Fix test name typo in PlainTableDBTest

PlainTable should avoid copying Get() results from immortal source. (#4924 )
Summary: https://github.com/facebook/rocksdb/pull/4053 avoids memcopy for Get() results if files are immortable (read-only DB, max_open_files=-1) and the file is ammaped. The same optimization is being applied to PlainTable here. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4924 Differential Revision: D13827749 Pulled By: siying fbshipit-source-id: 1f2cbfc530b40ce08ccd53f95f6e78de4d1c2f96
2026-11-07 03:22:34 -05:00 · 2019-01-25 17:31:06 -08:00 · 2019-01-25 17:12:19 -08:00 · 2019-01-24 11:25:27 -08:00 · 2019-01-24 10:34:30 -08:00 · 2019-01-23 21:46:10 -08:00
241 changed files with 6555 additions and 12724 deletions
@@ -71,10 +71,7 @@ install:
      CC=gcc-8 && CXX=g++-8;
    fi
  - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
-      mkdir cmake-dist && curl --silent --fail --show-error --location https://github.com/Kitware/CMake/releases/download/v3.14.5/cmake-3.14.5-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
-    fi
-  - if [[ "${JOB_NAME}" == java_test ]]; then
-      java -version && echo "JAVA_HOME=${JAVA_HOME}";
+      mkdir cmake-dist && curl -sfSL https://cmake.org/files/v3.8/cmake-3.8.1-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
    fi

 before_script:
@@ -94,7 +94,13 @@ else()
  if(WITH_ZLIB)
    find_package(zlib REQUIRED)
    add_definitions(-DZLIB)
-    include_directories(${ZLIB_INCLUDE_DIR})
+    if(ZLIB_INCLUDE_DIRS)
+      # CMake 3
+      include_directories(${ZLIB_INCLUDE_DIRS})
+    else()
+      # CMake 2
+      include_directories(${ZLIB_INCLUDE_DIR})
+    endif()
    list(APPEND THIRDPARTY_LIBS ${ZLIB_LIBRARIES})
  endif()

@@ -596,6 +602,7 @@ set(SOURCES
        util/comparator.cc
        util/compression_context_cache.cc
        util/concurrent_arena.cc
+        util/concurrent_task_limiter_impl.cc
        util/crc32c.cc
        util/delete_scheduler.cc
        util/dynamic_bloom.cc
@@ -636,18 +643,10 @@ set(SOURCES
        utilities/cassandra/format.cc
        utilities/cassandra/merge_operator.cc
        utilities/checkpoint/checkpoint_impl.cc
-        utilities/col_buf_decoder.cc
-        utilities/col_buf_encoder.cc
-        utilities/column_aware_encoding_util.cc
        utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc
-        utilities/date_tiered/date_tiered_db_impl.cc
        utilities/debug.cc
-        utilities/document/document_db.cc
-        utilities/document/json_document.cc
-        utilities/document/json_document_builder.cc
        utilities/env_mirror.cc
        utilities/env_timed.cc
-        utilities/geodb/geodb_impl.cc
        utilities/leveldb_options/leveldb_options.cc
        utilities/lua/rocks_lua_compaction_filter.cc
        utilities/memory/memory_util.cc
@@ -664,9 +663,7 @@ set(SOURCES
        utilities/persistent_cache/block_cache_tier_metadata.cc
        utilities/persistent_cache/persistent_cache_tier.cc
        utilities/persistent_cache/volatile_tier_impl.cc
-        utilities/redis/redis_lists.cc
        utilities/simulator_cache/sim_cache.cc
-        utilities/spatialdb/spatial_db.cc
        utilities/table_properties_collectors/compact_on_deletion_collector.cc
        utilities/trace/file_trace_reader_writer.cc
        utilities/transactions/optimistic_transaction_db_impl.cc
@@ -822,6 +819,7 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
    TARGETS ${ROCKSDB_SHARED_LIB}
    EXPORT RocksDBTargets
    COMPONENT runtime
+    ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}"
    RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}"
    LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
    INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
@@ -967,11 +965,6 @@ if(WITH_TESTS)
        utilities/cassandra/cassandra_row_merge_test.cc
        utilities/cassandra/cassandra_serialize_test.cc
        utilities/checkpoint/checkpoint_test.cc
-        utilities/column_aware_encoding_test.cc
-        utilities/date_tiered/date_tiered_test.cc
-        utilities/document/document_db_test.cc
-        utilities/document/json_document_test.cc
-        utilities/geodb/geodb_test.cc
        utilities/lua/rocks_lua_test.cc
        utilities/memory/memory_test.cc
        utilities/merge_operators/string_append/stringappend_test.cc
@@ -980,8 +973,6 @@ if(WITH_TESTS)
        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/simulator_cache/sim_cache_test.cc
        utilities/table_properties_collectors/compact_on_deletion_collector_test.cc
        utilities/transactions/optimistic_transaction_test.cc
@@ -1001,7 +992,6 @@ if(WITH_TESTS)
    db/range_del_aggregator_bench.cc
    tools/db_bench.cc
    table/table_reader_bench.cc
-    utilities/column_aware_encoding_exp.cc
    utilities/persistent_cache/hash_table_bench.cc)
  add_library(testharness OBJECT util/testharness.cc)
  foreach(sourcefile ${BENCHMARKS})
@@ -1,18 +1,32 @@
 # Rocksdb Change Log
-## 5.18.4 (3/3/2020)
-* Various fixes for ARM64 support (#6250)
-* Fix JEMALLOC_CXX_THROW macro missing from older Jemalloc versions, causing build failures on some platforms.
-
-## 5.18.3 (2/11/2019)
-### Bug Fixes
-* Fix possible LSM corruption when both range deletions and subcompactions are used. The symptom of this corruption is L1+ files overlapping in the user key space.
-
-## 5.18.2 (01/31/2019)
+## Unreleased
+### New Features
+* Enabled checkpoint on readonly db (DBImplReadOnly).
+* Make DB ignore dropped column families while committing results of atomic flush.
+* RocksDB may choose to preopen some files even if options.max_open_files != -1. This may make DB open slightly longer.
+* For users of dictionary compression with ZSTD v0.7.0+, we now reuse the same digested dictionary when compressing each of an SST file's data blocks for faster compression speeds.
+* For all users of dictionary compression who set `cache_index_and_filter_blocks == true`, we now store dictionary data used for decompression in the block cache for better control over memory usage. For users of ZSTD v1.1.4+ who compile with -DZSTD_STATIC_LINKING_ONLY, this includes a digested dictionary, which is used to increase decompression speed.

 ### Public API Change
+* CompactionPri = kMinOverlappingRatio also uses compensated file size, which boosts file with lots of tombstones to be compacted first.
+* Transaction::GetForUpdate is extended with a do_validate parameter with default value of true. If false it skips validating the snapshot before doing the read. Similarly ::Merge, ::Put, ::Delete, and ::SingleDelete are extended with assume_tracked with default value of false. If true it indicates that call is assumed to be after a ::GetForUpdate.
+* `TableProperties::num_entries` and `TableProperties::num_deletions` now also account for number of range tombstones.
+* Remove geodb, spatial_db, document_db, json_document, date_tiered_db, and redis_lists.
+* With "ldb ----try_load_options", when wal_dir specified by the option file doesn't exist, ignore it.
 * Change time resolution in FileOperationInfo.
 * Deleting Blob files also go through SStFileManager.

+### Bug Fixes
+* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
+* Fix a memory leak when files with range tombstones are read in mmap mode and block cache is enabled
+* Fix handling of corrupt range tombstone blocks such that corruptions cannot cause deleted keys to reappear
+* Lock free MultiGet
+* Fix incorrect `NotFound` point lookup result when querying the endpoint of a file that has been extended by a range tombstone.
+* Fix with pipelined write, write leaders's callback failure lead to the whole write group fail.
+
+### Change Default Options
+* Change options.compaction_pri's default to kMinOverlappingRatio
+
 ## 5.18.0 (11/30/2018)
 ### New Features
 * Introduced `JemallocNodumpAllocator` memory allocator. When being use, block cache will be excluded from core dump.
@@ -24,7 +38,6 @@
 * Add xxhash64 checksum support
 * Introduced `MemoryAllocator`, which lets the user specify custom memory allocator for block based table.
 * Improved `DeleteRange` to prevent read performance degradation. The feature is no longer marked as experimental.
-* Enabled checkpoint on readonly db (DBImplReadOnly).

 ### Public API Change
 * `DBOptions::use_direct_reads` now affects reads issued by `BackupEngine` on the database's SSTs.
@@ -41,8 +54,6 @@
 * Fixed Get correctness bug in the presence of range tombstones where merge operands covered by a range tombstone always result in NotFound.
 * Start populating `NO_FILE_CLOSES` ticker statistic, which was always zero previously.
 * The default value of NewBloomFilterPolicy()'s argument use_block_based_builder is changed to false. Note that this new default may cause large temp memory usage when building very large SST files.
-* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
-* Make DB ignore dropped column families while committing results of atomic flush.

 ## 5.17.0 (10/05/2018)
 ### Public API Change
@@ -137,12 +137,6 @@ CFLAGS +=  -DHAVE_POWER8
 HAVE_POWER8=1
 endif

-ifeq (,$(shell $(CXX) -fsyntax-only -march=armv8-a+crc -xc /dev/null 2>&1))
-CXXFLAGS += -march=armv8-a+crc
-CFLAGS += -march=armv8-a+crc
-ARMCRC_SOURCE=1
-endif
-
 # if we're compiling for release, compile without debug code (-DNDEBUG)
 ifeq ($(DEBUG_LEVEL),0)
 OPT += -DNDEBUG
@@ -204,7 +198,7 @@ include make_config.mk
 CLEAN_FILES += make_config.mk

 missing_make_config_paths := $(shell				\
-	grep "\/\S*" -o $(CURDIR)/make_config.mk | 		\
+	grep "\./\S*\|/\S*" -o $(CURDIR)/make_config.mk | 	\
 	while read path;					\
 		do [ -e $$path ] || echo $$path; 		\
 	done | sort | uniq)
@@ -409,7 +403,7 @@ BENCHTOOLOBJECTS = $(BENCH_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)

 ANALYZETOOLOBJECTS = $(ANALYZER_LIB_SOURCES:.cc=.o)

-EXPOBJECTS = $(EXP_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
+EXPOBJECTS = $(LIBOBJECTS) $(TESTUTIL)

 TESTS = \
 	db_basic_test \
@@ -488,7 +482,6 @@ TESTS = \
 	merger_test \
 	util_merge_operators_test \
 	options_file_test \
-	redis_test \
 	reduce_levels_test \
 	plain_table_db_test \
 	comparator_db_test \
@@ -502,12 +495,8 @@ TESTS = \
 	cassandra_row_merge_test \
 	cassandra_serialize_test \
 	ttl_test \
-	date_tiered_test \
 	backupable_db_test \
-	document_db_test \
-	json_document_test \
 	sim_cache_test \
-	spatial_db_test \
 	version_edit_test \
 	version_set_test \
 	compaction_picker_test \
@@ -519,7 +508,6 @@ TESTS = \
 	deletefile_test \
 	obsolete_files_test \
 	table_test \
-	geodb_test \
 	delete_scheduler_test \
 	options_test \
 	options_settable_test \
@@ -536,7 +524,6 @@ TESTS = \
 	compaction_job_test \
 	thread_list_test \
 	sst_dump_test \
-	column_aware_encoding_test \
 	compact_files_test \
 	optimistic_transaction_test \
 	write_callback_test \
@@ -610,7 +597,7 @@ 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 column_aware_encoding_exp persistent_cache_bench range_del_aggregator_bench
+BENCHMARKS = db_bench table_reader_bench cache_bench memtablerep_bench persistent_cache_bench range_del_aggregator_bench

 # if user didn't config LIBNAME, set the default
 ifeq ($(LIBNAME),)
@@ -1159,9 +1146,6 @@ cassandra_row_merge_test: utilities/cassandra/cassandra_row_merge_test.o utiliti
 cassandra_serialize_test: utilities/cassandra/cassandra_serialize_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-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)

@@ -1300,18 +1284,9 @@ backupable_db_test: utilities/backupable/backupable_db_test.o $(LIBOBJECTS) $(TE
 checkpoint_test: utilities/checkpoint/checkpoint_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-document_db_test: utilities/document/document_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
-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)

@@ -1329,9 +1304,6 @@ object_registry_test: utilities/object_registry_test.o $(LIBOBJECTS) $(TESTHARNE
 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-date_tiered_test: utilities/date_tiered/date_tiered_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 write_batch_with_index_test: utilities/write_batch_with_index/write_batch_with_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1458,9 +1430,6 @@ deletefile_test: db/deletefile_test.o $(LIBOBJECTS) $(TESTHARNESS)
 obsolete_files_test: db/obsolete_files_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-geodb_test: utilities/geodb/geodb_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 rocksdb_dump: tools/dump/rocksdb_dump.o $(LIBOBJECTS)
 	$(AM_LINK)

@@ -1509,9 +1478,6 @@ timer_queue_test: util/timer_queue_test.o $(LIBOBJECTS) $(TESTHARNESS)
 sst_dump_test: tools/sst_dump_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

-column_aware_encoding_test: utilities/column_aware_encoding_test.o $(TESTHARNESS) $(EXPOBJECTS)
-	$(AM_LINK)
-
 optimistic_transaction_test: utilities/transactions/optimistic_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1551,9 +1517,6 @@ sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 blob_dump: tools/blob_dump.o $(LIBOBJECTS)
 	$(AM_LINK)

-column_aware_encoding_exp: utilities/column_aware_encoding_exp.o $(EXPOBJECTS)
-	$(AM_LINK)
-
 repair_test: db/repair_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1641,7 +1604,7 @@ JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
 ifeq ($(PLATFORM), OS_SOLARIS)
 	ARCH := $(shell isainfo -b)
 else ifeq ($(PLATFORM), OS_OPENBSD)
-	ifneq (,$(filter amd64 ppc64 ppc64le arm64 aarch64 sparc64, $(MACHINE)))
+	ifneq (,$(filter $(MACHINE), amd64 arm64 sparc64))
 		ARCH := 64
 	else
 		ARCH := 32
@@ -1650,10 +1613,10 @@ else
 	ARCH := $(shell getconf LONG_BIT)
 endif

-ifneq (,$(filter ppc% arm64 aarch64 sparc64, $(MACHINE)))
-	ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE).so
+ifeq (,$(findstring ppc,$(MACHINE)))
+        ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
 else
-	ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
+        ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE).so
 endif
 ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux$(ARCH).jar
 ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
@@ -1666,15 +1629,15 @@ ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.6
 BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-BZIP2_DOWNLOAD_BASE ?= https://downloads.sourceforge.net/project/bzip2
-SNAPPY_VER ?= 1.1.8
-SNAPPY_SHA256 ?= 16b677f07832a612b0836178db7f374e414f94657c138e6993cbfc5dcc58651f
+BZIP2_DOWNLOAD_BASE ?= https://web.archive.org/web/20180624184835/http://www.bzip.org
+SNAPPY_VER ?= 1.1.7
+SNAPPY_SHA256 ?= 3dfa02e873ff51a11ee02b9ca391807f0c8ea0529a4924afa645fbf97163f9d4
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
-LZ4_VER ?= 1.9.2
-LZ4_SHA256 ?= 658ba6191fa44c92280d4aa2c271b0f4fbc0e34d249578dd05e50e76d0e5efcc
+LZ4_VER ?= 1.8.3
+LZ4_SHA256 ?= 33af5936ac06536805f9745e0b6d61da606a1f8b4cc5c04dd3cbaca3b9b4fc43
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.4.4
-ZSTD_SHA256 ?= a364f5162c7d1a455cc915e8e3cf5f4bd8b75d09bc0f53965b0c9ca1383c52c8
+ZSTD_VER ?= 1.3.7
+ZSTD_SHA256 ?= 5dd1e90eb16c25425880c8a91327f63de22891ffed082fcc17e5ae84fce0d5fb
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
 CURL_SSL_OPTS ?= --tlsv1

@@ -1714,7 +1677,7 @@ endif
 libz.a:
 	-rm -rf zlib-$(ZLIB_VER)
 ifeq (,$(wildcard ./zlib-$(ZLIB_VER).tar.gz))
-	curl --fail --output zlib-$(ZLIB_VER).tar.gz --location ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+	curl --output zlib-$(ZLIB_VER).tar.gz -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
 endif
 	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
@@ -1728,7 +1691,7 @@ endif
 libbz2.a:
 	-rm -rf bzip2-$(BZIP2_VER)
 ifeq (,$(wildcard ./bzip2-$(BZIP2_VER).tar.gz))
-	curl --fail --output bzip2-$(BZIP2_VER).tar.gz --location ${CURL_SSL_OPTS} ${BZIP2_DOWNLOAD_BASE}/bzip2-$(BZIP2_VER).tar.gz
+	curl --output bzip2-$(BZIP2_VER).tar.gz -L ${BZIP2_DOWNLOAD_BASE}/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
 endif
 	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
@@ -1742,7 +1705,7 @@ endif
 libsnappy.a:
 	-rm -rf snappy-$(SNAPPY_VER)
 ifeq (,$(wildcard ./snappy-$(SNAPPY_VER).tar.gz))
-	curl --fail --output snappy-$(SNAPPY_VER).tar.gz --location ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
+	curl --output snappy-$(SNAPPY_VER).tar.gz -L ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
 endif
 	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
@@ -1751,13 +1714,13 @@ endif
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
 	mkdir snappy-$(SNAPPY_VER)/build
-	cd snappy-$(SNAPPY_VER)/build && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' cmake -DCMAKE_POSITION_INDEPENDENT_CODE=ON .. && $(MAKE) ${SNAPPY_MAKE_TARGET}
+	cd snappy-$(SNAPPY_VER)/build && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' cmake .. && $(MAKE) ${SNAPPY_MAKE_TARGET}
 	cp snappy-$(SNAPPY_VER)/build/libsnappy.a .

 liblz4.a:
 	-rm -rf lz4-$(LZ4_VER)
 ifeq (,$(wildcard ./lz4-$(LZ4_VER).tar.gz))
-	curl --fail --output lz4-$(LZ4_VER).tar.gz --location ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
+	curl --output lz4-$(LZ4_VER).tar.gz -L ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
 endif
 	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
@@ -1771,7 +1734,7 @@ endif
 libzstd.a:
 	-rm -rf zstd-$(ZSTD_VER)
 ifeq (,$(wildcard ./zstd-$(ZSTD_VER).tar.gz))
-	curl --fail --output zstd-$(ZSTD_VER).tar.gz --location ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
+	curl --output zstd-$(ZSTD_VER).tar.gz -L ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
 endif
 	ZSTD_SHA256_ACTUAL=`$(SHA256_CMD) zstd-$(ZSTD_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZSTD_SHA256)" != "$$ZSTD_SHA256_ACTUAL" ]; then \
@@ -1863,14 +1826,6 @@ rocksdbjavastaticdockerppc64le:
 	fi
 	docker start -a rocksdb_linux_ppc64le-be

-rocksdbjavastaticdockerarm64v8:
-	mkdir -p java/target
-	DOCKER_LINUX_ARM64V8_CONTAINER=`docker ps -aqf name=rocksdb_linux_arm64v8-be`; \
-	if [ -z "$$DOCKER_LINUX_ARM64V8_CONTAINER" ]; then \
-		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_arm64v8-be evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
-	fi
-	docker start -a rocksdb_linux_arm64v8-be
-
 rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral

 rocksdbjavastaticpublishdocker: rocksdbjavastaticreleasedocker rocksdbjavastaticpublishcentral
@@ -1949,7 +1904,8 @@ commit_prereq: build_tools/rocksdb-lego-determinator \
 ifeq ($(PLATFORM), IOS)
 # For iOS, create universal object files to be used on both the simulator and
 # a device.
-PLATFORMSROOT=/Applications/Xcode.app/Contents/Developer/Platforms
+XCODEROOT=$(shell xcode-select -print-path)
+PLATFORMSROOT=$(XCODEROOT)/Platforms
 SIMULATORROOT=$(PLATFORMSROOT)/iPhoneSimulator.platform/Developer
 DEVICEROOT=$(PLATFORMSROOT)/iPhoneOS.platform/Developer
 IOSVERSION=$(shell defaults read $(PLATFORMSROOT)/iPhoneOS.platform/version CFBundleShortVersionString)
@@ -1986,7 +1942,7 @@ endif
 #  	Source files dependencies detection
 # ---------------------------------------------------------------------------

-all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(EXP_LIB_SOURCES) $(ANALYZER_LIB_SOURCES)
+all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(ANALYZER_LIB_SOURCES)
 DEPFILES = $(all_sources:.cc=.cc.d)

 # Add proper dependency support so changing a .h file forces a .cc file to
@@ -1,12 +1,10 @@
 load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+load("@fbcode_macros//build_defs:cpp_library.bzl", "cpp_library")
+load(":defs.bzl", "test_binary")

 REPO_PATH = package_name() + "/"

-BUCK_BINS = "buck-out/gen/" + REPO_PATH
-
-TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
-
-rocksdb_compiler_flags = [
+ROCKSDB_COMPILER_FLAGS = [
    "-fno-builtin-memcmp",
    "-DROCKSDB_PLATFORM_POSIX",
    "-DROCKSDB_LIB_IO_POSIX",
@@ -22,6 +20,7 @@ rocksdb_compiler_flags = [
    "-DBZIP2",
    "-DLZ4",
    "-DZSTD",
+    "-DZSTD_STATIC_LINKING_ONLY",
    "-DGFLAGS=gflags",
    "-DNUMA",
    "-DTBB",
@@ -33,7 +32,7 @@ rocksdb_compiler_flags = [
    "-Wnarrowing",
 ]

-rocksdb_external_deps = [
+ROCKSDB_EXTERNAL_DEPS = [
    ("bzip2", None, "bz2"),
    ("snappy", None, "snappy"),
    ("zlib", None, "z"),
@@ -45,13 +44,13 @@ rocksdb_external_deps = [
    ("googletest", None, "gtest"),
 ]

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

-rocksdb_arch_preprocessor_flags = {
+ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
    "x86_64": [
        "-DHAVE_SSE42",
        "-DHAVE_PCLMUL",
@@ -64,16 +63,15 @@ is_opt_mode = build_mode.startswith("opt")

 # -DNDEBUG is added by default in opt mode in fbcode. But adding it twice
 # doesn't harm and avoid forgetting to add it.
-if is_opt_mode:
-    rocksdb_compiler_flags.append("-DNDEBUG")
+ROCKSDB_COMPILER_FLAGS += (["-DNDEBUG"] if is_opt_mode else [])

 sanitizer = read_config("fbcode", "sanitizer")

 # Do not enable jemalloc if sanitizer presents. RocksDB will further detect
 # whether the binary is linked with jemalloc at runtime.
-if sanitizer == "":
-    rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
-    rocksdb_external_deps.append(("jemalloc", None, "headers"))
+ROCKSDB_COMPILER_FLAGS += (["-DROCKSDB_JEMALLOC"] if sanitizer == "" else [])
+
+ROCKSDB_EXTERNAL_DEPS += ([("jemalloc", None, "headers")] if sanitizer == "" else [])

 cpp_library(
    name = "rocksdb_lib",
@@ -217,6 +215,7 @@ cpp_library(
        "util/comparator.cc",
        "util/compression_context_cache.cc",
        "util/concurrent_arena.cc",
+        "util/concurrent_task_limiter_impl.cc",
        "util/crc32c.cc",
        "util/delete_scheduler.cc",
        "util/dynamic_bloom.cc",
@@ -258,14 +257,9 @@ cpp_library(
        "utilities/checkpoint/checkpoint_impl.cc",
        "utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc",
        "utilities/convenience/info_log_finder.cc",
-        "utilities/date_tiered/date_tiered_db_impl.cc",
        "utilities/debug.cc",
-        "utilities/document/document_db.cc",
-        "utilities/document/json_document.cc",
-        "utilities/document/json_document_builder.cc",
        "utilities/env_mirror.cc",
        "utilities/env_timed.cc",
-        "utilities/geodb/geodb_impl.cc",
        "utilities/leveldb_options/leveldb_options.cc",
        "utilities/lua/rocks_lua_compaction_filter.cc",
        "utilities/memory/memory_util.cc",
@@ -282,9 +276,7 @@ cpp_library(
        "utilities/persistent_cache/block_cache_tier_metadata.cc",
        "utilities/persistent_cache/persistent_cache_tier.cc",
        "utilities/persistent_cache/volatile_tier_impl.cc",
-        "utilities/redis/redis_lists.cc",
        "utilities/simulator_cache/sim_cache.cc",
-        "utilities/spatialdb/spatial_db.cc",
        "utilities/table_properties_collectors/compact_on_deletion_collector.cc",
        "utilities/trace/file_trace_reader_writer.cc",
        "utilities/transactions/optimistic_transaction.cc",
@@ -305,11 +297,11 @@ cpp_library(
        "utilities/write_batch_with_index/write_batch_with_index_internal.cc",
    ],
    auto_headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

 cpp_library(
@@ -322,16 +314,13 @@ cpp_library(
        "util/testharness.cc",
        "util/testutil.cc",
        "utilities/cassandra/test_utils.cc",
-        "utilities/col_buf_decoder.cc",
-        "utilities/col_buf_encoder.cc",
-        "utilities/column_aware_encoding_util.cc",
    ],
    auto_headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [":rocksdb_lib"],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

 cpp_library(
@@ -342,22 +331,22 @@ cpp_library(
        "util/testutil.cc",
    ],
    auto_headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [":rocksdb_lib"],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

 cpp_library(
    name = "env_basic_test_lib",
    srcs = ["env/env_basic_test.cc"],
    auto_headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [":rocksdb_test_lib"],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

 # [test_name, test_src, test_type]
@@ -447,11 +436,6 @@ ROCKS_TESTS = [
        "util/coding_test.cc",
        "serial",
    ],
-    [
-        "column_aware_encoding_test",
-        "utilities/column_aware_encoding_test.cc",
-        "serial",
-    ],
    [
        "column_family_test",
        "db/column_family_test.cc",
@@ -522,11 +506,6 @@ ROCKS_TESTS = [
        "table/data_block_hash_index_test.cc",
        "serial",
    ],
-    [
-        "date_tiered_test",
-        "utilities/date_tiered/date_tiered_test.cc",
-        "serial",
-    ],
    [
        "db_basic_test",
        "db/db_basic_test.cc",
@@ -687,11 +666,6 @@ ROCKS_TESTS = [
        "db/deletefile_test.cc",
        "serial",
    ],
-    [
-        "document_db_test",
-        "utilities/document/document_db_test.cc",
-        "serial",
-    ],
    [
        "dynamic_bloom_test",
        "util/dynamic_bloom_test.cc",
@@ -767,11 +741,6 @@ ROCKS_TESTS = [
        "table/full_filter_block_test.cc",
        "serial",
    ],
-    [
-        "geodb_test",
-        "utilities/geodb/geodb_test.cc",
-        "serial",
-    ],
    [
        "hash_table_test",
        "utilities/persistent_cache/hash_table_test.cc",
@@ -802,11 +771,6 @@ ROCKS_TESTS = [
        "monitoring/iostats_context_test.cc",
        "serial",
    ],
-    [
-        "json_document_test",
-        "utilities/document/json_document_test.cc",
-        "serial",
-    ],
    [
        "ldb_cmd_test",
        "tools/ldb_cmd_test.cc",
@@ -972,11 +936,6 @@ ROCKS_TESTS = [
        "util/slice_transform_test.cc",
        "serial",
    ],
-    [
-        "spatial_db_test",
-        "utilities/spatialdb/spatial_db_test.cc",
-        "serial",
-    ],
    [
        "sst_dump_test",
        "tools/sst_dump_test.cc",
@@ -1102,26 +1061,16 @@ ROCKS_TESTS = [
 # Generate a test rule for each entry in ROCKS_TESTS
 # Do not build the tests in opt mode, since SyncPoint and other test code
 # will not be included.
-if not is_opt_mode:
-    for test_cfg in ROCKS_TESTS:
-        test_name = test_cfg[0]
-        test_cc = test_cfg[1]
-        ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
-        test_bin = test_name + "_bin"
-
-        cpp_binary(
-            name = test_bin,
-            srcs = [test_cc],
-            arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-            compiler_flags = rocksdb_compiler_flags,
-            preprocessor_flags = rocksdb_preprocessor_flags,
-            deps = [":rocksdb_test_lib"],
-            external_deps = rocksdb_external_deps,
-        )
-
-        custom_unittest(
-            name = test_name,
-            command = [TEST_RUNNER, BUCK_BINS + test_bin],
-            type = ttype,
-            deps = [":" + test_bin],
-        )
+[
+    test_binary(
+        parallelism = parallelism,
+        rocksdb_arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+        rocksdb_compiler_flags = ROCKSDB_COMPILER_FLAGS,
+        rocksdb_external_deps = ROCKSDB_EXTERNAL_DEPS,
+        rocksdb_preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+        test_cc = test_cc,
+        test_name = test_name,
+    )
+    for test_name, test_cc, parallelism in ROCKS_TESTS
+    if not is_opt_mode
+]
@@ -1,75 +1,15 @@
 version: 1.0.{build}
-
 image: Visual Studio 2017
-
-environment:
-  JAVA_HOME: C:\Program Files\Java\jdk1.8.0
-  THIRDPARTY_HOME: $(APPVEYOR_BUILD_FOLDER)\thirdparty
-  SNAPPY_HOME: $(THIRDPARTY_HOME)\snappy-1.1.7
-  SNAPPY_INCLUDE: $(SNAPPY_HOME);$(SNAPPY_HOME)\build
-  SNAPPY_LIB_DEBUG: $(SNAPPY_HOME)\build\Debug\snappy.lib
-  SNAPPY_LIB_RELEASE: $(SNAPPY_HOME)\build\Release\snappy.lib
-  LZ4_HOME: $(THIRDPARTY_HOME)\lz4-1.8.3
-  LZ4_INCLUDE: $(LZ4_HOME)\lib
-  LZ4_LIB_DEBUG: $(LZ4_HOME)\visual\VS2010\bin\x64_Debug\liblz4_static.lib
-  LZ4_LIB_RELEASE: $(LZ4_HOME)\visual\VS2010\bin\x64_Release\liblz4_static.lib
-  ZSTD_HOME: $(THIRDPARTY_HOME)\zstd-1.4.0
-  ZSTD_INCLUDE: $(ZSTD_HOME)\lib;$(ZSTD_HOME)\lib\dictBuilder
-  ZSTD_LIB_DEBUG: $(ZSTD_HOME)\build\VS2010\bin\x64_Debug\libzstd_static.lib
-  ZSTD_LIB_RELEASE: $(ZSTD_HOME)\build\VS2010\bin\x64_Release\libzstd_static.lib
-  matrix:
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
-      CMAKE_GENERATOR: Visual Studio 14 Win64
-      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio 14.0\Common7\IDE\devenv.com
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
-      CMAKE_GENERATOR: Visual Studio 15 Win64
-      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\devenv.com
-
-install:
-  - md %THIRDPARTY_HOME%
-  - echo "Building Snappy dependency..."
-  - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output snappy-1.1.7.zip --location https://github.com/google/snappy/archive/1.1.7.zip
-  - unzip snappy-1.1.7.zip
-  - cd snappy-1.1.7
-  - mkdir build
-  - cd build
-  - cmake -G "%CMAKE_GENERATOR%" ..
-  - msbuild Snappy.sln /p:Configuration=Debug /p:Platform=x64
-  - msbuild Snappy.sln /p:Configuration=Release /p:Platform=x64
-  - echo "Building LZ4 dependency..."
-  - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output lz4-1.8.3.zip --location https://github.com/lz4/lz4/archive/v1.8.3.zip
-  - unzip lz4-1.8.3.zip
-  - cd lz4-1.8.3\visual\VS2010
-  - ps: $CMD="$Env:DEV_ENV"; & $CMD lz4.sln /upgrade
-  - msbuild lz4.sln /p:Configuration=Debug /p:Platform=x64
-  - msbuild lz4.sln /p:Configuration=Release /p:Platform=x64
-  - echo "Building ZStd dependency..."
-  - cd %THIRDPARTY_HOME%
-  - curl --fail --silent --show-error --output zstd-1.4.0.zip --location https://github.com/facebook/zstd/archive/v1.4.0.zip
-  - unzip zstd-1.4.0.zip
-  - cd zstd-1.4.0\build\VS2010
-  - ps: $CMD="$Env:DEV_ENV"; & $CMD zstd.sln /upgrade
-  - msbuild zstd.sln /p:Configuration=Debug /p:Platform=x64
-  - msbuild zstd.sln /p:Configuration=Release /p:Platform=x64
-
 before_build:
-  - md %APPVEYOR_BUILD_FOLDER%\build
-  - cd %APPVEYOR_BUILD_FOLDER%\build
-  - cmake -G "%CMAKE_GENERATOR%" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DLZ4=1 -DZSTD=1 -DXPRESS=1 -DJNI=1 ..
-  - cd ..
-
+- md %APPVEYOR_BUILD_FOLDER%\build
+- cd %APPVEYOR_BUILD_FOLDER%\build
+- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 -DJNI=1 ..
+- cd ..
 build:
  project: build\rocksdb.sln
  parallel: true
  verbosity: normal
-
 test:
-
 test_script:
-  - ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test,db_merge_operand_test -Concurrency 8
-
-on_failure:
-  - cmd: 7z a build-failed.zip %APPVEYOR_BUILD_FOLDER%\build\ && appveyor PushArtifact build-failed.zip
+- ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test -Concurrency 8

@@ -3,14 +3,11 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
 rocksdb_target_header = """load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+load(":defs.bzl", "test_binary")

 REPO_PATH = package_name() + "/"

-BUCK_BINS = "buck-out/gen/" + REPO_PATH
-
-TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
-
-rocksdb_compiler_flags = [
+ROCKSDB_COMPILER_FLAGS = [
    "-fno-builtin-memcmp",
    "-DROCKSDB_PLATFORM_POSIX",
    "-DROCKSDB_LIB_IO_POSIX",
@@ -37,7 +34,7 @@ rocksdb_compiler_flags = [
    "-Wnarrowing",
 ]

-rocksdb_external_deps = [
+ROCKSDB_EXTERNAL_DEPS = [
    ("bzip2", None, "bz2"),
    ("snappy", None, "snappy"),
    ("zlib", None, "z"),
@@ -49,13 +46,13 @@ rocksdb_external_deps = [
    ("googletest", None, "gtest"),
 ]

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

-rocksdb_arch_preprocessor_flags = {
+ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
    "x86_64": [
        "-DHAVE_SSE42",
        "-DHAVE_PCLMUL",
@@ -68,16 +65,15 @@ is_opt_mode = build_mode.startswith("opt")

 # -DNDEBUG is added by default in opt mode in fbcode. But adding it twice
 # doesn't harm and avoid forgetting to add it.
-if is_opt_mode:
-    rocksdb_compiler_flags.append("-DNDEBUG")
+ROCKSDB_COMPILER_FLAGS += (["-DNDEBUG"] if is_opt_mode else [])

 sanitizer = read_config("fbcode", "sanitizer")

 # Do not enable jemalloc if sanitizer presents. RocksDB will further detect
 # whether the binary is linked with jemalloc at runtime.
-if sanitizer == "":
-    rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
-    rocksdb_external_deps.append(("jemalloc", None, "headers"))
+ROCKSDB_COMPILER_FLAGS += (["-DROCKSDB_JEMALLOC"] if sanitizer == "" else [])
+
+ROCKSDB_EXTERNAL_DEPS += ([("jemalloc", None, "headers")] if sanitizer == "" else [])
 """


@@ -86,11 +82,11 @@ cpp_library(
    name = "{name}",
    srcs = [{srcs}],
    {headers_attr_prefix}headers = {headers},
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [{deps}],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 """

@@ -98,11 +94,11 @@ binary_template = """
 cpp_binary(
    name = "%s",
    srcs = [%s],
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
    deps = [%s],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 """

@@ -121,27 +117,17 @@ ROCKS_TESTS = [
 # Generate a test rule for each entry in ROCKS_TESTS
 # Do not build the tests in opt mode, since SyncPoint and other test code
 # will not be included.
-if not is_opt_mode:
-    for test_cfg in ROCKS_TESTS:
-        test_name = test_cfg[0]
-        test_cc = test_cfg[1]
-        ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
-        test_bin = test_name + "_bin"
-
-        cpp_binary(
-            name = test_bin,
-            srcs = [test_cc],
-            arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-            compiler_flags = rocksdb_compiler_flags,
-            preprocessor_flags = rocksdb_preprocessor_flags,
-            deps = [":rocksdb_test_lib"],
-            external_deps = rocksdb_external_deps,
-        )
-
-        custom_unittest(
-            name = test_name,
-            command = [TEST_RUNNER, BUCK_BINS + test_bin],
-            type = ttype,
-            deps = [":" + test_bin],
-        )
+[
+    test_binary(
+        parallelism = parallelism,
+        rocksdb_arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+        rocksdb_compiler_flags = ROCKSDB_COMPILER_FLAGS,
+        rocksdb_external_deps = ROCKSDB_EXTERNAL_DEPS,
+        rocksdb_preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+        test_cc = test_cc,
+        test_name = test_name,
+    )
+    for test_name, test_cc, parallelism in ROCKS_TESTS
+    if not is_opt_mode
+]
 """
@@ -97,7 +97,7 @@ function getSteps($applyDiff, $diffID, $username, $test) {
  }

  // fbcode is a sub-repo. We cannot patch until we add it to ignore otherwise
-  // Git thinks it is an uncommited change.
+  // Git thinks it is an uncommitted change.
  $fix_git_ignore = array(
    "name" => "Fix git ignore",
    "shell" => "echo fbcode >> .git/info/exclude",
@@ -53,13 +53,11 @@ if [ -z "$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=$COMPILE_WITH_TSAN
-    if [ -n "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
-      # we need this to build with MySQL. Don't use for other purposes.
-      source "$PWD/build_tools/fbcode_config4.8.1.sh"
-    elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_5xx" ]; then
+    if [ -z "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
      source "$PWD/build_tools/fbcode_config.sh"
    else
-      source "$PWD/build_tools/fbcode_config_platform007.sh"
+      # we need this to build with MySQL. Don't use for other purposes.
+      source "$PWD/build_tools/fbcode_config4.8.1.sh"
    fi
 fi

@@ -370,6 +368,16 @@ EOF
            # JEMALLOC can be enabled either using the flag (like here) or by
            # providing direct link to the jemalloc library
            WITH_JEMALLOC_FLAG=1
+            # check for JEMALLOC installed with HomeBrew
+            if [ "$PLATFORM" == "OS_MACOSX" ]; then
+                if hash brew 2>/dev/null && brew ls --versions jemalloc > /dev/null; then
+                    JEMALLOC_VER=$(brew ls --versions jemalloc | tail -n 1 | cut -f 2 -d ' ')
+                    JEMALLOC_INCLUDE="-I/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/include"
+                    JEMALLOC_LIB="/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/lib/libjemalloc_pic.a"
+                    PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS $JEMALLOC_LIB"
+                    JAVA_STATIC_LDFLAGS="$JAVA_STATIC_LDFLAGS $JEMALLOC_LIB"
+                fi
+            fi
        fi
    fi
    if ! test $JEMALLOC && ! test $ROCKSDB_DISABLE_TCMALLOC; then
@@ -527,8 +535,6 @@ if test -z "$PORTABLE"; then
  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
    # TODO: Handle this with approprite options.
    COMMON_FLAGS="$COMMON_FLAGS"
-  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^aarch64`"; then
-    COMMON_FLAGS="$COMMON_FLAGS"
  elif [ "$TARGET_OS" == "IOS" ]; then
    COMMON_FLAGS="$COMMON_FLAGS"
  elif [ "$TARGET_OS" != "AIX" ] && [ "$TARGET_OS" != "SunOS" ]; then
@@ -1,18 +0,0 @@
-GCC_BASE=/mnt/gvfs/third-party2/gcc/6e8e715624fd15256a7970073387793dfcf79b46/7.x/centos7-native/b2ef2b6
-CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/ef37e1faa1c29782abfac1ae65a291b9b7966f6d/stable/centos7-native/c9f9104
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/c67031f0f739ac61575a061518d6ef5038f99f90/7.x/platform007/5620abc
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/60d6f124a78798b73944f5ba87c2306ae3460153/2.26/platform007/f259413
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/22c2d65676fb7c23cfa797c4f6937f38b026f3cf/1.2.8/platform007/ca4da3d
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/907b498203d297947f3bb70b9466f47e100f1873/r131/platform007/ca4da3d
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/3ee276cbacfad3074e3f07bf826ac47f06970f4e/1.3.5/platform007/15a3614
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/9c910d36d6235cc40e8ff559358f1833452300ca/master/platform007/5b0f53e
-NUMA_BASE=/mnt/gvfs/third-party2/numa/9cbf2460284c669ed19c3ccb200a71f7dd7e53c7/2.0.11/platform007/ca4da3d
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/bf3d7497fe4e6d007354f0adffa16ce3003f8338/1.3/platform007/6f3e0a9
-TBB_BASE=/mnt/gvfs/third-party2/tbb/ff4e0b093534704d8abab678a4fd7f5ea7b094c7/2018_U5/platform007/ca4da3d
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/b5c4a61a5c483ba24722005ae07895971a2ac707/fb/platform007/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/92ff90349e2f43ea0a8246d8b1cf17b6869013e3/2.29.1/centos7-native/da39a3e
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/f3f697a28122e6bcd513273dd9c1ff23852fc59f/3.13.0/platform007/ca4da3d
-LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832
@@ -51,7 +51,7 @@ if test -z $PIC_BUILD; then
 else
  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
 fi
-CFLAGS+=" -DZSTD"
+CFLAGS+=" -DZSTD -DZSTD_STATIC_LINKING_ONLY"

 # location of gflags headers and libraries
 GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
@@ -1,157 +0,0 @@
-#!/bin/sh
-#
-# Set environment variables so that we can compile rocksdb using
-# fbcode settings.  It uses the latest g++ and clang compilers and also
-# uses jemalloc
-# Environment variables that change the behavior of this script:
-# PIC_BUILD -- if true, it will only take pic versions of libraries from fbcode. libraries that don't have pic variant will not be included
-
-
-BASEDIR=`dirname $BASH_SOURCE`
-source "$BASEDIR/dependencies_platform007.sh"
-
-CFLAGS=""
-
-# libgcc
-LIBGCC_INCLUDE="$LIBGCC_BASE/include/c++/7.3.0"
-LIBGCC_LIBS=" -L $LIBGCC_BASE/lib"
-
-# glibc
-GLIBC_INCLUDE="$GLIBC_BASE/include"
-GLIBC_LIBS=" -L $GLIBC_BASE/lib"
-
-# snappy
-SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include/"
-if test -z $PIC_BUILD; then
-  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy.a"
-else
-  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy_pic.a"
-fi
-CFLAGS+=" -DSNAPPY"
-
-if test -z $PIC_BUILD; then
-  # location of zlib headers and libraries
-  ZLIB_INCLUDE=" -I $ZLIB_BASE/include/"
-  ZLIB_LIBS=" $ZLIB_BASE/lib/libz.a"
-  CFLAGS+=" -DZLIB"
-
-  # location of bzip headers and libraries
-  BZIP_INCLUDE=" -I $BZIP2_BASE/include/"
-  BZIP_LIBS=" $BZIP2_BASE/lib/libbz2.a"
-  CFLAGS+=" -DBZIP2"
-
-  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
-  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
-  CFLAGS+=" -DLZ4"
-fi
-
-ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
-if test -z $PIC_BUILD; then
-  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
-else
-  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
-fi
-CFLAGS+=" -DZSTD"
-
-# location of gflags headers and libraries
-GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
-if test -z $PIC_BUILD; then
-  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags.a"
-else
-  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
-fi
-CFLAGS+=" -DGFLAGS=gflags"
-
-# location of jemalloc
-JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
-JEMALLOC_LIB=" $JEMALLOC_BASE/lib/libjemalloc.a"
-
-if test -z $PIC_BUILD; then
-  # location of numa
-  NUMA_INCLUDE=" -I $NUMA_BASE/include/"
-  NUMA_LIB=" $NUMA_BASE/lib/libnuma.a"
-  CFLAGS+=" -DNUMA"
-
-  # location of libunwind
-  LIBUNWIND="$LIBUNWIND_BASE/lib/libunwind.a"
-fi
-
-# location of TBB
-TBB_INCLUDE=" -isystem $TBB_BASE/include/"
-if test -z $PIC_BUILD; then
-  TBB_LIBS="$TBB_BASE/lib/libtbb.a"
-else
-  TBB_LIBS="$TBB_BASE/lib/libtbb_pic.a"
-fi
-CFLAGS+=" -DTBB"
-
-# use Intel SSE support for checksum calculations
-export USE_SSE=1
-export PORTABLE=1
-
-BINUTILS="$BINUTILS_BASE/bin"
-AR="$BINUTILS/ar"
-
-DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE"
-
-STDLIBS="-L $GCC_BASE/lib64"
-
-CLANG_BIN="$CLANG_BASE/bin"
-CLANG_LIB="$CLANG_BASE/lib"
-CLANG_SRC="$CLANG_BASE/../../src"
-
-CLANG_ANALYZER="$CLANG_BIN/clang++"
-CLANG_SCAN_BUILD="$CLANG_SRC/llvm/tools/clang/tools/scan-build/bin/scan-build"
-
-if [ -z "$USE_CLANG" ]; then
-  # gcc
-  CC="$GCC_BASE/bin/gcc"
-  CXX="$GCC_BASE/bin/g++"
-
-  CFLAGS+=" -B$BINUTILS/gold"
-  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
-  CFLAGS+=" -isystem $GLIBC_INCLUDE"
-  JEMALLOC=1
-else
-  # clang
-  CLANG_INCLUDE="$CLANG_LIB/clang/stable/include"
-  CC="$CLANG_BIN/clang"
-  CXX="$CLANG_BIN/clang++"
-
-  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include"
-
-  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x "
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x/x86_64-facebook-linux "
-  CFLAGS+=" -isystem $GLIBC_INCLUDE"
-  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
-  CFLAGS+=" -isystem $CLANG_INCLUDE"
-  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE/linux "
-  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE "
-  CFLAGS+=" -Wno-expansion-to-defined "
-  CXXFLAGS="-nostdinc++"
-fi
-
-CFLAGS+=" $DEPS_INCLUDE"
-CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
-CXXFLAGS+=" $CFLAGS"
-
-EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
-EXEC_LDFLAGS+=" -B$BINUTILS/gold"
-EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
-EXEC_LDFLAGS+=" $LIBUNWIND"
-EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
-# required by libtbb
-EXEC_LDFLAGS+=" -ldl"
-
-PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
-
-EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
-
-VALGRIND_VER="$VALGRIND_BASE/bin/"
-
-# lua not supported because it's on track for deprecation, I think
-LUA_PATH=
-LUA_LIB=
-
-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
@@ -377,7 +377,7 @@ function send_to_ods {
    echo >&2 "ERROR: Key $key doesn't have a value."
    return
  fi
-  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
+  curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
    --connect-timeout 60
 }

@@ -85,9 +85,7 @@ NON_SHM="TMPD=/tmp/rocksdb_test_tmp"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-# in gcc-5 there are known problems with TSAN like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71090.
-# using platform007 gives us gcc-8 or higher which has that bug fixed.
-TSAN="ROCKSDB_FBCODE_BUILD_WITH_PLATFORM007=1 COMPILE_WITH_TSAN=1"
+TSAN="COMPILE_WITH_TSAN=1"
 UBSAN="COMPILE_WITH_UBSAN=1"
 TSAN_CRASH='CRASH_TEST_EXT_ARGS="--compression_type=zstd --log2_keys_per_lock=22"'
 NON_TSAN_CRASH="CRASH_TEST_EXT_ARGS=--compression_type=zstd"
@@ -646,7 +644,7 @@ run_regression()

  # parameters: $1 -- key, $2 -- value
  function send_size_to_ods {
-    curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
+    curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
      --connect-timeout 60
  }

@@ -53,45 +53,6 @@ function get_lib_base()
  log_variable $__res_var
 }

-###########################################################
-#                platform007 dependencies                 #
-###########################################################
-
-OUTPUT="$BASEDIR/dependencies_platform007.sh"
-
-rm -f "$OUTPUT"
-touch "$OUTPUT"
-
-echo "Writing dependencies to $OUTPUT"
-
-# Compilers locations
-GCC_BASE=`readlink -f $TP2_LATEST/gcc/7.x/centos7-native/*/`
-CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos7-native/*/`
-
-log_variable GCC_BASE
-log_variable CLANG_BASE
-
-# Libraries locations
-get_lib_base libgcc     7.x     platform007
-get_lib_base glibc      2.26    platform007
-get_lib_base snappy     LATEST  platform007
-get_lib_base zlib       LATEST  platform007
-get_lib_base bzip2      LATEST  platform007
-get_lib_base lz4        LATEST  platform007
-get_lib_base zstd       LATEST  platform007
-get_lib_base gflags     LATEST  platform007
-get_lib_base jemalloc   LATEST  platform007
-get_lib_base numa       LATEST  platform007
-get_lib_base libunwind  LATEST  platform007
-get_lib_base tbb        LATEST  platform007
-
-get_lib_base kernel-headers fb platform007
-get_lib_base binutils   LATEST centos7-native
-get_lib_base valgrind   LATEST platform007
-get_lib_base lua        5.3.4  platform007
-
-git diff $OUTPUT
-
 ###########################################################
 #                   5.x dependencies                      #
 ###########################################################
@@ -1,21 +0,0 @@
-# - Find zlib
-# Find the zlib compression library and includes
-#
-# ZLIB_INCLUDE_DIR - where to find zlib.h, etc.
-# ZLIB_LIBRARIES - List of libraries when using zlib.
-# ZLIB_FOUND - True if zlib found.
-
-find_path(ZLIB_INCLUDE_DIR
-  NAMES zlib.h
-  HINTS ${ZLIB_ROOT_DIR}/include)
-
-find_library(ZLIB_LIBRARIES
-  NAMES z
-  HINTS ${ZLIB_ROOT_DIR}/lib)
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(zlib DEFAULT_MSG ZLIB_LIBRARIES ZLIB_INCLUDE_DIR)
-
-mark_as_advanced(
-  ZLIB_LIBRARIES
-  ZLIB_INCLUDE_DIR)
@@ -1784,11 +1784,11 @@ rocksdb_iterator_t* rocksdb_writebatch_wi_create_iterator_with_base(
 }

 rocksdb_iterator_t* rocksdb_writebatch_wi_create_iterator_with_base_cf(
-    rocksdb_writebatch_wi_t* wbwi,
-    rocksdb_iterator_t* base_iterator,
+    rocksdb_writebatch_wi_t* wbwi, rocksdb_iterator_t* base_iterator,
    rocksdb_column_family_handle_t* column_family) {
  rocksdb_iterator_t* result = new rocksdb_iterator_t;
-  result->rep = wbwi->rep->NewIteratorWithBase(column_family->rep, base_iterator->rep);
+  result->rep =
+      wbwi->rep->NewIteratorWithBase(column_family->rep, base_iterator->rep);
  delete base_iterator;
  return result;
 }
@@ -2067,6 +2067,15 @@ void rocksdb_set_options(
            db->rep->SetOptions(options_map));
    }

+void rocksdb_set_options_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* handle, int count, const char* const keys[], const char* const values[], char** errptr) {
+        std::unordered_map<std::string, std::string> options_map;
+        for (int i=0; i<count; i++)
+            options_map[keys[i]] = values[i];
+        SaveError(errptr,
+            db->rep->SetOptions(handle->rep, options_map));
+    }
+
 rocksdb_options_t* rocksdb_options_create() {
  return new rocksdb_options_t;
 }
@@ -917,7 +917,8 @@ int main(int argc, char** argv) {
    rocksdb_writebatch_wi_t* wbi = rocksdb_writebatch_wi_create(0, 1);
    rocksdb_writebatch_wi_put(wbi, "bar", 3, "b", 1);
    rocksdb_writebatch_wi_delete(wbi, "foo", 3);
-    rocksdb_iterator_t* iter = rocksdb_writebatch_wi_create_iterator_with_base(wbi, base_iter);
+    rocksdb_iterator_t* iter =
+        rocksdb_writebatch_wi_create_iterator_with_base(wbi, base_iter);
    CheckCondition(!rocksdb_iter_valid(iter));
    rocksdb_iter_seek_to_first(iter);
    CheckCondition(rocksdb_iter_valid(iter));
@@ -1527,7 +1528,7 @@ int main(int argc, char** argv) {
    const rocksdb_snapshot_t* snapshot;
    snapshot = rocksdb_transactiondb_create_snapshot(txn_db);
    rocksdb_readoptions_set_snapshot(roptions, snapshot);
-  
+
    rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hey", 3,  &err);
    CheckNoError(err);

@@ -129,14 +129,10 @@ Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
    }
  }
  if (cf_options.compression_opts.zstd_max_train_bytes > 0) {
-    if (!CompressionTypeSupported(CompressionType::kZSTD)) {
-      // Dictionary trainer is available since v0.6.1, but ZSTD was marked
-      // stable only since v0.8.0. For now we enable the feature in stable
-      // versions only.
+    if (!ZSTD_TrainDictionarySupported()) {
      return Status::InvalidArgument(
-          "zstd dictionary trainer cannot be used because " +
-          CompressionTypeToString(CompressionType::kZSTD) +
-          " is not linked with the binary.");
+          "zstd dictionary trainer cannot be used because ZSTD 1.1.3+ "
+          "is not linked with the binary.");
    }
    if (cf_options.compression_opts.max_dict_bytes == 0) {
      return Status::InvalidArgument(
@@ -383,6 +383,8 @@ class ColumnFamilyData {

  Directory* GetDataDir(size_t path_id) const;

+  ThreadLocalPtr* TEST_GetLocalSV() { return local_sv_.get(); }
+
 private:
  friend class ColumnFamilySet;
  ColumnFamilyData(uint32_t id, const std::string& name,
@@ -2042,7 +2042,7 @@ TEST_P(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
 }
 #endif  // !ROCKSDB_LITE

-#ifndef ROCKSDB_LITE  // Tailing interator not supported
+#ifndef ROCKSDB_LITE  // Tailing iterator not supported
 namespace {
 std::string IterStatus(Iterator* iter) {
  std::string result;
@@ -357,6 +357,51 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
  }
 }

+TEST_F(CompactFilesTest, GetCompactionJobInfo) {
+  Options options;
+  options.create_if_missing = true;
+  // Disable RocksDB background compaction.
+  options.compaction_style = kCompactionStyleNone;
+  options.level0_slowdown_writes_trigger = 1000;
+  options.level0_stop_writes_trigger = 1000;
+  options.write_buffer_size = 65536;
+  options.max_write_buffer_number = 2;
+  options.compression = kNoCompression;
+  options.max_compaction_bytes = 5000;
+
+  // 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 couple files
+  for (int i = 0; i < 500; ++i) {
+    db->Put(WriteOptions(), ToString(i), std::string(1000, 'a' + (i % 26)));
+  }
+  reinterpret_cast<DBImpl*>(db)->TEST_WaitForFlushMemTable();
+  auto l0_files_1 = collector->GetFlushedFiles();
+  CompactionOptions co;
+  co.compression = CompressionType::kLZ4Compression;
+  CompactionJobInfo compaction_job_info;
+  ASSERT_OK(
+      db->CompactFiles(co, l0_files_1, 0, -1, nullptr, &compaction_job_info));
+  ASSERT_EQ(compaction_job_info.base_input_level, 0);
+  ASSERT_EQ(compaction_job_info.cf_id, db->DefaultColumnFamily()->GetID());
+  ASSERT_EQ(compaction_job_info.cf_name, db->DefaultColumnFamily()->GetName());
+  ASSERT_EQ(compaction_job_info.compaction_reason,
+            CompactionReason::kManualCompaction);
+  ASSERT_EQ(compaction_job_info.compression, CompressionType::kLZ4Compression);
+  ASSERT_EQ(compaction_job_info.output_level, 0);
+  ASSERT_OK(compaction_job_info.status);
+  // no assertion failure
+  delete db;
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -373,8 +373,8 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
        if (user_cmp->Compare(user_key, f->largest.user_key()) <= 0) {
          // We've advanced far enough
          if (user_cmp->Compare(user_key, f->smallest.user_key()) >= 0) {
-            // Key falls in this file's range, so definitely
-            // exists beyond output level
+            // Key falls in this file's range, so it may
+            // exist beyond output level
            return false;
          }
          break;
@@ -9,6 +9,23 @@
 #include "port/likely.h"
 #include "rocksdb/listener.h"
 #include "table/internal_iterator.h"
+#include "util/sync_point.h"
+
+#define DEFINITELY_IN_SNAPSHOT(seq, snapshot)                       \
+  ((seq) <= (snapshot) &&                                           \
+   (snapshot_checker_ == nullptr ||                                 \
+    LIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
+           SnapshotCheckerResult::kInSnapshot)))
+
+#define DEFINITELY_NOT_IN_SNAPSHOT(seq, snapshot)                     \
+  ((seq) > (snapshot) ||                                              \
+   (snapshot_checker_ != nullptr &&                                   \
+    UNLIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
+             SnapshotCheckerResult::kNotInSnapshot)))
+
+#define IN_EARLIEST_SNAPSHOT(seq) \
+  ((seq) <= earliest_snapshot_ && \
+   (snapshot_checker_ == nullptr || LIKELY(IsInEarliestSnapshot(seq))))

 namespace rocksdb {

@@ -61,26 +78,28 @@ CompactionIterator::CompactionIterator(
      merge_out_iter_(merge_helper_),
      current_key_committed_(false) {
  assert(compaction_filter_ == nullptr || compaction_ != nullptr);
+  assert(snapshots_ != nullptr);
  bottommost_level_ =
      compaction_ == nullptr ? false : compaction_->bottommost_level();
  if (compaction_ != nullptr) {
    level_ptrs_ = std::vector<size_t>(compaction_->number_levels(), 0);
  }
-
  if (snapshots_->size() == 0) {
    // optimize for fast path if there are no snapshots
    visible_at_tip_ = true;
+    earliest_snapshot_iter_ = snapshots_->end();
    earliest_snapshot_ = kMaxSequenceNumber;
    latest_snapshot_ = 0;
  } else {
    visible_at_tip_ = false;
+    earliest_snapshot_iter_ = snapshots_->begin();
    earliest_snapshot_ = snapshots_->at(0);
    latest_snapshot_ = snapshots_->back();
  }
 #ifndef NDEBUG
  // findEarliestVisibleSnapshot assumes this ordering.
  for (size_t i = 1; i < snapshots_->size(); ++i) {
-    assert(snapshots_->at(i - 1) <= snapshots_->at(i));
+    assert(snapshots_->at(i - 1) < snapshots_->at(i));
  }
 #endif
  if (compaction_filter_ != nullptr) {
@@ -91,6 +110,7 @@ CompactionIterator::CompactionIterator(
    ignore_snapshots_ = false;
  }
  input_->SetPinnedItersMgr(&pinned_iters_mgr_);
+  TEST_SYNC_POINT_CALLBACK("CompactionIterator:AfterInit", compaction_.get());
 }

 CompactionIterator::~CompactionIterator() {
@@ -162,10 +182,7 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
  if (compaction_filter_ != nullptr &&
      (ikey_.type == kTypeValue || ikey_.type == kTypeBlobIndex) &&
      (visible_at_tip_ || ignore_snapshots_ ||
-       ikey_.sequence > latest_snapshot_ ||
-       (snapshot_checker_ != nullptr &&
-        UNLIKELY(!snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                  latest_snapshot_))))) {
+       DEFINITELY_NOT_IN_SNAPSHOT(ikey_.sequence, latest_snapshot_))) {
    // If the user has specified a compaction filter and the sequence
    // number is greater than any external snapshot, then invoke the
    // filter. If the return value of the compaction filter is true,
@@ -241,6 +258,7 @@ void CompactionIterator::NextFromInput() {
      valid_ = true;
      break;
    }
+    TEST_SYNC_POINT_CALLBACK("CompactionIterator:ProcessKV", &ikey_);

    // Update input statistics
    if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion) {
@@ -269,9 +287,7 @@ void CompactionIterator::NextFromInput() {
      has_outputted_key_ = false;
      current_user_key_sequence_ = kMaxSequenceNumber;
      current_user_key_snapshot_ = 0;
-      current_key_committed_ =
-          (snapshot_checker_ == nullptr ||
-           snapshot_checker_->IsInSnapshot(ikey_.sequence, kMaxSequenceNumber));
+      current_key_committed_ = KeyCommitted(ikey_.sequence);

      // Apply the compaction filter to the first committed version of the user
      // key.
@@ -293,8 +309,7 @@ void CompactionIterator::NextFromInput() {
      // to query snapshot_checker_ in that case.
      if (UNLIKELY(!current_key_committed_)) {
        assert(snapshot_checker_ != nullptr);
-        current_key_committed_ =
-            snapshot_checker_->IsInSnapshot(ikey_.sequence, kMaxSequenceNumber);
+        current_key_committed_ = KeyCommitted(ikey_.sequence);
        // Apply the compaction filter to the first committed version of the
        // user key.
        if (current_key_committed_) {
@@ -378,10 +393,8 @@ void CompactionIterator::NextFromInput() {
          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
        // Check whether the next key belongs to the same snapshot as the
        // SingleDelete.
-        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
-            (snapshot_checker_ != nullptr &&
-             UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
-                                                       prev_snapshot)))) {
+        if (prev_snapshot == 0 ||
+            DEFINITELY_NOT_IN_SNAPSHOT(next_ikey.sequence, prev_snapshot)) {
          if (next_ikey.type == kTypeSingleDeletion) {
            // We encountered two SingleDeletes in a row.  This could be due to
            // unexpected user input.
@@ -393,11 +406,8 @@ void CompactionIterator::NextFromInput() {
            ++iter_stats_.num_record_drop_obsolete;
            ++iter_stats_.num_single_del_mismatch;
          } else if (has_outputted_key_ ||
-                     (ikey_.sequence <= earliest_write_conflict_snapshot_ &&
-                      (snapshot_checker_ == nullptr ||
-                       LIKELY(snapshot_checker_->IsInSnapshot(
-                           ikey_.sequence,
-                           earliest_write_conflict_snapshot_))))) {
+                     DEFINITELY_IN_SNAPSHOT(
+                         ikey_.sequence, earliest_write_conflict_snapshot_)) {
            // Found a matching value, we can drop the single delete and the
            // value.  It is safe to drop both records since we've already
            // outputted a key in this snapshot, or there is no earlier
@@ -407,7 +417,8 @@ void CompactionIterator::NextFromInput() {
            // is an unexpected Merge or Delete.  We will compact it out
            // either way. We will maintain counts of how many mismatches
            // happened
-            if (next_ikey.type != kTypeValue) {
+            if (next_ikey.type != kTypeValue &&
+                next_ikey.type != kTypeBlobIndex) {
              ++iter_stats_.num_single_del_mismatch;
            }

@@ -444,10 +455,7 @@ 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 && ikey_.sequence <= earliest_snapshot_ &&
-            (snapshot_checker_ == nullptr ||
-             LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                    earliest_snapshot_))) &&
+        if (compaction_ != nullptr && IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_)) {
          // Key doesn't exist outside of this range.
@@ -466,24 +474,34 @@ void CompactionIterator::NextFromInput() {
      if (valid_) {
        at_next_ = true;
      }
-    } else if (last_snapshot == current_user_key_snapshot_) {
+    } else if (last_snapshot == current_user_key_snapshot_ ||
+               (last_snapshot > 0 &&
+                last_snapshot < current_user_key_snapshot_)) {
      // If the earliest snapshot is which this key is visible in
      // is the same as the visibility of a previous instance of the
      // same key, then this kv is not visible in any snapshot.
      // Hidden by an newer entry for same user key
-      // TODO(noetzli): 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_);
+
+      // Note2: if last_snapshot < current_user_key_snapshot, it can only
+      // mean last_snapshot is released between we process last value and
+      // this value, and findEarliestVisibleSnapshot returns the next snapshot
+      // as current_user_key_snapshot. In this case last value and current
+      // value are both in current_user_key_snapshot currently.
+      assert(last_snapshot == current_user_key_snapshot_ ||
+             (snapshot_checker_ != nullptr &&
+              snapshot_checker_->CheckInSnapshot(current_user_key_sequence_,
+                                                 last_snapshot) ==
+                  SnapshotCheckerResult::kSnapshotReleased));
+
      ++iter_stats_.num_record_drop_hidden;  // (A)
      input_->Next();
    } else if (compaction_ != nullptr && ikey_.type == kTypeDeletion &&
-               ikey_.sequence <= earliest_snapshot_ &&
-               (snapshot_checker_ == nullptr ||
-                LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                       earliest_snapshot_))) &&
+               IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
               ikeyNotNeededForIncrementalSnapshot() &&
               compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                          &level_ptrs_)) {
@@ -520,13 +538,10 @@ void CompactionIterator::NextFromInput() {
      input_->Next();
      // Skip over all versions of this key that happen to occur in the same snapshot
      // range as the delete
-      while (input_->Valid() &&
-             ParseInternalKey(input_->key(), &next_ikey) &&
+      while (input_->Valid() && ParseInternalKey(input_->key(), &next_ikey) &&
             cmp_->Equal(ikey_.user_key, next_ikey.user_key) &&
-             (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
-              (snapshot_checker_ != nullptr &&
-               UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
-                                                         prev_snapshot))))) {
+             (prev_snapshot == 0 ||
+              DEFINITELY_NOT_IN_SNAPSHOT(next_ikey.sequence, prev_snapshot))) {
        input_->Next();
      }
      // If you find you still need to output a row with this key, we need to output the
@@ -617,13 +632,9 @@ void CompactionIterator::PrepareOutput() {
  //
  // Can we do the same for levels above bottom level as long as
  // KeyNotExistsBeyondOutputLevel() return true?
-  if ((compaction_ != nullptr &&
-      !compaction_->allow_ingest_behind()) &&
-      ikeyNotNeededForIncrementalSnapshot() &&
-      bottommost_level_ && valid_ && ikey_.sequence <= earliest_snapshot_ &&
-      (snapshot_checker_ == nullptr || LIKELY(snapshot_checker_->IsInSnapshot(
-        ikey_.sequence, earliest_snapshot_))) &&
-      ikey_.type != kTypeMerge &&
+  if ((compaction_ != nullptr && !compaction_->allow_ingest_behind()) &&
+      ikeyNotNeededForIncrementalSnapshot() && bottommost_level_ && valid_ &&
+      IN_EARLIEST_SNAPSHOT(ikey_.sequence) && ikey_.type != kTypeMerge &&
      !cmp_->Equal(compaction_->GetLargestUserKey(), ikey_.user_key)) {
    assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
    ikey_.sequence = 0;
@@ -642,12 +653,23 @@ inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
    *prev_snapshot = *std::prev(snapshots_iter);
    assert(*prev_snapshot < in);
  }
+  if (snapshot_checker_ == nullptr) {
+    return snapshots_iter != snapshots_->end()
+      ? *snapshots_iter : kMaxSequenceNumber;
+  }
+  bool has_released_snapshot = !released_snapshots_.empty();
  for (; snapshots_iter != snapshots_->end(); ++snapshots_iter) {
    auto cur = *snapshots_iter;
    assert(in <= cur);
-    if (snapshot_checker_ == nullptr ||
-        snapshot_checker_->IsInSnapshot(in, cur)) {
+    // Skip if cur is in released_snapshots.
+    if (has_released_snapshot && released_snapshots_.count(cur) > 0) {
+      continue;
+    }
+    auto res = snapshot_checker_->CheckInSnapshot(in, cur);
+    if (res == SnapshotCheckerResult::kInSnapshot) {
      return cur;
+    } else if (res == SnapshotCheckerResult::kSnapshotReleased) {
+      released_snapshots_.insert(cur);
    }
    *prev_snapshot = cur;
  }
@@ -661,4 +683,30 @@ inline bool CompactionIterator::ikeyNotNeededForIncrementalSnapshot() {
         (ikey_.sequence < preserve_deletes_seqnum_);
 }

+bool CompactionIterator::IsInEarliestSnapshot(SequenceNumber sequence) {
+  assert(snapshot_checker_ != nullptr);
+  assert(earliest_snapshot_ == kMaxSequenceNumber ||
+         (earliest_snapshot_iter_ != snapshots_->end() &&
+          *earliest_snapshot_iter_ == earliest_snapshot_));
+  auto in_snapshot =
+      snapshot_checker_->CheckInSnapshot(sequence, earliest_snapshot_);
+  while (UNLIKELY(in_snapshot == SnapshotCheckerResult::kSnapshotReleased)) {
+    // Avoid the the current earliest_snapshot_ being return as
+    // earliest visible snapshot for the next value. So if a value's sequence
+    // is zero-ed out by PrepareOutput(), the next value will be compact out.
+    released_snapshots_.insert(earliest_snapshot_);
+    earliest_snapshot_iter_++;
+
+    if (earliest_snapshot_iter_ == snapshots_->end()) {
+      earliest_snapshot_ = kMaxSequenceNumber;
+    } else {
+      earliest_snapshot_ = *earliest_snapshot_iter_;
+    }
+    in_snapshot =
+        snapshot_checker_->CheckInSnapshot(sequence, earliest_snapshot_);
+  }
+  assert(in_snapshot != SnapshotCheckerResult::kSnapshotReleased);
+  return in_snapshot == SnapshotCheckerResult::kInSnapshot;
+}
+
 }  // namespace rocksdb
@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <deque>
 #include <string>
+#include <unordered_set>
 #include <vector>

 #include "db/compaction.h"
@@ -132,10 +133,25 @@ class CompactionIterator {
  // or seqnum be zero-ed out even if all other conditions for it are met.
  inline bool ikeyNotNeededForIncrementalSnapshot();

+  inline bool KeyCommitted(SequenceNumber sequence) {
+    return snapshot_checker_ == nullptr ||
+           snapshot_checker_->CheckInSnapshot(sequence, kMaxSequenceNumber) ==
+               SnapshotCheckerResult::kInSnapshot;
+  }
+
+  bool IsInEarliestSnapshot(SequenceNumber sequence);
+
  InternalIterator* input_;
  const Comparator* cmp_;
  MergeHelper* merge_helper_;
  const std::vector<SequenceNumber>* snapshots_;
+  // List of snapshots released during compaction.
+  // findEarliestVisibleSnapshot() find them out from return of
+  // snapshot_checker, and make sure they will not be returned as
+  // earliest visible snapshot of an older value.
+  // See WritePreparedTransactionTest::ReleaseSnapshotDuringCompaction3.
+  std::unordered_set<SequenceNumber> released_snapshots_;
+  std::vector<SequenceNumber>::const_iterator earliest_snapshot_iter_;
  const SequenceNumber earliest_write_conflict_snapshot_;
  const SnapshotChecker* const snapshot_checker_;
  Env* env_;
@@ -151,6 +167,7 @@ class CompactionIterator {
  bool visible_at_tip_;
  SequenceNumber earliest_snapshot_;
  SequenceNumber latest_snapshot_;
+
  bool ignore_snapshots_;

  // State
@@ -190,13 +190,17 @@ class TestSnapshotChecker : public SnapshotChecker {
      : last_committed_sequence_(last_committed_sequence),
        snapshots_(snapshots) {}

-  bool IsInSnapshot(SequenceNumber seq,
-                    SequenceNumber snapshot_seq) const override {
+  SnapshotCheckerResult CheckInSnapshot(
+      SequenceNumber seq, SequenceNumber snapshot_seq) const override {
    if (snapshot_seq == kMaxSequenceNumber) {
-      return seq <= last_committed_sequence_;
+      return seq <= last_committed_sequence_
+                 ? SnapshotCheckerResult::kInSnapshot
+                 : SnapshotCheckerResult::kNotInSnapshot;
    }
    assert(snapshots_.count(snapshot_seq) > 0);
-    return seq <= snapshots_.at(snapshot_seq);
+    return seq <= snapshots_.at(snapshot_seq)
+               ? SnapshotCheckerResult::kInSnapshot
+               : SnapshotCheckerResult::kNotInSnapshot;
  }

 private:
@@ -1189,10 +1189,12 @@ Status CompactionJob::FinishCompactionOutputFile(
    Slice lower_bound_guard, upper_bound_guard;
    std::string smallest_user_key;
    const Slice *lower_bound, *upper_bound;
+    bool lower_bound_from_sub_compact = false;
    if (sub_compact->outputs.size() == 1) {
      // For the first output table, include range tombstones before the min key
      // but after the subcompaction boundary.
      lower_bound = sub_compact->start;
+      lower_bound_from_sub_compact = true;
    } else if (meta->smallest.size() > 0) {
      // For subsequent output tables, only include range tombstones from min
      // key onwards since the previous file was extended to contain range
@@ -1204,19 +1206,10 @@ Status CompactionJob::FinishCompactionOutputFile(
      lower_bound = nullptr;
    }
    if (next_table_min_key != nullptr) {
-      // This may be the last file in the subcompaction in some cases, so we
-      // need to compare the end key of subcompaction with the next file start
-      // key. When the end key is chosen by the subcompaction, we know that
-      // it must be the biggest key in output file. Therefore, it is safe to
-      // use the smaller key as the upper bound of the output file, to ensure
-      // that there is no overlapping between different output files.
+      // This isn't the last file in the subcompaction, so extend until the next
+      // file starts.
      upper_bound_guard = ExtractUserKey(*next_table_min_key);
-      if (sub_compact->end != nullptr &&
-          ucmp->Compare(upper_bound_guard, *sub_compact->end) >= 0) {
-        upper_bound = sub_compact->end;
-      } else {
-        upper_bound = &upper_bound_guard;
-      }
+      upper_bound = &upper_bound_guard;
    } else {
      // This is the last file in the subcompaction, so extend until the
      // subcompaction ends.
@@ -1234,13 +1227,6 @@ Status CompactionJob::FinishCompactionOutputFile(
      has_overlapping_endpoints = false;
    }

-    // The end key of the subcompaction must be bigger or equal to the upper
-    // bound. If the end of subcompaction is null or the upper bound is null,
-    // it means that this file is the last file in the compaction. So there
-    // will be no overlapping between this file and others.
-    assert(sub_compact->end == nullptr ||
-           upper_bound == nullptr ||
-           ucmp->Compare(*upper_bound , *sub_compact->end) <= 0);
    auto it = range_del_agg->NewIterator(lower_bound, upper_bound,
                                         has_overlapping_endpoints);
    // Position the range tombstone output iterator. There may be tombstone
@@ -1286,11 +1272,24 @@ Status CompactionJob::FinishCompactionOutputFile(
        // (the max key in the previous table or subcompaction) in order for
        // files to appear key-space partitioned.
        //
-        // Choose lowest seqnum so this file's smallest internal key comes
-        // after the previous file's/subcompaction's largest. The fake seqnum
-        // is OK because the read path's file-picking code only considers user
-        // key.
-        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+        // When lower_bound is chosen by a subcompaction, we know that
+        // subcompactions over smaller keys cannot contain any keys at
+        // lower_bound. We also know that smaller subcompactions exist, because
+        // otherwise the subcompaction woud be unbounded on the left. As a
+        // result, we know that no other files on the output level will contain
+        // actual keys at lower_bound (an output file may have a largest key of
+        // lower_bound@kMaxSequenceNumber, but this only indicates a large range
+        // tombstone was truncated). Therefore, it is safe to use the
+        // tombstone's sequence number, to ensure that keys at lower_bound at
+        // lower levels are covered by truncated tombstones.
+        //
+        // If lower_bound was chosen by the smallest data key in the file,
+        // choose lowest seqnum so this file's smallest internal key comes after
+        // the previous file's largest. The fake seqnum is OK because the read
+        // path's file-picking code only considers user key.
+        smallest_candidate = InternalKey(
+            *lower_bound, lower_bound_from_sub_compact ? tombstone.seq_ : 0,
+            kTypeRangeDeletion);
      }
      InternalKey largest_candidate = tombstone.SerializeEndKey();
      if (upper_bound != nullptr &&
@@ -1312,9 +1311,23 @@ Status CompactionJob::FinishCompactionOutputFile(
        largest_candidate =
            InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
      }
+#ifndef NDEBUG
+      SequenceNumber smallest_ikey_seqnum = kMaxSequenceNumber;
+      if (meta->smallest.size() > 0) {
+        smallest_ikey_seqnum = GetInternalKeySeqno(meta->smallest.Encode());
+      }
+#endif
      meta->UpdateBoundariesForRange(smallest_candidate, largest_candidate,
                                     tombstone.seq_,
                                     cfd->internal_comparator());
+
+      // The smallest key in a file is used for range tombstone truncation, so
+      // it cannot have a seqnum of 0 (unless the smallest data key in a file
+      // has a seqnum of 0). Otherwise, the truncated tombstone may expose
+      // deleted keys at lower levels.
+      assert(smallest_ikey_seqnum == 0 ||
+             ExtractInternalKeyFooter(meta->smallest.Encode()) !=
+                 PackSequenceAndType(0, kTypeRangeDeletion));
    }
    meta->marked_for_compaction = sub_compact->builder->NeedCompact();
  }
@@ -456,8 +456,7 @@ TEST_F(CompactionJobTest, NonAssocMerge) {

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

  SetLastSequence(5U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -484,7 +483,7 @@ TEST_F(CompactionJobTest, MergeOperandFilter) {

  auto expected_results =
      mock::MakeMockFile({{KeyStr("a", 0U, kTypeValue), test::EncodeInt(8U)},
-                          {KeyStr("b", 2U, kTypeMerge), test::EncodeInt(2U)}});
+                          {KeyStr("b", 2U, kTypeValue), test::EncodeInt(2U)}});

  SetLastSequence(5U);
  auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -59,6 +59,9 @@ class CompactionPickerTest : public testing::Test {
        log_buffer_(InfoLogLevel::INFO_LEVEL, &logger_),
        file_num_(1),
        vstorage_(nullptr) {
+    // ioptions_.compaction_pri = kMinOverlappingRatio has its own set of
+    // tests to cover.
+    ioptions_.compaction_pri = kByCompensatedSize;
    fifo_options_.max_table_files_size = 1;
    mutable_cf_options_.RefreshDerivedOptions(ioptions_);
    ioptions_.cf_paths.emplace_back("dummy",
@@ -85,8 +88,8 @@ class CompactionPickerTest : public testing::Test {

  void Add(int level, uint32_t file_number, const char* smallest,
           const char* largest, uint64_t file_size = 1, uint32_t path_id = 0,
-           SequenceNumber smallest_seq = 100,
-           SequenceNumber largest_seq = 100) {
+           SequenceNumber smallest_seq = 100, SequenceNumber largest_seq = 100,
+           size_t compensated_file_size = 0) {
    assert(level < vstorage_->num_levels());
    FileMetaData* f = new FileMetaData;
    f->fd = FileDescriptor(file_number, path_id, file_size);
@@ -94,7 +97,8 @@ class CompactionPickerTest : public testing::Test {
    f->largest = InternalKey(largest, largest_seq, kTypeValue);
    f->fd.smallest_seqno = smallest_seq;
    f->fd.largest_seqno = largest_seq;
-    f->compensated_file_size = file_size;
+    f->compensated_file_size =
+        (compensated_file_size != 0) ? compensated_file_size : file_size;
    f->refs = 0;
    vstorage_->AddFile(level, f);
    files_.emplace_back(f);
@@ -617,6 +621,35 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping3) {
  ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
 }

+TEST_F(CompactionPickerTest, CompactionPriMinOverlapping4) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
+  mutable_cf_options_.max_bytes_for_level_base = 10000000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+
+  // file 7 and 8 over lap with the same file, but file 8 is smaller so
+  // it will be picked.
+  // Overlaps with file 26, 27. And the file is compensated so will be
+  // picked up.
+  Add(2, 6U, "150", "167", 60000000U, 0, 100, 100, 180000000U);
+  Add(2, 7U, "168", "169", 60000000U);  // Overlaps with file 27
+  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 28
+
+  Add(3, 26U, "160", "165", 60000000U);
+  // Boosted file size in output level is not considered.
+  Add(3, 27U, "166", "170", 60000000U, 0, 100, 100, 260000000U);
+  Add(3, 28U, "180", "400", 60000000U);
+  Add(3, 29U, "401", "500", 60000000U);
+  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(6U, 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) {
@@ -24,6 +24,8 @@
 #include "rocksdb/env.h"
 #include "rocksdb/table.h"
 #include "rocksdb/write_batch.h"
+#include "table/block_based_table_builder.h"
+#include "table/meta_blocks.h"
 #include "util/filename.h"
 #include "util/string_util.h"
 #include "util/testharness.h"
@@ -467,6 +469,39 @@ TEST_F(CorruptionTest, UnrelatedKeys) {
  ASSERT_EQ(Value(1000, &tmp2).ToString(), v);
 }

+TEST_F(CorruptionTest, RangeDeletionCorrupted) {
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  std::vector<LiveFileMetaData> metadata;
+  db_->GetLiveFilesMetaData(&metadata);
+  ASSERT_EQ(static_cast<size_t>(1), metadata.size());
+  std::string filename = dbname_ + metadata[0].name;
+
+  std::unique_ptr<RandomAccessFile> file;
+  ASSERT_OK(options_.env->NewRandomAccessFile(filename, &file, EnvOptions()));
+  std::unique_ptr<RandomAccessFileReader> file_reader(
+      new RandomAccessFileReader(std::move(file), filename));
+
+  uint64_t file_size;
+  ASSERT_OK(options_.env->GetFileSize(filename, &file_size));
+
+  BlockHandle range_del_handle;
+  ASSERT_OK(FindMetaBlock(
+      file_reader.get(), file_size, kBlockBasedTableMagicNumber,
+      ImmutableCFOptions(options_), kRangeDelBlock, &range_del_handle));
+
+  ASSERT_OK(TryReopen());
+  CorruptFile(filename, static_cast<int>(range_del_handle.offset()), 1);
+  // The test case does not fail on TryReopen because failure to preload table
+  // handlers is not considered critical.
+  ASSERT_OK(TryReopen());
+  std::string val;
+  // However, it does fail on any read involving that file since that file
+  // cannot be opened with a corrupt range deletion meta-block.
+  ASSERT_TRUE(db_->Get(ReadOptions(), "a", &val).IsCorruption());
+}
+
 TEST_F(CorruptionTest, FileSystemStateCorrupted) {
  for (int iter = 0; iter < 2; ++iter) {
    Options options;
@@ -6,6 +6,7 @@
 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
+// #include <iostream>
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
@@ -956,6 +957,184 @@ TEST_F(DBBasicTest, DBCloseFlushError) {
  Destroy(options);
 }

+TEST_F(DBBasicTest, MultiGetMultiCF) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  rocksdb::DBImpl* db = reinterpret_cast<DBImpl*>(db_);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (++get_sv_count == 2) {
+          // After MultiGet refs a couple of CFs, flush all CFs so MultiGet
+          // is forced to repeat the process
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                          "cf" + std::to_string(i) + "_val2"));
+          }
+        }
+        if (get_sv_count == 11) {
+          for (int i = 0; i < 8; ++i) {
+            auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                            db->GetColumnFamilyHandle(i))
+                            ->cfd();
+            ASSERT_EQ(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  values = MultiGet(cfs, keys);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j], "cf" + std::to_string(j) + "_val2");
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVObsolete);
+  }
+}
+
+TEST_F(DBBasicTest, MultiGetMultiCFMutex) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  int retries = 0;
+  bool last_try = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::LastTry", [&](void* /*arg*/) {
+        last_try = true;
+        rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (last_try) {
+          return;
+        }
+        if (++get_sv_count == 2) {
+          ++retries;
+          get_sv_count = 0;
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(
+                i, "cf" + std::to_string(i) + "_key",
+                "cf" + std::to_string(i) + "_val" + std::to_string(retries)));
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  values = MultiGet(cfs, keys);
+  ASSERT_TRUE(last_try);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j],
+              "cf" + std::to_string(j) + "_val" + std::to_string(retries));
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+  }
+}
+
+TEST_F(DBBasicTest, MultiGetMultiCFSnapshot) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  rocksdb::DBImpl* db = reinterpret_cast<DBImpl*>(db_);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (++get_sv_count == 2) {
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                          "cf" + std::to_string(i) + "_val2"));
+          }
+        }
+        if (get_sv_count == 8) {
+          for (int i = 0; i < 8; ++i) {
+            auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                            db->GetColumnFamilyHandle(i))
+                            ->cfd();
+            ASSERT_TRUE(
+                (cfd->TEST_GetLocalSV()->Get() == SuperVersion::kSVInUse) ||
+                (cfd->TEST_GetLocalSV()->Get() == SuperVersion::kSVObsolete));
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  const Snapshot* snapshot = db_->GetSnapshot();
+  values = MultiGet(cfs, keys, snapshot);
+  db_->ReleaseSnapshot(snapshot);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j], "cf" + std::to_string(j) + "_val");
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -173,7 +173,7 @@ TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
  delete iter;
  iter = nullptr;

-  // Release interators and access cache again.
+  // Release iterators and access cache again.
  for (size_t i = 0; i < kNumBlocks - 1; i++) {
    iterators[i].reset();
    CheckCacheCounters(options, 0, 0, 0, 0);
@@ -631,6 +631,74 @@ TEST_F(DBBlockCacheTest, CompressedCache) {
  }
 }

+TEST_F(DBBlockCacheTest, CacheCompressionDict) {
+  const int kNumFiles = 4;
+  const int kNumEntriesPerFile = 32;
+  const int kNumBytesPerEntry = 1024;
+
+  // Try all the available libraries that support dictionary compression
+  std::vector<CompressionType> compression_types;
+#ifdef ZLIB
+  compression_types.push_back(kZlibCompression);
+#endif  // ZLIB
+#if LZ4_VERSION_NUMBER >= 10400
+  compression_types.push_back(kLZ4Compression);
+  compression_types.push_back(kLZ4HCCompression);
+#endif  // LZ4_VERSION_NUMBER >= 10400
+#if ZSTD_VERSION_NUMBER >= 500
+  compression_types.push_back(kZSTD);
+#endif  // ZSTD_VERSION_NUMBER >= 500
+  Random rnd(301);
+  for (auto compression_type : compression_types) {
+    Options options = CurrentOptions();
+    options.compression = compression_type;
+    options.compression_opts.max_dict_bytes = 4096;
+    options.create_if_missing = true;
+    options.num_levels = 2;
+    options.statistics = rocksdb::CreateDBStatistics();
+    options.target_file_size_base = kNumEntriesPerFile * kNumBytesPerEntry;
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.block_cache.reset(new MockCache());
+    options.table_factory.reset(new BlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    for (int i = 0; i < kNumFiles; ++i) {
+      ASSERT_EQ(i, NumTableFilesAtLevel(0, 0));
+      for (int j = 0; j < kNumEntriesPerFile; ++j) {
+        std::string value = RandomString(&rnd, kNumBytesPerEntry);
+        ASSERT_OK(Put(Key(j * kNumFiles + i), value.c_str()));
+      }
+      ASSERT_OK(Flush());
+    }
+    dbfull()->TEST_WaitForCompact();
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(1));
+
+    // Seek to a key in a file. It should cause the SST's dictionary meta-block
+    // to be read.
+    RecordCacheCounters(options);
+    ASSERT_EQ(0,
+              TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD));
+    ASSERT_EQ(
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
+        0);
+    ReadOptions read_options;
+    ASSERT_NE("NOT_FOUND", Get(Key(kNumFiles * kNumEntriesPerFile - 1)));
+    // Two blocks missed/added: dictionary and data block
+    // One block hit: index since it's prefetched
+    CheckCacheCounters(options, 2 /* expected_misses */, 1 /* expected_hits */,
+                       2 /* expected_inserts */, 0 /* expected_failures */);
+    ASSERT_EQ(1,
+              TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD));
+    ASSERT_GT(
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
+        0);
+  }
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -10,8 +10,10 @@
 #include "db/db_test_util.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
+#include "rocksdb/concurrent_task_limiter.h"
 #include "rocksdb/experimental.h"
 #include "rocksdb/utilities/convenience.h"
+#include "util/concurrent_task_limiter_impl.h"
 #include "util/fault_injection_test_env.h"
 #include "util/sync_point.h"

@@ -412,6 +414,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {

  // Reopen the DB with stats-update disabled
  options.skip_stats_update_on_db_open = true;
+  options.max_open_files = 20;
  env_->random_file_open_counter_.store(0);
  Reopen(options);

@@ -437,7 +440,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  Options options = CurrentOptions();
  options.env = env_;
  options.new_table_reader_for_compaction_inputs = true;
-  options.max_open_files = 100;
+  options.max_open_files = 20;
  options.level0_file_num_compaction_trigger = 3;
  DestroyAndReopen(options);
  Random rnd(301);
@@ -466,15 +469,16 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
      Flush();
      dbfull()->TEST_WaitForCompact();
      // preloading iterator issues one table cache lookup and create
-      // a new table reader.
-      ASSERT_EQ(num_table_cache_lookup, 1);
+      // a new table reader, if not preloaded.
+      int old_num_table_cache_lookup = num_table_cache_lookup;
+      ASSERT_GE(num_table_cache_lookup, 1);
      ASSERT_EQ(num_new_table_reader, 1);

      num_table_cache_lookup = 0;
      num_new_table_reader = 0;
      ASSERT_EQ(Key(k), Get(Key(k)));
      // lookup iterator from table cache and no need to create a new one.
-      ASSERT_EQ(num_table_cache_lookup, 1);
+      ASSERT_EQ(old_num_table_cache_lookup + num_table_cache_lookup, 2);
      ASSERT_EQ(num_new_table_reader, 0);
    }
  }
@@ -487,7 +491,10 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  // a new table reader. One file is created for flush and one for compaction.
  // Compaction inputs make no table cache look-up for data/range deletion
  // iterators
-  ASSERT_EQ(num_table_cache_lookup, 2);
+  // May preload table cache too.
+  ASSERT_GE(num_table_cache_lookup, 2);
+  int old_num_table_cache_lookup2 = num_table_cache_lookup;
+
  // Create new iterator for:
  // (1) 1 for verifying flush results
  // (2) 3 for compaction input files
@@ -497,7 +504,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  num_table_cache_lookup = 0;
  num_new_table_reader = 0;
  ASSERT_EQ(Key(1), Get(Key(1)));
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  ASSERT_EQ(num_table_cache_lookup + old_num_table_cache_lookup2, 3);
  ASSERT_EQ(num_new_table_reader, 0);

  num_table_cache_lookup = 0;
@@ -509,14 +516,16 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  db_->CompactRange(cro, nullptr, nullptr);
  // Only verifying compaction outputs issues one table cache lookup
  // for both data block and range deletion block).
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  // May preload table cache too.
+  ASSERT_GE(num_table_cache_lookup, 1);
+  old_num_table_cache_lookup2 = num_table_cache_lookup;
  // One for compaction input, one for verifying compaction results.
  ASSERT_EQ(num_new_table_reader, 2);

  num_table_cache_lookup = 0;
  num_new_table_reader = 0;
  ASSERT_EQ(Key(1), Get(Key(1)));
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  ASSERT_EQ(num_table_cache_lookup + old_num_table_cache_lookup2, 2);
  ASSERT_EQ(num_new_table_reader, 0);

  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
@@ -1594,7 +1603,8 @@ TEST_F(DBCompactionTest, DeleteFileRange) {
  // 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(db_->CompactRange(compact_options, nullptr, nullptr));
+  dbfull()->TEST_WaitForCompact();

  ASSERT_OK(
      DeleteFilesInRange(db_, db_->DefaultColumnFamily(), nullptr, nullptr));
@@ -3890,6 +3900,192 @@ TEST_F(DBCompactionTest, CompactionHasEmptyOutput) {
  ASSERT_EQ(2, collector->num_ssts_creation_started());
 }

+TEST_F(DBCompactionTest, CompactionLimiter) {
+  const int kNumKeysPerFile = 10;
+  const int kMaxBackgroundThreads = 64;
+
+  struct CompactionLimiter {
+    std::string name;
+    int limit_tasks;
+    int max_tasks;
+    int tasks;
+    std::shared_ptr<ConcurrentTaskLimiter> limiter;
+  };
+
+  std::vector<CompactionLimiter> limiter_settings;
+  limiter_settings.push_back({"limiter_1", 1, 0, 0, nullptr});
+  limiter_settings.push_back({"limiter_2", 2, 0, 0, nullptr});
+  limiter_settings.push_back({"limiter_3", 3, 0, 0, nullptr});
+
+  for (auto& ls : limiter_settings) {
+    ls.limiter.reset(NewConcurrentTaskLimiter(ls.name, ls.limit_tasks));
+  }
+
+  std::shared_ptr<ConcurrentTaskLimiter> unique_limiter(
+    NewConcurrentTaskLimiter("unique_limiter", -1));
+
+  const char* cf_names[] = {"default", "0", "1", "2", "3", "4", "5",
+    "6", "7", "8", "9", "a", "b", "c", "d", "e", "f" };
+  const int cf_count = sizeof cf_names / sizeof cf_names[0];
+
+  std::unordered_map<std::string, CompactionLimiter*> cf_to_limiter;
+
+  Options options = CurrentOptions();
+  options.write_buffer_size = 110 * 1024;  // 110KB
+  options.arena_block_size = 4096;
+  options.num_levels = 3;
+  options.level0_file_num_compaction_trigger = 4;
+  options.level0_slowdown_writes_trigger = 64;
+  options.level0_stop_writes_trigger = 64;
+  options.max_background_jobs = kMaxBackgroundThreads; // Enough threads
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.max_write_buffer_number = 10; // Enough memtables
+  DestroyAndReopen(options);
+
+  std::vector<Options> option_vector;
+  option_vector.reserve(cf_count);
+
+  for (int cf = 0; cf < cf_count; cf++) {
+    ColumnFamilyOptions cf_opt(options);
+    if (cf == 0) {
+      // "Default" CF does't use compaction limiter
+      cf_opt.compaction_thread_limiter = nullptr;
+    } else if (cf == 1) {
+      // "1" CF uses bypass compaction limiter
+      unique_limiter->SetMaxOutstandingTask(-1);
+      cf_opt.compaction_thread_limiter = unique_limiter;
+    } else {
+      // Assign limiter by mod
+      auto& ls = limiter_settings[cf % 3];
+      cf_opt.compaction_thread_limiter = ls.limiter;
+      cf_to_limiter[cf_names[cf]] = &ls;
+    }
+    option_vector.emplace_back(DBOptions(options), cf_opt);
+  }
+
+  for (int cf = 1; cf < cf_count; cf++) {
+    CreateColumnFamilies({cf_names[cf]}, option_vector[cf]);
+  }
+
+  ReopenWithColumnFamilies(std::vector<std::string>(cf_names,
+                                                    cf_names + cf_count),
+                           option_vector);
+
+  port::Mutex mutex;
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "DBImpl::BackgroundCompaction:BeforeCompaction", [&](void* arg) {
+      const auto& cf_name = static_cast<ColumnFamilyData*>(arg)->GetName();
+      auto iter = cf_to_limiter.find(cf_name);
+      if (iter != cf_to_limiter.end()) {
+        MutexLock l(&mutex);
+        ASSERT_GE(iter->second->limit_tasks, ++iter->second->tasks);
+        iter->second->max_tasks = std::max(iter->second->max_tasks,
+              iter->second->limit_tasks);
+      }
+    });
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "DBImpl::BackgroundCompaction:AfterCompaction", [&](void* arg) {
+      const auto& cf_name = static_cast<ColumnFamilyData*>(arg)->GetName();
+      auto iter = cf_to_limiter.find(cf_name);
+      if (iter != cf_to_limiter.end()) {
+        MutexLock l(&mutex);
+        ASSERT_GE(--iter->second->tasks, 0);
+      }
+    });
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Block all compact threads in thread pool.
+  const size_t kTotalFlushTasks = kMaxBackgroundThreads / 4;
+  const size_t kTotalCompactTasks = kMaxBackgroundThreads - kTotalFlushTasks;
+  env_->SetBackgroundThreads((int)kTotalFlushTasks, Env::HIGH);
+  env_->SetBackgroundThreads((int)kTotalCompactTasks, Env::LOW);
+
+  test::SleepingBackgroundTask sleeping_compact_tasks[kTotalCompactTasks];
+
+  // Block all compaction threads in thread pool.
+  for (size_t i = 0; i < kTotalCompactTasks; i++) {
+    env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                   &sleeping_compact_tasks[i], Env::LOW);
+    sleeping_compact_tasks[i].WaitUntilSleeping();
+  }
+
+  int keyIndex = 0;
+
+  for (int n = 0; n < options.level0_file_num_compaction_trigger; n++) {
+    for (int cf = 0; cf < cf_count; cf++) {
+      for (int i = 0; i < kNumKeysPerFile; i++) {
+        ASSERT_OK(Put(cf, Key(keyIndex++), ""));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put(cf, "", ""));
+    }
+
+    for (int cf = 0; cf < cf_count; cf++) {
+      dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+    }
+  }
+
+  // Enough L0 files to trigger compaction
+  for (int cf = 0; cf < cf_count; cf++) {
+    ASSERT_EQ(NumTableFilesAtLevel(0, cf),
+      options.level0_file_num_compaction_trigger);
+  }
+
+  // Create more files for one column family, which triggers speed up
+  // condition, all 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(0, Key(i), ""));
+    }
+    // put extra key to trigger flush
+    ASSERT_OK(Put(0, "", ""));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
+    ASSERT_EQ(options.level0_file_num_compaction_trigger + num + 1,
+              NumTableFilesAtLevel(0, 0));
+  }
+
+  // All CFs are pending compaction
+  ASSERT_EQ(cf_count, env_->GetThreadPoolQueueLen(Env::LOW));
+
+  // Unblock all compaction threads
+  for (size_t i = 0; i < kTotalCompactTasks; i++) {
+    sleeping_compact_tasks[i].WakeUp();
+    sleeping_compact_tasks[i].WaitUntilDone();
+  }
+
+  for (int cf = 0; cf < cf_count; cf++) {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+  }
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Max outstanding compact tasks reached limit
+  for (auto& ls : limiter_settings) {
+    ASSERT_EQ(ls.limit_tasks, ls.max_tasks);
+    ASSERT_EQ(0, ls.limiter->GetOutstandingTask());
+  }
+
+  // test manual compaction under a fully throttled limiter
+  int cf_test = 1;
+  unique_limiter->SetMaxOutstandingTask(0);
+
+  // flush one more file to cf 1
+  for (int i = 0; i < kNumKeysPerFile; i++) {
+      ASSERT_OK(Put(cf_test, Key(keyIndex++), ""));
+  }
+  // put extra key to trigger flush
+  ASSERT_OK(Put(cf_test, "", ""));
+
+  dbfull()->TEST_WaitForFlushMemTable(handles_[cf_test]);
+  ASSERT_EQ(1, NumTableFilesAtLevel(0, cf_test));
+
+  Compact(cf_test, Key(0), Key(keyIndex));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+}
+
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                        ::testing::Values(std::make_tuple(1, true),
                                          std::make_tuple(1, false),
@@ -1216,6 +1216,7 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
                       PinnableSlice* pinnable_val, bool* value_found,
                       ReadCallback* callback, bool* is_blob_index) {
  assert(pinnable_val != nullptr);
+  PERF_CPU_TIMER_GUARD(get_cpu_nanos, env_);
  StopWatch sw(env_, stats_, DB_GET);
  PERF_TIMER_GUARD(get_snapshot_time);

@@ -1330,6 +1331,7 @@ std::vector<Status> DBImpl::MultiGet(
    const ReadOptions& read_options,
    const std::vector<ColumnFamilyHandle*>& column_family,
    const std::vector<Slice>& keys, std::vector<std::string>* values) {
+  PERF_CPU_TIMER_GUARD(get_cpu_nanos, env_);
  StopWatch sw(env_, stats_, DB_MULTIGET);
  PERF_TIMER_GUARD(get_snapshot_time);

@@ -1338,33 +1340,96 @@ std::vector<Status> DBImpl::MultiGet(
  struct MultiGetColumnFamilyData {
    ColumnFamilyData* cfd;
    SuperVersion* super_version;
+    MultiGetColumnFamilyData(ColumnFamilyData* cf, SuperVersion* sv)
+        : cfd(cf), super_version(sv) {}
  };
-  std::unordered_map<uint32_t, MultiGetColumnFamilyData*> multiget_cf_data;
-  // fill up and allocate outside of mutex
+  std::unordered_map<uint32_t, MultiGetColumnFamilyData> multiget_cf_data(
+      column_family.size());
  for (auto cf : column_family) {
    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(cf);
    auto cfd = cfh->cfd();
    if (multiget_cf_data.find(cfd->GetID()) == multiget_cf_data.end()) {
-      auto mgcfd = new MultiGetColumnFamilyData();
-      mgcfd->cfd = cfd;
-      multiget_cf_data.insert({cfd->GetID(), mgcfd});
+      multiget_cf_data.emplace(cfd->GetID(),
+                               MultiGetColumnFamilyData(cfd, nullptr));
    }
  }

-  mutex_.Lock();
-  if (read_options.snapshot != nullptr) {
-    snapshot =
-        reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
-  } else {
-    snapshot = last_seq_same_as_publish_seq_
-                   ? versions_->LastSequence()
-                   : versions_->LastPublishedSequence();
+  bool last_try = false;
+  {
+    // If we end up with the same issue of memtable geting sealed during 2
+    // consecutive retries, it means the write rate is very high. In that case
+    // its probably ok to take the mutex on the 3rd try so we can succeed for
+    // sure
+    static const int num_retries = 3;
+    for (auto i = 0; i < num_retries; ++i) {
+      last_try = (i == num_retries - 1);
+      bool retry = false;
+
+      if (i > 0) {
+        for (auto mgd_iter = multiget_cf_data.begin();
+             mgd_iter != multiget_cf_data.end(); ++mgd_iter) {
+          auto super_version = mgd_iter->second.super_version;
+          auto cfd = mgd_iter->second.cfd;
+          if (super_version != nullptr) {
+            ReturnAndCleanupSuperVersion(cfd, super_version);
+          }
+          mgd_iter->second.super_version = nullptr;
+        }
+      }
+
+      if (read_options.snapshot == nullptr) {
+        if (last_try) {
+          TEST_SYNC_POINT("DBImpl::MultiGet::LastTry");
+          // We're close to max number of retries. For the last retry,
+          // acquire the lock so we're sure to succeed
+          mutex_.Lock();
+        }
+        snapshot = last_seq_same_as_publish_seq_
+                       ? versions_->LastSequence()
+                       : versions_->LastPublishedSequence();
+      } else {
+        snapshot = reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                       ->number_;
+      }
+
+      for (auto mgd_iter = multiget_cf_data.begin();
+           mgd_iter != multiget_cf_data.end(); ++mgd_iter) {
+        if (!last_try) {
+          mgd_iter->second.super_version =
+              GetAndRefSuperVersion(mgd_iter->second.cfd);
+        } else {
+          mgd_iter->second.super_version =
+              mgd_iter->second.cfd->GetSuperVersion()->Ref();
+        }
+        TEST_SYNC_POINT("DBImpl::MultiGet::AfterRefSV");
+        if (read_options.snapshot != nullptr || last_try) {
+          // If user passed a snapshot, then we don't care if a memtable is
+          // sealed or compaction happens because the snapshot would ensure
+          // that older key versions are kept around. If this is the last
+          // retry, then we have the lock so nothing bad can happen
+          continue;
+        }
+        // We could get the earliest sequence number for the whole list of
+        // memtables, which will include immutable memtables as well, but that
+        // might be tricky to maintain in case we decide, in future, to do
+        // memtable compaction.
+        if (!last_try) {
+          auto seq =
+              mgd_iter->second.super_version->mem->GetEarliestSequenceNumber();
+          if (seq > snapshot) {
+            retry = true;
+            break;
+          }
+        }
+      }
+      if (!retry) {
+        if (last_try) {
+          mutex_.Unlock();
+        }
+        break;
+      }
+    }
  }
-  for (auto mgd_iter : multiget_cf_data) {
-    mgd_iter.second->super_version =
-        mgd_iter.second->cfd->GetSuperVersion()->Ref();
-  }
-  mutex_.Unlock();

  // Contain a list of merge operations if merge occurs.
  MergeContext merge_context;
@@ -1394,7 +1459,7 @@ std::vector<Status> DBImpl::MultiGet(
    auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
    assert(mgd_iter != multiget_cf_data.end());
    auto mgd = mgd_iter->second;
-    auto super_version = mgd->super_version;
+    auto super_version = mgd.super_version;
    bool skip_memtable =
        (read_options.read_tier == kPersistedTier &&
         has_unpersisted_data_.load(std::memory_order_relaxed));
@@ -1430,24 +1495,14 @@ std::vector<Status> DBImpl::MultiGet(
  PERF_TIMER_GUARD(get_post_process_time);
  autovector<SuperVersion*> superversions_to_delete;

-  // TODO(icanadi) do we need lock here or just around Cleanup()?
-  mutex_.Lock();
  for (auto mgd_iter : multiget_cf_data) {
    auto mgd = mgd_iter.second;
-    if (mgd->super_version->Unref()) {
-      mgd->super_version->Cleanup();
-      superversions_to_delete.push_back(mgd->super_version);
+    if (!last_try) {
+      ReturnAndCleanupSuperVersion(mgd.cfd, mgd.super_version);
+    } else {
+      mgd.cfd->GetSuperVersion()->Unref();
    }
  }
-  mutex_.Unlock();
-
-  for (auto td : superversions_to_delete) {
-    delete td;
-  }
-  for (auto mgd : multiget_cf_data) {
-    delete mgd.second;
-  }
-
  RecordTick(stats_, NUMBER_MULTIGET_CALLS);
  RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
  RecordTick(stats_, NUMBER_MULTIGET_KEYS_FOUND, num_found);
@@ -1866,7 +1921,7 @@ Status DBImpl::NewIterators(
  if (read_options.tailing) {
 #ifdef ROCKSDB_LITE
    return Status::InvalidArgument(
-        "Tailing interator not supported in RocksDB lite");
+        "Tailing iterator not supported in RocksDB lite");
 #else
    for (auto cfh : column_families) {
      auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
@@ -1905,21 +1960,32 @@ const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
 }
 #endif  // ROCKSDB_LITE

-SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
+SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
+                                      bool lock) {
  int64_t unix_time = 0;
  env_->GetCurrentTime(&unix_time);  // Ignore error
  SnapshotImpl* s = new SnapshotImpl;

-  InstrumentedMutexLock l(&mutex_);
+  if (lock) {
+    mutex_.Lock();
+  }
  // returns null if the underlying memtable does not support snapshot.
  if (!is_snapshot_supported_) {
+    if (lock) {
+      mutex_.Unlock();
+    }
    delete s;
    return nullptr;
  }
  auto snapshot_seq = last_seq_same_as_publish_seq_
                          ? versions_->LastSequence()
                          : versions_->LastPublishedSequence();
-  return snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
+  SnapshotImpl* snapshot =
+      snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
+  if (lock) {
+    mutex_.Unlock();
+  }
+  return snapshot;
 }

 void DBImpl::ReleaseSnapshot(const Snapshot* s) {
@@ -2387,9 +2453,9 @@ Status DBImpl::DeleteFile(std::string name) {
    status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                    &edit, &mutex_, directories_.GetDbDir());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_contexts[0],
-          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(cfd,
+                                         &job_context.superversion_contexts[0],
+                                         *cfd->GetLatestMutableCFOptions());
    }
    FindObsoleteFiles(&job_context, false);
  }  // lock released here
@@ -2472,9 +2538,9 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
    status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                    &edit, &mutex_, directories_.GetDbDir());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_contexts[0],
-          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(cfd,
+                                         &job_context.superversion_contexts[0],
+                                         *cfd->GetLatestMutableCFOptions());
    }
    for (auto* deleted_file : deleted_files) {
      deleted_file->being_compacted = false;
@@ -2864,11 +2930,15 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
 #ifndef ROCKSDB_LITE
  Status s;

-  versions_->options_file_number_ = versions_->NewFileNumber();
+  uint64_t options_file_number = versions_->NewFileNumber();
  std::string options_file_name =
-      OptionsFileName(GetName(), versions_->options_file_number_);
+      OptionsFileName(GetName(), options_file_number);
  // Retry if the file name happen to conflict with an existing one.
  s = GetEnv()->RenameFile(file_name, options_file_name);
+  if (s.ok()) {
+    InstrumentedMutexLock l(&mutex_);
+    versions_->options_file_number_ = options_file_number;
+  }

  if (0 == disable_delete_obsolete_files_) {
    DeleteObsoleteOptionsFiles();
@@ -2919,6 +2989,8 @@ void DumpRocksDBBuildVersion(Logger* log) {
                   ROCKSDB_MINOR, ROCKSDB_PATCH);
  ROCKS_LOG_HEADER(log, "Git sha %s", rocksdb_build_git_sha);
  ROCKS_LOG_HEADER(log, "Compile date %s", rocksdb_build_compile_date);
+#else
+    (void)log;  // ignore "-Wunused-parameter"
 #endif
 }

@@ -3164,8 +3236,7 @@ Status DBImpl::IngestExternalFile(
                                 &mutex_, directories_.GetDbDir());
    }
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(cfd, &sv_context, *mutable_cf_options,
-                                         FlushReason::kExternalFileIngestion);
+      InstallSuperVersionAndScheduleWork(cfd, &sv_context, *mutable_cf_options);
    }

    // Resume writes to the DB
@@ -65,6 +65,7 @@ class Arena;
 class ArenaWrappedDBIter;
 class MemTable;
 class TableCache;
+class TaskLimiterToken;
 class Version;
 class VersionEdit;
 class VersionSet;
@@ -191,13 +192,13 @@ class DBImpl : public DB {
                              const Slice* begin, const Slice* end) override;

  using DB::CompactFiles;
-  virtual Status CompactFiles(const CompactionOptions& compact_options,
-                              ColumnFamilyHandle* column_family,
-                              const std::vector<std::string>& input_file_names,
-                              const int output_level,
-                              const int output_path_id = -1,
-                              std::vector<std::string>* const output_file_names
-                              = nullptr) override;
+  virtual Status CompactFiles(
+      const CompactionOptions& compact_options,
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& input_file_names, const int output_level,
+      const int output_path_id = -1,
+      std::vector<std::string>* const output_file_names = nullptr,
+      CompactionJobInfo* compaction_job_info = nullptr) override;

  virtual Status PauseBackgroundWork() override;
  virtual Status ContinueBackgroundWork() override;
@@ -237,6 +238,13 @@ class DBImpl : public DB {
  virtual Status SyncWAL() override;

  virtual SequenceNumber GetLatestSequenceNumber() const override;
+  virtual SequenceNumber GetLastPublishedSequence() const {
+    if (last_seq_same_as_publish_seq_) {
+      return versions_->LastSequence();
+    } else {
+      return versions_->LastPublishedSequence();
+    }
+  }
  // REQUIRES: joined the main write queue if two_write_queues is disabled, and
  // the second write queue otherwise.
  virtual void SetLastPublishedSequence(SequenceNumber seq);
@@ -692,6 +700,12 @@ class DBImpl : public DB {

  void SetSnapshotChecker(SnapshotChecker* snapshot_checker);

+  // Fill JobContext with snapshot information needed by flush and compaction.
+  void GetSnapshotContext(JobContext* job_context,
+                          std::vector<SequenceNumber>* snapshot_seqs,
+                          SequenceNumber* earliest_write_conflict_snapshot,
+                          SnapshotChecker** snapshot_checker);
+
  // Not thread-safe.
  void SetRecoverableStatePreReleaseCallback(PreReleaseCallback* callback);

@@ -1050,7 +1064,8 @@ class DBImpl : public DB {
                          const std::vector<std::string>& input_file_names,
                          std::vector<std::string>* const output_file_names,
                          const int output_level, int output_path_id,
-                          JobContext* job_context, LogBuffer* log_buffer);
+                          JobContext* job_context, LogBuffer* log_buffer,
+                          CompactionJobInfo* compaction_job_info);

  // Wait for current IngestExternalFile() calls to finish.
  // REQUIRES: mutex_ held
@@ -1103,6 +1118,12 @@ class DBImpl : public DB {
                               const std::vector<CompactionInputFiles>& inputs,
                               bool* sfm_bookkeeping, LogBuffer* log_buffer);

+  // Request compaction tasks token from compaction thread limiter.
+  // It always succeeds if force = true or limiter is disable.
+  bool RequestCompactionToken(ColumnFamilyData* cfd, bool force,
+                              std::unique_ptr<TaskLimiterToken>* token,
+                              LogBuffer* log_buffer);
+
  // Schedule background tasks
  void StartTimedTasks();

@@ -1126,10 +1147,15 @@ class DBImpl : public DB {
  ColumnFamilyData* PopFirstFromCompactionQueue();
  FlushRequest PopFirstFromFlushQueue();

+  // Pick the first unthrottled compaction with task token from queue.
+  ColumnFamilyData* PickCompactionFromQueue(
+      std::unique_ptr<TaskLimiterToken>* token, LogBuffer* log_buffer);
+
  // helper function to call after some of the logs_ were synced
  void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);

-  SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary);
+  SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary,
+                                bool lock = true);

  uint64_t GetMaxTotalWalSize() const;

@@ -1352,7 +1378,7 @@ class DBImpl : public DB {
  // compacted. Consumers of these queues are flush and compaction threads. When
  // column family is put on this queue, we increase unscheduled_flushes_ and
  // unscheduled_compactions_. When these variables are bigger than zero, that
-  // means we need to schedule background threads for compaction and thread.
+  // means we need to schedule background threads for flush and compaction.
  // Once the background threads are scheduled, we decrease unscheduled_flushes_
  // and unscheduled_compactions_. That way we keep track of number of
  // compaction and flush threads we need to schedule. This scheduling is done
@@ -1419,6 +1445,8 @@ class DBImpl : public DB {
    // caller retains ownership of `manual_compaction_state` as it is reused
    // across background compactions.
    ManualCompactionState* manual_compaction_state;  // nullptr if non-manual
+    // task limiter token is requested during compaction picking.
+    std::unique_ptr<TaskLimiterToken> task_token;
  };
  std::deque<ManualCompactionState*> manual_compaction_dequeue_;

@@ -1458,7 +1486,7 @@ class DBImpl : public DB {
  std::atomic<bool> has_unpersisted_data_;

  // if an attempt was made to flush all column families that
-  // the oldest log depends on but uncommited data in the oldest
+  // the oldest log depends on but uncommitted data in the oldest
  // log prevents the log from being released.
  // We must attempt to free the dependent memtables again
  // at a later time after the transaction in the oldest
@@ -1527,8 +1555,7 @@ class DBImpl : public DB {
  // state needs flush or compaction.
  void InstallSuperVersionAndScheduleWork(
      ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-      const MutableCFOptions& mutable_cf_options,
-      FlushReason flush_reason = FlushReason::kOthers);
+      const MutableCFOptions& mutable_cf_options);

 #ifndef ROCKSDB_LITE
  using DB::GetPropertiesOfAllTables;
@@ -1553,6 +1580,13 @@ class DBImpl : public DB {
  bool ShouldntRunManualCompaction(ManualCompactionState* m);
  bool HaveManualCompaction(ColumnFamilyData* cfd);
  bool MCOverlap(ManualCompactionState* m, ManualCompactionState* m1);
+#ifndef ROCKSDB_LITE
+  void BuildCompactionJobInfo(const ColumnFamilyData* cfd, Compaction* c,
+                              const Status& st,
+                              const CompactionJobStats& compaction_job_stats,
+                              const int job_id, const Version* current,
+                              CompactionJobInfo* compaction_job_info) const;
+#endif

  bool ShouldPurge(uint64_t file_number) const;
  void MarkAsGrabbedForPurge(uint64_t file_number);
@@ -20,6 +20,7 @@
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/thread_status_updater.h"
 #include "monitoring/thread_status_util.h"
+#include "util/concurrent_task_limiter_impl.h"
 #include "util/sst_file_manager_impl.h"
 #include "util/sync_point.h"

@@ -59,6 +60,27 @@ bool DBImpl::EnoughRoomForCompaction(
  return enough_room;
 }

+bool DBImpl::RequestCompactionToken(ColumnFamilyData* cfd, bool force,
+                                    std::unique_ptr<TaskLimiterToken>* token,
+                                    LogBuffer* log_buffer) {
+  assert(*token == nullptr);
+  auto limiter = static_cast<ConcurrentTaskLimiterImpl*>(
+      cfd->ioptions()->compaction_thread_limiter.get());
+  if (limiter == nullptr) {
+    return true;
+  }
+  *token = limiter->GetToken(force);
+  if (*token != nullptr) {
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "Thread limiter [%s] increase [%s] compaction task, "
+                     "force: %s, tasks after: %d",
+                     limiter->GetName().c_str(), cfd->GetName().c_str(),
+                     force ? "true" : "false", limiter->GetOutstandingTask());
+    return true;
+  }
+  return false;
+}
+
 Status DBImpl::SyncClosedLogs(JobContext* job_context) {
  TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
  mutex_.AssertHeld();
@@ -115,14 +137,12 @@ Status DBImpl::FlushMemTableToOutputFile(
  assert(cfd->imm()->NumNotFlushed() != 0);
  assert(cfd->imm()->IsFlushPending());

+  std::vector<SequenceNumber> snapshot_seqs;
  SequenceNumber earliest_write_conflict_snapshot;
-  std::vector<SequenceNumber> snapshot_seqs =
-      snapshots_.GetAll(&earliest_write_conflict_snapshot);
+  SnapshotChecker* snapshot_checker;
+  GetSnapshotContext(job_context, &snapshot_seqs,
+                     &earliest_write_conflict_snapshot, &snapshot_checker);

-  auto snapshot_checker = snapshot_checker_.get();
-  if (use_custom_gc_ && snapshot_checker == nullptr) {
-    snapshot_checker = DisableGCSnapshotChecker::Instance();
-  }
  FlushJob flush_job(
      dbname_, cfd, immutable_db_options_, mutable_cf_options,
      nullptr /* memtable_id */, env_options_for_compaction_, versions_.get(),
@@ -265,14 +285,12 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
  }
 #endif /* !NDEBUG */

+  std::vector<SequenceNumber> snapshot_seqs;
  SequenceNumber earliest_write_conflict_snapshot;
-  std::vector<SequenceNumber> snapshot_seqs =
-      snapshots_.GetAll(&earliest_write_conflict_snapshot);
+  SnapshotChecker* snapshot_checker;
+  GetSnapshotContext(job_context, &snapshot_seqs,
+                     &earliest_write_conflict_snapshot, &snapshot_checker);

-  auto snapshot_checker = snapshot_checker_.get();
-  if (use_custom_gc_ && snapshot_checker == nullptr) {
-    snapshot_checker = DisableGCSnapshotChecker::Instance();
-  }
  autovector<Directory*> distinct_output_dirs;
  std::vector<FlushJob> jobs;
  std::vector<MutableCFOptions> all_mutable_cf_options;
@@ -310,18 +328,21 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
    jobs.back().PickMemTable();
  }

-  std::vector<FileMetaData> file_meta(num_cfs);
+  autovector<FileMetaData> file_meta;
  Status s;
  assert(num_cfs == static_cast<int>(jobs.size()));

-#ifndef ROCKSDB_LITE
  for (int i = 0; i != num_cfs; ++i) {
-    const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.at(i);
+    file_meta.emplace_back();
+
+#ifndef ROCKSDB_LITE
+    const MutableCFOptions& mutable_cf_options =
+        *cfds[i]->GetLatestMutableCFOptions();
    // may temporarily unlock and lock the mutex.
    NotifyOnFlushBegin(cfds[i], &file_meta[i], mutable_cf_options,
                       job_context->job_id, jobs[i].GetTableProperties());
-  }
 #endif /* !ROCKSDB_LITE */
+  }

  if (logfile_number_ > 0) {
    // TODO (yanqin) investigate whether we should sync the closed logs for
@@ -425,21 +446,19 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
    autovector<ColumnFamilyData*> tmp_cfds;
    autovector<const autovector<MemTable*>*> mems_list;
    autovector<const MutableCFOptions*> mutable_cf_options_list;
-    autovector<FileMetaData*> tmp_file_meta;
    for (int i = 0; i != num_cfs; ++i) {
      const auto& mems = jobs[i].GetMemTables();
      if (!cfds[i]->IsDropped() && !mems.empty()) {
        tmp_cfds.emplace_back(cfds[i]);
        mems_list.emplace_back(&mems);
        mutable_cf_options_list.emplace_back(&all_mutable_cf_options[i]);
-        tmp_file_meta.emplace_back(&file_meta[i]);
      }
    }

    s = InstallMemtableAtomicFlushResults(
        nullptr /* imm_lists */, tmp_cfds, mutable_cf_options_list, mems_list,
-        versions_.get(), &mutex_, tmp_file_meta,
-        &job_context->memtables_to_free, directories_.GetDbDir(), log_buffer);
+        versions_.get(), &mutex_, file_meta, &job_context->memtables_to_free,
+        directories_.GetDbDir(), log_buffer);
  }

  if (s.ok() || s.IsShutdownInProgress()) {
@@ -451,7 +470,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
      }
      InstallSuperVersionAndScheduleWork(cfds[i],
                                         &job_context->superversion_contexts[i],
-                                         all_mutable_cf_options[i]);
+                                         *cfds[i]->GetLatestMutableCFOptions());
      VersionStorageInfo::LevelSummaryStorage tmp;
      ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
                       cfds[i]->GetName().c_str(),
@@ -467,7 +486,8 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
      if (cfds[i]->IsDropped()) {
        continue;
      }
-      NotifyOnFlushCompleted(cfds[i], &file_meta[i], all_mutable_cf_options[i],
+      NotifyOnFlushCompleted(cfds[i], &file_meta[i],
+                             *cfds[i]->GetLatestMutableCFOptions(),
                             job_context->job_id, jobs[i].GetTableProperties());
      if (sfm) {
        std::string file_path = MakeTableFileName(
@@ -530,6 +550,7 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
  mutex_.Unlock();
  {
    FlushJobInfo info;
+    info.cf_id = cfd->GetID();
    info.cf_name = cfd->GetName();
    // TODO(yhchiang): make db_paths dynamic in case flush does not
    //                 go to L0 in the future.
@@ -581,6 +602,7 @@ void DBImpl::NotifyOnFlushCompleted(ColumnFamilyData* cfd,
  mutex_.Unlock();
  {
    FlushJobInfo info;
+    info.cf_id = cfd->GetID();
    info.cf_name = cfd->GetName();
    // TODO(yhchiang): make db_paths dynamic in case flush does not
    //                 go to L0 in the future.
@@ -754,7 +776,8 @@ Status DBImpl::CompactFiles(const CompactionOptions& compact_options,
                            ColumnFamilyHandle* column_family,
                            const std::vector<std::string>& input_file_names,
                            const int output_level, const int output_path_id,
-                            std::vector<std::string>* const output_file_names) {
+                            std::vector<std::string>* const output_file_names,
+                            CompactionJobInfo* compaction_job_info) {
 #ifdef ROCKSDB_LITE
  (void)compact_options;
  (void)column_family;
@@ -762,6 +785,7 @@ Status DBImpl::CompactFiles(const CompactionOptions& compact_options,
  (void)output_level;
  (void)output_path_id;
  (void)output_file_names;
+  (void)compaction_job_info;
  // not supported in lite version
  return Status::NotSupported("Not supported in ROCKSDB LITE");
 #else
@@ -793,7 +817,7 @@ Status DBImpl::CompactFiles(const CompactionOptions& compact_options,

    s = CompactFilesImpl(compact_options, cfd, current, input_file_names,
                         output_file_names, output_level, output_path_id,
-                         &job_context, &log_buffer);
+                         &job_context, &log_buffer, compaction_job_info);

    current->Unref();
  }
@@ -833,7 +857,8 @@ Status DBImpl::CompactFilesImpl(
    const CompactionOptions& compact_options, ColumnFamilyData* cfd,
    Version* version, const std::vector<std::string>& input_file_names,
    std::vector<std::string>* const output_file_names, const int output_level,
-    int output_path_id, JobContext* job_context, LogBuffer* log_buffer) {
+    int output_path_id, JobContext* job_context, LogBuffer* log_buffer,
+    CompactionJobInfo* compaction_job_info) {
  mutex_.AssertHeld();

  if (shutting_down_.load(std::memory_order_acquire)) {
@@ -907,18 +932,17 @@ Status DBImpl::CompactFilesImpl(
  // deletion compaction currently not allowed in CompactFiles.
  assert(!c->deletion_compaction());

+  std::vector<SequenceNumber> snapshot_seqs;
  SequenceNumber earliest_write_conflict_snapshot;
-  std::vector<SequenceNumber> snapshot_seqs =
-      snapshots_.GetAll(&earliest_write_conflict_snapshot);
+  SnapshotChecker* snapshot_checker;
+  GetSnapshotContext(job_context, &snapshot_seqs,
+                     &earliest_write_conflict_snapshot, &snapshot_checker);

  auto pending_outputs_inserted_elem =
      CaptureCurrentFileNumberInPendingOutputs();

-  auto snapshot_checker = snapshot_checker_.get();
-  if (use_custom_gc_ && snapshot_checker == nullptr) {
-    snapshot_checker = DisableGCSnapshotChecker::Instance();
-  }
  assert(is_snapshot_supported_ || snapshots_.empty());
+  CompactionJobStats compaction_job_stats;
  CompactionJob compaction_job(
      job_context->job_id, c.get(), immutable_db_options_,
      env_options_for_compaction_, versions_.get(), &shutting_down_,
@@ -928,19 +952,7 @@ Status DBImpl::CompactFilesImpl(
      snapshot_checker, table_cache_, &event_logger_,
      c->mutable_cf_options()->paranoid_file_checks,
      c->mutable_cf_options()->report_bg_io_stats, dbname_,
-      nullptr);  // Here we pass a nullptr for CompactionJobStats because
-                 // CompactFiles does not trigger OnCompactionCompleted(),
-                 // which is the only place where CompactionJobStats is
-                 // returned.  The idea of not triggering OnCompationCompleted()
-                 // is that CompactFiles runs in the caller thread, so the user
-                 // should always know when it completes.  As a result, it makes
-                 // less sense to notify the users something they should already
-                 // know.
-                 //
-                 // In the future, if we would like to add CompactionJobStats
-                 // support for CompactFiles, we should have CompactFiles API
-                 // pass a pointer of CompactionJobStats as the out-value
-                 // instead of using EventListener.
+      &compaction_job_stats);

  // Creating a compaction influences the compaction score because the score
  // takes running compactions into account (by skipping files that are already
@@ -961,9 +973,9 @@ Status DBImpl::CompactFilesImpl(

  Status status = compaction_job.Install(*c->mutable_cf_options());
  if (status.ok()) {
-    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_contexts[0],
-        *c->mutable_cf_options(), FlushReason::kManualCompaction);
+    InstallSuperVersionAndScheduleWork(c->column_family_data(),
+                                       &job_context->superversion_contexts[0],
+                                       *c->mutable_cf_options());
  }
  c->ReleaseCompactionFiles(s);
 #ifndef ROCKSDB_LITE
@@ -977,6 +989,11 @@ Status DBImpl::CompactFilesImpl(

  ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);

+  if (compaction_job_info != nullptr) {
+    BuildCompactionJobInfo(cfd, c.get(), s, compaction_job_stats,
+                           job_context->job_id, version, compaction_job_info);
+  }
+
  if (status.ok()) {
    // Done
  } else if (status.IsShutdownInProgress()) {
@@ -1119,36 +1136,8 @@ void DBImpl::NotifyOnCompactionCompleted(
  TEST_SYNC_POINT("DBImpl::NotifyOnCompactionCompleted::UnlockMutex");
  {
    CompactionJobInfo info;
-    info.cf_id = cfd->GetID();
-    info.cf_name = cfd->GetName();
-    info.status = st;
-    info.thread_id = env_->GetThreadID();
-    info.job_id = job_id;
-    info.base_input_level = c->start_level();
-    info.output_level = c->output_level();
-    info.stats = compaction_job_stats;
-    info.table_properties = c->GetOutputTableProperties();
-    info.compaction_reason = c->compaction_reason();
-    info.compression = c->output_compression();
-    for (size_t i = 0; i < c->num_input_levels(); ++i) {
-      for (const auto fmd : *c->inputs(i)) {
-        auto fn = TableFileName(c->immutable_cf_options()->cf_paths,
-                                fmd->fd.GetNumber(), fmd->fd.GetPathId());
-        info.input_files.push_back(fn);
-        if (info.table_properties.count(fn) == 0) {
-          std::shared_ptr<const TableProperties> tp;
-          auto s = current->GetTableProperties(&tp, fmd, &fn);
-          if (s.ok()) {
-            info.table_properties[fn] = tp;
-          }
-        }
-      }
-    }
-    for (const auto newf : c->edit()->GetNewFiles()) {
-      info.output_files.push_back(TableFileName(
-          c->immutable_cf_options()->cf_paths, newf.second.fd.GetNumber(),
-          newf.second.fd.GetPathId()));
-    }
+    BuildCompactionJobInfo(cfd, c, st, compaction_job_stats, job_id, current,
+                           &info);
    for (auto listener : immutable_db_options_.listeners) {
      listener->OnCompactionCompleted(this, info);
    }
@@ -1403,6 +1392,8 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
  ROCKS_LOG_INFO(immutable_db_options_.info_log,
                 "[%s] Manual compaction starting", cfd->GetName().c_str());

+  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
+                       immutable_db_options_.info_log.get());
  // We don't check bg_error_ here, because if we get the error in compaction,
  // the compaction will set manual.status to bg_error_ and set manual.done to
  // true.
@@ -1440,6 +1431,11 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
      ca->prepicked_compaction = new PrepickedCompaction;
      ca->prepicked_compaction->manual_compaction_state = &manual;
      ca->prepicked_compaction->compaction = compaction;
+      if (!RequestCompactionToken(
+              cfd, true, &ca->prepicked_compaction->task_token, &log_buffer)) {
+        // Don't throttle manual compaction, only count outstanding tasks.
+        assert(false);
+      }
      manual.incomplete = false;
      bg_compaction_scheduled_++;
      env_->Schedule(&DBImpl::BGWorkCompaction, ca, Env::Priority::LOW, this,
@@ -1448,6 +1444,7 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
    }
  }

+  log_buffer.FlushBufferToLog();
  assert(!manual.in_progress);
  assert(HasPendingManualCompaction());
  RemoveManualCompaction(&manual);
@@ -1883,6 +1880,32 @@ DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
  return flush_req;
 }

+ColumnFamilyData* DBImpl::PickCompactionFromQueue(
+    std::unique_ptr<TaskLimiterToken>* token, LogBuffer* log_buffer) {
+  assert(!compaction_queue_.empty());
+  assert(*token == nullptr);
+  autovector<ColumnFamilyData*> throttled_candidates;
+  ColumnFamilyData* cfd = nullptr;
+  while (!compaction_queue_.empty()) {
+    auto first_cfd = *compaction_queue_.begin();
+    compaction_queue_.pop_front();
+    assert(first_cfd->queued_for_compaction());
+    if (!RequestCompactionToken(first_cfd, false, token, log_buffer)) {
+      throttled_candidates.push_back(first_cfd);
+      continue;
+    }
+    cfd = first_cfd;
+    cfd->set_queued_for_compaction(false);
+    break;
+  }
+  // Add throttled compaction candidates back to queue in the original order.
+  for (auto iter = throttled_candidates.rbegin();
+       iter != throttled_candidates.rend(); ++iter) {
+    compaction_queue_.push_front(*iter);
+  }
+  return cfd;
+}
+
 void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
                                  FlushReason flush_reason) {
  if (flush_req.empty()) {
@@ -2141,7 +2164,12 @@ void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
    Status s = BackgroundCompaction(&made_progress, &job_context, &log_buffer,
                                    prepicked_compaction);
    TEST_SYNC_POINT("BackgroundCallCompaction:1");
-    if (!s.ok() && !s.IsShutdownInProgress()) {
+    if (s.IsBusy()) {
+      bg_cv_.SignalAll();  // In case a waiter can proceed despite the error
+      mutex_.Unlock();
+      env_->SleepForMicroseconds(10000); // prevent hot loop
+      mutex_.Lock();
+    } else if (!s.ok() && !s.IsShutdownInProgress()) {
      // Wait a little bit before retrying background compaction in
      // case this is an environmental problem and we do not want to
      // chew up resources for failed compactions for the duration of
@@ -2276,6 +2304,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    manual_compaction->in_progress = true;
  }

+  std::unique_ptr<TaskLimiterToken> task_token;
+
  // InternalKey manual_end_storage;
  // InternalKey* manual_end = &manual_end_storage;
  bool sfm_reserved_compact_space = false;
@@ -2327,17 +2357,23 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
      return Status::OK();
    }

-    // cfd is referenced here
-    auto cfd = PopFirstFromCompactionQueue();
+    auto cfd = PickCompactionFromQueue(&task_token, log_buffer);
+    if (cfd == nullptr) {
+      // Can't find any executable task from the compaction queue.
+      // All tasks have been throttled by compaction thread limiter.
+      ++unscheduled_compactions_;
+      return Status::Busy();
+    }
+
    // We unreference here because the following code will take a Ref() on
    // this cfd if it is going to use it (Compaction class holds a
    // reference).
    // This will all happen under a mutex so we don't have to be afraid of
    // somebody else deleting it.
    if (cfd->Unref()) {
-      delete cfd;
      // This was the last reference of the column family, so no need to
      // compact.
+      delete cfd;
      return Status::OK();
    }

@@ -2407,6 +2443,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
  } else if (c->deletion_compaction()) {
    // TODO(icanadi) Do we want to honor snapshots here? i.e. not delete old
    // file if there is alive snapshot pointing to it
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
+                             c->column_family_data());
    assert(c->num_input_files(1) == 0);
    assert(c->level() == 0);
    assert(c->column_family_data()->ioptions()->compaction_style ==
@@ -2423,15 +2461,19 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    status = versions_->LogAndApply(c->column_family_data(),
                                    *c->mutable_cf_options(), c->edit(),
                                    &mutex_, directories_.GetDbDir());
-    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_contexts[0],
-        *c->mutable_cf_options(), FlushReason::kAutoCompaction);
+    InstallSuperVersionAndScheduleWork(c->column_family_data(),
+                                       &job_context->superversion_contexts[0],
+                                       *c->mutable_cf_options());
    ROCKS_LOG_BUFFER(log_buffer, "[%s] Deleted %d files\n",
                     c->column_family_data()->GetName().c_str(),
                     c->num_input_files(0));
    *made_progress = true;
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
+                             c->column_family_data());
  } else if (!trivial_move_disallowed && c->IsTrivialMove()) {
    TEST_SYNC_POINT("DBImpl::BackgroundCompaction:TrivialMove");
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
+                             c->column_family_data());
    // Instrument for event update
    // TODO(yhchiang): add op details for showing trivial-move.
    ThreadStatusUtil::SetColumnFamily(
@@ -2473,9 +2515,9 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                                    *c->mutable_cf_options(), c->edit(),
                                    &mutex_, directories_.GetDbDir());
    // Use latest MutableCFOptions
-    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_contexts[0],
-        *c->mutable_cf_options(), FlushReason::kAutoCompaction);
+    InstallSuperVersionAndScheduleWork(c->column_family_data(),
+                                       &job_context->superversion_contexts[0],
+                                       *c->mutable_cf_options());

    VersionStorageInfo::LevelSummaryStorage tmp;
    c->column_family_data()->internal_stats()->IncBytesMoved(c->output_level(),
@@ -2497,6 +2539,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,

    // Clear Instrument
    ThreadStatusUtil::ResetThreadStatus();
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
+                             c->column_family_data());
  } else if (!is_prepicked && c->output_level() > 0 &&
             c->output_level() ==
                 c->column_family_data()
@@ -2514,22 +2558,23 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    ca->prepicked_compaction = new PrepickedCompaction;
    ca->prepicked_compaction->compaction = c.release();
    ca->prepicked_compaction->manual_compaction_state = nullptr;
+    // Transfer requested token, so it doesn't need to do it again.
+    ca->prepicked_compaction->task_token = std::move(task_token);
    ++bg_bottom_compaction_scheduled_;
    env_->Schedule(&DBImpl::BGWorkBottomCompaction, ca, Env::Priority::BOTTOM,
                   this, &DBImpl::UnscheduleCallback);
  } else {
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
+                             c->column_family_data());
    int output_level __attribute__((__unused__));
    output_level = c->output_level();
    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:NonTrivial",
                             &output_level);
+    std::vector<SequenceNumber> snapshot_seqs;
    SequenceNumber earliest_write_conflict_snapshot;
-    std::vector<SequenceNumber> snapshot_seqs =
-        snapshots_.GetAll(&earliest_write_conflict_snapshot);
-
-    auto snapshot_checker = snapshot_checker_.get();
-    if (use_custom_gc_ && snapshot_checker == nullptr) {
-      snapshot_checker = DisableGCSnapshotChecker::Instance();
-    }
+    SnapshotChecker* snapshot_checker;
+    GetSnapshotContext(job_context, &snapshot_seqs,
+                       &earliest_write_conflict_snapshot, &snapshot_checker);
    assert(is_snapshot_supported_ || snapshots_.empty());
    CompactionJob compaction_job(
        job_context->job_id, c.get(), immutable_db_options_,
@@ -2553,11 +2598,13 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,

    status = compaction_job.Install(*c->mutable_cf_options());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          c->column_family_data(), &job_context->superversion_contexts[0],
-          *c->mutable_cf_options(), FlushReason::kAutoCompaction);
+      InstallSuperVersionAndScheduleWork(c->column_family_data(),
+                                         &job_context->superversion_contexts[0],
+                                         *c->mutable_cf_options());
    }
    *made_progress = true;
+    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
+                             c->column_family_data());
  }
  if (c != nullptr) {
    c->ReleaseCompactionFiles(status);
@@ -2738,6 +2785,45 @@ bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {
  return true;
 }

+#ifndef ROCKSDB_LITE
+void DBImpl::BuildCompactionJobInfo(
+    const ColumnFamilyData* cfd, Compaction* c, const Status& st,
+    const CompactionJobStats& compaction_job_stats, const int job_id,
+    const Version* current, CompactionJobInfo* compaction_job_info) const {
+  assert(compaction_job_info != nullptr);
+  compaction_job_info->cf_id = cfd->GetID();
+  compaction_job_info->cf_name = cfd->GetName();
+  compaction_job_info->status = st;
+  compaction_job_info->thread_id = env_->GetThreadID();
+  compaction_job_info->job_id = job_id;
+  compaction_job_info->base_input_level = c->start_level();
+  compaction_job_info->output_level = c->output_level();
+  compaction_job_info->stats = compaction_job_stats;
+  compaction_job_info->table_properties = c->GetOutputTableProperties();
+  compaction_job_info->compaction_reason = c->compaction_reason();
+  compaction_job_info->compression = c->output_compression();
+  for (size_t i = 0; i < c->num_input_levels(); ++i) {
+    for (const auto fmd : *c->inputs(i)) {
+      auto fn = TableFileName(c->immutable_cf_options()->cf_paths,
+                              fmd->fd.GetNumber(), fmd->fd.GetPathId());
+      compaction_job_info->input_files.push_back(fn);
+      if (compaction_job_info->table_properties.count(fn) == 0) {
+        shared_ptr<const TableProperties> tp;
+        auto s = current->GetTableProperties(&tp, fmd, &fn);
+        if (s.ok()) {
+          compaction_job_info->table_properties[fn] = tp;
+        }
+      }
+    }
+  }
+  for (const auto& newf : c->edit()->GetNewFiles()) {
+    compaction_job_info->output_files.push_back(
+        TableFileName(c->immutable_cf_options()->cf_paths,
+                      newf.second.fd.GetNumber(), newf.second.fd.GetPathId()));
+  }
+}
+#endif
+
 // SuperVersionContext gets created and destructed outside of the lock --
 // we use this conveniently to:
 // * malloc one SuperVersion() outside of the lock -- new_superversion
@@ -2752,10 +2838,7 @@ bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {

 void DBImpl::InstallSuperVersionAndScheduleWork(
    ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-    const MutableCFOptions& mutable_cf_options,
-    FlushReason /* flush_reason */) {
-  // TODO(yanqin) investigate if 'flush_reason' can be removed since it's not
-  // used.
+    const MutableCFOptions& mutable_cf_options) {
  mutex_.AssertHeld();

  // Update max_total_in_memory_state_
@@ -2822,4 +2905,31 @@ void DBImpl::SetSnapshotChecker(SnapshotChecker* snapshot_checker) {
  assert(!snapshot_checker_);
  snapshot_checker_.reset(snapshot_checker);
 }
+
+void DBImpl::GetSnapshotContext(
+    JobContext* job_context, std::vector<SequenceNumber>* snapshot_seqs,
+    SequenceNumber* earliest_write_conflict_snapshot,
+    SnapshotChecker** snapshot_checker_ptr) {
+  mutex_.AssertHeld();
+  assert(job_context != nullptr);
+  assert(snapshot_seqs != nullptr);
+  assert(earliest_write_conflict_snapshot != nullptr);
+  assert(snapshot_checker_ptr != nullptr);
+
+  *snapshot_checker_ptr = snapshot_checker_.get();
+  if (use_custom_gc_ && *snapshot_checker_ptr == nullptr) {
+    *snapshot_checker_ptr = DisableGCSnapshotChecker::Instance();
+  }
+  if (*snapshot_checker_ptr != nullptr) {
+    // If snapshot_checker is used, that means the flush/compaction may
+    // contain values not visible to snapshot taken after
+    // flush/compaction job starts. Take a snapshot and it will appear
+    // in snapshot_seqs and force compaction iterator to consider such
+    // snapshots.
+    const Snapshot* job_snapshot =
+        GetSnapshotImpl(false /*write_conflict_boundary*/, false /*lock*/);
+    job_context->job_snapshot.reset(new ManagedSnapshot(this, job_snapshot));
+  }
+  *snapshot_seqs = snapshots_.GetAll(earliest_write_conflict_snapshot);
+}
 }  // namespace rocksdb
@@ -41,6 +41,8 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
      max_max_open_files = 0x400000;
    }
    ClipToRange(&result.max_open_files, 20, max_max_open_files);
+    TEST_SYNC_POINT_CALLBACK("SanitizeOptions::AfterChangeMaxOpenFiles",
+                             &result.max_open_files);
  }

  if (result.info_log == nullptr) {
@@ -419,7 +421,10 @@ Status DBImpl::Recover(
    // produced by an older version of rocksdb.
    std::vector<std::string> filenames;
    s = env_->GetChildren(immutable_db_options_.wal_dir, &filenames);
-    if (!s.ok()) {
+    if (s.IsNotFound()) {
+      return Status::InvalidArgument("wal_dir not found",
+                                     immutable_db_options_.wal_dir);
+    } else if (!s.ok()) {
      return s;
    }

@@ -77,8 +77,8 @@ class DBImplReadOnly : public DBImpl {
      ColumnFamilyHandle* /*column_family*/,
      const std::vector<std::string>& /*input_file_names*/,
      const int /*output_level*/, const int /*output_path_id*/ = -1,
-      std::vector<std::string>* const /*output_file_names*/ = nullptr
-      ) override {
+      std::vector<std::string>* const /*output_file_names*/ = nullptr,
+      CompactionJobInfo* /*compaction_job_info*/ = nullptr) override {
    return Status::NotSupported("Not supported operation in read only mode.");
  }

@@ -265,19 +265,18 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    // We're optimistic, updating the stats before we successfully
    // commit.  That lets us release our leader status early.
    auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count,
+    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count,
                      concurrent_update);
    RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
+    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
                      concurrent_update);
    RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-    stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
-                      concurrent_update);
+    stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
    RecordTick(stats_, WRITE_DONE_BY_SELF);
    auto write_done_by_other = write_group.size - 1;
    if (write_done_by_other > 0) {
-      stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
-                        write_done_by_other, concurrent_update);
+      stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
+                        concurrent_update);
      RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
    }
    MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -468,20 +467,20 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
    }

    auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count);
+    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count);
    RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size);
+    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size);
    RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
    MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);

    PERF_TIMER_STOP(write_pre_and_post_process_time);

-    if (w.ShouldWriteToWAL()) {
+    if (w.status.ok() && !write_options.disableWAL) {
      PERF_TIMER_GUARD(write_wal_time);
-      stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1);
+      stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
      RecordTick(stats_, WRITE_DONE_BY_SELF, 1);
      if (wal_write_group.size > 1) {
-        stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
+        stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
                          wal_write_group.size - 1);
        RecordTick(stats_, WRITE_DONE_BY_OTHER, wal_write_group.size - 1);
      }
@@ -505,7 +504,7 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
  WriteThread::WriteGroup memtable_write_group;
  if (w.state == WriteThread::STATE_MEMTABLE_WRITER_LEADER) {
    PERF_TIMER_GUARD(write_memtable_time);
-    assert(w.status.ok());
+    assert(w.ShouldWriteToMemtable());
    write_thread_.EnterAsMemTableWriter(&w, &memtable_write_group);
    if (memtable_write_group.size > 1 &&
        immutable_db_options_.allow_concurrent_memtable_write) {
@@ -592,16 +591,15 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
  // We're optimistic, updating the stats before we successfully
  // commit.  That lets us release our leader status early.
  auto stats = default_cf_internal_stats_;
-  stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
+  stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
                    concurrent_update);
  RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-  stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
-                    concurrent_update);
+  stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
  RecordTick(stats_, WRITE_DONE_BY_SELF);
  auto write_done_by_other = write_group.size - 1;
  if (write_done_by_other > 0) {
-    stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
-                      write_done_by_other, concurrent_update);
+    stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
+                      concurrent_update);
    RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
  }
  MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -910,12 +908,12 @@ Status DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
  if (status.ok()) {
    auto stats = default_cf_internal_stats_;
    if (need_log_sync) {
-      stats->AddDBStats(InternalStats::kIntStatsWalFileSynced, 1);
+      stats->AddDBStats(InternalStats::WAL_FILE_SYNCED, 1);
      RecordTick(stats_, WAL_FILE_SYNCED);
    }
-    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size);
+    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size);
    RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal);
+    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal);
    RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
  }
  return status;
@@ -961,10 +959,9 @@ Status DBImpl::ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
  if (status.ok()) {
    const bool concurrent = true;
    auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size,
-                      concurrent);
+    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size, concurrent);
    RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal,
+    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal,
                      concurrent);
    RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
  }
@@ -1051,22 +1048,22 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
  auto oldest_alive_log = alive_log_files_.begin()->number;
  bool flush_wont_release_oldest_log = false;
  if (allow_2pc()) {
-    auto oldest_log_with_uncommited_prep =
+    auto oldest_log_with_uncommitted_prep =
        logs_with_prep_tracker_.FindMinLogContainingOutstandingPrep();

-    assert(oldest_log_with_uncommited_prep == 0 ||
-           oldest_log_with_uncommited_prep >= oldest_alive_log);
-    if (oldest_log_with_uncommited_prep > 0 &&
-        oldest_log_with_uncommited_prep == oldest_alive_log) {
+    assert(oldest_log_with_uncommitted_prep == 0 ||
+           oldest_log_with_uncommitted_prep >= oldest_alive_log);
+    if (oldest_log_with_uncommitted_prep > 0 &&
+        oldest_log_with_uncommitted_prep == oldest_alive_log) {
      if (unable_to_release_oldest_log_) {
        // we already attempted to flush all column families dependent on
-        // the oldest alive log but the log still contained uncommited
+        // the oldest alive log but the log still contained uncommitted
        // transactions so there is still nothing that we can do.
        return status;
      } else {
        ROCKS_LOG_WARN(
            immutable_db_options_.info_log,
-            "Unable to release oldest log due to uncommited transaction");
+            "Unable to release oldest log due to uncommitted transaction");
        unable_to_release_oldest_log_ = true;
        flush_wont_release_oldest_log = true;
      }
@@ -1258,8 +1255,8 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
  }
  assert(!delayed || !write_options.no_slowdown);
  if (delayed) {
-    default_cf_internal_stats_->AddDBStats(
-        InternalStats::kIntStatsWriteStallMicros, time_delayed);
+    default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_STALL_MICROS,
+                                           time_delayed);
    RecordTick(stats_, STALL_MICROS, time_delayed);
  }

@@ -21,7 +21,8 @@ class TestReadCallback : public ReadCallback {
      : snapshot_checker_(snapshot_checker), snapshot_seq_(snapshot_seq) {}

  bool IsVisible(SequenceNumber seq) override {
-    return snapshot_checker_->IsInSnapshot(seq, snapshot_seq_);
+    return snapshot_checker_->CheckInSnapshot(seq, snapshot_seq_) ==
+           SnapshotCheckerResult::kInSnapshot;
  }

 private:
@@ -547,8 +548,15 @@ TEST_F(DBMergeOperatorTest, SnapshotCheckerAndReadCallback) {
  DestroyAndReopen(options);

  class TestSnapshotChecker : public SnapshotChecker {
-    bool IsInSnapshot(SequenceNumber seq,
-                      SequenceNumber snapshot_seq) const override {
+   public:
+    SnapshotCheckerResult CheckInSnapshot(
+        SequenceNumber seq, SequenceNumber snapshot_seq) const override {
+      return IsInSnapshot(seq, snapshot_seq)
+                 ? SnapshotCheckerResult::kInSnapshot
+                 : SnapshotCheckerResult::kNotInSnapshot;
+    }
+
+    bool IsInSnapshot(SequenceNumber seq, SequenceNumber snapshot_seq) const {
      switch (snapshot_seq) {
        case 0:
          return seq == 0;
@@ -245,15 +245,14 @@ void GetExpectedTableProperties(
  const int kDeletionCount = kTableCount * kDeletionsPerTable;
  const int kMergeCount = kTableCount * kMergeOperandsPerTable;
  const int kRangeDeletionCount = kTableCount * kRangeDeletionsPerTable;
-  const int kKeyCount = kPutCount + kDeletionCount + kMergeCount;
+  const int kKeyCount = kPutCount + kDeletionCount + kMergeCount + kRangeDeletionCount;
  const int kAvgSuccessorSize = kKeySize / 5;
  const int kEncodingSavePerKey = kKeySize / 4;
-  expected_tp->raw_key_size =
-      (kKeyCount + kRangeDeletionCount) * (kKeySize + 8);
+  expected_tp->raw_key_size = kKeyCount * (kKeySize + 8);
  expected_tp->raw_value_size =
      (kPutCount + kMergeCount + kRangeDeletionCount) * kValueSize;
  expected_tp->num_entries = kKeyCount;
-  expected_tp->num_deletions = kDeletionCount;
+  expected_tp->num_deletions = kDeletionCount + kRangeDeletionCount;
  expected_tp->num_merge_operands = kMergeCount;
  expected_tp->num_range_deletions = kRangeDeletionCount;
  expected_tp->num_data_blocks =
@@ -393,7 +392,15 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
  options.target_file_size_base = 98 << 10;
  options.max_write_buffer_number = 2;
  options.statistics = rocksdb::CreateDBStatistics();
-  options.max_open_files = 100;
+  options.max_open_files = 11;  // Make sure no proloading of table readers
+
+  // RocksDB sanitize max open files to at least 20. Modify it back.
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+        int* max_open_files = static_cast<int*>(arg);
+        *max_open_files = 11;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  BlockBasedTableOptions table_options;
  table_options.no_block_cache = true;
@@ -441,6 +448,7 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {

  // Reopen and issue iterating. See thee latency tracked
  ReopenWithColumnFamilies({"default", "pikachu"}, options);
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
  ASSERT_TRUE(dbfull()->GetProperty("rocksdb.cf-file-histogram", &prop));
  ASSERT_EQ(std::string::npos, prop.find("** Level 0 read latency histogram"));
  ASSERT_EQ(std::string::npos, prop.find("** Level 1 read latency histogram"));
@@ -1138,6 +1138,14 @@ TEST_F(DBRangeDelTest, KeyAtOverlappingEndpointReappears) {
  options.target_file_size_base = kFileBytes;
  Reopen(options);

+  // Push dummy data to L3 so that our actual test files on L0-L2
+  // will not be considered "bottommost" level, otherwise compaction
+  // may prevent us from creating overlapping user keys
+  // as on the bottommost layer MergeHelper
+  ASSERT_OK(db_->Merge(WriteOptions(), "key", "dummy"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  MoveFilesToLevel(3);
+
  Random rnd(301);
  const Snapshot* snapshot = nullptr;
  for (int i = 0; i < kNumFiles; ++i) {
@@ -1159,8 +1167,9 @@ TEST_F(DBRangeDelTest, KeyAtOverlappingEndpointReappears) {
  std::string value;
  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());

-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
-                              nullptr /* end_key */));
+  dbfull()->TEST_CompactRange(0 /* level */, nullptr /* begin */,
+                              nullptr /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  // Now we have multiple files at L1 all containing a single user key, thus
  // guaranteeing overlap in the file endpoints.
@@ -385,6 +385,7 @@ TEST_F(DBSSTTest, RateLimitedDelete) {

  // Compaction will move the 4 files in L0 to trash and create 1 L1 file
  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
  ASSERT_EQ("0,1", FilesPerLevel(0));

  uint64_t delete_start_time = env_->NowMicros();
@@ -2606,8 +2606,8 @@ class ModelDB : public DB {
      ColumnFamilyHandle* /*column_family*/,
      const std::vector<std::string>& /*input_file_names*/,
      const int /*output_level*/, const int /*output_path_id*/ = -1,
-      std::vector<std::string>* const /*output_file_names*/ = nullptr
-      ) override {
+      std::vector<std::string>* const /*output_file_names*/ = nullptr,
+      CompactionJobInfo* /*compaction_job_info*/ = nullptr) override {
    return Status::NotSupported("Not supported operation.");
  }

@@ -342,6 +342,7 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
    ASSERT_GE(cache->GetUsage(), 1024 * 1024);
    Close();
    options.write_buffer_manager.reset();
+    last_options_.write_buffer_manager.reset();
    ASSERT_LT(cache->GetUsage(), 1024 * 1024);
  }
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
@@ -1345,11 +1346,49 @@ TEST_F(DBTest2, FirstSnapshotTest) {
  db_->ReleaseSnapshot(s1);
 }

-class PinL0IndexAndFilterBlocksTest : public DBTestBase,
-                                      public testing::WithParamInterface<bool> {
+#ifndef ROCKSDB_LITE
+TEST_F(DBTest2, DuplicateSnapshot) {
+  Options options;
+  options = CurrentOptions(options);
+  std::vector<const Snapshot*> snapshots;
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  SequenceNumber oldest_ww_snap, first_ww_snap;
+
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+  snapshots.push_back(db_->GetSnapshot());
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+  snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
+  first_ww_snap = snapshots.back()->GetSequenceNumber();
+  Put("k", "v");  // inc seq
+  snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
+  snapshots.push_back(db_->GetSnapshot());
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+
+  {
+    InstrumentedMutexLock l(dbi->mutex());
+    auto seqs = dbi->snapshots().GetAll(&oldest_ww_snap);
+    ASSERT_EQ(seqs.size(), 4);  // duplicates are not counted
+    ASSERT_EQ(oldest_ww_snap, first_ww_snap);
+  }
+
+  for (auto s : snapshots) {
+    db_->ReleaseSnapshot(s);
+  }
+}
+#endif  // ROCKSDB_LITE
+
+class PinL0IndexAndFilterBlocksTest
+    : public DBTestBase,
+      public testing::WithParamInterface<std::tuple<bool, bool>> {
 public:
  PinL0IndexAndFilterBlocksTest() : DBTestBase("/db_pin_l0_index_bloom_test") {}
-  virtual void SetUp() override { infinite_max_files_ = GetParam(); }
+  virtual void SetUp() override {
+    infinite_max_files_ = std::get<0>(GetParam());
+    disallow_preload_ = std::get<1>(GetParam());
+  }

  void CreateTwoLevels(Options* options, bool close_afterwards) {
    if (infinite_max_files_) {
@@ -1386,6 +1425,7 @@ class PinL0IndexAndFilterBlocksTest : public DBTestBase,
  }

  bool infinite_max_files_;
+  bool disallow_preload_;
 };

 TEST_P(PinL0IndexAndFilterBlocksTest,
@@ -1476,12 +1516,27 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
  uint64_t im = TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS);
  uint64_t ih = TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT);

+  if (disallow_preload_) {
+    // Now we have two files. We narrow the max open files to allow 3 entries
+    // so that preloading SST files won't happen.
+    options.max_open_files = 13;
+    // RocksDB sanitize max open files to at least 20. Modify it back.
+    rocksdb::SyncPoint::GetInstance()->SetCallBack(
+        "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+          int* max_open_files = static_cast<int*>(arg);
+          *max_open_files = 13;
+        });
+  }
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
  // Reopen database. If max_open_files is set as -1, table readers will be
  // preloaded. This will trigger a BlockBasedTable::Open() and prefetch
  // L0 index and filter. Level 1's prefetching is disabled in DB::Open()
  TryReopenWithColumnFamilies({"default", "pikachu"}, options);

-  if (infinite_max_files_) {
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  if (!disallow_preload_) {
    // After reopen, cache miss are increased by one because we read (and only
    // read) filter and index on L0
    ASSERT_EQ(fm + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
@@ -1509,7 +1564,7 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {

  // this should be read from L1
  value = Get(1, "a");
-  if (infinite_max_files_) {
+  if (!disallow_preload_) {
    // In inifinite max files case, there's a cache miss in executing Get()
    // because index and filter are not prefetched before.
    ASSERT_EQ(fm + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
@@ -1527,10 +1582,45 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
    ASSERT_EQ(im + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
    ASSERT_EQ(ih + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
  }
+
+  // Force a full compaction to one single file. There will be a block
+  // cache read for both of index and filter. If prefetch doesn't explicitly
+  // happen, it will happen when verifying the file.
+  Compact(1, "a", "zzzzz");
+  dbfull()->TEST_WaitForCompact();
+
+  if (!disallow_preload_) {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  } else {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  }
+
+  // Bloom and index hit will happen when a Get() happens.
+  value = Get(1, "a");
+  if (!disallow_preload_) {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  } else {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 4, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  }
 }

 INSTANTIATE_TEST_CASE_P(PinL0IndexAndFilterBlocksTest,
-                        PinL0IndexAndFilterBlocksTest, ::testing::Bool());
+                        PinL0IndexAndFilterBlocksTest,
+                        ::testing::Values(std::make_tuple(true, false),
+                                          std::make_tuple(false, false),
+                                          std::make_tuple(false, true)));

 #ifndef ROCKSDB_LITE
 TEST_F(DBTest2, MaxCompactionBytesTest) {
@@ -1646,6 +1736,42 @@ class MockPersistentCache : public PersistentCache {
  const size_t max_size_ = 10 * 1024;  // 10KiB
 };

+#ifdef OS_LINUX
+// Make sure that in CPU time perf context counters, Env::NowCPUNanos()
+// is used, rather than Env::CPUNanos();
+TEST_F(DBTest2, TestPerfContextCpuTime) {
+  // force resizing table cache so table handle is not preloaded so that
+  // we can measure find_table_nanos during Get().
+  dbfull()->TEST_table_cache()->SetCapacity(0);
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+  env_->now_cpu_count_.store(0);
+
+  // CPU timing is not enabled with kEnableTimeExceptForMutex
+  SetPerfLevel(PerfLevel::kEnableTimeExceptForMutex);
+  ASSERT_EQ("bar", Get("foo"));
+  ASSERT_EQ(0, get_perf_context()->get_cpu_nanos);
+  ASSERT_EQ(0, env_->now_cpu_count_.load());
+
+  uint64_t kDummyAddonTime = uint64_t{1000000000000};
+
+  // Add time to NowNanos() reading.
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "TableCache::FindTable:0",
+      [&](void* /*arg*/) { env_->addon_time_.fetch_add(kDummyAddonTime); });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  SetPerfLevel(PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);
+  ASSERT_EQ("bar", Get("foo"));
+  ASSERT_EQ(env_->now_cpu_count_.load(), 2);
+  ASSERT_LT(get_perf_context()->get_cpu_nanos, kDummyAddonTime);
+  ASSERT_GT(get_perf_context()->find_table_nanos, kDummyAddonTime);
+
+  SetPerfLevel(PerfLevel::kDisable);
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+#endif  // OS_LINUX
+
 #ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
 TEST_F(DBTest2, PersistentCache) {
  int num_iter = 80;
@@ -3088,6 +3214,57 @@ TEST_F(DBTest2, TestCompactFiles) {
 }
 #endif  // ROCKSDB_LITE

+// TODO: figure out why this test fails in appveyor
+#ifndef OS_WIN
+TEST_F(DBTest2, MultiDBParallelOpenTest) {
+  const int kNumDbs = 2;
+  Options options = CurrentOptions();
+  std::vector<std::string> dbnames;
+  for (int i = 0; i < kNumDbs; ++i) {
+    dbnames.emplace_back(test::TmpDir(env_) + "/db" + ToString(i));
+    ASSERT_OK(DestroyDB(dbnames.back(), options));
+  }
+
+  // Verify empty DBs can be created in parallel
+  std::vector<std::thread> open_threads;
+  std::vector<DB*> dbs{static_cast<unsigned int>(kNumDbs), nullptr};
+  options.create_if_missing = true;
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads.emplace_back(
+        [&](int dbnum) {
+          ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
+        },
+        i);
+  }
+
+  // Now add some data and close, so next we can verify non-empty DBs can be
+  // recovered in parallel
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads[i].join();
+    ASSERT_OK(dbs[i]->Put(WriteOptions(), "xi", "gua"));
+    delete dbs[i];
+  }
+
+  // Verify non-empty DBs can be recovered in parallel
+  dbs.clear();
+  open_threads.clear();
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads.emplace_back(
+        [&](int dbnum) {
+          ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
+        },
+        i);
+  }
+
+  // Wait and cleanup
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads[i].join();
+    delete dbs[i];
+    ASSERT_OK(DestroyDB(dbnames[i], options));
+  }
+}
+#endif  // OS_WIN
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -604,6 +604,7 @@ Status DBTestBase::TryReopenWithColumnFamilies(
    column_families.push_back(ColumnFamilyDescriptor(cfs[i], options[i]));
  }
  DBOptions db_opts = DBOptions(options[0]);
+  last_options_ = options[0];
  return DB::Open(db_opts, dbname_, column_families, &handles_, &db_);
 }

@@ -762,6 +763,31 @@ std::string DBTestBase::Get(int cf, const std::string& k,
  return result;
 }

+std::vector<std::string> DBTestBase::MultiGet(std::vector<int> cfs,
+                                              const std::vector<std::string>& k,
+                                              const Snapshot* snapshot) {
+  ReadOptions options;
+  options.verify_checksums = true;
+  options.snapshot = snapshot;
+  std::vector<ColumnFamilyHandle*> handles;
+  std::vector<Slice> keys;
+  std::vector<std::string> result;
+
+  for (unsigned int i = 0; i < cfs.size(); ++i) {
+    handles.push_back(handles_[cfs[i]]);
+    keys.push_back(k[i]);
+  }
+  std::vector<Status> s = db_->MultiGet(options, handles, keys, &result);
+  for (unsigned int i = 0; i < s.size(); ++i) {
+    if (s[i].IsNotFound()) {
+      result[i] = "NOT_FOUND";
+    } else if (!s[i].ok()) {
+      result[i] = s[i].ToString();
+    }
+  }
+  return result;
+}
+
 Status DBTestBase::Get(const std::string& k, PinnableSlice* v) {
  ReadOptions options;
  options.verify_checksums = true;
@@ -503,6 +503,11 @@ class SpecialEnv : public EnvWrapper {
    return s;
  }

+  virtual uint64_t NowCPUNanos() override {
+    now_cpu_count_.fetch_add(1);
+    return target()->NowCPUNanos();
+  }
+
  virtual uint64_t NowNanos() override {
    return (time_elapse_only_sleep_ ? 0 : target()->NowNanos()) +
           addon_time_.load() * 1000;
@@ -572,6 +577,8 @@ class SpecialEnv : public EnvWrapper {

  std::atomic<int64_t> addon_time_;

+  std::atomic<int> now_cpu_count_;
+
  std::atomic<int> delete_count_;

  std::atomic<bool> time_elapse_only_sleep_;
@@ -833,6 +840,10 @@ class DBTestBase : public testing::Test {

  Status Get(const std::string& k, PinnableSlice* v);

+  std::vector<std::string> MultiGet(std::vector<int> cfs,
+                                    const std::vector<std::string>& k,
+                                    const Snapshot* snapshot = nullptr);
+
  uint64_t GetNumSnapshots();

  uint64_t GetTimeOldestSnapshots();
@@ -283,7 +283,31 @@ TEST_F(DBWALTest, RecoverWithTableHandle) {
    ASSERT_OK(Put(1, "bar", "v4"));
    ASSERT_OK(Flush(1));
    ASSERT_OK(Put(1, "big", std::string(100, 'a')));
-    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+
+    options = CurrentOptions();
+    const int kSmallMaxOpenFiles = 13;
+    if (option_config_ == kDBLogDir) {
+      // Use this option to check not preloading files
+      // Set the max open files to be small enough so no preload will
+      // happen.
+      options.max_open_files = kSmallMaxOpenFiles;
+      // RocksDB sanitize max open files to at least 20. Modify it back.
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+            int* max_open_files = static_cast<int*>(arg);
+            *max_open_files = kSmallMaxOpenFiles;
+          });
+
+    } else if (option_config_ == kWalDirAndMmapReads) {
+      // Use this option to check always loading all files.
+      options.max_open_files = 100;
+    } else {
+      options.max_open_files = -1;
+    }
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+    rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();

    std::vector<std::vector<FileMetaData>> files;
    dbfull()->TEST_GetFilesMetaData(handles_[1], &files);
@@ -294,10 +318,10 @@ TEST_F(DBWALTest, RecoverWithTableHandle) {
    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 {
+        if (options.max_open_files == kSmallMaxOpenFiles) {
          ASSERT_TRUE(file.table_reader_handle == nullptr);
+        } else {
+          ASSERT_TRUE(file.table_reader_handle != nullptr);
        }
      }
    }
@@ -17,7 +17,7 @@ namespace rocksdb {
 class ExternalSSTFileBasicTest : public DBTestBase,
                                 public ::testing::WithParamInterface<bool> {
 public:
-  ExternalSSTFileBasicTest() : DBTestBase("/external_sst_file_test") {
+  ExternalSSTFileBasicTest() : DBTestBase("/external_sst_file_basic_test") {
    sst_files_dir_ = dbname_ + "/sst_files/";
    DestroyAndRecreateExternalSSTFilesDir();
  }
@@ -308,9 +308,7 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
    k++;
  }
  HistogramData hist;
-  std::vector<FileMetaData> file_metas;
-  // Call reserve to avoid auto-resizing
-  file_metas.reserve(flush_jobs.size());
+  autovector<FileMetaData> file_metas;
  mutex_.Lock();
  for (auto& job : flush_jobs) {
    job.PickMemTable();
@@ -321,10 +319,6 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
    ASSERT_OK(job.Run(nullptr /**/, &meta));
    file_metas.emplace_back(meta);
  }
-  autovector<FileMetaData*> file_meta_ptrs;
-  for (auto& meta : file_metas) {
-    file_meta_ptrs.push_back(&meta);
-  }
  autovector<const autovector<MemTable*>*> mems_list;
  for (size_t i = 0; i != all_cfds.size(); ++i) {
    const auto& mems = flush_jobs[i].GetMemTables();
@@ -337,7 +331,7 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {

  Status s = InstallMemtableAtomicFlushResults(
      nullptr /* imm_lists */, all_cfds, mutable_cf_options_list, mems_list,
-      versions_.get(), &mutex_, file_meta_ptrs, &job_context.memtables_to_free,
+      versions_.get(), &mutex_, file_metas, &job_context.memtables_to_free,
      nullptr /* db_directory */, nullptr /* log_buffer */);
  ASSERT_OK(s);

@@ -371,17 +365,17 @@ TEST_F(FlushJobTest, Snapshots) {
  auto new_mem = cfd->ConstructNewMemtable(*cfd->GetLatestMutableCFOptions(),
                                           kMaxSequenceNumber);

-  std::vector<SequenceNumber> snapshots;
  std::set<SequenceNumber> snapshots_set;
  int keys = 10000;
  int max_inserts_per_keys = 8;

  Random rnd(301);
  for (int i = 0; i < keys / 2; ++i) {
-    snapshots.push_back(rnd.Uniform(keys * (max_inserts_per_keys / 2)) + 1);
-    snapshots_set.insert(snapshots.back());
+    snapshots_set.insert(rnd.Uniform(keys * (max_inserts_per_keys / 2)) + 1);
  }
-  std::sort(snapshots.begin(), snapshots.end());
+  // set has already removed the duplicate snapshots
+  std::vector<SequenceNumber> snapshots(snapshots_set.begin(),
+                                        snapshots_set.end());

  new_mem->Ref();
  SequenceNumber current_seqno = 0;
@@ -949,17 +949,14 @@ void InternalStats::DumpDBStats(std::string* value) {
           seconds_up, interval_seconds_up);
  value->append(buf);
  // Cumulative
-  uint64_t user_bytes_written =
-      GetDBStats(InternalStats::kIntStatsBytesWritten);
-  uint64_t num_keys_written =
-      GetDBStats(InternalStats::kIntStatsNumKeysWritten);
-  uint64_t write_other = GetDBStats(InternalStats::kIntStatsWriteDoneByOther);
-  uint64_t write_self = GetDBStats(InternalStats::kIntStatsWriteDoneBySelf);
-  uint64_t wal_bytes = GetDBStats(InternalStats::kIntStatsWalFileBytes);
-  uint64_t wal_synced = GetDBStats(InternalStats::kIntStatsWalFileSynced);
-  uint64_t write_with_wal = GetDBStats(InternalStats::kIntStatsWriteWithWal);
-  uint64_t write_stall_micros =
-      GetDBStats(InternalStats::kIntStatsWriteStallMicros);
+  uint64_t user_bytes_written = GetDBStats(InternalStats::BYTES_WRITTEN);
+  uint64_t num_keys_written = GetDBStats(InternalStats::NUMBER_KEYS_WRITTEN);
+  uint64_t write_other = GetDBStats(InternalStats::WRITE_DONE_BY_OTHER);
+  uint64_t write_self = GetDBStats(InternalStats::WRITE_DONE_BY_SELF);
+  uint64_t wal_bytes = GetDBStats(InternalStats::WAL_FILE_BYTES);
+  uint64_t wal_synced = GetDBStats(InternalStats::WAL_FILE_SYNCED);
+  uint64_t write_with_wal = GetDBStats(InternalStats::WRITE_WITH_WAL);
+  uint64_t write_stall_micros = GetDBStats(InternalStats::WRITE_STALL_MICROS);

  const int kHumanMicrosLen = 32;
  char human_micros[kHumanMicrosLen];
@@ -108,15 +108,15 @@ class InternalStats {
  };

  enum InternalDBStatsType {
-    kIntStatsWalFileBytes,
-    kIntStatsWalFileSynced,
-    kIntStatsBytesWritten,
-    kIntStatsNumKeysWritten,
-    kIntStatsWriteDoneByOther,
-    kIntStatsWriteDoneBySelf,
-    kIntStatsWriteWithWal,
-    kIntStatsWriteStallMicros,
-    kIntStatsNumMax,
+    WAL_FILE_BYTES,
+    WAL_FILE_SYNCED,
+    BYTES_WRITTEN,
+    NUMBER_KEYS_WRITTEN,
+    WRITE_DONE_BY_OTHER,
+    WRITE_DONE_BY_SELF,
+    WRITE_WITH_WAL,
+    WRITE_STALL_MICROS,
+    INTERNAL_DB_STATS_ENUM_MAX,
  };

  InternalStats(int num_levels, Env* env, ColumnFamilyData* cfd)
@@ -292,7 +292,7 @@ class InternalStats {
  };

  void Clear() {
-    for (int i = 0; i < kIntStatsNumMax; i++) {
+    for (int i = 0; i < INTERNAL_DB_STATS_ENUM_MAX; i++) {
      db_stats_[i].store(0);
    }
    for (int i = 0; i < INTERNAL_CF_STATS_ENUM_MAX; i++) {
@@ -382,7 +382,7 @@ class InternalStats {
  bool HandleBlockCacheStat(Cache** block_cache);

  // Per-DB stats
-  std::atomic<uint64_t> db_stats_[kIntStatsNumMax];
+  std::atomic<uint64_t> db_stats_[INTERNAL_DB_STATS_ENUM_MAX];
  // Per-ColumnFamily stats
  uint64_t cf_stats_value_[INTERNAL_CF_STATS_ENUM_MAX];
  uint64_t cf_stats_count_[INTERNAL_CF_STATS_ENUM_MAX];
@@ -580,15 +580,15 @@ class InternalStats {
  };

  enum InternalDBStatsType {
-    kIntStatsWalFileBytes,
-    kIntStatsWalFileSynced,
-    kIntStatsBytesWritten,
-    kIntStatsNumKeysWritten,
-    kIntStatsWriteDoneByOther,
-    kIntStatsWriteDoneBySelf,
-    kIntStatsWriteWithWal,
-    kIntStatsWriteStallMicros,
-    kIntStatsNumMax,
+    WAL_FILE_BYTES,
+    WAL_FILE_SYNCED,
+    BYTES_WRITTEN,
+    NUMBER_KEYS_WRITTEN,
+    WRITE_DONE_BY_OTHER,
+    WRITE_DONE_BY_SELF,
+    WRITE_WITH_WAL,
+    WRITE_STALL_MICROS,
+    INTERNAL_DB_STATS_ENUM_MAX,
  };

  InternalStats(int /*num_levels*/, Env* /*env*/, ColumnFamilyData* /*cfd*/) {}
@@ -175,6 +175,9 @@ struct JobContext {
  size_t num_alive_log_files = 0;
  uint64_t size_log_to_delete = 0;

+  // Snapshot taken before flush/compaction job.
+  std::unique_ptr<ManagedSnapshot> job_snapshot;
+
  explicit JobContext(int _job_id, bool create_superversion = false) {
    job_id = _job_id;
    manifest_file_number = 0;
@@ -204,6 +207,7 @@ struct JobContext {

    memtables_to_free.clear();
    logs_to_free.clear();
+    job_snapshot.reset();
  }

  ~JobContext() {
@@ -110,10 +110,10 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
  assert(!ShouldScheduleFlush());

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

@@ -414,7 +414,8 @@ InternalIterator* MemTable::NewIterator(const ReadOptions& read_options,

 FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIterator(
    const ReadOptions& read_options, SequenceNumber read_seq) {
-  if (read_options.ignore_range_deletions || is_range_del_table_empty_) {
+  if (read_options.ignore_range_deletions ||
+      is_range_del_table_empty_.load(std::memory_order_relaxed)) {
    return nullptr;
  }
  auto* unfragmented_iter = new MemTableIterator(
@@ -563,8 +564,8 @@ bool MemTable::Add(SequenceNumber s, ValueType type,
        !first_seqno_.compare_exchange_weak(cur_earliest_seqno, s)) {
    }
  }
-  if (is_range_del_table_empty_ && type == kTypeRangeDeletion) {
-    is_range_del_table_empty_ = false;
+  if (type == kTypeRangeDeletion) {
+    is_range_del_table_empty_.store(false, std::memory_order_relaxed);
  }
  UpdateOldestKeyTime();
  return true;
@@ -411,7 +411,7 @@ class MemTable {
  ConcurrentArena arena_;
  std::unique_ptr<MemTableRep> table_;
  std::unique_ptr<MemTableRep> range_del_table_;
-  bool is_range_del_table_empty_;
+  std::atomic_bool is_range_del_table_empty_;

  // Total data size of all data inserted
  std::atomic<uint64_t> data_size_;
@@ -533,7 +533,7 @@ Status InstallMemtableAtomicFlushResults(
    const autovector<ColumnFamilyData*>& cfds,
    const autovector<const MutableCFOptions*>& mutable_cf_options_list,
    const autovector<const autovector<MemTable*>*>& mems_list, VersionSet* vset,
-    InstrumentedMutex* mu, const autovector<FileMetaData*>& file_metas,
+    InstrumentedMutex* mu, const autovector<FileMetaData>& file_metas,
    autovector<MemTable*>* to_delete, Directory* db_directory,
    LogBuffer* log_buffer) {
  AutoThreadOperationStageUpdater stage_updater(
@@ -553,11 +553,10 @@ Status InstallMemtableAtomicFlushResults(
      assert((*mems_list[k])[0]->GetID() == imm->GetEarliestMemTableID());
    }
 #endif
-    assert(nullptr != file_metas[k]);
    for (size_t i = 0; i != mems_list[k]->size(); ++i) {
      assert(i == 0 || (*mems_list[k])[i]->GetEdits()->NumEntries() == 0);
      (*mems_list[k])[i]->SetFlushCompleted(true);
-      (*mems_list[k])[i]->SetFileNumber(file_metas[k]->fd.GetNumber());
+      (*mems_list[k])[i]->SetFileNumber(file_metas[k].fd.GetNumber());
    }
  }

@@ -123,7 +123,7 @@ class MemTableListVersion {
      const autovector<const MutableCFOptions*>& mutable_cf_options_list,
      const autovector<const autovector<MemTable*>*>& mems_list,
      VersionSet* vset, InstrumentedMutex* mu,
-      const autovector<FileMetaData*>& file_meta,
+      const autovector<FileMetaData>& file_meta,
      autovector<MemTable*>* to_delete, Directory* db_directory,
      LogBuffer* log_buffer);

@@ -301,7 +301,7 @@ class MemTableList {
      const autovector<const MutableCFOptions*>& mutable_cf_options_list,
      const autovector<const autovector<MemTable*>*>& mems_list,
      VersionSet* vset, InstrumentedMutex* mu,
-      const autovector<FileMetaData*>& file_meta,
+      const autovector<FileMetaData>& file_meta,
      autovector<MemTable*>* to_delete, Directory* db_directory,
      LogBuffer* log_buffer);

@@ -337,7 +337,7 @@ extern Status InstallMemtableAtomicFlushResults(
    const autovector<ColumnFamilyData*>& cfds,
    const autovector<const MutableCFOptions*>& mutable_cf_options_list,
    const autovector<const autovector<MemTable*>*>& mems_list, VersionSet* vset,
-    InstrumentedMutex* mu, const autovector<FileMetaData*>& file_meta,
+    InstrumentedMutex* mu, const autovector<FileMetaData>& file_meta,
    autovector<MemTable*>* to_delete, Directory* db_directory,
    LogBuffer* log_buffer);
 }  // namespace rocksdb
@@ -161,23 +161,18 @@ class MemTableListTest : public testing::Test {
      cfds.emplace_back(column_family_set->GetColumnFamily(cf_ids[i]));
      EXPECT_NE(nullptr, cfds[i]);
    }
-    std::vector<FileMetaData> file_metas;
-    file_metas.reserve(cf_ids.size());
+    autovector<FileMetaData> file_metas;
    for (size_t i = 0; i != cf_ids.size(); ++i) {
      FileMetaData meta;
      uint64_t file_num = file_number.fetch_add(1);
      meta.fd = FileDescriptor(file_num, 0, 0);
      file_metas.emplace_back(meta);
    }
-    autovector<FileMetaData*> file_meta_ptrs;
-    for (auto& meta : file_metas) {
-      file_meta_ptrs.push_back(&meta);
-    }
    InstrumentedMutex mutex;
    InstrumentedMutexLock l(&mutex);
    return InstallMemtableAtomicFlushResults(
        &lists, cfds, mutable_cf_options_list, mems_list, &versions, &mutex,
-        file_meta_ptrs, to_delete, nullptr, &log_buffer);
+        file_metas, to_delete, nullptr, &log_buffer);
  }
 };

@@ -166,9 +166,11 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
      break;
    } else if (stop_before > 0 && ikey.sequence <= stop_before &&
               LIKELY(snapshot_checker_ == nullptr ||
-                      snapshot_checker_->IsInSnapshot(ikey.sequence,
-                                                      stop_before))) {
-      // hit an entry that's visible by the previous snapshot, can't touch that
+                      snapshot_checker_->CheckInSnapshot(ikey.sequence,
+                                                         stop_before) !=
+                          SnapshotCheckerResult::kNotInSnapshot)) {
+      // hit an entry that's possibly visible by the previous snapshot, can't
+      // touch that
      break;
    }

@@ -280,22 +282,24 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
    return Status::OK();
  }

-  // We are sure we have seen this key's entire history if we are at the
-  // last level and exhausted all internal keys of this user key.
-  // NOTE: !iter->Valid() does not necessarily mean we hit the
-  // beginning of a user key, as versions of a user key might be
-  // split into multiple files (even files on the same level)
-  // and some files might not be included in the compaction/merge.
+  // We are sure we have seen this key's entire history if:
+  // at_bottom == true (this does not necessarily mean it is the bottommost
+  // layer, but rather that we are confident the key does not appear on any of
+  // the lower layers, at_bottom == false doesn't mean it does appear, just
+  // that we can't be sure, see Compaction::IsBottommostLevel for details)
+  // AND
+  // we have either encountered another key or end of key history on this
+  // layer.
  //
-  // There are also cases where we have seen the root of history of this
-  // key without being sure of it. Then, we simply miss the opportunity
+  // When these conditions are true we are able to merge all the keys
+  // using full merge.
+  //
+  // For these cases we are not sure about, we simply miss the opportunity
  // to combine the keys. Since VersionSet::SetupOtherInputs() always makes
  // sure that all merge-operands on the same level get compacted together,
  // this will simply lead to these merge operands moving to the next level.
-  //
-  // So, we only perform the following logic (to merge all operands together
-  // without a Put/Delete) if we are certain that we have seen the end of key.
-  bool surely_seen_the_beginning = hit_the_next_user_key && at_bottom;
+  bool surely_seen_the_beginning = (hit_the_next_user_key || !iter->Valid())
+                                    && at_bottom;
  if (surely_seen_the_beginning) {
    // do a final merge with nullptr as the existing value and say
    // bye to the merge type (it's now converted to a Put)
@@ -130,7 +130,7 @@ TEST_F(MergeHelperTest, SingleOperand) {

  AddKeyVal("a", 50, kTypeMerge, test::EncodeInt(1U));

-  ASSERT_TRUE(Run(31, true).IsMergeInProgress());
+  ASSERT_TRUE(Run(31, false).IsMergeInProgress());
  ASSERT_FALSE(iter_->Valid());
  ASSERT_EQ(test::KeyStr("a", 50, kTypeMerge), merge_helper_->keys()[0]);
  ASSERT_EQ(test::EncodeInt(1U), merge_helper_->values()[0]);
@@ -696,6 +696,10 @@ TEST_F(PerfContextTest, PerfContextByLevelGetSet) {
  PERF_COUNTER_BY_LEVEL_ADD(bloom_filter_useful, 1, 7);
  PERF_COUNTER_BY_LEVEL_ADD(bloom_filter_useful, 1, 7);
  PERF_COUNTER_BY_LEVEL_ADD(bloom_filter_full_true_positive, 1, 2);
+  PERF_COUNTER_BY_LEVEL_ADD(block_cache_hit_count, 1, 0);
+  PERF_COUNTER_BY_LEVEL_ADD(block_cache_hit_count, 5, 2);
+  PERF_COUNTER_BY_LEVEL_ADD(block_cache_miss_count, 2, 3);
+  PERF_COUNTER_BY_LEVEL_ADD(block_cache_miss_count, 4, 1);
  ASSERT_EQ(
      0, (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
  ASSERT_EQ(
@@ -706,6 +710,14 @@ TEST_F(PerfContextTest, PerfContextByLevelGetSet) {
                   .bloom_filter_full_positive);
  ASSERT_EQ(1, (*(get_perf_context()->level_to_perf_context))[2]
                   .bloom_filter_full_true_positive);
+  ASSERT_EQ(1, (*(get_perf_context()->level_to_perf_context))[0]
+                  .block_cache_hit_count);
+  ASSERT_EQ(5, (*(get_perf_context()->level_to_perf_context))[2]
+                  .block_cache_hit_count);
+  ASSERT_EQ(2, (*(get_perf_context()->level_to_perf_context))[3]
+                  .block_cache_miss_count);
+  ASSERT_EQ(4, (*(get_perf_context()->level_to_perf_context))[1]
+                  .block_cache_miss_count);
  std::string zero_excluded = get_perf_context()->ToString(true);
  ASSERT_NE(std::string::npos,
            zero_excluded.find("bloom_filter_useful = 1@level5, 2@level7"));
@@ -713,6 +725,10 @@ TEST_F(PerfContextTest, PerfContextByLevelGetSet) {
            zero_excluded.find("bloom_filter_full_positive = 1@level0"));
  ASSERT_NE(std::string::npos,
            zero_excluded.find("bloom_filter_full_true_positive = 1@level2"));
+  ASSERT_NE(std::string::npos,
+            zero_excluded.find("block_cache_hit_count = 1@level0, 5@level2"));
+  ASSERT_NE(std::string::npos,
+            zero_excluded.find("block_cache_miss_count = 4@level1, 2@level3"));
 }
 }  // namespace rocksdb

@@ -158,6 +158,8 @@ class PlainTableDBTest : public testing::Test,
    db_ = nullptr;
  }

+  bool mmap_mode() const { return mmap_mode_; }
+
  void DestroyAndReopen(Options* options = nullptr) {
    //Destroy using last options
    Destroy(&last_options_);
@@ -174,6 +176,12 @@ class PlainTableDBTest : public testing::Test,
    return DB::Open(*options, dbname_, db);
  }

+  Status ReopenForReadOnly(Options* options) {
+    delete db_;
+    db_ = nullptr;
+    return DB::OpenForReadOnly(*options, dbname_, &db_);
+  }
+
  Status TryReopen(Options* options = nullptr) {
    delete db_;
    db_ = nullptr;
@@ -547,6 +555,50 @@ TEST_P(PlainTableDBTest, Flush2) {
  }
 }

+TEST_P(PlainTableDBTest, Immortal) {
+  for (EncodingType encoding_type : {kPlain, kPrefix}) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.max_open_files = -1;
+    // Set only one bucket to force bucket conflict.
+    // Test index interval for the same prefix to be 1, 2 and 4
+    PlainTableOptions plain_table_options;
+    plain_table_options.hash_table_ratio = 0.75;
+    plain_table_options.index_sparseness = 16;
+    plain_table_options.user_key_len = kPlainTableVariableLength;
+    plain_table_options.bloom_bits_per_key = 10;
+    plain_table_options.encoding_type = encoding_type;
+    options.table_factory.reset(NewPlainTableFactory(plain_table_options));
+
+    DestroyAndReopen(&options);
+    ASSERT_OK(Put("0000000000000bar", "b"));
+    ASSERT_OK(Put("1000000000000foo", "v1"));
+    dbfull()->TEST_FlushMemTable();
+
+    int copied = 0;
+    rocksdb::SyncPoint::GetInstance()->SetCallBack(
+        "GetContext::SaveValue::PinSelf", [&](void* /*arg*/) { copied++; });
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+    ASSERT_EQ("b", Get("0000000000000bar"));
+    ASSERT_EQ("v1", Get("1000000000000foo"));
+    ASSERT_EQ(2, copied);
+    copied = 0;
+
+    Close();
+    ASSERT_OK(ReopenForReadOnly(&options));
+
+    ASSERT_EQ("b", Get("0000000000000bar"));
+    ASSERT_EQ("v1", Get("1000000000000foo"));
+    ASSERT_EQ("NOT_FOUND", Get("1000000000000bar"));
+    if (mmap_mode()) {
+      ASSERT_EQ(0, copied);
+    } else {
+      ASSERT_EQ(2, copied);
+    }
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  }
+}
+
 TEST_P(PlainTableDBTest, Iterator) {
  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
       huge_page_tlb_size += 2 * 1024 * 1024) {
@@ -470,8 +470,6 @@ CompactionRangeDelAggregator::NewIterator(const Slice* lower_bound,
      new TruncatedRangeDelMergingIter(icmp_, lower_bound, upper_bound,
                                       upper_bound_inclusive, parent_iters_));

-  // TODO: add tests where tombstone fragments can be outside of upper and lower
-  // bound range
  auto fragmented_tombstone_list =
      std::make_shared<FragmentedRangeTombstoneList>(
          std::move(merging_iter), *icmp_, true /* for_compaction */,
@@ -197,7 +197,8 @@ int main(int argc, char** argv) {
      // real workloads.
      for (int j = 0; j < FLAGS_num_range_tombstones; j++) {
        uint64_t start = rnd.Uniform(FLAGS_tombstone_start_upper_bound);
-        uint64_t end = start + static_cast<uint64_t>(std::max(1.0, normal_dist(random_gen)));
+        uint64_t end = static_cast<uint64_t>(
+            std::round(start + std::max(1.0, normal_dist(random_gen))));
        persistent_range_tombstones[j] = rocksdb::PersistentRangeTombstone(
            rocksdb::Key(start), rocksdb::Key(end), j);
      }
@@ -8,22 +8,30 @@

 namespace rocksdb {

-// Callback class that control GC of duplicate keys in flush/compaction
+enum class SnapshotCheckerResult : int {
+  kInSnapshot = 0,
+  kNotInSnapshot = 1,
+  // In case snapshot is released and the checker has no clue whether
+  // the given sequence is visible to the snapshot.
+  kSnapshotReleased = 2,
+};
+
+// Callback class that control GC of duplicate keys in flush/compaction.
 class SnapshotChecker {
 public:
  virtual ~SnapshotChecker() {}
-  virtual bool IsInSnapshot(SequenceNumber sequence,
-                            SequenceNumber snapshot_sequence) const = 0;
+  virtual SnapshotCheckerResult CheckInSnapshot(
+      SequenceNumber sequence, SequenceNumber snapshot_sequence) const = 0;
 };

 class DisableGCSnapshotChecker : public SnapshotChecker {
 public:
  virtual ~DisableGCSnapshotChecker() {}
-  virtual bool IsInSnapshot(
+  virtual SnapshotCheckerResult CheckInSnapshot(
      SequenceNumber /*sequence*/,
      SequenceNumber /*snapshot_sequence*/) const override {
-    // By returning false, we prevent all the values from being GCed
-    return false;
+    // By returning kNotInSnapshot, we prevent all the values from being GCed
+    return SnapshotCheckerResult::kNotInSnapshot;
  }
  static DisableGCSnapshotChecker* Instance() { return &instance_; }

@@ -41,8 +49,8 @@ class WritePreparedSnapshotChecker : public SnapshotChecker {
  explicit WritePreparedSnapshotChecker(WritePreparedTxnDB* txn_db);
  virtual ~WritePreparedSnapshotChecker() {}

-  virtual bool IsInSnapshot(SequenceNumber sequence,
-                            SequenceNumber snapshot_sequence) const override;
+  virtual SnapshotCheckerResult CheckInSnapshot(
+      SequenceNumber sequence, SequenceNumber snapshot_sequence) const override;

 private:
 #ifndef ROCKSDB_LITE
@@ -86,7 +86,7 @@ class SnapshotList {
  }

  // retrieve all snapshot numbers up until max_seq. They are sorted in
-  // ascending order.
+  // ascending order (with no duplicates).
  std::vector<SequenceNumber> GetAll(
      SequenceNumber* oldest_write_conflict_snapshot = nullptr,
      const SequenceNumber& max_seq = kMaxSequenceNumber) const {
@@ -104,7 +104,10 @@ class SnapshotList {
      if (s->next_->number_ > max_seq) {
        break;
      }
-      ret.push_back(s->next_->number_);
+      // Avoid duplicates
+      if (ret.empty() || ret.back() != s->next_->number_) {
+        ret.push_back(s->next_->number_);
+      }

      if (oldest_write_conflict_snapshot != nullptr &&
          *oldest_write_conflict_snapshot == kMaxSequenceNumber &&
@@ -147,7 +147,7 @@ Status TableCache::FindTable(const EnvOptions& env_options,
                             HistogramImpl* file_read_hist, bool skip_filters,
                             int level,
                             bool prefetch_index_and_filter_in_cache) {
-  PERF_TIMER_GUARD(find_table_nanos);
+  PERF_TIMER_GUARD_WITH_ENV(find_table_nanos, ioptions_.env);
  Status s;
  uint64_t number = fd.GetNumber();
  Slice key = GetSliceForFileNumber(&number);
@@ -122,6 +122,8 @@ class TableCache {
  // Release the handle from a cache
  void ReleaseHandle(Cache::Handle* handle);

+  Cache* get_cache() const { return cache_; }
+
  // Capacity of the backing Cache that indicates inifinite TableCache capacity.
  // For example when max_open_files is -1 we set the backing Cache to this.
  static const int kInfiniteCapacity = 0x400000;
@@ -366,8 +366,40 @@ class VersionBuilder::Rep {

  void LoadTableHandlers(InternalStats* internal_stats, int max_threads,
                         bool prefetch_index_and_filter_in_cache,
+                         bool is_initial_load,
                         const SliceTransform* prefix_extractor) {
    assert(table_cache_ != nullptr);
+
+    size_t table_cache_capacity = table_cache_->get_cache()->GetCapacity();
+    bool always_load = (table_cache_capacity == TableCache::kInfiniteCapacity);
+    size_t max_load = port::kMaxSizet;
+
+    if (!always_load) {
+      // If it is initial loading and not set to always laoding all the
+      // files, we only load up to kInitialLoadLimit files, to limit the
+      // time reopening the DB.
+      const size_t kInitialLoadLimit = 16;
+      size_t load_limit;
+      // If the table cache is not 1/4 full, we pin the table handle to
+      // file metadata to avoid the cache read costs when reading the file.
+      // The downside of pinning those files is that LRU won't be followed
+      // for those files. This doesn't matter much because if number of files
+      // of the DB excceeds table cache capacity, eventually no table reader
+      // will be pinned and LRU will be followed.
+      if (is_initial_load) {
+        load_limit = std::min(kInitialLoadLimit, table_cache_capacity / 4);
+      } else {
+        load_limit = table_cache_capacity / 4;
+      }
+
+      size_t table_cache_usage = table_cache_->get_cache()->GetUsage();
+      if (table_cache_usage >= load_limit) {
+        return;
+      } else {
+        max_load = load_limit - table_cache_usage;
+      }
+    }
+
    // <file metadata, level>
    std::vector<std::pair<FileMetaData*, int>> files_meta;
    for (int level = 0; level < num_levels_; level++) {
@@ -375,6 +407,12 @@ class VersionBuilder::Rep {
        auto* file_meta = file_meta_pair.second;
        assert(!file_meta->table_reader_handle);
        files_meta.emplace_back(file_meta, level);
+        if (files_meta.size() >= max_load) {
+          break;
+        }
+      }
+      if (files_meta.size() >= max_load) {
+        break;
      }
    }

@@ -452,9 +490,11 @@ void VersionBuilder::SaveTo(VersionStorageInfo* vstorage) {
 void VersionBuilder::LoadTableHandlers(InternalStats* internal_stats,
                                       int max_threads,
                                       bool prefetch_index_and_filter_in_cache,
+                                       bool is_initial_load,
                                       const SliceTransform* prefix_extractor) {
  rep_->LoadTableHandlers(internal_stats, max_threads,
-                          prefetch_index_and_filter_in_cache, prefix_extractor);
+                          prefetch_index_and_filter_in_cache, is_initial_load,
+                          prefix_extractor);
 }

 void VersionBuilder::MaybeAddFile(VersionStorageInfo* vstorage, int level,
@@ -35,6 +35,7 @@ class VersionBuilder {
  void SaveTo(VersionStorageInfo* vstorage);
  void LoadTableHandlers(InternalStats* internal_stats, int max_threads,
                         bool prefetch_index_and_filter_in_cache,
+                         bool is_initial_load,
                         const SliceTransform* prefix_extractor);
  void MaybeAddFile(VersionStorageInfo* vstorage, int level, FileMetaData* f);

@@ -263,7 +263,7 @@ const char* VersionEdit::DecodeNewFile4From(Slice* input) {
        break;
      }
      if (!GetLengthPrefixedSlice(input, &field)) {
-        return "new-file4 custom field lenth prefixed slice error";
+        return "new-file4 custom field length prefixed slice error";
      }
      switch (custom_tag) {
        case kPathId:
@@ -1887,9 +1887,9 @@ void SortFileByOverlappingRatio(
      next_level_it++;
    }

-    assert(file->fd.file_size != 0);
+    assert(file->compensated_file_size != 0);
    file_to_order[file->fd.GetNumber()] =
-        overlapping_bytes * 1024u / file->fd.file_size;
+        overlapping_bytes * 1024u / file->compensated_file_size;
  }

  std::sort(temp->begin(), temp->end(),
@@ -3013,9 +3013,7 @@ Status VersionSet::ProcessManifestWrites(
    mu->Unlock();

    TEST_SYNC_POINT("VersionSet::LogAndApply:WriteManifest");
-    if (!first_writer.edit_list.front()->IsColumnFamilyManipulation() &&
-        column_family_set_->get_table_cache()->GetCapacity() ==
-            TableCache::kInfiniteCapacity) {
+    if (!first_writer.edit_list.front()->IsColumnFamilyManipulation()) {
      for (int i = 0; i < static_cast<int>(versions.size()); ++i) {
        assert(!builder_guards.empty() &&
               builder_guards.size() == versions.size());
@@ -3025,6 +3023,7 @@ Status VersionSet::ProcessManifestWrites(
        builder_guards[i]->version_builder()->LoadTableHandlers(
            cfd->internal_stats(), cfd->ioptions()->optimize_filters_for_hits,
            true /* prefetch_index_and_filter_in_cache */,
+            false /* is_initial_load */,
            mutable_cf_options_ptrs[i]->prefix_extractor.get());
      }
    }
@@ -3682,15 +3681,13 @@ Status VersionSet::Recover(
      assert(builders_iter != builders.end());
      auto* builder = builders_iter->second->version_builder();

-      if (GetColumnFamilySet()->get_table_cache()->GetCapacity() ==
-          TableCache::kInfiniteCapacity) {
-        // unlimited table cache. Pre-load table handle now.
-        // Need to do it out of the mutex.
-        builder->LoadTableHandlers(
-            cfd->internal_stats(), db_options_->max_file_opening_threads,
-            false /* prefetch_index_and_filter_in_cache */,
-            cfd->GetLatestMutableCFOptions()->prefix_extractor.get());
-      }
+      // unlimited table cache. Pre-load table handle now.
+      // Need to do it out of the mutex.
+      builder->LoadTableHandlers(
+          cfd->internal_stats(), db_options_->max_file_opening_threads,
+          false /* prefetch_index_and_filter_in_cache */,
+          true /* is_initial_load */,
+          cfd->GetLatestMutableCFOptions()->prefix_extractor.get());

      Version* v = new Version(cfd, this, env_options_,
                               *cfd->GetLatestMutableCFOptions(),
@@ -127,7 +127,7 @@ class WriteThread {
    std::atomic<uint8_t> state;  // write under StateMutex() or pre-link
    WriteGroup* write_group;
    SequenceNumber sequence;  // the sequence number to use for the first key
-    Status status;            // status of memtable inserter
+    Status status;
    Status callback_status;   // status returned by callback->Callback()

    std::aligned_storage<sizeof(std::mutex)>::type state_mutex_bytes;
@@ -31,4 +31,11 @@
   }(document, 'script', 'facebook-jssdk'));
 </script>

-<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
+<script>!function(d,s,id){
+  var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';
+  if(!d.getElementById(id)){
+  js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';
+  fjs.parentNode.insertBefore(js,fjs);
+  }
+  }(document, 'script', 'twitter-wjs');
+</script>
@@ -2,4 +2,11 @@
  <a href="https://twitter.com/{{ include.href }}" class="twitter-follow-button pluginBlock" data-show-count="false">Follow @{{ include.href }}</a>
 </div>

-<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
+<script>!function(d,s,id){
+  var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';
+  if(!d.getElementById(id)){
+  js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';
+  fjs.parentNode.insertBefore(js,fjs);
+  }
+  }(document, 'script', 'twitter-wjs');
+</script>
@@ -1,4 +1,11 @@
 <div class="pluginWrapper twitterSharePlugin">
  <a href="https://twitter.com/share" class="twitter-share-button" data-hashtags="{{ site.title| replace: ' ', '' }}">Tweet</a>
 </div>
-<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
+<script>!function(d,s,id){
+  var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';
+  if(!d.getElementById(id)){js=d.createElement(s);
+  js.id=id;js.src=p+'://platform.twitter.com/widgets.js';
+  fjs.parentNode.insertBefore(js,fjs);
+}
+}(document, 'script', 'twitter-wjs');
+</script>
@@ -21,4 +21,11 @@
  fjs.parentNode.insertBefore(js, fjs);
 }(document, 'script', 'facebook-jssdk'));</script>

-<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
+<script>!function(d,s,id){
+  var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';
+  if(!d.getElementById(id)){
+  js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';
+  fjs.parentNode.insertBefore(js,fjs);
+  }
+  }(document, 'script', 'twitter-wjs');
+</script>
@@ -843,6 +843,15 @@ class PosixEnv : public Env {
 #endif
  }

+  virtual uint64_t NowCPUNanos() override {
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_AIX)
+    struct timespec ts;
+    clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts);
+    return static_cast<uint64_t>(ts.tv_sec) * 1000000000 + ts.tv_nsec;
+#endif
+    return 0;
+  }
+
  virtual void SleepForMicroseconds(int micros) override { usleep(micros); }

  virtual Status GetHostName(char* name, uint64_t len) override {
@@ -528,8 +528,8 @@ struct AdvancedColumnFamilyOptions {

  // If level compaction_style = kCompactionStyleLevel, for each level,
  // which files are prioritized to be picked to compact.
-  // Default: kByCompensatedSize
-  CompactionPri compaction_pri = kByCompensatedSize;
+  // Default: kMinOverlappingRatio
+  CompactionPri compaction_pri = kMinOverlappingRatio;

  // The options needed to support Universal Style compactions
  //
@@ -637,7 +637,6 @@ extern ROCKSDB_LIBRARY_API rocksdb_iterator_t* rocksdb_writebatch_wi_create_iter
    rocksdb_iterator_t* base_iterator,
    rocksdb_column_family_handle_t* cf);

-
 /* Block based table options */

 extern ROCKSDB_LIBRARY_API rocksdb_block_based_table_options_t*
@@ -729,6 +728,9 @@ extern ROCKSDB_LIBRARY_API void rocksdb_options_set_cuckoo_table_factory(
 extern ROCKSDB_LIBRARY_API void rocksdb_set_options(
    rocksdb_t* db, int count, const char* const keys[], const char* const values[], char** errptr);

+extern ROCKSDB_LIBRARY_API void rocksdb_set_options_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* handle, int count, const char* const keys[], const char* const values[], char** errptr);
+
 extern ROCKSDB_LIBRARY_API rocksdb_options_t* rocksdb_options_create();
 extern ROCKSDB_LIBRARY_API void rocksdb_options_destroy(rocksdb_options_t*);
 extern ROCKSDB_LIBRARY_API void rocksdb_options_increase_parallelism(
@@ -0,0 +1,47 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#pragma once
+
+#include "rocksdb/env.h"
+#include "rocksdb/statistics.h"
+
+namespace rocksdb {
+
+class ConcurrentTaskLimiter {
+ public:
+
+  virtual ~ConcurrentTaskLimiter() {}
+
+  // Returns a name that identifies this concurrent task limiter.
+  virtual const std::string& GetName() const = 0;
+
+  // Set max concurrent tasks.
+  // limit = 0 means no new task allowed.
+  // limit < 0 means no limitation.
+  virtual void SetMaxOutstandingTask(int32_t limit) = 0;
+
+  // Reset to unlimited max concurrent task.
+  virtual void ResetMaxOutstandingTask() = 0;
+
+  // Returns current outstanding task count.
+  virtual int32_t GetOutstandingTask() const = 0;
+};
+
+// Create a ConcurrentTaskLimiter that can be shared with mulitple CFs
+// across RocksDB instances to control concurrent tasks.
+//
+// @param name: Name of the limiter.
+// @param limit: max concurrent tasks.
+//        limit = 0 means no new task allowed.
+//        limit < 0 means no limitation.
+extern ConcurrentTaskLimiter* NewConcurrentTaskLimiter(
+    const std::string& name, int32_t limit);
+
+}  // namespace rocksdb
@@ -53,6 +53,9 @@ class WriteBatch;
 class Env;
 class EventListener;
 class TraceWriter;
+#ifdef ROCKSDB_LITE
+class CompactionJobInfo;
+#endif

 using std::unique_ptr;

@@ -834,18 +837,20 @@ class DB {
  virtual Status CompactFiles(
      const CompactionOptions& compact_options,
      ColumnFamilyHandle* column_family,
-      const std::vector<std::string>& input_file_names,
-      const int output_level, const int output_path_id = -1,
-      std::vector<std::string>* const output_file_names = nullptr) = 0;
+      const std::vector<std::string>& input_file_names, const int output_level,
+      const int output_path_id = -1,
+      std::vector<std::string>* const output_file_names = nullptr,
+      CompactionJobInfo* compaction_job_info = nullptr) = 0;

  virtual Status CompactFiles(
      const CompactionOptions& compact_options,
-      const std::vector<std::string>& input_file_names,
-      const int output_level, const int output_path_id = -1,
-      std::vector<std::string>* const output_file_names = nullptr) {
+      const std::vector<std::string>& input_file_names, const int output_level,
+      const int output_path_id = -1,
+      std::vector<std::string>* const output_file_names = nullptr,
+      CompactionJobInfo* compaction_job_info = nullptr) {
    return CompactFiles(compact_options, DefaultColumnFamily(),
                        input_file_names, output_level, output_path_id,
-                        output_file_names);
+                        output_file_names, compaction_job_info);
  }

  // This function will wait until all currently running background processes
@@ -384,6 +384,9 @@ class Env {
    return NowMicros() * 1000;
  }

+  // 0 indicates not supported.
+  virtual uint64_t NowCPUNanos() { return 0; }
+
  // Sleep/delay the thread for the prescribed number of micro-seconds.
  virtual void SleepForMicroseconds(int micros) = 0;

@@ -160,6 +160,8 @@ struct FileOperationInfo {
 };

 struct FlushJobInfo {
+  // the id of the column family
+  uint32_t cf_id;
  // the name of the column family
  std::string cf_name;
  // the path to the newly created file
@@ -108,6 +108,23 @@ class MergeOperator {
    Slice& existing_operand;
  };

+  // This function applies a stack of merge operands in chrionological order
+  // on top of an existing value. There are two ways in which this method is
+  // being used:
+  // a) During Get() operation, it used to calculate the final value of a key
+  // b) During compaction, in order to collapse some operands with the based
+  //    value.
+  //
+  // Note: The name of the method is somewhat misleading, as both in the cases
+  // of Get() or compaction it may be called on a subset of operands:
+  // K:    0    +1    +2    +7    +4     +5      2     +1     +2
+  //                              ^
+  //                              |
+  //                          snapshot
+  // In the example above, Get(K) operation will call FullMerge with a base
+  // value of 2 and operands [+1, +2]. Compaction process might decide to
+  // collapse the beginning of the history up to the snapshot by performing
+  // full Merge with base value of 0 and operands [+1, +2, +7, +3].
  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
                           MergeOperationOutput* merge_out) const;

@@ -182,11 +199,11 @@ class MergeOperator {
  //       consistent MergeOperator between DB opens.
  virtual const char* Name() const = 0;

-  // Determines whether the MergeOperator can be called with just a single
+  // Determines whether the PartialMerge can be called with just a single
  // merge operand.
-  // Override and return true for allowing a single operand. Both FullMergeV2
-  // and PartialMerge/PartialMergeMulti should be overridden and implemented
-  // correctly to handle a single operand.
+  // Override and return true for allowing a single operand. PartialMerge
+  // and PartialMergeMulti should be overridden and implemented
+  // correctly to properly handle a single operand.
  virtual bool AllowSingleOperand() const { return false; }

  // Allows to control when to invoke a full merge during Get.
@@ -34,6 +34,7 @@ class Cache;
 class CompactionFilter;
 class CompactionFilterFactory;
 class Comparator;
+class ConcurrentTaskLimiter;
 class Env;
 enum InfoLogLevel : unsigned char;
 class SstFileManager;
@@ -293,6 +294,14 @@ struct ColumnFamilyOptions : public AdvancedColumnFamilyOptions {
  // Default: empty
  std::vector<DbPath> cf_paths;

+  // Compaction concurrent thread limiter for the column family.
+  // If non-nullptr, use given concurrent thread limiter to control 
+  // the max outstanding compaction tasks. Limiter can be shared with
+  // multiple column families across db instances.
+  //
+  // Default: nullptr
+  std::shared_ptr<ConcurrentTaskLimiter> compaction_thread_limiter = nullptr;
+
  // Create ColumnFamilyOptions with default values for all fields
  ColumnFamilyOptions();
  // Create ColumnFamilyOptions from Options
@@ -35,6 +35,9 @@ struct PerfContextByLevel {
  // total nanos spent on reading data from SST files
  uint64_t get_from_table_nanos;

+  uint64_t block_cache_hit_count = 0;     // total number of block cache hits
+  uint64_t block_cache_miss_count = 0;    // total number of block cache misses
+
  void Reset(); // reset all performance counters to zero
 };

@@ -60,6 +63,12 @@ struct PerfContext {
  uint64_t block_read_count;          // total number of block reads (with IO)
  uint64_t block_read_byte;           // total number of bytes from block reads
  uint64_t block_read_time;           // total nanos spent on block reads
+  uint64_t block_cache_index_hit_count; // total number of index block hits
+  uint64_t index_block_read_count;      // total number of index block reads
+  uint64_t block_cache_filter_hit_count; // total number of filter block hits
+  uint64_t filter_block_read_count;     // total number of filter block reads
+  uint64_t compression_dict_block_read_count;  // total number of compression
+                                               // dictionary block reads
  uint64_t block_checksum_time;       // total nanos spent on block checksum
  uint64_t block_decompress_time;  // total nanos spent on block decompression

@@ -201,6 +210,9 @@ struct PerfContext {
  uint64_t env_lock_file_nanos;
  uint64_t env_unlock_file_nanos;
  uint64_t env_new_logger_nanos;
+
+  uint64_t get_cpu_nanos;
+
  std::map<uint32_t, PerfContextByLevel>* level_to_perf_context = nullptr;
  bool per_level_perf_context_enabled = false;
 };
@@ -17,8 +17,11 @@ enum PerfLevel : unsigned char {
  kEnableCount = 2,               // enable only count stats
  kEnableTimeExceptForMutex = 3,  // Other than count stats, also enable time
                                  // stats except for mutexes
-  kEnableTime = 4,                // enable count and time stats
-  kOutOfBounds = 5                // N.B. Must always be the last value!
+  // Other than time, also measure CPU time counters. Still don't measure
+  // time (neither wall time nor CPU time) for mutexes.
+  kEnableTimeAndCPUTimeExceptForMutex = 4,
+  kEnableTime = 5,  // enable count and time stats
+  kOutOfBounds = 6  // N.B. Must always be the last value!
 };

 // set the perf stats level for current thread
@@ -81,11 +81,11 @@ class RateLimiter {
  // Max bytes can be granted in a single burst
  virtual int64_t GetSingleBurstBytes() const = 0;

-  // Total bytes that go though rate limiter
+  // Total bytes that go through rate limiter
  virtual int64_t GetTotalBytesThrough(
      const Env::IOPriority pri = Env::IO_TOTAL) const = 0;

-  // Total # of requests that go though rate limiter
+  // Total # of requests that go through rate limiter
  virtual int64_t GetTotalRequests(
      const Env::IOPriority pri = Env::IO_TOTAL) const = 0;

@@ -5,11 +5,13 @@
 #ifndef ROCKSDB_LITE
 #pragma once

+#include "rocksdb/options.h"
+
 namespace rocksdb {

 class SSTDumpTool {
 public:
-  int Run(int argc, char** argv);
+  int Run(int argc, char** argv, Options options = Options());
 };

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

 #ifndef ROCKSDB_LITE

-#include "rocksdb/slice.h"
-#include "rocksdb/options.h"
 #include "rocksdb/iterator.h"
+#include "rocksdb/options.h"
+#include "rocksdb/slice.h"
 #include "rocksdb/table_properties.h"

 namespace rocksdb {
@@ -309,7 +309,7 @@ enum Tickers : uint32_t {
  // # of bytes in the blob files evicted because of BlobDB is full.
  BLOB_DB_FIFO_BYTES_EVICTED,

-  // These coutners indicate a performance issue in WritePrepared transactions.
+  // These counters indicate a performance issue in WritePrepared transactions.
  // We should not seem them ticking them much.
  // # of times prepare_mutex_ is acquired in the fast path.
  TXN_PREPARE_MUTEX_OVERHEAD,
@@ -320,12 +320,19 @@ enum Tickers : uint32_t {
  // # of times snapshot_mutex_ is acquired in the fast path.
  TXN_SNAPSHOT_MUTEX_OVERHEAD,

-  // Number of keys actually found in MultiGet calls (vs number requested by caller)
+  // Number of keys actually found in MultiGet calls (vs number requested by
+  // caller)
  // NUMBER_MULTIGET_KEYS_READ gives the number requested by caller
  NUMBER_MULTIGET_KEYS_FOUND,

  NO_ITERATOR_CREATED,  // number of iterators created
  NO_ITERATOR_DELETED,  // number of iterators deleted
+
+  BLOCK_CACHE_COMPRESSION_DICT_MISS,
+  BLOCK_CACHE_COMPRESSION_DICT_HIT,
+  BLOCK_CACHE_COMPRESSION_DICT_ADD,
+  BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT,
+  BLOCK_CACHE_COMPRESSION_DICT_BYTES_EVICT,
  TICKER_ENUM_MAX
 };

@@ -430,6 +437,7 @@ struct HistogramData {
  double max = 0.0;
  uint64_t count = 0;
  uint64_t sum = 0;
+  double min = 0.0;
 };

 enum StatsLevel {
@@ -62,6 +62,10 @@ struct BlockBasedTableOptions {
  // TODO(kailiu) Temporarily disable this feature by making the default value
  // to be false.
  //
+  // TODO(ajkr) we need to update names of variables controlling meta-block
+  // caching as they should now apply to range tombstone and compression
+  // dictionary meta-blocks, in addition to index and filter meta-blocks.
+  //
  // Indicating if we'd put index/filter blocks to the block cache.
  // If not specified, each "table reader" object will pre-load index/filter
  // block during table initialization.
@@ -454,8 +458,8 @@ class TableFactory {
  //     and cache the table object returned.
  // (2) SstFileDumper (for SST Dump) opens the table and dump the table
  //     contents using the iterator of the table.
-  // (3) DBImpl::IngestExternalFile() calls this function to read the contents of
-  //     the sst file it's attempting to add
+  // (3) DBImpl::IngestExternalFile() calls this function to read the contents
+  //     of the sst file it's attempting to add
  //
  // table_reader_options is a TableReaderOptions which contain all the
  //    needed parameters and configuration to open the table.
@@ -1,108 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#pragma once
-#ifndef ROCKSDB_LITE
-
-#include <map>
-#include <string>
-#include <vector>
-
-#include "rocksdb/db.h"
-
-namespace rocksdb {
-
-// Date tiered database is a wrapper of DB that implements
-// a simplified DateTieredCompactionStrategy by using multiple column famillies
-// as time windows.
-//
-// DateTieredDB provides an interface similar to DB, but it assumes that user
-// provides keys with last 8 bytes encoded as timestamp in seconds. DateTieredDB
-// is assigned with a TTL to declare when data should be deleted.
-//
-// DateTieredDB hides column families layer from standard RocksDB instance. It
-// uses multiple column families to manage time series data, each containing a
-// specific range of time. Column families are named by its maximum possible
-// timestamp. A column family is created automatically when data newer than
-// latest timestamp of all existing column families. The time range of a column
-// family is configurable by `column_family_interval`. By doing this, we
-// guarantee that compaction will only happen in a column family.
-//
-// DateTieredDB is assigned with a TTL. When all data in a column family are
-// expired (CF_Timestamp <= CUR_Timestamp - TTL), we directly drop the whole
-// column family.
-//
-// TODO(jhli): This is only a simplified version of DTCS. In a complete DTCS,
-// time windows can be merged over time, so that older time windows will have
-// larger time range. Also, compaction are executed only for adjacent SST files
-// to guarantee there is no time overlap between SST files.
-
-class DateTieredDB {
- public:
-  // Open a DateTieredDB whose name is `dbname`.
-  // Similar to DB::Open(), created database object is stored in dbptr.
-  //
-  // Two parameters can be configured: `ttl` to specify the length of time that
-  // keys should exist in the database, and `column_family_interval` to specify
-  // the time range of a column family interval.
-  //
-  // Open a read only database if read only is set as true.
-  // TODO(jhli): Should use an option object that includes ttl and
-  // column_family_interval.
-  static Status Open(const Options& options, const std::string& dbname,
-                     DateTieredDB** dbptr, int64_t ttl,
-                     int64_t column_family_interval, bool read_only = false);
-
-  explicit DateTieredDB() {}
-
-  virtual ~DateTieredDB() {}
-
-  // Wrapper for Put method. Similar to DB::Put(), but column family to be
-  // inserted is decided by the timestamp in keys, i.e. the last 8 bytes of user
-  // key. If key is already obsolete, it will not be inserted.
-  //
-  // When client put a key value pair in DateTieredDB, it assumes last 8 bytes
-  // of keys are encoded as timestamp. Timestamp is a 64-bit signed integer
-  // encoded as the number of seconds since 1970-01-01 00:00:00 (UTC) (Same as
-  // Env::GetCurrentTime()). Timestamp should be encoded in big endian.
-  virtual Status Put(const WriteOptions& options, const Slice& key,
-                     const Slice& val) = 0;
-
-  // Wrapper for Get method. Similar to DB::Get() but column family is decided
-  // by timestamp in keys. If key is already obsolete, it will not be found.
-  virtual Status Get(const ReadOptions& options, const Slice& key,
-                     std::string* value) = 0;
-
-  // Wrapper for Delete method. Similar to DB::Delete() but column family is
-  // decided by timestamp in keys. If key is already obsolete, return NotFound
-  // status.
-  virtual Status Delete(const WriteOptions& options, const Slice& key) = 0;
-
-  // Wrapper for KeyMayExist method. Similar to DB::KeyMayExist() but column
-  // family is decided by timestamp in keys. Return false when key is already
-  // obsolete.
-  virtual bool KeyMayExist(const ReadOptions& options, const Slice& key,
-                           std::string* value, bool* value_found = nullptr) = 0;
-
-  // Wrapper for Merge method. Similar to DB::Merge() but column family is
-  // decided by timestamp in keys.
-  virtual Status Merge(const WriteOptions& options, const Slice& key,
-                       const Slice& value) = 0;
-
-  // Create an iterator that hides low level details. This iterator internally
-  // merge results from all active time series column families. Note that
-  // column families are not deleted until all data are obsolete, so this
-  // iterator can possibly access obsolete key value pairs.
-  virtual Iterator* NewIterator(const ReadOptions& opts) = 0;
-
-  // Explicitly drop column families in which all keys are obsolete. This
-  // process is also inplicitly done in Put() operation.
-  virtual Status DropObsoleteColumnFamilies() = 0;
-
-  static const uint64_t kTSLength = sizeof(int64_t);  // size of timestamp
-};
-
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE
@@ -1,149 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#pragma once
-#ifndef ROCKSDB_LITE
-
-#include <string>
-#include <vector>
-
-#include "rocksdb/utilities/stackable_db.h"
-#include "rocksdb/utilities/json_document.h"
-#include "rocksdb/db.h"
-
-namespace rocksdb {
-
-// IMPORTANT: DocumentDB is a work in progress. It is unstable and we might
-// change the API without warning. Talk to RocksDB team before using this in
-// production ;)
-
-// DocumentDB is a layer on top of RocksDB that provides a very simple JSON API.
-// When creating a DB, you specify a list of indexes you want to keep on your
-// data. You can insert a JSON document to the DB, which is automatically
-// indexed. Every document added to the DB needs to have "_id" field which is
-// automatically indexed and is an unique primary key. All other indexes are
-// non-unique.
-
-// NOTE: field names in the JSON are NOT allowed to start with '$' or
-// contain '.'. We don't currently enforce that rule, but will start behaving
-// badly.
-
-// Cursor is what you get as a result of executing query. To get all
-// results from a query, call Next() on a Cursor while  Valid() returns true
-class Cursor {
- public:
-  Cursor() = default;
-  virtual ~Cursor() {}
-
-  virtual bool Valid() const = 0;
-  virtual void Next() = 0;
-  // Lifecycle of the returned JSONDocument is until the next Next() call
-  virtual const JSONDocument& document() const = 0;
-  virtual Status status() const = 0;
-
- private:
-  // No copying allowed
-  Cursor(const Cursor&);
-  void operator=(const Cursor&);
-};
-
-struct DocumentDBOptions {
-  int background_threads = 4;
-  uint64_t memtable_size = 128 * 1024 * 1024;    // 128 MB
-  uint64_t cache_size = 1 * 1024 * 1024 * 1024;  // 1 GB
-};
-
-// TODO(icanadi) Add `JSONDocument* info` parameter to all calls that can be
-// used by the caller to get more information about the call execution (number
-// of dropped records, number of updated records, etc.)
-class DocumentDB : public StackableDB {
- public:
-  struct IndexDescriptor {
-    // Currently, you can only define an index on a single field. To specify an
-    // index on a field X, set index description to JSON "{X: 1}"
-    // Currently the value needs to be 1, which means ascending.
-    // In the future, we plan to also support indexes on multiple keys, where
-    // you could mix ascending sorting (1) with descending sorting indexes (-1)
-    JSONDocument* description;
-    std::string name;
-  };
-
-  // Open DocumentDB with specified indexes. The list of indexes has to be
-  // complete, i.e. include all indexes present in the DB, except the primary
-  // key index.
-  // Otherwise, Open() will return an error
-  static Status Open(const DocumentDBOptions& options, const std::string& name,
-                     const std::vector<IndexDescriptor>& indexes,
-                     DocumentDB** db, bool read_only = false);
-
-  explicit DocumentDB(DB* db) : StackableDB(db) {}
-
-  // Create a new index. It will stop all writes for the duration of the call.
-  // All current documents in the DB are scanned and corresponding index entries
-  // are created
-  virtual Status CreateIndex(const WriteOptions& write_options,
-                             const IndexDescriptor& index) = 0;
-
-  // Drop an index. Client is responsible to make sure that index is not being
-  // used by currently executing queries
-  virtual Status DropIndex(const std::string& name) = 0;
-
-  // Insert a document to the DB. The document needs to have a primary key "_id"
-  // which can either be a string or an integer. Otherwise the write will fail
-  // with InvalidArgument.
-  virtual Status Insert(const WriteOptions& options,
-                        const JSONDocument& document) = 0;
-
-  // Deletes all documents matching a filter atomically
-  virtual Status Remove(const ReadOptions& read_options,
-                        const WriteOptions& write_options,
-                        const JSONDocument& query) = 0;
-
-  // Does this sequence of operations:
-  // 1. Find all documents matching a filter
-  // 2. For all documents, atomically:
-  // 2.1. apply the update operators
-  // 2.2. update the secondary indexes
-  //
-  // Currently only $set update operator is supported.
-  // Syntax is: {$set: {key1: value1, key2: value2, etc...}}
-  // This operator will change a document's key1 field to value1, key2 to
-  // value2, etc. New values will be set even if a document didn't have an entry
-  // for the specified key.
-  //
-  // You can not change a primary key of a document.
-  //
-  // Update example: Update({id: {$gt: 5}, $index: id}, {$set: {enabled: true}})
-  virtual Status Update(const ReadOptions& read_options,
-                        const WriteOptions& write_options,
-                        const JSONDocument& filter,
-                        const JSONDocument& updates) = 0;
-
-  // query has to be an array in which every element is an operator. Currently
-  // only $filter operator is supported. Syntax of $filter operator is:
-  // {$filter: {key1: condition1, key2: condition2, etc.}} where conditions can
-  // be either:
-  // 1) a single value in which case the condition is equality condition, or
-  // 2) a defined operators, like {$gt: 4}, which will match all documents that
-  // have key greater than 4.
-  //
-  // Supported operators are:
-  // 1) $gt -- greater than
-  // 2) $gte -- greater than or equal
-  // 3) $lt -- less than
-  // 4) $lte -- less than or equal
-  // If you want the filter to use an index, you need to specify it like this:
-  // {$filter: {...(conditions)..., $index: index_name}}
-  //
-  // Example query:
-  // * [{$filter: {name: John, age: {$gte: 18}, $index: age}}]
-  // will return all Johns whose age is greater or equal to 18 and it will use
-  // index "age" to satisfy the query.
-  virtual Cursor* Query(const ReadOptions& read_options,
-                        const JSONDocument& query) = 0;
-};
-
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE
@@ -1,114 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-//
-
-#ifndef ROCKSDB_LITE
-#pragma once
-#include <string>
-#include <vector>
-
-#include "rocksdb/utilities/stackable_db.h"
-#include "rocksdb/status.h"
-
-namespace rocksdb {
-
-//
-// Configurable options needed for setting up a Geo database
-//
-struct GeoDBOptions {
-  // Backup info and error messages will be written to info_log
-  // if non-nullptr.
-  // Default: nullptr
-  Logger* info_log;
-
-  explicit GeoDBOptions(Logger* _info_log = nullptr):info_log(_info_log) { }
-};
-
-//
-// A position in the earth's geoid
-//
-class GeoPosition {
- public:
-  double latitude;
-  double longitude;
-
-  explicit GeoPosition(double la = 0, double lo = 0) :
-    latitude(la), longitude(lo) {
-  }
-};
-
-//
-// Description of an object on the Geoid. It is located by a GPS location,
-// and is identified by the id. The value associated with this object is
-// an opaque string 'value'. Different objects identified by unique id's
-// can have the same gps-location associated with them.
-//
-class GeoObject {
- public:
-  GeoPosition position;
-  std::string id;
-  std::string value;
-
-  GeoObject() {}
-
-  GeoObject(const GeoPosition& pos, const std::string& i,
-            const std::string& val) :
-    position(pos), id(i), value(val) {
-  }
-};
-
-class GeoIterator {
- public:
-  GeoIterator() = default;
-  virtual ~GeoIterator() {}
-  virtual void Next() = 0;
-  virtual bool Valid() const = 0;
-  virtual const GeoObject& geo_object() = 0;
-  virtual Status status() const = 0;
-};
-
-//
-// Stack your DB with GeoDB to be able to get geo-spatial support
-//
-class GeoDB : public StackableDB {
- public:
-  // GeoDBOptions have to be the same as the ones used in a previous
-  // incarnation of the DB
-  //
-  // GeoDB owns the pointer `DB* db` now. You should not delete it or
-  // use it after the invocation of GeoDB
-  // GeoDB(DB* db, const GeoDBOptions& options) : StackableDB(db) {}
-  GeoDB(DB* db, const GeoDBOptions& /*options*/) : StackableDB(db) {}
-  virtual ~GeoDB() {}
-
-  // Insert a new object into the location database. The object is
-  // uniquely identified by the id. If an object with the same id already
-  // exists in the db, then the old one is overwritten by the new
-  // object being inserted here.
-  virtual Status Insert(const GeoObject& object) = 0;
-
-  // Retrieve the value of the object located at the specified GPS
-  // location and is identified by the 'id'.
-  virtual Status GetByPosition(const GeoPosition& pos,
-                               const Slice& id, std::string* value) = 0;
-
-  // Retrieve the value of the object identified by the 'id'. This method
-  // could be potentially slower than GetByPosition
-  virtual Status GetById(const Slice& id, GeoObject*  object) = 0;
-
-  // Delete the specified object
-  virtual Status Remove(const Slice& id) = 0;
-
-  // Returns an iterator for the items within a circular radius from the
-  // specified gps location. If 'number_of_values' is specified,
-  // then the iterator is capped to that number of objects.
-  // The radius is specified in 'meters'.
-  virtual GeoIterator* SearchRadial(const GeoPosition& pos,
-                                    double radius,
-                                    int number_of_values = INT_MAX) = 0;
-};
-
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE
@@ -1,195 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-#pragma once
-#ifndef ROCKSDB_LITE
-
-#include <deque>
-#include <map>
-#include <memory>
-#include <string>
-#include <unordered_map>
-#include <utility>
-#include <vector>
-
-#include "rocksdb/slice.h"
-
-// We use JSONDocument for DocumentDB API
-// Implementation inspired by folly::dynamic, rapidjson and fbson
-
-namespace fbson {
-  class FbsonValue;
-  class ObjectVal;
-  template <typename T>
-  class FbsonWriterT;
-  class FbsonOutStream;
-  typedef FbsonWriterT<FbsonOutStream> FbsonWriter;
-}  // namespace fbson
-
-namespace rocksdb {
-
-// NOTE: none of this is thread-safe
-class JSONDocument {
- public:
-  // return nullptr on parse failure
-  static JSONDocument* ParseJSON(const char* json);
-
-  enum Type {
-    kNull,
-    kArray,
-    kBool,
-    kDouble,
-    kInt64,
-    kObject,
-    kString,
-  };
-
-  /* implicit */ JSONDocument();  // null
-  /* implicit */ JSONDocument(bool b);
-  /* implicit */ JSONDocument(double d);
-  /* implicit */ JSONDocument(int8_t i);
-  /* implicit */ JSONDocument(int16_t i);
-  /* implicit */ JSONDocument(int32_t i);
-  /* implicit */ JSONDocument(int64_t i);
-  /* implicit */ JSONDocument(const std::string& s);
-  /* implicit */ JSONDocument(const char* s);
-  // constructs JSONDocument of specific type with default value
-  explicit JSONDocument(Type _type);
-
-  JSONDocument(const JSONDocument& json_document);
-
-  JSONDocument(JSONDocument&& json_document);
-
-  Type type() const;
-
-  // REQUIRES: IsObject()
-  bool Contains(const std::string& key) const;
-  // REQUIRES: IsObject()
-  // Returns non-owner object
-  JSONDocument operator[](const std::string& key) const;
-
-  // REQUIRES: IsArray() == true || IsObject() == true
-  size_t Count() const;
-
-  // REQUIRES: IsArray()
-  // Returns non-owner object
-  JSONDocument operator[](size_t i) const;
-
-  JSONDocument& operator=(JSONDocument jsonDocument);
-
-  bool IsNull() const;
-  bool IsArray() const;
-  bool IsBool() const;
-  bool IsDouble() const;
-  bool IsInt64() const;
-  bool IsObject() const;
-  bool IsString() const;
-
-  // REQUIRES: IsBool() == true
-  bool GetBool() const;
-  // REQUIRES: IsDouble() == true
-  double GetDouble() const;
-  // REQUIRES: IsInt64() == true
-  int64_t GetInt64() const;
-  // REQUIRES: IsString() == true
-  std::string GetString() const;
-
-  bool operator==(const JSONDocument& rhs) const;
-
-  bool operator!=(const JSONDocument& rhs) const;
-
-  JSONDocument Copy() const;
-
-  bool IsOwner() const;
-
-  std::string DebugString() const;
-
- private:
-  class ItemsIteratorGenerator;
-
- public:
-  // REQUIRES: IsObject()
-  ItemsIteratorGenerator Items() const;
-
-  // appends serialized object to dst
-  void Serialize(std::string* dst) const;
-  // returns nullptr if Slice doesn't represent valid serialized JSONDocument
-  static JSONDocument* Deserialize(const Slice& src);
-
- private:
-  friend class JSONDocumentBuilder;
-
-  JSONDocument(fbson::FbsonValue* val, bool makeCopy);
-
-  void InitFromValue(const fbson::FbsonValue* val);
-
-  // iteration on objects
-  class const_item_iterator {
-   private:
-    class Impl;
-   public:
-    typedef std::pair<std::string, JSONDocument> value_type;
-    explicit const_item_iterator(Impl* impl);
-    const_item_iterator(const_item_iterator&&);
-    const_item_iterator& operator++();
-    bool operator!=(const const_item_iterator& other);
-    value_type operator*();
-    ~const_item_iterator();
-   private:
-    friend class ItemsIteratorGenerator;
-    std::unique_ptr<Impl> it_;
-  };
-
-  class ItemsIteratorGenerator {
-   public:
-    explicit ItemsIteratorGenerator(const fbson::ObjectVal& object);
-    const_item_iterator begin() const;
-
-    const_item_iterator end() const;
-
-   private:
-    const fbson::ObjectVal& object_;
-  };
-
-  std::unique_ptr<char[]> data_;
-  mutable fbson::FbsonValue* value_;
-
-  // Our serialization format's first byte specifies the encoding version. That
-  // way, we can easily change our format while providing backwards
-  // compatibility. This constant specifies the current version of the
-  // serialization format
-  static const char kSerializationFormatVersion;
-};
-
-class JSONDocumentBuilder {
- public:
-  JSONDocumentBuilder();
-
-  explicit JSONDocumentBuilder(fbson::FbsonOutStream* out);
-
-  void Reset();
-
-  bool WriteStartArray();
-
-  bool WriteEndArray();
-
-  bool WriteStartObject();
-
-  bool WriteEndObject();
-
-  bool WriteKeyValue(const std::string& key, const JSONDocument& value);
-
-  bool WriteJSONDocument(const JSONDocument& value);
-
-  JSONDocument GetJSONDocument();
-
-  ~JSONDocumentBuilder();
-
- private:
-  std::unique_ptr<fbson::FbsonWriter> writer_;
-};
-
-}  // namespace rocksdb
-
-#endif  // ROCKSDB_LITE
@@ -1,261 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#pragma once
-#ifndef ROCKSDB_LITE
-
-#include <string>
-#include <vector>
-
-#include "rocksdb/db.h"
-#include "rocksdb/slice.h"
-#include "rocksdb/utilities/stackable_db.h"
-
-namespace rocksdb {
-namespace spatial {
-
-// NOTE: SpatialDB is experimental and we might change its API without warning.
-// Please talk to us before developing against SpatialDB API.
-//
-// SpatialDB is a support for spatial indexes built on top of RocksDB.
-// When creating a new SpatialDB, clients specifies a list of spatial indexes to
-// build on their data. Each spatial index is defined by the area and
-// granularity. If you're storing map data, different spatial index
-// granularities can be used for different zoom levels.
-//
-// Each element inserted into SpatialDB has:
-// * a bounding box, which determines how will the element be indexed
-// * string blob, which will usually be WKB representation of the polygon
-// (http://en.wikipedia.org/wiki/Well-known_text)
-// * feature set, which is a map of key-value pairs, where value can be null,
-// int, double, bool, string
-// * a list of indexes to insert the element in
-//
-// Each query is executed on a single spatial index. Query guarantees that it
-// will return all elements intersecting the specified bounding box, but it
-// might also return some extra non-intersecting elements.
-
-// Variant is a class that can be many things: null, bool, int, double or string
-// It is used to store different value types in FeatureSet (see below)
-struct Variant {
-  // Don't change the values here, they are persisted on disk
-  enum Type {
-    kNull = 0x0,
-    kBool = 0x1,
-    kInt = 0x2,
-    kDouble = 0x3,
-    kString = 0x4,
-  };
-
-  Variant() : type_(kNull) {}
-  /* implicit */ Variant(bool b) : type_(kBool) { data_.b = b; }
-  /* implicit */ Variant(uint64_t i) : type_(kInt) { data_.i = i; }
-  /* implicit */ Variant(double d) : type_(kDouble) { data_.d = d; }
-  /* implicit */ Variant(const std::string& s) : type_(kString) {
-    new (&data_.s) std::string(s);
-  }
-
-  Variant(const Variant& v) : type_(v.type_) { Init(v, data_); }
-
-  Variant& operator=(const Variant& v);
-
-  Variant(Variant&& rhs) : type_(kNull) { *this = std::move(rhs); }
-
-  Variant& operator=(Variant&& v);
-
-  ~Variant() { Destroy(type_, data_); }
-
-  Type type() const { return type_; }
-  bool get_bool() const { return data_.b; }
-  uint64_t get_int() const { return data_.i; }
-  double get_double() const { return data_.d; }
-  const std::string& get_string() const { return *GetStringPtr(data_); }
-
-  bool operator==(const Variant& other) const;
-  bool operator!=(const Variant& other) const { return !(*this == other); }
-
- private:
-  Type type_;
-
-  union Data {
-    bool b;
-    uint64_t i;
-    double d;
-    // Current version of MS compiler not C++11 compliant so can not put
-    // std::string
-    // however, even then we still need the rest of the maintenance.
-    char s[sizeof(std::string)];
-  } data_;
-
-  // Avoid type_punned aliasing problem
-  static std::string* GetStringPtr(Data& d) {
-    void* p = d.s;
-    return reinterpret_cast<std::string*>(p);
-  }
-
-  static const std::string* GetStringPtr(const Data& d) {
-    const void* p = d.s;
-    return reinterpret_cast<const std::string*>(p);
-  }
-
-  static void Init(const Variant&, Data&);
-
-  static void Destroy(Type t, Data& d) {
-    if (t == kString) {
-      using std::string;
-      GetStringPtr(d)->~string();
-    }
-  }
-};
-
-// FeatureSet is a map of key-value pairs. One feature set is associated with
-// each element in SpatialDB. It can be used to add rich data about the element.
-class FeatureSet {
- private:
-  typedef std::unordered_map<std::string, Variant> map;
-
- public:
-  class iterator {
-   public:
-    /* implicit */ iterator(const map::const_iterator itr) : itr_(itr) {}
-    iterator& operator++() {
-      ++itr_;
-      return *this;
-    }
-    bool operator!=(const iterator& other) { return itr_ != other.itr_; }
-    bool operator==(const iterator& other) { return itr_ == other.itr_; }
-    map::value_type operator*() { return *itr_; }
-
-   private:
-    map::const_iterator itr_;
-  };
-  FeatureSet() = default;
-
-  FeatureSet* Set(const std::string& key, const Variant& value);
-  bool Contains(const std::string& key) const;
-  // REQUIRES: Contains(key)
-  const Variant& Get(const std::string& key) const;
-  iterator Find(const std::string& key) const;
-
-  iterator begin() const { return map_.begin(); }
-  iterator end() const { return map_.end(); }
-
-  void Clear();
-  size_t Size() const { return map_.size(); }
-
-  void Serialize(std::string* output) const;
-  // REQUIRED: empty FeatureSet
-  bool Deserialize(const Slice& input);
-
-  std::string DebugString() const;
-
- private:
-  map map_;
-};
-
-// BoundingBox is a helper structure for defining rectangles representing
-// bounding boxes of spatial elements.
-template <typename T>
-struct BoundingBox {
-  T min_x, min_y, max_x, max_y;
-  BoundingBox() = default;
-  BoundingBox(T _min_x, T _min_y, T _max_x, T _max_y)
-      : min_x(_min_x), min_y(_min_y), max_x(_max_x), max_y(_max_y) {}
-
-  bool Intersects(const BoundingBox<T>& a) const {
-    return !(min_x > a.max_x || min_y > a.max_y || a.min_x > max_x ||
-             a.min_y > max_y);
-  }
-};
-
-struct SpatialDBOptions {
-  uint64_t cache_size = 1 * 1024 * 1024 * 1024LL;  // 1GB
-  int num_threads = 16;
-  bool bulk_load = true;
-};
-
-// Cursor is used to return data from the query to the client. To get all the
-// data from the query, just call Next() while Valid() is true
-class Cursor {
- public:
-  Cursor() = default;
-  virtual ~Cursor() {}
-
-  virtual bool Valid() const = 0;
-  // REQUIRES: Valid()
-  virtual void Next() = 0;
-
-  // Lifetime of the underlying storage until the next call to Next()
-  // REQUIRES: Valid()
-  virtual const Slice blob() = 0;
-  // Lifetime of the underlying storage until the next call to Next()
-  // REQUIRES: Valid()
-  virtual const FeatureSet& feature_set() = 0;
-
-  virtual Status status() const = 0;
-
- private:
-  // No copying allowed
-  Cursor(const Cursor&);
-  void operator=(const Cursor&);
-};
-
-// SpatialIndexOptions defines a spatial index that will be built on the data
-struct SpatialIndexOptions {
-  // Spatial indexes are referenced by names
-  std::string name;
-  // An area that is indexed. If the element is not intersecting with spatial
-  // index's bbox, it will not be inserted into the index
-  BoundingBox<double> bbox;
-  // tile_bits control the granularity of the spatial index. Each dimension of
-  // the bbox will be split into (1 << tile_bits) tiles, so there will be a
-  // total of (1 << tile_bits)^2 tiles. It is recommended to configure a size of
-  // each  tile to be approximately the size of the query on that spatial index
-  uint32_t tile_bits;
-  SpatialIndexOptions() {}
-  SpatialIndexOptions(const std::string& _name,
-                      const BoundingBox<double>& _bbox, uint32_t _tile_bits)
-      : name(_name), bbox(_bbox), tile_bits(_tile_bits) {}
-};
-
-class SpatialDB : public StackableDB {
- public:
-  // Creates the SpatialDB with specified list of indexes.
-  // REQUIRED: db doesn't exist
-  static Status Create(const SpatialDBOptions& options, const std::string& name,
-                       const std::vector<SpatialIndexOptions>& spatial_indexes);
-
-  // Open the existing SpatialDB.  The resulting db object will be returned
-  // through db parameter.
-  // REQUIRED: db was created using SpatialDB::Create
-  static Status Open(const SpatialDBOptions& options, const std::string& name,
-                     SpatialDB** db, bool read_only = false);
-
-  explicit SpatialDB(DB* db) : StackableDB(db) {}
-
-  // Insert the element into the DB. Element will be inserted into specified
-  // spatial_indexes, based on specified bbox.
-  // REQUIRES: spatial_indexes.size() > 0
-  virtual Status Insert(const WriteOptions& write_options,
-                        const BoundingBox<double>& bbox, const Slice& blob,
-                        const FeatureSet& feature_set,
-                        const std::vector<std::string>& spatial_indexes) = 0;
-
-  // Calling Compact() after inserting a bunch of elements should speed up
-  // reading. This is especially useful if you use SpatialDBOptions::bulk_load
-  // Num threads determines how many threads we'll use for compactions. Setting
-  // this to bigger number will use more IO and CPU, but finish faster
-  virtual Status Compact(int num_threads = 1) = 0;
-
-  // Query the specified spatial_index. Query will return all elements that
-  // intersect bbox, but it may also return some extra elements.
-  virtual Cursor* Query(const ReadOptions& read_options,
-                        const BoundingBox<double>& bbox,
-                        const std::string& spatial_index) = 0;
-};
-
-}  // namespace spatial
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE
@@ -218,12 +218,13 @@ class StackableDB : public DB {
  virtual Status CompactFiles(
      const CompactionOptions& compact_options,
      ColumnFamilyHandle* column_family,
-      const std::vector<std::string>& input_file_names,
-      const int output_level, const int output_path_id = -1,
-      std::vector<std::string>* const output_file_names = nullptr) override {
-    return db_->CompactFiles(
-        compact_options, column_family, input_file_names,
-        output_level, output_path_id, output_file_names);
+      const std::vector<std::string>& input_file_names, const int output_level,
+      const int output_path_id = -1,
+      std::vector<std::string>* const output_file_names = nullptr,
+      CompactionJobInfo* compaction_job_info = nullptr) override {
+    return db_->CompactFiles(compact_options, column_family, input_file_names,
+                             output_level, output_path_id, output_file_names,
+                             compaction_job_info);
  }

  virtual Status PauseBackgroundWork() override {
@@ -208,8 +208,10 @@ class Transaction {
  // Read this key and ensure that this transaction will only
  // be able to be committed if this key is not written outside this
  // transaction after it has first been read (or after the snapshot if a
-  // snapshot is set in this transaction).  The transaction behavior is the
-  // same regardless of whether the key exists or not.
+  // snapshot is set in this transaction and do_validate is true). If
+  // do_validate is false, ReadOptions::snapshot is expected to be nullptr so
+  // that GetForUpdate returns the latest committed value. The transaction
+  // behavior is the same regardless of whether the key exists or not.
  //
  // Note: Currently, this function will return Status::MergeInProgress
  // if the most recent write to the queried key in this batch is a Merge.
@@ -234,27 +236,31 @@ class Transaction {
  virtual Status GetForUpdate(const ReadOptions& options,
                              ColumnFamilyHandle* column_family,
                              const Slice& key, std::string* value,
-                              bool exclusive = true) = 0;
+                              bool exclusive = true,
+                              const bool do_validate = true) = 0;

  // An overload of the above method that receives a PinnableSlice
  // For backward compatibility a default implementation is provided
  virtual Status GetForUpdate(const ReadOptions& options,
                              ColumnFamilyHandle* column_family,
                              const Slice& key, PinnableSlice* pinnable_val,
-                              bool exclusive = true) {
+                              bool exclusive = true,
+                              const bool do_validate = true) {
    if (pinnable_val == nullptr) {
      std::string* null_str = nullptr;
-      return GetForUpdate(options, column_family, key, null_str, exclusive);
+      return GetForUpdate(options, column_family, key, null_str, exclusive,
+                          do_validate);
    } else {
      auto s = GetForUpdate(options, column_family, key,
-                            pinnable_val->GetSelf(), exclusive);
+                            pinnable_val->GetSelf(), exclusive, do_validate);
      pinnable_val->PinSelf();
      return s;
    }
  }

  virtual Status GetForUpdate(const ReadOptions& options, const Slice& key,
-                              std::string* value, bool exclusive = true) = 0;
+                              std::string* value, bool exclusive = true,
+                              const bool do_validate = true) = 0;

  virtual std::vector<Status> MultiGetForUpdate(
      const ReadOptions& options,
@@ -287,6 +293,9 @@ class Transaction {
  // functions in WriteBatch, but will also do conflict checking on the
  // keys being written.
  //
+  // assume_tracked=true expects the key be already tracked. If valid then it
+  // skips ValidateSnapshot. Returns error otherwise.
+  //
  // If this Transaction was created on an OptimisticTransactionDB, these
  // functions should always return Status::OK().
  //
@@ -299,28 +308,33 @@ class Transaction {
  //  (See max_write_buffer_number_to_maintain)
  // or other errors on unexpected failures.
  virtual Status Put(ColumnFamilyHandle* column_family, const Slice& key,
-                     const Slice& value) = 0;
+                     const Slice& value, const bool assume_tracked = false) = 0;
  virtual Status Put(const Slice& key, const Slice& value) = 0;
  virtual Status Put(ColumnFamilyHandle* column_family, const SliceParts& key,
-                     const SliceParts& value) = 0;
+                     const SliceParts& value,
+                     const bool assume_tracked = false) = 0;
  virtual Status Put(const SliceParts& key, const SliceParts& value) = 0;

  virtual Status Merge(ColumnFamilyHandle* column_family, const Slice& key,
-                       const Slice& value) = 0;
+                       const Slice& value,
+                       const bool assume_tracked = false) = 0;
  virtual Status Merge(const Slice& key, const Slice& value) = 0;

-  virtual Status Delete(ColumnFamilyHandle* column_family,
-                        const Slice& key) = 0;
+  virtual Status Delete(ColumnFamilyHandle* column_family, const Slice& key,
+                        const bool assume_tracked = false) = 0;
  virtual Status Delete(const Slice& key) = 0;
  virtual Status Delete(ColumnFamilyHandle* column_family,
-                        const SliceParts& key) = 0;
+                        const SliceParts& key,
+                        const bool assume_tracked = false) = 0;
  virtual Status Delete(const SliceParts& key) = 0;

  virtual Status SingleDelete(ColumnFamilyHandle* column_family,
-                              const Slice& key) = 0;
+                              const Slice& key,
+                              const bool assume_tracked = false) = 0;
  virtual Status SingleDelete(const Slice& key) = 0;
  virtual Status SingleDelete(ColumnFamilyHandle* column_family,
-                              const SliceParts& key) = 0;
+                              const SliceParts& key,
+                              const bool assume_tracked = false) = 0;
  virtual Status SingleDelete(const SliceParts& key) = 0;

  // PutUntracked() will write a Put to the batch of operations to be committed