Update history and bump patch number

Shared block cache in RocksJava
Summary: Changes to support sharing block cache using the Java API. Previously DB instances could share the block cache only when the same Options instance is passed to all the DB instances. But now, with this change, it is possible to explicitly create a cache and pass it to multiple options instances, to share the block cache. Implementing this for [Rocksandra](https://github.com/instagram/cassandra/tree/rocks_3.0), but this feature has been requested by many java api users over the years. Closes https://github.com/facebook/rocksdb/pull/3623 Differential Revision: D7305794 Pulled By: sagar0 fbshipit-source-id: 03e4e8ed7aeee6f88bada4a8365d4279ede2ad71
2026-11-06 11:02:36 -05:00 · 2018-03-23 11:39:50 -07:00 · 2018-03-23 11:33:26 -07:00 · 2018-03-23 11:15:55 -07:00 · 2018-03-16 14:09:20 -07:00 · 2018-03-16 14:08:01 -07:00
341 changed files with 21875 additions and 3450 deletions
@@ -14,7 +14,7 @@
 #        cd build
 # 3. Run cmake to generate project files for Windows, add more options to enable required third-party libraries.
 #    See thirdparty.inc for more information.
-#        sample command: cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
+#        sample command: cmake -G "Visual Studio 15 Win64" -DWITH_GFLAGS=1 -DWITH_SNAPPY=1 -DWITH_JEMALLOC=1 -DWITH_JNI=1 ..
 # 4. Then build the project in debug mode (you may want to add /m[:<N>] flag to run msbuild in <N> parallel threads
 #                                          or simply /m ot use all avail cores)
 #        msbuild rocksdb.sln
@@ -34,15 +34,25 @@

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

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

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

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

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

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

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

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

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

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

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

 include(CheckCXXSourceCompiles)
 if(NOT MSVC)
-  set(CMAKE_REQUIRED_FLAGS "-msse4.2")
+  set(CMAKE_REQUIRED_FLAGS "-msse4.2 -mpclmul")
 endif()
 CHECK_CXX_SOURCE_COMPILES("
 #include <cstdint>
 #include <nmmintrin.h>
+#include <wmmintrin.h>
 int main() {
  volatile uint32_t x = _mm_crc32_u32(0, 0);
+  const auto a = _mm_set_epi64x(0, 0);
+  const auto b = _mm_set_epi64x(0, 0);
+  const auto c = _mm_clmulepi64_si128(a, b, 0x00);
+  auto d = _mm_cvtsi128_si64(c);
 }
 " HAVE_SSE42)
 unset(CMAKE_REQUIRED_FLAGS)
@@ -249,13 +285,6 @@ if(WITH_UBSAN)
  endif()
 endif()

-find_package(gflags)
-if(gflags_FOUND)
-  add_definitions(-DGFLAGS=1)
-  include_directories(${gflags_INCLUDE_DIR})
-  list(APPEND THIRDPARTY_LIBS ${gflags_LIBRARIES})
-endif()
-
 find_package(NUMA)
 if(NUMA_FOUND)
  add_definitions(-DNUMA)
@@ -485,6 +514,7 @@ set(SOURCES
        table/block_based_table_factory.cc
        table/block_based_table_reader.cc
        table/block_builder.cc
+        table/block_fetcher.cc
        table/block_prefix_index.cc
        table/bloom_block.cc
        table/cuckoo_table_builder.cc
@@ -572,6 +602,7 @@ set(SOURCES
        utilities/leveldb_options/leveldb_options.cc
        utilities/lua/rocks_lua_compaction_filter.cc
        utilities/memory/memory_util.cc
+        utilities/merge_operators/bytesxor.cc
        utilities/merge_operators/max.cc
        utilities/merge_operators/put.cc
        utilities/merge_operators/string_append/stringappend.cc
@@ -608,10 +639,16 @@ if(HAVE_SSE42 AND NOT FORCE_SSE42)
  if(NOT MSVC)
 set_source_files_properties(
  util/crc32c.cc
-      PROPERTIES COMPILE_FLAGS "-msse4.2")
+      PROPERTIES COMPILE_FLAGS "-msse4.2 -mpclmul")
  endif()
 endif()

+if(HAVE_POWER8)
+  list(APPEND SOURCES
+    util/crc32c_ppc.c
+    util/crc32c_ppc_asm.S)
+endif(HAVE_POWER8)
+
 if(WIN32)
  list(APPEND SOURCES
    port/win/io_win.cc
@@ -706,7 +743,7 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
  set(package_config_destination ${CMAKE_INSTALL_LIBDIR}/cmake/rocksdb)

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

@@ -936,17 +973,18 @@ if(WITH_TESTS)

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

@@ -1,5 +1,61 @@
 # Rocksdb Change Log
-## Unreleased
+## 5.12.2 (3/23/2018)
+### Bug Fixes
+* Fsync after writing global seq number to the ingestion file in ExternalSstFileIngestionJob.
+
+### Java API Changes
+* Add `BlockBasedTableConfig.setBlockCache` to allow sharing a block cache across DB instances.
+
+## 5.12.1 (3/16/2018)
+### Public API Change
+* RocksDBOptionsParser::Parse()'s `ignore_unknown_options` argument will only be effective if the option file shows it is generated using a higher version of RocksDB than the current version.
+
+### New Features
+* Avoid unnecessarily flushing in `CompactRange()` when the range specified by the user does not overlap unflushed memtables.
+
+### Bug Fixes
+* Fix WAL corruption caused by race condition between user write thread and backup/checkpoint thread.
+
+## 5.12.0 (2/14/2018)
+### Public API Change
+* Iterator::SeekForPrev is now a pure virtual method. This is to prevent user who implement the Iterator interface fail to implement SeekForPrev by mistake.
+* Add `include_end` option to make the range end exclusive when `include_end == false` in `DeleteFilesInRange()`.
+* Add `CompactRangeOptions::allow_write_stall`, which makes `CompactRange` start working immediately, even if it causes user writes to stall. The default value is false, meaning we add delay to `CompactRange` calls until stalling can be avoided when possible. Note this delay is not present in previous RocksDB versions.
+* Creating checkpoint with empty directory now returns `Status::InvalidArgument`; previously, it returned `Status::IOError`.
+* Adds a BlockBasedTableOption to turn off index block compression.
+* Close() method now returns a status when closing a db.
+
+### New Features
+* Improve the performance of iterators doing long range scans by using readahead.
+* Add new function `DeleteFilesInRanges()` to delete files in multiple ranges at once for better performance.
+* FreeBSD build support for RocksDB and RocksJava.
+* Improved performance of long range scans with readahead.
+* Updated to and now continuously tested in Visual Studio 2017.
+
+### Bug Fixes
+* Fix `DisableFileDeletions()` followed by `GetSortedWalFiles()` to not return obsolete WAL files that `PurgeObsoleteFiles()` is going to delete.
+* Fix Handle error return from WriteBuffer() during WAL file close and DB close.
+* Fix advance reservation of arena block addresses.
+* Fix handling of empty string as checkpoint directory.
+
+## 5.11.0 (01/08/2018)
+### Public API Change
+* Add `autoTune` and `getBytesPerSecond()` to RocksJava RateLimiter
+
+### New Features
+* Add a new histogram stat called rocksdb.db.flush.micros for memtable flush.
+* Add "--use_txn" option to use transactional API in db_stress.
+* Disable onboard cache for compaction output in Windows platform.
+* Improve the performance of iterators doing long range scans by using readahead.
+
+### Bug Fixes
+* Fix a stack-use-after-scope bug in ForwardIterator.
+* Fix builds on platforms including Linux, Windows, and PowerPC.
+* Fix buffer overrun in backup engine for DBs with huge number of files.
+* Fix a mislabel bug for bottom-pri compaction threads.
+* Fix DB::Flush() keep waiting after flush finish under certain condition.
+
+## 5.10.0 (12/11/2017)
 ### Public API Change
 * When running `make` with environment variable `USE_SSE` set and `PORTABLE` unset, will use all machine features available locally. Previously this combination only compiled SSE-related features.

@@ -25,6 +81,7 @@
 * Return an error on write if write_options.sync = true and write_options.disableWAL = true to warn user of inconsistent options. Previously we will not write to WAL and not respecting the sync options in this case.

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

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

@@ -15,3 +15,4 @@ This is the list of all known third-party language bindings for RocksDB. If some
 * D programming language - https://github.com/b1naryth1ef/rocksdb
 * Erlang - https://gitlab.com/barrel-db/erlang-rocksdb
 * Elixir - https://github.com/urbint/rox
+* Nim - https://github.com/status-im/nim-rocksdb
@@ -232,9 +232,13 @@ endif
 # USAN doesn't work well with jemalloc. If we're compiling with USAN, we should use regular malloc.
 ifdef COMPILE_WITH_UBSAN
 	DISABLE_JEMALLOC=1
-	EXEC_LDFLAGS += -fsanitize=undefined
-	PLATFORM_CCFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
-	PLATFORM_CXXFLAGS += -fsanitize=undefined -DROCKSDB_UBSAN_RUN
+	# Suppress alignment warning because murmurhash relies on casting unaligned
+	# memory to integer. Fixing it may cause performance regression. 3-way crc32
+	# relies on it too, although it can be rewritten to eliminate with minimal
+	# performance regression.
+	EXEC_LDFLAGS += -fsanitize=undefined -fno-sanitize-recover=all
+	PLATFORM_CCFLAGS += -fsanitize=undefined -fno-sanitize-recover=all -DROCKSDB_UBSAN_RUN
+	PLATFORM_CXXFLAGS += -fsanitize=undefined -fno-sanitize-recover=all -DROCKSDB_UBSAN_RUN
 endif

 ifdef ROCKSDB_VALGRIND_RUN
@@ -305,6 +309,9 @@ LDFLAGS += $(LUA_LIB)

 endif

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

 CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers
@@ -341,6 +348,8 @@ ifeq ($(HAVE_POWER8),1)
 LIB_CC_OBJECTS = $(LIB_SOURCES:.cc=.o)
 LIBOBJECTS += $(LIB_SOURCES_C:.c=.o)
 LIBOBJECTS += $(LIB_SOURCES_ASM:.S=.o)
+else
+LIB_CC_OBJECTS = $(LIB_SOURCES:.cc=.o)
 endif

 LIBOBJECTS += $(TOOL_LIB_SOURCES:.cc=.o)
@@ -504,6 +513,7 @@ PARALLEL_TEST = \
 	backupable_db_test \
 	db_compaction_filter_test \
 	db_compaction_test \
+	db_merge_operator_test \
 	db_sst_test \
 	db_test \
 	db_universal_compaction_test \
@@ -515,8 +525,12 @@ PARALLEL_TEST = \
 	persistent_cache_test \
 	table_test \
 	transaction_test \
-	write_prepared_transaction_test
+	write_prepared_transaction_test \

+# options_settable_test doesn't pass with UBSAN as we use hack in the test
+ifdef COMPILE_WITH_UBSAN
+        TESTS := $(shell echo $(TESTS) | sed 's/\boptions_settable_test\b//g')
+endif
 SUBSET := $(TESTS)
 ifdef ROCKSDBTESTS_START
        SUBSET := $(shell echo $(SUBSET) | sed 's/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/')
@@ -1550,11 +1564,11 @@ BZIP2_DOWNLOAD_BASE ?= http://www.bzip.org
 SNAPPY_VER ?= 1.1.4
 SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
-LZ4_VER ?= 1.7.5
-LZ4_SHA256 ?= 0190cacd63022ccb86f44fa5041dc6c3804407ad61550ca21c382827319e7e7e
+LZ4_VER ?= 1.8.0
+LZ4_SHA256 ?= 2ca482ea7a9bb103603108b5a7510b7592b90158c151ff50a28f1ca8389fccf6
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.2.0
-ZSTD_SHA256 ?= 4a7e4593a3638276ca7f2a09dc4f38e674d8317bbea51626393ca73fc047cbfb
+ZSTD_VER ?= 1.3.3
+ZSTD_SHA256 ?= a77c47153ee7de02626c5b2a097005786b71688be61e9fb81806a011f90b297b
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive

 ifeq ($(PLATFORM), OS_MACOSX)
@@ -1568,7 +1582,7 @@ else
 endif
 endif
 ifeq ($(PLATFORM), OS_FREEBSD)
-	JAVA_INCLUDE += -I$(JAVA_HOME)/include/freebsd
+	JAVA_INCLUDE = -I$(JAVA_HOME)/include -I$(JAVA_HOME)/include/freebsd
 	ROCKSDBJNILIB = librocksdbjni-freebsd$(ARCH).so
 	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-freebsd$(ARCH).jar
 endif
@@ -1594,7 +1608,7 @@ libz.a:
 		exit 1; \
 	fi
 	tar xvzf zlib-$(ZLIB_VER).tar.gz
-	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC ${EXTRA_CFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --static && make
+	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC ${EXTRA_CFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --static && $(MAKE)
 	cp zlib-$(ZLIB_VER)/libz.a .

 libbz2.a:
@@ -1606,7 +1620,7 @@ libbz2.a:
 		exit 1; \
 	fi
 	tar xvzf bzip2-$(BZIP2_VER).tar.gz
-	cd bzip2-$(BZIP2_VER) && make CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64 ${EXTRA_CFLAGS}' AR='ar ${EXTRA_ARFLAGS}'
+	cd bzip2-$(BZIP2_VER) && $(MAKE) CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64 ${EXTRA_CFLAGS}' AR='ar ${EXTRA_ARFLAGS}'
 	cp bzip2-$(BZIP2_VER)/libbz2.a .

 libsnappy.a:
@@ -1619,7 +1633,7 @@ libsnappy.a:
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
 	cd snappy-$(SNAPPY_VER) && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --with-pic --enable-static --disable-shared
-	cd snappy-$(SNAPPY_VER) && make ${SNAPPY_MAKE_TARGET}
+	cd snappy-$(SNAPPY_VER) && $(MAKE) ${SNAPPY_MAKE_TARGET}
 	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .

 liblz4.a:
@@ -1632,7 +1646,7 @@ liblz4.a:
 		exit 1; \
 	fi
 	tar xvzf lz4-$(LZ4_VER).tar.gz
-	cd lz4-$(LZ4_VER)/lib && make CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
+	cd lz4-$(LZ4_VER)/lib && $(MAKE) CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
 	cp lz4-$(LZ4_VER)/lib/liblz4.a .

 libzstd.a:
@@ -1645,29 +1659,45 @@ libzstd.a:
 		exit 1; \
 	fi
 	tar xvzf zstd-$(ZSTD_VER).tar.gz
-	cd zstd-$(ZSTD_VER)/lib && make CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' all
+	cd zstd-$(ZSTD_VER)/lib && DESTDIR=. PREFIX= $(MAKE) CFLAGS='-fPIC -O2 ${EXTRA_CFLAGS}' install
 	cp zstd-$(ZSTD_VER)/lib/libzstd.a .

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

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

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

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

-rocksdbjavastatic: $(java_static_libobjects)
+rocksdbjavastatic: $(java_static_all_libobjects)
 	cd java;$(MAKE) javalib;
 	rm -f ./java/target/$(ROCKSDBJNILIB)
 	$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC \
 	  -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) \
-	  $(java_static_libobjects) $(COVERAGEFLAGS) \
+	  $(java_static_all_libobjects) $(COVERAGEFLAGS) \
 	  $(JAVA_COMPRESSIONS) $(JAVA_STATIC_LDFLAGS)
 	cd java/target;strip $(STRIPFLAGS) $(ROCKSDBJNILIB)
 	cd java;jar -cf target/$(ROCKSDB_JAR) HISTORY*.md
@@ -1728,7 +1758,7 @@ JAVA_C_LIBOBJECTS = $(patsubst %.c.o,jl/%.c.o,$(JAVA_C_OBJECTS))
 JAVA_ASM_LIBOBJECTS = $(patsubst %.S.o,jl/%.S.o,$(JAVA_ASM_OBJECTS))
 endif

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

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

-[![Build Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
-[![Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
+[![Linux/Mac Build Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
+[![Windows Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
+[![PPC64le Build Status](http://140.211.168.68:8080/buildStatus/icon?job=Rocksdb)](http://140.211.168.68:8080/job/Rocksdb)


 RocksDB is developed and maintained by Facebook Database Engineering Team.
@@ -157,6 +157,7 @@ cpp_library(
        "table/block_based_table_factory.cc",
        "table/block_based_table_reader.cc",
        "table/block_builder.cc",
+        "table/block_fetcher.cc",
        "table/block_prefix_index.cc",
        "table/bloom_block.cc",
        "table/cuckoo_table_builder.cc",
@@ -241,6 +242,7 @@ cpp_library(
        "utilities/leveldb_options/leveldb_options.cc",
        "utilities/lua/rocks_lua_compaction_filter.cc",
        "utilities/memory/memory_util.cc",
+        "utilities/merge_operators/bytesxor.cc",
        "utilities/merge_operators/max.cc",
        "utilities/merge_operators/put.cc",
        "utilities/merge_operators/string_append/stringappend.cc",
@@ -566,7 +568,7 @@ ROCKS_TESTS = [
    [
        "db_merge_operator_test",
        "db/db_merge_operator_test.cc",
-        "serial",
+        "parallel",
    ],
    [
        "db_options_test",
@@ -1016,7 +1018,7 @@ ROCKS_TESTS = [
    [
        "write_prepared_transaction_test",
        "utilities/transactions/write_prepared_transaction_test.cc",
-        "serial",
+        "parallel",
    ],
 ]

@@ -14,6 +14,11 @@ Vagrant.configure("2") do |config|
    box.vm.box = "chef/centos-6.5"
  end

+  config.vm.define "centos7" do |box|
+    box.vm.box = "centos/7"
+    box.vm.provision "shell", path: "build_tools/setup_centos7.sh"
+  end
+
  config.vm.define "FreeBSD10" do |box|
    box.vm.guest = :freebsd
    box.vm.box = "robin/freebsd-10"
@@ -1,9 +1,9 @@
 version: 1.0.{build}
-image: Visual Studio 2015
+image: Visual Studio 2017
 before_build:
 - md %APPVEYOR_BUILD_FOLDER%\build
 - cd %APPVEYOR_BUILD_FOLDER%\build
- cmake -G "Visual Studio 14 2015 Win64" -DOPTDBG=1 -DXPRESS=1 -DPORTABLE=1 ..
+- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 ..
 - cd ..
 build:
  project: build\rocksdb.sln
@@ -36,7 +36,7 @@ def parse_src_mk(repo_path):
 # get all .cc / .c files
 def get_cc_files(repo_path):
    cc_files = []
-    for root, dirnames, filenames in os.walk(repo_path):
+    for root, dirnames, filenames in os.walk(repo_path):  # noqa: B007 T25377293 Grandfathered in
        root = root[(len(repo_path) + 1):]
        if "java" in root:
            # Skip java
@@ -1,4 +1,5 @@
 #!/usr/bin/env bash
 # Create a tmp directory for the test to use
 TEST_DIR=$(mktemp -d /dev/shm/fbcode_rocksdb_XXXXXXX)
+# shellcheck disable=SC2068
 TEST_TMPDIR="$TEST_DIR" $@ && rm -rf "$TEST_DIR"
@@ -141,6 +141,7 @@ case "$TARGET_OS" in
        ;;
    FreeBSD)
        PLATFORM=OS_FREEBSD
+        CXX=clang++
        COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_FREEBSD"
        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
        # PORT_FILES=port/freebsd/freebsd_specific.cc
@@ -481,13 +482,16 @@ if test -z "$PORTABLE"; then
    COMMON_FLAGS="$COMMON_FLAGS -mcpu=$POWER -mtune=$POWER "
  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^s390x`"; then
    COMMON_FLAGS="$COMMON_FLAGS -march=z10 "
+  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
+    # TODO: Handle this with approprite options.
+    COMMON_FLAGS="$COMMON_FLAGS"
  elif [ "$TARGET_OS" != AIX ] && [ "$TARGET_OS" != SunOS ]; then
    COMMON_FLAGS="$COMMON_FLAGS -march=native "
  elif test "$USE_SSE"; then
-    COMMON_FLAGS="$COMMON_FLAGS -msse4.2"
+    COMMON_FLAGS="$COMMON_FLAGS -msse4.2 -mpclmul"
  fi
 elif test "$USE_SSE"; then
-  COMMON_FLAGS="$COMMON_FLAGS -msse4.2"
+  COMMON_FLAGS="$COMMON_FLAGS -msse4.2 -mpclmul"
 fi

 $CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
@@ -501,6 +505,24 @@ if [ "$?" = 0 ]; then
  COMMON_FLAGS="$COMMON_FLAGS -DHAVE_SSE42"
 elif test "$USE_SSE"; then
  echo "warning: USE_SSE specified but compiler could not use SSE intrinsics, disabling"
+  exit 1
+fi
+
+$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
+  #include <cstdint>
+  #include <wmmintrin.h>
+  int main() {
+    const auto a = _mm_set_epi64x(0, 0);
+    const auto b = _mm_set_epi64x(0, 0);
+    const auto c = _mm_clmulepi64_si128(a, b, 0x00);
+    auto d = _mm_cvtsi128_si64(c);
+  }
+EOF
+if [ "$?" = 0 ]; then
+  COMMON_FLAGS="$COMMON_FLAGS -DHAVE_PCLMUL"
+elif test "$USE_SSE"; then
+  echo "warning: USE_SSE specified but compiler could not use PCLMUL intrinsics, disabling"
+  exit 1
 fi

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

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

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

@@ -1,3 +1,4 @@
+# shellcheck disable=SC2148
 GCC_BASE=/mnt/gvfs/third-party2/gcc/8219ec1bcedf8ad9da05e121e193364de2cc4f61/5.x/centos6-native/c447969
 CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/64d8d58e3d84f8bde7a029763d4f5baf39d0d5b9/stable/centos6-native/6aaf4de
 LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/ba9be983c81de7299b59fe71950c664a84dcb5f8/5.x/gcc-5-glibc-2.23/339d858
@@ -1,3 +1,4 @@
+# shellcheck disable=SC2148
 GCC_BASE=/mnt/gvfs/third-party2/gcc/cf7d14c625ce30bae1a4661c2319c5a283e4dd22/4.8.1/centos6-native/cc6c9dc
 CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/8598c375b0e94e1448182eb3df034704144a838d/stable/centos6-native/3f16ddd
 LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/d6e0a7da6faba45f5e5b1638f9edd7afc2f34e7d/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc
@@ -1,3 +1,4 @@
+# shellcheck disable=SC1113
 #/usr/bin/env bash

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

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

+#
+# Report RocksDB lite binary size to scuba
+REPORT_LITE_BINARY_SIZE_COMMANDS="[
+    {
+        'name':'Rocksdb Lite Binary Size',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Report RocksDB Lite binary size to scuba',
+                'shell':'tools/report_lite_binary_size.sh',
+                'user':'root',
+            },
+        ],
+]"
+
 #
 # RocksDB stress/crash test
 #
@@ -728,6 +744,9 @@ case $1 in
  lite)
    echo $LITE_BUILD_COMMANDS
    ;;
+  report_lite_binary_size)
+    echo $REPORT_LITE_BINARY_SIZE_COMMANDS
+    ;;
  stress_crash)
    echo $STRESS_CRASH_TEST_COMMANDS
    ;;
@@ -0,0 +1,43 @@
+#!/bin/bash
+set -e
+
+ROCKSDB_VERSION="5.10.3"
+ZSTD_VERSION="1.1.3"
+
+echo "This script configures CentOS with everything needed to build and run RocksDB"
+
+yum update -y && yum install epel-release -y
+
+yum install -y \
+  wget \
+  gcc-c++ \
+  snappy snappy-devel \
+  zlib zlib-devel \
+  bzip2 bzip2-devel \
+  lz4-devel \
+  libasan \
+  gflags
+
+mkdir -pv /usr/local/rocksdb-${ROCKSDB_VERSION}
+ln -sfT /usr/local/rocksdb-${ROCKSDB_VERSION} /usr/local/rocksdb
+
+wget -qO /tmp/zstd-${ZSTD_VERSION}.tar.gz https://github.com/facebook/zstd/archive/v${ZSTD_VERSION}.tar.gz
+wget -qO /tmp/rocksdb-${ROCKSDB_VERSION}.tar.gz https://github.com/facebook/rocksdb/archive/v${ROCKSDB_VERSION}.tar.gz
+
+cd /tmp
+
+tar xzvf zstd-${ZSTD_VERSION}.tar.gz
+tar xzvf rocksdb-${ROCKSDB_VERSION}.tar.gz -C /usr/local/
+
+echo "Installing ZSTD..."
+pushd zstd-${ZSTD_VERSION}
+make && make install
+popd
+
+echo "Compiling RocksDB..."
+cd /usr/local/rocksdb
+chown -R vagrant:vagrant /usr/local/rocksdb/
+sudo -u vagrant make static_lib
+cd examples/
+sudo -u vagrant make all
+sudo -u vagrant ./c_simple_example
@@ -57,7 +57,7 @@ struct LRUHandle {
  // Include the following flags:
  //   in_cache:    whether this entry is referenced by the hash table.
  //   is_high_pri: whether this entry is high priority entry.
-  //   in_high_pro_pool: whether this entry is in high-pri pool.
+  //   in_high_pri_pool: whether this entry is in high-pri pool.
  char flags;

  uint32_t hash;     // Hash of key(); used for fast sharding and comparisons
@@ -72,7 +72,7 @@ def display_file_coverage(per_file_coverage, total_coverage):
    header_template = \
        "%" + str(max_file_name_length) + "s\t%s\t%s"
    separator = "-" * (max_file_name_length + 10 + 20)
-    print header_template % ("Filename", "Coverage", "Lines")
+    print header_template % ("Filename", "Coverage", "Lines")  # noqa: E999 T25377293 Grandfathered in
    print separator

    # -- Print body
@@ -131,7 +131,8 @@ Status BuildTable(
    MergeHelper merge(env, internal_comparator.user_comparator(),
                      ioptions.merge_operator, nullptr, ioptions.info_log,
                      true /* internal key corruption is not ok */,
-                      snapshots.empty() ? 0 : snapshots.back());
+                      snapshots.empty() ? 0 : snapshots.back(),
+                      snapshot_checker);

    CompactionIterator c_iter(
        iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
@@ -1999,6 +1999,11 @@ void rocksdb_options_optimize_universal_style_compaction(
  opt->rep.OptimizeUniversalStyleCompaction(memtable_memory_budget);
 }

+void rocksdb_options_set_allow_ingest_behind(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.allow_ingest_behind = v;
+}
+
 void rocksdb_options_set_compaction_filter(
    rocksdb_options_t* opt,
    rocksdb_compactionfilter_t* filter) {
@@ -31,6 +31,7 @@
 #include "monitoring/thread_status_util.h"
 #include "options/options_helper.h"
 #include "table/block_based_table_factory.h"
+#include "table/merging_iterator.h"
 #include "util/autovector.h"
 #include "util/compression.h"

@@ -344,12 +345,13 @@ void SuperVersionUnrefHandle(void* ptr) {
  // When latter happens, we are in ~ColumnFamilyData(), no get should happen as
  // well.
  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
-  if (sv->Unref()) {
-    sv->db_mutex->Lock();
-    sv->Cleanup();
-    sv->db_mutex->Unlock();
-    delete sv;
-  }
+  bool was_last_ref __attribute__((__unused__));
+  was_last_ref = sv->Unref();
+  // Thread-local SuperVersions can't outlive ColumnFamilyData::super_version_.
+  // This is important because we can't do SuperVersion cleanup here.
+  // That would require locking DB mutex, which would deadlock because
+  // SuperVersionUnrefHandle is called with locked ThreadLocalPtr mutex.
+  assert(!was_last_ref);
 }
 }  // anonymous namespace

@@ -381,11 +383,13 @@ ColumnFamilyData::ColumnFamilyData(
      next_(nullptr),
      prev_(nullptr),
      log_number_(0),
+      flush_reason_(FlushReason::kUnknown),
      column_family_set_(column_family_set),
      pending_flush_(false),
      pending_compaction_(false),
      prev_compaction_needed_bytes_(0),
-      allow_2pc_(db_options.allow_2pc) {
+      allow_2pc_(db_options.allow_2pc),
+      last_memtable_id_(0) {
  Ref();

  // Convert user defined table properties collector factories to internal ones.
@@ -468,7 +472,7 @@ ColumnFamilyData::~ColumnFamilyData() {
    local_sv_.reset();
    super_version_->db_mutex->Lock();

-    bool is_last_reference __attribute__((unused));
+    bool is_last_reference __attribute__((__unused__));
    is_last_reference = super_version_->Unref();
    assert(is_last_reference);
    super_version_->Cleanup();
@@ -479,7 +483,7 @@ ColumnFamilyData::~ColumnFamilyData() {
  if (dummy_versions_ != nullptr) {
    // List must be empty
    assert(dummy_versions_->TEST_Next() == dummy_versions_);
-    bool deleted __attribute__((unused));
+    bool deleted __attribute__((__unused__));
    deleted = dummy_versions_->Unref();
    assert(deleted);
  }
@@ -630,6 +634,41 @@ int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
 }
 }  // namespace

+std::pair<WriteStallCondition, ColumnFamilyData::WriteStallCause>
+ColumnFamilyData::GetWriteStallConditionAndCause(
+    int num_unflushed_memtables, int num_l0_files,
+    uint64_t num_compaction_needed_bytes,
+    const MutableCFOptions& mutable_cf_options) {
+  if (num_unflushed_memtables >= mutable_cf_options.max_write_buffer_number) {
+    return {WriteStallCondition::kStopped, WriteStallCause::kMemtableLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             num_l0_files >= mutable_cf_options.level0_stop_writes_trigger) {
+    return {WriteStallCondition::kStopped, WriteStallCause::kL0FileCountLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
+             num_compaction_needed_bytes >=
+                 mutable_cf_options.hard_pending_compaction_bytes_limit) {
+    return {WriteStallCondition::kStopped,
+            WriteStallCause::kPendingCompactionBytes};
+  } else if (mutable_cf_options.max_write_buffer_number > 3 &&
+             num_unflushed_memtables >=
+                 mutable_cf_options.max_write_buffer_number - 1) {
+    return {WriteStallCondition::kDelayed, WriteStallCause::kMemtableLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
+             num_l0_files >=
+                 mutable_cf_options.level0_slowdown_writes_trigger) {
+    return {WriteStallCondition::kDelayed, WriteStallCause::kL0FileCountLimit};
+  } else if (!mutable_cf_options.disable_auto_compactions &&
+             mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
+             num_compaction_needed_bytes >=
+                 mutable_cf_options.soft_pending_compaction_bytes_limit) {
+    return {WriteStallCondition::kDelayed,
+            WriteStallCause::kPendingCompactionBytes};
+  }
+  return {WriteStallCondition::kNormal, WriteStallCause::kNone};
+}
+
 WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
      const MutableCFOptions& mutable_cf_options) {
  auto write_stall_condition = WriteStallCondition::kNormal;
@@ -639,25 +678,29 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
    uint64_t compaction_needed_bytes =
        vstorage->estimated_compaction_needed_bytes();

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

-    if (imm()->NumNotFlushed() >= mutable_cf_options.max_write_buffer_number) {
+    if (write_stall_condition == WriteStallCondition::kStopped &&
+        write_stall_cause == WriteStallCause::kMemtableLimit) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stopping writes because we have %d immutable memtables "
          "(waiting for flush), max_write_buffer_number is set to %d",
          name_.c_str(), imm()->NumNotFlushed(),
          mutable_cf_options.max_write_buffer_number);
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               vstorage->l0_delay_trigger_count() >=
-                   mutable_cf_options.level0_stop_writes_trigger) {
+    } else if (write_stall_condition == WriteStallCondition::kStopped &&
+               write_stall_cause == WriteStallCause::kL0FileCountLimit) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::L0_FILE_COUNT_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
        internal_stats_->AddCFStats(
            InternalStats::LOCKED_L0_FILE_COUNT_LIMIT_STOPS, 1);
@@ -665,28 +708,23 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
      ROCKS_LOG_WARN(ioptions_.info_log,
                     "[%s] Stopping writes because we have %d level-0 files",
                     name_.c_str(), vstorage->l0_delay_trigger_count());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
-               compaction_needed_bytes >=
-                   mutable_cf_options.hard_pending_compaction_bytes_limit) {
+    } else if (write_stall_condition == WriteStallCondition::kStopped &&
+               write_stall_cause == WriteStallCause::kPendingCompactionBytes) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(
          InternalStats::PENDING_COMPACTION_BYTES_LIMIT_STOPS, 1);
-      write_stall_condition = WriteStallCondition::kStopped;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stopping writes because of estimated pending compaction "
          "bytes %" PRIu64,
          name_.c_str(), compaction_needed_bytes);
-    } else if (mutable_cf_options.max_write_buffer_number > 3 &&
-               imm()->NumNotFlushed() >=
-                   mutable_cf_options.max_write_buffer_number - 1) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kMemtableLimit) {
      write_controller_token_ =
          SetupDelay(write_controller, compaction_needed_bytes,
                     prev_compaction_needed_bytes_, was_stopped,
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_LIMIT_SLOWDOWNS, 1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stalling writes because we have %d immutable memtables "
@@ -695,10 +733,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
          name_.c_str(), imm()->NumNotFlushed(),
          mutable_cf_options.max_write_buffer_number,
          write_controller->delayed_write_rate());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
-               vstorage->l0_delay_trigger_count() >=
-                   mutable_cf_options.level0_slowdown_writes_trigger) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kL0FileCountLimit) {
      // L0 is the last two files from stopping.
      bool near_stop = vstorage->l0_delay_trigger_count() >=
                       mutable_cf_options.level0_stop_writes_trigger - 2;
@@ -708,7 +744,6 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::L0_FILE_COUNT_LIMIT_SLOWDOWNS,
                                  1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
        internal_stats_->AddCFStats(
            InternalStats::LOCKED_L0_FILE_COUNT_LIMIT_SLOWDOWNS, 1);
@@ -718,10 +753,8 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     "rate %" PRIu64,
                     name_.c_str(), vstorage->l0_delay_trigger_count(),
                     write_controller->delayed_write_rate());
-    } else if (!mutable_cf_options.disable_auto_compactions &&
-               mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
-               vstorage->estimated_compaction_needed_bytes() >=
-                   mutable_cf_options.soft_pending_compaction_bytes_limit) {
+    } else if (write_stall_condition == WriteStallCondition::kDelayed &&
+               write_stall_cause == WriteStallCause::kPendingCompactionBytes) {
      // If the distance to hard limit is less than 1/4 of the gap between soft
      // and
      // hard bytes limit, we think it is near stop and speed up the slowdown.
@@ -739,7 +772,6 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(
          InternalStats::PENDING_COMPACTION_BYTES_LIMIT_SLOWDOWNS, 1);
-      write_stall_condition = WriteStallCondition::kDelayed;
      ROCKS_LOG_WARN(
          ioptions_.info_log,
          "[%s] Stalling writes because of estimated pending compaction "
@@ -747,6 +779,7 @@ WriteStallCondition ColumnFamilyData::RecalculateWriteStallConditions(
          name_.c_str(), vstorage->estimated_compaction_needed_bytes(),
          write_controller->delayed_write_rate());
    } else {
+      assert(write_stall_condition == WriteStallCondition::kNormal);
      if (vstorage->l0_delay_trigger_count() >=
          GetL0ThresholdSpeedupCompaction(
              mutable_cf_options.level0_file_num_compaction_trigger,
@@ -849,6 +882,68 @@ bool ColumnFamilyData::RangeOverlapWithCompaction(
      smallest_user_key, largest_user_key, level);
 }

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

@@ -867,8 +962,7 @@ Compaction* ColumnFamilyData::CompactRange(

 SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(
    InstrumentedMutex* db_mutex) {
-  SuperVersion* sv = nullptr;
-  sv = GetThreadLocalSuperVersion(db_mutex);
+  SuperVersion* sv = GetThreadLocalSuperVersion(db_mutex);
  sv->Ref();
  if (!ReturnThreadLocalSuperVersion(sv)) {
    // This Unref() corresponds to the Ref() in GetThreadLocalSuperVersion()
@@ -882,7 +976,6 @@ SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(

 SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
    InstrumentedMutex* db_mutex) {
-  SuperVersion* sv = nullptr;
  // The SuperVersion is cached in thread local storage to avoid acquiring
  // mutex when SuperVersion does not change since the last use. When a new
  // SuperVersion is installed, the compaction or flush thread cleans up
@@ -901,7 +994,7 @@ SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
  // should only keep kSVInUse before ReturnThreadLocalSuperVersion call
  // (if no Scrape happens).
  assert(ptr != SuperVersion::kSVInUse);
-  sv = static_cast<SuperVersion*>(ptr);
+  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
  if (sv == SuperVersion::kSVObsolete ||
      sv->version_number != super_version_number_.load()) {
    RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_ACQUIRES);
@@ -965,6 +1058,12 @@ void ColumnFamilyData::InstallSuperVersion(
      RecalculateWriteStallConditions(mutable_cf_options);

  if (old_superversion != nullptr) {
+    // Reset SuperVersions cached in thread local storage.
+    // This should be done before old_superversion->Unref(). That's to ensure
+    // that local_sv_ never holds the last reference to SuperVersion, since
+    // it has no means to safely do SuperVersion cleanup.
+    ResetThreadLocalSuperVersions();
+
    if (old_superversion->mutable_cf_options.write_buffer_size !=
        mutable_cf_options.write_buffer_size) {
      mem_->UpdateWriteBufferSize(mutable_cf_options.write_buffer_size);
@@ -980,9 +1079,6 @@ void ColumnFamilyData::InstallSuperVersion(
      sv_context->superversions_to_free.push_back(old_superversion);
    }
  }
-
-  // Reset SuperVersions cached in thread local storage
-  ResetThreadLocalSuperVersions();
 }

 void ColumnFamilyData::ResetThreadLocalSuperVersions() {
@@ -994,10 +1090,12 @@ void ColumnFamilyData::ResetThreadLocalSuperVersions() {
      continue;
    }
    auto sv = static_cast<SuperVersion*>(ptr);
-    if (sv->Unref()) {
-      sv->Cleanup();
-      delete sv;
-    }
+    bool was_last_ref __attribute__((__unused__));
+    was_last_ref = sv->Unref();
+    // sv couldn't have been the last reference because
+    // ResetThreadLocalSuperVersions() is called before
+    // unref'ing super_version_.
+    assert(!was_last_ref);
  }
 }

@@ -1059,10 +1157,14 @@ ColumnFamilySet::~ColumnFamilySet() {
  while (column_family_data_.size() > 0) {
    // cfd destructor will delete itself from column_family_data_
    auto cfd = column_family_data_.begin()->second;
-    cfd->Unref();
+    bool last_ref __attribute__((__unused__));
+    last_ref = cfd->Unref();
+    assert(last_ref);
    delete cfd;
  }
-  dummy_cfd_->Unref();
+  bool dummy_last_ref __attribute__((__unused__));
+  dummy_last_ref = dummy_cfd_->Unref();
+  assert(dummy_last_ref);
  delete dummy_cfd_;
 }

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

 class Version;
 class VersionSet;
+class VersionStorageInfo;
 class MemTable;
 class MemTableListVersion;
 class CompactionPicker;
@@ -202,6 +203,10 @@ class ColumnFamilyData {
  void SetLogNumber(uint64_t log_number) { log_number_ = log_number; }
  uint64_t GetLogNumber() const { return log_number_; }

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

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

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

+  enum class WriteStallCause {
+    kNone,
+    kMemtableLimit,
+    kL0FileCountLimit,
+    kPendingCompactionBytes,
+  };
+  static std::pair<WriteStallCondition, WriteStallCause>
+  GetWriteStallConditionAndCause(int num_unflushed_memtables, int num_l0_files,
+                                 uint64_t num_compaction_needed_bytes,
+                                 const MutableCFOptions& mutable_cf_options);
+
  // Recalculate some small conditions, which are changed only during
  // compaction, adding new memtable and/or
  // recalculation of compaction score. These values are used in
@@ -400,6 +430,8 @@ class ColumnFamilyData {
  // recovered from
  uint64_t log_number_;

+  std::atomic<FlushReason> flush_reason_;
+
  // An object that keeps all the compaction stats
  // and picks the next compaction
  std::unique_ptr<CompactionPicker> compaction_picker_;
@@ -419,6 +451,9 @@ class ColumnFamilyData {

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

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

-// This test checks for automatic getting a conflict if there is a
-// manual which has not yet been scheduled.
-// The manual compaction waits in NotScheduled
-// We generate more files and then trigger an automatic compaction
-// This will wait because there is an unscheduled manual compaction.
-// Once the conflict is hit, the manual compaction starts and ends
-// Then another automatic will start and end.
-TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
-  Open();
-  CreateColumnFamilies({"one"});
-  ColumnFamilyOptions default_cf, one;
-  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.max_background_compactions = 3;
-
-  default_cf.compaction_style = kCompactionStyleLevel;
-  default_cf.num_levels = 3;
-  default_cf.write_buffer_size = 64 << 10;  // 64KB
-  default_cf.target_file_size_base = 30 << 10;
-  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
-  table_options.no_block_cache = true;
-  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
-
-  one.compaction_style = kCompactionStyleUniversal;
-
-  one.num_levels = 1;
-  // trigger compaction if there are >= 4 files
-  one.level0_file_num_compaction_trigger = 4;
-  one.write_buffer_size = 120000;
-
-  Reopen({default_cf, one});
-  // make sure all background compaction jobs can be scheduled
-  auto stop_token =
-      dbfull()->TEST_write_controler().GetCompactionPressureToken();
-
-  // SETUP column family "one" -- universal style
-  for (int i = 0; i < one.level0_file_num_compaction_trigger - 2; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-    AssertFilesPerLevel(ToString(i + 1), 1);
-  }
-  bool cf_1_1 = true;
-  bool cf_1_2 = true;
-  rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      {{"DBImpl::BackgroundCompaction()::Conflict",
-        "ColumnFamilyTest::ManualAutoCon:7"},
-       {"ColumnFamilyTest::ManualAutoCon:9",
-        "ColumnFamilyTest::ManualAutoCon:8"},
-       {"ColumnFamilyTest::ManualAutoCon:2",
-        "ColumnFamilyTest::ManualAutoCon:6"},
-       {"ColumnFamilyTest::ManualAutoCon:4",
-        "ColumnFamilyTest::ManualAutoCon:5"},
-       {"ColumnFamilyTest::ManualAutoCon:1",
-        "ColumnFamilyTest::ManualAutoCon:2"},
-       {"ColumnFamilyTest::ManualAutoCon:1",
-        "ColumnFamilyTest::ManualAutoCon:3"}});
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
-        if (cf_1_1) {
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:4");
-          cf_1_1 = false;
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:3");
-        } else if (cf_1_2) {
-          cf_1_2 = false;
-          TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:2");
-        }
-      });
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::RunManualCompaction:NotScheduled", [&](void* arg) {
-        InstrumentedMutex* mutex = static_cast<InstrumentedMutex*>(arg);
-        mutex->Unlock();
-        TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:9");
-        TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:7");
-        mutex->Lock();
-      });
-
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  rocksdb::port::Thread threads([&] {
-    CompactRangeOptions compact_options;
-    compact_options.exclusive_manual_compaction = false;
-    ASSERT_OK(
-        db_->CompactRange(compact_options, handles_[1], nullptr, nullptr));
-    TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:6");
-  });
-
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:8");
-  WaitForFlush(1);
-
-  // Add more L0 files and force automatic compaction
-  for (int i = 0; i < one.level0_file_num_compaction_trigger; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-    AssertFilesPerLevel(ToString(one.level0_file_num_compaction_trigger + i),
-                        1);
-  }
-
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:5");
-  // Add more L0 files and force automatic compaction
-  for (int i = 0; i < one.level0_file_num_compaction_trigger; ++i) {
-    PutRandomData(1, 10, 12000, true);
-    PutRandomData(1, 1, 10, true);
-    WaitForFlush(1);
-  }
-  TEST_SYNC_POINT("ColumnFamilyTest::ManualAutoCon:1");
-
-  threads.join();
-  WaitForCompaction();
-  // VERIFY compaction "one"
-  ASSERT_LE(NumTableFilesAtLevel(0, 1), 3);
-
-  // Compare against saved keys
-  std::set<std::string>::iterator key_iter = keys_.begin();
-  while (key_iter != keys_.end()) {
-    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
-    key_iter++;
-  }
-}
-
 // In this test, we generate enough files to trigger automatic compactions.
 // The automatic compaction waits in NonTrivial:AfterRun
 // We generate more files and then trigger an automatic compaction
@@ -405,20 +405,23 @@ void Compaction::Summary(char* output, int len) {
 uint64_t Compaction::OutputFilePreallocationSize() const {
  uint64_t preallocation_size = 0;

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

 std::unique_ptr<CompactionFilter> Compaction::CreateCompactionFilter() const {
@@ -184,8 +184,11 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
                                              Slice* skip_until) {
  if (compaction_filter_ != nullptr &&
      (ikey_.type == kTypeValue || ikey_.type == kTypeBlobIndex) &&
-      (visible_at_tip_ || ikey_.sequence > latest_snapshot_ ||
-       ignore_snapshots_)) {
+      (visible_at_tip_ || ignore_snapshots_ ||
+       ikey_.sequence > latest_snapshot_ ||
+       (snapshot_checker_ != nullptr &&
+        UNLIKELY(!snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                  latest_snapshot_))))) {
    // If the user has specified a compaction filter and the sequence
    // number is greater than any external snapshot, then invoke the
    // filter. If the return value of the compaction filter is true,
@@ -411,7 +414,10 @@ void CompactionIterator::NextFromInput() {
          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
        // Check whether the next key belongs to the same snapshot as the
        // SingleDelete.
-        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot) {
+        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
+            (snapshot_checker_ != nullptr &&
+             UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
+                                                       prev_snapshot)))) {
          if (next_ikey.type == kTypeSingleDeletion) {
            // We encountered two SingleDeletes in a row.  This could be due to
            // unexpected user input.
@@ -422,8 +428,12 @@ void CompactionIterator::NextFromInput() {
            // input_->Next().
            ++iter_stats_.num_record_drop_obsolete;
            ++iter_stats_.num_single_del_mismatch;
-          } else if ((ikey_.sequence <= earliest_write_conflict_snapshot_) ||
-                     has_outputted_key_) {
+          } else if (has_outputted_key_ ||
+                     (ikey_.sequence <= earliest_write_conflict_snapshot_ &&
+                      (snapshot_checker_ == nullptr ||
+                       LIKELY(snapshot_checker_->IsInSnapshot(
+                           ikey_.sequence,
+                           earliest_write_conflict_snapshot_))))) {
            // Found a matching value, we can drop the single delete and the
            // value.  It is safe to drop both records since we've already
            // outputted a key in this snapshot, or there is no earlier
@@ -471,6 +481,9 @@ void CompactionIterator::NextFromInput() {
        // comparison, so the value of has_current_user_key does not matter.
        has_current_user_key_ = false;
        if (compaction_ != nullptr && ikey_.sequence <= earliest_snapshot_ &&
+            (snapshot_checker_ == nullptr ||
+             LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                    earliest_snapshot_))) &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_)) {
          // Key doesn't exist outside of this range.
@@ -504,6 +517,9 @@ void CompactionIterator::NextFromInput() {
      input_->Next();
    } else if (compaction_ != nullptr && ikey_.type == kTypeDeletion &&
               ikey_.sequence <= earliest_snapshot_ &&
+               (snapshot_checker_ == nullptr ||
+                LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
+                                                       earliest_snapshot_))) &&
               ikeyNotNeededForIncrementalSnapshot() &&
               compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                          &level_ptrs_)) {
@@ -521,6 +537,11 @@ void CompactionIterator::NextFromInput() {
      //
      // Note:  Dropping this Delete will not affect TransactionDB
      // write-conflict checking since it is earlier than any snapshot.
+      //
+      // It seems that we can also drop deletion later than earliest snapshot
+      // given that:
+      // (1) The deletion is earlier than earliest_write_conflict_snapshot, and
+      // (2) No value exist earlier than the deletion.
      ++iter_stats_.num_record_drop_obsolete;
      if (!bottommost_level_) {
        ++iter_stats_.num_optimized_del_drop_obsolete;
@@ -601,9 +622,12 @@ void CompactionIterator::PrepareOutput() {
  // and the earliest snapshot is larger than this seqno
  // and the userkey differs from the last userkey in compaction
  // then we can squash the seqno to zero.
-
+  //
  // This is safe for TransactionDB write-conflict checking since transactions
  // only care about sequence number larger than any active snapshots.
+  //
+  // Can we do the same for levels above bottom level as long as
+  // KeyNotExistsBeyondOutputLevel() return true?
  if ((compaction_ != nullptr &&
      !compaction_->allow_ingest_behind()) &&
      ikeyNotNeededForIncrementalSnapshot() &&
@@ -9,8 +9,10 @@
 #include <vector>

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

 namespace rocksdb {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

 int main(int argc, char** argv) {
@@ -625,7 +625,8 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
      "[%s] compacted to: %s, MB/sec: %.1f rd, %.1f wr, level %d, "
      "files in(%d, %d) out(%d) "
      "MB in(%.1f, %.1f) out(%.1f), read-write-amplify(%.1f) "
-      "write-amplify(%.1f) %s, records in: %d, records dropped: %d\n",
+      "write-amplify(%.1f) %s, records in: %" PRIu64
+      ", records dropped: %" PRIu64 " output_compression: %s\n",
      cfd->GetName().c_str(), vstorage->LevelSummary(&tmp), bytes_read_per_sec,
      bytes_written_per_sec, compact_->compaction->output_level(),
      stats.num_input_files_in_non_output_levels,
@@ -634,20 +635,23 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
      stats.bytes_read_output_level / 1048576.0,
      stats.bytes_written / 1048576.0, read_write_amp, write_amp,
      status.ToString().c_str(), stats.num_input_records,
-      stats.num_dropped_records);
+      stats.num_dropped_records,
+      CompressionTypeToString(compact_->compaction->output_compression())
+          .c_str());

  UpdateCompactionJobStats(stats);

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

  if (compaction_job_stats_ != nullptr) {
    stream << "num_single_delete_mismatches"
@@ -743,8 +747,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      compaction_filter, db_options_.info_log.get(),
      false /* internal key corruption is expected */,
      existing_snapshots_.empty() ? 0 : existing_snapshots_.back(),
-      compact_->compaction->level(), db_options_.statistics.get(),
-      shutting_down_);
+      snapshot_checker_, compact_->compaction->level(),
+      db_options_.statistics.get(), shutting_down_);

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

@@ -175,6 +175,8 @@ TEST_F(CompactionPickerTest, Level1Trigger) {
 }

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

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

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

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

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
@@ -603,9 +603,9 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(

  CompactionReason compaction_reason;
  if (max_number_of_files_to_compact == UINT_MAX) {
-    compaction_reason = CompactionReason::kUniversalSortedRunNum;
-  } else {
    compaction_reason = CompactionReason::kUniversalSizeRatio;
+  } else {
+    compaction_reason = CompactionReason::kUniversalSortedRunNum;
  }
  return new Compaction(
      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
@@ -19,9 +19,17 @@ void CancelAllBackgroundWork(DB* db, bool wait) {
 }

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

 Status VerifySstFileChecksum(const Options& options,
@@ -9,6 +9,7 @@
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
+#include "util/fault_injection_test_env.h"
 #if !defined(ROCKSDB_LITE)
 #include "util/sync_point.h"
 #endif
@@ -570,19 +571,19 @@ TEST_F(DBBasicTest, CompactBetweenSnapshots) {

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

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

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

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

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

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

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

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

+TEST_F(DBBlockCacheTest, IteratorBlockCacheUsage) {
+  ReadOptions read_options;
+  read_options.fill_cache = false;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(0, 0, false);
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  ASSERT_EQ(0, cache->GetUsage());
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(0));
+  ASSERT_LT(0, cache->GetUsage());
+  delete iter;
+  iter = nullptr;
+  ASSERT_EQ(0, cache->GetUsage());
+}
+
 TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
  ReadOptions read_options;
  auto table_options = GetTableOptions();
@@ -1517,6 +1517,122 @@ TEST_F(DBCompactionTest, DeleteFileRange) {
  ASSERT_GT(old_num_files, new_num_files);
 }

+TEST_F(DBCompactionTest, DeleteFilesInRanges) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1024 * 1024;
+  options.max_bytes_for_level_multiplier = 2;
+  options.num_levels = 4;
+  options.max_background_compactions = 3;
+  options.disable_auto_compactions = true;
+
+  DestroyAndReopen(options);
+  int32_t value_size = 10 * 1024;  // 10 KB
+
+  Random rnd(301);
+  std::map<int32_t, std::string> values;
+
+  // file [0 => 100), [100 => 200), ... [900, 1000)
+  for (auto i = 0; i < 10; i++) {
+    for (auto j = 0; j < 100; j++) {
+      auto k = i * 100 + j;
+      values[k] = RandomString(&rnd, value_size);
+      ASSERT_OK(Put(Key(k), values[k]));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_EQ("10", FilesPerLevel(0));
+  CompactRangeOptions compact_options;
+  compact_options.change_level = true;
+  compact_options.target_level = 2;
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  ASSERT_EQ("0,0,10", FilesPerLevel(0));
+
+  // file [0 => 100), [200 => 300), ... [800, 900)
+  for (auto i = 0; i < 10; i+=2) {
+    for (auto j = 0; j < 100; j++) {
+      auto k = i * 100 + j;
+      ASSERT_OK(Put(Key(k), values[k]));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_EQ("5,0,10", FilesPerLevel(0));
+  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
+  ASSERT_EQ("0,5,10", FilesPerLevel(0));
+
+  // Delete files in range [0, 299] (inclusive)
+  {
+    auto begin_str1 = Key(0), end_str1 = Key(100);
+    auto begin_str2 = Key(100), end_str2 = Key(200);
+    auto begin_str3 = Key(200), end_str3 = Key(299);
+    Slice begin1(begin_str1), end1(end_str1);
+    Slice begin2(begin_str2), end2(end_str2);
+    Slice begin3(begin_str3), end3(end_str3);
+    std::vector<RangePtr> ranges;
+    ranges.push_back(RangePtr(&begin1, &end1));
+    ranges.push_back(RangePtr(&begin2, &end2));
+    ranges.push_back(RangePtr(&begin3, &end3));
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(),
+                                  ranges.data(), ranges.size()));
+    ASSERT_EQ("0,3,7", FilesPerLevel(0));
+
+    // Keys [0, 300) should not exist.
+    for (auto i = 0; i < 300; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+    for (auto i = 300; i < 1000; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+  }
+
+  // Delete files in range [600, 999) (exclusive)
+  {
+    auto begin_str1 = Key(600), end_str1 = Key(800);
+    auto begin_str2 = Key(700), end_str2 = Key(900);
+    auto begin_str3 = Key(800), end_str3 = Key(999);
+    Slice begin1(begin_str1), end1(end_str1);
+    Slice begin2(begin_str2), end2(end_str2);
+    Slice begin3(begin_str3), end3(end_str3);
+    std::vector<RangePtr> ranges;
+    ranges.push_back(RangePtr(&begin1, &end1));
+    ranges.push_back(RangePtr(&begin2, &end2));
+    ranges.push_back(RangePtr(&begin3, &end3));
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(),
+                                  ranges.data(), ranges.size(), false));
+    ASSERT_EQ("0,1,4", FilesPerLevel(0));
+
+    // Keys [600, 900) should not exist.
+    for (auto i = 600; i < 900; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+    for (auto i = 300; i < 600; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+    for (auto i = 900; i < 1000; i++) {
+      ASSERT_EQ(Get(Key(i)), values[i]);
+    }
+  }
+
+  // Delete all files.
+  {
+    RangePtr range;
+    ASSERT_OK(DeleteFilesInRanges(db_, db_->DefaultColumnFamily(), &range, 1));
+    ASSERT_EQ("", FilesPerLevel(0));
+
+    for (auto i = 0; i < 1000; i++) {
+      ReadOptions ropts;
+      std::string result;
+      auto s = db_->Get(ropts, Key(i), &result);
+      ASSERT_TRUE(s.IsNotFound());
+    }
+  }
+}
+
 TEST_F(DBCompactionTest, DeleteFileRangeFileEndpointsOverlapBug) {
  // regression test for #2833: groups of files whose user-keys overlap at the
  // endpoints could be split by `DeleteFilesInRange`. This caused old data to
@@ -2989,6 +3105,276 @@ TEST_F(DBCompactionTest, CompactBottomLevelFilesWithDeletions) {
  }
 }

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

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

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

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

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

@@ -123,13 +123,13 @@ CompressionType GetCompressionFlush(
 namespace {
 void DumpSupportInfo(Logger* logger) {
  ROCKS_LOG_HEADER(logger, "Compression algorithms supported:");
-  ROCKS_LOG_HEADER(logger, "\tSnappy supported: %d", Snappy_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZlib supported: %d", Zlib_Supported());
-  ROCKS_LOG_HEADER(logger, "\tBzip supported: %d", BZip2_Supported());
-  ROCKS_LOG_HEADER(logger, "\tLZ4 supported: %d", LZ4_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZSTDNotFinal supported: %d",
-                   ZSTDNotFinal_Supported());
-  ROCKS_LOG_HEADER(logger, "\tZSTD supported: %d", ZSTD_Supported());
+  for (auto& compression : OptionsHelper::compression_type_string_map) {
+    if (compression.second != kNoCompression &&
+        compression.second != kDisableCompressionOption) {
+      ROCKS_LOG_HEADER(logger, "\t%s supported: %d", compression.first.c_str(),
+                       CompressionTypeSupported(compression.second));
+    }
+  }
  ROCKS_LOG_HEADER(logger, "Fast CRC32 supported: %s",
                   crc32c::IsFastCrc32Supported().c_str());
 }
@@ -141,6 +141,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
               const bool seq_per_batch)
    : env_(options.env),
      dbname_(dbname),
+      own_info_log_(options.info_log == nullptr),
      initial_db_options_(SanitizeOptions(dbname, options)),
      immutable_db_options_(initial_db_options_),
      mutable_db_options_(initial_db_options_),
@@ -177,6 +178,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
      num_running_flushes_(0),
      bg_purge_scheduled_(0),
      disable_delete_obsolete_files_(0),
+      pending_purge_obsolete_files_(0),
      delete_obsolete_files_last_run_(env_->NowMicros()),
      last_stats_dump_time_microsec_(0),
      next_job_id_(1),
@@ -212,7 +214,8 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
      // requires a custom gc for compaction, we use that to set use_custom_gc_
      // as well.
      use_custom_gc_(seq_per_batch),
-      preserve_deletes_(options.preserve_deletes) {
+      preserve_deletes_(options.preserve_deletes),
+      closed_(false) {
  env_->GetAbsolutePath(dbname, &db_absolute_path_);

  // Reserve ten files or so for other uses and give the rest to TableCache.
@@ -255,7 +258,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
      if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
        cfd->Ref();
        mutex_.Unlock();
-        FlushMemTable(cfd, FlushOptions());
+        FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
        mutex_.Lock();
        cfd->Unref();
      }
@@ -275,7 +278,7 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
  }
 }

-DBImpl::~DBImpl() {
+Status DBImpl::CloseHelper() {
  // CancelAllBackgroundWork called with false means we just set the shutdown
  // marker. After this we do a variant of the waiting and unschedule work
  // (to consider: moving all the waiting into CancelAllBackgroundWork(true))
@@ -284,6 +287,7 @@ DBImpl::~DBImpl() {
      env_->UnSchedule(this, Env::Priority::BOTTOM);
  int compactions_unscheduled = env_->UnSchedule(this, Env::Priority::LOW);
  int flushes_unscheduled = env_->UnSchedule(this, Env::Priority::HIGH);
+  Status ret;
  mutex_.Lock();
  bg_bottom_compaction_scheduled_ -= bottom_compactions_unscheduled;
  bg_compaction_scheduled_ -= compactions_unscheduled;
@@ -291,7 +295,8 @@ DBImpl::~DBImpl() {

  // Wait for background work to finish
  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
-         bg_flush_scheduled_ || bg_purge_scheduled_) {
+         bg_flush_scheduled_ || bg_purge_scheduled_ ||
+         pending_purge_obsolete_files_) {
    TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
    bg_cv_.Wait();
  }
@@ -345,7 +350,18 @@ DBImpl::~DBImpl() {
    delete l;
  }
  for (auto& log : logs_) {
-    log.ClearWriter();
+    uint64_t log_number = log.writer->get_log_number();
+    Status s = log.ClearWriter();
+    if (!s.ok()) {
+      ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                     "Unable to Sync WAL file %s with error -- %s",
+                     LogFileName(immutable_db_options_.wal_dir, log_number).c_str(),
+                     s.ToString().c_str());
+      // Retain the first error
+      if (ret.ok()) {
+        ret = s;
+      }
+    }
  }
  logs_.clear();

@@ -378,6 +394,25 @@ DBImpl::~DBImpl() {

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

 void DBImpl::MaybeIgnoreError(Status* s) const {
@@ -523,6 +558,7 @@ Status DBImpl::SetOptions(ColumnFamilyHandle* column_family,

      persist_options_status = WriteOptionsFile(
          false /*need_mutex_lock*/, true /*need_enter_write_thread*/);
+      bg_cv_.SignalAll();
    }
  }
  sv_context.Clean();
@@ -1407,6 +1443,7 @@ Status DBImpl::DropColumnFamilyImpl(ColumnFamilyHandle* column_family) {
      }
      is_snapshot_supported_ = new_is_snapshot_supported;
    }
+    bg_cv_.SignalAll();
  }

  if (s.ok()) {
@@ -1510,7 +1547,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
                                            ColumnFamilyData* cfd,
                                            SequenceNumber snapshot,
                                            ReadCallback* read_callback,
-                                            bool allow_blob) {
+                                            bool allow_blob,
+                                            bool allow_refresh) {
  SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);

  // Try to generate a DB iterator tree in continuous memory area to be
@@ -1559,7 +1597,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
      env_, read_options, *cfd->ioptions(), snapshot,
      sv->mutable_cf_options.max_sequential_skip_in_iterations,
      sv->version_number, read_callback,
-      ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob);
+      ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob,
+      allow_refresh);

  InternalIterator* internal_iter =
      NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
@@ -1682,12 +1721,6 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
  delete casted_s;
 }

-bool DBImpl::HasActiveSnapshotInRange(SequenceNumber lower_bound,
-                                      SequenceNumber upper_bound) {
-  InstrumentedMutexLock l(&mutex_);
-  return snapshots_.HasSnapshotInRange(lower_bound, upper_bound);
-}
-
 #ifndef ROCKSDB_LITE
 Status DBImpl::GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
                                        TablePropertiesCollection* props) {
@@ -2051,8 +2084,7 @@ Status DBImpl::DeleteFile(std::string name) {
                      name.c_str());
      return Status::NotSupported("Delete only supported for archived logs");
    }
-    status =
-        env_->DeleteFile(immutable_db_options_.wal_dir + "/" + name.c_str());
+    status = wal_manager_.DeleteFile(name, number);
    if (!status.ok()) {
      ROCKS_LOG_ERROR(immutable_db_options_.info_log,
                      "DeleteFile %s failed -- %s.\n", name.c_str(),
@@ -2131,52 +2163,63 @@ Status DBImpl::DeleteFile(std::string name) {
  return status;
 }

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

    auto* vstorage = input_version->storage_info();
-    for (int i = 1; i < cfd->NumberLevels(); i++) {
-      if (vstorage->LevelFiles(i).empty() ||
-          !vstorage->OverlapInLevel(i, begin, end)) {
-        continue;
-      }
-      std::vector<FileMetaData*> level_files;
-      InternalKey begin_storage, end_storage, *begin_key, *end_key;
-      if (begin == nullptr) {
-        begin_key = nullptr;
-      } else {
-        begin_storage.SetMinPossibleForUserKey(*begin);
-        begin_key = &begin_storage;
-      }
-      if (end == nullptr) {
-        end_key = nullptr;
-      } else {
-        end_storage.SetMaxPossibleForUserKey(*end);
-        end_key = &end_storage;
-      }
-
-      vstorage->GetCleanInputsWithinInterval(i, begin_key, end_key,
-                                             &level_files, -1 /* hint_index */,
-                                             nullptr /* file_index */);
-      FileMetaData* level_file;
-      for (uint32_t j = 0; j < level_files.size(); j++) {
-        level_file = level_files[j];
-        if (level_file->being_compacted) {
+    for (size_t r = 0; r < n; r++) {
+      auto begin = ranges[r].start, end = ranges[r].limit;
+      for (int i = 1; i < cfd->NumberLevels(); i++) {
+        if (vstorage->LevelFiles(i).empty() ||
+            !vstorage->OverlapInLevel(i, begin, end)) {
          continue;
        }
-        edit.SetColumnFamily(cfd->GetID());
-        edit.DeleteFile(i, level_file->fd.GetNumber());
-        deleted_files.push_back(level_file);
-        level_file->being_compacted = true;
+        std::vector<FileMetaData*> level_files;
+        InternalKey begin_storage, end_storage, *begin_key, *end_key;
+        if (begin == nullptr) {
+          begin_key = nullptr;
+        } else {
+          begin_storage.SetMinPossibleForUserKey(*begin);
+          begin_key = &begin_storage;
+        }
+        if (end == nullptr) {
+          end_key = nullptr;
+        } else {
+          end_storage.SetMaxPossibleForUserKey(*end);
+          end_key = &end_storage;
+        }
+
+        vstorage->GetCleanInputsWithinInterval(i, begin_key, end_key,
+                                               &level_files, -1 /* hint_index */,
+                                               nullptr /* file_index */);
+        FileMetaData* level_file;
+        for (uint32_t j = 0; j < level_files.size(); j++) {
+          level_file = level_files[j];
+          if (level_file->being_compacted) {
+            continue;
+          }
+          if (deleted_files.find(level_file) != deleted_files.end()) {
+            continue;
+          }
+          if (!include_end && end != nullptr &&
+              cfd->user_comparator()->Compare(level_file->largest.user_key(), *end) == 0) {
+            continue;
+          }
+          edit.SetColumnFamily(cfd->GetID());
+          edit.DeleteFile(i, level_file->fd.GetNumber());
+          deleted_files.insert(level_file);
+          level_file->being_compacted = true;
+        }
      }
    }
    if (edit.GetDeletedFiles().empty()) {
@@ -2325,7 +2368,15 @@ Status DB::DestroyColumnFamilyHandle(ColumnFamilyHandle* column_family) {
  return Status::OK();
 }

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

 Status DB::ListColumnFamilies(const DBOptions& db_options,
                              const std::string& name,
@@ -2781,12 +2832,14 @@ Status DBImpl::IngestExternalFile(
    // Figure out if we need to flush the memtable first
    if (status.ok()) {
      bool need_flush = false;
-      status = ingestion_job.NeedsFlush(&need_flush);
+      status = ingestion_job.NeedsFlush(&need_flush, cfd->GetSuperVersion());
      TEST_SYNC_POINT_CALLBACK("DBImpl::IngestExternalFile:NeedFlush",
                               &need_flush);
      if (status.ok() && need_flush) {
        mutex_.Unlock();
-        status = FlushMemTable(cfd, FlushOptions(), true /* writes_stopped */);
+        status = FlushMemTable(cfd, FlushOptions(),
+                               FlushReason::kExternalFileIngestion,
+                               true /* writes_stopped */);
        mutex_.Lock();
      }
    }
@@ -148,7 +148,8 @@ class DBImpl : public DB {
                                      ColumnFamilyData* cfd,
                                      SequenceNumber snapshot,
                                      ReadCallback* read_callback,
-                                      bool allow_blob = false);
+                                      bool allow_blob = false,
+                                      bool allow_refresh = true);

  virtual const Snapshot* GetSnapshot() override;
  virtual void ReleaseSnapshot(const Snapshot* snapshot) override;
@@ -229,10 +230,6 @@ class DBImpl : public DB {

  virtual bool SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) override;

-  // Whether there is an active snapshot in range [lower_bound, upper_bound).
-  bool HasActiveSnapshotInRange(SequenceNumber lower_bound,
-                                SequenceNumber upper_bound);
-
 #ifndef ROCKSDB_LITE
  using DB::ResetStats;
  virtual Status ResetStats() override;
@@ -250,8 +247,9 @@ class DBImpl : public DB {
      const TransactionLogIterator::ReadOptions&
          read_options = TransactionLogIterator::ReadOptions()) override;
  virtual Status DeleteFile(std::string name) override;
-  Status DeleteFilesInRange(ColumnFamilyHandle* column_family,
-                            const Slice* begin, const Slice* end);
+  Status DeleteFilesInRanges(ColumnFamilyHandle* column_family,
+                             const RangePtr* ranges, size_t n,
+                             bool include_end = true);

  virtual void GetLiveFilesMetaData(
      std::vector<LiveFileMetaData>* metadata) override;
@@ -471,6 +469,8 @@ class DBImpl : public DB {
  // belong to live files are posibly removed. Also, removes all the
  // files in sst_delete_files and log_delete_files.
  // It is not necessary to hold the mutex when invoking this method.
+  // If FindObsoleteFiles() was run, we need to also run
+  // PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
  void PurgeObsoleteFiles(const JobContext& background_contet,
                          bool schedule_only = false);

@@ -614,10 +614,14 @@ class DBImpl : public DB {
                     std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
                     const bool seq_per_batch);

+  virtual Status Close() override;
+
 protected:
  Env* const env_;
  const std::string dbname_;
  unique_ptr<VersionSet> versions_;
+  // Flag to check whether we allocated and own the info log file
+  bool own_info_log_;
  const DBOptions initial_db_options_;
  const ImmutableDBOptions immutable_db_options_;
  MutableDBOptions mutable_db_options_;
@@ -669,10 +673,14 @@ class DBImpl : public DB {
  // batch which will be written to memtable later during the commit, and in
  // WritePrepared it is guaranteed since it will be used only for WAL markers
  // which will never be written to memtable.
+  // batch_cnt is expected to be non-zero in seq_per_batch mode and indicates
+  // the number of sub-patches. A sub-patch is a subset of the write batch that
+  // does not have duplicate keys.
  Status WriteImpl(const WriteOptions& options, WriteBatch* updates,
                   WriteCallback* callback = nullptr,
                   uint64_t* log_used = nullptr, uint64_t log_ref = 0,
                   bool disable_memtable = false, uint64_t* seq_used = nullptr,
+                   size_t batch_cnt = 0,
                   PreReleaseCallback* pre_release_callback = nullptr);

  Status PipelinedWriteImpl(const WriteOptions& options, WriteBatch* updates,
@@ -681,10 +689,13 @@ class DBImpl : public DB {
                            bool disable_memtable = false,
                            uint64_t* seq_used = nullptr);

+  // batch_cnt is expected to be non-zero in seq_per_batch mode and indicates
+  // the number of sub-patches. A sub-patch is a subset of the write batch that
+  // does not have duplicate keys.
  Status WriteImplWALOnly(const WriteOptions& options, WriteBatch* updates,
                          WriteCallback* callback = nullptr,
                          uint64_t* log_used = nullptr, uint64_t log_ref = 0,
-                          uint64_t* seq_used = nullptr,
+                          uint64_t* seq_used = nullptr, size_t batch_cnt = 0,
                          PreReleaseCallback* pre_release_callback = nullptr);

  uint64_t FindMinLogContainingOutstandingPrep();
@@ -693,6 +704,9 @@ class DBImpl : public DB {
  // The writer must be the leader in write_thread_ and holding mutex_
  Status WriteRecoverableState();

+  // Actual implementation of Close()
+  Status CloseImpl();
+
 private:
  friend class DB;
  friend class InternalStats;
@@ -808,10 +822,14 @@ class DBImpl : public DB {

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

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

  // REQUIRES: mutex locked
  Status SwitchWAL(WriteContext* write_context);
@@ -866,7 +884,7 @@ class DBImpl : public DB {
  ColumnFamilyData* GetColumnFamilyDataByName(const std::string& cf_name);

  void MaybeScheduleFlushOrCompaction();
-  void SchedulePendingFlush(ColumnFamilyData* cfd);
+  void SchedulePendingFlush(ColumnFamilyData* cfd, FlushReason flush_reason);
  void SchedulePendingCompaction(ColumnFamilyData* cfd);
  void SchedulePendingPurge(std::string fname, FileType type, uint64_t number,
                            uint32_t path_id, int job_id);
@@ -905,7 +923,7 @@ class DBImpl : public DB {
  // helper functions for adding and removing from flush & compaction queues
  void AddToCompactionQueue(ColumnFamilyData* cfd);
  ColumnFamilyData* PopFirstFromCompactionQueue();
-  void AddToFlushQueue(ColumnFamilyData* cfd);
+  void AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason);
  ColumnFamilyData* PopFirstFromFlushQueue();

  // helper function to call after some of the logs_ were synced
@@ -915,6 +933,8 @@ class DBImpl : public DB {

  uint64_t GetMaxTotalWalSize() const;

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

@@ -943,6 +963,10 @@ class DBImpl : public DB {
  // (i.e. whenever a flush is done, even if it didn't make any progress)
  // * whenever there is an error in background purge, flush or compaction
  // * whenever num_running_ingest_file_ goes to 0.
+  // * whenever pending_purge_obsolete_files_ goes to 0.
+  // * whenever disable_delete_obsolete_files_ goes to 0.
+  // * whenever SetOptions successfully updates options.
+  // * whenever a column family is dropped.
  InstrumentedCondVar bg_cv_;
  // Writes are protected by locking both mutex_ and log_write_mutex_, and reads
  // must be under either mutex_ or log_write_mutex_. Since after ::Open,
@@ -950,7 +974,7 @@ class DBImpl : public DB {
  // from the same write_thread_ without any locks.
  uint64_t logfile_number_;
  std::deque<uint64_t>
-      log_recycle_files;  // a list of log files that we can recycle
+      log_recycle_files_;  // a list of log files that we can recycle
  bool log_dir_synced_;
  // Without two_write_queues, read and writes to log_empty_ are protected by
  // mutex_. Since it is currently updated/read only in write_thread_, it can be
@@ -980,9 +1004,11 @@ class DBImpl : public DB {
      writer = nullptr;
      return w;
    }
-    void ClearWriter() {
+    Status ClearWriter() {
+      Status s = writer->WriteBuffer();
      delete writer;
      writer = nullptr;
+      return s;
    }

    uint64_t number;
@@ -1209,6 +1235,10 @@ class DBImpl : public DB {
  // without any synchronization
  int disable_delete_obsolete_files_;

+  // Number of times FindObsoleteFiles has found deletable files and the
+  // corresponding call to PurgeObsoleteFiles has not yet finished.
+  int pending_purge_obsolete_files_;
+
  // last time when DeleteObsoleteFiles with full scan was executed. Originaly
  // initialized with startup time.
  uint64_t delete_obsolete_files_last_run_;
@@ -1337,7 +1367,7 @@ class DBImpl : public DB {
    return Env::WLTH_SHORT;
  }

-  // When set, we use a seprate queue for writes that dont write to memtable. In
+  // When set, we use a separate queue for writes that dont write to memtable. In
  // 2PC these are the writes at Prepare phase.
  const bool two_write_queues_;
  const bool manual_wal_flush_;
@@ -1362,6 +1392,9 @@ class DBImpl : public DB {
  // is set to false.
  std::atomic<SequenceNumber> preserve_deletes_seqnum_;
  const bool preserve_deletes_;
+
+  // Flag to check whether Close() has been called on this DB
+  bool closed_;
 };

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

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

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

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

  // delete unnecessary files if any, this is done outside the mutex
-  if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
+  if (job_context.HaveSomethingToClean() ||
+      job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
    // Have to flush the info logs before bg_compaction_scheduled_--
    // because if bg_flush_scheduled_ becomes 0 and the lock is
    // released, the deconstructor of DB can kick in and destroy all the
@@ -809,7 +878,14 @@ int DBImpl::Level0StopWriteTrigger(ColumnFamilyHandle* column_family) {
 Status DBImpl::Flush(const FlushOptions& flush_options,
                     ColumnFamilyHandle* column_family) {
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-  return FlushMemTable(cfh->cfd(), flush_options);
+  ROCKS_LOG_INFO(immutable_db_options_.info_log, "[%s] Manual flush start.",
+                 cfh->GetName().c_str());
+  Status s =
+      FlushMemTable(cfh->cfd(), flush_options, FlushReason::kManualCompaction);
+  ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                 "[%s] Manual flush finished, status: %s\n",
+                 cfh->GetName().c_str(), s.ToString().c_str());
+  return s;
 }

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

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

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

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

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

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

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

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

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

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

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

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

@@ -1679,7 +1767,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    env_->Schedule(&DBImpl::BGWorkBottomCompaction, ca, Env::Priority::BOTTOM,
                   this, &DBImpl::UnscheduleCallback);
  } else {
-    int output_level  __attribute__((unused));
+    int output_level  __attribute__((__unused__));
    output_level = c->output_level();
    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:NonTrivial",
                             &output_level);
@@ -1913,7 +2001,7 @@ void DBImpl::InstallSuperVersionAndScheduleWork(

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

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

 Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
@@ -238,11 +238,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
    while (alive_log_files_.begin()->number < min_log_number) {
      auto& earliest = *alive_log_files_.begin();
      if (immutable_db_options_.recycle_log_file_num >
-          log_recycle_files.size()) {
+          log_recycle_files_.size()) {
        ROCKS_LOG_INFO(immutable_db_options_.info_log,
                       "adding log %" PRIu64 " to recycle list\n",
                       earliest.number);
-        log_recycle_files.push_back(earliest.number);
+        log_recycle_files_.push_back(earliest.number);
      } else {
        job_context->log_delete_files.push_back(earliest.number);
      }
@@ -283,8 +283,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  // We're just cleaning up for DB::Write().
  assert(job_context->logs_to_free.empty());
  job_context->logs_to_free = logs_to_free_;
-  job_context->log_recycle_files.assign(log_recycle_files.begin(),
-                                        log_recycle_files.end());
+  job_context->log_recycle_files.assign(log_recycle_files_.begin(),
+                                        log_recycle_files_.end());
+  if (job_context->HaveSomethingToDelete()) {
+    ++pending_purge_obsolete_files_;
+  }
  logs_to_free_.clear();
 }

@@ -342,15 +345,12 @@ void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
 // files in sst_delete_files and log_delete_files.
 // It is not necessary to hold the mutex when invoking this method.
 void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
+  TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:Begin");
  // we'd better have sth to delete
  assert(state.HaveSomethingToDelete());

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

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

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

  Status s = versions_->Recover(column_families, read_only);
@@ -64,7 +64,9 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
                         WriteBatch* my_batch, WriteCallback* callback,
                         uint64_t* log_used, uint64_t log_ref,
                         bool disable_memtable, uint64_t* seq_used,
+                         size_t batch_cnt,
                         PreReleaseCallback* pre_release_callback) {
+  assert(!seq_per_batch_ || batch_cnt != 0);
  if (my_batch == nullptr) {
    return Status::Corruption("Batch is nullptr!");
  }
@@ -76,6 +78,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
        "pipelined_writes is not compatible with concurrent prepares");
  }
  if (seq_per_batch_ && immutable_db_options_.enable_pipelined_write) {
+    // TODO(yiwu): update pipeline write with seq_per_batch and batch_cnt
    return Status::NotSupported(
        "pipelined_writes is not compatible with seq_per_batch");
  }
@@ -93,7 +96,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

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

  if (immutable_db_options_.enable_pipelined_write) {
@@ -103,7 +106,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

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

  if (!write_options.disableWAL) {
    RecordTick(stats_, WRITE_WITH_WAL);
@@ -122,7 +125,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
      w.status = WriteBatchInternal::InsertInto(
          &w, w.sequence, &column_family_memtables, &flush_scheduler_,
          write_options.ignore_missing_column_families, 0 /*log_number*/, this,
-          true /*concurrent_memtable_writes*/, seq_per_batch_);
+          true /*concurrent_memtable_writes*/, seq_per_batch_, w.batch_cnt);
    }

    if (write_thread_.CompleteParallelMemTableWriter(&w)) {
@@ -214,7 +217,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    uint64_t total_byte_size = 0;
    for (auto* writer : write_group) {
      if (writer->CheckCallback(this)) {
-        valid_batches++;
+        valid_batches += writer->batch_cnt;
        if (writer->ShouldWriteToMemtable()) {
          total_count += WriteBatchInternal::Count(writer->batch);
          parallel = parallel && !writer->batch->HasMerge();
@@ -303,7 +306,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
          }
          writer->sequence = next_sequence;
          if (seq_per_batch_) {
-            next_sequence++;
+            assert(writer->batch_cnt);
+            next_sequence += writer->batch_cnt;
          } else if (writer->ShouldWriteToMemtable()) {
            next_sequence += WriteBatchInternal::Count(writer->batch);
          }
@@ -323,7 +327,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
          w.status = WriteBatchInternal::InsertInto(
              &w, w.sequence, &column_family_memtables, &flush_scheduler_,
              write_options.ignore_missing_column_families, 0 /*log_number*/,
-              this, true /*concurrent_memtable_writes*/, seq_per_batch_);
+              this, true /*concurrent_memtable_writes*/, seq_per_batch_,
+              w.batch_cnt);
        }
      }
      if (seq_used != nullptr) {
@@ -515,15 +520,13 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
 Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
                                WriteBatch* my_batch, WriteCallback* callback,
                                uint64_t* log_used, uint64_t log_ref,
-                                uint64_t* seq_used,
+                                uint64_t* seq_used, size_t batch_cnt,
                                PreReleaseCallback* pre_release_callback) {
  Status status;
  PERF_TIMER_GUARD(write_pre_and_post_process_time);
  WriteThread::Writer w(write_options, my_batch, callback, log_ref,
-                        true /* disable_memtable */, pre_release_callback);
-  if (write_options.disableWAL) {
-    return status;
-  }
+                        true /* disable_memtable */, batch_cnt,
+                        pre_release_callback);
  RecordTick(stats_, WRITE_WITH_WAL);
  StopWatch write_sw(env_, immutable_db_options_.statistics.get(), DB_WRITE);

@@ -579,8 +582,22 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
  PERF_TIMER_GUARD(write_wal_time);
  // LastAllocatedSequence is increased inside WriteToWAL under
  // wal_write_mutex_ to ensure ordered events in WAL
-  size_t seq_inc = seq_per_batch_ ? write_group.size : 0 /*total_count*/;
-  status = ConcurrentWriteToWAL(write_group, log_used, &last_sequence, seq_inc);
+  size_t seq_inc = 0 /* total_count */;
+  if (seq_per_batch_) {
+    size_t total_batch_cnt = 0;
+    for (auto* writer : write_group) {
+      assert(writer->batch_cnt);
+      total_batch_cnt += writer->batch_cnt;
+    }
+    seq_inc = total_batch_cnt;
+  }
+  if (!write_options.disableWAL) {
+    status =
+        ConcurrentWriteToWAL(write_group, log_used, &last_sequence, seq_inc);
+  } else {
+    // Otherwise we inc seq number to do solely the seq allocation
+    last_sequence = versions_->FetchAddLastAllocatedSequence(seq_inc);
+  }
  auto curr_seq = last_sequence + 1;
  for (auto* writer : write_group) {
    if (writer->CallbackFailed()) {
@@ -588,11 +605,13 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
    }
    writer->sequence = curr_seq;
    if (seq_per_batch_) {
-      curr_seq++;
+      assert(writer->batch_cnt);
+      curr_seq += writer->batch_cnt;
    }
    // else seq advances only by memtable writes
  }
  if (status.ok() && write_options.sync) {
+    assert(!write_options.disableWAL);
    // Requesting sync with two_write_queues_ is expected to be very rare. We
    // hance provide a simple implementation that is not necessarily efficient.
    if (manual_wal_flush_) {
@@ -693,7 +712,7 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
  }

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

  if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
@@ -1000,7 +1019,7 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
        break;
      }
      cfd->imm()->FlushRequested();
-      SchedulePendingFlush(cfd);
+      SchedulePendingFlush(cfd, FlushReason::kWriteBufferManager);
    }
  }
  MaybeScheduleFlushOrCompaction();
@@ -1046,7 +1065,7 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
    status = SwitchMemtable(cfd_picked, write_context);
    if (status.ok()) {
      cfd_picked->imm()->FlushRequested();
-      SchedulePendingFlush(cfd_picked);
+      SchedulePendingFlush(cfd_picked, FlushReason::kWriteBufferFull);
      MaybeScheduleFlushOrCompaction();
    }
  }
@@ -1211,9 +1230,9 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  }
  uint64_t recycle_log_number = 0;
  if (creating_new_log && immutable_db_options_.recycle_log_file_num &&
-      !log_recycle_files.empty()) {
-    recycle_log_number = log_recycle_files.front();
-    log_recycle_files.pop_front();
+      !log_recycle_files_.empty()) {
+    recycle_log_number = log_recycle_files_.front();
+    log_recycle_files_.pop_front();
  }
  uint64_t new_log_number =
      creating_new_log ? versions_->NewFileNumber() : logfile_number_;
@@ -1288,6 +1307,29 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
                 ". Immutable memtables: %d.\n",
                 cfd->GetName().c_str(), new_log_number, num_imm_unflushed);
  mutex_.Lock();
+  if (s.ok() && creating_new_log) {
+    log_write_mutex_.Lock();
+    logfile_number_ = new_log_number;
+    assert(new_log != nullptr);
+    log_empty_ = true;
+    log_dir_synced_ = false;
+    if (!logs_.empty()) {
+      // Alway flush the buffer of the last log before switching to a new one
+      log::Writer* cur_log_writer = logs_.back().writer;
+      s = cur_log_writer->WriteBuffer();
+      if (!s.ok()) {
+        ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                       "[%s] Failed to switch from #%" PRIu64 " to #%" PRIu64
+                       "  WAL file -- %s\n",
+                       cfd->GetName().c_str(), cur_log_writer->get_log_number(),
+                       new_log_number);
+      }
+    }
+    logs_.emplace_back(logfile_number_, new_log);
+    alive_log_files_.push_back(LogFileNumberSize(logfile_number_));
+    log_write_mutex_.Unlock();
+  }
+
  if (!s.ok()) {
    // how do we fail if we're not creating new log?
    assert(creating_new_log);
@@ -1298,21 +1340,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
    }
    return s;
  }
-  if (creating_new_log) {
-    log_write_mutex_.Lock();
-    logfile_number_ = new_log_number;
-    assert(new_log != nullptr);
-    log_empty_ = true;
-    log_dir_synced_ = false;
-    if (!logs_.empty()) {
-      // Alway flush the buffer of the last log before switching to a new one
-      log::Writer* cur_log_writer = logs_.back().writer;
-      cur_log_writer->WriteBuffer();
-    }
-    logs_.emplace_back(logfile_number_, new_log);
-    alive_log_files_.push_back(LogFileNumberSize(logfile_number_));
-    log_write_mutex_.Unlock();
-  }
+
  for (auto loop_cfd : *versions_->GetColumnFamilySet()) {
    // all this is just optimization to delete logs that
    // are no longer needed -- if CF is empty, that means it
@@ -1367,7 +1395,10 @@ Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,
  // 8 bytes are taken by header, 4 bytes for count, 1 byte for type,
  // and we allocate 11 extra bytes for key length, as well as value length.
  WriteBatch batch(key.size() + value.size() + 24);
-  batch.Put(column_family, key, value);
+  Status s = batch.Put(column_family, key, value);
+  if (!s.ok()) {
+    return s;
+  }
  return Write(opt, &batch);
 }

@@ -1396,7 +1427,10 @@ Status DB::DeleteRange(const WriteOptions& opt,
 Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
                 const Slice& key, const Slice& value) {
  WriteBatch batch;
-  batch.Merge(column_family, key, value);
+  Status s = batch.Merge(column_family, key, value);
+  if (!s.ok()) {
+    return s;
+  }
  return Write(opt, &batch);
 }
 }  // namespace rocksdb
@@ -217,6 +217,9 @@ class DBIter final: public Iterator {
        *prop = "Iterator is not valid.";
      }
      return Status::OK();
+    } else if (prop_name == "rocksdb.iterator.internal-key") {
+      *prop = saved_key_.GetUserKey().ToString();
+      return Status::OK();
    }
    return Status::InvalidArgument("Undentified property.");
  }
@@ -611,10 +614,12 @@ void DBIter::MergeValuesNewToOld() {
  // Start the merge process by pushing the first operand
  merge_context_.PushOperand(iter_->value(),
                             iter_->IsValuePinned() /* operand_pinned */);
+  TEST_SYNC_POINT("DBIter::MergeValuesNewToOld:PushedFirstOperand");

  ParsedInternalKey ikey;
  Status s;
  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
+    TEST_SYNC_POINT("DBIter::MergeValuesNewToOld:SteppedToNextOperand");
    if (!ParseKey(&ikey)) {
      // skip corrupted key
      continue;
@@ -972,6 +977,17 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  // assume there is at least one parseable key for this user key
  ParsedInternalKey ikey;
  FindParseableKey(&ikey, kForward);
+  assert(iter_->Valid());
+  assert(user_comparator_->Equal(ikey.user_key, saved_key_.GetUserKey()));
+
+  // In case read_callback presents, the value we seek to may not be visible.
+  // Seek for the next value that's visible.
+  while (!IsVisible(ikey.sequence)) {
+    iter_->Next();
+    FindParseableKey(&ikey, kForward);
+    assert(iter_->Valid());
+    assert(user_comparator_->Equal(ikey.user_key, saved_key_.GetUserKey()));
+  }

  if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
      range_del_agg_.ShouldDelete(
@@ -1378,17 +1394,19 @@ void ArenaWrappedDBIter::Init(Env* env, const ReadOptions& read_options,
                              const SequenceNumber& sequence,
                              uint64_t max_sequential_skip_in_iteration,
                              uint64_t version_number,
-                              ReadCallback* read_callback, bool allow_blob) {
+                              ReadCallback* read_callback, bool allow_blob,
+                              bool allow_refresh) {
  auto mem = arena_.AllocateAligned(sizeof(DBIter));
  db_iter_ = new (mem)
      DBIter(env, read_options, cf_options, cf_options.user_comparator, nullptr,
             sequence, true, max_sequential_skip_in_iteration, read_callback,
             allow_blob);
  sv_number_ = version_number;
+  allow_refresh_ = allow_refresh;
 }

 Status ArenaWrappedDBIter::Refresh() {
-  if (cfd_ == nullptr || db_impl_ == nullptr) {
+  if (cfd_ == nullptr || db_impl_ == nullptr || !allow_refresh_) {
    return Status::NotSupported("Creating renew iterator is not allowed.");
  }
  assert(db_iter_ != nullptr);
@@ -1406,7 +1424,7 @@ Status ArenaWrappedDBIter::Refresh() {
    SuperVersion* sv = cfd_->GetReferencedSuperVersion(db_impl_->mutex());
    Init(env, read_options_, *(cfd_->ioptions()), latest_seq,
         sv->mutable_cf_options.max_sequential_skip_in_iterations,
-         cur_sv_number, read_callback_, allow_blob_);
+         cur_sv_number, read_callback_, allow_blob_, allow_refresh_);

    InternalIterator* internal_iter = db_impl_->NewInternalIterator(
        read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator());
@@ -1423,12 +1441,12 @@ ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    const ImmutableCFOptions& cf_options, const SequenceNumber& sequence,
    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
    ReadCallback* read_callback, DBImpl* db_impl, ColumnFamilyData* cfd,
-    bool allow_blob) {
+    bool allow_blob, bool allow_refresh) {
  ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
  iter->Init(env, read_options, cf_options, sequence,
             max_sequential_skip_in_iterations, version_number, read_callback,
-             allow_blob);
-  if (db_impl != nullptr && cfd != nullptr) {
+             allow_blob, allow_refresh);
+  if (db_impl != nullptr && cfd != nullptr && allow_refresh) {
    iter->StoreRefreshInfo(read_options, db_impl, cfd, read_callback,
                           allow_blob);
  }
@@ -73,7 +73,7 @@ class ArenaWrappedDBIter : public Iterator {
            const ImmutableCFOptions& cf_options,
            const SequenceNumber& sequence,
            uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
-            ReadCallback* read_callback, bool allow_blob);
+            ReadCallback* read_callback, bool allow_blob, bool allow_refresh);

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

 // Generate the arena wrapped iterator class.
@@ -104,6 +105,6 @@ extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    const ImmutableCFOptions& cf_options, const SequenceNumber& sequence,
    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
    ReadCallback* read_callback, DBImpl* db_impl = nullptr,
-    ColumnFamilyData* cfd = nullptr, bool allow_blob = false);
-
+    ColumnFamilyData* cfd = nullptr, bool allow_blob = false,
+    bool allow_refresh = true);
 }  // namespace rocksdb
@@ -9,6 +9,7 @@

 #include <functional>

+#include "db/db_iter.h"
 #include "db/db_test_util.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
@@ -17,9 +18,34 @@

 namespace rocksdb {

-class DBIteratorTest : public DBTestBase {
+// A dumb ReadCallback which saying every key is committed.
+class DummyReadCallback : public ReadCallback {
+  bool IsCommitted(SequenceNumber /*seq*/) { return true; }
+};
+
+// Test param:
+//   bool: whether to pass read_callback to NewIterator().
+class DBIteratorTest : public DBTestBase,
+                       public testing::WithParamInterface<bool> {
 public:
  DBIteratorTest() : DBTestBase("/db_iterator_test") {}
+
+  Iterator* NewIterator(const ReadOptions& read_options,
+                        ColumnFamilyHandle* column_family = nullptr) {
+    if (column_family == nullptr) {
+      column_family = db_->DefaultColumnFamily();
+    }
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family)->cfd();
+    SequenceNumber seq = read_options.snapshot != nullptr
+                             ? read_options.snapshot->GetSequenceNumber()
+                             : db_->GetLatestSequenceNumber();
+    bool use_read_callback = GetParam();
+    ReadCallback* read_callback = use_read_callback ? &read_callback_ : nullptr;
+    return dbfull()->NewIteratorImpl(read_options, cfd, seq, read_callback);
+  }
+
+ private:
+  DummyReadCallback read_callback_;
 };

 class FlushBlockEveryKeyPolicy : public FlushBlockPolicy {
@@ -50,28 +76,35 @@ class FlushBlockEveryKeyPolicyFactory : public FlushBlockPolicyFactory {
  }
 };

-TEST_F(DBIteratorTest, IteratorProperty) {
+TEST_P(DBIteratorTest, IteratorProperty) {
  // The test needs to be changed if kPersistedTier is supported in iterator.
  Options options = CurrentOptions();
  CreateAndReopenWithCF({"pikachu"}, options);
  Put(1, "1", "2");
+  Delete(1, "2");
  ReadOptions ropt;
  ropt.pin_data = false;
  {
-    unique_ptr<Iterator> iter(db_->NewIterator(ropt, handles_[1]));
+    unique_ptr<Iterator> iter(NewIterator(ropt, handles_[1]));
    iter->SeekToFirst();
    std::string prop_value;
    ASSERT_NOK(iter->GetProperty("non_existing.value", &prop_value));
    ASSERT_OK(iter->GetProperty("rocksdb.iterator.is-key-pinned", &prop_value));
    ASSERT_EQ("0", prop_value);
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.internal-key", &prop_value));
+    ASSERT_EQ("1", prop_value);
    iter->Next();
    ASSERT_OK(iter->GetProperty("rocksdb.iterator.is-key-pinned", &prop_value));
    ASSERT_EQ("Iterator is not valid.", prop_value);
+
+    // Get internal key at which the iteration stopped (tombstone in this case).
+    ASSERT_OK(iter->GetProperty("rocksdb.iterator.internal-key", &prop_value));
+    ASSERT_EQ("2", prop_value);
  }
  Close();
 }

-TEST_F(DBIteratorTest, PersistedTierOnIterator) {
+TEST_P(DBIteratorTest, PersistedTierOnIterator) {
  // The test needs to be changed if kPersistedTier is supported in iterator.
  Options options = CurrentOptions();
  CreateAndReopenWithCF({"pikachu"}, options);
@@ -87,7 +120,7 @@ TEST_F(DBIteratorTest, PersistedTierOnIterator) {
  Close();
 }

-TEST_F(DBIteratorTest, NonBlockingIteration) {
+TEST_P(DBIteratorTest, NonBlockingIteration) {
  do {
    ReadOptions non_blocking_opts, regular_opts;
    Options options = CurrentOptions();
@@ -99,7 +132,7 @@ TEST_F(DBIteratorTest, NonBlockingIteration) {

    // scan using non-blocking iterator. We should find it because
    // it is in memtable.
-    Iterator* iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    Iterator* iter = NewIterator(non_blocking_opts, handles_[1]);
    int count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      ASSERT_OK(iter->status());
@@ -116,7 +149,7 @@ TEST_F(DBIteratorTest, NonBlockingIteration) {
    // kvs. Neither does it do any IOs to storage.
    uint64_t numopen = TestGetTickerCount(options, NO_FILE_OPENS);
    uint64_t cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    iter = NewIterator(non_blocking_opts, handles_[1]);
    count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      count++;
@@ -133,7 +166,7 @@ TEST_F(DBIteratorTest, NonBlockingIteration) {
    // verify that we can find it via a non-blocking scan
    numopen = TestGetTickerCount(options, NO_FILE_OPENS);
    cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    iter = NewIterator(non_blocking_opts, handles_[1]);
    count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      ASSERT_OK(iter->status());
@@ -152,7 +185,7 @@ TEST_F(DBIteratorTest, NonBlockingIteration) {
 }

 #ifndef ROCKSDB_LITE
-TEST_F(DBIteratorTest, ManagedNonBlockingIteration) {
+TEST_P(DBIteratorTest, ManagedNonBlockingIteration) {
  do {
    ReadOptions non_blocking_opts, regular_opts;
    Options options = CurrentOptions();
@@ -165,7 +198,7 @@ TEST_F(DBIteratorTest, ManagedNonBlockingIteration) {

    // scan using non-blocking iterator. We should find it because
    // it is in memtable.
-    Iterator* iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    Iterator* iter = NewIterator(non_blocking_opts, handles_[1]);
    int count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      ASSERT_OK(iter->status());
@@ -182,7 +215,7 @@ TEST_F(DBIteratorTest, ManagedNonBlockingIteration) {
    // kvs. Neither does it do any IOs to storage.
    int64_t numopen = TestGetTickerCount(options, NO_FILE_OPENS);
    int64_t cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    iter = NewIterator(non_blocking_opts, handles_[1]);
    count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      count++;
@@ -199,7 +232,7 @@ TEST_F(DBIteratorTest, ManagedNonBlockingIteration) {
    // verify that we can find it via a non-blocking scan
    numopen = TestGetTickerCount(options, NO_FILE_OPENS);
    cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = db_->NewIterator(non_blocking_opts, handles_[1]);
+    iter = NewIterator(non_blocking_opts, handles_[1]);
    count = 0;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      ASSERT_OK(iter->status());
@@ -218,7 +251,7 @@ TEST_F(DBIteratorTest, ManagedNonBlockingIteration) {
 }
 #endif  // ROCKSDB_LITE

-TEST_F(DBIteratorTest, IterSeekBeforePrev) {
+TEST_P(DBIteratorTest, IterSeekBeforePrev) {
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
  dbfull()->Flush(FlushOptions());
@@ -226,7 +259,7 @@ TEST_F(DBIteratorTest, IterSeekBeforePrev) {
  ASSERT_OK(Put("1", "h"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("2", "j"));
-  auto iter = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());
  iter->Seek(Slice("c"));
  iter->Prev();
  iter->Seek(Slice("a"));
@@ -234,7 +267,7 @@ TEST_F(DBIteratorTest, IterSeekBeforePrev) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterSeekForPrevBeforeNext) {
+TEST_P(DBIteratorTest, IterSeekForPrevBeforeNext) {
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
  dbfull()->Flush(FlushOptions());
@@ -242,7 +275,7 @@ TEST_F(DBIteratorTest, IterSeekForPrevBeforeNext) {
  ASSERT_OK(Put("1", "h"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("2", "j"));
-  auto iter = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());
  iter->SeekForPrev(Slice("0"));
  iter->Next();
  iter->SeekForPrev(Slice("1"));
@@ -256,7 +289,7 @@ std::string MakeLongKey(size_t length, char c) {
 }
 }  // namespace

-TEST_F(DBIteratorTest, IterLongKeys) {
+TEST_P(DBIteratorTest, IterLongKeys) {
  ASSERT_OK(Put(MakeLongKey(20, 0), "0"));
  ASSERT_OK(Put(MakeLongKey(32, 2), "2"));
  ASSERT_OK(Put("a", "b"));
@@ -264,7 +297,7 @@ TEST_F(DBIteratorTest, IterLongKeys) {
  ASSERT_OK(Put(MakeLongKey(50, 1), "1"));
  ASSERT_OK(Put(MakeLongKey(127, 3), "3"));
  ASSERT_OK(Put(MakeLongKey(64, 4), "4"));
-  auto iter = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());

  // Create a key that needs to be skipped for Seq too new
  iter->Seek(MakeLongKey(20, 0));
@@ -286,7 +319,7 @@ TEST_F(DBIteratorTest, IterLongKeys) {
  ASSERT_EQ(IterStatus(iter), MakeLongKey(50, 1) + "->1");
  delete iter;

-  iter = db_->NewIterator(ReadOptions());
+  iter = NewIterator(ReadOptions());
  iter->Seek(MakeLongKey(50, 1));
  ASSERT_EQ(IterStatus(iter), MakeLongKey(50, 1) + "->1");
  iter->Next();
@@ -296,13 +329,13 @@ TEST_F(DBIteratorTest, IterLongKeys) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterNextWithNewerSeq) {
+TEST_P(DBIteratorTest, IterNextWithNewerSeq) {
  ASSERT_OK(Put("0", "0"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
  ASSERT_OK(Put("d", "e"));
-  auto iter = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());

  // Create a key that needs to be skipped for Seq too new
  for (uint64_t i = 0; i < last_options_.max_sequential_skip_in_iterations + 1;
@@ -322,13 +355,13 @@ TEST_F(DBIteratorTest, IterNextWithNewerSeq) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterPrevWithNewerSeq) {
+TEST_P(DBIteratorTest, IterPrevWithNewerSeq) {
  ASSERT_OK(Put("0", "0"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
  ASSERT_OK(Put("d", "e"));
-  auto iter = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());

  // Create a key that needs to be skipped for Seq too new
  for (uint64_t i = 0; i < last_options_.max_sequential_skip_in_iterations + 1;
@@ -353,14 +386,14 @@ TEST_F(DBIteratorTest, IterPrevWithNewerSeq) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterPrevWithNewerSeq2) {
+TEST_P(DBIteratorTest, IterPrevWithNewerSeq2) {
  ASSERT_OK(Put("0", "0"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
  ASSERT_OK(Put("e", "f"));
-  auto iter = db_->NewIterator(ReadOptions());
-  auto iter2 = db_->NewIterator(ReadOptions());
+  auto iter = NewIterator(ReadOptions());
+  auto iter2 = NewIterator(ReadOptions());
  iter->Seek(Slice("c"));
  iter2->SeekForPrev(Slice("d"));
  ASSERT_EQ(IterStatus(iter), "c->d");
@@ -382,10 +415,10 @@ TEST_F(DBIteratorTest, IterPrevWithNewerSeq2) {
  delete iter2;
 }

-TEST_F(DBIteratorTest, IterEmpty) {
+TEST_P(DBIteratorTest, IterEmpty) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);

    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "(invalid)");
@@ -403,11 +436,11 @@ TEST_F(DBIteratorTest, IterEmpty) {
  } while (ChangeCompactOptions());
 }

-TEST_F(DBIteratorTest, IterSingle) {
+TEST_P(DBIteratorTest, IterSingle) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "a", "va"));
-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);

    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "a->va");
@@ -454,13 +487,13 @@ TEST_F(DBIteratorTest, IterSingle) {
  } while (ChangeCompactOptions());
 }

-TEST_F(DBIteratorTest, IterMulti) {
+TEST_P(DBIteratorTest, IterMulti) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "a", "va"));
    ASSERT_OK(Put(1, "b", "vb"));
    ASSERT_OK(Put(1, "c", "vc"));
-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);

    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "a->va");
@@ -553,7 +586,7 @@ TEST_F(DBIteratorTest, IterMulti) {

 // Check that we can skip over a run of user keys
 // by using reseek rather than sequential scan
-TEST_F(DBIteratorTest, IterReseek) {
+TEST_P(DBIteratorTest, IterReseek) {
  anon::OptionsOverride options_override;
  options_override.skip_policy = kSkipNoSnapshot;
  Options options = CurrentOptions(options_override);
@@ -570,7 +603,7 @@ TEST_F(DBIteratorTest, IterReseek) {
  ASSERT_OK(Put(1, "a", "one"));
  ASSERT_OK(Put(1, "a", "two"));
  ASSERT_OK(Put(1, "b", "bone"));
-  Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+  Iterator* iter = NewIterator(ReadOptions(), handles_[1]);
  iter->SeekToFirst();
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_OF_RESEEKS_IN_ITERATION), 0);
  ASSERT_EQ(IterStatus(iter), "a->two");
@@ -582,7 +615,7 @@ TEST_F(DBIteratorTest, IterReseek) {
  // insert a total of three keys with same userkey and verify
  // that reseek is still not invoked.
  ASSERT_OK(Put(1, "a", "three"));
-  iter = db_->NewIterator(ReadOptions(), handles_[1]);
+  iter = NewIterator(ReadOptions(), handles_[1]);
  iter->SeekToFirst();
  ASSERT_EQ(IterStatus(iter), "a->three");
  iter->Next();
@@ -593,7 +626,7 @@ TEST_F(DBIteratorTest, IterReseek) {
  // insert a total of four keys with same userkey and verify
  // that reseek is invoked.
  ASSERT_OK(Put(1, "a", "four"));
-  iter = db_->NewIterator(ReadOptions(), handles_[1]);
+  iter = NewIterator(ReadOptions(), handles_[1]);
  iter->SeekToFirst();
  ASSERT_EQ(IterStatus(iter), "a->four");
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_OF_RESEEKS_IN_ITERATION), 0);
@@ -610,7 +643,7 @@ TEST_F(DBIteratorTest, IterReseek) {

  // Insert another version of b and assert that reseek is not invoked
  ASSERT_OK(Put(1, "b", "btwo"));
-  iter = db_->NewIterator(ReadOptions(), handles_[1]);
+  iter = NewIterator(ReadOptions(), handles_[1]);
  iter->SeekToLast();
  ASSERT_EQ(IterStatus(iter), "b->btwo");
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_OF_RESEEKS_IN_ITERATION),
@@ -625,7 +658,7 @@ TEST_F(DBIteratorTest, IterReseek) {
  // of b and 4 versions of a.
  ASSERT_OK(Put(1, "b", "bthree"));
  ASSERT_OK(Put(1, "b", "bfour"));
-  iter = db_->NewIterator(ReadOptions(), handles_[1]);
+  iter = NewIterator(ReadOptions(), handles_[1]);
  iter->SeekToLast();
  ASSERT_EQ(IterStatus(iter), "b->bfour");
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_OF_RESEEKS_IN_ITERATION),
@@ -639,7 +672,7 @@ TEST_F(DBIteratorTest, IterReseek) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterSmallAndLargeMix) {
+TEST_P(DBIteratorTest, IterSmallAndLargeMix) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "a", "va"));
@@ -648,7 +681,7 @@ TEST_F(DBIteratorTest, IterSmallAndLargeMix) {
    ASSERT_OK(Put(1, "d", std::string(100000, 'd')));
    ASSERT_OK(Put(1, "e", std::string(100000, 'e')));

-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);

    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "a->va");
@@ -680,7 +713,7 @@ TEST_F(DBIteratorTest, IterSmallAndLargeMix) {
  } while (ChangeCompactOptions());
 }

-TEST_F(DBIteratorTest, IterMultiWithDelete) {
+TEST_P(DBIteratorTest, IterMultiWithDelete) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "ka", "va"));
@@ -689,7 +722,7 @@ TEST_F(DBIteratorTest, IterMultiWithDelete) {
    ASSERT_OK(Delete(1, "kb"));
    ASSERT_EQ("NOT_FOUND", Get(1, "kb"));

-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);
    iter->Seek("kc");
    ASSERT_EQ(IterStatus(iter), "kc->vc");
    if (!CurrentOptions().merge_operator) {
@@ -706,7 +739,7 @@ TEST_F(DBIteratorTest, IterMultiWithDelete) {
  } while (ChangeOptions());
 }

-TEST_F(DBIteratorTest, IterPrevMaxSkip) {
+TEST_P(DBIteratorTest, IterPrevMaxSkip) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    for (int i = 0; i < 2; i++) {
@@ -736,7 +769,7 @@ TEST_F(DBIteratorTest, IterPrevMaxSkip) {
  } while (ChangeOptions(kSkipMergePut | kSkipNoSeekToLast));
 }

-TEST_F(DBIteratorTest, IterWithSnapshot) {
+TEST_P(DBIteratorTest, IterWithSnapshot) {
  anon::OptionsOverride options_override;
  options_override.skip_policy = kSkipNoSnapshot;
  do {
@@ -750,7 +783,7 @@ TEST_F(DBIteratorTest, IterWithSnapshot) {
    const Snapshot* snapshot = db_->GetSnapshot();
    ReadOptions options;
    options.snapshot = snapshot;
-    Iterator* iter = db_->NewIterator(options, handles_[1]);
+    Iterator* iter = NewIterator(options, handles_[1]);

    ASSERT_OK(Put(1, "key0", "val0"));
    // Put more values after the snapshot
@@ -803,13 +836,13 @@ TEST_F(DBIteratorTest, IterWithSnapshot) {
  } while (ChangeOptions(kSkipHashCuckoo));
 }

-TEST_F(DBIteratorTest, IteratorPinsRef) {
+TEST_P(DBIteratorTest, IteratorPinsRef) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    Put(1, "foo", "hello");

    // Get iterator that will yield the current contents of the DB.
-    Iterator* iter = db_->NewIterator(ReadOptions(), handles_[1]);
+    Iterator* iter = NewIterator(ReadOptions(), handles_[1]);

    // Write to force compactions
    Put(1, "foo", "newvalue1");
@@ -829,7 +862,7 @@ TEST_F(DBIteratorTest, IteratorPinsRef) {
  } while (ChangeCompactOptions());
 }

-TEST_F(DBIteratorTest, DBIteratorBoundTest) {
+TEST_P(DBIteratorTest, DBIteratorBoundTest) {
  Options options = CurrentOptions();
  options.env = env_;
  options.create_if_missing = true;
@@ -846,7 +879,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    ReadOptions ro;
    ro.iterate_upper_bound = nullptr;

-    std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+    std::unique_ptr<Iterator> iter(NewIterator(ro));

    iter->Seek("foo");

@@ -883,7 +916,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    Slice prefix("foo2");
    ro.iterate_upper_bound = &prefix;

-    std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+    std::unique_ptr<Iterator> iter(NewIterator(ro));

    iter->Seek("foo");

@@ -905,7 +938,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    Slice prefix("foo");
    ro.iterate_upper_bound = &prefix;

-    std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+    std::unique_ptr<Iterator> iter(NewIterator(ro));

    iter->SeekToLast();
    ASSERT_TRUE(iter->Valid());
@@ -929,7 +962,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    Slice upper_bound("g");
    ro.iterate_upper_bound = &upper_bound;

-    std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+    std::unique_ptr<Iterator> iter(NewIterator(ro));

    iter->Seek("foo");

@@ -962,7 +995,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    ReadOptions ro;
    ro.iterate_upper_bound = nullptr;

-    std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+    std::unique_ptr<Iterator> iter(NewIterator(ro));

    iter->Seek("b");
    ASSERT_TRUE(iter->Valid());
@@ -982,7 +1015,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    Slice prefix("c");
    ro.iterate_upper_bound = &prefix;

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

    get_perf_context()->Reset();

@@ -1003,7 +1036,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
  }
 }

-TEST_F(DBIteratorTest, DBIteratorBoundOptimizationTest) {
+TEST_P(DBIteratorTest, DBIteratorBoundOptimizationTest) {
  int upper_bound_hits = 0;
  Options options = CurrentOptions();
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -1031,7 +1064,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundOptimizationTest) {
  ReadOptions ro;
  ro.iterate_upper_bound = &ub;

-  std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+  std::unique_ptr<Iterator> iter(NewIterator(ro));

  iter->Seek("foo");
  ASSERT_TRUE(iter->Valid());
@@ -1049,7 +1082,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundOptimizationTest) {
 }
 // TODO(3.13): fix the issue of Seek() + Prev() which might not necessary
 //             return the biggest key which is smaller than the seek key.
-TEST_F(DBIteratorTest, PrevAfterAndNextAfterMerge) {
+TEST_P(DBIteratorTest, PrevAfterAndNextAfterMerge) {
  Options options;
  options.create_if_missing = true;
  options.merge_operator = MergeOperators::CreatePutOperator();
@@ -1062,7 +1095,7 @@ TEST_F(DBIteratorTest, PrevAfterAndNextAfterMerge) {
  db_->Merge(wopts, "2", "data2");
  db_->Merge(wopts, "3", "data3");

-  std::unique_ptr<Iterator> it(db_->NewIterator(ReadOptions()));
+  std::unique_ptr<Iterator> it(NewIterator(ReadOptions()));

  it->Seek("2");
  ASSERT_TRUE(it->Valid());
@@ -1158,7 +1191,7 @@ class DBIteratorTestForPinnedData : public DBIteratorTest {

    ReadOptions ro;
    ro.pin_data = true;
-    auto iter = db_->NewIterator(ro);
+    auto iter = NewIterator(ro);

    {
      // Test Seek to random keys
@@ -1250,25 +1283,25 @@ class DBIteratorTestForPinnedData : public DBIteratorTest {
 }
 };

-TEST_F(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedNormal) {
+TEST_P(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedNormal) {
  PinnedDataIteratorRandomized(TestConfig::NORMAL);
 }

-TEST_F(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedCLoseAndOpen) {
+TEST_P(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedCLoseAndOpen) {
  PinnedDataIteratorRandomized(TestConfig::CLOSE_AND_OPEN);
 }

-TEST_F(DBIteratorTestForPinnedData,
+TEST_P(DBIteratorTestForPinnedData,
       PinnedDataIteratorRandomizedCompactBeforeRead) {
  PinnedDataIteratorRandomized(TestConfig::COMPACT_BEFORE_READ);
 }

-TEST_F(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedFlush) {
+TEST_P(DBIteratorTestForPinnedData, PinnedDataIteratorRandomizedFlush) {
  PinnedDataIteratorRandomized(TestConfig::FLUSH_EVERY_1000);
 }

 #ifndef ROCKSDB_LITE
-TEST_F(DBIteratorTest, PinnedDataIteratorMultipleFiles) {
+TEST_P(DBIteratorTest, PinnedDataIteratorMultipleFiles) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
  table_options.use_delta_encoding = false;
@@ -1317,7 +1350,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorMultipleFiles) {

  ReadOptions ro;
  ro.pin_data = true;
-  auto iter = db_->NewIterator(ro);
+  auto iter = NewIterator(ro);

  std::vector<std::pair<Slice, std::string>> results;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
@@ -1339,7 +1372,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorMultipleFiles) {
 }
 #endif

-TEST_F(DBIteratorTest, PinnedDataIteratorMergeOperator) {
+TEST_P(DBIteratorTest, PinnedDataIteratorMergeOperator) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
  table_options.use_delta_encoding = false;
@@ -1372,7 +1405,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorMergeOperator) {

  ReadOptions ro;
  ro.pin_data = true;
-  auto iter = db_->NewIterator(ro);
+  auto iter = NewIterator(ro);

  std::vector<std::pair<Slice, std::string>> results;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
@@ -1399,7 +1432,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorMergeOperator) {
  delete iter;
 }

-TEST_F(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
+TEST_P(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
  table_options.use_delta_encoding = false;
@@ -1419,7 +1452,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {

  ReadOptions ro;
  ro.pin_data = true;
-  auto iter = db_->NewIterator(ro);
+  auto iter = NewIterator(ro);

  // Delete 50% of the keys and update the other 50%
  for (auto& kv : true_data) {
@@ -1449,7 +1482,7 @@ TEST_F(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
  delete iter;
 }

-TEST_F(DBIteratorTest, IterSeekForPrevCrossingFiles) {
+TEST_P(DBIteratorTest, IterSeekForPrevCrossingFiles) {
  Options options = CurrentOptions();
  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
  options.disable_auto_compactions = true;
@@ -1478,7 +1511,7 @@ TEST_F(DBIteratorTest, IterSeekForPrevCrossingFiles) {
  MoveFilesToLevel(1);
  {
    ReadOptions ro;
-    Iterator* iter = db_->NewIterator(ro);
+    Iterator* iter = NewIterator(ro);

    iter->SeekForPrev("a4");
    ASSERT_EQ(iter->key().ToString(), "a3");
@@ -1496,14 +1529,14 @@ TEST_F(DBIteratorTest, IterSeekForPrevCrossingFiles) {
  {
    ReadOptions ro;
    ro.prefix_same_as_start = true;
-    Iterator* iter = db_->NewIterator(ro);
+    Iterator* iter = NewIterator(ro);
    iter->SeekForPrev("c2");
    ASSERT_TRUE(!iter->Valid());
    delete iter;
  }
 }

-TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocks) {
+TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocks) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
  table_options.block_size = 1;  // every block will contain one entry
@@ -1545,7 +1578,7 @@ TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocks) {
  {
    ReadOptions ro;
    ro.fill_cache = false;
-    Iterator* iter = db_->NewIterator(ro);
+    Iterator* iter = NewIterator(ro);

    iter->SeekToLast();
    ASSERT_EQ(iter->key().ToString(), "key5");
@@ -1571,7 +1604,7 @@ TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocks) {
  }
 }

-TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
+TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
  Options options = CurrentOptions();
  options.merge_operator = MergeOperators::CreateStringAppendTESTOperator();
  options.disable_auto_compactions = true;
@@ -1647,7 +1680,7 @@ TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
  {
    ReadOptions ro;
    ro.fill_cache = false;
-    Iterator* iter = db_->NewIterator(ro);
+    Iterator* iter = NewIterator(ro);
    auto data_iter = true_data.rbegin();

    for (iter->SeekToLast(); iter->Valid(); iter->Prev()) {
@@ -1663,7 +1696,7 @@ TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
  {
    ReadOptions ro;
    ro.fill_cache = false;
-    Iterator* iter = db_->NewIterator(ro);
+    Iterator* iter = NewIterator(ro);
    auto data_iter = true_data.rbegin();

    int entries_right = 0;
@@ -1718,7 +1751,7 @@ TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocksRandomized) {
  }
 }

-TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {
+TEST_P(DBIteratorTest, IteratorWithLocalStatistics) {
  Options options = CurrentOptions();
  options.statistics = rocksdb::CreateDBStatistics();
  DestroyAndReopen(options);
@@ -1739,7 +1772,7 @@ TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {
  std::function<void()> reader_func_next = [&]() {
    SetPerfLevel(kEnableCount);
    get_perf_context()->Reset();
-    Iterator* iter = db_->NewIterator(ReadOptions());
+    Iterator* iter = NewIterator(ReadOptions());

    iter->SeekToFirst();
    // Seek will bump ITER_BYTES_READ
@@ -1766,7 +1799,7 @@ TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {

  std::function<void()> reader_func_prev = [&]() {
    SetPerfLevel(kEnableCount);
-    Iterator* iter = db_->NewIterator(ReadOptions());
+    Iterator* iter = NewIterator(ReadOptions());

    iter->SeekToLast();
    // Seek will bump ITER_BYTES_READ
@@ -1812,7 +1845,7 @@ TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {

 }

-TEST_F(DBIteratorTest, ReadAhead) {
+TEST_P(DBIteratorTest, ReadAhead) {
  Options options;
  env_->count_random_reads_ = true;
  options.env = env_;
@@ -1849,7 +1882,7 @@ TEST_F(DBIteratorTest, ReadAhead) {
  env_->random_read_bytes_counter_ = 0;
  options.statistics->setTickerCount(NO_FILE_OPENS, 0);
  ReadOptions read_options;
-  auto* iter = db_->NewIterator(read_options);
+  auto* iter = NewIterator(read_options);
  iter->SeekToFirst();
  int64_t num_file_opens = TestGetTickerCount(options, NO_FILE_OPENS);
  size_t bytes_read = env_->random_read_bytes_counter_;
@@ -1858,7 +1891,7 @@ TEST_F(DBIteratorTest, ReadAhead) {
  env_->random_read_bytes_counter_ = 0;
  options.statistics->setTickerCount(NO_FILE_OPENS, 0);
  read_options.readahead_size = 1024 * 10;
-  iter = db_->NewIterator(read_options);
+  iter = NewIterator(read_options);
  iter->SeekToFirst();
  int64_t num_file_opens_readahead = TestGetTickerCount(options, NO_FILE_OPENS);
  size_t bytes_read_readahead = env_->random_read_bytes_counter_;
@@ -1868,7 +1901,7 @@ TEST_F(DBIteratorTest, ReadAhead) {
  ASSERT_GT(bytes_read_readahead, read_options.readahead_size * 3);

  // Verify correctness.
-  iter = db_->NewIterator(read_options);
+  iter = NewIterator(read_options);
  int count = 0;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    ASSERT_EQ(value, iter->value());
@@ -1885,7 +1918,7 @@ TEST_F(DBIteratorTest, ReadAhead) {
 // Insert a key, create a snapshot iterator, overwrite key lots of times,
 // seek to a smaller key. Expect DBIter to fall back to a seek instead of
 // going through all the overwrites linearly.
-TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {
+TEST_P(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {
  Options options = CurrentOptions();
  options.env = env_;
  options.create_if_missing = true;
@@ -1900,7 +1933,7 @@ TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {

  // Create iterator.
  ReadOptions ro;
-  std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+  std::unique_ptr<Iterator> iter(NewIterator(ro));

  // Insert a lot.
  for (int i = 0; i < 100; ++i) {
@@ -1938,10 +1971,10 @@ TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {
                 NUMBER_OF_RESEEKS_IN_ITERATION));
 }

-TEST_F(DBIteratorTest, Refresh) {
+TEST_P(DBIteratorTest, Refresh) {
  ASSERT_OK(Put("x", "y"));

-  std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+  std::unique_ptr<Iterator> iter(NewIterator(ReadOptions()));
  iter->Seek(Slice("a"));
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(iter->key().compare(Slice("x")), 0);
@@ -1997,20 +2030,20 @@ TEST_F(DBIteratorTest, Refresh) {
  iter.reset();
 }

-TEST_F(DBIteratorTest, CreationFailure) {
+TEST_P(DBIteratorTest, CreationFailure) {
  SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::NewInternalIterator:StatusCallback", [](void* arg) {
        *(reinterpret_cast<Status*>(arg)) = Status::Corruption("test status");
      });
  SyncPoint::GetInstance()->EnableProcessing();

-  Iterator* iter = db_->NewIterator(ReadOptions());
+  Iterator* iter = NewIterator(ReadOptions());
  ASSERT_FALSE(iter->Valid());
  ASSERT_TRUE(iter->status().IsCorruption());
  delete iter;
 }

-TEST_F(DBIteratorTest, TableFilter) {
+TEST_P(DBIteratorTest, TableFilter) {
  ASSERT_OK(Put("a", "1"));
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("b", "2"));
@@ -2035,7 +2068,7 @@ TEST_F(DBIteratorTest, TableFilter) {
      }
      return true;
    };
-    auto iter = db_->NewIterator(opts);
+    auto iter = NewIterator(opts);
    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "a->1");
    iter->Next();
@@ -2061,7 +2094,7 @@ TEST_F(DBIteratorTest, TableFilter) {
    opts.table_filter = [](const TableProperties& props) {
      return props.num_entries != 2;
    };
-    auto iter = db_->NewIterator(opts);
+    auto iter = NewIterator(opts);
    iter->SeekToFirst();
    ASSERT_EQ(IterStatus(iter), "a->1");
    iter->Next();
@@ -2076,7 +2109,7 @@ TEST_F(DBIteratorTest, TableFilter) {
  }
 }

-TEST_F(DBIteratorTest, SkipStatistics) {
+TEST_P(DBIteratorTest, SkipStatistics) {
  Options options = CurrentOptions();
  options.statistics = rocksdb::CreateDBStatistics();
  DestroyAndReopen(options);
@@ -2098,7 +2131,7 @@ TEST_F(DBIteratorTest, SkipStatistics) {
  ASSERT_OK(Delete("e"));
  ASSERT_OK(Delete("f"));

-  Iterator* iter = db_->NewIterator(ReadOptions());
+  Iterator* iter = NewIterator(ReadOptions());
  int count = 0;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    ASSERT_OK(iter->status());
@@ -2109,7 +2142,7 @@ TEST_F(DBIteratorTest, SkipStatistics) {
  skip_count += 8; // 3 deletes + 3 original keys + 2 lower in sequence
  ASSERT_EQ(skip_count, TestGetTickerCount(options, NUMBER_ITER_SKIP));

-  iter = db_->NewIterator(ReadOptions());
+  iter = NewIterator(ReadOptions());
  count = 0;
  for (iter->SeekToLast(); iter->Valid(); iter->Prev()) {
    ASSERT_OK(iter->status());
@@ -2133,7 +2166,7 @@ TEST_F(DBIteratorTest, SkipStatistics) {
  Slice prefix("b");
  ro.iterate_upper_bound = &prefix;

-  iter = db_->NewIterator(ro);
+  iter = NewIterator(ro);
  count = 0;
  for(iter->Seek("aa"); iter->Valid(); iter->Next()) {
    ASSERT_OK(iter->status());
@@ -2144,7 +2177,7 @@ TEST_F(DBIteratorTest, SkipStatistics) {
  skip_count += 6; // 3 deletes + 3 original keys
  ASSERT_EQ(skip_count, TestGetTickerCount(options, NUMBER_ITER_SKIP));

-  iter = db_->NewIterator(ro);
+  iter = NewIterator(ro);
  count = 0;
  for(iter->SeekToLast(); iter->Valid(); iter->Prev()) {
    ASSERT_OK(iter->status());
@@ -2157,6 +2190,189 @@ TEST_F(DBIteratorTest, SkipStatistics) {
  ASSERT_EQ(skip_count, TestGetTickerCount(options, NUMBER_ITER_SKIP));
 }

+TEST_P(DBIteratorTest, SeekAfterHittingManyInternalKeys) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  ReadOptions ropts;
+  ropts.max_skippable_internal_keys = 2;
+
+  Put("1", "val_1");
+  // Add more tombstones than max_skippable_internal_keys so that Next() fails.
+  Delete("2");
+  Delete("3");
+  Delete("4");
+  Delete("5");
+  Put("6", "val_6");
+
+  unique_ptr<Iterator> iter(NewIterator(ropts));
+  iter->SeekToFirst();
+
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().ToString(), "1");
+  ASSERT_EQ(iter->value().ToString(), "val_1");
+
+  // This should fail as incomplete due to too many non-visible internal keys on
+  // the way to the next valid user key.
+  iter->Next();
+  ASSERT_TRUE(!iter->Valid());
+  ASSERT_TRUE(iter->status().IsIncomplete());
+
+  // Get the internal key at which Next() failed.
+  std::string prop_value;
+  ASSERT_OK(iter->GetProperty("rocksdb.iterator.internal-key", &prop_value));
+  ASSERT_EQ("4", prop_value);
+
+  // Create a new iterator to seek to the internal key.
+  unique_ptr<Iterator> iter2(NewIterator(ropts));
+  iter2->Seek(prop_value);
+  ASSERT_TRUE(iter2->Valid());
+  ASSERT_OK(iter2->status());
+
+  ASSERT_EQ(iter2->key().ToString(), "6");
+  ASSERT_EQ(iter2->value().ToString(), "val_6");
+}
+
+INSTANTIATE_TEST_CASE_P(DBIteratorTestInstance, DBIteratorTest,
+                        testing::Values(true, false));
+
+// Tests how DBIter work with ReadCallback
+class DBIteratorWithReadCallbackTest : public DBIteratorTest {};
+
+TEST_F(DBIteratorWithReadCallbackTest, ReadCallback) {
+  class TestReadCallback : public ReadCallback {
+   public:
+    explicit TestReadCallback(SequenceNumber last_visible_seq)
+        : last_visible_seq_(last_visible_seq) {}
+
+    bool IsCommitted(SequenceNumber seq) override {
+      return seq <= last_visible_seq_;
+    }
+
+   private:
+    SequenceNumber last_visible_seq_;
+  };
+
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_OK(Put("foo", "v2"));
+  ASSERT_OK(Put("foo", "v3"));
+  ASSERT_OK(Put("a", "va"));
+  ASSERT_OK(Put("z", "vz"));
+  SequenceNumber seq1 = db_->GetLatestSequenceNumber();
+  TestReadCallback callback1(seq1);
+  ASSERT_OK(Put("foo", "v4"));
+  ASSERT_OK(Put("foo", "v5"));
+  ASSERT_OK(Put("bar", "v7"));
+
+  SequenceNumber seq2 = db_->GetLatestSequenceNumber();
+  auto* cfd =
+      reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
+          ->cfd();
+  // The iterator are suppose to see data before seq1.
+  Iterator* iter =
+      dbfull()->NewIteratorImpl(ReadOptions(), cfd, seq2, &callback1);
+
+  // Seek
+  // The latest value of "foo" before seq1 is "v3"
+  iter->Seek("foo");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v3", iter->value());
+  // "bar" is not visible to the iterator. It will move on to the next key
+  // "foo".
+  iter->Seek("bar");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v3", iter->value());
+
+  // Next
+  // Seek to "a"
+  iter->Seek("a");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("va", iter->value());
+  // "bar" is not visible to the iterator. It will move on to the next key
+  // "foo".
+  iter->Next();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v3", iter->value());
+
+  // Prev
+  // Seek to "z"
+  iter->Seek("z");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("vz", iter->value());
+  // The previous key is "foo", which is visible to the iterator.
+  iter->Prev();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v3", iter->value());
+  // "bar" is not visible to the iterator. It will move on to the next key "a".
+  iter->Prev();  // skipping "bar"
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("a", iter->key());
+  ASSERT_EQ("va", iter->value());
+
+  // SeekForPrev
+  // The previous key is "foo", which is visible to the iterator.
+  iter->SeekForPrev("y");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v3", iter->value());
+  // "bar" is not visible to the iterator. It will move on to the next key "a".
+  iter->SeekForPrev("bar");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("a", iter->key());
+  ASSERT_EQ("va", iter->value());
+
+  delete iter;
+
+  // Prev beyond max_sequential_skip_in_iterations
+  uint64_t num_versions =
+      CurrentOptions().max_sequential_skip_in_iterations + 10;
+  for (uint64_t i = 0; i < num_versions; i++) {
+    ASSERT_OK(Put("bar", ToString(i)));
+  }
+  SequenceNumber seq3 = db_->GetLatestSequenceNumber();
+  TestReadCallback callback2(seq3);
+  ASSERT_OK(Put("bar", "v8"));
+  SequenceNumber seq4 = db_->GetLatestSequenceNumber();
+
+  // The iterator is suppose to see data before seq3.
+  iter = dbfull()->NewIteratorImpl(ReadOptions(), cfd, seq4, &callback2);
+  // Seek to "z", which is visible.
+  iter->Seek("z");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("vz", iter->value());
+  // Previous key is "foo" and the last value "v5" is visible.
+  iter->Prev();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("foo", iter->key());
+  ASSERT_EQ("v5", iter->value());
+  // Since the number of values of "bar" is more than
+  // max_sequential_skip_in_iterations, Prev() will ultimately fallback to
+  // seek in forward direction. Here we test the fallback seek is correct.
+  // The last visible value should be (num_versions - 1), as "v8" is not
+  // visible.
+  iter->Prev();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("bar", iter->key());
+  ASSERT_EQ(ToString(num_versions - 1), iter->value());
+
+  delete iter;
+}
+
 }  // namespace rocksdb

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

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

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

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

 namespace rocksdb {

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

 TEST_F(DBMergeOperatorTest, LimitMergeOperands) {
@@ -449,8 +482,152 @@ TEST_P(MergeOperatorPinningTest, TailingIterator) {
  writer_thread.join();
  reader_thread.join();
 }
+
+TEST_F(DBMergeOperatorTest, TailingIteratorMemtableUnrefedBySomeoneElse) {
+  Options options = CurrentOptions();
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  DestroyAndReopen(options);
+
+  // Overview of the test:
+  //  * There are two merge operands for the same key: one in an sst file,
+  //    another in a memtable.
+  //  * Seek a tailing iterator to this key.
+  //  * As part of the seek, the iterator will:
+  //      (a) first visit the operand in the memtable and tell ForwardIterator
+  //          to pin this operand, then
+  //      (b) move on to the operand in the sst file, then pass both operands
+  //          to merge operator.
+  //  * The memtable may get flushed and unreferenced by another thread between
+  //    (a) and (b). The test simulates it by flushing the memtable inside a
+  //    SyncPoint callback located between (a) and (b).
+  //  * In this case it's ForwardIterator's responsibility to keep the memtable
+  //    pinned until (b) is complete. There used to be a bug causing
+  //    ForwardIterator to not pin it in some circumstances. This test
+  //    reproduces it.
+
+  db_->Merge(WriteOptions(), "key", "sst");
+  db_->Flush(FlushOptions()); // Switch to SuperVersion A
+  db_->Merge(WriteOptions(), "key", "memtable");
+
+  // Pin SuperVersion A
+  std::unique_ptr<Iterator> someone_else(db_->NewIterator(ReadOptions()));
+
+  bool pushed_first_operand = false;
+  bool stepped_to_next_operand = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBIter::MergeValuesNewToOld:PushedFirstOperand", [&](void*) {
+        EXPECT_FALSE(pushed_first_operand);
+        pushed_first_operand = true;
+        db_->Flush(FlushOptions()); // Switch to SuperVersion B
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBIter::MergeValuesNewToOld:SteppedToNextOperand", [&](void*) {
+        EXPECT_FALSE(stepped_to_next_operand);
+        stepped_to_next_operand = true;
+        someone_else.reset(); // Unpin SuperVersion A
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  ReadOptions ro;
+  ro.tailing = true;
+  std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+  iter->Seek("key");
+
+  ASSERT_TRUE(iter->status().ok());
+  ASSERT_TRUE(iter->Valid());
+  EXPECT_EQ(std::string("sst,memtable"), iter->value().ToString());
+  EXPECT_TRUE(pushed_first_operand);
+  EXPECT_TRUE(stepped_to_next_operand);
+}
 #endif  // ROCKSDB_LITE

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

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

 TEST_F(DBOptionsTest, DeleteObsoleteFilesPeriodChange) {
@@ -376,6 +376,30 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

+TEST_F(DBSSTTest, OpenDBWithExistingTrash) {
+  Options options = CurrentOptions();
+
+  options.sst_file_manager.reset(
+      NewSstFileManager(env_, nullptr, "", 1024 * 1024 /* 1 MB/sec */));
+  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());
+
+  Destroy(last_options_);
+
+  // Add some trash files to the db directory so the DB can clean them up
+  env_->CreateDirIfMissing(dbname_);
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "001.sst.trash"));
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "002.sst.trash"));
+  ASSERT_OK(WriteStringToFile(env_, "abc", dbname_ + "/" + "003.sst.trash"));
+
+  // Reopen the DB and verify that it deletes existing trash files
+  ASSERT_OK(TryReopen(options));
+  sfm->WaitForEmptyTrash();
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "001.sst.trash"));
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "002.sst.trash"));
+  ASSERT_NOK(env_->FileExists(dbname_ + "/" + "003.sst.trash"));
+}
+
+
 // Create a DB with 2 db_paths, and generate multiple files in the 2
 // db_paths using CompactRangeOptions, make sure that files that were
 // deleted from first db_path were deleted using DeleteScheduler and
@@ -480,6 +480,36 @@ TEST_F(DBTest, SingleDeletePutFlush) {
                         kSkipUniversalCompaction | kSkipMergePut));
 }

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

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

  std::string value;
  Status s = db_->Get(ReadOptions(), key1, &value);
@@ -2191,6 +2223,8 @@ class ModelDB : public DB {
    batch.Put(cf, k, v);
    return Write(o, &batch);
  }
+  using DB::Close;
+  virtual Status Close() override { return Status::OK(); }
  using DB::Delete;
  virtual Status Delete(const WriteOptions& o, ColumnFamilyHandle* cf,
                        const Slice& key) override {
@@ -3488,12 +3522,16 @@ TEST_F(DBTest, GetThreadStatus) {
    const int kTestCount = 3;
    const unsigned int kHighPriCounts[kTestCount] = {3, 2, 5};
    const unsigned int kLowPriCounts[kTestCount] = {10, 15, 3};
+    const unsigned int kBottomPriCounts[kTestCount] = {2, 1, 4};
    for (int test = 0; test < kTestCount; ++test) {
      // Change the number of threads in high / low priority pool.
      env_->SetBackgroundThreads(kHighPriCounts[test], Env::HIGH);
      env_->SetBackgroundThreads(kLowPriCounts[test], Env::LOW);
+      env_->SetBackgroundThreads(kBottomPriCounts[test], Env::BOTTOM);
      // Wait to ensure the all threads has been registered
      unsigned int thread_type_counts[ThreadStatus::NUM_THREAD_TYPES];
+      // TODO(ajkr): it'd be better if SetBackgroundThreads returned only after
+      // all threads have been registered.
      // Try up to 60 seconds.
      for (int num_try = 0; num_try < 60000; num_try++) {
        env_->SleepForMicroseconds(1000);
@@ -3508,20 +3546,21 @@ TEST_F(DBTest, GetThreadStatus) {
        if (thread_type_counts[ThreadStatus::HIGH_PRIORITY] ==
                kHighPriCounts[test] &&
            thread_type_counts[ThreadStatus::LOW_PRIORITY] ==
-                kLowPriCounts[test]) {
+                kLowPriCounts[test] &&
+            thread_type_counts[ThreadStatus::BOTTOM_PRIORITY] ==
+                kBottomPriCounts[test]) {
          break;
        }
      }
-      // Verify the total number of threades
-      ASSERT_EQ(thread_type_counts[ThreadStatus::HIGH_PRIORITY] +
-                    thread_type_counts[ThreadStatus::LOW_PRIORITY],
-                kHighPriCounts[test] + kLowPriCounts[test]);
      // Verify the number of high-priority threads
      ASSERT_EQ(thread_type_counts[ThreadStatus::HIGH_PRIORITY],
                kHighPriCounts[test]);
      // Verify the number of low-priority threads
      ASSERT_EQ(thread_type_counts[ThreadStatus::LOW_PRIORITY],
                kLowPriCounts[test]);
+      // Verify the number of bottom-priority threads
+      ASSERT_EQ(thread_type_counts[ThreadStatus::BOTTOM_PRIORITY],
+                kBottomPriCounts[test]);
    }
    if (i == 0) {
      // repeat the test with multiple column families
@@ -4375,7 +4414,6 @@ TEST_F(DBTest, DynamicFIFOCompactionOptions) {
  ASSERT_EQ(dbfull()->GetOptions().compaction_options_fifo.allow_compaction,
            true);
 }
-#endif  // ROCKSDB_LITE

 TEST_F(DBTest, DynamicUniversalCompactionOptions) {
  Options options;
@@ -4444,6 +4482,7 @@ TEST_F(DBTest, DynamicUniversalCompactionOptions) {
      dbfull()->GetOptions().compaction_options_universal.allow_trivial_move,
      false);
 }
+#endif  // ROCKSDB_LITE

 TEST_F(DBTest, FileCreationRandomFailure) {
  Options options;
@@ -5049,55 +5088,84 @@ TEST_F(DBTest, PromoteL0Failure) {
  status = experimental::PromoteL0(db_, db_->DefaultColumnFamily());
  ASSERT_TRUE(status.IsInvalidArgument());
 }
-#endif  // ROCKSDB_LITE

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

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

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

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

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

 // Github issue #595
@@ -5317,12 +5385,15 @@ class WriteStallListener : public EventListener {
 public:
  WriteStallListener() : condition_(WriteStallCondition::kNormal) {}
  void OnStallConditionsChanged(const WriteStallInfo& info) override {
+    MutexLock l(&mutex_);
    condition_ = info.condition.cur;
  }
  bool CheckCondition(WriteStallCondition expected) {
+    MutexLock l(&mutex_);
    return expected == condition_;
  }
 private:
+  port::Mutex mutex_;
  WriteStallCondition condition_;
 };

@@ -5651,6 +5722,50 @@ TEST_F(DBTest, PauseBackgroundWorkTest) {
  // now it's done
  ASSERT_TRUE(done.load());
 }
+
+// Keep spawning short-living threads that create an iterator and quit.
+// Meanwhile in another thread keep flushing memtables.
+// This used to cause a deadlock.
+TEST_F(DBTest, ThreadLocalPtrDeadlock) {
+  std::atomic<int> flushes_done{0};
+  std::atomic<int> threads_destroyed{0};
+  auto done = [&] {
+    return flushes_done.load() > 10;
+  };
+
+  std::thread flushing_thread([&] {
+    for (int i = 0; !done(); ++i) {
+      ASSERT_OK(db_->Put(WriteOptions(), Slice("hi"),
+                         Slice(std::to_string(i).c_str())));
+      ASSERT_OK(db_->Flush(FlushOptions()));
+      int cnt = ++flushes_done;
+      fprintf(stderr, "Flushed %d times\n", cnt);
+    }
+  });
+
+  std::vector<std::thread> thread_spawning_threads(10);
+  for (auto& t: thread_spawning_threads) {
+    t = std::thread([&] {
+      while (!done()) {
+        {
+          std::thread tmp_thread([&] {
+            auto it = db_->NewIterator(ReadOptions());
+            delete it;
+          });
+          tmp_thread.join();
+        }
+        ++threads_destroyed;
+      }
+    });
+  }
+
+  for (auto& t: thread_spawning_threads) {
+    t.join();
+  }
+  flushing_thread.join();
+  fprintf(stderr, "Done. Flushed %d times, destroyed %d threads\n",
+          flushes_done.load(), threads_destroyed.load());
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1820,11 +1820,17 @@ TEST_F(DBTest2, ReadAmpBitmap) {

 #ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
 TEST_F(DBTest2, ReadAmpBitmapLiveInCacheAfterDBClose) {
-  if (dbname_.find("dev/shm") != std::string::npos) {
-    // /dev/shm dont support getting a unique file id, this mean that
-    // running this test on /dev/shm will fail because lru_cache will load
-    // the blocks again regardless of them being already in the cache
-    return;
+  {
+    const int kIdBufLen = 100;
+    char id_buf[kIdBufLen];
+    std::unique_ptr<RandomAccessFile> file;
+    env_->NewRandomAccessFile(dbname_, &file, EnvOptions());
+    if (file->GetUniqueId(id_buf, kIdBufLen) == 0) {
+      // fs holding db directory doesn't support getting a unique file id,
+      // this means that running this test will fail because lru_cache will load
+      // the blocks again regardless of them being already in the cache
+      return;
+    }
  }
  uint32_t bytes_per_bit[2] = {1, 16};
  for (size_t k = 0; k < 2; k++) {
@@ -2430,6 +2436,60 @@ TEST_F(DBTest2, ReadCallbackTest) {
  }
 }

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

 int main(int argc, char** argv) {
@@ -867,7 +867,6 @@ size_t DBTestBase::CountLiveFiles() {
  db_->GetLiveFilesMetaData(&metadata);
  return metadata.size();
 }
-#endif  // ROCKSDB_LITE

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

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

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

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

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

  // Returns true iff an entry that compares equal to key is in the list.
@@ -863,13 +863,13 @@ class DBTestBase : public testing::Test {
  size_t TotalLiveFiles(int cf = 0);

  size_t CountLiveFiles();
-#endif  // ROCKSDB_LITE

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

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

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

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

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

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

  std::string DumpSSTableList();

@@ -388,10 +388,16 @@ TEST_P(DBTestUniversalCompaction, DynamicUniversalCompactionSizeAmplification) {
  DestroyAndReopen(options);

  int total_picked_compactions = 0;
+  int total_size_amp_compactions = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "UniversalCompactionPicker::PickCompaction:Return", [&](void* arg) {
        if (arg) {
          total_picked_compactions++;
+          Compaction* c = static_cast<Compaction*>(arg);
+          if (c->compaction_reason() ==
+              CompactionReason::kUniversalSizeAmplification) {
+            total_size_amp_compactions++;
+          }
        }
      });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
@@ -445,6 +451,102 @@ TEST_P(DBTestUniversalCompaction, DynamicUniversalCompactionSizeAmplification) {
  // Verify that size amplification did happen
  ASSERT_EQ(NumSortedRuns(1), 1);
  ASSERT_EQ(total_picked_compactions, 1);
+  ASSERT_EQ(total_size_amp_compactions, 1);
+}
+
+TEST_P(DBTestUniversalCompaction, DynamicUniversalCompactionReadAmplification) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.num_levels = 1;
+  options.write_buffer_size = 100 << 10;     // 100KB
+  options.target_file_size_base = 32 << 10;  // 32KB
+  options.level0_file_num_compaction_trigger = 3;
+  // Initial setup of compaction_options_universal will prevent universal
+  // compaction from happening
+  options.compaction_options_universal.max_size_amplification_percent = 2000;
+  options.compaction_options_universal.size_ratio = 0;
+  options.compaction_options_universal.min_merge_width = 100;
+  DestroyAndReopen(options);
+
+  int total_picked_compactions = 0;
+  int total_size_ratio_compactions = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "UniversalCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        if (arg) {
+          total_picked_compactions++;
+          Compaction* c = static_cast<Compaction*>(arg);
+          if (c->compaction_reason() == CompactionReason::kUniversalSizeRatio) {
+            total_size_ratio_compactions++;
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  MutableCFOptions mutable_cf_options;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  Random rnd(301);
+  int key_idx = 0;
+
+  // Generate three files in Level 0. All files are approx the same size.
+  for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+    // Write 110KB (11 values, each 10K)
+    for (int i = 0; i < 11; i++) {
+      ASSERT_OK(Put(1, Key(key_idx), RandomString(&rnd, 10000)));
+      key_idx++;
+    }
+    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    ASSERT_EQ(NumSortedRuns(1), num + 1);
+  }
+  ASSERT_EQ(NumSortedRuns(1), options.level0_file_num_compaction_trigger);
+
+  // Flush whatever is remaining in memtable. This is typically small, about
+  // 30KB.
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  // Verify compaction did not happen
+  ASSERT_EQ(NumSortedRuns(1), options.level0_file_num_compaction_trigger + 1);
+  ASSERT_EQ(total_picked_compactions, 0);
+
+  ASSERT_OK(dbfull()->SetOptions(
+      handles_[1],
+      {{"compaction_options_universal",
+        "{min_merge_width=2;max_merge_width=2;size_ratio=100;}"}}));
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.min_merge_width,
+            2);
+  ASSERT_EQ(dbfull()
+                ->GetOptions(handles_[1])
+                .compaction_options_universal.max_merge_width,
+            2);
+  ASSERT_EQ(
+      dbfull()->GetOptions(handles_[1]).compaction_options_universal.size_ratio,
+      100);
+
+  ASSERT_OK(dbfull()->TEST_GetLatestMutableCFOptions(handles_[1],
+                                                     &mutable_cf_options));
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.size_ratio, 100);
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.min_merge_width, 2);
+  ASSERT_EQ(mutable_cf_options.compaction_options_universal.max_merge_width, 2);
+
+  dbfull()->TEST_WaitForCompact();
+
+  // Files in L0 are approx: 0.3 (30KB), 1, 1, 1.
+  // On compaction: the files are below the size amp threshold, so we
+  // fallthrough to checking read amp conditions. The configured size ratio is
+  // not big enough to take 0.3 into consideration. So the next files 1 and 1
+  // are compacted together first as they satisfy size ratio condition and
+  // (min_merge_width, max_merge_width) condition, to give out a file size of 2.
+  // Next, the newly generated 2 and the last file 1 are compacted together. So
+  // at the end: #sortedRuns = 2, #picked_compactions = 2, and all the picked
+  // ones are size ratio based compactions.
+  ASSERT_EQ(NumSortedRuns(1), 2);
+  // If max_merge_width had not been changed dynamically above, and if it
+  // continued to be the default value of UINIT_MAX, total_picked_compactions
+  // would have been 1.
+  ASSERT_EQ(total_picked_compactions, 2);
+  ASSERT_EQ(total_size_ratio_compactions, 2);
 }

 TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
@@ -39,52 +39,6 @@ TEST_P(DBWriteTest, SyncAndDisableWAL) {
  ASSERT_TRUE(dbfull()->Write(write_options, &batch).IsInvalidArgument());
 }

-// Sequence number should be return through input write batch.
-TEST_P(DBWriteTest, ReturnSeuqneceNumber) {
-  Random rnd(4422);
-  Open();
-  for (int i = 0; i < 100; i++) {
-    WriteBatch batch;
-    batch.Put("key" + ToString(i), test::RandomHumanReadableString(&rnd, 10));
-    ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
-    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(),
-              WriteBatchInternal::Sequence(&batch));
-  }
-}
-
-TEST_P(DBWriteTest, ReturnSeuqneceNumberMultiThreaded) {
-  constexpr size_t kThreads = 16;
-  constexpr size_t kNumKeys = 1000;
-  Open();
-  ASSERT_EQ(0, dbfull()->GetLatestSequenceNumber());
-  // Check each sequence is used once and only once.
-  std::vector<std::atomic_flag> flags(kNumKeys * kThreads + 1);
-  for (size_t i = 0; i < flags.size(); i++) {
-    flags[i].clear();
-  }
-  auto writer = [&](size_t id) {
-    Random rnd(4422 + static_cast<uint32_t>(id));
-    for (size_t k = 0; k < kNumKeys; k++) {
-      WriteBatch batch;
-      batch.Put("key" + ToString(id) + "-" + ToString(k),
-                test::RandomHumanReadableString(&rnd, 10));
-      ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
-      SequenceNumber sequence = WriteBatchInternal::Sequence(&batch);
-      ASSERT_GT(sequence, 0);
-      ASSERT_LE(sequence, kNumKeys * kThreads);
-      // The sequence isn't consumed by someone else.
-      ASSERT_FALSE(flags[sequence].test_and_set());
-    }
-  };
-  std::vector<port::Thread> threads;
-  for (size_t i = 0; i < kThreads; i++) {
-    threads.emplace_back(writer, i);
-  }
-  for (size_t i = 0; i < kThreads; i++) {
-    threads[i].join();
-  }
-}
-
 TEST_P(DBWriteTest, IOErrorOnWALWritePropagateToWriteThreadFollower) {
  constexpr int kNumThreads = 5;
  std::unique_ptr<FaultInjectionTestEnv> mock_env(
@@ -125,6 +125,26 @@ int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
  return r;
 }

+int InternalKeyComparator::CompareKeySeq(const Slice& akey,
+                                         const Slice& bkey) const {
+  // Order by:
+  //    increasing user key (according to user-supplied comparator)
+  //    decreasing sequence number
+  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
+  PERF_COUNTER_ADD(user_key_comparison_count, 1);
+  if (r == 0) {
+    // Shift the number to exclude the last byte which contains the value type
+    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8) >> 8;
+    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8) >> 8;
+    if (anum > bnum) {
+      r = -1;
+    } else if (anum < bnum) {
+      r = +1;
+    }
+  }
+  return r;
+}
+
 int InternalKeyComparator::Compare(const ParsedInternalKey& a,
                                   const ParsedInternalKey& b) const {
  // Order by:
@@ -162,6 +162,8 @@ class InternalKeyComparator

  virtual const char* Name() const override;
  virtual int Compare(const Slice& a, const Slice& b) const override;
+  // Same as Compare except that it excludes the value type from comparison
+  virtual int CompareKeySeq(const Slice& a, const Slice& b) const;
  virtual void FindShortestSeparator(std::string* start,
                                     const Slice& limit) const override;
  virtual void FindShortSuccessor(std::string* key) const override;
@@ -132,11 +132,15 @@ Status ExternalSstFileIngestionJob::Prepare(
  return status;
 }

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

@@ -167,7 +172,6 @@ Status ExternalSstFileIngestionJob::Run() {
  // It is safe to use this instead of LastAllocatedSequence since we are
  // the only active writer, and hence they are equal
  const SequenceNumber last_seqno = versions_->LastSequence();
-  SuperVersion* super_version = cfd_->GetSuperVersion();
  edit_.SetColumnFamily(cfd_->GetID());
  // The levels that the files will be ingested into

@@ -375,54 +379,6 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
  return status;
 }

-Status ExternalSstFileIngestionJob::IngestedFilesOverlapWithMemtables(
-    SuperVersion* sv, bool* overlap) {
-  *overlap = false;
-  // Create an InternalIterator over all memtables
-  Arena arena;
-  ReadOptions ro;
-  ro.total_order_seek = true;
-  MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(), &arena);
-  merge_iter_builder.AddIterator(sv->mem->NewIterator(ro, &arena));
-  sv->imm->AddIterators(ro, &merge_iter_builder);
-  ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
-
-  std::vector<InternalIterator*> memtable_range_del_iters;
-  auto* active_range_del_iter = sv->mem->NewRangeTombstoneIterator(ro);
-  if (active_range_del_iter != nullptr) {
-    memtable_range_del_iters.push_back(active_range_del_iter);
-  }
-  sv->imm->AddRangeTombstoneIterators(ro, &memtable_range_del_iters);
-  RangeDelAggregator range_del_agg(cfd_->internal_comparator(),
-                                   {} /* snapshots */,
-                                   false /* collapse_deletions */);
-  Status status;
-  {
-    std::unique_ptr<InternalIterator> memtable_range_del_iter(
-        NewMergingIterator(&cfd_->internal_comparator(),
-                           memtable_range_del_iters.empty()
-                               ? nullptr
-                               : &memtable_range_del_iters[0],
-                           static_cast<int>(memtable_range_del_iters.size())));
-    status = range_del_agg.AddTombstones(std::move(memtable_range_del_iter));
-  }
-  if (status.ok()) {
-    for (IngestedFileInfo& f : files_to_ingest_) {
-      status = IngestedFileOverlapWithIteratorRange(&f, memtable_iter.get(),
-                                                    overlap);
-      if (!status.ok() || *overlap == true) {
-        break;
-      }
-      if (range_del_agg.IsRangeOverlapped(f.smallest_user_key,
-                                          f.largest_user_key)) {
-        *overlap = true;
-        break;
-      }
-    }
-  }
-  return status;
-}
-
 Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
    SuperVersion* sv, bool force_global_seqno, CompactionStyle compaction_style,
    IngestedFileInfo* file_to_ingest, SequenceNumber* assigned_seqno) {
@@ -473,7 +429,10 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
                            return f1->largest_seqno < f2->largest_seqno;
                          }))
                ->largest_seqno;
-        if (level_largest_seqno != 0) {
+        // should only assign seqno to current level's largest seqno when
+        // the file fits
+        if (level_largest_seqno != 0 &&
+            IngestedFileFitInLevel(file_to_ingest, lvl)) {
          *assigned_seqno = level_largest_seqno;
        } else {
          continue;
@@ -550,6 +509,9 @@ Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
  std::string seqno_val;
  PutFixed64(&seqno_val, seqno);
  status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
+  if (status.ok()) {
+    status = rwfile->Fsync();
+  }
  if (status.ok()) {
    file_to_ingest->assigned_seqno = seqno;
  }
@@ -84,8 +84,12 @@ class ExternalSstFileIngestionJob {
  // Check if we need to flush the memtable before running the ingestion job
  // This will be true if the files we are ingesting are overlapping with any
  // key range in the memtable.
-  // REQUIRES: Mutex held
-  Status NeedsFlush(bool* flush_needed);
+  //
+  // @param super_version A referenced SuperVersion that will be held for the
+  //    duration of this function.
+  //
+  // Thread-safe
+  Status NeedsFlush(bool* flush_needed, SuperVersion* super_version);

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

-  // Check if the files we are ingesting overlap with any memtable.
-  // REQUIRES: Mutex held
-  Status IngestedFilesOverlapWithMemtables(SuperVersion* sv, bool* overlap);
-
  // Assign `file_to_ingest` the appropriate sequence number and  the lowest
  // possible level that it can be ingested to according to compaction_style.
  // REQUIRES: Mutex held
@@ -388,9 +388,10 @@ class SstFileWriterCollector : public TablePropertiesCollector {
  const char* Name() const override { return name_.c_str(); }

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

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

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

 namespace rocksdb {

+const char* GetFlushReasonString (FlushReason flush_reason) {
+  switch (flush_reason) {
+    case FlushReason::kUnknown:
+      return "Unknown";
+    case FlushReason::kGetLiveFiles:
+      return "Get Live Files";
+    case FlushReason::kShutDown:
+      return "Shut down";
+    case FlushReason::kExternalFileIngestion:
+      return "External File Ingestion";
+    case FlushReason::kManualCompaction:
+      return "Manual Compaction";
+    case FlushReason::kWriteBufferManager:
+      return "Write Buffer Manager";
+    case FlushReason::kWriteBufferFull:
+      return "Write Buffer Full";
+    case FlushReason::kTest:
+      return "Test";
+    case FlushReason::kSuperVersionChange:
+      return "SuperVersion Change";
+    default:
+      return "Invalid";
+  }
+}
+
+
 FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                   const ImmutableDBOptions& db_options,
                   const MutableCFOptions& mutable_cf_options,
@@ -209,6 +235,8 @@ Status FlushJob::Run(FileMetaData* file_meta) {
  auto stream = event_logger_->LogToBuffer(log_buffer_);
  stream << "job" << job_context_->job_id << "event"
         << "flush_finished";
+  stream << "output_compression"
+         << CompressionTypeToString(output_compression_);
  stream << "lsm_state";
  stream.StartArray();
  auto vstorage = cfd_->current()->storage_info();
@@ -276,12 +304,13 @@ Status FlushJob::WriteLevel0Table() {
      total_memory_usage += m->ApproximateMemoryUsage();
    }

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

    {
      ScopedArenaIterator iter(
@@ -362,6 +391,7 @@ Status FlushJob::WriteLevel0Table() {
  InternalStats::CompactionStats stats(1);
  stats.micros = db_options_.env->NowMicros() - start_micros;
  stats.bytes_written = meta_.fd.GetFileSize();
+  MeasureTime(stats_, FLUSH_TIME, stats.micros);
  cfd_->internal_stats()->AddCompactionStats(0 /* level */, stats);
  cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
                                     meta_.fd.GetFileSize());
@@ -40,6 +40,7 @@ class FlushJobTest : public testing::Test {
    EXPECT_OK(env_->CreateDirIfMissing(dbname_));
    db_options_.db_paths.emplace_back(dbname_,
                                      std::numeric_limits<uint64_t>::max());
+    db_options_.statistics = rocksdb::CreateDBStatistics();
    // TODO(icanadi) Remove this once we mock out VersionSet
    NewDB();
    std::vector<ColumnFamilyDescriptor> column_families;
@@ -138,16 +139,22 @@ TEST_F(FlushJobTest, NonEmpty) {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

 Slice MemTableRep::UserKey(const char* key) const {
@@ -281,8 +289,7 @@ class MemTableIterator : public InternalIterator {
 #ifndef NDEBUG
    // Assert that the MemTableIterator is never deleted while
    // Pinning is Enabled.
-    assert(!pinned_iters_mgr_ ||
-           (pinned_iters_mgr_ && !pinned_iters_mgr_->PinningEnabled()));
+    assert(!pinned_iters_mgr_ || !pinned_iters_mgr_->PinningEnabled());
 #endif
    if (arena_mode_) {
      iter_->~Iterator();
@@ -436,7 +443,7 @@ MemTable::MemTableStats MemTable::ApproximateStats(const Slice& start_ikey,
  return {entry_count * (data_size / n), entry_count};
 }

-void MemTable::Add(SequenceNumber s, ValueType type,
+bool MemTable::Add(SequenceNumber s, ValueType type,
                   const Slice& key, /* user key */
                   const Slice& value, bool allow_concurrent,
                   MemTablePostProcessInfo* post_process_info) {
@@ -471,9 +478,15 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    if (insert_with_hint_prefix_extractor_ != nullptr &&
        insert_with_hint_prefix_extractor_->InDomain(key_slice)) {
      Slice prefix = insert_with_hint_prefix_extractor_->Transform(key_slice);
-      table->InsertWithHint(handle, &insert_hints_[prefix]);
+      bool res = table->InsertKeyWithHint(handle, &insert_hints_[prefix]);
+      if (!res) {
+        return res;
+      }
    } else {
-      table->Insert(handle);
+      bool res = table->InsertKey(handle);
+      if (!res) {
+        return res;
+      }
    }

    // this is a bit ugly, but is the way to avoid locked instructions
@@ -506,7 +519,10 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    assert(post_process_info == nullptr);
    UpdateFlushState();
  } else {
-    table->InsertConcurrently(handle);
+    bool res = table->InsertKeyConcurrently(handle);
+    if (!res) {
+      return res;
+    }

    assert(post_process_info != nullptr);
    post_process_info->num_entries++;
@@ -536,6 +552,7 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    is_range_del_table_empty_ = false;
  }
  UpdateOldestKeyTime();
+  return true;
 }

 // Callback from MemTable::Get()
@@ -571,11 +588,12 @@ struct Saver {

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

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

  // entry format is:
  //    klength  varint32
@@ -791,8 +809,9 @@ void MemTable::Update(SequenceNumber seq,
      // Correct user key
      const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
      ValueType type;
-      SequenceNumber unused;
-      UnPackSequenceAndType(tag, &unused, &type);
+      SequenceNumber existing_seq;
+      UnPackSequenceAndType(tag, &existing_seq, &type);
+      assert(existing_seq != seq);
      if (type == kTypeValue) {
        Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
        uint32_t prev_size = static_cast<uint32_t>(prev_value.size());
@@ -815,7 +834,10 @@ void MemTable::Update(SequenceNumber seq,
  }

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

 bool MemTable::UpdateCallback(SequenceNumber seq,
@@ -167,7 +167,10 @@ class MemTable {
  //
  // REQUIRES: if allow_concurrent = false, external synchronization to prevent
  // simultaneous operations on the same MemTable.
-  void Add(SequenceNumber seq, ValueType type, const Slice& key,
+  //
+  // Returns false if MemTableRepFactory::CanHandleDuplicatedKey() is true and
+  // the <key, seq> already exists.
+  bool Add(SequenceNumber seq, ValueType type, const Slice& key,
           const Slice& value, bool allow_concurrent = false,
           MemTablePostProcessInfo* post_process_info = nullptr);

@@ -368,6 +371,11 @@ class MemTable {
    return oldest_key_time_.load(std::memory_order_relaxed);
  }

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

@@ -437,6 +445,9 @@ class MemTable {
  // Timestamp of oldest key
  std::atomic<uint64_t> oldest_key_time_;

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

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

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

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

  uint64_t GetMinLogContainingPrepSection();

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

 #include "db/merge_helper.h"

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

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

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

  // the scratch area that holds the result of MergeUntil
@@ -45,6 +45,10 @@ class SnapshotList {
    list_.prev_ = &list_;
    list_.next_ = &list_;
    list_.number_ = 0xFFFFFFFFL;      // placeholder marker, for debugging
+    // Set all the variables to make UBSAN happy.
+    list_.list_ = nullptr;
+    list_.unix_time_ = 0;
+    list_.is_write_conflict_boundary_ = false;
    count_ = 0;
  }

@@ -108,22 +112,6 @@ class SnapshotList {
    return ret;
  }

-  // Whether there is an active snapshot in range [lower_bound, upper_bound).
-  bool HasSnapshotInRange(SequenceNumber lower_bound,
-                          SequenceNumber upper_bound) {
-    if (empty()) {
-      return false;
-    }
-    const SnapshotImpl* s = &list_;
-    while (s->next_ != &list_) {
-      if (s->next_->number_ >= lower_bound) {
-        return s->next_->number_ < upper_bound;
-      }
-      s = s->next_;
-    }
-    return false;
-  }
-
  // get the sequence number of the most recent snapshot
  SequenceNumber GetNewest() {
    if (empty()) {
@@ -272,9 +272,8 @@ InternalIterator* TableCache::NewRangeTombstoneIterator(
    const InternalKeyComparator& icomparator, const FileDescriptor& fd,
    HistogramImpl* file_read_hist, bool skip_filters, int level) {
  Status s;
-  TableReader* table_reader = nullptr;
  Cache::Handle* handle = nullptr;
-  table_reader = fd.table_reader;
+  TableReader* table_reader = fd.table_reader;
  if (table_reader == nullptr) {
    s = FindTable(env_options, icomparator, fd, &handle,
                  options.read_tier == kBlockCacheTier /* no_io */,
@@ -60,7 +60,7 @@ InternalKeyPropertiesCollector::GetReadableProperties() const {
 namespace {

 uint64_t GetUint64Property(const UserCollectedProperties& props,
-                           const std::string property_name,
+                           const std::string& property_name,
                           bool* property_present) {
  auto pos = props.find(property_name);
  if (pos == props.end()) {
@@ -424,129 +424,261 @@ bool SomeFileOverlapsRange(
 }

 namespace {
-
-// An internal iterator.  For a given version/level pair, yields
-// information about the files in the level.  For a given entry, key()
-// is the largest key that occurs in the file, and value() is an
-// 16-byte value containing the file number and file size, both
-// encoded using EncodeFixed64.
-class LevelFileNumIterator : public InternalIterator {
+class LevelIterator final : public InternalIterator {
 public:
-  LevelFileNumIterator(const InternalKeyComparator& icmp,
-                       const LevelFilesBrief* flevel, bool should_sample)
-      : icmp_(icmp),
-        flevel_(flevel),
-        index_(static_cast<uint32_t>(flevel->num_files)),
-        current_value_(0, 0, 0),  // Marks as invalid
-        should_sample_(should_sample) {}
-  virtual bool Valid() const override { return index_ < flevel_->num_files; }
-  virtual void Seek(const Slice& target) override {
-    index_ = FindFile(icmp_, *flevel_, target);
-  }
-  virtual void SeekForPrev(const Slice& target) override {
-    SeekForPrevImpl(target, &icmp_);
-  }
-
-  virtual void SeekToFirst() override { index_ = 0; }
-  virtual void SeekToLast() override {
-    index_ = (flevel_->num_files == 0)
-                 ? 0
-                 : static_cast<uint32_t>(flevel_->num_files) - 1;
-  }
-  virtual void Next() override {
-    assert(Valid());
-    index_++;
-  }
-  virtual void Prev() override {
-    assert(Valid());
-    if (index_ == 0) {
-      index_ = static_cast<uint32_t>(flevel_->num_files);  // Marks as invalid
-    } else {
-      index_--;
-    }
-  }
-  Slice key() const override {
-    assert(Valid());
-    return flevel_->files[index_].largest_key;
-  }
-  Slice value() const override {
-    assert(Valid());
-
-    auto file_meta = flevel_->files[index_];
-    if (should_sample_) {
-      sample_file_read_inc(file_meta.file_metadata);
-    }
-    current_value_ = file_meta.fd;
-    return Slice(reinterpret_cast<const char*>(&current_value_),
-                 sizeof(FileDescriptor));
-  }
-  virtual Status status() const override { return Status::OK(); }
-
- private:
-  const InternalKeyComparator icmp_;
-  const LevelFilesBrief* flevel_;
-  uint32_t index_;
-  mutable FileDescriptor current_value_;
-  bool should_sample_;
-};
-
-class LevelFileIteratorState : public TwoLevelIteratorState {
- public:
-  // @param skip_filters Disables loading/accessing the filter block
-  LevelFileIteratorState(TableCache* table_cache,
-                         const ReadOptions& read_options,
-                         const EnvOptions& env_options,
-                         const InternalKeyComparator& icomparator,
-                         HistogramImpl* file_read_hist, bool for_compaction,
-                         bool prefix_enabled, bool skip_filters, int level,
-                         RangeDelAggregator* range_del_agg)
-      : TwoLevelIteratorState(prefix_enabled),
-        table_cache_(table_cache),
+  LevelIterator(TableCache* table_cache, const ReadOptions& read_options,
+                const EnvOptions& env_options,
+                const InternalKeyComparator& icomparator,
+                const LevelFilesBrief* flevel, bool should_sample,
+                HistogramImpl* file_read_hist, bool for_compaction,
+                bool skip_filters, int level, RangeDelAggregator* range_del_agg)
+      : table_cache_(table_cache),
        read_options_(read_options),
        env_options_(env_options),
        icomparator_(icomparator),
+        flevel_(flevel),
        file_read_hist_(file_read_hist),
+        should_sample_(should_sample),
        for_compaction_(for_compaction),
        skip_filters_(skip_filters),
+        file_index_(flevel_->num_files),
        level_(level),
-        range_del_agg_(range_del_agg) {}
+        range_del_agg_(range_del_agg),
+        pinned_iters_mgr_(nullptr) {
+    // Empty level is not supported.
+    assert(flevel_ != nullptr && flevel_->num_files > 0);
+  }

-  InternalIterator* NewSecondaryIterator(const Slice& meta_handle) override {
-    if (meta_handle.size() != sizeof(FileDescriptor)) {
-      return NewErrorInternalIterator(
-          Status::Corruption("FileReader invoked with unexpected value"));
+  virtual ~LevelIterator() { delete file_iter_.Set(nullptr); }
+
+  virtual void Seek(const Slice& target) override;
+  virtual void SeekForPrev(const Slice& target) override;
+  virtual void SeekToFirst() override;
+  virtual void SeekToLast() override;
+  virtual void Next() override;
+  virtual void Prev() override;
+
+  virtual bool Valid() const override { return file_iter_.Valid(); }
+  virtual Slice key() const override {
+    assert(Valid());
+    return file_iter_.key();
+  }
+  virtual Slice value() const override {
+    assert(Valid());
+    return file_iter_.value();
+  }
+  virtual Status status() const override {
+    // It'd be nice if status() returned a const Status& instead of a Status
+    if (!status_.ok()) {
+      return status_;
+    } else if (file_iter_.iter() != nullptr) {
+      return file_iter_.status();
    }
-    const FileDescriptor* fd =
-        reinterpret_cast<const FileDescriptor*>(meta_handle.data());
-    return table_cache_->NewIterator(
-        read_options_, env_options_, icomparator_, *fd, range_del_agg_,
-        nullptr /* don't need reference to table */, file_read_hist_,
-        for_compaction_, nullptr /* arena */, skip_filters_, level_);
+    return Status::OK();
+  }
+  virtual void SetPinnedItersMgr(
+      PinnedIteratorsManager* pinned_iters_mgr) override {
+    pinned_iters_mgr_ = pinned_iters_mgr;
+    if (file_iter_.iter()) {
+      file_iter_.SetPinnedItersMgr(pinned_iters_mgr);
+    }
+  }
+  virtual bool IsKeyPinned() const override {
+    return pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled() &&
+           file_iter_.iter() && file_iter_.IsKeyPinned();
+  }
+  virtual bool IsValuePinned() const override {
+    return pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled() &&
+           file_iter_.iter() && file_iter_.IsValuePinned();
  }

-  bool PrefixMayMatch(const Slice& internal_key) override {
-    return true;
+ private:
+  void SkipEmptyFileForward();
+  void SkipEmptyFileBackward();
+  void SetFileIterator(InternalIterator* iter);
+  void InitFileIterator(size_t new_file_index);
+
+  const Slice& file_smallest_key(size_t file_index) {
+    assert(file_index < flevel_->num_files);
+    return flevel_->files[file_index].smallest_key;
  }

-  bool KeyReachedUpperBound(const Slice& internal_key) override {
+  bool KeyReachedUpperBound(const Slice& internal_key) {
    return read_options_.iterate_upper_bound != nullptr &&
           icomparator_.user_comparator()->Compare(
               ExtractUserKey(internal_key),
               *read_options_.iterate_upper_bound) >= 0;
  }

- private:
+  InternalIterator* NewFileIterator() {
+    assert(file_index_ < flevel_->num_files);
+    auto file_meta = flevel_->files[file_index_];
+    if (should_sample_) {
+      sample_file_read_inc(file_meta.file_metadata);
+    }
+
+    return table_cache_->NewIterator(
+        read_options_, env_options_, icomparator_, file_meta.fd, range_del_agg_,
+        nullptr /* don't need reference to table */, file_read_hist_,
+        for_compaction_, nullptr /* arena */, skip_filters_, level_);
+  }
+
  TableCache* table_cache_;
  const ReadOptions read_options_;
  const EnvOptions& env_options_;
  const InternalKeyComparator& icomparator_;
+  const LevelFilesBrief* flevel_;
+  mutable FileDescriptor current_value_;
+
  HistogramImpl* file_read_hist_;
+  bool should_sample_;
  bool for_compaction_;
  bool skip_filters_;
+  size_t file_index_;
  int level_;
  RangeDelAggregator* range_del_agg_;
+  IteratorWrapper file_iter_;  // May be nullptr
+  PinnedIteratorsManager* pinned_iters_mgr_;
+  Status status_;
 };

+void LevelIterator::Seek(const Slice& target) {
+  size_t new_file_index = FindFile(icomparator_, *flevel_, target);
+
+  InitFileIterator(new_file_index);
+  if (file_iter_.iter() != nullptr) {
+    file_iter_.Seek(target);
+  }
+  SkipEmptyFileForward();
+}
+
+void LevelIterator::SeekForPrev(const Slice& target) {
+  size_t new_file_index = FindFile(icomparator_, *flevel_, target);
+  if (new_file_index >= flevel_->num_files) {
+    new_file_index = flevel_->num_files - 1;
+  }
+
+  InitFileIterator(new_file_index);
+  if (file_iter_.iter() != nullptr) {
+    file_iter_.SeekForPrev(target);
+    SkipEmptyFileBackward();
+  }
+}
+
+void LevelIterator::SeekToFirst() {
+  InitFileIterator(0);
+  if (file_iter_.iter() != nullptr) {
+    file_iter_.SeekToFirst();
+  }
+  SkipEmptyFileForward();
+}
+
+void LevelIterator::SeekToLast() {
+  InitFileIterator(flevel_->num_files - 1);
+  if (file_iter_.iter() != nullptr) {
+    file_iter_.SeekToLast();
+  }
+  SkipEmptyFileBackward();
+}
+
+void LevelIterator::Next() {
+  assert(Valid());
+  file_iter_.Next();
+  SkipEmptyFileForward();
+}
+
+void LevelIterator::Prev() {
+  assert(Valid());
+  file_iter_.Prev();
+  SkipEmptyFileBackward();
+}
+
+void LevelIterator::SkipEmptyFileForward() {
+  // For an error (IO error, checksum mismatch, etc), we skip the file
+  // and move to the next one and continue reading data.
+  // TODO this behavior is from LevelDB. We should revisit it.
+  while (file_iter_.iter() == nullptr ||
+         (!file_iter_.Valid() && !file_iter_.status().IsIncomplete())) {
+    if (file_iter_.iter() != nullptr && !file_iter_.Valid() &&
+        file_iter_.iter()->IsOutOfBound()) {
+      return;
+    }
+
+    // Move to next file
+    if (file_index_ >= flevel_->num_files - 1) {
+      // Already at the last file
+      SetFileIterator(nullptr);
+      return;
+    }
+    if (KeyReachedUpperBound(file_smallest_key(file_index_ + 1))) {
+      SetFileIterator(nullptr);
+      return;
+    }
+    InitFileIterator(file_index_ + 1);
+    if (file_iter_.iter() != nullptr) {
+      file_iter_.SeekToFirst();
+    }
+  }
+}
+
+void LevelIterator::SkipEmptyFileBackward() {
+  while (file_iter_.iter() == nullptr ||
+         (!file_iter_.Valid() && !file_iter_.status().IsIncomplete())) {
+    // Move to previous file
+    if (file_index_ == 0) {
+      // Already the first file
+      SetFileIterator(nullptr);
+      return;
+    }
+    InitFileIterator(file_index_ - 1);
+    if (file_iter_.iter() != nullptr) {
+      file_iter_.SeekToLast();
+    }
+  }
+}
+
+void LevelIterator::SetFileIterator(InternalIterator* iter) {
+  if (file_iter_.iter() != nullptr && status_.ok()) {
+    // TODO right now we don't invalidate the iterator even if the status is
+    // not OK. We should consider to do that so that it is harder for users to
+    // skip errors.
+    status_ = file_iter_.status();
+  }
+
+  if (pinned_iters_mgr_ && iter) {
+    iter->SetPinnedItersMgr(pinned_iters_mgr_);
+  }
+
+  InternalIterator* old_iter = file_iter_.Set(iter);
+  if (pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled()) {
+    pinned_iters_mgr_->PinIterator(old_iter);
+  } else {
+    delete old_iter;
+  }
+}
+
+void LevelIterator::InitFileIterator(size_t new_file_index) {
+  if (new_file_index >= flevel_->num_files) {
+    file_index_ = new_file_index;
+    SetFileIterator(nullptr);
+    return;
+  } else {
+    // If the file iterator shows incomplete, we try it again if users seek
+    // to the same file, as this time we may go to a different data block
+    // which is cached in block cache.
+    //
+    if (file_iter_.iter() != nullptr && !file_iter_.status().IsIncomplete() &&
+        new_file_index == file_index_) {
+      // file_iter_ is already constructed with this iterator, so
+      // no need to change anything
+    } else {
+      file_index_ = new_file_index;
+      InternalIterator* iter = NewFileIterator();
+      SetFileIterator(iter);
+    }
+  }
+}
+
 // A wrapper of version builder which references the current version in
 // constructor and unref it in the destructor.
 // Both of the constructor and destructor need to be called inside DB Mutex.
@@ -854,24 +986,18 @@ void Version::AddIteratorsForLevel(const ReadOptions& read_options,
        sample_file_read_inc(meta);
      }
    }
-  } else {
+  } else if (storage_info_.LevelFilesBrief(level).num_files > 0) {
    // For levels > 0, we can use a concatenating iterator that sequentially
    // walks through the non-overlapping files in the level, opening them
    // lazily.
-    auto* mem = arena->AllocateAligned(sizeof(LevelFileIteratorState));
-    auto* state = new (mem)
-        LevelFileIteratorState(cfd_->table_cache(), read_options, soptions,
-                               cfd_->internal_comparator(),
-                               cfd_->internal_stats()->GetFileReadHist(level),
-                               false /* for_compaction */,
-                               cfd_->ioptions()->prefix_extractor != nullptr,
-                               IsFilterSkipped(level), level, range_del_agg);
-    mem = arena->AllocateAligned(sizeof(LevelFileNumIterator));
-    auto* first_level_iter = new (mem) LevelFileNumIterator(
+    auto* mem = arena->AllocateAligned(sizeof(LevelIterator));
+    merge_iter_builder->AddIterator(new (mem) LevelIterator(
+        cfd_->table_cache(), read_options, soptions,
        cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level),
-        should_sample_file_read());
-    merge_iter_builder->AddIterator(
-        NewTwoLevelIterator(state, first_level_iter, arena, false));
+        should_sample_file_read(),
+        cfd_->internal_stats()->GetFileReadHist(level),
+        false /* for_compaction */, IsFilterSkipped(level), level,
+        range_del_agg));
  }
 }

@@ -980,10 +1106,12 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
      storage_info_.num_non_empty_levels_, &storage_info_.file_indexer_,
      user_comparator(), internal_comparator());
  FdWithKeyRange* f = fp.GetNextFile();
+
  while (f != nullptr) {
    if (get_context.sample()) {
      sample_file_read_inc(f->file_metadata);
    }
+
    *status = table_cache_->Get(
        read_options, *internal_comparator(), f->fd, ikey, &get_context,
        cfd_->internal_stats()->GetFileReadHist(fp.GetHitFileLevel()),
@@ -995,10 +1123,21 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
      return;
    }

+    // report the counters before returning
+    if (get_context.State() != GetContext::kNotFound &&
+        get_context.State() != GetContext::kMerge) {
+      for (uint32_t t = 0; t < Tickers::TICKER_ENUM_MAX; t++) {
+        if (get_context.tickers_value[t] > 0) {
+          RecordTick(db_statistics_, t, get_context.tickers_value[t]);
+        }
+      }
+    }
    switch (get_context.State()) {
      case GetContext::kNotFound:
        // Keep searching in other files
        break;
+      case GetContext::kMerge:
+        break;
      case GetContext::kFound:
        if (fp.GetHitFileLevel() == 0) {
          RecordTick(db_statistics_, GET_HIT_L0);
@@ -1015,8 +1154,6 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
      case GetContext::kCorrupt:
        *status = Status::Corruption("corrupted key for ", user_key);
        return;
-      case GetContext::kMerge:
-        break;
      case GetContext::kBlobIndex:
        ROCKS_LOG_ERROR(info_log_, "Encounter unexpected blob index.");
        *status = Status::NotSupported(
@@ -1027,6 +1164,11 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
    f = fp.GetNextFile();
  }

+  for (uint32_t t = 0; t < Tickers::TICKER_ENUM_MAX; t++) {
+    if (get_context.tickers_value[t] > 0) {
+      RecordTick(db_statistics_, t, get_context.tickers_value[t]);
+    }
+  }
  if (GetContext::kMerge == get_context.State()) {
    if (!merge_operator_) {
      *status =  Status::InvalidArgument(
@@ -1960,7 +2102,7 @@ void VersionStorageInfo::ExtendFileRangeOverlappingInterval(
 #endif
  *start_index = mid_index + 1;
  *end_index = mid_index;
-  int count __attribute__((unused));
+  int count __attribute__((__unused__));
  count = 0;

  // check backwards from 'mid' to lower indices
@@ -3716,17 +3858,13 @@ InternalIterator* VersionSet::MakeInputIterator(
        }
      } else {
        // Create concatenating iterator for the files from this level
-        list[num++] = NewTwoLevelIterator(
-            new LevelFileIteratorState(
-                cfd->table_cache(), read_options, env_options_compactions,
-                cfd->internal_comparator(),
-                nullptr /* no per level latency histogram */,
-                true /* for_compaction */, false /* prefix enabled */,
-                false /* skip_filters */, (int)which /* level */,
-                range_del_agg),
-            new LevelFileNumIterator(cfd->internal_comparator(),
-                                     c->input_levels(which),
-                                     false /* don't sample compaction */));
+        list[num++] = new LevelIterator(
+            cfd->table_cache(), read_options, env_options_compactions,
+            cfd->internal_comparator(), c->input_levels(which),
+            false /* should_sample */,
+            nullptr /* no per level latency histogram */,
+            true /* for_compaction */, false /* skip_filters */,
+            (int)which /* level */, range_del_agg);
      }
    }
  }
@@ -39,6 +39,15 @@ namespace rocksdb {

 #ifndef ROCKSDB_LITE

+Status WalManager::DeleteFile(const std::string& fname, uint64_t number) {
+  auto s = env_->DeleteFile(db_options_.wal_dir + "/" + fname);
+  if (s.ok()) {
+    MutexLock l(&read_first_record_cache_mutex_);
+    read_first_record_cache_.erase(number);
+  }
+  return s;
+}
+
 Status WalManager::GetSortedWalFiles(VectorLogPtr& files) {
  // First get sorted files in db dir, then get sorted files from archived
  // dir, to avoid a race condition where a log file is moved to archived
@@ -49,6 +49,8 @@ class WalManager {

  void ArchiveWALFile(const std::string& fname, uint64_t number);

+  Status DeleteFile(const std::string& fname, uint64_t number);
+
  Status TEST_ReadFirstRecord(const WalFileType type, const uint64_t number,
                              SequenceNumber* sequence) {
    return ReadFirstRecord(type, number, sequence);
@@ -15,10 +15,12 @@
 //    kTypeValue varstring varstring
 //    kTypeDeletion varstring
 //    kTypeSingleDeletion varstring
+//    kTypeRangeDeletion varstring varstring
 //    kTypeMerge varstring varstring
 //    kTypeColumnFamilyValue varint32 varstring varstring
-//    kTypeColumnFamilyDeletion varint32 varstring varstring
-//    kTypeColumnFamilySingleDeletion varint32 varstring varstring
+//    kTypeColumnFamilyDeletion varint32 varstring
+//    kTypeColumnFamilySingleDeletion varint32 varstring
+//    kTypeColumnFamilyRangeDeletion varint32 varstring varstring
 //    kTypeColumnFamilyMerge varint32 varstring varstring
 //    kTypeBeginPrepareXID varstring
 //    kTypeEndPrepareXID
@@ -144,6 +146,12 @@ WriteBatch::WriteBatch(const std::string& rep)
      max_bytes_(0),
      rep_(rep) {}

+WriteBatch::WriteBatch(std::string&& rep)
+    : save_points_(nullptr),
+      content_flags_(ContentFlags::DEFERRED),
+      max_bytes_(0),
+      rep_(std::move(rep)) {}
+
 WriteBatch::WriteBatch(const WriteBatch& src)
    : save_points_(src.save_points_),
      wal_term_point_(src.wal_term_point_),
@@ -394,14 +402,22 @@ Status WriteBatch::Iterate(Handler* handler) const {
  bool empty_batch = true;
  int found = 0;
  Status s;
-  while (s.ok() && !input.empty() && handler->Continue()) {
-    char tag = 0;
-    uint32_t column_family = 0;  // default
+  char tag = 0;
+  uint32_t column_family = 0;  // default
+  while ((s.ok() || UNLIKELY(s.IsTryAgain())) && !input.empty() &&
+         handler->Continue()) {
+    if (LIKELY(!s.IsTryAgain())) {
+      tag = 0;
+      column_family = 0;  // default

-    s = ReadRecordFromWriteBatch(&input, &tag, &column_family, &key, &value,
-                                 &blob, &xid);
-    if (!s.ok()) {
-      return s;
+      s = ReadRecordFromWriteBatch(&input, &tag, &column_family, &key, &value,
+                                   &blob, &xid);
+      if (!s.ok()) {
+        return s;
+      }
+    } else {
+      assert(s.IsTryAgain());
+      s = Status::OK();
    }

    switch (tag) {
@@ -410,47 +426,59 @@ Status WriteBatch::Iterate(Handler* handler) const {
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_PUT));
        s = handler->PutCF(column_family, key, value);
-        empty_batch = false;
-        found++;
+        if (LIKELY(s.ok())) {
+          empty_batch = false;
+          found++;
+        }
        break;
      case kTypeColumnFamilyDeletion:
      case kTypeDeletion:
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_DELETE));
        s = handler->DeleteCF(column_family, key);
-        empty_batch = false;
-        found++;
+        if (LIKELY(s.ok())) {
+          empty_batch = false;
+          found++;
+        }
        break;
      case kTypeColumnFamilySingleDeletion:
      case kTypeSingleDeletion:
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_SINGLE_DELETE));
        s = handler->SingleDeleteCF(column_family, key);
-        empty_batch = false;
-        found++;
+        if (LIKELY(s.ok())) {
+          empty_batch = false;
+          found++;
+        }
        break;
      case kTypeColumnFamilyRangeDeletion:
      case kTypeRangeDeletion:
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_DELETE_RANGE));
        s = handler->DeleteRangeCF(column_family, key, value);
-        empty_batch = false;
-        found++;
+        if (LIKELY(s.ok())) {
+          empty_batch = false;
+          found++;
+        }
        break;
      case kTypeColumnFamilyMerge:
      case kTypeMerge:
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_MERGE));
        s = handler->MergeCF(column_family, key, value);
-        empty_batch = false;
-        found++;
+        if (LIKELY(s.ok())) {
+          empty_batch = false;
+          found++;
+        }
        break;
      case kTypeColumnFamilyBlobIndex:
      case kTypeBlobIndex:
        assert(content_flags_.load(std::memory_order_relaxed) &
               (ContentFlags::DEFERRED | ContentFlags::HAS_BLOB_INDEX));
        s = handler->PutBlobIndexCF(column_family, key, value);
-        found++;
+        if (LIKELY(s.ok())) {
+          found++;
+        }
        break;
      case kTypeLogData:
        handler->LogData(blob);
@@ -547,6 +575,13 @@ size_t WriteBatchInternal::GetFirstOffset(WriteBatch* b) {

 Status WriteBatchInternal::Put(WriteBatch* b, uint32_t column_family_id,
                               const Slice& key, const Slice& value) {
+  if (key.size() > size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("key is too large");
+  }
+  if (value.size() > size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("value is too large");
+  }
+
  LocalSavePoint save(b);
  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
  if (column_family_id == 0) {
@@ -569,8 +604,33 @@ Status WriteBatch::Put(ColumnFamilyHandle* column_family, const Slice& key,
                                 value);
 }

+Status WriteBatchInternal::CheckSlicePartsLength(const SliceParts& key,
+                                                 const SliceParts& value) {
+  size_t total_key_bytes = 0;
+  for (int i = 0; i < key.num_parts; ++i) {
+    total_key_bytes += key.parts[i].size();
+  }
+  if (total_key_bytes >= size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("key is too large");
+  }
+
+  size_t total_value_bytes = 0;
+  for (int i = 0; i < value.num_parts; ++i) {
+    total_value_bytes += value.parts[i].size();
+  }
+  if (total_value_bytes >= size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("value is too large");
+  }
+  return Status::OK();
+}
+
 Status WriteBatchInternal::Put(WriteBatch* b, uint32_t column_family_id,
                               const SliceParts& key, const SliceParts& value) {
+  Status s = CheckSlicePartsLength(key, value);
+  if (!s.ok()) {
+    return s;
+  }
+
  LocalSavePoint save(b);
  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
  if (column_family_id == 0) {
@@ -787,6 +847,13 @@ Status WriteBatch::DeleteRange(ColumnFamilyHandle* column_family,

 Status WriteBatchInternal::Merge(WriteBatch* b, uint32_t column_family_id,
                                 const Slice& key, const Slice& value) {
+  if (key.size() > size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("key is too large");
+  }
+  if (value.size() > size_t{port::kMaxUint32}) {
+    return Status::InvalidArgument("value is too large");
+  }
+
  LocalSavePoint save(b);
  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
  if (column_family_id == 0) {
@@ -812,6 +879,11 @@ Status WriteBatch::Merge(ColumnFamilyHandle* column_family, const Slice& key,
 Status WriteBatchInternal::Merge(WriteBatch* b, uint32_t column_family_id,
                                 const SliceParts& key,
                                 const SliceParts& value) {
+  Status s = CheckSlicePartsLength(key, value);
+  if (!s.ok()) {
+    return s;
+  }
+
  LocalSavePoint save(b);
  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
  if (column_family_id == 0) {
@@ -1076,12 +1148,23 @@ class MemTableInserter : public WriteBatch::Handler {
      MaybeAdvanceSeq();
      return seek_status;
    }
+    Status ret_status;

    MemTable* mem = cf_mems_->GetMemTable();
    auto* moptions = mem->GetImmutableMemTableOptions();
+    // inplace_update_support is inconsistent with snapshots, and therefore with
+    // any kind of transactions including the ones that use seq_per_batch
+    assert(!seq_per_batch_ || !moptions->inplace_update_support);
    if (!moptions->inplace_update_support) {
-      mem->Add(sequence_, value_type, key, value, concurrent_memtable_writes_,
-               get_post_process_info(mem));
+      bool mem_res =
+          mem->Add(sequence_, value_type, key, value,
+                   concurrent_memtable_writes_, get_post_process_info(mem));
+      if (UNLIKELY(!mem_res)) {
+        assert(seq_per_batch_);
+        ret_status = Status::TryAgain("key+seq exists");
+        const bool BATCH_BOUNDRY = true;
+        MaybeAdvanceSeq(BATCH_BOUNDRY);
+      }
    } else if (moptions->inplace_callback == nullptr) {
      assert(!concurrent_memtable_writes_);
      mem->Update(sequence_, key, value);
@@ -1117,11 +1200,17 @@ class MemTableInserter : public WriteBatch::Handler {
                                                 value, &merged_value);
        if (status == UpdateStatus::UPDATED_INPLACE) {
          // prev_value is updated in-place with final value.
-          mem->Add(sequence_, value_type, key, Slice(prev_buffer, prev_size));
+          bool mem_res __attribute__((__unused__));
+          mem_res = mem->Add(
+              sequence_, value_type, key, Slice(prev_buffer, prev_size));
+          assert(mem_res);
          RecordTick(moptions->statistics, NUMBER_KEYS_WRITTEN);
        } else if (status == UpdateStatus::UPDATED) {
          // merged_value contains the final value.
-          mem->Add(sequence_, value_type, key, Slice(merged_value));
+          bool mem_res __attribute__((__unused__));
+          mem_res =
+              mem->Add(sequence_, value_type, key, Slice(merged_value));
+          assert(mem_res);
          RecordTick(moptions->statistics, NUMBER_KEYS_WRITTEN);
        }
      }
@@ -1131,7 +1220,7 @@ class MemTableInserter : public WriteBatch::Handler {
    // in memtable add/update.
    MaybeAdvanceSeq();
    CheckMemtableFull();
-    return Status::OK();
+    return ret_status;
  }

  virtual Status PutCF(uint32_t column_family_id, const Slice& key,
@@ -1141,12 +1230,20 @@ class MemTableInserter : public WriteBatch::Handler {

  Status DeleteImpl(uint32_t column_family_id, const Slice& key,
                    const Slice& value, ValueType delete_type) {
+    Status ret_status;
    MemTable* mem = cf_mems_->GetMemTable();
-    mem->Add(sequence_, delete_type, key, value, concurrent_memtable_writes_,
-             get_post_process_info(mem));
+    bool mem_res =
+        mem->Add(sequence_, delete_type, key, value,
+                 concurrent_memtable_writes_, get_post_process_info(mem));
+    if (UNLIKELY(!mem_res)) {
+      assert(seq_per_batch_);
+      ret_status = Status::TryAgain("key+seq exists");
+      const bool BATCH_BOUNDRY = true;
+      MaybeAdvanceSeq(BATCH_BOUNDRY);
+    }
    MaybeAdvanceSeq();
    CheckMemtableFull();
-    return Status::OK();
+    return ret_status;
  }

  virtual Status DeleteCF(uint32_t column_family_id,
@@ -1238,6 +1335,7 @@ class MemTableInserter : public WriteBatch::Handler {
      return seek_status;
    }

+    Status ret_status;
    MemTable* mem = cf_mems_->GetMemTable();
    auto* moptions = mem->GetImmutableMemTableOptions();
    bool perform_merge = false;
@@ -1293,18 +1391,30 @@ class MemTableInserter : public WriteBatch::Handler {
        perform_merge = false;
      } else {
        // 3) Add value to memtable
-        mem->Add(sequence_, kTypeValue, key, new_value);
+        bool mem_res = mem->Add(sequence_, kTypeValue, key, new_value);
+        if (UNLIKELY(!mem_res)) {
+          assert(seq_per_batch_);
+          ret_status = Status::TryAgain("key+seq exists");
+          const bool BATCH_BOUNDRY = true;
+          MaybeAdvanceSeq(BATCH_BOUNDRY);
+        }
      }
    }

    if (!perform_merge) {
      // Add merge operator to memtable
-      mem->Add(sequence_, kTypeMerge, key, value);
+      bool mem_res = mem->Add(sequence_, kTypeMerge, key, value);
+      if (UNLIKELY(!mem_res)) {
+        assert(seq_per_batch_);
+        ret_status = Status::TryAgain("key+seq exists");
+        const bool BATCH_BOUNDRY = true;
+        MaybeAdvanceSeq(BATCH_BOUNDRY);
+      }
    }

    MaybeAdvanceSeq();
    CheckMemtableFull();
-    return Status::OK();
+    return ret_status;
  }

  virtual Status PutBlobIndexCF(uint32_t column_family_id, const Slice& key,
@@ -1488,6 +1598,8 @@ Status WriteBatchInternal::InsertInto(
    if (!w->status.ok()) {
      return w->status;
    }
+    assert(!seq_per_batch || w->batch_cnt != 0);
+    assert(!seq_per_batch || inserter.sequence() - w->sequence == w->batch_cnt);
  }
  return Status::OK();
 }
@@ -1496,7 +1608,7 @@ Status WriteBatchInternal::InsertInto(
    WriteThread::Writer* writer, SequenceNumber sequence,
    ColumnFamilyMemTables* memtables, FlushScheduler* flush_scheduler,
    bool ignore_missing_column_families, uint64_t log_number, DB* db,
-    bool concurrent_memtable_writes, bool seq_per_batch) {
+    bool concurrent_memtable_writes, bool seq_per_batch, size_t batch_cnt) {
  assert(writer->ShouldWriteToMemtable());
  MemTableInserter inserter(sequence, memtables, flush_scheduler,
                            ignore_missing_column_families, log_number, db,
@@ -1505,6 +1617,8 @@ Status WriteBatchInternal::InsertInto(
  SetSequence(writer->batch, sequence);
  inserter.set_log_number_ref(writer->log_ref);
  Status s = writer->batch->Iterate(&inserter);
+  assert(!seq_per_batch || batch_cnt != 0);
+  assert(!seq_per_batch || inserter.sequence() - sequence == batch_cnt);
  if (concurrent_memtable_writes) {
    inserter.PostProcess();
  }
@@ -138,6 +138,9 @@ class WriteBatchInternal {

  static Status SetContents(WriteBatch* batch, const Slice& contents);

+  static Status CheckSlicePartsLength(const SliceParts& key,
+                                      const SliceParts& value);
+
  // Inserts batches[i] into memtable, for i in 0..num_batches-1 inclusive.
  //
  // If ignore_missing_column_families == true. WriteBatch
@@ -182,7 +185,7 @@ class WriteBatchInternal {
                           bool ignore_missing_column_families = false,
                           uint64_t log_number = 0, DB* db = nullptr,
                           bool concurrent_memtable_writes = false,
-                           bool seq_per_batch = false);
+                           bool seq_per_batch = false, size_t batch_cnt = 0);

  static Status Append(WriteBatch* dst, const WriteBatch* src,
                       const bool WAL_only = false);
@@ -291,7 +291,7 @@ TEST_F(WriteCallbackTest, WriteWithCallbackTest) {
                woptions.disableWAL = !enable_WAL;
                woptions.sync = enable_WAL;
                Status s;
-                if (seq_per_batch && two_queues) {
+                if (seq_per_batch) {
                  class PublishSeqCallback : public PreReleaseCallback {
                   public:
                    PublishSeqCallback(DBImpl* db_impl_in)
@@ -302,9 +302,13 @@ TEST_F(WriteCallbackTest, WriteWithCallbackTest) {
                    }
                    DBImpl* db_impl_;
                  } publish_seq_callback(db_impl);
-                  s = db_impl->WriteImpl(woptions, &write_op.write_batch_,
-                                         &write_op.callback_, nullptr, 0, false,
-                                         nullptr, &publish_seq_callback);
+                  // seq_per_batch requires a natural batch separator or Noop
+                  WriteBatchInternal::InsertNoop(&write_op.write_batch_);
+                  const size_t ONE_BATCH = 1;
+                  s = db_impl->WriteImpl(
+                      woptions, &write_op.write_batch_, &write_op.callback_,
+                      nullptr, 0, false, nullptr, ONE_BATCH,
+                      two_queues ? &publish_seq_callback : nullptr);
                } else {
                  s = db_impl->WriteWithCallback(
                      woptions, &write_op.write_batch_, &write_op.callback_);
@@ -118,6 +118,7 @@ class WriteThread {
    bool no_slowdown;
    bool disable_wal;
    bool disable_memtable;
+    size_t batch_cnt;  // if non-zero, number of sub-batches in the write batch
    PreReleaseCallback* pre_release_callback;
    uint64_t log_used;  // log number that this batch was inserted into
    uint64_t log_ref;   // log number that memtable insert should reference
@@ -128,6 +129,7 @@ class WriteThread {
    SequenceNumber sequence;  // the sequence number to use for the first key
    Status status;            // status of memtable inserter
    Status callback_status;   // status returned by callback->Callback()
+
    std::aligned_storage<sizeof(std::mutex)>::type state_mutex_bytes;
    std::aligned_storage<sizeof(std::condition_variable)>::type state_cv_bytes;
    Writer* link_older;  // read/write only before linking, or as leader
@@ -139,6 +141,7 @@ class WriteThread {
          no_slowdown(false),
          disable_wal(false),
          disable_memtable(false),
+          batch_cnt(0),
          pre_release_callback(nullptr),
          log_used(0),
          log_ref(0),
@@ -152,12 +155,14 @@ class WriteThread {

    Writer(const WriteOptions& write_options, WriteBatch* _batch,
           WriteCallback* _callback, uint64_t _log_ref, bool _disable_memtable,
+           size_t _batch_cnt = 0,
           PreReleaseCallback* _pre_release_callback = nullptr)
        : batch(_batch),
          sync(write_options.sync),
          no_slowdown(write_options.no_slowdown),
          disable_wal(write_options.disableWAL),
          disable_memtable(_disable_memtable),
+          batch_cnt(_batch_cnt),
          pre_release_callback(_pre_release_callback),
          log_used(0),
          log_ref(_log_ref),
@@ -105,13 +105,13 @@ GEM
      rb-fsevent (>= 0.9.3)
      rb-inotify (>= 0.9.7)
    mercenary (0.3.6)
-    mini_portile2 (2.1.0)
+    mini_portile2 (2.3.0)
    minima (2.0.0)
    minitest (5.9.1)
    multipart-post (2.0.0)
    net-dns (0.8.0)
-    nokogiri (1.6.8.1)
-      mini_portile2 (~> 2.1.0)
+    nokogiri (1.8.1)
+      mini_portile2 (~> 2.3.0)
    octokit (4.4.1)
      sawyer (~> 0.7.0, >= 0.5.3)
    pathutil (0.14.0)
@@ -0,0 +1,28 @@
+---
+title: Auto-tuned Rate Limiter
+layout: post
+author: ajkr
+category: blog
+---
+
+### Introduction
+
+Our rate limiter has been hard to configure since users need to pick a value that is low enough to prevent background I/O spikes, which can impact user-visible read/write latencies. Meanwhile, picking too low a value can cause memtables and L0 files to pile up, eventually leading to writes stalling. Tuning the rate limiter has been especially difficult for users whose DB instances have different workloads, or have workloads that vary over time, or commonly both.
+
+To address this, in RocksDB 5.9 we released a dynamic rate limiter that adjusts itself over time according to demand for background I/O. It can be enabled simply by passing `auto_tuned=true` in the `NewGenericRateLimiter()` call. In this case `rate_bytes_per_sec` will indicate the upper-bound of the window within which a rate limit will be picked dynamically. The chosen rate limit will be much lower unless absolutely necessary, so setting this to the device's maximum throughput is a reasonable choice on dedicated hosts.
+
+### Algorithm
+
+We use a simple multiplicative-increase, multiplicative-decrease algorithm. We measure demand for background I/O as the ratio of intervals where the rate limiter is drained. There are low and high watermarks for this ratio, which will trigger a change in rate limit when breached. The rate limit can move within a window bounded by the user-specified upper-bound, and a lower-bound that we derive internally. Users can expect this lower bound to be 1-2 orders of magnitude less than the provided upper-bound (so don't provide INT64_MAX as your upper-bound), although it's subject to change.
+
+### Benchmark Results
+
+Data is ingested at 10MB/s and the rate limiter was created with 1000MB/s as its upper bound. The dynamically chosen rate limit hovers around 125MB/s. The other clustering of points at 50MB/s is due to number of compaction threads being reduced to one when there's no compaction pressure. 
+
+![](/static/images/rate-limiter/write-KBps-series.png)
+
+![](/static/images/rate-limiter/auto-tuned-write-KBps-series.png)
+
+The following graph summarizes the above two time series graphs in CDF form. In particular, notice the p90 - p100 for background write rate are significantly lower with auto-tuned rate limiter enabled.
+
+![](/static/images/rate-limiter/write-KBps-cdf.png)
@@ -0,0 +1,37 @@
+---
+title: WritePrepared Transactions
+layout: post
+author: maysamyabandeh
+category: blog
+---
+
+RocksDB supports both optimistic and pessimistic concurrency controls. The pessimistic transactions make use of locks to provide isolation between the transactions. The default write policy in pessimistic transactions is _WriteCommitted_, which means that the data is written to the DB, i.e., the memtable, only after the transaction is committed. This policy simplified the implementation but came with some limitations in throughput, transaction size, and variety in supported isolation levels. In the below, we explain these in detail and present the other write policies, _WritePrepared_ and _WriteUnprepared_. We then dive into the design of _WritePrepared_ transactions.
+
+> _WritePrepared_ are to be announced as production-ready soon.
+
+### WriteCommitted, Pros and Cons
+
+With _WriteCommitted_ write policy, the data is written to the memtable only after the transaction commits. This greatly simplifies the read path as any data that is read by other transactions can be assumed to be committed. This write policy, however, implies that the writes are buffered in memory in the meanwhile. This makes memory a bottleneck for large transactions. The delay of the commit phase in 2PC (two-phase commit) also becomes noticeable since most of the work, i.e., writing to memtable, is done at the commit phase. When the commit of multiple transactions are done in a serial fashion, such as in 2PC implementation of MySQL, the lengthy commit latency becomes a major contributor to lower throughput. Moreover this write policy cannot provide weaker isolation levels, such as READ UNCOMMITTED, that could potentially provide higher throughput for some applications.
+
+### Alternatives: _WritePrepared_ and _WriteUnprepared_
+
+To tackle the lengthy commit issue, we should do memtable writes at earlier phases of 2PC so that the commit phase become lightweight and fast. 2PC is composed of Write stage, where the transaction `::Put` is invoked, the prepare phase, where `::Prepare` is invoked (upon which the DB promises to commit the transaction if later is requested), and commit phase, where `::Commit` is invoked and the transaction writes become visible to all readers. To make the commit phase lightweight, the memtable write could be done at either `::Prepare` or `::Put` stages, resulting into _WritePrepared_ and _WriteUnprepared_ write policies respectively. The downside is that when another transaction is reading data, it would need a way to tell apart which data is committed, and if they are, whether they are committed before the transaction's start, i.e., in the read snapshot of the transaction. _WritePrepared_ would still have the issue of buffering the data, which makes the memory the bottleneck for large transactions. It however provides a good milestone for transitioning from _WriteCommitted_ to _WriteUnprepared_ write policy. Here we explain the design of _WritePrepared_ policy. We will cover the changes that make the design to also supported _WriteUnprepared_ in an upcoming post.
+
+### _WritePrepared_ in a nutshell
+
+These are the primary design questions that needs to be addressed:
+1) How do we identify the key/values in the DB with transactions that wrote them?
+2) How do we figure if a key/value written by transaction Txn_w is in the read snapshot of the reading transaction Txn_r?
+3) How do we rollback the data written by aborted transactions?
+
+With _WritePrepared_, a transaction still buffers the writes in a write batch object in memory. When 2PC `::Prepare` is called, it writes the in-memory write batch to the WAL (write-ahead log) as well as to the memtable(s) (one memtable per column family); We reuse the existing notion of sequence numbers in RocksDB to tag all the key/values in the same write batch with the same sequence number, `prepare_seq`, which is also used as the identifier for the transaction. At commit time, it writes a commit marker to the WAL, whose sequence number, `commit_seq`, will be used as the commit timestamp of the transaction. Before releasing the commit sequence number to the readers, it stores a mapping from `prepare_seq` to `commit_seq` in an in-memory data structure that we call _CommitCache_. When a transaction reading values from the DB (tagged with `prepare_seq`) it makes use of the _CommitCache_ to figure if `commit_seq` of the value is in its read snapshot. To rollback an aborted transaction, we apply the status before the transaction by making another write that cancels out the writes of the aborted transaction.
+
+The _CommitCache_ is a lock-free data structure that caches the recent commit entries. Looking up the entries in the cache must be enough for almost all th transactions that commit in a timely manner. When evicting the older entries from the cache, it still maintains some other data structures to cover the corner cases for transactions that takes abnormally too long to finish. We will cover them in the design details below.
+
+### Preliminary Results
+The full experimental results are to be reported soon. Here we present the improvement in tps observed in some preliminary experiments with MyRocks:
+* sysbench update-noindex: 25%
+* sysbench read-write: 7.6%
+* linkbench: 3.7%
+
+Learn more [here](https://github.com/facebook/rocksdb/wiki/WritePrepared-Transactions).
@@ -0,0 +1,22 @@
+---
+title: RocksDB 5.10.2 Released!
+layout: post
+author: siying
+category: blog
+---
+
+### Public API Change
+* When running `make` with environment variable `USE_SSE` set and `PORTABLE` unset, will use all machine features available locally. Previously this combination only compiled SSE-related features.
+
+### New Features
+* CRC32C is now using the 3-way pipelined SSE algorithm `crc32c_3way` on supported platforms to improve performance. The system will choose to use this algorithm on supported platforms automatically whenever possible. If PCLMULQDQ is not supported it will fall back to the old Fast_CRC32 algorithm.
+* Provide lifetime hints when writing files on Linux. This reduces hardware write-amp on storage devices supporting multiple streams.
+* Add a DB stat, `NUMBER_ITER_SKIP`, which returns how many internal keys were skipped during iterations (e.g., due to being tombstones or duplicate versions of a key).
+* Add PerfContext counters, `key_lock_wait_count` and `key_lock_wait_time`, which measure the number of times transactions wait on key locks and total amount of time waiting.
+
+### Bug Fixes
+* Fix IOError on WAL write doesn't propagate to write group follower
+* Make iterator invalid on merge error.
+* Fix performance issue in `IngestExternalFile()` affecting databases with large number of SST files.
+* Fix possible corruption to LSM structure when `DeleteFilesInRange()` deletes a subset of files spanned by a `DeleteRange()` marker.
+* Fix DB::Flush() keep waiting after flush finish under certain condition.
@@ -73,9 +73,19 @@ RandomAccessFile::~RandomAccessFile() {
 WritableFile::~WritableFile() {
 }

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

+Status Logger::CloseImpl() { return Status::NotSupported(); }
+
 FileLock::~FileLock() {
 }

@@ -277,6 +277,21 @@ class HdfsLogger : public Logger {
  HdfsWritableFile* file_;
  uint64_t (*gettid_)();  // Return the thread id for the current thread

+  Status HdfsCloseHelper() {
+    ROCKS_LOG_DEBUG(mylog, "[hdfs] HdfsLogger closed %s\n",
+                    file_->getName().c_str());
+    Status s = file_->Close();
+    if (mylog != nullptr && mylog == this) {
+      mylog = nullptr;
+    }
+    return s;
+  }
+
+ protected:
+  virtual Status CloseImpl() override {
+    return HdfsCloseHelper();
+  }
+
 public:
  HdfsLogger(HdfsWritableFile* f, uint64_t (*gettid)())
      : file_(f), gettid_(gettid) {
@@ -285,11 +300,9 @@ class HdfsLogger : public Logger {
  }

  virtual ~HdfsLogger() {
-    ROCKS_LOG_DEBUG(mylog, "[hdfs] HdfsLogger closed %s\n",
-                    file_->getName().c_str());
-    delete file_;
-    if (mylog != nullptr && mylog == this) {
-      mylog = nullptr;
+    if (!closed_) {
+      closed_ = true;
+      HdfsCloseHelper();
    }
  }

@@ -763,7 +763,7 @@ class PosixEnv : public Env {

  virtual Status GetAbsolutePath(const std::string& db_path,
                                 std::string* output_path) override {
-    if (db_path.find('/') == 0) {
+    if (!db_path.empty() && db_path[0] == '/') {
      *output_path = db_path;
      return Status::OK();
    }
@@ -1475,6 +1475,74 @@ TEST_P(EnvPosixTestWithParam, PosixRandomRWFileRandomized) {
  env_->DeleteFile(path);
 }

+class TestEnv : public EnvWrapper {
+  public:
+    explicit TestEnv() : EnvWrapper(Env::Default()),
+                close_count(0) { }
+
+    class TestLogger : public Logger {
+      public:
+        using Logger::Logv;
+        TestLogger(TestEnv *env_ptr) : Logger() { env = env_ptr; }
+        ~TestLogger() {
+          if (!closed_) {
+            CloseHelper();
+          }
+        }
+        virtual void Logv(const char *format, va_list ap) override { };
+      protected:
+        virtual Status CloseImpl() override {
+          return CloseHelper();
+        }
+      private:
+        Status CloseHelper() {
+          env->CloseCountInc();;
+          return Status::OK();
+        }
+        TestEnv *env;
+    };
+
+    void CloseCountInc() { close_count++; }
+
+    int GetCloseCount() { return close_count; }
+
+    virtual Status NewLogger(const std::string& fname,
+                             shared_ptr<Logger>* result) {
+      result->reset(new TestLogger(this));
+      return Status::OK();
+    }
+
+  private:
+    int close_count;
+};
+
+class EnvTest : public testing::Test {
+};
+
+TEST_F(EnvTest, Close) {
+  TestEnv *env = new TestEnv();
+  std::shared_ptr<Logger> logger;
+  Status s;
+
+  s = env->NewLogger("", &logger);
+  ASSERT_EQ(s, Status::OK());
+  logger.get()->Close();
+  ASSERT_EQ(env->GetCloseCount(), 1);
+  // Call Close() again. CloseHelper() should not be called again
+  logger.get()->Close();
+  ASSERT_EQ(env->GetCloseCount(), 1);
+  logger.reset();
+  ASSERT_EQ(env->GetCloseCount(), 1);
+
+  s = env->NewLogger("", &logger);
+  ASSERT_EQ(s, Status::OK());
+  logger.reset();
+  ASSERT_EQ(env->GetCloseCount(), 2);
+
+  delete env;
+}
+
+
 INSTANTIATE_TEST_CASE_P(DefaultEnvWithoutDirectIO, EnvPosixTestWithParam,
                        ::testing::Values(std::pair<Env*, bool>(Env::Default(),
                                                                false)));
@@ -799,7 +799,7 @@ Status PosixWritableFile::Close() {
    // trim the extra space preallocated at the end of the file
    // NOTE(ljin): we probably don't want to surface failure as an IOError,
    // but it will be nice to log these errors.
-    int dummy __attribute__((unused));
+    int dummy __attribute__((__unused__));
    dummy = ftruncate(fd_, filesize_);
 #if defined(ROCKSDB_FALLOCATE_PRESENT) && !defined(TRAVIS)
    // in some file systems, ftruncate only trims trailing space if the
@@ -24,6 +24,7 @@
 #endif

 #include <atomic>
+#include "env/io_posix.h"
 #include "monitoring/iostats_context_imp.h"
 #include "rocksdb/env.h"
 #include "util/sync_point.h"
@@ -32,6 +33,15 @@ namespace rocksdb {

 class PosixLogger : public Logger {
 private:
+  Status PosixCloseHelper() {
+    int ret;
+
+    ret = fclose(file_);
+    if (ret) {
+      return IOError("Unable to close log file", "", ret);
+    }
+    return Status::OK();
+  }
  FILE* file_;
  uint64_t (*gettid_)();  // Return the thread id for the current thread
  std::atomic_size_t log_size_;
@@ -40,6 +50,12 @@ class PosixLogger : public Logger {
  std::atomic_uint_fast64_t last_flush_micros_;
  Env* env_;
  std::atomic<bool> flush_pending_;
+
+ protected:
+  virtual Status CloseImpl() override {
+    return PosixCloseHelper();
+  }
+
 public:
  PosixLogger(FILE* f, uint64_t (*gettid)(), Env* env,
              const InfoLogLevel log_level = InfoLogLevel::ERROR_LEVEL)
@@ -52,7 +68,10 @@ class PosixLogger : public Logger {
        env_(env),
        flush_pending_(false) {}
  virtual ~PosixLogger() {
-    fclose(file_);
+    if (!closed_) {
+      closed_ = true;
+      PosixCloseHelper();
+    }
  }
  virtual void Flush() override {
    TEST_SYNC_POINT("PosixLogger::Flush:Begin1");
@@ -1,3 +1,4 @@
+# shellcheck disable=SC2148
 export USE_HDFS=1
 export LD_LIBRARY_PATH=$JAVA_HOME/jre/lib/amd64/server:$JAVA_HOME/jre/lib/amd64:/usr/lib/hadoop/lib/native