ4_VER).tar.gz
-	cd lz4-$(LZ4_VER)/lib && make CFLAGS='-fPIC' all
+	cd lz4-$(LZ4_VER)/lib && make CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
 	cp lz4-$(LZ4_VER)/lib/liblz4.a .

 libzstd.a:
@@ -1514,7 +1596,7 @@ rocksdbjavastatic: $(java_static_libobjects)
 	  -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) \
 	  $(java_static_libobjects) $(COVERAGEFLAGS) \
 	  $(JAVA_COMPRESSIONS) $(JAVA_STATIC_LDFLAGS)
-	cd java/target;strip -S -x $(ROCKSDBJNILIB)
+	cd java/target;strip $(STRIPFLAGS) $(ROCKSDBJNILIB)
 	cd java;jar -cf target/$(ROCKSDB_JAR) HISTORY*.md
 	cd java/target;jar -uf $(ROCKSDB_JAR) $(ROCKSDBJNILIB)
 	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR) org/rocksdb/*.class org/rocksdb/util/*.class
@@ -1530,12 +1612,12 @@ rocksdbjavastaticrelease: rocksdbjavastatic
 rocksdbjavastaticreleasedocker: rocksdbjavastatic
 	DOCKER_LINUX_X64_CONTAINER=`docker ps -aqf name=rocksdb_linux_x64-be`; \
 	if [ -z "$$DOCKER_LINUX_X64_CONTAINER" ]; then \
-		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x64-be evolvedbinary/rocksjava:centos5_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
+		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x64-be evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
 	fi
 	docker start -a rocksdb_linux_x64-be
 	DOCKER_LINUX_X86_CONTAINER=`docker ps -aqf name=rocksdb_linux_x86-be`; \
 	if [ -z "$$DOCKER_LINUX_X86_CONTAINER" ]; then \
-		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x86-be evolvedbinary/rocksjava:centos5_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
+		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x86-be evolvedbinary/rocksjava:centos6_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
 	fi
 	docker start -a rocksdb_linux_x86-be
 	cd java;jar -cf target/$(ROCKSDB_JAR_ALL) HISTORY*.md
@@ -1,4 +1,8 @@
-REPO_PATH = "internal_repo_rocksdb/repo/"
+
+import os
+
+TARGETS_PATH = os.path.dirname(__file__)
+REPO_PATH = TARGETS_PATH[(TARGETS_PATH.find('fbcode/') + len('fbcode/')):] + "/"
 BUCK_BINS = "buck-out/gen/" + REPO_PATH
 TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
 rocksdb_compiler_flags = [
@@ -9,6 +13,10 @@ rocksdb_compiler_flags = [
  "-DROCKSDB_LIB_IO_POSIX",
  "-DROCKSDB_FALLOCATE_PRESENT",
  "-DROCKSDB_MALLOC_USABLE_SIZE",
+  "-DROCKSDB_RANGESYNC_PRESENT",
+  "-DROCKSDB_SCHED_GETCPU_PRESENT",
+  "-DROCKSDB_SUPPORT_THREAD_LOCAL",
+  "-DHAVE_SSE42",
  "-DOS_LINUX",
  # Flags to enable libs we include
  "-DSNAPPY",
@@ -31,7 +39,7 @@ rocksdb_external_deps = [
  ('lz4', None, 'lz4'),
  ('zstd', None),
  ('tbb', None),
-  ("numa", "2.0.8", "numa"),
+  ("numa", None, "numa"),
  ("googletest", None, "gtest"),
 ]

@@ -80,6 +88,7 @@ cpp_library(
      "db/internal_stats.cc",
      "db/log_reader.cc",
      "db/log_writer.cc",
+      "db/malloc_stats.cc",
      "db/managed_iterator.cc",
      "db/memtable.cc",
      "db/memtable_list.cc",
@@ -104,13 +113,14 @@ cpp_library(
      "env/env_hdfs.cc",
      "env/env_posix.cc",
      "env/io_posix.cc",
-      "env/memenv.cc",
+      "env/mock_env.cc",
+      "memtable/alloc_tracker.cc",
      "memtable/hash_cuckoo_rep.cc",
      "memtable/hash_linklist_rep.cc",
      "memtable/hash_skiplist_rep.cc",
-      "memtable/memtable_allocator.cc",
      "memtable/skiplistrep.cc",
      "memtable/vectorrep.cc",
+      "memtable/write_buffer_manager.cc",
      "monitoring/histogram.cc",
      "monitoring/histogram_windowing.cc",
      "monitoring/instrumented_mutex.cc",
@@ -201,10 +211,11 @@ cpp_library(
      "utilities/blob_db/blob_log_reader.cc",
      "utilities/blob_db/blob_log_writer.cc",
      "utilities/blob_db/blob_log_format.cc",
-      "utilities/checkpoint/checkpoint.cc",
+      "utilities/checkpoint/checkpoint_impl.cc",
      "utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc",
      "utilities/convenience/info_log_finder.cc",
      "utilities/date_tiered/date_tiered_db_impl.cc",
+      "utilities/debug.cc",
      "utilities/document/document_db.cc",
      "utilities/document/json_document.cc",
      "utilities/document/json_document_builder.cc",
@@ -214,6 +225,8 @@ cpp_library(
      "utilities/leveldb_options/leveldb_options.cc",
      "utilities/lua/rocks_lua_compaction_filter.cc",
      "utilities/memory/memory_util.cc",
+      "utilities/merge_operators/cassandra/format.cc",
+      "utilities/merge_operators/cassandra/merge_operator.cc",
      "utilities/merge_operators/max.cc",
      "utilities/merge_operators/put.cc",
      "utilities/merge_operators/string_append/stringappend.cc",
@@ -244,6 +257,7 @@ cpp_library(
      "tools/ldb_cmd.cc",
      "tools/ldb_tool.cc",
      "tools/sst_dump_tool.cc",
+      "utilities/blob_db/blob_dump_tool.cc",
    ],
    deps = [],
    preprocessor_flags = rocksdb_preprocessor_flags,
@@ -255,12 +269,12 @@ cpp_library(
    name = "rocksdb_test_lib",
    headers = AutoHeaders.RECURSIVE_GLOB,
    srcs = [
-      "env/mock_env.cc",
      "table/mock_table.cc",
      "util/fault_injection_test_env.cc",
      "util/testharness.cc",
      "util/testutil.cc",
      "db/db_test_util.cc",
+      "utilities/merge_operators/cassandra/test_utils.cc",
      "utilities/col_buf_encoder.cc",
      "utilities/col_buf_decoder.cc",
      "utilities/column_aware_encoding_util.cc",
@@ -295,162 +309,182 @@ cpp_library(
 )

 # [test_name, test_src, test_type]
-ROCKS_TESTS = [['merger_test', 'table/merger_test.cc', 'serial'],
- ['cache_test', 'cache/cache_test.cc', 'serial'],
- ['options_file_test', 'db/options_file_test.cc', 'serial'],
- ['compaction_picker_test', 'db/compaction_picker_test.cc', 'serial'],
- ['corruption_test', 'db/corruption_test.cc', 'serial'],
- ['thread_list_test', 'util/thread_list_test.cc', 'serial'],
- ['table_properties_collector_test',
-  'db/table_properties_collector_test.cc',
-  'serial'],
- ['document_db_test', 'utilities/document/document_db_test.cc', 'serial'],
- ['event_logger_test', 'util/event_logger_test.cc', 'serial'],
- ['coding_test', 'util/coding_test.cc', 'serial'],
- ['statistics_test', 'monitoring/statistics_test.cc', 'serial'],
- ['options_settable_test', 'options/options_settable_test.cc', 'serial'],
- ['sst_dump_test', 'tools/sst_dump_test.cc', 'serial'],
- ['column_aware_encoding_test',
-  'utilities/column_aware_encoding_test.cc',
-  'serial'],
- ['db_iterator_test', 'db/db_iterator_test.cc', 'serial'],
- ['db_sst_test', 'db/db_sst_test.cc', 'parallel'],
- ['geodb_test', 'utilities/geodb/geodb_test.cc', 'serial'],
- ['listener_test', 'db/listener_test.cc', 'serial'],
- ['write_callback_test', 'db/write_callback_test.cc', 'serial'],
- ['version_set_test', 'db/version_set_test.cc', 'serial'],
- ['full_filter_block_test', 'table/full_filter_block_test.cc', 'serial'],
- ['cleanable_test', 'table/cleanable_test.cc', 'serial'],
- ['checkpoint_test', 'utilities/checkpoint/checkpoint_test.cc', 'serial'],
- ['compact_files_test', 'db/compact_files_test.cc', 'serial'],
- ['db_options_test', 'db/db_options_test.cc', 'serial'],
- ['object_registry_test', 'utilities/object_registry_test.cc', 'serial'],
+ROCKS_TESTS = [['arena_test', 'util/arena_test.cc', 'serial'],
 ['auto_roll_logger_test', 'util/auto_roll_logger_test.cc', 'serial'],
- ['dbformat_test', 'db/dbformat_test.cc', 'serial'],
- ['write_batch_with_index_test',
-  'utilities/write_batch_with_index/write_batch_with_index_test.cc',
-  'serial'],
- ['json_document_test', 'utilities/document/json_document_test.cc', 'serial'],
- ['file_reader_writer_test', 'util/file_reader_writer_test.cc', 'serial'],
- ['repair_test', 'db/repair_test.cc', 'serial'],
- ['persistent_cache_test',
-  'utilities/persistent_cache/persistent_cache_test.cc',
-  'parallel'],
- ['db_bloom_filter_test', 'db/db_bloom_filter_test.cc', 'serial'],
- ['external_sst_file_basic_test',
-  'db/external_sst_file_basic_test.cc',
-  'serial'],
- ['options_test', 'options/options_test.cc', 'serial'],
- ['perf_context_test', 'db/perf_context_test.cc', 'serial'],
- ['db_block_cache_test', 'db/db_block_cache_test.cc', 'serial'],
- ['heap_test', 'util/heap_test.cc', 'serial'],
- ['db_test2', 'db/db_test2.cc', 'serial'],
- ['filelock_test', 'util/filelock_test.cc', 'serial'],
- ['write_controller_test', 'db/write_controller_test.cc', 'serial'],
- ['compaction_iterator_test', 'db/compaction_iterator_test.cc', 'serial'],
- ['spatial_db_test', 'utilities/spatialdb/spatial_db_test.cc', 'serial'],
- ['c_test', 'db/c_test.c', 'serial'],
- ['range_del_aggregator_test', 'db/range_del_aggregator_test.cc', 'serial'],
- ['date_tiered_test', 'utilities/date_tiered/date_tiered_test.cc', 'serial'],
- ['ldb_cmd_test', 'tools/ldb_cmd_test.cc', 'serial'],
- ['db_test', 'db/db_test.cc', 'parallel'],
- ['block_based_filter_block_test',
-  'table/block_based_filter_block_test.cc',
-  'serial'],
- ['merge_test', 'db/merge_test.cc', 'serial'],
- ['bloom_test', 'util/bloom_test.cc', 'serial'],
- ['block_test', 'table/block_test.cc', 'serial'],
- ['cuckoo_table_builder_test', 'table/cuckoo_table_builder_test.cc', 'serial'],
+ ['autovector_test', 'util/autovector_test.cc', 'serial'],
 ['backupable_db_test',
  'utilities/backupable/backupable_db_test.cc',
  'parallel'],
- ['db_flush_test', 'db/db_flush_test.cc', 'serial'],
- ['filename_test', 'db/filename_test.cc', 'serial'],
- ['cuckoo_table_reader_test', 'table/cuckoo_table_reader_test.cc', 'serial'],
- ['slice_transform_test', 'util/slice_transform_test.cc', 'serial'],
- ['cuckoo_table_db_test', 'db/cuckoo_table_db_test.cc', 'serial'],
- ['inlineskiplist_test', 'memtable/inlineskiplist_test.cc', 'parallel'],
- ['optimistic_transaction_test',
-  'utilities/transactions/optimistic_transaction_test.cc',
+ ['blob_db_test', 'utilities/blob_db/blob_db_test.cc', 'serial'],
+ ['block_based_filter_block_test',
+  'table/block_based_filter_block_test.cc',
  'serial'],
- ['hash_table_test',
-  'utilities/persistent_cache/hash_table_test.cc',
+ ['block_test', 'table/block_test.cc', 'serial'],
+ ['bloom_test', 'util/bloom_test.cc', 'serial'],
+ ['c_test', 'db/c_test.c', 'serial'],
+ ['cache_test', 'cache/cache_test.cc', 'serial'],
+ ['cassandra_format_test',
+  'utilities/merge_operators/cassandra/cassandra_format_test.cc',
  'serial'],
- ['db_dynamic_level_test', 'db/db_dynamic_level_test.cc', 'serial'],
- ['option_change_migration_test',
-  'utilities/option_change_migration/option_change_migration_test.cc',
+ ['cassandra_merge_test',
+  'utilities/merge_operators/cassandra/cassandra_merge_test.cc',
  'serial'],
- ['db_inplace_update_test', 'db/db_inplace_update_test.cc', 'serial'],
- ['autovector_test', 'util/autovector_test.cc', 'serial'],
- ['db_iter_test', 'db/db_iter_test.cc', 'serial'],
- ['flush_job_test', 'db/flush_job_test.cc', 'serial'],
- ['wal_manager_test', 'db/wal_manager_test.cc', 'serial'],
- ['write_batch_test', 'db/write_batch_test.cc', 'serial'],
- ['crc32c_test', 'util/crc32c_test.cc', 'serial'],
- ['rate_limiter_test', 'util/rate_limiter_test.cc', 'serial'],
- ['external_sst_file_test', 'db/external_sst_file_test.cc', 'parallel'],
- ['compaction_job_test', 'db/compaction_job_test.cc', 'serial'],
- ['mock_env_test', 'env/mock_env_test.cc', 'serial'],
- ['db_table_properties_test', 'db/db_table_properties_test.cc', 'serial'],
- ['db_compaction_test', 'db/db_compaction_test.cc', 'parallel'],
- ['arena_test', 'util/arena_test.cc', 'serial'],
- ['stringappend_test',
-  'utilities/merge_operators/string_append/stringappend_test.cc',
+ ['cassandra_row_merge_test',
+  'utilities/merge_operators/cassandra/cassandra_row_merge_test.cc',
  'serial'],
- ['reduce_levels_test', 'tools/reduce_levels_test.cc', 'serial'],
- ['prefix_test', 'db/prefix_test.cc', 'serial'],
- ['ttl_test', 'utilities/ttl/ttl_test.cc', 'serial'],
- ['merge_helper_test', 'db/merge_helper_test.cc', 'serial'],
- ['file_indexer_test', 'db/file_indexer_test.cc', 'serial'],
- ['memory_test', 'utilities/memory/memory_test.cc', 'serial'],
- ['log_test', 'db/log_test.cc', 'serial'],
- ['env_timed_test', 'utilities/env_timed_test.cc', 'serial'],
- ['deletefile_test', 'db/deletefile_test.cc', 'serial'],
- ['partitioned_filter_block_test',
-  'table/partitioned_filter_block_test.cc',
+ ['cassandra_serialize_test',
+  'utilities/merge_operators/cassandra/cassandra_serialize_test.cc',
  'serial'],
- ['comparator_db_test', 'db/comparator_db_test.cc', 'serial'],
- ['compaction_job_stats_test', 'db/compaction_job_stats_test.cc', 'serial'],
- ['thread_local_test', 'util/thread_local_test.cc', 'serial'],
- ['version_builder_test', 'db/version_builder_test.cc', 'serial'],
- ['db_range_del_test', 'db/db_range_del_test.cc', 'serial'],
- ['table_test', 'table/table_test.cc', 'parallel'],
- ['db_tailing_iter_test', 'db/db_tailing_iter_test.cc', 'serial'],
- ['db_compaction_filter_test', 'db/db_compaction_filter_test.cc', 'parallel'],
- ['options_util_test', 'utilities/options/options_util_test.cc', 'serial'],
- ['dynamic_bloom_test', 'util/dynamic_bloom_test.cc', 'serial'],
- ['db_basic_test', 'db/db_basic_test.cc', 'serial'],
- ['db_merge_operator_test', 'db/db_merge_operator_test.cc', 'serial'],
- ['manual_compaction_test', 'db/manual_compaction_test.cc', 'parallel'],
- ['delete_scheduler_test', 'util/delete_scheduler_test.cc', 'serial'],
- ['transaction_test', 'utilities/transactions/transaction_test.cc', 'serial'],
- ['db_io_failure_test', 'db/db_io_failure_test.cc', 'serial'],
- ['db_log_iter_test', 'db/db_log_iter_test.cc', 'serial'],
+ ['checkpoint_test', 'utilities/checkpoint/checkpoint_test.cc', 'serial'],
+ ['cleanable_test', 'table/cleanable_test.cc', 'serial'],
+ ['coding_test', 'util/coding_test.cc', 'serial'],
+ ['column_aware_encoding_test',
+  'utilities/column_aware_encoding_test.cc',
+  'serial'],
+ ['column_family_test', 'db/column_family_test.cc', 'serial'],
+ ['compact_files_test', 'db/compact_files_test.cc', 'serial'],
 ['compact_on_deletion_collector_test',
  'utilities/table_properties_collectors/compact_on_deletion_collector_test.cc',
  'serial'],
- ['env_test', 'env/env_test.cc', 'serial'],
- ['db_wal_test', 'db/db_wal_test.cc', 'parallel'],
- ['sim_cache_test', 'utilities/simulator_cache/sim_cache_test.cc', 'serial'],
+ ['compaction_iterator_test', 'db/compaction_iterator_test.cc', 'serial'],
+ ['compaction_job_stats_test', 'db/compaction_job_stats_test.cc', 'serial'],
+ ['compaction_job_test', 'db/compaction_job_test.cc', 'serial'],
+ ['compaction_picker_test', 'db/compaction_picker_test.cc', 'serial'],
+ ['comparator_db_test', 'db/comparator_db_test.cc', 'serial'],
+ ['corruption_test', 'db/corruption_test.cc', 'serial'],
+ ['crc32c_test', 'util/crc32c_test.cc', 'serial'],
+ ['cuckoo_table_builder_test', 'table/cuckoo_table_builder_test.cc', 'serial'],
+ ['cuckoo_table_db_test', 'db/cuckoo_table_db_test.cc', 'serial'],
+ ['cuckoo_table_reader_test', 'table/cuckoo_table_reader_test.cc', 'serial'],
+ ['date_tiered_test', 'utilities/date_tiered/date_tiered_test.cc', 'serial'],
+ ['db_basic_test', 'db/db_basic_test.cc', 'serial'],
+ ['db_block_cache_test', 'db/db_block_cache_test.cc', 'serial'],
+ ['db_bloom_filter_test', 'db/db_bloom_filter_test.cc', 'serial'],
+ ['db_compaction_filter_test', 'db/db_compaction_filter_test.cc', 'parallel'],
+ ['db_compaction_test', 'db/db_compaction_test.cc', 'parallel'],
+ ['db_dynamic_level_test', 'db/db_dynamic_level_test.cc', 'serial'],
+ ['db_encryption_test', 'db/db_encryption_test.cc', 'serial'],
+ ['db_flush_test', 'db/db_flush_test.cc', 'serial'],
+ ['db_inplace_update_test', 'db/db_inplace_update_test.cc', 'serial'],
+ ['db_io_failure_test', 'db/db_io_failure_test.cc', 'serial'],
+ ['db_iter_test', 'db/db_iter_test.cc', 'serial'],
+ ['db_iterator_test', 'db/db_iterator_test.cc', 'serial'],
+ ['db_log_iter_test', 'db/db_log_iter_test.cc', 'serial'],
 ['db_memtable_test', 'db/db_memtable_test.cc', 'serial'],
+ ['db_merge_operator_test', 'db/db_merge_operator_test.cc', 'serial'],
+ ['db_options_test', 'db/db_options_test.cc', 'serial'],
+ ['db_properties_test', 'db/db_properties_test.cc', 'serial'],
+ ['db_range_del_test', 'db/db_range_del_test.cc', 'serial'],
+ ['db_sst_test', 'db/db_sst_test.cc', 'parallel'],
+ ['db_statistics_test', 'db/db_statistics_test.cc', 'serial'],
+ ['db_table_properties_test', 'db/db_table_properties_test.cc', 'serial'],
+ ['db_tailing_iter_test', 'db/db_tailing_iter_test.cc', 'serial'],
+ ['db_test', 'db/db_test.cc', 'parallel'],
+ ['db_test2', 'db/db_test2.cc', 'serial'],
 ['db_universal_compaction_test',
  'db/db_universal_compaction_test.cc',
  'parallel'],
+ ['db_wal_test', 'db/db_wal_test.cc', 'parallel'],
+ ['db_write_test', 'db/db_write_test.cc', 'serial'],
+ ['dbformat_test', 'db/dbformat_test.cc', 'serial'],
+ ['delete_scheduler_test', 'util/delete_scheduler_test.cc', 'serial'],
+ ['deletefile_test', 'db/deletefile_test.cc', 'serial'],
+ ['document_db_test', 'utilities/document/document_db_test.cc', 'serial'],
+ ['dynamic_bloom_test', 'util/dynamic_bloom_test.cc', 'serial'],
+ ['env_basic_test', 'env/env_basic_test.cc', 'serial'],
+ ['env_test', 'env/env_test.cc', 'serial'],
+ ['env_timed_test', 'utilities/env_timed_test.cc', 'serial'],
+ ['event_logger_test', 'util/event_logger_test.cc', 'serial'],
+ ['external_sst_file_basic_test',
+  'db/external_sst_file_basic_test.cc',
+  'serial'],
+ ['external_sst_file_test', 'db/external_sst_file_test.cc', 'parallel'],
+ ['fault_injection_test', 'db/fault_injection_test.cc', 'parallel'],
+ ['file_indexer_test', 'db/file_indexer_test.cc', 'serial'],
+ ['file_reader_writer_test', 'util/file_reader_writer_test.cc', 'serial'],
+ ['filelock_test', 'util/filelock_test.cc', 'serial'],
+ ['filename_test', 'db/filename_test.cc', 'serial'],
+ ['flush_job_test', 'db/flush_job_test.cc', 'serial'],
+ ['full_filter_block_test', 'table/full_filter_block_test.cc', 'serial'],
+ ['geodb_test', 'utilities/geodb/geodb_test.cc', 'serial'],
+ ['hash_table_test',
+  'utilities/persistent_cache/hash_table_test.cc',
+  'serial'],
+ ['hash_test', 'util/hash_test.cc', 'serial'],
+ ['heap_test', 'util/heap_test.cc', 'serial'],
 ['histogram_test', 'monitoring/histogram_test.cc', 'serial'],
+ ['inlineskiplist_test', 'memtable/inlineskiplist_test.cc', 'parallel'],
+ ['iostats_context_test', 'monitoring/iostats_context_test.cc', 'serial'],
+ ['json_document_test', 'utilities/document/json_document_test.cc', 'serial'],
+ ['ldb_cmd_test', 'tools/ldb_cmd_test.cc', 'serial'],
+ ['listener_test', 'db/listener_test.cc', 'serial'],
+ ['log_test', 'db/log_test.cc', 'serial'],
+ ['lru_cache_test', 'cache/lru_cache_test.cc', 'serial'],
+ ['manual_compaction_test', 'db/manual_compaction_test.cc', 'parallel'],
+ ['memory_test', 'utilities/memory/memory_test.cc', 'serial'],
+ ['memtable_list_test', 'db/memtable_list_test.cc', 'serial'],
+ ['merge_helper_test', 'db/merge_helper_test.cc', 'serial'],
+ ['merge_test', 'db/merge_test.cc', 'serial'],
+ ['merger_test', 'table/merger_test.cc', 'serial'],
+ ['mock_env_test', 'env/mock_env_test.cc', 'serial'],
+ ['object_registry_test', 'utilities/object_registry_test.cc', 'serial'],
+ ['optimistic_transaction_test',
+  'utilities/transactions/optimistic_transaction_test.cc',
+  'serial'],
+ ['option_change_migration_test',
+  'utilities/option_change_migration/option_change_migration_test.cc',
+  'serial'],
+ ['options_file_test', 'db/options_file_test.cc', 'serial'],
+ ['options_settable_test', 'options/options_settable_test.cc', 'serial'],
+ ['options_test', 'options/options_test.cc', 'serial'],
+ ['options_util_test', 'utilities/options/options_util_test.cc', 'serial'],
+ ['partitioned_filter_block_test',
+  'table/partitioned_filter_block_test.cc',
+  'serial'],
+ ['perf_context_test', 'db/perf_context_test.cc', 'serial'],
+ ['persistent_cache_test',
+  'utilities/persistent_cache/persistent_cache_test.cc',
+  'parallel'],
+ ['plain_table_db_test', 'db/plain_table_db_test.cc', 'serial'],
+ ['prefix_test', 'db/prefix_test.cc', 'serial'],
+ ['range_del_aggregator_test', 'db/range_del_aggregator_test.cc', 'serial'],
+ ['rate_limiter_test', 'util/rate_limiter_test.cc', 'serial'],
+ ['reduce_levels_test', 'tools/reduce_levels_test.cc', 'serial'],
+ ['repair_test', 'db/repair_test.cc', 'serial'],
+ ['sim_cache_test', 'utilities/simulator_cache/sim_cache_test.cc', 'serial'],
+ ['skiplist_test', 'memtable/skiplist_test.cc', 'serial'],
+ ['slice_transform_test', 'util/slice_transform_test.cc', 'serial'],
+ ['spatial_db_test', 'utilities/spatialdb/spatial_db_test.cc', 'serial'],
+ ['sst_dump_test', 'tools/sst_dump_test.cc', 'serial'],
+ ['statistics_test', 'monitoring/statistics_test.cc', 'serial'],
+ ['stringappend_test',
+  'utilities/merge_operators/string_append/stringappend_test.cc',
+  'serial'],
+ ['table_properties_collector_test',
+  'db/table_properties_collector_test.cc',
+  'serial'],
+ ['table_test', 'table/table_test.cc', 'parallel'],
+ ['thread_list_test', 'util/thread_list_test.cc', 'serial'],
+ ['thread_local_test', 'util/thread_local_test.cc', 'serial'],
+ ['timer_queue_test', 'util/timer_queue_test.cc', 'serial'],
+ ['transaction_test', 'utilities/transactions/transaction_test.cc', 'serial'],
+ ['ttl_test', 'utilities/ttl/ttl_test.cc', 'serial'],
 ['util_merge_operators_test',
  'utilities/util_merge_operators_test.cc',
  'serial'],
- ['fault_injection_test', 'db/fault_injection_test.cc', 'parallel'],
- ['env_basic_test', 'env/env_basic_test.cc', 'serial'],
- ['iostats_context_test', 'monitoring/iostats_context_test.cc', 'serial'],
- ['memtable_list_test', 'db/memtable_list_test.cc', 'serial'],
- ['column_family_test', 'db/column_family_test.cc', 'serial'],
- ['db_properties_test', 'db/db_properties_test.cc', 'serial'],
+ ['version_builder_test', 'db/version_builder_test.cc', 'serial'],
 ['version_edit_test', 'db/version_edit_test.cc', 'serial'],
- ['skiplist_test', 'memtable/skiplist_test.cc', 'serial'],
- ['lru_cache_test', 'cache/lru_cache_test.cc', 'serial'],
- ['plain_table_db_test', 'db/plain_table_db_test.cc', 'serial'],
- ['db_statistics_test', 'db/db_statistics_test.cc', 'serial']]
+ ['version_set_test', 'db/version_set_test.cc', 'serial'],
+ ['wal_manager_test', 'db/wal_manager_test.cc', 'serial'],
+ ['write_batch_test', 'db/write_batch_test.cc', 'serial'],
+ ['write_batch_with_index_test',
+  'utilities/write_batch_with_index/write_batch_with_index_test.cc',
+  'serial'],
+ ['write_buffer_manager_test',
+  'memtable/write_buffer_manager_test.cc',
+  'serial'],
+ ['write_callback_test', 'db/write_callback_test.cc', 'serial'],
+ ['write_controller_test', 'db/write_controller_test.cc', 'serial']]


 # Generate a test rule for each entry in ROCKS_TESTS
@@ -475,3 +509,15 @@ for test_cfg in ROCKS_TESTS:
      deps = [":" + test_bin],
      command = [TEST_RUNNER, BUCK_BINS + test_bin]
    )
+
+custom_unittest(
+    name = "make_rocksdbjavastatic",
+    type = "simple",
+    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
+)
+
+custom_unittest(
+    name = "make_rocksdb_lite_release",
+    type = "simple",
+    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
+)
@@ -0,0 +1,2 @@
+arcanist_util are only used internaly, If you want to change it please check
+<internal_rocksdb_repo>/arcanist_util
@@ -84,7 +84,7 @@ def get_tests(repo_path):
            else:
                # we consumed all the parallel tests
                break
-
+    
    return tests


@@ -122,7 +122,7 @@ def generate_targets(repo_path):
        [":rocksdb_lib"])

    # test for every test we found in the Makefile
-    for test in tests:
+    for test in sorted(tests):
        match_src = [src for src in cc_files if ("/%s.c" % test) in src]
        if len(match_src) == 0:
            print(ColorString.warning("Cannot find .cc file for %s" % test))
@@ -2,7 +2,11 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
-rocksdb_target_header = """REPO_PATH = "internal_repo_rocksdb/repo/"
+rocksdb_target_header = """
+import os
+
+TARGETS_PATH = os.path.dirname(__file__)
+REPO_PATH = TARGETS_PATH[(TARGETS_PATH.find('fbcode/') + len('fbcode/')):] + "/"
 BUCK_BINS = "buck-out/gen/" + REPO_PATH
 TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
 rocksdb_compiler_flags = [
@@ -13,6 +17,10 @@ rocksdb_compiler_flags = [
  "-DROCKSDB_LIB_IO_POSIX",
  "-DROCKSDB_FALLOCATE_PRESENT",
  "-DROCKSDB_MALLOC_USABLE_SIZE",
+  "-DROCKSDB_RANGESYNC_PRESENT",
+  "-DROCKSDB_SCHED_GETCPU_PRESENT",
+  "-DROCKSDB_SUPPORT_THREAD_LOCAL",
+  "-DHAVE_SSE42",
  "-DOS_LINUX",
  # Flags to enable libs we include
  "-DSNAPPY",
@@ -35,7 +43,7 @@ rocksdb_external_deps = [
  ('lz4', None, 'lz4'),
  ('zstd', None),
  ('tbb', None),
-  ("numa", "2.0.8", "numa"),
+  ("numa", None, "numa"),
  ("googletest", None, "gtest"),
 ]

@@ -51,7 +51,14 @@ if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
    FBCODE_BUILD="true"
    # If we're compiling with TSAN we need pic build
    PIC_BUILD=$COMPILE_WITH_TSAN
-    source "$PWD/build_tools/fbcode_config.sh"
+    if [ -n "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
+      # we need this to build with MySQL. Don't use for other purposes.
+      source "$PWD/build_tools/fbcode_config4.8.1.sh"
+    elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_5xx" ]; then
+      source "$PWD/build_tools/fbcode_config.sh"
+    else
+      source "$PWD/build_tools/fbcode_config_platform007.sh"
+    fi
 fi

 # Delete existing output, if it exists
@@ -123,10 +130,17 @@ case "$TARGET_OS" in
        ;;
    SunOS)
        PLATFORM=OS_SOLARIS
-        COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_SOLARIS"
-        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
+        COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_SOLARIS -m64"
+        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt -static-libstdc++ -static-libgcc -m64"
        # PORT_FILES=port/sunos/sunos_specific.cc
        ;;
+    AIX)
+        PLATFORM=OS_AIX
+        CC=gcc
+        COMMON_FLAGS="$COMMON_FLAGS -maix64 -pthread -fno-builtin-memcmp -D_REENTRANT -DOS_AIX -D__STDC_FORMAT_MACROS"
+        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -pthread -lpthread -lrt -maix64 -static-libstdc++ -static-libgcc"
+        # PORT_FILES=port/aix/aix_specific.cc
+        ;;
    FreeBSD)
        PLATFORM=OS_FREEBSD
        COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_FREEBSD"
@@ -377,6 +391,29 @@ EOF
    if [ "$?" = 0 ]; then
        PROFILING_FLAGS=-pg
    fi
+
+    # Test whether sync_file_range is supported for compatibility with an old glibc
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <fcntl.h>
+      int main() {
+        int fd = open("/dev/null", 0);
+        sync_file_range(fd, 0, 1024, SYNC_FILE_RANGE_WRITE);
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_RANGESYNC_PRESENT"
+    fi
+
+    # Test whether sched_getcpu is supported
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <sched.h>
+      int main() {
+        int cpuid = sched_getcpu();
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SCHED_GETCPU_PRESENT"
+    fi
 fi

 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -407,14 +444,8 @@ if test "$USE_HDFS"; then
  JAVA_LDFLAGS="$JAVA_LDFLAGS $HDFS_LDFLAGS"
 fi

-if [ "$TARGET_OS" = FreeBSD -a "$TARGET_ARCHITECTURE" = i386 ]; then
-  # Intel SSE instructions breaks compilation on FreeBSD i386
-  unset USE_SSE
-fi
-
 if test "$USE_SSE"; then
-  # if Intel SSE instruction set is supported, set USE_SSE=1
-  COMMON_FLAGS="$COMMON_FLAGS -msse -msse4.2 "
+  COMMON_FLAGS="$COMMON_FLAGS -msse4.2"
 elif test -z "$PORTABLE"; then
  if test -n "`echo $TARGET_ARCHITECTURE | grep ^ppc64`"; then
    # Tune for this POWER processor, treating '+' models as base models
@@ -422,11 +453,41 @@ elif test -z "$PORTABLE"; then
    COMMON_FLAGS="$COMMON_FLAGS -mcpu=$POWER -mtune=$POWER "
  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^s390x`"; then
    COMMON_FLAGS="$COMMON_FLAGS -march=z10 "
-  else
+  elif [ "$TARGET_OS" != AIX ] && [ "$TARGET_OS" != SunOS ]; then
    COMMON_FLAGS="$COMMON_FLAGS -march=native "
  fi
 fi

+$CXX $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
+  #include <cstdint>
+  #include <nmmintrin.h>
+  int main() {
+    volatile uint32_t x = _mm_crc32_u32(0, 0);
+  }
+EOF
+if [ "$?" = 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"
+fi
+
+# iOS doesn't support thread-local storage, but this check would erroneously
+# succeed because the cross-compiler flags are added by the Makefile, not this
+# script.
+if [ "$PLATFORM" != IOS ]; then
+  $CXX $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
+  #if defined(_MSC_VER) && !defined(__thread)
+  #define __thread __declspec(thread)
+  #endif
+  int main() {
+    static __thread int tls;
+  }
+EOF
+  if [ "$?" = 0 ]; then
+    COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SUPPORT_THREAD_LOCAL"
+  fi
+fi
+
 PLATFORM_CCFLAGS="$PLATFORM_CCFLAGS $COMMON_FLAGS"
 PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS $COMMON_FLAGS"

@@ -1,19 +1,18 @@
-# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-GCC_BASE=/mnt/gvfs/third-party2/gcc/7331085db891a2ef4a88a48a751d834e8d68f4cb/7.x/centos7-native/b2ef2b6
-CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/963d9aeda70cc4779885b1277484fe7544a04e3e/9.0.0/platform007/9e92d53/
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/6ace84e956873d53638c738b6f65f3f469cca74c/7.x/platform007/5620abc
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/192b0f42d63dcf6210d6ceae387b49af049e6e0c/2.26/platform007/f259413
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2d9f0b9a4274cc21f61272a9e89bdb859bce8f1f/1.2.8/platform007/ca4da3d
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/0f607f8fc442ea7d6b876931b1898bb573d5e5da/1.9.1/platform007/ca4da3d
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/ca22bc441a4eb709e9e0b1f9fec9750fed7b31c5/1.4.x/platform007/15a3614
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/c26f08f47ac35fc31da2633b7da92d6b863246eb/master/platform007/c26c002
-NUMA_BASE=/mnt/gvfs/third-party2/numa/3f3fb57a5ccc5fd21c66416c0b83e0aa76a05376/2.0.11/platform007/ca4da3d
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/40c73d874898b386a71847f1b99115d93822d11f/1.4/platform007/6f3e0a9
-TBB_BASE=/mnt/gvfs/third-party2/tbb/4ce8e8dba77cdbd81b75d6f0c32fd7a1b76a11ec/2018_U5/platform007/ca4da3d
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/fb251ecd2f5ae16f8671f7014c246e52a748fe0b/fb/platform007/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/ab9f09bba370e7066cafd4eb59752db93f2e8312/2.29.1/platform007/15a3614
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d42d152a15636529b0861ec493927200ebebca8e/3.15.0/platform007/ca4da3d
-LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832
+GCC_BASE=/mnt/gvfs/third-party2/gcc/2928bb3ed95bf64f5b388ee88c30dc74710c3b35/5.x/centos6-native/f4950a1
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/a5fea028cb7ba43498976e1f8054b0b2e790c295/stable/centos6-native/6aaf4de
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/7a9099f6587ee4378c0b1fa32bb8934019d30ca4/5.x/gcc-5-glibc-2.23/339d858
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/3b7c6469854dfc7832a1c3cc5b86919a84e5f865/2.23/gcc-5-glibc-2.23/ca1d1c0
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-5-glibc-2.23/9bc6787
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/d7861abe6f0e27ab98c9303b95a662f0e4cdedb5/1.2.8/gcc-5-glibc-2.23/9bc6787
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/740325875f6729f42d28deaa2147b0854f3a347e/1.0.6/gcc-5-glibc-2.23/9bc6787
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0815d59804160c96caac5f27ca004f51af893dc6/r131/gcc-5-glibc-2.23/9bc6787
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/c15a4f5f619a2930478d01e2e34dc1e0652b0873/1.1.4/gcc-5-glibc-2.23/03859b5
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/f905a5e1032fb30c05db3d3752319857388c0c49/2.2.0/gcc-5-glibc-2.23/9bc6787
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/8d60633d822a2a55849c73db24e74a25e52b71db/master/gcc-5-glibc-2.23/1c32b4b
+NUMA_BASE=/mnt/gvfs/third-party2/numa/17c514c4d102a25ca15f4558be564eeed76f4b6a/2.0.8/gcc-5-glibc-2.23/9bc6787
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/8db74270cd6d0212ac92d69e7fc7beefe617d772/trunk/gcc-5-glibc-2.23/b1847cb
+TBB_BASE=/mnt/gvfs/third-party2/tbb/9d9a554877d0c5bef330fe818ab7178806dd316a/4.0_update2/gcc-5-glibc-2.23/9bc6787
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/90c9734afc5579c9d1db529fa788d09f97763b85/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/9e829389ef61b92c62de8748c80169aaf25ce1f0/2.26.1/centos6-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d7f4d4d86674a57668e3a96f76f0e17dd0eb8765/3.11.0/gcc-5-glibc-2.23/9bc6787
+LUA_BASE=/mnt/gvfs/third-party2/lua/61e4abf5813bbc39bc4f548757ccfcadde175a48/5.2.3/gcc-5-glibc-2.23/65372bd
@@ -0,0 +1,18 @@
+GCC_BASE=/mnt/gvfs/third-party2/gcc/6e8e715624fd15256a7970073387793dfcf79b46/7.x/centos7-native/b2ef2b6
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/ef37e1faa1c29782abfac1ae65a291b9b7966f6d/stable/centos7-native/c9f9104
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/c67031f0f739ac61575a061518d6ef5038f99f90/7.x/platform007/5620abc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/60d6f124a78798b73944f5ba87c2306ae3460153/2.26/platform007/f259413
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/22c2d65676fb7c23cfa797c4f6937f38b026f3cf/1.2.8/platform007/ca4da3d
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/907b498203d297947f3bb70b9466f47e100f1873/r131/platform007/ca4da3d
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/3ee276cbacfad3074e3f07bf826ac47f06970f4e/1.3.5/platform007/15a3614
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/9c910d36d6235cc40e8ff559358f1833452300ca/master/platform007/5b0f53e
+NUMA_BASE=/mnt/gvfs/third-party2/numa/9cbf2460284c669ed19c3ccb200a71f7dd7e53c7/2.0.11/platform007/ca4da3d
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/bf3d7497fe4e6d007354f0adffa16ce3003f8338/1.3/platform007/6f3e0a9
+TBB_BASE=/mnt/gvfs/third-party2/tbb/ff4e0b093534704d8abab678a4fd7f5ea7b094c7/2018_U5/platform007/ca4da3d
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/b5c4a61a5c483ba24722005ae07895971a2ac707/fb/platform007/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/92ff90349e2f43ea0a8246d8b1cf17b6869013e3/2.29.1/centos7-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/f3f697a28122e6bcd513273dd9c1ff23852fc59f/3.13.0/platform007/ca4da3d
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832
@@ -13,7 +13,7 @@ source "$BASEDIR/dependencies.sh"
 CFLAGS=""

 # libgcc
-LIBGCC_INCLUDE="$LIBGCC_BASE/include/c++/7.3.0"
+LIBGCC_INCLUDE="$LIBGCC_BASE/include"
 LIBGCC_LIBS=" -L $LIBGCC_BASE/lib"

 # glibc
@@ -43,15 +43,11 @@ if test -z $PIC_BUILD; then
  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
  CFLAGS+=" -DLZ4"
-fi

-ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
-if test -z $PIC_BUILD; then
+  ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
-else
-  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
+  CFLAGS+=" -DZSTD"
 fi
-CFLAGS+=" -DZSTD"

 # location of gflags headers and libraries
 GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
@@ -108,8 +104,8 @@ if [ -z "$USE_CLANG" ]; then
  CXX="$GCC_BASE/bin/g++"

  CFLAGS+=" -B$BINUTILS/gold"
-  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  JEMALLOC=1
 else
  # clang
@@ -120,8 +116,8 @@ else
  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include"

  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x "
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x/x86_64-facebook-linux "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/5.x "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/5.x/x86_64-facebook-linux "
  CFLAGS+=" -isystem $GLIBC_INCLUDE"
  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  CFLAGS+=" -isystem $CLANG_INCLUDE"
@@ -137,9 +133,9 @@ CXXFLAGS+=" $CFLAGS"

 EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
 EXEC_LDFLAGS+=" -B$BINUTILS/gold"
-EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/gcc-5-glibc-2.23/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
-EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/gcc-5-glibc-2.23/lib"
 # required by libtbb
 EXEC_LDFLAGS+=" -ldl"

@@ -149,4 +145,12 @@ EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GF

 VALGRIND_VER="$VALGRIND_BASE/bin/"

-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD 
+LUA_PATH="$LUA_BASE"
+
+if test -z $PIC_BUILD; then
+  LUA_LIB=" $LUA_PATH/lib/liblua.a"
+else
+  LUA_LIB=" $LUA_PATH/lib/liblua_pic.a"
+fi
+
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
@@ -93,7 +93,7 @@ else
 fi

 CFLAGS+=" $DEPS_INCLUDE"
-CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
 CFLAGS+=" -DSNAPPY -DGFLAGS=google -DZLIB -DBZIP2 -DLZ4 -DZSTD -DNUMA -DTBB"
 CXXFLAGS+=" $CFLAGS"

@@ -0,0 +1,157 @@
+#!/bin/sh
+#
+# Set environment variables so that we can compile rocksdb using
+# fbcode settings.  It uses the latest g++ and clang compilers and also
+# uses jemalloc
+# Environment variables that change the behavior of this script:
+# PIC_BUILD -- if true, it will only take pic versions of libraries from fbcode. libraries that don't have pic variant will not be included
+
+
+BASEDIR=`dirname $BASH_SOURCE`
+source "$BASEDIR/dependencies_platform007.sh"
+
+CFLAGS=""
+
+# libgcc
+LIBGCC_INCLUDE="$LIBGCC_BASE/include/c++/7.3.0"
+LIBGCC_LIBS=" -L $LIBGCC_BASE/lib"
+
+# glibc
+GLIBC_INCLUDE="$GLIBC_BASE/include"
+GLIBC_LIBS=" -L $GLIBC_BASE/lib"
+
+# snappy
+SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include/"
+if test -z $PIC_BUILD; then
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy.a"
+else
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy_pic.a"
+fi
+CFLAGS+=" -DSNAPPY"
+
+if test -z $PIC_BUILD; then
+  # location of zlib headers and libraries
+  ZLIB_INCLUDE=" -I $ZLIB_BASE/include/"
+  ZLIB_LIBS=" $ZLIB_BASE/lib/libz.a"
+  CFLAGS+=" -DZLIB"
+
+  # location of bzip headers and libraries
+  BZIP_INCLUDE=" -I $BZIP2_BASE/include/"
+  BZIP_LIBS=" $BZIP2_BASE/lib/libbz2.a"
+  CFLAGS+=" -DBZIP2"
+
+  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
+  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
+  CFLAGS+=" -DLZ4"
+fi
+
+ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
+if test -z $PIC_BUILD; then
+  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
+else
+  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
+fi
+CFLAGS+=" -DZSTD"
+
+# location of gflags headers and libraries
+GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
+if test -z $PIC_BUILD; then
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags.a"
+else
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
+fi
+CFLAGS+=" -DGFLAGS=gflags"
+
+# location of jemalloc
+JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
+JEMALLOC_LIB=" $JEMALLOC_BASE/lib/libjemalloc.a"
+
+if test -z $PIC_BUILD; then
+  # location of numa
+  NUMA_INCLUDE=" -I $NUMA_BASE/include/"
+  NUMA_LIB=" $NUMA_BASE/lib/libnuma.a"
+  CFLAGS+=" -DNUMA"
+
+  # location of libunwind
+  LIBUNWIND="$LIBUNWIND_BASE/lib/libunwind.a"
+fi
+
+# location of TBB
+TBB_INCLUDE=" -isystem $TBB_BASE/include/"
+if test -z $PIC_BUILD; then
+  TBB_LIBS="$TBB_BASE/lib/libtbb.a"
+else
+  TBB_LIBS="$TBB_BASE/lib/libtbb_pic.a"
+fi
+CFLAGS+=" -DTBB"
+
+# use Intel SSE support for checksum calculations
+export USE_SSE=1
+export PORTABLE=1
+
+BINUTILS="$BINUTILS_BASE/bin"
+AR="$BINUTILS/ar"
+
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE"
+
+STDLIBS="-L $GCC_BASE/lib64"
+
+CLANG_BIN="$CLANG_BASE/bin"
+CLANG_LIB="$CLANG_BASE/lib"
+CLANG_SRC="$CLANG_BASE/../../src"
+
+CLANG_ANALYZER="$CLANG_BIN/clang++"
+CLANG_SCAN_BUILD="$CLANG_SRC/llvm/tools/clang/tools/scan-build/bin/scan-build"
+
+if [ -z "$USE_CLANG" ]; then
+  # gcc
+  CC="$GCC_BASE/bin/gcc"
+  CXX="$GCC_BASE/bin/g++"
+
+  CFLAGS+=" -B$BINUTILS/gold"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  JEMALLOC=1
+else
+  # clang
+  CLANG_INCLUDE="$CLANG_LIB/clang/stable/include"
+  CC="$CLANG_BIN/clang"
+  CXX="$CLANG_BIN/clang++"
+
+  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include"
+
+  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x/x86_64-facebook-linux "
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $CLANG_INCLUDE"
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE/linux "
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE "
+  CFLAGS+=" -Wno-expansion-to-defined "
+  CXXFLAGS="-nostdinc++"
+fi
+
+CFLAGS+=" $DEPS_INCLUDE"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
+CXXFLAGS+=" $CFLAGS"
+
+EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
+EXEC_LDFLAGS+=" -B$BINUTILS/gold"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
+EXEC_LDFLAGS+=" $LIBUNWIND"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
+# required by libtbb
+EXEC_LDFLAGS+=" -ldl"
+
+PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
+
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
+
+VALGRIND_VER="$VALGRIND_BASE/bin/"
+
+# lua not supported because it's on track for deprecation, I think
+LUA_PATH=
+LUA_LIB=
+
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
@@ -99,6 +99,18 @@ function main() {

  make static_lib
  make install INSTALL_PATH=package
+
+  cd package
+
+  LIB_DIR=lib
+  if [[ -z "$ARCH" ]]; then
+      ARCH=$(getconf LONG_BIT)
+  fi
+  if [[ ("$FPM_OUTPUT" = "rpm") && ($ARCH -eq 64) ]]; then
+      mv lib lib64
+      LIB_DIR=lib64
+  fi
+
  fpm \
    -s dir \
    -t $FPM_OUTPUT \
@@ -110,7 +122,7 @@ function main() {
    --license BSD \
    --vendor Facebook \
    --description "RocksDB is an embeddable persistent key-value store for fast storage." \
-    package
+    include $LIB_DIR
 }

 main $@
@@ -85,8 +85,9 @@ NON_SHM="TMPD=/tmp/rocksdb_test_tmp"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-LITE="OPT=\"-DROCKSDB_LITE -g\""
-TSAN="COMPILE_WITH_TSAN=1"
+# in gcc-5 there are known problems with TSAN like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71090.
+# using platform007 gives us gcc-8 or higher which has that bug fixed.
+TSAN="ROCKSDB_FBCODE_BUILD_WITH_PLATFORM007=1 COMPILE_WITH_TSAN=1"
 UBSAN="COMPILE_WITH_UBSAN=1"
 DISABLE_JEMALLOC="DISABLE_JEMALLOC=1"
 HTTP_PROXY="https_proxy=http://fwdproxy.29.prn1:8080 http_proxy=http://fwdproxy.29.prn1:8080 ftp_proxy=http://fwdproxy.29.prn1:8080"
@@ -276,7 +276,7 @@ if($RunAll) {
      continue
    }

-    if($RunOnly.Contains($e)) {
+    if($RunOnly.Contains($bare_name)) {
      MakeAndAdd -token $e -HashTable $TestExes
    } else {
      $ListOfSuites += $bare_name
@@ -333,7 +333,7 @@ function RunJobs($Suites, $TestCmds, [int]$ConcurrencyVal)
            $Cases = @{}

            if($Suites.Count -gt 0) {
-              # Will teh first one
+              # Will the first one
              ForEach($e in $Suites.Keys) {
                $exe_name = $e
                $Cases = $Suites[$e]
@@ -53,7 +53,46 @@ function get_lib_base()
 }

 ###########################################################
-#                   4.9.x dependencies                    #
+#                platform007 dependencies                 #
+###########################################################
+
+OUTPUT="$BASEDIR/dependencies_platform007.sh"
+
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+echo "Writing dependencies to $OUTPUT"
+
+# Compilers locations
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/7.x/centos7-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos7-native/*/`
+
+log_variable GCC_BASE
+log_variable CLANG_BASE
+
+# Libraries locations
+get_lib_base libgcc     7.x     platform007
+get_lib_base glibc      2.26    platform007
+get_lib_base snappy     LATEST  platform007
+get_lib_base zlib       LATEST  platform007
+get_lib_base bzip2      LATEST  platform007
+get_lib_base lz4        LATEST  platform007
+get_lib_base zstd       LATEST  platform007
+get_lib_base gflags     LATEST  platform007
+get_lib_base jemalloc   LATEST  platform007
+get_lib_base numa       LATEST  platform007
+get_lib_base libunwind  LATEST  platform007
+get_lib_base tbb        LATEST  platform007
+
+get_lib_base kernel-headers fb platform007
+get_lib_base binutils   LATEST centos7-native
+get_lib_base valgrind   LATEST platform007
+get_lib_base lua        5.3.4  platform007
+
+git diff $OUTPUT
+
+###########################################################
+#                   5.x dependencies                      #
 ###########################################################

 OUTPUT="$BASEDIR/dependencies.sh"
@@ -64,16 +103,16 @@ touch "$OUTPUT"
 echo "Writing dependencies to $OUTPUT"

 # Compilers locations
-GCC_BASE=`readlink -f $TP2_LATEST/gcc/4.9.x/centos6-native/*/`
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/5.x/centos6-native/*/`
 CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos6-native/*/`

 log_variable GCC_BASE
 log_variable CLANG_BASE

 # Libraries locations
-get_lib_base libgcc     4.9.x
-get_lib_base glibc      2.20
-get_lib_base snappy     LATEST gcc-4.9-glibc-2.20
+get_lib_base libgcc     5.x
+get_lib_base glibc      2.23
+get_lib_base snappy     LATEST gcc-5-glibc-2.23
 get_lib_base zlib       LATEST
 get_lib_base bzip2      LATEST
 get_lib_base lz4        LATEST
@@ -82,12 +121,12 @@ get_lib_base gflags     LATEST
 get_lib_base jemalloc   LATEST
 get_lib_base numa       LATEST
 get_lib_base libunwind  LATEST
-get_lib_base tbb        4.0_update2 gcc-4.9-glibc-2.20
+get_lib_base tbb        4.0_update2 gcc-5-glibc-2.23

 get_lib_base kernel-headers LATEST 
 get_lib_base binutils   LATEST centos6-native 
-get_lib_base valgrind   3.10.0 gcc-4.9-glibc-2.20
-get_lib_base lua        5.2.3 gcc-4.9-glibc-2.20
+get_lib_base valgrind   3.10.0 gcc-5-glibc-2.23
+get_lib_base lua        5.2.3 gcc-5-glibc-2.23

 git diff $OUTPUT

@@ -475,6 +475,37 @@ void deleter(const Slice& key, void* value) {
 }
 }  // namespace

+TEST_P(CacheTest, ReleaseAndErase) {
+  std::shared_ptr<Cache> cache = NewCache(5, 0, false);
+  Cache::Handle* handle;
+  Status s = cache->Insert(EncodeKey(100), EncodeValue(100), 1,
+                           &CacheTest::Deleter, &handle);
+  ASSERT_TRUE(s.ok());
+  ASSERT_EQ(5U, cache->GetCapacity());
+  ASSERT_EQ(1U, cache->GetUsage());
+  ASSERT_EQ(0U, deleted_keys_.size());
+  auto erased = cache->Release(handle, true);
+  ASSERT_TRUE(erased);
+  // This tests that deleter has been called
+  ASSERT_EQ(1U, deleted_keys_.size());
+}
+
+TEST_P(CacheTest, ReleaseWithoutErase) {
+  std::shared_ptr<Cache> cache = NewCache(5, 0, false);
+  Cache::Handle* handle;
+  Status s = cache->Insert(EncodeKey(100), EncodeValue(100), 1,
+                           &CacheTest::Deleter, &handle);
+  ASSERT_TRUE(s.ok());
+  ASSERT_EQ(5U, cache->GetCapacity());
+  ASSERT_EQ(1U, cache->GetUsage());
+  ASSERT_EQ(0U, deleted_keys_.size());
+  auto erased = cache->Release(handle);
+  ASSERT_FALSE(erased);
+  // This tests that deleter is not called. When cache has free capacity it is
+  // not expected to immediately erase the released items.
+  ASSERT_EQ(0U, deleted_keys_.size());
+}
+
 TEST_P(CacheTest, SetCapacity) {
  // test1: increase capacity
  // lets create a cache with capacity 5,
@@ -26,7 +26,6 @@ std::shared_ptr<Cache> NewClockCache(size_t capacity, int num_shard_bits,
 #include <assert.h>
 #include <atomic>
 #include <deque>
-#include <limits>

 #include "tbb/concurrent_hash_map.h"

@@ -252,8 +251,11 @@ class ClockCacheShard : public CacheShard {
  //
  // Not necessary to hold mutex_ before being called.
  virtual bool Ref(Cache::Handle* handle) override;
-  virtual void Release(Cache::Handle* handle) override;
+  virtual bool Release(Cache::Handle* handle,
+                       bool force_erase = false) override;
  virtual void Erase(const Slice& key, uint32_t hash) override;
+  bool EraseAndConfirm(const Slice& key, uint32_t hash,
+                       CleanupContext* context);
  virtual size_t GetUsage() const override;
  virtual size_t GetPinnedUsage() const override;
  virtual void EraseUnRefEntries() override;
@@ -274,13 +276,17 @@ class ClockCacheShard : public CacheShard {
  // Decrease reference count of the entry. If this decreases the count to 0,
  // recycle the entry. If set_usage is true, also set the usage bit.
  //
+  // returns true if a value is erased.
+  //
  // Not necessary to hold mutex_ before being called.
-  void Unref(CacheHandle* handle, bool set_usage, CleanupContext* context);
+  bool Unref(CacheHandle* handle, bool set_usage, CleanupContext* context);

  // Unset in-cache bit of the entry. Recycle the handle if necessary.
  //
+  // returns true if a value is erased.
+  //
  // Has to hold mutex_ before being called.
-  void UnsetInCache(CacheHandle* handle, CleanupContext* context);
+  bool UnsetInCache(CacheHandle* handle, CleanupContext* context);

  // Put the handle back to recycle_ list, and put the value associated with
  // it into to-be-deleted list. It doesn't cleanup the key as it might be
@@ -432,7 +438,7 @@ bool ClockCacheShard::Ref(Cache::Handle* h) {
  return false;
 }

-void ClockCacheShard::Unref(CacheHandle* handle, bool set_usage,
+bool ClockCacheShard::Unref(CacheHandle* handle, bool set_usage,
                            CleanupContext* context) {
  if (set_usage) {
    handle->flags.fetch_or(kUsageBit, std::memory_order_relaxed);
@@ -451,9 +457,10 @@ void ClockCacheShard::Unref(CacheHandle* handle, bool set_usage,
      RecycleHandle(handle, context);
    }
  }
+  return context->to_delete_value.size();
 }

-void ClockCacheShard::UnsetInCache(CacheHandle* handle,
+bool ClockCacheShard::UnsetInCache(CacheHandle* handle,
                                   CleanupContext* context) {
  mutex_.AssertHeld();
  // Use acquire-release semantics as previous operations on the cache entry
@@ -465,6 +472,7 @@ void ClockCacheShard::UnsetInCache(CacheHandle* handle,
  if (InCache(flags) && CountRefs(flags) == 0) {
    RecycleHandle(handle, context);
  }
+  return context->to_delete_value.size();
 }

 bool ClockCacheShard::TryEvict(CacheHandle* handle, CleanupContext* context) {
@@ -618,27 +626,36 @@ Cache::Handle* ClockCacheShard::Lookup(const Slice& key, uint32_t hash) {
  return reinterpret_cast<Cache::Handle*>(handle);
 }

-void ClockCacheShard::Release(Cache::Handle* h) {
+bool ClockCacheShard::Release(Cache::Handle* h, bool force_erase) {
  CleanupContext context;
  CacheHandle* handle = reinterpret_cast<CacheHandle*>(h);
-  Unref(handle, true, &context);
+  bool erased = Unref(handle, true, &context);
+  if (force_erase && !erased) {
+    erased = EraseAndConfirm(handle->key, handle->hash, &context);
+  }
  Cleanup(context);
+  return erased;
 }

 void ClockCacheShard::Erase(const Slice& key, uint32_t hash) {
  CleanupContext context;
-  {
-    MutexLock l(&mutex_);
-    HashTable::accessor accessor;
-    if (table_.find(accessor, CacheKey(key, hash))) {
-      CacheHandle* handle = accessor->second;
-      table_.erase(accessor);
-      UnsetInCache(handle, &context);
-    }
-  }
+  EraseAndConfirm(key, hash, &context);
  Cleanup(context);
 }

+bool ClockCacheShard::EraseAndConfirm(const Slice& key, uint32_t hash,
+                                      CleanupContext* context) {
+  MutexLock l(&mutex_);
+  HashTable::accessor accessor;
+  bool erased = false;
+  if (table_.find(accessor, CacheKey(key, hash))) {
+    CacheHandle* handle = accessor->second;
+    table_.erase(accessor);
+    erased = UnsetInCache(handle, context);
+  }
+  return erased;
+}
+
 void ClockCacheShard::EraseUnRefEntries() {
  CleanupContext context;
  {
@@ -273,9 +273,9 @@ void LRUCacheShard::SetHighPriorityPoolRatio(double high_pri_pool_ratio) {
  MaintainPoolSize();
 }

-void LRUCacheShard::Release(Cache::Handle* handle) {
+bool LRUCacheShard::Release(Cache::Handle* handle, bool force_erase) {
  if (handle == nullptr) {
-    return;
+    return false;
  }
  LRUHandle* e = reinterpret_cast<LRUHandle*>(handle);
  bool last_reference = false;
@@ -287,10 +287,10 @@ void LRUCacheShard::Release(Cache::Handle* handle) {
    }
    if (e->refs == 1 && e->InCache()) {
      // The item is still in cache, and nobody else holds a reference to it
-      if (usage_ > capacity_) {
+      if (usage_ > capacity_ || force_erase) {
        // the cache is full
        // The LRU list must be empty since the cache is full
-        assert(lru_.next == &lru_);
+        assert(!(usage_ > capacity_) || lru_.next == &lru_);
        // take this opportunity and remove the item
        table_.Remove(e->key(), e->hash);
        e->SetInCache(false);
@@ -308,6 +308,7 @@ void LRUCacheShard::Release(Cache::Handle* handle) {
  if (last_reference) {
    e->Free();
  }
+  return last_reference;
 }

 Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
@@ -178,7 +178,8 @@ class LRUCacheShard : public CacheShard {
                        Cache::Priority priority) override;
  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash) override;
  virtual bool Ref(Cache::Handle* handle) override;
-  virtual void Release(Cache::Handle* handle) override;
+  virtual bool Release(Cache::Handle* handle,
+                       bool force_erase = false) override;
  virtual void Erase(const Slice& key, uint32_t hash) override;

  // Although in some platforms the update of size_t is atomic, to make sure
@@ -63,9 +63,9 @@ bool ShardedCache::Ref(Handle* handle) {
  return GetShard(Shard(hash))->Ref(handle);
 }

-void ShardedCache::Release(Handle* handle) {
+bool ShardedCache::Release(Handle* handle, bool force_erase) {
  uint32_t hash = GetHash(handle);
-  GetShard(Shard(hash))->Release(handle);
+  return GetShard(Shard(hash))->Release(handle, force_erase);
 }

 void ShardedCache::Erase(const Slice& key) {
@@ -30,7 +30,7 @@ class CacheShard {
                        Cache::Handle** handle, Cache::Priority priority) = 0;
  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash) = 0;
  virtual bool Ref(Cache::Handle* handle) = 0;
-  virtual void Release(Cache::Handle* handle) = 0;
+  virtual bool Release(Cache::Handle* handle, bool force_erase = false) = 0;
  virtual void Erase(const Slice& key, uint32_t hash) = 0;
  virtual void SetCapacity(size_t capacity) = 0;
  virtual void SetStrictCapacityLimit(bool strict_capacity_limit) = 0;
@@ -65,7 +65,7 @@ class ShardedCache : public Cache {
                        Handle** handle, Priority priority) override;
  virtual Handle* Lookup(const Slice& key, Statistics* stats) override;
  virtual bool Ref(Handle* handle) override;
-  virtual void Release(Handle* handle) override;
+  virtual bool Release(Handle* handle, bool force_erase = false) override;
  virtual void Erase(const Slice& key) override;
  virtual uint64_t NewId() override;
  virtual size_t GetCapacity() const override;
@@ -34,6 +34,9 @@
 #include "rocksdb/utilities/backupable_db.h"
 #include "rocksdb/utilities/write_batch_with_index.h"
 #include "utilities/merge_operators.h"
+#include "rocksdb/utilities/transaction.h"
+#include "rocksdb/utilities/transaction_db.h"
+#include "rocksdb/utilities/checkpoint.h"

 using rocksdb::BytewiseComparator;
 using rocksdb::Cache;
@@ -87,6 +90,12 @@ using rocksdb::RestoreOptions;
 using rocksdb::CompactRangeOptions;
 using rocksdb::RateLimiter;
 using rocksdb::NewGenericRateLimiter;
+using rocksdb::PinnableSlice;
+using rocksdb::TransactionDBOptions;
+using rocksdb::TransactionDB;
+using rocksdb::TransactionOptions;
+using rocksdb::Transaction;
+using rocksdb::Checkpoint;

 using std::shared_ptr;

@@ -125,6 +134,24 @@ struct rocksdb_envoptions_t      { EnvOptions        rep; };
 struct rocksdb_ingestexternalfileoptions_t  { IngestExternalFileOptions rep; };
 struct rocksdb_sstfilewriter_t   { SstFileWriter*    rep; };
 struct rocksdb_ratelimiter_t     { RateLimiter*      rep; };
+struct rocksdb_pinnableslice_t {
+  PinnableSlice rep;
+};
+struct rocksdb_transactiondb_options_t {
+  TransactionDBOptions rep;
+};
+struct rocksdb_transactiondb_t {
+  TransactionDB* rep;
+};
+struct rocksdb_transaction_options_t {
+  TransactionOptions rep;
+};
+struct rocksdb_transaction_t {
+  Transaction* rep;
+};
+struct rocksdb_checkpoint_t {
+  Checkpoint* rep;
+};

 struct rocksdb_compactionfiltercontext_t {
  CompactionFilter::Context rep;
@@ -537,6 +564,29 @@ void rocksdb_backup_engine_close(rocksdb_backup_engine_t* be) {
  delete be;
 }

+rocksdb_checkpoint_t* rocksdb_checkpoint_object_create(rocksdb_t* db,
+                                                       char** errptr) {
+  Checkpoint* checkpoint;
+  if (SaveError(errptr, Checkpoint::Create(db->rep, &checkpoint))) {
+    return nullptr;
+  }
+  rocksdb_checkpoint_t* result = new rocksdb_checkpoint_t;
+  result->rep = checkpoint;
+  return result;
+}
+
+void rocksdb_checkpoint_create(rocksdb_checkpoint_t* checkpoint,
+                               const char* checkpoint_dir,
+                               uint64_t log_size_for_flush, char** errptr) {
+  SaveError(errptr, checkpoint->rep->CreateCheckpoint(
+                        std::string(checkpoint_dir), log_size_for_flush));
+}
+
+void rocksdb_checkpoint_object_destroy(rocksdb_checkpoint_t* checkpoint) {
+  delete checkpoint->rep;
+  delete checkpoint;
+}
+
 void rocksdb_close(rocksdb_t* db) {
  delete db->rep;
  delete db;
@@ -1322,23 +1372,24 @@ void rocksdb_writebatch_put_log_data(
  b->rep.PutLogData(Slice(blob, len));
 }

+class H : public WriteBatch::Handler {
+ public:
+  void* state_;
+  void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
+  void (*deleted_)(void*, const char* k, size_t klen);
+  virtual void Put(const Slice& key, const Slice& value) override {
+    (*put_)(state_, key.data(), key.size(), value.data(), value.size());
+  }
+  virtual void Delete(const Slice& key) override {
+    (*deleted_)(state_, key.data(), key.size());
+  }
+};
+
 void rocksdb_writebatch_iterate(
    rocksdb_writebatch_t* b,
    void* state,
    void (*put)(void*, const char* k, size_t klen, const char* v, size_t vlen),
    void (*deleted)(void*, const char* k, size_t klen)) {
-  class H : public WriteBatch::Handler {
-   public:
-    void* state_;
-    void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
-    void (*deleted_)(void*, const char* k, size_t klen);
-    virtual void Put(const Slice& key, const Slice& value) override {
-      (*put_)(state_, key.data(), key.size(), value.data(), value.size());
-    }
-    virtual void Delete(const Slice& key) override {
-      (*deleted_)(state_, key.data(), key.size());
-    }
-  };
  H handler;
  handler.state_ = state;
  handler.put_ = put;
@@ -1360,6 +1411,10 @@ void rocksdb_writebatch_rollback_to_save_point(rocksdb_writebatch_t* b,
  SaveError(errptr, b->rep.RollbackToSavePoint());
 }

+void rocksdb_writebatch_pop_save_point(rocksdb_writebatch_t* b, char** errptr) {
+  SaveError(errptr, b->rep.PopSavePoint());
+}
+
 rocksdb_writebatch_wi_t* rocksdb_writebatch_wi_create(size_t reserved_bytes, unsigned char overwrite_key) {
  rocksdb_writebatch_wi_t* b = new rocksdb_writebatch_wi_t;
  b->rep = new WriteBatchWithIndex(BytewiseComparator(), reserved_bytes, overwrite_key);
@@ -1579,18 +1634,6 @@ void rocksdb_writebatch_wi_iterate(
    void* state,
    void (*put)(void*, const char* k, size_t klen, const char* v, size_t vlen),
    void (*deleted)(void*, const char* k, size_t klen)) {
-  class H : public WriteBatch::Handler {
-   public:
-    void* state_;
-    void (*put_)(void*, const char* k, size_t klen, const char* v, size_t vlen);
-    void (*deleted_)(void*, const char* k, size_t klen);
-    virtual void Put(const Slice& key, const Slice& value) override {
-      (*put_)(state_, key.data(), key.size(), value.data(), value.size());
-    }
-    virtual void Delete(const Slice& key) override {
-      (*deleted_)(state_, key.data(), key.size());
-    }
-  };
  H handler;
  handler.state_ = state;
  handler.put_ = put;
@@ -1983,6 +2026,10 @@ void rocksdb_options_set_max_open_files(rocksdb_options_t* opt, int n) {
  opt->rep.max_open_files = n;
 }

+void rocksdb_options_set_max_file_opening_threads(rocksdb_options_t* opt, int n) {
+  opt->rep.max_file_opening_threads = n;
+}
+
 void rocksdb_options_set_max_total_wal_size(rocksdb_options_t* opt, uint64_t n) {
  opt->rep.max_total_wal_size = n;
 }
@@ -2792,7 +2839,26 @@ void rocksdb_sstfilewriter_open(rocksdb_sstfilewriter_t* writer,
 void rocksdb_sstfilewriter_add(rocksdb_sstfilewriter_t* writer, const char* key,
                               size_t keylen, const char* val, size_t vallen,
                               char** errptr) {
-  SaveError(errptr, writer->rep->Add(Slice(key, keylen), Slice(val, vallen)));
+  SaveError(errptr, writer->rep->Put(Slice(key, keylen), Slice(val, vallen)));
+}
+
+void rocksdb_sstfilewriter_put(rocksdb_sstfilewriter_t* writer, const char* key,
+                               size_t keylen, const char* val, size_t vallen,
+                               char** errptr) {
+  SaveError(errptr, writer->rep->Put(Slice(key, keylen), Slice(val, vallen)));
+}
+
+void rocksdb_sstfilewriter_merge(rocksdb_sstfilewriter_t* writer,
+                                 const char* key, size_t keylen,
+                                 const char* val, size_t vallen,
+                                 char** errptr) {
+  SaveError(errptr, writer->rep->Merge(Slice(key, keylen), Slice(val, vallen)));
+}
+
+void rocksdb_sstfilewriter_delete(rocksdb_sstfilewriter_t* writer,
+                                  const char* key, size_t keylen,
+                                  char** errptr) {
+  SaveError(errptr, writer->rep->Delete(Slice(key, keylen)));
 }

 void rocksdb_sstfilewriter_finish(rocksdb_sstfilewriter_t* writer,
@@ -2889,20 +2955,21 @@ void rocksdb_slicetransform_destroy(rocksdb_slicetransform_t* st) {
  delete st;
 }

+struct Wrapper : public rocksdb_slicetransform_t {
+  const SliceTransform* rep_;
+  ~Wrapper() { delete rep_; }
+  const char* Name() const override { return rep_->Name(); }
+  Slice Transform(const Slice& src) const override {
+    return rep_->Transform(src);
+  }
+  bool InDomain(const Slice& src) const override {
+    return rep_->InDomain(src);
+  }
+  bool InRange(const Slice& src) const override { return rep_->InRange(src); }
+  static void DoNothing(void*) { }
+};
+
 rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t prefixLen) {
-  struct Wrapper : public rocksdb_slicetransform_t {
-    const SliceTransform* rep_;
-    ~Wrapper() { delete rep_; }
-    const char* Name() const override { return rep_->Name(); }
-    Slice Transform(const Slice& src) const override {
-      return rep_->Transform(src);
-    }
-    bool InDomain(const Slice& src) const override {
-      return rep_->InDomain(src);
-    }
-    bool InRange(const Slice& src) const override { return rep_->InRange(src); }
-    static void DoNothing(void*) { }
-  };
  Wrapper* wrapper = new Wrapper;
  wrapper->rep_ = rocksdb::NewFixedPrefixTransform(prefixLen);
  wrapper->state_ = nullptr;
@@ -2911,19 +2978,6 @@ rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t pref
 }

 rocksdb_slicetransform_t* rocksdb_slicetransform_create_noop() {
-  struct Wrapper : public rocksdb_slicetransform_t {
-    const SliceTransform* rep_;
-    ~Wrapper() { delete rep_; }
-    const char* Name() const override { return rep_->Name(); }
-    Slice Transform(const Slice& src) const override {
-      return rep_->Transform(src);
-    }
-    bool InDomain(const Slice& src) const override {
-      return rep_->InDomain(src);
-    }
-    bool InRange(const Slice& src) const override { return rep_->InRange(src); }
-    static void DoNothing(void*) { }
-  };
  Wrapper* wrapper = new Wrapper;
  wrapper->rep_ = rocksdb::NewNoopTransform();
  wrapper->state_ = nullptr;
@@ -3080,8 +3134,270 @@ void rocksdb_delete_file_in_range_cf(
          (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr)));
 }

+rocksdb_transactiondb_options_t* rocksdb_transactiondb_options_create() {
+  return new rocksdb_transactiondb_options_t;
+}
+
+void rocksdb_transactiondb_options_destroy(rocksdb_transactiondb_options_t* opt){
+  delete opt;
+}
+
+void rocksdb_transactiondb_options_set_max_num_locks(
+    rocksdb_transactiondb_options_t* opt, int64_t max_num_locks) {
+  opt->rep.max_num_locks = max_num_locks;
+}
+
+void rocksdb_transactiondb_options_set_num_stripes(
+    rocksdb_transactiondb_options_t* opt, size_t num_stripes) {
+  opt->rep.num_stripes = num_stripes;
+}
+
+void rocksdb_transactiondb_options_set_transaction_lock_timeout(
+    rocksdb_transactiondb_options_t* opt, int64_t txn_lock_timeout) {
+  opt->rep.transaction_lock_timeout = txn_lock_timeout;
+}
+
+void rocksdb_transactiondb_options_set_default_lock_timeout(
+    rocksdb_transactiondb_options_t* opt, int64_t default_lock_timeout) {
+  opt->rep.default_lock_timeout = default_lock_timeout;
+}
+
+rocksdb_transaction_options_t* rocksdb_transaction_options_create() {
+  return new rocksdb_transaction_options_t;
+}
+
+void rocksdb_transaction_options_destroy(rocksdb_transaction_options_t* opt) {
+  delete opt;
+}
+
+void rocksdb_transaction_options_set_set_snapshot(
+    rocksdb_transaction_options_t* opt, unsigned char v) {
+  opt->rep.set_snapshot = v;
+}
+
+void rocksdb_transaction_options_set_deadlock_detect(
+    rocksdb_transaction_options_t* opt, unsigned char v) {
+  opt->rep.deadlock_detect = v;
+}
+
+void rocksdb_transaction_options_set_lock_timeout(
+    rocksdb_transaction_options_t* opt, int64_t lock_timeout) {
+  opt->rep.lock_timeout = lock_timeout;
+}
+
+void rocksdb_transaction_options_set_expiration(
+    rocksdb_transaction_options_t* opt, int64_t expiration) {
+  opt->rep.expiration = expiration;
+}
+
+void rocksdb_transaction_options_set_deadlock_detect_depth(
+    rocksdb_transaction_options_t* opt, int64_t depth) {
+  opt->rep.deadlock_detect_depth = depth;
+}
+
+void rocksdb_transaction_options_set_max_write_batch_size(
+    rocksdb_transaction_options_t* opt, size_t size) {
+  opt->rep.max_write_batch_size = size;
+}
+
+rocksdb_transactiondb_t* rocksdb_transactiondb_open(
+    const rocksdb_options_t* options,
+    const rocksdb_transactiondb_options_t* txn_db_options, const char* name,
+    char** errptr) {
+  TransactionDB* txn_db;
+  if (SaveError(errptr, TransactionDB::Open(options->rep, txn_db_options->rep,
+                                            std::string(name), &txn_db))) {
+    return nullptr;
+  }
+  rocksdb_transactiondb_t* result = new rocksdb_transactiondb_t;
+  result->rep = txn_db;
+  return result;
+}
+
+const rocksdb_snapshot_t* rocksdb_transactiondb_create_snapshot(
+    rocksdb_transactiondb_t* txn_db) {
+  rocksdb_snapshot_t* result = new rocksdb_snapshot_t;
+  result->rep = txn_db->rep->GetSnapshot();
+  return result;
+}
+
+void rocksdb_transactiondb_release_snapshot(
+    rocksdb_transactiondb_t* txn_db, const rocksdb_snapshot_t* snapshot) {
+  txn_db->rep->ReleaseSnapshot(snapshot->rep);
+  delete snapshot;
+}
+
+rocksdb_transaction_t* rocksdb_transaction_begin(
+    rocksdb_transactiondb_t* txn_db,
+    const rocksdb_writeoptions_t* write_options,
+    const rocksdb_transaction_options_t* txn_options,
+    rocksdb_transaction_t* old_txn) {
+  if (old_txn == nullptr) {
+    rocksdb_transaction_t* result = new rocksdb_transaction_t;
+    result->rep = txn_db->rep->BeginTransaction(write_options->rep,
+                                                txn_options->rep, nullptr);
+    return result;
+  }
+  old_txn->rep = txn_db->rep->BeginTransaction(write_options->rep,
+                                                txn_options->rep, old_txn->rep);
+  return old_txn;
+}
+
+void rocksdb_transaction_commit(rocksdb_transaction_t* txn, char** errptr) {
+  SaveError(errptr, txn->rep->Commit());
+}
+
+void rocksdb_transaction_rollback(rocksdb_transaction_t* txn, char** errptr) {
+  SaveError(errptr, txn->rep->Rollback());
+}
+
+void rocksdb_transaction_destroy(rocksdb_transaction_t* txn) {
+  delete txn->rep;
+  delete txn;
+}
+
+//Read a key inside a transaction
+char* rocksdb_transaction_get(rocksdb_transaction_t* txn,
+                              const rocksdb_readoptions_t* options,
+                              const char* key, size_t klen, size_t* vlen,
+                              char** errptr) {
+  char* result = nullptr;
+  std::string tmp;
+  Status s = txn->rep->Get(options->rep, Slice(key, klen), &tmp);
+  if (s.ok()) {
+    *vlen = tmp.size();
+    result = CopyString(tmp);
+  } else {
+    *vlen = 0;
+    if (!s.IsNotFound()) {
+      SaveError(errptr, s);
+    }
+  }
+  return result;
+}
+
+// Read a key outside a transaction
+char* rocksdb_transactiondb_get(
+    rocksdb_transactiondb_t* txn_db,
+    const rocksdb_readoptions_t* options,
+    const char* key, size_t klen,
+    size_t* vlen,
+    char** errptr){
+  char* result = nullptr;
+  std::string tmp;
+  Status s = txn_db->rep->Get(options->rep, Slice(key, klen), &tmp);
+  if (s.ok()) {
+    *vlen = tmp.size();
+    result = CopyString(tmp);
+  } else {
+    *vlen = 0;
+    if (!s.IsNotFound()) {
+      SaveError(errptr, s);
+    }
+  }
+  return result;
+}
+
+// Put a key inside a transaction
+void rocksdb_transaction_put(rocksdb_transaction_t* txn, const char* key,
+                             size_t klen, const char* val, size_t vlen,
+                             char** errptr) {
+  SaveError(errptr, txn->rep->Put(Slice(key, klen), Slice(val, vlen)));
+}
+
+//Put a key outside a transaction
+void rocksdb_transactiondb_put(rocksdb_transactiondb_t* txn_db,
+                               const rocksdb_writeoptions_t* options,
+                               const char* key, size_t klen, const char* val,
+                               size_t vlen, char** errptr) {
+  SaveError(errptr,
+            txn_db->rep->Put(options->rep, Slice(key, klen), Slice(val, vlen)));
+}
+
+// Delete a key inside a transaction
+void rocksdb_transaction_delete(rocksdb_transaction_t* txn, const char* key,
+                                size_t klen, char** errptr) {
+  SaveError(errptr, txn->rep->Delete(Slice(key, klen)));
+}
+
+// Delete a key outside a transaction
+void rocksdb_transactiondb_delete(rocksdb_transactiondb_t* txn_db,
+                                  const rocksdb_writeoptions_t* options,
+                                  const char* key, size_t klen, char** errptr) {
+  SaveError(errptr, txn_db->rep->Delete(options->rep, Slice(key, klen)));
+}
+
+// Create an iterator inside a transaction
+rocksdb_iterator_t* rocksdb_transaction_create_iterator(
+    rocksdb_transaction_t* txn, const rocksdb_readoptions_t* options) {
+  rocksdb_iterator_t* result = new rocksdb_iterator_t;
+  result->rep = txn->rep->GetIterator(options->rep);
+  return result;
+}
+
+void rocksdb_transactiondb_close(rocksdb_transactiondb_t* txn_db) {
+  delete txn_db->rep;
+  delete txn_db;
+}
+
+rocksdb_checkpoint_t* rocksdb_transactiondb_checkpoint_object_create(
+    rocksdb_transactiondb_t* txn_db, char** errptr) {
+  Checkpoint* checkpoint;
+  if (SaveError(errptr, Checkpoint::Create(txn_db->rep, &checkpoint))) {
+    return nullptr;
+  }
+  rocksdb_checkpoint_t* result = new rocksdb_checkpoint_t;
+  result->rep = checkpoint;
+  return result;
+}
+
 void rocksdb_free(void* ptr) { free(ptr); }

+rocksdb_pinnableslice_t* rocksdb_get_pinned(
+    rocksdb_t* db, const rocksdb_readoptions_t* options, const char* key,
+    size_t keylen, char** errptr) {
+  rocksdb_pinnableslice_t* v = new (rocksdb_pinnableslice_t);
+  Status s = db->rep->Get(options->rep, db->rep->DefaultColumnFamily(),
+                          Slice(key, keylen), &v->rep);
+  if (!s.ok()) {
+    delete (v);
+    if (!s.IsNotFound()) {
+      SaveError(errptr, s);
+    }
+    return NULL;
+  }
+  return v;
+}
+
+rocksdb_pinnableslice_t* rocksdb_get_pinned_cf(
+    rocksdb_t* db, const rocksdb_readoptions_t* options,
+    rocksdb_column_family_handle_t* column_family, const char* key,
+    size_t keylen, char** errptr) {
+  rocksdb_pinnableslice_t* v = new (rocksdb_pinnableslice_t);
+  Status s = db->rep->Get(options->rep, column_family->rep, Slice(key, keylen),
+                          &v->rep);
+  if (!s.ok()) {
+    delete v;
+    if (!s.IsNotFound()) {
+      SaveError(errptr, s);
+    }
+    return NULL;
+  }
+  return v;
+}
+
+void rocksdb_pinnableslice_destroy(rocksdb_pinnableslice_t* v) { delete v; }
+
+const char* rocksdb_pinnableslice_value(const rocksdb_pinnableslice_t* v,
+                                        size_t* vlen) {
+  if (!v) {
+    *vlen = 0;
+    return NULL;
+  }
+
+  *vlen = v->rep.size();
+  return v->rep.data();
+}
 }  // end extern "C"

 #endif  // !ROCKSDB_LITE
@@ -20,7 +20,7 @@
 // Can not use port/port.h macros as this is a c file
 #ifdef OS_WIN

-#include <Windows.h>
+#include <windows.h>

 #define snprintf _snprintf

@@ -40,6 +40,7 @@ const char* phase = "";
 static char dbname[200];
 static char sstfilename[200];
 static char dbbackupname[200];
+static char dbcheckpointname[200];

 static void StartPhase(const char* name) {
  fprintf(stderr, "=== Test %s\n", name);
@@ -126,6 +127,32 @@ static void CheckGetCF(
  Free(&val);
 }

+static void CheckPinGet(rocksdb_t* db, const rocksdb_readoptions_t* options,
+                        const char* key, const char* expected) {
+  char* err = NULL;
+  size_t val_len;
+  const char* val;
+  rocksdb_pinnableslice_t* p;
+  p = rocksdb_get_pinned(db, options, key, strlen(key), &err);
+  CheckNoError(err);
+  val = rocksdb_pinnableslice_value(p, &val_len);
+  CheckEqual(expected, val, val_len);
+  rocksdb_pinnableslice_destroy(p);
+}
+
+static void CheckPinGetCF(rocksdb_t* db, const rocksdb_readoptions_t* options,
+                          rocksdb_column_family_handle_t* handle,
+                          const char* key, const char* expected) {
+  char* err = NULL;
+  size_t val_len;
+  const char* val;
+  rocksdb_pinnableslice_t* p;
+  p = rocksdb_get_pinned_cf(db, options, handle, key, strlen(key), &err);
+  CheckNoError(err);
+  val = rocksdb_pinnableslice_value(p, &val_len);
+  CheckEqual(expected, val, val_len);
+  rocksdb_pinnableslice_destroy(p);
+}

 static void CheckIter(rocksdb_iterator_t* iter,
                      const char* key, const char* val) {
@@ -292,6 +319,34 @@ static char* MergeOperatorPartialMerge(
  return result;
 }

+static void CheckTxnGet(
+        rocksdb_transaction_t* txn,
+        const rocksdb_readoptions_t* options,
+        const char* key,
+        const char* expected) {
+        char* err = NULL;
+        size_t val_len;
+        char* val;
+        val = rocksdb_transaction_get(txn, options, key, strlen(key), &val_len, &err);
+        CheckNoError(err);
+        CheckEqual(expected, val, val_len);
+        Free(&val);
+}
+
+static void CheckTxnDBGet(
+        rocksdb_transactiondb_t* txn_db,
+        const rocksdb_readoptions_t* options,
+        const char* key,
+        const char* expected) {
+        char* err = NULL;
+        size_t val_len;
+        char* val;
+        val = rocksdb_transactiondb_get(txn_db, options, key, strlen(key), &val_len, &err);
+        CheckNoError(err);
+        CheckEqual(expected, val, val_len);
+        Free(&val);
+}
+
 int main(int argc, char** argv) {
  rocksdb_t* db;
  rocksdb_comparator_t* cmp;
@@ -303,6 +358,10 @@ int main(int argc, char** argv) {
  rocksdb_readoptions_t* roptions;
  rocksdb_writeoptions_t* woptions;
  rocksdb_ratelimiter_t* rate_limiter;
+  rocksdb_transactiondb_t* txn_db;
+  rocksdb_transactiondb_options_t* txn_db_options;
+  rocksdb_transaction_t* txn;
+  rocksdb_transaction_options_t* txn_options;
  char* err = NULL;
  int run = -1;

@@ -316,6 +375,11 @@ int main(int argc, char** argv) {
           GetTempDir(),
           ((int) geteuid()));

+  snprintf(dbcheckpointname, sizeof(dbcheckpointname),
+           "%s/rocksdb_c_test-%d-checkpoint",
+           GetTempDir(),
+           ((int) geteuid()));
+
  snprintf(sstfilename, sizeof(sstfilename),
           "%s/rocksdb_c_test-%d-sst",
           GetTempDir(),
@@ -431,6 +495,36 @@ int main(int argc, char** argv) {
    rocksdb_backup_engine_close(be);
  }

+  StartPhase("checkpoint");
+  {
+    rocksdb_destroy_db(options, dbcheckpointname, &err);
+    CheckNoError(err);
+
+    rocksdb_checkpoint_t* checkpoint = rocksdb_checkpoint_object_create(db, &err);
+    CheckNoError(err);
+
+    rocksdb_checkpoint_create(checkpoint, dbcheckpointname, 0, &err);
+    CheckNoError(err);
+
+    // start a new database from the checkpoint
+    rocksdb_close(db);
+    rocksdb_options_set_error_if_exists(options, 0);
+    db = rocksdb_open(options, dbcheckpointname, &err);
+    CheckNoError(err);
+
+    CheckGet(db, roptions, "foo", "hello");
+
+    rocksdb_checkpoint_object_destroy(checkpoint);
+
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbcheckpointname, &err);
+    CheckNoError(err);
+
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+    rocksdb_options_set_error_if_exists(options, 1);
+  }
+
  StartPhase("compactall");
  rocksdb_compact_range(db, NULL, 0, NULL, 0);
  CheckGet(db, roptions, "foo", "hello");
@@ -475,11 +569,11 @@ int main(int argc, char** argv) {
    unlink(sstfilename);
    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk1", 5, "v1", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk1", 5, "v1", 2, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v2", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk2", 5, "v2", 2, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v3", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk3", 5, "v3", 2, &err);
    CheckNoError(err);
    rocksdb_sstfilewriter_finish(writer, &err);
    CheckNoError(err);
@@ -496,11 +590,11 @@ int main(int argc, char** argv) {
    unlink(sstfilename);
    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v4", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk2", 5, "v4", 2, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk22", 6, "v5", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk22", 6, "v5", 2, &err);
    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v6", 2, &err);
+    rocksdb_sstfilewriter_put(writer, "sstk3", 5, "v6", 2, &err);
    CheckNoError(err);
    rocksdb_sstfilewriter_finish(writer, &err);
    CheckNoError(err);
@@ -587,10 +681,13 @@ int main(int argc, char** argv) {
  {
    rocksdb_writebatch_t* wb = rocksdb_writebatch_create();
    rocksdb_writebatch_set_save_point(wb);
+    rocksdb_writebatch_set_save_point(wb);
    const char* k_list[2] = {"z", "ap"};
    const size_t k_sizes[2] = {1, 2};
    const char* v_list[3] = {"x", "y", "z"};
    const size_t v_sizes[3] = {1, 1, 1};
+    rocksdb_writebatch_pop_save_point(wb, &err);
+    CheckNoError(err);
    rocksdb_writebatch_putv(wb, 2, k_list, k_sizes, 3, v_list, v_sizes);
    rocksdb_writebatch_rollback_to_save_point(wb, &err);
    CheckNoError(err);
@@ -786,6 +883,13 @@ int main(int argc, char** argv) {
    }
  }

+  StartPhase("pin_get");
+  {
+    CheckPinGet(db, roptions, "box", "c");
+    CheckPinGet(db, roptions, "foo", "hello");
+    CheckPinGet(db, roptions, "notfound", NULL);
+  }
+
  StartPhase("approximate_sizes");
  {
    int i;
@@ -997,11 +1101,13 @@ int main(int argc, char** argv) {
    CheckNoError(err);

    CheckGetCF(db, roptions, handles[1], "foo", "hello");
+    CheckPinGetCF(db, roptions, handles[1], "foo", "hello");

    rocksdb_delete_cf(db, woptions, handles[1], "foo", 3, &err);
    CheckNoError(err);

    CheckGetCF(db, roptions, handles[1], "foo", NULL);
+    CheckPinGetCF(db, roptions, handles[1], "foo", NULL);

    rocksdb_writebatch_t* wb = rocksdb_writebatch_create();
    rocksdb_writebatch_put_cf(wb, handles[1], "baz", 3, "a", 1);
@@ -1014,6 +1120,9 @@ int main(int argc, char** argv) {
    CheckGetCF(db, roptions, handles[1], "baz", NULL);
    CheckGetCF(db, roptions, handles[1], "bar", NULL);
    CheckGetCF(db, roptions, handles[1], "box", "c");
+    CheckPinGetCF(db, roptions, handles[1], "baz", NULL);
+    CheckPinGetCF(db, roptions, handles[1], "bar", NULL);
+    CheckPinGetCF(db, roptions, handles[1], "box", "c");
    rocksdb_writebatch_destroy(wb);

    const char* keys[3] = { "box", "box", "barfooxx" };
@@ -1225,14 +1334,98 @@ int main(int argc, char** argv) {
    }
  }

-  // Simple sanity check that setting memtable rep works.
-  StartPhase("memtable_reps");
+  StartPhase("transactions");
  {
-    // Create database with vector memtable.
    rocksdb_close(db);
    rocksdb_destroy_db(options, dbname, &err);
    CheckNoError(err);

+    // open a TransactionDB
+    txn_db_options = rocksdb_transactiondb_options_create();
+    txn_options = rocksdb_transaction_options_create();
+    rocksdb_options_set_create_if_missing(options, 1);
+    txn_db = rocksdb_transactiondb_open(options, txn_db_options, dbname, &err);
+    CheckNoError(err);
+
+    // put outside a transaction
+    rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hello", 5, &err);
+    CheckNoError(err);
+    CheckTxnDBGet(txn_db, roptions, "foo", "hello");
+
+    // delete from outside transaction
+    rocksdb_transactiondb_delete(txn_db, woptions, "foo", 3, &err);
+    CheckNoError(err);
+    CheckTxnDBGet(txn_db, roptions, "foo", NULL);
+
+    // begin a transaction
+    txn = rocksdb_transaction_begin(txn_db, woptions, txn_options, NULL);
+    // put
+    rocksdb_transaction_put(txn, "foo", 3, "hello", 5, &err);
+    CheckNoError(err);
+    CheckTxnGet(txn, roptions, "foo", "hello");
+    // delete
+    rocksdb_transaction_delete(txn, "foo", 3, &err);
+    CheckNoError(err);
+    CheckTxnGet(txn, roptions, "foo", NULL);
+
+    rocksdb_transaction_put(txn, "foo", 3, "hello", 5, &err);
+    CheckNoError(err);
+
+    // read from outside transaction, before commit
+    CheckTxnDBGet(txn_db, roptions, "foo", NULL);
+
+    // commit
+    rocksdb_transaction_commit(txn, &err);
+    CheckNoError(err);
+
+    // read from outside transaction, after commit
+    CheckTxnDBGet(txn_db, roptions, "foo", "hello");
+
+    // reuse old transaction
+    txn = rocksdb_transaction_begin(txn_db, woptions, txn_options, txn);
+
+    // snapshot
+    const rocksdb_snapshot_t* snapshot;
+    snapshot = rocksdb_transactiondb_create_snapshot(txn_db);
+    rocksdb_readoptions_set_snapshot(roptions, snapshot);
+  
+    rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hey", 3,  &err);
+    CheckNoError(err);
+
+    CheckTxnDBGet(txn_db, roptions, "foo", "hello");
+    rocksdb_readoptions_set_snapshot(roptions, NULL);
+    rocksdb_transactiondb_release_snapshot(txn_db, snapshot);
+    CheckTxnDBGet(txn_db, roptions, "foo", "hey");
+
+    // iterate
+    rocksdb_transaction_put(txn, "bar", 3, "hi", 2, &err);
+    rocksdb_iterator_t* iter = rocksdb_transaction_create_iterator(txn, roptions);
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_seek_to_first(iter);
+    CheckCondition(rocksdb_iter_valid(iter));
+    CheckIter(iter, "bar", "hi");
+    rocksdb_iter_get_error(iter, &err);
+    CheckNoError(err);
+    rocksdb_iter_destroy(iter);
+
+    // rollback
+    rocksdb_transaction_rollback(txn, &err);
+    CheckNoError(err);
+    CheckTxnDBGet(txn_db, roptions, "bar", NULL);
+
+    // close and destroy
+    rocksdb_transaction_destroy(txn);
+    rocksdb_transactiondb_close(txn_db);
+    rocksdb_destroy_db(options, dbname, &err);
+    CheckNoError(err);
+    rocksdb_transaction_options_destroy(txn_options);
+    rocksdb_transactiondb_options_destroy(txn_db_options);
+  }
+
+  // Simple sanity check that setting memtable rep works.
+  StartPhase("memtable_reps");
+  {
+    // Create database with vector memtable.
    rocksdb_options_set_memtable_vector_rep(options);
    db = rocksdb_open(options, dbname, &err);
    CheckNoError(err);
@@ -170,6 +170,12 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
      result.num_levels < 2) {
    result.num_levels = 2;
  }
+
+  if (result.compaction_style == kCompactionStyleUniversal &&
+      db_options.allow_ingest_behind && result.num_levels < 3) {
+    result.num_levels = 3;
+  }
+
  if (result.max_write_buffer_number < 2) {
    result.max_write_buffer_number = 2;
  }
@@ -196,7 +202,6 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
    result.num_levels = 1;
    // since we delete level0 files in FIFO compaction when there are too many
    // of them, these options don't really mean anything
-    result.level0_file_num_compaction_trigger = std::numeric_limits<int>::max();
    result.level0_slowdown_writes_trigger = std::numeric_limits<int>::max();
    result.level0_stop_writes_trigger = std::numeric_limits<int>::max();
  }
@@ -342,6 +347,7 @@ ColumnFamilyData::ColumnFamilyData(
      dummy_versions_(_dummy_versions),
      current_(nullptr),
      refs_(0),
+      initialized_(false),
      dropped_(false),
      internal_comparator_(cf_options.comparator),
      initial_cf_options_(SanitizeOptions(db_options, cf_options)),
@@ -544,7 +550,7 @@ std::unique_ptr<WriteControllerToken> SetupDelay(
    // If DB just falled into the stop condition, we need to further reduce
    // the write rate to avoid the stop condition.
    if (penalize_stop) {
-      // Penalize the near stop or stop condition by more agressive slowdown.
+      // Penalize the near stop or stop condition by more aggressive slowdown.
      // This is to provide the long term slowdown increase signal.
      // The penalty is more than the reward of recovering to the normal
      // condition.
@@ -722,7 +728,7 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
              mutable_cf_options.level0_slowdown_writes_trigger)) {
        write_controller_token_ =
            write_controller->GetCompactionPressureToken();
-        ROCKS_LOG_WARN(
+        ROCKS_LOG_INFO(
            ioptions_.info_log,
            "[%s] Increasing compaction threads because we have %d level-0 "
            "files ",
@@ -736,7 +742,7 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
        write_controller_token_ =
            write_controller->GetCompactionPressureToken();
        if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0) {
-          ROCKS_LOG_WARN(
+          ROCKS_LOG_INFO(
              ioptions_.info_log,
              "[%s] Increasing compaction threads because of estimated pending "
              "compaction "
@@ -753,6 +759,12 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
        uint64_t write_rate = write_controller->delayed_write_rate();
        write_controller->set_delayed_write_rate(static_cast<uint64_t>(
            static_cast<double>(write_rate) * kDelayRecoverSlowdownRatio));
+        // Set the low pri limit to be 1/4 the delayed write rate.
+        // Note we don't reset this value even after delay condition is relased.
+        // Low-pri rate will continue to apply if there is a compaction
+        // pressure.
+        write_controller->low_pri_rate_limiter()->SetBytesPerSecond(write_rate /
+                                                                    4);
      }
    }
    prev_compaction_needed_bytes_ = compaction_needed_bytes;
@@ -777,9 +789,8 @@ uint64_t ColumnFamilyData::GetTotalSstFilesSize() const {

 MemTable* ColumnFamilyData::ConstructNewMemtable(
    const MutableCFOptions& mutable_cf_options, SequenceNumber earliest_seq) {
-  assert(current_ != nullptr);
  return new MemTable(internal_comparator_, ioptions_, mutable_cf_options,
-                      write_buffer_manager_, earliest_seq);
+                      write_buffer_manager_, earliest_seq, id_);
 }

 void ColumnFamilyData::CreateNewMemtable(
@@ -333,6 +333,10 @@ class ColumnFamilyData {
  void RecalculateWriteStallConditions(
      const MutableCFOptions& mutable_cf_options);

+  void set_initialized() { initialized_.store(true); }
+
+  bool initialized() const { return initialized_.load(); }
+
 private:
  friend class ColumnFamilySet;
  ColumnFamilyData(uint32_t id, const std::string& name,
@@ -349,6 +353,7 @@ class ColumnFamilyData {
  Version* current_;         // == dummy_versions->prev_

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

  const InternalKeyComparator internal_comparator_;
@@ -54,7 +54,7 @@ class EnvCounter : public EnvWrapper {
  }

 private:
-  int num_new_writable_file_;
+  std::atomic<int> num_new_writable_file_;
 };

 class ColumnFamilyTest : public testing::Test {
@@ -520,6 +520,30 @@ TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
  }
 }

+#ifndef ROCKSDB_LITE
+TEST_F(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
+  Open();
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::WriteOptionsFile:1",
+        "ColumnFamilyTest.CreateCFRaceWithGetAggProperty:1"},
+       {"ColumnFamilyTest.CreateCFRaceWithGetAggProperty:2",
+        "DBImpl::WriteOptionsFile:2"}});
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  rocksdb::port::Thread thread([&] { CreateColumnFamilies({"one"}); });
+
+  TEST_SYNC_POINT("ColumnFamilyTest.CreateCFRaceWithGetAggProperty:1");
+  uint64_t pv;
+  db_->GetAggregatedIntProperty(DB::Properties::kEstimateTableReadersMem, &pv);
+  TEST_SYNC_POINT("ColumnFamilyTest.CreateCFRaceWithGetAggProperty:2");
+
+  thread.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+#endif  // !ROCKSDB_LITE
+
 class FlushEmptyCFTestWithParam : public ColumnFamilyTest,
                                  public testing::WithParamInterface<bool> {
 public:
@@ -1379,7 +1403,6 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1403,6 +1426,9 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
  two.write_buffer_size = 100000;

  Reopen({default_cf, one, two});
+  // make sure all background compaction jobs can be scheduled
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();

  bool cf_1_1 = true;
  rocksdb::SyncPoint::GetInstance()->LoadDependency(
@@ -1462,7 +1488,6 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-  Close();
 }

 TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
@@ -1471,7 +1496,6 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1495,6 +1519,9 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
  two.write_buffer_size = 100000;

  Reopen({default_cf, one, two});
+  // 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) {
@@ -1557,7 +1584,6 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-  Close();
 }

 TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
@@ -1566,7 +1592,6 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1585,6 +1610,9 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
  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) {
@@ -1655,7 +1683,6 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-  Close();
 }

 TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
@@ -1664,7 +1691,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1683,6 +1709,9 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
  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) {
@@ -1744,7 +1773,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-  Close();
 }

 TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
@@ -1753,7 +1781,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1772,6 +1799,9 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  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" -- level style
  for (int i = 0; i < one.level0_file_num_compaction_trigger - 2; ++i) {
@@ -1833,7 +1863,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-  Close();
 }

 // This test checks for automatic getting a conflict if there is a
@@ -1849,7 +1878,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1868,6 +1896,9 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
  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) {
@@ -1952,8 +1983,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-
-  Close();
 }

 // In this test, we generate enough files to trigger automatic compactions.
@@ -1968,7 +1997,6 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
  db_options_.max_background_compactions = 3;
-  db_options_.base_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1987,6 +2015,9 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
  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();

  bool cf_1_1 = true;
  bool cf_1_2 = true;
@@ -2041,8 +2072,6 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
-
-  Close();
 }
 #endif  // !ROCKSDB_LITE

@@ -2495,7 +2524,6 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
 TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  const uint64_t kBaseRate = 800000u;
  db_options_.delayed_write_rate = kBaseRate;
-  db_options_.base_background_compactions = 2;
  db_options_.max_background_compactions = 6;

  Open({"default"});
@@ -2676,7 +2704,6 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
 }

 TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
-  db_options_.base_background_compactions = 2;
  db_options_.max_background_compactions = 6;
  Open({"default"});
  ColumnFamilyData* cfd =
@@ -2696,7 +2723,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  vstorage->TEST_set_estimated_compaction_needed_bytes(40);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(50);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2708,11 +2735,11 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  vstorage->TEST_set_estimated_compaction_needed_bytes(45);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(7);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(9);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2720,7 +2747,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  vstorage->set_l0_delay_trigger_count(6);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  // Speed up threshold = min(4 * 2, 4 + (12 - 4)/4) = 6
  mutable_cf_options.level0_file_num_compaction_trigger = 4;
@@ -2729,7 +2756,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  vstorage->set_l0_delay_trigger_count(5);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(7);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2737,7 +2764,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  vstorage->set_l0_delay_trigger_count(3);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }

 TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
@@ -2816,7 +2843,6 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
 }

 TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
-  db_options_.base_background_compactions = 2;
  db_options_.max_background_compactions = 6;
  column_family_options_.soft_pending_compaction_bytes_limit = 200;
  column_family_options_.hard_pending_compaction_bytes_limit = 2000;
@@ -2844,11 +2870,11 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {

  vstorage->TEST_set_estimated_compaction_needed_bytes(40);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(60);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

@@ -2866,7 +2892,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {

  vstorage1->TEST_set_estimated_compaction_needed_bytes(3);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(9);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2878,7 +2904,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {

  vstorage->set_l0_delay_trigger_count(0);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }

 #ifndef ROCKSDB_LITE
@@ -4,6 +4,7 @@
 //  (found in the LICENSE.Apache file in the root directory).

 #include "db/compaction_iterator.h"
+#include "rocksdb/listener.h"
 #include "table/internal_iterator.h"

 namespace rocksdb {
@@ -66,7 +67,9 @@ CompactionIterator::CompactionIterator(
      range_del_agg_(range_del_agg),
      compaction_(std::move(compaction)),
      compaction_filter_(compaction_filter),
+#ifndef ROCKSDB_LITE
      compaction_listener_(compaction_listener),
+#endif  // ROCKSDB_LITE
      shutting_down_(shutting_down),
      ignore_snapshots_(false),
      merge_out_iter_(merge_helper_) {
@@ -80,7 +83,7 @@ CompactionIterator::CompactionIterator(
  if (snapshots_->size() == 0) {
    // optimize for fast path if there are no snapshots
    visible_at_tip_ = true;
-    earliest_snapshot_ = last_sequence;
+    earliest_snapshot_ = kMaxSequenceNumber;
    latest_snapshot_ = 0;
  } else {
    visible_at_tip_ = false;
@@ -543,7 +546,8 @@ void CompactionIterator::PrepareOutput() {

  // This is safe for TransactionDB write-conflict checking since transactions
  // only care about sequence number larger than any active snapshots.
-  if (bottommost_level_ && valid_ && ikey_.sequence < earliest_snapshot_ &&
+  if ((compaction_ != nullptr && !compaction_->allow_ingest_behind()) &&
+      bottommost_level_ && valid_ && ikey_.sequence <= earliest_snapshot_ &&
      ikey_.type != kTypeMerge &&
      !cmp_->Equal(compaction_->GetLargestUserKey(), ikey_.user_key)) {
    assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
@@ -14,6 +14,7 @@
 #include "db/merge_helper.h"
 #include "db/pinned_iterators_manager.h"
 #include "db/range_del_aggregator.h"
+#include "options/cf_options.h"
 #include "rocksdb/compaction_filter.h"

 namespace rocksdb {
@@ -44,6 +45,9 @@ class CompactionIterator {
    virtual Slice GetLargestUserKey() const {
      return compaction_->GetLargestUserKey();
    }
+    virtual bool allow_ingest_behind() const {
+      return compaction_->immutable_cf_options()->allow_ingest_behind;
+    }

   protected:
    CompactionProxy() = default;
@@ -126,7 +130,9 @@ class CompactionIterator {
  RangeDelAggregator* range_del_agg_;
  std::unique_ptr<CompactionProxy> compaction_;
  const CompactionFilter* compaction_filter_;
+#ifndef ROCKSDB_LITE
  CompactionEventListener* compaction_listener_;
+#endif  // ROCKSDB_LITE
  const std::atomic<bool>* shutting_down_;
  bool bottommost_level_;
  bool valid_ = false;
@@ -154,6 +154,7 @@ class FakeCompaction : public CompactionIterator::CompactionProxy {
  virtual Slice GetLargestUserKey() const {
    return "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
  }
+  virtual bool allow_ingest_behind() const { return false; }

  bool key_not_exists_beyond_output_level = false;
 };
@@ -654,7 +654,6 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
  Options options;
  options.listeners.emplace_back(stats_checker);
  options.create_if_missing = true;
-  options.max_background_flushes = 0;
  // just enough setting to hold off auto-compaction.
  options.level0_file_num_compaction_trigger = kTestScale + 1;
  options.num_levels = 3;
@@ -876,7 +875,6 @@ TEST_P(CompactionJobStatsTest, DeletionStatsTest) {
  Options options;
  options.listeners.emplace_back(stats_checker);
  options.create_if_missing = true;
-  options.max_background_flushes = 0;
  options.level0_file_num_compaction_trigger = kTestScale+1;
  options.num_levels = 3;
  options.compression = kNoCompression;
@@ -36,6 +36,39 @@ uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) {
  }
  return sum;
 }
+
+bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
+                           size_t min_files_to_compact,
+                           uint64_t max_compact_bytes_per_del_file,
+                           CompactionInputFiles* comp_inputs) {
+  size_t compact_bytes = level_files[0]->fd.file_size;
+  size_t compact_bytes_per_del_file = port::kMaxSizet;
+  // compaction range will be [0, span_len).
+  size_t span_len;
+  // pull in files until the amount of compaction work per deleted file begins
+  // increasing.
+  size_t new_compact_bytes_per_del_file = 0;
+  for (span_len = 1; span_len < level_files.size(); ++span_len) {
+    compact_bytes += level_files[span_len]->fd.file_size;
+    new_compact_bytes_per_del_file = compact_bytes / span_len;
+    if (level_files[span_len]->being_compacted ||
+        new_compact_bytes_per_del_file > compact_bytes_per_del_file) {
+      break;
+    }
+    compact_bytes_per_del_file = new_compact_bytes_per_del_file;
+  }
+
+  if (span_len >= min_files_to_compact &&
+      new_compact_bytes_per_del_file < max_compact_bytes_per_del_file) {
+    assert(comp_inputs != nullptr);
+    comp_inputs->level = 0;
+    for (size_t i = 0; i < span_len; ++i) {
+      comp_inputs->files.push_back(level_files[i]);
+    }
+    return true;
+  }
+  return false;
+}
 }  // anonymous namespace

 // Determine compression type, based on user options, level of the output
@@ -494,7 +527,11 @@ Compaction* CompactionPicker::CompactRange(
    // files together to the last level.
    assert(vstorage->num_levels() > 1);
    // DBImpl::CompactRange() set output level to be the last level
-    assert(output_level == vstorage->num_levels() - 1);
+    if (ioptions_.allow_ingest_behind) {
+      assert(output_level == vstorage->num_levels() - 2);
+    } else {
+      assert(output_level == vstorage->num_levels() - 1);
+    }
    // DBImpl::RunManualCompaction will make full range for universal compaction
    assert(begin == nullptr);
    assert(end == nullptr);
@@ -1346,31 +1383,8 @@ bool LevelCompactionBuilder::PickIntraL0Compaction() {
    // resort to L0->L0 compaction yet.
    return false;
  }
-
-  size_t compact_bytes = level_files[0]->fd.file_size;
-  size_t compact_bytes_per_del_file = port::kMaxSizet;
-  // compaction range will be [0, span_len).
-  size_t span_len;
-  // pull in files until the amount of compaction work per deleted file begins
-  // increasing.
-  for (span_len = 1; span_len < level_files.size(); ++span_len) {
-    compact_bytes += level_files[span_len]->fd.file_size;
-    size_t new_compact_bytes_per_del_file = compact_bytes / span_len;
-    if (level_files[span_len]->being_compacted ||
-        new_compact_bytes_per_del_file > compact_bytes_per_del_file) {
-      break;
-    }
-    compact_bytes_per_del_file = new_compact_bytes_per_del_file;
-  }
-
-  if (span_len >= kMinFilesForIntraL0Compaction) {
-    start_level_inputs_.level = 0;
-    for (size_t i = 0; i < span_len; ++i) {
-      start_level_inputs_.files.push_back(level_files[i]);
-    }
-    return true;
-  }
-  return false;
+  return FindIntraL0Compaction(level_files, kMinFilesForIntraL0Compaction,
+                               port::kMaxUint64, &start_level_inputs_);
 }
 }  // namespace

@@ -1403,6 +1417,27 @@ Compaction* FIFOCompactionPicker::PickCompaction(
  if (total_size <= ioptions_.compaction_options_fifo.max_table_files_size ||
      level_files.size() == 0) {
    // total size not exceeded
+    if (ioptions_.compaction_options_fifo.allow_compaction &&
+        level_files.size() > 0) {
+      CompactionInputFiles comp_inputs;
+      if (FindIntraL0Compaction(
+              level_files,
+              mutable_cf_options
+                  .level0_file_num_compaction_trigger /* min_files_to_compact */,
+              mutable_cf_options.write_buffer_size, &comp_inputs)) {
+        Compaction* c = new Compaction(
+            vstorage, ioptions_, mutable_cf_options, {comp_inputs}, 0,
+            16 * 1024 * 1024 /* output file size limit */,
+            0 /* max compaction bytes, not applicable */,
+            0 /* output path ID */, mutable_cf_options.compression, {},
+            /* is manual */ false, vstorage->CompactionScore(0),
+            /* is deletion compaction */ false,
+            CompactionReason::kFIFOReduceNumFiles);
+        RegisterCompaction(c);
+        return c;
+      }
+    }
+
    ROCKS_LOG_BUFFER(log_buffer,
                     "[%s] FIFO compaction: nothing to do. Total size %" PRIu64
                     ", max size %" PRIu64 "\n",
@@ -400,6 +400,35 @@ TEST_F(CompactionPickerTest, NeedsCompactionUniversal) {
              vstorage_->CompactionScore(0) >= 1);
  }
 }
+
+TEST_F(CompactionPickerTest, CompactionUniversalIngestBehindReservedLevel) {
+  const uint64_t kFileSize = 100000;
+  NewVersionStorage(1, kCompactionStyleUniversal);
+  ioptions_.allow_ingest_behind = true;
+  ioptions_.num_levels = 3;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  // must return false when there's no files.
+  ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
+            false);
+
+  NewVersionStorage(3, kCompactionStyleUniversal);
+
+  Add(0, 1U, "150", "200", kFileSize, 0, 500, 550);
+  Add(0, 2U, "201", "250", kFileSize, 0, 401, 450);
+  Add(0, 4U, "260", "300", kFileSize, 0, 260, 300);
+  Add(1, 5U, "100", "151", kFileSize, 0, 200, 251);
+  Add(1, 3U, "301", "350", kFileSize, 0, 101, 150);
+  Add(2, 6U, "120", "200", kFileSize, 0, 20, 100);
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(
+      universal_compaction_picker.PickCompaction(
+          cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+
+  // output level should be the one above the bottom-most
+  ASSERT_EQ(1, compaction->output_level());
+}
 // Tests if the files can be trivially moved in multi level
 // universal compaction when allow_trivial_move option is set
 // In this test as the input files overlaps, they cannot
@@ -921,7 +950,7 @@ TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri3) {
 TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded1) {
  int num_levels = ioptions_.num_levels;
  ioptions_.level_compaction_dynamic_level_bytes = false;
-  mutable_cf_options_.level0_file_num_compaction_trigger = 3;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 4;
  mutable_cf_options_.max_bytes_for_level_base = 1000;
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
  NewVersionStorage(num_levels, kCompactionStyleLevel);
@@ -1011,7 +1040,8 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {

  // Set Last level size 50000
  // num_levels - 1 target 5000
-  // num_levels - 2 is base level with taret 500
+  // num_levels - 2 is base level with target 1000 (rounded up to
+  // max_bytes_for_level_base).
  Add(num_levels - 1, 10U, "400", "500", 50000);

  Add(0, 1U, "150", "200", 200);
@@ -1020,16 +1050,16 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {
  Add(0, 5U, "150", "200", 200);
  Add(0, 6U, "150", "200", 200);
  // num_levels - 3 is over target by 100 + 1000
-  Add(num_levels - 3, 7U, "400", "500", 300);
-  Add(num_levels - 3, 8U, "600", "700", 300);
+  Add(num_levels - 3, 7U, "400", "500", 550);
+  Add(num_levels - 3, 8U, "600", "700", 550);
  // num_levels - 2 is over target by 1100 + 200
  Add(num_levels - 2, 9U, "150", "200", 5200);

  UpdateVersionStorageInfo();

-  // Merging to the second last level: (5200 / 1600 + 1) * 1100
+  // Merging to the second last level: (5200 / 2100 + 1) * 1100
  // Merging to the last level: (50000 / 6300 + 1) * 1300
-  ASSERT_EQ(1600u + 4675u + 11617u,
+  ASSERT_EQ(2100u + 3823u + 11617u,
            vstorage_->estimated_compaction_needed_bytes());
 }

@@ -342,8 +342,6 @@ Compaction* UniversalCompactionPicker::PickCompaction(
      assert(f->smallest_seqno <= f->largest_seqno);
      if (is_first) {
        is_first = false;
-      } else {
-        assert(prev_smallest_seqno > f->largest_seqno);
      }
      prev_smallest_seqno = f->smallest_seqno;
    }
@@ -568,6 +566,13 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
    output_level = sorted_runs[first_index_after].level - 1;
  }

+  // last level is reserved for the files ingested behind
+  if (ioptions_.allow_ingest_behind &&
+      (output_level == vstorage->num_levels() - 1)) {
+    assert(output_level > 1);
+    output_level--;
+  }
+
  std::vector<CompactionInputFiles> inputs(vstorage->num_levels());
  for (size_t i = 0; i < inputs.size(); ++i) {
    inputs[i].level = start_level + static_cast<int>(i);
@@ -614,7 +619,7 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    VersionStorageInfo* vstorage, double score,
    const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
-  // percentage flexibilty while reducing size amplification
+  // percentage flexibility while reducing size amplification
  uint64_t ratio =
      ioptions_.compaction_options_universal.max_size_amplification_percent;

@@ -719,13 +724,20 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
                     cf_name.c_str(), file_num_buf);
  }

+  // output files at the bottom most level, unless it's reserved
+  int output_level = vstorage->num_levels() - 1;
+  // last level is reserved for the files ingested behind
+  if (ioptions_.allow_ingest_behind) {
+    assert(output_level > 1);
+    output_level--;
+  }
+
  return new Compaction(
      vstorage, ioptions_, mutable_cf_options, std::move(inputs),
-      vstorage->num_levels() - 1,
-      mutable_cf_options.MaxFileSizeForLevel(vstorage->num_levels() - 1),
+      output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
      GetCompressionType(ioptions_, vstorage, mutable_cf_options,
-                         vstorage->num_levels() - 1, 1),
+                         output_level, 1),
      /* grandparents */ {}, /* is manual */ false, score,
      false /* deletion_compaction */,
      CompactionReason::kUniversalSizeAmplification);
@@ -13,12 +13,12 @@
 namespace rocksdb {

 void CancelAllBackgroundWork(DB* db, bool wait) {
-  (dynamic_cast<DBImpl*>(db))->CancelAllBackgroundWork(wait);
+  (dynamic_cast<DBImpl*>(db->GetRootDB()))->CancelAllBackgroundWork(wait);
 }

 Status DeleteFilesInRange(DB* db, ColumnFamilyHandle* column_family,
                          const Slice* begin, const Slice* end) {
-  return (dynamic_cast<DBImpl*>(db))
+  return (dynamic_cast<DBImpl*>(db->GetRootDB()))
      ->DeleteFilesInRange(column_family, begin, end);
 }

@@ -320,8 +320,14 @@ TEST_F(CuckooTableDBTest, AdaptiveTable) {
 }  // namespace rocksdb

 int main(int argc, char** argv) {
-  ::testing::InitGoogleTest(&argc, argv);
-  return RUN_ALL_TESTS();
+  if (rocksdb::port::kLittleEndian) {
+    ::testing::InitGoogleTest(&argc, argv);
+    return RUN_ALL_TESTS();
+  }
+  else {
+    fprintf(stderr, "SKIPPED as Cuckoo table doesn't support Big Endian\n");
+    return 0;
+  }
 }

 #else
@@ -364,9 +364,9 @@ TEST_F(DBBasicTest, FLUSH) {
    ASSERT_OK(Flush(1));
    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v1"));

-    perf_context.Reset();
+    get_perf_context()->Reset();
    Get(1, "foo");
-    ASSERT_TRUE((int)perf_context.get_from_output_files_time > 0);
+    ASSERT_TRUE((int)get_perf_context()->get_from_output_files_time > 0);

    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_EQ("v1", Get(1, "foo"));
@@ -379,9 +379,9 @@ TEST_F(DBBasicTest, FLUSH) {

    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_EQ("v2", Get(1, "bar"));
-    perf_context.Reset();
+    get_perf_context()->Reset();
    ASSERT_EQ("v2", Get(1, "foo"));
-    ASSERT_TRUE((int)perf_context.get_from_output_files_time > 0);
+    ASSERT_TRUE((int)get_perf_context()->get_from_output_files_time > 0);

    writeOpt.disableWAL = false;
    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v3"));
@@ -686,12 +686,12 @@ class BloomStatsTestWithParam
    }
    options_.env = env_;

-    perf_context.Reset();
+    get_perf_context()->Reset();
    DestroyAndReopen(options_);
  }

  ~BloomStatsTestWithParam() {
-    perf_context.Reset();
+    get_perf_context()->Reset();
    Destroy(options_);
  }

@@ -724,33 +724,33 @@ TEST_P(BloomStatsTestWithParam, BloomStatsTest) {

  // check memtable bloom stats
  ASSERT_EQ(value1, Get(key1));
-  ASSERT_EQ(1, perf_context.bloom_memtable_hit_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_memtable_hit_count);
  ASSERT_EQ(value3, Get(key3));
-  ASSERT_EQ(2, perf_context.bloom_memtable_hit_count);
-  ASSERT_EQ(0, perf_context.bloom_memtable_miss_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_memtable_hit_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_memtable_miss_count);

  ASSERT_EQ("NOT_FOUND", Get(key2));
-  ASSERT_EQ(1, perf_context.bloom_memtable_miss_count);
-  ASSERT_EQ(2, perf_context.bloom_memtable_hit_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_memtable_miss_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_memtable_hit_count);

  // sanity checks
-  ASSERT_EQ(0, perf_context.bloom_sst_hit_count);
-  ASSERT_EQ(0, perf_context.bloom_sst_miss_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_sst_hit_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_sst_miss_count);

  Flush();

  // sanity checks
-  ASSERT_EQ(0, perf_context.bloom_sst_hit_count);
-  ASSERT_EQ(0, perf_context.bloom_sst_miss_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_sst_hit_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_sst_miss_count);

  // check SST bloom stats
  ASSERT_EQ(value1, Get(key1));
-  ASSERT_EQ(1, perf_context.bloom_sst_hit_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_sst_hit_count);
  ASSERT_EQ(value3, Get(key3));
-  ASSERT_EQ(2, perf_context.bloom_sst_hit_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_sst_hit_count);

  ASSERT_EQ("NOT_FOUND", Get(key2));
-  ASSERT_EQ(1, perf_context.bloom_sst_miss_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_sst_miss_count);
 }

 // Same scenario as in BloomStatsTest but using an iterator
@@ -771,21 +771,21 @@ TEST_P(BloomStatsTestWithParam, BloomStatsTestWithIter) {
  ASSERT_OK(iter->status());
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(value1, iter->value().ToString());
-  ASSERT_EQ(1, perf_context.bloom_memtable_hit_count);
-  ASSERT_EQ(0, perf_context.bloom_memtable_miss_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_memtable_hit_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_memtable_miss_count);

  iter->Seek(key3);
  ASSERT_OK(iter->status());
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(value3, iter->value().ToString());
-  ASSERT_EQ(2, perf_context.bloom_memtable_hit_count);
-  ASSERT_EQ(0, perf_context.bloom_memtable_miss_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_memtable_hit_count);
+  ASSERT_EQ(0, get_perf_context()->bloom_memtable_miss_count);

  iter->Seek(key2);
  ASSERT_OK(iter->status());
  ASSERT_TRUE(!iter->Valid());
-  ASSERT_EQ(1, perf_context.bloom_memtable_miss_count);
-  ASSERT_EQ(2, perf_context.bloom_memtable_hit_count);
+  ASSERT_EQ(1, get_perf_context()->bloom_memtable_miss_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_memtable_hit_count);

  Flush();

@@ -796,36 +796,19 @@ TEST_P(BloomStatsTestWithParam, BloomStatsTestWithIter) {
  ASSERT_OK(iter->status());
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(value1, iter->value().ToString());
-  if (partition_filters_) {
-    ASSERT_EQ(0, perf_context.bloom_sst_hit_count);   // no_io
-    ASSERT_EQ(0, perf_context.bloom_sst_miss_count);  // no_io
-  } else {
-    ASSERT_EQ(1, perf_context.bloom_sst_hit_count);
-  }
+  ASSERT_EQ(1, get_perf_context()->bloom_sst_hit_count);

  iter->Seek(key3);
  ASSERT_OK(iter->status());
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(value3, iter->value().ToString());
-  if (partition_filters_) {
-    ASSERT_EQ(0, perf_context.bloom_sst_hit_count);   // no_io
-    ASSERT_EQ(0, perf_context.bloom_sst_miss_count);  // no_io
-  } else {
-    ASSERT_EQ(2, perf_context.bloom_sst_hit_count);
-  }
+  ASSERT_EQ(2, get_perf_context()->bloom_sst_hit_count);

  iter->Seek(key2);
  ASSERT_OK(iter->status());
-  if (partition_filters_) {
-    // iter is still valid since filter did not reject the key2
-    ASSERT_TRUE(iter->Valid());
-    ASSERT_EQ(0, perf_context.bloom_sst_hit_count);   // no_io
-    ASSERT_EQ(0, perf_context.bloom_sst_miss_count);  // no_io
-  } else {
-    ASSERT_TRUE(!iter->Valid());
-    ASSERT_EQ(1, perf_context.bloom_sst_miss_count);
-    ASSERT_EQ(2, perf_context.bloom_sst_hit_count);
-  }
+  ASSERT_TRUE(!iter->Valid());
+  ASSERT_EQ(1, get_perf_context()->bloom_sst_miss_count);
+  ASSERT_EQ(2, get_perf_context()->bloom_sst_hit_count);
 }

 INSTANTIATE_TEST_CASE_P(BloomStatsTestWithParam, BloomStatsTestWithParam,
@@ -626,7 +626,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextManual) {
      iter->Next();
    }
    ASSERT_EQ(total, 700);
-    ASSERT_EQ(count, 2);
+    ASSERT_EQ(count, 1);
  }
 }
 #endif  // ROCKSDB_LITE
@@ -778,7 +778,7 @@ TEST_F(DBTestCompactionFilter, SkipUntil) {

  cfilter_skips = 0;
  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  // Numberof skips in tables: 2, 3, 3, 3.
+  // Number of skips in tables: 2, 3, 3, 3.
  ASSERT_EQ(11, cfilter_skips);

  for (int table = 0; table < 4; ++table) {
@@ -799,6 +799,43 @@ TEST_F(DBTestCompactionFilter, SkipUntil) {
  }
 }

+TEST_F(DBTestCompactionFilter, SkipUntilWithBloomFilter) {
+  BlockBasedTableOptions table_options;
+  table_options.whole_key_filtering = false;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(100, false));
+
+  Options options = CurrentOptions();
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  options.prefix_extractor.reset(NewCappedPrefixTransform(9));
+  options.compaction_filter_factory = std::make_shared<SkipEvenFilterFactory>();
+  options.disable_auto_compactions = true;
+  options.create_if_missing = true;
+  DestroyAndReopen(options);
+
+  Put("0000000010", "v10");
+  Put("0000000020", "v20"); // skipped
+  Put("0000000050", "v50");
+  Flush();
+
+  cfilter_skips = 0;
+  EXPECT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  EXPECT_EQ(1, cfilter_skips);
+
+  Status s;
+  std::string val;
+
+  s = db_->Get(ReadOptions(), "0000000010", &val);
+  ASSERT_OK(s);
+  EXPECT_EQ("v10", val);
+
+  s = db_->Get(ReadOptions(), "0000000020", &val);
+  EXPECT_TRUE(s.IsNotFound());
+
+  s = db_->Get(ReadOptions(), "0000000050", &val);
+  ASSERT_OK(s);
+  EXPECT_EQ("v50", val);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -519,7 +519,6 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
  options.level0_file_num_compaction_trigger = 2;
  options.level0_slowdown_writes_trigger = 20;
  options.soft_pending_compaction_bytes_limit = 1 << 30;  // Infinitely large
-  options.base_background_compactions = 1;
  options.max_background_compactions = 3;
  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));

@@ -1193,7 +1192,8 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
  }
 }

-TEST_F(DBCompactionTest, ManualPartialFill) {
+// Disable as the test is flaky.
+TEST_F(DBCompactionTest, DISABLED_ManualPartialFill) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -1227,6 +1227,9 @@ TEST_F(DBCompactionTest, ManualPartialFill) {
  options.max_background_compactions = 3;

  DestroyAndReopen(options);
+  // make sure all background compaction jobs can be scheduled
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();
  int32_t value_size = 10 * 1024;  // 10 KB

  // Add 2 non-overlapping files
@@ -1974,7 +1977,6 @@ TEST_P(DBCompactionTestWithParam, ManualCompaction) {

    if (iter == 0) {
      options = CurrentOptions();
-      options.max_background_flushes = 0;
      options.num_levels = 3;
      options.create_if_missing = true;
      options.statistics = rocksdb::CreateDBStatistics();
@@ -2420,15 +2422,12 @@ TEST_F(DBCompactionTest, SanitizeCompactionOptionsTest) {
  options.hard_pending_compaction_bytes_limit = 100;
  options.create_if_missing = true;
  DestroyAndReopen(options);
-  ASSERT_EQ(5, db_->GetOptions().base_background_compactions);
  ASSERT_EQ(100, db_->GetOptions().soft_pending_compaction_bytes_limit);

-  options.base_background_compactions = 4;
  options.max_background_compactions = 3;
  options.soft_pending_compaction_bytes_limit = 200;
  options.hard_pending_compaction_bytes_limit = 150;
  DestroyAndReopen(options);
-  ASSERT_EQ(3, db_->GetOptions().base_background_compactions);
  ASSERT_EQ(150, db_->GetOptions().soft_pending_compaction_bytes_limit);
 }

@@ -57,7 +57,6 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase) {
      Options options;
      options.env = env.get();
      options.create_if_missing = true;
-      options.db_write_buffer_size = 2048;
      options.write_buffer_size = 2048;
      options.max_write_buffer_number = 2;
      options.level0_file_num_compaction_trigger = 2;
@@ -127,7 +126,6 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {

  Options options = CurrentOptions();
  options.create_if_missing = true;
-  options.db_write_buffer_size = 204800;
  options.write_buffer_size = 20480;
  options.max_write_buffer_number = 2;
  options.level0_file_num_compaction_trigger = 2;
@@ -284,7 +282,6 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesCompactRange) {

  Options options = CurrentOptions();
  options.create_if_missing = true;
-  options.db_write_buffer_size = 2048;
  options.write_buffer_size = 2048;
  options.max_write_buffer_number = 2;
  options.level0_file_num_compaction_trigger = 2;
@@ -362,7 +359,6 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesCompactRange) {
 TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBaseInc) {
  Options options = CurrentOptions();
  options.create_if_missing = true;
-  options.db_write_buffer_size = 2048;
  options.write_buffer_size = 2048;
  options.max_write_buffer_number = 2;
  options.level0_file_num_compaction_trigger = 2;
@@ -375,6 +371,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBaseInc) {
  options.soft_rate_limit = 1.1;
  options.max_background_compactions = 2;
  options.num_levels = 5;
+  options.max_compaction_bytes = 100000000;

  DestroyAndReopen(options);

@@ -414,7 +411,6 @@ TEST_F(DBTestDynamicLevel, DISABLED_MigrateToDynamicLevelMaxBytesBase) {

  Options options;
  options.create_if_missing = true;
-  options.db_write_buffer_size = 2048;
  options.write_buffer_size = 2048;
  options.max_write_buffer_number = 8;
  options.level0_file_num_compaction_trigger = 4;
@@ -89,6 +89,43 @@ TEST_F(DBFlushTest, SyncFail) {
  Destroy(options);
 }

+TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
+  // Verify setting an empty high-pri (flush) thread pool causes flushes to be
+  // scheduled in the low-pri (compaction) thread pool.
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = 4;
+  options.memtable_factory.reset(new SpecialSkipListFactory(1));
+  Reopen(options);
+  env_->SetBackgroundThreads(0, Env::HIGH);
+
+  std::thread::id tid;
+  int num_flushes = 0, num_compactions = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BGWorkFlush", [&](void* arg) {
+        if (tid == std::thread::id()) {
+          tid = std::this_thread::get_id();
+        } else {
+          ASSERT_EQ(tid, std::this_thread::get_id());
+        }
+        ++num_flushes;
+      });
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BGWorkCompaction", [&](void* arg) {
+        ASSERT_EQ(tid, std::this_thread::get_id());
+        ++num_compactions;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(Put("key", "val"));
+  for (int i = 0; i < 4; ++i) {
+    ASSERT_OK(Put("key", "val"));
+    dbfull()->TEST_WaitForFlushMemTable();
+  }
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(4, num_flushes);
+  ASSERT_EQ(1, num_compactions);
+}
+
 TEST_P(DBFlushDirectIOTest, DirectIO) {
  Options options;
  options.create_if_missing = true;
@@ -16,9 +16,6 @@
 #ifdef OS_SOLARIS
 #include <alloca.h>
 #endif
-#ifdef ROCKSDB_JEMALLOC
-#include "jemalloc/jemalloc.h"
-#endif

 #include <algorithm>
 #include <climits>
@@ -44,6 +41,7 @@
 #include "db/job_context.h"
 #include "db/log_reader.h"
 #include "db/log_writer.h"
+#include "db/malloc_stats.h"
 #include "db/managed_iterator.h"
 #include "db/memtable.h"
 #include "db/memtable_list.h"
@@ -134,6 +132,8 @@ void DumpSupportInfo(Logger* logger) {
  ROCKS_LOG_HEADER(logger, "Fast CRC32 supported: %d",
                   crc32c::IsFastCrc32Supported());
 }
+
+int64_t kDefaultLowPriThrottledRate = 2 * 1024 * 1024;
 } // namespace

 DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
@@ -157,11 +157,13 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
      max_total_in_memory_state_(0),
      is_snapshot_supported_(true),
      write_buffer_manager_(immutable_db_options_.write_buffer_manager.get()),
-      write_thread_(immutable_db_options_.enable_write_thread_adaptive_yield
-                        ? immutable_db_options_.write_thread_max_yield_usec
-                        : 0,
-                    immutable_db_options_.write_thread_slow_yield_usec),
+      write_thread_(immutable_db_options_),
      write_controller_(mutable_db_options_.delayed_write_rate),
+      // Use delayed_write_rate as a base line to determine the initial
+      // low pri write rate limit. It may be adjusted later.
+      low_pri_write_rate_limiter_(NewGenericRateLimiter(std::min(
+          static_cast<int64_t>(mutable_db_options_.delayed_write_rate / 8),
+          kDefaultLowPriThrottledRate))),
      last_batch_group_size_(0),
      unscheduled_flushes_(0),
      unscheduled_compactions_(0),
@@ -190,9 +192,9 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)

  // Reserve ten files or so for other uses and give the rest to TableCache.
  // Give a large number for setting of "infinite" open files.
-  const int table_cache_size = (immutable_db_options_.max_open_files == -1)
-                                   ? 4194304
-                                   : immutable_db_options_.max_open_files - 10;
+  const int table_cache_size = (mutable_db_options_.max_open_files == -1)
+                                   ? TableCache::kInfiniteCapacity
+                                   : mutable_db_options_.max_open_files - 10;
  table_cache_ = NewLRUCache(table_cache_size,
                             immutable_db_options_.table_cache_numshardbits);

@@ -220,7 +222,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
      has_unpersisted_data_.load(std::memory_order_relaxed) &&
      !mutable_db_options_.avoid_flush_during_shutdown) {
    for (auto cfd : *versions_->GetColumnFamilySet()) {
-      if (!cfd->IsDropped() && !cfd->mem()->IsEmpty()) {
+      if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
        cfd->Ref();
        mutex_.Unlock();
        FlushMemTable(cfd, FlushOptions());
@@ -371,39 +373,6 @@ void DBImpl::PrintStatistics() {
  }
 }

-#ifndef ROCKSDB_LITE
-#ifdef ROCKSDB_JEMALLOC
-typedef struct {
-  char* cur;
-  char* end;
-} MallocStatus;
-
-static void GetJemallocStatus(void* mstat_arg, const char* status) {
-  MallocStatus* mstat = reinterpret_cast<MallocStatus*>(mstat_arg);
-  size_t status_len = status ? strlen(status) : 0;
-  size_t buf_size = (size_t)(mstat->end - mstat->cur);
-  if (!status_len || status_len > buf_size) {
-    return;
-  }
-
-  snprintf(mstat->cur, buf_size, "%s", status);
-  mstat->cur += status_len;
-}
-#endif  // ROCKSDB_JEMALLOC
-
-static void DumpMallocStats(std::string* stats) {
-#ifdef ROCKSDB_JEMALLOC
-  MallocStatus mstat;
-  const unsigned int kMallocStatusLen = 1000000;
-  std::unique_ptr<char[]> buf{new char[kMallocStatusLen + 1]};
-  mstat.cur = buf.get();
-  mstat.end = buf.get() + kMallocStatusLen;
-  je_malloc_stats_print(GetJemallocStatus, &mstat, "");
-  stats->append(buf.get());
-#endif  // ROCKSDB_JEMALLOC
-}
-#endif  // !ROCKSDB_LITE
-
 void DBImpl::MaybeDumpStats() {
  mutex_.Lock();
  unsigned int stats_dump_period_sec =
@@ -435,12 +404,17 @@ void DBImpl::MaybeDumpStats() {
      default_cf_internal_stats_->GetStringProperty(
          *db_property_info, DB::Properties::kDBStats, &stats);
      for (auto cfd : *versions_->GetColumnFamilySet()) {
-        cfd->internal_stats()->GetStringProperty(
-            *cf_property_info, DB::Properties::kCFStatsNoFileHistogram, &stats);
+        if (cfd->initialized()) {
+          cfd->internal_stats()->GetStringProperty(
+              *cf_property_info, DB::Properties::kCFStatsNoFileHistogram,
+              &stats);
+        }
      }
      for (auto cfd : *versions_->GetColumnFamilySet()) {
-        cfd->internal_stats()->GetStringProperty(
-            *cf_property_info, DB::Properties::kCFFileHistogram, &stats);
+        if (cfd->initialized()) {
+          cfd->internal_stats()->GetStringProperty(
+              *cf_property_info, DB::Properties::kCFFileHistogram, &stats);
+        }
      }
    }
    ROCKS_LOG_WARN(immutable_db_options_.info_log,
@@ -571,6 +545,9 @@ Status DBImpl::SetDBOptions(
      }

      write_controller_.set_max_delayed_write_rate(new_options.delayed_write_rate);
+      table_cache_.get()->SetCapacity(new_options.max_open_files == -1
+                                          ? TableCache::kInfiniteCapacity
+                                          : new_options.max_open_files - 10);

      mutable_db_options_ = new_options;

@@ -1234,6 +1211,8 @@ Status DBImpl::CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
        is_snapshot_supported_ = false;
      }

+      cfd->set_initialized();
+
      *handle = new ColumnFamilyHandleImpl(cfd, this, &mutex_);
      ROCKS_LOG_INFO(immutable_db_options_.info_log,
                     "Created column family [%s] (ID %u)",
@@ -1735,7 +1714,9 @@ bool DBImpl::GetIntPropertyInternal(ColumnFamilyData* cfd,
 Status DBImpl::ResetStats() {
  InstrumentedMutexLock l(&mutex_);
  for (auto* cfd : *versions_->GetColumnFamilySet()) {
-    cfd->internal_stats()->Clear();
+    if (cfd->initialized()) {
+      cfd->internal_stats()->Clear();
+    }
  }
  return Status::OK();
 }
@@ -1754,6 +1735,9 @@ bool DBImpl::GetAggregatedIntProperty(const Slice& property,
    InstrumentedMutexLock l(&mutex_);
    uint64_t value;
    for (auto* cfd : *versions_->GetColumnFamilySet()) {
+      if (!cfd->initialized()) {
+        continue;
+      }
      if (GetIntPropertyInternal(cfd, *property_info, true, &value)) {
        sum += value;
      } else {
@@ -2346,6 +2330,9 @@ Status DBImpl::WriteOptionsFile(bool need_mutex_lock,
      BuildDBOptions(immutable_db_options_, mutable_db_options_);
  mutex_.Unlock();

+  TEST_SYNC_POINT("DBImpl::WriteOptionsFile:1");
+  TEST_SYNC_POINT("DBImpl::WriteOptionsFile:2");
+
  std::string file_name =
      TempOptionsFileName(GetName(), versions_->NewFileNumber());
  Status s =
@@ -2481,7 +2468,7 @@ void DBImpl::EraseThreadStatusDbInfo() const {
 // A global method that can dump out the build version
 void DumpRocksDBBuildVersion(Logger * log) {
 #if !defined(IOS_CROSS_COMPILE)
-  // if we compile with Xcode, we don't run build_detect_vesion, so we don't
+  // if we compile with Xcode, we don't run build_detect_version, so we don't
  // generate util/build_version.cc
  ROCKS_LOG_HEADER(log, "RocksDB version: %d.%d.%d\n", ROCKSDB_MAJOR,
                   ROCKSDB_MINOR, ROCKSDB_PATCH);
@@ -2609,6 +2596,15 @@ Status DBImpl::IngestExternalFile(
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();

+  // Ingest should immediately fail if ingest_behind is requested,
+  // but the DB doesn't support it.
+  if (ingestion_options.ingest_behind) {
+    if (!immutable_db_options_.allow_ingest_behind) {
+      return Status::InvalidArgument(
+        "Can't ingest_behind file in DB with allow_ingest_behind=false");
+    }
+  }
+
  ExternalSstFileIngestionJob ingestion_job(env_, versions_.get(), cfd,
                                            immutable_db_options_, env_options_,
                                            &snapshots_, ingestion_options);
@@ -2652,7 +2648,8 @@ Status DBImpl::IngestExternalFile(
    if (status.ok()) {
      bool need_flush = false;
      status = ingestion_job.NeedsFlush(&need_flush);
-
+      TEST_SYNC_POINT_CALLBACK("DBImpl::IngestExternalFile:NeedFlush",
+                               &need_flush);
      if (status.ok() && need_flush) {
        mutex_.Unlock();
        status = FlushMemTable(cfd, FlushOptions(), true /* writes_stopped */);
@@ -387,9 +387,9 @@ class DBImpl : public DB {
  Status TEST_GetAllImmutableCFOptions(
      std::unordered_map<std::string, const ImmutableCFOptions*>* iopts_map);

-  // Return the lastest MutableCFOptions of of a column family
+  // Return the lastest MutableCFOptions of a column family
  Status TEST_GetLatestMutableCFOptions(ColumnFamilyHandle* column_family,
-                                        MutableCFOptions* mutable_cf_opitons);
+                                        MutableCFOptions* mutable_cf_options);

  Cache* TEST_table_cache() { return table_cache_.get(); }

@@ -399,12 +399,21 @@ class DBImpl : public DB {
  uint64_t TEST_FindMinPrepLogReferencedByMemTable();

  int TEST_BGCompactionsAllowed() const;
+  int TEST_BGFlushesAllowed() const;

 #endif  // NDEBUG

-  // Return maximum background compaction allowed to be scheduled based on
-  // compaction status.
-  int BGCompactionsAllowed() const;
+  struct BGJobLimits {
+    int max_flushes;
+    int max_compactions;
+  };
+  // Returns maximum background flushes and compactions allowed to be scheduled
+  BGJobLimits GetBGJobLimits() const;
+  // Need a static version that can be called during SanitizeOptions().
+  static BGJobLimits GetBGJobLimits(int max_background_flushes,
+                                    int max_background_compactions,
+                                    int max_background_jobs,
+                                    bool parallelize_compactions);

  // move logs pending closing from job_context to the DB queue and
  // schedule a purge
@@ -605,6 +614,11 @@ class DBImpl : public DB {
                   uint64_t* log_used = nullptr, uint64_t log_ref = 0,
                   bool disable_memtable = false);

+  Status PipelinedWriteImpl(const WriteOptions& options, WriteBatch* updates,
+                            WriteCallback* callback = nullptr,
+                            uint64_t* log_used = nullptr, uint64_t log_ref = 0,
+                            bool disable_memtable = false);
+
  uint64_t FindMinLogContainingOutstandingPrep();
  uint64_t FindMinPrepLogReferencedByMemTable();

@@ -706,6 +720,9 @@ class DBImpl : public DB {
  //            `num_bytes` going through.
  Status DelayWrite(uint64_t num_bytes, const WriteOptions& write_options);

+  Status ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
+                                      WriteBatch* my_batch);
+
  Status ScheduleFlushes(WriteContext* context);

  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context);
@@ -724,13 +741,19 @@ class DBImpl : public DB {
  Status HandleWriteBufferFull(WriteContext* write_context);

  // REQUIRES: mutex locked
-  Status PreprocessWrite(const WriteOptions& write_options, bool need_log_sync,
-                         bool* logs_getting_syned, WriteContext* write_context);
+  Status PreprocessWrite(const WriteOptions& write_options, bool* need_log_sync,
+                         WriteContext* write_context);

-  Status WriteToWAL(const autovector<WriteThread::Writer*>& write_group,
+  Status WriteToWAL(const WriteThread::WriteGroup& write_group,
                    log::Writer* log_writer, bool need_log_sync,
                    bool need_log_dir_sync, SequenceNumber sequence);

+  // Used by WriteImpl to update bg_error_ if paranoid check is enabled.
+  void ParanoidCheck(const Status& status);
+
+  // Used by WriteImpl to update bg_error_ in case of memtable insert error.
+  void MemTableInsertStatusCheck(const Status& memtable_insert_status);
+
 #ifndef ROCKSDB_LITE

  Status CompactFilesImpl(const CompactionOptions& compact_options,
@@ -919,6 +942,8 @@ class DBImpl : public DB {

  WriteController write_controller_;

+  unique_ptr<RateLimiter> low_pri_write_rate_limiter_;
+
  // Size of the last batch group. In slowdown mode, next write needs to
  // sleep if it uses up the quota.
  uint64_t last_batch_group_size_;
@@ -1092,7 +1117,7 @@ class DBImpl : public DB {
  // Indicate DB was opened successfully
  bool opened_successfully_;

-  // minmum log number still containing prepared data.
+  // minimum log number still containing prepared data.
  // this is used by FindObsoleteFiles to determine which
  // flushed logs we must keep around because they still
  // contain prepared data which has not been flushed or rolled back
@@ -1105,7 +1130,7 @@ class DBImpl : public DB {
  // to prepared_section_completed_ which maps LOG -> instance_count
  // since a log could contain multiple prepared sections
  //
-  // when trying to determine the minmum log still active we first
+  // when trying to determine the minimum log still active we first
  // consult min_log_with_prep_. while that root value maps to
  // a value > 0 in prepared_section_completed_ we decrement the
  // instance_count for that log and pop the root value in
@@ -1173,7 +1198,7 @@ extern DBOptions SanitizeOptions(const std::string& db, const DBOptions& src);
 extern CompressionType GetCompressionFlush(
    const ImmutableCFOptions& ioptions,
    const MutableCFOptions& mutable_cf_options);
- 
+
 // Fix user-supplied options to be reasonable
 template <class T, class V>
 static void ClipToRange(T* ptr, V minvalue, V maxvalue) {
@@ -282,10 +282,14 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
  if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal &&
      cfd->NumberLevels() > 1) {
    // Always compact all files together.
-    s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
-                            cfd->NumberLevels() - 1, options.target_path_id,
-                            begin, end, exclusive);
    final_output_level = cfd->NumberLevels() - 1;
+    // if bottom most level is reserved
+    if (immutable_db_options_.allow_ingest_behind) {
+      final_output_level--;
+    }
+    s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
+                            final_output_level, options.target_path_id,
+                            begin, end, exclusive);
  } else {
    for (int level = 0; level <= max_level_with_files; level++) {
      int output_level;
@@ -977,22 +981,22 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
    // DB is being deleted; no more background compactions
    return;
  }
-
-  while (unscheduled_flushes_ > 0 &&
-         bg_flush_scheduled_ < immutable_db_options_.max_background_flushes) {
+  auto bg_job_limits = GetBGJobLimits();
+  bool is_flush_pool_empty =
+    env_->GetBackgroundThreads(Env::Priority::HIGH) == 0;
+  while (!is_flush_pool_empty && unscheduled_flushes_ > 0 &&
+         bg_flush_scheduled_ < bg_job_limits.max_flushes) {
    unscheduled_flushes_--;
    bg_flush_scheduled_++;
    env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::HIGH, this);
  }

-  auto bg_compactions_allowed = BGCompactionsAllowed();
-
-  // special case -- if max_background_flushes == 0, then schedule flush on a
-  // compaction thread
-  if (immutable_db_options_.max_background_flushes == 0) {
+  // special case -- if high-pri (flush) thread pool is empty, then schedule
+  // flushes in low-pri (compaction) thread pool.
+  if (is_flush_pool_empty) {
    while (unscheduled_flushes_ > 0 &&
           bg_flush_scheduled_ + bg_compaction_scheduled_ <
-               bg_compactions_allowed) {
+               bg_job_limits.max_flushes) {
      unscheduled_flushes_--;
      bg_flush_scheduled_++;
      env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::LOW, this);
@@ -1010,7 +1014,7 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
    return;
  }

-  while (bg_compaction_scheduled_ < bg_compactions_allowed &&
+  while (bg_compaction_scheduled_ < bg_job_limits.max_compactions &&
         unscheduled_compactions_ > 0) {
    CompactionArg* ca = new CompactionArg;
    ca->db = this;
@@ -1022,13 +1026,35 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
  }
 }

-int DBImpl::BGCompactionsAllowed() const {
+DBImpl::BGJobLimits DBImpl::GetBGJobLimits() const {
  mutex_.AssertHeld();
-  if (write_controller_.NeedSpeedupCompaction()) {
-    return mutable_db_options_.max_background_compactions;
+  return GetBGJobLimits(immutable_db_options_.max_background_flushes,
+                        mutable_db_options_.max_background_compactions,
+                        mutable_db_options_.max_background_jobs,
+                        write_controller_.NeedSpeedupCompaction());
+}
+
+DBImpl::BGJobLimits DBImpl::GetBGJobLimits(int max_background_flushes,
+                                           int max_background_compactions,
+                                           int max_background_jobs,
+                                           bool parallelize_compactions) {
+  BGJobLimits res;
+  if (max_background_flushes == -1 && max_background_compactions == -1) {
+    // for our first stab implementing max_background_jobs, simply allocate a
+    // quarter of the threads to flushes.
+    res.max_flushes = std::max(1, max_background_jobs / 4);
+    res.max_compactions = std::max(1, max_background_jobs - res.max_flushes);
  } else {
-    return mutable_db_options_.base_background_compactions;
+    // compatibility code in case users haven't migrated to max_background_jobs,
+    // which automatically computes flush/compaction limits
+    res.max_flushes = std::max(1, max_background_flushes);
+    res.max_compactions = std::max(1, max_background_compactions);
  }
+  if (!parallelize_compactions) {
+    // throttle background compactions until we deem necessary
+    res.max_compactions = 1;
+  }
+  return res;
 }

 void DBImpl::AddToCompactionQueue(ColumnFamilyData* cfd) {
@@ -1150,13 +1176,15 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
  if (cfd != nullptr) {
    const MutableCFOptions mutable_cf_options =
        *cfd->GetLatestMutableCFOptions();
+    auto bg_job_limits = GetBGJobLimits();
    ROCKS_LOG_BUFFER(
        log_buffer,
        "Calling FlushMemTableToOutputFile with column "
-        "family [%s], flush slots available %d, compaction slots allowed %d, "
-        "compaction slots scheduled %d",
-        cfd->GetName().c_str(), immutable_db_options_.max_background_flushes,
-        bg_flush_scheduled_, BGCompactionsAllowed() - bg_compaction_scheduled_);
+        "family [%s], flush slots available %d, compaction slots available %d, "
+        "flush slots scheduled %d, compaction slots scheduled %d",
+        cfd->GetName().c_str(), bg_job_limits.max_flushes,
+        bg_job_limits.max_compactions, bg_flush_scheduled_,
+        bg_compaction_scheduled_);
    status = FlushMemTableToOutputFile(cfd, mutable_cf_options, made_progress,
                                       job_context, log_buffer);
    if (cfd->Unref()) {
@@ -1397,7 +1425,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
      // Can't compact right now, but try again later
      TEST_SYNC_POINT("DBImpl::BackgroundCompaction()::Conflict");

-      // Stay in the compaciton queue.
+      // Stay in the compaction queue.
      unscheduled_compactions_++;

      return Status::OK();
@@ -186,7 +186,12 @@ Status DBImpl::TEST_GetLatestMutableCFOptions(

 int DBImpl::TEST_BGCompactionsAllowed() const {
  InstrumentedMutexLock l(&mutex_);
-  return BGCompactionsAllowed();
+  return GetBGJobLimits().max_compactions;
+}
+
+int DBImpl::TEST_BGFlushesAllowed() const {
+  InstrumentedMutexLock l(&mutex_);
+  return GetBGJobLimits().max_flushes;
 }

 }  // namespace rocksdb
@@ -16,6 +16,7 @@
 #include "db/builder.h"
 #include "options/options_helper.h"
 #include "rocksdb/wal_filter.h"
+#include "util/rate_limiter.h"
 #include "util/sst_file_manager_impl.h"
 #include "util/sync_point.h"

@@ -36,7 +37,7 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
  if (result.max_open_files != -1) {
    int max_max_open_files = port::GetMaxOpenFiles();
    if (max_max_open_files == -1) {
-      max_max_open_files = 1000000;
+      max_max_open_files = 0x400000;
    }
    ClipToRange(&result.max_open_files, 20, max_max_open_files);
  }
@@ -53,15 +54,13 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
    result.write_buffer_manager.reset(
        new WriteBufferManager(result.db_write_buffer_size));
  }
-  if (result.base_background_compactions == -1) {
-    result.base_background_compactions = result.max_background_compactions;
-  }
-  if (result.base_background_compactions > result.max_background_compactions) {
-    result.base_background_compactions = result.max_background_compactions;
-  }
-  result.env->IncBackgroundThreadsIfNeeded(src.max_background_compactions,
+  auto bg_job_limits = DBImpl::GetBGJobLimits(result.max_background_flushes,
+                                              result.max_background_compactions,
+                                              result.max_background_jobs,
+                                              true /* parallelize_compactions */);
+  result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_compactions,
                                           Env::Priority::LOW);
-  result.env->IncBackgroundThreadsIfNeeded(src.max_background_flushes,
+  result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_flushes,
                                           Env::Priority::HIGH);

  if (result.rate_limiter.get() != nullptr) {
@@ -70,6 +69,15 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
    }
  }

+  if (result.delayed_write_rate == 0) {
+    if (result.rate_limiter.get() != nullptr) {
+      result.delayed_write_rate = result.rate_limiter->GetBytesPerSecond();
+    }
+    if (result.delayed_write_rate == 0) {
+      result.delayed_write_rate = 16 * 1024 * 1024;
+    }
+  }
+
  if (result.WAL_ttl_seconds > 0 || result.WAL_size_limit_MB > 0) {
    result.recycle_log_file_num = false;
  }
@@ -498,7 +506,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
    unique_ptr<SequentialFileReader> file_reader;
    {
      unique_ptr<SequentialFile> file;
-      status = env_->NewSequentialFile(fname, &file, env_options_);
+      status = env_->NewSequentialFile(fname, &file,
+                                       env_->OptimizeForLogRead(env_options_));
      if (!status.ok()) {
        MaybeIgnoreError(&status);
        if (!status.ok()) {
@@ -889,7 +898,7 @@ Status DB::Open(const Options& options, const std::string& dbname, DB** dbptr) {
  }
  return s;
 }
-  
+
 Status DB::Open(const DBOptions& db_options, const std::string& dbname,
                const std::vector<ColumnFamilyDescriptor>& column_families,
                std::vector<ColumnFamilyHandle*>* handles, DB** dbptr) {
@@ -65,6 +65,17 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  }

  Status status;
+  if (write_options.low_pri) {
+    status = ThrottleLowPriWritesIfNeeded(write_options, my_batch);
+    if (!status.ok()) {
+      return status;
+    }
+  }
+
+  if (immutable_db_options_.enable_pipelined_write) {
+    return PipelinedWriteImpl(write_options, my_batch, callback, log_used,
+                              log_ref, disable_memtable);
+  }

  PERF_TIMER_GUARD(write_pre_and_post_process_time);
  WriteThread::Writer w(write_options, my_batch, callback, log_ref,
@@ -77,25 +88,25 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  StopWatch write_sw(env_, immutable_db_options_.statistics.get(), DB_WRITE);

  write_thread_.JoinBatchGroup(&w);
-  if (w.state == WriteThread::STATE_PARALLEL_FOLLOWER) {
+  if (w.state == WriteThread::STATE_PARALLEL_MEMTABLE_WRITER) {
    // we are a non-leader in a parallel group
    PERF_TIMER_GUARD(write_memtable_time);

    if (w.ShouldWriteToMemtable()) {
      ColumnFamilyMemTablesImpl column_family_memtables(
          versions_->GetColumnFamilySet());
-      WriteBatchInternal::SetSequence(w.batch, w.sequence);
      w.status = WriteBatchInternal::InsertInto(
-          &w, &column_family_memtables, &flush_scheduler_,
+          &w, w.sequence, &column_family_memtables, &flush_scheduler_,
          write_options.ignore_missing_column_families, 0 /*log_number*/, this,
          true /*concurrent_memtable_writes*/);
    }

-    if (write_thread_.CompleteParallelWorker(&w)) {
-      // we're responsible for early exit
-      auto last_sequence = w.parallel_group->last_sequence;
+    if (write_thread_.CompleteParallelMemTableWriter(&w)) {
+      // we're responsible for exit batch group
+      auto last_sequence = w.write_group->last_sequence;
      versions_->SetLastSequence(last_sequence);
-      write_thread_.EarlyExitParallelGroup(&w);
+      MemTableInsertStatusCheck(w.status);
+      write_thread_.ExitAsBatchGroupFollower(&w);
    }
    assert(w.state == WriteThread::STATE_COMPLETED);
    // STATE_COMPLETED conditional below handles exit
@@ -117,16 +128,15 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  // when it finds suitable, and finish them in the same write batch.
  // This is how a write job could be done by the other writer.
  WriteContext write_context;
-  WriteThread::Writer* last_writer = &w;  // Dummy intial value
-  autovector<WriteThread::Writer*> write_group;
-  bool logs_getting_synced = false;
+  WriteThread::WriteGroup write_group;
+  bool in_parallel_group = false;
+  uint64_t last_sequence = versions_->LastSequence();

  mutex_.Lock();

  bool need_log_sync = !write_options.disableWAL && write_options.sync;
  bool need_log_dir_sync = need_log_sync && !log_dir_synced_;
-  status = PreprocessWrite(write_options, need_log_sync, &logs_getting_synced,
-                           &write_context);
+  status = PreprocessWrite(write_options, &need_log_sync, &write_context);
  log::Writer* cur_log_writer = logs_.back().writer;

  mutex_.Unlock();
@@ -136,9 +146,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  // and protects against concurrent loggers and concurrent writes
  // into memtables

-  bool exit_completed_early = false;
  last_batch_group_size_ =
-      write_thread_.EnterAsBatchGroupLeader(&w, &last_writer, &write_group);
+      write_thread_.EnterAsBatchGroupLeader(&w, &write_group);

  if (status.ok()) {
    // Rules for when we can update the memtable concurrently
@@ -153,32 +162,28 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    // relax rules 2 if we could prevent write batches from referring
    // more than once to a particular key.
    bool parallel = immutable_db_options_.allow_concurrent_memtable_write &&
-                    write_group.size() > 1;
+                    write_group.size > 1;
    int total_count = 0;
    uint64_t total_byte_size = 0;
-    for (auto writer : write_group) {
+    for (auto* writer : write_group) {
      if (writer->CheckCallback(this)) {
        if (writer->ShouldWriteToMemtable()) {
          total_count += WriteBatchInternal::Count(writer->batch);
          parallel = parallel && !writer->batch->HasMerge();
        }

-        if (writer->ShouldWriteToWAL()) {
-          total_byte_size = WriteBatchInternal::AppendedByteSize(
-              total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
-        }
+        total_byte_size = WriteBatchInternal::AppendedByteSize(
+            total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
      }
    }

-    uint64_t last_sequence = versions_->LastSequence();
    const SequenceNumber current_sequence = last_sequence + 1;
    last_sequence += total_count;

    // 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
-    // commit.  That lets us release our leader status early in
-    // some cases.
+    // commit.  That lets us release our leader status early.
    auto stats = default_cf_internal_stats_;
    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count);
    RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
@@ -186,7 +191,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
    stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
    RecordTick(stats_, WRITE_DONE_BY_SELF);
-    auto write_done_by_other = write_group.size() - 1;
+    auto write_done_by_other = write_group.size - 1;
    if (write_done_by_other > 0) {
      stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
                        write_done_by_other);
@@ -213,26 +218,23 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
      PERF_TIMER_GUARD(write_memtable_time);

      if (!parallel) {
-        status = WriteBatchInternal::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);
-
-        if (status.ok()) {
-          // There were no write failures. Set leader's status
-          // in case the write callback returned a non-ok status.
-          status = w.FinalStatus();
-        }
-
      } else {
-        WriteThread::ParallelGroup pg;
-        pg.leader = &w;
-        pg.last_writer = last_writer;
-        pg.last_sequence = last_sequence;
-        pg.early_exit_allowed = !need_log_sync;
-        pg.running.store(static_cast<uint32_t>(write_group.size()),
-                         std::memory_order_relaxed);
-        write_thread_.LaunchParallelFollowers(&pg, current_sequence);
+        SequenceNumber next_sequence = current_sequence;
+        for (auto* writer : write_group) {
+          if (writer->ShouldWriteToMemtable()) {
+            writer->sequence = next_sequence;
+            next_sequence += WriteBatchInternal::Count(writer->batch);
+          }
+        }
+        write_group.last_sequence = last_sequence;
+        write_group.running.store(static_cast<uint32_t>(write_group.size),
+                                  std::memory_order_relaxed);
+        write_thread_.LaunchParallelMemTableWriters(&write_group);
+        in_parallel_group = true;

        // Each parallel follower is doing each own writes. The leader should
        // also do its own.
@@ -240,73 +242,203 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
          ColumnFamilyMemTablesImpl column_family_memtables(
              versions_->GetColumnFamilySet());
          assert(w.sequence == current_sequence);
-          WriteBatchInternal::SetSequence(w.batch, w.sequence);
          w.status = WriteBatchInternal::InsertInto(
-              &w, &column_family_memtables, &flush_scheduler_,
+              &w, w.sequence, &column_family_memtables, &flush_scheduler_,
              write_options.ignore_missing_column_families, 0 /*log_number*/,
              this, true /*concurrent_memtable_writes*/);
        }
-
-        // CompleteParallelWorker returns true if this thread should
-        // handle exit, false means somebody else did
-        exit_completed_early = !write_thread_.CompleteParallelWorker(&w);
-        status = w.FinalStatus();
-      }
-
-      if (!exit_completed_early && w.status.ok()) {
-        versions_->SetLastSequence(last_sequence);
-        if (!need_log_sync) {
-          write_thread_.ExitAsBatchGroupLeader(&w, last_writer, w.status);
-          exit_completed_early = true;
-        }
-      }
-
-      // A non-OK status here indicates that the state implied by the
-      // WAL has diverged from the in-memory state.  This could be
-      // because of a corrupt write_batch (very bad), or because the
-      // client specified an invalid column family and didn't specify
-      // ignore_missing_column_families.
-      //
-      // Is setting bg_error_ enough here?  This will at least stop
-      // compaction and fail any further writes.
-      if (!status.ok() && !w.CallbackFailed()) {
-        mutex_.Lock();
-        if (bg_error_.ok()) {
-          bg_error_ = status;  // stop compaction & fail any further writes
-        }
-        mutex_.Unlock();
      }
    }
  }
  PERF_TIMER_START(write_pre_and_post_process_time);

+  if (!w.CallbackFailed()) {
+    ParanoidCheck(status);
+  }
+
+  if (need_log_sync) {
+    mutex_.Lock();
+    MarkLogsSynced(logfile_number_, need_log_dir_sync, status);
+    mutex_.Unlock();
+  }
+
+  bool should_exit_batch_group = true;
+  if (in_parallel_group) {
+    // CompleteParallelWorker returns true if this thread should
+    // handle exit, false means somebody else did
+    should_exit_batch_group = write_thread_.CompleteParallelMemTableWriter(&w);
+  }
+  if (should_exit_batch_group) {
+    versions_->SetLastSequence(last_sequence);
+    MemTableInsertStatusCheck(w.status);
+    write_thread_.ExitAsBatchGroupLeader(write_group, w.status);
+  }
+
+  if (status.ok()) {
+    status = w.FinalStatus();
+  }
+  return status;
+}
+
+Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
+                                  WriteBatch* my_batch, WriteCallback* callback,
+                                  uint64_t* log_used, uint64_t log_ref,
+                                  bool disable_memtable) {
+  PERF_TIMER_GUARD(write_pre_and_post_process_time);
+  StopWatch write_sw(env_, immutable_db_options_.statistics.get(), DB_WRITE);
+
+  WriteContext write_context;
+
+  WriteThread::Writer w(write_options, my_batch, callback, log_ref,
+                        disable_memtable);
+  write_thread_.JoinBatchGroup(&w);
+  if (w.state == WriteThread::STATE_GROUP_LEADER) {
+    WriteThread::WriteGroup wal_write_group;
+    if (w.callback && !w.callback->AllowWriteBatching()) {
+      write_thread_.WaitForMemTableWriters();
+    }
+    mutex_.Lock();
+    bool need_log_sync = !write_options.disableWAL && write_options.sync;
+    bool need_log_dir_sync = need_log_sync && !log_dir_synced_;
+    w.status = PreprocessWrite(write_options, &need_log_sync, &write_context);
+    log::Writer* cur_log_writer = logs_.back().writer;
+    mutex_.Unlock();
+
+    // This can set non-OK status if callback fail.
+    last_batch_group_size_ =
+        write_thread_.EnterAsBatchGroupLeader(&w, &wal_write_group);
+    const SequenceNumber current_sequence =
+        write_thread_.UpdateLastSequence(versions_->LastSequence()) + 1;
+    size_t total_count = 0;
+    size_t total_byte_size = 0;
+
+    if (w.status.ok()) {
+      SequenceNumber next_sequence = current_sequence;
+      for (auto writer : wal_write_group) {
+        if (writer->CheckCallback(this)) {
+          if (writer->ShouldWriteToMemtable()) {
+            writer->sequence = next_sequence;
+            size_t count = WriteBatchInternal::Count(writer->batch);
+            next_sequence += count;
+            total_count += count;
+          }
+          total_byte_size = WriteBatchInternal::AppendedByteSize(
+              total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
+        }
+      }
+      if (w.disable_wal) {
+        has_unpersisted_data_.store(true, std::memory_order_relaxed);
+      }
+      write_thread_.UpdateLastSequence(current_sequence + total_count - 1);
+    }
+
+    auto stats = default_cf_internal_stats_;
+    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count);
+    RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
+    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size);
+    RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
+
+    PERF_TIMER_STOP(write_pre_and_post_process_time);
+
+    if (w.ShouldWriteToWAL()) {
+      PERF_TIMER_GUARD(write_wal_time);
+      stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
+      RecordTick(stats_, WRITE_DONE_BY_SELF, 1);
+      if (wal_write_group.size > 1) {
+        stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
+                          wal_write_group.size - 1);
+        RecordTick(stats_, WRITE_DONE_BY_OTHER, wal_write_group.size - 1);
+      }
+      w.status = WriteToWAL(wal_write_group, cur_log_writer, need_log_sync,
+                            need_log_dir_sync, current_sequence);
+    }
+
+    if (!w.CallbackFailed()) {
+      ParanoidCheck(w.status);
+    }
+
+    if (need_log_sync) {
+      mutex_.Lock();
+      MarkLogsSynced(logfile_number_, need_log_dir_sync, w.status);
+      mutex_.Unlock();
+    }
+
+    write_thread_.ExitAsBatchGroupLeader(wal_write_group, w.status);
+  }
+
+  WriteThread::WriteGroup memtable_write_group;
+  if (w.state == WriteThread::STATE_MEMTABLE_WRITER_LEADER) {
+    PERF_TIMER_GUARD(write_memtable_time);
+    assert(w.status.ok());
+    write_thread_.EnterAsMemTableWriter(&w, &memtable_write_group);
+    if (memtable_write_group.size > 1 &&
+        immutable_db_options_.allow_concurrent_memtable_write) {
+      write_thread_.LaunchParallelMemTableWriters(&memtable_write_group);
+    } else {
+      memtable_write_group.status = WriteBatchInternal::InsertInto(
+          memtable_write_group, w.sequence, column_family_memtables_.get(),
+          &flush_scheduler_, write_options.ignore_missing_column_families,
+          0 /*log_number*/, this);
+      versions_->SetLastSequence(memtable_write_group.last_sequence);
+      write_thread_.ExitAsMemTableWriter(&w, memtable_write_group);
+    }
+  }
+
+  if (w.state == WriteThread::STATE_PARALLEL_MEMTABLE_WRITER) {
+    assert(w.ShouldWriteToMemtable());
+    ColumnFamilyMemTablesImpl column_family_memtables(
+        versions_->GetColumnFamilySet());
+    w.status = WriteBatchInternal::InsertInto(
+        &w, w.sequence, &column_family_memtables, &flush_scheduler_,
+        write_options.ignore_missing_column_families, 0 /*log_number*/, this,
+        true /*concurrent_memtable_writes*/);
+    if (write_thread_.CompleteParallelMemTableWriter(&w)) {
+      MemTableInsertStatusCheck(w.status);
+      versions_->SetLastSequence(w.write_group->last_sequence);
+      write_thread_.ExitAsMemTableWriter(&w, *w.write_group);
+    }
+  }
+
+  assert(w.state == WriteThread::STATE_COMPLETED);
+  if (log_used != nullptr) {
+    *log_used = w.log_used;
+  }
+
+  return w.FinalStatus();
+}
+
+void DBImpl::ParanoidCheck(const Status& status) {
+  // Is setting bg_error_ enough here?  This will at least stop
+  // compaction and fail any further writes.
  if (immutable_db_options_.paranoid_checks && !status.ok() &&
-      !w.CallbackFailed() && !status.IsBusy() && !status.IsIncomplete()) {
+      !status.IsBusy() && !status.IsIncomplete()) {
    mutex_.Lock();
    if (bg_error_.ok()) {
      bg_error_ = status;  // stop compaction & fail any further writes
    }
    mutex_.Unlock();
  }
+}

-  if (logs_getting_synced) {
+void DBImpl::MemTableInsertStatusCheck(const Status& status) {
+  // A non-OK status here indicates that the state implied by the
+  // WAL has diverged from the in-memory state.  This could be
+  // because of a corrupt write_batch (very bad), or because the
+  // client specified an invalid column family and didn't specify
+  // ignore_missing_column_families.
+  if (!status.ok()) {
    mutex_.Lock();
-    MarkLogsSynced(logfile_number_, need_log_dir_sync, status);
+    assert(bg_error_.ok());
+    bg_error_ = status;
    mutex_.Unlock();
  }
-
-  if (!exit_completed_early) {
-    write_thread_.ExitAsBatchGroupLeader(&w, last_writer, w.status);
-  }
-
-  return status;
 }

 Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
-                               bool need_log_sync, bool* logs_getting_synced,
+                               bool* need_log_sync,
                               WriteContext* write_context) {
  mutex_.AssertHeld();
-  assert(write_context != nullptr && logs_getting_synced != nullptr);
+  assert(write_context != nullptr && need_log_sync != nullptr);
  Status status;

  assert(!single_column_family_mode_ ||
@@ -326,7 +458,7 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
  }

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

  if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
@@ -343,7 +475,7 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
    status = DelayWrite(last_batch_group_size_, write_options);
  }

-  if (status.ok() && need_log_sync) {
+  if (status.ok() && *need_log_sync) {
    // Wait until the parallel syncs are finished. Any sync process has to sync
    // the front log too so it is enough to check the status of front()
    // We do a while loop since log_sync_cv_ is signalled when any sync is
@@ -363,26 +495,28 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
      // actually write to the WAL
      log.getting_synced = true;
    }
-    *logs_getting_synced = true;
+  } else {
+    *need_log_sync = false;
  }

  return status;
 }

-Status DBImpl::WriteToWAL(const autovector<WriteThread::Writer*>& write_group,
+Status DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
                          log::Writer* log_writer, bool need_log_sync,
                          bool need_log_dir_sync, SequenceNumber sequence) {
  Status status;

  WriteBatch* merged_batch = nullptr;
  size_t write_with_wal = 0;
-  if (write_group.size() == 1 && write_group[0]->ShouldWriteToWAL() &&
-      write_group[0]->batch->GetWalTerminationPoint().is_cleared()) {
+  auto* leader = write_group.leader;
+  if (write_group.size == 1 && leader->ShouldWriteToWAL() &&
+      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,
    // and the batch is not wanting to be truncated
-    merged_batch = write_group[0]->batch;
-    write_group[0]->log_used = logfile_number_;
+    merged_batch = leader->batch;
+    leader->log_used = logfile_number_;
    write_with_wal = 1;
  } else {
    // WAL needs all of the batches flattened into a single batch.
@@ -612,6 +746,34 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
  return bg_error_;
 }

+Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
+                                            WriteBatch* my_batch) {
+  assert(write_options.low_pri);
+  // This is called outside the DB mutex. Although it is safe to make the call,
+  // the consistency condition is not guaranteed to hold. It's OK to live with
+  // it in this case.
+  // If we need to speed compaction, it means the compaction is left behind
+  // and we start to limit low pri writes to a limit.
+  if (write_controller_.NeedSpeedupCompaction()) {
+    if (allow_2pc() && (my_batch->HasCommit() || my_batch->HasRollback())) {
+      // For 2PC, we only rate limit prepare, not commit.
+      return Status::OK();
+    }
+    if (write_options.no_slowdown) {
+      return Status::Incomplete();
+    } else {
+      assert(my_batch != nullptr);
+      // Rate limit those writes. The reason that we don't completely wait
+      // is that in case the write is heavy, low pri writes may never have
+      // a chance to run. Now we guarantee we are still slowly making
+      // progress.
+      write_controller_.low_pri_rate_limiter()->Request(my_batch->GetDataSize(),
+                                                        Env::IO_HIGH, nullptr);
+    }
+  }
+  return Status::OK();
+}
+
 Status DBImpl::ScheduleFlushes(WriteContext* context) {
  ColumnFamilyData* cfd;
  while ((cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
@@ -650,6 +812,12 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  log::Writer* new_log = nullptr;
  MemTable* new_mem = nullptr;

+  // In case of pipelined write is enabled, wait for all pending memtable
+  // writers.
+  if (immutable_db_options_.enable_pipelined_write) {
+    write_thread_.WaitForMemTableWriters();
+  }
+
  // 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);
@@ -560,7 +560,7 @@ void DBIter::MergeValuesNewToOld() {
      const Slice val = iter_->value();
      s = MergeHelper::TimedFullMerge(
          merge_operator_, ikey.user_key, &val, merge_context_.GetOperands(),
-          &saved_value_, logger_, statistics_, env_, &pinned_value_);
+          &saved_value_, logger_, statistics_, env_, &pinned_value_, true);
      if (!s.ok()) {
        status_ = s;
      }
@@ -585,7 +585,7 @@ void DBIter::MergeValuesNewToOld() {
  s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetUserKey(),
                                  nullptr, merge_context_.GetOperands(),
                                  &saved_value_, logger_, statistics_, env_,
-                                  &pinned_value_);
+                                  &pinned_value_, true);
  if (!s.ok()) {
    status_ = s;
  }
@@ -804,13 +804,13 @@ bool DBIter::FindValueForCurrentKey() {
        s = MergeHelper::TimedFullMerge(
            merge_operator_, saved_key_.GetUserKey(), nullptr,
            merge_context_.GetOperands(), &saved_value_, logger_, statistics_,
-            env_, &pinned_value_);
+            env_, &pinned_value_, true);
      } else {
        assert(last_not_merge_type == kTypeValue);
        s = MergeHelper::TimedFullMerge(
            merge_operator_, saved_key_.GetUserKey(), &pinned_value_,
            merge_context_.GetOperands(), &saved_value_, logger_, statistics_,
-            env_, &pinned_value_);
+            env_, &pinned_value_, true);
      }
      break;
    case kTypeValue:
@@ -882,7 +882,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
    s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetUserKey(),
                                    nullptr, merge_context_.GetOperands(),
                                    &saved_value_, logger_, statistics_, env_,
-                                    &pinned_value_);
+                                    &pinned_value_, true);
    // Make iter_ valid and point to saved_key_
    if (!iter_->Valid() ||
        !user_comparator_->Equal(ikey.user_key, saved_key_.GetUserKey())) {
@@ -900,7 +900,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetUserKey(),
                                  &val, merge_context_.GetOperands(),
                                  &saved_value_, logger_, statistics_, env_,
-                                  &pinned_value_);
+                                  &pinned_value_, true);
  valid_ = true;
  if (!s.ok()) {
    status_ = s;
@@ -352,11 +352,11 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    SetPerfLevel(kEnableCount);
    ASSERT_TRUE(GetPerfLevel() == kEnableCount);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    db_iter->SeekToLast();

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

    SetPerfLevel(kDisable);
@@ -471,11 +471,11 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    SetPerfLevel(kEnableCount);
    ASSERT_TRUE(GetPerfLevel() == kEnableCount);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    db_iter->SeekToLast();

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

    SetPerfLevel(kDisable);
@@ -972,11 +972,11 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    ASSERT_TRUE(iter->Valid());
    ASSERT_EQ(iter->key().compare(("b1")), 0);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    iter->Next();

    ASSERT_TRUE(iter->Valid());
-    ASSERT_EQ(static_cast<int>(perf_context.internal_delete_skipped_count), 2);
+    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_delete_skipped_count), 2);

    // now testing with iterate_bound
    Slice prefix("c");
@@ -984,7 +984,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {

    iter.reset(db_->NewIterator(ro));

-    perf_context.Reset();
+    get_perf_context()->Reset();

    iter->Seek("b");
    ASSERT_TRUE(iter->Valid());
@@ -999,12 +999,13 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    // even though the key is deleted
    // hence internal_delete_skipped_count should be 0
    ASSERT_TRUE(!iter->Valid());
-    ASSERT_EQ(static_cast<int>(perf_context.internal_delete_skipped_count), 0);
+    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_delete_skipped_count), 0);
  }
 }

 TEST_F(DBIteratorTest, DBIteratorBoundOptimizationTest) {
  int upper_bound_hits = 0;
+  Options options = CurrentOptions();
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "BlockBasedTable::BlockEntryIteratorState::KeyReachedUpperBound",
      [&upper_bound_hits](void* arg) {
@@ -1012,7 +1013,6 @@ TEST_F(DBIteratorTest, DBIteratorBoundOptimizationTest) {
        upper_bound_hits += (*static_cast<bool*>(arg) ? 1 : 0);
      });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  Options options = CurrentOptions();
  options.env = env_;
  options.create_if_missing = true;
  options.prefix_extractor = nullptr;
@@ -1886,7 +1886,7 @@ TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {
 #endif

  // Seek iterator to a smaller key.
-  perf_context.Reset();
+  get_perf_context()->Reset();
  iter->Seek("a");
  ASSERT_TRUE(iter->Valid());
  EXPECT_EQ("b", iter->key().ToString());
@@ -1894,17 +1894,17 @@ TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {

  // Check that the seek didn't do too much work.
  // Checks are not tight, just make sure that everything is well below 100.
-  EXPECT_LT(perf_context.internal_key_skipped_count, 4);
-  EXPECT_LT(perf_context.internal_recent_skipped_count, 8);
-  EXPECT_LT(perf_context.seek_on_memtable_count, 10);
-  EXPECT_LT(perf_context.next_on_memtable_count, 10);
-  EXPECT_LT(perf_context.prev_on_memtable_count, 10);
+  EXPECT_LT(get_perf_context()->internal_key_skipped_count, 4);
+  EXPECT_LT(get_perf_context()->internal_recent_skipped_count, 8);
+  EXPECT_LT(get_perf_context()->seek_on_memtable_count, 10);
+  EXPECT_LT(get_perf_context()->next_on_memtable_count, 10);
+  EXPECT_LT(get_perf_context()->prev_on_memtable_count, 10);

  // Check that iterator did something like what we expect.
-  EXPECT_EQ(perf_context.internal_delete_skipped_count, 0);
-  EXPECT_EQ(perf_context.internal_merge_count, 0);
-  EXPECT_GE(perf_context.internal_recent_skipped_count, 2);
-  EXPECT_GE(perf_context.seek_on_memtable_count, 2);
+  EXPECT_EQ(get_perf_context()->internal_delete_skipped_count, 0);
+  EXPECT_EQ(get_perf_context()->internal_merge_count, 0);
+  EXPECT_GE(get_perf_context()->internal_recent_skipped_count, 2);
+  EXPECT_GE(get_perf_context()->seek_on_memtable_count, 2);
  EXPECT_EQ(1, options.statistics->getTickerCount(
                 NUMBER_OF_RESEEKS_IN_ITERATION));
 }
@@ -21,7 +21,7 @@ class DBMemTableTest : public DBTestBase {

 class MockMemTableRep : public MemTableRep {
 public:
-  explicit MockMemTableRep(MemTableAllocator* allocator, MemTableRep* rep)
+  explicit MockMemTableRep(Allocator* allocator, MemTableRep* rep)
      : MemTableRep(allocator), rep_(rep), num_insert_with_hint_(0) {}

  virtual KeyHandle Allocate(const size_t len, char** buf) override {
@@ -72,7 +72,7 @@ class MockMemTableRep : public MemTableRep {
 class MockMemTableRepFactory : public MemTableRepFactory {
 public:
  virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator& cmp,
-                                         MemTableAllocator* allocator,
+                                         Allocator* allocator,
                                         const SliceTransform* transform,
                                         Logger* logger) override {
    SkipListFactory factory;
@@ -82,14 +82,27 @@ class MockMemTableRepFactory : public MemTableRepFactory {
    return mock_rep_;
  }

+  virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator& cmp,
+                                         Allocator* allocator,
+                                         const SliceTransform* transform,
+                                         Logger* logger,
+                                         uint32_t column_family_id) override {
+    last_column_family_id_ = column_family_id;
+    return CreateMemTableRep(cmp, allocator, transform, logger);
+  }
+
  virtual const char* Name() const override { return "MockMemTableRepFactory"; }

  MockMemTableRep* rep() { return mock_rep_; }

  bool IsInsertConcurrentlySupported() const override { return false; }

+  uint32_t GetLastColumnFamilyId() { return last_column_family_id_; }
+
 private:
  MockMemTableRep* mock_rep_;
+  // workaround since there's no port::kMaxUint32 yet.
+  uint32_t last_column_family_id_ = static_cast<uint32_t>(-1);
 };

 class TestPrefixExtractor : public SliceTransform {
@@ -155,6 +168,24 @@ TEST_F(DBMemTableTest, InsertWithHint) {
  ASSERT_EQ("vvv", Get("whitelisted"));
 }

+TEST_F(DBMemTableTest, ColumnFamilyId) {
+  // Verifies MemTableRepFactory is told the right column family id.
+  Options options;
+  options.allow_concurrent_memtable_write = false;
+  options.create_if_missing = true;
+  options.memtable_factory.reset(new MockMemTableRepFactory());
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  for (int cf = 0; cf < 2; ++cf) {
+    ASSERT_OK(Put(cf, "key", "val"));
+    ASSERT_OK(Flush(cf));
+    ASSERT_EQ(
+        cf, static_cast<MockMemTableRepFactory*>(options.memtable_factory.get())
+                ->GetLastColumnFamilyId());
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -18,28 +18,6 @@ class DBMergeOperatorTest : public DBTestBase {
  DBMergeOperatorTest() : DBTestBase("/db_merge_operator_test") {}
 };

-// A test merge operator mimics put but also fails if one of merge operands is
-// "corrupted".
-class TestPutOperator : public MergeOperator {
- public:
-  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
-                           MergeOperationOutput* merge_out) const override {
-    if (merge_in.existing_value != nullptr &&
-        *(merge_in.existing_value) == "corrupted") {
-      return false;
-    }
-    for (auto value : merge_in.operand_list) {
-      if (value == "corrupted") {
-        return false;
-      }
-    }
-    merge_out->existing_operand = merge_in.operand_list.back();
-    return true;
-  }
-
-  virtual const char* Name() const override { return "TestPutOperator"; }
-};
-
 TEST_F(DBMergeOperatorTest, MergeErrorOnRead) {
  Options options;
  options.create_if_missing = true;
@@ -15,7 +15,9 @@
 #include "db/db_test_util.h"
 #include "options/options_helper.h"
 #include "port/stack_trace.h"
+#include "rocksdb/cache.h"
 #include "rocksdb/convenience.h"
+#include "rocksdb/rate_limiter.h"
 #include "util/random.h"
 #include "util/sync_point.h"
 #include "util/testutil.h"
@@ -251,18 +253,44 @@ TEST_F(DBOptionsTest, SetOptionsMayTriggerCompaction) {
 TEST_F(DBOptionsTest, SetBackgroundCompactionThreads) {
  Options options;
  options.create_if_missing = true;
-  options.base_background_compactions = 1;  // default value
  options.max_background_compactions = 1;   // default value
  options.env = env_;
  Reopen(options);
  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
-  ASSERT_OK(dbfull()->SetDBOptions({{"base_background_compactions", "2"},
-                                    {"max_background_compactions", "3"}}));
-  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_OK(dbfull()->SetDBOptions({{"max_background_compactions", "3"}}));
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
  auto stop_token = dbfull()->TEST_write_controler().GetStopToken();
  ASSERT_EQ(3, dbfull()->TEST_BGCompactionsAllowed());
 }

+TEST_F(DBOptionsTest, SetBackgroundJobs) {
+  Options options;
+  options.create_if_missing = true;
+  options.max_background_jobs = 8;
+  options.env = env_;
+  Reopen(options);
+
+  for (int i = 0; i < 2; ++i) {
+    if (i > 0) {
+      options.max_background_jobs = 12;
+      ASSERT_OK(dbfull()->SetDBOptions(
+          {{"max_background_jobs",
+            std::to_string(options.max_background_jobs)}}));
+    }
+
+    ASSERT_EQ(options.max_background_jobs / 4,
+              dbfull()->TEST_BGFlushesAllowed());
+    ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
+
+    auto stop_token = dbfull()->TEST_write_controler().GetStopToken();
+
+    ASSERT_EQ(options.max_background_jobs / 4,
+              dbfull()->TEST_BGFlushesAllowed());
+    ASSERT_EQ(3 * options.max_background_jobs / 4,
+              dbfull()->TEST_BGCompactionsAllowed());
+  }
+}
+
 TEST_F(DBOptionsTest, AvoidFlushDuringShutdown) {
  Options options;
  options.create_if_missing = true;
@@ -383,6 +411,35 @@ TEST_F(DBOptionsTest, DeleteObsoleteFilesPeriodChange) {
  Close();
 }

+TEST_F(DBOptionsTest, MaxOpenFilesChange) {
+  SpecialEnv env(env_);
+  Options options;
+  options.max_open_files = -1;
+
+  Reopen(options);
+
+  Cache* tc = dbfull()->TEST_table_cache();
+
+  ASSERT_EQ(-1, dbfull()->GetDBOptions().max_open_files);
+  ASSERT_LT(2000, tc->GetCapacity());
+  ASSERT_OK(dbfull()->SetDBOptions({{"max_open_files", "1024"}}));
+  ASSERT_EQ(1024, dbfull()->GetDBOptions().max_open_files);
+  // examine the table cache (actual size should be 1014)
+  ASSERT_GT(1500, tc->GetCapacity());
+  Close();
+}
+
+TEST_F(DBOptionsTest, SanitizeDelayedWriteRate) {
+  Options options;
+  options.delayed_write_rate = 0;
+  Reopen(options);
+  ASSERT_EQ(16 * 1024 * 1024, dbfull()->GetDBOptions().delayed_write_rate);
+
+  options.rate_limiter.reset(NewGenericRateLimiter(31 * 1024 * 1024));
+  Reopen(options);
+  ASSERT_EQ(31 * 1024 * 1024, dbfull()->GetDBOptions().delayed_write_rate);
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -533,9 +533,9 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {
    ASSERT_TRUE(dbfull()->GetProperty(
        handles_[1], "rocksdb.num-entries-active-mem-table", &num));
    ASSERT_EQ(num, "1");
-    perf_context.Reset();
+    get_perf_context()->Reset();
    Get(1, "k1");
-    ASSERT_EQ(1, static_cast<int>(perf_context.get_from_memtable_count));
+    ASSERT_EQ(1, static_cast<int>(get_perf_context()->get_from_memtable_count));

    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "k2", big_value));
    ASSERT_TRUE(dbfull()->GetProperty(handles_[1],
@@ -548,12 +548,12 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {
        handles_[1], "rocksdb.num-entries-imm-mem-tables", &num));
    ASSERT_EQ(num, "1");

-    perf_context.Reset();
+    get_perf_context()->Reset();
    Get(1, "k1");
-    ASSERT_EQ(2, static_cast<int>(perf_context.get_from_memtable_count));
-    perf_context.Reset();
+    ASSERT_EQ(2, static_cast<int>(get_perf_context()->get_from_memtable_count));
+    get_perf_context()->Reset();
    Get(1, "k2");
-    ASSERT_EQ(1, static_cast<int>(perf_context.get_from_memtable_count));
+    ASSERT_EQ(1, static_cast<int>(get_perf_context()->get_from_memtable_count));

    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "k3", big_value));
    ASSERT_TRUE(dbfull()->GetProperty(
@@ -567,15 +567,15 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {
    ASSERT_TRUE(dbfull()->GetProperty(
        handles_[1], "rocksdb.num-entries-imm-mem-tables", &num));
    ASSERT_EQ(num, "2");
-    perf_context.Reset();
+    get_perf_context()->Reset();
    Get(1, "k2");
-    ASSERT_EQ(2, static_cast<int>(perf_context.get_from_memtable_count));
-    perf_context.Reset();
+    ASSERT_EQ(2, static_cast<int>(get_perf_context()->get_from_memtable_count));
+    get_perf_context()->Reset();
    Get(1, "k3");
-    ASSERT_EQ(1, static_cast<int>(perf_context.get_from_memtable_count));
-    perf_context.Reset();
+    ASSERT_EQ(1, static_cast<int>(get_perf_context()->get_from_memtable_count));
+    get_perf_context()->Reset();
    Get(1, "k1");
-    ASSERT_EQ(3, static_cast<int>(perf_context.get_from_memtable_count));
+    ASSERT_EQ(3, static_cast<int>(get_perf_context()->get_from_memtable_count));

    ASSERT_OK(Flush(1));
    ASSERT_TRUE(dbfull()->GetProperty(handles_[1],
@@ -670,7 +670,7 @@ TEST_F(DBPropertiesTest, DISABLED_GetProperty) {
  ASSERT_EQ(num, "0");
  ASSERT_TRUE(dbfull()->GetProperty("rocksdb.estimate-num-keys", &num));
  ASSERT_EQ(num, "1");
-  perf_context.Reset();
+  get_perf_context()->Reset();

  ASSERT_OK(dbfull()->Put(writeOpt, "k2", big_value));
  ASSERT_TRUE(dbfull()->GetProperty("rocksdb.num-immutable-mem-table", &num));
@@ -913,7 +913,7 @@ TEST_F(DBPropertiesTest, EstimatePendingCompBytes) {
  Flush();
  ASSERT_TRUE(dbfull()->GetIntProperty(
      "rocksdb.estimate-pending-compaction-bytes", &int_num));
-  ASSERT_EQ(int_num, 0U);
+  ASSERT_GT(int_num, 0U);

  ASSERT_OK(dbfull()->Put(writeOpt, "k3", big_value));
  Flush();
@@ -940,7 +940,6 @@ TEST_F(DBPropertiesTest, EstimateCompressionRatio) {
  Options options = CurrentOptions();
  options.compression_per_level = {kNoCompression, kSnappyCompression};
  options.disable_auto_compactions = true;
-  options.max_background_flushes = 0;
  options.num_levels = 2;
  Reopen(options);

@@ -1068,7 +1067,6 @@ class CountingDeleteTabPropCollectorFactory
 TEST_F(DBPropertiesTest, GetUserDefinedTableProperties) {
  Options options = CurrentOptions();
  options.level0_file_num_compaction_trigger = (1 << 30);
-  options.max_background_flushes = 0;
  options.table_properties_collector_factories.resize(1);
  std::shared_ptr<CountingUserTblPropCollectorFactory> collector_factory =
      std::make_shared<CountingUserTblPropCollectorFactory>(0);
@@ -1109,7 +1107,6 @@ TEST_F(DBPropertiesTest, GetUserDefinedTableProperties) {
 TEST_F(DBPropertiesTest, UserDefinedTablePropertiesContext) {
  Options options = CurrentOptions();
  options.level0_file_num_compaction_trigger = 3;
-  options.max_background_flushes = 0;
  options.table_properties_collector_factories.resize(1);
  std::shared_ptr<CountingUserTblPropCollectorFactory> collector_factory =
      std::make_shared<CountingUserTblPropCollectorFactory>(1);
@@ -1231,7 +1228,7 @@ TEST_F(DBPropertiesTest, TablePropertiesNeedCompactTest) {

  {
    SetPerfLevel(kEnableCount);
-    perf_context.Reset();
+    get_perf_context()->Reset();
    int c = 0;
    std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
    iter->Seek(Key(kMaxKey - 100));
@@ -1239,8 +1236,8 @@ TEST_F(DBPropertiesTest, TablePropertiesNeedCompactTest) {
      iter->Next();
    }
    ASSERT_EQ(c, 0);
-    ASSERT_LT(perf_context.internal_delete_skipped_count, 30u);
-    ASSERT_LT(perf_context.internal_key_skipped_count, 30u);
+    ASSERT_LT(get_perf_context()->internal_delete_skipped_count, 30u);
+    ASSERT_LT(get_perf_context()->internal_key_skipped_count, 30u);
    SetPerfLevel(kDisable);
  }
 }
@@ -1287,16 +1284,16 @@ TEST_F(DBPropertiesTest, NeedCompactHintPersistentTest) {
  ASSERT_EQ(NumTableFilesAtLevel(0), 0);
  {
    SetPerfLevel(kEnableCount);
-    perf_context.Reset();
+    get_perf_context()->Reset();
    int c = 0;
    std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
    for (iter->Seek(Key(0)); iter->Valid(); iter->Next()) {
      c++;
    }
    ASSERT_EQ(c, 2);
-    ASSERT_EQ(perf_context.internal_delete_skipped_count, 0);
+    ASSERT_EQ(get_perf_context()->internal_delete_skipped_count, 0);
    // We iterate every key twice. Is it a bug?
-    ASSERT_LE(perf_context.internal_key_skipped_count, 2);
+    ASSERT_LE(get_perf_context()->internal_key_skipped_count, 2);
    SetPerfLevel(kDisable);
  }
 }
@@ -789,6 +789,7 @@ TEST_F(DBRangeDelTest, IteratorIgnoresRangeDeletions) {
  db_->ReleaseSnapshot(snapshot);
 }

+#ifndef ROCKSDB_UBSAN_RUN
 TEST_F(DBRangeDelTest, TailingIteratorRangeTombstoneUnsupported) {
  db_->Put(WriteOptions(), "key", "val");
  // snapshot prevents key from being deleted during flush
@@ -815,6 +816,8 @@ TEST_F(DBRangeDelTest, TailingIteratorRangeTombstoneUnsupported) {
  }
  db_->ReleaseSnapshot(snapshot);
 }
+#endif  // !ROCKSDB_UBSAN_RUN
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -297,7 +297,7 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  std::vector<uint64_t> penalties;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DeleteScheduler::BackgroundEmptyTrash:Wait",
-      [&](void* arg) { penalties.push_back(*(static_cast<int*>(arg))); });
+      [&](void* arg) { penalties.push_back(*(static_cast<uint64_t*>(arg))); });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "InstrumentedCondVar::TimedWaitInternal", [&](void* arg) {
        // Turn timed wait into a simulated sleep
@@ -1,7 +1,7 @@
 //  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under the BSD-style license found in the
-//  LICENSE file in the root directory of this source tree. An additional grant
-//  of patent rights can be found in the PATENTS file in the same directory.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).

 #include <string>

@@ -60,7 +60,6 @@
 #include "util/file_reader_writer.h"
 #include "util/filename.h"
 #include "util/hash.h"
-#include "util/logging.h"
 #include "util/mutexlock.h"
 #include "util/rate_limiter.h"
 #include "util/string_util.h"
@@ -1439,7 +1438,6 @@ TEST_F(DBTest, UnremovableSingleDelete) {
 #ifndef ROCKSDB_LITE
 TEST_F(DBTest, DeletionMarkers1) {
  Options options = CurrentOptions();
-  options.max_background_flushes = 0;
  CreateAndReopenWithCF({"pikachu"}, options);
  Put(1, "foo", "v1");
  ASSERT_OK(Flush(1));
@@ -2765,13 +2763,56 @@ TEST_P(DBTestWithParam, FIFOCompactionTest) {
    }
  }
 }
+
+TEST_F(DBTest, FIFOCompactionTestWithCompaction) {
+  Options options;
+  options.compaction_style = kCompactionStyleFIFO;
+  options.write_buffer_size = 20 << 10;  // 20K
+  options.arena_block_size = 4096;
+  options.compaction_options_fifo.max_table_files_size = 1500 << 10;  // 1MB
+  options.compaction_options_fifo.allow_compaction = true;
+  options.level0_file_num_compaction_trigger = 6;
+  options.compression = kNoCompression;
+  options.create_if_missing = true;
+  options = CurrentOptions(options);
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+  for (int i = 0; i < 60; i++) {
+    // Generate and flush a file about 20KB.
+    for (int j = 0; j < 20; j++) {
+      ASSERT_OK(Put(ToString(i * 20 + j), RandomString(&rnd, 980)));
+    }
+    Flush();
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  }
+  // It should be compacted to 10 files.
+  ASSERT_EQ(NumTableFilesAtLevel(0), 10);
+
+  for (int i = 0; i < 60; i++) {
+    // Generate and flush a file about 10KB.
+    for (int j = 0; j < 20; j++) {
+      ASSERT_OK(Put(ToString(i * 20 + j + 2000), RandomString(&rnd, 980)));
+    }
+    Flush();
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  }
+
+  // It should be compacted to no more than 20 files.
+  ASSERT_GT(NumTableFilesAtLevel(0), 10);
+  ASSERT_LT(NumTableFilesAtLevel(0), 18);
+  // Size limit is still guaranteed.
+  ASSERT_LE(SizeAtLevel(0),
+            options.compaction_options_fifo.max_table_files_size);
+}
 #endif  // ROCKSDB_LITE

 #ifndef ROCKSDB_LITE
 /*
 * This test is not reliable enough as it heavily depends on disk behavior.
+ * Disable as it is flaky.
 */
-TEST_F(DBTest, RateLimitingTest) {
+TEST_F(DBTest, DISABLED_RateLimitingTest) {
  Options options = CurrentOptions();
  options.write_buffer_size = 1 << 20;  // 1MB
  options.level0_file_num_compaction_trigger = 2;
@@ -3030,9 +3071,9 @@ TEST_F(DBTest, DynamicMemtableOptions) {
                 Env::Priority::LOW);
  // Start from scratch and disable compaction/flush. Flush can only happen
  // during compaction but trigger is pretty high
-  options.max_background_flushes = 0;
  options.disable_auto_compactions = true;
  DestroyAndReopen(options);
+  env_->SetBackgroundThreads(0, Env::HIGH);

  // Put until writes are stopped, bounded by 256 puts. We should see stop at
  // ~128KB
@@ -3309,7 +3350,6 @@ TEST_P(DBTestWithParam, ThreadStatusSingleCompaction) {

 TEST_P(DBTestWithParam, PreShutdownManualCompaction) {
  Options options = CurrentOptions();
-  options.max_background_flushes = 0;
  options.max_subcompactions = max_subcompactions_;
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -3348,7 +3388,6 @@ TEST_P(DBTestWithParam, PreShutdownManualCompaction) {

    if (iter == 0) {
      options = CurrentOptions();
-      options.max_background_flushes = 0;
      options.num_levels = 3;
      options.create_if_missing = true;
      DestroyAndReopen(options);
@@ -3359,7 +3398,6 @@ TEST_P(DBTestWithParam, PreShutdownManualCompaction) {

 TEST_F(DBTest, PreShutdownFlush) {
  Options options = CurrentOptions();
-  options.max_background_flushes = 0;
  CreateAndReopenWithCF({"pikachu"}, options);
  ASSERT_OK(Put(1, "key", "value"));
  CancelAllBackgroundWork(db_);
@@ -3651,19 +3689,14 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
  Random rnd(301);
  Options options;
  options.create_if_missing = true;
-  options.db_write_buffer_size = 6000;
-  options.write_buffer_size = 6000;
+  options.db_write_buffer_size = 6000000;
+  options.write_buffer_size = 600000;
  options.max_write_buffer_number = 2;
  options.level0_file_num_compaction_trigger = 2;
  options.level0_slowdown_writes_trigger = 2;
  options.level0_stop_writes_trigger = 2;
  options.soft_pending_compaction_bytes_limit = 1024 * 1024;
-
-  // Use file size to distinguish levels
-  // L1: 10, L2: 20, L3 40, L4 80
-  // L0 is less than 30
-  options.target_file_size_base = 10;
-  options.target_file_size_multiplier = 2;
+  options.target_file_size_base = 20;

  options.level_compaction_dynamic_level_bytes = true;
  options.max_bytes_for_level_base = 200;
@@ -3700,10 +3733,11 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  for (int i = 0; i < 100; i++) {
-    ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 200)));
-
-    if (i % 25 == 0) {
-      dbfull()->TEST_WaitForFlushMemTable();
+    std::string value = RandomString(&rnd, 200);
+    ASSERT_OK(Put(Key(keys[i]), value));
+    if (i % 25 == 24) {
+      Flush();
+      dbfull()->TEST_WaitForCompact();
    }
  }

@@ -3744,7 +3778,8 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  for (int i = 101; i < 500; i++) {
-    ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 200)));
+    std::string value = RandomString(&rnd, 200);
+    ASSERT_OK(Put(Key(keys[i]), value));
    if (i % 100 == 99) {
      Flush();
      dbfull()->TEST_WaitForCompact();
@@ -5118,7 +5153,6 @@ TEST_F(DBTest, PauseBackgroundWorkTest) {
  // now it's done
  ASSERT_TRUE(done.load());
 }
-
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -165,20 +165,48 @@ TEST_F(DBTest2, MaxSuccessiveMergesChangeWithDBRecovery) {
 #ifndef ROCKSDB_LITE
 class DBTestSharedWriteBufferAcrossCFs
    : public DBTestBase,
-      public testing::WithParamInterface<bool> {
+      public testing::WithParamInterface<std::tuple<bool, bool>> {
 public:
  DBTestSharedWriteBufferAcrossCFs()
      : DBTestBase("/db_test_shared_write_buffer") {}
-  void SetUp() override { use_old_interface_ = GetParam(); }
+  void SetUp() override {
+    use_old_interface_ = std::get<0>(GetParam());
+    cost_cache_ = std::get<1>(GetParam());
+  }
  bool use_old_interface_;
+  bool cost_cache_;
 };

 TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
  Options options = CurrentOptions();
+  options.arena_block_size = 4096;
+
+  // Avoid undeterministic value by malloc_usable_size();
+  // Force arena block size to 1
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Arena::Arena:0", [&](void* arg) {
+        size_t* block_size = static_cast<size_t*>(arg);
+        *block_size = 1;
+      });
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Arena::AllocateNewBlock:0", [&](void* arg) {
+        std::pair<size_t*, size_t*>* pair =
+            static_cast<std::pair<size_t*, size_t*>*>(arg);
+        *std::get<0>(*pair) = *std::get<1>(*pair);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // The total soft write buffer size is about 105000
+  std::shared_ptr<Cache> cache = NewLRUCache(4 * 1024 * 1024, 2);
+  ASSERT_LT(cache->GetUsage(), 1024 * 1024);
+
  if (use_old_interface_) {
-    options.db_write_buffer_size = 100000;  // this is the real limit
+    options.db_write_buffer_size = 120000;  // this is the real limit
+  } else if (!cost_cache_) {
+    options.write_buffer_manager.reset(new WriteBufferManager(114285));
  } else {
-    options.write_buffer_manager.reset(new WriteBufferManager(100000));
+    options.write_buffer_manager.reset(new WriteBufferManager(114285, cache));
  }
  options.write_buffer_size = 500000;  // this is never hit
  CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
@@ -186,6 +214,13 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
  WriteOptions wo;
  wo.disableWAL = true;

+  std::function<void()> wait_flush = [&]() {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+    dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  };
+
  // Create some data and flush "default" and "nikitich" so that they
  // are newer CFs created.
  ASSERT_OK(Put(3, Key(1), DummyString(1), wo));
@@ -199,13 +234,20 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
            static_cast<uint64_t>(1));

  ASSERT_OK(Put(3, Key(1), DummyString(30000), wo));
+  if (cost_cache_) {
+    ASSERT_GE(cache->GetUsage(), 1024 * 1024);
+    ASSERT_LE(cache->GetUsage(), 2 * 1024 * 1024);
+  }
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(60000), wo));
+  if (cost_cache_) {
+    ASSERT_GE(cache->GetUsage(), 1024 * 1024);
+    ASSERT_LE(cache->GetUsage(), 2 * 1024 * 1024);
+  }
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
  // No flush should trigger
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  wait_flush();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(1));
@@ -219,11 +261,9 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {

  // Trigger a flush. Flushing "nikitich".
  ASSERT_OK(Put(3, Key(2), DummyString(30000), wo));
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  wait_flush();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(1));
@@ -237,12 +277,11 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {

  // Without hitting the threshold, no flush should trigger.
  ASSERT_OK(Put(2, Key(1), DummyString(30000), wo));
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  wait_flush();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(1));
@@ -257,14 +296,15 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
  // Hit the write buffer limit again. "default"
  // will have been flushed.
  ASSERT_OK(Put(2, Key(2), DummyString(10000), wo));
+  wait_flush();
  ASSERT_OK(Put(3, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  wait_flush();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(2));
@@ -279,13 +319,14 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
  // Trigger another flush. This time "dobrynia". "pikachu" should not
  // be flushed, althrough it was never flushed.
  ASSERT_OK(Put(1, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(80000), wo));
+  wait_flush();
  ASSERT_OK(Put(1, Key(1), DummyString(1), wo));
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  wait_flush();
+
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(2));
@@ -296,16 +337,45 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
              static_cast<uint64_t>(2));
  }
+  if (cost_cache_) {
+    ASSERT_GE(cache->GetUsage(), 1024 * 1024);
+    Close();
+    options.write_buffer_manager.reset();
+    ASSERT_LT(cache->GetUsage(), 1024 * 1024);
+  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

 INSTANTIATE_TEST_CASE_P(DBTestSharedWriteBufferAcrossCFs,
-                        DBTestSharedWriteBufferAcrossCFs, ::testing::Bool());
+                        DBTestSharedWriteBufferAcrossCFs,
+                        ::testing::Values(std::make_tuple(true, false),
+                                          std::make_tuple(false, false),
+                                          std::make_tuple(false, true)));

 TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
  std::string dbname2 = test::TmpDir(env_) + "/db_shared_wb_db2";
  Options options = CurrentOptions();
+  options.arena_block_size = 4096;
+  // Avoid undeterministic value by malloc_usable_size();
+  // Force arena block size to 1
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Arena::Arena:0", [&](void* arg) {
+        size_t* block_size = static_cast<size_t*>(arg);
+        *block_size = 1;
+      });
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Arena::AllocateNewBlock:0", [&](void* arg) {
+        std::pair<size_t*, size_t*>* pair =
+            static_cast<std::pair<size_t*, size_t*>*>(arg);
+        *std::get<0>(*pair) = *std::get<1>(*pair);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
  options.write_buffer_size = 500000;  // this is never hit
-  options.write_buffer_manager.reset(new WriteBufferManager(100000));
+  // Use a write buffer total size so that the soft limit is about
+  // 105000.
+  options.write_buffer_manager.reset(new WriteBufferManager(120000));
  CreateAndReopenWithCF({"cf1", "cf2"}, options);

  ASSERT_OK(DestroyDB(dbname2, options));
@@ -315,17 +385,25 @@ TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
  WriteOptions wo;
  wo.disableWAL = true;

+  std::function<void()> wait_flush = [&]() {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+    static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable();
+  };
+
  // Trigger a flush on cf2
  ASSERT_OK(Put(2, Key(1), DummyString(70000), wo));
+  wait_flush();
  ASSERT_OK(Put(0, Key(1), DummyString(20000), wo));
+  wait_flush();

  // Insert to DB2
  ASSERT_OK(db2->Put(wo, Key(2), DummyString(20000)));
+  wait_flush();

  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+  wait_flush();
  static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default") +
@@ -338,10 +416,9 @@ TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {

  // Triggering to flush another CF in DB1
  ASSERT_OK(db2->Put(wo, Key(2), DummyString(70000)));
+  wait_flush();
  ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+  wait_flush();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
              static_cast<uint64_t>(1));
@@ -355,10 +432,9 @@ TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {

  // Triggering flush in DB2.
  ASSERT_OK(db2->Put(wo, Key(3), DummyString(40000)));
+  wait_flush();
  ASSERT_OK(db2->Put(wo, Key(1), DummyString(1)));
-  dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
-  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+  wait_flush();
  static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable();
  {
    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
@@ -373,6 +449,8 @@ TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {

  delete db2;
  ASSERT_OK(DestroyDB(dbname2, options));
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

 namespace {
@@ -1080,7 +1158,6 @@ TEST_F(DBTest2, CompressionOptions) {
  options.max_bytes_for_level_multiplier = 2;
  options.num_levels = 7;
  options.max_background_compactions = 1;
-  options.base_background_compactions = 1;

  CompactionCompressionListener* listener =
      new CompactionCompressionListener(&options);
@@ -1157,6 +1234,9 @@ TEST_F(DBTest2, CompactionStall) {
  CompactionStallTestListener* listener = new CompactionStallTestListener();
  options.listeners.emplace_back(listener);
  DestroyAndReopen(options);
+  // make sure all background compaction jobs can be scheduled
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();

  Random rnd(301);

@@ -1516,6 +1596,7 @@ class MockPersistentCache : public PersistentCache {
  const size_t max_size_ = 10 * 1024;  // 10KiB
 };

+#ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
 TEST_F(DBTest2, PersistentCache) {
  int num_iter = 80;

@@ -1579,6 +1660,7 @@ TEST_F(DBTest2, PersistentCache) {
    }
  }
 }
+#endif // !OS_SOLARIS

 namespace {
 void CountSyncPoint() {
@@ -1716,6 +1798,7 @@ 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
@@ -1811,6 +1894,7 @@ TEST_F(DBTest2, ReadAmpBitmapLiveInCacheAfterDBClose) {
    }
  }
 }
+#endif // !OS_SOLARIS

 #ifndef ROCKSDB_LITE
 TEST_F(DBTest2, AutomaticCompactionOverlapManualCompaction) {
@@ -2100,6 +2184,56 @@ TEST_F(DBTest2, MemtableOnlyIterator) {
  ASSERT_EQ(1, count);
  delete it;
 }
+
+TEST_F(DBTest2, LowPriWrite) {
+  Options options = CurrentOptions();
+  // Compaction pressure should trigger since 6 files
+  options.level0_file_num_compaction_trigger = 4;
+  options.level0_slowdown_writes_trigger = 12;
+  options.level0_stop_writes_trigger = 30;
+  options.delayed_write_rate = 8 * 1024 * 1024;
+  Reopen(options);
+
+  std::atomic<int> rate_limit_count(0);
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "GenericRateLimiter::Request:1", [&](void* arg) {
+        rate_limit_count.fetch_add(1);
+        int64_t* rate_bytes_per_sec = static_cast<int64_t*>(arg);
+        ASSERT_EQ(1024 * 1024, *rate_bytes_per_sec);
+      });
+  // Block compaction
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"DBTest.LowPriWrite:0", "DBImpl::BGWorkCompaction"},
+  });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  WriteOptions wo;
+  for (int i = 0; i < 6; i++) {
+    wo.low_pri = false;
+    Put("", "", wo);
+    wo.low_pri = true;
+    Put("", "", wo);
+    Flush();
+  }
+  ASSERT_EQ(0, rate_limit_count.load());
+  wo.low_pri = true;
+  Put("", "", wo);
+  ASSERT_EQ(1, rate_limit_count.load());
+  wo.low_pri = false;
+  Put("", "", wo);
+  ASSERT_EQ(1, rate_limit_count.load());
+
+  TEST_SYNC_POINT("DBTest.LowPriWrite:0");
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  dbfull()->TEST_WaitForCompact();
+  wo.low_pri = true;
+  Put("", "", wo);
+  ASSERT_EQ(1, rate_limit_count.load());
+  wo.low_pri = false;
+  Put("", "", wo);
+  ASSERT_EQ(1, rate_limit_count.load());
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -43,9 +43,9 @@ SpecialEnv::SpecialEnv(Env* base)
 }

 DBTestBase::DBTestBase(const std::string path)
-    : option_config_(kDefault),
-      mem_env_(!getenv("MEM_ENV") ? nullptr : new MockEnv(Env::Default())),
-      env_(new SpecialEnv(mem_env_ ? mem_env_ : Env::Default())) {
+    : mem_env_(!getenv("MEM_ENV") ? nullptr : new MockEnv(Env::Default())),
+      env_(new SpecialEnv(mem_env_ ? mem_env_ : Env::Default())),
+      option_config_(kDefault) {
  env_->SetBackgroundThreads(1, Env::LOW);
  env_->SetBackgroundThreads(1, Env::HIGH);
  dbname_ = test::TmpDir(env_) + path;
@@ -245,27 +245,43 @@ bool DBTestBase::ChangeFilterOptions() {

 // Return the current option configuration.
 Options DBTestBase::CurrentOptions(
-    const anon::OptionsOverride& options_override) {
+    const anon::OptionsOverride& options_override) const {
+  return GetOptions(option_config_, GetDefaultOptions(), options_override);
+}
+
+Options DBTestBase::CurrentOptions(
+    const Options& default_options,
+    const anon::OptionsOverride& options_override) const {
+  return GetOptions(option_config_, default_options, options_override);
+}
+
+Options DBTestBase::GetDefaultOptions() {
  Options options;
  options.write_buffer_size = 4090 * 4096;
  options.target_file_size_base = 2 * 1024 * 1024;
  options.max_bytes_for_level_base = 10 * 1024 * 1024;
  options.max_open_files = 5000;
-  options.base_background_compactions = -1;
  options.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
  options.compaction_pri = CompactionPri::kByCompensatedSize;
-
-  return CurrentOptions(options, options_override);
+  return options;
 }

-Options DBTestBase::CurrentOptions(
-    const Options& defaultOptions,
-    const anon::OptionsOverride& options_override) {
+Options DBTestBase::GetOptions(
+    int option_config, const Options& default_options,
+    const anon::OptionsOverride& options_override) const {
  // this redundant copy is to minimize code change w/o having lint error.
-  Options options = defaultOptions;
+  Options options = default_options;
  BlockBasedTableOptions table_options;
  bool set_block_based_table_factory = true;
-  switch (option_config_) {
+#if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) &&  \
+  !defined(OS_AIX)
+  rocksdb::SyncPoint::GetInstance()->ClearCallBack(
+      "NewRandomAccessFile:O_DIRECT");
+  rocksdb::SyncPoint::GetInstance()->ClearCallBack(
+      "NewWritableFile:O_DIRECT");
+#endif
+
+  switch (option_config) {
 #ifndef ROCKSDB_LITE
    case kHashSkipList:
      options.prefix_extractor.reset(NewFixedPrefixTransform(1));
@@ -429,6 +445,30 @@ Options DBTestBase::CurrentOptions(
      options.enable_write_thread_adaptive_yield = true;
      break;
    }
+    case kDirectIO: {
+      options.use_direct_reads = true;
+      options.use_direct_io_for_flush_and_compaction = true;
+      options.compaction_readahead_size = 2 * 1024 * 1024;
+#if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
+    !defined(OS_AIX)
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "NewWritableFile:O_DIRECT", [&](void* arg) {
+            int* val = static_cast<int*>(arg);
+            *val &= ~O_DIRECT;
+          });
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
+            int* val = static_cast<int*>(arg);
+            *val &= ~O_DIRECT;
+          });
+      rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+#endif
+      break;
+    }
+    case kPipelinedWrite: {
+      options.enable_pipelined_write = true;
+      break;
+    }

    default:
      break;
@@ -49,10 +49,7 @@
 #include "util/mutexlock.h"

 #include "util/string_util.h"
-// SyncPoint is not supported in Released Windows Mode.
-#if !(defined NDEBUG) || !defined(OS_WIN)
 #include "util/sync_point.h"
-#endif  // !(defined NDEBUG) || !defined(OS_WIN)
 #include "util/testharness.h"
 #include "util/testutil.h"
 #include "utilities/merge_operators.h"
@@ -126,8 +123,8 @@ enum SkipPolicy { kSkipNone = 0, kSkipNoSnapshot = 1, kSkipNoPrefix = 2 };
 // A hacky skip list mem table that triggers flush after number of entries.
 class SpecialMemTableRep : public MemTableRep {
 public:
-  explicit SpecialMemTableRep(MemTableAllocator* allocator,
-                              MemTableRep* memtable, int num_entries_flush)
+  explicit SpecialMemTableRep(Allocator* allocator, MemTableRep* memtable,
+                              int num_entries_flush)
      : MemTableRep(allocator),
        memtable_(memtable),
        num_entries_flush_(num_entries_flush),
@@ -187,8 +184,9 @@ class SpecialSkipListFactory : public MemTableRepFactory {
  explicit SpecialSkipListFactory(int num_entries_flush)
      : num_entries_flush_(num_entries_flush) {}

+  using MemTableRepFactory::CreateMemTableRep;
  virtual MemTableRep* CreateMemTableRep(
-      const MemTableRep::KeyComparator& compare, MemTableAllocator* allocator,
+      const MemTableRep::KeyComparator& compare, Allocator* allocator,
      const SliceTransform* transform, Logger* logger) override {
    return new SpecialMemTableRep(
        allocator, factory_.CreateMemTableRep(compare, allocator, transform, 0),
@@ -586,10 +584,32 @@ class OnFileDeletionListener : public EventListener {
 };
 #endif

+// A test merge operator mimics put but also fails if one of merge operands is
+// "corrupted".
+class TestPutOperator : public MergeOperator {
+ public:
+  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
+                           MergeOperationOutput* merge_out) const override {
+    if (merge_in.existing_value != nullptr &&
+        *(merge_in.existing_value) == "corrupted") {
+      return false;
+    }
+    for (auto value : merge_in.operand_list) {
+      if (value == "corrupted") {
+        return false;
+      }
+    }
+    merge_out->existing_operand = merge_in.operand_list.back();
+    return true;
+  }
+
+  virtual const char* Name() const override { return "TestPutOperator"; }
+};
+
 class DBTestBase : public testing::Test {
- protected:
+ public:
  // Sequence of option configurations to try
-  enum OptionConfig {
+  enum OptionConfig : int {
    kDefault = 0,
    kBlockBasedTableWithPrefixHashIndex = 1,
    kBlockBasedTableWithWholeKeyHashIndex = 2,
@@ -620,14 +640,15 @@ class DBTestBase : public testing::Test {
    kRowCache = 27,
    kRecycleLogFiles = 28,
    kConcurrentSkipList = 29,
-    kEnd = 30,
-    kLevelSubcompactions = 31,
-    kUniversalSubcompactions = 32,
-    kBlockBasedTableWithIndexRestartInterval = 33,
-    kBlockBasedTableWithPartitionedIndex = 34,
-    kPartitionedFilterWithNewTableReaderForCompactions = 35,
+    kPipelinedWrite = 30,
+    kEnd = 31,
+    kDirectIO = 32,
+    kLevelSubcompactions = 33,
+    kUniversalSubcompactions = 34,
+    kBlockBasedTableWithIndexRestartInterval = 35,
+    kBlockBasedTableWithPartitionedIndex = 36,
+    kPartitionedFilterWithNewTableReaderForCompactions = 37,
  };
-  int option_config_;

 public:
  std::string dbname_;
@@ -638,6 +659,7 @@ class DBTestBase : public testing::Test {
  DB* db_;
  std::vector<ColumnFamilyHandle*> handles_;

+  int option_config_;
  Options last_options_;

  // Skip some options, as they may not be applicable to a specific test.
@@ -688,12 +710,19 @@ class DBTestBase : public testing::Test {
  bool ChangeFilterOptions();

  // Return the current option configuration.
-  Options CurrentOptions(
-      const anon::OptionsOverride& options_override = anon::OptionsOverride());
+  Options CurrentOptions(const anon::OptionsOverride& options_override =
+                             anon::OptionsOverride()) const;

-  Options CurrentOptions(
-      const Options& defaultOptions,
-      const anon::OptionsOverride& options_override = anon::OptionsOverride());
+  Options CurrentOptions(const Options& default_options,
+                         const anon::OptionsOverride& options_override =
+                             anon::OptionsOverride()) const;
+
+  static Options GetDefaultOptions();
+
+  Options GetOptions(int option_config,
+                     const Options& default_options = GetDefaultOptions(),
+                     const anon::OptionsOverride& options_override =
+                         anon::OptionsOverride()) const;

  DBImpl* dbfull() { return reinterpret_cast<DBImpl*>(db_); }

@@ -116,7 +116,7 @@ class DelayFilterFactory : public CompactionFilterFactory {
 };
 }  // namespace

-// Make sure we don't trigger a problem if the trigger conditon is given
+// Make sure we don't trigger a problem if the trigger condtion is given
 // to be 0, which is invalid.
 TEST_P(DBTestUniversalCompaction, UniversalCompactionSingleSortedRun) {
  Options options = CurrentOptions();
@@ -0,0 +1,80 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include <memory>
+#include <thread>
+#include <vector>
+#include "db/db_test_util.h"
+#include "db/write_batch_internal.h"
+#include "port/stack_trace.h"
+#include "util/sync_point.h"
+
+namespace rocksdb {
+
+// Test variations of WriteImpl.
+class DBWriteTest : public DBTestBase, public testing::WithParamInterface<int> {
+ public:
+  DBWriteTest() : DBTestBase("/db_write_test") {}
+
+  void Open() { DBTestBase::Reopen(GetOptions(GetParam())); }
+};
+
+// 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());
+    }
+  };
+  std::vector<port::Thread> threads;
+  for (size_t i = 0; i < kThreads; i++) {
+    threads.emplace_back(writer, i);
+  }
+  for (size_t i = 0; i < kThreads; i++) {
+    threads[i].join();
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(DBWriteTestInstance, DBWriteTest,
+                        testing::Values(DBTestBase::kDefault,
+                                        DBTestBase::kPipelinedWrite));
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -38,8 +38,10 @@ class ExternalSSTFileBasicTest : public DBTestBase {
  }

  Status GenerateAndAddExternalFile(
-      const Options options, std::vector<int> keys, int file_id,
+      const Options options, std::vector<int> keys,
+      const std::vector<ValueType>& value_types, int file_id,
      std::map<std::string, std::string>* true_data) {
+    assert(value_types.size() == 1 || keys.size() == value_types.size());
    std::string file_path = sst_files_dir_ + ToString(file_id);
    SstFileWriter sst_file_writer(EnvOptions(), options);

@@ -47,11 +49,28 @@ class ExternalSSTFileBasicTest : public DBTestBase {
    if (!s.ok()) {
      return s;
    }
-    for (int k : keys) {
-      std::string key = Key(k);
-      std::string value = Key(k) + ToString(file_id);
-      s = sst_file_writer.Add(key, value);
-      (*true_data)[key] = value;
+    for (size_t i = 0; i < keys.size(); i++) {
+      std::string key = Key(keys[i]);
+      std::string value = Key(keys[i]) + ToString(file_id);
+      ValueType value_type =
+          (value_types.size() == 1 ? value_types[0] : value_types[i]);
+      switch (value_type) {
+        case ValueType::kTypeValue:
+          s = sst_file_writer.Put(key, value);
+          (*true_data)[key] = value;
+          break;
+        case ValueType::kTypeMerge:
+          s = sst_file_writer.Merge(key, value);
+          // we only use TestPutOperator in this test
+          (*true_data)[key] = value;
+          break;
+        case ValueType::kTypeDeletion:
+          s = sst_file_writer.Delete(key);
+          true_data->erase(key);
+          break;
+        default:
+          return Status::InvalidArgument("Value type is not supported");
+      }
      if (!s.ok()) {
        sst_file_writer.Finish();
        return s;
@@ -64,10 +83,17 @@ class ExternalSSTFileBasicTest : public DBTestBase {
      ifo.allow_global_seqno = true;
      s = db_->IngestExternalFile({file_path}, ifo);
    }
-
    return s;
  }

+  Status GenerateAndAddExternalFile(
+      const Options options, std::vector<int> keys, const ValueType value_type,
+      int file_id, std::map<std::string, std::string>* true_data) {
+    return GenerateAndAddExternalFile(options, keys,
+                                      std::vector<ValueType>(1, value_type),
+                                      file_id, true_data);
+  }
+
  ~ExternalSSTFileBasicTest() { test::DestroyDir(env_, sst_files_dir_); }

 protected:
@@ -87,7 +113,7 @@ TEST_F(ExternalSSTFileBasicTest, Basic) {
  std::string file1 = sst_files_dir_ + "file1.sst";
  ASSERT_OK(sst_file_writer.Open(file1));
  for (int k = 0; k < 100; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file1_info;
  Status s = sst_file_writer.Finish(&file1_info);
@@ -101,7 +127,7 @@ TEST_F(ExternalSSTFileBasicTest, Basic) {
  ASSERT_EQ(file1_info.smallest_key, Key(0));
  ASSERT_EQ(file1_info.largest_key, Key(99));
  // sst_file_writer already finished, cannot add this value
-  s = sst_file_writer.Add(Key(100), "bad_val");
+  s = sst_file_writer.Put(Key(100), "bad_val");
  ASSERT_FALSE(s.ok()) << s.ToString();

  DestroyAndReopen(options);
@@ -126,7 +152,7 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
  std::string file1 = sst_files_dir_ + "file1.sst";
  ASSERT_OK(sst_file_writer.Open(file1));
  for (int k = 0; k < 100; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file1_info;
  Status s = sst_file_writer.Finish(&file1_info);
@@ -140,7 +166,7 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
  std::string file2 = sst_files_dir_ + "file2.sst";
  ASSERT_OK(sst_file_writer.Open(file2));
  for (int k = 100; k < 300; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file2_info;
  s = sst_file_writer.Finish(&file2_info);
@@ -154,7 +180,7 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
  std::string file3 = sst_files_dir_ + "file3.sst";
  ASSERT_OK(sst_file_writer.Open(file3));
  for (int k = 110; k < 125; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val_overlap"));
  }
  ExternalSstFileInfo file3_info;
  s = sst_file_writer.Finish(&file3_info);
@@ -182,92 +208,313 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
 }

 TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
-  Options options = CurrentOptions();
-  DestroyAndReopen(options);
-  std::map<std::string, std::string> true_data;
+  do {
+    Options options = CurrentOptions();
+    DestroyAndReopen(options);
+    std::map<std::string, std::string> true_data;

-  int file_id = 1;
+    int file_id = 1;

-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6}, file_id++,
-                                       &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
+                                         ValueType::kTypeValue, file_id++,
+                                         &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

-  ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13}, file_id++,
-                                       &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
+                                         ValueType::kTypeValue, file_id++,

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {1, 4, 6}, file_id++, &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
+                                         &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {11, 15, 19}, file_id++, &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 4, 6}, ValueType::kTypeValue, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {120, 130}, file_id++, &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {11, 15, 19}, ValueType::kTypeValue, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {1, 130}, file_id++, &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {120, 130}, ValueType::kTypeValue, file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

-  // Write some keys through normal write path
-  for (int i = 0; i < 50; i++) {
-    ASSERT_OK(Put(Key(i), "memtable"));
-    true_data[Key(i)] = "memtable";
-  }
-  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 130}, ValueType::kTypeValue, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {60, 61, 62}, file_id++, &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
+    // Write some keys through normal write path
+    for (int i = 0; i < 50; i++) {
+      ASSERT_OK(Put(Key(i), "memtable"));
+      true_data[Key(i)] = "memtable";
+    }
+    SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {40, 41, 42}, file_id++, &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {20, 30, 40}, file_id++, &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {40, 41, 42}, ValueType::kTypeValue, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);

-  const Snapshot* snapshot = db_->GetSnapshot();
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {20, 30, 40}, ValueType::kTypeValue, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);

-  // We will need a seqno for the file regardless if the file overwrite
-  // keys in the DB or not because we have a snapshot
-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {1000, 1002}, file_id++, &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
+    const Snapshot* snapshot = db_->GetSnapshot();

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {2000, 3002}, file_id++, &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
+    // We will need a seqno for the file regardless if the file overwrite
+    // keys in the DB or not because we have a snapshot
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1000, 1002}, ValueType::kTypeValue, file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);

-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
-                                       file_id++, &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {2000, 3002}, ValueType::kTypeValue, file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);

-  db_->ReleaseSnapshot(snapshot);
+    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
+                                         ValueType::kTypeValue, file_id++,
+                                         &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);

-  ASSERT_OK(
-      GenerateAndAddExternalFile(options, {5000, 5001}, file_id++, &true_data));
-  // No snapshot anymore, no need to assign a seqno
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+    db_->ReleaseSnapshot(snapshot);

-  size_t kcnt = 0;
-  VerifyDBFromMap(true_data, &kcnt, false);
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {5000, 5001}, ValueType::kTypeValue, file_id++, &true_data));
+    // No snapshot anymore, no need to assign a seqno
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+    size_t kcnt = 0;
+    VerifyDBFromMap(true_data, &kcnt, false);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
+  do {
+    Options options = CurrentOptions();
+    options.merge_operator.reset(new TestPutOperator());
+    DestroyAndReopen(options);
+    std::map<std::string, std::string> true_data;
+
+    int file_id = 1;
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
+                                         ValueType::kTypeValue, file_id++,
+                                         &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
+                                         ValueType::kTypeValue, file_id++,
+
+                                         &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 4, 6}, ValueType::kTypeMerge, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {11, 15, 19},
+                                         ValueType::kTypeDeletion, file_id++,
+                                         &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {120, 130}, ValueType::kTypeMerge, file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 130}, ValueType::kTypeDeletion, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+
+    // Write some keys through normal write path
+    for (int i = 0; i < 50; i++) {
+      ASSERT_OK(Put(Key(i), "memtable"));
+      true_data[Key(i)] = "memtable";
+    }
+    SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {40, 41, 42}, ValueType::kTypeMerge, file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {20, 30, 40},
+                                         ValueType::kTypeDeletion, file_id++,
+                                         &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
+
+    const Snapshot* snapshot = db_->GetSnapshot();
+
+    // We will need a seqno for the file regardless if the file overwrite
+    // keys in the DB or not because we have a snapshot
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1000, 1002}, ValueType::kTypeMerge, file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {2000, 3002}, ValueType::kTypeMerge, file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
+                                         ValueType::kTypeMerge, file_id++,
+                                         &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+    db_->ReleaseSnapshot(snapshot);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {5000, 5001}, ValueType::kTypeValue, file_id++, &true_data));
+    // No snapshot anymore, no need to assign a seqno
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+    size_t kcnt = 0;
+    VerifyDBFromMap(true_data, &kcnt, false);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
+  do {
+    Options options = CurrentOptions();
+    options.merge_operator.reset(new TestPutOperator());
+    DestroyAndReopen(options);
+    std::map<std::string, std::string> true_data;
+
+    int file_id = 1;
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 2, 3, 4, 5, 6},
+        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
+         ValueType::kTypeMerge, ValueType::kTypeValue, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {10, 11, 12, 13},
+        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
+         ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 4, 6}, {ValueType::kTypeDeletion, ValueType::kTypeValue,
+                             ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {11, 15, 19}, {ValueType::kTypeDeletion, ValueType::kTypeMerge,
+                                ValueType::kTypeValue},
+        file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {120, 130}, {ValueType::kTypeValue, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 130}, {ValueType::kTypeMerge, ValueType::kTypeDeletion},
+        file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+
+    // Write some keys through normal write path
+    for (int i = 0; i < 50; i++) {
+      ASSERT_OK(Put(Key(i), "memtable"));
+      true_data[Key(i)] = "memtable";
+    }
+    SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {60, 61, 62},
+        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
+        file_id++, &true_data));
+    // File dont overwrite any keys, No seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {40, 41, 42}, {ValueType::kTypeValue, ValueType::kTypeDeletion,
+                                ValueType::kTypeDeletion},
+        file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {20, 30, 40},
+        {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
+         ValueType::kTypeDeletion},
+        file_id++, &true_data));
+    // File overwrite some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
+
+    const Snapshot* snapshot = db_->GetSnapshot();
+
+    // We will need a seqno for the file regardless if the file overwrite
+    // keys in the DB or not because we have a snapshot
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1000, 1002}, {ValueType::kTypeValue, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {2000, 3002}, {ValueType::kTypeValue, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 20, 40, 100, 150},
+        {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
+         ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // A global seqno will be assigned anyway because of the snapshot
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+    db_->ReleaseSnapshot(snapshot);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {5000, 5001}, {ValueType::kTypeValue, ValueType::kTypeMerge},
+        file_id++, &true_data));
+    // No snapshot anymore, no need to assign a seqno
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+    size_t kcnt = 0;
+    VerifyDBFromMap(true_data, &kcnt, false);
+  } while (ChangeCompactOptions());
 }

 TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
@@ -276,7 +523,7 @@ TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {

  size_t total_fadvised_bytes = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "SstFileWriter::InvalidatePageCache", [&](void* arg) {
+      "SstFileWriter::Rep::InvalidatePageCache", [&](void* arg) {
        size_t fadvise_size = *(reinterpret_cast<size_t*>(arg));
        total_fadvised_bytes += fadvise_size;
      });
@@ -289,19 +536,18 @@ TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
      new SstFileWriter(EnvOptions(), options, nullptr, false));
  ASSERT_OK(sst_file_writer->Open(sst_file_path));
  for (int i = 0; i < kNumKeys; i++) {
-    ASSERT_OK(sst_file_writer->Add(Key(i), Key(i)));
+    ASSERT_OK(sst_file_writer->Put(Key(i), Key(i)));
  }
  ASSERT_OK(sst_file_writer->Finish());
  // fadvise disabled
  ASSERT_EQ(total_fadvised_bytes, 0);

-
  sst_file_path = sst_files_dir_ + "file_fadvise_enable.sst";
  sst_file_writer.reset(
      new SstFileWriter(EnvOptions(), options, nullptr, true));
  ASSERT_OK(sst_file_writer->Open(sst_file_path));
  for (int i = 0; i < kNumKeys; i++) {
-    ASSERT_OK(sst_file_writer->Add(Key(i), Key(i)));
+    ASSERT_OK(sst_file_writer->Put(Key(i), Key(i)));
  }
  ASSERT_OK(sst_file_writer->Finish());
  // fadvise enabled
@@ -311,6 +557,55 @@ TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

+TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  Reopen(options);
+
+  std::map<std::string, std::string> true_data;
+  int file_id = 1;
+  // 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) {
+      db_->Flush(FlushOptions());
+    }
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                               Key(50 * i), Key(50 * (i + 1))));
+  }
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+
+  // overlaps with L0 file but not memtable, so flush is skipped
+  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+  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));
+
+  // overlaps with memtable, so flush is triggered (thus file count increases by
+  // two at this step).
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {50, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
+      file_id++, &true_data));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
+  ASSERT_EQ(4, NumTableFilesAtLevel(0));
+
+  // snapshot unneeded now that both 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},
+      file_id++, &true_data));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
+  ASSERT_EQ(4, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -7,7 +7,9 @@

 #include "db/external_sst_file_ingestion_job.h"

+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif

 #include <inttypes.h>
 #include <algorithm>
@@ -154,34 +156,38 @@ Status ExternalSstFileIngestionJob::Run() {

  bool consumed_seqno = false;
  bool force_global_seqno = false;
-  const SequenceNumber last_seqno = versions_->LastSequence();
+
  if (ingestion_options_.snapshot_consistency && !db_snapshots_->empty()) {
    // We need to assign a global sequence number to all the files even
    // if the dont overlap with any ranges since we have snapshots
    force_global_seqno = true;
  }
-
+  const SequenceNumber last_seqno = versions_->LastSequence();
  SuperVersion* super_version = cfd_->GetSuperVersion();
  edit_.SetColumnFamily(cfd_->GetID());
  // The levels that the files will be ingested into
+
  for (IngestedFileInfo& f : files_to_ingest_) {
-    bool overlap_with_db = false;
-    status = AssignLevelForIngestedFile(super_version, &f, &overlap_with_db);
-    if (!status.ok()) {
-      return status;
-    }
-
-    if (overlap_with_db || force_global_seqno) {
-      status = AssignGlobalSeqnoForIngestedFile(&f, last_seqno + 1);
-      consumed_seqno = true;
+    SequenceNumber assigned_seqno = 0;
+    if (ingestion_options_.ingest_behind) {
+      status = CheckLevelForIngestedBehindFile(&f);
    } else {
-      status = AssignGlobalSeqnoForIngestedFile(&f, 0);
+      status = AssignLevelAndSeqnoForIngestedFile(
+         super_version, force_global_seqno, cfd_->ioptions()->compaction_style,
+         &f, &assigned_seqno);
+    }
+    if (!status.ok()) {
+      return status;
+    }
+    status = AssignGlobalSeqnoForIngestedFile(&f, assigned_seqno);
+    TEST_SYNC_POINT_CALLBACK("ExternalSstFileIngestionJob::Run",
+                             &assigned_seqno);
+    if (assigned_seqno == last_seqno + 1) {
+      consumed_seqno = true;
    }
-
    if (!status.ok()) {
      return status;
    }
-
    edit_.AddFile(f.picked_level, f.fd.GetNumber(), f.fd.GetPathId(),
                  f.fd.GetFileSize(), f.smallest_internal_key(),
                  f.largest_internal_key(), f.assigned_seqno, f.assigned_seqno,
@@ -373,6 +379,17 @@ Status ExternalSstFileIngestionJob::IngestedFilesOverlapWithMemtables(
  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_) {
@@ -381,22 +398,37 @@ Status ExternalSstFileIngestionJob::IngestedFilesOverlapWithMemtables(
    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::AssignLevelForIngestedFile(
-    SuperVersion* sv, IngestedFileInfo* file_to_ingest, bool* overlap_with_db) {
-  *overlap_with_db = false;
+Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
+    SuperVersion* sv, bool force_global_seqno, CompactionStyle compaction_style,
+    IngestedFileInfo* file_to_ingest, SequenceNumber* assigned_seqno) {
+  Status status;
+  *assigned_seqno = 0;
+  const SequenceNumber last_seqno = versions_->LastSequence();
+  if (force_global_seqno) {
+    *assigned_seqno = last_seqno + 1;
+    if (compaction_style == kCompactionStyleUniversal) {
+      file_to_ingest->picked_level = 0;
+      return status;
+    }
+  }

+  bool overlap_with_db = false;
  Arena arena;
  ReadOptions ro;
  ro.total_order_seek = true;
-
-  Status status;
  int target_level = 0;
  auto* vstorage = cfd_->current()->storage_info();
+
  for (int lvl = 0; lvl < cfd_->NumberLevels(); lvl++) {
    if (lvl > 0 && lvl < vstorage->base_level()) {
      continue;
@@ -404,44 +436,81 @@ Status ExternalSstFileIngestionJob::AssignLevelForIngestedFile(

    if (vstorage->NumLevelFiles(lvl) > 0) {
      bool overlap_with_level = false;
-      MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(),
-                                              &arena);
-      RangeDelAggregator range_del_agg(cfd_->internal_comparator(),
-                                       {} /* snapshots */);
-      sv->current->AddIteratorsForLevel(ro, env_options_, &merge_iter_builder,
-                                        lvl, &range_del_agg);
-      if (!range_del_agg.IsEmpty()) {
-        return Status::NotSupported(
-            "file ingestion with range tombstones is currently unsupported");
-      }
-      ScopedArenaIterator level_iter(merge_iter_builder.Finish());
-
-      status = IngestedFileOverlapWithIteratorRange(
-          file_to_ingest, level_iter.get(), &overlap_with_level);
+      status = IngestedFileOverlapWithLevel(sv, file_to_ingest, lvl,
+        &overlap_with_level);
      if (!status.ok()) {
        return status;
      }
-
      if (overlap_with_level) {
        // We must use L0 or any level higher than `lvl` to be able to overwrite
        // the keys that we overlap with in this level, We also need to assign
        // this file a seqno to overwrite the existing keys in level `lvl`
-        *overlap_with_db = true;
+        overlap_with_db = true;
        break;
      }
+
+      if (compaction_style == kCompactionStyleUniversal && lvl != 0) {
+        const std::vector<FileMetaData*>& level_files =
+            vstorage->LevelFiles(lvl);
+        const SequenceNumber level_largest_seqno =
+            (*max_element(level_files.begin(), level_files.end(),
+                          [](FileMetaData* f1, FileMetaData* f2) {
+                            return f1->largest_seqno < f2->largest_seqno;
+                          }))
+                ->largest_seqno;
+        if (level_largest_seqno != 0) {
+          *assigned_seqno = level_largest_seqno;
+        } else {
+          continue;
+        }
+      }
+    } else if (compaction_style == kCompactionStyleUniversal) {
+      continue;
    }

    // We dont overlap with any keys in this level, but we still need to check
    // if our file can fit in it
-
    if (IngestedFileFitInLevel(file_to_ingest, lvl)) {
      target_level = lvl;
    }
  }
+ TEST_SYNC_POINT_CALLBACK(
+      "ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile",
+      &overlap_with_db);
  file_to_ingest->picked_level = target_level;
+  if (overlap_with_db && *assigned_seqno == 0) {
+    *assigned_seqno = last_seqno + 1;
+  }
  return status;
 }

+Status ExternalSstFileIngestionJob::CheckLevelForIngestedBehindFile(
+    IngestedFileInfo* file_to_ingest) {
+  auto* vstorage = cfd_->current()->storage_info();
+  // first check if new files fit in the bottommost level
+  int bottom_lvl = cfd_->NumberLevels() - 1;
+  if(!IngestedFileFitInLevel(file_to_ingest, bottom_lvl)) {
+    return Status::InvalidArgument(
+      "Can't ingest_behind file as it doesn't fit "
+      "at the bottommost level!");
+  }
+
+  // second check if despite allow_ingest_behind=true we still have 0 seqnums
+  // at some upper level
+  for (int lvl = 0; lvl < cfd_->NumberLevels() - 1; lvl++) {
+    for (auto file : vstorage->LevelFiles(lvl)) {
+      if (file->smallest_seqno == 0) {
+        return Status::InvalidArgument(
+          "Can't ingest_behind file as despite allow_ingest_behind=true "
+          "there are files with 0 seqno in database at upper levels!");
+      }
+    }
+  }
+
+  file_to_ingest->picked_level = bottom_lvl;
+  return Status::OK();
+}
+
 Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
    IngestedFileInfo* file_to_ingest, SequenceNumber seqno) {
  if (file_to_ingest->original_seqno == seqno) {
@@ -500,6 +569,34 @@ 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) {
@@ -528,6 +625,35 @@ bool ExternalSstFileIngestionJob::IngestedFileFitInLevel(
  return true;
 }

+Status ExternalSstFileIngestionJob::IngestedFileOverlapWithLevel(
+    SuperVersion* sv, IngestedFileInfo* file_to_ingest, int lvl,
+    bool* overlap_with_level) {
+  Arena arena;
+  ReadOptions ro;
+  ro.total_order_seek = true;
+  MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(),
+                                          &arena);
+  sv->current->AddIteratorsForLevel(ro, env_options_, &merge_iter_builder, lvl,
+                                    nullptr /* 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);
+  }
+  return status;
+}
+
 }  // namespace rocksdb

 #endif  // !ROCKSDB_LITE
@@ -95,7 +95,7 @@ class ExternalSstFileIngestionJob {
  // REQUIRES: Mutex held
  void UpdateStats();

-  // Cleanup after successfull/failed job
+  // Cleanup after successful/failed job
  void Cleanup(const Status& status);

  VersionEdit* edit() { return &edit_; }
@@ -114,12 +114,20 @@ class ExternalSstFileIngestionJob {
  // REQUIRES: Mutex held
  Status IngestedFilesOverlapWithMemtables(SuperVersion* sv, bool* overlap);

-  // Assign `file_to_ingest` the lowest possible level that it can
-  // be ingested to.
+  // 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
-  Status AssignLevelForIngestedFile(SuperVersion* sv,
-                                    IngestedFileInfo* file_to_ingest,
-                                    bool* overlap_with_db);
+  Status AssignLevelAndSeqnoForIngestedFile(SuperVersion* sv,
+                                            bool force_global_seqno,
+                                            CompactionStyle compaction_style,
+                                            IngestedFileInfo* file_to_ingest,
+                                            SequenceNumber* assigned_seqno);
+
+  // File that we want to ingest behind always goes to the lowest level;
+  // we just check that it fits in the level, that DB allows ingest_behind,
+  // and that we don't have 0 seqnums at the upper levels.
+  // REQUIRES: Mutex held
+  Status CheckLevelForIngestedBehindFile(IngestedFileInfo* file_to_ingest);

  // Set the file global sequence number to `seqno`
  Status AssignGlobalSeqnoForIngestedFile(IngestedFileInfo* file_to_ingest,
@@ -131,6 +139,18 @@ 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,
+    IngestedFileInfo* file_to_ingest, int lvl, bool* overlap_with_level);
+
  // Check if `file_to_ingest` can fit in level `level`
  // REQUIRES: Mutex held
  bool IngestedFileFitInLevel(const IngestedFileInfo* file_to_ingest,
@@ -61,7 +61,7 @@ class ExternalSSTFileTest : public DBTestBase {
      return s;
    }
    for (auto& entry : data) {
-      s = sst_file_writer.Add(entry.first, entry.second);
+      s = sst_file_writer.Put(entry.first, entry.second);
      if (!s.ok()) {
        sst_file_writer.Finish();
        return s;
@@ -88,6 +88,68 @@ class ExternalSSTFileTest : public DBTestBase {
    return s;
  }

+  Status GenerateAndAddExternalFileIngestBehind(
+      const Options options, const IngestExternalFileOptions ifo,
+      std::vector<std::pair<std::string, std::string>> data, int file_id = -1,
+      bool sort_data = false,
+      std::map<std::string, std::string>* true_data = nullptr,
+      ColumnFamilyHandle* cfh = nullptr) {
+    // Generate a file id if not provided
+    if (file_id == -1) {
+      file_id = last_file_id_ + 1;
+      last_file_id_++;
+    }
+
+    // Sort data if asked to do so
+    if (sort_data) {
+      std::sort(data.begin(), data.end(),
+                [&](const std::pair<std::string, std::string>& e1,
+                    const std::pair<std::string, std::string>& e2) {
+                  return options.comparator->Compare(e1.first, e2.first) < 0;
+                });
+      auto uniq_iter = std::unique(
+          data.begin(), data.end(),
+          [&](const std::pair<std::string, std::string>& e1,
+              const std::pair<std::string, std::string>& e2) {
+            return options.comparator->Compare(e1.first, e2.first) == 0;
+          });
+      data.resize(uniq_iter - data.begin());
+    }
+    std::string file_path = sst_files_dir_ + ToString(file_id);
+    SstFileWriter sst_file_writer(EnvOptions(), options, cfh);
+
+    Status s = sst_file_writer.Open(file_path);
+    if (!s.ok()) {
+      return s;
+    }
+    for (auto& entry : data) {
+      s = sst_file_writer.Put(entry.first, entry.second);
+      if (!s.ok()) {
+        sst_file_writer.Finish();
+        return s;
+      }
+    }
+    s = sst_file_writer.Finish();
+
+    if (s.ok()) {
+      if (cfh) {
+        s = db_->IngestExternalFile(cfh, {file_path}, ifo);
+      } else {
+        s = db_->IngestExternalFile({file_path}, ifo);
+      }
+    }
+
+    if (s.ok() && true_data) {
+      for (auto& entry : data) {
+        (*true_data)[entry.first] = entry.second;
+      }
+    }
+
+    return s;
+  }
+
+
+
  Status GenerateAndAddExternalFile(
      const Options options, std::vector<std::pair<int, std::string>> data,
      int file_id = -1, bool allow_global_seqno = false, bool sort_data = false,
@@ -147,7 +209,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
    std::string file1 = sst_files_dir_ + "file1.sst";
    ASSERT_OK(sst_file_writer.Open(file1));
    for (int k = 0; k < 100; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    ExternalSstFileInfo file1_info;
    Status s = sst_file_writer.Finish(&file1_info);
@@ -161,17 +223,17 @@ TEST_F(ExternalSSTFileTest, Basic) {
    ASSERT_EQ(file1_info.smallest_key, Key(0));
    ASSERT_EQ(file1_info.largest_key, Key(99));
    // sst_file_writer already finished, cannot add this value
-    s = sst_file_writer.Add(Key(100), "bad_val");
+    s = sst_file_writer.Put(Key(100), "bad_val");
    ASSERT_FALSE(s.ok()) << s.ToString();

    // file2.sst (100 => 199)
    std::string file2 = sst_files_dir_ + "file2.sst";
    ASSERT_OK(sst_file_writer.Open(file2));
    for (int k = 100; k < 200; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    // Cannot add this key because it's not after last added key
-    s = sst_file_writer.Add(Key(99), "bad_val");
+    s = sst_file_writer.Put(Key(99), "bad_val");
    ASSERT_FALSE(s.ok()) << s.ToString();
    ExternalSstFileInfo file2_info;
    s = sst_file_writer.Finish(&file2_info);
@@ -186,7 +248,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
    std::string file3 = sst_files_dir_ + "file3.sst";
    ASSERT_OK(sst_file_writer.Open(file3));
    for (int k = 195; k < 300; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val_overlap"));
    }
    ExternalSstFileInfo file3_info;
    s = sst_file_writer.Finish(&file3_info);
@@ -204,7 +266,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
    std::string file4 = sst_files_dir_ + "file4.sst";
    ASSERT_OK(sst_file_writer.Open(file4));
    for (int k = 30; k < 40; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val_overlap"));
    }
    ExternalSstFileInfo file4_info;
    s = sst_file_writer.Finish(&file4_info);
@@ -218,7 +280,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
    std::string file5 = sst_files_dir_ + "file5.sst";
    ASSERT_OK(sst_file_writer.Open(file5));
    for (int k = 400; k < 500; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    ExternalSstFileInfo file5_info;
    s = sst_file_writer.Finish(&file5_info);
@@ -315,8 +377,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
      ASSERT_EQ(Get(Key(k)), value);
    }
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipUniversalCompaction |
-                         kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }
 class SstFileWriterCollector : public TablePropertiesCollector {
 public:
@@ -381,7 +442,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
    std::string file1 = sst_files_dir_ + "file1.sst";
    ASSERT_OK(sst_file_writer.Open(file1));
    for (int k = 0; k < 100; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    ExternalSstFileInfo file1_info;
    Status s = sst_file_writer.Finish(&file1_info);
@@ -391,17 +452,17 @@ TEST_F(ExternalSSTFileTest, AddList) {
    ASSERT_EQ(file1_info.smallest_key, Key(0));
    ASSERT_EQ(file1_info.largest_key, Key(99));
    // sst_file_writer already finished, cannot add this value
-    s = sst_file_writer.Add(Key(100), "bad_val");
+    s = sst_file_writer.Put(Key(100), "bad_val");
    ASSERT_FALSE(s.ok()) << s.ToString();

    // file2.sst (100 => 199)
    std::string file2 = sst_files_dir_ + "file2.sst";
    ASSERT_OK(sst_file_writer.Open(file2));
    for (int k = 100; k < 200; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    // Cannot add this key because it's not after last added key
-    s = sst_file_writer.Add(Key(99), "bad_val");
+    s = sst_file_writer.Put(Key(99), "bad_val");
    ASSERT_FALSE(s.ok()) << s.ToString();
    ExternalSstFileInfo file2_info;
    s = sst_file_writer.Finish(&file2_info);
@@ -416,7 +477,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
    std::string file3 = sst_files_dir_ + "file3.sst";
    ASSERT_OK(sst_file_writer.Open(file3));
    for (int k = 195; k < 200; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val_overlap"));
    }
    ExternalSstFileInfo file3_info;
    s = sst_file_writer.Finish(&file3_info);
@@ -431,7 +492,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
    std::string file4 = sst_files_dir_ + "file4.sst";
    ASSERT_OK(sst_file_writer.Open(file4));
    for (int k = 30; k < 40; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val_overlap"));
    }
    ExternalSstFileInfo file4_info;
    s = sst_file_writer.Finish(&file4_info);
@@ -445,7 +506,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
    std::string file5 = sst_files_dir_ + "file5.sst";
    ASSERT_OK(sst_file_writer.Open(file5));
    for (int k = 200; k < 300; k++) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
    }
    ExternalSstFileInfo file5_info;
    s = sst_file_writer.Finish(&file5_info);
@@ -556,8 +617,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
      ASSERT_EQ(Get(Key(k)), value);
    }
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipUniversalCompaction |
-                         kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }

 TEST_F(ExternalSSTFileTest, AddListAtomicity) {
@@ -577,7 +637,7 @@ TEST_F(ExternalSSTFileTest, AddListAtomicity) {
      files[i] = sst_files_dir_ + "file" + std::to_string(i) + ".sst";
      ASSERT_OK(sst_file_writer.Open(files[i]));
      for (int k = i * 100; k < (i + 1) * 100; k++) {
-        ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+        ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
      }
      Status s = sst_file_writer.Finish(&files_info[i]);
      ASSERT_TRUE(s.ok()) << s.ToString();
@@ -599,8 +659,7 @@ TEST_F(ExternalSSTFileTest, AddListAtomicity) {
      ASSERT_EQ(Get(Key(k)), value);
    }
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipUniversalCompaction |
-                         kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }
 // This test reporduce a bug that can happen in some cases if the DB started
 // purging obsolete files when we are adding an external sst file.
@@ -615,7 +674,7 @@ TEST_F(ExternalSSTFileTest, PurgeObsoleteFilesBug) {
  ASSERT_OK(s);
  for (int i = 0; i < 500; i++) {
    std::string k = Key(i);
-    s = sst_file_writer.Add(k, k + "_val");
+    s = sst_file_writer.Put(k, k + "_val");
    ASSERT_OK(s);
  }

@@ -658,7 +717,7 @@ TEST_F(ExternalSSTFileTest, SkipSnapshot) {
  std::string file1 = sst_files_dir_ + "file1.sst";
  ASSERT_OK(sst_file_writer.Open(file1));
  for (int k = 0; k < 100; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file1_info;
  Status s = sst_file_writer.Finish(&file1_info);
@@ -672,7 +731,7 @@ TEST_F(ExternalSSTFileTest, SkipSnapshot) {
  std::string file2 = sst_files_dir_ + "file2.sst";
  ASSERT_OK(sst_file_writer.Open(file2));
  for (int k = 100; k < 300; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file2_info;
  s = sst_file_writer.Finish(&file2_info);
@@ -702,7 +761,7 @@ TEST_F(ExternalSSTFileTest, SkipSnapshot) {
  std::string file3 = sst_files_dir_ + "file3.sst";
  ASSERT_OK(sst_file_writer.Open(file3));
  for (int k = 300; k < 400; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+    ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
  }
  ExternalSstFileInfo file3_info;
  s = sst_file_writer.Finish(&file3_info);
@@ -748,7 +807,7 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {
      ASSERT_OK(sst_file_writer.Open(file_names[file_idx]));

      for (int k = range_start; k < range_end; k++) {
-        ASSERT_OK(sst_file_writer.Add(Key(k), Key(k)));
+        ASSERT_OK(sst_file_writer.Put(Key(k), Key(k)));
      }

      Status s = sst_file_writer.Finish();
@@ -831,12 +890,31 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {

    fprintf(stderr, "Verified %d values\n", num_files * keys_per_file);
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipUniversalCompaction |
-                         kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }

 TEST_F(ExternalSSTFileTest, OverlappingRanges) {
  Random rnd(301);
+  int picked_level = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "ExternalSstFileIngestionJob::Run", [&picked_level](void* arg) {
+      ASSERT_TRUE(arg != nullptr);
+      picked_level = *(static_cast<int*>(arg));
+    });
+  bool need_flush = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "DBImpl::IngestExternalFile:NeedFlush", [&need_flush](void* arg) {
+      ASSERT_TRUE(arg != nullptr);
+      need_flush = *(static_cast<bool*>(arg));
+    });
+  bool overlap_with_db = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile",
+      [&overlap_with_db](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        overlap_with_db = *(static_cast<bool*>(arg));
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
  do {
    Options options = CurrentOptions();
    DestroyAndReopen(options);
@@ -880,7 +958,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
        std::string file_name = sst_files_dir_ + env_->GenerateUniqueId();
        ASSERT_OK(sst_file_writer.Open(file_name));
        for (int k = range_start; k <= range_end; k++) {
-          s = sst_file_writer.Add(Key(k), range_val);
+          s = sst_file_writer.Put(Key(k), range_val);
          ASSERT_OK(s);
        }
        ExternalSstFileInfo file_info;
@@ -889,15 +967,27 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {

        // Insert the generated file
        s = DeprecatedAddFile({file_name});
-
        auto it = true_data.lower_bound(Key(range_start));
-        if (it != true_data.end() && it->first <= Key(range_end)) {
-          // This range overlap with data already exist in DB
-          ASSERT_NOK(s);
-          failed_add_file++;
+        if (option_config_ != kUniversalCompaction &&
+            option_config_ != kUniversalCompactionMultiLevel) {
+          if (it != true_data.end() && it->first <= Key(range_end)) {
+            // This range overlap with data already exist in DB
+            ASSERT_NOK(s);
+            failed_add_file++;
+          } else {
+            ASSERT_OK(s);
+            success_add_file++;
+          }
        } else {
-          ASSERT_OK(s);
-          success_add_file++;
+          if ((it != true_data.end() && it->first <= Key(range_end)) ||
+              need_flush || picked_level > 0 || overlap_with_db) {
+            // This range overlap with data already exist in DB
+            ASSERT_NOK(s);
+            failed_add_file++;
+          } else {
+            ASSERT_OK(s);
+            success_add_file++;
+          }
        }
      }

@@ -930,8 +1020,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
    }
    printf("keys/values verified\n");
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipUniversalCompaction |
-                         kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }

 TEST_F(ExternalSSTFileTest, PickedLevel) {
@@ -1267,7 +1356,7 @@ TEST_F(ExternalSSTFileTest, AddExternalSstFileWithCustomCompartor) {
    int range_end = i * 10;
    int range_start = range_end + 15;
    for (int k = (range_start - 1); k >= range_end; k--) {
-      ASSERT_OK(sst_file_writer.Add(Key(k), Key(k)));
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k)));
    }
    ExternalSstFileInfo file_info;
    ASSERT_OK(sst_file_writer.Finish(&file_info));
@@ -1357,12 +1446,12 @@ TEST_F(ExternalSSTFileTest, SstFileWriterNonSharedKeys) {

  std::string suffix(100, 'X');
  ASSERT_OK(sst_file_writer.Open(file_path));
-  ASSERT_OK(sst_file_writer.Add("A" + suffix, "VAL"));
-  ASSERT_OK(sst_file_writer.Add("BB" + suffix, "VAL"));
-  ASSERT_OK(sst_file_writer.Add("CC" + suffix, "VAL"));
-  ASSERT_OK(sst_file_writer.Add("CXD" + suffix, "VAL"));
-  ASSERT_OK(sst_file_writer.Add("CZZZ" + suffix, "VAL"));
-  ASSERT_OK(sst_file_writer.Add("ZAAAX" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("A" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("BB" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("CC" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("CXD" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("CZZZ" + suffix, "VAL"));
+  ASSERT_OK(sst_file_writer.Put("ZAAAX" + suffix, "VAL"));

  ASSERT_OK(sst_file_writer.Finish());
  ASSERT_OK(DeprecatedAddFile({file_path}));
@@ -1657,22 +1746,22 @@ TEST_F(ExternalSSTFileTest, FileWithCFInfo) {
  // default_cf.sst
  const std::string cf_default_sst = sst_files_dir_ + "/default_cf.sst";
  ASSERT_OK(sfw_default.Open(cf_default_sst));
-  ASSERT_OK(sfw_default.Add("K1", "V1"));
-  ASSERT_OK(sfw_default.Add("K2", "V2"));
+  ASSERT_OK(sfw_default.Put("K1", "V1"));
+  ASSERT_OK(sfw_default.Put("K2", "V2"));
  ASSERT_OK(sfw_default.Finish());

  // cf1.sst
  const std::string cf1_sst = sst_files_dir_ + "/cf1.sst";
  ASSERT_OK(sfw_cf1.Open(cf1_sst));
-  ASSERT_OK(sfw_cf1.Add("K3", "V1"));
-  ASSERT_OK(sfw_cf1.Add("K4", "V2"));
+  ASSERT_OK(sfw_cf1.Put("K3", "V1"));
+  ASSERT_OK(sfw_cf1.Put("K4", "V2"));
  ASSERT_OK(sfw_cf1.Finish());

  // cf_unknown.sst
  const std::string unknown_sst = sst_files_dir_ + "/cf_unknown.sst";
  ASSERT_OK(sfw_unknown.Open(unknown_sst));
-  ASSERT_OK(sfw_unknown.Add("K5", "V1"));
-  ASSERT_OK(sfw_unknown.Add("K6", "V2"));
+  ASSERT_OK(sfw_unknown.Put("K5", "V1"));
+  ASSERT_OK(sfw_unknown.Put("K6", "V2"));
  ASSERT_OK(sfw_unknown.Finish());

  IngestExternalFileOptions ifo;
@@ -1773,7 +1862,7 @@ TEST_F(ExternalSSTFileTest, SnapshotInconsistencyBug) {
  SstFileWriter sst_file_writer(EnvOptions(), options);
  ASSERT_OK(sst_file_writer.Open(sst_file_path));
  for (int i = 0; i < kNumKeys; i++) {
-    ASSERT_OK(sst_file_writer.Add(Key(i), Key(i) + "_V2"));
+    ASSERT_OK(sst_file_writer.Put(Key(i), Key(i) + "_V2"));
  }
  ASSERT_OK(sst_file_writer.Finish());

@@ -1788,6 +1877,69 @@ TEST_F(ExternalSSTFileTest, SnapshotInconsistencyBug) {

  db_->ReleaseSnapshot(snap);
 }
+
+TEST_F(ExternalSSTFileTest, IngestBehind) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.num_levels = 3;
+  options.disable_auto_compactions = false;
+  DestroyAndReopen(options);
+  std::vector<std::pair<std::string, std::string>> file_data;
+  std::map<std::string, std::string> true_data;
+
+  // Insert 100 -> 200 into the memtable
+  for (int i = 100; i <= 200; i++) {
+    ASSERT_OK(Put(Key(i), "memtable"));
+    true_data[Key(i)] = "memtable";
+  }
+
+  // Insert 100 -> 200 using IngestExternalFile
+  file_data.clear();
+  for (int i = 0; i <= 20; i++) {
+    file_data.emplace_back(Key(i), "ingest_behind");
+  }
+
+  IngestExternalFileOptions ifo;
+  ifo.allow_global_seqno = true;
+  ifo.ingest_behind = true;
+
+  // Can't ingest behind since allow_ingest_behind isn't set to true
+  ASSERT_NOK(GenerateAndAddExternalFileIngestBehind(options, ifo,
+                                                   file_data, -1, false,
+                                                   &true_data));
+
+  options.allow_ingest_behind = true;
+  // check that we still can open the DB, as num_levels should be
+  // sanitized to 3
+  options.num_levels = 2;
+  DestroyAndReopen(options);
+
+  options.num_levels = 3;
+  DestroyAndReopen(options);
+  // Insert 100 -> 200 into the memtable
+  for (int i = 100; i <= 200; i++) {
+    ASSERT_OK(Put(Key(i), "memtable"));
+    true_data[Key(i)] = "memtable";
+  }
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  // Universal picker should go at second from the bottom level
+  ASSERT_EQ("0,1", FilesPerLevel());
+  ASSERT_OK(GenerateAndAddExternalFileIngestBehind(options, ifo,
+                                                   file_data, -1, false,
+                                                   &true_data));
+  ASSERT_EQ("0,1,1", FilesPerLevel());
+  // this time ingest should fail as the file doesn't fit to the bottom level
+  ASSERT_NOK(GenerateAndAddExternalFileIngestBehind(options, ifo,
+                                                   file_data, -1, false,
+                                                   &true_data));
+  ASSERT_EQ("0,1,1", FilesPerLevel());
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  // bottom level should be empty
+  ASSERT_EQ("0,1", FilesPerLevel());
+
+  size_t kcnt = 0;
+  VerifyDBFromMap(true_data, &kcnt, false);
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -228,10 +228,12 @@ class FaultInjectionTest : public testing::Test,
    return Status::OK();
  }

-#if __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 9)
-__attribute__((__no_sanitize__("undefined")))
-#elif __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9)
+#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 {
@@ -256,13 +258,15 @@ __attribute__((__no_sanitize_undefined__))

  void CloseDB() {
    delete db_;
-    db_ = NULL;
+    db_ = nullptr;
  }

  Status OpenDB() {
    CloseDB();
    env_->ResetState();
-    return DB::Open(options_, dbname_, &db_);
+    Status s = DB::Open(options_, dbname_, &db_);
+    assert(db_ != nullptr);
+    return s;
  }

  void DeleteAllData() {
@@ -13,11 +13,9 @@ namespace rocksdb {

 void FlushScheduler::ScheduleFlush(ColumnFamilyData* cfd) {
 #ifndef NDEBUG
-  {
-    std::lock_guard<std::mutex> lock(checking_mutex_);
-    assert(checking_set_.count(cfd) == 0);
-    checking_set_.insert(cfd);
-  }
+  std::lock_guard<std::mutex> lock(checking_mutex_);
+  assert(checking_set_.count(cfd) == 0);
+  checking_set_.insert(cfd);
 #endif  // NDEBUG
  cfd->Ref();
 // Suppress false positive clang analyzer warnings.
@@ -34,8 +32,11 @@ void FlushScheduler::ScheduleFlush(ColumnFamilyData* cfd) {
 }

 ColumnFamilyData* FlushScheduler::TakeNextColumnFamily() {
+#ifndef NDEBUG
+  std::lock_guard<std::mutex> lock(checking_mutex_);
+#endif  // NDEBUG
  while (true) {
-    if (Empty()) {
+    if (head_.load(std::memory_order_relaxed) == nullptr) {
      return nullptr;
    }

@@ -46,11 +47,9 @@ ColumnFamilyData* FlushScheduler::TakeNextColumnFamily() {
    delete node;

 #ifndef NDEBUG
-    {
-      auto iter = checking_set_.find(cfd);
-      assert(iter != checking_set_.end());
-      checking_set_.erase(iter);
-    }
+    auto iter = checking_set_.find(cfd);
+    assert(iter != checking_set_.end());
+    checking_set_.erase(iter);
 #endif  // NDEBUG

    if (!cfd->IsDropped()) {
@@ -66,8 +65,13 @@ ColumnFamilyData* FlushScheduler::TakeNextColumnFamily() {
 }

 bool FlushScheduler::Empty() {
+#ifndef NDEBUG
+  std::lock_guard<std::mutex> lock(checking_mutex_);
+#endif  // NDEBUG
  auto rv = head_.load(std::memory_order_relaxed) == nullptr;
+#ifndef NDEBUG
  assert(rv == checking_set_.empty());
+#endif  // NDEBUG
  return rv;
 }

@@ -78,7 +82,7 @@ void FlushScheduler::Clear() {
      delete cfd;
    }
  }
-  assert(Empty());
+  assert(head_.load(std::memory_order_relaxed) == nullptr);
 }

 }  // namespace rocksdb
@@ -1137,6 +1137,7 @@ void InternalStats::DumpCFStatsNoFileHistogram(std::string* value) {
           total_stall_count - cf_stats_snapshot_.stall_count);
  value->append(buf);

+  cf_stats_snapshot_.seconds_up = seconds_up;
  cf_stats_snapshot_.ingest_bytes_flush = flush_ingest;
  cf_stats_snapshot_.ingest_bytes_addfile = add_file_ingest;
  cf_stats_snapshot_.ingest_files_addfile = ingest_files_addfile;
@@ -49,6 +49,7 @@ struct DBPropertyInfo {
 extern const DBPropertyInfo* GetPropertyInfo(const Slice& property);

 #ifndef ROCKSDB_LITE
+#undef SCORE
 enum class LevelStatType {
  INVALID = 0,
  NUM_FILES,
@@ -0,0 +1,52 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#include "db/malloc_stats.h"
+
+#ifndef ROCKSDB_LITE
+#include <memory>
+#include <string.h>
+
+namespace rocksdb {
+
+#ifdef ROCKSDB_JEMALLOC
+#include "jemalloc/jemalloc.h"
+
+typedef struct {
+  char* cur;
+  char* end;
+} MallocStatus;
+
+static void GetJemallocStatus(void* mstat_arg, const char* status) {
+  MallocStatus* mstat = reinterpret_cast<MallocStatus*>(mstat_arg);
+  size_t status_len = status ? strlen(status) : 0;
+  size_t buf_size = (size_t)(mstat->end - mstat->cur);
+  if (!status_len || status_len > buf_size) {
+    return;
+  }
+
+  snprintf(mstat->cur, buf_size, "%s", status);
+  mstat->cur += status_len;
+}
+#endif  // ROCKSDB_JEMALLOC
+
+void DumpMallocStats(std::string* stats) {
+#ifdef ROCKSDB_JEMALLOC
+  MallocStatus mstat;
+  const unsigned int kMallocStatusLen = 1000000;
+  std::unique_ptr<char[]> buf{new char[kMallocStatusLen + 1]};
+  mstat.cur = buf.get();
+  mstat.end = buf.get() + kMallocStatusLen;
+  je_malloc_stats_print(GetJemallocStatus, &mstat, "");
+  stats->append(buf.get());
+#endif  // ROCKSDB_JEMALLOC
+}
+
+}
+#endif  // !ROCKSDB_LITE
@@ -0,0 +1,22 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#pragma once
+
+#ifndef ROCKSDB_LITE
+
+#include <string>
+
+namespace rocksdb {
+
+void DumpMallocStats(std::string*);
+
+}
+
+#endif  // !ROCKSDB_LITE
@@ -11,7 +11,6 @@

 #include <memory>
 #include <algorithm>
-#include <limits>

 #include "db/dbformat.h"
 #include "db/merge_context.h"
@@ -35,7 +34,6 @@
 #include "util/memory_usage.h"
 #include "util/murmurhash.h"
 #include "util/mutexlock.h"
-#include "util/stop_watch.h"

 namespace rocksdb {

@@ -61,20 +59,24 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
                   const ImmutableCFOptions& ioptions,
                   const MutableCFOptions& mutable_cf_options,
                   WriteBufferManager* write_buffer_manager,
-                   SequenceNumber latest_seq)
+                   SequenceNumber latest_seq, uint32_t column_family_id)
    : comparator_(cmp),
      moptions_(ioptions, mutable_cf_options),
      refs_(0),
      kArenaBlockSize(OptimizeBlockSize(moptions_.arena_block_size)),
-      arena_(moptions_.arena_block_size,
-             mutable_cf_options.memtable_huge_page_size),
-      allocator_(&arena_, write_buffer_manager),
+      mem_tracker_(write_buffer_manager),
+      arena_(
+          moptions_.arena_block_size,
+          (write_buffer_manager != nullptr && write_buffer_manager->enabled())
+              ? &mem_tracker_
+              : nullptr,
+          mutable_cf_options.memtable_huge_page_size),
      table_(ioptions.memtable_factory->CreateMemTableRep(
-          comparator_, &allocator_, ioptions.prefix_extractor,
-          ioptions.info_log)),
+          comparator_, &arena_, ioptions.prefix_extractor, ioptions.info_log,
+          column_family_id)),
      range_del_table_(SkipListFactory().CreateMemTableRep(
-          comparator_, &allocator_, nullptr /* transform */,
-          ioptions.info_log)),
+          comparator_, &arena_, nullptr /* transform */, ioptions.info_log,
+          column_family_id)),
      is_range_del_table_empty_(true),
      data_size_(0),
      num_entries_(0),
@@ -101,13 +103,16 @@ MemTable::MemTable(const InternalKeyComparator& cmp,

  if (prefix_extractor_ && moptions_.memtable_prefix_bloom_bits > 0) {
    prefix_bloom_.reset(new DynamicBloom(
-        &allocator_, moptions_.memtable_prefix_bloom_bits,
-        ioptions.bloom_locality, 6 /* hard coded 6 probes */, nullptr,
-        moptions_.memtable_huge_page_size, ioptions.info_log));
+        &arena_, moptions_.memtable_prefix_bloom_bits, ioptions.bloom_locality,
+        6 /* hard coded 6 probes */, nullptr, moptions_.memtable_huge_page_size,
+        ioptions.info_log));
  }
 }

-MemTable::~MemTable() { assert(refs_ == 0); }
+MemTable::~MemTable() {
+  mem_tracker_.FreeMem();
+  assert(refs_ == 0);
+}

 size_t MemTable::ApproximateMemoryUsage() {
  autovector<size_t> usages = {arena_.ApproximateMemoryUsage(),
@@ -576,7 +581,7 @@ static bool SaveValue(void* arg, const char* entry) {
          *(s->status) = MergeHelper::TimedFullMerge(
              merge_operator, s->key->user_key(), &v,
              merge_context->GetOperands(), s->value, s->logger, s->statistics,
-              s->env_);
+              s->env_, nullptr /* result_operand */, true);
        } else if (s->value != nullptr) {
          s->value->assign(v.data(), v.size());
        }
@@ -593,7 +598,7 @@ static bool SaveValue(void* arg, const char* entry) {
          *(s->status) = MergeHelper::TimedFullMerge(
              merge_operator, s->key->user_key(), nullptr,
              merge_context->GetOperands(), s->value, s->logger, s->statistics,
-              s->env_);
+              s->env_, nullptr /* result_operand */, true);
        } else {
          *(s->status) = Status::NotFound();
        }
@@ -18,12 +18,12 @@
 #include "db/dbformat.h"
 #include "db/range_del_aggregator.h"
 #include "db/version_edit.h"
-#include "memtable/memtable_allocator.h"
 #include "monitoring/instrumented_mutex.h"
 #include "options/cf_options.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
+#include "util/allocator.h"
 #include "util/concurrent_arena.h"
 #include "util/dynamic_bloom.h"
 #include "util/hash.h"
@@ -101,7 +101,7 @@ class MemTable {
                    const ImmutableCFOptions& ioptions,
                    const MutableCFOptions& mutable_cf_options,
                    WriteBufferManager* write_buffer_manager,
-                    SequenceNumber earliest_seq);
+                    SequenceNumber earliest_seq, uint32_t column_family_id);

  // Do not delete this MemTable unless Unref() indicates it not in use.
  ~MemTable();
@@ -317,7 +317,7 @@ class MemTable {
  // write anything to this MemTable().  (Ie. do not call Add() or Update()).
  void MarkImmutable() {
    table_->MarkReadOnly();
-    allocator_.DoneAllocating();
+    mem_tracker_.DoneAllocating();
  }

  // return true if the current MemTableRep supports merge operator.
@@ -359,8 +359,8 @@ class MemTable {
  const MemTableOptions moptions_;
  int refs_;
  const size_t kArenaBlockSize;
+  AllocTracker mem_tracker_;
  ConcurrentArena arena_;
-  MemTableAllocator allocator_;
  unique_ptr<MemTableRep> table_;
  unique_ptr<MemTableRep> range_del_table_;
  bool is_range_del_table_empty_;
@@ -166,6 +166,18 @@ Status MemTableListVersion::AddRangeTombstoneIterators(
  return Status::OK();
 }

+Status MemTableListVersion::AddRangeTombstoneIterators(
+    const ReadOptions& read_opts,
+    std::vector<InternalIterator*>* range_del_iters) {
+  for (auto& m : memlist_) {
+    auto* range_del_iter = m->NewRangeTombstoneIterator(read_opts);
+    if (range_del_iter != nullptr) {
+      range_del_iters->push_back(range_del_iter);
+    }
+  }
+  return Status::OK();
+}
+
 void MemTableListVersion::AddIterators(
    const ReadOptions& options, std::vector<InternalIterator*>* iterator_list,
    Arena* arena) {
@@ -82,6 +82,9 @@ class MemTableListVersion {

  Status AddRangeTombstoneIterators(const ReadOptions& read_opts, Arena* arena,
                                    RangeDelAggregator* range_del_agg);
+  Status AddRangeTombstoneIterators(
+      const ReadOptions& read_opts,
+      std::vector<InternalIterator*>* range_del_iters);

  void AddIterators(const ReadOptions& options,
                    std::vector<InternalIterator*>* iterator_list,
@@ -134,7 +134,7 @@ TEST_F(MemTableListTest, GetTest) {

  WriteBufferManager wb(options.db_write_buffer_size);
  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
-                               kMaxSequenceNumber);
+                               kMaxSequenceNumber, 0 /* column_family_id */);
  mem->Ref();

  // Write some keys to this memtable.
@@ -173,7 +173,7 @@ TEST_F(MemTableListTest, GetTest) {
  // Create another memtable and write some keys to it
  WriteBufferManager wb2(options.db_write_buffer_size);
  MemTable* mem2 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb2,
-                                kMaxSequenceNumber);
+                                kMaxSequenceNumber, 0 /* column_family_id */);
  mem2->Ref();

  mem2->Add(++seq, kTypeDeletion, "key1", "");
@@ -241,7 +241,7 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {

  WriteBufferManager wb(options.db_write_buffer_size);
  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
-                               kMaxSequenceNumber);
+                               kMaxSequenceNumber, 0 /* column_family_id */);
  mem->Ref();

  // Write some keys to this memtable.
@@ -319,7 +319,7 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {
  // Create another memtable and write some keys to it
  WriteBufferManager wb2(options.db_write_buffer_size);
  MemTable* mem2 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb2,
-                                kMaxSequenceNumber);
+                                kMaxSequenceNumber, 0 /* column_family_id */);
  mem2->Ref();

  mem2->Add(++seq, kTypeDeletion, "key1", "");
@@ -344,7 +344,7 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {
  // Add a third memtable to push the first memtable out of the history
  WriteBufferManager wb3(options.db_write_buffer_size);
  MemTable* mem3 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb3,
-                                kMaxSequenceNumber);
+                                kMaxSequenceNumber, 0 /* column_family_id */);
  mem3->Ref();
  list.Add(mem3, &to_delete);
  ASSERT_EQ(1, list.NumNotFlushed());
@@ -418,7 +418,7 @@ TEST_F(MemTableListTest, FlushPendingTest) {
  MutableCFOptions mutable_cf_options(options);
  for (int i = 0; i < num_tables; i++) {
    MemTable* mem = new MemTable(cmp, ioptions, mutable_cf_options, &wb,
-                                 kMaxSequenceNumber);
+                                 kMaxSequenceNumber, 0 /* column_family_id */);
    mem->Ref();

    std::string value;
@@ -23,7 +23,8 @@ Status MergeHelper::TimedFullMerge(const MergeOperator* merge_operator,
                                   const std::vector<Slice>& operands,
                                   std::string* result, Logger* logger,
                                   Statistics* statistics, Env* env,
-                                   Slice* result_operand) {
+                                   Slice* result_operand,
+                                   bool update_num_ops_stats) {
  assert(merge_operator != nullptr);

  if (operands.size() == 0) {
@@ -32,6 +33,11 @@ Status MergeHelper::TimedFullMerge(const MergeOperator* merge_operator,
    return Status::OK();
  }

+  if (update_num_ops_stats) {
+    MeasureTime(statistics, READ_NUM_MERGE_OPERANDS,
+                static_cast<uint64_t>(operands.size()));
+  }
+
  bool success;
  Slice tmp_result_operand(nullptr, 0);
  const MergeOperator::MergeOperationInput merge_in(key, value, operands,
@@ -54,6 +54,8 @@ class MergeHelper {
  // Wrapper around MergeOperator::FullMergeV2() that records perf statistics.
  // Result of merge will be written to result if status returned is OK.
  // If operands is empty, the value will simply be copied to result.
+  // Set `update_num_ops_stats` to true if it is from a user read, so that
+  // the latency is sensitive.
  // Returns one of the following statuses:
  // - OK: Entries were successfully merged.
  // - Corruption: Merge operator reported unsuccessful merge.
@@ -62,7 +64,8 @@ class MergeHelper {
                               const std::vector<Slice>& operands,
                               std::string* result, Logger* logger,
                               Statistics* statistics, Env* env,
-                               Slice* result_operand = nullptr);
+                               Slice* result_operand = nullptr,
+                               bool update_num_ops_stats = false);

  // Merge entries until we hit
  //     - a corrupted key
@@ -141,7 +141,9 @@ class Counters {
  // mapped to a levedb Put
  bool set(const std::string& key, uint64_t value) {
    // just treat the internal rep of int64 as the string
-    Slice slice((char *)&value, sizeof(value));
+    char buf[sizeof(value)];
+    EncodeFixed64(buf, value);
+    Slice slice(buf, sizeof(value));
    auto s = db_->Put(put_option_, key, slice);

    if (s.ok()) {
@@ -375,7 +377,9 @@ void testSingleBatchSuccessiveMerge(DB* db, size_t max_num_merges,

  Slice key("BatchSuccessiveMerge");
  uint64_t merge_value = 1;
-  Slice merge_value_slice((char *)&merge_value, sizeof(merge_value));
+  char buf[sizeof(merge_value)];
+  EncodeFixed64(buf, merge_value);
+  Slice merge_value_slice(buf, sizeof(merge_value));

  // Create the batch
  WriteBatch batch;
@@ -87,13 +87,13 @@ TEST_F(PerfContextTest, SeekIntoDeletion) {
    std::string key = "k" + ToString(i);
    std::string value;

-    perf_context.Reset();
+    get_perf_context()->Reset();
    StopWatchNano timer(Env::Default());
    timer.Start();
    auto status = db->Get(read_options, key, &value);
    auto elapsed_nanos = timer.ElapsedNanos();
    ASSERT_TRUE(status.IsNotFound());
-    hist_get.Add(perf_context.user_key_comparison_count);
+    hist_get.Add(get_perf_context()->user_key_comparison_count);
    hist_get_time.Add(elapsed_nanos);
  }

@@ -106,19 +106,19 @@ TEST_F(PerfContextTest, SeekIntoDeletion) {
    HistogramImpl hist_seek_to_first;
    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));

-    perf_context.Reset();
+    get_perf_context()->Reset();
    StopWatchNano timer(Env::Default(), true);
    iter->SeekToFirst();
-    hist_seek_to_first.Add(perf_context.user_key_comparison_count);
+    hist_seek_to_first.Add(get_perf_context()->user_key_comparison_count);
    auto elapsed_nanos = timer.ElapsedNanos();

    if (FLAGS_verbose) {
      std::cout << "SeekToFirst uesr key comparison: \n"
                << hist_seek_to_first.ToString()
-                << "ikey skipped: " << perf_context.internal_key_skipped_count
+                << "ikey skipped: " << get_perf_context()->internal_key_skipped_count
                << "\n"
                << "idelete skipped: "
-                << perf_context.internal_delete_skipped_count << "\n"
+                << get_perf_context()->internal_delete_skipped_count << "\n"
                << "elapsed: " << elapsed_nanos << "\n";
    }
  }
@@ -128,26 +128,26 @@ TEST_F(PerfContextTest, SeekIntoDeletion) {
    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
    std::string key = "k" + ToString(i);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    StopWatchNano timer(Env::Default(), true);
    iter->Seek(key);
    auto elapsed_nanos = timer.ElapsedNanos();
-    hist_seek.Add(perf_context.user_key_comparison_count);
+    hist_seek.Add(get_perf_context()->user_key_comparison_count);
    if (FLAGS_verbose) {
-      std::cout << "seek cmp: " << perf_context.user_key_comparison_count
-                << " ikey skipped " << perf_context.internal_key_skipped_count
+      std::cout << "seek cmp: " << get_perf_context()->user_key_comparison_count
+                << " ikey skipped " << get_perf_context()->internal_key_skipped_count
                << " idelete skipped "
-                << perf_context.internal_delete_skipped_count
+                << get_perf_context()->internal_delete_skipped_count
                << " elapsed: " << elapsed_nanos << "ns\n";
    }

-    perf_context.Reset();
+    get_perf_context()->Reset();
    ASSERT_TRUE(iter->Valid());
    StopWatchNano timer2(Env::Default(), true);
    iter->Next();
    auto elapsed_nanos2 = timer2.ElapsedNanos();
    if (FLAGS_verbose) {
-      std::cout << "next cmp: " << perf_context.user_key_comparison_count
+      std::cout << "next cmp: " << get_perf_context()->user_key_comparison_count
                << "elapsed: " << elapsed_nanos2 << "ns\n";
    }
  }
@@ -263,18 +263,18 @@ void ProfileQueries(bool enabled_time = false) {

    std::vector<std::string> values;

-    perf_context.Reset();
+    get_perf_context()->Reset();
    db->Put(write_options, key, value);
    if (++num_mutex_waited > 3) {
 #ifndef NDEBUG
      ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 0U);
 #endif
    }
-    hist_write_pre_post.Add(perf_context.write_pre_and_post_process_time);
-    hist_write_wal_time.Add(perf_context.write_wal_time);
-    hist_write_memtable_time.Add(perf_context.write_memtable_time);
-    hist_put.Add(perf_context.user_key_comparison_count);
-    total_db_mutex_nanos += perf_context.db_mutex_lock_nanos;
+    hist_write_pre_post.Add(get_perf_context()->write_pre_and_post_process_time);
+    hist_write_wal_time.Add(get_perf_context()->write_wal_time);
+    hist_write_memtable_time.Add(get_perf_context()->write_memtable_time);
+    hist_put.Add(get_perf_context()->user_key_comparison_count);
+    total_db_mutex_nanos += get_perf_context()->db_mutex_lock_nanos;
  }
 #ifndef NDEBUG
  ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 0U);
@@ -291,24 +291,24 @@ void ProfileQueries(bool enabled_time = false) {
    std::vector<Slice> multiget_keys = {Slice(key)};
    std::vector<std::string> values;

-    perf_context.Reset();
+    get_perf_context()->Reset();
    ASSERT_OK(db->Get(read_options, key, &value));
    ASSERT_EQ(expected_value, value);
-    hist_get_snapshot.Add(perf_context.get_snapshot_time);
-    hist_get_memtable.Add(perf_context.get_from_memtable_time);
-    hist_get_files.Add(perf_context.get_from_output_files_time);
-    hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
-    hist_get_post_process.Add(perf_context.get_post_process_time);
-    hist_get.Add(perf_context.user_key_comparison_count);
+    hist_get_snapshot.Add(get_perf_context()->get_snapshot_time);
+    hist_get_memtable.Add(get_perf_context()->get_from_memtable_time);
+    hist_get_files.Add(get_perf_context()->get_from_output_files_time);
+    hist_num_memtable_checked.Add(get_perf_context()->get_from_memtable_count);
+    hist_get_post_process.Add(get_perf_context()->get_post_process_time);
+    hist_get.Add(get_perf_context()->user_key_comparison_count);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    db->MultiGet(read_options, multiget_keys, &values);
-    hist_mget_snapshot.Add(perf_context.get_snapshot_time);
-    hist_mget_memtable.Add(perf_context.get_from_memtable_time);
-    hist_mget_files.Add(perf_context.get_from_output_files_time);
-    hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
-    hist_mget_post_process.Add(perf_context.get_post_process_time);
-    hist_mget.Add(perf_context.user_key_comparison_count);
+    hist_mget_snapshot.Add(get_perf_context()->get_snapshot_time);
+    hist_mget_memtable.Add(get_perf_context()->get_from_memtable_time);
+    hist_mget_files.Add(get_perf_context()->get_from_output_files_time);
+    hist_mget_num_memtable_checked.Add(get_perf_context()->get_from_memtable_count);
+    hist_mget_post_process.Add(get_perf_context()->get_post_process_time);
+    hist_mget.Add(get_perf_context()->user_key_comparison_count);
  }

  if (FLAGS_verbose) {
@@ -392,24 +392,24 @@ void ProfileQueries(bool enabled_time = false) {
    std::vector<Slice> multiget_keys = {Slice(key)};
    std::vector<std::string> values;

-    perf_context.Reset();
+    get_perf_context()->Reset();
    ASSERT_OK(db->Get(read_options, key, &value));
    ASSERT_EQ(expected_value, value);
-    hist_get_snapshot.Add(perf_context.get_snapshot_time);
-    hist_get_memtable.Add(perf_context.get_from_memtable_time);
-    hist_get_files.Add(perf_context.get_from_output_files_time);
-    hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
-    hist_get_post_process.Add(perf_context.get_post_process_time);
-    hist_get.Add(perf_context.user_key_comparison_count);
+    hist_get_snapshot.Add(get_perf_context()->get_snapshot_time);
+    hist_get_memtable.Add(get_perf_context()->get_from_memtable_time);
+    hist_get_files.Add(get_perf_context()->get_from_output_files_time);
+    hist_num_memtable_checked.Add(get_perf_context()->get_from_memtable_count);
+    hist_get_post_process.Add(get_perf_context()->get_post_process_time);
+    hist_get.Add(get_perf_context()->user_key_comparison_count);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    db->MultiGet(read_options, multiget_keys, &values);
-    hist_mget_snapshot.Add(perf_context.get_snapshot_time);
-    hist_mget_memtable.Add(perf_context.get_from_memtable_time);
-    hist_mget_files.Add(perf_context.get_from_output_files_time);
-    hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
-    hist_mget_post_process.Add(perf_context.get_post_process_time);
-    hist_mget.Add(perf_context.user_key_comparison_count);
+    hist_mget_snapshot.Add(get_perf_context()->get_snapshot_time);
+    hist_mget_memtable.Add(get_perf_context()->get_from_memtable_time);
+    hist_mget_files.Add(get_perf_context()->get_from_output_files_time);
+    hist_mget_num_memtable_checked.Add(get_perf_context()->get_from_memtable_count);
+    hist_mget_post_process.Add(get_perf_context()->get_post_process_time);
+    hist_mget.Add(get_perf_context()->user_key_comparison_count);
  }

  if (FLAGS_verbose) {
@@ -512,13 +512,13 @@ TEST_F(PerfContextTest, SeekKeyComparison) {
    std::string key = "k" + ToString(i);
    std::string value = "v" + ToString(i);

-    perf_context.Reset();
+    get_perf_context()->Reset();
    timer.Start();
    db->Put(write_options, key, value);
    auto put_time = timer.ElapsedNanos();
    hist_put_time.Add(put_time);
-    hist_wal_time.Add(perf_context.write_wal_time);
-    hist_time_diff.Add(put_time - perf_context.write_wal_time);
+    hist_wal_time.Add(get_perf_context()->write_wal_time);
+    hist_time_diff.Add(put_time - get_perf_context()->write_wal_time);
  }

  if (FLAGS_verbose) {
@@ -535,18 +535,18 @@ TEST_F(PerfContextTest, SeekKeyComparison) {
    std::string value = "v" + ToString(i);

    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
-    perf_context.Reset();
+    get_perf_context()->Reset();
    iter->Seek(key);
    ASSERT_TRUE(iter->Valid());
    ASSERT_EQ(iter->value().ToString(), value);
-    hist_seek.Add(perf_context.user_key_comparison_count);
+    hist_seek.Add(get_perf_context()->user_key_comparison_count);
  }

  std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
  for (iter->SeekToFirst(); iter->Valid();) {
-    perf_context.Reset();
+    get_perf_context()->Reset();
    iter->Next();
-    hist_next.Add(perf_context.user_key_comparison_count);
+    hist_next.Add(get_perf_context()->user_key_comparison_count);
  }

  if (FLAGS_verbose) {
@@ -564,16 +564,16 @@ TEST_F(PerfContextTest, DBMutexLockCounter) {
    mutex.Lock();
    rocksdb::port::Thread child_thread([&] {
      SetPerfLevel(perf_level);
-      perf_context.Reset();
-      ASSERT_EQ(perf_context.db_mutex_lock_nanos, 0);
+      get_perf_context()->Reset();
+      ASSERT_EQ(get_perf_context()->db_mutex_lock_nanos, 0);
      mutex.Lock();
      mutex.Unlock();
      if (perf_level == PerfLevel::kEnableTimeExceptForMutex ||
          stats_code[c] != DB_MUTEX_WAIT_MICROS) {
-        ASSERT_EQ(perf_context.db_mutex_lock_nanos, 0);
+        ASSERT_EQ(get_perf_context()->db_mutex_lock_nanos, 0);
      } else {
        // increment the counter only when it's a DB Mutex
-        ASSERT_GT(perf_context.db_mutex_lock_nanos, 0);
+        ASSERT_GT(get_perf_context()->db_mutex_lock_nanos, 0);
      }
    });
    Env::Default()->SleepForMicroseconds(100);
@@ -589,28 +589,28 @@ TEST_F(PerfContextTest, FalseDBMutexWait) {
  for (int c = 0; c < 2; ++c) {
    InstrumentedMutex mutex(nullptr, Env::Default(), stats_code[c]);
    InstrumentedCondVar lock(&mutex);
-    perf_context.Reset();
+    get_perf_context()->Reset();
    mutex.Lock();
    lock.TimedWait(100);
    mutex.Unlock();
    if (stats_code[c] == static_cast<int>(DB_MUTEX_WAIT_MICROS)) {
      // increment the counter only when it's a DB Mutex
-      ASSERT_GT(perf_context.db_condition_wait_nanos, 0);
+      ASSERT_GT(get_perf_context()->db_condition_wait_nanos, 0);
    } else {
-      ASSERT_EQ(perf_context.db_condition_wait_nanos, 0);
+      ASSERT_EQ(get_perf_context()->db_condition_wait_nanos, 0);
    }
  }
 }

 TEST_F(PerfContextTest, ToString) {
-  perf_context.Reset();
-  perf_context.block_read_count = 12345;
+  get_perf_context()->Reset();
+  get_perf_context()->block_read_count = 12345;

-  std::string zero_included = perf_context.ToString();
+  std::string zero_included = get_perf_context()->ToString();
  ASSERT_NE(std::string::npos, zero_included.find("= 0"));
  ASSERT_NE(std::string::npos, zero_included.find("= 12345"));

-  std::string zero_excluded = perf_context.ToString(true);
+  std::string zero_excluded = get_perf_context()->ToString(true);
  ASSERT_EQ(std::string::npos, zero_excluded.find("= 0"));
  ASSERT_NE(std::string::npos, zero_excluded.find("= 12345"));
 }
@@ -631,21 +631,36 @@ TEST_F(PerfContextTest, MergeOperatorTime) {
  ASSERT_OK(db->Merge(WriteOptions(), "k1", "val4"));

  SetPerfLevel(kEnableTime);
-  perf_context.Reset();
+  get_perf_context()->Reset();
  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
-  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+#ifdef OS_SOLARIS
+  for (int i = 0; i < 100; i++) {
+    ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  }
+#endif
+  EXPECT_GT(get_perf_context()->merge_operator_time_nanos, 0);

  ASSERT_OK(db->Flush(FlushOptions()));

-  perf_context.Reset();
+  get_perf_context()->Reset();
  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
-  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+#ifdef OS_SOLARIS
+  for (int i = 0; i < 100; i++) {
+    ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  }
+#endif
+  EXPECT_GT(get_perf_context()->merge_operator_time_nanos, 0);

  ASSERT_OK(db->CompactRange(CompactRangeOptions(), nullptr, nullptr));

-  perf_context.Reset();
+  get_perf_context()->Reset();
  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
-  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+#ifdef OS_SOLARIS
+  for (int i = 0; i < 100; i++) {
+    ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  }
+#endif
+  EXPECT_GT(get_perf_context()->merge_operator_time_nanos, 0);

  delete db;
 }
@@ -32,6 +32,7 @@ int main() {
 #include "util/string_util.h"
 #include "util/testharness.h"
 #include "utilities/merge_operators.h"
+#include "util/coding.h"

 using GFLAGS::ParseCommandLineFlags;

@@ -65,12 +66,16 @@ struct TestKey {
 };

 // return a slice backed by test_key
-inline Slice TestKeyToSlice(const TestKey& test_key) {
-  return Slice((const char*)&test_key, sizeof(test_key));
+inline Slice TestKeyToSlice(std::string &s, const TestKey& test_key) {
+  s.clear();
+  PutFixed64(&s, test_key.prefix);
+  PutFixed64(&s, test_key.sorted);
+  return Slice(s.c_str(), s.size());
 }

-inline const TestKey* SliceToTestKey(const Slice& slice) {
-  return (const TestKey*)slice.data();
+inline const TestKey SliceToTestKey(const Slice& slice) {
+  return TestKey(DecodeFixed64(slice.data()),
+    DecodeFixed64(slice.data() + 8));
 }

 class TestKeyComparator : public Comparator {
@@ -79,8 +84,10 @@ class TestKeyComparator : public Comparator {
  // Compare needs to be aware of the possibility of a and/or b is
  // prefix only
  virtual int Compare(const Slice& a, const Slice& b) const override {
-    const TestKey* key_a = SliceToTestKey(a);
-    const TestKey* key_b = SliceToTestKey(b);
+    const TestKey kkey_a = SliceToTestKey(a);
+    const TestKey kkey_b = SliceToTestKey(b);
+    const TestKey *key_a = &kkey_a;
+    const TestKey *key_b = &kkey_b;
    if (key_a->prefix != key_b->prefix) {
      if (key_a->prefix < key_b->prefix) return -1;
      if (key_a->prefix > key_b->prefix) return 1;
@@ -111,7 +118,8 @@ class TestKeyComparator : public Comparator {
  }

  bool operator()(const TestKey& a, const TestKey& b) const {
-    return Compare(TestKeyToSlice(a), TestKeyToSlice(b)) < 0;
+    std::string sa, sb;
+    return Compare(TestKeyToSlice(sa, a), TestKeyToSlice(sb, b)) < 0;
  }

  virtual const char* Name() const override {
@@ -128,30 +136,35 @@ namespace {
 void PutKey(DB* db, WriteOptions write_options, uint64_t prefix,
            uint64_t suffix, const Slice& value) {
  TestKey test_key(prefix, suffix);
-  Slice key = TestKeyToSlice(test_key);
+  std::string s;
+  Slice key = TestKeyToSlice(s, test_key);
  ASSERT_OK(db->Put(write_options, key, value));
 }

 void PutKey(DB* db, WriteOptions write_options, const TestKey& test_key,
            const Slice& value) {
-  Slice key = TestKeyToSlice(test_key);
+  std::string s;
+  Slice key = TestKeyToSlice(s, test_key);
  ASSERT_OK(db->Put(write_options, key, value));
 }

 void MergeKey(DB* db, WriteOptions write_options, const TestKey& test_key,
              const Slice& value) {
-  Slice key = TestKeyToSlice(test_key);
+  std::string s;
+  Slice key = TestKeyToSlice(s, test_key);
  ASSERT_OK(db->Merge(write_options, key, value));
 }

 void DeleteKey(DB* db, WriteOptions write_options, const TestKey& test_key) {
-  Slice key = TestKeyToSlice(test_key);
+  std::string s;
+  Slice key = TestKeyToSlice(s, test_key);
  ASSERT_OK(db->Delete(write_options, key));
 }

 void SeekIterator(Iterator* iter, uint64_t prefix, uint64_t suffix) {
  TestKey test_key(prefix, suffix);
-  Slice key = TestKeyToSlice(test_key);
+  std::string s;
+  Slice key = TestKeyToSlice(s, test_key);
  iter->Seek(key);
 }

@@ -160,7 +173,8 @@ const std::string kNotFoundResult = "NOT_FOUND";
 std::string Get(DB* db, const ReadOptions& read_options, uint64_t prefix,
                uint64_t suffix) {
  TestKey test_key(prefix, suffix);
-  Slice key = TestKeyToSlice(test_key);
+  std::string s2;
+  Slice key = TestKeyToSlice(s2, test_key);

  std::string result;
  Status s = db->Get(read_options, key, &result);
@@ -527,7 +541,9 @@ TEST_F(PrefixTest, PrefixValid) {
      PutKey(db.get(), write_options, 12345, 9, v19);
      PutKey(db.get(), write_options, 12346, 8, v16);
      db->Flush(FlushOptions());
-      db->Delete(write_options, TestKeyToSlice(TestKey(12346, 8)));
+      TestKey test_key(12346, 8);
+      std::string s;
+      db->Delete(write_options, TestKeyToSlice(s, test_key));
      db->Flush(FlushOptions());
      read_options.prefix_same_as_start = true;
      std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
@@ -583,14 +599,15 @@ TEST_F(PrefixTest, DynamicPrefixIterator) {
       for (uint64_t sorted = 0; sorted < FLAGS_items_per_prefix; sorted++) {
        TestKey test_key(prefix, sorted);

-        Slice key = TestKeyToSlice(test_key);
+        std::string s;
+        Slice key = TestKeyToSlice(s, test_key);
        std::string value(FLAGS_value_size, 0);

-        perf_context.Reset();
+        get_perf_context()->Reset();
        StopWatchNano timer(Env::Default(), true);
        ASSERT_OK(db->Put(write_options, key, value));
        hist_put_time.Add(timer.ElapsedNanos());
-        hist_put_comparison.Add(perf_context.user_key_comparison_count);
+        hist_put_comparison.Add(get_perf_context()->user_key_comparison_count);
      }
    }

@@ -605,10 +622,11 @@ TEST_F(PrefixTest, DynamicPrefixIterator) {

    for (auto prefix : prefixes) {
      TestKey test_key(prefix, FLAGS_items_per_prefix / 2);
-      Slice key = TestKeyToSlice(test_key);
+      std::string s;
+      Slice key = TestKeyToSlice(s, test_key);
      std::string value = "v" + ToString(0);

-      perf_context.Reset();
+      get_perf_context()->Reset();
      StopWatchNano timer(Env::Default(), true);
      auto key_prefix = options.prefix_extractor->Transform(key);
      uint64_t total_keys = 0;
@@ -622,7 +640,7 @@ TEST_F(PrefixTest, DynamicPrefixIterator) {
        total_keys++;
      }
      hist_seek_time.Add(timer.ElapsedNanos());
-      hist_seek_comparison.Add(perf_context.user_key_comparison_count);
+      hist_seek_comparison.Add(get_perf_context()->user_key_comparison_count);
      ASSERT_EQ(total_keys, FLAGS_items_per_prefix - FLAGS_items_per_prefix/2);
    }

@@ -639,13 +657,14 @@ TEST_F(PrefixTest, DynamicPrefixIterator) {
         prefix < FLAGS_total_prefixes + 10000;
         prefix++) {
      TestKey test_key(prefix, 0);
-      Slice key = TestKeyToSlice(test_key);
+      std::string s;
+      Slice key = TestKeyToSlice(s, test_key);

-      perf_context.Reset();
+      get_perf_context()->Reset();
      StopWatchNano timer(Env::Default(), true);
      iter->Seek(key);
      hist_no_seek_time.Add(timer.ElapsedNanos());
-      hist_no_seek_comparison.Add(perf_context.user_key_comparison_count);
+      hist_no_seek_comparison.Add(get_perf_context()->user_key_comparison_count);
      ASSERT_TRUE(!iter->Valid());
    }

@@ -725,8 +744,8 @@ TEST_F(PrefixTest, PrefixSeekModePrev) {
      ASSERT_TRUE(iter->Valid());
      if (FLAGS_enable_print) {
        std::cout << "round " << prefix
-                  << " iter: " << SliceToTestKey(iter->key())->prefix
-                  << SliceToTestKey(iter->key())->sorted
+                  << " iter: " << SliceToTestKey(iter->key()).prefix
+                  << SliceToTestKey(iter->key()).sorted
                  << " | map: " << it->first.prefix << it->first.sorted << " | "
                  << iter->value().ToString() << " " << it->second << std::endl;
      }
@@ -747,16 +766,16 @@ TEST_F(PrefixTest, PrefixSeekModePrev) {
          }
        }
        if (!iter->Valid() ||
-            SliceToTestKey(iter->key())->prefix != stored_prefix) {
+            SliceToTestKey(iter->key()).prefix != stored_prefix) {
          break;
        }
-        stored_prefix = SliceToTestKey(iter->key())->prefix;
+        stored_prefix = SliceToTestKey(iter->key()).prefix;
        ASSERT_TRUE(iter->Valid());
        ASSERT_NE(it, whole_map.end());
        ASSERT_EQ(iter->value(), it->second);
        if (FLAGS_enable_print) {
-          std::cout << "iter: " << SliceToTestKey(iter->key())->prefix
-                    << SliceToTestKey(iter->key())->sorted
+          std::cout << "iter: " << SliceToTestKey(iter->key()).prefix
+                    << SliceToTestKey(iter->key()).sorted
                    << " | map: " << it->first.prefix << it->first.sorted
                    << " | " << iter->value().ToString() << " " << it->second
                    << std::endl;
@@ -810,7 +829,8 @@ TEST_F(PrefixTest, PrefixSeekModePrev3) {
  options.write_buffer_size = 1024 * 1024;
  std::string v14("v14");
  TestKey upper_bound_key = TestKey(1, 5);
-  Slice upper_bound = TestKeyToSlice(upper_bound_key);
+  std::string s;
+  Slice upper_bound = TestKeyToSlice(s, upper_bound_key);

  {
    DestroyDB(kDbName, Options());
@@ -314,7 +314,8 @@ class Repairer {
    // Open the log file
    std::string logname = LogFileName(dbname_, log);
    unique_ptr<SequentialFile> lfile;
-    Status status = env_->NewSequentialFile(logname, &lfile, env_options_);
+    Status status = env_->NewSequentialFile(
+        logname, &lfile, env_->OptimizeForLogRead(env_options_));
    if (!status.ok()) {
      return status;
    }
@@ -260,6 +260,44 @@ InternalIterator* TableCache::NewIterator(
  return result;
 }

+InternalIterator* TableCache::NewRangeTombstoneIterator(
+    const ReadOptions& options, const EnvOptions& env_options,
+    const InternalKeyComparator& icomparator, const FileDescriptor& fd,
+    HistogramImpl* file_read_hist, bool skip_filters, int level) {
+  Status s;
+  TableReader* table_reader = nullptr;
+  Cache::Handle* handle = nullptr;
+  table_reader = fd.table_reader;
+  if (table_reader == nullptr) {
+    s = FindTable(env_options, icomparator, fd, &handle,
+                  options.read_tier == kBlockCacheTier /* no_io */,
+                  true /* record read_stats */, file_read_hist, skip_filters,
+                  level);
+    if (s.ok()) {
+      table_reader = GetTableReaderFromHandle(handle);
+    }
+  }
+  InternalIterator* result = nullptr;
+  if (s.ok()) {
+    result = table_reader->NewRangeTombstoneIterator(options);
+    if (result != nullptr) {
+      if (handle != nullptr) {
+        result->RegisterCleanup(&UnrefEntry, cache_, handle);
+      }
+    }
+  }
+  if (result == nullptr && handle != nullptr) {
+    // the range deletion block didn't exist, or there was a failure between
+    // getting handle and getting iterator.
+    ReleaseHandle(handle);
+  }
+  if (!s.ok()) {
+    assert(result == nullptr);
+    result = NewErrorInternalIterator(s);
+  }
+  return result;
+}
+
 Status TableCache::Get(const ReadOptions& options,
                       const InternalKeyComparator& internal_comparator,
                       const FileDescriptor& fd, const Slice& k,
@@ -58,6 +58,12 @@ class TableCache {
      HistogramImpl* file_read_hist = nullptr, bool for_compaction = false,
      Arena* arena = nullptr, bool skip_filters = false, int level = -1);

+  InternalIterator* NewRangeTombstoneIterator(
+      const ReadOptions& options, const EnvOptions& toptions,
+      const InternalKeyComparator& internal_comparator,
+      const FileDescriptor& file_fd, HistogramImpl* file_read_hist,
+      bool skip_filters, int level);
+
  // If a seek to internal key "k" in specified file finds an entry,
  // call (*handle_result)(arg, found_key, found_value) repeatedly until
  // it returns false.
@@ -115,6 +121,10 @@ class TableCache {
  // Release the handle from a cache
  void ReleaseHandle(Cache::Handle* handle);

+  // Capacity of the backing Cache that indicates inifinite TableCache capacity.
+  // For example when max_open_files is -1 we set the backing Cache to this.
+  static const int kInfiniteCapacity = 0x400000;
+
 private:
  // Build a table reader
  Status GetTableReader(const EnvOptions& env_options,
@@ -45,17 +45,18 @@ Status TransactionLogIteratorImpl::OpenLogFile(
  Env* env = options_->env;
  unique_ptr<SequentialFile> file;
  Status s;
+  EnvOptions optimized_env_options = env->OptimizeForLogRead(soptions_);
  if (logFile->Type() == kArchivedLogFile) {
    std::string fname = ArchivedLogFileName(dir_, logFile->LogNumber());
-    s = env->NewSequentialFile(fname, &file, soptions_);
+    s = env->NewSequentialFile(fname, &file, optimized_env_options);
  } else {
    std::string fname = LogFileName(dir_, logFile->LogNumber());
-    s = env->NewSequentialFile(fname, &file, soptions_);
+    s = env->NewSequentialFile(fname, &file, optimized_env_options);
    if (!s.ok()) {
      //  If cannot open file in DB directory.
      //  Try the archive dir, as it could have moved in the meanwhile.
      fname = ArchivedLogFileName(dir_, logFile->LogNumber());
-      s = env->NewSequentialFile(fname, &file, soptions_);
+      s = env->NewSequentialFile(fname, &file, optimized_env_options);
    }
  }
  if (s.ok()) {
@@ -134,8 +134,8 @@ class FilePicker {

        // Do key range filtering of files or/and fractional cascading if:
        // (1) not all the files are in level 0, or
-        // (2) there are more than 3 Level 0 files
-        // If there are only 3 or less level 0 files in the system, we skip
+        // (2) there are more than 3 current level files
+        // If there are only 3 or less current level files in the system, we skip
        // the key range filtering. In this case, more likely, the system is
        // highly tuned to minimize number of tables queried by each query,
        // so it is unlikely that key range filtering is more efficient than
@@ -863,6 +863,22 @@ void Version::AddIteratorsForLevel(const ReadOptions& read_options,
  }
 }

+void Version::AddRangeDelIteratorsForLevel(
+    const ReadOptions& read_options, const EnvOptions& soptions, int level,
+    std::vector<InternalIterator*>* range_del_iters) {
+  range_del_iters->clear();
+  for (size_t i = 0; i < storage_info_.LevelFilesBrief(level).num_files; i++) {
+    const auto& file = storage_info_.LevelFilesBrief(level).files[i];
+    auto* range_del_iter = cfd_->table_cache()->NewRangeTombstoneIterator(
+        read_options, soptions, cfd_->internal_comparator(), file.fd,
+        cfd_->internal_stats()->GetFileReadHist(level),
+        false /* skip_filters */, level);
+    if (range_del_iter != nullptr) {
+      range_del_iters->push_back(range_del_iter);
+    }
+  }
+}
+
 VersionStorageInfo::VersionStorageInfo(
    const InternalKeyComparator* internal_comparator,
    const Comparator* user_comparator, int levels,
@@ -1014,7 +1030,8 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
    std::string* str_value = value != nullptr ? value->GetSelf() : nullptr;
    *status = MergeHelper::TimedFullMerge(
        merge_operator_, user_key, nullptr, merge_context->GetOperands(),
-        str_value, info_log_, db_statistics_, env_);
+        str_value, info_log_, db_statistics_, env_,
+        nullptr /* result_operand */, true);
    if (LIKELY(value != nullptr)) {
      value->PinSelf();
    }
@@ -1126,8 +1143,11 @@ void Version::UpdateAccumulatedStats(bool update_stats) {
          storage_info_.UpdateAccumulatedStats(file_meta);
          // when option "max_open_files" is -1, all the file metadata has
          // already been read, so MaybeInitializeFileMetaData() won't incur
-          // any I/O cost.
-          if (vset_->db_options_->max_open_files == -1) {
+          // any I/O cost. "max_open_files=-1" means that the table cache passed
+          // to the VersionSet and then to the ColumnFamilySet has a size of
+          // TableCache::kInfiniteCapacity
+          if (vset_->GetColumnFamilySet()->get_table_cache()->GetCapacity() ==
+              TableCache::kInfiniteCapacity) {
            continue;
          }
          if (++init_count >= kMaxInitCount) {
@@ -1211,15 +1231,18 @@ void VersionStorageInfo::EstimateCompactionBytesNeeded(
  // accumulated bytes.

  uint64_t bytes_compact_to_next_level = 0;
+  uint64_t level_size = 0;
+  for (auto* f : files_[0]) {
+    level_size += f->fd.GetFileSize();
+  }
  // Level 0
  bool level0_compact_triggered = false;
-  if (static_cast<int>(files_[0].size()) >
-      mutable_cf_options.level0_file_num_compaction_trigger) {
+  if (static_cast<int>(files_[0].size()) >=
+          mutable_cf_options.level0_file_num_compaction_trigger ||
+      level_size >= mutable_cf_options.max_bytes_for_level_base) {
    level0_compact_triggered = true;
-    for (auto* f : files_[0]) {
-      bytes_compact_to_next_level += f->fd.GetFileSize();
-    }
-    estimated_compaction_needed_bytes_ = bytes_compact_to_next_level;
+    estimated_compaction_needed_bytes_ = level_size;
+    bytes_compact_to_next_level = level_size;
  } else {
    estimated_compaction_needed_bytes_ = 0;
  }
@@ -1227,7 +1250,7 @@ void VersionStorageInfo::EstimateCompactionBytesNeeded(
  // Level 1 and up.
  uint64_t bytes_next_level = 0;
  for (int level = base_level(); level <= MaxInputLevel(); level++) {
-    uint64_t level_size = 0;
+    level_size = 0;
    if (bytes_next_level > 0) {
 #ifndef NDEBUG
      uint64_t level_size2 = 0;
@@ -1314,6 +1337,13 @@ void VersionStorageInfo::ComputeCompactionScore(
        score =
            static_cast<double>(total_size) /
            immutable_cf_options.compaction_options_fifo.max_table_files_size;
+        if (immutable_cf_options.compaction_options_fifo.allow_compaction) {
+          score = std::max(
+              static_cast<double>(num_sorted_runs) /
+                  mutable_cf_options.level0_file_num_compaction_trigger,
+              score);
+        }
+
      } else {
        score = static_cast<double>(num_sorted_runs) /
                mutable_cf_options.level0_file_num_compaction_trigger;
@@ -2115,7 +2145,11 @@ void VersionStorageInfo::CalculateBaseBytes(const ImmutableCFOptions& ioptions,
          level_size = MultiplyCheckOverflow(
              level_size, options.max_bytes_for_level_multiplier);
        }
-        level_max_bytes_[i] = level_size;
+        // Don't set any level below base_bytes_max. Otherwise, the LSM can
+        // assume an hourglass shape where L1+ sizes are smaller than L0. This
+        // causes compaction scoring, which depends on level sizes, to favor L1+
+        // at the expense of L0, which may fill up and stall.
+        level_max_bytes_[i] = std::max(level_size, base_bytes_max);
      }
    }
  }
@@ -2385,7 +2419,8 @@ Status VersionSet::LogAndApply(ColumnFamilyData* column_family_data,

    TEST_SYNC_POINT("VersionSet::LogAndApply:WriteManifest");
    if (!w.edit_list.front()->IsColumnFamilyManipulation() &&
-        db_options_->max_open_files == -1) {
+        this->GetColumnFamilySet()->get_table_cache()->GetCapacity() ==
+            TableCache::kInfiniteCapacity) {
      // unlimited table cache. Pre-load table handle now.
      // Need to do it out of the mutex.
      builder_guard->version_builder()->LoadTableHandlers(
@@ -2620,7 +2655,7 @@ Status VersionSet::Recover(
  {
    unique_ptr<SequentialFile> manifest_file;
    s = env_->NewSequentialFile(manifest_filename, &manifest_file,
-                                env_options_);
+                                env_->OptimizeForManifestRead(env_options_));
    if (!s.ok()) {
      return s;
    }
@@ -2654,6 +2689,9 @@ Status VersionSet::Recover(
  default_cf_edit.SetColumnFamily(0);
  ColumnFamilyData* default_cfd =
      CreateColumnFamily(default_cf_iter->second, &default_cf_edit);
+  // In recovery, nobody else can access it, so it's fine to set it to be
+  // initialized earlier.
+  default_cfd->set_initialized();
  builders.insert({0, new BaseReferencedVersionBuilder(default_cfd)});

  {
@@ -2700,6 +2738,7 @@ Status VersionSet::Recover(
              {edit.column_family_, edit.column_family_name_});
        } else {
          cfd = CreateColumnFamily(cf_options->second, &edit);
+          cfd->set_initialized();
          builders.insert(
              {edit.column_family_, new BaseReferencedVersionBuilder(cfd)});
        }
@@ -2827,11 +2866,13 @@ Status VersionSet::Recover(
      if (cfd->IsDropped()) {
        continue;
      }
+      assert(cfd->initialized());
      auto builders_iter = builders.find(cfd->GetID());
      assert(builders_iter != builders.end());
      auto* builder = builders_iter->second->version_builder();

-      if (db_options_->max_open_files == -1) {
+      if (GetColumnFamilySet()->get_table_cache()->GetCapacity() ==
+          TableCache::kInfiniteCapacity) {
        // unlimited table cache. Pre-load table handle now.
        // Need to do it out of the mutex.
        builder->LoadTableHandlers(
@@ -3046,7 +3087,8 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
  Status s;
  {
    unique_ptr<SequentialFile> file;
-    s = options.env->NewSequentialFile(dscname, &file, env_options_);
+    s = options.env->NewSequentialFile(
+        dscname, &file, env_->OptimizeForManifestRead(env_options_));
    if (!s.ok()) {
      return s;
    }
@@ -3109,6 +3151,7 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
          break;
        }
        cfd = CreateColumnFamily(ColumnFamilyOptions(options), &edit);
+        cfd->set_initialized();
        builders.insert(
            {edit.column_family_, new BaseReferencedVersionBuilder(cfd)});
      } else if (edit.is_column_family_drop_) {
@@ -3252,6 +3295,7 @@ Status VersionSet::WriteSnapshot(log::Writer* log) {
    if (cfd->IsDropped()) {
      continue;
    }
+    assert(cfd->initialized());
    {
      // Store column family info
      VersionEdit edit;
@@ -3418,6 +3462,9 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
  // pre-calculate space requirement
  int64_t total_files = 0;
  for (auto cfd : *column_family_set_) {
+    if (!cfd->initialized()) {
+      continue;
+    }
    Version* dummy_versions = cfd->dummy_versions();
    for (Version* v = dummy_versions->next_; v != dummy_versions;
         v = v->next_) {
@@ -3432,6 +3479,9 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
  live_list->reserve(live_list->size() + static_cast<size_t>(total_files));

  for (auto cfd : *column_family_set_) {
+    if (!cfd->initialized()) {
+      continue;
+    }
    auto* current = cfd->current();
    bool found_current = false;
    Version* dummy_versions = cfd->dummy_versions();
@@ -3456,9 +3506,11 @@ InternalIterator* VersionSet::MakeInputIterator(
  ReadOptions read_options;
  read_options.verify_checksums = true;
  read_options.fill_cache = false;
-  if (c->ShouldFormSubcompactions()) {
-    read_options.total_order_seek = true;
-  }
+  // Compaction iterators shouldn't be confined to a single prefix.
+  // Compactions use Seek() for
+  // (a) concurrent compactions,
+  // (b) CompactionFilter::Decision::kRemoveAndSkipUntil.
+  read_options.total_order_seek = true;

  // Level-0 files have to be merged together.  For other levels,
  // we will make a concatenating iterator per level.
@@ -3557,6 +3609,9 @@ Status VersionSet::GetMetadataForFile(uint64_t number, int* filelevel,
                                      FileMetaData** meta,
                                      ColumnFamilyData** cfd) {
  for (auto cfd_iter : *column_family_set_) {
+    if (!cfd_iter->initialized()) {
+      continue;
+    }
    Version* version = cfd_iter->current();
    const auto* vstorage = version->storage_info();
    for (int level = 0; level < vstorage->num_levels(); level++) {
@@ -3575,7 +3630,7 @@ Status VersionSet::GetMetadataForFile(uint64_t number, int* filelevel,

 void VersionSet::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
  for (auto cfd : *column_family_set_) {
-    if (cfd->IsDropped()) {
+    if (cfd->IsDropped() || !cfd->initialized()) {
      continue;
    }
    for (int level = 0; level < cfd->NumberLevels(); level++) {
@@ -461,6 +461,10 @@ class Version {
                            MergeIteratorBuilder* merger_iter_builder,
                            int level, RangeDelAggregator* range_del_agg);

+  void AddRangeDelIteratorsForLevel(
+      const ReadOptions& read_options, const EnvOptions& soptions, int level,
+      std::vector<InternalIterator*>* range_del_iters);
+
  // Lookup the value for key.  If found, store it in *val and
  // return OK.  Else return a non-OK status.
  // Uses *operands to store merge_operator operations to apply later.
@@ -166,13 +166,13 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamic) {
  Add(5, 2U, "3", "4", 550U);
  vstorage_.CalculateBaseBytes(ioptions_, mutable_cf_options_);
  ASSERT_EQ(0, logger_->log_count);
-  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 210U);
+  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 1000U);
  ASSERT_EQ(vstorage_.base_level(), 4);

  Add(4, 3U, "3", "4", 550U);
  vstorage_.CalculateBaseBytes(ioptions_, mutable_cf_options_);
  ASSERT_EQ(0, logger_->log_count);
-  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 210U);
+  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 1000U);
  ASSERT_EQ(vstorage_.base_level(), 4);

  Add(3, 4U, "3", "4", 250U);
@@ -180,7 +180,7 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamic) {
  vstorage_.CalculateBaseBytes(ioptions_, mutable_cf_options_);
  ASSERT_EQ(1, logger_->log_count);
  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 1005U);
-  ASSERT_EQ(vstorage_.MaxBytesForLevel(3), 201U);
+  ASSERT_EQ(vstorage_.MaxBytesForLevel(3), 1000U);
  ASSERT_EQ(vstorage_.base_level(), 3);

  Add(1, 6U, "3", "4", 5U);
@@ -191,7 +191,7 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamic) {
  ASSERT_GT(vstorage_.MaxBytesForLevel(4), 1005U);
  ASSERT_GT(vstorage_.MaxBytesForLevel(3), 1005U);
  ASSERT_EQ(vstorage_.MaxBytesForLevel(2), 1005U);
-  ASSERT_EQ(vstorage_.MaxBytesForLevel(1), 201U);
+  ASSERT_EQ(vstorage_.MaxBytesForLevel(1), 1000U);
  ASSERT_EQ(vstorage_.base_level(), 1);
 }

@@ -229,7 +229,7 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamicLargeLevel) {
  ASSERT_EQ(vstorage_.MaxBytesForLevel(5), 3000U * kOneGB);
  ASSERT_EQ(vstorage_.MaxBytesForLevel(4), 300U * kOneGB);
  ASSERT_EQ(vstorage_.MaxBytesForLevel(3), 30U * kOneGB);
-  ASSERT_EQ(vstorage_.MaxBytesForLevel(2), 3U * kOneGB);
+  ASSERT_EQ(vstorage_.MaxBytesForLevel(2), 10U * kOneGB);
  ASSERT_EQ(vstorage_.base_level(), 2);
  ASSERT_EQ(0, logger_->log_count);
 }
@@ -431,7 +431,8 @@ Status WalManager::ReadFirstLine(const std::string& fname,
  };

  std::unique_ptr<SequentialFile> file;
-  Status status = env_->NewSequentialFile(fname, &file, env_options_);
+  Status status = env_->NewSequentialFile(
+      fname, &file, env_->OptimizeForLogRead(env_options_));
  unique_ptr<SequentialFileReader> file_reader(
      new SequentialFileReader(std::move(file)));

@@ -801,6 +801,17 @@ Status WriteBatch::RollbackToSavePoint() {
  return Status::OK();
 }

+Status WriteBatch::PopSavePoint() {
+  if (save_points_ == nullptr || save_points_->stack.size() == 0) {
+    return Status::NotFound();
+  }
+
+  // Pop the most recent savepoint off the stack
+  save_points_->stack.pop();
+
+  return Status::OK();
+}
+
 class MemTableInserter : public WriteBatch::Handler {

  SequenceNumber sequence_;
@@ -839,24 +850,24 @@ class MemTableInserter : public WriteBatch::Handler {

 public:
  // cf_mems should not be shared with concurrent inserters
-  MemTableInserter(SequenceNumber sequence, ColumnFamilyMemTables* cf_mems,
-                   FlushScheduler* flush_scheduler,
-                   bool ignore_missing_column_families,
-                   uint64_t recovering_log_number, DB* db,
-                   bool concurrent_memtable_writes,
-                   bool* has_valid_writes = nullptr)
-      : sequence_(sequence),
-        cf_mems_(cf_mems),
-        flush_scheduler_(flush_scheduler),
-        ignore_missing_column_families_(ignore_missing_column_families),
-        recovering_log_number_(recovering_log_number),
-        log_number_ref_(0),
-        db_(reinterpret_cast<DBImpl*>(db)),
-        concurrent_memtable_writes_(concurrent_memtable_writes),
-        post_info_created_(false),
-        has_valid_writes_(has_valid_writes),
-        rebuilding_trx_(nullptr) {
-    assert(cf_mems_);
+ MemTableInserter(SequenceNumber _sequence, ColumnFamilyMemTables* cf_mems,
+                  FlushScheduler* flush_scheduler,
+                  bool ignore_missing_column_families,
+                  uint64_t recovering_log_number, DB* db,
+                  bool concurrent_memtable_writes,
+                  bool* has_valid_writes = nullptr)
+     : sequence_(_sequence),
+       cf_mems_(cf_mems),
+       flush_scheduler_(flush_scheduler),
+       ignore_missing_column_families_(ignore_missing_column_families),
+       recovering_log_number_(recovering_log_number),
+       log_number_ref_(0),
+       db_(reinterpret_cast<DBImpl*>(db)),
+       concurrent_memtable_writes_(concurrent_memtable_writes),
+       post_info_created_(false),
+       has_valid_writes_(has_valid_writes),
+       rebuilding_trx_(nullptr) {
+   assert(cf_mems_);
  }

  ~MemTableInserter() {
@@ -871,7 +882,7 @@ public:

  void set_log_number_ref(uint64_t log) { log_number_ref_ = log; }

-  SequenceNumber get_final_sequence() const { return sequence_; }
+  SequenceNumber sequence() const { return sequence_; }

  void PostProcess() {
    assert(concurrent_memtable_writes_);
@@ -1277,19 +1288,21 @@ public:
 // 2) During Write(), in a single-threaded write thread
 // 3) During Write(), in a concurrent context where memtables has been cloned
 // The reason is that it calls memtables->Seek(), which has a stateful cache
-Status WriteBatchInternal::InsertInto(
-    const autovector<WriteThread::Writer*>& writers, SequenceNumber sequence,
-    ColumnFamilyMemTables* memtables, FlushScheduler* flush_scheduler,
-    bool ignore_missing_column_families, uint64_t recovery_log_number, DB* db,
-    bool concurrent_memtable_writes) {
+Status WriteBatchInternal::InsertInto(WriteThread::WriteGroup& write_group,
+                                      SequenceNumber sequence,
+                                      ColumnFamilyMemTables* memtables,
+                                      FlushScheduler* flush_scheduler,
+                                      bool ignore_missing_column_families,
+                                      uint64_t recovery_log_number, DB* db,
+                                      bool concurrent_memtable_writes) {
  MemTableInserter inserter(sequence, memtables, flush_scheduler,
                            ignore_missing_column_families, recovery_log_number,
                            db, concurrent_memtable_writes);
-  for (size_t i = 0; i < writers.size(); i++) {
-    auto w = writers[i];
+  for (auto w : write_group) {
    if (!w->ShouldWriteToMemtable()) {
      continue;
    }
+    SetSequence(w->batch, inserter.sequence());
    inserter.set_log_number_ref(w->log_ref);
    w->status = w->batch->Iterate(&inserter);
    if (!w->status.ok()) {
@@ -1300,16 +1313,17 @@ Status WriteBatchInternal::InsertInto(
 }

 Status WriteBatchInternal::InsertInto(WriteThread::Writer* writer,
+                                      SequenceNumber sequence,
                                      ColumnFamilyMemTables* memtables,
                                      FlushScheduler* flush_scheduler,
                                      bool ignore_missing_column_families,
                                      uint64_t log_number, DB* db,
                                      bool concurrent_memtable_writes) {
-  MemTableInserter inserter(WriteBatchInternal::Sequence(writer->batch),
-                            memtables, flush_scheduler,
+  assert(writer->ShouldWriteToMemtable());
+  MemTableInserter inserter(sequence, memtables, flush_scheduler,
                            ignore_missing_column_families, log_number, db,
                            concurrent_memtable_writes);
-  assert(writer->ShouldWriteToMemtable());
+  SetSequence(writer->batch, sequence);
  inserter.set_log_number_ref(writer->log_ref);
  Status s = writer->batch->Iterate(&inserter);
  if (concurrent_memtable_writes) {
@@ -1323,13 +1337,12 @@ Status WriteBatchInternal::InsertInto(
    FlushScheduler* flush_scheduler, bool ignore_missing_column_families,
    uint64_t log_number, DB* db, bool concurrent_memtable_writes,
    SequenceNumber* last_seq_used, bool* has_valid_writes) {
-  MemTableInserter inserter(WriteBatchInternal::Sequence(batch), memtables,
-                            flush_scheduler, ignore_missing_column_families,
-                            log_number, db, concurrent_memtable_writes,
-                            has_valid_writes);
+  MemTableInserter inserter(Sequence(batch), memtables, flush_scheduler,
+                            ignore_missing_column_families, log_number, db,
+                            concurrent_memtable_writes, has_valid_writes);
  Status s = batch->Iterate(&inserter);
  if (last_seq_used != nullptr) {
-    *last_seq_used = inserter.get_final_sequence();
+    *last_seq_used = inserter.sequence();
  }
  if (concurrent_memtable_writes) {
    inserter.PostProcess();
@@ -151,7 +151,7 @@ class WriteBatchInternal {
  //
  // Under concurrent use, the caller is responsible for making sure that
  // the memtables object itself is thread-local.
-  static Status InsertInto(const autovector<WriteThread::Writer*>& batches,
+  static Status InsertInto(WriteThread::WriteGroup& write_group,
                           SequenceNumber sequence,
                           ColumnFamilyMemTables* memtables,
                           FlushScheduler* flush_scheduler,
@@ -170,7 +170,7 @@ class WriteBatchInternal {
                           SequenceNumber* last_seq_used = nullptr,
                           bool* has_valid_writes = nullptr);

-  static Status InsertInto(WriteThread::Writer* writer,
+  static Status InsertInto(WriteThread::Writer* writer, SequenceNumber sequence,
                           ColumnFamilyMemTables* memtables,
                           FlushScheduler* flush_scheduler,
                           bool ignore_missing_column_families = false,
@@ -32,7 +32,7 @@ static std::string PrintContents(WriteBatch* b) {
  ImmutableCFOptions ioptions(options);
  WriteBufferManager wb(options.db_write_buffer_size);
  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
-                               kMaxSequenceNumber);
+                               kMaxSequenceNumber, 0 /* column_family_id */);
  mem->Ref();
  std::string state;
  ColumnFamilyMemTablesDefault cf_mems_default(mem);
@@ -859,6 +859,19 @@ TEST_F(WriteBatchTest, SavePointTest) {
  s = batch2.RollbackToSavePoint();
  ASSERT_TRUE(s.IsNotFound());
  ASSERT_EQ("", PrintContents(&batch2));
+
+  WriteBatch batch3;
+
+  s = batch3.PopSavePoint();
+  ASSERT_TRUE(s.IsNotFound());
+  ASSERT_EQ("", PrintContents(&batch3));
+
+  batch3.SetSavePoint();
+  batch3.Delete("A");
+
+  s = batch3.PopSavePoint();
+  ASSERT_OK(s);
+  ASSERT_EQ("Delete(A)@0", PrintContents(&batch3));
 }

 TEST_F(WriteBatchTest, MemoryLimitTest) {