URI-based Env selection for db_bench

Summary: Added an option, --env_uri. When provided, it is used as an argument to NewEnvFromUri(), which instantiates an Env based on it. Test Plan: built a simple binary that registers ChrootEnv for prefix "/", then ran: $ ./tmp --env_uri /tmp/ --db /abcde /tmp/ is the chroot directory and /abcde is the db_name. Then I verified db_bench uses /tmp/abcde Reviewers: sdong, kradhakrishnan, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D59325
duplicate line
2026-11-06 11:02:36 -05:00 · 2016-06-09 17:53:03 -07:00 · 2016-06-09 17:30:52 -07:00 · 2016-06-09 17:26:17 -07:00 · 2016-06-09 16:56:17 -07:00 · 2016-06-09 11:10:35 -07:00
774 changed files with 50231 additions and 26124 deletions
@@ -42,6 +42,7 @@ unity.a
 tags
 rocksdb_dump
 rocksdb_undump
+db_test2

 java/out
 java/target
@@ -1,24 +1,29 @@
 sudo: false
 language: cpp
+os:
+  - linux
+  - osx
+compiler:
+  - clang

-matrix:
-  include:
-    - os: linux
-      compiler: clang
-      env: COMPILER=clang++-3.6
-      addons:
-         apt:
-            sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
-            packages: ['clang-3.6', 'clang-format-3.6', 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl']
-    - os: osx
-      compiler: clang
+addons:
+   apt:
+      sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
+      packages: ['clang-3.6' , 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl']
+env:
+  # Run all tests before db_block_cache_test (db_test, db_test2)
+  - JOB_NAME=unittests ROCKSDBTESTS_END=db_block_cache_test
+  # Run all tests starting from db_block_cache_test (db_block_cache_test, db_iter_test, ...)
+  - JOB_NAME=unittests ROCKSDBTESTS_START=db_block_cache_test
+  # Run java tests
+  - JOB_NAME=java_test
+  # Build ROCKSDB_LITE
+  - JOB_NAME=lite_build

 install:
  # Build gflags
  # TODO(noetzli): Remove when gflags available through Travis
  - pushd /tmp/ && curl -L https://github.com/gflags/gflags/archive/v2.1.2.tar.gz -o gflags.tar.gz && tar xfz gflags.tar.gz && cd gflags-2.1.2 && cmake . && make && popd
-  # Download clang-format-diff.py to check source code formatting
-  - pushd /tmp/ && curl -L http://llvm.org/svn/llvm-project/cfe/trunk/tools/clang-format/clang-format-diff.py -o clang-format-diff.py && chmod +x clang-format-diff.py && popd

 before_script:
  # Add gflags to include/library paths
@@ -26,16 +31,15 @@ before_script:
  - export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/tmp/gflags-2.1.2/lib"
  - export LIBRARY_PATH="$LIBRARY_PATH:/tmp/gflags-2.1.2/lib"
  - export CPLUS_INCLUDE_PATH="$CPLUS_INCLUDE_PATH:/tmp/gflags-2.1.2/include"
-  - if [ -n "${COMPILER}" ]; then CXX=${COMPILER}; fi
+  - if [[ "${TRAVIS_OS_NAME}" == 'linux' ]]; then CXX=clang++-3.6; fi
  - if [[ "${TRAVIS_OS_NAME}" == 'osx' ]]; then brew install gflags snappy; fi
+  # Limit the maximum number of open file descriptors to 2000
  - ulimit -n 2000 || true

-# Lousy hack to disable use and testing of fallocate, which doesn't behave quite
-# as EnvPosixTest::AllocateTest expects within the Travis OpenVZ environment.
 script:
-  - if [[ "${TRAVIS_OS_NAME}" == 'linux' ]]; then OPT=-DTRAVIS CLANG_FORMAT_DIFF=/tmp/clang-format-diff.py make format || true; fi
-  - OPT=-DTRAVIS V=1 make -j4 check && OPT=-DTRAVIS V=1 make clean jclean rocksdbjava jtest && make clean && OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 check
-
+  - if [[ "${JOB_NAME}" == 'unittests' ]]; then OPT=-DTRAVIS V=1 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
 notifications:
    email:
      - leveldb@fb.com
@@ -13,8 +13,8 @@
 #        cd build
 # 3. Run cmake to generate project files for Windows, add more options to enable required third-party libraries.
 #    See thirdparty.inc for more information.
-#        sample command: cmake -G "Visual Studio 12 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 ..
-#        OR for VS Studio 15 cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 ..
+#        sample command: cmake -G "Visual Studio 12 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
+#        OR for VS Studio 15 cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
 # 4. Then build the project in debug mode (you may want to add /m[:<N>] flag to run msbuild in <N> parallel threads
 #                                          or simply /m ot use all avail cores)
 #        msbuild rocksdb.sln
@@ -30,10 +30,9 @@ cmake_minimum_required(VERSION 2.6)
 project(rocksdb)

 include(${CMAKE_CURRENT_SOURCE_DIR}/thirdparty.inc)
-
-execute_process(COMMAND $ENV{COMSPEC} " /C date /T" OUTPUT_VARIABLE DATE)
-execute_process(COMMAND $ENV{COMSPEC} " /C time /T" OUTPUT_VARIABLE TIME)
-string(REGEX REPLACE "(..)/(..)/..(..).*" "\\1/\\2/\\3" DATE ${DATE})
+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)
+string(REGEX REPLACE "(..)_(..)_..(..).*" "\\1/\\2/\\3" DATE ${DATE})
 string(REGEX REPLACE "(..):(.....).*" " \\1:\\2" TIME ${TIME})
 string(CONCAT GIT_DATE_TIME ${DATE} ${TIME})

@@ -60,7 +59,7 @@ add_custom_target(GenerateBuildVersion DEPENDS ${BUILD_VERSION_CC})

 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi /nologo  /EHsc /GS /Gd /GR /GF /fp:precise /Zc:wchar_t /Zc:forScope /errorReport:queue")

-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W3 /WX /wd4127 /wd4244 /wd4267 /wd4800 /wd4996")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W3 /WX /wd4127 /wd4800 /wd4996 /wd4351")

 # Used to run CI build and tests so we can run faster
 set(OPTIMIZE_DEBUG_DEFAULT 0)        # Debug build is unoptimized by default use -DOPTDBG=1 to optimize
@@ -83,7 +82,7 @@ set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /MD")
 set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /DEBUG")
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /DEBUG")

-add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64)
+add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64 -DNOMINMAX)

 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
@@ -91,7 +90,7 @@ include_directories(${PROJECT_SOURCE_DIR}/third-party/gtest-1.7.0/fused-src)

 set(ROCKSDB_LIBS rocksdblib${ARTIFACT_SUFFIX})
 set(THIRDPARTY_LIBS ${THIRDPARTY_LIBS} gtest)
-set(SYSTEM_LIBS Shlwapi.lib Rpcrt4.lib)
+set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)

 set(LIBS ${ROCKSDB_LIBS} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})

@@ -99,6 +98,7 @@ add_subdirectory(third-party/gtest-1.7.0/fused-src/gtest)

 # Main library source code
 set(SOURCES
+        db/auto_roll_logger.cc
        db/builder.cc
        db/c.cc
        db/column_family.cc
@@ -114,6 +114,7 @@ set(SOURCES
        db/db_impl_debug.cc
        db/db_impl_experimental.cc
        db/db_impl_readonly.cc
+        db/db_info_dumper.cc
        db/db_iter.cc
        db/event_helpers.cc
        db/experimental.cc
@@ -145,13 +146,19 @@ set(SOURCES
        db/write_batch_base.cc
        db/write_controller.cc
        db/write_thread.cc
+        db/xfunc_test_points.cc
        memtable/hash_cuckoo_rep.cc
        memtable/hash_linklist_rep.cc
        memtable/hash_skiplist_rep.cc
+        memtable/skiplistrep.cc
+        memtable/vectorrep.cc
        port/stack_trace.cc
+        port/win/io_win.cc
        port/win/env_win.cc
+        port/win/env_default.cc
        port/win/port_win.cc
        port/win/win_logger.cc
+        port/win/xpress_win.cc
        table/adaptive_table_factory.cc
        table/block.cc
        table/block_based_filter_block.cc
@@ -159,7 +166,6 @@ set(SOURCES
        table/block_based_table_factory.cc
        table/block_based_table_reader.cc
        table/block_builder.cc
-        table/block_hash_index.cc
        table/block_prefix_index.cc
        table/bloom_block.cc
        table/cuckoo_table_builder.cc
@@ -173,36 +179,39 @@ set(SOURCES
        table/merger.cc
        table/sst_file_writer.cc
        table/meta_blocks.cc
-        table/mock_table.cc
        table/plain_table_builder.cc
        table/plain_table_factory.cc
        table/plain_table_index.cc
        table/plain_table_key_coding.cc
        table/plain_table_reader.cc
+        table/persistent_cache_helper.cc
        table/table_properties.cc
        table/two_level_iterator.cc
        tools/sst_dump_tool.cc
+        tools/db_bench_tool.cc
        tools/dump/db_dump_tool.cc
        util/arena.cc
-        util/auto_roll_logger.cc
        util/bloom.cc
        util/build_version.cc
        util/cache.cc
        util/coding.cc
        util/compaction_job_stats_impl.cc
        util/comparator.cc
+        util/concurrent_arena.cc
        util/crc32c.cc
-        util/db_info_dumper.cc
-        util/delete_scheduler_impl.cc
+        util/delete_scheduler.cc
        util/dynamic_bloom.cc
        util/env.cc
+        util/env_chroot.cc
        util/env_hdfs.cc
        util/event_logger.cc
        util/file_util.cc
        util/file_reader_writer.cc
+        util/sst_file_manager_impl.cc
        util/filter_policy.cc
        util/hash.cc
        util/histogram.cc
+        util/histogram_windowing.cc
        util/instrumented_mutex.cc
        util/iostats_context.cc
        tools/ldb_cmd.cc
@@ -210,11 +219,9 @@ set(SOURCES
        util/logging.cc
        util/log_buffer.cc
        util/memenv.cc
-        util/mock_env.cc
        util/murmurhash.cc
        util/mutable_cf_options.cc
        util/options.cc
-        util/options_builder.cc
        util/options_helper.cc
        util/options_parser.cc
        util/options_sanity_check.cc
@@ -222,7 +229,6 @@ set(SOURCES
        util/perf_level.cc
        util/random.cc
        util/rate_limiter.cc
-        util/skiplistrep.cc
        util/slice.cc
        util/statistics.cc
        util/status.cc
@@ -232,18 +238,22 @@ set(SOURCES
        util/testharness.cc
        util/testutil.cc
        util/thread_local.cc
+        util/threadpool.cc
        util/thread_status_impl.cc
        util/thread_status_updater.cc
        util/thread_status_util.cc
        util/thread_status_util_debug.cc
-        util/vectorrep.cc
+        util/transaction_test_util.cc
        util/xfunc.cc
        util/xxhash.cc
        utilities/backupable/backupable_db.cc
        utilities/checkpoint/checkpoint.cc
+        utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc
        utilities/document/document_db.cc
        utilities/document/json_document.cc
        utilities/document/json_document_builder.cc
+        utilities/env_mirror.cc
+        utilities/env_registry.cc
        utilities/flashcache/flashcache.cc
        utilities/geodb/geodb_impl.cc
        utilities/leveldb_options/leveldb_options.cc
@@ -251,8 +261,11 @@ set(SOURCES
        utilities/merge_operators/string_append/stringappend.cc
        utilities/merge_operators/string_append/stringappend2.cc
        utilities/merge_operators/put.cc
+        utilities/merge_operators/max.cc
        utilities/merge_operators/uint64add.cc
        utilities/options/options_util.cc
+        utilities/persistent_cache/persistent_cache_tier.cc
+        utilities/persistent_cache/volatile_tier_impl.cc
        utilities/redis/redis_lists.cc
        utilities/spatialdb/spatial_db.cc
        utilities/table_properties_collectors/compact_on_deletion_collector.cc
@@ -273,6 +286,9 @@ set(SOURCES
 # and linked to tests. Add test only code that is not #ifdefed for Release here.
 set(TESTUTIL_SOURCE
    db/db_test_util.cc
+    table/mock_table.cc
+    util/mock_env.cc
+    util/fault_injection_test_env.cc
    util/thread_status_updater_debug.cc
 )

@@ -285,41 +301,62 @@ set_target_properties(rocksdb${ARTIFACT_SUFFIX} PROPERTIES COMPILE_FLAGS "-DROCK
 add_dependencies(rocksdb${ARTIFACT_SUFFIX} GenerateBuildVersion)
 target_link_libraries(rocksdb${ARTIFACT_SUFFIX} ${LIBS})

+if (DEFINED JNI)
+  if (${JNI} EQUAL 1)
+    message(STATUS "JNI library is enabled")
+    add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/java)
+  else()
+    message(STATUS "JNI library is disabled")
+  endif()
+else()
+  message(STATUS "JNI library is disabled")
+endif()
+
 set(APPS
-        db/db_bench.cc
+        tools/db_bench.cc
        db/memtablerep_bench.cc
        table/table_reader_bench.cc
        tools/db_stress.cc
        tools/write_stress.cc
        tools/db_repl_stress.cc
+        tools/ldb.cc
        tools/sst_dump.cc
        tools/dump/rocksdb_dump.cc
        tools/dump/rocksdb_undump.cc
        util/cache_bench.cc
+        utilities/persistent_cache/hash_table_bench.cc
 )

 set(C_TESTS db/c_test.c)

 set(TESTS
+        db/auto_roll_logger_test.cc
        db/column_family_test.cc
        db/compact_files_test.cc
+        db/compaction_iterator_test.cc
        db/compaction_job_test.cc
        db/compaction_job_stats_test.cc
        db/compaction_picker_test.cc
        db/comparator_db_test.cc
        db/corruption_test.cc
        db/cuckoo_table_db_test.cc
-        db/db_iter_test.cc
-        db/db_test.cc
        db/db_compaction_filter_test.cc
        db/db_compaction_test.cc
        db/db_dynamic_level_test.cc
        db/db_inplace_update_test.cc
+        db/db_iter_test.cc
        db/db_log_iter_test.cc
+        db/db_properties_test.cc
+        db/db_table_properties_test.cc
+        db/db_tailing_iter_test.cc
+        db/db_test.cc
+        db/db_test2.cc
+        db/db_block_cache_test.cc
+        db/db_bloom_filter_test.cc
+        db/db_iterator_test.cc
+        db/db_sst_test.cc
        db/db_universal_compaction_test.cc
        db/db_wal_test.cc
-        db/db_tailing_iter_test.cc
-        db/db_table_properties_test.cc
        db/dbformat_test.cc
        db/deletefile_test.cc
        db/fault_injection_test.cc
@@ -337,6 +374,7 @@ set(TESTS
        db/perf_context_test.cc
        db/plain_table_db_test.cc
        db/prefix_test.cc
+        db/repair_test.cc
        db/skiplist_test.cc
        db/table_properties_collector_test.cc
        db/version_builder_test.cc
@@ -346,8 +384,8 @@ set(TESTS
        db/write_batch_test.cc
        db/write_callback_test.cc
        db/write_controller_test.cc
+        db/db_io_failure_test.cc
        table/block_based_filter_block_test.cc
-        table/block_hash_index_test.cc
        table/block_test.cc
        table/cuckoo_table_builder_test.cc
        table/cuckoo_table_reader_test.cc
@@ -360,20 +398,22 @@ set(TESTS
        tools/sst_dump_test.cc
        util/arena_test.cc
        util/autovector_test.cc
-        util/auto_roll_logger_test.cc
        util/bloom_test.cc
        util/cache_test.cc
        util/coding_test.cc
        util/crc32c_test.cc
+        util/delete_scheduler_test.cc
        util/dynamic_bloom_test.cc
+        util/env_basic_test.cc
        util/env_test.cc
        util/event_logger_test.cc
        util/filelock_test.cc
        util/file_reader_writer_test.cc
        util/heap_test.cc
        util/histogram_test.cc
-        util/memenv_test.cc
+        util/iostats_context_test.cc
        util/mock_env_test.cc
+        util/options_settable_test.cc
        util/options_test.cc
        util/rate_limiter_test.cc
        util/slice_transform_test.cc
@@ -383,10 +423,13 @@ set(TESTS
        utilities/checkpoint/checkpoint_test.cc
        utilities/document/document_db_test.cc
        utilities/document/json_document_test.cc
+        utilities/env_registry_test.cc
        utilities/geodb/geodb_test.cc
        utilities/memory/memory_test.cc
        utilities/merge_operators/string_append/stringappend_test.cc
        utilities/options/options_util_test.cc
+        utilities/persistent_cache/hash_table_test.cc
+        utilities/persistent_cache/persistent_cache_test.cc
        utilities/redis/redis_lists_test.cc
        utilities/spatialdb/spatial_db_test.cc
        utilities/table_properties_collectors/compact_on_deletion_collector_test.cc
@@ -0,0 +1,14 @@
+# RocksDB default options change log
+## 4.8.0 (5/2/2016)
+* options.max_open_files changes from 5000 to -1. It improves performance, but users need to set file descriptor limit to be large enough and watch memory usage for index and bloom filters.
+* options.base_background_compactions changes from max_background_compactions to 1. When users set higher max_background_compactions but the write throughput is not high, the writes are less spiky to disks.
+* options.wal_recovery_mode changes from kTolerateCorruptedTailRecords to kPointInTimeRecovery. Avoid some false positive when file system or hardware reorder the writes for file data and metadata.
+
+## 4.7.0 (4/8/2016)
+* options.write_buffer_size changes from 4MB to 64MB.
+* options.target_file_size_base changes from 2MB to 64MB.
+* options.max_bytes_for_level_base changes from 10MB to 256MB.
+* options.soft_pending_compaction_bytes_limit changes from 0 (disabled) to 64GB.
+* options.hard_pending_compaction_bytes_limit changes from 0 (disabled) to 256GB.
+* table_cache_numshardbits changes from 4 to 6.
+* max_file_opening_threads changes from 1 to 16.
@@ -1,4 +1,62 @@
 # Rocksdb Change Log
+## Unreleased
+
+## 4.9.0 (6/9/2016)
+### Public API changes
+* Add bottommost_compression option, This option can be used to set a specific compression algorithm for the bottommost level (Last level containing files in the DB).
+* Introduce CompactionJobInfo::compression, This field state the compression algorithm used to generate the output files of the compaction.
+* Deprecate BlockBaseTableOptions.hash_index_allow_collision=false
+* Deprecate options builder (GetOptions()).
+
+### New Features
+* Introduce NewSimCache() in rocksdb/utilities/sim_cache.h. This function creates a block cache that is able to give simulation results (mainly hit rate) of simulating block behavior with a configurable cache size.
+
+## 4.8.0 (5/2/2016)
+### Public API Change
+* Allow preset compression dictionary for improved compression of block-based tables. This is supported for zlib, zstd, and lz4. The compression dictionary's size is configurable via CompressionOptions::max_dict_bytes.
+* Delete deprecated classes for creating backups (BackupableDB) and restoring from backups (RestoreBackupableDB). Now, BackupEngine should be used for creating backups, and BackupEngineReadOnly should be used for restorations. For more details, see https://github.com/facebook/rocksdb/wiki/How-to-backup-RocksDB%3F
+* Expose estimate of per-level compression ratio via DB property: "rocksdb.compression-ratio-at-levelN".
+* Added EventListener::OnTableFileCreationStarted. EventListener::OnTableFileCreated will be called on failure case. User can check creation status via TableFileCreationInfo::status.
+
+### New Features
+* Add ReadOptions::readahead_size. If non-zero, NewIterator will create a new table reader which performs reads of the given size.
+
+## 4.7.0 (4/8/2016)
+### Public API Change
+* rename options compaction_measure_io_stats to report_bg_io_stats and include flush too.
+* Change some default options. Now default options will optimize for server-workloads. Also enable slowdown and full stop triggers for pending compaction bytes. These changes may cause sub-optimal performance or significant increase of resource usage. To avoid these risks, users can open existing RocksDB with options extracted from RocksDB option files. See https://github.com/facebook/rocksdb/wiki/RocksDB-Options-File for how to use RocksDB option files. Or you can call Options.OldDefaults() to recover old defaults. DEFAULT_OPTIONS_HISTORY.md will track change history of default options.
+
+## 4.6.0 (3/10/2016)
+### Public API Changes
+* Change default of BlockBasedTableOptions.format_version to 2. It means default DB created by 4.6 or up cannot be opened by RocksDB version 3.9 or earlier.
+* Added strict_capacity_limit option to NewLRUCache. If the flag is set to true, insert to cache will fail if no enough capacity can be free. Signiture of Cache::Insert() is updated accordingly.
+* Tickers [NUMBER_DB_NEXT, NUMBER_DB_PREV, NUMBER_DB_NEXT_FOUND, NUMBER_DB_PREV_FOUND, ITER_BYTES_READ] are not updated immediately. The are updated when the Iterator is deleted.
+* Add monotonically increasing counter (DB property "rocksdb.current-super-version-number") that increments upon any change to the LSM tree.
+
+### New Features
+* Add CompactionPri::kMinOverlappingRatio, a compaction picking mode friendly to write amplification.
+* Deprecate Iterator::IsKeyPinned() and replace it with Iterator::GetProperty() with prop_name="rocksdb.iterator.is.key.pinned"
+
+## 4.5.0 (2/5/2016)
+### Public API Changes
+* Add a new perf context level between kEnableCount and kEnableTime. Level 2 now does not include timers for mutexes.
+* Statistics of mutex operation durations will not be measured by default. If you want to have them enabled, you need to set Statistics::stats_level_ to kAll.
+* DBOptions::delete_scheduler and NewDeleteScheduler() are removed, please use DBOptions::sst_file_manager and NewSstFileManager() instead
+
+### New Features
+* ldb tool now supports operations to non-default column families.
+* Add kPersistedTier to ReadTier.  This option allows Get and MultiGet to read only the persited data and skip mem-tables if writes were done with disableWAL = true.
+* Add DBOptions::sst_file_manager. Use NewSstFileManager() in include/rocksdb/sst_file_manager.h to create a SstFileManager that can be used to track the total size of SST files and control the SST files deletion rate.
+
+## 4.4.0 (1/14/2016)
+### Public API Changes
+* Change names in CompactionPri and add a new one.
+* Deprecate options.soft_rate_limit and add options.soft_pending_compaction_bytes_limit.
+* If options.max_write_buffer_number > 3, writes will be slowed down when writing to the last write buffer to delay a full stop.
+* Introduce CompactionJobInfo::compaction_reason, this field include the reason to trigger the compaction.
+* After slow down is triggered, if estimated pending compaction bytes keep increasing, slowdown more.
+* Increase default options.delayed_write_rate to 2MB/s.
+* Added a new parameter --path to ldb tool. --path accepts the name of either MANIFEST, SST or a WAL file. Either --db or --path can be used when calling ldb.

 ## 4.3.0 (12/8/2015)
 ### New Features
@@ -12,8 +70,8 @@

 ## 4.2.0 (11/9/2015)
 ### New Features
-* Introduce CreateLoggerFromOptions(), this function create a Logger for provided DBOptions. 
-* Add GetAggregatedIntProperty(), which returns the sum of the GetIntProperty of all the column families. 
+* Introduce CreateLoggerFromOptions(), this function create a Logger for provided DBOptions.
+* Add GetAggregatedIntProperty(), which returns the sum of the GetIntProperty of all the column families.
 * Add MemoryUtil in rocksdb/utilities/memory.h.  It currently offers a way to get the memory usage by type from a list rocksdb instances.

 ### Public API Changes
@@ -210,7 +268,7 @@
 * Support Multiple DB paths in universal style compactions
 * Add feature of storing plain table index and bloom filter in SST file.
 * CompactRange() will never output compacted files to level 0. This used to be the case when all the compaction input files were at level 0.
-* Added iterate_upper_bound to define the extent upto which the forward iterator will return entries. This will prevent iterating over delete markers and overwritten entries for edge cases where you want to break out the iterator anyways. This may improve perfomance in case there are a large number of delete markers or overwritten entries.
+* Added iterate_upper_bound to define the extent upto which the forward iterator will return entries. This will prevent iterating over delete markers and overwritten entries for edge cases where you want to break out the iterator anyways. This may improve performance in case there are a large number of delete markers or overwritten entries.

 ### Public API changes
 * DBOptions.db_paths now is a vector of a DBPath structure which indicates both of path and target size
@@ -21,7 +21,7 @@ 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). If you want to build a portable binary, add 'PORTABLE=1' before
+(-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`

 ## Dependencies
@@ -0,0 +1,11 @@
+This is the list of all known third-party language bindings for RocksDB. If something is missing, please open a pull request to add it.
+
+* Java - https://github.com/facebook/rocksdb/tree/master/java
+* Python - http://pyrocksdb.readthedocs.org/en/latest/
+* Perl - https://metacpan.org/pod/RocksDB
+* Node.js - https://npmjs.org/package/rocksdb
+* Go - https://github.com/tecbot/gorocksdb
+* Ruby - http://rubygems.org/gems/rocksdb-ruby
+* Haskell - https://hackage.haskell.org/package/rocksdb-haskell
+* PHP - https://github.com/Photonios/rocksdb-php
+* C# - https://github.com/warrenfalk/rocksdb-sharp
@@ -2,7 +2,7 @@ BSD License

 For rocksdb software

-Copyright (c) 2014, Facebook, Inc.
+Copyright (c) 2011-present, Facebook, Inc.
 All rights reserved.
 ---------------------------------------------------------------------

@@ -15,7 +15,9 @@ ARFLAGS = rs

 # Transform parallel LOG output into something more readable.
 perl_command = perl -n \
-  -e '@a=split("\t",$$_,-1); $$t=$$a[8]; $$t =~ s,^\./,,;'		\
+  -e '@a=split("\t",$$_,-1); $$t=$$a[8];'				\
+  -e '$$t =~ /.*if\s\[\[\s"(.*?\.[\w\/]+)/ and $$t=$$1;'		\
+  -e '$$t =~ s,^\./,,;'							\
  -e '$$t =~ s, >.*,,; chomp $$t;'					\
  -e '$$t =~ /.*--gtest_filter=(.*?\.[\w\/]+)/ and $$t=$$1;'		\
  -e 'printf "%7.3f %s %s\n", $$a[3], $$a[6] == 0 ? "PASS" : "FAIL", $$t'
@@ -84,7 +86,8 @@ endif
 # compile with -O2 if debug level is not 2
 ifneq ($(DEBUG_LEVEL), 2)
 OPT += -O2 -fno-omit-frame-pointer
-ifneq ($(MACHINE),ppc64) # ppc64 doesn't support -momit-leaf-frame-pointer
+# Skip for archs that don't support -momit-leaf-frame-pointer
+ifeq (,$(shell $(CXX) -fsyntax-only -momit-leaf-frame-pointer -xc /dev/null 2>&1))
 OPT += -momit-leaf-frame-pointer
 endif
 endif
@@ -143,6 +146,9 @@ else
 OPT += -DNDEBUG
 endif

+ifeq ($(PLATFORM), OS_SOLARIS)
+	PLATFORM_CXXFLAGS += -D _GLIBCXX_USE_C99
+endif
 ifneq ($(filter -DROCKSDB_LITE,$(OPT)),)
 	# found
 	CFLAGS += -fno-exceptions
@@ -165,12 +171,22 @@ ifdef COMPILE_WITH_TSAN
 	PLATFORM_CXXFLAGS += -fsanitize=thread -fPIC -DROCKSDB_TSAN_RUN
        # Turn off -pg when enabling TSAN testing, because that induces
        # a link failure.  TODO: find the root cause
-	pg =
-else
-	pg = -pg
+	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
 endif

 ifndef DISABLE_JEMALLOC
+	ifdef JEMALLOC
+		PLATFORM_CXXFLAGS += "-DROCKSDB_JEMALLOC"
+		PLATFORM_CCFLAGS +=  "-DROCKSDB_JEMALLOC"
+	endif
 	EXEC_LDFLAGS := $(JEMALLOC_LIB) $(EXEC_LDFLAGS)
 	PLATFORM_CXXFLAGS += $(JEMALLOC_INCLUDE)
 	PLATFORM_CCFLAGS += $(JEMALLOC_INCLUDE)
@@ -187,6 +203,10 @@ default: all
 WARNING_FLAGS = -W -Wextra -Wall -Wsign-compare -Wshadow \
  -Wno-unused-parameter

+ifndef DISABLE_WARNING_AS_ERROR
+	WARNING_FLAGS += -Werror
+endif
+
 CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers

@@ -233,19 +253,27 @@ VALGRIND_VER := $(join $(VALGRIND_VER),valgrind)

 VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full

+BENCHTOOLOBJECTS = $(BENCH_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
+
 TESTS = \
 	db_test \
+	db_test2 \
+	db_block_cache_test \
+	db_bloom_filter_test \
 	db_iter_test \
 	db_log_iter_test \
 	db_compaction_filter_test \
 	db_compaction_test \
 	db_dynamic_level_test \
 	db_inplace_update_test \
+	db_iterator_test \
+	db_sst_test \
 	db_tailing_iter_test \
 	db_universal_compaction_test \
 	db_wal_test \
+	db_io_failure_test \
+	db_properties_test \
 	db_table_properties_test \
-	block_hash_index_test \
 	autovector_test \
 	column_family_test \
 	table_properties_collector_test \
@@ -262,6 +290,7 @@ TESTS = \
 	crc32c_test \
 	slice_transform_test \
 	dbformat_test \
+	env_basic_test \
 	env_test \
 	fault_injection_test \
 	filelock_test \
@@ -269,17 +298,18 @@ TESTS = \
 	file_reader_writer_test \
 	block_based_filter_block_test \
 	full_filter_block_test \
+	hash_table_test \
 	histogram_test \
 	inlineskiplist_test \
 	log_test \
 	manual_compaction_test \
-	memenv_test \
 	mock_env_test \
 	memtable_list_test \
 	merge_helper_test \
 	memory_test \
 	merge_test \
 	merger_test \
+	util_merge_operators_test \
 	options_file_test \
 	redis_test \
 	reduce_levels_test \
@@ -292,6 +322,7 @@ TESTS = \
 	backupable_db_test \
 	document_db_test \
 	json_document_test \
+	sim_cache_test \
 	spatial_db_test \
 	version_edit_test \
 	version_set_test \
@@ -308,6 +339,7 @@ TESTS = \
 	rate_limiter_test \
 	delete_scheduler_test \
 	options_test \
+	options_settable_test \
 	options_util_test \
 	event_logger_test \
 	cuckoo_table_builder_test \
@@ -328,9 +360,30 @@ TESTS = \
 	compact_on_deletion_collector_test \
 	compaction_job_stats_test \
 	transaction_test \
-	ldb_cmd_test
+	ldb_cmd_test \
+	iostats_context_test \
+	persistent_cache_test \

-SUBSET :=  $(shell echo $(TESTS) |sed s/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/)
+PARALLEL_TEST = \
+	backupable_db_test \
+	compact_on_deletion_collector_test \
+	db_compaction_filter_test \
+	db_compaction_test \
+	db_test \
+	db_universal_compaction_test \
+	fault_injection_test \
+	inlineskiplist_test \
+	manual_compaction_test \
+	table_test
+
+SUBSET := $(TESTS)
+ifdef ROCKSDBTESTS_START
+        SUBSET := $(shell echo $(SUBSET) | sed 's/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/')
+endif
+
+ifdef ROCKSDBTESTS_END
+        SUBSET := $(shell echo $(SUBSET) | sed 's/$(ROCKSDBTESTS_END).*//')
+endif

 TOOLS = \
 	sst_dump \
@@ -342,6 +395,9 @@ TOOLS = \
 	rocksdb_dump \
 	rocksdb_undump

+TEST_LIBS = \
+	librocksdb_env_basic_test.a
+
 # TODO: add back forward_iterator_bench, after making it build in all environemnts.
 BENCHMARKS = db_bench table_reader_bench cache_bench memtablerep_bench

@@ -355,6 +411,7 @@ else
 endif
 endif
 LIBRARY = ${LIBNAME}.a
+TOOLS_LIBRARY = ${LIBNAME}_tools.a

 ROCKSDB_MAJOR = $(shell egrep "ROCKSDB_MAJOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
 ROCKSDB_MINOR = $(shell egrep "ROCKSDB_MINOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
@@ -406,10 +463,10 @@ endif  # PLATFORM_SHARED_EXT
 .PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests package \
 	release tags valgrind_check whitebox_crash_test format static_lib shared_lib all \
 	dbg rocksdbjavastatic rocksdbjava install install-static install-shared uninstall \
-	analyze tools
+	analyze tools tools_lib


-all: $(LIBRARY) $(BENCHMARKS) tools $(TESTS)
+all: $(LIBRARY) $(BENCHMARKS) tools tools_lib test_libs $(TESTS)

 static_lib: $(LIBRARY)

@@ -417,6 +474,10 @@ shared_lib: $(SHARED)

 tools: $(TOOLS)

+tools_lib: $(TOOLS_LIBRARY)
+
+test_libs: $(TEST_LIBS)
+
 dbg: $(LIBRARY) $(BENCHMARKS) tools $(TESTS)

 # creates static library and programs
@@ -431,7 +492,26 @@ coverage:
        # Delete intermediate files
 	find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;

-# Extract the names of its tests by running db_test with --gtest_list_tests.
+ifneq (,$(filter check parallel_check,$(MAKECMDGOALS)),)
+# Use /dev/shm if it has the sticky bit set (otherwise, /tmp),
+# 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 $$f || f=/tmp;			\
+  perl -le 'use File::Temp "tempdir";'					\
+    -e 'print tempdir("'$$f'/rocksdb.XXXX", CLEANUP => 0)')
+endif
+endif
+
+# Run all tests in parallel, accumulating per-test logs in t/log-*.
+#
+# Each t/run-* file is a tiny generated bourne shell script that invokes one of
+# sub-tests. Why use a file for this?  Because that makes the invocation of
+# parallel below simpler, which in turn makes the parsing of parallel's
+# LOG simpler (the latter is for live monitoring as parallel
+# tests run).
+#
+# Test names are extracted by running tests with --gtest_list_tests.
 # This filter removes the "#"-introduced comments, and expands to
 # fully-qualified names by changing input like this:
 #
@@ -449,52 +529,33 @@ coverage:
 #   MultiThreaded/MultiThreadedDBTest.MultiThreaded/0
 #   MultiThreaded/MultiThreadedDBTest.MultiThreaded/1
 #
-test_names = \
-  ./db_test --gtest_list_tests						\
-    | perl -n								\
-      -e 's/ *\#.*//;'							\
-      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'			\
-      -e 'print qq! $$p$$2!'

-ifeq ($(MAKECMDGOALS),check)
-# Use /dev/shm if it has the sticky bit set (otherwise, /tmp),
-# and create a randomly-named rocksdb.XXXX directory therein.
-# We'll use that directory in the "make check" rules.
-ifeq ($(TMPD),)
-TMPD := $(shell f=/dev/shm; test -k $$f || f=/tmp;			\
-  perl -le 'use File::Temp "tempdir";'					\
-    -e 'print tempdir("'$$f'/rocksdb.XXXX", CLEANUP => 0)')
-endif
-endif
+parallel_tests = $(patsubst %,parallel_%,$(PARALLEL_TEST))
+.PHONY: gen_parallel_tests $(parallel_tests)
+$(parallel_tests): $(PARALLEL_TEST)
+	$(AM_V_at)TEST_BINARY=$(patsubst parallel_%,%,$@); \
+  TEST_NAMES=` \
+    ./$$TEST_BINARY --gtest_list_tests \
+    | perl -n \
+      -e 's/ *\#.*//;' \
+      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'	\
+      -e 'print qq! $$p$$2!'`; \
+	for TEST_NAME in $$TEST_NAMES; do \
+		TEST_SCRIPT=t/run-$$TEST_BINARY-$${TEST_NAME//\//-}; \
+		echo "  GEN     " $$TEST_SCRIPT; \
+    printf '%s\n' \
+      '#!/bin/sh' \
+      "d=\$(TMPD)$$TEST_SCRIPT" \
+      'mkdir -p $$d' \
+			"TEST_TMPDIR=\$$d ./$$TEST_BINARY --gtest_filter=$$TEST_NAME" \
+		> $$TEST_SCRIPT; \
+		chmod a=rx $$TEST_SCRIPT; \
+	done

-ifneq ($(T),)
-
-# Run all tests in parallel, accumulating per-test logs in t/log-*.
-
-# t_sanitized is each $(T) with "-" in place of each "/".
-t_sanitized = $(subst /,-,$(T))
-
-# t_run is each sanitized name with a leading "t/".
-t_run = $(patsubst %,t/%,$(t_sanitized))
-
-# Each t_run file is a tiny generated bourne shell script
-# that invokes one of db_tests's sub-tests. Why use a file
-# for this?  Because that makes the invocation of parallel
-# below simpler, which in turn makes the parsing of parallel's
-# LOG simpler (the latter is for live monitoring as parallel
-# tests run).
-filter = --gtest_filter=$(subst -,/,$(@F))
-$(t_run): Makefile db_test
-	$(AM_V_GEN)mkdir -p t
-	$(AM_V_at)rm -f $@ $@-t
-	$(AM_V_at)printf '%s\n'						\
-	    '#!/bin/sh'							\
-	    'd=$(TMPD)/$(@F)'						\
-	    'mkdir -p $$d'						\
-	    'TEST_TMPDIR=$$d ./db_test $(filter)'			\
-	  > $@-t
-	$(AM_V_at)chmod a=rx $@-t
-	$(AM_V_at)mv $@-t $@
+gen_parallel_tests:
+	$(AM_V_at)mkdir -p t
+	$(AM_V_at)rm -f t/run-*
+	$(MAKE) $(parallel_tests)

 # Reorder input lines (which are one per test) so that the
 # longest-running tests appear first in the output.
@@ -513,7 +574,7 @@ $(t_run): Makefile db_test
 # 107.816 PASS t/DBTest.EncodeDecompressedBlockSizeTest
 #
 slow_test_regexp = \
-  ^t/DBTest\.(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$
+	^t/run-table_test-HarnessTest.Randomized$$|^t/run-db_test-.*(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$
 prioritize_long_running_tests =						\
  perl -pe 's,($(slow_test_regexp)),100 $$1,'				\
    | sort -k1,1gr							\
@@ -524,26 +585,43 @@ prioritize_long_running_tests =						\
 # Run with "make J=200% check" to run two parallel jobs per core.
 # The default is to run one job per core (J=100%).
 # See "man parallel" for its "-j ..." option.
-J = 100%
+J ?= 100%

 # Use this regexp to select the subset of tests whose names match.
 tests-regexp = .

+t_run = $(wildcard t/run-*)
 .PHONY: check_0
-check_0: $(t_run)
+check_0:
+	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD); \
+	printf '%s\n' ''						\
+	  'To monitor subtest <duration,pass/fail,name>,'		\
+	  '  run "make watch-log" in a separate window' '';		\
+	test -t 1 && eta=--eta || eta=; \
+	{ \
+		printf './%s\n' $(filter-out $(PARALLEL_TEST),$(TESTS)); \
+		printf '%s\n' $(t_run); \
+	} \
+	  | $(prioritize_long_running_tests)				\
+	  | grep -E '$(tests-regexp)'					\
+	  | parallel -j$(J) --joblog=LOG $$eta --gnu '{} >& t/log-{/}'
+
+.PHONY: valgrind_check_0
+valgrind_check_0:
 	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD);				\
 	printf '%s\n' ''						\
 	  'To monitor subtest <duration,pass/fail,name>,'		\
 	  '  run "make watch-log" in a separate window' '';		\
 	test -t 1 && eta=--eta || eta=;					\
 	{								\
-	  printf './%s\n' $(filter-out db_test, $(TESTS));		\
+	  printf './%s\n' $(filter-out $(PARALLEL_TEST) %skiplist_test options_settable_test, $(TESTS));		\
 	  printf '%s\n' $(t_run);					\
 	}								\
 	  | $(prioritize_long_running_tests)				\
 	  | grep -E '$(tests-regexp)'					\
-	  | parallel -j$(J) --joblog=LOG $$eta --gnu '{} >& t/log-{/}'
-endif
+	  | parallel -j$(J) --joblog=LOG $$eta --gnu \
+      'if [[ "{}" == "./"* ]] ; then $(DRIVER) {} >& t/valgrind_log-{/}; ' \
+      'else {} >& t/valgrind_log-{/}; fi'

 CLEAN_FILES += t LOG $(TMPD)

@@ -560,11 +638,11 @@ watch-log:
 # If J != 1 and GNU parallel is installed, run the tests in parallel,
 # via the check_0 rule above.  Otherwise, run them sequentially.
 check: all
+	$(MAKE) gen_parallel_tests
 	$(AM_V_GEN)if test "$(J)" != 1                                  \
 	    && (parallel --gnu --help 2>/dev/null) |                    \
 	        grep -q 'GNU Parallel';                                 \
 	then                                                            \
-	    t=$$($(test_names));                                        \
 	    $(MAKE) T="$$t" TMPD=$(TMPD) check_0;                       \
 	else                                                            \
 	    for t in $(TESTS); do                                       \
@@ -586,7 +664,7 @@ ldb_tests: ldb
 crash_test: whitebox_crash_test blackbox_crash_test

 blackbox_crash_test: db_stress
-	python -u tools/db_crashtest.py --simple blackbox 
+	python -u tools/db_crashtest.py --simple blackbox
 	python -u tools/db_crashtest.py blackbox

 whitebox_crash_test: db_stress
@@ -603,14 +681,80 @@ asan_crash_test:
 	COMPILE_WITH_ASAN=1 $(MAKE) crash_test
 	$(MAKE) clean

+ubsan_check:
+	$(MAKE) clean
+	COMPILE_WITH_UBSAN=1 $(MAKE) check -j32
+	$(MAKE) clean
+
+ubsan_crash_test:
+	$(MAKE) clean
+	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test
+	$(MAKE) clean
+
 valgrind_check: $(TESTS)
-	for t in $(filter-out %skiplist_test,$(TESTS)); do \
-		$(VALGRIND_VER) $(VALGRIND_OPTS) ./$$t; \
+	$(MAKE) gen_parallel_tests
+	$(AM_V_GEN)if test "$(J)" != 1                                  \
+	    && (parallel --gnu --help 2>/dev/null) |                    \
+	        grep -q 'GNU Parallel';                                 \
+	then                                                            \
+      $(MAKE) TMPD=$(TMPD)                                        \
+      DRIVER="$(VALGRIND_VER) $(VALGRIND_OPTS)" valgrind_check_0; \
+	else                                                            \
+		for t in $(filter-out %skiplist_test options_settable_test,$(TESTS)); do \
+			$(VALGRIND_VER) $(VALGRIND_OPTS) ./$$t; \
+			ret_code=$$?; \
+			if [ $$ret_code -ne 0 ]; then \
+				exit $$ret_code; \
+			fi; \
+		done; \
+	fi
+
+
+ifneq ($(PAR_TEST),)
+parloop:
+	ret_bad=0;							\
+	for t in $(PAR_TEST); do		\
+		echo "===== Running $$t in parallel $(NUM_PAR)";\
+		if [ $(db_test) -eq 1 ]; then \
+			seq $(J) | v="$$t" parallel --gnu 's=$(TMPD)/rdb-{};  export TEST_TMPDIR=$$s;' \
+				'timeout 2m ./db_test --gtest_filter=$$v >> $$s/log-{} 2>1'; \
+		else\
+			seq $(J) | v="./$$t" parallel --gnu 's=$(TMPD)/rdb-{};' \
+			     'export TEST_TMPDIR=$$s; timeout 10m $$v >> $$s/log-{} 2>1'; \
+		fi; \
 		ret_code=$$?; \
 		if [ $$ret_code -ne 0 ]; then \
-			exit $$ret_code; \
+			ret_bad=$$ret_code; \
+			echo $$t exited with $$ret_code; \
 		fi; \
-	done
+	done; \
+	exit $$ret_bad;
+endif
+
+test_names = \
+  ./db_test --gtest_list_tests						\
+    | perl -n								\
+      -e 's/ *\#.*//;'							\
+      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'			\
+      -e 'print qq! $$p$$2!'
+
+parallel_check: $(TESTS)
+	$(AM_V_GEN)if test "$(J)" > 1                                  \
+	    && (parallel --gnu --help 2>/dev/null) |                    \
+	        grep -q 'GNU Parallel';                                 \
+	then                                                            \
+	    echo Running in parallel $(J);			\
+	else                                                            \
+	    echo "Need to have GNU Parallel and J > 1"; exit 1;		\
+	fi;								\
+	ret_bad=0;							\
+	echo $(J);\
+	echo Test Dir: $(TMPD); \
+        seq $(J) | parallel --gnu 's=$(TMPD)/rdb-{}; rm -rf $$s; mkdir $$s'; \
+	$(MAKE)  PAR_TEST="$(shell $(test_names))" TMPD=$(TMPD) \
+		J=$(J) db_test=1 parloop; \
+	$(MAKE) PAR_TEST="$(filter-out db_test, $(TESTS))" \
+		TMPD=$(TMPD) J=$(J) db_test=0 parloop;

 analyze: clean
 	$(CLANG_SCAN_BUILD) --use-analyzer=$(CLANG_ANALYZER) \
@@ -648,7 +792,7 @@ clean:

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

 format:
 	build_tools/format-diff.sh
@@ -663,7 +807,15 @@ $(LIBRARY): $(LIBOBJECTS)
 	$(AM_V_AR)rm -f $@
 	$(AM_V_at)$(AR) $(ARFLAGS) $@ $(LIBOBJECTS)

-db_bench: db/db_bench.o $(LIBOBJECTS) $(TESTUTIL)
+$(TOOLS_LIBRARY): $(BENCH_SOURCES:.cc=.o) $(TOOL_SOURCES:.cc=.o) $(LIB_SOURCES:.cc=.o) $(TESTUTIL)
+	$(AM_V_AR)rm -f $@
+	$(AM_V_at)$(AR) $(ARFLAGS) $@ $^
+
+librocksdb_env_basic_test.a: util/env_basic_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_V_AR)rm -f $@
+	$(AM_V_at)$(AR) $(ARFLAGS) $@ $^
+
+db_bench: tools/db_bench.o $(BENCHTOOLOBJECTS)
 	$(AM_LINK)

 cache_bench: util/cache_bench.o $(LIBOBJECTS) $(TESTUTIL)
@@ -672,9 +824,6 @@ cache_bench: util/cache_bench.o $(LIBOBJECTS) $(TESTUTIL)
 memtablerep_bench: db/memtablerep_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)

-block_hash_index_test: table/block_hash_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 db_stress: tools/db_stress.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)

@@ -720,6 +869,9 @@ stringappend_test: utilities/merge_operators/string_append/stringappend_test.o $
 redis_test: utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+hash_table_test: utilities/persistent_cache/hash_table_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 histogram_test: util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -738,6 +890,15 @@ slice_transform_test: util/slice_transform_test.o $(LIBOBJECTS) $(TESTHARNESS)
 db_test: db/db_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_test2: db/db_test2.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_block_cache_test: db/db_block_cache_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_bloom_filter_test: db/db_bloom_filter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_log_iter_test: db/db_log_iter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -753,6 +914,12 @@ db_dynamic_level_test: db/db_dynamic_level_test.o db/db_test_util.o $(LIBOBJECTS
 db_inplace_update_test: db/db_inplace_update_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_iterator_test: db/db_iterator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_sst_test: db/db_sst_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_tailing_iter_test: db/db_tailing_iter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -765,11 +932,17 @@ db_universal_compaction_test: db/db_universal_compaction_test.o db/db_test_util.
 db_wal_test: db/db_wal_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_io_failure_test: db/db_io_failure_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_properties_test: db/db_properties_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_table_properties_test: db/db_table_properties_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 log_write_bench: util/log_write_bench.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK) $(pg)
+	$(AM_LINK) $(PROFILING_FLAGS)

 plain_table_db_test: db/plain_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
@@ -778,7 +951,7 @@ comparator_db_test: db/comparator_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 table_reader_bench: table/table_reader_bench.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK) $(pg)
+	$(AM_LINK) $(PROFILING_FLAGS)

 perf_context_test: db/perf_context_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
@@ -798,9 +971,18 @@ document_db_test: utilities/document/document_db_test.o $(LIBOBJECTS) $(TESTHARN
 json_document_test: utilities/document/json_document_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+sim_cache_test: utilities/simulator_cache/sim_cache_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 spatial_db_test: utilities/spatialdb/spatial_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+env_mirror_test: utilities/env_mirror_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+env_registry_test: utilities/env_registry_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -828,6 +1010,9 @@ wal_manager_test: db/wal_manager_test.o $(LIBOBJECTS) $(TESTHARNESS)
 dbformat_test: db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+env_basic_test: util/env_basic_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 env_test: util/env_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -903,6 +1088,9 @@ merge_test: db/merge_test.o $(LIBOBJECTS) $(TESTHARNESS)
 merger_test: table/merger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+util_merge_operators_test: utilities/util_merge_operators_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 options_file_test: db/options_file_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -927,7 +1115,7 @@ cuckoo_table_reader_test: table/cuckoo_table_reader_test.o $(LIBOBJECTS) $(TESTH
 cuckoo_table_db_test: db/cuckoo_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-listener_test: db/listener_test.o $(LIBOBJECTS) $(TESTHARNESS)
+listener_test: db/listener_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 thread_list_test: util/thread_list_test.o $(LIBOBJECTS) $(TESTHARNESS)
@@ -939,18 +1127,21 @@ compact_files_test: db/compact_files_test.o $(LIBOBJECTS) $(TESTHARNESS)
 options_test: util/options_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+options_settable_test: util/options_settable_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 options_util_test: utilities/options/options_util_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_bench_tool_test: tools/db_bench_tool_test.o $(BENCHTOOLOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 event_logger_test: util/event_logger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 sst_dump_test: tools/sst_dump_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-memenv_test : util/memenv_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 optimistic_transaction_test: utilities/transactions/optimistic_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -963,7 +1154,7 @@ manual_compaction_test: db/manual_compaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 filelock_test: util/filelock_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-auto_roll_logger_test: util/auto_roll_logger_test.o $(LIBOBJECTS) $(TESTHARNESS)
+auto_roll_logger_test: db/auto_roll_logger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 memtable_list_test: db/memtable_list_test.o $(LIBOBJECTS) $(TESTHARNESS)
@@ -981,12 +1172,21 @@ transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TEST
 sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 	$(AM_LINK)

+repair_test: db/repair_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 ldb_cmd_test: tools/ldb_cmd_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 ldb: tools/ldb.o $(LIBOBJECTS)
 	$(AM_LINK)

+iostats_context_test: util/iostats_context_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
+persistent_cache_test: utilities/persistent_cache/persistent_cache_test.o  db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1029,7 +1229,11 @@ install: install-static
 # ---------------------------------------------------------------------------

 JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
-ARCH := $(shell getconf LONG_BIT)
+ifeq ($(PLATFORM), OS_SOLARIS)
+	ARCH := $(shell isainfo -b)
+else
+	ARCH := $(shell getconf LONG_BIT)
+endif
 ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
 ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux$(ARCH).jar
 ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
@@ -1037,14 +1241,19 @@ ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PA
 ROCKSDB_SOURCES_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar

 ifeq ($(PLATFORM), OS_MACOSX)
-ROCKSDBJNILIB = librocksdbjni-osx.jnilib
-ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar
+	ROCKSDBJNILIB = librocksdbjni-osx.jnilib
+	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar
 ifneq ("$(wildcard $(JAVA_HOME)/include/darwin)","")
 	JAVA_INCLUDE = -I$(JAVA_HOME)/include -I $(JAVA_HOME)/include/darwin
 else
 	JAVA_INCLUDE = -I/System/Library/Frameworks/JavaVM.framework/Headers/
 endif
 endif
+ifeq ($(PLATFORM), OS_SOLARIS)
+	ROCKSDBJNILIB = librocksdbjni-solaris$(ARCH).so
+	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-solaris$(ARCH).jar
+	JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/solaris
+endif

 libz.a:
 	-rm -rf zlib-1.2.8
@@ -1138,11 +1347,10 @@ jtest: rocksdbjava
 jdb_bench:
 	cd java;$(MAKE) db_bench;

-commit-prereq:
-	$(MAKE) clean && $(MAKE) all check;
+commit_prereq: build_tools/rocksdb-lego-determinator \
+               build_tools/precommit_checker.py
+	J=$(J) build_tools/precommit_checker.py unit unit_481 clang_unit tsan asan ubsan lite
 	$(MAKE) clean && $(MAKE) jclean && $(MAKE) rocksdbjava;
-	$(MAKE) clean && USE_CLANG=1 $(MAKE) all;
-	$(MAKE) clean && OPT=-DROCKSDB_LITE $(MAKE) static_lib;

 xfunc:
 	for xftest in $(XFUNC_TESTS); do \
@@ -9,7 +9,7 @@ At Facebook, we use RocksDB as a backend for many different stateful services. W
 ## LinkedIn
 Two different use cases at Linkedin are using RocksDB as a storage engine:

-1. LinkedIn's follow feed for storing user's activities
+1. LinkedIn's follow feed for storing user's activities. Check out the blog post: https://engineering.linkedin.com/blog/2016/03/followfeed--linkedin-s-feed-made-faster-and-smarter
 2. Apache Samza, open source framework for stream processing

 Learn more about those use cases in a Tech Talk by Ankit Gupta and Naveen Somasundaram: http://www.youtube.com/watch?v=plqVp_OnSzg
@@ -37,3 +37,25 @@ Check out our RocksDB Protobuf merge operator at: https://github.com/vladb38/roc

 ## Santanader UK/Cloudera Profession Services
 Check out their blog post: http://blog.cloudera.com/blog/2015/08/inside-santanders-near-real-time-data-ingest-architecture/
+
+## Airbnb
+Airbnb is using RocksDB as a storage engine for their personalized search service. You can learn more about it here: https://www.youtube.com/watch?v=ASQ6XMtogMs
+
+## Pinterest
+Pinterest's Object Retrieval System uses RocksDB for storage: https://www.youtube.com/watch?v=MtFEVEs_2Vo
+
+## Smyte
+[Smyte](https://www.smyte.com/) uses RocksDB as the storage layer for their core key-value storage, high-performance counters and time-windowed HyperLogLog services.
+
+## Rakuten Marketing
+[Rakuten Marketing](https://marketing.rakuten.com/) uses RocksDB as the disk cache layer for the real-time bidding service in their Performance DSP.
+
+## VWO, Wingify
+[VWO's](https://vwo.com/) Smart Code checker and URL helper uses RocksDB to store all the URLs where VWO's Smart Code is installed.
+
+## quasardb
+[quasardb](https://www.quasardb.net) is a high-performance, distributed, transactional key-value database that integrates well with in-memory analytics engines such as Apache Spark. 
+quasardb uses a heavily tuned RocksDB as its persistence layer.
+
+## Netflix
+[Netflix](http://techblog.netflix.com/2016/05/application-data-caching-using-ssds.html) uses RocksDB on spinning disks to cache application data.
@@ -2,7 +2,7 @@ version: 1.0.{build}
 before_build:
 - md %APPVEYOR_BUILD_FOLDER%\build
 - cd %APPVEYOR_BUILD_FOLDER%\build
- cmake -G "Visual Studio 12 Win64" -DOPTDBG=1 ..
+- cmake -G "Visual Studio 12 Win64" -DOPTDBG=1 -DXPRESS=1 ..
 - cd ..
 build:
  project: build\rocksdb.sln
@@ -10,6 +10,7 @@ build:
  verbosity: minimal
 test:
 test_script:
- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test -Exclude DBTest.Randomized,DBTest.FileCreationRandomFailure -Concurrency 18
+- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test2 -Concurrency 8
+- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test -Exclude DBTest.GroupCommitTest -Concurrency 10
 - ps: build_tools\run_ci_db_test.ps1 -Run env_test -Concurrency 1

@@ -10,13 +10,185 @@ class FacebookArcanistConfiguration extends ArcanistConfiguration {
                                 ArcanistBaseWorkflow $workflow,
                                 $error_code) {
    if ($command == 'diff' && !$workflow->isRawDiffSource()) {
-      $this->maybePushToJenkins($workflow);
+      $this->startTestsInJenkins($workflow);
+      $this->startTestsInSandcastle($workflow);
    }
  }

+  //////////////////////////////////////////////////////////////////////
+  /*  Run tests in sandcastle */
+  function postURL($diffID, $url) {
+    $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
+           . '"name":"click here for sandcastle tests for D' . $diffID . '", '
+           . '"link":"' . $url . '"}\' | '
+           . 'http_proxy=fwdproxy.any.facebook.com:8080 '
+           . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
+           . 'differential.updateunitresults';
+    shell_exec($cmd);
+  }
+
+  function updateTestCommand($diffID, $test, $status) {
+    $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
+           . '"name":"' . $test . '", '
+           . '"result":"' . $status . '"}\' | '
+           . 'http_proxy=fwdproxy.any.facebook.com:8080 '
+           . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
+           . 'differential.updateunitresults';
+    return $cmd;
+  }
+
+  function updateTest($diffID, $test) {
+    shell_exec($this->updateTestCommand($diffID, $test, "waiting"));
+  }
+
+  function getSteps($diffID, $username, $test) {
+    $arcrc_content = exec("cat ~/.arcrc | gzip -f | base64 -w0");
+
+    // Sandcastle machines don't have arc setup. We copy the user certificate
+    // and authenticate using that in sandcastle
+    $setup = array(
+      "name" => "Setup arcrc",
+      "shell" => "echo " . $arcrc_content . " | base64 --decode"
+                 . " | gzip -d > ~/.arcrc",
+      "user" => "root"
+    );
+
+    // arc demands certain permission on its config
+    $fix_permission = array(
+      "name" => "Fix environment",
+      "shell" => "chmod 600 ~/.arcrc",
+      "user" => "root"
+    );
+
+    // fbcode is a sub-repo. We cannot patch until we add it to ignore otherwise
+    // git thinks it is uncommited change
+    $fix_git_ignore = array(
+      "name" => "Fix git ignore",
+      "shell" => "echo fbcode >> .git/info/exclude",
+      "user" => "root"
+    );
+
+    // Patch the code (keep your fingures crossed)
+    $patch = array(
+      "name" => "Patch " . $diffID,
+      "shell" => "HTTPS_PROXY=fwdproxy:8080 arc --arcrc-file ~/.arcrc "
+                  . "patch --diff " . $diffID,
+      "user" => "root"
+    );
+
+    // Clean up the user arc config we are using
+    $cleanup = array(
+      "name" => "Arc cleanup",
+      "shell" => "rm -f ~/.arcrc",
+      "user" => "root"
+    );
+
+    // Construct the steps in the order of execution
+    $steps[] = $setup;
+    $steps[] = $fix_permission;
+    $steps[] = $fix_git_ignore;
+    $steps[] = $patch;
+
+    // Run the actual command
+    $this->updateTest($diffID, $test);
+    $cmd = $this->updateTestCommand($diffID, $test, "running") . ";"
+           . "./build_tools/precommit_checker.py " . $test
+           . "; exit_code=$?; ([[ \$exit_code -eq 0 ]] &&"
+           . $this->updateTestCommand($diffID, $test, "pass") . ")"
+           . "||" . $this->updateTestCommand($diffID, $test, "fail")
+           . "; cat /tmp/precommit-check.log"
+           . "; for f in `ls t/log-*`; do echo \$f; cat \$f; done;"
+           . "[[ \$exit_code -eq 0 ]]";
+
+    $run_test = array(
+      "name" => "Run " . $test,
+      "shell" => $cmd,
+      "user" => "root",
+    );
+
+    $steps[] = $run_test;
+    $steps[] = $cleanup;
+
+    return $steps;
+  }
+
+  function startTestsInSandcastle($workflow) {
+    // extract information we need from workflow or CLI
+    $diffID = $workflow->getDiffId();
+    $username = exec("whoami");
+
+    if ($diffID == null || $username == null) {
+      // there is no diff and we can't extract username
+      // we cannot schedule sandcasstle job
+      return;
+    }
+
+    if (strcmp(getenv("ROCKSDB_CHECK_ALL"), 1) == 0) {
+      // extract all tests from the CI definition
+      $output = file_get_contents("build_tools/rocksdb-lego-determinator");
+      preg_match_all('/[ ]{2}([a-zA-Z0-9_]+)[\)]{1}/', $output, $matches);
+      $tests = $matches[1];
+    } else {
+      // manually list of tests we want to run in sandcastle
+      $tests = array(
+        "unit", "unit_481", "clang_unit", "tsan", "asan", "lite_test", "valgrind"
+      );
+    }
+
+    // construct a job definition for each test and add it to the master plan
+    foreach ($tests as $test) {
+      $arg[] = array(
+        "name" => "RocksDB diff " . $diffID . " test " . $test,
+        "steps" => $this->getSteps($diffID, $username, $test)
+      );
+    }
+
+    // we cannot submit the parallel execution master plan to sandcastle
+    // we need supply the job plan as a determinator
+    // so we construct a small job that will spit out the master job plan
+    // which sandcastle will parse and execute
+    // Why compress ? Otherwise we run over the max string size.
+    $cmd = "echo " . base64_encode(json_encode($arg))
+           . " | gzip -f | base64 -w0";
+    $arg_encoded = shell_exec($cmd);
+
+    $command = array(
+      "name" => "Run diff " . $diffID . "for user " . $username,
+      "steps" => array()
+    );
+
+    $command["steps"][] = array(
+      "name" => "Generate determinator",
+      "shell" => "echo " . $arg_encoded . " | base64 --decode | gzip -d"
+                 . " | base64 --decode",
+      "determinator" => true,
+      "user" => "root"
+    );
+
+    // submit to sandcastle
+    $url = 'https://interngraph.intern.facebook.com/sandcastle/generate?'
+            .'command=SandcastleUniversalCommand'
+            .'&vcs=rocksdb-git&revision=origin%2Fmaster&type=lego'
+            .'&user=krad&alias=rocksdb-precommit'
+            .'&command-args=' . urlencode(json_encode($command));
+
+    $cmd = 'https_proxy= HTTPS_PROXY= curl -s -k -F app=659387027470559 '
+            . '-F token=AeO_3f2Ya3TujjnxGD4 "' . $url . '"';
+
+    $output = shell_exec($cmd);
+
+    // extract sandcastle URL from the response
+    preg_match('/url": "(.+)"/', $output, $sandcastle_url);
+
+    echo "\nSandcastle URL: " . $sandcastle_url[1] . "\n";
+
+    // Ask phabricator to display it on the diff UI
+    $this->postURL($diffID, $sandcastle_url[1]);
+  }
+
  //////////////////////////////////////////////////////////////////////
  /* Send off builds to jenkins */
-  function maybePushToJenkins($workflow) {
+  function startTestsInJenkins($workflow) {
    $diffID = $workflow->getDiffID();
    if ($diffID === null) {
      return;
@@ -31,5 +203,4 @@ class FacebookArcanistConfiguration extends ArcanistConfiguration {
               ."buildWithParameters?token=AUTH&DIFF_ID=$diffID";
    system("curl --noproxy '*' \"$url\" > /dev/null 2>&1");
  }
-
 }
@@ -1,5 +1,5 @@
 #!/usr/bin/python
-# Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+# 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.
@@ -2714,7 +2714,7 @@ def CheckSpacing(filename, clean_lines, linenum, nesting_state, error):

    # Look for < that is not surrounded by spaces.  This is only
    # triggered if both sides are missing spaces, even though
-    # technically should should flag if at least one side is missing a
+    # technically should flag if at least one side is missing a
    # space.  This is done to avoid some false positives with shifts.
    match = Search(r'[^\s<]<([^\s=<].*)', reduced_line)
    if (match and
@@ -52,7 +52,12 @@ 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 "$PWD/build_tools/fbcode_config.sh"
+    if [ -z "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
+      source "$PWD/build_tools/fbcode_config.sh"
+    else
+      # we need this to build with MySQL. Don't use for other purposes.
+      source "$PWD/build_tools/fbcode_config4.8.1.sh"
+    fi
 fi

 # Delete existing output, if it exists
@@ -184,17 +189,19 @@ if [ "$CROSS_COMPILE" = "true" -o "$FBCODE_BUILD" = "true" ]; then
    # Also don't need any compilation tests if compiling on fbcode
    true
 else
-    # Test whether fallocate is available
-    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
-      #include <fcntl.h>
-      #include <linux/falloc.h>
-      int main() {
-	int fd = open("/dev/null", 0);
-  fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, 0, 1024);
-      }
+    if ! test $ROCKSDB_DISABLE_FALLOCATE; then
+        # Test whether fallocate is available
+        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <fcntl.h>
+          #include <linux/falloc.h>
+          int main() {
+      int fd = open("/dev/null", 0);
+      fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, 0, 1024);
+          }
 EOF
-    if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_FALLOCATE_PRESENT"
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_FALLOCATE_PRESENT"
+        fi
    fi

    # Test whether Snappy library is installed
@@ -292,13 +299,14 @@ EOF

    # Test whether jemalloc is available
    if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null -ljemalloc \
-      2>/dev/null; then 
+      2>/dev/null; then
        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ljemalloc"
        JAVA_LDFLAGS="$JAVA_LDFLAGS -ljemalloc"
+        JEMALLOC=1
    else
        # jemalloc is not available. Let's try tcmalloc
        if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null \
-          -ltcmalloc 2>/dev/null; then 
+          -ltcmalloc 2>/dev/null; then
            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltcmalloc"
            JAVA_LDFLAGS="$JAVA_LDFLAGS -ltcmalloc"
        fi
@@ -315,6 +323,55 @@ EOF
    if [ "$?" = 0 ]; then
        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
    fi
+
+    # Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <pthread.h>
+      int main() {
+        int x = PTHREAD_MUTEX_ADAPTIVE_NP;
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_PTHREAD_ADAPTIVE_MUTEX"
+    fi
+
+    # Test whether backtrace is available
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <execinfo.h>>
+      int main() {
+        void* frames[1];
+        backtrace_symbols(frames, backtrace(frames, 1));
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+    else
+        # Test whether execinfo library is installed
+        $CXX $CFLAGS -lexecinfo -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <execinfo.h>
+          int main() {
+            void* frames[1];
+            backtrace_symbols(frames, backtrace(frames, 1));
+          }
+EOF
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lexecinfo"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lexecinfo"
+        fi
+    fi
+
+    # Test if -pg is supported
+    $CXX $CFLAGS -pg -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      int main() {
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        PROFILING_FLAGS=-pg
+    fi
 fi

 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -354,7 +411,15 @@ if test "$USE_SSE"; then
  # if Intel SSE instruction set is supported, set USE_SSE=1
  COMMON_FLAGS="$COMMON_FLAGS -msse -msse4.2 "
 elif test -z "$PORTABLE"; then
-  COMMON_FLAGS="$COMMON_FLAGS -march=native "
+  if test -n "`echo $TARGET_ARCHITECTURE | grep ^ppc64`"; then
+    # Tune for this POWER processor, treating '+' models as base models
+    POWER=`LD_SHOW_AUXV=1 /bin/true | grep AT_PLATFORM | grep -E -o power[0-9]+`
+    COMMON_FLAGS="$COMMON_FLAGS -mcpu=$POWER -mtune=$POWER "
+  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^s390x`"; then
+    COMMON_FLAGS="$COMMON_FLAGS -march=z10 "
+  else
+    COMMON_FLAGS="$COMMON_FLAGS -march=native "
+  fi
 fi

 PLATFORM_CCFLAGS="$PLATFORM_CCFLAGS $COMMON_FLAGS"
@@ -387,3 +452,7 @@ echo "ROCKSDB_MINOR=$ROCKSDB_MINOR" >> "$OUTPUT"
 echo "ROCKSDB_PATCH=$ROCKSDB_PATCH" >> "$OUTPUT"
 echo "CLANG_SCAN_BUILD=$CLANG_SCAN_BUILD" >> "$OUTPUT"
 echo "CLANG_ANALYZER=$CLANG_ANALYZER" >> "$OUTPUT"
+echo "PROFILING_FLAGS=$PROFILING_FLAGS" >> "$OUTPUT"
+if test -n "$JEMALLOC"; then
+  echo "JEMALLOC=1" >> "$OUTPUT"
+fi
@@ -1,19 +1,16 @@
-# 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/ebc96bc2fb751b5a0300b8d91a95bdf24ac1d88b/4.9.x/centos6-native/108cf83
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/b91de48a4974ec839946d824402b098d43454cef/stable/centos6-native/7aaccbe
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/53e0eac8911888a105aa98b9a35fe61cf1d8b278/4.9.x/gcc-4.9-glibc-2.20/024dbc3
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/ee36ac9a72dfac4a995f1b215bb4c0fc8a0f6f91/2.20/gcc-4.9-glibc-2.20/500e281
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-4.9-glibc-2.20/e9936bf
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2b24f1e99454f9ca7b0301720f94836dae1bf71b/1.2.8/gcc-5-glibc-2.23/9bc6787
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/af7c14c9b652cdd5ec34eadd25c3f38a9b306c5d/1.0.6/gcc-5-glibc-2.23/9bc6787
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0e790b441e2d9acd68d51e1d2e028f88c6a79ddf/r131/gcc-5-glibc-2.23/9bc6787
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/1408484d03b15492aa54b10356104e9dc22e1cc5/0.6.1/gcc-5-glibc-2.23/9bc6787
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/5a5c7a6608cb32f1e1e7f814023d5bdfbd136370/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/a4c2adecffcaa68d5585d06be2252e3efa52555b/master/gcc-5-glibc-2.23/1c32b4b
+NUMA_BASE=/mnt/gvfs/third-party2/numa/1abc0d3c01743b854676423cf2d3629912f34930/2.0.8/gcc-5-glibc-2.23/9bc6787
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/ad576de2a1ea560c4d3434304f0fc4e079bede42/trunk/gcc-5-glibc-2.23/b1847cb
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/f24be37d170e04be6e469af487644d4d62e1c6c1/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/9d65c666b9adf8f2a989fd4b98a9a5e7d3afa233/2.26/centos6-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/9cee5a3628dc9d4b93897972c58eba865e25b270/3.10.0/gcc-4.9-glibc-2.20/e9936bf
@@ -0,0 +1,16 @@
+GCC_BASE=/mnt/gvfs/third-party2/gcc/ebc96bc2fb751b5a0300b8d91a95bdf24ac1d88b/4.8.1/centos6-native/cc6c9dc
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/b91de48a4974ec839946d824402b098d43454cef/stable/centos6-native/7aaccbe
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/53e0eac8911888a105aa98b9a35fe61cf1d8b278/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/ee36ac9a72dfac4a995f1b215bb4c0fc8a0f6f91/2.17/gcc-4.8.1-glibc-2.17/99df8fc
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-4.8.1-glibc-2.17/c3f970a
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2b24f1e99454f9ca7b0301720f94836dae1bf71b/1.2.8/gcc-4.8.1-glibc-2.17/c3f970a
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/af7c14c9b652cdd5ec34eadd25c3f38a9b306c5d/1.0.6/gcc-4.8.1-glibc-2.17/c3f970a
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0e790b441e2d9acd68d51e1d2e028f88c6a79ddf/r131/gcc-4.8.1-glibc-2.17/c3f970a
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/1408484d03b15492aa54b10356104e9dc22e1cc5/0.6.1/gcc-4.8.1-glibc-2.17/c3f970a
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/5a5c7a6608cb32f1e1e7f814023d5bdfbd136370/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/a4c2adecffcaa68d5585d06be2252e3efa52555b/master/gcc-4.8.1-glibc-2.17/8d31e51
+NUMA_BASE=/mnt/gvfs/third-party2/numa/1abc0d3c01743b854676423cf2d3629912f34930/2.0.8/gcc-4.8.1-glibc-2.17/c3f970a
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/ad576de2a1ea560c4d3434304f0fc4e079bede42/trunk/gcc-4.8.1-glibc-2.17/675d945
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/f24be37d170e04be6e469af487644d4d62e1c6c1/4.0.9-36_fbk5_2933_gd092e3f/gcc-4.8.1-glibc-2.17/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/9d65c666b9adf8f2a989fd4b98a9a5e7d3afa233/2.26/centos6-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/9cee5a3628dc9d4b93897972c58eba865e25b270/3.8.1/gcc-4.8.1-glibc-2.17/c3f970a
@@ -0,0 +1,164 @@
+# Copyright (c) 2016-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.
+
+'''Filter for error messages in test output:
+    - Receives merged stdout/stderr from test on stdin
+    - Finds patterns of known error messages for test name (first argument)
+    - Prints those error messages to stdout
+'''
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import re
+import sys
+
+
+class ErrorParserBase(object):
+    def parse_error(self, line):
+        '''Parses a line of test output. If it contains an error, returns a
+        formatted message describing the error; otherwise, returns None.
+        Subclasses must override this method.
+        '''
+        raise NotImplementedError
+
+
+class GTestErrorParser(ErrorParserBase):
+    '''A parser that remembers the last test that began running so it can print
+    that test's name upon detecting failure.
+    '''
+    _GTEST_NAME_PATTERN = re.compile(r'\[ RUN      \] (\S+)$')
+    # format: '<filename or "unknown file">:<line #>: Failure'
+    _GTEST_FAIL_PATTERN = re.compile(r'(unknown file|\S+:\d+): Failure$')
+
+    def __init__(self):
+        self._last_gtest_name = 'Unknown test'
+
+    def parse_error(self, line):
+        gtest_name_match = self._GTEST_NAME_PATTERN.match(line)
+        if gtest_name_match:
+            self._last_gtest_name = gtest_name_match.group(1)
+            return None
+        gtest_fail_match = self._GTEST_FAIL_PATTERN.match(line)
+        if gtest_fail_match:
+            return '%s failed: %s' % (
+                    self._last_gtest_name, gtest_fail_match.group(1))
+        return None
+
+
+class MatchErrorParser(ErrorParserBase):
+    '''A simple parser that returns the whole line if it matches the pattern.
+    '''
+    def __init__(self, pattern):
+        self._pattern = re.compile(pattern)
+
+    def parse_error(self, line):
+        if self._pattern.match(line):
+            return line
+        return None
+
+
+class CompilerErrorParser(MatchErrorParser):
+    def __init__(self):
+        # format: '<filename>:<line #>:<column #>: error: <error msg>'
+        super(CompilerErrorParser, self).__init__(r'\S+:\d+:\d+: error:')
+
+
+class ScanBuildErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(ScanBuildErrorParser, self).__init__(
+                r'scan-build: \d+ bugs found.$')
+
+
+class DbCrashErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(DbCrashErrorParser, self).__init__(r'\*\*\*.*\^$|TEST FAILED.')
+
+
+class WriteStressErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(WriteStressErrorParser, self).__init__(
+                r'ERROR: write_stress died with exitcode=\d+')
+
+
+class AsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(AsanErrorParser, self).__init__(
+                r'==\d+==ERROR: AddressSanitizer:')
+
+
+class UbsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        # format: '<filename>:<line #>:<column #>: runtime error: <error msg>'
+        super(UbsanErrorParser, self).__init__(r'\S+:\d+:\d+: runtime error:')
+
+
+class ValgrindErrorParser(MatchErrorParser):
+    def __init__(self):
+        # just grab the summary, valgrind doesn't clearly distinguish errors
+        # from other log messages.
+        super(ValgrindErrorParser, self).__init__(r'==\d+== ERROR SUMMARY:')
+
+
+class CompatErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(CompatErrorParser, self).__init__(r'==== .*[Ee]rror.* ====$')
+
+
+class TsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(TsanErrorParser, self).__init__(r'WARNING: ThreadSanitizer:')
+
+
+_TEST_NAME_TO_PARSERS = {
+    'punit': [CompilerErrorParser, GTestErrorParser],
+    'unit': [CompilerErrorParser, GTestErrorParser],
+    'unit_481': [CompilerErrorParser, GTestErrorParser],
+    'clang_unit': [CompilerErrorParser, GTestErrorParser],
+    'clang_analyze': [CompilerErrorParser, ScanBuildErrorParser],
+    'code_cov': [CompilerErrorParser, GTestErrorParser],
+    'unity': [CompilerErrorParser, GTestErrorParser],
+    'lite': [CompilerErrorParser],
+    'lite_test': [CompilerErrorParser, GTestErrorParser],
+    'stress_crash': [CompilerErrorParser, DbCrashErrorParser],
+    'write_stress': [CompilerErrorParser, WriteStressErrorParser],
+    'asan': [CompilerErrorParser, GTestErrorParser, AsanErrorParser],
+    'asan_crash': [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
+    'ubsan': [CompilerErrorParser, GTestErrorParser, UbsanErrorParser],
+    'ubsan_crash': [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
+    'valgrind': [CompilerErrorParser, GTestErrorParser, ValgrindErrorParser],
+    'tsan': [CompilerErrorParser, GTestErrorParser, TsanErrorParser],
+    'format_compatible': [CompilerErrorParser, CompatErrorParser],
+    'run_format_compatible': [CompilerErrorParser, CompatErrorParser],
+    'no_compression': [CompilerErrorParser, GTestErrorParser],
+    'run_no_compression': [CompilerErrorParser, GTestErrorParser],
+    'regression': [CompilerErrorParser],
+    'run_regression': [CompilerErrorParser],
+}
+
+
+def main():
+    if len(sys.argv) != 2:
+        return 'Usage: %s <test name>' % sys.argv[0]
+    test_name = sys.argv[1]
+    if test_name not in _TEST_NAME_TO_PARSERS:
+        return 'Unknown test name: %s' % test_name
+
+    error_parsers = []
+    for parser_cls in _TEST_NAME_TO_PARSERS[test_name]:
+        error_parsers.append(parser_cls())
+
+    for line in sys.stdin:
+        line = line.strip()
+        for error_parser in error_parsers:
+            error_msg = error_parser.parse_error(line)
+            if error_msg is not None:
+                print(error_msg)
+
+
+if __name__ == '__main__':
+    sys.exit(main())
@@ -13,8 +13,8 @@ source "$BASEDIR/dependencies.sh"
 CFLAGS=""

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

 # glibc
 GLIBC_INCLUDE="$GLIBC_BASE/include"
@@ -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/"
@@ -60,7 +56,7 @@ if test -z $PIC_BUILD; then
 else
  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
 fi
-CFLAGS+=" -DGFLAGS=gflags"
+CFLAGS+=" -DGFLAGS=google"

 # location of jemalloc
 JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
@@ -76,22 +72,13 @@ if test -z $PIC_BUILD; then
  LIBUNWIND="$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

 BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"

-DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE"
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE"

 STDLIBS="-L $GCC_BASE/lib64"

@@ -100,7 +87,7 @@ 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"
+CLANG_SCAN_BUILD="$CLANG_SRC/clang/tools/scan-build/scan-build"

 if [ -z "$USE_CLANG" ]; then
  # gcc
@@ -108,8 +95,8 @@ if [ -z "$USE_CLANG" ]; then
  CXX="$GCC_BASE/bin/g++"

  CFLAGS+=" -B$BINUTILS/gold"
-  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  JEMALLOC=1
 else
  # clang
@@ -120,8 +107,8 @@ else
  KERNEL_HEADERS_INCLUDE="$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++/4.9.x "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/4.9.x/x86_64-facebook-linux "
  CFLAGS+=" -isystem $GLIBC_INCLUDE"
  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
  CFLAGS+=" -isystem $CLANG_INCLUDE"
@@ -132,21 +119,18 @@ else
 fi

 CFLAGS+=" $DEPS_INCLUDE"
-CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE"
 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=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/gcc-4.9-glibc-2.20/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
-EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
-# required by libtbb
-EXEC_LDFLAGS+=" -ldl"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/gcc-4.9-glibc-2.20/lib"

 PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"

-EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS"

 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 
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD
@@ -4,85 +4,78 @@
 # fbcode settings.  It uses the latest g++ compiler and also
 # uses jemalloc

+BASEDIR=`dirname $BASH_SOURCE`
+source "$BASEDIR/dependencies_4.8.1.sh"
+
 # location of libgcc
-LIBGCC_BASE="/mnt/gvfs/third-party2/libgcc/d00277f4559e261ed0a81f30f23c0ce5564e359e/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc"
 LIBGCC_INCLUDE="$LIBGCC_BASE/include"
 LIBGCC_LIBS=" -L $LIBGCC_BASE/libs"

 # location of glibc
-GLIBC_REV=0600c95b31226b5e535614c590677d87c62d8016
-GLIBC_INCLUDE="/mnt/gvfs/third-party2/glibc/$GLIBC_REV/2.17/gcc-4.8.1-glibc-2.17/99df8fc/include"
-GLIBC_LIBS=" -L /mnt/gvfs/third-party2/glibc/$GLIBC_REV/2.17/gcc-4.8.1-glibc-2.17/99df8fc/lib"
+GLIBC_INCLUDE="$GLIBC_BASE/include"
+GLIBC_LIBS=" -L $GLIBC_BASE/lib"

 # location of snappy headers and libraries
-SNAPPY_REV=cbf6f1f209e5bd160bdc5d971744e039f36b1566
-SNAPPY_INCLUDE=" -I /mnt/gvfs/third-party2/snappy/$SNAPPY_REV/1.1.3/gcc-4.8.1-glibc-2.17/c3f970a/include"
-SNAPPY_LIBS=" /mnt/gvfs/third-party2/snappy/$SNAPPY_REV/1.1.3/gcc-4.8.1-glibc-2.17/c3f970a/lib/libsnappy.a"
+SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include"
+SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy.a"

 # location of zlib headers and libraries
-ZLIB_REV=6d39cb54708049f527e713ad19f2aadb9d3667e8
-ZLIB_INCLUDE=" -I /mnt/gvfs/third-party2/zlib/$ZLIB_REV/1.2.8/gcc-4.8.1-glibc-2.17/c3f970a/include"
-ZLIB_LIBS=" /mnt/gvfs/third-party2/zlib/$ZLIB_REV/1.2.8/gcc-4.8.1-glibc-2.17/c3f970a/lib/libz.a"
+ZLIB_INCLUDE=" -I $ZLIB_BASE/include"
+ZLIB_LIBS=" $ZLIB_BASE/lib/libz.a"

 # location of bzip headers and libraries
-BZIP_REV=d6c789bfc2ec4c51a63d66df2878926b8158cde8
-BZIP_INCLUDE=" -I /mnt/gvfs/third-party2/bzip2/$BZIP_REV/1.0.6/gcc-4.8.1-glibc-2.17/c3f970a/include/"
-BZIP_LIBS=" /mnt/gvfs/third-party2/bzip2/$BZIP_REV/1.0.6/gcc-4.8.1-glibc-2.17/c3f970a/lib/libbz2.a"
+BZIP2_INCLUDE=" -I $BZIP2_BASE/include/"
+BZIP2_LIBS=" $BZIP2_BASE/lib/libbz2.a"

-LZ4_REV=6858fac689e0f92e584224d91bdb0e39f6c8320d
-LZ4_INCLUDE=" -I /mnt/gvfs/third-party2/lz4/$LZ4_REV/r131/gcc-4.8.1-glibc-2.17/c3f970a/include"
-LZ4_LIBS=" /mnt/gvfs/third-party2/lz4/$LZ4_REV/r131/gcc-4.8.1-glibc-2.17/c3f970a/lib/liblz4.a"
+LZ4_INCLUDE=" -I $LZ4_BASE/include"
+LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"

-ZSTD_REV=810b81b4705def5243e998b54701f3c504e4009e
-ZSTD_INCLUDE=" -I /mnt/gvfs/third-party2/zstd/$ZSTD_REV/0.4.2/gcc-4.8.1-glibc-2.17/c3f970a/include"
-ZSTD_LIBS=" /mnt/gvfs/third-party2/zstd/$ZSTD_REV/0.4.2/gcc-4.8.1-glibc-2.17/c3f970a/lib/libzstd.a"
+ZSTD_INCLUDE=" -I $ZSTD_BASE/include"
+ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"

 # location of gflags headers and libraries
-GFLAGS_REV=c7275a4ceae0aca0929e56964a31dafc53c1ee96
-GFLAGS_INCLUDE=" -I /mnt/gvfs/third-party2/gflags/$GFLAGS_REV/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a/include/"
-GFLAGS_LIBS=" /mnt/gvfs/third-party2/gflags/$GFLAGS_REV/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a/lib/libgflags.a"
+GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
+GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags.a"

 # location of jemalloc
-JEMALLOC_REV=c370265e58c4b6602e798df23335a1e9913dae52
-JEMALLOC_INCLUDE=" -I /mnt/gvfs/third-party2/jemalloc/$JEMALLOC_REV/4.0.3/gcc-4.8.1-glibc-2.17/8d31e51/include"
-JEMALLOC_LIB="/mnt/gvfs/third-party2/jemalloc/$JEMALLOC_REV/4.0.3/gcc-4.8.1-glibc-2.17/8d31e51/lib/libjemalloc.a"
+JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include"
+JEMALLOC_LIB="$JEMALLOC_BASE/lib/libjemalloc.a"

 # location of numa
-NUMA_REV=ae54a5ed22cdabb1c6446dce4e8ffae5b4446d73
-NUMA_INCLUDE=" -I /mnt/gvfs/third-party2/numa/$NUMA_REV/2.0.8/gcc-4.8.1-glibc-2.17/c3f970a/include/"
-NUMA_LIB=" /mnt/gvfs/third-party2/numa/$NUMA_REV/2.0.8/gcc-4.8.1-glibc-2.17/c3f970a/lib/libnuma.a"
+NUMA_INCLUDE=" -I $NUMA_BASE/include/"
+NUMA_LIB=" $NUMA_BASE/lib/libnuma.a"

 # location of libunwind
-LIBUNWIND_REV=121f1a75c4414683aea8c70b761bfaf187f7c1a3
-LIBUNWIND="/mnt/gvfs/third-party2/libunwind/$LIBUNWIND_REV/trunk/gcc-4.8.1-glibc-2.17/675d945/lib/libunwind.a"
+LIBUNWIND="$LIBUNWIND_BASE/lib/libunwind.a"

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

-BINUTILS="/mnt/gvfs/third-party2/binutils/75670d0d8ef4891fd1ec2a7513ef01cd002c823b/2.25/centos6-native/da39a3e/bin"
+BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"

-DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE"
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP2_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE"

-GCC_BASE="/mnt/gvfs/third-party2/gcc/c0064002d2609ab649603f769f0bd110bbe48029/4.8.1/centos6-native/cc6c9dc"
 STDLIBS="-L $GCC_BASE/lib64"

 if [ -z "$USE_CLANG" ]; then
  # gcc
  CC="$GCC_BASE/bin/gcc"
  CXX="$GCC_BASE/bin/g++"
-  
+
  CFLAGS="-B$BINUTILS/gold -m64 -mtune=generic"
  CFLAGS+=" -isystem $GLIBC_INCLUDE"
  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  JEMALLOC=1
 else
-  # clang 
-  CLANG_BASE="/mnt/gvfs/third-party2/clang/ab054e9a490a8fd4537c0b6ec56e5c91c0f81c91/3.4"
-  CLANG_INCLUDE="$CLANG_BASE/gcc-4.8.1-glibc-2.17/fb0f730/lib/clang/3.4/include"
-  CC="$CLANG_BASE/centos6-native/9cefd8a/bin/clang"
-  CXX="$CLANG_BASE/centos6-native/9cefd8a/bin/clang++"
+  # clang
+  CLANG_BIN="$CLANG_BASE/bin"
+  CLANG_LIB="$CLANG_BASE/lib"
+  CLANG_INCLUDE="$CLANG_LIB/clang/*/include"
+  CC="$CLANG_BIN/clang"
+  CXX="$CLANG_BIN/clang++"

-  KERNEL_HEADERS_INCLUDE="/mnt/gvfs/third-party2/kernel-headers/1a48835975c66d30e47770ec419758ed3b9ba010/3.10.62-62_fbk17_03959_ge29cc63/gcc-4.8.1-glibc-2.17/da39a3e/include/"
+  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include/"

  CFLAGS="-B$BINUTILS/gold -nostdinc -nostdlib"
  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/4.8.1 "
@@ -100,16 +93,15 @@ CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PR
 CFLAGS+=" -DSNAPPY -DGFLAGS=google -DZLIB -DBZIP2 -DLZ4 -DZSTD -DNUMA"
 CXXFLAGS+=" $CFLAGS"

-EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB"
+EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP2_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB"
 EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/gcc-4.8.1-glibc-2.17/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
 EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/gcc-4.8.1-glibc-2.17/lib"

 PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"

-EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS"
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP2_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS"

-VALGRIND_REV=af85c56f424cd5edfc2c97588299b44ecdec96bb
-VALGRIND_VER="/mnt/gvfs/third-party2/valgrind/$VALGRIND_REV/3.8.1/gcc-4.8.1-glibc-2.17/c3f970a/bin/"
+VALGRIND_VER="$VALGRIND_BASE/bin/"

 export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE
@@ -1,5 +1,5 @@
 #!/bin/bash
-#  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+#  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.
@@ -10,7 +10,7 @@ then
  GIT="git"
 fi

-# Print out the colored progress info so that it can be brainlessly 
+# Print out the colored progress info so that it can be brainlessly
 # distinguished by users.
 function title() {
  echo -e "\033[1;32m$*\033[0m"
@@ -0,0 +1,198 @@
+#!/usr/local/fbcode/gcc-4.8.1-glibc-2.17-fb/bin/python2.7
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+import argparse
+import commands
+import subprocess
+import sys
+import re
+import os
+import time
+
+#
+# Simple logger
+#
+
+
+class Log:
+
+    LOG_FILE = "/tmp/precommit-check.log"
+
+    def __init__(self):
+        self.filename = Log.LOG_FILE
+        self.f = open(self.filename, 'w+', 0)
+
+    def caption(self, str):
+        line = "\n##### %s #####\n" % str
+        if self.f:
+            self.f.write("%s \n" % line)
+        else:
+            print(line)
+
+    def error(self, str):
+        data = "\n\n##### ERROR ##### %s" % str
+        if self.f:
+            self.f.write("%s \n" % data)
+        else:
+            print(data)
+
+    def log(self, str):
+        if self.f:
+            self.f.write("%s \n" % str)
+        else:
+            print(str)
+
+#
+# Shell Environment
+#
+
+
+class Env(object):
+
+    def __init__(self, tests):
+        self.tests = tests
+        self.log = Log()
+
+    def shell(self, cmd, path=os.getcwd()):
+        if path:
+            os.chdir(path)
+
+        self.log.log("==== shell session ===========================")
+        self.log.log("%s> %s" % (path, cmd))
+        status = subprocess.call("cd %s; %s" % (path, cmd), shell=True,
+                                 stdout=self.log.f, stderr=self.log.f)
+        self.log.log("status = %s" % status)
+        self.log.log("============================================== \n\n")
+        return status
+
+    def GetOutput(self, cmd, path=os.getcwd()):
+        if path:
+            os.chdir(path)
+
+        self.log.log("==== shell session ===========================")
+        self.log.log("%s> %s" % (path, cmd))
+        status, out = commands.getstatusoutput(cmd)
+        self.log.log("status = %s" % status)
+        self.log.log("out = %s" % out)
+        self.log.log("============================================== \n\n")
+        return status, out
+
+#
+# Pre-commit checker
+#
+
+
+class PreCommitChecker(Env):
+
+    def __init__(self, tests):
+        Env.__init__(self, tests)
+
+    #
+    #   Get commands for a given job from the determinator file
+    #
+    def get_commands(self, test):
+        status, out = self.GetOutput(
+            "build_tools/rocksdb-lego-determinator %s" % test, ".")
+        return status, out
+
+    #
+    # Run a specific CI job
+    #
+    def run_test(self, test):
+        self.log.caption("Running test %s locally" % test)
+
+        # get commands for the CI job determinator
+        status, cmds = self.get_commands(test)
+        if status != 0:
+            self.log.error("Error getting commands for test %s" % test)
+            return False
+
+        # Parse the JSON to extract the commands to run
+        cmds = re.findall("'shell':'([^\']*)'", cmds)
+
+        if len(cmds) == 0:
+            self.log.log("No commands found")
+            return False
+
+        # Run commands
+        for cmd in cmds:
+            # Replace J=<..> with the local environment variable
+            if "J" in os.environ:
+                cmd = cmd.replace("J=1", "J=%s" % os.environ["J"])
+                cmd = cmd.replace("make ", "make -j%s " % os.environ["J"])
+            # Run the command
+            status = self.shell(cmd, ".")
+            if status != 0:
+                self.log.error("Error running command %s for test %s"
+                               % (cmd, test))
+                return False
+
+        return True
+
+    #
+    # Run specified CI jobs
+    #
+    def run_tests(self):
+        if not self.tests:
+            self.log.error("Invalid args. Please provide tests")
+            return False
+
+        self.print_separator()
+        self.print_row("TEST", "RESULT")
+        self.print_separator()
+
+        for test in self.tests:
+            start_time = time.time()
+            self.print_test(test)
+            result = self.run_test(test)
+            elapsed_min = (time.time() - start_time) / 60
+            if not result:
+                self.log.error("Error running test %s" % test)
+                self.print_result("FAIL (%dm)" % elapsed_min)
+                return False
+            self.print_result("PASS (%dm)" % elapsed_min)
+
+        self.print_separator()
+        return True
+
+    #
+    # Print a line
+    #
+    def print_separator(self):
+        print("".ljust(60, "-"))
+
+    #
+    # Print two colums
+    #
+    def print_row(self, c0, c1):
+        print("%s%s" % (c0.ljust(40), c1.ljust(20)))
+
+    def print_test(self, test):
+        print(test.ljust(40), end="")
+        sys.stdout.flush()
+
+    def print_result(self, result):
+        print(result.ljust(20))
+
+#
+# Main
+#
+parser = argparse.ArgumentParser(description='RocksDB pre-commit checker.')
+
+# <test ....>
+parser.add_argument('test', nargs='+',
+                    help='CI test(s) to run. e.g: unit punit asan tsan ubsan')
+
+print("Please follow log %s" % Log.LOG_FILE)
+
+args = parser.parse_args()
+checker = PreCommitChecker(args.test)
+
+if not checker.run_tests():
+    print("Error running tests. Please check log file %s" % Log.LOG_FILE)
+    sys.exit(1)
+
+sys.exit(0)
@@ -243,7 +243,7 @@ make release
    --bloom_bits=10 \
    --num=$((NUM / 4)) \
    --reads=$((NUM / 4)) \
-    --writes_per_second=1000 \
+    --benchmark_write_rate_limit=$(( 110 * 1024 )) \
    --write_buffer_size=100000000 \
    --cache_size=6442450944 \
    --cache_numshardbits=6 \
@@ -329,7 +329,7 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --use_existing_db=1 \
    --duration=600 \
    --threads=32 \
-    --writes_per_second=81920 > ${STAT_FILE}.readwhilewriting_in_ram
+    --benchmark_write_rate_limit=9502720 > ${STAT_FILE}.readwhilewriting_in_ram

 # Seekrandomwhilewriting
 ./db_bench \
@@ -342,7 +342,7 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --use_tailing_iterator=1 \
    --duration=600 \
    --threads=32 \
-    --writes_per_second=81920 > ${STAT_FILE}.seekwhilewriting_in_ram
+    --benchmark_write_rate_limit=9502720 > ${STAT_FILE}.seekwhilewriting_in_ram

 # measure fillseq with bunch of column families
 ./db_bench \
@@ -65,10 +65,13 @@ SHM="TEST_TMPDIR=/dev/shm/rocksdb"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-LITE="OPT=-DROCKSDB_LITE"
+LITE="OPT=\"-DROCKSDB_LITE -g\""
 TSAN="COMPILE_WITH_TSAN=1"
+UBSAN="COMPILE_WITH_UBSAN=1"
 DISABLE_JEMALLOC="DISABLE_JEMALLOC=1"
-PARSER="'parser':'egrep \'Failure|^#|Abort\''"
+HTTP_PROXY="https_proxy=http://fwdproxy.29.prn1:8080 http_proxy=http://fwdproxy.29.prn1:8080 ftp_proxy=http://fwdproxy.29.prn1:8080"
+SETUP_JAVA_ENV="export $HTTP_PROXY; export JAVA_HOME=/usr/local/jdk-7u10-64/; export PATH=\$PATH:\$JAVA_HOME/bin"
+PARSER="'parser':'python build_tools/error_filter.py $1'"

 ARTIFACTS=" 'artifacts': [
    {
@@ -254,7 +257,7 @@ LITE_BUILD_COMMANDS="[
            $CLEANUP_ENV,
            {
                'name':'Build RocksDB debug version',
-                'shell':'$LITE $DEBUG make J=1 static_lib',
+                'shell':'$LITE make J=1 static_lib',
                'user':'root',
                $PARSER
            },
@@ -274,7 +277,7 @@ LITE_UNIT_TEST_COMMANDS="[
            $CLEANUP_ENV,
            {
                'name':'Build RocksDB debug version',
-                'shell':'$SHM $LITE $DEBUG make J=1 check',
+                'shell':'$SHM $LITE make J=1 check',
                'user':'root',
                $PARSER
            },
@@ -376,6 +379,48 @@ ASAN_CRASH_TEST_COMMANDS="[
    }
 ]"

+#
+# RocksDB test under undefined behavior sanitizer
+#
+UBSAN_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb Unit Test under UBSAN',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Test RocksDB debug under UBSAN',
+                'shell':'set -o pipefail && $SHM $UBSAN $DEBUG make J=1 ubsan_check',
+                'user':'root',
+                $PARSER
+            }
+        ],
+        $REPORT
+    }
+]"
+
+#
+# RocksDB crash testing under udnefined behavior sanitizer
+#
+UBSAN_CRASH_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb crash test under UBSAN',
+        'oncall':'$ONCALL',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build and run RocksDB debug ubsan_crash_test',
+                'timeout': 86400,
+                'shell':'$SHM $DEBUG make J=1 ubsan_crash_test',
+                'user':'root',
+                $PARSER
+            },
+        ],
+        $REPORT
+    }
+]"
+
 #
 # RocksDB unit test under valgrind
 #
@@ -462,9 +507,9 @@ run_format_compatible()
  if [ -e "build_detect_platform" ]
  then
    sed "s/tcmalloc/nothingnothingnothing/g" build_detect_platform > $TEST_TMPDIR/temp_build_file
-    rm -rf build_detect_platform 
+    rm -rf build_detect_platform
    cp $TEST_TMPDIR/temp_build_file build_detect_platform
-    chmod +x build_detect_platform 
+    chmod +x build_detect_platform
  fi

  make ldb -j32
@@ -592,6 +637,27 @@ REGRESSION_COMMANDS="[
    }
 ]"

+#
+# RocksDB Java build
+#
+JAVA_BUILD_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb Java Build',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build RocksDB for Java',
+                'shell':'$SETUP_JAVA_ENV; $SHM make rocksdbjava',
+                'user':'root',
+                $PARSER
+            },
+        ],
+        $REPORT
+    }
+]"
+
+
 case $1 in
  punit)
    echo $PARALLEL_UNIT_TEST_COMMANDS
@@ -632,6 +698,12 @@ case $1 in
  asan_crash)
    echo $ASAN_CRASH_TEST_COMMANDS
    ;;
+  ubsan)
+    echo $UBSAN_TEST_COMMANDS
+    ;;
+  ubsan_crash)
+    echo $UBSAN_CRASH_TEST_COMMANDS
+    ;;
  valgrind)
    echo $VALGRIND_TEST_COMMANDS
    ;;
@@ -659,6 +731,9 @@ case $1 in
  run_regression)
    run_regression
    ;;
+  java_build)
+    echo $JAVA_BUILD_TEST_COMMANDS
+    ;;
  *)
    echo "Invalid determinator command"
    ;;
@@ -1,5 +1,5 @@
 # This script enables you running RocksDB tests by running
-# All the tests in paralell and utilizing all the cores
+# All the tests in parallel and utilizing all the cores
 # For db_test the script first lists and parses the tests
 # and then fires them up in parallel using async PS Job functionality
 # Run the script from the enlistment
@@ -9,7 +9,7 @@ Param(
  [string]$WorkFolder = "",  # Direct tests to use that folder
  [int]$Limit = -1, # -1 means run all otherwise limit for testing purposes
  [string]$Exclude = "", # Expect a comma separated list, no spaces
-  [string]$Run = "db_test",  # Run db_test|tests|testname1,testname2...
+  [string]$Run = "db_test",  # Run db_test|db_test2|tests|testname1,testname2...
   # Number of async tasks that would run concurrently. Recommend a number below 64.
   # However, CPU utlization really depends on the storage media. Recommend ram based disk.
   # a value of 1 will run everything serially
@@ -46,15 +46,25 @@ if($WorkFolder -eq "") {
  $Env:TEST_TMPDIR = $WorkFolder
 }

+Write-Output "Root: $RootFolder, WorkFolder: $WorkFolder"
+
 # Use JEMALLOC executables
-if($EnableJE) {
-    $db_test = -Join ($BinariesFolder, "db_test_je.exe")
-} else {
-    $db_test = -Join ($BinariesFolder, "db_test.exe")
+if($Run -ceq "db_test" -or
+   $Run -ceq "db_test2" ) {
+
+   $file_name = $Run
+
+   if($EnableJE) {
+     $file_name += "_je"
+   }
+
+   $file_name += ".exe"
+
+   $db_test = -Join ($BinariesFolder, $file_name)
+
+   Write-Output "Binaries: $BinariesFolder db_test: $db_test"
 }

-Write-Output "Root: $RootFolder, WorkFolder: $WorkFolder"
-Write-Output "Binaries: $BinariesFolder db_test: $db_test"

 #Exclusions that we do not want to run
 $ExcludeTests = New-Object System.Collections.Generic.HashSet[string]
@@ -153,7 +163,8 @@ function Discover-TestBinaries([string]$Pattern, $HashTable) {

 $TestsToRun = [ordered]@{}

-if($Run -ceq "db_test") {
+if($Run -ceq "db_test" -or
+   $Run -ceq "db_test2") {
    Normalize-DbTests -HashTable $TestsToRun
 } elseif($Run -ceq "tests") {
    if($EnableJE) {
@@ -225,7 +236,8 @@ function RunJobs($TestToLog, [int]$ConcurrencyVal, [bool]$AddForRerun)
            Write-Host "Starting $k"
            $log_path = ($TestToLog.$k)

-            if($Run -ceq "db_test") {
+            if($Run -ceq "db_test" -or
+               $Run -ceq "db_test2") {
              $job = Start-Job -Name $k -ScriptBlock $InvokeTestCase -ArgumentList @($db_test,$k,$log_path)
            } else {
              [string]$Exe =  -Join ($BinariesFolder, $k)
@@ -258,7 +270,7 @@ function RunJobs($TestToLog, [int]$ConcurrencyVal, [bool]$AddForRerun)
            Write-Warning $message
            $log_content | Write-Warning
        } else {
-            # Scan the log. If we find PASSED and no occurence of FAILED
+            # Scan the log. If we find PASSED and no occurrence of FAILED
            # then it is a success
            [bool]$pass_found = $false
            ForEach($l in $log_content) {
@@ -315,4 +327,4 @@ if(!$success) {
    exit 12345
 }

- 
+ 
@@ -0,0 +1,127 @@
+#!/bin/sh
+#
+# Update dependencies.sh file with the latest avaliable versions
+
+BASEDIR=$(dirname $0)
+OUTPUT=""
+
+function log_variable()
+{
+  echo "$1=${!1}" >> "$OUTPUT"
+}
+
+
+TP2_LATEST="/mnt/vol/engshare/fbcode/third-party2"
+## $1 => lib name
+## $2 => lib version (if not provided, will try to pick latest)
+## $3 => platform (if not provided, will try to pick latest gcc)
+##
+## get_lib_base will set a variable named ${LIB_NAME}_BASE to the lib location
+function get_lib_base()
+{
+  local lib_name=$1
+  local lib_version=$2
+  local lib_platform=$3
+
+  local result="$TP2_LATEST/$lib_name/"
+  
+  # Lib Version
+  if [ -z "$lib_version" ] || [ "$lib_version" = "LATEST" ]; then
+    # version is not provided, use latest
+    result=`ls -dr1v $result/*/ | head -n1`
+  else
+    result="$result/$lib_version/"
+  fi
+  
+  # Lib Platform
+  if [ -z "$lib_platform" ]; then
+    # platform is not provided, use latest gcc
+    result=`ls -dr1v $result/gcc-*[^fb]/ | head -n1`
+  else
+    result="$result/$lib_platform/"
+  fi
+  
+  result=`ls -1d $result/*/ | head -n1`
+  
+  # lib_name => LIB_NAME_BASE
+  local __res_var=${lib_name^^}"_BASE"
+  __res_var=`echo $__res_var | tr - _`
+  # LIB_NAME_BASE=$result
+  eval $__res_var=`readlink -f $result`
+  
+  log_variable $__res_var
+}
+
+###########################################################
+#                   4.9.x dependencies                    #
+###########################################################
+
+OUTPUT="$BASEDIR/dependencies.sh"
+
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+echo "Writing dependencies to $OUTPUT"
+
+# Compilers locations
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/4.9.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 zlib       LATEST
+get_lib_base bzip2      LATEST
+get_lib_base lz4        LATEST
+get_lib_base zstd       LATEST
+get_lib_base gflags     LATEST
+get_lib_base jemalloc   LATEST
+get_lib_base numa       LATEST
+get_lib_base libunwind  LATEST
+
+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
+
+git diff $OUTPUT
+
+###########################################################
+#                   4.8.1 dependencies                    #
+###########################################################
+
+OUTPUT="$BASEDIR/dependencies_4.8.1.sh"
+
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+echo "Writing 4.8.1 dependencies to $OUTPUT"
+
+# Compilers locations
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/4.8.1/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     4.8.1  gcc-4.8.1-glibc-2.17
+get_lib_base glibc      2.17   gcc-4.8.1-glibc-2.17  
+get_lib_base snappy     LATEST gcc-4.8.1-glibc-2.17
+get_lib_base zlib       LATEST gcc-4.8.1-glibc-2.17
+get_lib_base bzip2      LATEST gcc-4.8.1-glibc-2.17
+get_lib_base lz4        LATEST gcc-4.8.1-glibc-2.17
+get_lib_base zstd       LATEST gcc-4.8.1-glibc-2.17
+get_lib_base gflags     LATEST gcc-4.8.1-glibc-2.17
+get_lib_base jemalloc   LATEST gcc-4.8.1-glibc-2.17
+get_lib_base numa       LATEST gcc-4.8.1-glibc-2.17
+get_lib_base libunwind  LATEST gcc-4.8.1-glibc-2.17
+
+get_lib_base kernel-headers LATEST gcc-4.8.1-glibc-2.17 
+get_lib_base binutils   LATEST centos6-native 
+get_lib_base valgrind   3.8.1  gcc-4.8.1-glibc-2.17
+
+git diff $OUTPUT
@@ -1,18 +1,18 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
 //
-#include "util/auto_roll_logger.h"
+#include "db/auto_roll_logger.h"
 #include "util/mutexlock.h"

-using namespace std;
-
 namespace rocksdb {

 // -- AutoRollLogger
 Status AutoRollLogger::ResetLogger() {
+  TEST_SYNC_POINT("AutoRollLogger::ResetLogger:BeforeNewLogger");
  status_ = env_->NewLogger(log_fname_, &logger_);
+  TEST_SYNC_POINT("AutoRollLogger::ResetLogger:AfterNewLogger");

  if (!status_.ok()) {
    return status_;
@@ -32,12 +32,21 @@ Status AutoRollLogger::ResetLogger() {
 }

 void AutoRollLogger::RollLogFile() {
-  std::string old_fname = OldInfoLogFileName(
-      dbname_, env_->NowMicros(), db_absolute_path_, db_log_dir_);
+  // This function is called when log is rotating. Two rotations
+  // can happen quickly (NowMicro returns same value). To not overwrite
+  // previous log file we increment by one micro second and try again.
+  uint64_t now = env_->NowMicros();
+  std::string old_fname;
+  do {
+    old_fname = OldInfoLogFileName(
+      dbname_, now, db_absolute_path_, db_log_dir_);
+    now++;
+  } while (env_->FileExists(old_fname).ok());
  env_->RenameFile(log_fname_, old_fname);
 }

-string AutoRollLogger::ValistToString(const char* format, va_list args) const {
+std::string AutoRollLogger::ValistToString(const char* format,
+                                           va_list args) const {
  // Any log messages longer than 1024 will get truncated.
  // The user is responsible for chopping longer messages into multi line log
  static const int MAXBUFFERSIZE = 1024;
@@ -102,7 +111,7 @@ void AutoRollLogger::LogHeader(const char* format, va_list args) {
  // strings
  va_list tmp;
  va_copy(tmp, args);
-  string data = ValistToString(format, tmp);
+  std::string data = ValistToString(format, tmp);
  va_end(tmp);

  MutexLock l(&mutex_);
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -13,6 +13,8 @@
 #include "db/filename.h"
 #include "port/port.h"
 #include "port/util_logger.h"
+#include "util/sync_point.h"
+#include "util/mutexlock.h"

 namespace rocksdb {

@@ -53,11 +55,26 @@ class AutoRollLogger : public Logger {
    return status_;
  }

-  size_t GetLogFileSize() const override { return logger_->GetLogFileSize(); }
+  size_t GetLogFileSize() const override {
+    std::shared_ptr<Logger> logger;
+    {
+      MutexLock l(&mutex_);
+      // pin down the current logger_ instance before releasing the mutex.
+      logger = logger_;
+    }
+    return logger->GetLogFileSize();
+  }

  void Flush() override {
-    if (logger_) {
-      logger_->Flush();
+    std::shared_ptr<Logger> logger;
+    {
+      MutexLock l(&mutex_);
+      // pin down the current logger_ instance before releasing the mutex.
+      logger = logger_;
+    }
+    TEST_SYNC_POINT("AutoRollLogger::Flush:PinnedLogger");
+    if (logger) {
+      logger->Flush();
    }
  }

@@ -101,7 +118,7 @@ class AutoRollLogger : public Logger {
  uint64_t ctime_;
  uint64_t cached_now_access_count;
  uint64_t call_NowMicros_every_N_records_;
-  port::Mutex mutex_;
+  mutable port::Mutex mutex_;
 };

 // Facade to craete logger automatically
@@ -1,23 +1,24 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
 //
 #include <string>
+#include <thread>
 #include <vector>
 #include <cmath>
 #include <iostream>
 #include <fstream>
 #include <iterator>
 #include <algorithm>
+#include "db/auto_roll_logger.h"
+#include "port/port.h"
+#include "util/sync_point.h"
 #include "util/testharness.h"
-#include "util/auto_roll_logger.h"
 #include "rocksdb/db.h"
 #include <sys/stat.h>
 #include <errno.h>

-using namespace std;
-
 namespace rocksdb {

 class AutoRollLoggerTest : public testing::Test {
@@ -37,23 +38,22 @@ class AutoRollLoggerTest : public testing::Test {
    Env::Default()->CreateDir(kTestDir);
  }

-  void RollLogFileBySizeTest(AutoRollLogger* logger,
-                             size_t log_max_size,
-                             const string& log_message);
-  uint64_t RollLogFileByTimeTest(AutoRollLogger* logger,
-                                 size_t time,
-                                 const string& log_message);
+  void RollLogFileBySizeTest(AutoRollLogger* logger, size_t log_max_size,
+                             const std::string& log_message);
+  uint64_t RollLogFileByTimeTest(AutoRollLogger* logger, size_t time,
+                                 const std::string& log_message);

-  static const string kSampleMessage;
-  static const string kTestDir;
-  static const string kLogFile;
+  static const std::string kSampleMessage;
+  static const std::string kTestDir;
+  static const std::string kLogFile;
  static Env* env;
 };

-const string AutoRollLoggerTest::kSampleMessage(
+const std::string AutoRollLoggerTest::kSampleMessage(
    "this is the message to be written to the log file!!");
-const string AutoRollLoggerTest::kTestDir(test::TmpDir() + "/db_log_test");
-const string AutoRollLoggerTest::kLogFile(test::TmpDir() + "/db_log_test/LOG");
+const std::string AutoRollLoggerTest::kTestDir(test::TmpDir() + "/db_log_test");
+const std::string AutoRollLoggerTest::kLogFile(test::TmpDir() +
+                                               "/db_log_test/LOG");
 Env* AutoRollLoggerTest::env = Env::Default();

 // In this test we only want to Log some simple log message with
@@ -83,7 +83,7 @@ void GetFileCreateTime(const std::string& fname, uint64_t* file_ctime) {

 void AutoRollLoggerTest::RollLogFileBySizeTest(AutoRollLogger* logger,
                                               size_t log_max_size,
-                                               const string& log_message) {
+                                               const std::string& log_message) {
  logger->SetInfoLogLevel(InfoLogLevel::INFO_LEVEL);
  // measure the size of each message, which is supposed
  // to be equal or greater than log_message.size()
@@ -108,7 +108,7 @@ void AutoRollLoggerTest::RollLogFileBySizeTest(AutoRollLogger* logger,
 }

 uint64_t AutoRollLoggerTest::RollLogFileByTimeTest(
-    AutoRollLogger* logger, size_t time, const string& log_message) {
+    AutoRollLogger* logger, size_t time, const std::string& log_message) {
  uint64_t expected_create_time;
  uint64_t actual_create_time;
  uint64_t total_log_size;
@@ -260,7 +260,60 @@ TEST_F(AutoRollLoggerTest, CreateLoggerFromOptions) {
      auto_roll_logger, options.log_file_time_to_roll,
      kSampleMessage + ":CreateLoggerFromOptions - both");
 }
-#endif
+
+TEST_F(AutoRollLoggerTest, LogFlushWhileRolling) {
+  DBOptions options;
+  shared_ptr<Logger> logger;
+
+  InitTestDb();
+  options.max_log_file_size = 1024 * 5;
+  ASSERT_OK(CreateLoggerFromOptions(kTestDir, options, &logger));
+  AutoRollLogger* auto_roll_logger =
+      dynamic_cast<AutoRollLogger*>(logger.get());
+  ASSERT_TRUE(auto_roll_logger);
+  std::thread flush_thread;
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      // Need to pin the old logger before beginning the roll, as rolling grabs
+      // the mutex, which would prevent us from accessing the old logger.
+      {"AutoRollLogger::Flush:PinnedLogger",
+       "AutoRollLoggerTest::LogFlushWhileRolling:PreRollAndPostThreadInit"},
+      // Need to finish the flush thread init before this callback because the
+      // callback accesses flush_thread.get_id() in order to apply certain sync
+      // points only to the flush thread.
+      {"AutoRollLoggerTest::LogFlushWhileRolling:PreRollAndPostThreadInit",
+       "AutoRollLoggerTest::LogFlushWhileRolling:FlushCallbackBegin"},
+      // Need to reset logger at this point in Flush() to exercise a race
+      // condition case, which is executing the flush with the pinned (old)
+      // logger after the roll has cut over to a new logger.
+      {"AutoRollLoggerTest::LogFlushWhileRolling:FlushCallback1",
+       "AutoRollLogger::ResetLogger:BeforeNewLogger"},
+      {"AutoRollLogger::ResetLogger:AfterNewLogger",
+       "AutoRollLoggerTest::LogFlushWhileRolling:FlushCallback2"},
+  });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "PosixLogger::Flush:BeginCallback", [&](void* arg) {
+        TEST_SYNC_POINT(
+            "AutoRollLoggerTest::LogFlushWhileRolling:FlushCallbackBegin");
+        if (std::this_thread::get_id() == flush_thread.get_id()) {
+          TEST_SYNC_POINT(
+              "AutoRollLoggerTest::LogFlushWhileRolling:FlushCallback1");
+          TEST_SYNC_POINT(
+              "AutoRollLoggerTest::LogFlushWhileRolling:FlushCallback2");
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  flush_thread = std::thread([&]() { auto_roll_logger->Flush(); });
+  TEST_SYNC_POINT(
+      "AutoRollLoggerTest::LogFlushWhileRolling:PreRollAndPostThreadInit");
+  RollLogFileBySizeTest(auto_roll_logger, options.max_log_file_size,
+                        kSampleMessage + ":LogFlushWhileRolling");
+  flush_thread.join();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+#endif  // OS_WIN

 TEST_F(AutoRollLoggerTest, InfoLogLevel) {
  InitTestDb();
@@ -305,13 +358,13 @@ TEST_F(AutoRollLoggerTest, InfoLogLevel) {

 // Test the logger Header function for roll over logs
 // We expect the new logs creates as roll over to carry the headers specified
-static std::vector<string> GetOldFileNames(const string& path) {
-  std::vector<string> ret;
+static std::vector<std::string> GetOldFileNames(const std::string& path) {
+  std::vector<std::string> ret;

-  const string dirname = path.substr(/*start=*/ 0, path.find_last_of("/"));
-  const string fname = path.substr(path.find_last_of("/") + 1);
+  const std::string dirname = path.substr(/*start=*/0, path.find_last_of("/"));
+  const std::string fname = path.substr(path.find_last_of("/") + 1);

-  std::vector<string> children;
+  std::vector<std::string> children;
  Env::Default()->GetChildren(dirname, &children);

  // We know that the old log files are named [path]<something>
@@ -326,12 +379,13 @@ static std::vector<string> GetOldFileNames(const string& path) {
 }

 // Return the number of lines where a given pattern was found in the file
-static size_t GetLinesCount(const string& fname, const string& pattern) {
-  stringstream ssbuf;
-  string line;
+static size_t GetLinesCount(const std::string& fname,
+                            const std::string& pattern) {
+  std::stringstream ssbuf;
+  std::string line;
  size_t count = 0;

-  ifstream inFile(fname.c_str());
+  std::ifstream inFile(fname.c_str());
  ssbuf << inFile.rdbuf();

  while (getline(ssbuf, line)) {
@@ -370,7 +424,7 @@ TEST_F(AutoRollLoggerTest, LogHeaderTest) {
      }
    }

-    const string newfname = logger.TEST_log_fname();
+    const std::string newfname = logger.TEST_log_fname();

    // Log enough data to cause a roll over
    int i = 0;
@@ -402,7 +456,16 @@ TEST_F(AutoRollLoggerTest, LogHeaderTest) {
 TEST_F(AutoRollLoggerTest, LogFileExistence) {
  rocksdb::DB* db;
  rocksdb::Options options;
-  string deleteCmd = "rm -rf " + kTestDir;
+#ifdef OS_WIN
+  // Replace all slashes in the path so windows CompSpec does not
+  // become confused
+  std::string testDir(kTestDir);
+  std::replace_if(testDir.begin(), testDir.end(),
+    [](char ch) { return ch == '/'; }, '\\');
+  std::string deleteCmd = "if exist " + testDir + " rd /s /q " + testDir;
+#else
+  std::string deleteCmd = "rm -rf " + kTestDir;
+#endif
  ASSERT_EQ(system(deleteCmd.c_str()), 0);
  options.max_log_file_size = 100 * 1024 * 1024;
  options.create_if_missing = true;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -15,6 +15,7 @@

 #include "db/compaction_iterator.h"
 #include "db/dbformat.h"
+#include "db/event_helpers.h"
 #include "db/filename.h"
 #include "db/internal_stats.h"
 #include "db/merge_helper.h"
@@ -41,28 +42,39 @@ TableBuilder* NewTableBuilder(
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
-    uint32_t column_family_id, WritableFileWriter* file,
-    const CompressionType compression_type,
-    const CompressionOptions& compression_opts, const bool skip_filters) {
+    uint32_t column_family_id, const std::string& column_family_name,
+    WritableFileWriter* file, const CompressionType compression_type,
+    const CompressionOptions& compression_opts,
+    const std::string* compression_dict, const bool skip_filters) {
+  assert((column_family_id ==
+          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
+         column_family_name.empty());
  return ioptions.table_factory->NewTableBuilder(
      TableBuilderOptions(ioptions, internal_comparator,
                          int_tbl_prop_collector_factories, compression_type,
-                          compression_opts, skip_filters),
+                          compression_opts, compression_dict, skip_filters,
+                          column_family_name),
      column_family_id, file);
 }

 Status BuildTable(
    const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
-    const EnvOptions& env_options, TableCache* table_cache,
-    InternalIterator* iter, FileMetaData* meta,
+    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    TableCache* table_cache, InternalIterator* iter, FileMetaData* meta,
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
-    uint32_t column_family_id, std::vector<SequenceNumber> snapshots,
+    uint32_t column_family_id, const std::string& column_family_name,
+    std::vector<SequenceNumber> snapshots,
+    SequenceNumber earliest_write_conflict_snapshot,
    const CompressionType compression,
    const CompressionOptions& compression_opts, bool paranoid_file_checks,
-    InternalStats* internal_stats, const Env::IOPriority io_priority,
-    TableProperties* table_properties) {
+    InternalStats* internal_stats, TableFileCreationReason reason,
+    EventLogger* event_logger, int job_id, const Env::IOPriority io_priority,
+    TableProperties* table_properties, int level) {
+  assert((column_family_id ==
+          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
+         column_family_name.empty());
  // Reports the IOStats for flush for every following bytes.
  const size_t kReportFlushIOStatsEvery = 1048576;
  Status s;
@@ -71,6 +83,12 @@ Status BuildTable(

  std::string fname = TableFileName(ioptions.db_paths, meta->fd.GetNumber(),
                                    meta->fd.GetPathId());
+#ifndef ROCKSDB_LITE
+  EventHelpers::NotifyTableFileCreationStarted(
+      ioptions.listeners, dbname, column_family_name, fname, job_id, reason);
+#endif  // !ROCKSDB_LITE
+  TableProperties tp;
+
  if (iter->Valid()) {
    TableBuilder* builder;
    unique_ptr<WritableFileWriter> file_writer;
@@ -78,6 +96,9 @@ Status BuildTable(
      unique_ptr<WritableFile> file;
      s = NewWritableFile(env, fname, &file, env_options);
      if (!s.ok()) {
+        EventHelpers::LogAndNotifyTableFileCreationFinished(
+            event_logger, ioptions.listeners, dbname, column_family_name, fname,
+            job_id, meta->fd, tp, reason, s);
        return s;
      }
      file->SetIOPriority(io_priority);
@@ -86,17 +107,19 @@ Status BuildTable(

      builder = NewTableBuilder(
          ioptions, internal_comparator, int_tbl_prop_collector_factories,
-          column_family_id, file_writer.get(), compression, compression_opts);
+          column_family_id, column_family_name, file_writer.get(), compression,
+          compression_opts);
    }

    MergeHelper merge(env, internal_comparator.user_comparator(),
                      ioptions.merge_operator, nullptr, ioptions.info_log,
-                      ioptions.min_partial_merge_operands,
+                      mutable_cf_options.min_partial_merge_operands,
                      true /* internal key corruption is not ok */,
                      snapshots.empty() ? 0 : snapshots.back());

    CompactionIterator c_iter(iter, internal_comparator.user_comparator(),
-                              &merge, kMaxSequenceNumber, &snapshots, env,
+                              &merge, kMaxSequenceNumber, &snapshots,
+                              earliest_write_conflict_snapshot, env,
                              true /* internal key corruption is not ok */);
    c_iter.SeekToFirst();
    for (; c_iter.Valid(); c_iter.Next()) {
@@ -123,11 +146,13 @@ Status BuildTable(
    }

    if (s.ok() && !empty) {
-      meta->fd.file_size = builder->FileSize();
+      uint64_t file_size = builder->FileSize();
+      meta->fd.file_size = file_size;
      meta->marked_for_compaction = builder->NeedCompact();
      assert(meta->fd.GetFileSize() > 0);
+      tp = builder->GetTableProperties();
      if (table_properties) {
-        *table_properties = builder->GetTableProperties();
+        *table_properties = tp;
      }
    }
    delete builder;
@@ -147,7 +172,8 @@ Status BuildTable(
          ReadOptions(), env_options, internal_comparator, meta->fd, nullptr,
          (internal_stats == nullptr) ? nullptr
                                      : internal_stats->GetFileReadHist(0),
-          false));
+          false /* for_compaction */, nullptr /* arena */,
+          false /* skip_filter */, level));
      s = it->status();
      if (s.ok() && paranoid_file_checks) {
        for (it->SeekToFirst(); it->Valid(); it->Next()) {
@@ -165,6 +191,12 @@ Status BuildTable(
  if (!s.ok() || meta->fd.GetFileSize() == 0) {
    env->DeleteFile(fname);
  }
+
+  // Output to event logger and fire events.
+  EventHelpers::LogAndNotifyTableFileCreationFinished(
+      event_logger, ioptions.listeners, dbname, column_family_name, fname,
+      job_id, meta->fd, tp, reason, s);
+
  return s;
 }

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -12,11 +12,14 @@
 #include "db/table_properties_collector.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
-#include "rocksdb/status.h"
-#include "rocksdb/types.h"
-#include "rocksdb/options.h"
 #include "rocksdb/immutable_options.h"
+#include "rocksdb/listener.h"
+#include "rocksdb/options.h"
+#include "rocksdb/status.h"
 #include "rocksdb/table_properties.h"
+#include "rocksdb/types.h"
+#include "util/event_logger.h"
+#include "util/mutable_cf_options.h"

 namespace rocksdb {

@@ -33,14 +36,20 @@ class WritableFileWriter;
 class InternalStats;
 class InternalIterator;

+// @param column_family_name Name of the column family that is also identified
+//    by column_family_id, or empty string if unknown. It must outlive the
+//    TableBuilder returned by this function.
+// @param compression_dict Data for presetting the compression library's
+//    dictionary, or nullptr.
 TableBuilder* NewTableBuilder(
    const ImmutableCFOptions& options,
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
-    uint32_t column_family_id, WritableFileWriter* file,
-    const CompressionType compression_type,
+    uint32_t column_family_id, const std::string& column_family_name,
+    WritableFileWriter* file, const CompressionType compression_type,
    const CompressionOptions& compression_opts,
+    const std::string* compression_dict = nullptr,
    const bool skip_filters = false);

 // Build a Table file from the contents of *iter.  The generated file
@@ -48,18 +57,24 @@ TableBuilder* NewTableBuilder(
 // *meta will be filled with metadata about the generated table.
 // If no data is present in *iter, meta->file_size will be set to
 // zero, and no Table file will be produced.
+//
+// @param column_family_name Name of the column family that is also identified
+//    by column_family_id, or empty string if unknown.
 extern Status BuildTable(
    const std::string& dbname, Env* env, const ImmutableCFOptions& options,
-    const EnvOptions& env_options, TableCache* table_cache,
-    InternalIterator* iter, FileMetaData* meta,
+    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    TableCache* table_cache, InternalIterator* iter, FileMetaData* meta,
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
-    uint32_t column_family_id, std::vector<SequenceNumber> snapshots,
+    uint32_t column_family_id, const std::string& column_family_name,
+    std::vector<SequenceNumber> snapshots,
+    SequenceNumber earliest_write_conflict_snapshot,
    const CompressionType compression,
    const CompressionOptions& compression_opts, bool paranoid_file_checks,
-    InternalStats* internal_stats,
+    InternalStats* internal_stats, TableFileCreationReason reason,
+    EventLogger* event_logger = nullptr, int job_id = 0,
    const Env::IOPriority io_priority = Env::IO_HIGH,
-    TableProperties* table_properties = nullptr);
+    TableProperties* table_properties = nullptr, int level = -1);

 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -446,6 +446,12 @@ void rocksdb_backup_engine_create_new_backup(rocksdb_backup_engine_t* be,
  SaveError(errptr, be->rep->CreateNewBackup(db->rep));
 }

+void rocksdb_backup_engine_purge_old_backups(rocksdb_backup_engine_t* be,
+                                             uint32_t num_backups_to_keep,
+                                             char** errptr) {
+  SaveError(errptr, be->rep->PurgeOldBackups(num_backups_to_keep));
+}
+
 rocksdb_restore_options_t* rocksdb_restore_options_create() {
  return new rocksdb_restore_options_t;
 }
@@ -825,6 +831,31 @@ rocksdb_iterator_t* rocksdb_create_iterator_cf(
  return result;
 }

+void rocksdb_create_iterators(
+    rocksdb_t *db,
+    rocksdb_readoptions_t* opts,
+    rocksdb_column_family_handle_t** column_families,
+    rocksdb_iterator_t** iterators,
+    size_t size,
+    char** errptr) {
+  std::vector<ColumnFamilyHandle*> column_families_vec;
+  for (size_t i = 0; i < size; i++) {
+    column_families_vec.push_back(column_families[i]->rep);
+  }
+
+  std::vector<Iterator*> res;
+  Status status = db->rep->NewIterators(opts->rep, column_families_vec, &res);
+  assert(res.size() == size);
+  if (SaveError(errptr, status)) {
+    return;
+  }
+
+  for (size_t i = 0; i < size; i++) {
+    iterators[i] = new rocksdb_iterator_t;
+    iterators[i]->rep = res[i];
+  }
+}
+
 const rocksdb_snapshot_t* rocksdb_create_snapshot(
    rocksdb_t* db) {
  rocksdb_snapshot_t* result = new rocksdb_snapshot_t;
@@ -1288,6 +1319,11 @@ void rocksdb_block_based_options_set_cache_index_and_filter_blocks(
  options->rep.cache_index_and_filter_blocks = v;
 }

+void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
+}
+
 void rocksdb_block_based_options_set_skip_table_builder_flush(
    rocksdb_block_based_table_options_t* options, unsigned char v) {
  options->rep.skip_table_builder_flush = v;
@@ -1388,6 +1424,11 @@ void rocksdb_options_set_compaction_filter_factory(
      std::shared_ptr<CompactionFilterFactory>(factory);
 }

+void rocksdb_options_compaction_readahead_size(
+    rocksdb_options_t* opt, size_t s) {
+  opt->rep.compaction_readahead_size = s;
+}
+
 void rocksdb_options_set_comparator(
    rocksdb_options_t* opt,
    rocksdb_comparator_t* cmp) {
@@ -1531,11 +1572,13 @@ void rocksdb_options_set_compression_per_level(rocksdb_options_t* opt,
  }
 }

-void rocksdb_options_set_compression_options(
-    rocksdb_options_t* opt, int w_bits, int level, int strategy) {
+void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
+                                             int level, int strategy,
+                                             int max_dict_bytes) {
  opt->rep.compression_opts.window_bits = w_bits;
  opt->rep.compression_opts.level = level;
  opt->rep.compression_opts.strategy = strategy;
+  opt->rep.compression_opts.max_dict_bytes = max_dict_bytes;
 }

 void rocksdb_options_set_prefix_extractor(
@@ -1766,6 +1809,11 @@ void rocksdb_options_set_memtable_prefix_bloom_probes(
  opt->rep.memtable_prefix_bloom_probes = v;
 }

+void rocksdb_options_set_memtable_prefix_bloom_huge_page_tlb_size(
+    rocksdb_options_t* opt, size_t v) {
+  opt->rep.memtable_prefix_bloom_huge_page_tlb_size = v;
+}
+
 void rocksdb_options_set_hash_skip_list_rep(
    rocksdb_options_t *opt, size_t bucket_count,
    int32_t skiplist_height, int32_t skiplist_branching_factor) {
@@ -1827,6 +1875,11 @@ void rocksdb_options_set_inplace_update_num_locks(
  opt->rep.inplace_update_num_locks = v;
 }

+void rocksdb_options_set_report_bg_io_stats(
+    rocksdb_options_t* opt, int v) {
+  opt->rep.report_bg_io_stats = v;
+}
+
 void rocksdb_options_set_compaction_style(rocksdb_options_t *opt, int style) {
  opt->rep.compaction_style = static_cast<rocksdb::CompactionStyle>(style);
 }
@@ -1965,7 +2018,7 @@ void rocksdb_filterpolicy_destroy(rocksdb_filterpolicy_t* filter) {
  delete filter;
 }

-rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(int bits_per_key) {
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_format(int bits_per_key, bool original_format) {
  // Make a rocksdb_filterpolicy_t, but override all of its methods so
  // they delegate to a NewBloomFilterPolicy() instead of user
  // supplied C functions.
@@ -1983,13 +2036,21 @@ rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(int bits_per_key) {
    static void DoNothing(void*) { }
  };
  Wrapper* wrapper = new Wrapper;
-  wrapper->rep_ = NewBloomFilterPolicy(bits_per_key);
+  wrapper->rep_ = NewBloomFilterPolicy(bits_per_key, original_format);
  wrapper->state_ = nullptr;
  wrapper->delete_filter_ = nullptr;
  wrapper->destructor_ = &Wrapper::DoNothing;
  return wrapper;
 }

+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_full(int bits_per_key) {
+  return rocksdb_filterpolicy_create_bloom_format(bits_per_key, false);
+}
+
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(int bits_per_key) {
+  return rocksdb_filterpolicy_create_bloom_format(bits_per_key, true);
+}
+
 rocksdb_mergeoperator_t* rocksdb_mergeoperator_create(
    void* state, void (*destructor)(void*),
    char* (*full_merge)(void*, const char* key, size_t key_length,
@@ -2066,6 +2127,11 @@ void rocksdb_readoptions_set_tailing(
  opt->rep.tailing = v;
 }

+void rocksdb_readoptions_set_readahead_size(
+    rocksdb_readoptions_t* opt, size_t v) {
+  opt->rep.readahead_size = v;
+}
+
 rocksdb_writeoptions_t* rocksdb_writeoptions_create() {
  return new rocksdb_writeoptions_t;
 }
@@ -2107,6 +2173,10 @@ void rocksdb_cache_destroy(rocksdb_cache_t* cache) {
  delete cache;
 }

+void rocksdb_cache_set_capacity(rocksdb_cache_t* cache, size_t capacity) {
+  cache->rep->SetCapacity(capacity);
+}
+
 rocksdb_env_t* rocksdb_create_default_env() {
  rocksdb_env_t* result = new rocksdb_env_t;
  result->rep = Env::Default();
@@ -2114,6 +2184,13 @@ rocksdb_env_t* rocksdb_create_default_env() {
  return result;
 }

+rocksdb_env_t* rocksdb_create_mem_env() {
+  rocksdb_env_t* result = new rocksdb_env_t;
+  result->rep = rocksdb::NewMemEnv(Env::Default());
+  result->is_default = false;
+  return result;
+}
+
 void rocksdb_env_set_background_threads(rocksdb_env_t* env, int n) {
  env->rep->SetBackgroundThreads(n);
 }
@@ -2325,6 +2402,31 @@ void rocksdb_get_options_from_string(const rocksdb_options_t* base_options,
                                 &new_options->rep));
 }

+void rocksdb_delete_file_in_range(rocksdb_t* db, const char* start_key,
+                                  size_t start_key_len, const char* limit_key,
+                                  size_t limit_key_len, char** errptr) {
+  Slice a, b;
+  SaveError(
+      errptr,
+      DeleteFilesInRange(
+          db->rep, db->rep->DefaultColumnFamily(),
+          (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+          (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr)));
+}
+
+void rocksdb_delete_file_in_range_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* column_family,
+    const char* start_key, size_t start_key_len, const char* limit_key,
+    size_t limit_key_len, char** errptr) {
+  Slice a, b;
+  SaveError(
+      errptr,
+      DeleteFilesInRange(
+          db->rep, column_family->rep,
+          (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+          (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr)));
+}
+
 void rocksdb_free(void* ptr) { free(ptr); }

 }  // end extern "C"
@@ -322,7 +322,7 @@ int main(int argc, char** argv) {
  rocksdb_options_set_block_based_table_factory(options, table_options);

  rocksdb_options_set_compression(options, rocksdb_no_compression);
-  rocksdb_options_set_compression_options(options, -14, -1, 0);
+  rocksdb_options_set_compression_options(options, -14, -1, 0, 0);
  int compression_levels[] = {rocksdb_no_compression, rocksdb_no_compression,
                              rocksdb_no_compression, rocksdb_no_compression};
  rocksdb_options_set_compression_per_level(options, compression_levels, 4);
@@ -365,6 +365,24 @@ int main(int argc, char** argv) {
    rocksdb_backup_engine_create_new_backup(be, db, &err);
    CheckNoError(err);

+    // need a change to trigger a new backup
+    rocksdb_delete(db, woptions, "does-not-exist", 14, &err);
+    CheckNoError(err);
+
+    rocksdb_backup_engine_create_new_backup(be, db, &err);
+    CheckNoError(err);
+
+    const rocksdb_backup_engine_info_t* bei = rocksdb_backup_engine_get_backup_info(be);
+    CheckCondition(rocksdb_backup_engine_info_count(bei) > 1);
+    rocksdb_backup_engine_info_destroy(bei);
+
+    rocksdb_backup_engine_purge_old_backups(be, 1, &err);
+    CheckNoError(err);
+
+    bei = rocksdb_backup_engine_get_backup_info(be);
+    CheckCondition(rocksdb_backup_engine_info_count(bei) == 1);
+    rocksdb_backup_engine_info_destroy(bei);
+
    rocksdb_delete(db, woptions, "foo", 3, &err);
    CheckNoError(err);

@@ -768,6 +786,30 @@ int main(int argc, char** argv) {
    CheckNoError(err);
    rocksdb_iter_destroy(iter);

+    rocksdb_column_family_handle_t* iters_cf_handles[2] = { handles[0], handles[1] };
+    rocksdb_iterator_t* iters_handles[2];
+    rocksdb_create_iterators(db, roptions, iters_cf_handles, iters_handles, 2, &err);
+    CheckNoError(err);
+
+    iter = iters_handles[0];
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_seek_to_first(iter);
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_destroy(iter);
+
+    iter = iters_handles[1];
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_seek_to_first(iter);
+    CheckCondition(rocksdb_iter_valid(iter));
+
+    for (i = 0; rocksdb_iter_valid(iter) != 0; rocksdb_iter_next(iter)) {
+      i++;
+    }
+    CheckCondition(i == 1);
+    rocksdb_iter_get_error(iter, &err);
+    CheckNoError(err);
+    rocksdb_iter_destroy(iter);
+
    rocksdb_drop_column_family(db, handles[1], &err);
    CheckNoError(err);
    for (i = 0; i < 2; i++) {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -68,6 +68,20 @@ const std::string& ColumnFamilyHandleImpl::GetName() const {
  return cfd()->GetName();
 }

+Status ColumnFamilyHandleImpl::GetDescriptor(ColumnFamilyDescriptor* desc) {
+#ifndef ROCKSDB_LITE
+  // accessing mutable cf-options requires db mutex.
+  InstrumentedMutexLock l(mutex_);
+  *desc = ColumnFamilyDescriptor(
+      cfd()->GetName(),
+      BuildColumnFamilyOptions(*cfd()->options(),
+                               *cfd()->GetLatestMutableCFOptions()));
+  return Status::OK();
+#else
+  return Status::NotSupported();
+#endif  // !ROCKSDB_LITE
+}
+
 const Comparator* ColumnFamilyHandleImpl::user_comparator() const {
  return cfd()->user_comparator();
 }
@@ -110,18 +124,33 @@ Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
  return Status::OK();
 }

+Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
+  if (cf_options.inplace_update_support) {
+    return Status::InvalidArgument(
+        "In-place memtable updates (inplace_update_support) is not compatible "
+        "with concurrent writes (allow_concurrent_memtable_write)");
+  }
+  if (cf_options.filter_deletes) {
+    return Status::InvalidArgument(
+        "Delete filtering (filter_deletes) is not compatible with concurrent "
+        "memtable writes (allow_concurrent_memtable_writes)");
+  }
+  if (!cf_options.memtable_factory->IsInsertConcurrentlySupported()) {
+    return Status::InvalidArgument(
+        "Memtable doesn't concurrent writes (allow_concurrent_memtable_write)");
+  }
+  return Status::OK();
+}
+
 ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
                                    const InternalKeyComparator* icmp,
                                    const ColumnFamilyOptions& src) {
  ColumnFamilyOptions result = src;
  result.comparator = icmp;
-#ifdef OS_MACOSX
-  // TODO(icanadi) make write_buffer_size uint64_t instead of size_t
-  ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, ((size_t)1) << 30);
-#else
-  ClipToRange(&result.write_buffer_size,
-              ((size_t)64) << 10, ((size_t)64) << 30);
-#endif
+  size_t clamp_max = std::conditional<
+      sizeof(size_t) == 4, std::integral_constant<size_t, 0xffffffff>,
+      std::integral_constant<uint64_t, 64ull << 30>>::type::value;
+  ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, clamp_max);
  // if user sets arena_block_size, we trust user to use this value. Otherwise,
  // calculate a proper value from writer_buffer_size;
  if (result.arena_block_size <= 0) {
@@ -211,6 +240,17 @@ ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
         result.level0_slowdown_writes_trigger,
         result.level0_file_num_compaction_trigger);
  }
+
+  if (result.soft_pending_compaction_bytes_limit == 0) {
+    result.soft_pending_compaction_bytes_limit =
+        result.hard_pending_compaction_bytes_limit;
+  } else if (result.hard_pending_compaction_bytes_limit > 0 &&
+             result.soft_pending_compaction_bytes_limit >
+                 result.hard_pending_compaction_bytes_limit) {
+    result.soft_pending_compaction_bytes_limit =
+        result.hard_pending_compaction_bytes_limit;
+  }
+
  if (result.level_compaction_dynamic_level_bytes) {
    if (result.compaction_style != kCompactionStyleLevel ||
        db_options.db_paths.size() > 1U) {
@@ -315,7 +355,8 @@ ColumnFamilyData::ColumnFamilyData(
      log_number_(0),
      column_family_set_(column_family_set),
      pending_flush_(false),
-      pending_compaction_(false) {
+      pending_compaction_(false),
+      prev_compaction_needed_bytes_(0) {
  Ref();

  // Convert user defined table properties collector factories to internal ones.
@@ -433,13 +474,81 @@ void ColumnFamilyData::SetDropped() {
  column_family_set_->RemoveColumnFamily(this);
 }

+const double kSlowdownRatio = 1.2;
+
+namespace {
+std::unique_ptr<WriteControllerToken> SetupDelay(
+    uint64_t max_write_rate, WriteController* write_controller,
+    uint64_t compaction_needed_bytes, uint64_t prev_compaction_neeed_bytes,
+    bool auto_comapctions_disabled) {
+  const uint64_t kMinWriteRate = 1024u;  // Minimum write rate 1KB/s.
+
+  uint64_t write_rate = write_controller->delayed_write_rate();
+
+  if (auto_comapctions_disabled) {
+    // When auto compaction is disabled, always use the value user gave.
+    write_rate = max_write_rate;
+  } else if (write_controller->NeedsDelay() && max_write_rate > kMinWriteRate) {
+    // If user gives rate less than kMinWriteRate, don't adjust it.
+    //
+    // If already delayed, need to adjust based on previous compaction debt.
+    // When there are two or more column families require delay, we always
+    // increase or reduce write rate based on information for one single
+    // column family. It is likely to be OK but we can improve if there is a
+    // problem.
+    // Ignore compaction_needed_bytes = 0 case because compaction_needed_bytes
+    // is only available in level-based compaction
+    //
+    // If the compaction debt stays the same as previously, we also further slow
+    // down. It usually means a mem table is full. It's mainly for the case
+    // where both of flush and compaction are much slower than the speed we
+    // insert to mem tables, so we need to actively slow down before we get
+    // feedback signal from compaction and flushes to avoid the full stop
+    // because of hitting the max write buffer number.
+    if (prev_compaction_neeed_bytes > 0 &&
+        prev_compaction_neeed_bytes <= compaction_needed_bytes) {
+      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) /
+                                         kSlowdownRatio);
+      if (write_rate < kMinWriteRate) {
+        write_rate = kMinWriteRate;
+      }
+    } else if (prev_compaction_neeed_bytes > compaction_needed_bytes) {
+      // We are speeding up by ratio of kSlowdownRatio when we have paid
+      // compaction debt. But we'll never speed up to faster than the write rate
+      // given by users.
+      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) *
+                                         kSlowdownRatio);
+      if (write_rate > max_write_rate) {
+        write_rate = max_write_rate;
+      }
+    }
+  }
+  return write_controller->GetDelayToken(write_rate);
+}
+
+int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
+                                    int level0_slowdown_writes_trigger) {
+  // SanitizeOptions() ensures it.
+  assert(level0_file_num_compaction_trigger <= level0_slowdown_writes_trigger);
+
+  // 1/4 of the way between L0 compaction trigger threshold and slowdown
+  // condition.
+  // Or twice as compaction trigger, if it is smaller.
+  return std::min(level0_file_num_compaction_trigger * 2,
+                  level0_file_num_compaction_trigger +
+                      (level0_slowdown_writes_trigger -
+                       level0_file_num_compaction_trigger) /
+                          4);
+}
+}  // namespace
+
 void ColumnFamilyData::RecalculateWriteStallConditions(
      const MutableCFOptions& mutable_cf_options) {
  if (current_ != nullptr) {
    auto* vstorage = current_->storage_info();
-    const double score = vstorage->max_compaction_score();
-    const int max_level = vstorage->max_compaction_score_level();
    auto write_controller = column_family_set_->write_controller_;
+    uint64_t compaction_needed_bytes =
+        vstorage->estimated_compaction_needed_bytes();

    if (imm()->NumNotFlushed() >= mutable_cf_options.max_write_buffer_number) {
      write_controller_token_ = write_controller->GetStopToken();
@@ -461,37 +570,88 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
          "[%s] Stopping writes because we have %d level-0 files",
          name_.c_str(), vstorage->l0_delay_trigger_count());
    } else if (mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
-               vstorage->estimated_compaction_needed_bytes() >=
+               compaction_needed_bytes >=
                   mutable_cf_options.hard_pending_compaction_bytes_limit) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(
          InternalStats::HARD_PENDING_COMPACTION_BYTES_LIMIT, 1);
      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
-          "[%s] Stopping writes because estimated pending compaction "
-          "bytes exceed %" PRIu64,
-          name_.c_str(), vstorage->estimated_compaction_needed_bytes());
+          "[%s] Stopping writes because of estimated pending compaction "
+          "bytes %" PRIu64,
+          name_.c_str(), compaction_needed_bytes);
+    } else if (mutable_cf_options.max_write_buffer_number > 3 &&
+               imm()->NumNotFlushed() >=
+                   mutable_cf_options.max_write_buffer_number - 1) {
+      write_controller_token_ =
+          SetupDelay(ioptions_.delayed_write_rate, write_controller,
+                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+                     mutable_cf_options.disable_auto_compactions);
+      internal_stats_->AddCFStats(InternalStats::MEMTABLE_SLOWDOWN, 1);
+      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
+          "[%s] Stalling writes because we have %d immutable memtables "
+          "(waiting for flush), max_write_buffer_number is set to %d "
+          "rate %" PRIu64,
+          name_.c_str(), imm()->NumNotFlushed(),
+          mutable_cf_options.max_write_buffer_number,
+          write_controller->delayed_write_rate());
    } else if (mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
               vstorage->l0_delay_trigger_count() >=
                   mutable_cf_options.level0_slowdown_writes_trigger) {
-      write_controller_token_ = write_controller->GetDelayToken();
+      write_controller_token_ =
+          SetupDelay(ioptions_.delayed_write_rate, write_controller,
+                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::LEVEL0_SLOWDOWN_TOTAL, 1);
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
        internal_stats_->AddCFStats(
            InternalStats::LEVEL0_SLOWDOWN_WITH_COMPACTION, 1);
      }
      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
-          "[%s] Stalling writes because we have %d level-0 files",
-          name_.c_str(), vstorage->l0_delay_trigger_count());
-    } else if (mutable_cf_options.soft_rate_limit > 0.0 &&
-               score > mutable_cf_options.soft_rate_limit) {
-      write_controller_token_ = write_controller->GetDelayToken();
-      internal_stats_->RecordLevelNSlowdown(max_level, true);
+          "[%s] Stalling writes because we have %d level-0 files "
+          "rate %" PRIu64,
+          name_.c_str(), vstorage->l0_delay_trigger_count(),
+          write_controller->delayed_write_rate());
+    } else if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
+               vstorage->estimated_compaction_needed_bytes() >=
+                   mutable_cf_options.soft_pending_compaction_bytes_limit) {
+      write_controller_token_ =
+          SetupDelay(ioptions_.delayed_write_rate, write_controller,
+                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+                     mutable_cf_options.disable_auto_compactions);
+      internal_stats_->AddCFStats(
+          InternalStats::SOFT_PENDING_COMPACTION_BYTES_LIMIT, 1);
      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
-          "[%s] Stalling writes because we hit soft limit on level %d",
-          name_.c_str(), max_level);
+          "[%s] Stalling writes because of estimated pending compaction "
+          "bytes %" PRIu64 " rate %" PRIu64,
+          name_.c_str(), vstorage->estimated_compaction_needed_bytes(),
+          write_controller->delayed_write_rate());
+    } else if (vstorage->l0_delay_trigger_count() >=
+               GetL0ThresholdSpeedupCompaction(
+                   mutable_cf_options.level0_file_num_compaction_trigger,
+                   mutable_cf_options.level0_slowdown_writes_trigger)) {
+      write_controller_token_ = write_controller->GetCompactionPressureToken();
+      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
+          "[%s] Increasing compaction threads because we have %d level-0 "
+          "files ",
+          name_.c_str(), vstorage->l0_delay_trigger_count());
+    } else if (vstorage->estimated_compaction_needed_bytes() >=
+               mutable_cf_options.soft_pending_compaction_bytes_limit / 4) {
+      // Increase compaction threads if bytes needed for compaction exceeds
+      // 1/4 of threshold for slowing down.
+      // If soft pending compaction byte limit is not set, always speed up
+      // compaction.
+      write_controller_token_ = write_controller->GetCompactionPressureToken();
+      if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0) {
+        Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
+            "[%s] Increasing compaction threads because of estimated pending "
+            "compaction "
+            "bytes %" PRIu64,
+            name_.c_str(), vstorage->estimated_compaction_needed_bytes());
+      }
    } else {
      write_controller_token_.reset();
    }
+    prev_compaction_needed_bytes_ = compaction_needed_bytes;
  }
 }

@@ -545,13 +705,12 @@ const int ColumnFamilyData::kCompactAllLevels = -1;
 const int ColumnFamilyData::kCompactToBaseLevel = -2;

 Compaction* ColumnFamilyData::CompactRange(
-    const MutableCFOptions& mutable_cf_options,
-    int input_level, int output_level, uint32_t output_path_id,
-    const InternalKey* begin, const InternalKey* end,
-    InternalKey** compaction_end) {
+    const MutableCFOptions& mutable_cf_options, int input_level,
+    int output_level, uint32_t output_path_id, const InternalKey* begin,
+    const InternalKey* end, InternalKey** compaction_end, bool* conflict) {
  auto* result = compaction_picker_->CompactRange(
      GetName(), mutable_cf_options, current_->storage_info(), input_level,
-      output_level, output_path_id, begin, end, compaction_end);
+      output_level, output_path_id, begin, end, compaction_end, conflict);
  if (result != nullptr) {
    result->SetInputVersion(current_);
  }
@@ -837,13 +996,6 @@ ColumnFamilyHandle* ColumnFamilyMemTablesImpl::GetColumnFamilyHandle() {
  return &handle_;
 }

-void ColumnFamilyMemTablesImpl::CheckMemtableFull() {
-  if (current_ != nullptr && current_->mem()->ShouldScheduleFlush()) {
-    flush_scheduler_->ScheduleFlush(current_);
-    current_->mem()->MarkFlushScheduled();
-  }
-}
-
 uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family) {
  uint32_t column_family_id = 0;
  if (column_family != nullptr) {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -19,12 +19,10 @@
 #include "db/write_controller.h"
 #include "db/table_cache.h"
 #include "db/table_properties_collector.h"
-#include "db/flush_scheduler.h"
 #include "rocksdb/compaction_job_stats.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/options.h"
-#include "util/instrumented_mutex.h"
 #include "util/mutable_cf_options.h"
 #include "util/thread_local.h"

@@ -44,6 +42,8 @@ class LogBuffer;
 class InstrumentedMutex;
 class InstrumentedMutexLock;

+extern const double kSlowdownRatio;
+
 // ColumnFamilyHandleImpl is the class that clients use to access different
 // column families. It has non-trivial destructor, which gets called when client
 // is done using the column family
@@ -59,6 +59,7 @@ class ColumnFamilyHandleImpl : public ColumnFamilyHandle {

  virtual uint32_t GetID() const override;
  virtual const std::string& GetName() const override;
+  virtual Status GetDescriptor(ColumnFamilyDescriptor* desc) override;

 private:
  ColumnFamilyData* cfd_;
@@ -132,6 +133,9 @@ struct SuperVersion {

 extern Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options);

+extern Status CheckConcurrentWritesSupported(
+    const ColumnFamilyOptions& cf_options);
+
 extern ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
                                           const InternalKeyComparator* icmp,
                                           const ColumnFamilyOptions& src);
@@ -156,14 +160,16 @@ class ColumnFamilyData {
  // thread-safe
  const std::string& GetName() const { return name_; }

-  // Ref() can only be called whily holding a DB mutex or during a
-  // single-threaded write.
+  // Ref() can only be called from a context where the caller can guarantee
+  // that ColumnFamilyData is alive (while holding a non-zero ref already,
+  // holding a DB mutex, or as the leader in a write batch group).
  void Ref() { refs_.fetch_add(1, std::memory_order_relaxed); }
-  // will just decrease reference count to 0, but will not delete it. returns
-  // true if the ref count was decreased to zero. in that case, it can be
-  // deleted by the caller immediately, or later, by calling
-  // FreeDeadColumnFamilies()
-  // Unref() can only be called while holding a DB mutex
+
+  // Unref decreases the reference count, but does not handle deletion
+  // when the count goes to 0.  If this method returns true then the
+  // caller should delete the instance immediately, or later, by calling
+  // FreeDeadColumnFamilies().  Unref() can only be called while holding
+  // a DB mutex, or during single-threaded recovery.
  bool Unref() {
    int old_refs = refs_.fetch_sub(1, std::memory_order_relaxed);
    assert(old_refs > 0);
@@ -203,7 +209,7 @@ class ColumnFamilyData {
  const ImmutableCFOptions* ioptions() const { return &ioptions_; }
  // REQUIRES: DB mutex held
  // This returns the MutableCFOptions used by current SuperVersion
-  // You shoul use this API to reference MutableCFOptions most of the time.
+  // You should use this API to reference MutableCFOptions most of the time.
  const MutableCFOptions* GetCurrentMutableCFOptions() const {
    return &(super_version_->mutable_cf_options);
  }
@@ -224,7 +230,7 @@ class ColumnFamilyData {
  MemTable* mem() { return mem_; }
  Version* current() { return current_; }
  Version* dummy_versions() { return dummy_versions_; }
-  void SetCurrent(Version* current);
+  void SetCurrent(Version* _current);
  uint64_t GetNumLiveVersions() const;  // REQUIRE: DB mutex held
  uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
  void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }
@@ -249,11 +255,11 @@ class ColumnFamilyData {
  // A flag to tell a manual compaction's output is base level.
  static const int kCompactToBaseLevel;
  // REQUIRES: DB mutex held
-  Compaction* CompactRange(
-      const MutableCFOptions& mutable_cf_options,
-      int input_level, int output_level, uint32_t output_path_id,
-      const InternalKey* begin, const InternalKey* end,
-      InternalKey** compaction_end);
+  Compaction* CompactRange(const MutableCFOptions& mutable_cf_options,
+                           int input_level, int output_level,
+                           uint32_t output_path_id, const InternalKey* begin,
+                           const InternalKey* end, InternalKey** compaction_end,
+                           bool* manual_conflict);

  CompactionPicker* compaction_picker() { return compaction_picker_.get(); }
  // thread-safe
@@ -305,6 +311,14 @@ class ColumnFamilyData {
  bool pending_flush() { return pending_flush_; }
  bool pending_compaction() { return pending_compaction_; }

+  // Recalculate some small conditions, which are changed only during
+  // compaction, adding new memtable and/or
+  // recalculation of compaction score. These values are used in
+  // DBImpl::MakeRoomForWrite function to decide, if it need to make
+  // a write stall
+  void RecalculateWriteStallConditions(
+      const MutableCFOptions& mutable_cf_options);
+
 private:
  friend class ColumnFamilySet;
  ColumnFamilyData(uint32_t id, const std::string& name,
@@ -314,14 +328,6 @@ class ColumnFamilyData {
                   const DBOptions* db_options, const EnvOptions& env_options,
                   ColumnFamilySet* column_family_set);

-  // Recalculate some small conditions, which are changed only during
-  // compaction, adding new memtable and/or
-  // recalculation of compaction score. These values are used in
-  // DBImpl::MakeRoomForWrite function to decide, if it need to make
-  // a write stall
-  void RecalculateWriteStallConditions(
-      const MutableCFOptions& mutable_cf_options);
-
  uint32_t id_;
  const std::string name_;
  Version* dummy_versions_;  // Head of circular doubly-linked list of versions.
@@ -382,6 +388,8 @@ class ColumnFamilyData {
  // If true --> this ColumnFamily is currently present in
  // DBImpl::compaction_queue_
  bool pending_compaction_;
+
+  uint64_t prev_compaction_needed_bytes_;
 };

 // ColumnFamilySet has interesting thread-safety requirements
@@ -457,6 +465,8 @@ class ColumnFamilySet {
  // Don't call while iterating over ColumnFamilySet
  void FreeDeadColumnFamilies();

+  Cache* get_table_cache() { return table_cache_; }
+
 private:
  friend class ColumnFamilyData;
  // helper function that gets called from cfd destructor
@@ -493,15 +503,18 @@ class ColumnFamilySet {
 // memtables of different column families (specified by ID in the write batch)
 class ColumnFamilyMemTablesImpl : public ColumnFamilyMemTables {
 public:
-  explicit ColumnFamilyMemTablesImpl(ColumnFamilySet* column_family_set,
-                                     FlushScheduler* flush_scheduler)
-      : column_family_set_(column_family_set),
-        current_(nullptr),
-        flush_scheduler_(flush_scheduler) {}
+  explicit ColumnFamilyMemTablesImpl(ColumnFamilySet* column_family_set)
+      : column_family_set_(column_family_set), current_(nullptr) {}
+
+  // Constructs a ColumnFamilyMemTablesImpl equivalent to one constructed
+  // with the arguments used to construct *orig.
+  explicit ColumnFamilyMemTablesImpl(ColumnFamilyMemTablesImpl* orig)
+      : column_family_set_(orig->column_family_set_), current_(nullptr) {}

  // sets current_ to ColumnFamilyData with column_family_id
  // returns false if column family doesn't exist
-  // REQUIRES: under a DB mutex OR from a write thread
+  // REQUIRES: use this function of DBImpl::column_family_memtables_ should be
+  //           under a DB mutex OR from a write thread
  bool Seek(uint32_t column_family_id) override;

  // Returns log number of the selected column family
@@ -509,20 +522,23 @@ class ColumnFamilyMemTablesImpl : public ColumnFamilyMemTables {
  uint64_t GetLogNumber() const override;

  // REQUIRES: Seek() called first
-  // REQUIRES: under a DB mutex OR from a write thread
+  // REQUIRES: use this function of DBImpl::column_family_memtables_ should be
+  //           under a DB mutex OR from a write thread
  virtual MemTable* GetMemTable() const override;

  // Returns column family handle for the selected column family
-  // REQUIRES: under a DB mutex OR from a write thread
+  // REQUIRES: use this function of DBImpl::column_family_memtables_ should be
+  //           under a DB mutex OR from a write thread
  virtual ColumnFamilyHandle* GetColumnFamilyHandle() override;

-  // REQUIRES: under a DB mutex OR from a write thread
-  virtual void CheckMemtableFull() override;
+  // Cannot be called while another thread is calling Seek().
+  // REQUIRES: use this function of DBImpl::column_family_memtables_ should be
+  //           under a DB mutex OR from a write thread
+  virtual ColumnFamilyData* current() override { return current_; }

 private:
  ColumnFamilySet* column_family_set_;
  ColumnFamilyData* current_;
-  FlushScheduler* flush_scheduler_;
  ColumnFamilyHandleInternal handle_;
 };

@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -7,6 +7,7 @@

 #include <mutex>
 #include <string>
+#include <thread>
 #include <vector>

 #include "rocksdb/db.h"
@@ -107,6 +108,7 @@ TEST_F(CompactFilesTest, L0ConflictsFiles) {
      break;
    }
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
  delete db;
 }

@@ -141,9 +143,6 @@ TEST_F(CompactFilesTest, ObsoleteFiles) {
  }

  auto l0_files = collector->GetFlushedFiles();
-  CompactionOptions compact_opt;
-  compact_opt.compression = kNoCompression;
-  compact_opt.output_file_size_limit = kWriteBufferSize * 5;
  ASSERT_OK(db->CompactFiles(CompactionOptions(), l0_files, 1));

  // verify all compaction input files are deleted
@@ -153,6 +152,62 @@ TEST_F(CompactFilesTest, ObsoleteFiles) {
  delete db;
 }

+TEST_F(CompactFilesTest, CapturingPendingFiles) {
+  Options options;
+  options.create_if_missing = true;
+  // Disable RocksDB background compaction.
+  options.compaction_style = kCompactionStyleNone;
+  // Always do full scans for obsolete files (needed to reproduce the issue).
+  options.delete_obsolete_files_period_micros = 0;
+
+  // Add listener.
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+
+  DB* db = nullptr;
+  DestroyDB(db_name_, options);
+  Status s = DB::Open(options, db_name_, &db);
+  assert(s.ok());
+  assert(db);
+
+  // Create 5 files.
+  for (int i = 0; i < 5; ++i) {
+    db->Put(WriteOptions(), "key" + ToString(i), "value");
+    db->Flush(FlushOptions());
+  }
+
+  auto l0_files = collector->GetFlushedFiles();
+  EXPECT_EQ(5, l0_files.size());
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"CompactFilesImpl:2", "CompactFilesTest.CapturingPendingFiles:0"},
+      {"CompactFilesTest.CapturingPendingFiles:1", "CompactFilesImpl:3"},
+  });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Start compacting files.
+  std::thread compaction_thread(
+      [&] { EXPECT_OK(db->CompactFiles(CompactionOptions(), l0_files, 1)); });
+
+  // In the meantime flush another file.
+  TEST_SYNC_POINT("CompactFilesTest.CapturingPendingFiles:0");
+  db->Put(WriteOptions(), "key5", "value");
+  db->Flush(FlushOptions());
+  TEST_SYNC_POINT("CompactFilesTest.CapturingPendingFiles:1");
+
+  compaction_thread.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  delete db;
+
+  // Make sure we can reopen the DB.
+  s = DB::Open(options, db_name_, &db);
+  ASSERT_TRUE(s.ok());
+  assert(db);
+  delete db;
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -93,7 +93,7 @@ Status CompactedDBImpl::Init(const Options& options) {
  mutex_.Lock();
  ColumnFamilyDescriptor cf(kDefaultColumnFamilyName,
                            ColumnFamilyOptions(options));
-  Status s = Recover({ cf }, true /* read only */, false);
+  Status s = Recover({cf}, true /* read only */, false, true);
  if (s.ok()) {
    cfd_ = reinterpret_cast<ColumnFamilyHandleImpl*>(
              DefaultColumnFamily())->cfd();
@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -16,8 +16,8 @@
 #include <inttypes.h>
 #include <vector>

-#include "rocksdb/compaction_filter.h"
 #include "db/column_family.h"
+#include "rocksdb/compaction_filter.h"
 #include "util/logging.h"
 #include "util/sync_point.h"

@@ -46,7 +46,7 @@ void Compaction::GetBoundaryKeys(
    Slice* largest_user_key) {
  bool initialized = false;
  const Comparator* ucmp = vstorage->InternalComparator()->user_comparator();
-  for (uint32_t i = 0; i < inputs.size(); ++i) {
+  for (size_t i = 0; i < inputs.size(); ++i) {
    if (inputs[i].files.empty()) {
      continue;
    }
@@ -128,8 +128,8 @@ bool Compaction::TEST_IsBottommostLevel(
 bool Compaction::IsFullCompaction(
    VersionStorageInfo* vstorage,
    const std::vector<CompactionInputFiles>& inputs) {
-  int num_files_in_compaction = 0;
-  int total_num_files = 0;
+  size_t num_files_in_compaction = 0;
+  size_t total_num_files = 0;
  for (int l = 0; l < vstorage->num_levels(); l++) {
    total_num_files += vstorage->NumLevelFiles(l);
  }
@@ -147,7 +147,8 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
                       uint32_t _output_path_id, CompressionType _compression,
                       std::vector<FileMetaData*> _grandparents,
                       bool _manual_compaction, double _score,
-                       bool _deletion_compaction)
+                       bool _deletion_compaction,
+                       CompactionReason _compaction_reason)
    : start_level_(_inputs[0].level),
      output_level_(_output_level),
      max_output_file_size_(_target_file_size),
@@ -161,14 +162,15 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
      deletion_compaction_(_deletion_compaction),
      inputs_(std::move(_inputs)),
      grandparents_(std::move(_grandparents)),
-      grandparent_index_(0),
-      seen_key_(false),
-      overlapped_bytes_(0),
      score_(_score),
      bottommost_level_(IsBottommostLevel(output_level_, vstorage, inputs_)),
      is_full_compaction_(IsFullCompaction(vstorage, inputs_)),
-      is_manual_compaction_(_manual_compaction) {
+      is_manual_compaction_(_manual_compaction),
+      compaction_reason_(_compaction_reason) {
  MarkFilesBeingCompacted(true);
+  if (is_manual_compaction_) {
+    compaction_reason_ = CompactionReason::kManualCompaction;
+  }

 #ifndef NDEBUG
  for (size_t i = 1; i < inputs_.size(); ++i) {
@@ -184,6 +186,9 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
                                &arena_);
    }
  }
+
+  Slice smallest_user_key;
+  GetBoundaryKeys(vstorage, inputs_, &smallest_user_key, &largest_user_key_);
 }

 Compaction::~Compaction() {
@@ -198,9 +203,11 @@ Compaction::~Compaction() {
 }

 bool Compaction::InputCompressionMatchesOutput() const {
-  int base_level = input_version_->storage_info()->base_level();
-  bool matches = (GetCompressionType(*cfd_->ioptions(), start_level_,
-                                     base_level) == output_compression_);
+  VersionStorageInfo* vstorage = input_version_->storage_info();
+  int base_level = vstorage->base_level();
+  bool matches =
+      (GetCompressionType(*cfd_->ioptions(), vstorage, mutable_cf_options_,
+                          start_level_, base_level) == output_compression_);
  if (matches) {
    TEST_SYNC_POINT("Compaction::InputCompressionMatchesOutput:Matches");
    return true;
@@ -282,38 +289,12 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
  return true;
 }

-bool Compaction::ShouldStopBefore(const Slice& internal_key) {
-  // Scan to find earliest grandparent file that contains key.
-  const InternalKeyComparator* icmp = &cfd_->internal_comparator();
-  while (grandparent_index_ < grandparents_.size() &&
-      icmp->Compare(internal_key,
-                    grandparents_[grandparent_index_]->largest.Encode()) > 0) {
-    if (seen_key_) {
-      overlapped_bytes_ += grandparents_[grandparent_index_]->fd.GetFileSize();
-    }
-    assert(grandparent_index_ + 1 >= grandparents_.size() ||
-           icmp->Compare(grandparents_[grandparent_index_]->largest.Encode(),
-                         grandparents_[grandparent_index_+1]->smallest.Encode())
-                         < 0);
-    grandparent_index_++;
-  }
-  seen_key_ = true;
-
-  if (overlapped_bytes_ > max_grandparent_overlap_bytes_) {
-    // Too much overlap for current output; start new output
-    overlapped_bytes_ = 0;
-    return true;
-  } else {
-    return false;
-  }
-}
-
 // Mark (or clear) each file that is being compacted
 void Compaction::MarkFilesBeingCompacted(bool mark_as_compacted) {
  for (size_t i = 0; i < num_input_levels(); i++) {
-    for (unsigned int j = 0; j < inputs_[i].size(); j++) {
-      assert(mark_as_compacted ? !inputs_[i][j]->being_compacted :
-                                  inputs_[i][j]->being_compacted);
+    for (size_t j = 0; j < inputs_[i].size(); j++) {
+      assert(mark_as_compacted ? !inputs_[i][j]->being_compacted
+                               : inputs_[i][j]->being_compacted);
      inputs_[i][j]->being_compacted = mark_as_compacted;
    }
  }
@@ -371,7 +352,7 @@ int InputSummary(const std::vector<FileMetaData*>& files, char* output,
                 int len) {
  *output = '\0';
  int write = 0;
-  for (unsigned int i = 0; i < files.size(); i++) {
+  for (size_t i = 0; i < files.size(); i++) {
    int sz = len - write;
    int ret;
    char sztxt[16];
@@ -388,10 +369,8 @@ int InputSummary(const std::vector<FileMetaData*>& files, char* output,

 void Compaction::Summary(char* output, int len) {
  int write =
-      snprintf(output, len, "Base version %" PRIu64
-                            " Base level %d, inputs: [",
-               input_version_->GetVersionNumber(),
-               start_level_);
+      snprintf(output, len, "Base version %" PRIu64 " Base level %d, inputs: [",
+               input_version_->GetVersionNumber(), start_level_);
  if (write < 0 || write >= len) {
    return;
  }
@@ -413,7 +392,7 @@ void Compaction::Summary(char* output, int len) {
  snprintf(output + write, len - write, "]");
 }

-uint64_t Compaction::OutputFilePreallocationSize() {
+uint64_t Compaction::OutputFilePreallocationSize() const {
  uint64_t preallocation_size = 0;

  if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel ||
@@ -428,7 +407,7 @@ uint64_t Compaction::OutputFilePreallocationSize() {
  }
  // Over-estimate slightly so we don't end up just barely crossing
  // the threshold
-  return preallocation_size * 1.1;
+  return preallocation_size + (preallocation_size / 10);
 }

 std::unique_ptr<CompactionFilter> Compaction::CreateCompactionFilter() const {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -41,7 +41,8 @@ class Compaction {
             uint32_t output_path_id, CompressionType compression,
             std::vector<FileMetaData*> grandparents,
             bool manual_compaction = false, double score = -1,
-             bool deletion_compaction = false);
+             bool deletion_compaction = false,
+             CompactionReason compaction_reason = CompactionReason::kUnknown);

  // No copying allowed
  Compaction(const Compaction&) = delete;
@@ -103,7 +104,7 @@ class Compaction {
  }

  // Returns the LevelFilesBrief of the specified compaction input level.
-  LevelFilesBrief* input_levels(size_t compaction_input_level) {
+  const LevelFilesBrief* input_levels(size_t compaction_input_level) const {
    return &input_levels_[compaction_input_level];
  }

@@ -131,12 +132,10 @@ class Compaction {
  bool KeyNotExistsBeyondOutputLevel(const Slice& user_key,
                                     std::vector<size_t>* level_ptrs) const;

-  // Returns true iff we should stop building the current output
-  // before processing "internal_key".
-  bool ShouldStopBefore(const Slice& internal_key);
-
  // Clear all files to indicate that they are not being compacted
  // Delete this compaction from the list of running compactions.
+  //
+  // Requirement: DB mutex held
  void ReleaseCompactionFiles(Status status);

  // Returns the summary of the compaction in "output" with maximum "len"
@@ -148,13 +147,13 @@ class Compaction {
  double score() const { return score_; }

  // Is this compaction creating a file in the bottom most level?
-  bool bottommost_level() { return bottommost_level_; }
+  bool bottommost_level() const { return bottommost_level_; }

  // Does this compaction include all sst files?
-  bool is_full_compaction() { return is_full_compaction_; }
+  bool is_full_compaction() const { return is_full_compaction_; }

  // Was this compaction triggered manually by the client?
-  bool is_manual_compaction() { return is_manual_compaction_; }
+  bool is_manual_compaction() const { return is_manual_compaction_; }

  // Used when allow_trivial_move option is set in
  // Universal compaction. If all the input files are
@@ -167,19 +166,21 @@ class Compaction {
  // Used when allow_trivial_move option is set in
  // Universal compaction. Returns true, if the input files
  // are non-overlapping and can be trivially moved.
-  bool is_trivial_move() { return is_trivial_move_; }
+  bool is_trivial_move() const { return is_trivial_move_; }

  // How many total levels are there?
  int number_levels() const { return number_levels_; }

  // Return the MutableCFOptions that should be used throughout the compaction
  // procedure
-  const MutableCFOptions* mutable_cf_options() { return &mutable_cf_options_; }
+  const MutableCFOptions* mutable_cf_options() const {
+    return &mutable_cf_options_;
+  }

  // Returns the size in bytes that the output file should be preallocated to.
  // In level compaction, that is max_file_size_. In universal compaction, that
  // is the sum of all input file sizes.
-  uint64_t OutputFilePreallocationSize();
+  uint64_t OutputFilePreallocationSize() const;

  void SetInputVersion(Version* input_version);

@@ -218,6 +219,18 @@ class Compaction {
    output_table_properties_ = std::move(tp);
  }

+  Slice GetLargestUserKey() const { return largest_user_key_; }
+
+  CompactionReason compaction_reason() { return compaction_reason_; }
+
+  const std::vector<FileMetaData*>& grandparents() const {
+    return grandparents_;
+  }
+
+  uint64_t max_grandparent_overlap_bytes() const {
+    return max_grandparent_overlap_bytes_;
+  }
+
 private:
  // mark (or clear) all files that are being compacted
  void MarkFilesBeingCompacted(bool mark_as_compacted);
@@ -258,13 +271,9 @@ class Compaction {
  // A copy of inputs_, organized more closely in memory
  autovector<LevelFilesBrief, 2> input_levels_;

-  // State used to check for number of of overlapping grandparent files
+  // State used to check for number of overlapping grandparent files
  // (grandparent == "output_level_ + 1")
  std::vector<FileMetaData*> grandparents_;
-  size_t grandparent_index_;   // Index in grandparent_starts_
-  bool seen_key_;              // Some output key has been seen
-  uint64_t overlapped_bytes_;  // Bytes of overlap between current output
-                               // and grandparent files
  const double score_;         // score that was used to pick this compaction.

  // Is this compaction creating a file in the bottom most level?
@@ -284,6 +293,12 @@ class Compaction {

  // table properties of output files
  TablePropertiesCollection output_table_properties_;
+
+  // largest user keys in compaction
+  Slice largest_user_key_;
+
+  // Reason for compaction
+  CompactionReason compaction_reason_;
 };

 // Utility function
@@ -1,6 +1,6 @@
 // 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.
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -13,12 +13,14 @@ namespace rocksdb {
 CompactionIterator::CompactionIterator(
    InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
    SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
-    Env* env, bool expect_valid_internal_key, Compaction* compaction,
+    SequenceNumber earliest_write_conflict_snapshot, Env* env,
+    bool expect_valid_internal_key, const Compaction* compaction,
    const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
    : input_(input),
      cmp_(cmp),
      merge_helper_(merge_helper),
      snapshots_(snapshots),
+      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      env_(env),
      expect_valid_internal_key_(expect_valid_internal_key),
      compaction_(compaction),
@@ -95,6 +97,11 @@ void CompactionIterator::Next() {
    NextFromInput();
  }

+  if (valid_) {
+    // Record that we've ouputted a record for the current key.
+    has_outputted_key_ = true;
+  }
+
  PrepareOutput();
 }

@@ -141,8 +148,10 @@ void CompactionIterator::NextFromInput() {
      key_ = current_key_.SetKey(key_, &ikey_);
      current_user_key_ = ikey_.user_key;
      has_current_user_key_ = true;
+      has_outputted_key_ = false;
      current_user_key_sequence_ = kMaxSequenceNumber;
      current_user_key_snapshot_ = 0;
+
      // apply the compaction filter to the first occurrence of the user key
      if (compaction_filter_ != nullptr && ikey_.type == kTypeValue &&
          (visible_at_tip_ || ikey_.sequence > latest_snapshot_ ||
@@ -196,31 +205,99 @@ void CompactionIterator::NextFromInput() {
        visible_at_tip_ ? visible_at_tip_ : findEarliestVisibleSnapshot(
                                                ikey_.sequence, &prev_snapshot);

-    if (ikey_.type == kTypeSingleDeletion) {
+    if (clear_and_output_next_key_) {
+      // In the previous iteration we encountered a single delete that we could
+      // not compact out.  We will keep this Put, but can drop it's data.
+      // (See Optimization 3, below.)
+      assert(ikey_.type == kTypeValue);
+      assert(current_user_key_snapshot_ == last_snapshot);
+
+      value_.clear();
+      valid_ = true;
+      clear_and_output_next_key_ = false;
+    } else if (ikey_.type == kTypeSingleDeletion) {
+      // We can compact out a SingleDelete if:
+      // 1) We encounter the corresponding PUT -OR- we know that this key
+      //    doesn't appear past this output level
+      // =AND=
+      // 2) We've already returned a record in this snapshot -OR-
+      //    there are no earlier earliest_write_conflict_snapshot.
+      //
+      // Rule 1 is needed for SingleDelete correctness.  Rule 2 is needed to
+      // allow Transactions to do write-conflict checking (if we compacted away
+      // all keys, then we wouldn't know that a write happened in this
+      // snapshot).  If there is no earlier snapshot, then we know that there
+      // are no active transactions that need to know about any writes.
+      //
+      // Optimization 3:
+      // If we encounter a SingleDelete followed by a PUT and Rule 2 is NOT
+      // true, then we must output a SingleDelete.  In this case, we will decide
+      // to also output the PUT.  While we are compacting less by outputting the
+      // PUT now, hopefully this will lead to better compaction in the future
+      // when Rule 2 is later true (Ie, We are hoping we can later compact out
+      // both the SingleDelete and the Put, while we couldn't if we only
+      // outputted the SingleDelete now).
+      // In this case, we can save space by removing the PUT's value as it will
+      // never be read.
+      //
+      // Deletes and Merges are not supported on the same key that has a
+      // SingleDelete as it is not possible to correctly do any partial
+      // compaction of such a combination of operations.  The result of mixing
+      // those operations for a given key is documented as being undefined.  So
+      // we can choose how to handle such a combinations of operations.  We will
+      // try to compact out as much as we can in these cases.
+
+      // The easiest way to process a SingleDelete during iteration is to peek
+      // ahead at the next key.
      ParsedInternalKey next_ikey;
      input_->Next();

-      // Check whether the current key is valid, not corrupt and the same
+      // Check whether the next key exists, is not corrupt, and is the same key
      // as the single delete.
      if (input_->Valid() && ParseInternalKey(input_->key(), &next_ikey) &&
          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
-        // Mixing single deletes and merges is not supported. Consecutive
-        // single deletes are not valid.
-        if (next_ikey.type != kTypeValue) {
-          assert(false);
-          status_ =
-              Status::InvalidArgument("Put expected after single delete.");
-          break;
-        }
-
-        // Check whether the current key belongs to the same snapshot as the
-        // single delete.
+        // Check whether the next key belongs to the same snapshot as the
+        // SingleDelete.
        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot) {
-          // Found the matching value, we can drop the single delete and the
-          // value.
-          ++iter_stats_.num_record_drop_hidden;
-          ++iter_stats_.num_record_drop_obsolete;
-          input_->Next();
+          if (next_ikey.type == kTypeSingleDeletion) {
+            // We encountered two SingleDeletes in a row.  This could be due to
+            // unexpected user input.
+            // Skip the first SingleDelete and let the next iteration decide how
+            // to handle the second SingleDelete
+
+            // First SingleDelete has been skipped since we already called
+            // input_->Next().
+            ++iter_stats_.num_record_drop_obsolete;
+          } else if ((ikey_.sequence <= earliest_write_conflict_snapshot_) ||
+                     has_outputted_key_) {
+            // Found a matching value, we can drop the single delete and the
+            // value.  It is safe to drop both records since we've already
+            // outputted a key in this snapshot, or there is no earlier
+            // snapshot (Rule 2 above).
+
+            // Note: it doesn't matter whether the second key is a Put or if it
+            // is an unexpected Merge or Delete.  We will compact it out
+            // either way.
+            ++iter_stats_.num_record_drop_hidden;
+            ++iter_stats_.num_record_drop_obsolete;
+            // Already called input_->Next() once.  Call it a second time to
+            // skip past the second key.
+            input_->Next();
+          } else {
+            // Found a matching value, but we cannot drop both keys since
+            // there is an earlier snapshot and we need to leave behind a record
+            // to know that a write happened in this snapshot (Rule 2 above).
+            // Clear the value and output the SingleDelete. (The value will be
+            // outputted on the next iteration.)
+            ++iter_stats_.num_record_drop_hidden;
+
+            // Setting valid_ to true will output the current SingleDelete
+            valid_ = true;
+
+            // Set up the Put to be outputted in the next iteration.
+            // (Optimization 3).
+            clear_and_output_next_key_ = true;
+          }
        } else {
          // We hit the next snapshot without hitting a put, so the iterator
          // returns the single delete.
@@ -235,11 +312,14 @@ void CompactionIterator::NextFromInput() {
        // iteration. If the next key is corrupt, we return before the
        // comparison, so the value of has_current_user_key does not matter.
        has_current_user_key_ = false;
-        if (compaction_ != nullptr &&
+        if (compaction_ != nullptr && ikey_.sequence <= earliest_snapshot_ &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_)) {
+          // Key doesn't exist outside of this range.
+          // Can compact out this SingleDelete.
          ++iter_stats_.num_record_drop_obsolete;
        } else {
+          // Output SingleDelete
          valid_ = true;
        }
      }
@@ -253,6 +333,10 @@ void CompactionIterator::NextFromInput() {
      // same key, then this kv is not visible in any snapshot.
      // Hidden by an newer entry for same user key
      // TODO: why not > ?
+      //
+      // Note: Dropping this key will not affect TransactionDB write-conflict
+      // checking since there has already been a record returned for this key
+      // in this snapshot.
      assert(last_sequence >= current_user_key_sequence_);
      ++iter_stats_.num_record_drop_hidden;  // (A)
      input_->Next();
@@ -271,6 +355,9 @@ void CompactionIterator::NextFromInput() {
      //     smaller sequence numbers will be dropped in the next
      //     few iterations of this loop (by rule (A) above).
      // Therefore this deletion marker is obsolete and can be dropped.
+      //
+      // Note:  Dropping this Delete will not affect TransactionDB
+      // write-conflict checking since it is earlier than any snapshot.
      ++iter_stats_.num_record_drop_obsolete;
      input_->Next();
    } else if (ikey_.type == kTypeMerge) {
@@ -319,9 +406,14 @@ void CompactionIterator::PrepareOutput() {
  // Zeroing out the sequence number leads to better compression.
  // If this is the bottommost level (no files in lower levels)
  // and the earliest snapshot is larger than this seqno
+  // and the userkey differs from the last userkey in compaction
  // then we can squash the seqno to zero.
+
+  // This is safe for TransactionDB write-conflict checking since transactions
+  // only care about sequence number larger than any active snapshots.
  if (bottommost_level_ && valid_ && ikey_.sequence < earliest_snapshot_ &&
-      ikey_.type != kTypeMerge) {
+      ikey_.type != kTypeMerge &&
+      !cmp_->Equal(compaction_->GetLargestUserKey(), ikey_.user_key)) {
    assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
    ikey_.sequence = 0;
    current_key_.UpdateInternalKey(0, ikey_.type);
@@ -331,15 +423,15 @@ void CompactionIterator::PrepareOutput() {
 inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
    SequenceNumber in, SequenceNumber* prev_snapshot) {
  assert(snapshots_->size());
-  SequenceNumber prev __attribute__((unused)) = 0;
+  SequenceNumber prev __attribute__((__unused__)) = kMaxSequenceNumber;
  for (const auto cur : *snapshots_) {
-    assert(prev <= cur);
+    assert(prev == kMaxSequenceNumber || prev <= cur);
    if (cur >= in) {
-      *prev_snapshot = prev;
+      *prev_snapshot = prev == kMaxSequenceNumber ? 0 : prev;
      return cur;
    }
    prev = cur;
-    assert(prev);
+    assert(prev < kMaxSequenceNumber);
  }
  *prev_snapshot = prev;
  return kMaxSequenceNumber;
@@ -1,6 +1,6 @@
 // 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.
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -39,9 +39,10 @@ class CompactionIterator {
 public:
  CompactionIterator(InternalIterator* input, const Comparator* cmp,
                     MergeHelper* merge_helper, SequenceNumber last_sequence,
-                     std::vector<SequenceNumber>* snapshots, Env* env,
+                     std::vector<SequenceNumber>* snapshots,
+                     SequenceNumber earliest_write_conflict_snapshot, Env* env,
                     bool expect_valid_internal_key,
-                     Compaction* compaction = nullptr,
+                     const Compaction* compaction = nullptr,
                     const CompactionFilter* compaction_filter = nullptr,
                     LogBuffer* log_buffer = nullptr);

@@ -88,9 +89,10 @@ class CompactionIterator {
  const Comparator* cmp_;
  MergeHelper* merge_helper_;
  const std::vector<SequenceNumber>* snapshots_;
+  const SequenceNumber earliest_write_conflict_snapshot_;
  Env* env_;
  bool expect_valid_internal_key_;
-  Compaction* compaction_;
+  const Compaction* compaction_;
  const CompactionFilter* compaction_filter_;
  LogBuffer* log_buffer_;
  bool bottommost_level_;
@@ -125,6 +127,14 @@ class CompactionIterator {
  Slice current_user_key_;
  SequenceNumber current_user_key_sequence_;
  SequenceNumber current_user_key_snapshot_;
+
+  // True if the iterator has already returned a record for the current key.
+  bool has_outputted_key_ = false;
+
+  // truncated the value of the next key and output it without applying any
+  // compaction rules.  This is used for outputting a put after a single delete.
+  bool clear_and_output_next_key_ = false;
+
  MergeOutputIterator merge_out_iter_;
  std::string compaction_filter_value_;
  // "level_ptrs" holds indices that remember which file of an associated
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -20,9 +20,9 @@ class CompactionIteratorTest : public testing::Test {
                                        nullptr, 0U, false, 0));
    iter_.reset(new test::VectorIterator(ks, vs));
    iter_->SeekToFirst();
-    c_iter_.reset(new CompactionIterator(iter_.get(), cmp_, merge_helper_.get(),
-                                         last_sequence, &snapshots_,
-                                         Env::Default(), false));
+    c_iter_.reset(new CompactionIterator(
+        iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
+        kMaxSequenceNumber, Env::Default(), false));
  }

  const Comparator* cmp_;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -16,12 +16,13 @@
 #include <inttypes.h>
 #include <algorithm>
 #include <functional>
-#include <vector>
-#include <memory>
 #include <list>
+#include <memory>
+#include <random>
 #include <set>
 #include <thread>
 #include <utility>
+#include <vector>

 #include "db/builder.h"
 #include "db/db_iter.h"
@@ -51,6 +52,7 @@
 #include "util/iostats_context_imp.h"
 #include "util/log_buffer.h"
 #include "util/logging.h"
+#include "util/sst_file_manager_impl.h"
 #include "util/mutexlock.h"
 #include "util/perf_context_imp.h"
 #include "util/stop_watch.h"
@@ -62,7 +64,7 @@ namespace rocksdb {

 // Maintains state for each sub-compaction
 struct CompactionJob::SubcompactionState {
-  Compaction* compaction;
+  const Compaction* compaction;
  std::unique_ptr<CompactionIterator> c_iter;

  // The boundaries of the key-range this compaction is interested in. No two
@@ -103,6 +105,14 @@ struct CompactionJob::SubcompactionState {
  uint64_t num_output_records;
  CompactionJobStats compaction_job_stats;
  uint64_t approx_size;
+  // An index that used to speed up ShouldStopBefore().
+  size_t grandparent_index = 0;
+  // The number of bytes overlapping between the current output and
+  // grandparent files used in ShouldStopBefore().
+  uint64_t overlapped_bytes = 0;
+  // A flag determine whether the key has been seen in ShouldStopBefore()
+  bool seen_key = false;
+  std::string compression_dict;

  SubcompactionState(Compaction* c, Slice* _start, Slice* _end,
                     uint64_t size = 0)
@@ -114,7 +124,11 @@ struct CompactionJob::SubcompactionState {
        total_bytes(0),
        num_input_records(0),
        num_output_records(0),
-        approx_size(size) {
+        approx_size(size),
+        grandparent_index(0),
+        overlapped_bytes(0),
+        seen_key(false),
+        compression_dict() {
    assert(compaction != nullptr);
  }

@@ -133,6 +147,10 @@ struct CompactionJob::SubcompactionState {
    num_output_records = std::move(o.num_output_records);
    compaction_job_stats = std::move(o.compaction_job_stats);
    approx_size = std::move(o.approx_size);
+    grandparent_index = std::move(o.grandparent_index);
+    overlapped_bytes = std::move(o.overlapped_bytes);
+    seen_key = std::move(o.seen_key);
+    compression_dict = std::move(o.compression_dict);
    return *this;
  }

@@ -140,6 +158,38 @@ struct CompactionJob::SubcompactionState {
  SubcompactionState(const SubcompactionState&) = delete;

  SubcompactionState& operator=(const SubcompactionState&) = delete;
+
+  // Returns true iff we should stop building the current output
+  // before processing "internal_key".
+  bool ShouldStopBefore(const Slice& internal_key) {
+    const InternalKeyComparator* icmp =
+        &compaction->column_family_data()->internal_comparator();
+    const std::vector<FileMetaData*>& grandparents = compaction->grandparents();
+
+    // Scan to find earliest grandparent file that contains key.
+    while (grandparent_index < grandparents.size() &&
+           icmp->Compare(internal_key,
+                         grandparents[grandparent_index]->largest.Encode()) >
+               0) {
+      if (seen_key) {
+        overlapped_bytes += grandparents[grandparent_index]->fd.GetFileSize();
+      }
+      assert(grandparent_index + 1 >= grandparents.size() ||
+             icmp->Compare(
+                 grandparents[grandparent_index]->largest.Encode(),
+                 grandparents[grandparent_index + 1]->smallest.Encode()) <= 0);
+      grandparent_index++;
+    }
+    seen_key = true;
+
+    if (overlapped_bytes > compaction->max_grandparent_overlap_bytes()) {
+      // Too much overlap for current output; start new output
+      overlapped_bytes = 0;
+      return true;
+    }
+
+    return false;
+  }
 };

 // Maintains state for the entire compaction
@@ -211,7 +261,9 @@ CompactionJob::CompactionJob(
    const EnvOptions& env_options, VersionSet* versions,
    std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
    Directory* db_directory, Directory* output_directory, Statistics* stats,
+    InstrumentedMutex* db_mutex, Status* db_bg_error,
    std::vector<SequenceNumber> existing_snapshots,
+    SequenceNumber earliest_write_conflict_snapshot,
    std::shared_ptr<Cache> table_cache, EventLogger* event_logger,
    bool paranoid_file_checks, bool measure_io_stats, const std::string& dbname,
    CompactionJobStats* compaction_job_stats)
@@ -229,13 +281,18 @@ CompactionJob::CompactionJob(
      db_directory_(db_directory),
      output_directory_(output_directory),
      stats_(stats),
+      db_mutex_(db_mutex),
+      db_bg_error_(db_bg_error),
      existing_snapshots_(std::move(existing_snapshots)),
+      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      table_cache_(std::move(table_cache)),
      event_logger_(event_logger),
      paranoid_file_checks_(paranoid_file_checks),
      measure_io_stats_(measure_io_stats) {
  assert(log_buffer_ != nullptr);
-  ThreadStatusUtil::SetColumnFamily(compact_->compaction->column_family_data());
+  const auto* cfd = compact_->compaction->column_family_data();
+  ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
+                                    cfd->options()->enable_thread_tracking);
  ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_COMPACTION);
  ReportStartedCompaction(compaction);
 }
@@ -247,8 +304,9 @@ CompactionJob::~CompactionJob() {

 void CompactionJob::ReportStartedCompaction(
    Compaction* compaction) {
-  ThreadStatusUtil::SetColumnFamily(
-      compact_->compaction->column_family_data());
+  const auto* cfd = compact_->compaction->column_family_data();
+  ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
+                                    cfd->options()->enable_thread_tracking);

  ThreadStatusUtil::SetThreadOperationProperty(
      ThreadStatus::COMPACTION_JOB_ID,
@@ -332,11 +390,6 @@ struct RangeWithSize {
      : range(a, b), size(s) {}
 };

-bool SliceCompare(const Comparator* cmp, const Slice& a, const Slice& b) {
-  // Returns true if a < b
-  return cmp->Compare(ExtractUserKey(a), ExtractUserKey(b)) < 0;
-}
-
 // Generates a histogram representing potential divisions of key ranges from
 // the input. It adds the starting and/or ending keys of certain input files
 // to the working set and then finds the approximate size of data in between
@@ -345,14 +398,13 @@ bool SliceCompare(const Comparator* cmp, const Slice& a, const Slice& b) {
 void CompactionJob::GenSubcompactionBoundaries() {
  auto* c = compact_->compaction;
  auto* cfd = c->column_family_data();
-  std::set<Slice, std::function<bool(const Slice& a, const Slice& b)> > bounds(
-      std::bind(&SliceCompare, cfd->user_comparator(), std::placeholders::_1,
-                std::placeholders::_2));
+  const Comparator* cfd_comparator = cfd->user_comparator();
+  std::vector<Slice> bounds;
  int start_lvl = c->start_level();
  int out_lvl = c->output_level();

  // Add the starting and/or ending key of certain input files as a potential
-  // boundary (because we're inserting into a set, it avoids duplicates)
+  // boundary
  for (size_t lvl_idx = 0; lvl_idx < c->num_input_levels(); lvl_idx++) {
    int lvl = c->level(lvl_idx);
    if (lvl >= start_lvl && lvl <= out_lvl) {
@@ -360,34 +412,44 @@ void CompactionJob::GenSubcompactionBoundaries() {
      size_t num_files = flevel->num_files;

      if (num_files == 0) {
-        break;
+        continue;
      }

      if (lvl == 0) {
        // For level 0 add the starting and ending key of each file since the
        // files may have greatly differing key ranges (not range-partitioned)
        for (size_t i = 0; i < num_files; i++) {
-          bounds.emplace(flevel->files[i].smallest_key);
-          bounds.emplace(flevel->files[i].largest_key);
+          bounds.emplace_back(flevel->files[i].smallest_key);
+          bounds.emplace_back(flevel->files[i].largest_key);
        }
      } else {
        // For all other levels add the smallest/largest key in the level to
        // encompass the range covered by that level
-        bounds.emplace(flevel->files[0].smallest_key);
-        bounds.emplace(flevel->files[num_files - 1].largest_key);
+        bounds.emplace_back(flevel->files[0].smallest_key);
+        bounds.emplace_back(flevel->files[num_files - 1].largest_key);
        if (lvl == out_lvl) {
          // For the last level include the starting keys of all files since
          // the last level is the largest and probably has the widest key
          // range. Since it's range partitioned, the ending key of one file
          // and the starting key of the next are very close (or identical).
          for (size_t i = 1; i < num_files; i++) {
-            bounds.emplace(flevel->files[i].smallest_key);
+            bounds.emplace_back(flevel->files[i].smallest_key);
          }
        }
      }
    }
  }

+  std::sort(bounds.begin(), bounds.end(),
+    [cfd_comparator] (const Slice& a, const Slice& b) -> bool {
+      return cfd_comparator->Compare(ExtractUserKey(a), ExtractUserKey(b)) < 0;
+    });
+  // Remove duplicated entries from bounds
+  bounds.erase(std::unique(bounds.begin(), bounds.end(),
+    [cfd_comparator] (const Slice& a, const Slice& b) -> bool {
+      return cfd_comparator->Compare(ExtractUserKey(a), ExtractUserKey(b)) == 0;
+    }), bounds.end());
+
  // Combine consecutive pairs of boundaries into ranges with an approximate
  // size of data covered by keys in that range
  uint64_t sum = 0;
@@ -409,9 +471,9 @@ void CompactionJob::GenSubcompactionBoundaries() {

  // Group the ranges into subcompactions
  const double min_file_fill_percent = 4.0 / 5;
-  uint64_t max_output_files = std::ceil(
-      sum / min_file_fill_percent /
-      cfd->GetCurrentMutableCFOptions()->MaxFileSizeForLevel(out_lvl));
+  uint64_t max_output_files = static_cast<uint64_t>(
+      std::ceil(sum / min_file_fill_percent /
+                c->mutable_cf_options()->MaxFileSizeForLevel(out_lvl)));
  uint64_t subcompactions =
      std::min({static_cast<uint64_t>(ranges.size()),
                static_cast<uint64_t>(db_options_.max_subcompactions),
@@ -424,12 +486,12 @@ void CompactionJob::GenSubcompactionBoundaries() {
    // sizes becomes >= the expected mean size of a subcompaction
    sum = 0;
    for (size_t i = 0; i < ranges.size() - 1; i++) {
+      sum += ranges[i].size;
      if (subcompactions == 1) {
        // If there's only one left to schedule then it goes to the end so no
        // need to put an end boundary
-        break;
+        continue;
      }
-      sum += ranges[i].size;
      if (sum >= mean) {
        boundaries_.emplace_back(ExtractUserKey(ranges[i].range.limit));
        sizes_.emplace_back(sum);
@@ -509,18 +571,17 @@ Status CompactionJob::Run() {
  return status;
 }

-Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options,
-                              InstrumentedMutex* db_mutex) {
+Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_INSTALL);
-  db_mutex->AssertHeld();
+  db_mutex_->AssertHeld();
  Status status = compact_->status;
  ColumnFamilyData* cfd = compact_->compaction->column_family_data();
  cfd->internal_stats()->AddCompactionStats(
      compact_->compaction->output_level(), compaction_stats_);

  if (status.ok()) {
-    status = InstallCompactionResults(mutable_cf_options, db_mutex);
+    status = InstallCompactionResults(mutable_cf_options);
  }
  VersionStorageInfo::LevelSummaryStorage tmp;
  auto vstorage = cfd->current()->storage_info();
@@ -608,6 +669,31 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  }

  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  const MutableCFOptions* mutable_cf_options =
+      sub_compact->compaction->mutable_cf_options();
+
+  // To build compression dictionary, we sample the first output file, assuming
+  // it'll reach the maximum length, and then use the dictionary for compressing
+  // subsequent output files. The dictionary may be less than max_dict_bytes if
+  // the first output file's length is less than the maximum.
+  const int kSampleLenShift = 6;  // 2^6 = 64-byte samples
+  std::set<size_t> sample_begin_offsets;
+  if (bottommost_level_ &&
+      cfd->ioptions()->compression_opts.max_dict_bytes > 0) {
+    const size_t kMaxSamples =
+        cfd->ioptions()->compression_opts.max_dict_bytes >> kSampleLenShift;
+    const size_t kOutFileLen = mutable_cf_options->MaxFileSizeForLevel(
+        compact_->compaction->output_level());
+    if (kOutFileLen != port::kMaxSizet) {
+      const size_t kOutFileNumSamples = kOutFileLen >> kSampleLenShift;
+      Random64 generator{versions_->NewFileNumber()};
+      for (size_t i = 0; i < kMaxSamples; ++i) {
+        sample_begin_offsets.insert(generator.Uniform(kOutFileNumSamples)
+                                    << kSampleLenShift);
+      }
+    }
+  }
+
  auto compaction_filter = cfd->ioptions()->compaction_filter;
  std::unique_ptr<CompactionFilter> compaction_filter_from_factory = nullptr;
  if (compaction_filter == nullptr) {
@@ -618,7 +704,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  MergeHelper merge(
      env_, cfd->user_comparator(), cfd->ioptions()->merge_operator,
      compaction_filter, db_options_.info_log.get(),
-      cfd->ioptions()->min_partial_merge_operands,
+      mutable_cf_options->min_partial_merge_operands,
      false /* internal key corruption is expected */,
      existing_snapshots_.empty() ? 0 : existing_snapshots_.back(),
      compact_->compaction->level(), db_options_.statistics.get());
@@ -638,11 +724,18 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  Status status;
  sub_compact->c_iter.reset(new CompactionIterator(
      input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
-      &existing_snapshots_, env_, false, sub_compact->compaction,
-      compaction_filter));
+      &existing_snapshots_, earliest_write_conflict_snapshot_, env_, false,
+      sub_compact->compaction, compaction_filter));
  auto c_iter = sub_compact->c_iter.get();
  c_iter->SeekToFirst();
  const auto& c_iter_stats = c_iter->iter_stats();
+  auto sample_begin_offset_iter = sample_begin_offsets.cbegin();
+  // data_begin_offset and compression_dict are only valid while generating
+  // dictionary from the first output file.
+  size_t data_begin_offset = 0;
+  std::string compression_dict;
+  compression_dict.reserve(cfd->ioptions()->compression_opts.max_dict_bytes);
+
  // TODO(noetzli): check whether we could check !shutting_down_->... only
  // only occasionally (see diff D42687)
  while (status.ok() && !shutting_down_->load(std::memory_order_acquire) &&
@@ -657,7 +750,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    if (end != nullptr &&
        cfd->user_comparator()->Compare(c_iter->user_key(), *end) >= 0) {
      break;
-    } else if (sub_compact->compaction->ShouldStopBefore(key) &&
+    } else if (sub_compact->ShouldStopBefore(key) &&
               sub_compact->builder != nullptr) {
      status = FinishCompactionOutputFile(input->status(), sub_compact);
      if (!status.ok()) {
@@ -686,6 +779,55 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
        key, c_iter->ikey().sequence);
    sub_compact->num_output_records++;

+    if (sub_compact->outputs.size() == 1) {  // first output file
+      // Check if this key/value overlaps any sample intervals; if so, appends
+      // overlapping portions to the dictionary.
+      for (const auto& data_elmt : {key, value}) {
+        size_t data_end_offset = data_begin_offset + data_elmt.size();
+        while (sample_begin_offset_iter != sample_begin_offsets.cend() &&
+               *sample_begin_offset_iter < data_end_offset) {
+          size_t sample_end_offset =
+              *sample_begin_offset_iter + (1 << kSampleLenShift);
+          // Invariant: Because we advance sample iterator while processing the
+          // data_elmt containing the sample's last byte, the current sample
+          // cannot end before the current data_elmt.
+          assert(data_begin_offset < sample_end_offset);
+
+          size_t data_elmt_copy_offset, data_elmt_copy_len;
+          if (*sample_begin_offset_iter <= data_begin_offset) {
+            // The sample starts before data_elmt starts, so take bytes starting
+            // at the beginning of data_elmt.
+            data_elmt_copy_offset = 0;
+          } else {
+            // data_elmt starts before the sample starts, so take bytes starting
+            // at the below offset into data_elmt.
+            data_elmt_copy_offset =
+                *sample_begin_offset_iter - data_begin_offset;
+          }
+          if (sample_end_offset <= data_end_offset) {
+            // The sample ends before data_elmt ends, so take as many bytes as
+            // needed.
+            data_elmt_copy_len =
+                sample_end_offset - (data_begin_offset + data_elmt_copy_offset);
+          } else {
+            // data_elmt ends before the sample ends, so take all remaining
+            // bytes in data_elmt.
+            data_elmt_copy_len =
+                data_end_offset - (data_begin_offset + data_elmt_copy_offset);
+          }
+          compression_dict.append(&data_elmt.data()[data_elmt_copy_offset],
+                                  data_elmt_copy_len);
+          if (sample_end_offset > data_end_offset) {
+            // Didn't finish sample. Try to finish it with the next data_elmt.
+            break;
+          }
+          // Next sample may require bytes from same data_elmt.
+          sample_begin_offset_iter++;
+        }
+        data_begin_offset = data_end_offset;
+      }
+    }
+
    // Close output file if it is big enough
    // TODO(aekmekji): determine if file should be closed earlier than this
    // during subcompactions (i.e. if output size, estimated by input size, is
@@ -694,8 +836,12 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    if (sub_compact->builder->FileSize() >=
        sub_compact->compaction->max_output_file_size()) {
      status = FinishCompactionOutputFile(input->status(), sub_compact);
+      if (sub_compact->outputs.size() == 1) {
+        // Use dictionary from first output file for compression of subsequent
+        // files.
+        sub_compact->compression_dict = std::move(compression_dict);
+      }
    }
-
    c_iter->Next();
  }

@@ -808,9 +954,10 @@ Status CompactionJob::FinishCompactionOutputFile(
  }
  sub_compact->outfile.reset();

+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  TableProperties tp;
  if (s.ok() && current_entries > 0) {
    // Verify that the table is usable
-    ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
    InternalIterator* iter = cfd->table_cache()->NewIterator(
        ReadOptions(), env_options_, cfd->internal_comparator(), meta->fd,
        nullptr, cfd->internal_stats()->GetFileReadHist(
@@ -824,35 +971,51 @@ Status CompactionJob::FinishCompactionOutputFile(
    }

    delete iter;
+
+    // Output to event logger and fire events.
    if (s.ok()) {
-      auto tp = sub_compact->builder->GetTableProperties();
+      tp = sub_compact->builder->GetTableProperties();
      sub_compact->current_output()->table_properties =
          std::make_shared<TableProperties>(tp);
-      TableFileCreationInfo info(std::move(tp));
-      info.db_name = dbname_;
-      info.cf_name = cfd->GetName();
-      info.file_path =
-          TableFileName(cfd->ioptions()->db_paths, meta->fd.GetNumber(),
-                        meta->fd.GetPathId());
-      info.file_size = meta->fd.GetFileSize();
-      info.job_id = job_id_;
      Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
          "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
          " keys, %" PRIu64 " bytes%s",
          cfd->GetName().c_str(), job_id_, output_number, current_entries,
          current_bytes,
          meta->marked_for_compaction ? " (need compaction)" : "");
-      EventHelpers::LogAndNotifyTableFileCreation(
-          event_logger_, cfd->ioptions()->listeners, meta->fd, info);
    }
  }
+  std::string fname = TableFileName(db_options_.db_paths, meta->fd.GetNumber(),
+                                    meta->fd.GetPathId());
+  EventHelpers::LogAndNotifyTableFileCreationFinished(
+      event_logger_, cfd->ioptions()->listeners, dbname_, cfd->GetName(), fname,
+      job_id_, meta->fd, tp, TableFileCreationReason::kCompaction, s);
+
+  // Report new file to SstFileManagerImpl
+  auto sfm =
+      static_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
+  if (sfm && meta->fd.GetPathId() == 0) {
+    auto fn = TableFileName(cfd->ioptions()->db_paths, meta->fd.GetNumber(),
+                            meta->fd.GetPathId());
+    sfm->OnAddFile(fn);
+    if (sfm->IsMaxAllowedSpaceReached()) {
+      InstrumentedMutexLock l(db_mutex_);
+      if (db_bg_error_->ok()) {
+        s = Status::IOError("Max allowed space was reached");
+        *db_bg_error_ = s;
+        TEST_SYNC_POINT(
+            "CompactionJob::FinishCompactionOutputFile:MaxAllowedSpaceReached");
+      }
+    }
+  }
+
  sub_compact->builder.reset();
  return s;
 }

 Status CompactionJob::InstallCompactionResults(
-    const MutableCFOptions& mutable_cf_options, InstrumentedMutex* db_mutex) {
-  db_mutex->AssertHeld();
+    const MutableCFOptions& mutable_cf_options) {
+  db_mutex_->AssertHeld();

  auto* compaction = compact_->compaction;
  // paranoia: verify that the files that we started with
@@ -887,7 +1050,7 @@ Status CompactionJob::InstallCompactionResults(
  }
  return versions_->LogAndApply(compaction->column_family_data(),
                                mutable_cf_options, compaction->edit(),
-                                db_mutex, db_directory_);
+                                db_mutex_, db_directory_);
 }

 void CompactionJob::RecordCompactionIOStats() {
@@ -907,10 +1070,17 @@ Status CompactionJob::OpenCompactionOutputFile(
  assert(sub_compact->builder == nullptr);
  // no need to lock because VersionSet::next_file_number_ is atomic
  uint64_t file_number = versions_->NewFileNumber();
-  // Make the output file
-  unique_ptr<WritableFile> writable_file;
  std::string fname = TableFileName(db_options_.db_paths, file_number,
                                    sub_compact->compaction->output_path_id());
+  // Fire events.
+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+#ifndef ROCKSDB_LITE
+  EventHelpers::NotifyTableFileCreationStarted(
+      cfd->ioptions()->listeners, dbname_, cfd->GetName(), fname, job_id_,
+      TableFileCreationReason::kCompaction);
+#endif  // !ROCKSDB_LITE
+  // Make the output file
+  unique_ptr<WritableFile> writable_file;
  Status s = NewWritableFile(env_, fname, &writable_file, env_options_);
  if (!s.ok()) {
    Log(InfoLogLevel::ERROR_LEVEL, db_options_.info_log,
@@ -919,8 +1089,13 @@ Status CompactionJob::OpenCompactionOutputFile(
        sub_compact->compaction->column_family_data()->GetName().c_str(),
        job_id_, file_number, s.ToString().c_str());
    LogFlush(db_options_.info_log);
+    EventHelpers::LogAndNotifyTableFileCreationFinished(
+        event_logger_, cfd->ioptions()->listeners, dbname_, cfd->GetName(),
+        fname, job_id_, FileDescriptor(), TableProperties(),
+        TableFileCreationReason::kCompaction, s);
    return s;
  }
+
  SubcompactionState::Output out;
  out.meta.fd =
      FileDescriptor(file_number, sub_compact->compaction->output_path_id(), 0);
@@ -933,7 +1108,6 @@ Status CompactionJob::OpenCompactionOutputFile(
  sub_compact->outfile.reset(
      new WritableFileWriter(std::move(writable_file), env_options_));

-  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
  // If the Column family flag is to only optimize filters for hits,
  // we can skip creating filters if this is the bottommost_level where
  // data is going to be found
@@ -941,9 +1115,10 @@ Status CompactionJob::OpenCompactionOutputFile(
      cfd->ioptions()->optimize_filters_for_hits && bottommost_level_;
  sub_compact->builder.reset(NewTableBuilder(
      *cfd->ioptions(), cfd->internal_comparator(),
-      cfd->int_tbl_prop_collector_factories(), cfd->GetID(),
+      cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
      sub_compact->outfile.get(), sub_compact->compaction->output_compression(),
-      cfd->ioptions()->compression_opts, skip_filters));
+      cfd->ioptions()->compression_opts, &sub_compact->compression_dict,
+      skip_filters));
  LogFlush(db_options_.info_log);
  return s;
 }
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -56,8 +56,10 @@ class CompactionJob {
                const EnvOptions& env_options, VersionSet* versions,
                std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
                Directory* db_directory, Directory* output_directory,
-                Statistics* stats,
+                Statistics* stats, InstrumentedMutex* db_mutex,
+                Status* db_bg_error,
                std::vector<SequenceNumber> existing_snapshots,
+                SequenceNumber earliest_write_conflict_snapshot,
                std::shared_ptr<Cache> table_cache, EventLogger* event_logger,
                bool paranoid_file_checks, bool measure_io_stats,
                const std::string& dbname,
@@ -76,8 +78,7 @@ class CompactionJob {
  Status Run();

  // REQUIRED: mutex held
-  Status Install(const MutableCFOptions& mutable_cf_options,
-                 InstrumentedMutex* db_mutex);
+  Status Install(const MutableCFOptions& mutable_cf_options);

 private:
  struct SubcompactionState;
@@ -94,8 +95,7 @@ class CompactionJob {

  Status FinishCompactionOutputFile(const Status& input_status,
                                    SubcompactionState* sub_compact);
-  Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options,
-                                  InstrumentedMutex* db_mutex);
+  Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options);
  void RecordCompactionIOStats();
  Status OpenCompactionOutputFile(SubcompactionState* sub_compact);
  void CleanupCompaction();
@@ -122,6 +122,7 @@ class CompactionJob {
  const std::string& dbname_;
  const DBOptions& db_options_;
  const EnvOptions& env_options_;
+
  Env* env_;
  VersionSet* versions_;
  std::atomic<bool>* shutting_down_;
@@ -129,11 +130,19 @@ class CompactionJob {
  Directory* db_directory_;
  Directory* output_directory_;
  Statistics* stats_;
+  InstrumentedMutex* db_mutex_;
+  Status* db_bg_error_;
  // If there were two snapshots with seq numbers s1 and
  // s2 and s1 < s2, and if we find two instances of a key k1 then lies
  // entirely within s1 and s2, then the earlier version of k1 can be safely
  // deleted because that version is not visible in any snapshot.
  std::vector<SequenceNumber> existing_snapshots_;
+
+  // This is the earliest snapshot that could be used for write-conflict
+  // checking by a transaction.  For any user-key newer than this snapshot, we
+  // should make sure not to remove evidence that a write occurred.
+  SequenceNumber earliest_write_conflict_snapshot_;
+
  std::shared_ptr<Cache> table_cache_;

  EventLogger* event_logger_;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -551,8 +551,9 @@ uint64_t EstimatedFileSize(
  const size_t kFooterSize = 512;

  uint64_t data_size =
+    static_cast<uint64_t>(
      num_records * (key_size + value_size * compression_ratio +
-                     kPerKeyOverhead);
+                     kPerKeyOverhead));

  return data_size + kFooterSize
         + num_records * bloom_bits_per_key / 8      // filter block
@@ -623,7 +624,10 @@ CompressionType GetAnyCompression() {
    return kBZip2Compression;
  } else if (LZ4_Supported()) {
    return kLZ4Compression;
+  } else if (XPRESS_Supported()) {
+    return kXpressCompression;
  }
+
  return kNoCompression;
 }

@@ -659,7 +663,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
  options.max_subcompactions = max_subcompactions_;
  options.bytes_per_sync = 512 * 1024;

-  options.compaction_measure_io_stats = true;
+  options.report_bg_io_stats = true;
  for (int test = 0; test < 2; ++test) {
    DestroyAndReopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -162,9 +162,12 @@ class CompactionJobTest : public testing::Test {
        auto key = ToString(i * kMatchingKeys + k);
        auto value = ToString(i * kKeysPerFile + k);
        InternalKey internal_key(key, ++sequence_number, kTypeValue);
+
        // This is how the key will look like once it's written in bottommost
        // file
-        InternalKey bottommost_internal_key(key, 0, kTypeValue);
+        InternalKey bottommost_internal_key(
+            key, (key == "9999") ? sequence_number : 0, kTypeValue);
+
        if (corrupt_id(k)) {
          test::CorruptKeyType(&internal_key);
          test::CorruptKeyType(&bottommost_internal_key);
@@ -217,9 +220,11 @@ class CompactionJobTest : public testing::Test {
    cfd_ = versions_->GetColumnFamilySet()->GetDefault();
  }

-  void RunCompaction(const std::vector<std::vector<FileMetaData*>>& input_files,
-                     const stl_wrappers::KVMap& expected_results,
-                     const std::vector<SequenceNumber>& snapshots = {}) {
+  void RunCompaction(
+      const std::vector<std::vector<FileMetaData*>>& input_files,
+      const stl_wrappers::KVMap& expected_results,
+      const std::vector<SequenceNumber>& snapshots = {},
+      SequenceNumber earliest_write_conflict_snapshot = kMaxSequenceNumber) {
    auto cfd = versions_->GetColumnFamilySet()->GetDefault();

    size_t num_input_files = 0;
@@ -243,11 +248,11 @@ class CompactionJobTest : public testing::Test {
    LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
    mutex_.Lock();
    EventLogger event_logger(db_options_.info_log.get());
-    CompactionJob compaction_job(0, &compaction, db_options_, env_options_,
-                                 versions_.get(), &shutting_down_, &log_buffer,
-                                 nullptr, nullptr, nullptr, snapshots,
-                                 table_cache_, &event_logger, false, false,
-                                 dbname_, &compaction_job_stats_);
+    CompactionJob compaction_job(
+        0, &compaction, db_options_, env_options_, versions_.get(),
+        &shutting_down_, &log_buffer, nullptr, nullptr, nullptr, &mutex_,
+        &bg_error_, snapshots, earliest_write_conflict_snapshot, table_cache_,
+        &event_logger, false, false, dbname_, &compaction_job_stats_);

    VerifyInitializationOfCompactionJobStats(compaction_job_stats_);

@@ -257,8 +262,7 @@ class CompactionJobTest : public testing::Test {
    s = compaction_job.Run();
    ASSERT_OK(s);
    mutex_.Lock();
-    ASSERT_OK(compaction_job.Install(*cfd->GetLatestMutableCFOptions(),
-                                     &mutex_));
+    ASSERT_OK(compaction_job.Install(*cfd->GetLatestMutableCFOptions()));
    mutex_.Unlock();

    if (expected_results.size() == 0) {
@@ -290,6 +294,7 @@ class CompactionJobTest : public testing::Test {
  ColumnFamilyData* cfd_;
  std::unique_ptr<CompactionFilter> compaction_filter_;
  std::shared_ptr<MergeOperator> merge_op_;
+  Status bg_error_;
 };

 TEST_F(CompactionJobTest, Simple) {
@@ -346,7 +351,7 @@ TEST_F(CompactionJobTest, SimpleOverwrite) {

  auto expected_results =
      mock::MakeMockFile({{KeyStr("a", 0U, kTypeValue), "val2"},
-                          {KeyStr("b", 0U, kTypeValue), "val3"}});
+                          {KeyStr("b", 4U, kTypeValue), "val3"}});

  SetLastSequence(4U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -399,7 +404,7 @@ TEST_F(CompactionJobTest, SimpleMerge) {

  auto expected_results =
      mock::MakeMockFile({{KeyStr("a", 0U, kTypeValue), "3,4,5"},
-                          {KeyStr("b", 0U, kTypeValue), "1,2"}});
+                          {KeyStr("b", 2U, kTypeValue), "1,2"}});

  SetLastSequence(5U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -559,34 +564,144 @@ TEST_F(CompactionJobTest, SimpleSingleDelete) {
 TEST_F(CompactionJobTest, SingleDeleteSnapshots) {
  NewDB();

-  auto file1 = mock::MakeMockFile({{KeyStr("A", 12U, kTypeSingleDeletion), ""},
-                                   {KeyStr("a", 12U, kTypeSingleDeletion), ""},
-                                   {KeyStr("b", 21U, kTypeSingleDeletion), ""},
-                                   {KeyStr("c", 22U, kTypeSingleDeletion), ""},
-                                   {KeyStr("d", 9U, kTypeSingleDeletion), ""}});
+  auto file1 = mock::MakeMockFile({
+      {KeyStr("A", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("a", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("b", 21U, kTypeSingleDeletion), ""},
+      {KeyStr("c", 22U, kTypeSingleDeletion), ""},
+      {KeyStr("d", 9U, kTypeSingleDeletion), ""},
+      {KeyStr("f", 21U, kTypeSingleDeletion), ""},
+      {KeyStr("j", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("j", 9U, kTypeSingleDeletion), ""},
+      {KeyStr("k", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("k", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("l", 3U, kTypeSingleDeletion), ""},
+      {KeyStr("l", 2U, kTypeSingleDeletion), ""},
+  });
  AddMockFile(file1);

-  auto file2 = mock::MakeMockFile({{KeyStr("0", 2U, kTypeSingleDeletion), ""},
-                                   {KeyStr("a", 11U, kTypeValue), "val1"},
-                                   {KeyStr("b", 11U, kTypeValue), "val2"},
-                                   {KeyStr("c", 21U, kTypeValue), "val3"},
-                                   {KeyStr("d", 8U, kTypeValue), "val4"},
-                                   {KeyStr("e", 2U, kTypeSingleDeletion), ""}});
+  auto file2 = mock::MakeMockFile({
+      {KeyStr("0", 2U, kTypeSingleDeletion), ""},
+      {KeyStr("a", 11U, kTypeValue), "val1"},
+      {KeyStr("b", 11U, kTypeValue), "val2"},
+      {KeyStr("c", 21U, kTypeValue), "val3"},
+      {KeyStr("d", 8U, kTypeValue), "val4"},
+      {KeyStr("e", 2U, kTypeSingleDeletion), ""},
+      {KeyStr("f", 1U, kTypeValue), "val1"},
+      {KeyStr("g", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("h", 2U, kTypeSingleDeletion), ""},
+      {KeyStr("m", 12U, kTypeValue), "val1"},
+      {KeyStr("m", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("m", 8U, kTypeValue), "val2"},
+  });
  AddMockFile(file2);

-  auto file3 = mock::MakeMockFile({{KeyStr("A", 1U, kTypeValue), "val"},
-                                   {KeyStr("e", 1U, kTypeValue), "val"}});
+  auto file3 = mock::MakeMockFile({
+      {KeyStr("A", 1U, kTypeValue), "val"},
+      {KeyStr("e", 1U, kTypeValue), "val"},
+  });
  AddMockFile(file3, 2);

-  auto expected_results =
-      mock::MakeMockFile({{KeyStr("A", 12U, kTypeSingleDeletion), ""},
-                          {KeyStr("b", 21U, kTypeSingleDeletion), ""},
-                          {KeyStr("b", 11U, kTypeValue), "val2"},
-                          {KeyStr("e", 2U, kTypeSingleDeletion), ""}});
+  auto expected_results = mock::MakeMockFile({
+      {KeyStr("A", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("a", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("a", 11U, kTypeValue), ""},
+      {KeyStr("b", 21U, kTypeSingleDeletion), ""},
+      {KeyStr("b", 11U, kTypeValue), "val2"},
+      {KeyStr("c", 22U, kTypeSingleDeletion), ""},
+      {KeyStr("c", 21U, kTypeValue), ""},
+      {KeyStr("e", 2U, kTypeSingleDeletion), ""},
+      {KeyStr("f", 21U, kTypeSingleDeletion), ""},
+      {KeyStr("f", 1U, kTypeValue), "val1"},
+      {KeyStr("g", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("j", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("k", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("m", 12U, kTypeValue), "val1"},
+      {KeyStr("m", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("m", 8U, kTypeValue), "val2"},
+  });

  SetLastSequence(22U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
-  RunCompaction({files}, expected_results, {10U, 20U});
+  RunCompaction({files}, expected_results, {10U, 20U}, 10U);
+}
+
+TEST_F(CompactionJobTest, EarliestWriteConflictSnapshot) {
+  NewDB();
+
+  // Test multiple snapshots where the earliest snapshot is not a
+  // write-conflic-snapshot.
+
+  auto file1 = mock::MakeMockFile({
+      {KeyStr("A", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("A", 23U, kTypeValue), "val"},
+      {KeyStr("B", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("B", 23U, kTypeValue), "val"},
+      {KeyStr("D", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 32U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 31U, kTypeValue), "val"},
+      {KeyStr("G", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 23U, kTypeValue), "val2"},
+      {KeyStr("H", 31U, kTypeValue), "val"},
+      {KeyStr("H", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("H", 23U, kTypeValue), "val"},
+      {KeyStr("I", 35U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 34U, kTypeValue), "val2"},
+      {KeyStr("I", 33U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 32U, kTypeValue), "val3"},
+      {KeyStr("I", 31U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 34U, kTypeValue), "val"},
+      {KeyStr("J", 33U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 25U, kTypeValue), "val2"},
+      {KeyStr("J", 24U, kTypeSingleDeletion), ""},
+  });
+  AddMockFile(file1);
+
+  auto file2 = mock::MakeMockFile({
+      {KeyStr("A", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("A", 13U, kTypeValue), "val2"},
+      {KeyStr("C", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("C", 13U, kTypeValue), "val"},
+      {KeyStr("E", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("F", 4U, kTypeSingleDeletion), ""},
+      {KeyStr("F", 3U, kTypeValue), "val"},
+      {KeyStr("G", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 13U, kTypeValue), "val3"},
+      {KeyStr("H", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("H", 13U, kTypeValue), "val2"},
+      {KeyStr("I", 13U, kTypeValue), "val4"},
+      {KeyStr("I", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 11U, kTypeValue), "val5"},
+      {KeyStr("J", 15U, kTypeValue), "val3"},
+      {KeyStr("J", 14U, kTypeSingleDeletion), ""},
+  });
+  AddMockFile(file2);
+
+  auto expected_results = mock::MakeMockFile({
+      {KeyStr("A", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("A", 23U, kTypeValue), ""},
+      {KeyStr("B", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("B", 23U, kTypeValue), ""},
+      {KeyStr("D", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("E", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 32U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 31U, kTypeValue), ""},
+      {KeyStr("H", 31U, kTypeValue), "val"},
+      {KeyStr("I", 35U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 34U, kTypeValue), ""},
+      {KeyStr("I", 31U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 13U, kTypeValue), "val4"},
+      {KeyStr("J", 34U, kTypeValue), "val"},
+      {KeyStr("J", 33U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 25U, kTypeValue), "val2"},
+      {KeyStr("J", 24U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 15U, kTypeValue), "val3"},
+      {KeyStr("J", 14U, kTypeSingleDeletion), ""},
+  });
+
+  SetLastSequence(24U);
+  auto files = cfd_->current()->storage_info()->LevelFiles(0);
+  RunCompaction({files}, expected_results, {10U, 20U, 30U}, 20U);
 }

 TEST_F(CompactionJobTest, SingleDeleteZeroSeq) {
@@ -604,7 +719,7 @@ TEST_F(CompactionJobTest, SingleDeleteZeroSeq) {
  AddMockFile(file2);

  auto expected_results = mock::MakeMockFile({
-      {KeyStr("dummy", 0U, kTypeValue), "val2"},
+      {KeyStr("dummy", 5U, kTypeValue), "val2"},
  });

  SetLastSequence(22U);
@@ -616,9 +731,22 @@ TEST_F(CompactionJobTest, MultiSingleDelete) {
  // Tests three scenarios involving multiple single delete/put pairs:
  //
  // A: Put Snapshot SDel Put SDel -> Put Snapshot SDel
-  // B: Put SDel Put SDel -> (Removed)
+  // B: Snapshot Put SDel Put SDel Snapshot -> Snapshot SDel Snapshot
  // C: SDel Put SDel Snapshot Put -> Snapshot Put
-  // D: (Put) SDel Snapshot Put SDel -> (Put) SDel Snapshot
+  // D: (Put) SDel Snapshot Put SDel -> (Put) SDel Snapshot SDel
+  // E: Put SDel Snapshot Put SDel -> Snapshot SDel
+  // F: Put SDel Put Sdel Snapshot -> removed
+  // G: Snapshot SDel Put SDel Put -> Snapshot Put SDel
+  // H: (Put) Put SDel Put Sdel Snapshot -> Removed
+  // I: (Put) Snapshot Put SDel Put SDel -> SDel
+  // J: Put Put SDel Put SDel SDel Snapshot Put Put SDel SDel Put
+  //      -> Snapshot Put
+  // K: SDel SDel Put SDel Put Put Snapshot SDel Put SDel SDel Put SDel
+  //      -> Snapshot Put Snapshot SDel
+  // L: SDel Put Del Put SDel Snapshot Del Put Del SDel Put SDel
+  //      -> Snapshot SDel
+  // M: (Put) SDel Put Del Put SDel Snapshot Put Del SDel Put SDel Del
+  //      -> SDel Snapshot Del
  NewDB();

  auto file1 = mock::MakeMockFile({
@@ -630,6 +758,34 @@ TEST_F(CompactionJobTest, MultiSingleDelete) {
      {KeyStr("C", 14U, kTypeValue), "val3"},
      {KeyStr("D", 12U, kTypeSingleDeletion), ""},
      {KeyStr("D", 11U, kTypeValue), "val4"},
+      {KeyStr("G", 15U, kTypeValue), "val"},
+      {KeyStr("G", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("G", 13U, kTypeValue), "val"},
+      {KeyStr("I", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 13U, kTypeValue), "val"},
+      {KeyStr("J", 15U, kTypeValue), "val"},
+      {KeyStr("J", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 13U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 12U, kTypeValue), "val"},
+      {KeyStr("J", 11U, kTypeValue), "val"},
+      {KeyStr("K", 16U, kTypeSingleDeletion), ""},
+      {KeyStr("K", 15U, kTypeValue), "val1"},
+      {KeyStr("K", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("K", 13U, kTypeSingleDeletion), ""},
+      {KeyStr("K", 12U, kTypeValue), "val2"},
+      {KeyStr("K", 11U, kTypeSingleDeletion), ""},
+      {KeyStr("L", 16U, kTypeSingleDeletion), ""},
+      {KeyStr("L", 15U, kTypeValue), "val"},
+      {KeyStr("L", 14U, kTypeSingleDeletion), ""},
+      {KeyStr("L", 13U, kTypeDeletion), ""},
+      {KeyStr("L", 12U, kTypeValue), "val"},
+      {KeyStr("L", 11U, kTypeDeletion), ""},
+      {KeyStr("M", 16U, kTypeDeletion), ""},
+      {KeyStr("M", 15U, kTypeSingleDeletion), ""},
+      {KeyStr("M", 14U, kTypeValue), "val"},
+      {KeyStr("M", 13U, kTypeSingleDeletion), ""},
+      {KeyStr("M", 12U, kTypeDeletion), ""},
+      {KeyStr("M", 11U, kTypeValue), "val"},
  });
  AddMockFile(file1);

@@ -641,24 +797,87 @@ TEST_F(CompactionJobTest, MultiSingleDelete) {
      {KeyStr("C", 9U, kTypeValue), "val6"},
      {KeyStr("C", 8U, kTypeSingleDeletion), ""},
      {KeyStr("D", 10U, kTypeSingleDeletion), ""},
+      {KeyStr("E", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("E", 11U, kTypeValue), "val"},
+      {KeyStr("E", 5U, kTypeSingleDeletion), ""},
+      {KeyStr("E", 4U, kTypeValue), "val"},
+      {KeyStr("F", 6U, kTypeSingleDeletion), ""},
+      {KeyStr("F", 5U, kTypeValue), "val"},
+      {KeyStr("F", 4U, kTypeSingleDeletion), ""},
+      {KeyStr("F", 3U, kTypeValue), "val"},
+      {KeyStr("G", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("H", 6U, kTypeSingleDeletion), ""},
+      {KeyStr("H", 5U, kTypeValue), "val"},
+      {KeyStr("H", 4U, kTypeSingleDeletion), ""},
+      {KeyStr("H", 3U, kTypeValue), "val"},
+      {KeyStr("I", 12U, kTypeSingleDeletion), ""},
+      {KeyStr("I", 11U, kTypeValue), "val"},
+      {KeyStr("J", 6U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 5U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 4U, kTypeValue), "val"},
+      {KeyStr("J", 3U, kTypeSingleDeletion), ""},
+      {KeyStr("J", 2U, kTypeValue), "val"},
+      {KeyStr("K", 8U, kTypeValue), "val3"},
+      {KeyStr("K", 7U, kTypeValue), "val4"},
+      {KeyStr("K", 6U, kTypeSingleDeletion), ""},
+      {KeyStr("K", 5U, kTypeValue), "val5"},
+      {KeyStr("K", 2U, kTypeSingleDeletion), ""},
+      {KeyStr("K", 1U, kTypeSingleDeletion), ""},
+      {KeyStr("L", 5U, kTypeSingleDeletion), ""},
+      {KeyStr("L", 4U, kTypeValue), "val"},
+      {KeyStr("L", 3U, kTypeDeletion), ""},
+      {KeyStr("L", 2U, kTypeValue), "val"},
+      {KeyStr("L", 1U, kTypeSingleDeletion), ""},
+      {KeyStr("M", 10U, kTypeSingleDeletion), ""},
+      {KeyStr("M", 7U, kTypeValue), "val"},
+      {KeyStr("M", 5U, kTypeDeletion), ""},
+      {KeyStr("M", 4U, kTypeValue), "val"},
+      {KeyStr("M", 3U, kTypeSingleDeletion), ""},
  });
  AddMockFile(file2);

  auto file3 = mock::MakeMockFile({
-      {KeyStr("D", 11U, kTypeValue), "val"},
+      {KeyStr("D", 1U, kTypeValue), "val"},
+      {KeyStr("H", 1U, kTypeValue), "val"},
+      {KeyStr("I", 2U, kTypeValue), "val"},
  });
  AddMockFile(file3, 2);

-  auto expected_results = mock::MakeMockFile({
-      {KeyStr("A", 12U, kTypeSingleDeletion), ""},
-      {KeyStr("A", 10U, kTypeValue), "val"},
-      {KeyStr("C", 14U, kTypeValue), "val3"},
-      {KeyStr("D", 10U, kTypeSingleDeletion), ""},
+  auto file4 = mock::MakeMockFile({
+      {KeyStr("M", 1U, kTypeValue), "val"},
  });
+  AddMockFile(file4, 2);
+
+  auto expected_results =
+      mock::MakeMockFile({{KeyStr("A", 14U, kTypeSingleDeletion), ""},
+                          {KeyStr("A", 13U, kTypeValue), ""},
+                          {KeyStr("A", 12U, kTypeSingleDeletion), ""},
+                          {KeyStr("A", 10U, kTypeValue), "val"},
+                          {KeyStr("B", 14U, kTypeSingleDeletion), ""},
+                          {KeyStr("B", 13U, kTypeValue), ""},
+                          {KeyStr("C", 14U, kTypeValue), "val3"},
+                          {KeyStr("D", 12U, kTypeSingleDeletion), ""},
+                          {KeyStr("D", 11U, kTypeValue), ""},
+                          {KeyStr("D", 10U, kTypeSingleDeletion), ""},
+                          {KeyStr("E", 12U, kTypeSingleDeletion), ""},
+                          {KeyStr("E", 11U, kTypeValue), ""},
+                          {KeyStr("G", 15U, kTypeValue), "val"},
+                          {KeyStr("G", 12U, kTypeSingleDeletion), ""},
+                          {KeyStr("I", 14U, kTypeSingleDeletion), ""},
+                          {KeyStr("I", 13U, kTypeValue), ""},
+                          {KeyStr("J", 15U, kTypeValue), "val"},
+                          {KeyStr("K", 16U, kTypeSingleDeletion), ""},
+                          {KeyStr("K", 15U, kTypeValue), ""},
+                          {KeyStr("K", 11U, kTypeSingleDeletion), ""},
+                          {KeyStr("K", 8U, kTypeValue), "val3"},
+                          {KeyStr("L", 16U, kTypeSingleDeletion), ""},
+                          {KeyStr("L", 15U, kTypeValue), ""},
+                          {KeyStr("M", 16U, kTypeDeletion), ""},
+                          {KeyStr("M", 3U, kTypeSingleDeletion), ""}});

  SetLastSequence(22U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
-  RunCompaction({files}, expected_results, {10U});
+  RunCompaction({files}, expected_results, {10U}, 10U);
 }

 // This test documents the behavior where a corrupt key follows a deletion or a
@@ -684,7 +903,7 @@ TEST_F(CompactionJobTest, CorruptionAfterDeletion) {
      mock::MakeMockFile({{test::KeyStr("A", 0U, kTypeValue), "val3"},
                          {test::KeyStr("a", 0U, kTypeValue, true), "val"},
                          {test::KeyStr("b", 0U, kTypeValue, true), "val"},
-                          {test::KeyStr("c", 0U, kTypeValue), "val2"}});
+                          {test::KeyStr("c", 1U, kTypeValue), "val2"}});

  SetLastSequence(6U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -70,7 +70,8 @@ struct UserKeyComparator {
 };

 typedef std::priority_queue<InputFileInfo, std::vector<InputFileInfo>,
-                            UserKeyComparator> SmallestKeyHeap;
+                            UserKeyComparator>
+    SmallestKeyHeap;

 // This function creates the heap that is used to find if the files are
 // overlapping during universal compaction when the allow_trivial_move
@@ -109,13 +110,22 @@ SmallestKeyHeap create_level_heap(Compaction* c, const Comparator* ucmp) {
 // matter what the values of the other two parameters are.
 // Otherwise, the compression type is determined based on options and level.
 CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
+                                   const VersionStorageInfo* vstorage,
+                                   const MutableCFOptions& mutable_cf_options,
                                   int level, int base_level,
                                   const bool enable_compression) {
  if (!enable_compression) {
    // disable compression
    return kNoCompression;
  }
-  // If the use has specified a different compression level for each level,
+
+  // If bottommost_compression is set and we are compacting to the
+  // bottommost level then we should use it.
+  if (ioptions.bottommost_compression != kDisableCompressionOption &&
+      level > base_level && level >= (vstorage->num_non_empty_levels() - 1)) {
+    return ioptions.bottommost_compression;
+  }
+  // If the user has specified a different compression level for each level,
  // then pick the compression for that level.
  if (!ioptions.compression_per_level.empty()) {
    assert(level == 0 || level >= base_level);
@@ -129,7 +139,7 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
    // specified compression levels, use the last value.
    return ioptions.compression_per_level[std::max(0, std::min(idx, n))];
  } else {
-    return ioptions.compression;
+    return mutable_cf_options.compression;
  }
 }

@@ -141,7 +151,8 @@ CompactionPicker::~CompactionPicker() {}

 // Delete this compaction from the list of running compactions.
 void CompactionPicker::ReleaseCompactionFiles(Compaction* c, Status status) {
-  if (c->start_level() == 0) {
+  if (c->start_level() == 0 ||
+      ioptions_.compaction_style == kCompactionStyleUniversal) {
    level0_compactions_in_progress_.erase(c);
  }
  if (!status.ok()) {
@@ -189,10 +200,9 @@ void CompactionPicker::GetRange(const CompactionInputFiles& inputs1,
    InternalKey smallest1, smallest2, largest1, largest2;
    GetRange(inputs1, &smallest1, &largest1);
    GetRange(inputs2, &smallest2, &largest2);
-    *smallest = icmp_->Compare(smallest1, smallest2) < 0 ?
-                smallest1 : smallest2;
-    *largest = icmp_->Compare(largest1, largest2) < 0 ?
-               largest2 : largest1;
+    *smallest =
+        icmp_->Compare(smallest1, smallest2) < 0 ? smallest1 : smallest2;
+    *largest = icmp_->Compare(largest1, largest2) < 0 ? largest2 : largest1;
  }
 }

@@ -243,7 +253,7 @@ bool CompactionPicker::ExpandWhileOverlapping(const std::string& cf_name,
 // Returns true if any one of specified files are being compacted
 bool CompactionPicker::FilesInCompaction(
    const std::vector<FileMetaData*>& files) {
-  for (unsigned int i = 0; i < files.size(); i++) {
+  for (size_t i = 0; i < files.size(); i++) {
    if (files[i]->being_compacted) {
      return true;
    }
@@ -257,9 +267,9 @@ Compaction* CompactionPicker::FormCompaction(
    VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options,
    uint32_t output_path_id) {
  uint64_t max_grandparent_overlap_bytes =
-      output_level + 1 < vstorage->num_levels() ?
-          mutable_cf_options.MaxGrandParentOverlapBytes(output_level + 1) :
-          std::numeric_limits<uint64_t>::max();
+      output_level + 1 < vstorage->num_levels()
+          ? mutable_cf_options.MaxGrandParentOverlapBytes(output_level + 1)
+          : std::numeric_limits<uint64_t>::max();
  assert(input_files.size());

  // TODO(rven ): we might be able to run concurrent level 0 compaction
@@ -283,8 +293,7 @@ Compaction* CompactionPicker::FormCompaction(

 Status CompactionPicker::GetCompactionInputsFromFileNumbers(
    std::vector<CompactionInputFiles>* input_files,
-    std::unordered_set<uint64_t>* input_set,
-    const VersionStorageInfo* vstorage,
+    std::unordered_set<uint64_t>* input_set, const VersionStorageInfo* vstorage,
    const CompactionOptions& compact_options) const {
  if (input_set->size() == 0U) {
    return Status::InvalidArgument(
@@ -322,8 +331,8 @@ Status CompactionPicker::GetCompactionInputsFromFileNumbers(
    return Status::InvalidArgument(message);
  }

-  for (int level = first_non_empty_level;
-       level <= last_non_empty_level; ++level) {
+  for (int level = first_non_empty_level; level <= last_non_empty_level;
+       ++level) {
    matched_input_files[level].level = level;
    input_files->emplace_back(std::move(matched_input_files[level]));
  }
@@ -331,8 +340,6 @@ Status CompactionPicker::GetCompactionInputsFromFileNumbers(
  return Status::OK();
 }

-
-
 // Returns true if any one of the parent files are being compacted
 bool CompactionPicker::RangeInCompaction(VersionStorageInfo* vstorage,
                                         const InternalKey* smallest,
@@ -455,7 +462,7 @@ Compaction* CompactionPicker::CompactRange(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    VersionStorageInfo* vstorage, int input_level, int output_level,
    uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
-    InternalKey** compaction_end) {
+    InternalKey** compaction_end, bool* manual_conflict) {
  // CompactionPickerFIFO has its own implementation of compact range
  assert(ioptions_.compaction_style != kCompactionStyleFIFO);

@@ -481,6 +488,12 @@ Compaction* CompactionPicker::CompactRange(
      return nullptr;
    }

+    if ((start_level == 0) && (!level0_compactions_in_progress_.empty())) {
+      *manual_conflict = true;
+      // Only one level 0 compaction allowed
+      return nullptr;
+    }
+
    std::vector<CompactionInputFiles> inputs(vstorage->num_levels() -
                                             start_level);
    for (int level = start_level; level < vstorage->num_levels(); level++) {
@@ -489,13 +502,22 @@ Compaction* CompactionPicker::CompactRange(
      for (FileMetaData* f : vstorage->LevelFiles(level)) {
        files.push_back(f);
      }
+      if (FilesInCompaction(files)) {
+        *manual_conflict = true;
+        return nullptr;
+      }
    }
-    return new Compaction(
+    Compaction* c = new Compaction(
        vstorage, mutable_cf_options, std::move(inputs), output_level,
        mutable_cf_options.MaxFileSizeForLevel(output_level),
        /* max_grandparent_overlap_bytes */ LLONG_MAX, output_path_id,
-        GetCompressionType(ioptions_, output_level, 1),
+        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
+                           output_level, 1),
        /* grandparents */ {}, /* is manual */ true);
+    if (start_level == 0) {
+      level0_compactions_in_progress_.insert(c);
+    }
+    return c;
  }

  CompactionInputFiles inputs;
@@ -514,13 +536,20 @@ Compaction* CompactionPicker::CompactRange(
    return nullptr;
  }

+  if ((input_level == 0) && (!level0_compactions_in_progress_.empty())) {
+    // Only one level 0 compaction allowed
+    TEST_SYNC_POINT("CompactionPicker::CompactRange:Conflict");
+    *manual_conflict = true;
+    return nullptr;
+  }
+
  // Avoid compacting too much in one shot in case the range is large.
  // But we cannot do this for level-0 since level-0 files can overlap
  // and we must not pick one file and drop another older file if the
  // two files overlap.
  if (input_level > 0) {
    const uint64_t limit = mutable_cf_options.MaxFileSizeForLevel(input_level) *
-      mutable_cf_options.source_compaction_factor;
+                           mutable_cf_options.source_compaction_factor;
    uint64_t total = 0;
    for (size_t i = 0; i + 1 < inputs.size(); ++i) {
      uint64_t s = inputs[i]->compensated_file_size;
@@ -536,9 +565,10 @@ Compaction* CompactionPicker::CompactRange(
  assert(output_path_id < static_cast<uint32_t>(ioptions_.db_paths.size()));

  if (ExpandWhileOverlapping(cf_name, vstorage, &inputs) == false) {
-    // manual compaction is currently single-threaded, so it should never
+    // manual compaction is now multi-threaded, so it can
    // happen that ExpandWhileOverlapping fails
-    assert(false);
+    // we handle it higher in RunManualCompaction
+    *manual_conflict = true;
    return nullptr;
  }

@@ -557,9 +587,10 @@ Compaction* CompactionPicker::CompactRange(
    int parent_index = -1;
    if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage, &inputs,
                          &output_level_inputs, &parent_index, -1)) {
-      // manual compaction is currently single-threaded, so it should never
+      // manual compaction is now multi-threaded, so it can
      // happen that SetupOtherInputs fails
-      assert(false);
+      // we handle it higher in RunManualCompaction
+      *manual_conflict = true;
      return nullptr;
    }
  }
@@ -568,6 +599,12 @@ Compaction* CompactionPicker::CompactRange(
  if (!output_level_inputs.empty()) {
    compaction_inputs.push_back(output_level_inputs);
  }
+  for (size_t i = 0; i < compaction_inputs.size(); i++) {
+    if (FilesInCompaction(compaction_inputs[i].files)) {
+      *manual_conflict = true;
+      return nullptr;
+    }
+  }

  std::vector<FileMetaData*> grandparents;
  GetGrandparents(vstorage, inputs, output_level_inputs, &grandparents);
@@ -576,19 +613,29 @@ Compaction* CompactionPicker::CompactRange(
      mutable_cf_options.MaxFileSizeForLevel(output_level),
      mutable_cf_options.MaxGrandParentOverlapBytes(input_level),
      output_path_id,
-      GetCompressionType(ioptions_, output_level, vstorage->base_level()),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         vstorage->base_level()),
      std::move(grandparents), /* is manual compaction */ true);

  TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
+  if (input_level == 0) {
+    level0_compactions_in_progress_.insert(compaction);
+  }
+
+  // Creating a compaction influences the compaction score because the score
+  // takes running compactions into account (by skipping files that are already
+  // being compacted). Since we just changed compaction score, we recalculate it
+  // here
+  vstorage->ComputeCompactionScore(mutable_cf_options);
+
  return compaction;
 }

 #ifndef ROCKSDB_LITE
 namespace {
 // Test whether two files have overlapping key-ranges.
-bool HaveOverlappingKeyRanges(
-    const Comparator* c,
-    const SstFileMetaData& a, const SstFileMetaData& b) {
+bool HaveOverlappingKeyRanges(const Comparator* c, const SstFileMetaData& a,
+                              const SstFileMetaData& b) {
  if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
    if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
      // b.smallestkey <= a.smallestkey <= b.largestkey
@@ -612,9 +659,8 @@ bool HaveOverlappingKeyRanges(
 }  // namespace

 Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
-      std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const {
+    std::unordered_set<uint64_t>* input_files,
+    const ColumnFamilyMetaData& cf_meta, const int output_level) const {
  auto& levels = cf_meta.levels;
  auto comparator = icmp_->user_comparator();

@@ -667,18 +713,17 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
      // has overlapping key-range with other non-compaction input
      // files in the same level.
      while (first_included > 0) {
-        if (comparator->Compare(
-                current_files[first_included - 1].largestkey,
-                current_files[first_included].smallestkey) < 0) {
+        if (comparator->Compare(current_files[first_included - 1].largestkey,
+                                current_files[first_included].smallestkey) <
+            0) {
          break;
        }
        first_included--;
      }

      while (last_included < static_cast<int>(current_files.size()) - 1) {
-        if (comparator->Compare(
-                current_files[last_included + 1].smallestkey,
-                current_files[last_included].largestkey) > 0) {
+        if (comparator->Compare(current_files[last_included + 1].smallestkey,
+                                current_files[last_included].largestkey) > 0) {
          break;
        }
        last_included++;
@@ -688,33 +733,31 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
    // include all files between the first and the last compaction input files.
    for (int f = first_included; f <= last_included; ++f) {
      if (current_files[f].being_compacted) {
-        return Status::Aborted(
-            "Necessary compaction input file " + current_files[f].name +
-            " is currently being compacted.");
+        return Status::Aborted("Necessary compaction input file " +
+                               current_files[f].name +
+                               " is currently being compacted.");
      }
-      input_files->insert(
-          TableFileNameToNumber(current_files[f].name));
+      input_files->insert(TableFileNameToNumber(current_files[f].name));
    }

    // update smallest and largest key
    if (l == 0) {
      for (int f = first_included; f <= last_included; ++f) {
-        if (comparator->Compare(
-            smallestkey, current_files[f].smallestkey) > 0) {
+        if (comparator->Compare(smallestkey, current_files[f].smallestkey) >
+            0) {
          smallestkey = current_files[f].smallestkey;
        }
-        if (comparator->Compare(
-            largestkey, current_files[f].largestkey) < 0) {
+        if (comparator->Compare(largestkey, current_files[f].largestkey) < 0) {
          largestkey = current_files[f].largestkey;
        }
      }
    } else {
-      if (comparator->Compare(
-          smallestkey, current_files[first_included].smallestkey) > 0) {
+      if (comparator->Compare(smallestkey,
+                              current_files[first_included].smallestkey) > 0) {
        smallestkey = current_files[first_included].smallestkey;
      }
-      if (comparator->Compare(
-          largestkey, current_files[last_included].largestkey) < 0) {
+      if (comparator->Compare(largestkey,
+                              current_files[last_included].largestkey) < 0) {
        largestkey = current_files[last_included].largestkey;
      }
    }
@@ -731,16 +774,15 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
    // time and not by key
    for (int m = std::max(l, 1); m <= output_level; ++m) {
      for (auto& next_lv_file : levels[m].files) {
-        if (HaveOverlappingKeyRanges(
-            comparator, aggregated_file_meta, next_lv_file)) {
+        if (HaveOverlappingKeyRanges(comparator, aggregated_file_meta,
+                                     next_lv_file)) {
          if (next_lv_file.being_compacted) {
            return Status::Aborted(
                "File " + next_lv_file.name +
                " that has overlapping key range with one of the compaction "
                " input file is currently being compacted.");
          }
-          input_files->insert(
-              TableFileNameToNumber(next_lv_file.name));
+          input_files->insert(TableFileNameToNumber(next_lv_file.name));
        }
      }
    }
@@ -750,28 +792,25 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(

 Status CompactionPicker::SanitizeCompactionInputFiles(
    std::unordered_set<uint64_t>* input_files,
-    const ColumnFamilyMetaData& cf_meta,
-    const int output_level) const {
+    const ColumnFamilyMetaData& cf_meta, const int output_level) const {
  assert(static_cast<int>(cf_meta.levels.size()) - 1 ==
         cf_meta.levels[cf_meta.levels.size() - 1].level);
  if (output_level >= static_cast<int>(cf_meta.levels.size())) {
    return Status::InvalidArgument(
        "Output level for column family " + cf_meta.name +
        " must between [0, " +
-        ToString(cf_meta.levels[cf_meta.levels.size() - 1].level) +
-        "].");
+        ToString(cf_meta.levels[cf_meta.levels.size() - 1].level) + "].");
  }

  if (output_level > MaxOutputLevel()) {
    return Status::InvalidArgument(
        "Exceed the maximum output level defined by "
        "the current compaction algorithm --- " +
-            ToString(MaxOutputLevel()));
+        ToString(MaxOutputLevel()));
  }

  if (output_level < 0) {
-    return Status::InvalidArgument(
-        "Output level cannot be negative.");
+    return Status::InvalidArgument("Output level cannot be negative.");
  }

  if (input_files->size() == 0) {
@@ -779,8 +818,8 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
        "A compaction must contain at least one file.");
  }

-  Status s = SanitizeCompactionInputFilesForAllLevels(
-      input_files, cf_meta, output_level);
+  Status s = SanitizeCompactionInputFilesForAllLevels(input_files, cf_meta,
+                                                      output_level);

  if (!s.ok()) {
    return s;
@@ -794,10 +833,9 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
      for (auto file_meta : level_meta.files) {
        if (file_num == TableFileNameToNumber(file_meta.name)) {
          if (file_meta.being_compacted) {
-            return Status::Aborted(
-                "Specified compaction input file " +
-                MakeTableFileName("", file_num) +
-                " is already being compacted.");
+            return Status::Aborted("Specified compaction input file " +
+                                   MakeTableFileName("", file_num) +
+                                   " is already being compacted.");
          }
          found = true;
          break;
@@ -809,8 +847,7 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
    }
    if (!found) {
      return Status::InvalidArgument(
-          "Specified compaction input file " +
-          MakeTableFileName("", file_num) +
+          "Specified compaction input file " + MakeTableFileName("", file_num) +
          " does not exist in column family " + cf_meta.name + ".");
    }
  }
@@ -819,8 +856,8 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
 }
 #endif  // !ROCKSDB_LITE

-bool LevelCompactionPicker::NeedsCompaction(const VersionStorageInfo* vstorage)
-    const {
+bool LevelCompactionPicker::NeedsCompaction(
+    const VersionStorageInfo* vstorage) const {
  if (!vstorage->FilesMarkedForCompaction().empty()) {
    return true;
  }
@@ -884,22 +921,40 @@ Compaction* LevelCompactionPicker::PickCompaction(
  int base_index = -1;
  CompactionInputFiles inputs;
  double score = 0;
+  CompactionReason compaction_reason = CompactionReason::kUnknown;

  // Find the compactions by size on all levels.
+  bool skipped_l0 = false;
  for (int i = 0; i < NumberLevels() - 1; i++) {
    score = vstorage->CompactionScore(i);
    level = vstorage->CompactionScoreLevel(i);
    assert(i == 0 || score <= vstorage->CompactionScore(i - 1));
    if (score >= 1) {
+      if (skipped_l0 && level == vstorage->base_level()) {
+        // If L0->base_level compaction is pending, don't schedule further
+        // compaction from base level. Otherwise L0->base_level compaction
+        // may starve.
+        continue;
+      }
      output_level = (level == 0) ? vstorage->base_level() : level + 1;
      if (PickCompactionBySize(vstorage, level, output_level, &inputs,
                               &parent_index, &base_index) &&
          ExpandWhileOverlapping(cf_name, vstorage, &inputs)) {
        // found the compaction!
+        if (level == 0) {
+          // L0 score = `num L0 files` / `level0_file_num_compaction_trigger`
+          compaction_reason = CompactionReason::kLevelL0FilesNum;
+        } else {
+          // L1+ score = `Level files size` / `MaxBytesForLevel`
+          compaction_reason = CompactionReason::kLevelMaxLevelSize;
+        }
        break;
      } else {
        // didn't find the compaction, clear the inputs
        inputs.clear();
+        if (level == 0) {
+          skipped_l0 = true;
+        }
      }
    }
  }
@@ -912,6 +967,9 @@ Compaction* LevelCompactionPicker::PickCompaction(
    parent_index = base_index = -1;
    PickFilesMarkedForCompactionExperimental(cf_name, vstorage, &inputs, &level,
                                             &output_level);
+    if (!inputs.empty()) {
+      compaction_reason = CompactionReason::kFilesMarkedForCompaction;
+    }
  }
  if (inputs.empty()) {
    return nullptr;
@@ -945,7 +1003,7 @@ Compaction* LevelCompactionPicker::PickCompaction(
  CompactionInputFiles output_level_inputs;
  output_level_inputs.level = output_level;
  if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage, &inputs,
-                   &output_level_inputs, &parent_index, base_index)) {
+                        &output_level_inputs, &parent_index, base_index)) {
    return nullptr;
  }

@@ -961,8 +1019,10 @@ Compaction* LevelCompactionPicker::PickCompaction(
      mutable_cf_options.MaxFileSizeForLevel(output_level),
      mutable_cf_options.MaxGrandParentOverlapBytes(level),
      GetPathId(ioptions_, mutable_cf_options, output_level),
-      GetCompressionType(ioptions_, output_level, vstorage->base_level()),
-      std::move(grandparents), is_manual, score);
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         vstorage->base_level()),
+      std::move(grandparents), is_manual, score,
+      false /* deletion_compaction */, compaction_reason);

  // If it's level 0 compaction, make sure we don't execute any other level 0
  // compactions in parallel
@@ -974,11 +1034,7 @@ Compaction* LevelCompactionPicker::PickCompaction(
  // takes running compactions into account (by skipping files that are already
  // being compacted). Since we just changed compaction score, we recalculate it
  // here
-  {  // this piece of code recomputes compaction score
-    CompactionOptionsFIFO dummy_compaction_options_fifo;
-    vstorage->ComputeCompactionScore(mutable_cf_options,
-                                     dummy_compaction_options_fifo);
-  }
+  vstorage->ComputeCompactionScore(mutable_cf_options);

  TEST_SYNC_POINT_CALLBACK("LevelCompactionPicker::PickCompaction:Return", c);

@@ -1033,6 +1089,7 @@ bool LevelCompactionPicker::PickCompactionBySize(VersionStorageInfo* vstorage,
  // could be made better by looking at key-ranges that are
  // being compacted at level 0.
  if (level == 0 && !level0_compactions_in_progress_.empty()) {
+    TEST_SYNC_POINT("LevelCompactionPicker::PickCompactionBySize:0");
    return false;
  }

@@ -1109,18 +1166,19 @@ void UniversalCompactionPicker::SortedRun::Dump(char* out_buf,
 }

 void UniversalCompactionPicker::SortedRun::DumpSizeInfo(
-    char* out_buf, size_t out_buf_size, int sorted_run_count) const {
+    char* out_buf, size_t out_buf_size, size_t sorted_run_count) const {
  if (level == 0) {
    assert(file != nullptr);
    snprintf(out_buf, out_buf_size,
-             "file %" PRIu64
-             "[%d] "
+             "file %" PRIu64 "[%" ROCKSDB_PRIszt
+             "] "
             "with size %" PRIu64 " (compensated size %" PRIu64 ")",
             file->fd.GetNumber(), sorted_run_count, file->fd.GetFileSize(),
             file->compensated_file_size);
  } else {
    snprintf(out_buf, out_buf_size,
-             "level %d[%d] "
+             "level %d[%" ROCKSDB_PRIszt
+             "] "
             "with size %" PRIu64 " (compensated size %" PRIu64 ")",
             level, sorted_run_count, size, compensated_file_size);
  }
@@ -1371,8 +1429,9 @@ uint32_t UniversalCompactionPicker::GetPathId(
  // considered in this algorithm. So the target size can be violated in
  // that case. We need to improve it.
  uint64_t accumulated_size = 0;
-  uint64_t future_size = file_size *
-    (100 - ioptions.compaction_options_universal.size_ratio) / 100;
+  uint64_t future_size =
+      file_size * (100 - ioptions.compaction_options_universal.size_ratio) /
+      100;
  uint32_t p = 0;
  assert(!ioptions.db_paths.empty());
  for (; p < ioptions.db_paths.size() - 1; p++) {
@@ -1396,22 +1455,27 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    unsigned int max_number_of_files_to_compact,
    const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
  unsigned int min_merge_width =
-    ioptions_.compaction_options_universal.min_merge_width;
+      ioptions_.compaction_options_universal.min_merge_width;
  unsigned int max_merge_width =
-    ioptions_.compaction_options_universal.max_merge_width;
+      ioptions_.compaction_options_universal.max_merge_width;

  const SortedRun* sr = nullptr;
  bool done = false;
-  int start_index = 0;
+  size_t start_index = 0;
  unsigned int candidate_count = 0;

-  unsigned int max_files_to_compact = std::min(max_merge_width,
-                                       max_number_of_files_to_compact);
+  unsigned int max_files_to_compact =
+      std::min(max_merge_width, max_number_of_files_to_compact);
  min_merge_width = std::max(min_merge_width, 2U);

+  // Caller checks the size before executing this function. This invariant is
+  // important because otherwise we may have a possible integer underflow when
+  // dealing with unsigned types.
+  assert(sorted_runs.size() > 0);
+
  // Considers a candidate file only if it is smaller than the
  // total size accumulated so far.
-  for (unsigned int loop = 0; loop < sorted_runs.size(); loop++) {
+  for (size_t loop = 0; loop < sorted_runs.size(); loop++) {
    candidate_count = 0;

    // Skip files that are already being compacted
@@ -1443,7 +1507,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    }

    // Check if the succeeding files need compaction.
-    for (unsigned int i = loop + 1;
+    for (size_t i = loop + 1;
         candidate_count < max_files_to_compact && i < sorted_runs.size();
         i++) {
      const SortedRun* succeeding_sr = &sorted_runs[i];
@@ -1485,7 +1549,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
      done = true;
      break;
    } else {
-      for (unsigned int i = loop;
+      for (size_t i = loop;
           i < loop + candidate_count && i < sorted_runs.size(); i++) {
        const SortedRun* skipping_sr = &sorted_runs[i];
        char file_num_buf[256];
@@ -1498,7 +1562,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
  if (!done || candidate_count <= 1) {
    return nullptr;
  }
-  unsigned int first_index_after = start_index + candidate_count;
+  size_t first_index_after = start_index + candidate_count;
  // Compression is enabled if files compacted earlier already reached
  // size ratio of compression.
  bool enable_compression = true;
@@ -1513,7 +1577,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    uint64_t older_file_size = 0;
    for (size_t i = sorted_runs.size() - 1; i >= first_index_after; i--) {
      older_file_size += sorted_runs[i].size;
-      if (older_file_size * 100L >= total_size * (long) ratio_to_compress) {
+      if (older_file_size * 100L >= total_size * (long)ratio_to_compress) {
        enable_compression = false;
        break;
      }
@@ -1539,7 +1603,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
  for (size_t i = 0; i < inputs.size(); ++i) {
    inputs[i].level = start_level + static_cast<int>(i);
  }
-  for (unsigned int i = start_index; i < first_index_after; i++) {
+  for (size_t i = start_index; i < first_index_after; i++) {
    auto& picking_sr = sorted_runs[i];
    if (picking_sr.level == 0) {
      FileMetaData* picking_file = picking_sr.file;
@@ -1556,11 +1620,19 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
                file_num_buf);
  }

+  CompactionReason compaction_reason;
+  if (max_number_of_files_to_compact == UINT_MAX) {
+    compaction_reason = CompactionReason::kUniversalSortedRunNum;
+  } else {
+    compaction_reason = CompactionReason::kUniversalSizeRatio;
+  }
  return new Compaction(
      vstorage, mutable_cf_options, std::move(inputs), output_level,
      mutable_cf_options.MaxFileSizeForLevel(output_level), LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, start_level, 1, enable_compression),
-      /* grandparents */ {}, /* is manual */ false, score);
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, start_level,
+                         1, enable_compression),
+      /* grandparents */ {}, /* is manual */ false, score,
+      false /* deletion_compaction */, compaction_reason);
 }

 // Look at overall size amplification. If size amplification
@@ -1574,19 +1646,19 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
    VersionStorageInfo* vstorage, double score,
    const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
  // percentage flexibilty while reducing size amplification
-  uint64_t ratio = ioptions_.compaction_options_universal.
-                     max_size_amplification_percent;
+  uint64_t ratio =
+      ioptions_.compaction_options_universal.max_size_amplification_percent;

  unsigned int candidate_count = 0;
  uint64_t candidate_size = 0;
-  unsigned int start_index = 0;
+  size_t start_index = 0;
  const SortedRun* sr = nullptr;

  // Skip files that are already being compacted
-  for (unsigned int loop = 0; loop < sorted_runs.size() - 1; loop++) {
+  for (size_t loop = 0; loop < sorted_runs.size() - 1; loop++) {
    sr = &sorted_runs[loop];
    if (!sr->being_compacted) {
-      start_index = loop;         // Consider this as the first candidate.
+      start_index = loop;  // Consider this as the first candidate.
      break;
    }
    char file_num_buf[kFormatFileNumberBufSize];
@@ -1598,18 +1670,19 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
  }

  if (sr == nullptr) {
-    return nullptr;             // no candidate files
+    return nullptr;  // no candidate files
  }
  {
    char file_num_buf[kFormatFileNumberBufSize];
    sr->Dump(file_num_buf, sizeof(file_num_buf), true);
-    LogToBuffer(log_buffer, "[%s] Universal: First candidate %s[%d] %s",
+    LogToBuffer(log_buffer,
+                "[%s] Universal: First candidate %s[%" ROCKSDB_PRIszt "] %s",
                cf_name.c_str(), file_num_buf, start_index,
                " to reduce size amp.\n");
  }

  // keep adding up all the remaining files
-  for (unsigned int loop = start_index; loop < sorted_runs.size() - 1; loop++) {
+  for (size_t loop = start_index; loop < sorted_runs.size() - 1; loop++) {
    sr = &sorted_runs[loop];
    if (sr->being_compacted) {
      char file_num_buf[kFormatFileNumberBufSize];
@@ -1649,7 +1722,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(

  // Estimate total file size
  uint64_t estimated_total_size = 0;
-  for (unsigned int loop = start_index; loop < sorted_runs.size(); loop++) {
+  for (size_t loop = start_index; loop < sorted_runs.size(); loop++) {
    estimated_total_size += sorted_runs[loop].size;
  }
  uint32_t path_id = GetPathId(ioptions_, estimated_total_size);
@@ -1660,7 +1733,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
    inputs[i].level = start_level + static_cast<int>(i);
  }
  // We always compact all the files, so always compress.
-  for (unsigned int loop = start_index; loop < sorted_runs.size(); loop++) {
+  for (size_t loop = start_index; loop < sorted_runs.size(); loop++) {
    auto& picking_sr = sorted_runs[loop];
    if (picking_sr.level == 0) {
      FileMetaData* f = picking_sr.file;
@@ -1672,7 +1745,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
      }
    }
    char file_num_buf[256];
-    sr->DumpSizeInfo(file_num_buf, sizeof(file_num_buf), loop);
+    picking_sr.DumpSizeInfo(file_num_buf, sizeof(file_num_buf), loop);
    LogToBuffer(log_buffer, "[%s] Universal: size amp picking %s",
                cf_name.c_str(), file_num_buf);
  }
@@ -1682,12 +1755,15 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
      vstorage->num_levels() - 1,
      mutable_cf_options.MaxFileSizeForLevel(vstorage->num_levels() - 1),
      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, vstorage->num_levels() - 1, 1),
-      /* grandparents */ {}, /* is manual */ false, score);
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options,
+                         vstorage->num_levels() - 1, 1),
+      /* grandparents */ {}, /* is manual */ false, score,
+      false /* deletion_compaction */,
+      CompactionReason::kUniversalSizeAmplification);
 }

-bool FIFOCompactionPicker::NeedsCompaction(const VersionStorageInfo* vstorage)
-    const {
+bool FIFOCompactionPicker::NeedsCompaction(
+    const VersionStorageInfo* vstorage) const {
  const int kLevel0 = 0;
  return vstorage->CompactionScore(kLevel0) >= 1;
 }
@@ -1742,7 +1818,7 @@ Compaction* FIFOCompactionPicker::PickCompaction(
  Compaction* c = new Compaction(
      vstorage, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
      kNoCompression, {}, /* is manual */ false, vstorage->CompactionScore(0),
-      /* is deletion compaction */ true);
+      /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
  level0_compactions_in_progress_.insert(c);
  return c;
 }
@@ -1751,7 +1827,7 @@ Compaction* FIFOCompactionPicker::CompactRange(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    VersionStorageInfo* vstorage, int input_level, int output_level,
    uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
-    InternalKey** compaction_end) {
+    InternalKey** compaction_end, bool* manual_conflict) {
  assert(input_level == 0);
  assert(output_level == 0);
  *compaction_end = nullptr;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -60,29 +60,28 @@ class CompactionPicker {
      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
      VersionStorageInfo* vstorage, int input_level, int output_level,
      uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
-      InternalKey** compaction_end);
+      InternalKey** compaction_end, bool* manual_conflict);

  // The maximum allowed output level.  Default value is NumberLevels() - 1.
-  virtual int MaxOutputLevel() const {
-    return NumberLevels() - 1;
-  }
+  virtual int MaxOutputLevel() const { return NumberLevels() - 1; }

  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const = 0;

-  // Sanitize the input set of compaction input files.
-  // When the input parameters do not describe a valid compaction, the
-  // function will try to fix the input_files by adding necessary
-  // files.  If it's not possible to conver an invalid input_files
-  // into a valid one by adding more files, the function will return a
-  // non-ok status with specific reason.
+// Sanitize the input set of compaction input files.
+// When the input parameters do not describe a valid compaction, the
+// function will try to fix the input_files by adding necessary
+// files.  If it's not possible to conver an invalid input_files
+// into a valid one by adding more files, the function will return a
+// non-ok status with specific reason.
 #ifndef ROCKSDB_LITE
-  Status SanitizeCompactionInputFiles(
-      std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const;
+  Status SanitizeCompactionInputFiles(std::unordered_set<uint64_t>* input_files,
+                                      const ColumnFamilyMetaData& cf_meta,
+                                      const int output_level) const;
 #endif  // ROCKSDB_LITE

  // Free up the files that participated in a compaction
+  //
+  // Requirement: DB mutex held
  void ReleaseCompactionFiles(Compaction* c, Status status);

  // Returns true if any one of the specified files are being compacted
@@ -121,15 +120,15 @@ class CompactionPicker {
  // Stores the minimal range that covers all entries in inputs in
  // *smallest, *largest.
  // REQUIRES: inputs is not empty
-  void GetRange(const CompactionInputFiles& inputs,
-                InternalKey* smallest, InternalKey* largest);
+  void GetRange(const CompactionInputFiles& inputs, InternalKey* smallest,
+                InternalKey* largest);

  // Stores the minimal range that covers all entries in inputs1 and inputs2
  // in *smallest, *largest.
  // REQUIRES: inputs is not empty
  void GetRange(const CompactionInputFiles& inputs1,
-                const CompactionInputFiles& inputs2,
-                InternalKey* smallest, InternalKey* largest);
+                const CompactionInputFiles& inputs2, InternalKey* smallest,
+                InternalKey* largest);

  // Add more files to the inputs on "level" to make sure that
  // no newer version of a key is compacted to "level+1" while leaving an older
@@ -164,13 +163,12 @@ class CompactionPicker {

  const ImmutableCFOptions& ioptions_;

-  // A helper function to SanitizeCompactionInputFiles() that
-  // sanitizes "input_files" by adding necessary files.
+// A helper function to SanitizeCompactionInputFiles() that
+// sanitizes "input_files" by adding necessary files.
 #ifndef ROCKSDB_LITE
  virtual Status SanitizeCompactionInputFilesForAllLevels(
      std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const;
+      const ColumnFamilyMetaData& cf_meta, const int output_level) const;
 #endif  // ROCKSDB_LITE

  // Keeps track of all compactions that are running on Level0.
@@ -190,8 +188,8 @@ class LevelCompactionPicker : public CompactionPicker {
                                     VersionStorageInfo* vstorage,
                                     LogBuffer* log_buffer) override;

-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;

  // Pick a path ID to place a newly generated file, with its level
  static uint32_t GetPathId(const ImmutableCFOptions& ioptions,
@@ -230,8 +228,8 @@ class UniversalCompactionPicker : public CompactionPicker {

  virtual int MaxOutputLevel() const override { return NumberLevels() - 1; }

-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;

 private:
  struct SortedRun {
@@ -251,7 +249,7 @@ class UniversalCompactionPicker : public CompactionPicker {

    // sorted_run_count is added into the string to print
    void DumpSizeInfo(char* out_buf, size_t out_buf_size,
-                      int sorted_run_count) const;
+                      size_t sorted_run_count) const;

    int level;
    // `file` Will be null for level > 0. For level = 0, the sorted run is
@@ -302,22 +300,20 @@ class FIFOCompactionPicker : public CompactionPicker {
      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
      VersionStorageInfo* vstorage, int input_level, int output_level,
      uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
-      InternalKey** compaction_end) override;
+      InternalKey** compaction_end, bool* manual_conflict) override;

  // The maximum allowed output level.  Always returns 0.
-  virtual int MaxOutputLevel() const override {
-    return 0;
-  }
+  virtual int MaxOutputLevel() const override { return 0; }

-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;
 };

 class NullCompactionPicker : public CompactionPicker {
 public:
  NullCompactionPicker(const ImmutableCFOptions& ioptions,
-                       const InternalKeyComparator* icmp) :
-      CompactionPicker(ioptions, icmp) {}
+                       const InternalKeyComparator* icmp)
+      : CompactionPicker(ioptions, icmp) {}
  virtual ~NullCompactionPicker() {}

  // Always return "nullptr"
@@ -329,23 +325,27 @@ class NullCompactionPicker : public CompactionPicker {
  }

  // Always return "nullptr"
-  Compaction* CompactRange(
-      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-      VersionStorageInfo* vstorage, int input_level, int output_level,
-      uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
-      InternalKey** compaction_end) override {
+  Compaction* CompactRange(const std::string& cf_name,
+                           const MutableCFOptions& mutable_cf_options,
+                           VersionStorageInfo* vstorage, int input_level,
+                           int output_level, uint32_t output_path_id,
+                           const InternalKey* begin, const InternalKey* end,
+                           InternalKey** compaction_end,
+                           bool* manual_conflict) override {
    return nullptr;
  }

  // Always returns false.
-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override {
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override {
    return false;
  }
 };
 #endif  // !ROCKSDB_LITE

 CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
+                                   const VersionStorageInfo* vstorage,
+                                   const MutableCFOptions& mutable_cf_options,
                                   int level, int base_level,
                                   const bool enable_compression = true);

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -79,7 +79,7 @@ class CompactionPickerTest : public testing::Test {
  }

  void Add(int level, uint32_t file_number, const char* smallest,
-           const char* largest, uint64_t file_size = 0, uint32_t path_id = 0,
+           const char* largest, uint64_t file_size = 1, uint32_t path_id = 0,
           SequenceNumber smallest_seq = 100,
           SequenceNumber largest_seq = 100) {
    assert(level < vstorage_->num_levels());
@@ -119,7 +119,7 @@ class CompactionPickerTest : public testing::Test {
    vstorage_->UpdateNumNonEmptyLevels();
    vstorage_->GenerateFileIndexer();
    vstorage_->GenerateLevelFilesBrief();
-    vstorage_->ComputeCompactionScore(mutable_cf_options_, fifo_options_);
+    vstorage_->ComputeCompactionScore(mutable_cf_options_);
    vstorage_->GenerateLevel0NonOverlapping();
    vstorage_->SetFinalized();
  }
@@ -195,6 +195,7 @@ TEST_F(CompactionPickerTest, LevelMaxScore) {
  NewVersionStorage(6, kCompactionStyleLevel);
  mutable_cf_options_.target_file_size_base = 10000000;
  mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
  Add(0, 1U, "150", "200", 1000000000U);
  // Level 1 score 1.2
  Add(1, 66U, "150", "200", 6000000U);
@@ -319,6 +320,7 @@ TEST_F(CompactionPickerTest, Level0TriggerDynamic4) {
  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
  mutable_cf_options_.max_bytes_for_level_base = 200;
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+
  NewVersionStorage(num_levels, kCompactionStyleLevel);
  Add(0, 1U, "150", "200");
  Add(0, 2U, "200", "250");
@@ -351,6 +353,7 @@ TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
  mutable_cf_options_.max_bytes_for_level_base = 200;
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
  NewVersionStorage(num_levels, kCompactionStyleLevel);
  Add(0, 1U, "150", "200");
  Add(num_levels - 1, 3U, "200", "250", 300U);
@@ -366,11 +369,9 @@ TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
  ASSERT_TRUE(compaction.get() != nullptr);
  ASSERT_EQ(1U, compaction->num_input_files(0));
-  ASSERT_EQ(6U, compaction->input(0, 0)->fd.GetNumber());
-  ASSERT_EQ(2U, compaction->num_input_files(1));
-  ASSERT_EQ(3U, compaction->input(1, 0)->fd.GetNumber());
-  ASSERT_EQ(4U, compaction->input(1, 1)->fd.GetNumber());
-  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_EQ(5U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(0, compaction->num_input_files(1));
+  ASSERT_EQ(1U, compaction->num_input_levels());
  ASSERT_EQ(num_levels - 1, compaction->output_level());
 }

@@ -466,7 +467,6 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {
  fifo_options_.max_table_files_size = kMaxSize;
  ioptions_.compaction_options_fifo = fifo_options_;
  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
-
  UpdateVersionStorageInfo();
  // must return false when there's no files.
  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), false);
@@ -487,6 +487,90 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {
 }
 #endif  // ROCKSDB_LITE

+TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.target_file_size_base = 10000000;
+  mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
+  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
+
+  Add(2, 6U, "150", "179", 50000000U);
+  Add(2, 7U, "180", "220", 50000000U);
+  Add(2, 8U, "321", "400", 50000000U);  // File not overlapping
+  Add(2, 9U, "721", "800", 50000000U);
+
+  Add(3, 26U, "150", "170", 260000000U);
+  Add(3, 27U, "171", "179", 260000000U);
+  Add(3, 28U, "191", "220", 260000000U);
+  Add(3, 29U, "221", "300", 260000000U);
+  Add(3, 30U, "750", "900", 260000000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  // Pick file 8 because it overlaps with 0 files on level 3.
+  ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.target_file_size_base = 10000000;
+  mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
+  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
+
+  Add(2, 6U, "150", "175",
+      60000000U);  // Overlaps with file 26, 27, total size 521M
+  Add(2, 7U, "176", "200", 60000000U);  // Overlaps with file 27, 28, total size
+                                        // 520M, the smalelst overlapping
+  Add(2, 8U, "201", "300",
+      60000000U);  // Overlaps with file 28, 29, total size 521M
+
+  Add(3, 26U, "100", "110", 261000000U);
+  Add(3, 26U, "150", "170", 261000000U);
+  Add(3, 27U, "171", "179", 260000000U);
+  Add(3, 28U, "191", "220", 260000000U);
+  Add(3, 29U, "221", "300", 261000000U);
+  Add(3, 30U, "321", "400", 261000000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  // Picking file 7 because overlapping ratio is the biggest.
+  ASSERT_EQ(7U, compaction->input(0, 0)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, CompactionPriMinOverlapping3) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.max_bytes_for_level_base = 10000000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
+
+  // file 7 and 8 over lap with the same file, but file 8 is smaller so
+  // it will be picked.
+  Add(2, 6U, "150", "167", 60000000U);  // Overlaps with file 26, 27
+  Add(2, 7U, "168", "169", 60000000U);  // Overlaps with file 27
+  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 28, but the file
+                                        // itself is larger. Should be picked.
+
+  Add(3, 26U, "160", "165", 260000000U);
+  Add(3, 27U, "166", "170", 260000000U);
+  Add(3, 28U, "180", "400", 260000000U);
+  Add(3, 29U, "401", "500", 260000000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  // Picking file 8 because overlapping ratio is the biggest.
+  ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
+}
+
 // This test exhibits the bug where we don't properly reset parent_index in
 // PickCompaction()
 TEST_F(CompactionPickerTest, ParentIndexResetBug) {
@@ -514,6 +598,8 @@ TEST_F(CompactionPickerTest, ParentIndexResetBug) {
 // ranges (with different sequence numbers) in the input files.
 TEST_F(CompactionPickerTest, OverlappingUserKeys) {
  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.compaction_pri = kByCompensatedSize;
+
  Add(1, 1U, "100", "150", 1U);
  // Overlapping user keys
  Add(1, 2U, "200", "400", 1U);
@@ -580,6 +666,101 @@ TEST_F(CompactionPickerTest, OverlappingUserKeys3) {
  ASSERT_EQ(7U, compaction->input(1, 1)->fd.GetNumber());
 }

+TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri1) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_bytes_for_level_base = 900000000U;
+
+  // 6 L0 files, score 3.
+  Add(0, 1U, "000", "400", 1U);
+  Add(0, 2U, "001", "400", 1U, 0, 0);
+  Add(0, 3U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 31U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 32U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 33U, "001", "400", 1000000000U, 0, 0);
+
+  // L1 total size 2GB, score 2.2. If one file being comapcted, score 1.1.
+  Add(1, 4U, "050", "300", 1000000000U, 0, 0);
+  file_map_[4u].first->being_compacted = true;
+  Add(1, 5U, "301", "350", 1000000000U, 0, 0);
+
+  // Output level overlaps with the beginning and the end of the chain
+  Add(2, 6U, "050", "100", 1U);
+  Add(2, 7U, "300", "400", 1U);
+
+  // No compaction should be scheduled, if L0 has higher priority than L1
+  // but L0->L1 compaction is blocked by a file in L1 being compacted.
+  UpdateVersionStorageInfo();
+  ASSERT_EQ(0, vstorage_->CompactionScoreLevel(0));
+  ASSERT_EQ(1, vstorage_->CompactionScoreLevel(1));
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() == nullptr);
+}
+
+TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri2) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_bytes_for_level_base = 900000000U;
+
+  // 6 L0 files, score 3.
+  Add(0, 1U, "000", "400", 1U);
+  Add(0, 2U, "001", "400", 1U, 0, 0);
+  Add(0, 3U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 31U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 32U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 33U, "001", "400", 1000000000U, 0, 0);
+
+  // L1 total size 2GB, score 2.2. If one file being comapcted, score 1.1.
+  Add(1, 4U, "050", "300", 1000000000U, 0, 0);
+  Add(1, 5U, "301", "350", 1000000000U, 0, 0);
+
+  // Output level overlaps with the beginning and the end of the chain
+  Add(2, 6U, "050", "100", 1U);
+  Add(2, 7U, "300", "400", 1U);
+
+  // If no file in L1 being compacted, L0->L1 compaction will be scheduled.
+  UpdateVersionStorageInfo();  // being_compacted flag is cleared here.
+  ASSERT_EQ(0, vstorage_->CompactionScoreLevel(0));
+  ASSERT_EQ(1, vstorage_->CompactionScoreLevel(1));
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+}
+
+TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri3) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_bytes_for_level_base = 900000000U;
+
+  // 6 L0 files, score 3.
+  Add(0, 1U, "000", "400", 1U);
+  Add(0, 2U, "001", "400", 1U, 0, 0);
+  Add(0, 3U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 31U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 32U, "001", "400", 1000000000U, 0, 0);
+  Add(0, 33U, "001", "400", 1000000000U, 0, 0);
+
+  // L1 score more than 6.
+  Add(1, 4U, "050", "300", 1000000000U, 0, 0);
+  file_map_[4u].first->being_compacted = true;
+  Add(1, 5U, "301", "350", 1000000000U, 0, 0);
+  Add(1, 51U, "351", "400", 6000000000U, 0, 0);
+
+  // Output level overlaps with the beginning and the end of the chain
+  Add(2, 6U, "050", "100", 1U);
+  Add(2, 7U, "300", "400", 1U);
+
+  // If score in L1 is larger than L0, L1 compaction goes through despite
+  // there is pending L0 compaction.
+  UpdateVersionStorageInfo();
+  ASSERT_EQ(1, vstorage_->CompactionScoreLevel(0));
+  ASSERT_EQ(0, vstorage_->CompactionScoreLevel(1));
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+}
+
 TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded1) {
  int num_levels = ioptions_.num_levels;
  ioptions_.level_compaction_dynamic_level_bytes = false;
@@ -594,19 +775,22 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded1) {
  Add(1, 4U, "400", "500", 600);
  Add(1, 5U, "600", "700", 600);
  // Level 2 is less than target 10000 even added size of level 1
+  // Size ratio of L2/L1 is 9600 / 1200 = 8
  Add(2, 6U, "150", "200", 2500);
  Add(2, 7U, "201", "210", 2000);
-  Add(2, 8U, "300", "310", 2500);
+  Add(2, 8U, "300", "310", 2600);
  Add(2, 9U, "400", "500", 2500);
  // Level 3 exceeds target 100,000 of 1000
  Add(3, 10U, "400", "500", 101000);
-  // Level 4 exceeds target 1,000,000 of 500 after adding size from level 3
-  Add(4, 11U, "400", "500", 999500);
-  Add(5, 11U, "400", "500", 8000000);
+  // Level 4 exceeds target 1,000,000 by 900 after adding size from level 3
+  // Size ratio L4/L3 is 9.9
+  // After merge from L3, L4 size is 1000900
+  Add(4, 11U, "400", "500", 999900);
+  Add(5, 11U, "400", "500", 8007200);

  UpdateVersionStorageInfo();

-  ASSERT_EQ(2200u + 11000u + 5500u,
+  ASSERT_EQ(200u * 9u + 10900u + 900u * 9,
            vstorage_->estimated_compaction_needed_bytes());
 }

@@ -622,17 +806,42 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded2) {
  Add(0, 4U, "150", "200", 200);
  Add(0, 5U, "150", "200", 200);
  Add(0, 6U, "150", "200", 200);
-  // Level 1 is over target by
+  // Level 1 size will be 1400 after merging with L0
  Add(1, 7U, "400", "500", 200);
  Add(1, 8U, "600", "700", 200);
  // Level 2 is less than target 10000 even added size of level 1
-  Add(2, 9U, "150", "200", 9500);
+  Add(2, 9U, "150", "200", 9100);
+  // Level 3 over the target, but since level 4 is empty, we assume it will be
+  // a trivial move.
  Add(3, 10U, "400", "500", 101000);

  UpdateVersionStorageInfo();

-  ASSERT_EQ(1400u + 4400u + 11000u,
-            vstorage_->estimated_compaction_needed_bytes());
+  // estimated L1->L2 merge: 400 * (9100.0 / 1400.0 + 1.0)
+  ASSERT_EQ(1400u + 3000u, vstorage_->estimated_compaction_needed_bytes());
+}
+
+TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded3) {
+  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_.max_bytes_for_level_base = 1000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  NewVersionStorage(num_levels, kCompactionStyleLevel);
+  Add(0, 1U, "150", "200", 2000);
+  Add(0, 2U, "150", "200", 2000);
+  Add(0, 4U, "150", "200", 2000);
+  Add(0, 5U, "150", "200", 2000);
+  Add(0, 6U, "150", "200", 1000);
+  // Level 1 size will be 10000 after merging with L0
+  Add(1, 7U, "400", "500", 500);
+  Add(1, 8U, "600", "700", 500);
+
+  Add(2, 9U, "150", "200", 10000);
+
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(10000u + 18000u, vstorage_->estimated_compaction_needed_bytes());
 }

 TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {
@@ -656,12 +865,14 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {
  // 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);
-  // Level 2 is over target by 1100 + 100
-  Add(num_levels - 2, 9U, "150", "200", 5100);
+  // num_levels - 2 is over target by 1100 + 200
+  Add(num_levels - 2, 9U, "150", "200", 5200);

  UpdateVersionStorageInfo();

-  ASSERT_EQ(1600u + 12100u + 13200u,
+  // Merging to the second last level: (5200 / 1600 + 1) * 1100
+  // Merging to the last level: (50000 / 6300 + 1) * 1300
+  ASSERT_EQ(1600u + 4675u + 11617u,
            vstorage_->estimated_compaction_needed_bytes());
 }

@@ -1,6 +1,6 @@
 // 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.
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -18,6 +18,13 @@ namespace rocksdb {
 void CancelAllBackgroundWork(DB* db, bool wait) {
  (dynamic_cast<DBImpl*>(db))->CancelAllBackgroundWork(wait);
 }
+
+Status DeleteFilesInRange(DB* db, ColumnFamilyHandle* column_family,
+                          const Slice* begin, const Slice* end) {
+  return (dynamic_cast<DBImpl*>(db))
+      ->DeleteFilesInRange(column_family, begin, end);
+}
+
 }  // namespace rocksdb

 #endif  // ROCKSDB_LITE
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -11,6 +11,7 @@

 #include "rocksdb/db.h"

+#include <inttypes.h>
 #include <errno.h>
 #include <fcntl.h>
 #include <sys/stat.h>
@@ -41,6 +42,7 @@ class CorruptionTest : public testing::Test {

  CorruptionTest() {
    tiny_cache_ = NewLRUCache(100);
+    options_.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
    options_.env = &env_;
    dbname_ = test::TmpDir() + "/corruption_test";
    DestroyDB(dbname_, options_);
@@ -104,8 +106,8 @@ class CorruptionTest : public testing::Test {
  }

  void Check(int min_expected, int max_expected) {
-    unsigned int next_expected = 0;
-    int missed = 0;
+    uint64_t next_expected = 0;
+    uint64_t missed = 0;
    int bad_keys = 0;
    int bad_values = 0;
    int correct = 0;
@@ -126,7 +128,7 @@ class CorruptionTest : public testing::Test {
        continue;
      }
      missed += (key - next_expected);
-      next_expected = static_cast<unsigned int>(key + 1);
+      next_expected = key + 1;
      if (iter->value() != Value(static_cast<int>(key), &value_space)) {
        bad_values++;
      } else {
@@ -136,8 +138,9 @@ class CorruptionTest : public testing::Test {
    delete iter;

    fprintf(stderr,
-            "expected=%d..%d; got=%d; bad_keys=%d; bad_values=%d; missed=%d\n",
-            min_expected, max_expected, correct, bad_keys, bad_values, missed);
+      "expected=%d..%d; got=%d; bad_keys=%d; bad_values=%d; missed=%llu\n",
+            min_expected, max_expected, correct, bad_keys, bad_values,
+            static_cast<unsigned long long>(missed));
    ASSERT_LE(min_expected, correct);
    ASSERT_GE(max_expected, correct);
  }
@@ -183,7 +186,7 @@ class CorruptionTest : public testing::Test {
    FileType type;
    std::string fname;
    int picked_number = -1;
-    for (unsigned int i = 0; i < filenames.size(); i++) {
+    for (size_t i = 0; i < filenames.size(); i++) {
      if (ParseFileName(filenames[i], &number, &type) &&
          type == filetype &&
          static_cast<int>(number) > picked_number) {  // Pick latest file
@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -0,0 +1,495 @@
+//  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.
+//
+// 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 <cstdlib>
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+
+namespace rocksdb {
+
+class DBBlockCacheTest : public DBTestBase {
+ private:
+  size_t miss_count_ = 0;
+  size_t hit_count_ = 0;
+  size_t insert_count_ = 0;
+  size_t failure_count_ = 0;
+  size_t compressed_miss_count_ = 0;
+  size_t compressed_hit_count_ = 0;
+  size_t compressed_insert_count_ = 0;
+  size_t compressed_failure_count_ = 0;
+
+ public:
+  const size_t kNumBlocks = 10;
+  const size_t kValueSize = 100;
+
+  DBBlockCacheTest() : DBTestBase("/db_block_cache_test") {}
+
+  BlockBasedTableOptions GetTableOptions() {
+    BlockBasedTableOptions table_options;
+    // Set a small enough block size so that each key-value get its own block.
+    table_options.block_size = 1;
+    return table_options;
+  }
+
+  Options GetOptions(const BlockBasedTableOptions& table_options) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    // options.compression = kNoCompression;
+    options.statistics = rocksdb::CreateDBStatistics();
+    options.table_factory.reset(new BlockBasedTableFactory(table_options));
+    return options;
+  }
+
+  void InitTable(const Options& options) {
+    std::string value(kValueSize, 'a');
+    for (size_t i = 0; i < kNumBlocks; i++) {
+      ASSERT_OK(Put(ToString(i), value.c_str()));
+    }
+  }
+
+  void RecordCacheCounters(const Options& options) {
+    miss_count_ = TestGetTickerCount(options, BLOCK_CACHE_MISS);
+    hit_count_ = TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    insert_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD);
+    failure_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
+    compressed_miss_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
+    compressed_hit_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
+    compressed_insert_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
+    compressed_failure_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
+  }
+
+  void CheckCacheCounters(const Options& options, size_t expected_misses,
+                          size_t expected_hits, size_t expected_inserts,
+                          size_t expected_failures) {
+    size_t new_miss_count = TestGetTickerCount(options, BLOCK_CACHE_MISS);
+    size_t new_hit_count = TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    size_t new_insert_count = TestGetTickerCount(options, BLOCK_CACHE_ADD);
+    size_t new_failure_count =
+        TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
+    ASSERT_EQ(miss_count_ + expected_misses, new_miss_count);
+    ASSERT_EQ(hit_count_ + expected_hits, new_hit_count);
+    ASSERT_EQ(insert_count_ + expected_inserts, new_insert_count);
+    ASSERT_EQ(failure_count_ + expected_failures, new_failure_count);
+    miss_count_ = new_miss_count;
+    hit_count_ = new_hit_count;
+    insert_count_ = new_insert_count;
+    failure_count_ = new_failure_count;
+  }
+
+  void CheckCompressedCacheCounters(const Options& options,
+                                    size_t expected_misses,
+                                    size_t expected_hits,
+                                    size_t expected_inserts,
+                                    size_t expected_failures) {
+    size_t new_miss_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
+    size_t new_hit_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
+    size_t new_insert_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
+    size_t new_failure_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
+    ASSERT_EQ(compressed_miss_count_ + expected_misses, new_miss_count);
+    ASSERT_EQ(compressed_hit_count_ + expected_hits, new_hit_count);
+    ASSERT_EQ(compressed_insert_count_ + expected_inserts, new_insert_count);
+    ASSERT_EQ(compressed_failure_count_ + expected_failures, new_failure_count);
+    compressed_miss_count_ = new_miss_count;
+    compressed_hit_count_ = new_hit_count;
+    compressed_insert_count_ = new_insert_count;
+    compressed_failure_count_ = new_failure_count;
+  }
+};
+
+TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
+  ReadOptions read_options;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(0, 0, false);
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  // Load blocks into cache.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 1, 0, 1, 0);
+    iterators[i].reset(iter);
+  }
+  size_t usage = cache->GetUsage();
+  ASSERT_LT(0, usage);
+  cache->SetCapacity(usage);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+
+  // Test with strict capacity limit.
+  cache->SetStrictCapacityLimit(true);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_TRUE(iter->status().IsIncomplete());
+  CheckCacheCounters(options, 1, 0, 0, 1);
+  delete iter;
+  iter = nullptr;
+
+  // Release interators and access cache again.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iterators[i].reset();
+    CheckCacheCounters(options, 0, 0, 0, 0);
+  }
+  ASSERT_EQ(0, cache->GetPinnedUsage());
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 0, 1, 0, 0);
+    iterators[i].reset(iter);
+  }
+}
+
+#ifdef SNAPPY
+TEST_F(DBBlockCacheTest, TestWithCompressedBlockCache) {
+  ReadOptions read_options;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  options.compression = CompressionType::kSnappyCompression;
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(0, 0, false);
+  std::shared_ptr<Cache> compressed_cache = NewLRUCache(0, 0, false);
+  table_options.block_cache = cache;
+  table_options.block_cache_compressed = compressed_cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  // Load blocks into cache.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 1, 0, 1, 0);
+    CheckCompressedCacheCounters(options, 1, 0, 1, 0);
+    iterators[i].reset(iter);
+  }
+  size_t usage = cache->GetUsage();
+  ASSERT_LT(0, usage);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+  size_t compressed_usage = compressed_cache->GetUsage();
+  ASSERT_LT(0, compressed_usage);
+  // Compressed block cache cannot be pinned.
+  ASSERT_EQ(0, compressed_cache->GetPinnedUsage());
+
+  // Set strict capacity limit flag. Now block will only load into compressed
+  // block cache.
+  cache->SetCapacity(usage);
+  cache->SetStrictCapacityLimit(true);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+  // compressed_cache->SetCapacity(compressed_usage);
+  compressed_cache->SetCapacity(0);
+  // compressed_cache->SetStrictCapacityLimit(true);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_TRUE(iter->status().IsIncomplete());
+  CheckCacheCounters(options, 1, 0, 0, 1);
+  CheckCompressedCacheCounters(options, 1, 0, 1, 0);
+  delete iter;
+  iter = nullptr;
+
+  // Clear strict capacity limit flag. This time we shall hit compressed block
+  // cache.
+  cache->SetStrictCapacityLimit(false);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_OK(iter->status());
+  CheckCacheCounters(options, 1, 0, 1, 0);
+  CheckCompressedCacheCounters(options, 0, 1, 0, 0);
+  delete iter;
+  iter = nullptr;
+}
+#endif  // SNAPPY
+
+#ifndef ROCKSDB_LITE
+
+// Make sure that when options.block_cache is set, after a new table is
+// created its index/filter blocks are added to block cache.
+TEST_F(DBBlockCacheTest, IndexAndFilterBlocksOfNewTableAddedToCache) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = true;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "key", "val"));
+  // Create a new table.
+  ASSERT_OK(Flush(1));
+
+  // index/filter blocks added to block cache right after table creation.
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(2, /* only index/filter were added */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+  ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
+  uint64_t int_num;
+  ASSERT_TRUE(
+      dbfull()->GetIntProperty("rocksdb.estimate-table-readers-mem", &int_num));
+  ASSERT_EQ(int_num, 0U);
+
+  // Make sure filter block is in cache.
+  std::string value;
+  ReadOptions ropt;
+  db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
+
+  // Miss count should remain the same.
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  // Make sure index block is in cache.
+  auto index_block_hit = TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT);
+  value = Get(1, "key");
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(index_block_hit + 1,
+            TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  value = Get(1, "key");
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(index_block_hit + 2,
+            TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+}
+
+TEST_F(DBBlockCacheTest, IndexAndFilterBlocksStats) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = true;
+  // 200 bytes are enough to hold the first two blocks
+  std::shared_ptr<Cache> cache = NewLRUCache(200, 0, false);
+  table_options.block_cache = cache;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "key", "val"));
+  // Create a new table
+  ASSERT_OK(Flush(1));
+  size_t index_bytes_insert =
+      TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT);
+  size_t filter_bytes_insert =
+      TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT);
+  ASSERT_GT(index_bytes_insert, 0);
+  ASSERT_GT(filter_bytes_insert, 0);
+  ASSERT_EQ(cache->GetUsage(), index_bytes_insert + filter_bytes_insert);
+  // set the cache capacity to the current usage
+  cache->SetCapacity(index_bytes_insert + filter_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT), 0);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT), 0);
+  ASSERT_OK(Put(1, "key2", "val"));
+  // Create a new table
+  ASSERT_OK(Flush(1));
+  // cache evicted old index and block entries
+  ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT),
+            index_bytes_insert);
+  ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT),
+            filter_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT),
+            index_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT),
+            filter_bytes_insert);
+}
+
+TEST_F(DBBlockCacheTest, ParanoidFileChecks) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.level0_file_num_compaction_trigger = 2;
+  options.paranoid_file_checks = true;
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = false;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "1_key", "val"));
+  ASSERT_OK(Put(1, "9_key", "val"));
+  // Create a new table.
+  ASSERT_OK(Flush(1));
+  ASSERT_EQ(1, /* read and cache data block */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+  ASSERT_OK(Put(1, "1_key2", "val2"));
+  ASSERT_OK(Put(1, "9_key2", "val2"));
+  // Create a new SST file. This will further trigger a compaction
+  // and generate another file.
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(3, /* Totally 3 files created up to now */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+  // After disabling options.paranoid_file_checks. NO further block
+  // is added after generating a new file.
+  ASSERT_OK(
+      dbfull()->SetOptions(handles_[1], {{"paranoid_file_checks", "false"}}));
+
+  ASSERT_OK(Put(1, "1_key3", "val3"));
+  ASSERT_OK(Put(1, "9_key3", "val3"));
+  ASSERT_OK(Flush(1));
+  ASSERT_OK(Put(1, "1_key4", "val4"));
+  ASSERT_OK(Put(1, "9_key4", "val4"));
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(3, /* Totally 3 files created up to now */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+}
+
+TEST_F(DBBlockCacheTest, CompressedCache) {
+  if (!Snappy_Supported()) {
+    return;
+  }
+  int num_iter = 80;
+
+  // Run this test three iterations.
+  // Iteration 1: only a uncompressed block cache
+  // Iteration 2: only a compressed block cache
+  // Iteration 3: both block cache and compressed cache
+  // Iteration 4: both block cache and compressed cache, but DB is not
+  // compressed
+  for (int iter = 0; iter < 4; iter++) {
+    Options options = CurrentOptions();
+    options.write_buffer_size = 64 * 1024;  // small write buffer
+    options.statistics = rocksdb::CreateDBStatistics();
+
+    BlockBasedTableOptions table_options;
+    switch (iter) {
+      case 0:
+        // only uncompressed block cache
+        table_options.block_cache = NewLRUCache(8 * 1024);
+        table_options.block_cache_compressed = nullptr;
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 1:
+        // no block cache, only compressed cache
+        table_options.no_block_cache = true;
+        table_options.block_cache = nullptr;
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 2:
+        // both compressed and uncompressed block cache
+        table_options.block_cache = NewLRUCache(1024);
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 3:
+        // both block cache and compressed cache, but DB is not compressed
+        // also, make block cache sizes bigger, to trigger block cache hits
+        table_options.block_cache = NewLRUCache(1024 * 1024);
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        options.compression = kNoCompression;
+        break;
+      default:
+        ASSERT_TRUE(false);
+    }
+    CreateAndReopenWithCF({"pikachu"}, options);
+    // default column family doesn't have block cache
+    Options no_block_cache_opts;
+    no_block_cache_opts.statistics = options.statistics;
+    no_block_cache_opts = CurrentOptions(no_block_cache_opts);
+    BlockBasedTableOptions table_options_no_bc;
+    table_options_no_bc.no_block_cache = true;
+    no_block_cache_opts.table_factory.reset(
+        NewBlockBasedTableFactory(table_options_no_bc));
+    ReopenWithColumnFamilies(
+        {"default", "pikachu"},
+        std::vector<Options>({no_block_cache_opts, options}));
+
+    Random rnd(301);
+
+    // Write 8MB (80 values, each 100K)
+    ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
+    std::vector<std::string> values;
+    std::string str;
+    for (int i = 0; i < num_iter; i++) {
+      if (i % 4 == 0) {  // high compression ratio
+        str = RandomString(&rnd, 1000);
+      }
+      values.push_back(str);
+      ASSERT_OK(Put(1, Key(i), values[i]));
+    }
+
+    // flush all data from memtable so that reads are from block cache
+    ASSERT_OK(Flush(1));
+
+    for (int i = 0; i < num_iter; i++) {
+      ASSERT_EQ(Get(1, Key(i)), values[i]);
+    }
+
+    // check that we triggered the appropriate code paths in the cache
+    switch (iter) {
+      case 0:
+        // only uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 1:
+        // no block cache, only compressed cache
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 2:
+        // both compressed and uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 3:
+        // both compressed and uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_HIT), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        // compressed doesn't have any hits since blocks are not compressed on
+        // storage
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT), 0);
+        break;
+      default:
+        ASSERT_TRUE(false);
+    }
+
+    options.create_if_missing = true;
+    DestroyAndReopen(options);
+  }
+}
+
+#endif  // ROCKSDB_LITE
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -368,11 +368,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
 // entries are deleted. The compaction should create bunch of 'DeleteFile'
 // entries in VersionEdit, but none of the 'AddFile's.
 TEST_F(DBTestCompactionFilter, CompactionFilterDeletesAll) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_filter_factory = std::make_shared<DeleteFilterFactory>();
  options.disable_auto_compactions = true;
  options.create_if_missing = true;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  // put some data
@@ -400,11 +399,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterDeletesAll) {

 TEST_F(DBTestCompactionFilter, CompactionFilterWithValueChange) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.num_levels = 3;
    options.compaction_filter_factory =
      std::make_shared<ChangeFilterFactory>();
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Write 100K+1 keys, these are written to a few files
@@ -467,8 +465,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterWithMergeOperator) {
  PutFixed64(&three, 3);
  PutFixed64(&four, 4);

-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
  options.create_if_missing = true;
  options.merge_operator = MergeOperators::CreateUInt64AddOperator();
  options.num_levels = 3;
@@ -583,7 +580,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextManual) {
      iter->Next();
    }
    ASSERT_EQ(total, 700);
-    ASSERT_EQ(count, 1);
+    ASSERT_EQ(count, 2);
  }
 }
 #endif  // ROCKSDB_LITE
@@ -621,11 +618,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextCfId) {
 // Compaction filters should only be applied to records that are newer than the
 // latest snapshot. This test inserts records and applies a delete filter.
 TEST_F(DBTestCompactionFilter, CompactionFilterSnapshot) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_filter_factory = std::make_shared<DeleteFilterFactory>();
  options.disable_auto_compactions = true;
  options.create_if_missing = true;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  // Put some data.
@@ -659,11 +655,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterSnapshot) {
 // records newer than the snapshot will also be processed
 TEST_F(DBTestCompactionFilter, CompactionFilterIgnoreSnapshot) {
  std::string five = ToString(5);
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_filter_factory = std::make_shared<DeleteISFilterFactory>();
  options.disable_auto_compactions = true;
  options.create_if_missing = true;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  // Put some data.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -10,6 +10,7 @@
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/experimental.h"
+#include "rocksdb/utilities/convenience.h"
 #include "util/sync_point.h"
 namespace rocksdb {

@@ -21,11 +22,13 @@ class DBCompactionTest : public DBTestBase {
  DBCompactionTest() : DBTestBase("/db_compaction_test") {}
 };

-class DBCompactionTestWithParam : public DBTestBase,
-                                public testing::WithParamInterface<uint32_t> {
+class DBCompactionTestWithParam
+    : public DBTestBase,
+      public testing::WithParamInterface<std::tuple<uint32_t, bool>> {
 public:
  DBCompactionTestWithParam() : DBTestBase("/db_compaction_test") {
-    max_subcompactions_ = GetParam();
+    max_subcompactions_ = std::get<0>(GetParam());
+    exclusive_manual_compaction_ = std::get<1>(GetParam());
  }

  // Required if inheriting from testing::WithParamInterface<>
@@ -33,43 +36,41 @@ class DBCompactionTestWithParam : public DBTestBase,
  static void TearDownTestCase() {}

  uint32_t max_subcompactions_;
+  bool exclusive_manual_compaction_;
 };

 namespace {
-class OnFileDeletionListener : public EventListener {
+
+class FlushedFileCollector : public EventListener {
 public:
-  OnFileDeletionListener() :
-      matched_count_(0),
-      expected_file_name_("") {}
+  FlushedFileCollector() {}
+  ~FlushedFileCollector() {}

-  void SetExpectedFileName(
-      const std::string file_name) {
-    expected_file_name_ = file_name;
+  virtual void OnFlushCompleted(DB* db, const FlushJobInfo& info) override {
+    std::lock_guard<std::mutex> lock(mutex_);
+    flushed_files_.push_back(info.file_path);
  }

-  void VerifyMatchedCount(size_t expected_value) {
-    ASSERT_EQ(matched_count_, expected_value);
-  }
-
-  void OnTableFileDeleted(
-      const TableFileDeletionInfo& info) override {
-    if (expected_file_name_ != "") {
-      ASSERT_EQ(expected_file_name_, info.file_path);
-      expected_file_name_ = "";
-      matched_count_++;
+  std::vector<std::string> GetFlushedFiles() {
+    std::lock_guard<std::mutex> lock(mutex_);
+    std::vector<std::string> result;
+    for (auto fname : flushed_files_) {
+      result.push_back(fname);
    }
+    return result;
  }

+  void ClearFlushedFiles() { flushed_files_.clear(); }
+
 private:
-  size_t matched_count_;
-  std::string expected_file_name_;
+  std::vector<std::string> flushed_files_;
+  std::mutex mutex_;
 };

 static const int kCDTValueSize = 1000;
 static const int kCDTKeysPerBuffer = 4;
 static const int kCDTNumLevels = 8;
-Options DeletionTriggerOptions() {
-  Options options;
+Options DeletionTriggerOptions(Options options) {
  options.compression = kNoCompression;
  options.write_buffer_size = kCDTKeysPerBuffer * (kCDTValueSize + 24);
  options.min_write_buffer_number_to_merge = 1;
@@ -160,7 +161,7 @@ const SstFileMetaData* PickFileRandomly(
      auto result = rand->Uniform(file_id);
      return &(level_meta.files[result]);
    }
-    file_id -= level_meta.files.size();
+    file_id -= static_cast<uint32_t>(level_meta.files.size());
  }
  assert(false);
  return nullptr;
@@ -172,7 +173,7 @@ const SstFileMetaData* PickFileRandomly(
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
  for (int tid = 0; tid < 3; ++tid) {
    uint64_t db_size[2];
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
    options.max_subcompactions = max_subcompactions_;

    if (tid == 1) {
@@ -215,8 +216,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
  // the compaction behavior when there are many of deletion entries.
  // The test will need to be updated if the internal behavior changes.

-  Options options = DeletionTriggerOptions();
-  options = CurrentOptions(options);
+  Options options = DeletionTriggerOptions(CurrentOptions());
  options.env = env_;
  DestroyAndReopen(options);
  Random rnd(301);
@@ -258,8 +258,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
 }

 TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
  options.env = env_;
  options.new_table_reader_for_compaction_inputs = true;
  options.max_open_files = 100;
@@ -348,7 +347,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTriggerReopen) {
  for (int tid = 0; tid < 2; ++tid) {
    uint64_t db_size[3];
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
    options.max_subcompactions = max_subcompactions_;

    if (tid == 1) {
@@ -405,7 +404,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTriggerReopen) {
 TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
  uint64_t db_size[3];
  for (int test = 0; test < 2; ++test) {
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
    options.skip_stats_update_on_db_open = (test == 0);

    env_->random_read_counter_.Reset();
@@ -463,14 +462,13 @@ TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
 TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
  const int kNumKeysPerFile = 100;

-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
  options.num_levels = 3;
  options.level0_file_num_compaction_trigger = 3;
  options.max_subcompactions = max_subcompactions_;
  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
-  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);
@@ -503,11 +501,107 @@ TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
  ASSERT_EQ(NumTableFilesAtLevel(1, 1), 1);
 }

+TEST_F(DBCompactionTest, BGCompactionsAllowed) {
+  // Create several column families. Make compaction triggers in all of them
+  // and see number of compactions scheduled to be less than allowed.
+  const int kNumKeysPerFile = 100;
+
+  Options options = CurrentOptions();
+  options.write_buffer_size = 110 << 10;  // 110KB
+  options.arena_block_size = 4 << 10;
+  options.num_levels = 3;
+  // Should speed up compaction when there are 4 files.
+  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));
+
+  // Block all threads in thread pool.
+  const size_t kTotalTasks = 4;
+  env_->SetBackgroundThreads(4, Env::LOW);
+  test::SleepingBackgroundTask sleeping_tasks[kTotalTasks];
+  for (size_t i = 0; i < kTotalTasks; i++) {
+    env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                   &sleeping_tasks[i], Env::Priority::LOW);
+    sleeping_tasks[i].WaitUntilSleeping();
+  }
+
+  CreateAndReopenWithCF({"one", "two", "three"}, options);
+
+  Random rnd(301);
+  for (int cf = 0; cf < 4; cf++) {
+    for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+      for (int i = 0; i < kNumKeysPerFile; i++) {
+        ASSERT_OK(Put(cf, Key(i), ""));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put(cf, "", ""));
+      dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+      ASSERT_EQ(NumTableFilesAtLevel(0, cf), num + 1);
+    }
+  }
+
+  // Now all column families qualify compaction but only one should be
+  // scheduled, because no column family hits speed up condition.
+  ASSERT_EQ(1, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
+
+  // Create two more files for one column family, which triggers speed up
+  // condition, three compactions will be scheduled.
+  for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+    for (int i = 0; i < kNumKeysPerFile; i++) {
+      ASSERT_OK(Put(2, Key(i), ""));
+    }
+    // put extra key to trigger flush
+    ASSERT_OK(Put(2, "", ""));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+    ASSERT_EQ(options.level0_file_num_compaction_trigger + num + 1,
+              NumTableFilesAtLevel(0, 2));
+  }
+  ASSERT_EQ(3, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
+
+  // Unblock all threads to unblock all compactions.
+  for (size_t i = 0; i < kTotalTasks; i++) {
+    sleeping_tasks[i].WakeUp();
+    sleeping_tasks[i].WaitUntilDone();
+  }
+  dbfull()->TEST_WaitForCompact();
+
+  // Verify number of compactions allowed will come back to 1.
+
+  for (size_t i = 0; i < kTotalTasks; i++) {
+    sleeping_tasks[i].Reset();
+    env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                   &sleeping_tasks[i], Env::Priority::LOW);
+    sleeping_tasks[i].WaitUntilSleeping();
+  }
+  for (int cf = 0; cf < 4; cf++) {
+    for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+      for (int i = 0; i < kNumKeysPerFile; i++) {
+        ASSERT_OK(Put(cf, Key(i), ""));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put(cf, "", ""));
+      dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+      ASSERT_EQ(NumTableFilesAtLevel(0, cf), num + 1);
+    }
+  }
+
+  // Now all column families qualify compaction but only one should be
+  // scheduled, because no column family hits speed up condition.
+  ASSERT_EQ(1, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
+
+  for (size_t i = 0; i < kTotalTasks; i++) {
+    sleeping_tasks[i].WakeUp();
+    sleeping_tasks[i].WaitUntilDone();
+  }
+}
+
 TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {
-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 100000000;        // Large write buffer
  options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);
@@ -534,9 +628,8 @@ TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {

 TEST_F(DBCompactionTest, MinorCompactionsHappen) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.write_buffer_size = 10000;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    const int N = 500;
@@ -560,14 +653,66 @@ TEST_F(DBCompactionTest, MinorCompactionsHappen) {
  } while (ChangeCompactOptions());
 }

+TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleLevel;
+  options.level0_file_num_compaction_trigger = 3;
+
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+
+  // compaction options
+  CompactionOptions compact_opt;
+  compact_opt.compression = kNoCompression;
+  compact_opt.output_file_size_limit = 4096;
+  const size_t key_len =
+    static_cast<size_t>(compact_opt.output_file_size_limit) / 5;
+
+  DestroyAndReopen(options);
+
+  std::vector<const Snapshot*> snaps;
+
+  // create first file and flush to l0
+  for (auto& key : {"1", "2", "3", "3", "3", "3"}) {
+    Put(key, std::string(key_len, 'A'));
+    snaps.push_back(dbfull()->GetSnapshot());
+  }
+  Flush();
+  dbfull()->TEST_WaitForFlushMemTable();
+
+  // create second file and flush to l0
+  for (auto& key : {"3", "4", "5", "6", "7", "8"}) {
+    Put(key, std::string(key_len, 'A'));
+    snaps.push_back(dbfull()->GetSnapshot());
+  }
+  Flush();
+  dbfull()->TEST_WaitForFlushMemTable();
+
+  // move both files down to l1
+  dbfull()->CompactFiles(compact_opt, collector->GetFlushedFiles(), 1);
+
+  // release snap so that first instance of key(3) can have seqId=0
+  for (auto snap : snaps) {
+    dbfull()->ReleaseSnapshot(snap);
+  }
+
+  // create 3 files in l0 so to trigger compaction
+  for (int i = 0; i < options.level0_file_num_compaction_trigger; i++) {
+    Put("2", std::string(1, 'A'));
+    Flush();
+    dbfull()->TEST_WaitForFlushMemTable();
+  }
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_OK(Put("", ""));
+}
+
 // Check that writes done during a memtable compaction are recovered
 // if the database is shutdown during the memtable compaction.
 TEST_F(DBCompactionTest, RecoverDuringMemtableCompaction) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.env = env_;
-    options.write_buffer_size = 1000000;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Trigger a long memtable compaction and reopen the database during it
@@ -591,10 +736,9 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 100000000;
  options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t num_keys = 80;
@@ -617,8 +761,11 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
  ASSERT_EQ(metadata.size(), 1U);
  LiveFileMetaData level0_file = metadata[0];  // L0 file meta

+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = exclusive_manual_compaction_;
+
  // Compaction will initiate a trivial move from L0 to L1
-  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  dbfull()->CompactRange(cro, nullptr, nullptr);

  // File moved From L0 to L1
  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);  // 0 files in L0
@@ -670,7 +817,7 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {

  Random rnd(301);
  std::map<int32_t, std::string> values;
-  for (uint32_t i = 0; i < ranges.size(); i++) {
+  for (size_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      values[j] = RandomString(&rnd, value_size);
      ASSERT_OK(Put(Key(j), values[j]));
@@ -682,14 +829,17 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
  ASSERT_EQ(level0_files, ranges.size());    // Multiple files in L0
  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0);  // No files in L1

+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = exclusive_manual_compaction_;
+
  // Since data is non-overlapping we expect compaction to initiate
  // a trivial move
-  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  db_->CompactRange(cro, nullptr, nullptr);
  // We expect that all the files were trivially moved from L0 to L1
  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);
  ASSERT_EQ(NumTableFilesAtLevel(1, 0) /* level1_files */, level0_files);

-  for (uint32_t i = 0; i < ranges.size(); i++) {
+  for (size_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      ASSERT_EQ(Get(Key(j)), values[j]);
    }
@@ -713,7 +863,7 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
    {500, 560},  // this range overlap with the next one
    {551, 599},
  };
-  for (uint32_t i = 0; i < ranges.size(); i++) {
+  for (size_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      values[j] = RandomString(&rnd, value_size);
      ASSERT_OK(Put(Key(j), values[j]));
@@ -721,9 +871,9 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
    ASSERT_OK(Flush());
  }

-  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  db_->CompactRange(cro, nullptr, nullptr);

-  for (uint32_t i = 0; i < ranges.size(); i++) {
+  for (size_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      ASSERT_EQ(Get(Key(j)), values[j]);
    }
@@ -777,6 +927,7 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveTargetLevel) {
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 6;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  // 2 files in L6
  ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel(0));
@@ -792,6 +943,386 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveTargetLevel) {
  }
 }

+TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
+  int32_t trivial_move = 0;
+  int32_t non_trivial_move = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:TrivialMove",
+      [&](void* arg) { trivial_move++; });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:NonTrivial",
+      [&](void* arg) { non_trivial_move++; });
+  bool first = true;
+  bool second = true;
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBCompaction::ManualPartial:4", "DBCompaction::ManualPartial:1"},
+       {"DBCompaction::ManualPartial:2", "DBCompaction::ManualPartial:3"},
+       {"DBCompaction::ManualPartial:5",
+        "DBImpl::BackgroundCompaction:NonTrivial:AfterRun"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+        if (first) {
+          TEST_SYNC_POINT("DBCompaction::ManualPartial:4");
+          first = false;
+          TEST_SYNC_POINT("DBCompaction::ManualPartial:3");
+        } else if (second) {
+          TEST_SYNC_POINT("DBCompaction::ManualPartial:2");
+        }
+      });
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1024 * 1024;
+  options.num_levels = 7;
+  options.max_subcompactions = max_subcompactions_;
+  options.level0_file_num_compaction_trigger = 3;
+  options.max_background_compactions = 3;
+  options.target_file_size_base = 1 << 23;  // 8 MB
+
+  DestroyAndReopen(options);
+  int32_t value_size = 10 * 1024;  // 10 KB
+
+  // Add 2 non-overlapping files
+  Random rnd(301);
+  std::map<int32_t, std::string> values;
+
+  // file 1 [0 => 100]
+  for (int32_t i = 0; i < 100; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // file 2 [100 => 300]
+  for (int32_t i = 100; i < 300; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 2 files in L0
+  ASSERT_EQ("2", FilesPerLevel(0));
+  CompactRangeOptions compact_options;
+  compact_options.change_level = true;
+  compact_options.target_level = 6;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
+  // Trivial move the two non-overlapping files to level 6
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  // 2 files in L6
+  ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel(0));
+
+  ASSERT_EQ(trivial_move, 1);
+  ASSERT_EQ(non_trivial_move, 0);
+
+  // file 3 [ 0 => 200]
+  for (int32_t i = 0; i < 200; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 1 files in L0
+  ASSERT_EQ("1,0,0,0,0,0,2", FilesPerLevel(0));
+  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr, false));
+  ASSERT_OK(dbfull()->TEST_CompactRange(1, nullptr, nullptr, nullptr, false));
+  ASSERT_OK(dbfull()->TEST_CompactRange(2, nullptr, nullptr, nullptr, false));
+  ASSERT_OK(dbfull()->TEST_CompactRange(3, nullptr, nullptr, nullptr, false));
+  ASSERT_OK(dbfull()->TEST_CompactRange(4, nullptr, nullptr, nullptr, false));
+  // 2 files in L6, 1 file in L5
+  ASSERT_EQ("0,0,0,0,0,1,2", FilesPerLevel(0));
+
+  ASSERT_EQ(trivial_move, 6);
+  ASSERT_EQ(non_trivial_move, 0);
+
+  std::thread threads([&] {
+    compact_options.change_level = false;
+    compact_options.exclusive_manual_compaction = false;
+    std::string begin_string = Key(0);
+    std::string end_string = Key(199);
+    Slice begin(begin_string);
+    Slice end(end_string);
+    ASSERT_OK(db_->CompactRange(compact_options, &begin, &end));
+  });
+
+  TEST_SYNC_POINT("DBCompaction::ManualPartial:1");
+  // file 4 [300 => 400)
+  for (int32_t i = 300; i <= 400; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // file 5 [400 => 500)
+  for (int32_t i = 400; i <= 500; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // file 6 [500 => 600)
+  for (int32_t i = 500; i <= 600; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 3 files in L0
+  ASSERT_EQ("3,0,0,0,0,1,2", FilesPerLevel(0));
+  TEST_SYNC_POINT("DBCompaction::ManualPartial:5");
+
+  // 1 file in L6, 1 file in L1
+  dbfull()->TEST_WaitForFlushMemTable();
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ("0,1,0,0,0,0,1", FilesPerLevel(0));
+  threads.join();
+
+  for (int32_t i = 0; i < 600; i++) {
+    ASSERT_EQ(Get(Key(i)), values[i]);
+  }
+}
+
+TEST_F(DBCompactionTest, ManualPartialFill) {
+  int32_t trivial_move = 0;
+  int32_t non_trivial_move = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:TrivialMove",
+      [&](void* arg) { trivial_move++; });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:NonTrivial",
+      [&](void* arg) { non_trivial_move++; });
+  bool first = true;
+  bool second = true;
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBCompaction::PartialFill:4", "DBCompaction::PartialFill:1"},
+       {"DBCompaction::PartialFill:2", "DBCompaction::PartialFill:3"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
+        if (first) {
+          TEST_SYNC_POINT("DBCompaction::PartialFill:4");
+          first = false;
+          TEST_SYNC_POINT("DBCompaction::PartialFill:3");
+        } else if (second) {
+        }
+      });
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1024 * 1024;
+  options.max_bytes_for_level_multiplier = 2;
+  options.num_levels = 4;
+  options.level0_file_num_compaction_trigger = 3;
+  options.max_background_compactions = 3;
+
+  DestroyAndReopen(options);
+  int32_t value_size = 10 * 1024;  // 10 KB
+
+  // Add 2 non-overlapping files
+  Random rnd(301);
+  std::map<int32_t, std::string> values;
+
+  // file 1 [0 => 100]
+  for (int32_t i = 0; i < 100; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // file 2 [100 => 300]
+  for (int32_t i = 100; i < 300; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 2 files in L0
+  ASSERT_EQ("2", FilesPerLevel(0));
+  CompactRangeOptions compact_options;
+  compact_options.change_level = true;
+  compact_options.target_level = 2;
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  // 2 files in L2
+  ASSERT_EQ("0,0,2", FilesPerLevel(0));
+
+  ASSERT_EQ(trivial_move, 1);
+  ASSERT_EQ(non_trivial_move, 0);
+
+  // file 3 [ 0 => 200]
+  for (int32_t i = 0; i < 200; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 2 files in L2, 1 in L0
+  ASSERT_EQ("1,0,2", FilesPerLevel(0));
+  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr, false));
+  // 2 files in L2, 1 in L1
+  ASSERT_EQ("0,1,2", FilesPerLevel(0));
+
+  ASSERT_EQ(trivial_move, 2);
+  ASSERT_EQ(non_trivial_move, 0);
+
+  std::thread threads([&] {
+    compact_options.change_level = false;
+    compact_options.exclusive_manual_compaction = false;
+    std::string begin_string = Key(0);
+    std::string end_string = Key(199);
+    Slice begin(begin_string);
+    Slice end(end_string);
+    ASSERT_OK(db_->CompactRange(compact_options, &begin, &end));
+  });
+
+  TEST_SYNC_POINT("DBCompaction::PartialFill:1");
+  // Many files 4 [300 => 4300)
+  for (int32_t i = 0; i <= 5; i++) {
+    for (int32_t j = 300; j < 4300; j++) {
+      if (j == 2300) {
+        ASSERT_OK(Flush());
+        dbfull()->TEST_WaitForFlushMemTable();
+      }
+      values[j] = RandomString(&rnd, value_size);
+      ASSERT_OK(Put(Key(j), values[j]));
+    }
+  }
+
+  // Verify level sizes
+  uint64_t target_size = 4 * options.max_bytes_for_level_base;
+  for (int32_t i = 1; i < options.num_levels; i++) {
+    ASSERT_LE(SizeAtLevel(i), target_size);
+    target_size *= options.max_bytes_for_level_multiplier;
+  }
+
+  TEST_SYNC_POINT("DBCompaction::PartialFill:2");
+  dbfull()->TEST_WaitForFlushMemTable();
+  dbfull()->TEST_WaitForCompact();
+  threads.join();
+
+  for (int32_t i = 0; i < 4300; i++) {
+    ASSERT_EQ(Get(Key(i)), values[i]);
+  }
+}
+
+TEST_F(DBCompactionTest, DeleteFileRange) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1024 * 1024;
+  options.max_bytes_for_level_multiplier = 2;
+  options.num_levels = 4;
+  options.level0_file_num_compaction_trigger = 3;
+  options.max_background_compactions = 3;
+
+  DestroyAndReopen(options);
+  int32_t value_size = 10 * 1024;  // 10 KB
+
+  // Add 2 non-overlapping files
+  Random rnd(301);
+  std::map<int32_t, std::string> values;
+
+  // file 1 [0 => 100]
+  for (int32_t i = 0; i < 100; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // file 2 [100 => 300]
+  for (int32_t i = 100; i < 300; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // 2 files in L0
+  ASSERT_EQ("2", FilesPerLevel(0));
+  CompactRangeOptions compact_options;
+  compact_options.change_level = true;
+  compact_options.target_level = 2;
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  // 2 files in L2
+  ASSERT_EQ("0,0,2", FilesPerLevel(0));
+
+  // file 3 [ 0 => 200]
+  for (int32_t i = 0; i < 200; i++) {
+    values[i] = RandomString(&rnd, value_size);
+    ASSERT_OK(Put(Key(i), values[i]));
+  }
+  ASSERT_OK(Flush());
+
+  // Many files 4 [300 => 4300)
+  for (int32_t i = 0; i <= 5; i++) {
+    for (int32_t j = 300; j < 4300; j++) {
+      if (j == 2300) {
+        ASSERT_OK(Flush());
+        dbfull()->TEST_WaitForFlushMemTable();
+      }
+      values[j] = RandomString(&rnd, value_size);
+      ASSERT_OK(Put(Key(j), values[j]));
+    }
+  }
+  ASSERT_OK(Flush());
+  dbfull()->TEST_WaitForFlushMemTable();
+  dbfull()->TEST_WaitForCompact();
+
+  // Verify level sizes
+  uint64_t target_size = 4 * options.max_bytes_for_level_base;
+  for (int32_t i = 1; i < options.num_levels; i++) {
+    ASSERT_LE(SizeAtLevel(i), target_size);
+    target_size *= options.max_bytes_for_level_multiplier;
+  }
+
+  size_t old_num_files = CountFiles();
+  std::string begin_string = Key(1000);
+  std::string end_string = Key(2000);
+  Slice begin(begin_string);
+  Slice end(end_string);
+  ASSERT_OK(DeleteFilesInRange(db_, db_->DefaultColumnFamily(), &begin, &end));
+
+  int32_t deleted_count = 0;
+  for (int32_t i = 0; i < 4300; i++) {
+    if (i < 1000 || i > 2000) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    } else {
+      ReadOptions roptions;
+      std::string result;
+      Status s = db_->Get(roptions, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound() || s.ok());
+      if (s.IsNotFound()) {
+        deleted_count++;
+      }
+    }
+  }
+  ASSERT_GT(deleted_count, 0);
+  begin_string = Key(5000);
+  end_string = Key(6000);
+  Slice begin1(begin_string);
+  Slice end1(end_string);
+  // Try deleting files in range which contain no keys
+  ASSERT_OK(
+      DeleteFilesInRange(db_, db_->DefaultColumnFamily(), &begin1, &end1));
+
+  // Push data from level 0 to level 1 to force all data to be deleted
+  // Note that we don't delete level 0 files
+  compact_options.change_level = true;
+  compact_options.target_level = 1;
+  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
+
+  ASSERT_OK(
+      DeleteFilesInRange(db_, db_->DefaultColumnFamily(), nullptr, nullptr));
+
+  int32_t deleted_count2 = 0;
+  for (int32_t i = 0; i < 4300; i++) {
+    ReadOptions roptions;
+    std::string result;
+    Status s = db_->Get(roptions, Key(i), &result);
+    ASSERT_TRUE(s.IsNotFound());
+    deleted_count2++;
+  }
+  ASSERT_GT(deleted_count2, deleted_count);
+  size_t new_num_files = CountFiles();
+  ASSERT_GT(old_num_files, new_num_files);
+}
+
 TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
@@ -803,10 +1334,9 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 100000000;
  options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t value_size = 10 * 1024;  // 10 KB
@@ -825,6 +1355,7 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 3;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 1);
@@ -838,8 +1369,10 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
  ASSERT_OK(Flush());

  ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = exclusive_manual_compaction_;
  // Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
  ASSERT_EQ("0,0,0,2", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 4);
  ASSERT_EQ(non_trivial_move, 0);
@@ -1092,7 +1625,7 @@ TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
  int max_key_universal_insert = 600;

  // Stage 1: generate a db with level compaction
-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
  options.num_levels = 4;
@@ -1102,7 +1635,6 @@ TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
  options.target_file_size_base = 200 << 10;  // 200KB
  options.target_file_size_multiplier = 1;
  options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  for (int i = 0; i <= max_key_level_insert; i++) {
@@ -1143,6 +1675,7 @@ TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
  compact_options.target_level = 0;
  compact_options.bottommost_level_compaction =
      BottommostLevelCompaction::kForce;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  dbfull()->CompactRange(compact_options, handles_[1], nullptr, nullptr);

  // Only 1 file in L0
@@ -1276,7 +1809,9 @@ TEST_P(DBCompactionTestWithParam, ManualCompaction) {

    uint64_t prev_block_cache_add =
        options.statistics->getTickerCount(BLOCK_CACHE_ADD);
-    db_->CompactRange(CompactRangeOptions(), handles_[1], nullptr, nullptr);
+    CompactRangeOptions cro;
+    cro.exclusive_manual_compaction = exclusive_manual_compaction_;
+    db_->CompactRange(cro, handles_[1], nullptr, nullptr);
    // Verify manual compaction doesn't fill block cache
    ASSERT_EQ(prev_block_cache_add,
              options.statistics->getTickerCount(BLOCK_CACHE_ADD));
@@ -1355,6 +1890,7 @@ TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[0].path));
    CompactRangeOptions compact_options;
    compact_options.target_path_id = 1;
+    compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
    db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);

    ASSERT_EQ("0,1", FilesPerLevel(1));
@@ -1423,7 +1959,7 @@ TEST_P(DBCompactionTestWithParam, DISABLED_CompactFilesOnLevelCompaction) {
    std::set<std::string> overlapping_file_names;
    std::vector<std::string> compaction_input_file_names;
    for (int f = 0; f < file_picked; ++f) {
-      int level;
+      int level = 0;
      auto file_meta = PickFileRandomly(cf_meta, &rnd, &level);
      compaction_input_file_names.push_back(file_meta->name);
      GetOverlappingFileNumbersForLevelCompaction(
@@ -1723,6 +2259,25 @@ TEST_P(DBCompactionTestWithParam, CompressLevelCompaction) {
  Destroy(options);
 }

+TEST_F(DBCompactionTest, SanitizeCompactionOptionsTest) {
+  Options options = CurrentOptions();
+  options.max_background_compactions = 5;
+  options.soft_pending_compaction_bytes_limit = 0;
+  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);
+}
+
 // This tests for a bug that could cause two level0 compactions running
 // concurrently
 // TODO(aekmekji): Make sure that the reason this fails when run with
@@ -1794,10 +2349,9 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.write_buffer_size = 100000000;
  options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t value_size = 10 * 1024;  // 10 KB
@@ -1867,7 +2421,61 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
 }

 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
-                        ::testing::Values(1, 4));
+                        ::testing::Values(std::make_tuple(1, true),
+                                          std::make_tuple(1, false),
+                                          std::make_tuple(4, true),
+                                          std::make_tuple(4, false)));
+
+class CompactionPriTest : public DBTestBase,
+                          public testing::WithParamInterface<uint32_t> {
+ public:
+  CompactionPriTest() : DBTestBase("/compaction_pri_test") {
+    compaction_pri_ = GetParam();
+  }
+
+  // Required if inheriting from testing::WithParamInterface<>
+  static void SetUpTestCase() {}
+  static void TearDownTestCase() {}
+
+  uint32_t compaction_pri_;
+};
+
+TEST_P(CompactionPriTest, Test) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 16 * 1024;
+  options.compaction_pri = static_cast<CompactionPri>(compaction_pri_);
+  options.hard_pending_compaction_bytes_limit = 256 * 1024;
+  options.max_bytes_for_level_base = 64 * 1024;
+  options.max_bytes_for_level_multiplier = 4;
+  options.compression = kNoCompression;
+
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+  const int kNKeys = 5000;
+  int keys[kNKeys];
+  for (int i = 0; i < kNKeys; i++) {
+    keys[i] = i;
+  }
+  std::random_shuffle(std::begin(keys), std::end(keys));
+
+  for (int i = 0; i < kNKeys; i++) {
+    ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 102)));
+  }
+
+  dbfull()->TEST_WaitForCompact();
+  for (int i = 0; i < kNKeys; i++) {
+    ASSERT_NE("NOT_FOUND", Get(Key(i)));
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    CompactionPriTest, CompactionPriTest,
+    ::testing::Values(CompactionPri::kByCompensatedSize,
+                      CompactionPri::kOldestLargestSeqFirst,
+                      CompactionPri::kOldestSmallestSeqFirst,
+                      CompactionPri::kMinOverlappingRatio));
+
 #endif // !defined(ROCKSDB_LITE)
 }  // namespace rocksdb

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -125,15 +125,15 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {

  Options options = CurrentOptions();
  options.create_if_missing = true;
-  options.db_write_buffer_size = 2048;
-  options.write_buffer_size = 2048;
+  options.db_write_buffer_size = 204800;
+  options.write_buffer_size = 20480;
  options.max_write_buffer_number = 2;
  options.level0_file_num_compaction_trigger = 2;
  options.level0_slowdown_writes_trigger = 9999;
  options.level0_stop_writes_trigger = 9999;
-  options.target_file_size_base = 2048;
+  options.target_file_size_base = 9102;
  options.level_compaction_dynamic_level_bytes = true;
-  options.max_bytes_for_level_base = 10240;
+  options.max_bytes_for_level_base = 40960;
  options.max_bytes_for_level_multiplier = 4;
  options.max_background_compactions = 2;
  options.num_levels = 5;
@@ -154,10 +154,10 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
  ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
  ASSERT_EQ(4U, int_prop);

-  // Put about 7K to L0
+  // Put about 28K to L0
  for (int i = 0; i < 70; i++) {
    ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
  }
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "false"},
@@ -167,14 +167,14 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
  ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
  ASSERT_EQ(4U, int_prop);

-  // Insert extra about 3.5K to L0. After they are compacted to L4, base level
+  // Insert extra about 28K to L0. After they are compacted to L4, base level
  // should be changed to L3.
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "true"},
  }));
  for (int i = 0; i < 70; i++) {
    ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
  }

  ASSERT_OK(dbfull()->SetOptions({
@@ -199,10 +199,10 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "true"},
  }));
-  // Write about 10K more
+  // Write about 40K more
  for (int i = 0; i < 100; i++) {
    ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
  }
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "false"},
@@ -218,15 +218,15 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {

  // Trigger a condition that the compaction changes base level and L0->Lbase
  // happens at the same time.
-  // We try to make last levels' targets to be 10K, 40K, 160K, add triggers
+  // We try to make last levels' targets to be 40K, 160K, 640K, add triggers
  // another compaction from 40K->160K.
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "true"},
  }));
-  // Write about 150K more
-  for (int i = 0; i < 1350; i++) {
+  // Write about 600K more
+  for (int i = 0; i < 1500; i++) {
    ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
  }
  ASSERT_OK(dbfull()->SetOptions({
      {"disable_auto_compactions", "false"},
@@ -236,29 +236,39 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
  ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
  ASSERT_EQ(2U, int_prop);

+  // A manual compaction will trigger the base level to become L2
  // Keep Writing data until base level changed 2->1. There will be L0->L2
  // compaction going on at the same time.
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  for (int attempt = 0; attempt <= 20; attempt++) {
-    // Write about 5K more data with two flushes. It should be flush to level 2
-    // but when it is applied, base level is already 1.
-    for (int i = 0; i < 50; i++) {
-      ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                    RandomString(&rnd, 80)));
-    }
-    Flush();
-
-    ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
-    if (int_prop == 2U) {
-      env_->SleepForMicroseconds(50000);
-    } else {
-      break;
-    }
-  }
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();

-  env_->SleepForMicroseconds(200000);
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"CompactionJob::Run():Start", "DynamicLevelMaxBytesBase2:0"},
+      {"DynamicLevelMaxBytesBase2:1", "CompactionJob::Run():End"},
+      {"DynamicLevelMaxBytesBase2:compact_range_finish",
+       "FlushJob::WriteLevel0Table"},
+  });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::thread thread([this] {
+    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_start");
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_finish");
+  });
+
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:0");
+  for (int i = 0; i < 2; i++) {
+    ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
+                  RandomString(&rnd, 380)));
+  }
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:1");
+
+  Flush();
+
+  thread.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();

  ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
  ASSERT_EQ(1U, int_prop);
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -10,8 +10,10 @@

 #include <atomic>
 #include <deque>
+#include <functional>
 #include <limits>
 #include <list>
+#include <queue>
 #include <set>
 #include <string>
 #include <utility>
@@ -55,6 +57,7 @@ class Arena;
 class WriteCallback;
 struct JobContext;
 struct ExternalSstFileInfo;
+struct MemTableInfo;

 class DBImpl : public DB {
 public:
@@ -188,6 +191,8 @@ class DBImpl : public DB {
      const TransactionLogIterator::ReadOptions&
          read_options = TransactionLogIterator::ReadOptions()) override;
  virtual Status DeleteFile(std::string name) override;
+  Status DeleteFilesInRange(ColumnFamilyHandle* column_family,
+                            const Slice* begin, const Slice* end);

  virtual void GetLiveFilesMetaData(
      std::vector<LiveFileMetaData>* metadata) override;
@@ -226,13 +231,30 @@ class DBImpl : public DB {
                                                   bool include_history);

  // For a given key, check to see if there are any records for this key
-  // in the memtables, including memtable history.
-
-  // On success, *seq will contain the sequence number for the
-  // latest such change or kMaxSequenceNumber if no records were present.
-  // Returns OK on success, other status on error reading memtables.
-  Status GetLatestSequenceForKeyFromMemtable(SuperVersion* sv, const Slice& key,
-                                             SequenceNumber* seq);
+  // in the memtables, including memtable history.  If cache_only is false,
+  // SST files will also be checked.
+  //
+  // If a key is found, *found_record_for_key will be set to true and
+  // *seq will be set to the stored sequence number for the latest
+  // operation on this key or kMaxSequenceNumber if unknown.
+  // If no key is found, *found_record_for_key will be set to false.
+  //
+  // Note: If cache_only=false, it is possible for *seq to be set to 0 if
+  // the sequence number has been cleared from the record.  If the caller is
+  // holding an active db snapshot, we know the missing sequence must be less
+  // than the snapshot's sequence number (sequence numbers are only cleared
+  // when there are no earlier active snapshots).
+  //
+  // If NotFound is returned and found_record_for_key is set to false, then no
+  // record for this key was found.  If the caller is holding an active db
+  // snapshot, we know that no key could have existing after this snapshot
+  // (since we do not compact keys that have an earlier snapshot).
+  //
+  // Returns OK or NotFound on success,
+  // other status on unexpected error.
+  Status GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
+                                 bool cache_only, SequenceNumber* seq,
+                                 bool* found_record_for_key);

  using DB::AddFile;
  virtual Status AddFile(ColumnFamilyHandle* column_family,
@@ -243,6 +265,12 @@ class DBImpl : public DB {

 #endif  // ROCKSDB_LITE

+  // Similar to GetSnapshot(), but also lets the db know that this snapshot
+  // will be used for transaction write-conflict checking.  The DB can then
+  // make sure not to compact any keys that would prevent a write-conflict from
+  // being detected.
+  const Snapshot* GetSnapshotForWriteConflictBoundary();
+
  // checks if all live files exist on file system and that their file sizes
  // match to our in-memory records
  virtual Status CheckConsistency();
@@ -252,6 +280,7 @@ class DBImpl : public DB {
  Status RunManualCompaction(ColumnFamilyData* cfd, int input_level,
                             int output_level, uint32_t output_path_id,
                             const Slice* begin, const Slice* end,
+                             bool exclusive,
                             bool disallow_trivial_move = false);

  // Return an internal iterator over the current state of the database.
@@ -270,7 +299,8 @@ class DBImpl : public DB {
                           bool disallow_trivial_move = false);

  // Force current memtable contents to be flushed.
-  Status TEST_FlushMemTable(bool wait = true);
+  Status TEST_FlushMemTable(bool wait = true,
+                            ColumnFamilyHandle* cfh = nullptr);

  // Wait for memtable compaction
  Status TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family = nullptr);
@@ -315,14 +345,23 @@ 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
+  Status TEST_GetLatestMutableCFOptions(ColumnFamilyHandle* column_family,
+                                        MutableCFOptions* mutable_cf_opitons);
+
  Cache* TEST_table_cache() { return table_cache_.get(); }

-  const WriteController& TEST_write_controler() const {
-    return write_controller_;
-  }
+  WriteController& TEST_write_controler() { return write_controller_; }
+
+  uint64_t TEST_FindMinLogContainingOutstandingPrep();
+  uint64_t TEST_FindMinPrepLogReferencedByMemTable();

 #endif  // NDEBUG

+  // Return maximum background compaction alowed to be scheduled based on
+  // compaction status.
+  int BGCompactionsAllowed() const;
+
  // Returns the list of live files in 'live' and the list
  // of all files in the filesystem in 'candidate_files'.
  // If force == false and the last call was less than
@@ -393,12 +432,70 @@ class DBImpl : public DB {
    return num_running_compactions_;
  }

+  // hollow transactions shell used for recovery.
+  // these will then be passed to TransactionDB so that
+  // locks can be reacquired before writing can resume.
+  struct RecoveredTransaction {
+    uint64_t log_number_;
+    std::string name_;
+    WriteBatch* batch_;
+    explicit RecoveredTransaction(const uint64_t log, const std::string& name,
+                                  WriteBatch* batch)
+        : log_number_(log), name_(name), batch_(batch) {}
+
+    ~RecoveredTransaction() { delete batch_; }
+  };
+
+  bool allow_2pc() const { return db_options_.allow_2pc; }
+
+  std::unordered_map<std::string, RecoveredTransaction*>
+  recovered_transactions() {
+    return recovered_transactions_;
+  }
+
+  RecoveredTransaction* GetRecoveredTransaction(const std::string& name) {
+    auto it = recovered_transactions_.find(name);
+    if (it == recovered_transactions_.end()) {
+      return nullptr;
+    } else {
+      return it->second;
+    }
+  }
+
+  void InsertRecoveredTransaction(const uint64_t log, const std::string& name,
+                                  WriteBatch* batch) {
+    recovered_transactions_[name] = new RecoveredTransaction(log, name, batch);
+    MarkLogAsContainingPrepSection(log);
+  }
+
+  void DeleteRecoveredTransaction(const std::string& name) {
+    auto it = recovered_transactions_.find(name);
+    assert(it != recovered_transactions_.end());
+    auto* trx = it->second;
+    recovered_transactions_.erase(it);
+    MarkLogAsHavingPrepSectionFlushed(trx->log_number_);
+    delete trx;
+  }
+
+  void DeleteAllRecoveredTransactions() {
+    for (auto it = recovered_transactions_.begin();
+         it != recovered_transactions_.end(); it++) {
+      delete it->second;
+    }
+    recovered_transactions_.clear();
+  }
+
+  void MarkLogAsHavingPrepSectionFlushed(uint64_t log);
+  void MarkLogAsContainingPrepSection(uint64_t log);
+
 protected:
  Env* const env_;
  const std::string dbname_;
  unique_ptr<VersionSet> versions_;
  const DBOptions db_options_;
  Statistics* stats_;
+  std::unordered_map<std::string, RecoveredTransaction*>
+      recovered_transactions_;

  InternalIterator* NewInternalIterator(const ReadOptions&,
                                        ColumnFamilyData* cfd,
@@ -424,6 +521,8 @@ class DBImpl : public DB {
                                   Compaction *c, const Status &st,
                                   const CompactionJobStats& job_stats,
                                   int job_id);
+  void NotifyOnMemTableSealed(ColumnFamilyData* cfd, 
+                              const MemTableInfo& mem_table_info);

  void NewThreadStatusCfInfo(ColumnFamilyData* cfd) const;

@@ -432,11 +531,17 @@ class DBImpl : public DB {
  void EraseThreadStatusDbInfo() const;

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

 private:
  friend class DB;
  friend class InternalStats;
+  friend class TransactionImpl;
 #ifndef ROCKSDB_LITE
  friend class ForwardIterator;
 #endif
@@ -455,7 +560,8 @@ class DBImpl : public DB {
  // amount of work to recover recently logged updates.  Any changes to
  // be made to the descriptor are added to *edit.
  Status Recover(const std::vector<ColumnFamilyDescriptor>& column_families,
-                 bool read_only = false, bool error_if_log_file_exist = false);
+                 bool read_only = false, bool error_if_log_file_exist = false,
+                 bool error_if_data_exists_in_logs = false);

  void MaybeIgnoreError(Status* s) const;

@@ -466,10 +572,11 @@ class DBImpl : public DB {

  // Background process needs to call
  //     auto x = CaptureCurrentFileNumberInPendingOutputs()
+  //     auto file_num = versions_->NewFileNumber();
  //     <do something>
  //     ReleaseFileNumberFromPendingOutputs(x)
-  // This will protect any temporary files created while <do something> is
-  // executing from being deleted.
+  // This will protect any file with number `file_num` or greater from being
+  // deleted while <do something> is running.
  // -----------
  // This function will capture current file number and append it to
  // pending_outputs_. This will prevent any background process to delete any
@@ -494,7 +601,7 @@ class DBImpl : public DB {
                         SequenceNumber* max_sequence, bool read_only);

  // The following two methods are used to flush a memtable to
-  // storage. The first one is used atdatabase RecoveryTime (when the
+  // storage. The first one is used at database RecoveryTime (when the
  // database is opened) and is heavyweight because it holds the mutex
  // for the entire period. The second method WriteLevel0Table supports
  // concurrent flush memtables to storage.
@@ -515,9 +622,6 @@ class DBImpl : public DB {
  // Wait for memtable flushed
  Status WaitForFlushMemTable(ColumnFamilyData* cfd);

-  void RecordFlushIOStats();
-  void RecordCompactionIOStats();
-
 #ifndef ROCKSDB_LITE
  Status CompactFilesImpl(
      const CompactionOptions& compact_options, ColumnFamilyData* cfd,
@@ -531,12 +635,13 @@ class DBImpl : public DB {
  void MaybeScheduleFlushOrCompaction();
  void SchedulePendingFlush(ColumnFamilyData* cfd);
  void SchedulePendingCompaction(ColumnFamilyData* cfd);
-  static void BGWorkCompaction(void* db);
+  static void BGWorkCompaction(void* arg);
  static void BGWorkFlush(void* db);
-  void BackgroundCallCompaction();
+  static void UnscheduleCallback(void* arg);
+  void BackgroundCallCompaction(void* arg);
  void BackgroundCallFlush();
  Status BackgroundCompaction(bool* madeProgress, JobContext* job_context,
-                              LogBuffer* log_buffer);
+                              LogBuffer* log_buffer, void* m = 0);
  Status BackgroundFlush(bool* madeProgress, JobContext* job_context,
                         LogBuffer* log_buffer);

@@ -564,6 +669,8 @@ class DBImpl : public DB {
  // helper function to call after some of the logs_ were synced
  void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);

+  const Snapshot* GetSnapshotImpl(bool is_write_conflict_boundary);
+
  // table_cache_ provides its own synchronization
  std::shared_ptr<Cache> table_cache_;

@@ -580,7 +687,7 @@ class DBImpl : public DB {
  std::atomic<bool> shutting_down_;
  // This condition variable is signaled on these conditions:
  // * whenever bg_compaction_scheduled_ goes down to 0
-  // * if bg_manual_only_ > 0, whenever a compaction finishes, even if it hasn't
+  // * if AnyManualCompaction, whenever a compaction finishes, even if it hasn't
  // made any progress
  // * whenever a compaction made any progress
  // * whenever bg_flush_scheduled_ value decreases (i.e. whenever a flush is
@@ -738,11 +845,6 @@ class DBImpl : public DB {
  // stores the number of compactions are currently running
  int num_running_compactions_;

-  // If non-zero, MaybeScheduleFlushOrCompaction() will only schedule manual
-  // compactions (if manual_compaction_ is not null). This mechanism enables
-  // manual compactions to wait until all other compactions are finished.
-  int bg_manual_only_;
-
  // number of background memtable flush jobs, submitted to the HIGH pool
  int bg_flush_scheduled_;

@@ -755,15 +857,25 @@ class DBImpl : public DB {
    int input_level;
    int output_level;
    uint32_t output_path_id;
-    bool done;
    Status status;
+    bool done;
    bool in_progress;             // compaction request being processed?
+    bool incomplete;              // only part of requested range compacted
+    bool exclusive;               // current behavior of only one manual
+    bool disallow_trivial_move;   // Force actual compaction to run
    const InternalKey* begin;     // nullptr means beginning of key range
    const InternalKey* end;       // nullptr means end of key range
+    InternalKey* manual_end;      // how far we are compacting
    InternalKey tmp_storage;      // Used to keep track of compaction progress
-    bool disallow_trivial_move;   // Force actual compaction to run
+    InternalKey tmp_storage1;     // Used to keep track of compaction progress
+    Compaction* compaction;
+  };
+  std::deque<ManualCompaction*> manual_compaction_dequeue_;
+
+  struct CompactionArg {
+    DBImpl* db;
+    ManualCompaction* m;
  };
-  ManualCompaction* manual_compaction_;

  // Have we encountered a background error in paranoid mode?
  Status bg_error_;
@@ -786,7 +898,10 @@ class DBImpl : public DB {
  // they're unique
  std::atomic<int> next_job_id_;

-  bool flush_on_destroy_; // Used when disableWAL is true.
+  // A flag indicating whether the current rocksdb database has any
+  // data that is not yet persisted into either WAL or SST file.
+  // Used when disableWAL is true.
+  bool has_unpersisted_data_;

  static const int KEEP_LOG_FILE_NUM = 1000;
  // MSVC version 1800 still does not have constexpr for ::max()
@@ -804,15 +919,40 @@ class DBImpl : public DB {
  // Unified interface for logging events
  EventLogger event_logger_;

-  // A value of >0 temporarily disables scheduling of background work
+  // A value of > 0 temporarily disables scheduling of background work
  int bg_work_paused_;

+  // A value of > 0 temporarily disables scheduling of background compaction
+  int bg_compaction_paused_;
+
  // Guard against multiple concurrent refitting
  bool refitting_level_;

  // Indicate DB was opened successfully
  bool opened_successfully_;

+  // minmum 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
+  std::priority_queue<uint64_t, std::vector<uint64_t>, std::greater<uint64_t>>
+      min_log_with_prep_;
+
+  // to be used in conjunction with min_log_with_prep_.
+  // once a transaction with data in log L is committed or rolled back
+  // rather than removing the value from the heap we add that value
+  // 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
+  // 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
+  // min_log_with_prep_. This will work the same as a min_heap
+  // where we are deleteing arbitrary elements and the up heaping.
+  std::unordered_map<uint64_t, uint64_t> prepared_section_completed_;
+  std::mutex prep_heap_mutex_;
+
  // No copying allowed
  DBImpl(const DBImpl&);
  void operator=(const DBImpl&);
@@ -857,9 +997,16 @@ class DBImpl : public DB {
                 bool* value_found = nullptr);

  bool GetIntPropertyInternal(ColumnFamilyData* cfd,
-                              DBPropertyType property_type,
-                              bool need_out_of_mutex, bool is_locked,
-                              uint64_t* value);
+                              const DBPropertyInfo& property_info,
+                              bool is_locked, uint64_t* value);
+
+  bool HasPendingManualCompaction();
+  bool HasExclusiveManualCompaction();
+  void AddManualCompaction(ManualCompaction* m);
+  void RemoveManualCompaction(ManualCompaction* m);
+  bool ShouldntRunManualCompaction(ManualCompaction* m);
+  bool HaveManualCompaction(ColumnFamilyData* cfd);
+  bool MCOverlap(ManualCompaction* m, ManualCompaction* m1);
 };

 // Sanitize db options.  The caller should delete result.info_log if
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -70,14 +70,21 @@ Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
       cfd->ioptions()->compaction_style == kCompactionStyleFIFO)
          ? level
          : level + 1;
-  return RunManualCompaction(cfd, level, output_level, 0, begin, end,
+  return RunManualCompaction(cfd, level, output_level, 0, begin, end, true,
                             disallow_trivial_move);
 }

-Status DBImpl::TEST_FlushMemTable(bool wait) {
+Status DBImpl::TEST_FlushMemTable(bool wait, ColumnFamilyHandle* cfh) {
  FlushOptions fo;
  fo.wait = wait;
-  return FlushMemTable(default_cf_handle_->cfd(), fo);
+  ColumnFamilyData* cfd;
+  if (cfh == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfhi = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh);
+    cfd = cfhi->cfd();
+  }
+  return FlushMemTable(cfd, fo);
 }

 Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
@@ -154,5 +161,22 @@ Status DBImpl::TEST_GetAllImmutableCFOptions(
  return Status::OK();
 }

+uint64_t DBImpl::TEST_FindMinLogContainingOutstandingPrep() {
+  return FindMinLogContainingOutstandingPrep();
+}
+
+uint64_t DBImpl::TEST_FindMinPrepLogReferencedByMemTable() {
+  return FindMinPrepLogReferencedByMemTable();
+}
+
+Status DBImpl::TEST_GetLatestMutableCFOptions(
+    ColumnFamilyHandle* column_family, MutableCFOptions* mutable_cf_options) {
+  InstrumentedMutexLock l(&mutex_);
+
+  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  *mutable_cf_options = *cfh->cfd()->GetLatestMutableCFOptions();
+  return Status::OK();
+}
+
 }  // namespace rocksdb
 #endif  // NDEBUG
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -49,8 +49,7 @@ Status DBImpl::SuggestCompactRange(ColumnFamilyHandle* column_family,
    }
    // Since we have some more files to compact, we should also recompute
    // compaction score
-    vstorage->ComputeCompactionScore(*cfd->GetLatestMutableCFOptions(),
-                                     CompactionOptionsFIFO());
+    vstorage->ComputeCompactionScore(*cfd->GetLatestMutableCFOptions());
    SchedulePendingCompaction(cfd);
    MaybeScheduleFlushOrCompaction();
  }
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -54,10 +54,11 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
  auto db_iter = NewArenaWrappedDbIterator(
      env_, *cfd->ioptions(), cfd->user_comparator(),
      (read_options.snapshot != nullptr
-           ? reinterpret_cast<const SnapshotImpl*>(
-                read_options.snapshot)->number_
+           ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                 ->number_
           : latest_snapshot),
-      super_version->mutable_cf_options.max_sequential_skip_in_iterations);
+      super_version->mutable_cf_options.max_sequential_skip_in_iterations,
+      super_version->version_number);
  auto internal_iter = NewInternalIterator(
      read_options, cfd, super_version, db_iter->GetArena());
  db_iter->SetIterUnderDBIter(internal_iter);
@@ -81,10 +82,11 @@ Status DBImplReadOnly::NewIterators(
    auto* db_iter = NewArenaWrappedDbIterator(
        env_, *cfd->ioptions(), cfd->user_comparator(),
        (read_options.snapshot != nullptr
-            ? reinterpret_cast<const SnapshotImpl*>(
-                  read_options.snapshot)->number_
-            : latest_snapshot),
-        sv->mutable_cf_options.max_sequential_skip_in_iterations);
+             ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                   ->number_
+             : latest_snapshot),
+        sv->mutable_cf_options.max_sequential_skip_in_iterations,
+        sv->version_number);
    auto* internal_iter = NewInternalIterator(
        read_options, cfd, sv, db_iter->GetArena());
    db_iter->SetIterUnderDBIter(internal_iter);
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -7,6 +7,8 @@
 #define __STDC_FORMAT_MACROS
 #endif

+#include "db/db_info_dumper.h"
+
 #include <inttypes.h>
 #include <stdio.h>
 #include <string>
@@ -16,7 +18,6 @@
 #include "db/filename.h"
 #include "rocksdb/options.h"
 #include "rocksdb/env.h"
-#include "util/db_info_dumper.h"

 namespace rocksdb {

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -18,12 +18,11 @@ class DBTestInPlaceUpdate : public DBTestBase {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdate) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;
    options.env = env_;
    options.write_buffer_size = 100000;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller size
@@ -41,12 +40,11 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdate) {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateLargeNewValue) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;
    options.env = env_;
    options.write_buffer_size = 100000;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of larger size
@@ -64,7 +62,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateLargeNewValue) {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;

@@ -72,7 +70,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerSize;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller size
@@ -92,7 +89,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;

@@ -100,7 +97,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerVarintSize;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller varint size
@@ -120,7 +116,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;

@@ -128,7 +124,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceLargerSize;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of larger size
@@ -146,7 +141,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {

 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackNoAction) {
  do {
-    Options options;
+    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.inplace_update_support = true;

@@ -154,7 +149,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackNoAction) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceNoAction;
-    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Callback function requests no actions from db
@@ -0,0 +1,259 @@
+//  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.
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+
+namespace rocksdb {
+
+class DBIOFailureTest : public DBTestBase {
+ public:
+  DBIOFailureTest() : DBTestBase("/db_io_failure_test") {}
+};
+
+#ifndef ROCKSDB_LITE
+// Check that number of files does not grow when writes are dropped
+TEST_F(DBIOFailureTest, DropWrites) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.paranoid_checks = false;
+    Reopen(options);
+
+    ASSERT_OK(Put("foo", "v1"));
+    ASSERT_EQ("v1", Get("foo"));
+    Compact("a", "z");
+    const size_t num_files = CountFiles();
+    // Force out-of-space errors
+    env_->drop_writes_.store(true, std::memory_order_release);
+    env_->sleep_counter_.Reset();
+    env_->no_sleep_ = true;
+    for (int i = 0; i < 5; i++) {
+      if (option_config_ != kUniversalCompactionMultiLevel &&
+          option_config_ != kUniversalSubcompactions) {
+        for (int level = 0; level < dbfull()->NumberLevels(); level++) {
+          if (level > 0 && level == dbfull()->NumberLevels() - 1) {
+            break;
+          }
+          dbfull()->TEST_CompactRange(level, nullptr, nullptr, nullptr,
+                                      true /* disallow trivial move */);
+        }
+      } else {
+        dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+      }
+    }
+
+    std::string property_value;
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("5", property_value);
+
+    env_->drop_writes_.store(false, std::memory_order_release);
+    ASSERT_LT(CountFiles(), num_files + 3);
+
+    // Check that compaction attempts slept after errors
+    // TODO @krad: Figure out why ASSERT_EQ 5 keeps failing in certain compiler
+    // versions
+    ASSERT_GE(env_->sleep_counter_.Read(), 4);
+  } while (ChangeCompactOptions());
+}
+
+// Check background error counter bumped on flush failures.
+TEST_F(DBIOFailureTest, DropWritesFlush) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.max_background_flushes = 1;
+    Reopen(options);
+
+    ASSERT_OK(Put("foo", "v1"));
+    // Force out-of-space errors
+    env_->drop_writes_.store(true, std::memory_order_release);
+
+    std::string property_value;
+    // Background error count is 0 now.
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("0", property_value);
+
+    dbfull()->TEST_FlushMemTable(true);
+
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("1", property_value);
+
+    env_->drop_writes_.store(false, std::memory_order_release);
+  } while (ChangeCompactOptions());
+}
+#endif  // ROCKSDB_LITE
+
+// Check that CompactRange() returns failure if there is not enough space left
+// on device
+TEST_F(DBIOFailureTest, NoSpaceCompactRange) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.disable_auto_compactions = true;
+    Reopen(options);
+
+    // generate 5 tables
+    for (int i = 0; i < 5; ++i) {
+      ASSERT_OK(Put(Key(i), Key(i) + "v"));
+      ASSERT_OK(Flush());
+    }
+
+    // Force out-of-space errors
+    env_->no_space_.store(true, std::memory_order_release);
+
+    Status s = dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                                           true /* disallow trivial move */);
+    ASSERT_TRUE(s.IsIOError());
+
+    env_->no_space_.store(false, std::memory_order_release);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBIOFailureTest, NonWritableFileSystem) {
+  do {
+    Options options = CurrentOptions();
+    options.write_buffer_size = 4096;
+    options.arena_block_size = 4096;
+    options.env = env_;
+    Reopen(options);
+    ASSERT_OK(Put("foo", "v1"));
+    env_->non_writeable_rate_.store(100);
+    std::string big(100000, 'x');
+    int errors = 0;
+    for (int i = 0; i < 20; i++) {
+      if (!Put("foo", big).ok()) {
+        errors++;
+        env_->SleepForMicroseconds(100000);
+      }
+    }
+    ASSERT_GT(errors, 0);
+    env_->non_writeable_rate_.store(0);
+  } while (ChangeCompactOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBIOFailureTest, ManifestWriteError) {
+  // Test for the following problem:
+  // (a) Compaction produces file F
+  // (b) Log record containing F is written to MANIFEST file, but Sync() fails
+  // (c) GC deletes F
+  // (d) After reopening DB, reads fail since deleted F is named in log record
+
+  // We iterate twice.  In the second iteration, everything is the
+  // same except the log record never makes it to the MANIFEST file.
+  for (int iter = 0; iter < 2; iter++) {
+    std::atomic<bool>* error_type = (iter == 0) ? &env_->manifest_sync_error_
+                                                : &env_->manifest_write_error_;
+
+    // Insert foo=>bar mapping
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.create_if_missing = true;
+    options.error_if_exists = false;
+    options.paranoid_checks = true;
+    DestroyAndReopen(options);
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Memtable compaction (will succeed)
+    Flush();
+    ASSERT_EQ("bar", Get("foo"));
+    const int last = 2;
+    MoveFilesToLevel(2);
+    ASSERT_EQ(NumTableFilesAtLevel(last), 1);  // foo=>bar is now in last level
+
+    // Merging compaction (will fail)
+    error_type->store(true, std::memory_order_release);
+    dbfull()->TEST_CompactRange(last, nullptr, nullptr);  // Should fail
+    ASSERT_EQ("bar", Get("foo"));
+
+    error_type->store(false, std::memory_order_release);
+
+    // Since paranoid_checks=true, writes should fail
+    ASSERT_NOK(Put("foo2", "bar2"));
+
+    // Recovery: should not lose data
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Try again with paranoid_checks=false
+    Close();
+    options.paranoid_checks = false;
+    Reopen(options);
+
+    // Merging compaction (will fail)
+    error_type->store(true, std::memory_order_release);
+    dbfull()->TEST_CompactRange(last, nullptr, nullptr);  // Should fail
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Recovery: should not lose data
+    error_type->store(false, std::memory_order_release);
+    Reopen(options);
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Since paranoid_checks=false, writes should succeed
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_EQ("bar", Get("foo"));
+    ASSERT_EQ("bar2", Get("foo2"));
+  }
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBIOFailureTest, PutFailsParanoid) {
+  // Test the following:
+  // (a) A random put fails in paranoid mode (simulate by sync fail)
+  // (b) All other puts have to fail, even if writes would succeed
+  // (c) All of that should happen ONLY if paranoid_checks = true
+
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  options.error_if_exists = false;
+  options.paranoid_checks = true;
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+  Status s;
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(1, "foo1", "bar1"));
+  // simulate error
+  env_->log_write_error_.store(true, std::memory_order_release);
+  s = Put(1, "foo2", "bar2");
+  ASSERT_TRUE(!s.ok());
+  env_->log_write_error_.store(false, std::memory_order_release);
+  s = Put(1, "foo3", "bar3");
+  // the next put should fail, too
+  ASSERT_TRUE(!s.ok());
+  // but we're still able to read
+  ASSERT_EQ("bar", Get(1, "foo"));
+
+  // do the same thing with paranoid checks off
+  options.paranoid_checks = false;
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(1, "foo1", "bar1"));
+  // simulate error
+  env_->log_write_error_.store(true, std::memory_order_release);
+  s = Put(1, "foo2", "bar2");
+  ASSERT_TRUE(!s.ok());
+  env_->log_write_error_.store(false, std::memory_order_release);
+  s = Put(1, "foo3", "bar3");
+  // the next put should NOT fail
+  ASSERT_TRUE(s.ok());
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -13,18 +13,21 @@
 #include <string>
 #include <limits>

-#include "db/filename.h"
 #include "db/dbformat.h"
+#include "db/filename.h"
+#include "db/merge_context.h"
+#include "db/pinned_iterators_manager.h"
 #include "port/port.h"
 #include "rocksdb/env.h"
-#include "rocksdb/options.h"
 #include "rocksdb/iterator.h"
 #include "rocksdb/merge_operator.h"
+#include "rocksdb/options.h"
 #include "table/internal_iterator.h"
 #include "util/arena.h"
 #include "util/logging.h"
 #include "util/mutexlock.h"
 #include "util/perf_context_imp.h"
+#include "util/string_util.h"

 namespace rocksdb {

@@ -59,11 +62,49 @@ class DBIter: public Iterator {
    kReverse
  };

+  // LocalStatistics contain Statistics counters that will be aggregated per
+  // each iterator instance and then will be sent to the global statistics when
+  // the iterator is destroyed.
+  //
+  // The purpose of this approach is to avoid perf regression happening
+  // when multiple threads bump the atomic counters from a DBIter::Next().
+  struct LocalStatistics {
+    explicit LocalStatistics() { ResetCounters(); }
+
+    void ResetCounters() {
+      next_count_ = 0;
+      next_found_count_ = 0;
+      prev_count_ = 0;
+      prev_found_count_ = 0;
+      bytes_read_ = 0;
+    }
+
+    void BumpGlobalStatistics(Statistics* global_statistics) {
+      RecordTick(global_statistics, NUMBER_DB_NEXT, next_count_);
+      RecordTick(global_statistics, NUMBER_DB_NEXT_FOUND, next_found_count_);
+      RecordTick(global_statistics, NUMBER_DB_PREV, prev_count_);
+      RecordTick(global_statistics, NUMBER_DB_PREV_FOUND, prev_found_count_);
+      RecordTick(global_statistics, ITER_BYTES_READ, bytes_read_);
+      ResetCounters();
+    }
+
+    // Map to Tickers::NUMBER_DB_NEXT
+    uint64_t next_count_;
+    // Map to Tickers::NUMBER_DB_NEXT_FOUND
+    uint64_t next_found_count_;
+    // Map to Tickers::NUMBER_DB_PREV
+    uint64_t prev_count_;
+    // Map to Tickers::NUMBER_DB_PREV_FOUND
+    uint64_t prev_found_count_;
+    // Map to Tickers::ITER_BYTES_READ
+    uint64_t bytes_read_;
+  };
+
  DBIter(Env* env, const ImmutableCFOptions& ioptions, const Comparator* cmp,
         InternalIterator* iter, SequenceNumber s, bool arena_mode,
-         uint64_t max_sequential_skip_in_iterations,
+         uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
         const Slice* iterate_upper_bound = nullptr,
-         bool prefix_same_as_start = false)
+         bool prefix_same_as_start = false, bool pin_data = false)
      : arena_mode_(arena_mode),
        env_(env),
        logger_(ioptions.info_log),
@@ -75,14 +116,25 @@ class DBIter: public Iterator {
        valid_(false),
        current_entry_is_merged_(false),
        statistics_(ioptions.statistics),
+        version_number_(version_number),
        iterate_upper_bound_(iterate_upper_bound),
-        prefix_same_as_start_(prefix_same_as_start) {
+        prefix_same_as_start_(prefix_same_as_start),
+        pin_thru_lifetime_(pin_data) {
    RecordTick(statistics_, NO_ITERATORS);
    prefix_extractor_ = ioptions.prefix_extractor;
    max_skip_ = max_sequential_skip_in_iterations;
+    if (pin_thru_lifetime_) {
+      pinned_iters_mgr_.StartPinning();
+    }
+    if (iter_) {
+      iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
+    }
  }
  virtual ~DBIter() {
+    // Release pinned data if any
+    pinned_iters_mgr_.ReleasePinnedIterators();
    RecordTick(statistics_, NO_ITERATORS, -1);
+    local_stats_.BumpGlobalStatistics(statistics_);
    if (!arena_mode_) {
      delete iter_;
    } else {
@@ -92,6 +144,7 @@ class DBIter: public Iterator {
  virtual void SetIter(InternalIterator* iter) {
    assert(iter_ == nullptr);
    iter_ = iter;
+    iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
  }
  virtual bool Valid() const override { return valid_; }
  virtual Slice key() const override {
@@ -100,8 +153,13 @@ class DBIter: public Iterator {
  }
  virtual Slice value() const override {
    assert(valid_);
-    return (direction_ == kForward && !current_entry_is_merged_) ?
-      iter_->value() : saved_value_;
+    if (current_entry_is_merged_) {
+      return saved_value_;
+    } else if (direction_ == kReverse) {
+      return pinned_value_;
+    } else {
+      return iter_->value();
+    }
  }
  virtual Status status() const override {
    if (status_.ok()) {
@@ -111,6 +169,28 @@ class DBIter: public Iterator {
    }
  }

+  virtual Status GetProperty(std::string prop_name,
+                             std::string* prop) override {
+    if (prop == nullptr) {
+      return Status::InvalidArgument("prop is nullptr");
+    }
+    if (prop_name == "rocksdb.iterator.super-version-number") {
+      // First try to pass the value returned from inner iterator.
+      if (!iter_->GetProperty(prop_name, prop).ok()) {
+        *prop = ToString(version_number_);
+      }
+      return Status::OK();
+    } else if (prop_name == "rocksdb.iterator.is-key-pinned") {
+      if (valid_) {
+        *prop = (pin_thru_lifetime_ && saved_key_.IsKeyPinned()) ? "1" : "0";
+      } else {
+        *prop = "Iterator is not valid.";
+      }
+      return Status::OK();
+    }
+    return Status::InvalidArgument("Undentified property.");
+  }
+
  virtual void Next() override;
  virtual void Prev() override;
  virtual void Seek(const Slice& target) override;
@@ -125,11 +205,26 @@ class DBIter: public Iterator {
  bool FindValueForCurrentKeyUsingSeek();
  void FindPrevUserKey();
  void FindNextUserKey();
-  inline void FindNextUserEntry(bool skipping);
-  void FindNextUserEntryInternal(bool skipping);
+  inline void FindNextUserEntry(bool skipping, bool prefix_check);
+  void FindNextUserEntryInternal(bool skipping, bool prefix_check);
  bool ParseKey(ParsedInternalKey* key);
  void MergeValuesNewToOld();

+  // Temporarily pin the blocks that we encounter until ReleaseTempPinnedData()
+  // is called
+  void TempPinData() {
+    if (!pin_thru_lifetime_) {
+      pinned_iters_mgr_.StartPinning();
+    }
+  }
+
+  // Release blocks pinned by TempPinData()
+  void ReleaseTempPinnedData() {
+    if (!pin_thru_lifetime_) {
+      pinned_iters_mgr_.ReleasePinnedIterators();
+    }
+  }
+
  inline void ClearSavedValue() {
    if (saved_value_.capacity() > 1048576) {
      std::string empty;
@@ -151,14 +246,24 @@ class DBIter: public Iterator {
  Status status_;
  IterKey saved_key_;
  std::string saved_value_;
+  Slice pinned_value_;
  Direction direction_;
  bool valid_;
  bool current_entry_is_merged_;
  Statistics* statistics_;
  uint64_t max_skip_;
+  uint64_t version_number_;
  const Slice* iterate_upper_bound_;
-  IterKey prefix_start_;
-  bool prefix_same_as_start_;
+  IterKey prefix_start_buf_;
+  Slice prefix_start_key_;
+  const bool prefix_same_as_start_;
+  // Means that we will pin all data blocks we read as long the Iterator
+  // is not deleted, will be true if ReadOptions::pin_data is true
+  const bool pin_thru_lifetime_;
+  // List of operands for merge operator.
+  MergeContext merge_context_;
+  LocalStatistics local_stats_;
+  PinnedIteratorsManager pinned_iters_mgr_;

  // No copying allowed
  DBIter(const DBIter&);
@@ -181,6 +286,8 @@ void DBIter::Next() {
  assert(valid_);

  if (direction_ == kReverse) {
+    // We only pin blocks when doing kReverse
+    ReleaseTempPinnedData();
    FindNextUserKey();
    direction_ = kForward;
    if (!iter_->Valid()) {
@@ -196,24 +303,19 @@ void DBIter::Next() {
    PERF_COUNTER_ADD(internal_key_skipped_count, 1);
  }

+  if (statistics_ != nullptr) {
+    local_stats_.next_count_++;
+  }
  // Now we point to the next internal position, for both of merge and
  // not merge cases.
  if (!iter_->Valid()) {
    valid_ = false;
    return;
  }
-  FindNextUserEntry(true /* skipping the current user key */);
-  if (statistics_ != nullptr) {
-    RecordTick(statistics_, NUMBER_DB_NEXT);
-    if (valid_) {
-      RecordTick(statistics_, NUMBER_DB_NEXT_FOUND);
-      RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
-    }
-  }
-  if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
-      prefix_extractor_->Transform(saved_key_.GetKey())
-              .compare(prefix_start_.GetKey()) != 0) {
-    valid_ = false;
+  FindNextUserEntry(true /* skipping the current user key */, prefix_same_as_start_);
+  if (statistics_ != nullptr && valid_) {
+    local_stats_.next_found_count_++;
+    local_stats_.bytes_read_ += (key().size() + value().size());
  }
 }

@@ -225,13 +327,18 @@ void DBIter::Next() {
 //
 // NOTE: In between, saved_key_ can point to a user key that has
 //       a delete marker
-inline void DBIter::FindNextUserEntry(bool skipping) {
+//
+// The prefix_check parameter controls whether we check the iterated
+// keys against the prefix of the seeked key. Set to false when
+// performing a seek without a key (e.g. SeekToFirst). Set to
+// prefix_same_as_start_ for other iterations.
+inline void DBIter::FindNextUserEntry(bool skipping, bool prefix_check) {
  PERF_TIMER_GUARD(find_next_user_entry_time);
-  FindNextUserEntryInternal(skipping);
+  FindNextUserEntryInternal(skipping, prefix_check);
 }

 // Actual implementation of DBIter::FindNextUserEntry()
-void DBIter::FindNextUserEntryInternal(bool skipping) {
+void DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
  // Loop until we hit an acceptable entry to yield
  assert(iter_->Valid());
  assert(direction_ == kForward);
@@ -242,7 +349,12 @@ void DBIter::FindNextUserEntryInternal(bool skipping) {

    if (ParseKey(&ikey)) {
      if (iterate_upper_bound_ != nullptr &&
-          ikey.user_key.compare(*iterate_upper_bound_) >= 0) {
+          user_comparator_->Compare(ikey.user_key, *iterate_upper_bound_) >= 0) {
+        break;
+      }
+
+      if (prefix_extractor_ && prefix_check &&
+          prefix_extractor_->Transform(ikey.user_key).compare(prefix_start_key_) != 0) {
        break;
      }

@@ -257,18 +369,24 @@ void DBIter::FindNextUserEntryInternal(bool skipping) {
            case kTypeSingleDeletion:
              // Arrange to skip all upcoming entries for this key since
              // they are hidden by this deletion.
-              saved_key_.SetKey(ikey.user_key);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
              skipping = true;
              num_skipped = 0;
              PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
              break;
            case kTypeValue:
              valid_ = true;
-              saved_key_.SetKey(ikey.user_key);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
              return;
            case kTypeMerge:
              // By now, we are sure the current ikey is going to yield a value
-              saved_key_.SetKey(ikey.user_key);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
              current_entry_is_merged_ = true;
              valid_ = true;
              MergeValuesNewToOld();  // Go to a different state machine
@@ -314,9 +432,9 @@ void DBIter::MergeValuesNewToOld() {
    return;
  }

+  merge_context_.Clear();
  // Start the merge process by pushing the first operand
-  std::deque<std::string> operands;
-  operands.push_front(iter_->value().ToString());
+  merge_context_.PushOperand(iter_->value());

  ParsedInternalKey ikey;
  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
@@ -341,7 +459,8 @@ void DBIter::MergeValuesNewToOld() {
      {
        StopWatchNano timer(env_, statistics_ != nullptr);
        PERF_TIMER_GUARD(merge_operator_time_nanos);
-        user_merge_operator_->FullMerge(ikey.user_key, &val, operands,
+        user_merge_operator_->FullMerge(ikey.user_key, &val,
+                                        merge_context_.GetOperands(),
                                        &saved_value_, logger_);
        RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                   timer.ElapsedNanos());
@@ -353,7 +472,7 @@ void DBIter::MergeValuesNewToOld() {
      // hit a merge, add the value as an operand and run associative merge.
      // when complete, add result to operands and continue.
      const Slice& val = iter_->value();
-      operands.push_front(val.ToString());
+      merge_context_.PushOperand(val);
    } else {
      assert(false);
    }
@@ -366,8 +485,9 @@ void DBIter::MergeValuesNewToOld() {
    // a deletion marker.
    // feed null as the existing value to the merge operator, such that
    // client can differentiate this scenario and do things accordingly.
-    user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
-                                    &saved_value_, logger_);
+    user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
+                                    merge_context_.GetOperands(), &saved_value_,
+                                    logger_);
    RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
  }
 }
@@ -377,17 +497,18 @@ void DBIter::Prev() {
  if (direction_ == kForward) {
    ReverseToBackward();
  }
+  ReleaseTempPinnedData();
  PrevInternal();
  if (statistics_ != nullptr) {
-    RecordTick(statistics_, NUMBER_DB_PREV);
+    local_stats_.prev_count_++;
    if (valid_) {
-      RecordTick(statistics_, NUMBER_DB_PREV_FOUND);
-      RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
+      local_stats_.prev_found_count_++;
+      local_stats_.bytes_read_ += (key().size() + value().size());
    }
  }
  if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
      prefix_extractor_->Transform(saved_key_.GetKey())
-              .compare(prefix_start_.GetKey()) != 0) {
+              .compare(prefix_start_key_) != 0) {
    valid_ = false;
  }
 }
@@ -428,7 +549,8 @@ void DBIter::PrevInternal() {
  ParsedInternalKey ikey;

  while (iter_->Valid()) {
-    saved_key_.SetKey(ExtractUserKey(iter_->key()));
+    saved_key_.SetKey(ExtractUserKey(iter_->key()),
+                      !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
    if (FindValueForCurrentKey()) {
      valid_ = true;
      if (!iter_->Valid()) {
@@ -458,8 +580,8 @@ void DBIter::PrevInternal() {
 // saved_value_
 bool DBIter::FindValueForCurrentKey() {
  assert(iter_->Valid());
-  // Contains operands for merge operator.
-  std::deque<std::string> operands;
+  merge_context_.Clear();
+  current_entry_is_merged_ = false;
  // last entry before merge (could be kTypeDeletion, kTypeSingleDeletion or
  // kTypeValue)
  ValueType last_not_merge_type = kTypeDeletion;
@@ -479,19 +601,21 @@ bool DBIter::FindValueForCurrentKey() {
    last_key_entry_type = ikey.type;
    switch (last_key_entry_type) {
      case kTypeValue:
-        operands.clear();
-        saved_value_ = iter_->value().ToString();
+        merge_context_.Clear();
+        ReleaseTempPinnedData();
+        TempPinData();
+        pinned_value_ = iter_->value();
        last_not_merge_type = kTypeValue;
        break;
      case kTypeDeletion:
      case kTypeSingleDeletion:
-        operands.clear();
+        merge_context_.Clear();
        last_not_merge_type = last_key_entry_type;
        PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
        break;
      case kTypeMerge:
        assert(user_merge_operator_ != nullptr);
-        operands.push_back(iter_->value().ToString());
+        merge_context_.PushOperandBack(iter_->value());
        break;
      default:
        assert(false);
@@ -510,22 +634,23 @@ bool DBIter::FindValueForCurrentKey() {
      valid_ = false;
      return false;
    case kTypeMerge:
+      current_entry_is_merged_ = true;
      if (last_not_merge_type == kTypeDeletion) {
        StopWatchNano timer(env_, statistics_ != nullptr);
        PERF_TIMER_GUARD(merge_operator_time_nanos);
-        user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
+        user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
+                                        merge_context_.GetOperands(),
                                        &saved_value_, logger_);
        RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                   timer.ElapsedNanos());
      } else {
        assert(last_not_merge_type == kTypeValue);
-        std::string last_put_value = saved_value_;
-        Slice temp_slice(last_put_value);
        {
          StopWatchNano timer(env_, statistics_ != nullptr);
          PERF_TIMER_GUARD(merge_operator_time_nanos);
-          user_merge_operator_->FullMerge(saved_key_.GetKey(), &temp_slice,
-                                          operands, &saved_value_, logger_);
+          user_merge_operator_->FullMerge(saved_key_.GetKey(), &pinned_value_,
+                                          merge_context_.GetOperands(),
+                                          &saved_value_, logger_);
          RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                     timer.ElapsedNanos());
        }
@@ -558,7 +683,9 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  if (ikey.type == kTypeValue || ikey.type == kTypeDeletion ||
      ikey.type == kTypeSingleDeletion) {
    if (ikey.type == kTypeValue) {
-      saved_value_ = iter_->value().ToString();
+      ReleaseTempPinnedData();
+      TempPinData();
+      pinned_value_ = iter_->value();
      valid_ = true;
      return true;
    }
@@ -568,11 +695,12 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {

  // kTypeMerge. We need to collect all kTypeMerge values and save them
  // in operands
-  std::deque<std::string> operands;
+  current_entry_is_merged_ = true;
+  merge_context_.Clear();
  while (iter_->Valid() &&
         user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) &&
         ikey.type == kTypeMerge) {
-    operands.push_front(iter_->value().ToString());
+    merge_context_.PushOperand(iter_->value());
    iter_->Next();
    FindParseableKey(&ikey, kForward);
  }
@@ -583,7 +711,8 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
    {
      StopWatchNano timer(env_, statistics_ != nullptr);
      PERF_TIMER_GUARD(merge_operator_time_nanos);
-      user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
+      user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
+                                      merge_context_.GetOperands(),
                                      &saved_value_, logger_);
      RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
    }
@@ -601,8 +730,9 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  {
    StopWatchNano timer(env_, statistics_ != nullptr);
    PERF_TIMER_GUARD(merge_operator_time_nanos);
-    user_merge_operator_->FullMerge(saved_key_.GetKey(), &val, operands,
-                                    &saved_value_, logger_);
+    user_merge_operator_->FullMerge(saved_key_.GetKey(), &val,
+                                    merge_context_.GetOperands(), &saved_value_,
+                                    logger_);
    RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
  }
  valid_ = true;
@@ -668,6 +798,7 @@ void DBIter::FindParseableKey(ParsedInternalKey* ikey, Direction direction) {

 void DBIter::Seek(const Slice& target) {
  StopWatch sw(env_, statistics_, DB_SEEK);
+  ReleaseTempPinnedData();
  saved_key_.Clear();
  // now savved_key is used to store internal key.
  saved_key_.SetInternalKey(target, sequence_);
@@ -679,9 +810,15 @@ void DBIter::Seek(const Slice& target) {

  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
+    if (prefix_extractor_ && prefix_same_as_start_) {
+      prefix_start_key_ = prefix_extractor_->Transform(target);
+    }
    direction_ = kForward;
    ClearSavedValue();
-    FindNextUserEntry(false /* not skipping */);
+    FindNextUserEntry(false /* not skipping */, prefix_same_as_start_);
+    if (!valid_) {
+      prefix_start_key_.clear();
+    }
    if (statistics_ != nullptr) {
      if (valid_) {
        RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
@@ -692,7 +829,8 @@ void DBIter::Seek(const Slice& target) {
    valid_ = false;
  }
  if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(target));
+    prefix_start_buf_.SetKey(prefix_start_key_);
+    prefix_start_key_ = prefix_start_buf_.GetKey();
  }
 }

@@ -703,6 +841,7 @@ void DBIter::SeekToFirst() {
    max_skip_ = std::numeric_limits<uint64_t>::max();
  }
  direction_ = kForward;
+  ReleaseTempPinnedData();
  ClearSavedValue();

  {
@@ -712,7 +851,7 @@ void DBIter::SeekToFirst() {

  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
-    FindNextUserEntry(false /* not skipping */);
+    FindNextUserEntry(false /* not skipping */, false /* no prefix check */);
    if (statistics_ != nullptr) {
      if (valid_) {
        RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
@@ -723,7 +862,8 @@ void DBIter::SeekToFirst() {
    valid_ = false;
  }
  if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_buf_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_key_ = prefix_start_buf_.GetKey();
  }
 }

@@ -734,6 +874,7 @@ void DBIter::SeekToLast() {
    max_skip_ = std::numeric_limits<uint64_t>::max();
  }
  direction_ = kReverse;
+  ReleaseTempPinnedData();
  ClearSavedValue();

  {
@@ -744,7 +885,7 @@ void DBIter::SeekToLast() {
  // it will seek to the last key before the
  // ReadOptions.iterate_upper_bound
  if (iter_->Valid() && iterate_upper_bound_ != nullptr) {
-    saved_key_.SetKey(*iterate_upper_bound_);
+    saved_key_.SetKey(*iterate_upper_bound_, false /* copy */);
    std::string last_key;
    AppendInternalKey(&last_key,
                      ParsedInternalKey(saved_key_.GetKey(), kMaxSequenceNumber,
@@ -771,7 +912,8 @@ void DBIter::SeekToLast() {
    }
  }
  if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_buf_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_key_ = prefix_start_buf_.GetKey();
  }
 }

@@ -780,11 +922,14 @@ Iterator* NewDBIterator(Env* env, const ImmutableCFOptions& ioptions,
                        InternalIterator* internal_iter,
                        const SequenceNumber& sequence,
                        uint64_t max_sequential_skip_in_iterations,
+                        uint64_t version_number,
                        const Slice* iterate_upper_bound,
-                        bool prefix_same_as_start) {
-  return new DBIter(env, ioptions, user_key_comparator, internal_iter, sequence,
-                    false, max_sequential_skip_in_iterations,
-                    iterate_upper_bound, prefix_same_as_start);
+                        bool prefix_same_as_start, bool pin_data) {
+  DBIter* db_iter =
+      new DBIter(env, ioptions, user_key_comparator, internal_iter, sequence,
+                 false, max_sequential_skip_in_iterations, version_number,
+                 iterate_upper_bound, prefix_same_as_start, pin_data);
+  return db_iter;
 }

 ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
@@ -806,6 +951,10 @@ inline void ArenaWrappedDBIter::Prev() { db_iter_->Prev(); }
 inline Slice ArenaWrappedDBIter::key() const { return db_iter_->key(); }
 inline Slice ArenaWrappedDBIter::value() const { return db_iter_->value(); }
 inline Status ArenaWrappedDBIter::status() const { return db_iter_->status(); }
+inline Status ArenaWrappedDBIter::GetProperty(std::string prop_name,
+                                              std::string* prop) {
+  return db_iter_->GetProperty(prop_name, prop);
+}
 void ArenaWrappedDBIter::RegisterCleanup(CleanupFunction function, void* arg1,
                                         void* arg2) {
  db_iter_->RegisterCleanup(function, arg1, arg2);
@@ -814,15 +963,16 @@ void ArenaWrappedDBIter::RegisterCleanup(CleanupFunction function, void* arg1,
 ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    Env* env, const ImmutableCFOptions& ioptions,
    const Comparator* user_key_comparator, const SequenceNumber& sequence,
-    uint64_t max_sequential_skip_in_iterations,
-    const Slice* iterate_upper_bound, bool prefix_same_as_start) {
+    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
+    const Slice* iterate_upper_bound, bool prefix_same_as_start,
+    bool pin_data) {
  ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
  Arena* arena = iter->GetArena();
  auto mem = arena->AllocateAligned(sizeof(DBIter));
  DBIter* db_iter =
      new (mem) DBIter(env, ioptions, user_key_comparator, nullptr, sequence,
-                       true, max_sequential_skip_in_iterations,
-                       iterate_upper_bound, prefix_same_as_start);
+                       true, max_sequential_skip_in_iterations, version_number,
+                       iterate_upper_bound, prefix_same_as_start, pin_data);

  iter->SetDBIter(db_iter);

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -9,6 +9,7 @@

 #pragma once
 #include <stdint.h>
+#include <string>
 #include "rocksdb/db.h"
 #include "rocksdb/iterator.h"
 #include "db/dbformat.h"
@@ -24,13 +25,12 @@ class InternalIterator;
 // Return a new iterator that converts internal keys (yielded by
 // "*internal_iter") that were live at the specified "sequence" number
 // into appropriate user keys.
-extern Iterator* NewDBIterator(Env* env, const ImmutableCFOptions& options,
-                               const Comparator* user_key_comparator,
-                               InternalIterator* internal_iter,
-                               const SequenceNumber& sequence,
-                               uint64_t max_sequential_skip_in_iterations,
-                               const Slice* iterate_upper_bound = nullptr,
-                               bool prefix_same_as_start = false);
+extern Iterator* NewDBIterator(
+    Env* env, const ImmutableCFOptions& options,
+    const Comparator* user_key_comparator, InternalIterator* internal_iter,
+    const SequenceNumber& sequence, uint64_t max_sequential_skip_in_iterations,
+    uint64_t version_number, const Slice* iterate_upper_bound = nullptr,
+    bool prefix_same_as_start = false, bool pin_data = false);

 // A wrapper iterator which wraps DB Iterator and the arena, with which the DB
 // iterator is supposed be allocated. This class is used as an entry point of
@@ -63,6 +63,7 @@ class ArenaWrappedDBIter : public Iterator {
  virtual Status status() const override;

  void RegisterCleanup(CleanupFunction function, void* arg1, void* arg2);
+  virtual Status GetProperty(std::string prop_name, std::string* prop) override;

 private:
  DBIter* db_iter_;
@@ -73,8 +74,8 @@ class ArenaWrappedDBIter : public Iterator {
 extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    Env* env, const ImmutableCFOptions& options,
    const Comparator* user_key_comparator, const SequenceNumber& sequence,
-    uint64_t max_sequential_skip_in_iterations,
+    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
    const Slice* iterate_upper_bound = nullptr,
-    bool prefix_same_as_start = false);
+    bool prefix_same_as_start = false, bool pin_data = false);

 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -181,10 +181,9 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    internal_iter->AddPut("b", "val_b");
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 10,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        10, options.max_sequential_skip_in_iterations, 0));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -215,7 +214,7 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations));
+        10, options.max_sequential_skip_in_iterations, 0));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -241,7 +240,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        10, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -273,7 +273,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        10, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -308,7 +309,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        10, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(!db_iter->Valid());
@@ -337,7 +339,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        7, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        7, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    SetPerfLevel(kEnableCount);
    ASSERT_TRUE(GetPerfLevel() == kEnableCount);
@@ -374,7 +377,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        4, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        4, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -399,7 +403,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        10, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(!db_iter->Valid());
@@ -421,7 +426,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        10, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -456,7 +462,8 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        7, options.max_sequential_skip_in_iterations, ro.iterate_upper_bound));
+        7, options.max_sequential_skip_in_iterations, 0,
+        ro.iterate_upper_bound));

    SetPerfLevel(kEnableCount);
    ASSERT_TRUE(GetPerfLevel() == kEnableCount);
@@ -482,10 +489,9 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    internal_iter->AddPut("b", "val_b");
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 10,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        10, options.max_sequential_skip_in_iterations, 0));

    db_iter->SeekToFirst();
    ASSERT_TRUE(db_iter->Valid());
@@ -524,10 +530,9 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    internal_iter->AddPut("b", "val_b");
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 2,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        2, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "b");
@@ -555,10 +560,9 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    internal_iter->AddPut("c", "val_c");
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 10,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        10, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "c");
@@ -583,10 +587,9 @@ TEST_F(DBIteratorTest, DBIteratorEmpty) {
    TestIterator* internal_iter = new TestIterator(BytewiseComparator());
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 0,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        0, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(!db_iter->Valid());
  }
@@ -595,10 +598,9 @@ TEST_F(DBIteratorTest, DBIteratorEmpty) {
    TestIterator* internal_iter = new TestIterator(BytewiseComparator());
    internal_iter->Finish();

-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 0,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        0, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToFirst();
    ASSERT_TRUE(!db_iter->Valid());
  }
@@ -617,10 +619,9 @@ TEST_F(DBIteratorTest, DBIteratorUseSkipCountSkips) {
  }
  internal_iter->Finish();

-  std::unique_ptr<Iterator> db_iter(
-      NewDBIterator(env_, ImmutableCFOptions(options),
-                    BytewiseComparator(), internal_iter, 2,
-                    options.max_sequential_skip_in_iterations));
+  std::unique_ptr<Iterator> db_iter(NewDBIterator(
+      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 2,
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToLast();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "c");
@@ -659,9 +660,8 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {

      options.statistics = rocksdb::CreateDBStatistics();
      std::unique_ptr<Iterator> db_iter(NewDBIterator(
-          env_, ImmutableCFOptions(options),
-          BytewiseComparator(), internal_iter, i + 2,
-          options.max_sequential_skip_in_iterations));
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, i + 2, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(db_iter->Valid());

@@ -695,9 +695,8 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
      internal_iter->Finish();

      std::unique_ptr<Iterator> db_iter(NewDBIterator(
-          env_, ImmutableCFOptions(options),
-          BytewiseComparator(), internal_iter, i + 2,
-          options.max_sequential_skip_in_iterations));
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, i + 2, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(db_iter->Valid());

@@ -724,9 +723,8 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
      internal_iter->Finish();

      std::unique_ptr<Iterator> db_iter(NewDBIterator(
-          env_, ImmutableCFOptions(options),
-          BytewiseComparator(), internal_iter, 202,
-          options.max_sequential_skip_in_iterations));
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, 202, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(db_iter->Valid());

@@ -756,10 +754,9 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
      }
      internal_iter->AddPut("c", "200");
      internal_iter->Finish();
-      std::unique_ptr<Iterator> db_iter(
-          NewDBIterator(env_, ImmutableCFOptions(options),
-                        BytewiseComparator(), internal_iter, i,
-                        options.max_sequential_skip_in_iterations));
+      std::unique_ptr<Iterator> db_iter(NewDBIterator(
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, i, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(!db_iter->Valid());

@@ -773,10 +770,9 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
    }
    internal_iter->AddPut("c", "200");
    internal_iter->Finish();
-    std::unique_ptr<Iterator> db_iter(
-        NewDBIterator(env_, ImmutableCFOptions(options),
-                      BytewiseComparator(), internal_iter, 200,
-                      options.max_sequential_skip_in_iterations));
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        200, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "c");
@@ -809,9 +805,8 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
      internal_iter->Finish();

      std::unique_ptr<Iterator> db_iter(NewDBIterator(
-          env_, ImmutableCFOptions(options),
-          BytewiseComparator(), internal_iter, i + 2,
-          options.max_sequential_skip_in_iterations));
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, i + 2, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(db_iter->Valid());

@@ -844,9 +839,8 @@ TEST_F(DBIteratorTest, DBIteratorUseSkip) {
      internal_iter->Finish();

      std::unique_ptr<Iterator> db_iter(NewDBIterator(
-          env_, ImmutableCFOptions(options),
-          BytewiseComparator(), internal_iter, i + 2,
-          options.max_sequential_skip_in_iterations));
+          env_, ImmutableCFOptions(options), BytewiseComparator(),
+          internal_iter, i + 2, options.max_sequential_skip_in_iterations, 0));
      db_iter->SeekToLast();
      ASSERT_TRUE(db_iter->Valid());

@@ -887,7 +881,7 @@ TEST_F(DBIteratorTest, DBIterator1) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 1,
-      options.max_sequential_skip_in_iterations));
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -913,7 +907,7 @@ TEST_F(DBIteratorTest, DBIterator2) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 0,
-      options.max_sequential_skip_in_iterations));
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -936,7 +930,7 @@ TEST_F(DBIteratorTest, DBIterator3) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 2,
-      options.max_sequential_skip_in_iterations));
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -958,7 +952,7 @@ TEST_F(DBIteratorTest, DBIterator4) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 4,
-      options.max_sequential_skip_in_iterations));
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -987,7 +981,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        0, options.max_sequential_skip_in_iterations));
+        0, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1009,7 +1003,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        1, options.max_sequential_skip_in_iterations));
+        1, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1031,7 +1025,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        2, options.max_sequential_skip_in_iterations));
+        2, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1053,7 +1047,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        3, options.max_sequential_skip_in_iterations));
+        3, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1075,7 +1069,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        4, options.max_sequential_skip_in_iterations));
+        4, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1097,7 +1091,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        5, options.max_sequential_skip_in_iterations));
+        5, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1119,7 +1113,7 @@ TEST_F(DBIteratorTest, DBIterator5) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        6, options.max_sequential_skip_in_iterations));
+        6, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1145,7 +1139,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        0, options.max_sequential_skip_in_iterations));
+        0, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1167,7 +1161,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        1, options.max_sequential_skip_in_iterations));
+        1, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1189,7 +1183,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        2, options.max_sequential_skip_in_iterations));
+        2, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1211,7 +1205,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        3, options.max_sequential_skip_in_iterations));
+        3, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(!db_iter->Valid());
  }
@@ -1229,7 +1223,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        4, options.max_sequential_skip_in_iterations));
+        4, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1251,7 +1245,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        5, options.max_sequential_skip_in_iterations));
+        5, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1273,7 +1267,7 @@ TEST_F(DBIteratorTest, DBIterator6) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        6, options.max_sequential_skip_in_iterations));
+        6, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1311,7 +1305,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        0, options.max_sequential_skip_in_iterations));
+        0, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1345,7 +1339,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        2, options.max_sequential_skip_in_iterations));
+        2, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1385,7 +1379,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        4, options.max_sequential_skip_in_iterations));
+        4, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1425,7 +1419,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        5, options.max_sequential_skip_in_iterations));
+        5, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1470,7 +1464,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        6, options.max_sequential_skip_in_iterations));
+        6, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1516,7 +1510,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        7, options.max_sequential_skip_in_iterations));
+        7, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1556,7 +1550,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        9, options.max_sequential_skip_in_iterations));
+        9, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1602,7 +1596,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        13, options.max_sequential_skip_in_iterations));
+        13, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1649,7 +1643,7 @@ TEST_F(DBIteratorTest, DBIterator7) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        14, options.max_sequential_skip_in_iterations));
+        14, options.max_sequential_skip_in_iterations, 0));
    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());

@@ -1678,7 +1672,7 @@ TEST_F(DBIteratorTest, DBIterator8) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-      10, options.max_sequential_skip_in_iterations));
+      10, options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToLast();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "b");
@@ -1707,7 +1701,7 @@ TEST_F(DBIteratorTest, DBIterator9) {

    std::unique_ptr<Iterator> db_iter(NewDBIterator(
        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-        10, options.max_sequential_skip_in_iterations));
+        10, options.max_sequential_skip_in_iterations, 0));

    db_iter->SeekToLast();
    ASSERT_TRUE(db_iter->Valid());
@@ -1754,7 +1748,7 @@ TEST_F(DBIteratorTest, DBIterator10) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-      10, options.max_sequential_skip_in_iterations));
+      10, options.max_sequential_skip_in_iterations, 0));

  db_iter->Seek("c");
  ASSERT_TRUE(db_iter->Valid());
@@ -1778,9 +1772,9 @@ TEST_F(DBIteratorTest, SeekToLastOccurrenceSeq0) {
  internal_iter->AddPut("b", "2");
  internal_iter->Finish();

-  std::unique_ptr<Iterator> db_iter(NewDBIterator(
-      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
-      10, 0 /* force seek */));
+  std::unique_ptr<Iterator> db_iter(
+      NewDBIterator(env_, ImmutableCFOptions(options), BytewiseComparator(),
+                    internal_iter, 10, 0 /* force seek */, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1807,7 +1801,7 @@ TEST_F(DBIteratorTest, DBIterator11) {

  std::unique_ptr<Iterator> db_iter(NewDBIterator(
      env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter, 1,
-      options.max_sequential_skip_in_iterations));
+      options.max_sequential_skip_in_iterations, 0));
  db_iter->SeekToFirst();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "a");
@@ -1832,7 +1826,7 @@ TEST_F(DBIteratorTest, DBIterator12) {

  std::unique_ptr<Iterator> db_iter(
      NewDBIterator(env_, ImmutableCFOptions(options), BytewiseComparator(),
-                    internal_iter, 10, 0));
+                    internal_iter, 10, 0, 0));
  db_iter->SeekToLast();
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "c");
@@ -1874,7 +1868,7 @@ class DBIterWithMergeIterTest : public testing::Test {
    db_iter_.reset(NewDBIterator(env_, ImmutableCFOptions(options_),
                                 BytewiseComparator(), merge_iter,
                                 8 /* read data earlier than seqId 8 */,
-                                 3 /* max iterators before reseek */));
+                                 3 /* max iterators before reseek */, 0));
  }

  Env* env_;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -27,7 +27,7 @@ class DBTestXactLogIterator : public DBTestBase {
    Status status = dbfull()->GetUpdatesSince(seq, &iter);
    EXPECT_OK(status);
    EXPECT_TRUE(iter->Valid());
-    return std::move(iter);
+    return iter;
  }
 };

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -215,6 +215,34 @@ TEST_F(DBTablePropertiesTest, GetPropertiesOfTablesInRange) {
    TestGetPropertiesOfTablesInRange(std::move(ranges));
  }
 }
+
+TEST_F(DBTablePropertiesTest, GetColumnFamilyNameProperty) {
+  std::string kExtraCfName = "pikachu";
+  CreateAndReopenWithCF({kExtraCfName}, CurrentOptions());
+
+  // Create one table per CF, then verify it was created with the column family
+  // name property.
+  for (int cf = 0; cf < 2; ++cf) {
+    Put(cf, "key", "val");
+    Flush(cf);
+
+    TablePropertiesCollection fname_to_props;
+    ASSERT_OK(db_->GetPropertiesOfAllTables(handles_[cf], &fname_to_props));
+    ASSERT_EQ(1U, fname_to_props.size());
+
+    std::string expected_cf_name;
+    if (cf > 0) {
+      expected_cf_name = kExtraCfName;
+    } else {
+      expected_cf_name = kDefaultColumnFamilyName;
+    }
+    ASSERT_EQ(expected_cf_name,
+              fname_to_props.begin()->second->column_family_name);
+    ASSERT_EQ(cf, static_cast<uint32_t>(
+                      fname_to_props.begin()->second->column_family_id));
+  }
+}
+
 }  // namespace rocksdb

 #endif  // ROCKSDB_LITE
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -168,7 +168,7 @@ TEST_F(DBTestTailingIterator, TailingIteratorTrimSeekToNext) {
    char buf3[32];
    char buf4[32];
    snprintf(buf1, sizeof(buf1), "00a0%016d", i * 5);
-    snprintf(buf3, sizeof(buf1), "00b0%016d", i * 5);
+    snprintf(buf3, sizeof(buf3), "00b0%016d", i * 5);

    Slice key(buf1, 20);
    ASSERT_OK(Put(1, key, value));
@@ -181,7 +181,7 @@ TEST_F(DBTestTailingIterator, TailingIteratorTrimSeekToNext) {
      if (i == 299) {
        file_iters_deleted = true;
      }
-      snprintf(buf4, sizeof(buf1), "00a0%016d", i * 5 / 2);
+      snprintf(buf4, sizeof(buf4), "00a0%016d", i * 5 / 2);
      Slice target(buf4, 20);
      iterh->Seek(target);
      ASSERT_TRUE(iter->Valid());
@@ -654,6 +654,51 @@ TEST_F(DBTestTailingIterator, ManagedTailingIteratorSeekToSame) {
  ASSERT_EQ(found, iter->key().ToString());
 }

+TEST_F(DBTestTailingIterator, ForwardIteratorVersionProperty) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 1000;
+
+  ReadOptions read_options;
+  read_options.tailing = true;
+
+  Put("foo", "bar");
+
+  uint64_t v1, v2, v3, v4;
+  {
+    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+    iter->Seek("foo");
+    std::string prop_value;
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
+                                &prop_value));
+    v1 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
+
+    Put("foo1", "bar1");
+    Flush();
+
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
+                                &prop_value));
+    v2 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
+
+    iter->Seek("f");
+
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
+                                &prop_value));
+    v3 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
+
+    ASSERT_EQ(v1, v2);
+    ASSERT_GT(v3, v2);
+  }
+
+  {
+    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+    iter->Seek("foo");
+    std::string prop_value;
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
+                                &prop_value));
+    v4 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
+  }
+  ASSERT_EQ(v3, v4);
+}
 }  // namespace rocksdb

 #endif  // !defined(ROCKSDB_LITE)
@@ -1,4 +1,4 @@
-// Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+// 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.
@@ -73,63 +73,69 @@ DBTestBase::~DBTestBase() {
  delete env_;
 }

-// Switch to a fresh database with the next option configuration to
-// test.  Return false if there are no more configurations to test.
-bool DBTestBase::ChangeOptions(int skip_mask) {
-  for (option_config_++; option_config_ < kEnd; option_config_++) {
+bool DBTestBase::ShouldSkipOptions(int option_config, int skip_mask) {
 #ifdef ROCKSDB_LITE
    // These options are not supported in ROCKSDB_LITE
-    if (option_config_ == kHashSkipList ||
-        option_config_ == kPlainTableFirstBytePrefix ||
-        option_config_ == kPlainTableCappedPrefix ||
-        option_config_ == kPlainTableCappedPrefixNonMmap ||
-        option_config_ == kPlainTableAllBytesPrefix ||
-        option_config_ == kVectorRep || option_config_ == kHashLinkList ||
-        option_config_ == kHashCuckoo ||
-        option_config_ == kUniversalCompaction ||
-        option_config_ == kUniversalCompactionMultiLevel ||
-        option_config_ == kUniversalSubcompactions ||
-        option_config_ == kFIFOCompaction) {
-      continue;
+  if (option_config == kHashSkipList ||
+      option_config == kPlainTableFirstBytePrefix ||
+      option_config == kPlainTableCappedPrefix ||
+      option_config == kPlainTableCappedPrefixNonMmap ||
+      option_config == kPlainTableAllBytesPrefix ||
+      option_config == kVectorRep || option_config == kHashLinkList ||
+      option_config == kHashCuckoo || option_config == kUniversalCompaction ||
+      option_config == kUniversalCompactionMultiLevel ||
+      option_config == kUniversalSubcompactions ||
+      option_config == kFIFOCompaction ||
+      option_config == kConcurrentSkipList) {
+    return true;
    }
 #endif

    if ((skip_mask & kSkipDeletesFilterFirst) &&
-        option_config_ == kDeletesFilterFirst) {
-      continue;
+        option_config == kDeletesFilterFirst) {
+      return true;
    }
    if ((skip_mask & kSkipUniversalCompaction) &&
-        (option_config_ == kUniversalCompaction ||
-         option_config_ == kUniversalCompactionMultiLevel)) {
-      continue;
+        (option_config == kUniversalCompaction ||
+         option_config == kUniversalCompactionMultiLevel)) {
+      return true;
    }
-    if ((skip_mask & kSkipMergePut) && option_config_ == kMergePut) {
-      continue;
+    if ((skip_mask & kSkipMergePut) && option_config == kMergePut) {
+      return true;
    }
    if ((skip_mask & kSkipNoSeekToLast) &&
-        (option_config_ == kHashLinkList || option_config_ == kHashSkipList)) {
-      continue;
+        (option_config == kHashLinkList || option_config == kHashSkipList)) {
+      return true;
    }
    if ((skip_mask & kSkipPlainTable) &&
-        (option_config_ == kPlainTableAllBytesPrefix ||
-         option_config_ == kPlainTableFirstBytePrefix ||
-         option_config_ == kPlainTableCappedPrefix ||
-         option_config_ == kPlainTableCappedPrefixNonMmap)) {
-      continue;
+        (option_config == kPlainTableAllBytesPrefix ||
+         option_config == kPlainTableFirstBytePrefix ||
+         option_config == kPlainTableCappedPrefix ||
+         option_config == kPlainTableCappedPrefixNonMmap)) {
+      return true;
    }
    if ((skip_mask & kSkipHashIndex) &&
-        (option_config_ == kBlockBasedTableWithPrefixHashIndex ||
-         option_config_ == kBlockBasedTableWithWholeKeyHashIndex)) {
-      continue;
+        (option_config == kBlockBasedTableWithPrefixHashIndex ||
+         option_config == kBlockBasedTableWithWholeKeyHashIndex)) {
+      return true;
    }
-    if ((skip_mask & kSkipHashCuckoo) && (option_config_ == kHashCuckoo)) {
-      continue;
+    if ((skip_mask & kSkipHashCuckoo) && (option_config == kHashCuckoo)) {
+      return true;
    }
-    if ((skip_mask & kSkipFIFOCompaction) &&
-        option_config_ == kFIFOCompaction) {
-      continue;
+    if ((skip_mask & kSkipFIFOCompaction) && option_config == kFIFOCompaction) {
+      return true;
    }
-    if ((skip_mask & kSkipMmapReads) && option_config_ == kWalDirAndMmapReads) {
+    if ((skip_mask & kSkipMmapReads) && option_config == kWalDirAndMmapReads) {
+      return true;
+    }
+    return false;
+}
+
+// Switch to a fresh database with the next option configuration to
+// test.  Return false if there are no more configurations to test.
+bool DBTestBase::ChangeOptions(int skip_mask) {
+  for (option_config_++; option_config_ < kEnd; option_config_++) {
+    if (ShouldSkipOptions(option_config_, skip_mask)) {
      continue;
    }
    break;
@@ -204,6 +210,13 @@ Options DBTestBase::CurrentOptions(
    const anon::OptionsOverride& options_override) {
  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);
 }

@@ -295,6 +308,7 @@ Options DBTestBase::CurrentOptions(
    case kPerfOptions:
      options.soft_rate_limit = 2.0;
      options.delayed_write_rate = 8 * 1024 * 1024;
+      options.report_bg_io_stats = true;
      // TODO(3.13) -- test more options
      break;
    case kDeletesFilterFirst:
@@ -333,6 +347,10 @@ Options DBTestBase::CurrentOptions(
      options.prefix_extractor.reset(NewNoopTransform());
      break;
    }
+    case kBlockBasedTableWithIndexRestartInterval: {
+      table_options.index_block_restart_interval = 8;
+      break;
+    }
    case kOptimizeFiltersForHits: {
      options.optimize_filters_for_hits = true;
      set_block_based_table_factory = true;
@@ -356,6 +374,11 @@ Options DBTestBase::CurrentOptions(
      options.max_subcompactions = 4;
      break;
    }
+    case kConcurrentSkipList: {
+      options.allow_concurrent_memtable_write = true;
+      options.enable_write_thread_adaptive_yield = true;
+      break;
+    }

    default:
      break;
@@ -661,14 +684,14 @@ uint64_t DBTestBase::SizeAtLevel(int level) {
  return sum;
 }

-int DBTestBase::TotalLiveFiles(int cf) {
+size_t DBTestBase::TotalLiveFiles(int cf) {
  ColumnFamilyMetaData cf_meta;
  if (cf == 0) {
    db_->GetColumnFamilyMetaData(&cf_meta);
  } else {
    db_->GetColumnFamilyMetaData(handles_[cf], &cf_meta);
  }
-  int num_files = 0;
+  size_t num_files = 0;
  for (auto& level : cf_meta.levels) {
    num_files += level.files.size();
  }
@@ -696,6 +719,22 @@ int DBTestBase::NumTableFilesAtLevel(int level, int cf) {
  return atoi(property.c_str());
 }

+double DBTestBase::CompressionRatioAtLevel(int level, int cf) {
+  std::string property;
+  if (cf == 0) {
+    // default cfd
+    EXPECT_TRUE(db_->GetProperty(
+        "rocksdb.compression-ratio-at-level" + NumberToString(level),
+        &property));
+  } else {
+    EXPECT_TRUE(db_->GetProperty(
+        handles_[cf],
+        "rocksdb.compression-ratio-at-level" + NumberToString(level),
+        &property));
+  }
+  return std::stod(property);
+}
+
 int DBTestBase::TotalTableFiles(int cf, int levels) {
  if (levels == -1) {
    levels = CurrentOptions().num_levels;
@@ -1000,4 +1039,59 @@ void DBTestBase::CopyFile(const std::string& source,
  ASSERT_OK(destfile->Close());
 }

+std::unordered_map<std::string, uint64_t> DBTestBase::GetAllSSTFiles(
+    uint64_t* total_size) {
+  std::unordered_map<std::string, uint64_t> res;
+
+  if (total_size) {
+    *total_size = 0;
+  }
+  std::vector<std::string> files;
+  env_->GetChildren(dbname_, &files);
+  for (auto& file_name : files) {
+    uint64_t number;
+    FileType type;
+    std::string file_path = dbname_ + "/" + file_name;
+    if (ParseFileName(file_name, &number, &type) && type == kTableFile) {
+      uint64_t file_size = 0;
+      env_->GetFileSize(file_path, &file_size);
+      res[file_path] = file_size;
+      if (total_size) {
+        *total_size += file_size;
+      }
+    }
+  }
+  return res;
+}
+
+std::vector<std::uint64_t> DBTestBase::ListTableFiles(Env* env,
+                                                      const std::string& path) {
+  std::vector<std::string> files;
+  std::vector<uint64_t> file_numbers;
+  env->GetChildren(path, &files);
+  uint64_t number;
+  FileType type;
+  for (size_t i = 0; i < files.size(); ++i) {
+    if (ParseFileName(files[i], &number, &type)) {
+      if (type == kTableFile) {
+        file_numbers.push_back(number);
+      }
+    }
+  }
+  return file_numbers;
+}
+
+#ifndef ROCKSDB_LITE
+uint64_t DBTestBase::GetNumberOfSstFilesForColumnFamily(
+    DB* db, std::string column_family_name) {
+  std::vector<LiveFileMetaData> metadata;
+  db->GetLiveFilesMetaData(&metadata);
+  uint64_t result = 0;
+  for (auto& fileMetadata : metadata) {
+    result += (fileMetadata.column_family_name == column_family_name);
+  }
+  return result;
+}
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-// Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+// 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.
@@ -19,6 +19,7 @@
 #endif

 #include <algorithm>
+#include <map>
 #include <set>
 #include <string>
 #include <thread>
@@ -27,7 +28,6 @@
 #include <vector>

 #include "db/db_impl.h"
-#include "db/db_test_util.h"
 #include "db/dbformat.h"
 #include "db/filename.h"
 #include "memtable/hash_linklist_rep.h"
@@ -39,6 +39,7 @@
 #include "rocksdb/filter_policy.h"
 #include "rocksdb/options.h"
 #include "rocksdb/slice.h"
+#include "rocksdb/statistics.h"
 #include "rocksdb/table.h"
 #include "rocksdb/utilities/checkpoint.h"
 #include "table/block_based_table_factory.h"
@@ -357,23 +358,30 @@ class SpecialEnv : public EnvWrapper {
    class CountingFile : public RandomAccessFile {
     public:
      CountingFile(unique_ptr<RandomAccessFile>&& target,
-                   anon::AtomicCounter* counter)
-          : target_(std::move(target)), counter_(counter) {}
+                   anon::AtomicCounter* counter,
+                   std::atomic<size_t>* bytes_read)
+          : target_(std::move(target)),
+            counter_(counter),
+            bytes_read_(bytes_read) {}
      virtual Status Read(uint64_t offset, size_t n, Slice* result,
                          char* scratch) const override {
        counter_->Increment();
-        return target_->Read(offset, n, result, scratch);
+        Status s = target_->Read(offset, n, result, scratch);
+        *bytes_read_ += result->size();
+        return s;
      }

     private:
      unique_ptr<RandomAccessFile> target_;
      anon::AtomicCounter* counter_;
+      std::atomic<size_t>* bytes_read_;
    };

    Status s = target()->NewRandomAccessFile(f, r, soptions);
    random_file_open_counter_++;
    if (s.ok() && count_random_reads_) {
-      r->reset(new CountingFile(std::move(*r), &random_read_counter_));
+      r->reset(new CountingFile(std::move(*r), &random_read_counter_,
+                                &random_read_bytes_counter_));
    }
    return s;
  }
@@ -463,6 +471,7 @@ class SpecialEnv : public EnvWrapper {

  bool count_random_reads_;
  anon::AtomicCounter random_read_counter_;
+  std::atomic<size_t> random_read_bytes_counter_;
  std::atomic<int> random_file_open_counter_;

  bool count_sequential_reads_;
@@ -491,6 +500,33 @@ class SpecialEnv : public EnvWrapper {
  std::atomic<bool> is_wal_sync_thread_safe_{true};
 };

+#ifndef ROCKSDB_LITE
+class OnFileDeletionListener : public EventListener {
+ public:
+  OnFileDeletionListener() : matched_count_(0), expected_file_name_("") {}
+
+  void SetExpectedFileName(const std::string file_name) {
+    expected_file_name_ = file_name;
+  }
+
+  void VerifyMatchedCount(size_t expected_value) {
+    ASSERT_EQ(matched_count_, expected_value);
+  }
+
+  void OnTableFileDeleted(const TableFileDeletionInfo& info) override {
+    if (expected_file_name_ != "") {
+      ASSERT_EQ(expected_file_name_, info.file_path);
+      expected_file_name_ = "";
+      matched_count_++;
+    }
+  }
+
+ private:
+  size_t matched_count_;
+  std::string expected_file_name_;
+};
+#endif
+
 class DBTestBase : public testing::Test {
 protected:
  // Sequence of option configurations to try
@@ -525,9 +561,11 @@ class DBTestBase : public testing::Test {
    kOptimizeFiltersForHits = 27,
    kRowCache = 28,
    kRecycleLogFiles = 29,
-    kLevelSubcompactions = 30,
-    kUniversalSubcompactions = 31,
-    kEnd = 30
+    kConcurrentSkipList = 30,
+    kEnd = 31,
+    kLevelSubcompactions = 31,
+    kUniversalSubcompactions = 32,
+    kBlockBasedTableWithIndexRestartInterval = 33,
  };
  int option_config_;

@@ -573,6 +611,8 @@ class DBTestBase : public testing::Test {
    return std::string(buf);
  }

+  static bool ShouldSkipOptions(int option_config, int skip_mask = kNoSkip);
+
  // Switch to a fresh database with the next option configuration to
  // test.  Return false if there are no more configurations to test.
  bool ChangeOptions(int skip_mask = kNoSkip);
@@ -661,13 +701,15 @@ class DBTestBase : public testing::Test {

  uint64_t SizeAtLevel(int level);

-  int TotalLiveFiles(int cf = 0);
+  size_t TotalLiveFiles(int cf = 0);

  size_t CountLiveFiles();
 #endif  // ROCKSDB_LITE

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

+  double CompressionRatioAtLevel(int level, int cf = 0);
+
  int TotalTableFiles(int cf = 0, int levels = -1);

  // Return spread of files per level
@@ -749,6 +791,20 @@ class DBTestBase : public testing::Test {

  void CopyFile(const std::string& source, const std::string& destination,
                uint64_t size = 0);
+
+  std::unordered_map<std::string, uint64_t> GetAllSSTFiles(
+      uint64_t* total_size = nullptr);
+
+  std::vector<std::uint64_t> ListTableFiles(Env* env, const std::string& path);
+
+#ifndef ROCKSDB_LITE
+  uint64_t GetNumberOfSstFilesForColumnFamily(DB* db,
+                                              std::string column_family_name);
+#endif  // ROCKSDB_LITE
+
+  uint64_t TestGetTickerCount(const Options& options, Tickers ticker_type) {
+    return options.statistics->getTickerCount(ticker_type);
+  }
 };

 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -22,12 +22,16 @@ static std::string CompressibleString(Random* rnd, int len) {

 class DBTestUniversalCompactionBase
    : public DBTestBase,
-      public ::testing::WithParamInterface<int> {
+      public ::testing::WithParamInterface<std::tuple<int, bool>> {
 public:
  explicit DBTestUniversalCompactionBase(
      const std::string& path) : DBTestBase(path) {}
-  virtual void SetUp() override { num_levels_ = GetParam(); }
+  virtual void SetUp() override {
+    num_levels_ = std::get<0>(GetParam());
+    exclusive_manual_compaction_ = std::get<1>(GetParam());
+  }
  int num_levels_;
+  bool exclusive_manual_compaction_;
 };

 class DBTestUniversalCompaction : public DBTestUniversalCompactionBase {
@@ -115,8 +119,7 @@ class DelayFilterFactory : public CompactionFilterFactory {
 // Make sure we don't trigger a problem if the trigger conditon is given
 // to be 0, which is invalid.
 TEST_P(DBTestUniversalCompaction, UniversalCompactionSingleSortedRun) {
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();

  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = num_levels_;
@@ -150,6 +153,71 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSingleSortedRun) {
  }
 }

+TEST_P(DBTestUniversalCompaction, OptimizeFiltersForHits) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.compaction_options_universal.size_ratio = 5;
+  options.num_levels = num_levels_;
+  options.write_buffer_size = 105 << 10;  // 105KB
+  options.arena_block_size = 4 << 10;
+  options.target_file_size_base = 32 << 10;  // 32KB
+  // trigger compaction if there are >= 4 files
+  options.level0_file_num_compaction_trigger = 4;
+  BlockBasedTableOptions bbto;
+  bbto.cache_index_and_filter_blocks = true;
+  bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
+  bbto.whole_key_filtering = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+  options.optimize_filters_for_hits = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.memtable_factory.reset(new SpecialSkipListFactory(3));
+
+  DestroyAndReopen(options);
+
+  // block compaction from happening
+  env_->SetBackgroundThreads(1, Env::LOW);
+  test::SleepingBackgroundTask sleeping_task_low;
+  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
+                 Env::Priority::LOW);
+
+  for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+    Put(Key(num * 10), "val");
+    if (num) {
+      dbfull()->TEST_WaitForFlushMemTable();
+    }
+    Put(Key(30 + num * 10), "val");
+    Put(Key(60 + num * 10), "val");
+  }
+  Put("", "");
+  dbfull()->TEST_WaitForFlushMemTable();
+
+  // Query set of non existing keys
+  for (int i = 5; i < 90; i += 10) {
+    ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+  }
+
+  // Make sure bloom filter is used at least once.
+  ASSERT_GT(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+  auto prev_counter = TestGetTickerCount(options, BLOOM_FILTER_USEFUL);
+
+  // Make sure bloom filter is used for all but the last L0 file when looking
+  // up a non-existent key that's in the range of all L0 files.
+  ASSERT_EQ(Get(Key(35)), "NOT_FOUND");
+  ASSERT_EQ(prev_counter + NumTableFilesAtLevel(0) - 1,
+            TestGetTickerCount(options, BLOOM_FILTER_USEFUL));
+  prev_counter = TestGetTickerCount(options, BLOOM_FILTER_USEFUL);
+
+  // Unblock compaction and wait it for happening.
+  sleeping_task_low.WakeUp();
+  dbfull()->TEST_WaitForCompact();
+
+  // The same queries will not trigger bloom filter
+  for (int i = 5; i < 90; i += 10) {
+    ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+  }
+  ASSERT_EQ(prev_counter, TestGetTickerCount(options, BLOOM_FILTER_USEFUL));
+}
+
 // TODO(kailiu) The tests on UniversalCompaction has some issues:
 //  1. A lot of magic numbers ("11" or "12").
 //  2. Made assumption on the memtable flush conditions, which may change from
@@ -262,13 +330,12 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrigger) {
 }

 TEST_P(DBTestUniversalCompaction, UniversalCompactionSizeAmplification) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = num_levels_;
  options.write_buffer_size = 100 << 10;     // 100KB
  options.target_file_size_base = 32 << 10;  // 32KB
  options.level0_file_num_compaction_trigger = 3;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -313,12 +380,12 @@ TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
  ChangeCompactOptions();
  Options options;
  options.create_if_missing = true;
-  options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
  options.compaction_style = kCompactionStyleLevel;
  options.num_levels = 1;
  options.target_file_size_base = options.write_buffer_size;
  options.compression = kNoCompression;
  options = CurrentOptions(options);
+  options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
  CreateAndReopenWithCF({"pikachu"}, options);
  ASSERT_EQ(options.compaction_style, kCompactionStyleUniversal);
  Random rnd(301);
@@ -376,12 +443,11 @@ TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
 }

 TEST_P(DBTestUniversalCompaction, UniversalCompactionTargetLevel) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.write_buffer_size = 100 << 10;     // 100KB
  options.num_levels = 7;
  options.disable_auto_compactions = true;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  // Generate 3 overlapping files
@@ -406,6 +472,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTargetLevel) {
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 4;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  db_->CompactRange(compact_options, nullptr, nullptr);
  ASSERT_EQ("0,0,0,0,1", FilesPerLevel(0));
 }
@@ -420,14 +487,13 @@ class DBTestUniversalCompactionMultiLevels
 };

 TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionMultiLevels) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = num_levels_;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 8;
  options.max_background_compactions = 3;
  options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  // Trigger compaction if size amplification exceeds 110%
@@ -463,7 +529,7 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
      });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.compaction_options_universal.allow_trivial_move = true;
  options.num_levels = 3;
@@ -471,7 +537,6 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
  options.level0_file_num_compaction_trigger = 3;
  options.max_background_compactions = 2;
  options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -498,7 +563,8 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {

 INSTANTIATE_TEST_CASE_P(DBTestUniversalCompactionMultiLevels,
                        DBTestUniversalCompactionMultiLevels,
-                        ::testing::Values(3, 20));
+                        ::testing::Combine(::testing::Values(3, 20),
+                                           ::testing::Bool()));

 class DBTestUniversalCompactionParallel :
    public DBTestUniversalCompactionBase {
@@ -509,7 +575,7 @@ class DBTestUniversalCompactionParallel :
 };

 TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = num_levels_;
  options.write_buffer_size = 1 << 10;  // 1KB
@@ -518,7 +584,6 @@ TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
  options.max_background_flushes = 3;
  options.target_file_size_base = 1 * 1024;
  options.compaction_options_universal.max_size_amplification_percent = 110;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -571,10 +636,11 @@ TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {

 INSTANTIATE_TEST_CASE_P(DBTestUniversalCompactionParallel,
                        DBTestUniversalCompactionParallel,
-                        ::testing::Values(1, 10));
+                        ::testing::Combine(::testing::Values(1, 10),
+                                           ::testing::Bool()));

 TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.write_buffer_size = 105 << 10;    // 105KB
  options.arena_block_size = 4 << 10;       // 4KB
@@ -582,7 +648,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
  options.level0_file_num_compaction_trigger = 4;
  options.num_levels = num_levels_;
  options.compaction_options_universal.compression_size_percent = -1;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -695,14 +760,13 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio1) {
    return;
  }

-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.write_buffer_size = 100 << 10;     // 100KB
  options.target_file_size_base = 32 << 10;  // 32KB
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = num_levels_;
  options.compaction_options_universal.compression_size_percent = 70;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  Random rnd(301);
@@ -763,14 +827,13 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio2) {
  if (!Snappy_Supported()) {
    return;
  }
-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.write_buffer_size = 100 << 10;     // 100KB
  options.target_file_size_base = 32 << 10;  // 32KB
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = num_levels_;
  options.compaction_options_universal.compression_size_percent = 95;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);

  Random rnd(301);
@@ -806,7 +869,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
      });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.compaction_options_universal.allow_trivial_move = true;
  options.num_levels = 2;
@@ -814,7 +877,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
  options.level0_file_num_compaction_trigger = 3;
  options.max_background_compactions = 1;
  options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -851,7 +913,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  Options options;
+  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.compaction_options_universal.allow_trivial_move = true;
  options.num_levels = 15;
@@ -859,7 +921,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
  options.level0_file_num_compaction_trigger = 8;
  options.max_background_compactions = 4;
  options.target_file_size_base = 64 * 1024;
-  options = CurrentOptions(options);
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -885,7 +946,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
 }

 TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
-  Options options;
+  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_, 300 * 1024);
  options.db_paths.emplace_back(dbname_ + "_2", 300 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 500 * 1024);
@@ -894,11 +955,10 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleUniversal;
  options.compaction_options_universal.size_ratio = 5;
-  options.write_buffer_size = 110 << 10;  // 105KB
+  options.write_buffer_size = 111 << 10;  // 114KB
  options.arena_block_size = 4 << 10;
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 1;
-  options = CurrentOptions(options);

  std::vector<std::string> filenames;
  env_->GetChildren(options.db_paths[1].path, &filenames);
@@ -1025,16 +1085,12 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
  for (int i = 0; i <= max_key1; i++) {
    // each value is 10K
    ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
  }
  ASSERT_OK(Flush(1));
  dbfull()->TEST_WaitForCompact();

-  int non_level0_num_files = 0;
-  for (int i = 1; i < options.num_levels; i++) {
-    non_level0_num_files += NumTableFilesAtLevel(i, 1);
-  }
-  ASSERT_EQ(non_level0_num_files, 0);
-
  // Stage 2: reopen with universal compaction, num_levels=4
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = 4;
@@ -1047,6 +1103,8 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
  for (int i = max_key1 + 1; i <= max_key2; i++) {
    // each value is 10K
    ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
  }
  ASSERT_OK(Flush(1));
  dbfull()->TEST_WaitForCompact();
@@ -1063,6 +1121,7 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 0;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  dbfull()->CompactRange(compact_options, handles_[1], nullptr, nullptr);
  // Need to restart it once to remove higher level records in manifest.
  ReopenWithColumnFamilies({"default", "pikachu"}, options);
@@ -1076,6 +1135,8 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
  for (int i = max_key2 + 1; i <= max_key3; i++) {
    // each value is 10K
    ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
  }
  ASSERT_OK(Flush(1));
  dbfull()->TEST_WaitForCompact();
@@ -1087,19 +1148,17 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSecondPathRatio) {
  if (!Snappy_Supported()) {
    return;
  }
-  Options options;
+  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_, 500 * 1024);
  options.db_paths.emplace_back(dbname_ + "_2", 1024 * 1024 * 1024);
  options.compaction_style = kCompactionStyleUniversal;
  options.compaction_options_universal.size_ratio = 5;
-  options.write_buffer_size = 110 << 10;  // 105KB
-  options.arena_block_size = 4 * 1024;
+  options.write_buffer_size = 111 << 10;  // 114KB
  options.arena_block_size = 4 << 10;
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 1;
  options.memtable_factory.reset(
      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
-  options = CurrentOptions(options);

  std::vector<std::string> filenames;
  env_->GetChildren(options.db_paths[1].path, &filenames);
@@ -1186,7 +1245,8 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSecondPathRatio) {
 }

 INSTANTIATE_TEST_CASE_P(UniversalCompactionNumLevels, DBTestUniversalCompaction,
-                        ::testing::Values(1, 3, 5));
+                        ::testing::Combine(::testing::Values(1, 3, 5),
+                                           ::testing::Bool()));

 class DBTestUniversalManualCompactionOutputPathId
    : public DBTestUniversalCompactionBase {
@@ -1218,6 +1278,7 @@ TEST_P(DBTestUniversalManualCompactionOutputPathId,
  // Full compaction to DB path 0
  CompactRangeOptions compact_options;
  compact_options.target_path_id = 1;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);
  ASSERT_EQ(1, TotalLiveFiles(1));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
@@ -1240,6 +1301,7 @@ TEST_P(DBTestUniversalManualCompactionOutputPathId,

  // Full compaction to DB path 0
  compact_options.target_path_id = 0;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);
  ASSERT_EQ(1, TotalLiveFiles(1));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[0].path));
@@ -1247,13 +1309,15 @@ TEST_P(DBTestUniversalManualCompactionOutputPathId,

  // Fail when compacting to an invalid path ID
  compact_options.target_path_id = 2;
+  compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
  ASSERT_TRUE(db_->CompactRange(compact_options, handles_[1], nullptr, nullptr)
                  .IsInvalidArgument());
 }

 INSTANTIATE_TEST_CASE_P(DBTestUniversalManualCompactionOutputPathId,
                        DBTestUniversalManualCompactionOutputPathId,
-                        ::testing::Values(1, 8));
+                        ::testing::Combine(::testing::Values(1, 8),
+                                           ::testing::Bool()));

 }  // namespace rocksdb

@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -127,6 +127,637 @@ TEST_F(DBWALTest, SyncWALNotWaitWrite) {
  ASSERT_EQ(Get("foo2"), "bar2");
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
+
+TEST_F(DBWALTest, Recover) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "baz", "v5"));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v1", Get(1, "foo"));
+
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+    ASSERT_OK(Put(1, "bar", "v2"));
+    ASSERT_OK(Put(1, "foo", "v3"));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v3", Get(1, "foo"));
+    ASSERT_OK(Put(1, "foo", "v4"));
+    ASSERT_EQ("v4", Get(1, "foo"));
+    ASSERT_EQ("v2", Get(1, "bar"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+  } while (ChangeOptions());
+}
+
+TEST_F(DBWALTest, RecoverWithTableHandle) {
+  do {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.disable_auto_compactions = true;
+    DestroyAndReopen(options);
+    CreateAndReopenWithCF({"pikachu"}, options);
+
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "bar", "v2"));
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(Put(1, "foo", "v3"));
+    ASSERT_OK(Put(1, "bar", "v4"));
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(Put(1, "big", std::string(100, 'a')));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+
+    std::vector<std::vector<FileMetaData>> files;
+    dbfull()->TEST_GetFilesMetaData(handles_[1], &files);
+    size_t total_files = 0;
+    for (const auto& level : files) {
+      total_files += level.size();
+    }
+    ASSERT_EQ(total_files, 3);
+    for (const auto& level : files) {
+      for (const auto& file : level) {
+        if (kInfiniteMaxOpenFiles == option_config_) {
+          ASSERT_TRUE(file.table_reader_handle != nullptr);
+        } else {
+          ASSERT_TRUE(file.table_reader_handle == nullptr);
+        }
+      }
+    }
+  } while (ChangeOptions());
+}
+
+TEST_F(DBWALTest, IgnoreRecoveredLog) {
+  std::string backup_logs = dbname_ + "/backup_logs";
+
+  // delete old files in backup_logs directory
+  env_->CreateDirIfMissing(backup_logs);
+  std::vector<std::string> old_files;
+  env_->GetChildren(backup_logs, &old_files);
+  for (auto& file : old_files) {
+    if (file != "." && file != "..") {
+      env_->DeleteFile(backup_logs + "/" + file);
+    }
+  }
+
+  do {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.merge_operator = MergeOperators::CreateUInt64AddOperator();
+    options.wal_dir = dbname_ + "/logs";
+    DestroyAndReopen(options);
+
+    // fill up the DB
+    std::string one, two;
+    PutFixed64(&one, 1);
+    PutFixed64(&two, 2);
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("foo"), Slice(one)));
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("foo"), Slice(one)));
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("bar"), Slice(one)));
+
+    // copy the logs to backup
+    std::vector<std::string> logs;
+    env_->GetChildren(options.wal_dir, &logs);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(options.wal_dir + "/" + log, backup_logs + "/" + log);
+      }
+    }
+
+    // recover the DB
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+    Close();
+
+    // copy the logs from backup back to wal dir
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+      }
+    }
+    // this should ignore the log files, recovery should not happen again
+    // if the recovery happens, the same merge operator would be called twice,
+    // leading to incorrect results
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+    Close();
+    Destroy(options);
+    Reopen(options);
+    Close();
+
+    // copy the logs from backup back to wal dir
+    env_->CreateDirIfMissing(options.wal_dir);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+      }
+    }
+    // assert that we successfully recovered only from logs, even though we
+    // destroyed the DB
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+
+    // Recovery will fail if DB directory doesn't exist.
+    Destroy(options);
+    // copy the logs from backup back to wal dir
+    env_->CreateDirIfMissing(options.wal_dir);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+        // we won't be needing this file no more
+        env_->DeleteFile(backup_logs + "/" + log);
+      }
+    }
+    Status s = TryReopen(options);
+    ASSERT_TRUE(!s.ok());
+  } while (ChangeOptions(kSkipHashCuckoo));
+}
+
+TEST_F(DBWALTest, RecoveryWithEmptyLog) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "foo", "v2"));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v3"));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v3", Get(1, "foo"));
+  } while (ChangeOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBWALTest, RecoverWithLargeLog) {
+  do {
+    {
+      Options options = CurrentOptions();
+      CreateAndReopenWithCF({"pikachu"}, options);
+      ASSERT_OK(Put(1, "big1", std::string(200000, '1')));
+      ASSERT_OK(Put(1, "big2", std::string(200000, '2')));
+      ASSERT_OK(Put(1, "small3", std::string(10, '3')));
+      ASSERT_OK(Put(1, "small4", std::string(10, '4')));
+      ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
+    }
+
+    // Make sure that if we re-open with a small write buffer size that
+    // we flush table files in the middle of a large log file.
+    Options options;
+    options.write_buffer_size = 100000;
+    options = CurrentOptions(options);
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    ASSERT_EQ(NumTableFilesAtLevel(0, 1), 3);
+    ASSERT_EQ(std::string(200000, '1'), Get(1, "big1"));
+    ASSERT_EQ(std::string(200000, '2'), Get(1, "big2"));
+    ASSERT_EQ(std::string(10, '3'), Get(1, "small3"));
+    ASSERT_EQ(std::string(10, '4'), Get(1, "small4"));
+    ASSERT_GT(NumTableFilesAtLevel(0, 1), 1);
+  } while (ChangeCompactOptions());
+}
+
+// In https://reviews.facebook.net/D20661 we change
+// recovery behavior: previously for each log file each column family
+// memtable was flushed, even it was empty. Now it's changed:
+// we try to create the smallest number of table files by merging
+// updates from multiple logs
+TEST_F(DBWALTest, RecoverCheckFileAmountWithSmallWriteBuffer) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 5000000;
+  CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
+
+  // Since we will reopen DB with smaller write_buffer_size,
+  // each key will go to new SST file
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  // Make 'dobrynia' to be flushed and new WAL file to be created
+  ASSERT_OK(Put(2, Key(10), DummyString(7500000)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(1));
+    // Make sure 'dobrynia' was flushed: check sst files amount
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(1));
+  }
+  // New WAL file
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+
+  options.write_buffer_size = 4096;
+  options.arena_block_size = 4096;
+  ReopenWithColumnFamilies({"default", "pikachu", "dobrynia", "nikitich"},
+                           options);
+  {
+    // No inserts => default is empty
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
+              static_cast<uint64_t>(0));
+    // First 4 keys goes to separate SSTs + 1 more SST for 2 smaller keys
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
+              static_cast<uint64_t>(5));
+    // 1 SST for big key + 1 SST for small one
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(2));
+    // 1 SST for all keys
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(1));
+  }
+}
+
+// In https://reviews.facebook.net/D20661 we change
+// recovery behavior: previously for each log file each column family
+// memtable was flushed, even it wasn't empty. Now it's changed:
+// we try to create the smallest number of table files by merging
+// updates from multiple logs
+TEST_F(DBWALTest, RecoverCheckFileAmount) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 100000;
+  options.arena_block_size = 4 * 1024;
+  CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
+
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  // Make 'nikitich' memtable to be flushed
+  ASSERT_OK(Put(3, Key(10), DummyString(1002400)));
+  ASSERT_OK(Put(3, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  // 4 memtable are not flushed, 1 sst file
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(1));
+  }
+  // Memtable for 'nikitich' has flushed, new WAL file has opened
+  // 4 memtable still not flushed
+
+  // Write to new WAL file
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  // Fill up 'nikitich' one more time
+  ASSERT_OK(Put(3, Key(10), DummyString(1002400)));
+  // make it flush
+  ASSERT_OK(Put(3, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  // There are still 4 memtable not flushed, and 2 sst tables
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(2));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(2));
+  }
+
+  ReopenWithColumnFamilies({"default", "pikachu", "dobrynia", "nikitich"},
+                           options);
+  {
+    std::vector<uint64_t> table_files = ListTableFiles(env_, dbname_);
+    // Check, that records for 'default', 'dobrynia' and 'pikachu' from
+    // first, second and third WALs  went to the same SST.
+    // So, there is 6 SSTs: three  for 'nikitich', one for 'default', one for
+    // 'dobrynia', one for 'pikachu'
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
+              static_cast<uint64_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(3));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
+              static_cast<uint64_t>(1));
+  }
+}
+
+TEST_F(DBWALTest, SyncMultipleLogs) {
+  const uint64_t kNumBatches = 2;
+  const int kBatchSize = 1000;
+
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.write_buffer_size = 4096;
+  Reopen(options);
+
+  WriteBatch batch;
+  WriteOptions wo;
+  wo.sync = true;
+
+  for (uint64_t b = 0; b < kNumBatches; b++) {
+    batch.Clear();
+    for (int i = 0; i < kBatchSize; i++) {
+      batch.Put(Key(i), DummyString(128));
+    }
+
+    dbfull()->Write(wo, &batch);
+  }
+
+  ASSERT_OK(dbfull()->SyncWAL());
+}
+
+//
+// Test WAL recovery for the various modes available
+//
+class RecoveryTestHelper {
+ public:
+  // Number of WAL files to generate
+  static const int kWALFilesCount = 10;
+  // Starting number for the WAL file name like 00010.log
+  static const int kWALFileOffset = 10;
+  // Keys to be written per WAL file
+  static const int kKeysPerWALFile = 1024;
+  // Size of the value
+  static const int kValueSize = 10;
+
+  // Create WAL files with values filled in
+  static void FillData(DBWALTest* test, const Options& options,
+                       const size_t wal_count, size_t* count) {
+    const DBOptions& db_options = options;
+
+    *count = 0;
+
+    shared_ptr<Cache> table_cache = NewLRUCache(50000, 16);
+    EnvOptions env_options;
+    WriteBuffer write_buffer(db_options.db_write_buffer_size);
+
+    unique_ptr<VersionSet> versions;
+    unique_ptr<WalManager> wal_manager;
+    WriteController write_controller;
+
+    versions.reset(new VersionSet(test->dbname_, &db_options, env_options,
+                                  table_cache.get(), &write_buffer,
+                                  &write_controller));
+
+    wal_manager.reset(new WalManager(db_options, env_options));
+
+    std::unique_ptr<log::Writer> current_log_writer;
+
+    for (size_t j = kWALFileOffset; j < wal_count + kWALFileOffset; j++) {
+      uint64_t current_log_number = j;
+      std::string fname = LogFileName(test->dbname_, current_log_number);
+      unique_ptr<WritableFile> file;
+      ASSERT_OK(db_options.env->NewWritableFile(fname, &file, env_options));
+      unique_ptr<WritableFileWriter> file_writer(
+          new WritableFileWriter(std::move(file), env_options));
+      current_log_writer.reset(
+          new log::Writer(std::move(file_writer), current_log_number,
+                          db_options.recycle_log_file_num > 0));
+
+      for (int i = 0; i < kKeysPerWALFile; i++) {
+        std::string key = "key" + ToString((*count)++);
+        std::string value = test->DummyString(kValueSize);
+        assert(current_log_writer.get() != nullptr);
+        uint64_t seq = versions->LastSequence() + 1;
+        WriteBatch batch;
+        batch.Put(key, value);
+        WriteBatchInternal::SetSequence(&batch, seq);
+        current_log_writer->AddRecord(WriteBatchInternal::Contents(&batch));
+        versions->SetLastSequence(seq);
+      }
+    }
+  }
+
+  // Recreate and fill the store with some data
+  static size_t FillData(DBWALTest* test, Options* options) {
+    options->create_if_missing = true;
+    test->DestroyAndReopen(*options);
+    test->Close();
+
+    size_t count = 0;
+    FillData(test, *options, kWALFilesCount, &count);
+    return count;
+  }
+
+  // Read back all the keys we wrote and return the number of keys found
+  static size_t GetData(DBWALTest* test) {
+    size_t count = 0;
+    for (size_t i = 0; i < kWALFilesCount * kKeysPerWALFile; i++) {
+      if (test->Get("key" + ToString(i)) != "NOT_FOUND") {
+        ++count;
+      }
+    }
+    return count;
+  }
+
+  // Manuall corrupt the specified WAL
+  static void CorruptWAL(DBWALTest* test, const Options& options,
+                         const double off, const double len,
+                         const int wal_file_id, const bool trunc = false) {
+    Env* env = options.env;
+    std::string fname = LogFileName(test->dbname_, wal_file_id);
+    uint64_t size;
+    ASSERT_OK(env->GetFileSize(fname, &size));
+    ASSERT_GT(size, 0);
+#ifdef OS_WIN
+    // Windows disk cache behaves differently. When we truncate
+    // the original content is still in the cache due to the original
+    // handle is still open. Generally, in Windows, one prohibits
+    // shared access to files and it is not needed for WAL but we allow
+    // it to induce corruption at various tests.
+    test->Close();
+#endif
+    if (trunc) {
+      ASSERT_EQ(0, truncate(fname.c_str(), static_cast<int64_t>(size * off)));
+    } else {
+      InduceCorruption(fname, static_cast<size_t>(size * off),
+                       static_cast<size_t>(size * len));
+    }
+  }
+
+  // Overwrite data with 'a' from offset for length len
+  static void InduceCorruption(const std::string& filename, size_t offset,
+                               size_t len) {
+    ASSERT_GT(len, 0U);
+
+    int fd = open(filename.c_str(), O_RDWR);
+
+    // On windows long is 32-bit
+    ASSERT_LE(offset, std::numeric_limits<long>::max());
+
+    ASSERT_GT(fd, 0);
+    ASSERT_EQ(offset, lseek(fd, static_cast<long>(offset), SEEK_SET));
+
+    void* buf = alloca(len);
+    memset(buf, 'a', len);
+    ASSERT_EQ(len, write(fd, buf, static_cast<unsigned int>(len)));
+
+    close(fd);
+  }
+};
+
+// Test scope:
+// - We expect to open the data store when there is incomplete trailing writes
+// at the end of any of the logs
+// - We do not expect to open the data store for corruption
+TEST_F(DBWALTest, kTolerateCorruptedTailRecords) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Corruption offset position */
+      for (int j = jstart; j < jend; j++) { /* WAL file */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+        // test checksum failure or parsing
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, /*wal=*/j, trunc);
+
+        if (trunc) {
+          options.wal_recovery_mode =
+              WALRecoveryMode::kTolerateCorruptedTailRecords;
+          options.create_if_missing = false;
+          ASSERT_OK(TryReopen(options));
+          const size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+          ASSERT_TRUE(i == 0 || recovered_row_count > 0);
+          ASSERT_LT(recovered_row_count, row_count);
+        } else {
+          options.wal_recovery_mode =
+              WALRecoveryMode::kTolerateCorruptedTailRecords;
+          ASSERT_NOK(TryReopen(options));
+        }
+      }
+    }
+  }
+}
+
+// Test scope:
+// We don't expect the data store to be opened if there is any corruption
+// (leading, middle or trailing -- incomplete writes or corruption)
+TEST_F(DBWALTest, kAbsoluteConsistency) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  // Verify clean slate behavior
+  Options options = CurrentOptions();
+  const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+  options.wal_recovery_mode = WALRecoveryMode::kAbsoluteConsistency;
+  options.create_if_missing = false;
+  ASSERT_OK(TryReopen(options));
+  ASSERT_EQ(RecoveryTestHelper::GetData(this), row_count);
+
+  for (auto trunc : {true, false}) { /* Corruption style */
+    for (int i = 0; i < 4; i++) {    /* Corruption offset position */
+      if (trunc && i == 0) {
+        continue;
+      }
+
+      for (int j = jstart; j < jend; j++) { /* wal files */
+        // fill with new date
+        RecoveryTestHelper::FillData(this, &options);
+        // corrupt the wal
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+        // verify
+        options.wal_recovery_mode = WALRecoveryMode::kAbsoluteConsistency;
+        options.create_if_missing = false;
+        ASSERT_NOK(TryReopen(options));
+      }
+    }
+  }
+}
+
+// Test scope:
+// - We expect to open data store under all circumstances
+// - We expect only data upto the point where the first error was encountered
+TEST_F(DBWALTest, kPointInTimeRecovery) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+  const int maxkeys =
+      RecoveryTestHelper::kWALFilesCount * RecoveryTestHelper::kKeysPerWALFile;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Offset of corruption */
+      for (int j = jstart; j < jend; j++) { /* WAL file */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+
+        // Corrupt the wal
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+
+        // Verify
+        options.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
+        options.create_if_missing = false;
+        ASSERT_OK(TryReopen(options));
+
+        // Probe data for invariants
+        size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+        ASSERT_LT(recovered_row_count, row_count);
+
+        bool expect_data = true;
+        for (size_t k = 0; k < maxkeys; ++k) {
+          bool found = Get("key" + ToString(i)) != "NOT_FOUND";
+          if (expect_data && !found) {
+            expect_data = false;
+          }
+          ASSERT_EQ(found, expect_data);
+        }
+
+        const size_t min = RecoveryTestHelper::kKeysPerWALFile *
+                           (j - RecoveryTestHelper::kWALFileOffset);
+        ASSERT_GE(recovered_row_count, min);
+        if (!trunc && i != 0) {
+          const size_t max = RecoveryTestHelper::kKeysPerWALFile *
+                             (j - RecoveryTestHelper::kWALFileOffset + 1);
+          ASSERT_LE(recovered_row_count, max);
+        }
+      }
+    }
+  }
+}
+
+// Test scope:
+// - We expect to open the data store under all scenarios
+// - We expect to have recovered records past the corruption zone
+TEST_F(DBWALTest, kSkipAnyCorruptedRecords) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Corruption offset */
+      for (int j = jstart; j < jend; j++) { /* wal files */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+
+        // Corrupt the WAL
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+
+        // Verify behavior
+        options.wal_recovery_mode = WALRecoveryMode::kSkipAnyCorruptedRecords;
+        options.create_if_missing = false;
+        ASSERT_OK(TryReopen(options));
+
+        // Probe data for invariants
+        size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+        ASSERT_LT(recovered_row_count, row_count);
+
+        if (!trunc) {
+          ASSERT_TRUE(i != 0 || recovered_row_count > 0);
+        }
+      }
+    }
+  }
+}
+
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -39,6 +39,11 @@ enum ValueType : unsigned char {
  kTypeColumnFamilyMerge = 0x6,     // WAL only.
  kTypeSingleDeletion = 0x7,
  kTypeColumnFamilySingleDeletion = 0x8,  // WAL only.
+  kTypeBeginPrepareXID = 0x9,             // WAL only.
+  kTypeEndPrepareXID = 0xA,               // WAL only.
+  kTypeCommitXID = 0xB,                   // WAL only.
+  kTypeRollbackXID = 0xC,                 // WAL only.
+  kTypeNoop = 0xD,                        // WAL only.
  kMaxValue = 0x7F                        // Not used for storing records.
 };

@@ -271,7 +276,8 @@ inline LookupKey::~LookupKey() {

 class IterKey {
 public:
-  IterKey() : key_(space_), buf_size_(sizeof(space_)), key_size_(0) {}
+  IterKey()
+      : buf_(space_), buf_size_(sizeof(space_)), key_(buf_), key_size_(0) {}

  ~IterKey() { ResetBuffer(); }

@@ -293,31 +299,41 @@ class IterKey {
  void TrimAppend(const size_t shared_len, const char* non_shared_data,
                  const size_t non_shared_len) {
    assert(shared_len <= key_size_);
-
    size_t total_size = shared_len + non_shared_len;
-    if (total_size <= buf_size_) {
-      key_size_ = total_size;
-    } else {
+
+    if (IsKeyPinned() /* key is not in buf_ */) {
+      // Copy the key from external memory to buf_ (copy shared_len bytes)
+      EnlargeBufferIfNeeded(total_size);
+      memcpy(buf_, key_, shared_len);
+    } else if (total_size > buf_size_) {
      // Need to allocate space, delete previous space
      char* p = new char[total_size];
      memcpy(p, key_, shared_len);

-      if (key_ != space_) {
-        delete[] key_;
+      if (buf_ != space_) {
+        delete[] buf_;
      }

-      key_ = p;
-      key_size_ = total_size;
+      buf_ = p;
      buf_size_ = total_size;
    }

-    memcpy(key_ + shared_len, non_shared_data, non_shared_len);
+    memcpy(buf_ + shared_len, non_shared_data, non_shared_len);
+    key_ = buf_;
+    key_size_ = total_size;
  }

-  Slice SetKey(const Slice& key) {
+  Slice SetKey(const Slice& key, bool copy = true) {
    size_t size = key.size();
-    EnlargeBufferIfNeeded(size);
-    memcpy(key_, key.data(), size);
+    if (copy) {
+      // Copy key to buf_
+      EnlargeBufferIfNeeded(size);
+      memcpy(buf_, key.data(), size);
+      key_ = buf_;
+    } else {
+      // Update key_ to point to external memory
+      key_ = key.data();
+    }
    key_size_ = size;
    return Slice(key_, key_size_);
  }
@@ -335,11 +351,14 @@ class IterKey {
  // Update the sequence number in the internal key.  Guarantees not to
  // invalidate slices to the key (and the user key).
  void UpdateInternalKey(uint64_t seq, ValueType t) {
+    assert(!IsKeyPinned());
    assert(key_size_ >= 8);
    uint64_t newval = (seq << 8) | t;
-    EncodeFixed64(&key_[key_size_ - 8], newval);
+    EncodeFixed64(&buf_[key_size_ - 8], newval);
  }

+  bool IsKeyPinned() const { return (key_ != buf_); }
+
  void SetInternalKey(const Slice& key_prefix, const Slice& user_key,
                      SequenceNumber s,
                      ValueType value_type = kValueTypeForSeek) {
@@ -347,10 +366,12 @@ class IterKey {
    size_t usize = user_key.size();
    EnlargeBufferIfNeeded(psize + usize + sizeof(uint64_t));
    if (psize > 0) {
-      memcpy(key_, key_prefix.data(), psize);
+      memcpy(buf_, key_prefix.data(), psize);
    }
-    memcpy(key_ + psize, user_key.data(), usize);
-    EncodeFixed64(key_ + usize + psize, PackSequenceAndType(s, value_type));
+    memcpy(buf_ + psize, user_key.data(), usize);
+    EncodeFixed64(buf_ + usize + psize, PackSequenceAndType(s, value_type));
+
+    key_ = buf_;
    key_size_ = psize + usize + sizeof(uint64_t);
  }

@@ -377,20 +398,22 @@ class IterKey {
  void EncodeLengthPrefixedKey(const Slice& key) {
    auto size = key.size();
    EnlargeBufferIfNeeded(size + static_cast<size_t>(VarintLength(size)));
-    char* ptr = EncodeVarint32(key_, static_cast<uint32_t>(size));
+    char* ptr = EncodeVarint32(buf_, static_cast<uint32_t>(size));
    memcpy(ptr, key.data(), size);
+    key_ = buf_;
  }

 private:
-  char* key_;
+  char* buf_;
  size_t buf_size_;
+  const char* key_;
  size_t key_size_;
  char space_[32];  // Avoid allocation for short keys

  void ResetBuffer() {
-    if (key_ != space_) {
-      delete[] key_;
-      key_ = space_;
+    if (buf_ != space_) {
+      delete[] buf_;
+      buf_ = space_;
    }
    buf_size_ = sizeof(space_);
    key_size_ = 0;
@@ -407,7 +430,7 @@ class IterKey {
    if (key_size > buf_size_) {
      // Need to enlarge the buffer.
      ResetBuffer();
-      key_ = new char[key_size];
+      buf_ = new char[key_size];
      buf_size_ = key_size;
    }
  }
@@ -447,6 +470,12 @@ class InternalKeySliceTransform : public SliceTransform {
  const SliceTransform* const transform_;
 };

+// Read the key of a record from a write batch.
+// if this record represent the default column family then cf_record
+// must be passed as false, otherwise it must be passed as true.
+extern bool ReadKeyFromWriteBatchEntry(Slice* input, Slice* key,
+                                       bool cf_record);
+
 // Read record from a write batch piece from input.
 // tag, column_family, key, value and blob are return values. Callers own the
 // Slice they point to.
@@ -454,5 +483,5 @@ class InternalKeySliceTransform : public SliceTransform {
 // input will be advanced to after the record.
 extern Status ReadRecordFromWriteBatch(Slice* input, char* tag,
                                       uint32_t* column_family, Slice* key,
-                                       Slice* value, Slice* blob);
+                                       Slice* value, Slice* blob, Slice* xid);
 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -92,6 +92,30 @@ TEST_F(FormatTest, InternalKeyShortSeparator) {
            Shorten(IKey("foo", 100, kTypeValue),
                    IKey("hello", 200, kTypeValue)));

+  ASSERT_EQ(IKey("ABC2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("ABC1AAAAA", 100, kTypeValue),
+                    IKey("ABC2ABB", 200, kTypeValue)));
+
+  ASSERT_EQ(IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("AAA1AAA", 100, kTypeValue),
+                    IKey("AAA2AA", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+      Shorten(IKey("AAA1AAA", 100, kTypeValue), IKey("AAA4", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA1B", kMaxSequenceNumber, kValueTypeForSeek),
+      Shorten(IKey("AAA1AAA", 100, kTypeValue), IKey("AAA2", 200, kTypeValue)));
+
+  ASSERT_EQ(IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("AAA1AAA", 100, kTypeValue),
+                    IKey("AAA2A", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA1", 100, kTypeValue),
+      Shorten(IKey("AAA1", 100, kTypeValue), IKey("AAA2", 200, kTypeValue)));
+
  // When start user key is prefix of limit user key
  ASSERT_EQ(IKey("foo", 100, kTypeValue),
            Shorten(IKey("foo", 100, kTypeValue),
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -37,6 +37,7 @@ class DeleteFileTest : public testing::Test {
  DeleteFileTest() {
    db_ = nullptr;
    env_ = Env::Default();
+    options_.delete_obsolete_files_period_micros = 0;  // always do full purge
    options_.enable_thread_tracking = true;
    options_.write_buffer_size = 1024*1024*1000;
    options_.target_file_size_base = 1024*1024*1000;
@@ -74,6 +75,7 @@ class DeleteFileTest : public testing::Test {

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

  void AddKeys(int numkeys, int startkey = 0) {
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -8,7 +8,10 @@
 namespace rocksdb {

 namespace {
-inline double SafeDivide(double a, double b) { return b == 0.0 ? 0 : a / b; }
+template<class T>
+inline T SafeDivide(T a, T b) {
+  return b == 0 ? 0 : a / b;
+}
 }  // namespace

 void EventHelpers::AppendCurrentTime(JSONWriter* jwriter) {
@@ -17,57 +20,83 @@ void EventHelpers::AppendCurrentTime(JSONWriter* jwriter) {
                  std::chrono::system_clock::now().time_since_epoch()).count();
 }

-void EventHelpers::LogAndNotifyTableFileCreation(
+#ifndef ROCKSDB_LITE
+void EventHelpers::NotifyTableFileCreationStarted(
+    const std::vector<std::shared_ptr<EventListener>>& listeners,
+    const std::string& db_name, const std::string& cf_name,
+    const std::string& file_path, int job_id, TableFileCreationReason reason) {
+  TableFileCreationBriefInfo info;
+  info.db_name = db_name;
+  info.cf_name = cf_name;
+  info.file_path = file_path;
+  info.job_id = job_id;
+  info.reason = reason;
+  for (auto& listener : listeners) {
+    listener->OnTableFileCreationStarted(info);
+  }
+}
+#endif  // !ROCKSDB_LITE
+
+void EventHelpers::LogAndNotifyTableFileCreationFinished(
    EventLogger* event_logger,
    const std::vector<std::shared_ptr<EventListener>>& listeners,
-    const FileDescriptor& fd, const TableFileCreationInfo& info) {
-  assert(event_logger);
-  JSONWriter jwriter;
-  AppendCurrentTime(&jwriter);
-  jwriter << "cf_name" << info.cf_name
-          << "job" << info.job_id
-          << "event" << "table_file_creation"
-          << "file_number" << fd.GetNumber()
-          << "file_size" << fd.GetFileSize();
+    const std::string& db_name, const std::string& cf_name,
+    const std::string& file_path, int job_id, const FileDescriptor& fd,
+    const TableProperties& table_properties, TableFileCreationReason reason,
+    const Status& s) {
+  if (s.ok() && event_logger) {
+    JSONWriter jwriter;
+    AppendCurrentTime(&jwriter);
+    jwriter << "cf_name" << cf_name << "job" << job_id << "event"
+            << "table_file_creation"
+            << "file_number" << fd.GetNumber() << "file_size"
+            << fd.GetFileSize();

-  // table_properties
-  {
-    jwriter << "table_properties";
-    jwriter.StartObject();
+    // table_properties
+    {
+      jwriter << "table_properties";
+      jwriter.StartObject();

-    // basic properties:
-    jwriter << "data_size" << info.table_properties.data_size
-            << "index_size" << info.table_properties.index_size
-            << "filter_size" << info.table_properties.filter_size
-            << "raw_key_size" << info.table_properties.raw_key_size
-            << "raw_average_key_size" << SafeDivide(
-                info.table_properties.raw_key_size,
-                info.table_properties.num_entries)
-            << "raw_value_size" << info.table_properties.raw_value_size
-            << "raw_average_value_size" << SafeDivide(
-               info.table_properties.raw_value_size,
-               info.table_properties.num_entries)
-            << "num_data_blocks" << info.table_properties.num_data_blocks
-            << "num_entries" << info.table_properties.num_entries
-            << "filter_policy_name" <<
-                info.table_properties.filter_policy_name;
+      // basic properties:
+      jwriter << "data_size" << table_properties.data_size << "index_size"
+              << table_properties.index_size << "filter_size"
+              << table_properties.filter_size << "raw_key_size"
+              << table_properties.raw_key_size << "raw_average_key_size"
+              << SafeDivide(table_properties.raw_key_size,
+                            table_properties.num_entries)
+              << "raw_value_size" << table_properties.raw_value_size
+              << "raw_average_value_size"
+              << SafeDivide(table_properties.raw_value_size,
+                            table_properties.num_entries)
+              << "num_data_blocks" << table_properties.num_data_blocks
+              << "num_entries" << table_properties.num_entries
+              << "filter_policy_name" << table_properties.filter_policy_name;

-    // user collected properties
-    for (const auto& prop : info.table_properties.readable_properties) {
-      jwriter << prop.first << prop.second;
+      // user collected properties
+      for (const auto& prop : table_properties.readable_properties) {
+        jwriter << prop.first << prop.second;
+      }
+      jwriter.EndObject();
    }
    jwriter.EndObject();
-  }
-  jwriter.EndObject();

-  event_logger->Log(jwriter);
+    event_logger->Log(jwriter);
+  }

 #ifndef ROCKSDB_LITE
  if (listeners.size() == 0) {
    return;
  }
-
-  for (auto listener : listeners) {
+  TableFileCreationInfo info;
+  info.db_name = db_name;
+  info.cf_name = cf_name;
+  info.file_path = file_path;
+  info.file_size = fd.file_size;
+  info.job_id = job_id;
+  info.table_properties = table_properties;
+  info.reason = reason;
+  info.status = s;
+  for (auto& listener : listeners) {
    listener->OnTableFileCreated(info);
  }
 #endif  // !ROCKSDB_LITE
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -19,15 +19,30 @@ namespace rocksdb {
 class EventHelpers {
 public:
  static void AppendCurrentTime(JSONWriter* json_writer);
-  static void LogAndNotifyTableFileCreation(
+#ifndef ROCKSDB_LITE
+  static void NotifyTableFileCreationStarted(
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      const std::string& db_name, const std::string& cf_name,
+      const std::string& file_path, int job_id, TableFileCreationReason reason);
+#endif  // !ROCKSDB_LITE
+  static void LogAndNotifyTableFileCreationFinished(
      EventLogger* event_logger,
      const std::vector<std::shared_ptr<EventListener>>& listeners,
-      const FileDescriptor& fd, const TableFileCreationInfo& info);
+      const std::string& db_name, const std::string& cf_name,
+      const std::string& file_path, int job_id, const FileDescriptor& fd,
+      const TableProperties& table_properties, TableFileCreationReason reason,
+      const Status& s);
  static void LogAndNotifyTableFileDeletion(
      EventLogger* event_logger, int job_id,
      uint64_t file_number, const std::string& file_path,
      const Status& status, const std::string& db_name,
      const std::vector<std::shared_ptr<EventListener>>& listeners);
+
+ private:
+  static void LogAndNotifyTableFileCreation(
+      EventLogger* event_logger,
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      const FileDescriptor& fd, const TableFileCreationInfo& info);
 };

 }  // namespace rocksdb
@@ -1,4 +1,4 @@
-//  Copyright (c) 2014, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2015, Facebook, Inc.  All rights reserved.
+//  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.
@@ -11,8 +11,6 @@
 // the last "sync". It then checks for data loss errors by purposely dropping
 // file data (or entire files) not protected by a "sync".

-#include <map>
-#include <set>
 #include "db/db_impl.h"
 #include "db/filename.h"
 #include "db/log_format.h"
@@ -22,6 +20,7 @@
 #include "rocksdb/env.h"
 #include "rocksdb/table.h"
 #include "rocksdb/write_batch.h"
+#include "util/fault_injection_test_env.h"
 #include "util/logging.h"
 #include "util/mock_env.h"
 #include "util/mutexlock.h"
@@ -35,401 +34,6 @@ static const int kValueSize = 1000;
 static const int kMaxNumValues = 2000;
 static const size_t kNumIterations = 3;

-class TestWritableFile;
-class FaultInjectionTestEnv;
-
-namespace {
-
-// Assume a filename, and not a directory name like "/foo/bar/"
-static std::string GetDirName(const std::string filename) {
-  size_t found = filename.find_last_of("/\\");
-  if (found == std::string::npos) {
-    return "";
-  } else {
-    return filename.substr(0, found);
-  }
-}
-
-// Trim the tailing "/" in the end of `str`
-static std::string TrimDirname(const std::string& str) {
-  size_t found = str.find_last_not_of("/");
-  if (found == std::string::npos) {
-    return str;
-  }
-  return str.substr(0, found + 1);
-}
-
-// Return pair <parent directory name, file name> of a full path.
-static std::pair<std::string, std::string> GetDirAndName(
-    const std::string& name) {
-  std::string dirname = GetDirName(name);
-  std::string fname = name.substr(dirname.size() + 1);
-  return std::make_pair(dirname, fname);
-}
-
-// A basic file truncation function suitable for this test.
-Status Truncate(Env* env, const std::string& filename, uint64_t length) {
-  unique_ptr<SequentialFile> orig_file;
-  const EnvOptions options;
-  Status s = env->NewSequentialFile(filename, &orig_file, options);
-  if (!s.ok()) {
-    fprintf(stderr, "Cannot truncate file %s: %s\n", filename.c_str(),
-            s.ToString().c_str());
-    return s;
-  }
-
-  std::unique_ptr<char[]> scratch(new char[length]);
-  rocksdb::Slice result;
-  s = orig_file->Read(length, &result, scratch.get());
-#ifdef OS_WIN
-  orig_file.reset();
-#endif
-  if (s.ok()) {
-    std::string tmp_name = GetDirName(filename) + "/truncate.tmp";
-    unique_ptr<WritableFile> tmp_file;
-    s = env->NewWritableFile(tmp_name, &tmp_file, options);
-    if (s.ok()) {
-      s = tmp_file->Append(result);
-      if (s.ok()) {
-        s = env->RenameFile(tmp_name, filename);
-      } else {
-        fprintf(stderr, "Cannot rename file %s to %s: %s\n", tmp_name.c_str(),
-                filename.c_str(), s.ToString().c_str());
-        env->DeleteFile(tmp_name);
-      }
-    }
-  }
-  if (!s.ok()) {
-    fprintf(stderr, "Cannot truncate file %s: %s\n", filename.c_str(),
-            s.ToString().c_str());
-  }
-
-  return s;
-}
-
-struct FileState {
-  std::string filename_;
-  ssize_t pos_;
-  ssize_t pos_at_last_sync_;
-  ssize_t pos_at_last_flush_;
-
-  explicit FileState(const std::string& filename)
-      : filename_(filename),
-        pos_(-1),
-        pos_at_last_sync_(-1),
-        pos_at_last_flush_(-1) { }
-
-  FileState() : pos_(-1), pos_at_last_sync_(-1), pos_at_last_flush_(-1) {}
-
-  bool IsFullySynced() const { return pos_ <= 0 || pos_ == pos_at_last_sync_; }
-
-  Status DropUnsyncedData(Env* env) const;
-
-  Status DropRandomUnsyncedData(Env* env, Random* rand) const;
-};
-
-}  // anonymous namespace
-
-// A wrapper around WritableFileWriter* file
-// is written to or sync'ed.
-class TestWritableFile : public WritableFile {
- public:
-  explicit TestWritableFile(const std::string& fname,
-                            unique_ptr<WritableFile>&& f,
-                            FaultInjectionTestEnv* env);
-  virtual ~TestWritableFile();
-  virtual Status Append(const Slice& data) override;
-  virtual Status Truncate(uint64_t size) override { return target_->Truncate(size); }
-  virtual Status Close() override;
-  virtual Status Flush() override;
-  virtual Status Sync() override;
-  virtual bool IsSyncThreadSafe() const override { return true; }
-
- private:
-  FileState state_;
-  unique_ptr<WritableFile> target_;
-  bool writable_file_opened_;
-  FaultInjectionTestEnv* env_;
-};
-
-class TestDirectory : public Directory {
- public:
-  explicit TestDirectory(FaultInjectionTestEnv* env, std::string dirname,
-                         Directory* dir)
-      : env_(env), dirname_(dirname), dir_(dir) {}
-  ~TestDirectory() {}
-
-  virtual Status Fsync() override;
-
- private:
-  FaultInjectionTestEnv* env_;
-  std::string dirname_;
-  unique_ptr<Directory> dir_;
-};
-
-class FaultInjectionTestEnv : public EnvWrapper {
- public:
-  explicit FaultInjectionTestEnv(Env* base)
-      : EnvWrapper(base),
-        filesystem_active_(true) {}
-  virtual ~FaultInjectionTestEnv() { }
-
-  Status NewDirectory(const std::string& name,
-                      unique_ptr<Directory>* result) override {
-    unique_ptr<Directory> r;
-    Status s = target()->NewDirectory(name, &r);
-    EXPECT_OK(s);
-    if (!s.ok()) {
-      return s;
-    }
-    result->reset(new TestDirectory(this, TrimDirname(name), r.release()));
-    return Status::OK();
-  }
-
-  Status NewWritableFile(const std::string& fname,
-                         unique_ptr<WritableFile>* result,
-                         const EnvOptions& soptions) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    // Not allow overwriting files
-    Status s = target()->FileExists(fname);
-    if (s.ok()) {
-      return Status::Corruption("File already exists.");
-    } else if (!s.IsNotFound()) {
-      assert(s.IsIOError());
-      return s;
-    }
-    s = target()->NewWritableFile(fname, result, soptions);
-    if (s.ok()) {
-      result->reset(new TestWritableFile(fname, std::move(*result), this));
-      // WritableFileWriter* file is opened
-      // again then it will be truncated - so forget our saved state.
-      UntrackFile(fname);
-      MutexLock l(&mutex_);
-      open_files_.insert(fname);
-      auto dir_and_name = GetDirAndName(fname);
-      auto& list = dir_to_new_files_since_last_sync_[dir_and_name.first];
-      list.insert(dir_and_name.second);
-    }
-    return s;
-  }
-
-  virtual Status DeleteFile(const std::string& f) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    Status s = EnvWrapper::DeleteFile(f);
-    if (!s.ok()) {
-      fprintf(stderr, "Cannot delete file %s: %s\n", f.c_str(),
-              s.ToString().c_str());
-    }
-    EXPECT_OK(s);
-    if (s.ok()) {
-      UntrackFile(f);
-    }
-    return s;
-  }
-
-  virtual Status RenameFile(const std::string& s,
-                            const std::string& t) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    Status ret = EnvWrapper::RenameFile(s, t);
-
-    if (ret.ok()) {
-      MutexLock l(&mutex_);
-      if (db_file_state_.find(s) != db_file_state_.end()) {
-        db_file_state_[t] = db_file_state_[s];
-        db_file_state_.erase(s);
-      }
-
-      auto sdn = GetDirAndName(s);
-      auto tdn = GetDirAndName(t);
-      if (dir_to_new_files_since_last_sync_[sdn.first].erase(sdn.second) != 0) {
-        auto& tlist = dir_to_new_files_since_last_sync_[tdn.first];
-        assert(tlist.find(tdn.second) == tlist.end());
-        tlist.insert(tdn.second);
-      }
-    }
-
-    return ret;
-  }
-
-  void WritableFileClosed(const FileState& state) {
-    MutexLock l(&mutex_);
-    if (open_files_.find(state.filename_) != open_files_.end()) {
-      db_file_state_[state.filename_] = state;
-      open_files_.erase(state.filename_);
-    }
-  }
-
-  // For every file that is not fully synced, make a call to `func` with
-  // FileState of the file as the parameter.
-  Status DropFileData(std::function<Status(Env*, FileState)> func) {
-    Status s;
-    MutexLock l(&mutex_);
-    for (std::map<std::string, FileState>::const_iterator it =
-             db_file_state_.begin();
-         s.ok() && it != db_file_state_.end(); ++it) {
-      const FileState& state = it->second;
-      if (!state.IsFullySynced()) {
-        s = func(target(), state);
-      }
-    }
-    return s;
-  }
-
-  Status DropUnsyncedFileData() {
-    return DropFileData([&](Env* env, const FileState& state) {
-      return state.DropUnsyncedData(env);
-    });
-  }
-
-  Status DropRandomUnsyncedFileData(Random* rnd) {
-    return DropFileData([&](Env* env, const FileState& state) {
-      return state.DropRandomUnsyncedData(env, rnd);
-    });
-  }
-
-  Status DeleteFilesCreatedAfterLastDirSync() {
-    // Because DeleteFile access this container make a copy to avoid deadlock
-    std::map<std::string, std::set<std::string>> map_copy;
-    {
-      MutexLock l(&mutex_);
-      map_copy.insert(dir_to_new_files_since_last_sync_.begin(),
-                      dir_to_new_files_since_last_sync_.end());
-    }
-
-    for (auto& pair : map_copy) {
-      for (std::string name : pair.second) {
-        Status s = DeleteFile(pair.first + "/" + name);
-        if (!s.ok()) {
-          return s;
-        }
-      }
-    }
-    return Status::OK();
-  }
-  void ResetState() {
-    MutexLock l(&mutex_);
-    db_file_state_.clear();
-    dir_to_new_files_since_last_sync_.clear();
-    SetFilesystemActiveNoLock(true);
-  }
-
-  void UntrackFile(const std::string& f) {
-    MutexLock l(&mutex_);
-    auto dir_and_name = GetDirAndName(f);
-    dir_to_new_files_since_last_sync_[dir_and_name.first].erase(
-        dir_and_name.second);
-    db_file_state_.erase(f);
-    open_files_.erase(f);
-  }
-
-  void SyncDir(const std::string& dirname) {
-    MutexLock l(&mutex_);
-    dir_to_new_files_since_last_sync_.erase(dirname);
-  }
-
-  // Setting the filesystem to inactive is the test equivalent to simulating a
-  // system reset. Setting to inactive will freeze our saved filesystem state so
-  // that it will stop being recorded. It can then be reset back to the state at
-  // the time of the reset.
-  bool IsFilesystemActive() {
-    MutexLock l(&mutex_);
-    return filesystem_active_;
-  }
-  void SetFilesystemActiveNoLock(bool active) { filesystem_active_ = active; }
-  void SetFilesystemActive(bool active) {
-    MutexLock l(&mutex_);
-    SetFilesystemActiveNoLock(active);
-  }
-  void AssertNoOpenFile() { ASSERT_TRUE(open_files_.empty()); }
-
- private:
-  port::Mutex mutex_;
-  std::map<std::string, FileState> db_file_state_;
-  std::set<std::string> open_files_;
-  std::unordered_map<std::string, std::set<std::string>>
-      dir_to_new_files_since_last_sync_;
-  bool filesystem_active_;  // Record flushes, syncs, writes
-};
-
-Status FileState::DropUnsyncedData(Env* env) const {
-  ssize_t sync_pos = pos_at_last_sync_ == -1 ? 0 : pos_at_last_sync_;
-  return Truncate(env, filename_, sync_pos);
-}
-
-Status FileState::DropRandomUnsyncedData(Env* env, Random* rand) const {
-  ssize_t sync_pos = pos_at_last_sync_ == -1 ? 0 : pos_at_last_sync_;
-  assert(pos_ >= sync_pos);
-  int range = static_cast<int>(pos_ - sync_pos);
-  uint64_t truncated_size =
-      static_cast<uint64_t>(sync_pos) + rand->Uniform(range);
-  return Truncate(env, filename_, truncated_size);
-}
-
-Status TestDirectory::Fsync() {
-  env_->SyncDir(dirname_);
-  return dir_->Fsync();
-}
-
-TestWritableFile::TestWritableFile(const std::string& fname,
-                                   unique_ptr<WritableFile>&& f,
-                                   FaultInjectionTestEnv* env)
-      : state_(fname),
-        target_(std::move(f)),
-        writable_file_opened_(true),
-        env_(env) {
-  assert(target_ != nullptr);
-  state_.pos_ = 0;
-}
-
-TestWritableFile::~TestWritableFile() {
-  if (writable_file_opened_) {
-    Close();
-  }
-}
-
-Status TestWritableFile::Append(const Slice& data) {
-  if (!env_->IsFilesystemActive()) {
-    return Status::Corruption("Not Active");
-  }
-  Status s = target_->Append(data);
-  if (s.ok()) {
-    state_.pos_ += data.size();
-  }
-  return s;
-}
-
-Status TestWritableFile::Close() {
-  writable_file_opened_ = false;
-  Status s = target_->Close();
-  if (s.ok()) {
-    env_->WritableFileClosed(state_);
-  }
-  return s;
-}
-
-Status TestWritableFile::Flush() {
-  Status s = target_->Flush();
-  if (s.ok() && env_->IsFilesystemActive()) {
-    state_.pos_at_last_flush_ = state_.pos_;
-  }
-  return s;
-}
-
-Status TestWritableFile::Sync() {
-  if (!env_->IsFilesystemActive()) {
-    return Status::OK();
-  }
-  // No need to actual sync.
-  state_.pos_at_last_sync_ = state_.pos_;
-  return Status::OK();
-}
-
 class FaultInjectionTest : public testing::Test,
                           public testing::WithParamInterface<bool> {
 protected:
@@ -645,18 +249,17 @@ class FaultInjectionTest : public testing::Test,
    return test::RandomString(&r, kValueSize, storage);
  }

-  Status OpenDB() {
-    delete db_;
-    db_ = NULL;
-    env_->ResetState();
-    return DB::Open(options_, dbname_, &db_);
-  }
-
  void CloseDB() {
    delete db_;
    db_ = NULL;
  }

+  Status OpenDB() {
+    CloseDB();
+    env_->ResetState();
+    return DB::Open(options_, dbname_, &db_);
+  }
+
  void DeleteAllData() {
    Iterator* iter = db_->NewIterator(ReadOptions());
    WriteOptions options;
@@ -733,7 +336,7 @@ class FaultInjectionTest : public testing::Test,
  }

  void WaitCompactionFinish() {
-    static_cast<DBImpl*>(db_)->TEST_WaitForCompact();
+    static_cast<DBImpl*>(db_->GetRootDB())->TEST_WaitForCompact();
    ASSERT_OK(db_->Put(WriteOptions(), "", ""));
  }
 };
@@ -784,6 +387,7 @@ TEST_P(FaultInjectionTest, WriteOptionSyncTest) {
  // Block the job queue to prevent flush job from running.
  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
                 Env::Priority::HIGH);
+  sleeping_task_low.WaitUntilSleeping();

  WriteOptions write_options;
  write_options.sync = false;
@@ -867,6 +471,7 @@ TEST_P(FaultInjectionTest, ManualLogSyncTest) {
  // Block the job queue to prevent flush job from running.
  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
                 Env::Priority::HIGH);
+  sleeping_task_low.WaitUntilSleeping();

  WriteOptions write_options;
  write_options.sync = false;
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -26,9 +26,9 @@ struct FileLevel;
 // The file tree structure in Version is prebuilt and the range of each file
 // is known. On Version::Get(), it uses binary search to find a potential file
 // and then check if a target key can be found in the file by comparing the key
-// to each file's smallest and largest key. The results of these comparisions
+// to each file's smallest and largest key. The results of these comparisons
 // can be reused beyond checking if a key falls into a file's range.
-// With some pre-calculated knowledge, each key comparision that has been done
+// With some pre-calculated knowledge, each key comparison that has been done
 // can serve as a hint to narrow down further searches: if a key compared to
 // be smaller than a file's smallest or largest, that comparison can be used
 // to find out the right bound of next binary search. Similarly, if a key
@@ -48,7 +48,7 @@ class FileIndexer {
  size_t LevelIndexSize(size_t level) const;

  // Return a file index range in the next level to search for a key based on
-  // smallest and largest key comparision for the current file specified by
+  // smallest and largest key comparison for the current file specified by
  // level and file_index. When *left_index < *right_index, both index should
  // be valid and fit in the vector size.
  void GetNextLevelIndex(const size_t level, const size_t file_index,
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -1,4 +1,4 @@
-//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
+//  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.
@@ -62,10 +62,11 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                   InstrumentedMutex* db_mutex,
                   std::atomic<bool>* shutting_down,
                   std::vector<SequenceNumber> existing_snapshots,
+                   SequenceNumber earliest_write_conflict_snapshot,
                   JobContext* job_context, LogBuffer* log_buffer,
                   Directory* db_directory, Directory* output_file_directory,
                   CompressionType output_compression, Statistics* stats,
-                   EventLogger* event_logger)
+                   EventLogger* event_logger, bool measure_io_stats)
    : dbname_(dbname),
      cfd_(cfd),
      db_options_(db_options),
@@ -75,13 +76,15 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
      db_mutex_(db_mutex),
      shutting_down_(shutting_down),
      existing_snapshots_(std::move(existing_snapshots)),
+      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      job_context_(job_context),
      log_buffer_(log_buffer),
      db_directory_(db_directory),
      output_file_directory_(output_file_directory),
      output_compression_(output_compression),
      stats_(stats),
-      event_logger_(event_logger) {
+      event_logger_(event_logger),
+      measure_io_stats_(measure_io_stats) {
  // Update the thread status to indicate flush.
  ReportStartedFlush();
  TEST_SYNC_POINT("FlushJob::FlushJob()");
@@ -92,7 +95,8 @@ FlushJob::~FlushJob() {
 }

 void FlushJob::ReportStartedFlush() {
-  ThreadStatusUtil::SetColumnFamily(cfd_);
+  ThreadStatusUtil::SetColumnFamily(cfd_, cfd_->ioptions()->env,
+                                    cfd_->options()->enable_thread_tracking);
  ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_FLUSH);
  ThreadStatusUtil::SetThreadOperationProperty(
      ThreadStatus::COMPACTION_JOB_ID,
@@ -111,13 +115,30 @@ void FlushJob::ReportFlushInputSize(const autovector<MemTable*>& mems) {
 }

 void FlushJob::RecordFlushIOStats() {
-  ThreadStatusUtil::SetThreadOperationProperty(
+  RecordTick(stats_, FLUSH_WRITE_BYTES, IOSTATS(bytes_written));
+  ThreadStatusUtil::IncreaseThreadOperationProperty(
      ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
+  IOSTATS_RESET(bytes_written);
 }

 Status FlushJob::Run(FileMetaData* file_meta) {
  AutoThreadOperationStageUpdater stage_run(
      ThreadStatus::STAGE_FLUSH_RUN);
+  // I/O measurement variables
+  PerfLevel prev_perf_level = PerfLevel::kEnableTime;
+  uint64_t prev_write_nanos = 0;
+  uint64_t prev_fsync_nanos = 0;
+  uint64_t prev_range_sync_nanos = 0;
+  uint64_t prev_prepare_write_nanos = 0;
+  if (measure_io_stats_) {
+    prev_perf_level = GetPerfLevel();
+    SetPerfLevel(PerfLevel::kEnableTime);
+    prev_write_nanos = IOSTATS(write_nanos);
+    prev_fsync_nanos = IOSTATS(fsync_nanos);
+    prev_range_sync_nanos = IOSTATS(range_sync_nanos);
+    prev_prepare_write_nanos = IOSTATS(prepare_write_nanos);
+  }
+
  // Save the contents of the earliest memtable as a new Table
  FileMetaData meta;
  autovector<MemTable*> mems;
@@ -176,6 +197,19 @@ Status FlushJob::Run(FileMetaData* file_meta) {
    stream << vstorage->NumLevelFiles(level);
  }
  stream.EndArray();
+  stream << "immutable_memtables" << cfd_->imm()->NumNotFlushed();
+
+  if (measure_io_stats_) {
+    if (prev_perf_level != PerfLevel::kEnableTime) {
+      SetPerfLevel(prev_perf_level);
+    }
+    stream << "file_write_nanos" << (IOSTATS(write_nanos) - prev_write_nanos);
+    stream << "file_range_sync_nanos"
+           << (IOSTATS(range_sync_nanos) - prev_range_sync_nanos);
+    stream << "file_fsync_nanos" << (IOSTATS(fsync_nanos) - prev_fsync_nanos);
+    stream << "file_prepare_write_nanos"
+           << (IOSTATS(prepare_write_nanos) - prev_prepare_write_nanos);
+  }

  return s;
 }
@@ -220,7 +254,6 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
                         << total_num_deletes << "memory_usage"
                         << total_memory_usage;

-    TableFileCreationInfo info;
    {
      ScopedArenaIterator iter(
          NewMergingIterator(&cfd_->internal_comparator(), &memtables[0],
@@ -231,15 +264,16 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,

      TEST_SYNC_POINT_CALLBACK("FlushJob::WriteLevel0Table:output_compression",
                               &output_compression_);
-      s = BuildTable(dbname_, db_options_.env, *cfd_->ioptions(), env_options_,
-                     cfd_->table_cache(), iter.get(), meta,
-                     cfd_->internal_comparator(),
-                     cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
-                     existing_snapshots_, output_compression_,
-                     cfd_->ioptions()->compression_opts,
-                     mutable_cf_options_.paranoid_file_checks,
-                     cfd_->internal_stats(), Env::IO_HIGH, &table_properties_);
-      info.table_properties = table_properties_;
+      s = BuildTable(
+          dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
+          env_options_, cfd_->table_cache(), iter.get(), meta,
+          cfd_->internal_comparator(), cfd_->int_tbl_prop_collector_factories(),
+          cfd_->GetID(), cfd_->GetName(), existing_snapshots_,
+          earliest_write_conflict_snapshot_, output_compression_,
+          cfd_->ioptions()->compression_opts,
+          mutable_cf_options_.paranoid_file_checks, cfd_->internal_stats(),
+          TableFileCreationReason::kFlush, event_logger_, job_context_->job_id,
+          Env::IO_HIGH, &table_properties_, 0 /* level */);
      LogFlush(db_options_.info_log);
    }
    Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
@@ -250,24 +284,10 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
        meta->fd.GetFileSize(), s.ToString().c_str(),
        meta->marked_for_compaction ? " (needs compaction)" : "");

-    // output to event logger
-    if (s.ok()) {
-      info.db_name = dbname_;
-      info.cf_name = cfd_->GetName();
-      info.file_path = TableFileName(db_options_.db_paths,
-                                     meta->fd.GetNumber(),
-                                     meta->fd.GetPathId());
-      info.file_size = meta->fd.GetFileSize();
-      info.job_id = job_context_->job_id;
-      EventHelpers::LogAndNotifyTableFileCreation(
-          event_logger_, db_options_.listeners,
-          meta->fd, info);
-      TEST_SYNC_POINT("FlushJob::LogAndNotifyTableFileCreation()");
-    }
-
    if (!db_options_.disableDataSync && output_file_directory_ != nullptr) {
      output_file_directory_->Fsync();
    }
+    TEST_SYNC_POINT("FlushJob::WriteLevel0Table");
    db_mutex_->Lock();
  }
  base->Unref();
@@ -289,13 +309,14 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
                  meta->marked_for_compaction);
  }

+  // Note that here we treat flush as level 0 compaction in internal stats
  InternalStats::CompactionStats stats(1);
  stats.micros = db_options_.env->NowMicros() - start_micros;
  stats.bytes_written = meta->fd.GetFileSize();
  cfd_->internal_stats()->AddCompactionStats(0 /* level */, stats);
  cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
                                     meta->fd.GetFileSize());
-  RecordTick(stats_, COMPACT_WRITE_BYTES, meta->fd.GetFileSize());
+  RecordFlushIOStats();
  return s;
 }