Fix build on MSVC

Summary:
Fixes some RocksJava regressions recently introduced, whereby RocksJava would not build on JDK 7.
These should have been visible on Travis-CI!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4768

Differential Revision: D13418173

Pulled By: sagar0

fbshipit-source-id: 57bf223188887f84d9e072031af2e0d2c8a69c30

.travis.yml

@@ -71,7 +71,10 @@ install:
       CC=gcc-8 && CXX=g++-8;
     fi
   - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
-      mkdir cmake-dist && curl -sfSL https://cmake.org/files/v3.8/cmake-3.8.1-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
+      mkdir cmake-dist && curl --silent --fail --show-error --location https://github.com/Kitware/CMake/releases/download/v3.14.5/cmake-3.14.5-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
+    fi
+  - if [[ "${JOB_NAME}" == java_test ]]; then
+      java -version && echo "JAVA_HOME=${JAVA_HOME}";
     fi
 
 before_script:

CMakeLists.txt

@@ -504,7 +504,6 @@ set(SOURCES
         db/merge_helper.cc
         db/merge_operator.cc
         db/range_del_aggregator.cc
-        db/range_del_aggregator_v2.cc
         db/range_tombstone_fragmenter.cc
         db/repair.cc
         db/snapshot_impl.cc
@@ -907,7 +906,6 @@ if(WITH_TESTS)
         db/plain_table_db_test.cc
         db/prefix_test.cc
         db/range_del_aggregator_test.cc
-        db/range_del_aggregator_v2_test.cc
         db/range_tombstone_fragmenter_test.cc
         db/repair_test.cc
         db/table_properties_collector_test.cc

HISTORY.md

@@ -1,11 +1,17 @@
 # Rocksdb Change Log
-## Unreleased
+## 5.18.4 (3/3/2020)
-### New Features
+* Various fixes for ARM64 support (#6250)
+* Fix JEMALLOC_CXX_THROW macro missing from older Jemalloc versions, causing build failures on some platforms.
+
+## 5.18.3 (2/11/2019)
+### Bug Fixes
+* Fix possible LSM corruption when both range deletions and subcompactions are used. The symptom of this corruption is L1+ files overlapping in the user key space.
+
+## 5.18.2 (01/31/2019)
 
 ### Public API Change
-
+* Change time resolution in FileOperationInfo.
-### Bug Fixes
+* Deleting Blob files also go through SStFileManager.
-* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
 
 ## 5.18.0 (11/30/2018)
 ### New Features
@@ -18,6 +24,7 @@
 * Add xxhash64 checksum support
 * Introduced `MemoryAllocator`, which lets the user specify custom memory allocator for block based table.
 * Improved `DeleteRange` to prevent read performance degradation. The feature is no longer marked as experimental.
+* Enabled checkpoint on readonly db (DBImplReadOnly).
 
 ### Public API Change
 * `DBOptions::use_direct_reads` now affects reads issued by `BackupEngine` on the database's SSTs.
@@ -34,6 +41,8 @@
 * Fixed Get correctness bug in the presence of range tombstones where merge operands covered by a range tombstone always result in NotFound.
 * Start populating `NO_FILE_CLOSES` ticker statistic, which was always zero previously.
 * The default value of NewBloomFilterPolicy()'s argument use_block_based_builder is changed to false. Note that this new default may cause large temp memory usage when building very large SST files.
+* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
+* Make DB ignore dropped column families while committing results of atomic flush.
 
 ## 5.17.0 (10/05/2018)
 ### Public API Change

INSTALL.md

@@ -43,6 +43,8 @@ to build a portable binary, add `PORTABLE=1` before your make commands, like thi
       command line flags processing. You can compile rocksdb library even
       if you don't have gflags installed.
 
+* If you wish to build the RocksJava static target, then cmake is required for building Snappy.
+
 ## Supported platforms
 
 * **Linux - Ubuntu**

Makefile

@@ -137,6 +137,12 @@ CFLAGS +=  -DHAVE_POWER8
 HAVE_POWER8=1
 endif
 
+ifeq (,$(shell $(CXX) -fsyntax-only -march=armv8-a+crc -xc /dev/null 2>&1))
+CXXFLAGS += -march=armv8-a+crc
+CFLAGS += -march=armv8-a+crc
+ARMCRC_SOURCE=1
+endif
+
 # if we're compiling for release, compile without debug code (-DNDEBUG)
 ifeq ($(DEBUG_LEVEL),0)
 OPT += -DNDEBUG
@@ -543,7 +549,6 @@ TESTS = \
 	persistent_cache_test \
 	statistics_test \
 	lua_test \
-	range_del_aggregator_test \
 	lru_cache_test \
 	object_registry_test \
 	repair_test \
@@ -554,7 +559,7 @@ TESTS = \
 	trace_analyzer_test \
 	repeatable_thread_test \
 	range_tombstone_fragmenter_test \
-	range_del_aggregator_v2_test \
+	range_del_aggregator_test \
 	sst_file_reader_test \
 
 PARALLEL_TEST = \
@@ -1588,9 +1593,6 @@ repeatable_thread_test: util/repeatable_thread_test.o $(LIBOBJECTS) $(TESTHARNES
 range_tombstone_fragmenter_test: db/range_tombstone_fragmenter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-range_del_aggregator_v2_test: db/range_del_aggregator_v2_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 sst_file_reader_test: table/sst_file_reader_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1639,7 +1641,7 @@ JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
 ifeq ($(PLATFORM), OS_SOLARIS)
 	ARCH := $(shell isainfo -b)
 else ifeq ($(PLATFORM), OS_OPENBSD)
-	ifneq (,$(filter $(MACHINE), amd64 arm64 sparc64))
+	ifneq (,$(filter amd64 ppc64 ppc64le arm64 aarch64 sparc64, $(MACHINE)))
 		ARCH := 64
 	else
 		ARCH := 32
@@ -1648,10 +1650,10 @@ else
 	ARCH := $(shell getconf LONG_BIT)
 endif
 
-ifeq (,$(findstring ppc,$(MACHINE)))
+ifneq (,$(filter ppc% arm64 aarch64 sparc64, $(MACHINE)))
-        ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
-else
 	ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE).so
+else
+	ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
 endif
 ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux$(ARCH).jar
 ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
@@ -1664,15 +1666,15 @@ ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.6
 BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-BZIP2_DOWNLOAD_BASE ?= https://web.archive.org/web/20180624184835/http://www.bzip.org
+BZIP2_DOWNLOAD_BASE ?= https://downloads.sourceforge.net/project/bzip2
-SNAPPY_VER ?= 1.1.4
+SNAPPY_VER ?= 1.1.8
-SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
+SNAPPY_SHA256 ?= 16b677f07832a612b0836178db7f374e414f94657c138e6993cbfc5dcc58651f
-SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
+SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
-LZ4_VER ?= 1.8.0
+LZ4_VER ?= 1.9.2
-LZ4_SHA256 ?= 2ca482ea7a9bb103603108b5a7510b7592b90158c151ff50a28f1ca8389fccf6
+LZ4_SHA256 ?= 658ba6191fa44c92280d4aa2c271b0f4fbc0e34d249578dd05e50e76d0e5efcc
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.3.3
+ZSTD_VER ?= 1.4.4
-ZSTD_SHA256 ?= a77c47153ee7de02626c5b2a097005786b71688be61e9fb81806a011f90b297b
+ZSTD_SHA256 ?= a364f5162c7d1a455cc915e8e3cf5f4bd8b75d09bc0f53965b0c9ca1383c52c8
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
 CURL_SSL_OPTS ?= --tlsv1
 
@@ -1711,7 +1713,9 @@ endif
 
 libz.a:
 	-rm -rf zlib-$(ZLIB_VER)
-	curl -O -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+ifeq (,$(wildcard ./zlib-$(ZLIB_VER).tar.gz))
+	curl --fail --output zlib-$(ZLIB_VER).tar.gz --location ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+endif
 	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
 		echo zlib-$(ZLIB_VER).tar.gz checksum mismatch, expected=\"$(ZLIB_SHA256)\" actual=\"$$ZLIB_SHA256_ACTUAL\"; \
@@ -1723,7 +1727,9 @@ libz.a:
 
 libbz2.a:
 	-rm -rf bzip2-$(BZIP2_VER)
-	curl -O -L ${BZIP2_DOWNLOAD_BASE}/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
+ifeq (,$(wildcard ./bzip2-$(BZIP2_VER).tar.gz))
+	curl --fail --output bzip2-$(BZIP2_VER).tar.gz --location ${CURL_SSL_OPTS} ${BZIP2_DOWNLOAD_BASE}/bzip2-$(BZIP2_VER).tar.gz
+endif
 	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
 		echo bzip2-$(BZIP2_VER).tar.gz checksum mismatch, expected=\"$(BZIP2_SHA256)\" actual=\"$$BZIP2_SHA256_ACTUAL\"; \
@@ -1735,21 +1741,24 @@ libbz2.a:
 
 libsnappy.a:
 	-rm -rf snappy-$(SNAPPY_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER)/snappy-$(SNAPPY_VER).tar.gz
+ifeq (,$(wildcard ./snappy-$(SNAPPY_VER).tar.gz))
+	curl --fail --output snappy-$(SNAPPY_VER).tar.gz --location ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
+endif
 	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
 		echo snappy-$(SNAPPY_VER).tar.gz checksum mismatch, expected=\"$(SNAPPY_SHA256)\" actual=\"$$SNAPPY_SHA256_ACTUAL\"; \
 		exit 1; \
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
-	cd snappy-$(SNAPPY_VER) && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --with-pic --enable-static --disable-shared
+	mkdir snappy-$(SNAPPY_VER)/build
-	cd snappy-$(SNAPPY_VER) && $(MAKE) ${SNAPPY_MAKE_TARGET}
+	cd snappy-$(SNAPPY_VER)/build && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' cmake -DCMAKE_POSITION_INDEPENDENT_CODE=ON .. && $(MAKE) ${SNAPPY_MAKE_TARGET}
-	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .
+	cp snappy-$(SNAPPY_VER)/build/libsnappy.a .
 
 liblz4.a:
 	-rm -rf lz4-$(LZ4_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
+ifeq (,$(wildcard ./lz4-$(LZ4_VER).tar.gz))
-	mv v$(LZ4_VER).tar.gz lz4-$(LZ4_VER).tar.gz
+	curl --fail --output lz4-$(LZ4_VER).tar.gz --location ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
+endif
 	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
 		echo lz4-$(LZ4_VER).tar.gz checksum mismatch, expected=\"$(LZ4_SHA256)\" actual=\"$$LZ4_SHA256_ACTUAL\"; \
@@ -1761,8 +1770,9 @@ liblz4.a:
 
 libzstd.a:
 	-rm -rf zstd-$(ZSTD_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
+ifeq (,$(wildcard ./zstd-$(ZSTD_VER).tar.gz))
-	mv v$(ZSTD_VER).tar.gz zstd-$(ZSTD_VER).tar.gz
+	curl --fail --output zstd-$(ZSTD_VER).tar.gz --location ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
+endif
 	ZSTD_SHA256_ACTUAL=`$(SHA256_CMD) zstd-$(ZSTD_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZSTD_SHA256)" != "$$ZSTD_SHA256_ACTUAL" ]; then \
 		echo zstd-$(ZSTD_VER).tar.gz checksum mismatch, expected=\"$(ZSTD_SHA256)\" actual=\"$$ZSTD_SHA256_ACTUAL\"; \
@@ -1853,6 +1863,14 @@ rocksdbjavastaticdockerppc64le:
 	fi
 	docker start -a rocksdb_linux_ppc64le-be
 
+rocksdbjavastaticdockerarm64v8:
+	mkdir -p java/target
+	DOCKER_LINUX_ARM64V8_CONTAINER=`docker ps -aqf name=rocksdb_linux_arm64v8-be`; \
+	if [ -z "$$DOCKER_LINUX_ARM64V8_CONTAINER" ]; then \
+		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_arm64v8-be evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
+	fi
+	docker start -a rocksdb_linux_arm64v8-be
+
 rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral
 
 rocksdbjavastaticpublishdocker: rocksdbjavastaticreleasedocker rocksdbjavastaticpublishcentral

TARGETS

@@ -67,13 +67,11 @@ is_opt_mode = build_mode.startswith("opt")
 if is_opt_mode:
     rocksdb_compiler_flags.append("-DNDEBUG")
 
-default_allocator = read_config("fbcode", "default_allocator")
-
 sanitizer = read_config("fbcode", "sanitizer")
 
-# Let RocksDB aware of jemalloc existence.
+# Do not enable jemalloc if sanitizer presents. RocksDB will further detect
-# Do not enable it if sanitizer presents.
+# whether the binary is linked with jemalloc at runtime.
-if is_opt_mode and default_allocator.startswith("jemalloc") and sanitizer == "":
+if sanitizer == "":
     rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
     rocksdb_external_deps.append(("jemalloc", None, "headers"))
 
@@ -124,7 +122,6 @@ cpp_library(
         "db/merge_helper.cc",
         "db/merge_operator.cc",
         "db/range_del_aggregator.cc",
-        "db/range_del_aggregator_v2.cc",
         "db/range_tombstone_fragmenter.cc",
         "db/repair.cc",
         "db/snapshot_impl.cc",
@@ -935,11 +932,6 @@ ROCKS_TESTS = [
         "db/range_del_aggregator_test.cc",
         "serial",
     ],
-    [
-        "range_del_aggregator_v2_test",
-        "db/range_del_aggregator_v2_test.cc",
-        "serial",
-    ],
     [
         "range_tombstone_fragmenter_test",
         "db/range_tombstone_fragmenter_test.cc",

appveyor.yml

@@ -1,15 +1,75 @@
 version: 1.0.{build}
+
 image: Visual Studio 2017
+
+environment:
+  JAVA_HOME: C:\Program Files\Java\jdk1.8.0
+  THIRDPARTY_HOME: $(APPVEYOR_BUILD_FOLDER)\thirdparty
+  SNAPPY_HOME: $(THIRDPARTY_HOME)\snappy-1.1.7
+  SNAPPY_INCLUDE: $(SNAPPY_HOME);$(SNAPPY_HOME)\build
+  SNAPPY_LIB_DEBUG: $(SNAPPY_HOME)\build\Debug\snappy.lib
+  SNAPPY_LIB_RELEASE: $(SNAPPY_HOME)\build\Release\snappy.lib
+  LZ4_HOME: $(THIRDPARTY_HOME)\lz4-1.8.3
+  LZ4_INCLUDE: $(LZ4_HOME)\lib
+  LZ4_LIB_DEBUG: $(LZ4_HOME)\visual\VS2010\bin\x64_Debug\liblz4_static.lib
+  LZ4_LIB_RELEASE: $(LZ4_HOME)\visual\VS2010\bin\x64_Release\liblz4_static.lib
+  ZSTD_HOME: $(THIRDPARTY_HOME)\zstd-1.4.0
+  ZSTD_INCLUDE: $(ZSTD_HOME)\lib;$(ZSTD_HOME)\lib\dictBuilder
+  ZSTD_LIB_DEBUG: $(ZSTD_HOME)\build\VS2010\bin\x64_Debug\libzstd_static.lib
+  ZSTD_LIB_RELEASE: $(ZSTD_HOME)\build\VS2010\bin\x64_Release\libzstd_static.lib
+  matrix:
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
+      CMAKE_GENERATOR: Visual Studio 14 Win64
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio 14.0\Common7\IDE\devenv.com
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
+      CMAKE_GENERATOR: Visual Studio 15 Win64
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\devenv.com
+
+install:
+  - md %THIRDPARTY_HOME%
+  - echo "Building Snappy dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output snappy-1.1.7.zip --location https://github.com/google/snappy/archive/1.1.7.zip
+  - unzip snappy-1.1.7.zip
+  - cd snappy-1.1.7
+  - mkdir build
+  - cd build
+  - cmake -G "%CMAKE_GENERATOR%" ..
+  - msbuild Snappy.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild Snappy.sln /p:Configuration=Release /p:Platform=x64
+  - echo "Building LZ4 dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output lz4-1.8.3.zip --location https://github.com/lz4/lz4/archive/v1.8.3.zip
+  - unzip lz4-1.8.3.zip
+  - cd lz4-1.8.3\visual\VS2010
+  - ps: $CMD="$Env:DEV_ENV"; & $CMD lz4.sln /upgrade
+  - msbuild lz4.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild lz4.sln /p:Configuration=Release /p:Platform=x64
+  - echo "Building ZStd dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output zstd-1.4.0.zip --location https://github.com/facebook/zstd/archive/v1.4.0.zip
+  - unzip zstd-1.4.0.zip
+  - cd zstd-1.4.0\build\VS2010
+  - ps: $CMD="$Env:DEV_ENV"; & $CMD zstd.sln /upgrade
+  - msbuild zstd.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild zstd.sln /p:Configuration=Release /p:Platform=x64
+
 before_build:
-- md %APPVEYOR_BUILD_FOLDER%\build
+  - md %APPVEYOR_BUILD_FOLDER%\build
-- cd %APPVEYOR_BUILD_FOLDER%\build
+  - cd %APPVEYOR_BUILD_FOLDER%\build
-- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 ..
+  - cmake -G "%CMAKE_GENERATOR%" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DLZ4=1 -DZSTD=1 -DXPRESS=1 -DJNI=1 ..
-- cd ..
+  - cd ..
+
 build:
   project: build\rocksdb.sln
   parallel: true
   verbosity: normal
-test:
+
-test_script:
+test:
-- ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test -Concurrency 8
+
+test_script:
+  - ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test,db_merge_operand_test -Concurrency 8
+
+on_failure:
+  - cmd: 7z a build-failed.zip %APPVEYOR_BUILD_FOLDER%\build\ && appveyor PushArtifact build-failed.zip
 

buckifier/targets_cfg.py

@@ -71,13 +71,11 @@ is_opt_mode = build_mode.startswith("opt")
 if is_opt_mode:
     rocksdb_compiler_flags.append("-DNDEBUG")
 
-default_allocator = read_config("fbcode", "default_allocator")
-
 sanitizer = read_config("fbcode", "sanitizer")
 
-# Let RocksDB aware of jemalloc existence.
+# Do not enable jemalloc if sanitizer presents. RocksDB will further detect
-# Do not enable it if sanitizer presents.
+# whether the binary is linked with jemalloc at runtime.
-if is_opt_mode and default_allocator.startswith("jemalloc") and sanitizer == "":
+if sanitizer == "":
     rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
     rocksdb_external_deps.append(("jemalloc", None, "headers"))
 """

build_tools/build_detect_platform

@@ -53,11 +53,13 @@ if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
     FBCODE_BUILD="true"
     # If we're compiling with TSAN we need pic build
     PIC_BUILD=$COMPILE_WITH_TSAN
-    if [ -z "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
+    if [ -n "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
-      source "$PWD/build_tools/fbcode_config.sh"
-    else
       # we need this to build with MySQL. Don't use for other purposes.
       source "$PWD/build_tools/fbcode_config4.8.1.sh"
+    elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_5xx" ]; then
+      source "$PWD/build_tools/fbcode_config.sh"
+    else
+      source "$PWD/build_tools/fbcode_config_platform007.sh"
     fi
 fi
 
@@ -525,6 +527,8 @@ if test -z "$PORTABLE"; then
   elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
     # TODO: Handle this with approprite options.
     COMMON_FLAGS="$COMMON_FLAGS"
+  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^aarch64`"; then
+    COMMON_FLAGS="$COMMON_FLAGS"
   elif [ "$TARGET_OS" == "IOS" ]; then
     COMMON_FLAGS="$COMMON_FLAGS"
   elif [ "$TARGET_OS" != "AIX" ] && [ "$TARGET_OS" != "SunOS" ]; then

build_tools/dependencies_platform007.sh

@@ -0,0 +1,18 @@
+GCC_BASE=/mnt/gvfs/third-party2/gcc/6e8e715624fd15256a7970073387793dfcf79b46/7.x/centos7-native/b2ef2b6
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/ef37e1faa1c29782abfac1ae65a291b9b7966f6d/stable/centos7-native/c9f9104
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/c67031f0f739ac61575a061518d6ef5038f99f90/7.x/platform007/5620abc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/60d6f124a78798b73944f5ba87c2306ae3460153/2.26/platform007/f259413
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/22c2d65676fb7c23cfa797c4f6937f38b026f3cf/1.2.8/platform007/ca4da3d
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/907b498203d297947f3bb70b9466f47e100f1873/r131/platform007/ca4da3d
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/3ee276cbacfad3074e3f07bf826ac47f06970f4e/1.3.5/platform007/15a3614
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/9c910d36d6235cc40e8ff559358f1833452300ca/master/platform007/5b0f53e
+NUMA_BASE=/mnt/gvfs/third-party2/numa/9cbf2460284c669ed19c3ccb200a71f7dd7e53c7/2.0.11/platform007/ca4da3d
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/bf3d7497fe4e6d007354f0adffa16ce3003f8338/1.3/platform007/6f3e0a9
+TBB_BASE=/mnt/gvfs/third-party2/tbb/ff4e0b093534704d8abab678a4fd7f5ea7b094c7/2018_U5/platform007/ca4da3d
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/b5c4a61a5c483ba24722005ae07895971a2ac707/fb/platform007/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/92ff90349e2f43ea0a8246d8b1cf17b6869013e3/2.29.1/centos7-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/f3f697a28122e6bcd513273dd9c1ff23852fc59f/3.13.0/platform007/ca4da3d
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832

build_tools/fbcode_config_platform007.sh

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

build_tools/regression_build_test.sh

@@ -377,7 +377,7 @@ function send_to_ods {
     echo >&2 "ERROR: Key $key doesn't have a value."
     return
   fi
-  curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
+  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
     --connect-timeout 60
 }
 

build_tools/rocksdb-lego-determinator

@@ -85,7 +85,9 @@ NON_SHM="TMPD=/tmp/rocksdb_test_tmp"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-TSAN="COMPILE_WITH_TSAN=1"
+# in gcc-5 there are known problems with TSAN like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71090.
+# using platform007 gives us gcc-8 or higher which has that bug fixed.
+TSAN="ROCKSDB_FBCODE_BUILD_WITH_PLATFORM007=1 COMPILE_WITH_TSAN=1"
 UBSAN="COMPILE_WITH_UBSAN=1"
 TSAN_CRASH='CRASH_TEST_EXT_ARGS="--compression_type=zstd --log2_keys_per_lock=22"'
 NON_TSAN_CRASH="CRASH_TEST_EXT_ARGS=--compression_type=zstd"
@@ -644,7 +646,7 @@ run_regression()
 
   # parameters: $1 -- key, $2 -- value
   function send_size_to_ods {
-    curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
+    curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
       --connect-timeout 60
   }
 

build_tools/update_dependencies.sh

@@ -53,6 +53,45 @@ function get_lib_base()
   log_variable $__res_var
 }
 
+###########################################################
+#                platform007 dependencies                 #
+###########################################################
+
+OUTPUT="$BASEDIR/dependencies_platform007.sh"
+
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+echo "Writing dependencies to $OUTPUT"
+
+# Compilers locations
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/7.x/centos7-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos7-native/*/`
+
+log_variable GCC_BASE
+log_variable CLANG_BASE
+
+# Libraries locations
+get_lib_base libgcc     7.x     platform007
+get_lib_base glibc      2.26    platform007
+get_lib_base snappy     LATEST  platform007
+get_lib_base zlib       LATEST  platform007
+get_lib_base bzip2      LATEST  platform007
+get_lib_base lz4        LATEST  platform007
+get_lib_base zstd       LATEST  platform007
+get_lib_base gflags     LATEST  platform007
+get_lib_base jemalloc   LATEST  platform007
+get_lib_base numa       LATEST  platform007
+get_lib_base libunwind  LATEST  platform007
+get_lib_base tbb        LATEST  platform007
+
+get_lib_base kernel-headers fb platform007
+get_lib_base binutils   LATEST centos7-native
+get_lib_base valgrind   LATEST platform007
+get_lib_base lua        5.3.4  platform007
+
+git diff $OUTPUT
+
 ###########################################################
 #                   5.x dependencies                      #
 ###########################################################

db/builder.cc

@@ -18,6 +18,7 @@
 #include "db/event_helpers.h"
 #include "db/internal_stats.h"
 #include "db/merge_helper.h"
+#include "db/range_del_aggregator.h"
 #include "db/table_cache.h"
 #include "db/version_edit.h"
 #include "monitoring/iostats_context_imp.h"
@@ -65,8 +66,9 @@ Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
     const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
     TableCache* table_cache, InternalIterator* iter,
-    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
-    const InternalKeyComparator& internal_comparator,
+        range_del_iters,
+    FileMetaData* meta, const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
     uint32_t column_family_id, const std::string& column_family_name,
@@ -86,12 +88,10 @@ Status BuildTable(
   Status s;
   meta->fd.file_size = 0;
   iter->SeekToFirst();
-  std::unique_ptr<RangeDelAggregator> range_del_agg(
+  std::unique_ptr<CompactionRangeDelAggregator> range_del_agg(
-      new RangeDelAggregator(internal_comparator, snapshots));
+      new CompactionRangeDelAggregator(&internal_comparator, snapshots));
-  s = range_del_agg->AddTombstones(std::move(range_del_iter));
+  for (auto& range_del_iter : range_del_iters) {
-  if (!s.ok()) {
+    range_del_agg->AddTombstones(std::move(range_del_iter));
-    // may be non-ok if a range tombstone key is unparsable
-    return s;
   }
 
   std::string fname = TableFileName(ioptions.cf_paths, meta->fd.GetNumber(),
@@ -158,8 +158,10 @@ Status BuildTable(
       }
     }
 
-    for (auto it = range_del_agg->NewIterator(); it->Valid(); it->Next()) {
+    auto range_del_it = range_del_agg->NewIterator();
-      auto tombstone = it->Tombstone();
+    for (range_del_it->SeekToFirst(); range_del_it->Valid();
+         range_del_it->Next()) {
+      auto tombstone = range_del_it->Tombstone();
       auto kv = tombstone.Serialize();
       builder->Add(kv.first.Encode(), kv.second);
       meta->UpdateBoundariesForRange(kv.first, tombstone.SerializeEndKey(),

db/builder.h

@@ -9,6 +9,7 @@
 #include <string>
 #include <utility>
 #include <vector>
+#include "db/range_tombstone_fragmenter.h"
 #include "db/table_properties_collector.h"
 #include "options/cf_options.h"
 #include "rocksdb/comparator.h"
@@ -65,8 +66,9 @@ extern Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& options,
     const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
     TableCache* table_cache, InternalIterator* iter,
-    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
-    const InternalKeyComparator& internal_comparator,
+        range_del_iters,
+    FileMetaData* meta, const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
     uint32_t column_family_id, const std::string& column_family_name,

db/column_family.cc

@@ -25,7 +25,7 @@
 #include "db/db_impl.h"
 #include "db/internal_stats.h"
 #include "db/job_context.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "db/table_properties_collector.h"
 #include "db/version_set.h"
 #include "db/write_controller.h"
@@ -945,7 +945,7 @@ Status ColumnFamilyData::RangesOverlapWithMemtables(
   ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
 
   auto read_seq = super_version->current->version_set()->LastSequence();
-  RangeDelAggregatorV2 range_del_agg(&internal_comparator_, read_seq);
+  ReadRangeDelAggregator range_del_agg(&internal_comparator_, read_seq);
   auto* active_range_del_iter =
       super_version->mem->NewRangeTombstoneIterator(read_opts, read_seq);
   range_del_agg.AddTombstones(

db/compacted_db_impl.h

@@ -67,10 +67,11 @@ class CompactedDBImpl : public DBImpl {
   virtual Status EnableFileDeletions(bool /*force*/) override {
     return Status::NotSupported("Not supported in compacted db mode.");
   }
-  virtual Status GetLiveFiles(std::vector<std::string>&,
+  virtual Status GetLiveFiles(std::vector<std::string>& ret,
-                              uint64_t* /*manifest_file_size*/,
+                              uint64_t* manifest_file_size,
-                              bool /*flush_memtable*/ = true) override {
+                              bool /*flush_memtable*/) override {
-    return Status::NotSupported("Not supported in compacted db mode.");
+    return DBImpl::GetLiveFiles(ret, manifest_file_size,
+                                false /* flush_memtable */);
   }
   using DBImpl::Flush;
   virtual Status Flush(const FlushOptions& /*options*/,

db/compaction_iterator.cc

@@ -18,7 +18,7 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, Env* env,
     bool report_detailed_time, bool expect_valid_internal_key,
-    RangeDelAggregator* range_del_agg, const Compaction* compaction,
+    CompactionRangeDelAggregator* range_del_agg, const Compaction* compaction,
     const CompactionFilter* compaction_filter,
     const std::atomic<bool>* shutting_down,
     const SequenceNumber preserve_deletes_seqnum)
@@ -36,7 +36,7 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, Env* env,
     bool report_detailed_time, bool expect_valid_internal_key,
-    RangeDelAggregator* range_del_agg,
+    CompactionRangeDelAggregator* range_del_agg,
     std::unique_ptr<CompactionProxy> compaction,
     const CompactionFilter* compaction_filter,
     const std::atomic<bool>* shutting_down,

db/compaction_iterator.h

@@ -64,7 +64,7 @@ class CompactionIterator {
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* snapshot_checker, Env* env,
                      bool report_detailed_time, bool expect_valid_internal_key,
-                     RangeDelAggregator* range_del_agg,
+                     CompactionRangeDelAggregator* range_del_agg,
                      const Compaction* compaction = nullptr,
                      const CompactionFilter* compaction_filter = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -77,7 +77,7 @@ class CompactionIterator {
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* snapshot_checker, Env* env,
                      bool report_detailed_time, bool expect_valid_internal_key,
-                     RangeDelAggregator* range_del_agg,
+                     CompactionRangeDelAggregator* range_del_agg,
                      std::unique_ptr<CompactionProxy> compaction,
                      const CompactionFilter* compaction_filter = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -141,7 +141,7 @@ class CompactionIterator {
   Env* env_;
   bool report_detailed_time_;
   bool expect_valid_internal_key_;
-  RangeDelAggregator* range_del_agg_;
+  CompactionRangeDelAggregator* range_del_agg_;
   std::unique_ptr<CompactionProxy> compaction_;
   const CompactionFilter* compaction_filter_;
   const std::atomic<bool>* shutting_down_;

db/compaction_iterator_test.cc

@@ -221,10 +221,15 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
       MergeOperator* merge_op = nullptr, CompactionFilter* filter = nullptr,
       bool bottommost_level = false,
       SequenceNumber earliest_write_conflict_snapshot = kMaxSequenceNumber) {
-    std::unique_ptr<InternalIterator> range_del_iter(
+    std::unique_ptr<InternalIterator> unfragmented_range_del_iter(
         new test::VectorIterator(range_del_ks, range_del_vs));
-    range_del_agg_.reset(new RangeDelAggregator(icmp_, snapshots_));
+    auto tombstone_list = std::make_shared<FragmentedRangeTombstoneList>(
-    ASSERT_OK(range_del_agg_->AddTombstones(std::move(range_del_iter)));
+        std::move(unfragmented_range_del_iter), icmp_);
+    std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(
+        new FragmentedRangeTombstoneIterator(tombstone_list, icmp_,
+                                             kMaxSequenceNumber));
+    range_del_agg_.reset(new CompactionRangeDelAggregator(&icmp_, snapshots_));
+    range_del_agg_->AddTombstones(std::move(range_del_iter));
 
     std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
     if (filter || bottommost_level) {
@@ -292,7 +297,7 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
   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<CompactionRangeDelAggregator> range_del_agg_;
   std::unique_ptr<SnapshotChecker> snapshot_checker_;
   std::atomic<bool> shutting_down_{false};
   FakeCompaction* compaction_proxy_;

db/compaction_job.cc

@@ -36,7 +36,7 @@
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
 #include "db/merge_helper.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "db/version_set.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
@@ -805,15 +805,13 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
 void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   assert(sub_compact != nullptr);
   ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
-  RangeDelAggregatorV2 range_del_agg_v2(&cfd->internal_comparator(),
+  CompactionRangeDelAggregator range_del_agg(&cfd->internal_comparator(),
-                                        kMaxSequenceNumber /* upper_bound */);
+                                             existing_snapshots_);
-  auto* range_del_agg =
-      range_del_agg_v2.DelegateToRangeDelAggregator(existing_snapshots_);
 
   // Although the v2 aggregator is what the level iterator(s) know about,
   // the AddTombstones calls will be propagated down to the v1 aggregator.
   std::unique_ptr<InternalIterator> input(versions_->MakeInputIterator(
-      sub_compact->compaction, &range_del_agg_v2, env_optiosn_for_read_));
+      sub_compact->compaction, &range_del_agg, env_optiosn_for_read_));
 
   AutoThreadOperationStageUpdater stage_updater(
       ThreadStatus::STAGE_COMPACTION_PROCESS_KV);
@@ -902,8 +900,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
       input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
       &existing_snapshots_, earliest_write_conflict_snapshot_,
       snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_), false,
-      range_del_agg, sub_compact->compaction, compaction_filter, shutting_down_,
+      &range_del_agg, sub_compact->compaction, compaction_filter,
-      preserve_deletes_seqnum_));
+      shutting_down_, preserve_deletes_seqnum_));
   auto c_iter = sub_compact->c_iter.get();
   c_iter->SeekToFirst();
   if (c_iter->Valid() && sub_compact->compaction->output_level() != 0) {
@@ -1041,7 +1039,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
       }
       CompactionIterationStats range_del_out_stats;
       status =
-          FinishCompactionOutputFile(input_status, sub_compact, range_del_agg,
+          FinishCompactionOutputFile(input_status, sub_compact, &range_del_agg,
                                      &range_del_out_stats, next_key);
       RecordDroppedKeys(range_del_out_stats,
                         &sub_compact->compaction_job_stats);
@@ -1092,8 +1090,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   if (status.ok() && sub_compact->builder == nullptr &&
-      sub_compact->outputs.size() == 0 &&
+      sub_compact->outputs.size() == 0 && !range_del_agg.IsEmpty()) {
-      !range_del_agg->IsEmpty()) {
     // handle subcompaction containing only range deletions
     status = OpenCompactionOutputFile(sub_compact);
   }
@@ -1102,7 +1099,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   // close the output file.
   if (sub_compact->builder != nullptr) {
     CompactionIterationStats range_del_out_stats;
-    Status s = FinishCompactionOutputFile(status, sub_compact, range_del_agg,
+    Status s = FinishCompactionOutputFile(status, sub_compact, &range_del_agg,
                                           &range_del_out_stats);
     if (status.ok()) {
       status = s;
@@ -1168,7 +1165,7 @@ void CompactionJob::RecordDroppedKeys(
 
 Status CompactionJob::FinishCompactionOutputFile(
     const Status& input_status, SubcompactionState* sub_compact,
-    RangeDelAggregator* range_del_agg,
+    CompactionRangeDelAggregator* range_del_agg,
     CompactionIterationStats* range_del_out_stats,
     const Slice* next_table_min_key /* = nullptr */) {
   AutoThreadOperationStageUpdater stage_updater(
@@ -1207,10 +1204,19 @@ Status CompactionJob::FinishCompactionOutputFile(
       lower_bound = nullptr;
     }
     if (next_table_min_key != nullptr) {
-      // This isn't the last file in the subcompaction, so extend until the next
+      // This may be the last file in the subcompaction in some cases, so we
-      // file starts.
+      // need to compare the end key of subcompaction with the next file start
+      // key. When the end key is chosen by the subcompaction, we know that
+      // it must be the biggest key in output file. Therefore, it is safe to
+      // use the smaller key as the upper bound of the output file, to ensure
+      // that there is no overlapping between different output files.
       upper_bound_guard = ExtractUserKey(*next_table_min_key);
+      if (sub_compact->end != nullptr &&
+          ucmp->Compare(upper_bound_guard, *sub_compact->end) >= 0) {
+        upper_bound = sub_compact->end;
+      } else {
         upper_bound = &upper_bound_guard;
+      }
     } else {
       // This is the last file in the subcompaction, so extend until the
       // subcompaction ends.
@@ -1220,11 +1226,6 @@ Status CompactionJob::FinishCompactionOutputFile(
     if (existing_snapshots_.size() > 0) {
       earliest_snapshot = existing_snapshots_[0];
     }
-    auto it = range_del_agg->NewIterator();
-    if (lower_bound != nullptr) {
-      it->Seek(*lower_bound);
-    }
-
     bool has_overlapping_endpoints;
     if (upper_bound != nullptr && meta->largest.size() > 0) {
       has_overlapping_endpoints =
@@ -1232,6 +1233,24 @@ Status CompactionJob::FinishCompactionOutputFile(
     } else {
       has_overlapping_endpoints = false;
     }
+
+    // The end key of the subcompaction must be bigger or equal to the upper
+    // bound. If the end of subcompaction is null or the upper bound is null,
+    // it means that this file is the last file in the compaction. So there
+    // will be no overlapping between this file and others.
+    assert(sub_compact->end == nullptr ||
+           upper_bound == nullptr ||
+           ucmp->Compare(*upper_bound , *sub_compact->end) <= 0);
+    auto it = range_del_agg->NewIterator(lower_bound, upper_bound,
+                                         has_overlapping_endpoints);
+    // Position the range tombstone output iterator. There may be tombstone
+    // fragments that are entirely out of range, so make sure that we do not
+    // include those.
+    if (lower_bound != nullptr) {
+      it->Seek(*lower_bound);
+    } else {
+      it->SeekToFirst();
+    }
     for (; it->Valid(); it->Next()) {
       auto tombstone = it->Tombstone();
       if (upper_bound != nullptr) {
@@ -1257,6 +1276,8 @@ Status CompactionJob::FinishCompactionOutputFile(
       }
 
       auto kv = tombstone.Serialize();
+      assert(lower_bound == nullptr ||
+             ucmp->Compare(*lower_bound, kv.second) < 0);
       sub_compact->builder->Add(kv.first.Encode(), kv.second);
       InternalKey smallest_candidate = std::move(kv.first);
       if (lower_bound != nullptr &&

db/compaction_job.h

@@ -29,8 +29,8 @@
 #include "db/version_edit.h"
 #include "db/write_controller.h"
 #include "db/write_thread.h"
-#include "options/db_options.h"
 #include "options/cf_options.h"
+#include "options/db_options.h"
 #include "port/port.h"
 #include "rocksdb/compaction_filter.h"
 #include "rocksdb/compaction_job_stats.h"
@@ -104,7 +104,7 @@ class CompactionJob {
 
   Status FinishCompactionOutputFile(
       const Status& input_status, SubcompactionState* sub_compact,
-      RangeDelAggregator* range_del_agg,
+      CompactionRangeDelAggregator* range_del_agg,
       CompactionIterationStats* range_del_out_stats,
       const Slice* next_table_min_key = nullptr);
   Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options);

db/db_compaction_filter_test.cc

@@ -340,7 +340,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
   Arena arena;
   {
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregatorV2 range_del_agg(&icmp,
+    ReadRangeDelAggregator range_del_agg(&icmp,
                                          kMaxSequenceNumber /* upper_bound */);
     ScopedArenaIterator iter(dbfull()->NewInternalIterator(
         &arena, &range_del_agg, kMaxSequenceNumber, handles_[1]));
@@ -430,7 +430,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
   count = 0;
   {
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregatorV2 range_del_agg(&icmp,
+    ReadRangeDelAggregator range_del_agg(&icmp,
                                          kMaxSequenceNumber /* upper_bound */);
     ScopedArenaIterator iter(dbfull()->NewInternalIterator(
         &arena, &range_del_agg, kMaxSequenceNumber, handles_[1]));
@@ -648,7 +648,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextManual) {
     int total = 0;
     Arena arena;
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregatorV2 range_del_agg(&icmp,
+    ReadRangeDelAggregator range_del_agg(&icmp,
                                          kMaxSequenceNumber /* snapshots */);
     ScopedArenaIterator iter(dbfull()->NewInternalIterator(
         &arena, &range_del_agg, kMaxSequenceNumber));

db/db_flush_test.cc

@@ -407,6 +407,87 @@ TEST_P(DBAtomicFlushTest, AtomicFlushRollbackSomeJobs) {
   Destroy(options);
 }
 
+TEST_P(DBAtomicFlushTest, FlushMultipleCFs_DropSomeBeforeRequestFlush) {
+  bool atomic_flush = GetParam();
+  if (!atomic_flush) {
+    return;
+  }
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = atomic_flush;
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  std::vector<int> cf_ids;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_OK(Put(cf_id, "key", "value", wopts));
+    cf_ids.push_back(cf_id);
+  }
+  ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+  ASSERT_TRUE(Flush(cf_ids).IsShutdownInProgress());
+  Destroy(options);
+}
+
+TEST_P(DBAtomicFlushTest,
+       FlushMultipleCFs_DropSomeAfterScheduleFlushBeforeFlushJobRun) {
+  bool atomic_flush = GetParam();
+  if (!atomic_flush) {
+    return;
+  }
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = atomic_flush;
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::AtomicFlushMemTables:AfterScheduleFlush",
+        "DBAtomicFlushTest::BeforeDropCF"},
+       {"DBAtomicFlushTest::AfterDropCF",
+        "DBImpl::BackgroundCallFlush:start"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_OK(Put(cf_id, "key", "value", wopts));
+  }
+  port::Thread user_thread([&]() {
+    TEST_SYNC_POINT("DBAtomicFlushTest::BeforeDropCF");
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+    TEST_SYNC_POINT("DBAtomicFlushTest::AfterDropCF");
+  });
+  FlushOptions flush_opts;
+  flush_opts.wait = true;
+  ASSERT_OK(dbfull()->Flush(flush_opts, handles_));
+  user_thread.join();
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_EQ("value", Get(cf_id, "key"));
+  }
+
+  ReopenWithColumnFamilies({kDefaultColumnFamilyName, "eevee"}, options);
+  num_cfs = handles_.size();
+  ASSERT_EQ(2, num_cfs);
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_EQ("value", Get(cf_id, "key"));
+  }
+  Destroy(options);
+}
+
 INSTANTIATE_TEST_CASE_P(DBFlushDirectIOTest, DBFlushDirectIOTest,
                         testing::Bool());
 

db/db_impl.cc

@@ -45,7 +45,6 @@
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
 #include "db/merge_helper.h"
-#include "db/range_del_aggregator.h"
 #include "db/range_tombstone_fragmenter.h"
 #include "db/table_cache.h"
 #include "db/table_properties_collector.h"
@@ -221,7 +220,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       preserve_deletes_(options.preserve_deletes),
       closed_(false),
       error_handler_(this, immutable_db_options_, &mutex_),
-      atomic_flush_commit_in_progress_(false) {
+      atomic_flush_install_cv_(&mutex_) {
   // !batch_per_trx_ implies seq_per_batch_ because it is only unset for
   // WriteUnprepared, which should use seq_per_batch_.
   assert(batch_per_txn_ || seq_per_batch_);
@@ -1033,7 +1032,7 @@ bool DBImpl::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) {
 }
 
 InternalIterator* DBImpl::NewInternalIterator(
-    Arena* arena, RangeDelAggregatorV2* range_del_agg, SequenceNumber sequence,
+    Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
     ColumnFamilyHandle* column_family) {
   ColumnFamilyData* cfd;
   if (column_family == nullptr) {
@@ -1150,10 +1149,12 @@ static void CleanupIteratorState(void* arg1, void* /*arg2*/) {
 }
 }  // namespace
 
-InternalIterator* DBImpl::NewInternalIterator(
+InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
-    const ReadOptions& read_options, ColumnFamilyData* cfd,
+                                              ColumnFamilyData* cfd,
-    SuperVersion* super_version, Arena* arena,
+                                              SuperVersion* super_version,
-    RangeDelAggregatorV2* range_del_agg, SequenceNumber sequence) {
+                                              Arena* arena,
+                                              RangeDelAggregator* range_del_agg,
+                                              SequenceNumber sequence) {
   InternalIterator* internal_iter;
   assert(arena != nullptr);
   assert(range_del_agg != nullptr);

db/db_impl.h

@@ -31,7 +31,7 @@
 #include "db/log_writer.h"
 #include "db/logs_with_prep_tracker.h"
 #include "db/pre_release_callback.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "db/read_callback.h"
 #include "db/snapshot_checker.h"
 #include "db/snapshot_impl.h"
@@ -374,8 +374,8 @@ class DBImpl : public DB {
   // The keys of this iterator are internal keys (see format.h).
   // The returned iterator should be deleted when no longer needed.
   InternalIterator* NewInternalIterator(
-      Arena* arena, RangeDelAggregatorV2* range_del_agg,
+      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
-      SequenceNumber sequence, ColumnFamilyHandle* column_family = nullptr);
+      ColumnFamilyHandle* column_family = nullptr);
 
   LogsWithPrepTracker* logs_with_prep_tracker() {
     return &logs_with_prep_tracker_;
@@ -578,12 +578,9 @@ class DBImpl : public DB {
 
   const WriteController& write_controller() { return write_controller_; }
 
-  InternalIterator* NewInternalIterator(const ReadOptions&,
+  InternalIterator* NewInternalIterator(
-                                        ColumnFamilyData* cfd,
+      const ReadOptions&, ColumnFamilyData* cfd, SuperVersion* super_version,
-                                        SuperVersion* super_version,
+      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence);
-                                        Arena* arena,
-                                        RangeDelAggregatorV2* range_del_agg,
-                                        SequenceNumber sequence);
 
   // hollow transactions shell used for recovery.
   // these will then be passed to TransactionDB so that
@@ -1613,15 +1610,16 @@ class DBImpl : public DB {
 
   ErrorHandler error_handler_;
 
-  // True if the DB is committing atomic flush.
+  // Conditional variable to coordinate installation of atomic flush results.
-  // TODO (yanqin) the current impl assumes that the entire DB belongs to
+  // With atomic flush, each bg thread installs the result of flushing multiple
-  // a single atomic flush group. In the future we need to add a new class
+  // column families, and different threads can flush different column
-  // (struct) similar to the following to make it more general.
+  // families. It's difficult to rely on one thread to perform batch
-  // struct AtomicFlushGroup {
+  // installation for all threads. This is different from the non-atomic flush
-  //   bool commit_in_progress_;
+  // case.
-  //   std::vector<MemTableList*> imm_lists;
+  // atomic_flush_install_cv_ makes sure that threads install atomic flush
-  // };
+  // results sequentially. Flush results of memtables with lower IDs get
-  bool atomic_flush_commit_in_progress_;
+  // installed to MANIFEST first.
+  InstrumentedCondVar atomic_flush_install_cv_;
 };
 
 extern Options SanitizeOptions(const std::string& db,

db/db_impl_compaction_flush.cc

@@ -219,20 +219,24 @@ Status DBImpl::FlushMemTablesToOutputFiles(
     return AtomicFlushMemTablesToOutputFiles(bg_flush_args, made_progress,
                                              job_context, log_buffer);
   }
-  Status s;
+  Status status;
   for (auto& arg : bg_flush_args) {
     ColumnFamilyData* cfd = arg.cfd_;
-    const MutableCFOptions& mutable_cf_options =
+    MutableCFOptions mutable_cf_options = *cfd->GetLatestMutableCFOptions();
-        *cfd->GetLatestMutableCFOptions();
     SuperVersionContext* superversion_context = arg.superversion_context_;
-    s = FlushMemTableToOutputFile(cfd, mutable_cf_options, made_progress,
+    Status s = FlushMemTableToOutputFile(cfd, mutable_cf_options, made_progress,
                                          job_context, superversion_context,
                                          log_buffer);
     if (!s.ok()) {
+      status = s;
+      if (!s.IsShutdownInProgress()) {
+        // At this point, DB is not shutting down, nor is cfd dropped.
+        // Something is wrong, thus we break out of the loop.
         break;
       }
     }
-  return s;
+  }
+  return status;
 }
 
 /*
@@ -271,7 +275,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
   }
   autovector<Directory*> distinct_output_dirs;
   std::vector<FlushJob> jobs;
+  std::vector<MutableCFOptions> all_mutable_cf_options;
   int num_cfs = static_cast<int>(cfds.size());
+  all_mutable_cf_options.reserve(num_cfs);
   for (int i = 0; i < num_cfs; ++i) {
     auto cfd = cfds[i];
     Directory* data_dir = GetDataDir(cfd, 0U);
@@ -290,8 +296,8 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
       distinct_output_dirs.emplace_back(data_dir);
     }
 
-    const MutableCFOptions& mutable_cf_options =
+    all_mutable_cf_options.emplace_back(*cfd->GetLatestMutableCFOptions());
-        *cfd->GetLatestMutableCFOptions();
+    const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back();
     const uint64_t* max_memtable_id = &(bg_flush_args[i].max_memtable_id_);
     jobs.emplace_back(
         dbname_, cfds[i], immutable_db_options_, mutable_cf_options,
@@ -304,21 +310,18 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     jobs.back().PickMemTable();
   }
 
-  autovector<FileMetaData> file_meta;
+  std::vector<FileMetaData> file_meta(num_cfs);
   Status s;
   assert(num_cfs == static_cast<int>(jobs.size()));
 
-  for (int i = 0; i != num_cfs; ++i) {
-    file_meta.emplace_back();
-
 #ifndef ROCKSDB_LITE
-    const MutableCFOptions& mutable_cf_options =
+  for (int i = 0; i != num_cfs; ++i) {
-        *cfds[i]->GetLatestMutableCFOptions();
+    const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.at(i);
     // may temporarily unlock and lock the mutex.
     NotifyOnFlushBegin(cfds[i], &file_meta[i], mutable_cf_options,
                        job_context->job_id, jobs[i].GetTableProperties());
-#endif /* !ROCKSDB_LITE */
   }
+#endif /* !ROCKSDB_LITE */
 
   if (logfile_number_ > 0) {
     // TODO (yanqin) investigate whether we should sync the closed logs for
@@ -331,8 +334,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
   autovector<std::pair<bool, Status>> exec_status;
   for (int i = 0; i != num_cfs; ++i) {
     // Initially all jobs are not executed, with status OK.
-    std::pair<bool, Status> elem(false, Status::OK());
+    exec_status.emplace_back(false, Status::OK());
-    exec_status.emplace_back(elem);
   }
 
   if (s.ok()) {
@@ -341,10 +343,6 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
       exec_status[i].second =
           jobs[i].Run(&logs_with_prep_tracker_, &file_meta[i]);
       exec_status[i].first = true;
-      if (!exec_status[i].second.ok()) {
-        s = exec_status[i].second;
-        break;
-      }
     }
     if (num_cfs > 1) {
       TEST_SYNC_POINT(
@@ -352,17 +350,27 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
       TEST_SYNC_POINT(
           "DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:2");
     }
-    if (s.ok()) {
     exec_status[0].second =
         jobs[0].Run(&logs_with_prep_tracker_, &file_meta[0]);
     exec_status[0].first = true;
-      if (!exec_status[0].second.ok()) {
+
-        s = exec_status[0].second;
+    Status error_status;
+    for (const auto& e : exec_status) {
+      if (!e.second.ok()) {
+        s = e.second;
+        if (!e.second.IsShutdownInProgress()) {
+          // If a flush job did not return OK, and the CF is not dropped, and
+          // the DB is not shutting down, then we have to return this result to
+          // caller later.
+          error_status = e.second;
         }
       }
     }
 
-  if (s.ok()) {
+    s = error_status.ok() ? s : error_status;
+  }
+
+  if (s.ok() || s.IsShutdownInProgress()) {
     // Sync on all distinct output directories.
     for (auto dir : distinct_output_dirs) {
       if (dir != nullptr) {
@@ -372,37 +380,78 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
         }
       }
     }
+  }
 
   if (s.ok()) {
+    auto wait_to_install_func = [&]() {
+      bool ready = true;
+      for (size_t i = 0; i != cfds.size(); ++i) {
+        const auto& mems = jobs[i].GetMemTables();
+        if (cfds[i]->IsDropped()) {
+          // If the column family is dropped, then do not wait.
+          continue;
+        } else if (!mems.empty() &&
+                   cfds[i]->imm()->GetEarliestMemTableID() < mems[0]->GetID()) {
+          // If a flush job needs to install the flush result for mems and
+          // mems[0] is not the earliest memtable, it means another thread must
+          // be installing flush results for the same column family, then the
+          // current thread needs to wait.
+          ready = false;
+          break;
+        } else if (mems.empty() && cfds[i]->imm()->GetEarliestMemTableID() <=
+                                       bg_flush_args[i].max_memtable_id_) {
+          // If a flush job does not need to install flush results, then it has
+          // to wait until all memtables up to max_memtable_id_ (inclusive) are
+          // installed.
+          ready = false;
+          break;
+        }
+      }
+      return ready;
+    };
+
+    bool resuming_from_bg_err = error_handler_.IsDBStopped();
+    while ((!error_handler_.IsDBStopped() ||
+            error_handler_.GetRecoveryError().ok()) &&
+           !wait_to_install_func()) {
+      atomic_flush_install_cv_.Wait();
+    }
+
+    s = resuming_from_bg_err ? error_handler_.GetRecoveryError()
+                             : error_handler_.GetBGError();
+  }
+
+  if (s.ok()) {
+    autovector<ColumnFamilyData*> tmp_cfds;
     autovector<const autovector<MemTable*>*> mems_list;
+    autovector<const MutableCFOptions*> mutable_cf_options_list;
+    autovector<FileMetaData*> tmp_file_meta;
     for (int i = 0; i != num_cfs; ++i) {
       const auto& mems = jobs[i].GetMemTables();
+      if (!cfds[i]->IsDropped() && !mems.empty()) {
+        tmp_cfds.emplace_back(cfds[i]);
         mems_list.emplace_back(&mems);
-      }
+        mutable_cf_options_list.emplace_back(&all_mutable_cf_options[i]);
-      autovector<ColumnFamilyData*> all_cfds;
+        tmp_file_meta.emplace_back(&file_meta[i]);
-      autovector<MemTableList*> imm_lists;
-      autovector<const MutableCFOptions*> mutable_cf_options_list;
-      for (auto cfd : *versions_->GetColumnFamilySet()) {
-        all_cfds.emplace_back(cfd);
-        imm_lists.emplace_back(cfd->imm());
-        mutable_cf_options_list.emplace_back(cfd->GetLatestMutableCFOptions());
-      }
-
-      s = MemTableList::TryInstallMemtableFlushResults(
-          imm_lists, all_cfds, mutable_cf_options_list, mems_list,
-          &atomic_flush_commit_in_progress_, &logs_with_prep_tracker_,
-          versions_.get(), &mutex_, file_meta, &job_context->memtables_to_free,
-          directories_.GetDbDir(), log_buffer);
       }
     }
 
-  if (s.ok()) {
+    s = InstallMemtableAtomicFlushResults(
+        nullptr /* imm_lists */, tmp_cfds, mutable_cf_options_list, mems_list,
+        versions_.get(), &mutex_, tmp_file_meta,
+        &job_context->memtables_to_free, directories_.GetDbDir(), log_buffer);
+  }
+
+  if (s.ok() || s.IsShutdownInProgress()) {
     assert(num_cfs ==
            static_cast<int>(job_context->superversion_contexts.size()));
     for (int i = 0; i != num_cfs; ++i) {
+      if (cfds[i]->IsDropped()) {
+        continue;
+      }
       InstallSuperVersionAndScheduleWork(cfds[i],
                                          &job_context->superversion_contexts[i],
-                                         *cfds[i]->GetLatestMutableCFOptions());
+                                         all_mutable_cf_options[i]);
       VersionStorageInfo::LevelSummaryStorage tmp;
       ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
                        cfds[i]->GetName().c_str(),
@@ -415,8 +464,10 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     auto sfm = static_cast<SstFileManagerImpl*>(
         immutable_db_options_.sst_file_manager.get());
     for (int i = 0; i != num_cfs; ++i) {
-      NotifyOnFlushCompleted(cfds[i], &file_meta[i],
+      if (cfds[i]->IsDropped()) {
-                             *cfds[i]->GetLatestMutableCFOptions(),
+        continue;
+      }
+      NotifyOnFlushCompleted(cfds[i], &file_meta[i], all_mutable_cf_options[i],
                              job_context->job_id, jobs[i].GetTableProperties());
       if (sfm) {
         std::string file_path = MakeTableFileName(
@@ -434,7 +485,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
 #endif  // ROCKSDB_LITE
   }
 
-  if (!s.ok()) {
+  // Need to undo atomic flush if something went wrong, i.e. s is not OK and
+  // it is not because of CF drop.
+  if (!s.ok() && !s.IsShutdownInProgress()) {
     // Have to cancel the flush jobs that have NOT executed because we need to
     // unref the versions.
     for (int i = 0; i != num_cfs; ++i) {
@@ -442,7 +495,6 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
         jobs[i].Cancel();
       }
     }
-    if (!s.IsShutdownInProgress()) {
     for (int i = 0; i != num_cfs; ++i) {
       if (exec_status[i].first && exec_status[i].second.ok()) {
         auto& mems = jobs[i].GetMemTables();
@@ -453,7 +505,6 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     Status new_bg_error = s;
     error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
   }
-  }
 
   return s;
 }
@@ -1407,6 +1458,7 @@ Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
 void DBImpl::GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
                                   FlushRequest* req) {
   assert(req != nullptr);
+  req->reserve(cfds.size());
   for (const auto cfd : cfds) {
     if (nullptr == cfd) {
       // cfd may be null, see DBImpl::ScheduleFlushes
@@ -1440,12 +1492,17 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
       write_thread_.EnterUnbatched(&w, &mutex_);
     }
 
+    if (!cfd->mem()->IsEmpty() || !cached_recoverable_state_empty_.load()) {
+      s = SwitchMemtable(cfd, &context);
+    }
+
+    if (s.ok()) {
       if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
           !cached_recoverable_state_empty_.load()) {
-      s = SwitchMemtable(cfd, &context);
         flush_memtable_id = cfd->imm()->GetLatestMemTableID();
         flush_req.emplace_back(cfd, flush_memtable_id);
       }
+    }
 
     if (s.ok() && !flush_req.empty()) {
       for (auto& elem : flush_req) {
@@ -1518,6 +1575,9 @@ Status DBImpl::AtomicFlushMemTables(
       }
     }
     for (auto cfd : cfds) {
+      if (cfd->mem()->IsEmpty() && cached_recoverable_state_empty_.load()) {
+        continue;
+      }
       cfd->Ref();
       s = SwitchMemtable(cfd, &context);
       cfd->Unref();
@@ -1539,6 +1599,7 @@ Status DBImpl::AtomicFlushMemTables(
       write_thread_.ExitUnbatched(&w);
     }
   }
+  TEST_SYNC_POINT("DBImpl::AtomicFlushMemTables:AfterScheduleFlush");
 
   if (s.ok() && flush_options.wait) {
     autovector<const uint64_t*> flush_memtable_ids;
@@ -2046,6 +2107,7 @@ void DBImpl::BackgroundCallFlush() {
     bg_flush_scheduled_--;
     // See if there's more work to be done
     MaybeScheduleFlushOrCompaction();
+    atomic_flush_install_cv_.SignalAll();
     bg_cv_.SignalAll();
     // IMPORTANT: there should be no code after calling SignalAll. This call may
     // signal the DB destructor that it's OK to proceed with destruction. In

db/db_impl_open.cc

@@ -23,8 +23,7 @@
 #include "util/sync_point.h"
 
 namespace rocksdb {
-Options SanitizeOptions(const std::string& dbname,
+Options SanitizeOptions(const std::string& dbname, const Options& src) {
-                        const Options& src) {
   auto db_options = SanitizeOptions(dbname, DBOptions(src));
   ImmutableDBOptions immutable_db_options(db_options);
   auto cf_options =
@@ -56,10 +55,9 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.write_buffer_manager.reset(
         new WriteBufferManager(result.db_write_buffer_size));
   }
-  auto bg_job_limits = DBImpl::GetBGJobLimits(result.max_background_flushes,
+  auto bg_job_limits = DBImpl::GetBGJobLimits(
-                                              result.max_background_compactions,
+      result.max_background_flushes, result.max_background_compactions,
-                                              result.max_background_jobs,
+      result.max_background_jobs, true /* parallelize_compactions */);
-                                              true /* parallelize_compactions */);
   result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_compactions,
                                            Env::Priority::LOW);
   result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_flushes,
@@ -107,14 +105,12 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
   }
 
-  if (result.use_direct_reads &&
+  if (result.use_direct_reads && result.compaction_readahead_size == 0) {
-      result.compaction_readahead_size == 0) {
     TEST_SYNC_POINT_CALLBACK("SanitizeOptions:direct_io", nullptr);
     result.compaction_readahead_size = 1024 * 1024 * 2;
   }
 
-  if (result.compaction_readahead_size > 0 ||
+  if (result.compaction_readahead_size > 0 || result.use_direct_reads) {
-      result.use_direct_reads) {
     result.new_table_reader_for_compaction_inputs = true;
   }
 
@@ -258,8 +254,7 @@ Status DBImpl::NewDB() {
   return s;
 }
 
-Status DBImpl::CreateAndNewDirectory(
+Status DBImpl::CreateAndNewDirectory(Env* env, const std::string& dirname,
-    Env* env, const std::string& dirname,
                                      std::unique_ptr<Directory>* directory) {
   // We call CreateDirIfMissing() as the directory may already exist (if we
   // are reopening a DB), when this happens we don't want creating the
@@ -341,8 +336,8 @@ Status DBImpl::Recover(
       }
     } else if (s.ok()) {
       if (immutable_db_options_.error_if_exists) {
-        return Status::InvalidArgument(
+        return Status::InvalidArgument(dbname_,
-            dbname_, "exists (error_if_exists is true)");
+                                       "exists (error_if_exists is true)");
       }
     } else {
       // Unexpected error reading file
@@ -479,6 +474,28 @@ Status DBImpl::Recover(
     }
   }
 
+  if (read_only) {
+    // If we are opening as read-only, we need to update options_file_number_
+    // to reflect the most recent OPTIONS file. It does not matter for regular
+    // read-write db instance because options_file_number_ will later be
+    // updated to versions_->NewFileNumber() in RenameTempFileToOptionsFile.
+    std::vector<std::string> file_names;
+    if (s.ok()) {
+      s = env_->GetChildren(GetName(), &file_names);
+    }
+    if (s.ok()) {
+      uint64_t number = 0;
+      uint64_t options_file_number = 0;
+      FileType type;
+      for (const auto& fname : file_names) {
+        if (ParseFileName(fname, &number, &type) && type == kOptionsFile) {
+          options_file_number = std::max(number, options_file_number);
+        }
+      }
+      versions_->options_file_number_ = options_file_number;
+    }
+  }
+
   return s;
 }
 
@@ -527,8 +544,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
     std::map<std::string, uint32_t> cf_name_id_map;
     std::map<uint32_t, uint64_t> cf_lognumber_map;
     for (auto cfd : *versions_->GetColumnFamilySet()) {
-      cf_name_id_map.insert(
+      cf_name_id_map.insert(std::make_pair(cfd->GetName(), cfd->GetID()));
-        std::make_pair(cfd->GetName(), cfd->GetID()));
       cf_lognumber_map.insert(
           std::make_pair(cfd->GetID(), cfd->GetLogNumber()));
     }
@@ -880,8 +896,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
       // VersionSet::next_file_number_ always to be strictly greater than any
       // log number
       versions_->MarkFileNumberUsed(max_log_number + 1);
-      status = versions_->LogAndApply(
+      status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
-          cfd, *cfd->GetLatestMutableCFOptions(), edit, &mutex_);
+                                      edit, &mutex_);
       if (!status.ok()) {
         // Recovery failed
         break;
@@ -994,12 +1010,17 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
       if (use_custom_gc_ && snapshot_checker == nullptr) {
         snapshot_checker = DisableGCSnapshotChecker::Instance();
       }
+      std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+          range_del_iters;
+      auto range_del_iter =
+          mem->NewRangeTombstoneIterator(ro, kMaxSequenceNumber);
+      if (range_del_iter != nullptr) {
+        range_del_iters.emplace_back(range_del_iter);
+      }
       s = BuildTable(
           dbname_, env_, *cfd->ioptions(), mutable_cf_options,
           env_options_for_compaction_, cfd->table_cache(), iter.get(),
-          std::unique_ptr<InternalIterator>(
+          std::move(range_del_iters), &meta, cfd->internal_comparator(),
-              mem->NewRangeTombstoneIterator(ro, versions_->LastSequence())),
-          &meta, cfd->internal_comparator(),
           cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
           snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
           GetCompressionFlush(*cfd->ioptions(), mutable_cf_options),
@@ -1033,8 +1054,8 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
   stats.bytes_written = meta.fd.GetFileSize();
   stats.num_output_files = 1;
   cfd->internal_stats()->AddCompactionStats(level, stats);
-  cfd->internal_stats()->AddCFStats(
+  cfd->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
-      InternalStats::BYTES_FLUSHED, meta.fd.GetFileSize());
+                                    meta.fd.GetFileSize());
   RecordTick(stats_, COMPACT_WRITE_BYTES, meta.fd.GetFileSize());
   return s;
 }
@@ -1227,7 +1248,8 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
           !cfd->mem()->IsMergeOperatorSupported()) {
         s = Status::InvalidArgument(
             "The memtable of column family %s does not support merge operator "
-            "its options.merge_operator is non-null", cfd->GetName().c_str());
+            "its options.merge_operator is non-null",
+            cfd->GetName().c_str());
       }
       if (!s.ok()) {
         break;

db/db_impl_readonly.cc

@@ -9,7 +9,6 @@
 #include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/merge_context.h"
-#include "db/range_del_aggregator.h"
 #include "monitoring/perf_context_imp.h"
 
 namespace rocksdb {

db/db_impl_readonly.h

@@ -89,10 +89,11 @@ class DBImplReadOnly : public DBImpl {
   virtual Status EnableFileDeletions(bool /*force*/) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
-  virtual Status GetLiveFiles(std::vector<std::string>&,
+  virtual Status GetLiveFiles(std::vector<std::string>& ret,
-                              uint64_t* /*manifest_file_size*/,
+                              uint64_t* manifest_file_size,
-                              bool /*flush_memtable*/ = true) override {
+                              bool /*flush_memtable*/) override {
-    return Status::NotSupported("Not supported operation in read only mode.");
+    return DBImpl::GetLiveFiles(ret, manifest_file_size,
+                                false /* flush_memtable */);
   }
 
   using DBImpl::Flush;

db/db_impl_write.cc

@@ -265,18 +265,19 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     // We're optimistic, updating the stats before we successfully
     // commit.  That lets us release our leader status early.
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count,
+    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count,
                       concurrent_update);
     RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
+    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
                       concurrent_update);
     RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-    stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
+    stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
+                      concurrent_update);
     RecordTick(stats_, WRITE_DONE_BY_SELF);
     auto write_done_by_other = write_group.size - 1;
     if (write_done_by_other > 0) {
-      stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
+      stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
-                        concurrent_update);
+                        write_done_by_other, concurrent_update);
       RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
     }
     MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -467,9 +468,9 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
     }
 
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count);
+    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count);
     RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size);
+    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size);
     RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
     MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
 
@@ -477,10 +478,10 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
 
     if (w.ShouldWriteToWAL()) {
       PERF_TIMER_GUARD(write_wal_time);
-      stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
+      stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1);
       RecordTick(stats_, WRITE_DONE_BY_SELF, 1);
       if (wal_write_group.size > 1) {
-        stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
+        stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
                           wal_write_group.size - 1);
         RecordTick(stats_, WRITE_DONE_BY_OTHER, wal_write_group.size - 1);
       }
@@ -591,15 +592,16 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
   // We're optimistic, updating the stats before we successfully
   // commit.  That lets us release our leader status early.
   auto stats = default_cf_internal_stats_;
-  stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
+  stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
                     concurrent_update);
   RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-  stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
+  stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
+                    concurrent_update);
   RecordTick(stats_, WRITE_DONE_BY_SELF);
   auto write_done_by_other = write_group.size - 1;
   if (write_done_by_other > 0) {
-    stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
+    stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
-                      concurrent_update);
+                      write_done_by_other, concurrent_update);
     RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
   }
   MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -908,12 +910,12 @@ Status DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
   if (status.ok()) {
     auto stats = default_cf_internal_stats_;
     if (need_log_sync) {
-      stats->AddDBStats(InternalStats::WAL_FILE_SYNCED, 1);
+      stats->AddDBStats(InternalStats::kIntStatsWalFileSynced, 1);
       RecordTick(stats_, WAL_FILE_SYNCED);
     }
-    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size);
+    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size);
     RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal);
+    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal);
     RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
   }
   return status;
@@ -959,9 +961,10 @@ Status DBImpl::ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
   if (status.ok()) {
     const bool concurrent = true;
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size, concurrent);
+    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size,
+                      concurrent);
     RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal,
+    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal,
                       concurrent);
     RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
   }
@@ -1255,8 +1258,8 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
   }
   assert(!delayed || !write_options.no_slowdown);
   if (delayed) {
-    default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_STALL_MICROS,
+    default_cf_internal_stats_->AddDBStats(
-                                           time_delayed);
+        InternalStats::kIntStatsWriteStallMicros, time_delayed);
     RecordTick(stats_, STALL_MICROS, time_delayed);
   }
 

db/db_iter.cc

@@ -171,7 +171,7 @@ class DBIter final: public Iterator {
     iter_ = iter;
     iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
   }
-  virtual RangeDelAggregatorV2* GetRangeDelAggregator() {
+  virtual ReadRangeDelAggregator* GetRangeDelAggregator() {
     return &range_del_agg_;
   }
 
@@ -341,7 +341,7 @@ class DBIter final: public Iterator {
   const bool total_order_seek_;
   // List of operands for merge operator.
   MergeContext merge_context_;
-  RangeDelAggregatorV2 range_del_agg_;
+  ReadRangeDelAggregator range_del_agg_;
   LocalStatistics local_stats_;
   PinnedIteratorsManager pinned_iters_mgr_;
   ReadCallback* read_callback_;
@@ -1479,7 +1479,7 @@ Iterator* NewDBIterator(Env* env, const ReadOptions& read_options,
 
 ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
 
-RangeDelAggregatorV2* ArenaWrappedDBIter::GetRangeDelAggregator() {
+ReadRangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
   return db_iter_->GetRangeDelAggregator();
 }
 

db/db_iter.h

@@ -12,7 +12,7 @@
 #include <string>
 #include "db/db_impl.h"
 #include "db/dbformat.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "options/cf_options.h"
 #include "rocksdb/db.h"
 #include "rocksdb/iterator.h"
@@ -48,7 +48,7 @@ class ArenaWrappedDBIter : public Iterator {
   // Get the arena to be used to allocate memory for DBIter to be wrapped,
   // as well as child iterators in it.
   virtual Arena* GetArena() { return &arena_; }
-  virtual RangeDelAggregatorV2* GetRangeDelAggregator();
+  virtual ReadRangeDelAggregator* GetRangeDelAggregator();
 
   // Set the internal iterator wrapped inside the DB Iterator. Usually it is
   // a merging iterator.

db/db_memtable_test.cc

@@ -8,6 +8,7 @@
 
 #include "db/db_test_util.h"
 #include "db/memtable.h"
+#include "db/range_del_aggregator.h"
 #include "port/stack_trace.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/slice_transform.h"
@@ -135,7 +136,8 @@ TEST_F(DBMemTableTest, DuplicateSeq) {
   MergeContext merge_context;
   Options options;
   InternalKeyComparator ikey_cmp(options.comparator);
-  RangeDelAggregator range_del_agg(ikey_cmp, {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&ikey_cmp,
+                                       kMaxSequenceNumber /* upper_bound */);
 
   // Create a MemTable
   InternalKeyComparator cmp(BytewiseComparator());

db/db_range_del_test.cc

@@ -1041,11 +1041,16 @@ TEST_F(DBRangeDelTest, RangeTombstoneEndKeyAsSstableUpperBound) {
     //   L2:
     //     [key000000#1,1, key000000#1,1]
     //     [key000002#6,1, key000004#72057594037927935,15]
+    //
+    // At the same time, verify the compaction does not cause the key at the
+    // endpoint (key000002#6,1) to disappear.
+    ASSERT_EQ(value, Get(Key(2)));
     auto begin_str = Key(3);
     const rocksdb::Slice begin = begin_str;
     dbfull()->TEST_CompactRange(1, &begin, nullptr);
     ASSERT_EQ(1, NumTableFilesAtLevel(1));
     ASSERT_EQ(2, NumTableFilesAtLevel(2));
+    ASSERT_EQ(value, Get(Key(2)));
   }
 
   {

db/db_test_util.cc

@@ -814,7 +814,7 @@ std::string DBTestBase::AllEntriesFor(const Slice& user_key, int cf) {
   Arena arena;
   auto options = CurrentOptions();
   InternalKeyComparator icmp(options.comparator);
-  RangeDelAggregatorV2 range_del_agg(&icmp,
+  ReadRangeDelAggregator range_del_agg(&icmp,
                                        kMaxSequenceNumber /* upper_bound */);
   ScopedArenaIterator iter;
   if (cf == 0) {
@@ -1227,7 +1227,7 @@ void DBTestBase::validateNumberOfEntries(int numValues, int cf) {
   Arena arena;
   auto options = CurrentOptions();
   InternalKeyComparator icmp(options.comparator);
-  RangeDelAggregatorV2 range_del_agg(&icmp,
+  ReadRangeDelAggregator range_del_agg(&icmp,
                                        kMaxSequenceNumber /* upper_bound */);
   // This should be defined after range_del_agg so that it destructs the
   // assigned iterator before it range_del_agg is already destructed.
@@ -1437,7 +1437,7 @@ void DBTestBase::VerifyDBInternal(
     std::vector<std::pair<std::string, std::string>> true_data) {
   Arena arena;
   InternalKeyComparator icmp(last_options_.comparator);
-  RangeDelAggregatorV2 range_del_agg(&icmp,
+  ReadRangeDelAggregator range_del_agg(&icmp,
                                        kMaxSequenceNumber /* upper_bound */);
   auto iter =
       dbfull()->NewInternalIterator(&arena, &range_del_agg, kMaxSequenceNumber);

db/flush_job.cc

@@ -24,14 +24,15 @@
 #include "db/event_helpers.h"
 #include "db/log_reader.h"
 #include "db/log_writer.h"
+#include "db/memtable.h"
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
+#include "db/range_tombstone_fragmenter.h"
 #include "db/version_set.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/thread_status_util.h"
 #include "port/port.h"
-#include "db/memtable.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/statistics.h"
@@ -295,7 +296,8 @@ Status FlushJob::WriteLevel0Table() {
     // memtable and its associated range deletion memtable, respectively, at
     // corresponding indexes.
     std::vector<InternalIterator*> memtables;
-    std::vector<InternalIterator*> range_del_iters;
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+        range_del_iters;
     ReadOptions ro;
     ro.total_order_seek = true;
     Arena arena;
@@ -308,9 +310,9 @@ Status FlushJob::WriteLevel0Table() {
           cfd_->GetName().c_str(), job_context_->job_id, m->GetNextLogNumber());
       memtables.push_back(m->NewIterator(ro, &arena));
       auto* range_del_iter =
-          m->NewRangeTombstoneIterator(ro, versions_->LastSequence());
+          m->NewRangeTombstoneIterator(ro, kMaxSequenceNumber);
       if (range_del_iter != nullptr) {
-        range_del_iters.push_back(range_del_iter);
+        range_del_iters.emplace_back(range_del_iter);
       }
       total_num_entries += m->num_entries();
       total_num_deletes += m->num_deletes();
@@ -329,10 +331,6 @@ Status FlushJob::WriteLevel0Table() {
       ScopedArenaIterator iter(
           NewMergingIterator(&cfd_->internal_comparator(), &memtables[0],
                              static_cast<int>(memtables.size()), &arena));
-      std::unique_ptr<InternalIterator> range_del_iter(NewMergingIterator(
-          &cfd_->internal_comparator(),
-          range_del_iters.empty() ? nullptr : &range_del_iters[0],
-          static_cast<int>(range_del_iters.size())));
       ROCKS_LOG_INFO(db_options_.info_log,
                      "[%s] [JOB %d] Level-0 flush table #%" PRIu64 ": started",
                      cfd_->GetName().c_str(), job_context_->job_id,
@@ -358,7 +356,7 @@ Status FlushJob::WriteLevel0Table() {
       s = BuildTable(
           dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
           env_options_, cfd_->table_cache(), iter.get(),
-          std::move(range_del_iter), &meta_, cfd_->internal_comparator(),
+          std::move(range_del_iters), &meta_, cfd_->internal_comparator(),
           cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
           cfd_->GetName(), existing_snapshots_,
           earliest_write_conflict_snapshot_, snapshot_checker_,

db/flush_job_test.cc

@@ -279,7 +279,6 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
           *cfd->GetLatestMutableCFOptions(), kMaxSequenceNumber);
       mem->SetID(i);
       mem->Ref();
-      mem->TEST_AtomicFlushSequenceNumber() = 123;
 
       for (size_t j = 0; j != num_keys_per_memtable; ++j) {
         std::string key(ToString(j + i * num_keys_per_memtable));
@@ -309,7 +308,9 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
     k++;
   }
   HistogramData hist;
-  autovector<FileMetaData> file_metas;
+  std::vector<FileMetaData> file_metas;
+  // Call reserve to avoid auto-resizing
+  file_metas.reserve(flush_jobs.size());
   mutex_.Lock();
   for (auto& job : flush_jobs) {
     job.PickMemTable();
@@ -320,23 +321,23 @@ TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
     ASSERT_OK(job.Run(nullptr /**/, &meta));
     file_metas.emplace_back(meta);
   }
+  autovector<FileMetaData*> file_meta_ptrs;
+  for (auto& meta : file_metas) {
+    file_meta_ptrs.push_back(&meta);
+  }
   autovector<const autovector<MemTable*>*> mems_list;
   for (size_t i = 0; i != all_cfds.size(); ++i) {
     const auto& mems = flush_jobs[i].GetMemTables();
     mems_list.push_back(&mems);
   }
-  autovector<MemTableList*> imm_lists;
   autovector<const MutableCFOptions*> mutable_cf_options_list;
   for (auto cfd : all_cfds) {
-    imm_lists.push_back(cfd->imm());
     mutable_cf_options_list.push_back(cfd->GetLatestMutableCFOptions());
   }
 
-  bool atomic_flush_commit_in_progress = false;
+  Status s = InstallMemtableAtomicFlushResults(
-  Status s = MemTableList::TryInstallMemtableFlushResults(
+      nullptr /* imm_lists */, all_cfds, mutable_cf_options_list, mems_list,
-      imm_lists, all_cfds, mutable_cf_options_list, mems_list,
+      versions_.get(), &mutex_, file_meta_ptrs, &job_context.memtables_to_free,
-      &atomic_flush_commit_in_progress, nullptr /* logs_prep_tracker */,
-      versions_.get(), &mutex_, file_metas, &job_context.memtables_to_free,
       nullptr /* db_directory */, nullptr /* log_buffer */);
   ASSERT_OK(s);
 

db/forward_iterator.cc

@@ -15,7 +15,7 @@
 #include "db/db_iter.h"
 #include "db/dbformat.h"
 #include "db/job_context.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "db/range_tombstone_fragmenter.h"
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
@@ -73,7 +73,7 @@ class ForwardLevelIterator : public InternalIterator {
       delete file_iter_;
     }
 
-    RangeDelAggregatorV2 range_del_agg(&cfd_->internal_comparator(),
+    ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
                                          kMaxSequenceNumber /* upper_bound */);
     file_iter_ = cfd_->table_cache()->NewIterator(
         read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
@@ -610,7 +610,7 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
     // New
     sv_ = cfd_->GetReferencedSuperVersion(&(db_->mutex_));
   }
-  RangeDelAggregatorV2 range_del_agg(&cfd_->internal_comparator(),
+  ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
                                        kMaxSequenceNumber /* upper_bound */);
   mutable_iter_ = sv_->mem->NewIterator(read_options_, &arena_);
   sv_->imm->AddIterators(read_options_, &imm_iters_, &arena_);
@@ -669,7 +669,7 @@ void ForwardIterator::RenewIterators() {
 
   mutable_iter_ = svnew->mem->NewIterator(read_options_, &arena_);
   svnew->imm->AddIterators(read_options_, &imm_iters_, &arena_);
-  RangeDelAggregatorV2 range_del_agg(&cfd_->internal_comparator(),
+  ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
                                        kMaxSequenceNumber /* upper_bound */);
   if (!read_options_.ignore_range_deletions) {
     std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(

db/internal_stats.cc

@@ -949,14 +949,17 @@ void InternalStats::DumpDBStats(std::string* value) {
            seconds_up, interval_seconds_up);
   value->append(buf);
   // Cumulative
-  uint64_t user_bytes_written = GetDBStats(InternalStats::BYTES_WRITTEN);
+  uint64_t user_bytes_written =
-  uint64_t num_keys_written = GetDBStats(InternalStats::NUMBER_KEYS_WRITTEN);
+      GetDBStats(InternalStats::kIntStatsBytesWritten);
-  uint64_t write_other = GetDBStats(InternalStats::WRITE_DONE_BY_OTHER);
+  uint64_t num_keys_written =
-  uint64_t write_self = GetDBStats(InternalStats::WRITE_DONE_BY_SELF);
+      GetDBStats(InternalStats::kIntStatsNumKeysWritten);
-  uint64_t wal_bytes = GetDBStats(InternalStats::WAL_FILE_BYTES);
+  uint64_t write_other = GetDBStats(InternalStats::kIntStatsWriteDoneByOther);
-  uint64_t wal_synced = GetDBStats(InternalStats::WAL_FILE_SYNCED);
+  uint64_t write_self = GetDBStats(InternalStats::kIntStatsWriteDoneBySelf);
-  uint64_t write_with_wal = GetDBStats(InternalStats::WRITE_WITH_WAL);
+  uint64_t wal_bytes = GetDBStats(InternalStats::kIntStatsWalFileBytes);
-  uint64_t write_stall_micros = GetDBStats(InternalStats::WRITE_STALL_MICROS);
+  uint64_t wal_synced = GetDBStats(InternalStats::kIntStatsWalFileSynced);
+  uint64_t write_with_wal = GetDBStats(InternalStats::kIntStatsWriteWithWal);
+  uint64_t write_stall_micros =
+      GetDBStats(InternalStats::kIntStatsWriteStallMicros);
 
   const int kHumanMicrosLen = 32;
   char human_micros[kHumanMicrosLen];

db/internal_stats.h

@@ -108,15 +108,15 @@ class InternalStats {
   };
 
   enum InternalDBStatsType {
-    WAL_FILE_BYTES,
+    kIntStatsWalFileBytes,
-    WAL_FILE_SYNCED,
+    kIntStatsWalFileSynced,
-    BYTES_WRITTEN,
+    kIntStatsBytesWritten,
-    NUMBER_KEYS_WRITTEN,
+    kIntStatsNumKeysWritten,
-    WRITE_DONE_BY_OTHER,
+    kIntStatsWriteDoneByOther,
-    WRITE_DONE_BY_SELF,
+    kIntStatsWriteDoneBySelf,
-    WRITE_WITH_WAL,
+    kIntStatsWriteWithWal,
-    WRITE_STALL_MICROS,
+    kIntStatsWriteStallMicros,
-    INTERNAL_DB_STATS_ENUM_MAX,
+    kIntStatsNumMax,
   };
 
   InternalStats(int num_levels, Env* env, ColumnFamilyData* cfd)
@@ -292,7 +292,7 @@ class InternalStats {
   };
 
   void Clear() {
-    for (int i = 0; i < INTERNAL_DB_STATS_ENUM_MAX; i++) {
+    for (int i = 0; i < kIntStatsNumMax; i++) {
       db_stats_[i].store(0);
     }
     for (int i = 0; i < INTERNAL_CF_STATS_ENUM_MAX; i++) {
@@ -382,7 +382,7 @@ class InternalStats {
   bool HandleBlockCacheStat(Cache** block_cache);
 
   // Per-DB stats
-  std::atomic<uint64_t> db_stats_[INTERNAL_DB_STATS_ENUM_MAX];
+  std::atomic<uint64_t> db_stats_[kIntStatsNumMax];
   // Per-ColumnFamily stats
   uint64_t cf_stats_value_[INTERNAL_CF_STATS_ENUM_MAX];
   uint64_t cf_stats_count_[INTERNAL_CF_STATS_ENUM_MAX];
@@ -580,15 +580,15 @@ class InternalStats {
   };
 
   enum InternalDBStatsType {
-    WAL_FILE_BYTES,
+    kIntStatsWalFileBytes,
-    WAL_FILE_SYNCED,
+    kIntStatsWalFileSynced,
-    BYTES_WRITTEN,
+    kIntStatsBytesWritten,
-    NUMBER_KEYS_WRITTEN,
+    kIntStatsNumKeysWritten,
-    WRITE_DONE_BY_OTHER,
+    kIntStatsWriteDoneByOther,
-    WRITE_DONE_BY_SELF,
+    kIntStatsWriteDoneBySelf,
-    WRITE_WITH_WAL,
+    kIntStatsWriteWithWal,
-    WRITE_STALL_MICROS,
+    kIntStatsWriteStallMicros,
-    INTERNAL_DB_STATS_ENUM_MAX,
+    kIntStatsNumMax,
   };
 
   InternalStats(int /*num_levels*/, Env* /*env*/, ColumnFamilyData* /*cfd*/) {}

db/listener_test.cc

@@ -905,6 +905,7 @@ class TestFileOperationListener : public EventListener {
     if (info.status.ok()) {
       ++file_reads_success_;
     }
+    ReportDuration(info);
   }
 
   void OnFileWriteFinish(const FileOperationInfo& info) override {
@@ -912,6 +913,7 @@ class TestFileOperationListener : public EventListener {
     if (info.status.ok()) {
       ++file_writes_success_;
     }
+    ReportDuration(info);
   }
 
   bool ShouldBeNotifiedOnFileIO() override { return true; }
@@ -920,6 +922,13 @@ class TestFileOperationListener : public EventListener {
   std::atomic<size_t> file_reads_success_;
   std::atomic<size_t> file_writes_;
   std::atomic<size_t> file_writes_success_;
+
+ private:
+  void ReportDuration(const FileOperationInfo& info) const {
+    auto duration = std::chrono::duration_cast<std::chrono::nanoseconds>(
+        info.finish_timestamp - info.start_timestamp);
+    ASSERT_GT(duration.count(), 0);
+  }
 };
 
 TEST_F(EventListenerTest, OnFileOperationTest) {

db/malloc_stats.cc

@@ -13,17 +13,16 @@
 #include <memory>
 #include <string.h>
 
+#include "port/jemalloc_helper.h"
+
+
 namespace rocksdb {
 
 #ifdef ROCKSDB_JEMALLOC
-#ifdef __FreeBSD__
+
-#include <malloc_np.h>
-#else
-#include "jemalloc/jemalloc.h"
 #ifdef JEMALLOC_NO_RENAME
 #define malloc_stats_print je_malloc_stats_print
 #endif
-#endif
 
 typedef struct {
   char* cur;
@@ -41,10 +40,10 @@ static void GetJemallocStatus(void* mstat_arg, const char* status) {
   snprintf(mstat->cur, buf_size, "%s", status);
   mstat->cur += status_len;
 }
-#endif  // ROCKSDB_JEMALLOC
-
-#ifdef ROCKSDB_JEMALLOC
 void DumpMallocStats(std::string* stats) {
+  if (!HasJemalloc()) {
+    return;
+  }
   MallocStatus mstat;
   const unsigned int kMallocStatusLen = 1000000;
   std::unique_ptr<char[]> buf{new char[kMallocStatusLen + 1]};
@@ -56,5 +55,5 @@ void DumpMallocStats(std::string* stats) {
 #else
 void DumpMallocStats(std::string*) {}
 #endif  // ROCKSDB_JEMALLOC
-}
+}  // namespace rocksdb
 #endif  // !ROCKSDB_LITE

db/memtable.cc

@@ -428,7 +428,7 @@ FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIterator(
           comparator_.comparator);
 
   auto* fragmented_iter = new FragmentedRangeTombstoneIterator(
-      fragmented_tombstone_list, read_seq, comparator_.comparator);
+      fragmented_tombstone_list, comparator_.comparator, read_seq);
   return fragmented_iter;
 }
 

db/memtable.h

@@ -386,14 +386,16 @@ class MemTable {
 
   uint64_t GetID() const { return id_; }
 
-  SequenceNumber& TEST_AtomicFlushSequenceNumber() {
+  void SetFlushCompleted(bool completed) { flush_completed_ = completed; }
-    return atomic_flush_seqno_;
+
+  uint64_t GetFileNumber() const { return file_number_; }
+
+  void SetFileNumber(uint64_t file_num) { file_number_ = file_num; }
+
+  void SetFlushInProgress(bool in_progress) {
+    flush_in_progress_ = in_progress;
   }
 
-  void TEST_SetFlushCompleted(bool completed) { flush_completed_ = completed; }
-
-  void TEST_SetFileNumber(uint64_t file_num) { file_number_ = file_num; }
-
  private:
   enum FlushStateEnum { FLUSH_NOT_REQUESTED, FLUSH_REQUESTED, FLUSH_SCHEDULED };
 

db/memtable_list.cc

@@ -159,7 +159,7 @@ bool MemTableListVersion::GetFromList(
 
 Status MemTableListVersion::AddRangeTombstoneIterators(
     const ReadOptions& read_opts, Arena* /*arena*/,
-    RangeDelAggregatorV2* range_del_agg) {
+    RangeDelAggregator* range_del_agg) {
   assert(range_del_agg != nullptr);
   for (auto& m : memlist_) {
     // Using kMaxSequenceNumber is OK because these are immutable memtables.
@@ -260,228 +260,6 @@ void MemTableListVersion::TrimHistory(autovector<MemTable*>* to_delete) {
   }
 }
 
-// Try to record multiple successful flush to the MANIFEST as an atomic unit.
-// This function may just return Status::OK if there has already been
-// a concurrent thread performing actual recording.
-Status MemTableList::TryInstallMemtableFlushResults(
-    autovector<MemTableList*>& imm_lists,
-    const autovector<ColumnFamilyData*>& cfds,
-    const autovector<const MutableCFOptions*>& mutable_cf_options_list,
-    const autovector<const autovector<MemTable*>*>& mems_list,
-    bool* atomic_flush_commit_in_progress, LogsWithPrepTracker* prep_tracker,
-    VersionSet* vset, InstrumentedMutex* mu,
-    const autovector<FileMetaData>& file_metas,
-    autovector<MemTable*>* to_delete, Directory* db_directory,
-    LogBuffer* log_buffer) {
-  AutoThreadOperationStageUpdater stage_updater(
-      ThreadStatus::STAGE_MEMTABLE_INSTALL_FLUSH_RESULTS);
-  mu->AssertHeld();
-
-  for (size_t k = 0; k != mems_list.size(); ++k) {
-    for (size_t i = 0; i != mems_list[k]->size(); ++i) {
-      assert(i == 0 || (*mems_list[k])[i]->GetEdits()->NumEntries() == 0);
-      (*mems_list[k])[i]->flush_completed_ = true;
-      (*mems_list[k])[i]->file_number_ = file_metas[k].fd.GetNumber();
-    }
-  }
-
-  assert(atomic_flush_commit_in_progress != nullptr);
-  Status s;
-  if (*atomic_flush_commit_in_progress) {
-    // If the function reaches here, there must be a concurrent thread that
-    // have already started recording to MANIFEST. Therefore we should just
-    // return Status::OK and let the othe thread finish writing to MANIFEST on
-    // our behalf.
-    return s;
-  }
-
-  // If the function reaches here, the current thread will start writing to
-  // MANIFEST. It may record to MANIFEST the flush results of other flushes.
-  *atomic_flush_commit_in_progress = true;
-
-  auto comp = [&imm_lists](size_t lh, size_t rh) {
-    const auto& memlist1 = imm_lists[lh]->current_->memlist_;
-    const auto& memlist2 = imm_lists[rh]->current_->memlist_;
-    auto it1 = memlist1.rbegin();
-    auto it2 = memlist2.rbegin();
-    return (*it1)->atomic_flush_seqno_ > (*it2)->atomic_flush_seqno_;
-  };
-  // The top of the heap is the memtable with smallest atomic_flush_seqno_.
-  std::priority_queue<size_t, std::vector<size_t>, decltype(comp)> heap(comp);
-  // Sequence number of the oldest unfinished atomic flush.
-  SequenceNumber min_unfinished_seqno = kMaxSequenceNumber;
-  // Populate the heap with first element of each imm iff. it has been
-  // flushed to storage, i.e. flush_completed_ is true.
-  size_t num = imm_lists.size();
-  assert(num == cfds.size());
-  for (size_t i = 0; i != num; ++i) {
-    std::list<MemTable*>& memlist = imm_lists[i]->current_->memlist_;
-    if (memlist.empty()) {
-      continue;
-    }
-    auto it = memlist.rbegin();
-    if ((*it)->flush_completed_) {
-      heap.emplace(i);
-    } else if (min_unfinished_seqno > (*it)->atomic_flush_seqno_) {
-      min_unfinished_seqno = (*it)->atomic_flush_seqno_;
-    }
-  }
-
-  while (s.ok() && !heap.empty()) {
-    autovector<size_t> batch;
-    SequenceNumber seqno = kMaxSequenceNumber;
-    // Pop from the heap the memtables that belong to the same atomic flush,
-    // namely their atomic_flush_seqno_ are equal.
-    do {
-      size_t pos = heap.top();
-      const auto& memlist = imm_lists[pos]->current_->memlist_;
-      MemTable* mem = *(memlist.rbegin());
-      if (seqno == kMaxSequenceNumber) {
-        // First mem in this batch.
-        seqno = mem->atomic_flush_seqno_;
-        batch.emplace_back(pos);
-        heap.pop();
-      } else if (mem->atomic_flush_seqno_ == seqno) {
-        // mem has the same atomic_flush_seqno_, thus in the same atomic flush.
-        batch.emplace_back(pos);
-        heap.pop();
-      } else if (mem->atomic_flush_seqno_ > seqno) {
-        // mem belongs to another atomic flush with higher seqno, break the
-        // loop.
-        break;
-      }
-    } while (!heap.empty());
-    if (seqno >= min_unfinished_seqno) {
-      // If there is an older, unfinished atomic flush, then we should not
-      // proceed.
-      TEST_SYNC_POINT_CALLBACK(
-          "MemTableList::TryInstallMemtableFlushResults:"
-          "HasOlderUnfinishedAtomicFlush:0",
-          nullptr);
-      break;
-    }
-
-    // Found the earliest, complete atomic flush. No earlier atomic flush is
-    // pending. Therefore ready to record it to the MANIFEST.
-    uint32_t num_entries = 0;
-    autovector<ColumnFamilyData*> tmp_cfds;
-    autovector<const MutableCFOptions*> tmp_mutable_cf_options_list;
-    std::vector<autovector<MemTable*>> memtables_to_flush;
-    autovector<autovector<VersionEdit*>> edit_lists;
-    for (auto pos : batch) {
-      tmp_cfds.emplace_back(cfds[pos]);
-      tmp_mutable_cf_options_list.emplace_back(mutable_cf_options_list[pos]);
-      const auto& memlist = imm_lists[pos]->current_->memlist_;
-      uint64_t batch_file_number = 0;
-      autovector<MemTable*> tmp_mems;
-      autovector<VersionEdit*> edits;
-      for (auto it = memlist.rbegin(); it != memlist.rend(); ++it) {
-        MemTable* m = *it;
-        if (!m->flush_completed_ ||
-            (it != memlist.rbegin() && m->file_number_ != batch_file_number)) {
-          break;
-        }
-        if (it == memlist.rbegin()) {
-          batch_file_number = m->file_number_;
-          edits.push_back(m->GetEdits());
-          ++num_entries;
-        }
-        tmp_mems.push_back(m);
-      }
-      edit_lists.push_back(edits);
-      memtables_to_flush.push_back(tmp_mems);
-    }
-    TEST_SYNC_POINT_CALLBACK(
-        "MemTableList::TryInstallMemtableFlushResults:FoundBatchToCommit:0",
-        &num_entries);
-
-    // Mark the version edits as an atomic group
-    uint32_t remaining = num_entries;
-    for (auto& edit_list : edit_lists) {
-      assert(edit_list.size() == 1);
-      edit_list[0]->MarkAtomicGroup(--remaining);
-    }
-    assert(remaining == 0);
-
-    size_t batch_sz = batch.size();
-    assert(batch_sz > 0);
-    assert(batch_sz == memtables_to_flush.size());
-    assert(batch_sz == tmp_cfds.size());
-    assert(batch_sz == edit_lists.size());
-
-    if (vset->db_options()->allow_2pc) {
-      for (size_t i = 0; i != batch_sz; ++i) {
-        auto& edit_list = edit_lists[i];
-        assert(!edit_list.empty());
-        edit_list.back()->SetMinLogNumberToKeep(
-            PrecomputeMinLogNumberToKeep(vset, *tmp_cfds[i], edit_list,
-                                         memtables_to_flush[i], prep_tracker));
-      }
-    }
-    // this can release and reacquire the mutex.
-    s = vset->LogAndApply(tmp_cfds, tmp_mutable_cf_options_list, edit_lists, mu,
-                          db_directory);
-
-    for (const auto pos : batch) {
-      imm_lists[pos]->InstallNewVersion();
-    }
-
-    if (s.ok()) {
-      for (size_t i = 0; i != batch_sz; ++i) {
-        if (tmp_cfds[i]->IsDropped()) {
-          continue;
-        }
-        size_t pos = batch[i];
-        for (auto m : memtables_to_flush[i]) {
-          assert(m->file_number_ > 0);
-          uint64_t mem_id = m->GetID();
-          ROCKS_LOG_BUFFER(log_buffer,
-                           "[%s] Level-0 commit table #%" PRIu64
-                           ": memtable #%" PRIu64 " done",
-                           tmp_cfds[i]->GetName().c_str(), m->file_number_,
-                           mem_id);
-          imm_lists[pos]->current_->Remove(m, to_delete);
-        }
-      }
-    } else {
-      for (size_t i = 0; i != batch_sz; ++i) {
-        size_t pos = batch[i];
-        for (auto m : memtables_to_flush[i]) {
-          uint64_t mem_id = m->GetID();
-          ROCKS_LOG_BUFFER(log_buffer,
-                           "[%s] Level-0 commit table #%" PRIu64
-                           ": memtable #%" PRIu64 " failed",
-                           tmp_cfds[i]->GetName().c_str(), m->file_number_,
-                           mem_id);
-          m->flush_completed_ = false;
-          m->flush_in_progress_ = false;
-          m->edit_.Clear();
-          m->file_number_ = 0;
-          imm_lists[pos]->num_flush_not_started_++;
-        }
-        imm_lists[pos]->imm_flush_needed.store(true, std::memory_order_release);
-      }
-    }
-    // Adjust the heap AFTER installing new MemTableListVersions because the
-    // compare function 'comp' needs to capture the most up-to-date state of
-    // imm_lists.
-    for (auto pos : batch) {
-      const auto& memlist = imm_lists[pos]->current_->memlist_;
-      if (!memlist.empty()) {
-        MemTable* mem = *(memlist.rbegin());
-        if (mem->flush_completed_) {
-          heap.emplace(pos);
-        } else if (min_unfinished_seqno > mem->atomic_flush_seqno_) {
-          min_unfinished_seqno = mem->atomic_flush_seqno_;
-        }
-      }
-    }
-  }
-
-  *atomic_flush_commit_in_progress = false;
-  return s;
-}
-
 // Returns true if there is at least one memtable on which flush has
 // not yet started.
 bool MemTableList::IsFlushPending() const {
@@ -749,4 +527,106 @@ uint64_t MemTableList::PrecomputeMinLogContainingPrepSection(
   return min_log;
 }
 
+// Commit a successful atomic flush in the manifest file.
+Status InstallMemtableAtomicFlushResults(
+    const autovector<MemTableList*>* imm_lists,
+    const autovector<ColumnFamilyData*>& cfds,
+    const autovector<const MutableCFOptions*>& mutable_cf_options_list,
+    const autovector<const autovector<MemTable*>*>& mems_list, VersionSet* vset,
+    InstrumentedMutex* mu, const autovector<FileMetaData*>& file_metas,
+    autovector<MemTable*>* to_delete, Directory* db_directory,
+    LogBuffer* log_buffer) {
+  AutoThreadOperationStageUpdater stage_updater(
+      ThreadStatus::STAGE_MEMTABLE_INSTALL_FLUSH_RESULTS);
+  mu->AssertHeld();
+
+  size_t num = mems_list.size();
+  assert(cfds.size() == num);
+  if (imm_lists != nullptr) {
+    assert(imm_lists->size() == num);
+  }
+  for (size_t k = 0; k != num; ++k) {
+#ifndef NDEBUG
+    const auto* imm =
+        (imm_lists == nullptr) ? cfds[k]->imm() : imm_lists->at(k);
+    if (!mems_list[k]->empty()) {
+      assert((*mems_list[k])[0]->GetID() == imm->GetEarliestMemTableID());
+    }
+#endif
+    assert(nullptr != file_metas[k]);
+    for (size_t i = 0; i != mems_list[k]->size(); ++i) {
+      assert(i == 0 || (*mems_list[k])[i]->GetEdits()->NumEntries() == 0);
+      (*mems_list[k])[i]->SetFlushCompleted(true);
+      (*mems_list[k])[i]->SetFileNumber(file_metas[k]->fd.GetNumber());
+    }
+  }
+
+  Status s;
+
+  autovector<autovector<VersionEdit*>> edit_lists;
+  uint32_t num_entries = 0;
+  for (const auto mems : mems_list) {
+    assert(mems != nullptr);
+    autovector<VersionEdit*> edits;
+    assert(!mems->empty());
+    edits.emplace_back((*mems)[0]->GetEdits());
+    ++num_entries;
+    edit_lists.emplace_back(edits);
+  }
+  // Mark the version edits as an atomic group
+  for (auto& edits : edit_lists) {
+    assert(edits.size() == 1);
+    edits[0]->MarkAtomicGroup(--num_entries);
+  }
+  assert(0 == num_entries);
+
+  // this can release and reacquire the mutex.
+  s = vset->LogAndApply(cfds, mutable_cf_options_list, edit_lists, mu,
+                        db_directory);
+
+  for (size_t k = 0; k != cfds.size(); ++k) {
+    auto* imm = (imm_lists == nullptr) ? cfds[k]->imm() : imm_lists->at(k);
+    imm->InstallNewVersion();
+  }
+
+  if (s.ok() || s.IsShutdownInProgress()) {
+    for (size_t i = 0; i != cfds.size(); ++i) {
+      if (cfds[i]->IsDropped()) {
+        continue;
+      }
+      auto* imm = (imm_lists == nullptr) ? cfds[i]->imm() : imm_lists->at(i);
+      for (auto m : *mems_list[i]) {
+        assert(m->GetFileNumber() > 0);
+        uint64_t mem_id = m->GetID();
+        ROCKS_LOG_BUFFER(log_buffer,
+                         "[%s] Level-0 commit table #%" PRIu64
+                         ": memtable #%" PRIu64 " done",
+                         cfds[i]->GetName().c_str(), m->GetFileNumber(),
+                         mem_id);
+        imm->current_->Remove(m, to_delete);
+      }
+    }
+  } else {
+    for (size_t i = 0; i != cfds.size(); ++i) {
+      auto* imm = (imm_lists == nullptr) ? cfds[i]->imm() : imm_lists->at(i);
+      for (auto m : *mems_list[i]) {
+        uint64_t mem_id = m->GetID();
+        ROCKS_LOG_BUFFER(log_buffer,
+                         "[%s] Level-0 commit table #%" PRIu64
+                         ": memtable #%" PRIu64 " failed",
+                         cfds[i]->GetName().c_str(), m->GetFileNumber(),
+                         mem_id);
+        m->SetFlushCompleted(false);
+        m->SetFlushInProgress(false);
+        m->GetEdits()->Clear();
+        m->SetFileNumber(0);
+        imm->num_flush_not_started_++;
+      }
+      imm->imm_flush_needed.store(true, std::memory_order_release);
+    }
+  }
+
+  return s;
+}
+
 }  // namespace rocksdb

db/memtable_list.h

@@ -15,7 +15,7 @@
 #include "db/dbformat.h"
 #include "db/logs_with_prep_tracker.h"
 #include "db/memtable.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "monitoring/instrumented_mutex.h"
 #include "rocksdb/db.h"
 #include "rocksdb/iterator.h"
@@ -31,6 +31,7 @@ class ColumnFamilyData;
 class InternalKeyComparator;
 class InstrumentedMutex;
 class MergeIteratorBuilder;
+class MemTableList;
 
 // keeps a list of immutable memtables in a vector. the list is immutable
 // if refcount is bigger than one. It is used as a state for Get() and
@@ -91,7 +92,7 @@ class MemTableListVersion {
   }
 
   Status AddRangeTombstoneIterators(const ReadOptions& read_opts, Arena* arena,
-                                    RangeDelAggregatorV2* range_del_agg);
+                                    RangeDelAggregator* range_del_agg);
 
   void AddIterators(const ReadOptions& options,
                     std::vector<InternalIterator*>* iterator_list,
@@ -114,6 +115,18 @@ class MemTableListVersion {
   SequenceNumber GetEarliestSequenceNumber(bool include_history = false) const;
 
  private:
+  friend class MemTableList;
+
+  friend Status InstallMemtableAtomicFlushResults(
+      const autovector<MemTableList*>* imm_lists,
+      const autovector<ColumnFamilyData*>& cfds,
+      const autovector<const MutableCFOptions*>& mutable_cf_options_list,
+      const autovector<const autovector<MemTable*>*>& mems_list,
+      VersionSet* vset, InstrumentedMutex* mu,
+      const autovector<FileMetaData*>& file_meta,
+      autovector<MemTable*>* to_delete, Directory* db_directory,
+      LogBuffer* log_buffer);
+
   // REQUIRE: m is an immutable memtable
   void Add(MemTable* m, autovector<MemTable*>* to_delete);
   // REQUIRE: m is an immutable memtable
@@ -132,8 +145,6 @@ class MemTableListVersion {
 
   void UnrefMemTable(autovector<MemTable*>* to_delete, MemTable* m);
 
-  friend class MemTableList;
-
   // Immutable MemTables that have not yet been flushed.
   std::list<MemTable*> memlist_;
 
@@ -163,18 +174,6 @@ class MemTableListVersion {
 // write thread.)
 class MemTableList {
  public:
-  // Commit a successful atomic flush in the manifest file
-  static Status TryInstallMemtableFlushResults(
-      autovector<MemTableList*>& imm_lists,
-      const autovector<ColumnFamilyData*>& cfds,
-      const autovector<const MutableCFOptions*>& mutable_cf_options_list,
-      const autovector<const autovector<MemTable*>*>& mems_list,
-      bool* atomic_flush_commit_in_progress, LogsWithPrepTracker* prep_tracker,
-      VersionSet* vset, InstrumentedMutex* mu,
-      const autovector<FileMetaData>& file_meta,
-      autovector<MemTable*>* to_delete, Directory* db_directory,
-      LogBuffer* log_buffer);
-
   // A list of memtables.
   explicit MemTableList(int min_write_buffer_number_to_merge,
                         int max_write_buffer_number_to_maintain)
@@ -296,6 +295,16 @@ class MemTableList {
   }
 
  private:
+  friend Status InstallMemtableAtomicFlushResults(
+      const autovector<MemTableList*>* imm_lists,
+      const autovector<ColumnFamilyData*>& cfds,
+      const autovector<const MutableCFOptions*>& mutable_cf_options_list,
+      const autovector<const autovector<MemTable*>*>& mems_list,
+      VersionSet* vset, InstrumentedMutex* mu,
+      const autovector<FileMetaData*>& file_meta,
+      autovector<MemTable*>* to_delete, Directory* db_directory,
+      LogBuffer* log_buffer);
+
   // DB mutex held
   void InstallNewVersion();
 
@@ -317,4 +326,18 @@ class MemTableList {
   size_t current_memory_usage_;
 };
 
+// Installs memtable atomic flush results.
+// In most cases, imm_lists is nullptr, and the function simply uses the
+// immutable memtable lists associated with the cfds. There are unit tests that
+// installs flush results for external immutable memtable lists other than the
+// cfds' own immutable memtable lists, e.g. MemTableLIstTest. In this case,
+// imm_lists parameter is not nullptr.
+extern Status InstallMemtableAtomicFlushResults(
+    const autovector<MemTableList*>* imm_lists,
+    const autovector<ColumnFamilyData*>& cfds,
+    const autovector<const MutableCFOptions*>& mutable_cf_options_list,
+    const autovector<const autovector<MemTable*>*>& mems_list, VersionSet* vset,
+    InstrumentedMutex* mu, const autovector<FileMetaData*>& file_meta,
+    autovector<MemTable*>* to_delete, Directory* db_directory,
+    LogBuffer* log_buffer);
 }  // namespace rocksdb

db/memtable_list_test.cc

@@ -8,7 +8,6 @@
 #include <string>
 #include <vector>
 #include "db/merge_context.h"
-#include "db/range_del_aggregator.h"
 #include "db/version_set.h"
 #include "db/write_controller.h"
 #include "rocksdb/db.h"
@@ -86,17 +85,46 @@ class MemTableListTest : public testing::Test {
   Status Mock_InstallMemtableFlushResults(
       MemTableList* list, const MutableCFOptions& mutable_cf_options,
       const autovector<MemTable*>& m, autovector<MemTable*>* to_delete) {
-    autovector<MemTableList*> lists;
+    // Create a mock Logger
-    lists.emplace_back(list);
+    test::NullLogger logger;
-    autovector<const autovector<MemTable*>*> mems_list;
+    LogBuffer log_buffer(DEBUG_LEVEL, &logger);
-    mems_list.emplace_back(&m);
+
-    return Mock_InstallMemtableFlushResults(
+    CreateDB();
-        lists, {0} /* cf_ids */, {&mutable_cf_options}, mems_list, to_delete);
+    // Create a mock VersionSet
+    DBOptions db_options;
+    ImmutableDBOptions immutable_db_options(db_options);
+    EnvOptions env_options;
+    std::shared_ptr<Cache> table_cache(NewLRUCache(50000, 16));
+    WriteBufferManager write_buffer_manager(db_options.db_write_buffer_size);
+    WriteController write_controller(10000000u);
+
+    VersionSet versions(dbname, &immutable_db_options, env_options,
+                        table_cache.get(), &write_buffer_manager,
+                        &write_controller);
+    std::vector<ColumnFamilyDescriptor> cf_descs;
+    cf_descs.emplace_back(kDefaultColumnFamilyName, ColumnFamilyOptions());
+    cf_descs.emplace_back("one", ColumnFamilyOptions());
+    cf_descs.emplace_back("two", ColumnFamilyOptions());
+
+    EXPECT_OK(versions.Recover(cf_descs, false));
+
+    // Create mock default ColumnFamilyData
+    auto column_family_set = versions.GetColumnFamilySet();
+    LogsWithPrepTracker dummy_prep_tracker;
+    auto cfd = column_family_set->GetDefault();
+    EXPECT_TRUE(nullptr != cfd);
+    uint64_t file_num = file_number.fetch_add(1);
+    // Create dummy mutex.
+    InstrumentedMutex mutex;
+    InstrumentedMutexLock l(&mutex);
+    return list->TryInstallMemtableFlushResults(
+        cfd, mutable_cf_options, m, &dummy_prep_tracker, &versions, &mutex,
+        file_num, to_delete, nullptr, &log_buffer);
   }
 
   // Calls MemTableList::InstallMemtableFlushResults() and sets up all
   // structures needed to call this function.
-  Status Mock_InstallMemtableFlushResults(
+  Status Mock_InstallMemtableAtomicFlushResults(
       autovector<MemTableList*>& lists, const autovector<uint32_t>& cf_ids,
       const autovector<const MutableCFOptions*>& mutable_cf_options_list,
       const autovector<const autovector<MemTable*>*>& mems_list,
@@ -128,44 +156,28 @@ class MemTableListTest : public testing::Test {
     auto column_family_set = versions.GetColumnFamilySet();
 
     LogsWithPrepTracker dummy_prep_tracker;
-    if (1 == cf_ids.size()) {
-      auto cfd = column_family_set->GetColumnFamily(cf_ids[0]);
-      EXPECT_TRUE(nullptr != cfd);
-      EXPECT_EQ(1, lists.size());
-      MemTableList* list = lists[0];
-      EXPECT_EQ(1, mutable_cf_options_list.size());
-      const MutableCFOptions& mutable_cf_options =
-          *(mutable_cf_options_list.at(0));
-      const autovector<MemTable*>* mems = mems_list.at(0);
-      EXPECT_TRUE(nullptr != mems);
-
-      uint64_t file_num = file_number.fetch_add(1);
-      // Create dummy mutex.
-      InstrumentedMutex mutex;
-      InstrumentedMutexLock l(&mutex);
-      return list->TryInstallMemtableFlushResults(
-          cfd, mutable_cf_options, *mems, &dummy_prep_tracker, &versions,
-          &mutex, file_num, to_delete, nullptr, &log_buffer);
-    }
     autovector<ColumnFamilyData*> cfds;
     for (int i = 0; i != static_cast<int>(cf_ids.size()); ++i) {
       cfds.emplace_back(column_family_set->GetColumnFamily(cf_ids[i]));
       EXPECT_NE(nullptr, cfds[i]);
     }
-    autovector<FileMetaData> file_metas;
+    std::vector<FileMetaData> file_metas;
+    file_metas.reserve(cf_ids.size());
     for (size_t i = 0; i != cf_ids.size(); ++i) {
       FileMetaData meta;
       uint64_t file_num = file_number.fetch_add(1);
       meta.fd = FileDescriptor(file_num, 0, 0);
       file_metas.emplace_back(meta);
     }
-    bool atomic_flush_commit_in_progress = false;
+    autovector<FileMetaData*> file_meta_ptrs;
+    for (auto& meta : file_metas) {
+      file_meta_ptrs.push_back(&meta);
+    }
     InstrumentedMutex mutex;
     InstrumentedMutexLock l(&mutex);
-    return MemTableList::TryInstallMemtableFlushResults(
+    return InstallMemtableAtomicFlushResults(
-        lists, cfds, mutable_cf_options_list, mems_list,
+        &lists, cfds, mutable_cf_options_list, mems_list, &versions, &mutex,
-        &atomic_flush_commit_in_progress, &dummy_prep_tracker, &versions,
+        file_meta_ptrs, to_delete, nullptr, &log_buffer);
-        &mutex, file_metas, to_delete, nullptr, &log_buffer);
   }
 };
 
@@ -731,18 +743,28 @@ TEST_F(MemTableListTest, FlushPendingTest) {
   to_delete.clear();
 }
 
-TEST_F(MemTableListTest, FlushMultipleCFsTest) {
+TEST_F(MemTableListTest, EmptyAtomicFlusTest) {
+  autovector<MemTableList*> lists;
+  autovector<uint32_t> cf_ids;
+  autovector<const MutableCFOptions*> options_list;
+  autovector<const autovector<MemTable*>*> to_flush;
+  autovector<MemTable*> to_delete;
+  Status s = Mock_InstallMemtableAtomicFlushResults(lists, cf_ids, options_list,
+                                                    to_flush, &to_delete);
+  ASSERT_OK(s);
+  ASSERT_TRUE(to_delete.empty());
+}
+
+TEST_F(MemTableListTest, AtomicFlusTest) {
   const int num_cfs = 3;
-  const int num_tables_per_cf = 5;
+  const int num_tables_per_cf = 2;
   SequenceNumber seq = 1;
-  Status s;
 
   auto factory = std::make_shared<SkipListFactory>();
   options.memtable_factory = factory;
   ImmutableCFOptions ioptions(options);
   InternalKeyComparator cmp(BytewiseComparator());
   WriteBufferManager wb(options.db_write_buffer_size);
-  autovector<MemTable*> to_delete;
 
   // Create MemTableLists
   int min_write_buffer_number_to_merge = 3;
@@ -783,135 +805,72 @@ TEST_F(MemTableListTest, FlushMultipleCFsTest) {
   std::vector<autovector<MemTable*>> flush_candidates(num_cfs);
 
   // Nothing to flush
-  for (int i = 0; i != num_cfs; ++i) {
+  for (auto i = 0; i != num_cfs; ++i) {
-    auto list = lists[i];
+    auto* list = lists[i];
     ASSERT_FALSE(list->IsFlushPending());
     ASSERT_FALSE(list->imm_flush_needed.load(std::memory_order_acquire));
     list->PickMemtablesToFlush(nullptr /* memtable_id */, &flush_candidates[i]);
-    ASSERT_EQ(0, static_cast<int>(flush_candidates[i].size()));
+    ASSERT_EQ(0, flush_candidates[i].size());
   }
-
   // Request flush even though there is nothing to flush
-  for (int i = 0; i != num_cfs; ++i) {
+  for (auto i = 0; i != num_cfs; ++i) {
-    auto list = lists[i];
+    auto* list = lists[i];
     list->FlushRequested();
     ASSERT_FALSE(list->IsFlushPending());
     ASSERT_FALSE(list->imm_flush_needed.load(std::memory_order_acquire));
   }
-
+  autovector<MemTable*> to_delete;
-  // Add tables to column families
+  // Add tables to the immutable memtalbe lists associated with column families
-  for (int i = 0; i != num_cfs; ++i) {
+  for (auto i = 0; i != num_cfs; ++i) {
-    for (int j = 0; j != num_tables_per_cf; ++j) {
+    for (auto j = 0; j != num_tables_per_cf; ++j) {
       lists[i]->Add(tables[i][j], &to_delete);
     }
     ASSERT_EQ(num_tables_per_cf, lists[i]->NumNotFlushed());
     ASSERT_TRUE(lists[i]->IsFlushPending());
     ASSERT_TRUE(lists[i]->imm_flush_needed.load(std::memory_order_acquire));
   }
-
+  std::vector<uint64_t> flush_memtable_ids = {1, 1, 0};
-  autovector<const autovector<MemTable*>*> to_flush;
-  std::vector<uint64_t> prev_memtable_ids;
-  // For each column family, determine the memtables to flush
-  for (int k = 0; k != 4; ++k) {
-    std::vector<uint64_t> flush_memtable_ids;
-    if (0 == k) {
   //          +----+
-      // list[0]: |0  1|  2 3 4
+  // list[0]: |0  1|
-      // list[1]: |0  1|  2 3 4
+  // list[1]: |0  1|
   //          | +--+
-      // list[2]: |0| 1   2 3 4
+  // list[2]: |0| 1
   //          +-+
-      flush_memtable_ids = {1, 1, 0};
-    } else if (1 == k) {
-      //          +----+ +---+
-      // list[0]: |0  1| |2 3|  4
-      // list[1]: |0  1| |2 3|  4
-      //          | +--+ +---+
-      // list[2]: |0| 1   2 3   4
-      //          +-+
-      flush_memtable_ids = {3, 3, 0};
-    } else if (2 == k) {
-      //          +-----+ +---+
-      // list[0]: |0   1| |2 3|  4
-      // list[1]: |0   1| |2 3|  4
-      //          | +---+ +---+
-      //          | | +-------+
-      // list[2]: |0| |1   2 3|  4
-      //          +-+ +-------+
-      flush_memtable_ids = {3, 3, 3};
-    } else {
-      //          +-----+ +---+ +-+
-      // list[0]: |0   1| |2 3| |4|
-      // list[1]: |0   1| |2 3| |4|
-      //          | +---+ +---+ | |
-      //          | | +-------+ | |
-      // list[2]: |0| |1   2 3| |4|
-      //          +-+ +-------+ +-+
-      flush_memtable_ids = {4, 4, 4};
-    }
-    assert(num_cfs == static_cast<int>(flush_memtable_ids.size()));
-
   // Pick memtables to flush
-    for (int i = 0; i != num_cfs; ++i) {
+  for (auto i = 0; i != num_cfs; ++i) {
     flush_candidates[i].clear();
     lists[i]->PickMemtablesToFlush(&flush_memtable_ids[i],
                                    &flush_candidates[i]);
-      for (auto mem : flush_candidates[i]) {
+    ASSERT_EQ(flush_memtable_ids[i] - 0 + 1,
-        mem->TEST_AtomicFlushSequenceNumber() = SequenceNumber(k);
+              static_cast<uint64_t>(flush_candidates[i].size()));
   }
-      if (prev_memtable_ids.empty()) {
+  autovector<MemTableList*> tmp_lists;
-        ASSERT_EQ(flush_memtable_ids[i] - 0 + 1, flush_candidates[i].size());
+  autovector<uint32_t> tmp_cf_ids;
-      } else {
+  autovector<const MutableCFOptions*> tmp_options_list;
-        ASSERT_EQ(flush_memtable_ids[i] - prev_memtable_ids[i],
+  autovector<const autovector<MemTable*>*> to_flush;
-                  flush_candidates[i].size());
+  for (auto i = 0; i != num_cfs; ++i) {
-      }
+    if (!flush_candidates[i].empty()) {
-      ASSERT_EQ(num_tables_per_cf, lists[i]->NumNotFlushed());
+      to_flush.push_back(&flush_candidates[i]);
-      ASSERT_FALSE(lists[i]->HasFlushRequested());
+      tmp_lists.push_back(lists[i]);
-      if (flush_memtable_ids[i] == num_tables_per_cf - 1) {
+      tmp_cf_ids.push_back(i);
-        ASSERT_FALSE(
+      tmp_options_list.push_back(mutable_cf_options_list[i]);
-            lists[i]->imm_flush_needed.load(std::memory_order_acquire));
-      } else {
-        ASSERT_TRUE(lists[i]->imm_flush_needed.load(std::memory_order_acquire));
     }
   }
-    prev_memtable_ids = flush_memtable_ids;
+  Status s = Mock_InstallMemtableAtomicFlushResults(
-
+      tmp_lists, tmp_cf_ids, tmp_options_list, to_flush, &to_delete);
-    if (k < 3) {
-      for (const auto& mems : flush_candidates) {
-        uint64_t file_num = file_number.fetch_add(1);
-        for (auto m : mems) {
-          m->TEST_SetFlushCompleted(true);
-          m->TEST_SetFileNumber(file_num);
-        }
-      }
-    }
-
-    if (k == 0) {
-      // Rollback first pick of tables
-      for (int i = 0; i != num_cfs; ++i) {
-        auto list = lists[i];
-        const auto& mems = flush_candidates[i];
-        for (auto m : mems) {
-          m->TEST_SetFileNumber(0);
-        }
-        list->RollbackMemtableFlush(flush_candidates[i], 0);
-        ASSERT_TRUE(list->IsFlushPending());
-        ASSERT_TRUE(list->imm_flush_needed.load(std::memory_order_acquire));
-      }
-      prev_memtable_ids.clear();
-    }
-
-    if (k == 3) {
-      for (int i = 0; i != num_cfs; ++i) {
-        to_flush.emplace_back(&flush_candidates[i]);
-      }
-    }
-  }
-
-  s = Mock_InstallMemtableFlushResults(lists, cf_ids, mutable_cf_options_list,
-                                       to_flush, &to_delete);
   ASSERT_OK(s);
 
+  for (auto i = 0; i != num_cfs; ++i) {
+    for (auto j = 0; j != num_tables_per_cf; ++j) {
+      if (static_cast<uint64_t>(j) <= flush_memtable_ids[i]) {
+        ASSERT_LT(0, tables[i][j]->GetFileNumber());
+      }
+    }
+    ASSERT_EQ(
+        static_cast<size_t>(num_tables_per_cf) - flush_candidates[i].size(),
+        lists[i]->NumNotFlushed());
+  }
+
   to_delete.clear();
   for (auto list : lists) {
     list->current()->Unref(&to_delete);
@@ -933,126 +892,6 @@ TEST_F(MemTableListTest, FlushMultipleCFsTest) {
     ASSERT_EQ(m, m->Unref());
     delete m;
   }
-  to_delete.clear();
-}
-
-TEST_F(MemTableListTest, HasOlderAtomicFlush) {
-  const size_t num_cfs = 3;
-  const size_t num_memtables_per_cf = 2;
-  SequenceNumber seq = 1;
-  Status s;
-
-  auto factory = std::make_shared<SkipListFactory>();
-  options.memtable_factory = factory;
-  ImmutableCFOptions ioptions(options);
-  InternalKeyComparator cmp(BytewiseComparator());
-  WriteBufferManager wb(options.db_write_buffer_size);
-  autovector<MemTable*> to_delete;
-
-  // Create MemTableLists
-  int min_write_buffer_number_to_merge = 3;
-  int max_write_buffer_number_to_maintain = 7;
-  autovector<MemTableList*> lists;
-  for (size_t i = 0; i != num_cfs; ++i) {
-    lists.emplace_back(new MemTableList(min_write_buffer_number_to_merge,
-                                        max_write_buffer_number_to_maintain));
-  }
-
-  autovector<uint32_t> cf_ids;
-  std::vector<std::vector<MemTable*>> tables;
-  autovector<const MutableCFOptions*> mutable_cf_options_list;
-  uint32_t cf_id = 0;
-  for (size_t k = 0; k != num_cfs; ++k) {
-    std::vector<MemTable*> elem;
-    mutable_cf_options_list.emplace_back(new MutableCFOptions(options));
-    uint64_t memtable_id = 0;
-    for (int i = 0; i != num_memtables_per_cf; ++i) {
-      MemTable* mem =
-          new MemTable(cmp, ioptions, *(mutable_cf_options_list.back()), &wb,
-                       kMaxSequenceNumber, cf_id);
-      mem->SetID(memtable_id++);
-      mem->Ref();
-
-      std::string value;
-
-      mem->Add(++seq, kTypeValue, "key1", ToString(i));
-      mem->Add(++seq, kTypeValue, "keyN" + ToString(i), "valueN");
-      mem->Add(++seq, kTypeValue, "keyX" + ToString(i), "value");
-      mem->Add(++seq, kTypeValue, "keyM" + ToString(i), "valueM");
-      mem->Add(++seq, kTypeDeletion, "keyX" + ToString(i), "");
-
-      elem.push_back(mem);
-    }
-    tables.emplace_back(elem);
-    cf_ids.push_back(cf_id++);
-  }
-
-  // Add tables to column families' immutable memtable lists
-  for (size_t i = 0; i != num_cfs; ++i) {
-    for (size_t j = 0; j != num_memtables_per_cf; ++j) {
-      lists[i]->Add(tables[i][j], &to_delete);
-    }
-    lists[i]->FlushRequested();
-    ASSERT_EQ(num_memtables_per_cf, lists[i]->NumNotFlushed());
-    ASSERT_TRUE(lists[i]->IsFlushPending());
-    ASSERT_TRUE(lists[i]->imm_flush_needed.load(std::memory_order_acquire));
-  }
-  std::vector<autovector<MemTable*>> flush_candidates(num_cfs);
-  for (size_t i = 0; i != num_cfs; ++i) {
-    lists[i]->PickMemtablesToFlush(nullptr, &flush_candidates[i]);
-    for (auto m : flush_candidates[i]) {
-      m->TEST_AtomicFlushSequenceNumber() = 123;
-    }
-    lists[i]->RollbackMemtableFlush(flush_candidates[i], 0);
-  }
-  uint64_t memtable_id = num_memtables_per_cf - 1;
-  autovector<MemTable*> other_flush_candidates;
-  lists[0]->PickMemtablesToFlush(&memtable_id, &other_flush_candidates);
-  for (auto m : other_flush_candidates) {
-    m->TEST_AtomicFlushSequenceNumber() = 124;
-    m->TEST_SetFlushCompleted(true);
-    m->TEST_SetFileNumber(1);
-  }
-  autovector<const autovector<MemTable*>*> to_flush;
-  to_flush.emplace_back(&other_flush_candidates);
-  bool has_older_unfinished_atomic_flush = false;
-  bool found_batch_to_commit = false;
-
-  SyncPoint::GetInstance()->SetCallBack(
-      "MemTableList::TryInstallMemtableFlushResults:"
-      "HasOlderUnfinishedAtomicFlush:0",
-      [&](void* /*arg*/) { has_older_unfinished_atomic_flush = true; });
-  SyncPoint::GetInstance()->SetCallBack(
-      "MemTableList::TryInstallMemtableFlushResults:FoundBatchToCommit:0",
-      [&](void* /*arg*/) { found_batch_to_commit = true; });
-  SyncPoint::GetInstance()->EnableProcessing();
-
-  s = Mock_InstallMemtableFlushResults(lists, cf_ids, mutable_cf_options_list,
-                                       to_flush, &to_delete);
-  ASSERT_OK(s);
-  ASSERT_TRUE(has_older_unfinished_atomic_flush);
-  ASSERT_FALSE(found_batch_to_commit);
-
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-
-  ASSERT_TRUE(to_delete.empty());
-  for (auto list : lists) {
-    list->current()->Unref(&to_delete);
-    delete list;
-  }
-  lists.clear();
-  ASSERT_EQ(num_cfs * num_memtables_per_cf, to_delete.size());
-  for (auto m : to_delete) {
-    m->Ref();
-    ASSERT_EQ(m, m->Unref());
-    delete m;
-  }
-  to_delete.clear();
-  for (auto& opts : mutable_cf_options_list) {
-    delete opts;
-    opts = nullptr;
-  }
-  mutable_cf_options_list.clear();
 }
 
 }  // namespace rocksdb

db/merge_context.h

@@ -79,7 +79,8 @@ class MergeContext {
     return GetOperandsDirectionForward();
   }
 
-  // Return all the operands in the order as they were merged (passed to FullMerge or FullMergeV2)
+  // Return all the operands in the order as they were merged (passed to
+  // FullMerge or FullMergeV2)
   const std::vector<Slice>& GetOperandsDirectionForward() {
     if (!operand_list_) {
       return empty_operand_list;
@@ -89,7 +90,8 @@ class MergeContext {
     return *operand_list_;
   }
 
-  // Return all the operands in the reversed order relative to how they were merged (passed to FullMerge or FullMergeV2)
+  // Return all the operands in the reversed order relative to how they were
+  // merged (passed to FullMerge or FullMergeV2)
   const std::vector<Slice>& GetOperandsDirectionBackward() {
     if (!operand_list_) {
       return empty_operand_list;

db/merge_helper.cc

@@ -110,8 +110,11 @@ Status MergeHelper::TimedFullMerge(const MergeOperator* merge_operator,
 //       keys_ stores the list of keys encountered while merging.
 //       operands_ stores the list of merge operands encountered while merging.
 //       keys_[i] corresponds to operands_[i] for each i.
+//
+// TODO: Avoid the snapshot stripe map lookup in CompactionRangeDelAggregator
+// and just pass the StripeRep corresponding to the stripe being merged.
 Status MergeHelper::MergeUntil(InternalIterator* iter,
-                               RangeDelAggregator* range_del_agg,
+                               CompactionRangeDelAggregator* range_del_agg,
                                const SequenceNumber stop_before,
                                const bool at_bottom) {
   // Get a copy of the internal key, before it's invalidated by iter->Next()

db/merge_helper.h

@@ -78,7 +78,7 @@ class MergeHelper {
   //
   // REQUIRED: The first key in the input is not corrupted.
   Status MergeUntil(InternalIterator* iter,
-                    RangeDelAggregator* range_del_agg = nullptr,
+                    CompactionRangeDelAggregator* range_del_agg = nullptr,
                     const SequenceNumber stop_before = 0,
                     const bool at_bottom = false);
 

db/range_del_aggregator.h

@@ -1,10 +1,12 @@
-//  Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
+//  Copyright (c) 2018-present, Facebook, Inc.  All rights reserved.
 //  This source code is licensed under both the GPLv2 (found in the
 //  COPYING file in the root directory) and Apache 2.0 License
 //  (found in the LICENSE.Apache file in the root directory).
 
 #pragma once
 
+#include <algorithm>
+#include <iterator>
 #include <list>
 #include <map>
 #include <set>
@@ -14,220 +16,416 @@
 #include "db/compaction_iteration_stats.h"
 #include "db/dbformat.h"
 #include "db/pinned_iterators_manager.h"
+#include "db/range_del_aggregator.h"
+#include "db/range_tombstone_fragmenter.h"
 #include "db/version_edit.h"
 #include "include/rocksdb/comparator.h"
 #include "include/rocksdb/types.h"
 #include "table/internal_iterator.h"
 #include "table/scoped_arena_iterator.h"
 #include "table/table_builder.h"
+#include "util/heap.h"
 #include "util/kv_map.h"
 
 namespace rocksdb {
 
-// RangeDelMaps maintain position across calls to ShouldDelete. The caller may
+class TruncatedRangeDelIterator {
-// wish to specify a mode to optimize positioning the iterator during the next
-// call to ShouldDelete. The non-kFullScan modes are only available when
-// deletion collapsing is enabled.
-//
-// For example, if we invoke Next() on an iterator, kForwardTraversal should be
-// specified to advance one-by-one through deletions until one is found with its
-// interval containing the key. This will typically be faster than doing a full
-// binary search (kBinarySearch).
-enum class RangeDelPositioningMode {
-  kFullScan,  // used iff collapse_deletions_ == false
-  kForwardTraversal,
-  kBackwardTraversal,
-  kBinarySearch,
-};
-
-// TruncatedRangeTombstones are a slight generalization of regular
-// RangeTombstones that can represent truncations caused by SST boundaries.
-// Instead of using user keys to represent the start and end keys, they instead
-// use internal keys, whose sequence number indicates the sequence number of
-// the smallest/largest SST key (in the case where a tombstone is untruncated,
-// the sequence numbers will be kMaxSequenceNumber for both start and end
-// keys). Like RangeTombstones, TruncatedRangeTombstone are also
-// end-key-exclusive.
-struct TruncatedRangeTombstone {
-  TruncatedRangeTombstone(const ParsedInternalKey& sk,
-                          const ParsedInternalKey& ek, SequenceNumber s)
-      : start_key_(sk), end_key_(ek), seq_(s) {}
-
-  RangeTombstone Tombstone() const {
-    // The RangeTombstone returned here can cover less than the
-    // TruncatedRangeTombstone when its end key has a seqnum that is not
-    // kMaxSequenceNumber. Since this method is only used by RangeDelIterators
-    // (which in turn are only used during flush/compaction), we avoid this
-    // problem by using truncation boundaries spanning multiple SSTs, which
-    // are selected in a way that guarantee a clean break at the end key.
-    assert(end_key_.sequence == kMaxSequenceNumber);
-    return RangeTombstone(start_key_.user_key, end_key_.user_key, seq_);
-  }
-
-  ParsedInternalKey start_key_;
-  ParsedInternalKey end_key_;
-  SequenceNumber seq_;
-};
-
-// A RangeDelIterator iterates over range deletion tombstones.
-class RangeDelIterator {
  public:
-  virtual ~RangeDelIterator() = default;
+  TruncatedRangeDelIterator(
+      std::unique_ptr<FragmentedRangeTombstoneIterator> iter,
+      const InternalKeyComparator* icmp, const InternalKey* smallest,
+      const InternalKey* largest);
 
-  virtual bool Valid() const = 0;
+  bool Valid() const;
-  virtual void Next() = 0;
-  // NOTE: the Slice passed to this method must be a user key.
-  virtual void Seek(const Slice& target) = 0;
-  virtual void Seek(const ParsedInternalKey& target) = 0;
-  virtual RangeTombstone Tombstone() const = 0;
-};
 
-// A RangeDelMap keeps track of range deletion tombstones within a snapshot
+  void Next();
-// stripe.
+  void Prev();
-//
-// RangeDelMaps are used internally by RangeDelAggregator. They are not intended
-// to be used directly.
-class RangeDelMap {
- public:
-  virtual ~RangeDelMap() = default;
 
-  virtual bool ShouldDelete(const ParsedInternalKey& parsed,
+  void InternalNext();
-                            RangeDelPositioningMode mode) = 0;
-  virtual bool IsRangeOverlapped(const ParsedInternalKey& start,
-                                 const ParsedInternalKey& end) = 0;
-  virtual void InvalidatePosition() = 0;
 
-  virtual size_t Size() const = 0;
+  // Seeks to the tombstone with the highest viisble sequence number that covers
-  bool IsEmpty() const { return Size() == 0; }
+  // target (a user key). If no such tombstone exists, the position will be at
+  // the earliest tombstone that ends after target.
+  void Seek(const Slice& target);
 
-  virtual void AddTombstone(TruncatedRangeTombstone tombstone) = 0;
+  // Seeks to the tombstone with the highest viisble sequence number that covers
-  virtual std::unique_ptr<RangeDelIterator> NewIterator() = 0;
+  // target (a user key). If no such tombstone exists, the position will be at
-};
+  // the latest tombstone that starts before target.
+  void SeekForPrev(const Slice& target);
 
-// A RangeDelAggregator aggregates range deletion tombstones as they are
+  void SeekToFirst();
-// encountered in memtables/SST files. It provides methods that check whether a
+  void SeekToLast();
-// key is covered by range tombstones or write the relevant tombstones to a new
-// SST file.
-class RangeDelAggregator {
- public:
-  // @param snapshots These are used to organize the tombstones into snapshot
-  //    stripes, which is the seqnum range between consecutive snapshots,
-  //    including the higher snapshot and excluding the lower one. Currently,
-  //    this is used by ShouldDelete() to prevent deletion of keys that are
-  //    covered by range tombstones in other snapshot stripes. This constructor
-  //    is used for writes (flush/compaction). All DB snapshots are provided
-  //    such that no keys are removed that are uncovered according to any DB
-  //    snapshot.
-  // Note this overload does not lazily initialize Rep.
-  RangeDelAggregator(const InternalKeyComparator& icmp,
-                     const std::vector<SequenceNumber>& snapshots,
-                     bool collapse_deletions = true);
 
-  // @param upper_bound Similar to snapshots above, except with a single
+  ParsedInternalKey start_key() const {
-  //    snapshot, which allows us to store the snapshot on the stack and defer
+    return (smallest_ == nullptr ||
-  //    initialization of heap-allocating members (in Rep) until the first range
+            icmp_->Compare(*smallest_, iter_->parsed_start_key()) <= 0)
-  //    deletion is encountered. This constructor is used in case of reads (get/
+               ? iter_->parsed_start_key()
-  //    iterator), for which only the user snapshot (upper_bound) is provided
+               : *smallest_;
-  //    such that the seqnum space is divided into two stripes. Only the older
-  //    stripe will be used by ShouldDelete().
-  RangeDelAggregator(const InternalKeyComparator& icmp,
-                     SequenceNumber upper_bound,
-                     bool collapse_deletions = false);
-
-  // Returns whether the key should be deleted, which is the case when it is
-  // covered by a range tombstone residing in the same snapshot stripe.
-  // @param mode If collapse_deletions_ is true, this dictates how we will find
-  //             the deletion whose interval contains this key. Otherwise, its
-  //             value must be kFullScan indicating linear scan from beginning.
-  bool ShouldDelete(
-      const ParsedInternalKey& parsed,
-      RangeDelPositioningMode mode = RangeDelPositioningMode::kFullScan) {
-    if (rep_ == nullptr) {
-      return false;
   }
-    return ShouldDeleteImpl(parsed, mode);
+
+  ParsedInternalKey end_key() const {
+    return (largest_ == nullptr ||
+            icmp_->Compare(iter_->parsed_end_key(), *largest_) <= 0)
+               ? iter_->parsed_end_key()
+               : *largest_;
   }
-  bool ShouldDelete(
-      const Slice& internal_key,
-      RangeDelPositioningMode mode = RangeDelPositioningMode::kFullScan) {
-    if (rep_ == nullptr) {
-      return false;
-    }
-    return ShouldDeleteImpl(internal_key, mode);
-  }
-  bool ShouldDeleteImpl(const ParsedInternalKey& parsed,
-                        RangeDelPositioningMode mode);
-  bool ShouldDeleteImpl(const Slice& internal_key,
-                        RangeDelPositioningMode mode);
 
-  // Checks whether range deletions cover any keys between `start` and `end`,
+  SequenceNumber seq() const { return iter_->seq(); }
-  // inclusive.
-  //
-  // @param start User key representing beginning of range to check for overlap.
-  // @param end User key representing end of range to check for overlap. This
-  //     argument is inclusive, so the existence of a range deletion covering
-  //     `end` causes this to return true.
-  bool IsRangeOverlapped(const Slice& start, const Slice& end);
 
-  // Adds tombstones to the tombstone aggregation structure maintained by this
+  std::map<SequenceNumber, std::unique_ptr<TruncatedRangeDelIterator>>
-  // object. Tombstones are truncated to smallest and largest. If smallest (or
+  SplitBySnapshot(const std::vector<SequenceNumber>& snapshots);
-  // largest) is null, it is not used for truncation. When adding range
-  // tombstones present in an sstable, smallest and largest should be set to
-  // the smallest and largest keys from the sstable file metadata. Note that
-  // tombstones end keys are exclusive while largest is inclusive.
-  // @return non-OK status if any of the tombstone keys are corrupted.
-  Status AddTombstones(std::unique_ptr<InternalIterator> input,
-                       const InternalKey* smallest = nullptr,
-                       const InternalKey* largest = nullptr);
 
-  // Resets iterators maintained across calls to ShouldDelete(). This may be
+  SequenceNumber upper_bound() const { return iter_->upper_bound(); }
-  // called when the tombstones change, or the owner may call explicitly, e.g.,
-  // if it's an iterator that just seeked to an arbitrary position. The effect
-  // of invalidation is that the following call to ShouldDelete() will binary
-  // search for its tombstone.
-  void InvalidateRangeDelMapPositions();
 
-  bool IsEmpty();
+  SequenceNumber lower_bound() const { return iter_->lower_bound(); }
-  bool AddFile(uint64_t file_number);
-
-  // Create a new iterator over the range deletion tombstones in all of the
-  // snapshot stripes in this aggregator. Tombstones are presented in start key
-  // order. Tombstones with the same start key are presented in arbitrary order.
-  //
-  // The iterator is invalidated after any call to AddTombstones. It is the
-  // caller's responsibility to avoid using invalid iterators.
-  std::unique_ptr<RangeDelIterator> NewIterator();
 
  private:
-  // Maps snapshot seqnum -> map of tombstones that fall in that stripe, i.e.,
+  std::unique_ptr<FragmentedRangeTombstoneIterator> iter_;
-  // their seqnums are greater than the next smaller snapshot's seqnum, and the
+  const InternalKeyComparator* icmp_;
-  // corresponding index into the list of snapshots. Each entry is lazily
+  const ParsedInternalKey* smallest_ = nullptr;
-  // initialized.
+  const ParsedInternalKey* largest_ = nullptr;
-  typedef std::map<SequenceNumber,
+  std::list<ParsedInternalKey> pinned_bounds_;
-                   std::pair<std::unique_ptr<RangeDelMap>, size_t>>
-      StripeMap;
 
-  struct Rep {
+  const InternalKey* smallest_ikey_;
-    std::vector<SequenceNumber> snapshots_;
+  const InternalKey* largest_ikey_;
-    StripeMap stripe_map_;
+};
-    PinnedIteratorsManager pinned_iters_mgr_;
+
-    std::list<std::string> pinned_slices_;
+struct SeqMaxComparator {
-    std::set<uint64_t> added_files_;
+  bool operator()(const TruncatedRangeDelIterator* a,
+                  const TruncatedRangeDelIterator* b) const {
+    return a->seq() > b->seq();
+  }
+};
+
+struct StartKeyMinComparator {
+  explicit StartKeyMinComparator(const InternalKeyComparator* c) : icmp(c) {}
+
+  bool operator()(const TruncatedRangeDelIterator* a,
+                  const TruncatedRangeDelIterator* b) const {
+    return icmp->Compare(a->start_key(), b->start_key()) > 0;
+  }
+
+  const InternalKeyComparator* icmp;
+};
+
+class ForwardRangeDelIterator {
+ public:
+  explicit ForwardRangeDelIterator(const InternalKeyComparator* icmp);
+
+  bool ShouldDelete(const ParsedInternalKey& parsed);
+  void Invalidate();
+
+  void AddNewIter(TruncatedRangeDelIterator* iter,
+                  const ParsedInternalKey& parsed) {
+    iter->Seek(parsed.user_key);
+    PushIter(iter, parsed);
+    assert(active_iters_.size() == active_seqnums_.size());
+  }
+
+  size_t UnusedIdx() const { return unused_idx_; }
+  void IncUnusedIdx() { unused_idx_++; }
+
+ private:
+  using ActiveSeqSet =
+      std::multiset<TruncatedRangeDelIterator*, SeqMaxComparator>;
+
+  struct EndKeyMinComparator {
+    explicit EndKeyMinComparator(const InternalKeyComparator* c) : icmp(c) {}
+
+    bool operator()(const ActiveSeqSet::const_iterator& a,
+                    const ActiveSeqSet::const_iterator& b) const {
+      return icmp->Compare((*a)->end_key(), (*b)->end_key()) > 0;
+    }
+
+    const InternalKeyComparator* icmp;
   };
-  // Initializes rep_ lazily. This aggregator object is constructed for every
-  // read, so expensive members should only be created when necessary, i.e.,
-  // once the first range deletion is encountered.
-  void InitRep(const std::vector<SequenceNumber>& snapshots);
 
-  std::unique_ptr<RangeDelMap> NewRangeDelMap();
+  void PushIter(TruncatedRangeDelIterator* iter,
-  RangeDelMap* GetRangeDelMapIfExists(SequenceNumber seq);
+                const ParsedInternalKey& parsed) {
-  RangeDelMap& GetRangeDelMap(SequenceNumber seq);
+    if (!iter->Valid()) {
+      // The iterator has been fully consumed, so we don't need to add it to
+      // either of the heaps.
+      return;
+    }
+    int cmp = icmp_->Compare(parsed, iter->start_key());
+    if (cmp < 0) {
+      PushInactiveIter(iter);
+    } else {
+      PushActiveIter(iter);
+    }
+  }
 
+  void PushActiveIter(TruncatedRangeDelIterator* iter) {
+    auto seq_pos = active_seqnums_.insert(iter);
+    active_iters_.push(seq_pos);
+  }
+
+  TruncatedRangeDelIterator* PopActiveIter() {
+    auto active_top = active_iters_.top();
+    auto iter = *active_top;
+    active_iters_.pop();
+    active_seqnums_.erase(active_top);
+    return iter;
+  }
+
+  void PushInactiveIter(TruncatedRangeDelIterator* iter) {
+    inactive_iters_.push(iter);
+  }
+
+  TruncatedRangeDelIterator* PopInactiveIter() {
+    auto* iter = inactive_iters_.top();
+    inactive_iters_.pop();
+    return iter;
+  }
+
+  const InternalKeyComparator* icmp_;
+  size_t unused_idx_;
+  ActiveSeqSet active_seqnums_;
+  BinaryHeap<ActiveSeqSet::const_iterator, EndKeyMinComparator> active_iters_;
+  BinaryHeap<TruncatedRangeDelIterator*, StartKeyMinComparator> inactive_iters_;
+};
+
+class ReverseRangeDelIterator {
+ public:
+  explicit ReverseRangeDelIterator(const InternalKeyComparator* icmp);
+
+  bool ShouldDelete(const ParsedInternalKey& parsed);
+  void Invalidate();
+
+  void AddNewIter(TruncatedRangeDelIterator* iter,
+                  const ParsedInternalKey& parsed) {
+    iter->SeekForPrev(parsed.user_key);
+    PushIter(iter, parsed);
+    assert(active_iters_.size() == active_seqnums_.size());
+  }
+
+  size_t UnusedIdx() const { return unused_idx_; }
+  void IncUnusedIdx() { unused_idx_++; }
+
+ private:
+  using ActiveSeqSet =
+      std::multiset<TruncatedRangeDelIterator*, SeqMaxComparator>;
+
+  struct EndKeyMaxComparator {
+    explicit EndKeyMaxComparator(const InternalKeyComparator* c) : icmp(c) {}
+
+    bool operator()(const TruncatedRangeDelIterator* a,
+                    const TruncatedRangeDelIterator* b) const {
+      return icmp->Compare(a->end_key(), b->end_key()) < 0;
+    }
+
+    const InternalKeyComparator* icmp;
+  };
+  struct StartKeyMaxComparator {
+    explicit StartKeyMaxComparator(const InternalKeyComparator* c) : icmp(c) {}
+
+    bool operator()(const ActiveSeqSet::const_iterator& a,
+                    const ActiveSeqSet::const_iterator& b) const {
+      return icmp->Compare((*a)->start_key(), (*b)->start_key()) < 0;
+    }
+
+    const InternalKeyComparator* icmp;
+  };
+
+  void PushIter(TruncatedRangeDelIterator* iter,
+                const ParsedInternalKey& parsed) {
+    if (!iter->Valid()) {
+      // The iterator has been fully consumed, so we don't need to add it to
+      // either of the heaps.
+    } else if (icmp_->Compare(iter->end_key(), parsed) <= 0) {
+      PushInactiveIter(iter);
+    } else {
+      PushActiveIter(iter);
+    }
+  }
+
+  void PushActiveIter(TruncatedRangeDelIterator* iter) {
+    auto seq_pos = active_seqnums_.insert(iter);
+    active_iters_.push(seq_pos);
+  }
+
+  TruncatedRangeDelIterator* PopActiveIter() {
+    auto active_top = active_iters_.top();
+    auto iter = *active_top;
+    active_iters_.pop();
+    active_seqnums_.erase(active_top);
+    return iter;
+  }
+
+  void PushInactiveIter(TruncatedRangeDelIterator* iter) {
+    inactive_iters_.push(iter);
+  }
+
+  TruncatedRangeDelIterator* PopInactiveIter() {
+    auto* iter = inactive_iters_.top();
+    inactive_iters_.pop();
+    return iter;
+  }
+
+  const InternalKeyComparator* icmp_;
+  size_t unused_idx_;
+  ActiveSeqSet active_seqnums_;
+  BinaryHeap<ActiveSeqSet::const_iterator, StartKeyMaxComparator> active_iters_;
+  BinaryHeap<TruncatedRangeDelIterator*, EndKeyMaxComparator> inactive_iters_;
+};
+
+enum class RangeDelPositioningMode { kForwardTraversal, kBackwardTraversal };
+class RangeDelAggregator {
+ public:
+  explicit RangeDelAggregator(const InternalKeyComparator* icmp)
+      : icmp_(icmp) {}
+  virtual ~RangeDelAggregator() {}
+
+  virtual void AddTombstones(
+      std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
+      const InternalKey* smallest = nullptr,
+      const InternalKey* largest = nullptr) = 0;
+
+  bool ShouldDelete(const Slice& key, RangeDelPositioningMode mode) {
+    ParsedInternalKey parsed;
+    if (!ParseInternalKey(key, &parsed)) {
+      return false;
+    }
+    return ShouldDelete(parsed, mode);
+  }
+  virtual bool ShouldDelete(const ParsedInternalKey& parsed,
+                            RangeDelPositioningMode mode) = 0;
+
+  virtual void InvalidateRangeDelMapPositions() = 0;
+
+  virtual bool IsEmpty() const = 0;
+
+  bool AddFile(uint64_t file_number) {
+    return files_seen_.insert(file_number).second;
+  }
+
+ protected:
+  class StripeRep {
+   public:
+    StripeRep(const InternalKeyComparator* icmp, SequenceNumber upper_bound,
+              SequenceNumber lower_bound)
+        : icmp_(icmp),
+          forward_iter_(icmp),
+          reverse_iter_(icmp),
+          upper_bound_(upper_bound),
+          lower_bound_(lower_bound) {}
+
+    void AddTombstones(std::unique_ptr<TruncatedRangeDelIterator> input_iter) {
+      iters_.push_back(std::move(input_iter));
+    }
+
+    bool IsEmpty() const { return iters_.empty(); }
+
+    bool ShouldDelete(const ParsedInternalKey& parsed,
+                      RangeDelPositioningMode mode);
+
+    void Invalidate() {
+      InvalidateForwardIter();
+      InvalidateReverseIter();
+    }
+
+    bool IsRangeOverlapped(const Slice& start, const Slice& end);
+
+   private:
+    bool InStripe(SequenceNumber seq) const {
+      return lower_bound_ <= seq && seq <= upper_bound_;
+    }
+
+    void InvalidateForwardIter() { forward_iter_.Invalidate(); }
+
+    void InvalidateReverseIter() { reverse_iter_.Invalidate(); }
+
+    const InternalKeyComparator* icmp_;
+    std::vector<std::unique_ptr<TruncatedRangeDelIterator>> iters_;
+    ForwardRangeDelIterator forward_iter_;
+    ReverseRangeDelIterator reverse_iter_;
     SequenceNumber upper_bound_;
-  std::unique_ptr<Rep> rep_;
+    SequenceNumber lower_bound_;
-  const InternalKeyComparator& icmp_;
+  };
-  // collapse range deletions so they're binary searchable
+
-  const bool collapse_deletions_;
+  const InternalKeyComparator* icmp_;
+
+ private:
+  std::set<uint64_t> files_seen_;
+};
+
+class ReadRangeDelAggregator : public RangeDelAggregator {
+ public:
+  ReadRangeDelAggregator(const InternalKeyComparator* icmp,
+                         SequenceNumber upper_bound)
+      : RangeDelAggregator(icmp),
+        rep_(icmp, upper_bound, 0 /* lower_bound */) {}
+  ~ReadRangeDelAggregator() override {}
+
+  using RangeDelAggregator::ShouldDelete;
+  void AddTombstones(
+      std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
+      const InternalKey* smallest = nullptr,
+      const InternalKey* largest = nullptr) override;
+
+  bool ShouldDelete(const ParsedInternalKey& parsed,
+                    RangeDelPositioningMode mode) override;
+
+  bool IsRangeOverlapped(const Slice& start, const Slice& end);
+
+  void InvalidateRangeDelMapPositions() override { rep_.Invalidate(); }
+
+  bool IsEmpty() const override { return rep_.IsEmpty(); }
+
+ private:
+  StripeRep rep_;
+};
+
+class CompactionRangeDelAggregator : public RangeDelAggregator {
+ public:
+  CompactionRangeDelAggregator(const InternalKeyComparator* icmp,
+                               const std::vector<SequenceNumber>& snapshots)
+      : RangeDelAggregator(icmp), snapshots_(&snapshots) {}
+  ~CompactionRangeDelAggregator() override {}
+
+  void AddTombstones(
+      std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
+      const InternalKey* smallest = nullptr,
+      const InternalKey* largest = nullptr) override;
+
+  using RangeDelAggregator::ShouldDelete;
+  bool ShouldDelete(const ParsedInternalKey& parsed,
+                    RangeDelPositioningMode mode) override;
+
+  bool IsRangeOverlapped(const Slice& start, const Slice& end);
+
+  void InvalidateRangeDelMapPositions() override {
+    for (auto& rep : reps_) {
+      rep.second.Invalidate();
+    }
+  }
+
+  bool IsEmpty() const override {
+    for (const auto& rep : reps_) {
+      if (!rep.second.IsEmpty()) {
+        return false;
+      }
+    }
+    return true;
+  }
+
+  // Creates an iterator over all the range tombstones in the aggregator, for
+  // use in compaction. Nullptr arguments indicate that the iterator range is
+  // unbounded.
+  // NOTE: the boundaries are used for optimization purposes to reduce the
+  // number of tombstones that are passed to the fragmenter; they do not
+  // guarantee that the resulting iterator only contains range tombstones that
+  // cover keys in the provided range. If required, these bounds must be
+  // enforced during iteration.
+  std::unique_ptr<FragmentedRangeTombstoneIterator> NewIterator(
+      const Slice* lower_bound = nullptr, const Slice* upper_bound = nullptr,
+      bool upper_bound_inclusive = false);
+
+ private:
+  std::vector<std::unique_ptr<TruncatedRangeDelIterator>> parent_iters_;
+  std::map<SequenceNumber, StripeRep> reps_;
+
+  const std::vector<SequenceNumber>* snapshots_;
 };
 
 }  // namespace rocksdb

db/range_del_aggregator_bench.cc

@@ -20,7 +20,6 @@ int main() {
 #include <vector>
 
 #include "db/range_del_aggregator.h"
-#include "db/range_del_aggregator_v2.h"
 #include "db/range_tombstone_fragmenter.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
@@ -48,8 +47,6 @@ DEFINE_double(tombstone_width_mean, 100.0, "average range tombstone width");
 DEFINE_double(tombstone_width_stddev, 0.0,
               "standard deviation of range tombstone width");
 
-DEFINE_bool(use_collapsed, true, "use the collapsed range tombstone map");
-
 DEFINE_int32(seed, 0, "random number generator seed");
 
 DEFINE_int32(should_deletes_per_run, 1, "number of ShouldDelete calls per run");
@@ -57,8 +54,6 @@ DEFINE_int32(should_deletes_per_run, 1, "number of ShouldDelete calls per run");
 DEFINE_int32(add_tombstones_per_run, 1,
              "number of AddTombstones calls per run");
 
-DEFINE_bool(use_v2_aggregator, false, "benchmark RangeDelAggregatorV2");
-
 namespace {
 
 struct Stats {
@@ -187,14 +182,10 @@ int main(int argc, char** argv) {
         std::vector<rocksdb::PersistentRangeTombstone>(
             FLAGS_num_range_tombstones);
   }
-  auto mode = FLAGS_use_collapsed
+  auto mode = rocksdb::RangeDelPositioningMode::kForwardTraversal;
-                  ? rocksdb::RangeDelPositioningMode::kForwardTraversal
-                  : rocksdb::RangeDelPositioningMode::kFullScan;
 
   for (int i = 0; i < FLAGS_num_runs; i++) {
-    rocksdb::RangeDelAggregator range_del_agg(icmp, {} /* snapshots */,
+    rocksdb::ReadRangeDelAggregator range_del_agg(
-                                              FLAGS_use_collapsed);
-    rocksdb::RangeDelAggregatorV2 range_del_agg_v2(
         &icmp, rocksdb::kMaxSequenceNumber /* upper_bound */);
 
     std::vector<std::unique_ptr<rocksdb::FragmentedRangeTombstoneList> >
@@ -206,7 +197,7 @@ int main(int argc, char** argv) {
       // real workloads.
       for (int j = 0; j < FLAGS_num_range_tombstones; j++) {
         uint64_t start = rnd.Uniform(FLAGS_tombstone_start_upper_bound);
-        uint64_t end = start + std::max(1.0, normal_dist(random_gen));
+        uint64_t end = start + static_cast<uint64_t>(std::max(1.0, normal_dist(random_gen)));
         persistent_range_tombstones[j] = rocksdb::PersistentRangeTombstone(
             rocksdb::Key(start), rocksdb::Key(end), j);
       }
@@ -220,20 +211,13 @@ int main(int argc, char** argv) {
       std::unique_ptr<rocksdb::FragmentedRangeTombstoneIterator>
           fragmented_range_del_iter(
               new rocksdb::FragmentedRangeTombstoneIterator(
-                  fragmented_range_tombstone_lists.back().get(),
+                  fragmented_range_tombstone_lists.back().get(), icmp,
-                  rocksdb::kMaxSequenceNumber, icmp));
+                  rocksdb::kMaxSequenceNumber));
 
-      if (FLAGS_use_v2_aggregator) {
+      rocksdb::StopWatchNano stop_watch_add_tombstones(rocksdb::Env::Default(),
-        rocksdb::StopWatchNano stop_watch_add_tombstones(
+                                                       true /* auto_start */);
-            rocksdb::Env::Default(), true /* auto_start */);
+      range_del_agg.AddTombstones(std::move(fragmented_range_del_iter));
-        range_del_agg_v2.AddTombstones(std::move(fragmented_range_del_iter));
       stats.time_add_tombstones += stop_watch_add_tombstones.ElapsedNanos();
-      } else {
-        rocksdb::StopWatchNano stop_watch_add_tombstones(
-            rocksdb::Env::Default(), true /* auto_start */);
-        range_del_agg.AddTombstones(std::move(range_del_iter));
-        stats.time_add_tombstones += stop_watch_add_tombstones.ElapsedNanos();
-      }
     }
 
     rocksdb::ParsedInternalKey parsed_key;
@@ -247,18 +231,10 @@ int main(int argc, char** argv) {
       std::string key_string = rocksdb::Key(first_key + j);
       parsed_key.user_key = key_string;
 
-      uint64_t call_time;
-      if (FLAGS_use_v2_aggregator) {
-        rocksdb::StopWatchNano stop_watch_should_delete(rocksdb::Env::Default(),
-                                                        true /* auto_start */);
-        range_del_agg_v2.ShouldDelete(parsed_key, mode);
-        call_time = stop_watch_should_delete.ElapsedNanos();
-      } else {
       rocksdb::StopWatchNano stop_watch_should_delete(rocksdb::Env::Default(),
                                                       true /* auto_start */);
       range_del_agg.ShouldDelete(parsed_key, mode);
-        call_time = stop_watch_should_delete.ElapsedNanos();
+      uint64_t call_time = stop_watch_should_delete.ElapsedNanos();
-      }
 
       if (j == 0) {
         stats.time_first_should_delete += call_time;

db/range_del_aggregator_v2.cc

@@ -1,295 +0,0 @@
-//  Copyright (c) 2018-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#include "db/range_del_aggregator_v2.h"
-
-#include "db/compaction_iteration_stats.h"
-#include "db/dbformat.h"
-#include "db/pinned_iterators_manager.h"
-#include "db/range_del_aggregator.h"
-#include "db/range_tombstone_fragmenter.h"
-#include "db/version_edit.h"
-#include "include/rocksdb/comparator.h"
-#include "include/rocksdb/types.h"
-#include "table/internal_iterator.h"
-#include "table/scoped_arena_iterator.h"
-#include "table/table_builder.h"
-#include "util/heap.h"
-#include "util/kv_map.h"
-#include "util/vector_iterator.h"
-
-namespace rocksdb {
-
-TruncatedRangeDelIterator::TruncatedRangeDelIterator(
-    std::unique_ptr<FragmentedRangeTombstoneIterator> iter,
-    const InternalKeyComparator* icmp, const InternalKey* smallest,
-    const InternalKey* largest)
-    : iter_(std::move(iter)), icmp_(icmp) {
-  if (smallest != nullptr) {
-    pinned_bounds_.emplace_back();
-    auto& parsed_smallest = pinned_bounds_.back();
-    if (!ParseInternalKey(smallest->Encode(), &parsed_smallest)) {
-      assert(false);
-    }
-    smallest_ = &parsed_smallest;
-  }
-  if (largest != nullptr) {
-    pinned_bounds_.emplace_back();
-    auto& parsed_largest = pinned_bounds_.back();
-    if (!ParseInternalKey(largest->Encode(), &parsed_largest)) {
-      assert(false);
-    }
-    if (parsed_largest.type == kTypeRangeDeletion &&
-        parsed_largest.sequence == kMaxSequenceNumber) {
-      // The file boundary has been artificially extended by a range tombstone.
-      // We do not need to adjust largest to properly truncate range
-      // tombstones that extend past the boundary.
-    } else if (parsed_largest.sequence == 0) {
-      // The largest key in the sstable has a sequence number of 0. Since we
-      // guarantee that no internal keys with the same user key and sequence
-      // number can exist in a DB, we know that the largest key in this sstable
-      // cannot exist as the smallest key in the next sstable. This further
-      // implies that no range tombstone in this sstable covers largest;
-      // otherwise, the file boundary would have been artificially extended.
-      //
-      // Therefore, we will never truncate a range tombstone at largest, so we
-      // can leave it unchanged.
-    } else {
-      // The same user key may straddle two sstable boundaries. To ensure that
-      // the truncated end key can cover the largest key in this sstable, reduce
-      // its sequence number by 1.
-      parsed_largest.sequence -= 1;
-    }
-    largest_ = &parsed_largest;
-  }
-}
-
-bool TruncatedRangeDelIterator::Valid() const {
-  return iter_->Valid() &&
-         (smallest_ == nullptr ||
-          icmp_->Compare(*smallest_, iter_->parsed_end_key()) < 0) &&
-         (largest_ == nullptr ||
-          icmp_->Compare(iter_->parsed_start_key(), *largest_) < 0);
-}
-
-void TruncatedRangeDelIterator::Next() { iter_->TopNext(); }
-
-void TruncatedRangeDelIterator::Prev() { iter_->TopPrev(); }
-
-// NOTE: target is a user key
-void TruncatedRangeDelIterator::Seek(const Slice& target) {
-  if (largest_ != nullptr &&
-      icmp_->Compare(*largest_, ParsedInternalKey(target, kMaxSequenceNumber,
-                                                  kTypeRangeDeletion)) <= 0) {
-    iter_->Invalidate();
-    return;
-  }
-  iter_->Seek(target);
-}
-
-// NOTE: target is a user key
-void TruncatedRangeDelIterator::SeekForPrev(const Slice& target) {
-  if (smallest_ != nullptr &&
-      icmp_->Compare(ParsedInternalKey(target, 0, kTypeRangeDeletion),
-                     *smallest_) < 0) {
-    iter_->Invalidate();
-    return;
-  }
-  iter_->SeekForPrev(target);
-}
-
-void TruncatedRangeDelIterator::SeekToFirst() { iter_->SeekToTopFirst(); }
-
-void TruncatedRangeDelIterator::SeekToLast() { iter_->SeekToTopLast(); }
-
-ForwardRangeDelIterator::ForwardRangeDelIterator(
-    const InternalKeyComparator* icmp,
-    const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters)
-    : icmp_(icmp),
-      iters_(iters),
-      unused_idx_(0),
-      active_seqnums_(SeqMaxComparator()),
-      active_iters_(EndKeyMinComparator(icmp)),
-      inactive_iters_(StartKeyMinComparator(icmp)) {}
-
-bool ForwardRangeDelIterator::ShouldDelete(const ParsedInternalKey& parsed) {
-  assert(iters_ != nullptr);
-  // Pick up previously unseen iterators.
-  for (auto it = std::next(iters_->begin(), unused_idx_); it != iters_->end();
-       ++it, ++unused_idx_) {
-    auto& iter = *it;
-    iter->Seek(parsed.user_key);
-    PushIter(iter.get(), parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  // Move active iterators that end before parsed.
-  while (!active_iters_.empty() &&
-         icmp_->Compare((*active_iters_.top())->end_key(), parsed) <= 0) {
-    TruncatedRangeDelIterator* iter = PopActiveIter();
-    do {
-      iter->Next();
-    } while (iter->Valid() && icmp_->Compare(iter->end_key(), parsed) <= 0);
-    PushIter(iter, parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  // Move inactive iterators that start before parsed.
-  while (!inactive_iters_.empty() &&
-         icmp_->Compare(inactive_iters_.top()->start_key(), parsed) <= 0) {
-    TruncatedRangeDelIterator* iter = PopInactiveIter();
-    while (iter->Valid() && icmp_->Compare(iter->end_key(), parsed) <= 0) {
-      iter->Next();
-    }
-    PushIter(iter, parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  return active_seqnums_.empty()
-             ? false
-             : (*active_seqnums_.begin())->seq() > parsed.sequence;
-}
-
-void ForwardRangeDelIterator::Invalidate() {
-  unused_idx_ = 0;
-  active_iters_.clear();
-  active_seqnums_.clear();
-  inactive_iters_.clear();
-}
-
-ReverseRangeDelIterator::ReverseRangeDelIterator(
-    const InternalKeyComparator* icmp,
-    const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters)
-    : icmp_(icmp),
-      iters_(iters),
-      unused_idx_(0),
-      active_seqnums_(SeqMaxComparator()),
-      active_iters_(StartKeyMaxComparator(icmp)),
-      inactive_iters_(EndKeyMaxComparator(icmp)) {}
-
-bool ReverseRangeDelIterator::ShouldDelete(const ParsedInternalKey& parsed) {
-  assert(iters_ != nullptr);
-  // Pick up previously unseen iterators.
-  for (auto it = std::next(iters_->begin(), unused_idx_); it != iters_->end();
-       ++it, ++unused_idx_) {
-    auto& iter = *it;
-    iter->SeekForPrev(parsed.user_key);
-    PushIter(iter.get(), parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  // Move active iterators that start after parsed.
-  while (!active_iters_.empty() &&
-         icmp_->Compare(parsed, (*active_iters_.top())->start_key()) < 0) {
-    TruncatedRangeDelIterator* iter = PopActiveIter();
-    do {
-      iter->Prev();
-    } while (iter->Valid() && icmp_->Compare(parsed, iter->start_key()) < 0);
-    PushIter(iter, parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  // Move inactive iterators that end after parsed.
-  while (!inactive_iters_.empty() &&
-         icmp_->Compare(parsed, inactive_iters_.top()->end_key()) < 0) {
-    TruncatedRangeDelIterator* iter = PopInactiveIter();
-    while (iter->Valid() && icmp_->Compare(parsed, iter->start_key()) < 0) {
-      iter->Prev();
-    }
-    PushIter(iter, parsed);
-    assert(active_iters_.size() == active_seqnums_.size());
-  }
-
-  return active_seqnums_.empty()
-             ? false
-             : (*active_seqnums_.begin())->seq() > parsed.sequence;
-}
-
-void ReverseRangeDelIterator::Invalidate() {
-  unused_idx_ = 0;
-  active_iters_.clear();
-  active_seqnums_.clear();
-  inactive_iters_.clear();
-}
-
-RangeDelAggregatorV2::RangeDelAggregatorV2(const InternalKeyComparator* icmp,
-                                           SequenceNumber /* upper_bound */)
-    : icmp_(icmp), forward_iter_(icmp, &iters_), reverse_iter_(icmp, &iters_) {}
-
-void RangeDelAggregatorV2::AddTombstones(
-    std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
-    const InternalKey* smallest, const InternalKey* largest) {
-  if (input_iter == nullptr || input_iter->empty()) {
-    return;
-  }
-  if (wrapped_range_del_agg != nullptr) {
-    wrapped_range_del_agg->AddTombstones(std::move(input_iter), smallest,
-                                         largest);
-    // TODO: this eats the status of the wrapped call; may want to propagate it
-    return;
-  }
-  iters_.emplace_back(new TruncatedRangeDelIterator(std::move(input_iter),
-                                                    icmp_, smallest, largest));
-}
-
-bool RangeDelAggregatorV2::ShouldDelete(const ParsedInternalKey& parsed,
-                                        RangeDelPositioningMode mode) {
-  if (wrapped_range_del_agg != nullptr) {
-    return wrapped_range_del_agg->ShouldDelete(parsed, mode);
-  }
-
-  switch (mode) {
-    case RangeDelPositioningMode::kForwardTraversal:
-      reverse_iter_.Invalidate();
-      return forward_iter_.ShouldDelete(parsed);
-    case RangeDelPositioningMode::kBackwardTraversal:
-      forward_iter_.Invalidate();
-      return reverse_iter_.ShouldDelete(parsed);
-    default:
-      assert(false);
-      return false;
-  }
-}
-
-bool RangeDelAggregatorV2::IsRangeOverlapped(const Slice& start,
-                                             const Slice& end) {
-  assert(wrapped_range_del_agg == nullptr);
-  InvalidateRangeDelMapPositions();
-
-  // Set the internal start/end keys so that:
-  // - if start_ikey has the same user key and sequence number as the current
-  // end key, start_ikey will be considered greater; and
-  // - if end_ikey has the same user key and sequence number as the current
-  // start key, end_ikey will be considered greater.
-  ParsedInternalKey start_ikey(start, kMaxSequenceNumber,
-                               static_cast<ValueType>(0));
-  ParsedInternalKey end_ikey(end, 0, static_cast<ValueType>(0));
-  for (auto& iter : iters_) {
-    bool checked_candidate_tombstones = false;
-    for (iter->SeekForPrev(start);
-         iter->Valid() && icmp_->Compare(iter->start_key(), end_ikey) <= 0;
-         iter->Next()) {
-      checked_candidate_tombstones = true;
-      if (icmp_->Compare(start_ikey, iter->end_key()) < 0 &&
-          icmp_->Compare(iter->start_key(), end_ikey) <= 0) {
-        return true;
-      }
-    }
-
-    if (!checked_candidate_tombstones) {
-      // Do an additional check for when the end of the range is the begin key
-      // of a tombstone, which we missed earlier since SeekForPrev'ing to the
-      // start was invalid.
-      iter->SeekForPrev(end);
-      if (iter->Valid() && icmp_->Compare(start_ikey, iter->end_key()) < 0 &&
-          icmp_->Compare(iter->start_key(), end_ikey) <= 0) {
-        return true;
-      }
-    }
-  }
-  return false;
-}
-
-}  // namespace rocksdb

db/range_del_aggregator_v2.h

@@ -1,295 +0,0 @@
-//  Copyright (c) 2018-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#pragma once
-
-#include <list>
-#include <map>
-#include <set>
-#include <string>
-#include <vector>
-
-#include "db/compaction_iteration_stats.h"
-#include "db/dbformat.h"
-#include "db/pinned_iterators_manager.h"
-#include "db/range_del_aggregator.h"
-#include "db/range_tombstone_fragmenter.h"
-#include "db/version_edit.h"
-#include "include/rocksdb/comparator.h"
-#include "include/rocksdb/types.h"
-#include "table/internal_iterator.h"
-#include "table/scoped_arena_iterator.h"
-#include "table/table_builder.h"
-#include "util/heap.h"
-#include "util/kv_map.h"
-
-namespace rocksdb {
-
-class RangeDelAggregatorV2;
-
-class TruncatedRangeDelIterator {
- public:
-  TruncatedRangeDelIterator(
-      std::unique_ptr<FragmentedRangeTombstoneIterator> iter,
-      const InternalKeyComparator* icmp, const InternalKey* smallest,
-      const InternalKey* largest);
-
-  bool Valid() const;
-
-  void Next();
-  void Prev();
-
-  // Seeks to the tombstone with the highest viisble sequence number that covers
-  // target (a user key). If no such tombstone exists, the position will be at
-  // the earliest tombstone that ends after target.
-  void Seek(const Slice& target);
-
-  // Seeks to the tombstone with the highest viisble sequence number that covers
-  // target (a user key). If no such tombstone exists, the position will be at
-  // the latest tombstone that starts before target.
-  void SeekForPrev(const Slice& target);
-
-  void SeekToFirst();
-  void SeekToLast();
-
-  ParsedInternalKey start_key() const {
-    return (smallest_ == nullptr ||
-            icmp_->Compare(*smallest_, iter_->parsed_start_key()) <= 0)
-               ? iter_->parsed_start_key()
-               : *smallest_;
-  }
-
-  ParsedInternalKey end_key() const {
-    return (largest_ == nullptr ||
-            icmp_->Compare(iter_->parsed_end_key(), *largest_) <= 0)
-               ? iter_->parsed_end_key()
-               : *largest_;
-  }
-
-  SequenceNumber seq() const { return iter_->seq(); }
-
- private:
-  std::unique_ptr<FragmentedRangeTombstoneIterator> iter_;
-  const InternalKeyComparator* icmp_;
-  const ParsedInternalKey* smallest_ = nullptr;
-  const ParsedInternalKey* largest_ = nullptr;
-  std::list<ParsedInternalKey> pinned_bounds_;
-};
-
-struct SeqMaxComparator {
-  bool operator()(const TruncatedRangeDelIterator* a,
-                  const TruncatedRangeDelIterator* b) const {
-    return a->seq() > b->seq();
-  }
-};
-
-class ForwardRangeDelIterator {
- public:
-  ForwardRangeDelIterator(
-      const InternalKeyComparator* icmp,
-      const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters);
-
-  bool ShouldDelete(const ParsedInternalKey& parsed);
-  void Invalidate();
-
- private:
-  using ActiveSeqSet =
-      std::multiset<TruncatedRangeDelIterator*, SeqMaxComparator>;
-
-  struct StartKeyMinComparator {
-    explicit StartKeyMinComparator(const InternalKeyComparator* c) : icmp(c) {}
-
-    bool operator()(const TruncatedRangeDelIterator* a,
-                    const TruncatedRangeDelIterator* b) const {
-      return icmp->Compare(a->start_key(), b->start_key()) > 0;
-    }
-
-    const InternalKeyComparator* icmp;
-  };
-  struct EndKeyMinComparator {
-    explicit EndKeyMinComparator(const InternalKeyComparator* c) : icmp(c) {}
-
-    bool operator()(const ActiveSeqSet::const_iterator& a,
-                    const ActiveSeqSet::const_iterator& b) const {
-      return icmp->Compare((*a)->end_key(), (*b)->end_key()) > 0;
-    }
-
-    const InternalKeyComparator* icmp;
-  };
-
-  void PushIter(TruncatedRangeDelIterator* iter,
-                const ParsedInternalKey& parsed) {
-    if (!iter->Valid()) {
-      // The iterator has been fully consumed, so we don't need to add it to
-      // either of the heaps.
-    } else if (icmp_->Compare(parsed, iter->start_key()) < 0) {
-      PushInactiveIter(iter);
-    } else {
-      PushActiveIter(iter);
-    }
-  }
-
-  void PushActiveIter(TruncatedRangeDelIterator* iter) {
-    auto seq_pos = active_seqnums_.insert(iter);
-    active_iters_.push(seq_pos);
-  }
-
-  TruncatedRangeDelIterator* PopActiveIter() {
-    auto active_top = active_iters_.top();
-    auto iter = *active_top;
-    active_iters_.pop();
-    active_seqnums_.erase(active_top);
-    return iter;
-  }
-
-  void PushInactiveIter(TruncatedRangeDelIterator* iter) {
-    inactive_iters_.push(iter);
-  }
-
-  TruncatedRangeDelIterator* PopInactiveIter() {
-    auto* iter = inactive_iters_.top();
-    inactive_iters_.pop();
-    return iter;
-  }
-
-  const InternalKeyComparator* icmp_;
-  const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters_;
-  size_t unused_idx_;
-  ActiveSeqSet active_seqnums_;
-  BinaryHeap<ActiveSeqSet::const_iterator, EndKeyMinComparator> active_iters_;
-  BinaryHeap<TruncatedRangeDelIterator*, StartKeyMinComparator> inactive_iters_;
-};
-
-class ReverseRangeDelIterator {
- public:
-  ReverseRangeDelIterator(
-      const InternalKeyComparator* icmp,
-      const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters);
-
-  bool ShouldDelete(const ParsedInternalKey& parsed);
-  void Invalidate();
-
- private:
-  using ActiveSeqSet =
-      std::multiset<TruncatedRangeDelIterator*, SeqMaxComparator>;
-
-  struct EndKeyMaxComparator {
-    explicit EndKeyMaxComparator(const InternalKeyComparator* c) : icmp(c) {}
-
-    bool operator()(const TruncatedRangeDelIterator* a,
-                    const TruncatedRangeDelIterator* b) const {
-      return icmp->Compare(a->end_key(), b->end_key()) < 0;
-    }
-
-    const InternalKeyComparator* icmp;
-  };
-  struct StartKeyMaxComparator {
-    explicit StartKeyMaxComparator(const InternalKeyComparator* c) : icmp(c) {}
-
-    bool operator()(const ActiveSeqSet::const_iterator& a,
-                    const ActiveSeqSet::const_iterator& b) const {
-      return icmp->Compare((*a)->start_key(), (*b)->start_key()) < 0;
-    }
-
-    const InternalKeyComparator* icmp;
-  };
-
-  void PushIter(TruncatedRangeDelIterator* iter,
-                const ParsedInternalKey& parsed) {
-    if (!iter->Valid()) {
-      // The iterator has been fully consumed, so we don't need to add it to
-      // either of the heaps.
-    } else if (icmp_->Compare(iter->end_key(), parsed) <= 0) {
-      PushInactiveIter(iter);
-    } else {
-      PushActiveIter(iter);
-    }
-  }
-
-  void PushActiveIter(TruncatedRangeDelIterator* iter) {
-    auto seq_pos = active_seqnums_.insert(iter);
-    active_iters_.push(seq_pos);
-  }
-
-  TruncatedRangeDelIterator* PopActiveIter() {
-    auto active_top = active_iters_.top();
-    auto iter = *active_top;
-    active_iters_.pop();
-    active_seqnums_.erase(active_top);
-    return iter;
-  }
-
-  void PushInactiveIter(TruncatedRangeDelIterator* iter) {
-    inactive_iters_.push(iter);
-  }
-
-  TruncatedRangeDelIterator* PopInactiveIter() {
-    auto* iter = inactive_iters_.top();
-    inactive_iters_.pop();
-    return iter;
-  }
-
-  const InternalKeyComparator* icmp_;
-  const std::vector<std::unique_ptr<TruncatedRangeDelIterator>>* iters_;
-  size_t unused_idx_;
-  ActiveSeqSet active_seqnums_;
-  BinaryHeap<ActiveSeqSet::const_iterator, StartKeyMaxComparator> active_iters_;
-  BinaryHeap<TruncatedRangeDelIterator*, EndKeyMaxComparator> inactive_iters_;
-};
-
-class RangeDelAggregatorV2 {
- public:
-  RangeDelAggregatorV2(const InternalKeyComparator* icmp,
-                       SequenceNumber upper_bound);
-
-  void AddTombstones(
-      std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter,
-      const InternalKey* smallest = nullptr,
-      const InternalKey* largest = nullptr);
-
-  bool ShouldDelete(const ParsedInternalKey& parsed,
-                    RangeDelPositioningMode mode);
-
-  bool IsRangeOverlapped(const Slice& start, const Slice& end);
-
-  void InvalidateRangeDelMapPositions() {
-    forward_iter_.Invalidate();
-    reverse_iter_.Invalidate();
-  }
-
-  bool IsEmpty() const { return iters_.empty(); }
-  bool AddFile(uint64_t file_number) {
-    return files_seen_.insert(file_number).second;
-  }
-
-  // Adaptor method to pass calls through to an old-style RangeDelAggregator.
-  // Will be removed once this new version supports an iterator that can be used
-  // during flush/compaction.
-  RangeDelAggregator* DelegateToRangeDelAggregator(
-      const std::vector<SequenceNumber>& snapshots) {
-    wrapped_range_del_agg.reset(new RangeDelAggregator(
-        *icmp_, snapshots, true /* collapse_deletions */));
-    return wrapped_range_del_agg.get();
-  }
-
-  std::unique_ptr<RangeDelIterator> NewIterator() {
-    assert(wrapped_range_del_agg != nullptr);
-    return wrapped_range_del_agg->NewIterator();
-  }
-
- private:
-  const InternalKeyComparator* icmp_;
-
-  std::vector<std::unique_ptr<TruncatedRangeDelIterator>> iters_;
-  std::set<uint64_t> files_seen_;
-
-  ForwardRangeDelIterator forward_iter_;
-  ReverseRangeDelIterator reverse_iter_;
-
-  // TODO: remove once V2 supports exposing tombstone iterators
-  std::unique_ptr<RangeDelAggregator> wrapped_range_del_agg;
-};
-
-}  // namespace rocksdb

db/range_del_aggregator_v2_test.cc

@@ -1,469 +0,0 @@
-//  Copyright (c) 2018-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under both the GPLv2 (found in the
-//  COPYING file in the root directory) and Apache 2.0 License
-//  (found in the LICENSE.Apache file in the root directory).
-
-#include "db/range_del_aggregator_v2.h"
-
-#include <memory>
-#include <string>
-#include <vector>
-
-#include "db/db_test_util.h"
-#include "db/dbformat.h"
-#include "db/range_tombstone_fragmenter.h"
-#include "util/testutil.h"
-
-namespace rocksdb {
-
-class RangeDelAggregatorV2Test : public testing::Test {};
-
-namespace {
-
-static auto bytewise_icmp = InternalKeyComparator(BytewiseComparator());
-
-std::unique_ptr<InternalIterator> MakeRangeDelIter(
-    const std::vector<RangeTombstone>& range_dels) {
-  std::vector<std::string> keys, values;
-  for (const auto& range_del : range_dels) {
-    auto key_and_value = range_del.Serialize();
-    keys.push_back(key_and_value.first.Encode().ToString());
-    values.push_back(key_and_value.second.ToString());
-  }
-  return std::unique_ptr<test::VectorIterator>(
-      new test::VectorIterator(keys, values));
-}
-
-std::vector<std::unique_ptr<FragmentedRangeTombstoneList>>
-MakeFragmentedTombstoneLists(
-    const std::vector<std::vector<RangeTombstone>>& range_dels_list) {
-  std::vector<std::unique_ptr<FragmentedRangeTombstoneList>> fragment_lists;
-  for (const auto& range_dels : range_dels_list) {
-    auto range_del_iter = MakeRangeDelIter(range_dels);
-    fragment_lists.emplace_back(new FragmentedRangeTombstoneList(
-        std::move(range_del_iter), bytewise_icmp));
-  }
-  return fragment_lists;
-}
-
-struct TruncatedIterScanTestCase {
-  ParsedInternalKey start;
-  ParsedInternalKey end;
-  SequenceNumber seq;
-};
-
-struct TruncatedIterSeekTestCase {
-  Slice target;
-  ParsedInternalKey start;
-  ParsedInternalKey end;
-  SequenceNumber seq;
-  bool invalid;
-};
-
-struct ShouldDeleteTestCase {
-  ParsedInternalKey lookup_key;
-  bool result;
-};
-
-struct IsRangeOverlappedTestCase {
-  Slice start;
-  Slice end;
-  bool result;
-};
-
-ParsedInternalKey UncutEndpoint(const Slice& s) {
-  return ParsedInternalKey(s, kMaxSequenceNumber, kTypeRangeDeletion);
-}
-
-ParsedInternalKey InternalValue(const Slice& key, SequenceNumber seq) {
-  return ParsedInternalKey(key, seq, kTypeValue);
-}
-
-void VerifyIterator(
-    TruncatedRangeDelIterator* iter, const InternalKeyComparator& icmp,
-    const std::vector<TruncatedIterScanTestCase>& expected_range_dels) {
-  // Test forward iteration.
-  iter->SeekToFirst();
-  for (size_t i = 0; i < expected_range_dels.size(); i++, iter->Next()) {
-    ASSERT_TRUE(iter->Valid());
-    EXPECT_EQ(0, icmp.Compare(iter->start_key(), expected_range_dels[i].start));
-    EXPECT_EQ(0, icmp.Compare(iter->end_key(), expected_range_dels[i].end));
-    EXPECT_EQ(expected_range_dels[i].seq, iter->seq());
-  }
-  EXPECT_FALSE(iter->Valid());
-
-  // Test reverse iteration.
-  iter->SeekToLast();
-  std::vector<TruncatedIterScanTestCase> reverse_expected_range_dels(
-      expected_range_dels.rbegin(), expected_range_dels.rend());
-  for (size_t i = 0; i < reverse_expected_range_dels.size();
-       i++, iter->Prev()) {
-    ASSERT_TRUE(iter->Valid());
-    EXPECT_EQ(0, icmp.Compare(iter->start_key(),
-                              reverse_expected_range_dels[i].start));
-    EXPECT_EQ(
-        0, icmp.Compare(iter->end_key(), reverse_expected_range_dels[i].end));
-    EXPECT_EQ(reverse_expected_range_dels[i].seq, iter->seq());
-  }
-  EXPECT_FALSE(iter->Valid());
-}
-
-void VerifySeek(TruncatedRangeDelIterator* iter,
-                const InternalKeyComparator& icmp,
-                const std::vector<TruncatedIterSeekTestCase>& test_cases) {
-  for (const auto& test_case : test_cases) {
-    iter->Seek(test_case.target);
-    if (test_case.invalid) {
-      ASSERT_FALSE(iter->Valid());
-    } else {
-      ASSERT_TRUE(iter->Valid());
-      EXPECT_EQ(0, icmp.Compare(iter->start_key(), test_case.start));
-      EXPECT_EQ(0, icmp.Compare(iter->end_key(), test_case.end));
-      EXPECT_EQ(test_case.seq, iter->seq());
-    }
-  }
-}
-
-void VerifySeekForPrev(
-    TruncatedRangeDelIterator* iter, const InternalKeyComparator& icmp,
-    const std::vector<TruncatedIterSeekTestCase>& test_cases) {
-  for (const auto& test_case : test_cases) {
-    iter->SeekForPrev(test_case.target);
-    if (test_case.invalid) {
-      ASSERT_FALSE(iter->Valid());
-    } else {
-      ASSERT_TRUE(iter->Valid());
-      EXPECT_EQ(0, icmp.Compare(iter->start_key(), test_case.start));
-      EXPECT_EQ(0, icmp.Compare(iter->end_key(), test_case.end));
-      EXPECT_EQ(test_case.seq, iter->seq());
-    }
-  }
-}
-
-void VerifyShouldDelete(RangeDelAggregatorV2* range_del_agg,
-                        const std::vector<ShouldDeleteTestCase>& test_cases) {
-  for (const auto& test_case : test_cases) {
-    EXPECT_EQ(
-        test_case.result,
-        range_del_agg->ShouldDelete(
-            test_case.lookup_key, RangeDelPositioningMode::kForwardTraversal));
-  }
-  for (auto it = test_cases.rbegin(); it != test_cases.rend(); ++it) {
-    const auto& test_case = *it;
-    EXPECT_EQ(
-        test_case.result,
-        range_del_agg->ShouldDelete(
-            test_case.lookup_key, RangeDelPositioningMode::kBackwardTraversal));
-  }
-}
-
-void VerifyIsRangeOverlapped(
-    RangeDelAggregatorV2* range_del_agg,
-    const std::vector<IsRangeOverlappedTestCase>& test_cases) {
-  for (const auto& test_case : test_cases) {
-    EXPECT_EQ(test_case.result,
-              range_del_agg->IsRangeOverlapped(test_case.start, test_case.end));
-  }
-}
-
-}  // namespace
-
-TEST_F(RangeDelAggregatorV2Test, EmptyTruncatedIter) {
-  auto range_del_iter = MakeRangeDelIter({});
-  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
-                                             bytewise_icmp);
-  std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-      new FragmentedRangeTombstoneIterator(&fragment_list, kMaxSequenceNumber,
-                                           bytewise_icmp));
-
-  TruncatedRangeDelIterator iter(std::move(input_iter), &bytewise_icmp, nullptr,
-                                 nullptr);
-
-  iter.SeekToFirst();
-  ASSERT_FALSE(iter.Valid());
-
-  iter.SeekToLast();
-  ASSERT_FALSE(iter.Valid());
-}
-
-TEST_F(RangeDelAggregatorV2Test, UntruncatedIter) {
-  auto range_del_iter =
-      MakeRangeDelIter({{"a", "e", 10}, {"e", "g", 8}, {"j", "n", 4}});
-  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
-                                             bytewise_icmp);
-  std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-      new FragmentedRangeTombstoneIterator(&fragment_list, kMaxSequenceNumber,
-                                           bytewise_icmp));
-
-  TruncatedRangeDelIterator iter(std::move(input_iter), &bytewise_icmp, nullptr,
-                                 nullptr);
-
-  VerifyIterator(&iter, bytewise_icmp,
-                 {{UncutEndpoint("a"), UncutEndpoint("e"), 10},
-                  {UncutEndpoint("e"), UncutEndpoint("g"), 8},
-                  {UncutEndpoint("j"), UncutEndpoint("n"), 4}});
-
-  VerifySeek(
-      &iter, bytewise_icmp,
-      {{"d", UncutEndpoint("a"), UncutEndpoint("e"), 10},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("j"), UncutEndpoint("n"), 4},
-       {"n", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */},
-       {"", UncutEndpoint("a"), UncutEndpoint("e"), 10}});
-
-  VerifySeekForPrev(
-      &iter, bytewise_icmp,
-      {{"d", UncutEndpoint("a"), UncutEndpoint("e"), 10},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"n", UncutEndpoint("j"), UncutEndpoint("n"), 4},
-       {"", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, UntruncatedIterWithSnapshot) {
-  auto range_del_iter =
-      MakeRangeDelIter({{"a", "e", 10}, {"e", "g", 8}, {"j", "n", 4}});
-  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
-                                             bytewise_icmp);
-  std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-      new FragmentedRangeTombstoneIterator(&fragment_list, 9 /* snapshot */,
-                                           bytewise_icmp));
-
-  TruncatedRangeDelIterator iter(std::move(input_iter), &bytewise_icmp, nullptr,
-                                 nullptr);
-
-  VerifyIterator(&iter, bytewise_icmp,
-                 {{UncutEndpoint("e"), UncutEndpoint("g"), 8},
-                  {UncutEndpoint("j"), UncutEndpoint("n"), 4}});
-
-  VerifySeek(
-      &iter, bytewise_icmp,
-      {{"d", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("j"), UncutEndpoint("n"), 4},
-       {"n", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */},
-       {"", UncutEndpoint("e"), UncutEndpoint("g"), 8}});
-
-  VerifySeekForPrev(
-      &iter, bytewise_icmp,
-      {{"d", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"n", UncutEndpoint("j"), UncutEndpoint("n"), 4},
-       {"", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, TruncatedIter) {
-  auto range_del_iter =
-      MakeRangeDelIter({{"a", "e", 10}, {"e", "g", 8}, {"j", "n", 4}});
-  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
-                                             bytewise_icmp);
-  std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-      new FragmentedRangeTombstoneIterator(&fragment_list, kMaxSequenceNumber,
-                                           bytewise_icmp));
-
-  InternalKey smallest("d", 7, kTypeValue);
-  InternalKey largest("m", 9, kTypeValue);
-  TruncatedRangeDelIterator iter(std::move(input_iter), &bytewise_icmp,
-                                 &smallest, &largest);
-
-  VerifyIterator(&iter, bytewise_icmp,
-                 {{InternalValue("d", 7), UncutEndpoint("e"), 10},
-                  {UncutEndpoint("e"), UncutEndpoint("g"), 8},
-                  {UncutEndpoint("j"), InternalValue("m", 8), 4}});
-
-  VerifySeek(
-      &iter, bytewise_icmp,
-      {{"d", InternalValue("d", 7), UncutEndpoint("e"), 10},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("j"), InternalValue("m", 8), 4},
-       {"n", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */},
-       {"", InternalValue("d", 7), UncutEndpoint("e"), 10}});
-
-  VerifySeekForPrev(
-      &iter, bytewise_icmp,
-      {{"d", InternalValue("d", 7), UncutEndpoint("e"), 10},
-       {"e", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"ia", UncutEndpoint("e"), UncutEndpoint("g"), 8},
-       {"n", UncutEndpoint("j"), InternalValue("m", 8), 4},
-       {"", UncutEndpoint(""), UncutEndpoint(""), 0, true /* invalid */}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, SingleIterInAggregator) {
-  auto range_del_iter = MakeRangeDelIter({{"a", "e", 10}, {"c", "g", 8}});
-  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
-                                             bytewise_icmp);
-  std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-      new FragmentedRangeTombstoneIterator(&fragment_list, kMaxSequenceNumber,
-                                           bytewise_icmp));
-
-  RangeDelAggregatorV2 range_del_agg(&bytewise_icmp, kMaxSequenceNumber);
-  range_del_agg.AddTombstones(std::move(input_iter));
-
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("a", 19), false},
-                                      {InternalValue("b", 9), true},
-                                      {InternalValue("d", 9), true},
-                                      {InternalValue("e", 7), true},
-                                      {InternalValue("g", 7), false}});
-
-  VerifyIsRangeOverlapped(&range_del_agg, {{"", "_", false},
-                                           {"_", "a", true},
-                                           {"a", "c", true},
-                                           {"d", "f", true},
-                                           {"g", "l", false}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, MultipleItersInAggregator) {
-  auto fragment_lists = MakeFragmentedTombstoneLists(
-      {{{"a", "e", 10}, {"c", "g", 8}},
-       {{"a", "b", 20}, {"h", "i", 25}, {"ii", "j", 15}}});
-
-  RangeDelAggregatorV2 range_del_agg(&bytewise_icmp, kMaxSequenceNumber);
-  for (const auto& fragment_list : fragment_lists) {
-    std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-        new FragmentedRangeTombstoneIterator(
-            fragment_list.get(), kMaxSequenceNumber, bytewise_icmp));
-    range_del_agg.AddTombstones(std::move(input_iter));
-  }
-
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("a", 19), true},
-                                      {InternalValue("b", 19), false},
-                                      {InternalValue("b", 9), true},
-                                      {InternalValue("d", 9), true},
-                                      {InternalValue("e", 7), true},
-                                      {InternalValue("g", 7), false},
-                                      {InternalValue("h", 24), true},
-                                      {InternalValue("i", 24), false},
-                                      {InternalValue("ii", 14), true},
-                                      {InternalValue("j", 14), false}});
-
-  VerifyIsRangeOverlapped(&range_del_agg, {{"", "_", false},
-                                           {"_", "a", true},
-                                           {"a", "c", true},
-                                           {"d", "f", true},
-                                           {"g", "l", true},
-                                           {"x", "y", false}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, MultipleItersInAggregatorWithUpperBound) {
-  auto fragment_lists = MakeFragmentedTombstoneLists(
-      {{{"a", "e", 10}, {"c", "g", 8}},
-       {{"a", "b", 20}, {"h", "i", 25}, {"ii", "j", 15}}});
-
-  RangeDelAggregatorV2 range_del_agg(&bytewise_icmp, 19);
-  for (const auto& fragment_list : fragment_lists) {
-    std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-        new FragmentedRangeTombstoneIterator(fragment_list.get(),
-                                             19 /* snapshot */, bytewise_icmp));
-    range_del_agg.AddTombstones(std::move(input_iter));
-  }
-
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("a", 19), false},
-                                      {InternalValue("a", 9), true},
-                                      {InternalValue("b", 9), true},
-                                      {InternalValue("d", 9), true},
-                                      {InternalValue("e", 7), true},
-                                      {InternalValue("g", 7), false},
-                                      {InternalValue("h", 24), false},
-                                      {InternalValue("i", 24), false},
-                                      {InternalValue("ii", 14), true},
-                                      {InternalValue("j", 14), false}});
-
-  VerifyIsRangeOverlapped(&range_del_agg, {{"", "_", false},
-                                           {"_", "a", true},
-                                           {"a", "c", true},
-                                           {"d", "f", true},
-                                           {"g", "l", true},
-                                           {"x", "y", false}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, MultipleTruncatedItersInAggregator) {
-  auto fragment_lists = MakeFragmentedTombstoneLists(
-      {{{"a", "z", 10}}, {{"a", "z", 10}}, {{"a", "z", 10}}});
-  std::vector<std::pair<InternalKey, InternalKey>> iter_bounds = {
-      {InternalKey("a", 4, kTypeValue),
-       InternalKey("m", kMaxSequenceNumber, kTypeRangeDeletion)},
-      {InternalKey("m", 20, kTypeValue),
-       InternalKey("x", kMaxSequenceNumber, kTypeRangeDeletion)},
-      {InternalKey("x", 5, kTypeValue), InternalKey("zz", 30, kTypeValue)}};
-
-  RangeDelAggregatorV2 range_del_agg(&bytewise_icmp, 19);
-  for (size_t i = 0; i < fragment_lists.size(); i++) {
-    const auto& fragment_list = fragment_lists[i];
-    const auto& bounds = iter_bounds[i];
-    std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-        new FragmentedRangeTombstoneIterator(fragment_list.get(),
-                                             19 /* snapshot */, bytewise_icmp));
-    range_del_agg.AddTombstones(std::move(input_iter), &bounds.first,
-                                &bounds.second);
-  }
-
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("a", 10), false},
-                                      {InternalValue("a", 9), false},
-                                      {InternalValue("a", 4), true},
-                                      {InternalValue("m", 10), false},
-                                      {InternalValue("m", 9), true},
-                                      {InternalValue("x", 10), false},
-                                      {InternalValue("x", 9), false},
-                                      {InternalValue("x", 5), true},
-                                      {InternalValue("z", 9), false}});
-
-  VerifyIsRangeOverlapped(&range_del_agg, {{"", "_", false},
-                                           {"_", "a", true},
-                                           {"a", "n", true},
-                                           {"l", "x", true},
-                                           {"w", "z", true},
-                                           {"zzz", "zz", false},
-                                           {"zz", "zzz", false}});
-}
-
-TEST_F(RangeDelAggregatorV2Test, MultipleTruncatedItersInAggregatorSameLevel) {
-  auto fragment_lists = MakeFragmentedTombstoneLists(
-      {{{"a", "z", 10}}, {{"a", "z", 10}}, {{"a", "z", 10}}});
-  std::vector<std::pair<InternalKey, InternalKey>> iter_bounds = {
-      {InternalKey("a", 4, kTypeValue),
-       InternalKey("m", kMaxSequenceNumber, kTypeRangeDeletion)},
-      {InternalKey("m", 20, kTypeValue),
-       InternalKey("x", kMaxSequenceNumber, kTypeRangeDeletion)},
-      {InternalKey("x", 5, kTypeValue), InternalKey("zz", 30, kTypeValue)}};
-
-  RangeDelAggregatorV2 range_del_agg(&bytewise_icmp, 19);
-
-  auto add_iter_to_agg = [&](size_t i) {
-    std::unique_ptr<FragmentedRangeTombstoneIterator> input_iter(
-        new FragmentedRangeTombstoneIterator(fragment_lists[i].get(),
-                                             19 /* snapshot */, bytewise_icmp));
-    range_del_agg.AddTombstones(std::move(input_iter), &iter_bounds[i].first,
-                                &iter_bounds[i].second);
-  };
-
-  add_iter_to_agg(0);
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("a", 10), false},
-                                      {InternalValue("a", 9), false},
-                                      {InternalValue("a", 4), true}});
-
-  add_iter_to_agg(1);
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("m", 10), false},
-                                      {InternalValue("m", 9), true}});
-
-  add_iter_to_agg(2);
-  VerifyShouldDelete(&range_del_agg, {{InternalValue("x", 10), false},
-                                      {InternalValue("x", 9), false},
-                                      {InternalValue("x", 5), true},
-                                      {InternalValue("z", 9), false}});
-
-  VerifyIsRangeOverlapped(&range_del_agg, {{"", "_", false},
-                                           {"_", "a", true},
-                                           {"a", "n", true},
-                                           {"l", "x", true},
-                                           {"w", "z", true},
-                                           {"zzz", "zz", false},
-                                           {"zz", "zzz", false}});
-}
-
-}  // namespace rocksdb
-
-int main(int argc, char** argv) {
-  ::testing::InitGoogleTest(&argc, argv);
-  return RUN_ALL_TESTS();
-}

db/range_tombstone_fragmenter.cc

@@ -20,7 +20,8 @@ namespace rocksdb {
 
 FragmentedRangeTombstoneList::FragmentedRangeTombstoneList(
     std::unique_ptr<InternalIterator> unfragmented_tombstones,
-    const InternalKeyComparator& icmp) {
+    const InternalKeyComparator& icmp, bool for_compaction,
+    const std::vector<SequenceNumber>& snapshots) {
   if (unfragmented_tombstones == nullptr) {
     return;
   }
@@ -43,7 +44,8 @@ FragmentedRangeTombstoneList::FragmentedRangeTombstoneList(
     }
   }
   if (is_sorted) {
-    FragmentTombstones(std::move(unfragmented_tombstones), icmp);
+    FragmentTombstones(std::move(unfragmented_tombstones), icmp, for_compaction,
+                       snapshots);
     return;
   }
 
@@ -61,12 +63,13 @@ FragmentedRangeTombstoneList::FragmentedRangeTombstoneList(
   // VectorIterator implicitly sorts by key during construction.
   auto iter = std::unique_ptr<VectorIterator>(
       new VectorIterator(std::move(keys), std::move(values), &icmp));
-  FragmentTombstones(std::move(iter), icmp);
+  FragmentTombstones(std::move(iter), icmp, for_compaction, snapshots);
 }
 
 void FragmentedRangeTombstoneList::FragmentTombstones(
     std::unique_ptr<InternalIterator> unfragmented_tombstones,
-    const InternalKeyComparator& icmp) {
+    const InternalKeyComparator& icmp, bool for_compaction,
+    const std::vector<SequenceNumber>& snapshots) {
   Slice cur_start_key(nullptr, 0);
   auto cmp = ParsedInternalKeyComparator(&icmp);
 
@@ -117,10 +120,38 @@ void FragmentedRangeTombstoneList::FragmentTombstones(
       }
       std::sort(seqnums_to_flush.begin(), seqnums_to_flush.end(),
                 std::greater<SequenceNumber>());
+
       size_t start_idx = tombstone_seqs_.size();
       size_t end_idx = start_idx + seqnums_to_flush.size();
+
+      if (for_compaction) {
+        // Drop all tombstone seqnums that are not preserved by a snapshot.
+        SequenceNumber next_snapshot = kMaxSequenceNumber;
+        for (auto seq : seqnums_to_flush) {
+          if (seq <= next_snapshot) {
+            // This seqnum is visible by a lower snapshot.
+            tombstone_seqs_.push_back(seq);
+            seq_set_.insert(seq);
+            auto upper_bound_it =
+                std::lower_bound(snapshots.begin(), snapshots.end(), seq);
+            if (upper_bound_it == snapshots.begin()) {
+              // This seqnum is the topmost one visible by the earliest
+              // snapshot. None of the seqnums below it will be visible, so we
+              // can skip them.
+              break;
+            }
+            next_snapshot = *std::prev(upper_bound_it);
+          }
+        }
+        end_idx = tombstone_seqs_.size();
+      } else {
+        // The fragmentation is being done for reads, so preserve all seqnums.
         tombstone_seqs_.insert(tombstone_seqs_.end(), seqnums_to_flush.begin(),
                                seqnums_to_flush.end());
+        seq_set_.insert(seqnums_to_flush.begin(), seqnums_to_flush.end());
+      }
+
+      assert(start_idx < end_idx);
       tombstones_.emplace_back(cur_start_key, cur_end_key, start_idx, end_idx);
 
       cur_start_key = cur_end_key;
@@ -143,6 +174,11 @@ void FragmentedRangeTombstoneList::FragmentTombstones(
     const Slice& ikey = unfragmented_tombstones->key();
     Slice tombstone_start_key = ExtractUserKey(ikey);
     SequenceNumber tombstone_seq = GetInternalKeySeqno(ikey);
+    if (!unfragmented_tombstones->IsKeyPinned()) {
+      pinned_slices_.emplace_back(tombstone_start_key.data(),
+                                  tombstone_start_key.size());
+      tombstone_start_key = pinned_slices_.back();
+    }
     no_tombstones = false;
 
     Slice tombstone_end_key = unfragmented_tombstones->value();
@@ -157,13 +193,7 @@ void FragmentedRangeTombstoneList::FragmentTombstones(
       // this new start key.
       flush_current_tombstones(tombstone_start_key);
     }
-    if (unfragmented_tombstones->IsKeyPinned()) {
     cur_start_key = tombstone_start_key;
-    } else {
-      pinned_slices_.emplace_back(tombstone_start_key.data(),
-                                  tombstone_start_key.size());
-      cur_start_key = pinned_slices_.back();
-    }
 
     cur_end_keys.emplace(tombstone_end_key, tombstone_seq, kTypeRangeDeletion);
   }
@@ -178,33 +208,41 @@ void FragmentedRangeTombstoneList::FragmentTombstones(
   }
 }
 
+bool FragmentedRangeTombstoneList::ContainsRange(SequenceNumber lower,
+                                                 SequenceNumber upper) const {
+  auto seq_it = seq_set_.lower_bound(lower);
+  return seq_it != seq_set_.end() && *seq_it <= upper;
+}
+
 FragmentedRangeTombstoneIterator::FragmentedRangeTombstoneIterator(
-    const FragmentedRangeTombstoneList* tombstones, SequenceNumber snapshot,
+    const FragmentedRangeTombstoneList* tombstones,
-    const InternalKeyComparator& icmp)
+    const InternalKeyComparator& icmp, SequenceNumber _upper_bound,
+    SequenceNumber _lower_bound)
     : tombstone_start_cmp_(icmp.user_comparator()),
       tombstone_end_cmp_(icmp.user_comparator()),
+      icmp_(&icmp),
       ucmp_(icmp.user_comparator()),
       tombstones_(tombstones),
-      snapshot_(snapshot) {
+      upper_bound_(_upper_bound),
+      lower_bound_(_lower_bound) {
   assert(tombstones_ != nullptr);
-  pos_ = tombstones_->end();
+  Invalidate();
-  pinned_pos_ = tombstones_->end();
 }
 
 FragmentedRangeTombstoneIterator::FragmentedRangeTombstoneIterator(
     const std::shared_ptr<const FragmentedRangeTombstoneList>& tombstones,
-    SequenceNumber snapshot, const InternalKeyComparator& icmp)
+    const InternalKeyComparator& icmp, SequenceNumber _upper_bound,
+    SequenceNumber _lower_bound)
     : tombstone_start_cmp_(icmp.user_comparator()),
       tombstone_end_cmp_(icmp.user_comparator()),
+      icmp_(&icmp),
       ucmp_(icmp.user_comparator()),
       tombstones_ref_(tombstones),
       tombstones_(tombstones_ref_.get()),
-      snapshot_(snapshot) {
+      upper_bound_(_upper_bound),
+      lower_bound_(_lower_bound) {
   assert(tombstones_ != nullptr);
-  pos_ = tombstones_->end();
+  Invalidate();
-  seq_pos_ = tombstones_->seq_end();
-  pinned_pos_ = tombstones_->end();
-  pinned_seq_pos_ = tombstones_->seq_end();
 }
 
 void FragmentedRangeTombstoneIterator::SeekToFirst() {
@@ -220,7 +258,7 @@ void FragmentedRangeTombstoneIterator::SeekToTopFirst() {
   pos_ = tombstones_->begin();
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
   ScanForwardToVisibleTombstone();
 }
 
@@ -237,7 +275,7 @@ void FragmentedRangeTombstoneIterator::SeekToTopLast() {
   pos_ = std::prev(tombstones_->end());
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
   ScanBackwardToVisibleTombstone();
 }
 
@@ -270,7 +308,7 @@ void FragmentedRangeTombstoneIterator::SeekToCoveringTombstone(
   }
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
 }
 
 void FragmentedRangeTombstoneIterator::SeekForPrevToCoveringTombstone(
@@ -289,25 +327,28 @@ void FragmentedRangeTombstoneIterator::SeekForPrevToCoveringTombstone(
   --pos_;
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
 }
 
 void FragmentedRangeTombstoneIterator::ScanForwardToVisibleTombstone() {
   while (pos_ != tombstones_->end() &&
-         seq_pos_ == tombstones_->seq_iter(pos_->seq_end_idx)) {
+         (seq_pos_ == tombstones_->seq_iter(pos_->seq_end_idx) ||
+          *seq_pos_ < lower_bound_)) {
     ++pos_;
     if (pos_ == tombstones_->end()) {
+      Invalidate();
       return;
     }
     seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                                 tombstones_->seq_iter(pos_->seq_end_idx),
-                                snapshot_, std::greater<SequenceNumber>());
+                                upper_bound_, std::greater<SequenceNumber>());
   }
 }
 
 void FragmentedRangeTombstoneIterator::ScanBackwardToVisibleTombstone() {
   while (pos_ != tombstones_->end() &&
-         seq_pos_ == tombstones_->seq_iter(pos_->seq_end_idx)) {
+         (seq_pos_ == tombstones_->seq_iter(pos_->seq_end_idx) ||
+          *seq_pos_ < lower_bound_)) {
     if (pos_ == tombstones_->begin()) {
       Invalidate();
       return;
@@ -315,7 +356,7 @@ void FragmentedRangeTombstoneIterator::ScanBackwardToVisibleTombstone() {
     --pos_;
     seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                                 tombstones_->seq_iter(pos_->seq_end_idx),
-                                snapshot_, std::greater<SequenceNumber>());
+                                upper_bound_, std::greater<SequenceNumber>());
   }
 }
 
@@ -333,14 +374,13 @@ void FragmentedRangeTombstoneIterator::TopNext() {
   }
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
   ScanForwardToVisibleTombstone();
 }
 
 void FragmentedRangeTombstoneIterator::Prev() {
   if (seq_pos_ == tombstones_->seq_begin()) {
-    pos_ = tombstones_->end();
+    Invalidate();
-    seq_pos_ = tombstones_->seq_end();
     return;
   }
   --seq_pos_;
@@ -358,7 +398,7 @@ void FragmentedRangeTombstoneIterator::TopPrev() {
   --pos_;
   seq_pos_ = std::lower_bound(tombstones_->seq_iter(pos_->seq_start_idx),
                               tombstones_->seq_iter(pos_->seq_end_idx),
-                              snapshot_, std::greater<SequenceNumber>());
+                              upper_bound_, std::greater<SequenceNumber>());
   ScanBackwardToVisibleTombstone();
 }
 
@@ -372,4 +412,27 @@ SequenceNumber FragmentedRangeTombstoneIterator::MaxCoveringTombstoneSeqnum(
   return ValidPos() && ucmp_->Compare(start_key(), user_key) <= 0 ? seq() : 0;
 }
 
+std::map<SequenceNumber, std::unique_ptr<FragmentedRangeTombstoneIterator>>
+FragmentedRangeTombstoneIterator::SplitBySnapshot(
+    const std::vector<SequenceNumber>& snapshots) {
+  std::map<SequenceNumber, std::unique_ptr<FragmentedRangeTombstoneIterator>>
+      splits;
+  SequenceNumber lower = 0;
+  SequenceNumber upper;
+  for (size_t i = 0; i <= snapshots.size(); i++) {
+    if (i >= snapshots.size()) {
+      upper = kMaxSequenceNumber;
+    } else {
+      upper = snapshots[i];
+    }
+    if (tombstones_->ContainsRange(lower, upper)) {
+      splits.emplace(upper, std::unique_ptr<FragmentedRangeTombstoneIterator>(
+                                new FragmentedRangeTombstoneIterator(
+                                    tombstones_, *icmp_, upper, lower)));
+    }
+    lower = upper + 1;
+  }
+  return splits;
+}
+
 }  // namespace rocksdb

db/range_tombstone_fragmenter.h

@@ -7,6 +7,7 @@
 
 #include <list>
 #include <memory>
+#include <set>
 #include <string>
 #include <vector>
 
@@ -38,7 +39,8 @@ struct FragmentedRangeTombstoneList {
   };
   FragmentedRangeTombstoneList(
       std::unique_ptr<InternalIterator> unfragmented_tombstones,
-      const InternalKeyComparator& icmp);
+      const InternalKeyComparator& icmp, bool for_compaction = false,
+      const std::vector<SequenceNumber>& snapshots = {});
 
   std::vector<RangeTombstoneStack>::const_iterator begin() const {
     return tombstones_.begin();
@@ -60,7 +62,11 @@ struct FragmentedRangeTombstoneList {
     return tombstone_seqs_.end();
   }
 
-  bool empty() const { return tombstones_.size() == 0; }
+  bool empty() const { return tombstones_.empty(); }
+
+  // Returns true if the stored tombstones contain with one with a sequence
+  // number in [lower, upper].
+  bool ContainsRange(SequenceNumber lower, SequenceNumber upper) const;
 
  private:
   // Given an ordered range tombstone iterator unfragmented_tombstones,
@@ -68,10 +74,12 @@ struct FragmentedRangeTombstoneList {
   // tombstones_ and tombstone_seqs_.
   void FragmentTombstones(
       std::unique_ptr<InternalIterator> unfragmented_tombstones,
-      const InternalKeyComparator& icmp);
+      const InternalKeyComparator& icmp, bool for_compaction,
+      const std::vector<SequenceNumber>& snapshots);
 
   std::vector<RangeTombstoneStack> tombstones_;
   std::vector<SequenceNumber> tombstone_seqs_;
+  std::set<SequenceNumber> seq_set_;
   std::list<std::string> pinned_slices_;
   PinnedIteratorsManager pinned_iters_mgr_;
 };
@@ -88,11 +96,13 @@ struct FragmentedRangeTombstoneList {
 class FragmentedRangeTombstoneIterator : public InternalIterator {
  public:
   FragmentedRangeTombstoneIterator(
-      const FragmentedRangeTombstoneList* tombstones, SequenceNumber snapshot,
+      const FragmentedRangeTombstoneList* tombstones,
-      const InternalKeyComparator& icmp);
+      const InternalKeyComparator& icmp, SequenceNumber upper_bound,
+      SequenceNumber lower_bound = 0);
   FragmentedRangeTombstoneIterator(
       const std::shared_ptr<const FragmentedRangeTombstoneList>& tombstones,
-      SequenceNumber snapshot, const InternalKeyComparator& icmp);
+      const InternalKeyComparator& icmp, SequenceNumber upper_bound,
+      SequenceNumber lower_bound = 0);
 
   void SeekToFirst() override;
   void SeekToLast() override;
@@ -136,8 +146,7 @@ class FragmentedRangeTombstoneIterator : public InternalIterator {
     seq_pos_ = tombstones_->seq_end();
   }
 
-  // TODO: implement properly
+  RangeTombstone Tombstone() const {
-  RangeTombstone tombstone() const {
     return RangeTombstone(start_key(), end_key(), seq());
   }
   Slice start_key() const { return pos_->start_key; }
@@ -151,12 +160,24 @@ class FragmentedRangeTombstoneIterator : public InternalIterator {
     return ParsedInternalKey(pos_->end_key, kMaxSequenceNumber,
                              kTypeRangeDeletion);
   }
-  ParsedInternalKey internal_key() const {
-    return ParsedInternalKey(pos_->start_key, *seq_pos_, kTypeRangeDeletion);
-  }
 
   SequenceNumber MaxCoveringTombstoneSeqnum(const Slice& user_key);
 
+  // Splits the iterator into n+1 iterators (where n is the number of
+  // snapshots), each providing a view over a "stripe" of sequence numbers. The
+  // iterators are keyed by the upper bound of their ranges (the provided
+  // snapshots + kMaxSequenceNumber).
+  //
+  // NOTE: the iterators in the returned map are no longer valid if their
+  // parent iterator is deleted, since they do not modify the refcount of the
+  // underlying tombstone list. Therefore, this map should be deleted before
+  // the parent iterator.
+  std::map<SequenceNumber, std::unique_ptr<FragmentedRangeTombstoneIterator>>
+  SplitBySnapshot(const std::vector<SequenceNumber>& snapshots);
+
+  SequenceNumber upper_bound() const { return upper_bound_; }
+  SequenceNumber lower_bound() const { return lower_bound_; }
+
  private:
   using RangeTombstoneStack = FragmentedRangeTombstoneList::RangeTombstoneStack;
 
@@ -217,10 +238,12 @@ class FragmentedRangeTombstoneIterator : public InternalIterator {
 
   const RangeTombstoneStackStartComparator tombstone_start_cmp_;
   const RangeTombstoneStackEndComparator tombstone_end_cmp_;
+  const InternalKeyComparator* icmp_;
   const Comparator* ucmp_;
   std::shared_ptr<const FragmentedRangeTombstoneList> tombstones_ref_;
   const FragmentedRangeTombstoneList* tombstones_;
-  SequenceNumber snapshot_;
+  SequenceNumber upper_bound_;
+  SequenceNumber lower_bound_;
   std::vector<RangeTombstoneStack>::const_iterator pos_;
   std::vector<SequenceNumber>::const_iterator seq_pos_;
   mutable std::vector<RangeTombstoneStack>::const_iterator pinned_pos_;

db/range_tombstone_fragmenter_test.cc

@@ -29,15 +29,26 @@ std::unique_ptr<InternalIterator> MakeRangeDelIter(
       new test::VectorIterator(keys, values));
 }
 
+void CheckIterPosition(const RangeTombstone& tombstone,
+                       const FragmentedRangeTombstoneIterator* iter) {
+  // Test InternalIterator interface.
+  EXPECT_EQ(tombstone.start_key_, ExtractUserKey(iter->key()));
+  EXPECT_EQ(tombstone.end_key_, iter->value());
+  EXPECT_EQ(tombstone.seq_, iter->seq());
+
+  // Test FragmentedRangeTombstoneIterator interface.
+  EXPECT_EQ(tombstone.start_key_, iter->start_key());
+  EXPECT_EQ(tombstone.end_key_, iter->end_key());
+  EXPECT_EQ(tombstone.seq_, GetInternalKeySeqno(iter->key()));
+}
+
 void VerifyFragmentedRangeDels(
     FragmentedRangeTombstoneIterator* iter,
     const std::vector<RangeTombstone>& expected_tombstones) {
   iter->SeekToFirst();
-  for (size_t i = 0; i < expected_tombstones.size() && iter->Valid();
+  for (size_t i = 0; i < expected_tombstones.size(); i++, iter->Next()) {
-       i++, iter->Next()) {
+    ASSERT_TRUE(iter->Valid());
-    EXPECT_EQ(iter->start_key(), expected_tombstones[i].start_key_);
+    CheckIterPosition(expected_tombstones[i], iter);
-    EXPECT_EQ(iter->value(), expected_tombstones[i].end_key_);
-    EXPECT_EQ(iter->seq(), expected_tombstones[i].seq_);
   }
   EXPECT_FALSE(iter->Valid());
 }
@@ -46,11 +57,9 @@ void VerifyVisibleTombstones(
     FragmentedRangeTombstoneIterator* iter,
     const std::vector<RangeTombstone>& expected_tombstones) {
   iter->SeekToTopFirst();
-  for (size_t i = 0; i < expected_tombstones.size() && iter->Valid();
+  for (size_t i = 0; i < expected_tombstones.size(); i++, iter->TopNext()) {
-       i++, iter->TopNext()) {
+    ASSERT_TRUE(iter->Valid());
-    EXPECT_EQ(iter->start_key(), expected_tombstones[i].start_key_);
+    CheckIterPosition(expected_tombstones[i], iter);
-    EXPECT_EQ(iter->value(), expected_tombstones[i].end_key_);
-    EXPECT_EQ(iter->seq(), expected_tombstones[i].seq_);
   }
   EXPECT_FALSE(iter->Valid());
 }
@@ -69,9 +78,7 @@ void VerifySeek(FragmentedRangeTombstoneIterator* iter,
       ASSERT_FALSE(iter->Valid());
     } else {
       ASSERT_TRUE(iter->Valid());
-      EXPECT_EQ(testcase.expected_position.start_key_, iter->start_key());
+      CheckIterPosition(testcase.expected_position, iter);
-      EXPECT_EQ(testcase.expected_position.end_key_, iter->value());
-      EXPECT_EQ(testcase.expected_position.seq_, iter->seq());
     }
   }
 }
@@ -84,9 +91,7 @@ void VerifySeekForPrev(FragmentedRangeTombstoneIterator* iter,
       ASSERT_FALSE(iter->Valid());
     } else {
       ASSERT_TRUE(iter->Valid());
-      EXPECT_EQ(testcase.expected_position.start_key_, iter->start_key());
+      CheckIterPosition(testcase.expected_position, iter);
-      EXPECT_EQ(testcase.expected_position.end_key_, iter->value());
-      EXPECT_EQ(testcase.expected_position.seq_, iter->seq());
     }
   }
 }
@@ -112,8 +117,10 @@ TEST_F(RangeTombstoneFragmenterTest, NonOverlappingTombstones) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(&iter, {{"a", "b", 10}, {"c", "d", 5}});
   VerifyMaxCoveringTombstoneSeqnum(&iter,
                                    {{"", 0}, {"a", 10}, {"b", 0}, {"c", 5}});
@@ -124,8 +131,10 @@ TEST_F(RangeTombstoneFragmenterTest, OverlappingTombstones) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(
       &iter, {{"a", "c", 10}, {"c", "e", 15}, {"c", "e", 10}, {"e", "g", 15}});
   VerifyMaxCoveringTombstoneSeqnum(&iter,
@@ -138,8 +147,10 @@ TEST_F(RangeTombstoneFragmenterTest, ContiguousTombstones) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(
       &iter, {{"a", "c", 10}, {"c", "e", 20}, {"c", "e", 5}, {"e", "g", 15}});
   VerifyMaxCoveringTombstoneSeqnum(&iter,
@@ -152,8 +163,10 @@ TEST_F(RangeTombstoneFragmenterTest, RepeatedStartAndEndKey) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(&iter,
                             {{"a", "c", 10}, {"a", "c", 7}, {"a", "c", 3}});
   VerifyMaxCoveringTombstoneSeqnum(&iter, {{"a", 10}, {"b", 10}, {"c", 0}});
@@ -165,8 +178,10 @@ TEST_F(RangeTombstoneFragmenterTest, RepeatedStartKeyDifferentEndKeys) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(&iter, {{"a", "c", 10},
                                     {"a", "c", 7},
                                     {"a", "c", 3},
@@ -186,8 +201,10 @@ TEST_F(RangeTombstoneFragmenterTest, RepeatedStartKeyMixedEndKeys) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter.upper_bound());
   VerifyFragmentedRangeDels(&iter, {{"a", "c", 30},
                                     {"a", "c", 20},
                                     {"a", "c", 10},
@@ -211,16 +228,16 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKey) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter1(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter1(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         kMaxSequenceNumber);
-  FragmentedRangeTombstoneIterator iter2(&fragment_list, 9 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter2(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         9 /* upper_bound */);
-  FragmentedRangeTombstoneIterator iter3(&fragment_list, 7 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter3(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         7 /* upper_bound */);
-  FragmentedRangeTombstoneIterator iter4(&fragment_list, 5 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter4(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         5 /* upper_bound */);
-  FragmentedRangeTombstoneIterator iter5(&fragment_list, 3 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter5(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         3 /* upper_bound */);
   for (auto* iter : {&iter1, &iter2, &iter3, &iter4, &iter5}) {
     VerifyFragmentedRangeDels(iter, {{"a", "c", 10},
                                      {"c", "e", 10},
@@ -234,6 +251,8 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKey) {
                                      {"l", "n", 4}});
   }
 
+  ASSERT_EQ(0, iter1.lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, iter1.upper_bound());
   VerifyVisibleTombstones(&iter1, {{"a", "c", 10},
                                    {"c", "e", 10},
                                    {"e", "g", 8},
@@ -243,6 +262,8 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKey) {
   VerifyMaxCoveringTombstoneSeqnum(
       &iter1, {{"a", 10}, {"c", 10}, {"e", 8}, {"i", 0}, {"j", 4}, {"m", 4}});
 
+  ASSERT_EQ(0, iter2.lower_bound());
+  ASSERT_EQ(9, iter2.upper_bound());
   VerifyVisibleTombstones(&iter2, {{"c", "e", 8},
                                    {"e", "g", 8},
                                    {"g", "i", 6},
@@ -251,6 +272,8 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKey) {
   VerifyMaxCoveringTombstoneSeqnum(
       &iter2, {{"a", 0}, {"c", 8}, {"e", 8}, {"i", 0}, {"j", 4}, {"m", 4}});
 
+  ASSERT_EQ(0, iter3.lower_bound());
+  ASSERT_EQ(7, iter3.upper_bound());
   VerifyVisibleTombstones(&iter3, {{"c", "e", 6},
                                    {"e", "g", 6},
                                    {"g", "i", 6},
@@ -259,10 +282,14 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKey) {
   VerifyMaxCoveringTombstoneSeqnum(
       &iter3, {{"a", 0}, {"c", 6}, {"e", 6}, {"i", 0}, {"j", 4}, {"m", 4}});
 
+  ASSERT_EQ(0, iter4.lower_bound());
+  ASSERT_EQ(5, iter4.upper_bound());
   VerifyVisibleTombstones(&iter4, {{"j", "l", 4}, {"l", "n", 4}});
   VerifyMaxCoveringTombstoneSeqnum(
       &iter4, {{"a", 0}, {"c", 0}, {"e", 0}, {"i", 0}, {"j", 4}, {"m", 4}});
 
+  ASSERT_EQ(0, iter5.lower_bound());
+  ASSERT_EQ(3, iter5.upper_bound());
   VerifyVisibleTombstones(&iter5, {{"j", "l", 2}});
   VerifyMaxCoveringTombstoneSeqnum(
       &iter5, {{"a", 0}, {"c", 0}, {"e", 0}, {"i", 0}, {"j", 2}, {"m", 0}});
@@ -277,8 +304,10 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKeyUnordered) {
 
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
-  FragmentedRangeTombstoneIterator iter(&fragment_list, 9 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        9 /* upper_bound */);
+  ASSERT_EQ(0, iter.lower_bound());
+  ASSERT_EQ(9, iter.upper_bound());
   VerifyFragmentedRangeDels(&iter, {{"a", "c", 10},
                                     {"c", "e", 10},
                                     {"c", "e", 8},
@@ -293,6 +322,116 @@ TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKeyUnordered) {
       &iter, {{"a", 0}, {"c", 8}, {"e", 8}, {"i", 0}, {"j", 4}, {"m", 4}});
 }
 
+TEST_F(RangeTombstoneFragmenterTest, OverlapAndRepeatedStartKeyForCompaction) {
+  auto range_del_iter = MakeRangeDelIter({{"a", "e", 10},
+                                          {"j", "n", 4},
+                                          {"c", "i", 6},
+                                          {"c", "g", 8},
+                                          {"j", "l", 2}});
+
+  FragmentedRangeTombstoneList fragment_list(
+      std::move(range_del_iter), bytewise_icmp, true /* for_compaction */,
+      {} /* snapshots */);
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
+                                        kMaxSequenceNumber /* upper_bound */);
+  VerifyFragmentedRangeDels(&iter, {{"a", "c", 10},
+                                    {"c", "e", 10},
+                                    {"e", "g", 8},
+                                    {"g", "i", 6},
+                                    {"j", "l", 4},
+                                    {"l", "n", 4}});
+}
+
+TEST_F(RangeTombstoneFragmenterTest,
+       OverlapAndRepeatedStartKeyForCompactionWithSnapshot) {
+  auto range_del_iter = MakeRangeDelIter({{"a", "e", 10},
+                                          {"j", "n", 4},
+                                          {"c", "i", 6},
+                                          {"c", "g", 8},
+                                          {"j", "l", 2}});
+
+  FragmentedRangeTombstoneList fragment_list(
+      std::move(range_del_iter), bytewise_icmp, true /* for_compaction */,
+      {20, 9} /* upper_bounds */);
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
+                                        kMaxSequenceNumber /* upper_bound */);
+  VerifyFragmentedRangeDels(&iter, {{"a", "c", 10},
+                                    {"c", "e", 10},
+                                    {"c", "e", 8},
+                                    {"e", "g", 8},
+                                    {"g", "i", 6},
+                                    {"j", "l", 4},
+                                    {"l", "n", 4}});
+}
+
+TEST_F(RangeTombstoneFragmenterTest, IteratorSplitNoSnapshots) {
+  auto range_del_iter = MakeRangeDelIter({{"a", "e", 10},
+                                          {"j", "n", 4},
+                                          {"c", "i", 6},
+                                          {"c", "g", 8},
+                                          {"j", "l", 2}});
+
+  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
+                                             bytewise_icmp);
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
+                                        kMaxSequenceNumber /* upper_bound */);
+
+  auto split_iters = iter.SplitBySnapshot({} /* snapshots */);
+  ASSERT_EQ(1, split_iters.size());
+
+  auto* split_iter = split_iters[kMaxSequenceNumber].get();
+  ASSERT_EQ(0, split_iter->lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, split_iter->upper_bound());
+  VerifyVisibleTombstones(split_iter, {{"a", "c", 10},
+                                       {"c", "e", 10},
+                                       {"e", "g", 8},
+                                       {"g", "i", 6},
+                                       {"j", "l", 4},
+                                       {"l", "n", 4}});
+}
+
+TEST_F(RangeTombstoneFragmenterTest, IteratorSplitWithSnapshots) {
+  auto range_del_iter = MakeRangeDelIter({{"a", "e", 10},
+                                          {"j", "n", 4},
+                                          {"c", "i", 6},
+                                          {"c", "g", 8},
+                                          {"j", "l", 2}});
+
+  FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
+                                             bytewise_icmp);
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
+                                        kMaxSequenceNumber /* upper_bound */);
+
+  auto split_iters = iter.SplitBySnapshot({3, 5, 7, 9} /* snapshots */);
+  ASSERT_EQ(5, split_iters.size());
+
+  auto* split_iter1 = split_iters[3].get();
+  ASSERT_EQ(0, split_iter1->lower_bound());
+  ASSERT_EQ(3, split_iter1->upper_bound());
+  VerifyVisibleTombstones(split_iter1, {{"j", "l", 2}});
+
+  auto* split_iter2 = split_iters[5].get();
+  ASSERT_EQ(4, split_iter2->lower_bound());
+  ASSERT_EQ(5, split_iter2->upper_bound());
+  VerifyVisibleTombstones(split_iter2, {{"j", "l", 4}, {"l", "n", 4}});
+
+  auto* split_iter3 = split_iters[7].get();
+  ASSERT_EQ(6, split_iter3->lower_bound());
+  ASSERT_EQ(7, split_iter3->upper_bound());
+  VerifyVisibleTombstones(split_iter3,
+                          {{"c", "e", 6}, {"e", "g", 6}, {"g", "i", 6}});
+
+  auto* split_iter4 = split_iters[9].get();
+  ASSERT_EQ(8, split_iter4->lower_bound());
+  ASSERT_EQ(9, split_iter4->upper_bound());
+  VerifyVisibleTombstones(split_iter4, {{"c", "e", 8}, {"e", "g", 8}});
+
+  auto* split_iter5 = split_iters[kMaxSequenceNumber].get();
+  ASSERT_EQ(10, split_iter5->lower_bound());
+  ASSERT_EQ(kMaxSequenceNumber, split_iter5->upper_bound());
+  VerifyVisibleTombstones(split_iter5, {{"a", "c", 10}, {"c", "e", 10}});
+}
+
 TEST_F(RangeTombstoneFragmenterTest, SeekStartKey) {
   // Same tombstones as OverlapAndRepeatedStartKey.
   auto range_del_iter = MakeRangeDelIter({{"a", "e", 10},
@@ -304,8 +443,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekStartKey) {
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
 
-  FragmentedRangeTombstoneIterator iter1(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter1(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         kMaxSequenceNumber);
   VerifySeek(
       &iter1,
       {{"a", {"a", "c", 10}}, {"e", {"e", "g", 8}}, {"l", {"l", "n", 4}}});
@@ -313,8 +452,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekStartKey) {
       &iter1,
       {{"a", {"a", "c", 10}}, {"e", {"e", "g", 8}}, {"l", {"l", "n", 4}}});
 
-  FragmentedRangeTombstoneIterator iter2(&fragment_list, 3 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter2(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         3 /* upper_bound */);
   VerifySeek(&iter2, {{"a", {"j", "l", 2}},
                       {"e", {"j", "l", 2}},
                       {"l", {}, true /* out of range */}});
@@ -334,8 +473,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekCovered) {
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
 
-  FragmentedRangeTombstoneIterator iter1(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter1(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         kMaxSequenceNumber);
   VerifySeek(
       &iter1,
       {{"b", {"a", "c", 10}}, {"f", {"e", "g", 8}}, {"m", {"l", "n", 4}}});
@@ -343,8 +482,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekCovered) {
       &iter1,
       {{"b", {"a", "c", 10}}, {"f", {"e", "g", 8}}, {"m", {"l", "n", 4}}});
 
-  FragmentedRangeTombstoneIterator iter2(&fragment_list, 3 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter2(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         3 /* upper_bound */);
   VerifySeek(&iter2, {{"b", {"j", "l", 2}},
                       {"f", {"j", "l", 2}},
                       {"m", {}, true /* out of range */}});
@@ -364,8 +503,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekEndKey) {
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
 
-  FragmentedRangeTombstoneIterator iter1(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter1(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         kMaxSequenceNumber);
   VerifySeek(&iter1, {{"c", {"c", "e", 10}},
                       {"g", {"g", "i", 6}},
                       {"i", {"j", "l", 4}},
@@ -375,8 +514,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekEndKey) {
                              {"i", {"g", "i", 6}},
                              {"n", {"l", "n", 4}}});
 
-  FragmentedRangeTombstoneIterator iter2(&fragment_list, 3 /* snapshot */,
+  FragmentedRangeTombstoneIterator iter2(&fragment_list, bytewise_icmp,
-                                         bytewise_icmp);
+                                         3 /* upper_bound */);
   VerifySeek(&iter2, {{"c", {"j", "l", 2}},
                       {"g", {"j", "l", 2}},
                       {"i", {"j", "l", 2}},
@@ -398,8 +537,8 @@ TEST_F(RangeTombstoneFragmenterTest, SeekOutOfBounds) {
   FragmentedRangeTombstoneList fragment_list(std::move(range_del_iter),
                                              bytewise_icmp);
 
-  FragmentedRangeTombstoneIterator iter(&fragment_list, kMaxSequenceNumber,
+  FragmentedRangeTombstoneIterator iter(&fragment_list, bytewise_icmp,
-                                        bytewise_icmp);
+                                        kMaxSequenceNumber);
   VerifySeek(&iter, {{"", {"a", "c", 10}}, {"z", {}, true /* out of range */}});
   VerifySeekForPrev(&iter,
                     {{"", {}, true /* out of range */}, {"z", {"l", "n", 4}}});

db/repair.cc

@@ -417,11 +417,16 @@ class Repairer {
       SnapshotChecker* snapshot_checker = DisableGCSnapshotChecker::Instance();
 
       auto write_hint = cfd->CalculateSSTWriteHint(0);
+      std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+          range_del_iters;
+      auto range_del_iter =
+          mem->NewRangeTombstoneIterator(ro, kMaxSequenceNumber);
+      if (range_del_iter != nullptr) {
+        range_del_iters.emplace_back(range_del_iter);
+      }
       status = BuildTable(
           dbname_, env_, *cfd->ioptions(), *cfd->GetLatestMutableCFOptions(),
-          env_options_, table_cache_, iter.get(),
+          env_options_, table_cache_, iter.get(), std::move(range_del_iters),
-          std::unique_ptr<InternalIterator>(
-              mem->NewRangeTombstoneIterator(ro, vset_.LastSequence())),
           &meta, cfd->internal_comparator(),
           cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
           {}, kMaxSequenceNumber, snapshot_checker, kNoCompression,

db/table_cache.cc

@@ -185,7 +185,7 @@ Status TableCache::FindTable(const EnvOptions& env_options,
 InternalIterator* TableCache::NewIterator(
     const ReadOptions& options, const EnvOptions& env_options,
     const InternalKeyComparator& icomparator, const FileMetaData& file_meta,
-    RangeDelAggregatorV2* range_del_agg, const SliceTransform* prefix_extractor,
+    RangeDelAggregator* range_del_agg, const SliceTransform* prefix_extractor,
     TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
     bool for_compaction, Arena* arena, bool skip_filters, int level,
     const InternalKey* smallest_compaction_key,

db/table_cache.h

@@ -15,7 +15,7 @@
 #include <stdint.h>
 
 #include "db/dbformat.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "options/cf_options.h"
 #include "port/port.h"
 #include "rocksdb/cache.h"
@@ -52,7 +52,7 @@ class TableCache {
   InternalIterator* NewIterator(
       const ReadOptions& options, const EnvOptions& toptions,
       const InternalKeyComparator& internal_comparator,
-      const FileMetaData& file_meta, RangeDelAggregatorV2* range_del_agg,
+      const FileMetaData& file_meta, RangeDelAggregator* range_del_agg,
       const SliceTransform* prefix_extractor = nullptr,
       TableReader** table_reader_ptr = nullptr,
       HistogramImpl* file_read_hist = nullptr, bool for_compaction = false,

db/version_set.cc

@@ -301,17 +301,28 @@ class FilePicker {
         // On Level-n (n>=1), files are sorted. Binary search to find the
         // earliest file whose largest key >= ikey. Search left bound and
         // right bound are used to narrow the range.
-        if (search_left_bound_ == search_right_bound_) {
+        if (search_left_bound_ <= search_right_bound_) {
-          start_index = search_left_bound_;
-        } else if (search_left_bound_ < search_right_bound_) {
           if (search_right_bound_ == FileIndexer::kLevelMaxIndex) {
             search_right_bound_ =
                 static_cast<int32_t>(curr_file_level_->num_files) - 1;
           }
+          // `search_right_bound_` is an inclusive upper-bound, but since it was
+          // determined based on user key, it is still possible the lookup key
+          // falls to the right of `search_right_bound_`'s corresponding file.
+          // So, pass a limit one higher, which allows us to detect this case.
           start_index =
               FindFileInRange(*internal_comparator_, *curr_file_level_, ikey_,
                               static_cast<uint32_t>(search_left_bound_),
-                              static_cast<uint32_t>(search_right_bound_));
+                              static_cast<uint32_t>(search_right_bound_) + 1);
+          if (start_index == search_right_bound_ + 1) {
+            // `ikey_` comes after `search_right_bound_`. The lookup key does
+            // not exist on this level, so let's skip this level and do a full
+            // binary search on the next level.
+            search_left_bound_ = 0;
+            search_right_bound_ = FileIndexer::kLevelMaxIndex;
+            curr_level_++;
+            continue;
+          }
         } else {
           // search_left_bound > search_right_bound, key does not exist in
           // this level. Since no comparison is done in this level, it will
@@ -459,7 +470,7 @@ class LevelIterator final : public InternalIterator {
       const EnvOptions& env_options, const InternalKeyComparator& icomparator,
       const LevelFilesBrief* flevel, const SliceTransform* prefix_extractor,
       bool should_sample, HistogramImpl* file_read_hist, bool for_compaction,
-      bool skip_filters, int level, RangeDelAggregatorV2* range_del_agg,
+      bool skip_filters, int level, RangeDelAggregator* range_del_agg,
       const std::vector<AtomicCompactionUnitBoundary>* compaction_boundaries =
           nullptr)
       : table_cache_(table_cache),
@@ -571,7 +582,7 @@ class LevelIterator final : public InternalIterator {
   bool skip_filters_;
   size_t file_index_;
   int level_;
-  RangeDelAggregatorV2* range_del_agg_;
+  RangeDelAggregator* range_del_agg_;
   IteratorWrapper file_iter_;  // May be nullptr
   PinnedIteratorsManager* pinned_iters_mgr_;
 
@@ -985,7 +996,7 @@ double VersionStorageInfo::GetEstimatedCompressionRatioAtLevel(
 void Version::AddIterators(const ReadOptions& read_options,
                            const EnvOptions& soptions,
                            MergeIteratorBuilder* merge_iter_builder,
-                           RangeDelAggregatorV2* range_del_agg) {
+                           RangeDelAggregator* range_del_agg) {
   assert(storage_info_.finalized_);
 
   for (int level = 0; level < storage_info_.num_non_empty_levels(); level++) {
@@ -998,7 +1009,7 @@ void Version::AddIteratorsForLevel(const ReadOptions& read_options,
                                    const EnvOptions& soptions,
                                    MergeIteratorBuilder* merge_iter_builder,
                                    int level,
-                                   RangeDelAggregatorV2* range_del_agg) {
+                                   RangeDelAggregator* range_del_agg) {
   assert(storage_info_.finalized_);
   if (level >= storage_info_.num_non_empty_levels()) {
     // This is an empty level
@@ -1057,7 +1068,7 @@ Status Version::OverlapWithLevelIterator(const ReadOptions& read_options,
 
   Arena arena;
   Status status;
-  RangeDelAggregatorV2 range_del_agg(&icmp,
+  ReadRangeDelAggregator range_del_agg(&icmp,
                                        kMaxSequenceNumber /* upper_bound */);
 
   *overlap = false;
@@ -2849,6 +2860,7 @@ Status VersionSet::ProcessManifestWrites(
     batch_edits.push_back(first_writer.edit_list.front());
   } else {
     auto it = manifest_writers_.cbegin();
+    size_t group_start = std::numeric_limits<size_t>::max();
     while (it != manifest_writers_.cend()) {
       if ((*it)->edit_list.front()->IsColumnFamilyManipulation()) {
         // no group commits for column family add or drop
@@ -2857,7 +2869,36 @@ Status VersionSet::ProcessManifestWrites(
       last_writer = *(it++);
       assert(last_writer != nullptr);
       assert(last_writer->cfd != nullptr);
-      if (last_writer->cfd != nullptr && last_writer->cfd->IsDropped()) {
+      if (last_writer->cfd->IsDropped()) {
+        // If we detect a dropped CF at this point, and the corresponding
+        // version edits belong to an atomic group, then we need to find out
+        // the preceding version edits in the same atomic group, and update
+        // their `remaining_entries_` member variable because we are NOT going
+        // to write the version edits' of dropped CF to the MANIFEST. If we
+        // don't update, then Recover can report corrupted atomic group because
+        // the `remaining_entries_` do not match.
+        if (!batch_edits.empty()) {
+          if (batch_edits.back()->is_in_atomic_group_ &&
+              batch_edits.back()->remaining_entries_ > 0) {
+            assert(group_start < batch_edits.size());
+            const auto& edit_list = last_writer->edit_list;
+            size_t k = 0;
+            while (k < edit_list.size()) {
+              if (!edit_list[k]->is_in_atomic_group_) {
+                break;
+              } else if (edit_list[k]->remaining_entries_ == 0) {
+                ++k;
+                break;
+              }
+              ++k;
+            }
+            for (auto i = group_start; i < batch_edits.size(); ++i) {
+              assert(static_cast<uint32_t>(k) <=
+                     batch_edits.back()->remaining_entries_);
+              batch_edits[i]->remaining_entries_ -= static_cast<uint32_t>(k);
+            }
+          }
+        }
         continue;
       }
       // We do a linear search on versions because versions is small.
@@ -2888,6 +2929,15 @@ Status VersionSet::ProcessManifestWrites(
       }
       assert(builder != nullptr);  // make checker happy
       for (const auto& e : last_writer->edit_list) {
+        if (e->is_in_atomic_group_) {
+          if (batch_edits.empty() || !batch_edits.back()->is_in_atomic_group_ ||
+              (batch_edits.back()->is_in_atomic_group_ &&
+               batch_edits.back()->remaining_entries_ == 0)) {
+            group_start = batch_edits.size();
+          }
+        } else if (group_start != std::numeric_limits<size_t>::max()) {
+          group_start = std::numeric_limits<size_t>::max();
+        }
         LogAndApplyHelper(last_writer->cfd, builder, version, e, mu);
         batch_edits.push_back(e);
       }
@@ -2900,6 +2950,42 @@ Status VersionSet::ProcessManifestWrites(
     }
   }
 
+#ifndef NDEBUG
+  // Verify that version edits of atomic groups have correct
+  // remaining_entries_.
+  size_t k = 0;
+  while (k < batch_edits.size()) {
+    while (k < batch_edits.size() && !batch_edits[k]->is_in_atomic_group_) {
+      ++k;
+    }
+    if (k == batch_edits.size()) {
+      break;
+    }
+    size_t i = k;
+    while (i < batch_edits.size()) {
+      if (!batch_edits[i]->is_in_atomic_group_) {
+        break;
+      }
+      assert(i - k + batch_edits[i]->remaining_entries_ ==
+             batch_edits[k]->remaining_entries_);
+      if (batch_edits[i]->remaining_entries_ == 0) {
+        ++i;
+        break;
+      }
+      ++i;
+    }
+    assert(batch_edits[i - 1]->is_in_atomic_group_);
+    assert(0 == batch_edits[i - 1]->remaining_entries_);
+    std::vector<VersionEdit*> tmp;
+    for (size_t j = k; j != i; ++j) {
+      tmp.emplace_back(batch_edits[j]);
+    }
+    TEST_SYNC_POINT_CALLBACK(
+        "VersionSet::ProcessManifestWrites:CheckOneAtomicGroup", &tmp);
+    k = i;
+  }
+#endif  // NDEBUG
+
   uint64_t new_manifest_file_size = 0;
   Status s;
 
@@ -3205,7 +3291,7 @@ Status VersionSet::LogAndApply(
     if (!manifest_writers_.empty()) {
       manifest_writers_.front()->cv.Signal();
     }
-    return Status::OK();
+    return Status::ShutdownInProgress();
   }
 
   return ProcessManifestWrites(writers, mu, db_directory, new_descriptor_log,
@@ -4253,7 +4339,7 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
 }
 
 InternalIterator* VersionSet::MakeInputIterator(
-    const Compaction* c, RangeDelAggregatorV2* range_del_agg,
+    const Compaction* c, RangeDelAggregator* range_del_agg,
     const EnvOptions& env_options_compactions) {
   auto cfd = c->column_family_data();
   ReadOptions read_options;

db/version_set.h

@@ -34,7 +34,7 @@
 #include "db/dbformat.h"
 #include "db/file_indexer.h"
 #include "db/log_reader.h"
-#include "db/range_del_aggregator_v2.h"
+#include "db/range_del_aggregator.h"
 #include "db/read_callback.h"
 #include "db/table_cache.h"
 #include "db/version_builder.h"
@@ -538,11 +538,11 @@ class Version {
   // REQUIRES: This version has been saved (see VersionSet::SaveTo)
   void AddIterators(const ReadOptions&, const EnvOptions& soptions,
                     MergeIteratorBuilder* merger_iter_builder,
-                    RangeDelAggregatorV2* range_del_agg);
+                    RangeDelAggregator* range_del_agg);
 
   void AddIteratorsForLevel(const ReadOptions&, const EnvOptions& soptions,
                             MergeIteratorBuilder* merger_iter_builder,
-                            int level, RangeDelAggregatorV2* range_del_agg);
+                            int level, RangeDelAggregator* range_del_agg);
 
   Status OverlapWithLevelIterator(const ReadOptions&, const EnvOptions&,
                                   const Slice& smallest_user_key,
@@ -935,7 +935,7 @@ class VersionSet {
   // Create an iterator that reads over the compaction inputs for "*c".
   // The caller should delete the iterator when no longer needed.
   InternalIterator* MakeInputIterator(
-      const Compaction* c, RangeDelAggregatorV2* range_del_agg,
+      const Compaction* c, RangeDelAggregator* range_del_agg,
       const EnvOptions& env_options_compactions);
 
   // Add all files listed in any live version to *live.

db/version_set_test.cc

@@ -605,9 +605,13 @@ TEST_F(FindLevelFileTest, LevelOverlappingFiles) {
   ASSERT_TRUE(Overlaps("600", "700"));
 }
 
-class VersionSetTest : public testing::Test {
+class VersionSetTestBase {
  public:
-  VersionSetTest()
+  const static std::string kColumnFamilyName1;
+  const static std::string kColumnFamilyName2;
+  const static std::string kColumnFamilyName3;
+
+  VersionSetTestBase()
       : env_(Env::Default()),
         dbname_(test::PerThreadDBPath("version_set_test")),
         db_options_(),
@@ -635,8 +639,9 @@ class VersionSetTest : public testing::Test {
     new_db.SetNextFile(2);
     new_db.SetLastSequence(0);
 
-    const std::vector<std::string> cf_names = {kDefaultColumnFamilyName,
+    const std::vector<std::string> cf_names = {
-                                               "alice", "bob"};
+        kDefaultColumnFamilyName, kColumnFamilyName1, kColumnFamilyName2,
+        kColumnFamilyName3};
     const int kInitialNumOfCfs = static_cast<int>(cf_names.size());
     autovector<VersionEdit> new_cfs;
     uint64_t last_seq = 1;
@@ -711,6 +716,15 @@ class VersionSetTest : public testing::Test {
   std::shared_ptr<mock::MockTableFactory> mock_table_factory_;
 };
 
+const std::string VersionSetTestBase::kColumnFamilyName1 = "alice";
+const std::string VersionSetTestBase::kColumnFamilyName2 = "bob";
+const std::string VersionSetTestBase::kColumnFamilyName3 = "charles";
+
+class VersionSetTest : public VersionSetTestBase, public testing::Test {
+ public:
+  VersionSetTest() : VersionSetTestBase() {}
+};
+
 TEST_F(VersionSetTest, SameColumnFamilyGroupCommit) {
   NewDB();
   const int kGroupSize = 5;
@@ -958,6 +972,126 @@ TEST_F(VersionSetTest, HandleIncorrectAtomicGroupSize) {
             versions_->GetColumnFamilySet()->NumberOfColumnFamilies());
   EXPECT_TRUE(incorrect_group_size);
 }
+
+class VersionSetTestDropOneCF : public VersionSetTestBase,
+                                public testing::TestWithParam<std::string> {
+ public:
+  VersionSetTestDropOneCF() : VersionSetTestBase() {}
+};
+
+// This test simulates the following execution sequence
+// Time  thread1                  bg_flush_thr
+//  |                             Prepare version edits (e1,e2,e3) for atomic
+//  |                             flush cf1, cf2, cf3
+//  |    Enqueue e to drop cfi
+//  |    to manifest_writers_
+//  |                             Enqueue (e1,e2,e3) to manifest_writers_
+//  |
+//  |    Apply e,
+//  |    cfi.IsDropped() is true
+//  |                             Apply (e1,e2,e3),
+//  |                             since cfi.IsDropped() == true, we need to
+//  |                             drop ei and write the rest to MANIFEST.
+//  V
+//
+//  Repeat the test for i = 1, 2, 3 to simulate dropping the first, middle and
+//  last column family in an atomic group.
+TEST_P(VersionSetTestDropOneCF, HandleDroppedColumnFamilyInAtomicGroup) {
+  std::vector<ColumnFamilyDescriptor> column_families;
+  SequenceNumber last_seqno;
+  std::unique_ptr<log::Writer> log_writer;
+  PrepareManifest(&column_families, &last_seqno, &log_writer);
+  Status s = SetCurrentFile(env_, dbname_, 1, nullptr);
+  ASSERT_OK(s);
+
+  EXPECT_OK(versions_->Recover(column_families, false /* read_only */));
+  EXPECT_EQ(column_families.size(),
+            versions_->GetColumnFamilySet()->NumberOfColumnFamilies());
+
+  const int kAtomicGroupSize = 3;
+  const std::vector<std::string> non_default_cf_names = {
+      kColumnFamilyName1, kColumnFamilyName2, kColumnFamilyName3};
+
+  // Drop one column family
+  VersionEdit drop_cf_edit;
+  drop_cf_edit.DropColumnFamily();
+  const std::string cf_to_drop_name(GetParam());
+  auto cfd_to_drop =
+      versions_->GetColumnFamilySet()->GetColumnFamily(cf_to_drop_name);
+  ASSERT_NE(nullptr, cfd_to_drop);
+  // Increase its refcount because cfd_to_drop is used later, and we need to
+  // prevent it from being deleted.
+  cfd_to_drop->Ref();
+  drop_cf_edit.SetColumnFamily(cfd_to_drop->GetID());
+  mutex_.Lock();
+  s = versions_->LogAndApply(cfd_to_drop,
+                             *cfd_to_drop->GetLatestMutableCFOptions(),
+                             &drop_cf_edit, &mutex_);
+  mutex_.Unlock();
+  ASSERT_OK(s);
+
+  std::vector<VersionEdit> edits(kAtomicGroupSize);
+  uint32_t remaining = kAtomicGroupSize;
+  size_t i = 0;
+  autovector<ColumnFamilyData*> cfds;
+  autovector<const MutableCFOptions*> mutable_cf_options_list;
+  autovector<autovector<VersionEdit*>> edit_lists;
+  for (const auto& cf_name : non_default_cf_names) {
+    auto cfd = (cf_name != cf_to_drop_name)
+                   ? versions_->GetColumnFamilySet()->GetColumnFamily(cf_name)
+                   : cfd_to_drop;
+    ASSERT_NE(nullptr, cfd);
+    cfds.push_back(cfd);
+    mutable_cf_options_list.emplace_back(cfd->GetLatestMutableCFOptions());
+    edits[i].SetColumnFamily(cfd->GetID());
+    edits[i].SetLogNumber(0);
+    edits[i].SetNextFile(2);
+    edits[i].MarkAtomicGroup(--remaining);
+    edits[i].SetLastSequence(last_seqno++);
+    autovector<VersionEdit*> tmp_edits;
+    tmp_edits.push_back(&edits[i]);
+    edit_lists.emplace_back(tmp_edits);
+    ++i;
+  }
+  int called = 0;
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionSet::ProcessManifestWrites:CheckOneAtomicGroup", [&](void* arg) {
+        std::vector<VersionEdit*>* tmp_edits =
+            reinterpret_cast<std::vector<VersionEdit*>*>(arg);
+        EXPECT_EQ(kAtomicGroupSize - 1, tmp_edits->size());
+        for (const auto e : *tmp_edits) {
+          bool found = false;
+          for (const auto& e2 : edits) {
+            if (&e2 == e) {
+              found = true;
+              break;
+            }
+          }
+          ASSERT_TRUE(found);
+        }
+        ++called;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  mutex_.Lock();
+  s = versions_->LogAndApply(cfds, mutable_cf_options_list, edit_lists,
+                             &mutex_);
+  mutex_.Unlock();
+  ASSERT_OK(s);
+  ASSERT_EQ(1, called);
+  if (cfd_to_drop->Unref()) {
+    delete cfd_to_drop;
+    cfd_to_drop = nullptr;
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    AtomicGroup, VersionSetTestDropOneCF,
+    testing::Values(VersionSetTestBase::kColumnFamilyName1,
+                    VersionSetTestBase::kColumnFamilyName2,
+                    VersionSetTestBase::kColumnFamilyName3));
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

include/rocksdb/listener.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <chrono>
 #include <memory>
 #include <string>
 #include <unordered_map>
@@ -144,13 +145,18 @@ struct TableFileDeletionInfo {
 };
 
 struct FileOperationInfo {
+  using TimePoint = std::chrono::time_point<std::chrono::system_clock,
+                                            std::chrono::nanoseconds>;
+
   const std::string& path;
   uint64_t offset;
   size_t length;
-  time_t start_timestamp;
+  const TimePoint& start_timestamp;
-  time_t finish_timestamp;
+  const TimePoint& finish_timestamp;
   Status status;
-  FileOperationInfo(const std::string& _path) : path(_path) {}
+  FileOperationInfo(const std::string& _path, const TimePoint& start,
+                    const TimePoint& finish)
+      : path(_path), start_timestamp(start), finish_timestamp(finish) {}
 };
 
 struct FlushJobInfo {

include/rocksdb/perf_context.h

@@ -201,8 +201,8 @@ struct PerfContext {
   uint64_t env_lock_file_nanos;
   uint64_t env_unlock_file_nanos;
   uint64_t env_new_logger_nanos;
-  std::map<uint32_t, PerfContextByLevel>* level_to_perf_context;
+  std::map<uint32_t, PerfContextByLevel>* level_to_perf_context = nullptr;
-  bool per_level_perf_context_enabled;
+  bool per_level_perf_context_enabled = false;
 };
 
 // Get Thread-local PerfContext object pointer

include/rocksdb/version.h

@@ -6,7 +6,7 @@
 
 #define ROCKSDB_MAJOR 5
 #define ROCKSDB_MINOR 18
-#define ROCKSDB_PATCH 0
+#define ROCKSDB_PATCH 4
 
 // Do not use these. We made the mistake of declaring macros starting with
 // double underscore. Now we have to live with our choice. We'll deprecate these

java/CMakeLists.txt

@@ -313,17 +313,19 @@ if(NOT EXISTS ${JAVA_TEST_LIBDIR})
   file(MAKE_DIRECTORY mkdir ${JAVA_TEST_LIBDIR})
 endif()
 
-if (DEFINED CUSTOM_REPO_URL)
+if (DEFINED CUSTOM_DEPS_URL)
-  set(SEARCH_REPO_URL ${CUSTOM_REPO_URL}/)
+  set(DEPS_URL ${CUSTOM_DEPS_URL}/)
-  set(CENTRAL_REPO_URL ${CUSTOM_REPO_URL}/)
 else ()
-  set(SEARCH_REPO_URL "http://search.maven.org/remotecontent?filepath=")
+  # This is a URL for artifacts from a "fake" release on pdillinger's fork,
-  set(CENTRAL_REPO_URL "http://central.maven.org/maven2/")
+  # so as not to put binaries in git (ew). We should move to hosting these
+  # under the facebook account on github, or something else more reliable
+  # than maven.org, which has been failing frequently from Travis.
+  set(DEPS_URL "https://github.com/pdillinger/rocksdb/releases/download/v6.6.x-java-deps")
 endif()
 
 if(NOT EXISTS ${JAVA_JUNIT_JAR})
   message("Downloading ${JAVA_JUNIT_JAR}")
-  file(DOWNLOAD ${SEARCH_REPO_URL}junit/junit/4.12/junit-4.12.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
+  file(DOWNLOAD ${DEPS_URL}/junit-4.12.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
   list(GET downloadStatus 0 error_code)
   if(NOT error_code EQUAL 0)
     message(FATAL_ERROR "Failed downloading ${JAVA_JUNIT_JAR}")
@@ -332,7 +334,7 @@ if(NOT EXISTS ${JAVA_JUNIT_JAR})
 endif()
 if(NOT EXISTS ${JAVA_HAMCR_JAR})
   message("Downloading ${JAVA_HAMCR_JAR}")
-  file(DOWNLOAD ${SEARCH_REPO_URL}org/hamcrest/hamcrest-core/1.3/hamcrest-core-1.3.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
+  file(DOWNLOAD ${DEPS_URL}/hamcrest-core-1.3.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
   list(GET downloadStatus 0 error_code)
   if(NOT error_code EQUAL 0)
     message(FATAL_ERROR "Failed downloading ${JAVA_HAMCR_JAR}")
@@ -341,7 +343,7 @@ if(NOT EXISTS ${JAVA_HAMCR_JAR})
 endif()
 if(NOT EXISTS ${JAVA_MOCKITO_JAR})
   message("Downloading ${JAVA_MOCKITO_JAR}")
-  file(DOWNLOAD ${SEARCH_REPO_URL}org/mockito/mockito-all/1.10.19/mockito-all-1.10.19.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
+  file(DOWNLOAD ${DEPS_URL}/mockito-all-1.10.19.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
   list(GET downloadStatus 0 error_code)
   if(NOT error_code EQUAL 0)
     message(FATAL_ERROR "Failed downloading ${JAVA_MOCKITO_JAR}")
@@ -350,7 +352,7 @@ if(NOT EXISTS ${JAVA_MOCKITO_JAR})
 endif()
 if(NOT EXISTS ${JAVA_CGLIB_JAR})
   message("Downloading ${JAVA_CGLIB_JAR}")
-  file(DOWNLOAD ${SEARCH_REPO_URL}cglib/cglib/2.2.2/cglib-2.2.2.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
+  file(DOWNLOAD ${DEPS_URL}/cglib-2.2.2.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
   list(GET downloadStatus 0 error_code)
   if(NOT error_code EQUAL 0)
     message(FATAL_ERROR "Failed downloading ${JAVA_CGLIB_JAR}")
@@ -359,7 +361,7 @@ if(NOT EXISTS ${JAVA_CGLIB_JAR})
 endif()
 if(NOT EXISTS ${JAVA_ASSERTJ_JAR})
   message("Downloading ${JAVA_ASSERTJ_JAR}")
-  file(DOWNLOAD ${CENTRAL_REPO_URL}org/assertj/assertj-core/1.7.1/assertj-core-1.7.1.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
+  file(DOWNLOAD ${DEPS_URL}/assertj-core-1.7.1.jar ${JAVA_TMP_JAR} STATUS downloadStatus)
   list(GET downloadStatus 0 error_code)
   if(NOT error_code EQUAL 0)
     message(FATAL_ERROR "Failed downloading ${JAVA_ASSERTJ_JAR}")

java/Makefile

@@ -192,8 +192,11 @@ ifneq ($(DEBUG_LEVEL),0)
 	JAVAC_ARGS = -Xlint:deprecation -Xlint:unchecked
 endif
 
-SEARCH_REPO_URL?=http://search.maven.org/remotecontent?filepath=
+# This is a URL for artifacts from a "fake" release on pdillinger's fork,
-CENTRAL_REPO_URL?=http://central.maven.org/maven2/
+# so as not to put binaries in git (ew). We should move to hosting these
+# under the facebook account on github, or something else more reliable
+# than maven.org, which has been failing frequently from Travis.
+DEPS_URL?=https://github.com/pdillinger/rocksdb/releases/download/v6.6.x-java-deps
 
 clean:
 	$(AM_V_at)rm -rf include/*
@@ -250,11 +253,11 @@ optimistic_transaction_sample: java
 
 resolve_test_deps:
 	test -d "$(JAVA_TEST_LIBDIR)" || mkdir -p "$(JAVA_TEST_LIBDIR)"
-	test -s "$(JAVA_JUNIT_JAR)" || cp $(MVN_LOCAL)/junit/junit/4.12/junit-4.12.jar $(JAVA_TEST_LIBDIR) || curl -k -L -o $(JAVA_JUNIT_JAR) $(SEARCH_REPO_URL)junit/junit/4.12/junit-4.12.jar
+	test -s "$(JAVA_JUNIT_JAR)" || cp $(MVN_LOCAL)/junit/junit/4.12/junit-4.12.jar $(JAVA_TEST_LIBDIR) || curl --fail --insecure --output $(JAVA_JUNIT_JAR) --location $(DEPS_URL)/junit-4.12.jar
-	test -s "$(JAVA_HAMCR_JAR)" || cp $(MVN_LOCAL)/org/hamcrest/hamcrest-core/1.3/hamcrest-core-1.3.jar $(JAVA_TEST_LIBDIR) || curl -k -L -o $(JAVA_HAMCR_JAR) $(SEARCH_REPO_URL)org/hamcrest/hamcrest-core/1.3/hamcrest-core-1.3.jar
+	test -s "$(JAVA_HAMCR_JAR)" || cp $(MVN_LOCAL)/org/hamcrest/hamcrest-core/1.3/hamcrest-core-1.3.jar $(JAVA_TEST_LIBDIR) || curl --fail --insecure --output $(JAVA_HAMCR_JAR) --location $(DEPS_URL)/hamcrest-core-1.3.jar
-	test -s "$(JAVA_MOCKITO_JAR)" || cp $(MVN_LOCAL)/org/mockito/mockito-all/1.10.19/mockito-all-1.10.19.jar $(JAVA_TEST_LIBDIR) || curl -k -L -o "$(JAVA_MOCKITO_JAR)" $(SEARCH_REPO_URL)org/mockito/mockito-all/1.10.19/mockito-all-1.10.19.jar
+	test -s "$(JAVA_MOCKITO_JAR)" || cp $(MVN_LOCAL)/org/mockito/mockito-all/1.10.19/mockito-all-1.10.19.jar $(JAVA_TEST_LIBDIR) || curl --fail --insecure --output "$(JAVA_MOCKITO_JAR)" --location $(DEPS_URL)/mockito-all-1.10.19.jar
-	test -s "$(JAVA_CGLIB_JAR)" || cp $(MVN_LOCAL)/cglib/cglib/2.2.2/cglib-2.2.2.jar $(JAVA_TEST_LIBDIR) || curl -k -L -o "$(JAVA_CGLIB_JAR)" $(SEARCH_REPO_URL)cglib/cglib/2.2.2/cglib-2.2.2.jar
+	test -s "$(JAVA_CGLIB_JAR)" || cp $(MVN_LOCAL)/cglib/cglib/2.2.2/cglib-2.2.2.jar $(JAVA_TEST_LIBDIR) || curl --fail --insecure --output "$(JAVA_CGLIB_JAR)" --location $(DEPS_URL)/cglib-2.2.2.jar
-	test -s "$(JAVA_ASSERTJ_JAR)" || cp $(MVN_LOCAL)/org/assertj/assertj-core/1.7.1/assertj-core-1.7.1.jar $(JAVA_TEST_LIBDIR) || curl -k -L -o "$(JAVA_ASSERTJ_JAR)" $(CENTRAL_REPO_URL)org/assertj/assertj-core/1.7.1/assertj-core-1.7.1.jar
+	test -s "$(JAVA_ASSERTJ_JAR)" || cp $(MVN_LOCAL)/org/assertj/assertj-core/1.7.1/assertj-core-1.7.1.jar $(JAVA_TEST_LIBDIR) || curl --fail --insecure --output "$(JAVA_ASSERTJ_JAR)" --location $(DEPS_URL)/assertj-core-1.7.1.jar
 
 java_test: java resolve_test_deps
 	$(AM_V_GEN)mkdir -p $(TEST_CLASSES)

java/crossbuild/docker-build-linux-centos.sh

@@ -8,11 +8,21 @@ cd /rocksdb-local
 
 # Use scl devtoolset if available (i.e. CentOS <7)
 if hash scl 2>/dev/null; then
+	if scl --list | grep -q 'devtoolset-7'; then
+		scl enable devtoolset-7 'make jclean clean'
+		scl enable devtoolset-7 'PORTABLE=1 make -j2 rocksdbjavastatic'
+
+	elif scl --list | grep -q 'devtoolset-2'; then
 		scl enable devtoolset-2 'make jclean clean'
-	scl enable devtoolset-2 'PORTABLE=1 make -j8 rocksdbjavastatic'
+		scl enable devtoolset-2 'PORTABLE=1 make -j2 rocksdbjavastatic'
+
+	else
+		echo "Could not find devtoolset"
+		exit 1;
+	fi
 else
 	make jclean clean
-        PORTABLE=1 make -j8 rocksdbjavastatic
+        PORTABLE=1 make -j2 rocksdbjavastatic
 fi
 
 cp java/target/librocksdbjni-linux*.so java/target/rocksdbjni-*-linux*.jar /rocksdb-host/java/target

java/rocksjni/loggerjnicallback.cc

@@ -131,7 +131,6 @@ void LoggerJniCallback::Logv(const InfoLogLevel log_level, const char* format,
     }
 
     assert(format != nullptr);
-    assert(ap != nullptr);
     const std::unique_ptr<char[]> msg = format_str(format, ap);
 
     // pass msg to java callback handler

java/src/main/java/org/rocksdb/MemoryUsageType.java

@@ -54,9 +54,9 @@ public enum MemoryUsageType {
    *     cannot be found
    */
   public static MemoryUsageType getMemoryUsageType(final byte byteIdentifier) {
-    for (final MemoryUsageType MemoryUsageType : MemoryUsageType.values()) {
+    for (final MemoryUsageType memoryUsageType : MemoryUsageType.values()) {
-      if (MemoryUsageType.getValue() == byteIdentifier) {
+      if (memoryUsageType.getValue() == byteIdentifier) {
-        return MemoryUsageType;
+        return memoryUsageType;
       }
     }
 
@@ -64,7 +64,7 @@ public enum MemoryUsageType {
         "Illegal value provided for MemoryUsageType.");
   }
 
-  private MemoryUsageType(byte value) {
+  MemoryUsageType(byte value) {
     value_ = value;
   }
 

java/src/main/java/org/rocksdb/util/Environment.java

@@ -4,6 +4,10 @@ public class Environment {
   private static String OS = System.getProperty("os.name").toLowerCase();
   private static String ARCH = System.getProperty("os.arch").toLowerCase();
 
+  public static boolean isAarch64() {
+    return ARCH.contains("aarch64");
+  }
+
   public static boolean isPowerPC() {
     return ARCH.contains("ppc");
   }
@@ -59,7 +63,7 @@ public class Environment {
   public static String getJniLibraryName(final String name) {
     if (isUnix()) {
       final String arch = is64Bit() ? "64" : "32";
-      if(isPowerPC()) {
+      if(isPowerPC() || isAarch64()) {
         return String.format("%sjni-linux-%s", name, ARCH);
       } else if(isS390x()) {
         return String.format("%sjni-linux%s", name, ARCH);

java/src/test/java/org/rocksdb/MergeTest.java

@@ -46,13 +46,13 @@ public class MergeTest {
   }
 
   private byte[] longToByteArray(long l) {
-    ByteBuffer buf = ByteBuffer.allocate(Long.BYTES);
+    ByteBuffer buf = ByteBuffer.allocate(Long.SIZE / Byte.SIZE);
     buf.putLong(l);
     return buf.array();
   }
 
   private long longFromByteArray(byte[] a) {
-    ByteBuffer buf = ByteBuffer.allocate(Long.BYTES);
+    ByteBuffer buf = ByteBuffer.allocate(Long.SIZE / Byte.SIZE);
     buf.put(a);
     buf.flip();
     return buf.getLong();

java/src/test/java/org/rocksdb/util/EnvironmentTest.java

@@ -130,6 +130,24 @@ public class EnvironmentTest {
       isEqualTo("librocksdbjni.dll");
   }
 
+  @Test
+  public void aarch64() {
+    setEnvironmentClassFields("Linux", "aarch64");
+    assertThat(Environment.isUnix()).isTrue();
+    assertThat(Environment.isAarch64()).isTrue();
+    assertThat(Environment.is64Bit()).isTrue();
+    assertThat(Environment.getJniLibraryExtension()).
+        isEqualTo(".so");
+    assertThat(Environment.getSharedLibraryName("rocksdb")).
+        isEqualTo("rocksdbjni");
+    assertThat(Environment.getJniLibraryName("rocksdb")).
+        isEqualTo("rocksdbjni-linux-aarch64");
+    assertThat(Environment.getJniLibraryFileName("rocksdb")).
+        isEqualTo("librocksdbjni-linux-aarch64.so");
+    assertThat(Environment.getSharedLibraryFileName("rocksdb")).
+        isEqualTo("librocksdbjni.so");
+  }
+
   private void setEnvironmentClassFields(String osName,
       String osArch) {
     setEnvironmentClassField(OS_FIELD_NAME, osName);

port/jemalloc_helper.h

@@ -0,0 +1,53 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#ifdef ROCKSDB_JEMALLOC
+#ifdef __FreeBSD__
+#include <malloc_np.h>
+#else
+#include <jemalloc/jemalloc.h>
+#endif
+
+#ifndef JEMALLOC_CXX_THROW
+#define JEMALLOC_CXX_THROW
+#endif
+
+// Declare non-standard jemalloc APIs as weak symbols. We can null-check these
+// symbols to detect whether jemalloc is linked with the binary.
+extern "C" void* mallocx(size_t, int) __attribute__((__weak__));
+extern "C" void* rallocx(void*, size_t, int) __attribute__((__weak__));
+extern "C" size_t xallocx(void*, size_t, size_t, int) __attribute__((__weak__));
+extern "C" size_t sallocx(const void*, int) __attribute__((__weak__));
+extern "C" void dallocx(void*, int) __attribute__((__weak__));
+extern "C" void sdallocx(void*, size_t, int) __attribute__((__weak__));
+extern "C" size_t nallocx(size_t, int) __attribute__((__weak__));
+extern "C" int mallctl(const char*, void*, size_t*, void*, size_t)
+    __attribute__((__weak__));
+extern "C" int mallctlnametomib(const char*, size_t*, size_t*)
+    __attribute__((__weak__));
+extern "C" int mallctlbymib(const size_t*, size_t, void*, size_t*, void*,
+                            size_t) __attribute__((__weak__));
+extern "C" void malloc_stats_print(void (*)(void*, const char*), void*,
+                                   const char*) __attribute__((__weak__));
+extern "C" size_t malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void*)
+    JEMALLOC_CXX_THROW __attribute__((__weak__));
+
+// Check if Jemalloc is linked with the binary. Note the main program might be
+// using a different memory allocator even this method return true.
+// It is loosely based on folly::usingJEMalloc(), minus the check that actually
+// allocate memory and see if it is through jemalloc, to handle the dlopen()
+// case:
+// https://github.com/facebook/folly/blob/76cf8b5841fb33137cfbf8b224f0226437c855bc/folly/memory/Malloc.h#L147
+static inline bool HasJemalloc() {
+  return mallocx != nullptr && rallocx != nullptr && xallocx != nullptr &&
+         sallocx != nullptr && dallocx != nullptr && sdallocx != nullptr &&
+         nallocx != nullptr && mallctl != nullptr &&
+         mallctlnametomib != nullptr && mallctlbymib != nullptr &&
+         malloc_stats_print != nullptr && malloc_usable_size != nullptr;
+}
+
+#endif  // ROCKSDB_JEMALLOC

src.mk

@@ -44,7 +44,6 @@ LIB_SOURCES =                                                   \
   db/merge_helper.cc                                            \
   db/merge_operator.cc                                          \
   db/range_del_aggregator.cc                                    \
-  db/range_del_aggregator_v2.cc                                 \
   db/range_tombstone_fragmenter.cc                              \
   db/repair.cc                                                  \
   db/snapshot_impl.cc                                           \
@@ -223,6 +222,11 @@ LIB_SOURCES =                                                   \
   utilities/write_batch_with_index/write_batch_with_index.cc    \
   utilities/write_batch_with_index/write_batch_with_index_internal.cc    \
 
+ifeq ($(ARMCRC_SOURCE),1)
+LIB_SOURCES +=\
+  util/crc32c_arm64.cc
+endif
+
 ifeq (,$(shell $(CXX) -fsyntax-only -maltivec -xc /dev/null 2>&1))
 LIB_SOURCES_ASM =\
   util/crc32c_ppc_asm.S
@@ -334,7 +338,6 @@ MAIN_SOURCES =                                                          \
   db/repair_test.cc                                                     \
   db/range_del_aggregator_test.cc                                       \
   db/range_del_aggregator_bench.cc                                      \
-  db/range_del_aggregator_v2_test.cc                                    \
   db/range_tombstone_fragmenter_test.cc                                 \
   db/table_properties_collector_test.cc                                 \
   db/util_merge_operators_test.cc                                       \

table/block_based_table_reader.cc

@@ -2348,7 +2348,7 @@ FragmentedRangeTombstoneIterator* BlockBasedTable::NewRangeTombstoneIterator(
     snapshot = read_options.snapshot->GetSequenceNumber();
   }
   return new FragmentedRangeTombstoneIterator(
-      rep_->fragmented_range_dels, snapshot, rep_->internal_comparator);
+      rep_->fragmented_range_dels, rep_->internal_comparator, snapshot);
 }
 
 InternalIterator* BlockBasedTable::NewUnfragmentedRangeTombstoneIterator(

util/crc32c.cc

@@ -18,6 +18,8 @@
 #include "util/coding.h"
 #include "util/util.h"
 
+#include "util/crc32c_arm64.h"
+
 #ifdef __powerpc64__
 #include "util/crc32c_ppc.h"
 #include "util/crc32c_ppc_constants.h"
@@ -396,6 +398,8 @@ uint32_t ExtendImpl(uint32_t crc, const char* buf, size_t size) {
   return static_cast<uint32_t>(l ^ 0xffffffffu);
 }
 
+// Detect if ARM64 CRC or not.
+#ifndef HAVE_ARM64_CRC
 // Detect if SS42 or not.
 #ifndef HAVE_POWER8
 
@@ -434,6 +438,7 @@ static bool isPCLMULQDQ() {
 }
 
 #endif  // HAVE_POWER8
+#endif  // HAVE_ARM64_CRC
 
 typedef uint32_t (*Function)(uint32_t, const char*, size_t);
 
@@ -463,6 +468,11 @@ static bool isAltiVec() {
 }
 #endif
 
+#if defined(__linux__) && defined(HAVE_ARM64_CRC)
+uint32_t ExtendARMImpl(uint32_t crc, const char *buf, size_t size) {
+  return crc32c_arm64(crc, (const unsigned char *)buf, size);
+}
+#endif
 
 std::string IsFastCrc32Supported() {
   bool has_fast_crc = false;
@@ -478,6 +488,14 @@ std::string IsFastCrc32Supported() {
   has_fast_crc = false;
   arch = "PPC";
 #endif
+#elif defined(__linux__) && defined(HAVE_ARM64_CRC)
+  if (crc32c_runtime_check()) {
+    has_fast_crc = true;
+    arch = "Arm64";
+  } else {
+    has_fast_crc = false;
+    arch = "Arm64";
+  }
 #else
   has_fast_crc = isSSE42();
   arch = "x86";
@@ -1200,7 +1218,15 @@ uint32_t crc32c_3way(uint32_t crc, const char* buf, size_t len) {
 #endif //HAVE_SSE42 && HAVE_PCLMUL
 
 static inline Function Choose_Extend() {
-#ifndef HAVE_POWER8
+#ifdef HAVE_POWER8
+  return isAltiVec() ? ExtendPPCImpl : ExtendImpl<Slow_CRC32>;
+#elif defined(__linux__) && defined(HAVE_ARM64_CRC)
+  if(crc32c_runtime_check()) {
+    return ExtendARMImpl;
+  } else {
+    return ExtendImpl<Slow_CRC32>;
+  }
+#else
   if (isSSE42()) {
     if (isPCLMULQDQ()) {
 #if defined HAVE_SSE42  && defined HAVE_PCLMUL && !defined NO_THREEWAY_CRC32C
@@ -1216,8 +1242,6 @@ static inline Function Choose_Extend() {
   else {
     return ExtendImpl<Slow_CRC32>;
   }
-#else  //HAVE_POWER8
-  return isAltiVec() ? ExtendPPCImpl : ExtendImpl<Slow_CRC32>;
 #endif
 }
 

util/crc32c_arm64.cc

@@ -0,0 +1,56 @@
+//  Copyright (c) 2018, Arm Limited and affiliates. All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "util/crc32c_arm64.h"
+
+#if defined(__linux__) && defined(HAVE_ARM64_CRC)
+
+#include <asm/hwcap.h>
+#include <sys/auxv.h>
+#ifndef HWCAP_CRC32
+#define HWCAP_CRC32 (1 << 7)
+#endif
+uint32_t crc32c_runtime_check(void) {
+  uint64_t auxv = getauxval(AT_HWCAP);
+  return (auxv & HWCAP_CRC32) != 0;
+}
+
+uint32_t crc32c_arm64(uint32_t crc, unsigned char const *data,
+                             unsigned len) {
+  const uint8_t *buf1;
+  const uint16_t *buf2;
+  const uint32_t *buf4;
+  const uint64_t *buf8;
+
+  int64_t length = (int64_t)len;
+
+  crc ^= 0xffffffff;
+  buf8 = (const uint64_t *)data;
+  while ((length -= sizeof(uint64_t)) >= 0) {
+    crc = __crc32cd(crc, *buf8++);
+  }
+
+  /* The following is more efficient than the straight loop */
+  buf4 = (const uint32_t *)buf8;
+  if (length & sizeof(uint32_t)) {
+    crc = __crc32cw(crc, *buf4++);
+    length -= 4;
+  }
+
+  buf2 = (const uint16_t *)buf4;
+  if (length & sizeof(uint16_t)) {
+    crc = __crc32ch(crc, *buf2++);
+    length -= 2;
+  }
+
+  buf1 = (const uint8_t *)buf2;
+  if (length & sizeof(uint8_t))
+    crc = __crc32cb(crc, *buf1);
+
+  crc ^= 0xffffffff;
+  return crc;
+}
+
+#endif

util/crc32c_arm64.h

@@ -0,0 +1,21 @@
+//  Copyright (c) 2018, Arm Limited and affiliates. All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#ifndef UTIL_CRC32C_ARM64_H
+#define UTIL_CRC32C_ARM64_H
+
+#include <inttypes.h>
+
+#if defined(__aarch64__) || defined(__AARCH64__)
+#ifdef __ARM_FEATURE_CRC32
+#define HAVE_ARM64_CRC
+#include <arm_acle.h>
+extern uint32_t crc32c_arm64(uint32_t crc, unsigned char const *data, unsigned len);
+extern uint32_t crc32c_runtime_check(void);
+#endif
+#endif
+
+
+#endif

util/delete_scheduler.cc

@@ -52,11 +52,12 @@ DeleteScheduler::~DeleteScheduler() {
 }
 
 Status DeleteScheduler::DeleteFile(const std::string& file_path,
-                                   const std::string& dir_to_sync) {
+                                   const std::string& dir_to_sync,
+                                   const bool force_bg) {
   Status s;
-  if (rate_bytes_per_sec_.load() <= 0 ||
+  if (rate_bytes_per_sec_.load() <= 0 || (!force_bg &&
       total_trash_size_.load() >
-          sst_file_manager_->GetTotalSize() * max_trash_db_ratio_.load()) {
+          sst_file_manager_->GetTotalSize() * max_trash_db_ratio_.load())) {
     // Rate limiting is disabled or trash size makes up more than
     // max_trash_db_ratio_ (default 25%) of the total DB size
     TEST_SYNC_POINT("DeleteScheduler::DeleteFile");

util/delete_scheduler.h

@@ -46,8 +46,11 @@ class DeleteScheduler {
     rate_bytes_per_sec_.store(bytes_per_sec);
   }
 
-  // Mark file as trash directory and schedule it's deletion
+  // Mark file as trash directory and schedule it's deletion. If force_bg is
-  Status DeleteFile(const std::string& fname, const std::string& dir_to_sync);
+  // set, it forces the file to always be deleted in the background thread,
+  // except when rate limiting is disabled
+  Status DeleteFile(const std::string& fname, const std::string& dir_to_sync,
+      const bool force_bg = false);
 
   // Wait for all files being deleteing in the background to finish or for
   // destructor to be called.

util/file_reader_writer.cc

@@ -99,17 +99,18 @@ Status RandomAccessFileReader::Read(uint64_t offset, size_t n, Slice* result,
         }
         Slice tmp;
 
-        time_t start_ts = 0;
+        FileOperationInfo::TimePoint start_ts;
         uint64_t orig_offset = 0;
         if (ShouldNotifyListeners()) {
-          start_ts = std::chrono::system_clock::to_time_t(
+          start_ts = std::chrono::system_clock::now();
-              std::chrono::system_clock::now());
           orig_offset = aligned_offset + buf.CurrentSize();
         }
         s = file_->Read(aligned_offset + buf.CurrentSize(), allowed, &tmp,
                         buf.Destination());
         if (ShouldNotifyListeners()) {
-          NotifyOnFileReadFinish(orig_offset, tmp.size(), start_ts, s);
+          auto finish_ts = std::chrono::system_clock::now();
+          NotifyOnFileReadFinish(orig_offset, tmp.size(), start_ts, finish_ts,
+                                 s);
         }
 
         buf.Size(buf.CurrentSize() + tmp.size());
@@ -145,16 +146,17 @@ Status RandomAccessFileReader::Read(uint64_t offset, size_t n, Slice* result,
         Slice tmp_result;
 
 #ifndef ROCKSDB_LITE
-        time_t start_ts = 0;
+        FileOperationInfo::TimePoint start_ts;
         if (ShouldNotifyListeners()) {
-          start_ts = std::chrono::system_clock::to_time_t(
+          start_ts = std::chrono::system_clock::now();
-              std::chrono::system_clock::now());
         }
 #endif
         s = file_->Read(offset + pos, allowed, &tmp_result, scratch + pos);
 #ifndef ROCKSDB_LITE
         if (ShouldNotifyListeners()) {
-          NotifyOnFileReadFinish(offset + pos, tmp_result.size(), start_ts, s);
+          auto finish_ts = std::chrono::system_clock::now();
+          NotifyOnFileReadFinish(offset + pos, tmp_result.size(), start_ts,
+                                 finish_ts, s);
         }
 #endif
 
@@ -442,18 +444,18 @@ Status WritableFileWriter::WriteBuffered(const char* data, size_t size) {
       TEST_SYNC_POINT("WritableFileWriter::Flush:BeforeAppend");
 
 #ifndef ROCKSDB_LITE
-      time_t start_ts = 0;
+      FileOperationInfo::TimePoint start_ts;
       uint64_t old_size = writable_file_->GetFileSize();
       if (ShouldNotifyListeners()) {
-        start_ts = std::chrono::system_clock::to_time_t(
+        start_ts = std::chrono::system_clock::now();
-            std::chrono::system_clock::now());
         old_size = next_write_offset_;
       }
 #endif
       s = writable_file_->Append(Slice(src, allowed));
 #ifndef ROCKSDB_LITE
       if (ShouldNotifyListeners()) {
-        NotifyOnFileWriteFinish(old_size, allowed, start_ts, s);
+        auto finish_ts = std::chrono::system_clock::now();
+        NotifyOnFileWriteFinish(old_size, allowed, start_ts, finish_ts, s);
       }
 #endif
       if (!s.ok()) {
@@ -518,15 +520,15 @@ Status WritableFileWriter::WriteDirect() {
     {
       IOSTATS_TIMER_GUARD(write_nanos);
       TEST_SYNC_POINT("WritableFileWriter::Flush:BeforeAppend");
-      time_t start_ts(0);
+      FileOperationInfo::TimePoint start_ts;
       if (ShouldNotifyListeners()) {
-        start_ts = std::chrono::system_clock::to_time_t(
+        start_ts = std::chrono::system_clock::now();
-            std::chrono::system_clock::now());
       }
       // direct writes must be positional
       s = writable_file_->PositionedAppend(Slice(src, size), write_offset);
       if (ShouldNotifyListeners()) {
-        NotifyOnFileWriteFinish(write_offset, size, start_ts, s);
+        auto finish_ts = std::chrono::system_clock::now();
+        NotifyOnFileWriteFinish(write_offset, size, start_ts, finish_ts, s);
       }
       if (!s.ok()) {
         buf_.Size(file_advance + leftover_tail);

util/file_reader_writer.h

@@ -64,15 +64,13 @@ class SequentialFileReader {
 class RandomAccessFileReader {
  private:
 #ifndef ROCKSDB_LITE
-  void NotifyOnFileReadFinish(uint64_t offset, size_t length, time_t start_ts,
+  void NotifyOnFileReadFinish(uint64_t offset, size_t length,
+                              const FileOperationInfo::TimePoint& start_ts,
+                              const FileOperationInfo::TimePoint& finish_ts,
                               const Status& status) const {
-    FileOperationInfo info(file_name_);
+    FileOperationInfo info(file_name_, start_ts, finish_ts);
     info.offset = offset;
     info.length = length;
-    info.start_timestamp = start_ts;
-    time_t finish_ts =
-        std::chrono::system_clock::to_time_t(std::chrono::system_clock::now());
-    info.finish_timestamp = finish_ts;
     info.status = status;
 
     for (auto& listener : listeners_) {
@@ -157,15 +155,13 @@ class RandomAccessFileReader {
 class WritableFileWriter {
  private:
 #ifndef ROCKSDB_LITE
-  void NotifyOnFileWriteFinish(uint64_t offset, size_t length, time_t start_ts,
+  void NotifyOnFileWriteFinish(uint64_t offset, size_t length,
+                               const FileOperationInfo::TimePoint& start_ts,
+                               const FileOperationInfo::TimePoint& finish_ts,
                                const Status& status) {
-    FileOperationInfo info(file_name_);
+    FileOperationInfo info(file_name_, start_ts, finish_ts);
     info.offset = offset;
     info.length = length;
-    info.start_timestamp = start_ts;
-    time_t finish_ts =
-        std::chrono::system_clock::to_time_t(std::chrono::system_clock::now());
-    info.finish_timestamp = finish_ts;
     info.status = status;
 
     for (auto& listener : listeners_) {

util/file_util.cc

@@ -89,16 +89,23 @@ Status CreateFile(Env* env, const std::string& destination,
 
 Status DeleteSSTFile(const ImmutableDBOptions* db_options,
                      const std::string& fname, const std::string& dir_to_sync) {
+  return DeleteDBFile(db_options, fname, dir_to_sync, false);
+}
+
+Status DeleteDBFile(const ImmutableDBOptions* db_options,
+                     const std::string& fname, const std::string& dir_to_sync,
+                     const bool force_bg) {
 #ifndef ROCKSDB_LITE
   auto sfm =
       static_cast<SstFileManagerImpl*>(db_options->sst_file_manager.get());
   if (sfm) {
-    return sfm->ScheduleFileDeletion(fname, dir_to_sync);
+    return sfm->ScheduleFileDeletion(fname, dir_to_sync, force_bg);
   } else {
     return db_options->env->DeleteFile(fname);
   }
 #else
   (void)dir_to_sync;
+  (void)force_bg;
   // SstFileManager is not supported in ROCKSDB_LITE
   return db_options->env->DeleteFile(fname);
 #endif

util/file_util.h

@@ -25,4 +25,9 @@ extern Status DeleteSSTFile(const ImmutableDBOptions* db_options,
                             const std::string& fname,
                             const std::string& path_to_sync);
 
+extern Status DeleteDBFile(const ImmutableDBOptions* db_options,
+                            const std::string& fname,
+                            const std::string& path_to_sync,
+                            const bool force_bg);
+
 }  // namespace rocksdb

util/heap.h

@@ -92,9 +92,7 @@ class BinaryHeap {
     reset_root_cmp_cache();
   }
 
-  bool empty() const {
+  bool empty() const { return data_.empty(); }
-    return data_.empty();
-  }
 
   size_t size() const { return data_.size(); }
 

util/jemalloc_nodump_allocator.cc

@@ -133,12 +133,16 @@ Status NewJemallocNodumpAllocator(
     JemallocAllocatorOptions& options,
     std::shared_ptr<MemoryAllocator>* memory_allocator) {
   *memory_allocator = nullptr;
-#ifndef ROCKSDB_JEMALLOC_NODUMP_ALLOCATOR
+  Status unsupported = Status::NotSupported(
-  (void) options;
-  return Status::NotSupported(
       "JemallocNodumpAllocator only available with jemalloc version >= 5 "
       "and MADV_DONTDUMP is available.");
+#ifndef ROCKSDB_JEMALLOC_NODUMP_ALLOCATOR
+  (void)options;
+  return unsupported;
 #else
+  if (!HasJemalloc()) {
+    return unsupported;
+  }
   if (memory_allocator == nullptr) {
     return Status::InvalidArgument("memory_allocator must be non-null.");
   }

util/jemalloc_nodump_allocator.h

@@ -8,6 +8,7 @@
 #include <atomic>
 #include <vector>
 
+#include "port/jemalloc_helper.h"
 #include "port/port.h"
 #include "rocksdb/memory_allocator.h"
 #include "util/core_local.h"
@@ -15,7 +16,6 @@
 
 #if defined(ROCKSDB_JEMALLOC) && defined(ROCKSDB_PLATFORM_POSIX)
 
-#include <jemalloc/jemalloc.h>
 #include <sys/mman.h>
 
 #if (JEMALLOC_VERSION_MAJOR >= 5) && defined(MADV_DONTDUMP)

util/sst_file_manager_impl.cc

@@ -402,8 +402,11 @@ bool SstFileManagerImpl::CancelErrorRecovery(ErrorHandler* handler) {
 }
 
 Status SstFileManagerImpl::ScheduleFileDeletion(
-    const std::string& file_path, const std::string& path_to_sync) {
+    const std::string& file_path, const std::string& path_to_sync,
-  return delete_scheduler_.DeleteFile(file_path, path_to_sync);
+    const bool force_bg) {
+  TEST_SYNC_POINT("SstFileManagerImpl::ScheduleFileDeletion");
+  return delete_scheduler_.DeleteFile(file_path, path_to_sync,
+                                      force_bg);
 }
 
 void SstFileManagerImpl::WaitForEmptyTrash() {

util/sst_file_manager_impl.h

@@ -111,9 +111,12 @@ class SstFileManagerImpl : public SstFileManager {
   // not guaranteed
   bool CancelErrorRecovery(ErrorHandler* db);
 
-  // Mark file as trash and schedule it's deletion.
+  // Mark file as trash and schedule it's deletion. If force_bg is set, it
+  // forces the file to be deleting in the background regardless of DB size,
+  // except when rate limited delete is disabled
   virtual Status ScheduleFileDeletion(const std::string& file_path,
-                                      const std::string& dir_to_sync);
+                                      const std::string& dir_to_sync,
+                                      const bool force_bg = false);
 
   // Wait for all files being deleteing in the background to finish or for
   // destructor to be called.

utilities/blob_db/blob_db_impl.cc

@@ -26,6 +26,7 @@
 #include "util/cast_util.h"
 #include "util/crc32c.h"
 #include "util/file_reader_writer.h"
+#include "util/file_util.h"
 #include "util/filename.h"
 #include "util/logging.h"
 #include "util/mutexlock.h"
@@ -1745,7 +1746,8 @@ std::pair<bool, int64_t> BlobDBImpl::DeleteObsoleteFiles(bool aborted) {
                    bfile->PathName().c_str());
 
     blob_files_.erase(bfile->BlobFileNumber());
-    Status s = env_->DeleteFile(bfile->PathName());
+    Status s = DeleteDBFile(&(db_impl_->immutable_db_options()),
+                             bfile->PathName(), blob_dir_, true);
     if (!s.ok()) {
       ROCKS_LOG_ERROR(db_options_.info_log,
                       "File failed to be deleted as obsolete %s",
@@ -1835,7 +1837,7 @@ Status DestroyBlobDB(const std::string& dbname, const Options& options,
     uint64_t number;
     FileType type;
     if (ParseFileName(f, &number, &type) && type == kBlobFile) {
-      Status del = env->DeleteFile(blobdir + "/" + f);
+      Status del = DeleteDBFile(&soptions, blobdir + "/" + f, blobdir, true);
       if (status.ok() && !del.ok()) {
         status = del;
       }

utilities/blob_db/blob_db_test.cc

@@ -18,6 +18,7 @@
 #include "util/cast_util.h"
 #include "util/fault_injection_test_env.h"
 #include "util/random.h"
+#include "util/sst_file_manager_impl.h"
 #include "util/string_util.h"
 #include "util/sync_point.h"
 #include "util/testharness.h"
@@ -762,6 +763,52 @@ TEST_F(BlobDBTest, ReadWhileGC) {
   }
 }
 
+TEST_F(BlobDBTest, SstFileManager) {
+  // run the same test for Get(), MultiGet() and Iterator each.
+  std::shared_ptr<SstFileManager> sst_file_manager(
+      NewSstFileManager(mock_env_.get()));
+  sst_file_manager->SetDeleteRateBytesPerSecond(1);
+  SstFileManagerImpl *sfm =
+      static_cast<SstFileManagerImpl *>(sst_file_manager.get());
+
+  BlobDBOptions bdb_options;
+  bdb_options.min_blob_size = 0;
+  Options db_options;
+
+  int files_deleted_directly = 0;
+  int files_scheduled_to_delete = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "SstFileManagerImpl::ScheduleFileDeletion",
+      [&](void * /*arg*/) { files_scheduled_to_delete++; });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DeleteScheduler::DeleteFile",
+      [&](void * /*arg*/) { files_deleted_directly++; });
+  SyncPoint::GetInstance()->EnableProcessing();
+  db_options.sst_file_manager = sst_file_manager;
+
+  Open(bdb_options, db_options);
+
+  // Create one obselete file and clean it.
+  blob_db_->Put(WriteOptions(), "foo", "bar");
+  auto blob_files = blob_db_impl()->TEST_GetBlobFiles();
+  ASSERT_EQ(1, blob_files.size());
+  std::shared_ptr<BlobFile> bfile = blob_files[0];
+  ASSERT_OK(blob_db_impl()->TEST_CloseBlobFile(bfile));
+  GCStats gc_stats;
+  ASSERT_OK(blob_db_impl()->TEST_GCFileAndUpdateLSM(bfile, &gc_stats));
+  blob_db_impl()->TEST_DeleteObsoleteFiles();
+
+  // Even if SSTFileManager is not set, DB is creating a dummy one.
+  ASSERT_EQ(1, files_scheduled_to_delete);
+  ASSERT_EQ(0, files_deleted_directly);
+  Destroy();
+  // Make sure that DestroyBlobDB() also goes through delete scheduler.
+  ASSERT_GE(2, files_scheduled_to_delete);
+  ASSERT_EQ(0, files_deleted_directly);
+  SyncPoint::GetInstance()->DisableProcessing();
+  sfm->WaitForEmptyTrash();
+}
+
 TEST_F(BlobDBTest, SnapshotAndGarbageCollection) {
   BlobDBOptions bdb_options;
   bdb_options.min_blob_size = 0;

utilities/checkpoint/checkpoint_test.cc

@@ -164,6 +164,16 @@ class CheckpointTest : public testing::Test {
     return DB::OpenForReadOnly(options, dbname_, &db_);
   }
 
+  Status ReadOnlyReopenWithColumnFamilies(const std::vector<std::string>& cfs,
+                                          const Options& options) {
+    std::vector<ColumnFamilyDescriptor> column_families;
+    for (const auto& cf : cfs) {
+      column_families.emplace_back(cf, options);
+    }
+    return DB::OpenForReadOnly(options, dbname_, column_families, &handles_,
+                               &db_);
+  }
+
   Status TryReopen(const Options& options) {
     Close();
     last_options_ = options;
@@ -612,6 +622,69 @@ TEST_F(CheckpointTest, CheckpointWithUnsyncedDataDropped) {
   db_ = nullptr;
 }
 
+TEST_F(CheckpointTest, CheckpointReadOnlyDB) {
+  ASSERT_OK(Put("foo", "foo_value"));
+  ASSERT_OK(Flush());
+  Close();
+  Options options = CurrentOptions();
+  ASSERT_OK(ReadOnlyReopen(options));
+  Checkpoint* checkpoint = nullptr;
+  ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
+  ASSERT_OK(checkpoint->CreateCheckpoint(snapshot_name_));
+  delete checkpoint;
+  checkpoint = nullptr;
+  Close();
+  DB* snapshot_db = nullptr;
+  ASSERT_OK(DB::Open(options, snapshot_name_, &snapshot_db));
+  ReadOptions read_opts;
+  std::string get_result;
+  ASSERT_OK(snapshot_db->Get(read_opts, "foo", &get_result));
+  ASSERT_EQ("foo_value", get_result);
+  delete snapshot_db;
+}
+
+TEST_F(CheckpointTest, CheckpointReadOnlyDBWithMultipleColumnFamilies) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  for (int i = 0; i != 3; ++i) {
+    ASSERT_OK(Put(i, "foo", "foo_value"));
+    ASSERT_OK(Flush(i));
+  }
+  Close();
+  Status s = ReadOnlyReopenWithColumnFamilies(
+      {kDefaultColumnFamilyName, "pikachu", "eevee"}, options);
+  ASSERT_OK(s);
+  Checkpoint* checkpoint = nullptr;
+  ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
+  ASSERT_OK(checkpoint->CreateCheckpoint(snapshot_name_));
+  delete checkpoint;
+  checkpoint = nullptr;
+  Close();
+
+  std::vector<ColumnFamilyDescriptor> column_families{
+      {kDefaultColumnFamilyName, options},
+      {"pikachu", options},
+      {"eevee", options}};
+  DB* snapshot_db = nullptr;
+  std::vector<ColumnFamilyHandle*> snapshot_handles;
+  s = DB::Open(options, snapshot_name_, column_families, &snapshot_handles,
+               &snapshot_db);
+  ASSERT_OK(s);
+  ReadOptions read_opts;
+  for (int i = 0; i != 3; ++i) {
+    std::string get_result;
+    s = snapshot_db->Get(read_opts, snapshot_handles[i], "foo", &get_result);
+    ASSERT_OK(s);
+    ASSERT_EQ("foo_value", get_result);
+  }
+
+  for (auto snapshot_h : snapshot_handles) {
+    delete snapshot_h;
+  }
+  snapshot_handles.clear();
+  delete snapshot_db;
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

utilities/debug.cc

@@ -19,7 +19,7 @@ Status GetAllKeyVersions(DB* db, Slice begin_key, Slice end_key,
 
   DBImpl* idb = static_cast<DBImpl*>(db->GetRootDB());
   auto icmp = InternalKeyComparator(idb->GetOptions().comparator);
-  RangeDelAggregatorV2 range_del_agg(&icmp,
+  ReadRangeDelAggregator range_del_agg(&icmp,
                                        kMaxSequenceNumber /* upper_bound */);
   Arena arena;
   ScopedArenaIterator iter(