remove warning as error.

Add one more #include<functional>
Add some include<functional>
2026-11-06 11:02:36 -05:00 · 2019-11-05 11:06:51 -08:00 · 2019-11-05 11:06:00 -08:00 · 2019-10-31 14:17:26 -07:00 · 2019-10-31 14:12:48 -07:00 · 2019-10-31 13:58:06 -07:00
417 changed files with 21309 additions and 6514 deletions
@@ -1,6 +1,6 @@
 {
  "project_id" : "rocksdb",
-  "conduit_uri" : "https://reviews.facebook.net/",
+  "conduit_uri" : "https://phabricator.fb.com/api/",
  "copyright_holder" : "Facebook",
  "load" : [
    "arcanist_util"
@@ -37,6 +37,7 @@ util/build_version.cc
 build_tools/VALGRIND_LOGS/
 coverage/COVERAGE_REPORT
 .gdbhistory
+.gdb_history
 package/
 .phutil_module_cache
 unity.a
@@ -69,3 +70,4 @@ LOG
 db_logs/
 tp2/
 fbcode/
+fbcode
@@ -5,6 +5,8 @@ os:
  - osx
 compiler:
  - clang
+jdk:
+  - oraclejdk7

 addons:
   apt:
@@ -1,7 +1,7 @@
 # This cmake build is for Windows 64-bit only.
 #
 # Prerequisites:
-#     You must have Visual Studio 2013 Update 4 installed. Start the Developer Command Prompt window that is a part of Visual Studio installation.
+#     You must have at least Visual Studio 2015 Update 3. Start the Developer Command Prompt window that is a part of Visual Studio installation.
 #     Run the build commands from within the Developer Command Prompt window to have paths to the compiler and runtime libraries set.
 #     You must have git.exe in your %PATH% environment variable.
 #
@@ -13,8 +13,7 @@
 #        cd build
 # 3. Run cmake to generate project files for Windows, add more options to enable required third-party libraries.
 #    See thirdparty.inc for more information.
-#        sample command: cmake -G "Visual Studio 12 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
-#        OR for VS Studio 15 cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
+#        sample command: cmake -G "Visual Studio 14 Win64" -DGFLAGS=1 -DSNAPPY=1 -DJEMALLOC=1 -DJNI=1 ..
 # 4. Then build the project in debug mode (you may want to add /m[:<N>] flag to run msbuild in <N> parallel threads
 #                                          or simply /m ot use all avail cores)
 #        msbuild rocksdb.sln
@@ -29,37 +28,117 @@
 cmake_minimum_required(VERSION 2.6)
 project(rocksdb)

-include(${CMAKE_CURRENT_SOURCE_DIR}/thirdparty.inc)
-execute_process(COMMAND powershell -Command "Get-Date -format MM_dd_yyyy" OUTPUT_VARIABLE DATE)
-execute_process(COMMAND powershell -Command "Get-Date -format HH:mm:ss" OUTPUT_VARIABLE TIME)
-string(REGEX REPLACE "(..)_(..)_..(..).*" "\\1/\\2/\\3" DATE "${DATE}")
-string(REGEX REPLACE "(..):(.....).*" " \\1:\\2" TIME "${TIME}")
-string(CONCAT GIT_DATE_TIME ${DATE} ${TIME})
+if(POLICY CMP0042)
+  cmake_policy(SET CMP0042 NEW)
+endif()
+
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/modules/")
+
+if(WIN32)
+  include(${CMAKE_CURRENT_SOURCE_DIR}/thirdparty.inc)
+else()
+  option(WITH_JEMALLOC "build with JeMalloc" OFF)
+  if(WITH_JEMALLOC)
+    find_package(JeMalloc REQUIRED)
+    add_definitions(-DROCKSDB_JEMALLOC)
+    include_directories(${JEMALLOC_INCLUDE_DIR})
+  endif()
+  if(CMAKE_SYSTEM_NAME STREQUAL "FreeBSD")
+    # FreeBSD has jemaloc as default malloc
+    add_definitions(-DROCKSDB_JEMALLOC)
+    set(WITH_JEMALLOC ON)
+  endif()
+  option(WITH_SNAPPY "build with SNAPPY" OFF)
+  if(WITH_SNAPPY)
+    find_package(snappy REQUIRED)
+    add_definitions(-DSNAPPY)
+    include_directories(${SNAPPY_INCLUDE_DIR})
+    list(APPEND THIRDPARTY_LIBS ${SNAPPY_LIBRARIES})
+  endif()
+endif()
+
+if(WIN32)
+  execute_process(COMMAND powershell -Command "Get-Date -format MM_dd_yyyy" OUTPUT_VARIABLE DATE)
+  execute_process(COMMAND powershell -Command "Get-Date -format HH:mm:ss" OUTPUT_VARIABLE TIME)
+  string(REGEX REPLACE "(..)_(..)_..(..).*" "\\1/\\2/\\3" DATE "${DATE}")
+  string(REGEX REPLACE "(..):(.....).*" " \\1:\\2" TIME "${TIME}")
+  set(GIT_DATE_TIME "${DATE} ${TIME}")
+else()
+  execute_process(COMMAND date "+%Y/%m/%d %H:%M:%S" OUTPUT_VARIABLE DATETIME)
+  string(REGEX REPLACE "\n" "" DATETIME ${DATETIME})
+  set(GIT_DATE_TIME "${DATETIME}")
+endif()

 find_package(Git)

 if (GIT_FOUND AND EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/.git")
+  if(WIN32)
    execute_process(COMMAND $ENV{COMSPEC} /C ${GIT_EXECUTABLE} -C ${CMAKE_CURRENT_SOURCE_DIR} rev-parse HEAD OUTPUT_VARIABLE GIT_SHA)
+  else()
+    execute_process(COMMAND ${GIT_EXECUTABLE} -C ${CMAKE_CURRENT_SOURCE_DIR} rev-parse HEAD OUTPUT_VARIABLE GIT_SHA)
+  endif()
 else()
-    set(GIT_SHA 0)
+  set(GIT_SHA 0)
 endif()

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

-set(BUILD_VERSION_CC ${CMAKE_CURRENT_SOURCE_DIR}/util/build_version.cc)
+set(BUILD_VERSION_CC ${CMAKE_BINARY_DIR}/build_version.cc)
+configure_file(util/build_version.cc.in ${BUILD_VERSION_CC} @ONLY)
+add_library(build_version OBJECT ${BUILD_VERSION_CC})
+target_include_directories(build_version PRIVATE
+  ${CMAKE_CURRENT_SOURCE_DIR}/util)
+if(WIN32)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi /nologo  /EHsc /GS /Gd /GR /GF /fp:precise /Zc:wchar_t /Zc:forScope /errorReport:queue")

-add_custom_command(OUTPUT ${BUILD_VERSION_CC}
-    COMMAND echo "#include \"build_version.h\"" > ${BUILD_VERSION_CC}
-    COMMAND echo "const char* rocksdb_build_git_sha = \"rocksdb_build_git_sha:${GIT_SHA}\";" >> ${BUILD_VERSION_CC}
-    COMMAND echo "const char* rocksdb_build_git_datetime = \"rocksdb_build_git_datetime:${GIT_DATE_TIME}\";" >> ${BUILD_VERSION_CC}
-    COMMAND echo const char* rocksdb_build_compile_date = __DATE__\; >> ${BUILD_VERSION_CC}
-)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W3 /wd4127 /wd4800 /wd4996 /wd4351")
+else()
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wextra -Wall")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wsign-compare -Wshadow -Wno-unused-parameter -Wno-unused-variable -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer -momit-leaf-frame-pointer")
+endif()

-add_custom_target(GenerateBuildVersion DEPENDS ${BUILD_VERSION_CC})
+option(FAIL_ON_WARNINGS "Treat compile warnings as errors" ON)
+if(FAIL_ON_WARNINGS)
+  if(MSVC)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /WX")
+  else() # assume GCC
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror")
+  endif()
+endif()

-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi /nologo  /EHsc /GS /Gd /GR /GF /fp:precise /Zc:wchar_t /Zc:forScope /errorReport:queue")
+option(WITH_ASAN "build with ASAN" OFF)
+if(WITH_ASAN)
+  add_definitions(-DROCKSDB_TSAN_RUN)
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=address")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=address")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=address")
+  if(WITH_JEMALLOC)
+    message(FATAL "ASAN does not work well with JeMalloc")
+  endif()
+endif()

-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W3 /WX /wd4127 /wd4800 /wd4996 /wd4351")
+option(WITH_TSAN "build with TSAN" OFF)
+if(WITH_TSAN)
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=thread -pie")
+  add_definitions(-DROCKSDB_TSAN_RUN)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=thread -fPIC")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=thread -fPIC")
+  if(WITH_JEMALLOC)
+    message(FATAL "TSAN does not work well with JeMalloc")
+  endif()
+endif()
+
+option(WITH_UBSAN "build with UBSAN" OFF)
+if(WITH_UBSAN)
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=undefined")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=undefined")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fsanitize=undefined")
+  if(WITH_JEMALLOC)
+    message(FATAL "UBSAN does not work well with JeMalloc")
+  endif()
+endif()

 # Used to run CI build and tests so we can run faster
 set(OPTIMIZE_DEBUG_DEFAULT 0)        # Debug build is unoptimized by default use -DOPTDBG=1 to optimize
@@ -70,33 +149,102 @@ else()
   set(OPTIMIZE_DEBUG ${OPTIMIZE_DEBUG_DEFAULT})
 endif()

-if((${OPTIMIZE_DEBUG} EQUAL 1))
-   message(STATUS "Debug optimization is enabled")
-   set(CMAKE_CXX_FLAGS_DEBUG "/Oxt /MDd")
-else()
-   set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm /MDd")
+if(WIN32)
+  if((${OPTIMIZE_DEBUG} EQUAL 1))
+    message(STATUS "Debug optimization is enabled")
+    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt /MDd")
+  else()
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm /MDd")
+  endif()
+  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /MD")
+
+  set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /DEBUG")
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /DEBUG")
 endif()

-set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /MD")
+if(CMAKE_COMPILER_IS_GNUCXX)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-builtin-memcmp")
+endif()

-set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /DEBUG")
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /DEBUG")
+option(ROCKSDB_LITE "Build RocksDBLite version" OFF)
+if(ROCKSDB_LITE)
+  add_definitions(-DROCKSDB_LITE)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-exceptions")
+endif()

-add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64 -DNOMINMAX)
+if(CMAKE_SYSTEM_NAME MATCHES "Cygwin")
+  add_definitions(-fno-builtin-memcmp -DCYGWIN)
+elseif(CMAKE_SYSTEM_NAME MATCHES "Darwin")
+  add_definitions(-DOS_MACOSX)
+  if(CMAKE_SYSTEM_PROCESSOR MATCHES arm)
+    add_definitions(-DIOS_CROSS_COMPILE -DROCKSDB_LITE)
+    # no debug info for IOS, that will make our library big
+    add_definitions(-DNDEBUG)
+  endif()
+elseif(CMAKE_SYSTEM_NAME MATCHES "Linux")
+  add_definitions(-DOS_LINUX)
+elseif(CMAKE_SYSTEM_NAME MATCHES "SunOS")
+  add_definitions(-DOS_SOLARIS)
+elseif(CMAKE_SYSTEM_NAME MATCHES "FreeBSD")
+  add_definitions(-DOS_FREEBSD)
+elseif(CMAKE_SYSTEM_NAME MATCHES "NetBSD")
+  add_definitions(-DOS_NETBSD)
+elseif(CMAKE_SYSTEM_NAME MATCHES "OpenBSD")
+  add_definitions(-DOS_OPENBSD)
+elseif(CMAKE_SYSTEM_NAME MATCHES "DragonFly")
+  add_definitions(-DOS_DRAGONFLYBSD)
+elseif(CMAKE_SYSTEM_NAME MATCHES "Android")
+  add_definitions(-DOS_ANDROID)
+elseif(CMAKE_SYSTEM_NAME MATCHES "Windows")
+  add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64 -DNOMINMAX)
+endif()
+
+if(NOT WIN32)
+  add_definitions(-DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX)
+endif()
+
+option(WITH_FALLOCATE "build with fallocate" ON)
+
+if(WITH_FALLOCATE)
+  include(CheckCSourceCompiles)
+  CHECK_C_SOURCE_COMPILES("
+#include <fcntl.h>
+#include <linux/falloc.h>
+int main() {
+ int fd = open(\"/dev/null\", 0);
+ fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, 0, 1024);
+}
+" HAVE_FALLOCATE)
+  if(HAVE_FALLOCATE)
+    add_definitions(-DROCKSDB_FALLOCATE_PRESENT)
+  endif()
+endif()
+
+include(CheckFunctionExists)
+CHECK_FUNCTION_EXISTS(malloc_usable_size HAVE_MALLOC_USABLE_SIZE)
+if(HAVE_MALLOC_USABLE_SIZE)
+  add_definitions(-DROCKSDB_MALLOC_USABLE_SIZE)
+endif()

 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
-include_directories(${PROJECT_SOURCE_DIR}/third-party/gtest-1.7.0/fused-src)
+include_directories(SYSTEM ${PROJECT_SOURCE_DIR}/third-party/gtest-1.7.0/fused-src)
+
+find_package(Threads REQUIRED)
+if(WIN32)
+  set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)
+else()
+  set(SYSTEM_LIBS ${CMAKE_THREAD_LIBS_INIT})
+endif()

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

 add_subdirectory(third-party/gtest-1.7.0/fused-src/gtest)
+add_subdirectory(tools)

 # Main library source code
+
 set(SOURCES
        db/auto_roll_logger.cc
        db/builder.cc
@@ -113,11 +261,11 @@ set(SOURCES
        db/db_impl.cc
        db/db_impl_debug.cc
        db/db_impl_experimental.cc
-        db/db_impl_add_file.cc
        db/db_impl_readonly.cc
        db/db_info_dumper.cc
        db/db_iter.cc
        db/event_helpers.cc
+        db/external_sst_file_ingestion_job.cc
        db/experimental.cc
        db/filename.cc
        db/file_indexer.cc
@@ -133,6 +281,7 @@ set(SOURCES
        db/memtable_list.cc
        db/merge_helper.cc
        db/merge_operator.cc
+        db/range_del_aggregator.cc
        db/repair.cc
        db/snapshot_impl.cc
        db/table_cache.cc
@@ -153,12 +302,6 @@ set(SOURCES
        memtable/skiplistrep.cc
        memtable/vectorrep.cc
        port/stack_trace.cc
-        port/win/io_win.cc
-        port/win/env_win.cc
-        port/win/env_default.cc
-        port/win/port_win.cc
-        port/win/win_logger.cc
-        port/win/xpress_win.cc
        table/adaptive_table_factory.cc
        table/block.cc
        table/block_based_filter_block.cc
@@ -192,7 +335,6 @@ set(SOURCES
        tools/dump/db_dump_tool.cc
        util/arena.cc
        util/bloom.cc
-        util/build_version.cc
        util/cf_options.cc
        util/clock_cache.cc
        util/coding.cc
@@ -200,6 +342,7 @@ set(SOURCES
        util/comparator.cc
        util/concurrent_arena.cc
        util/crc32c.cc
+        util/db_options.cc
        util/delete_scheduler.cc
        util/dynamic_bloom.cc
        util/env.cc
@@ -237,7 +380,6 @@ set(SOURCES
        util/status_message.cc
        util/string_util.cc
        util/sync_point.cc
-        util/testharness.cc
        util/testutil.cc
        util/thread_local.cc
        util/threadpool_imp.cc
@@ -258,9 +400,9 @@ set(SOURCES
        utilities/document/json_document_builder.cc
        utilities/env_mirror.cc
        utilities/env_registry.cc
-        utilities/flashcache/flashcache.cc
        utilities/geodb/geodb_impl.cc
        utilities/leveldb_options/leveldb_options.cc
+        utilities/lua/rocks_lua_compaction_filter.cc
        utilities/memory/memory_util.cc
        utilities/merge_operators/string_append/stringappend.cc
        utilities/merge_operators/string_append/stringappend2.cc
@@ -292,56 +434,56 @@ set(SOURCES
        utilities/col_buf_encoder.cc
        utilities/col_buf_decoder.cc
        utilities/column_aware_encoding_util.cc
-)
+        $<TARGET_OBJECTS:build_version>)

-# For test util library that is build only in DEBUG mode
-# and linked to tests. Add test only code that is not #ifdefed for Release here.
-set(TESTUTIL_SOURCE
-    db/db_test_util.cc
-    table/mock_table.cc
-    util/mock_env.cc
-    util/fault_injection_test_env.cc
-    util/thread_status_updater_debug.cc
-)
+if(WIN32)
+  list(APPEND SOURCES
+    port/win/io_win.cc
+    port/win/env_win.cc
+    port/win/env_default.cc
+    port/win/port_win.cc
+    port/win/win_logger.cc
+    port/win/xpress_win.cc)
+else()
+  list(APPEND SOURCES
+    port/port_posix.cc
+    util/env_posix.cc
+    util/io_posix.cc)
+endif()

-add_library(rocksdblib${ARTIFACT_SUFFIX} ${SOURCES})
-set_target_properties(rocksdblib${ARTIFACT_SUFFIX} PROPERTIES COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/rocksdblib${ARTIFACT_SUFFIX}.pdb")
-add_dependencies(rocksdblib${ARTIFACT_SUFFIX} GenerateBuildVersion)
+option(WITH_LIBRADOS "Build with librados" OFF)
+if(WITH_LIBRADOS)
+  list(APPEND SOURCES
+    utilities/env_librados.cc)
+  list(APPEND THIRDPARTY_LIBS rados)
+endif()
+
+add_library(rocksdblib${ARTIFACT_SUFFIX} STATIC ${SOURCES})
+target_link_libraries(rocksdblib${ARTIFACT_SUFFIX}
+  ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
+
+if(WIN32)
+  set_target_properties(rocksdblib${ARTIFACT_SUFFIX} PROPERTIES
+    COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/rocksdblib${ARTIFACT_SUFFIX}.pdb")
+endif()

 add_library(rocksdb${ARTIFACT_SUFFIX} SHARED ${SOURCES})
-set_target_properties(rocksdb${ARTIFACT_SUFFIX} PROPERTIES COMPILE_FLAGS "-DROCKSDB_DLL -DROCKSDB_LIBRARY_EXPORTS /Fd${CMAKE_CFG_INTDIR}/rocksdb${ARTIFACT_SUFFIX}.pdb")
-add_dependencies(rocksdb${ARTIFACT_SUFFIX} GenerateBuildVersion)
-target_link_libraries(rocksdb${ARTIFACT_SUFFIX} ${LIBS})
+target_link_libraries(rocksdb${ARTIFACT_SUFFIX}
+  ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})

-if (DEFINED JNI)
-  if (${JNI} EQUAL 1)
-    message(STATUS "JNI library is enabled")
-    add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/java)
-  else()
-    message(STATUS "JNI library is disabled")
-  endif()
+if(WIN32)
+  set_target_properties(rocksdb${ARTIFACT_SUFFIX} PROPERTIES
+    COMPILE_FLAGS "-DROCKSDB_DLL -DROCKSDB_LIBRARY_EXPORTS /Fd${CMAKE_CFG_INTDIR}/rocksdb${ARTIFACT_SUFFIX}.pdb")
+endif()
+
+option(WITH_JNI "build with JNI" OFF)
+if(WITH_JNI OR JNI)
+  message(STATUS "JNI library is enabled")
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/java)
 else()
  message(STATUS "JNI library is disabled")
 endif()

-set(APPS
-        tools/db_bench.cc
-        db/memtablerep_bench.cc
-        table/table_reader_bench.cc
-        tools/db_stress.cc
-        tools/write_stress.cc
-        tools/db_repl_stress.cc
-        tools/ldb.cc
-        tools/sst_dump.cc
-        tools/dump/rocksdb_dump.cc
-        tools/dump/rocksdb_undump.cc
-        util/cache_bench.cc
-        utilities/persistent_cache/hash_table_bench.cc
-        utilities/column_aware_encoding_exp.cc
-)
-
-set(C_TESTS db/c_test.c)
-
 set(TESTS
        db/auto_roll_logger_test.cc
        db/column_family_test.cc
@@ -359,6 +501,8 @@ set(TESTS
        db/db_inplace_update_test.cc
        db/db_iter_test.cc
        db/db_log_iter_test.cc
+        db/db_memtable_test.cc
+        db/db_merge_operator_test.cc
        db/db_options_test.cc
        db/db_properties_test.cc
        db/db_table_properties_test.cc
@@ -369,6 +513,7 @@ set(TESTS
        db/db_bloom_filter_test.cc
        db/db_flush_test.cc
        db/db_iterator_test.cc
+        db/db_range_del_test.cc
        db/db_sst_test.cc
        db/external_sst_file_test.cc
        db/db_universal_compaction_test.cc
@@ -408,7 +553,6 @@ set(TESTS
        table/full_filter_block_test.cc
        table/merger_test.cc
        table/table_test.cc
-        tools/db_sanity_test.cc
        tools/ldb_cmd_test.cc
        tools/reduce_levels_test.cc
        tools/sst_dump_test.cc
@@ -459,21 +603,44 @@ set(TESTS
        utilities/ttl/ttl_test.cc
        utilities/write_batch_with_index/write_batch_with_index_test.cc
        utilities/column_aware_encoding_test.cc
+        utilities/lua/rocks_lua_test.cc
 )
+if(WITH_LIBRADOS)
+  list(APPEND TESTS utilities/env_librados_test.cc)
+endif()

-set(EXES ${APPS})
-
-foreach(sourcefile ${EXES})
-    string(REPLACE ".cc" "" exename ${sourcefile})
-    string(REGEX REPLACE "^((.+)/)+" "" exename ${exename})
-    add_executable(${exename}${ARTIFACT_SUFFIX} ${sourcefile})
-    target_link_libraries(${exename}${ARTIFACT_SUFFIX} ${LIBS})
-endforeach(sourcefile ${EXES})
+set(BENCHMARKS
+  tools/db_bench.cc
+  table/table_reader_bench.cc
+  util/cache_bench.cc
+  db/memtablerep_bench.cc
+  utilities/column_aware_encoding_exp.cc
+  utilities/persistent_cache/hash_table_bench.cc)
+add_library(testharness OBJECT util/testharness.cc)
+foreach(sourcefile ${BENCHMARKS})
+  get_filename_component(exename ${sourcefile} NAME_WE)
+  add_executable(${exename}${ARTIFACT_SUFFIX} ${sourcefile}
+    $<TARGET_OBJECTS:testharness>)
+  target_link_libraries(${exename}${ARTIFACT_SUFFIX} ${LIBS} gtest)
+endforeach(sourcefile ${BENCHMARKS})

+# For test util library that is build only in DEBUG mode
+# and linked to tests. Add test only code that is not #ifdefed for Release here.
+set(TESTUTIL_SOURCE
+    db/db_test_util.cc
+    table/mock_table.cc
+    util/mock_env.cc
+    util/fault_injection_test_env.cc
+    util/thread_status_updater_debug.cc
+)
 # test utilities are only build in debug
+enable_testing()
+add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND})
 set(TESTUTILLIB testutillib${ARTIFACT_SUFFIX})
 add_library(${TESTUTILLIB} STATIC ${TESTUTIL_SOURCE})
-set_target_properties(${TESTUTILLIB} PROPERTIES COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/testutillib${ARTIFACT_SUFFIX}.pdb")
+if(WIN32)
+  set_target_properties(${TESTUTILLIB} PROPERTIES COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/testutillib${ARTIFACT_SUFFIX}.pdb")
+endif()
 set_target_properties(${TESTUTILLIB}
      PROPERTIES EXCLUDE_FROM_DEFAULT_BUILD_RELEASE 1
      EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
@@ -484,18 +651,23 @@ set_target_properties(${TESTUTILLIB}
 set(TEST_EXES ${TESTS})

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

 # C executables must link to a shared object
+set(C_TESTS db/c_test.c)
 set(C_TEST_EXES ${C_TESTS})

 foreach(sourcefile ${C_TEST_EXES})
@@ -508,4 +680,6 @@ foreach(sourcefile ${C_TEST_EXES})
      EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
      )
    target_link_libraries(${exename}${ARTIFACT_SUFFIX} rocksdb${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX})
+    add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
+    add_dependencies(check ${exename}${ARTIFACT_SUFFIX})
 endforeach(sourcefile ${C_TEST_EXES})
@@ -1,4 +1,8 @@
 # RocksDB default options change log
+## 5.0 (11/17/2016)
+* Options::allow_concurrent_memtable_write and Options::enable_write_thread_adaptive_yield are now true by default
+* Options.level0_stop_writes_trigger default value changes from 24 to 32.
+
 ## 4.8.0 (5/2/2016)
 * options.max_open_files changes from 5000 to -1. It improves performance, but users need to set file descriptor limit to be large enough and watch memory usage for index and bloom filters.
 * options.base_background_compactions changes from max_background_compactions to 1. When users set higher max_background_compactions but the write throughput is not high, the writes are less spiky to disks.
@@ -1,5 +1,41 @@
 # Rocksdb Change Log
 ## Unreleased
+### Bug fixes
+* Fix the data corruption bug in the case that concurrent memtable write is enabled and 2PC is used.
+* OptimizeForPointLookup() doesn't work with the default DB setting of allow_concurrent_memtable_write=true. Fix it.
+* Fix a 2PC related bug where WAL files size grow too large.
+* Fix the bug that if 2PC is enabled, checkpoints may loss some recent transactions.
+* When file copying is needed when creating checkpoints or bulk loading files, fsync the file after the file copying.
+
+## 5.0.0 (11/17/2016)
+### Public API Change
+* Options::max_bytes_for_level_multiplier is now a double along with all getters and setters.
+* Support dynamically change `delayed_write_rate` and `max_total_wal_size` options via SetDBOptions().
+* Introduce DB::DeleteRange for optimized deletion of large ranges of contiguous keys.
+* Support dynamically change `delayed_write_rate` option via SetDBOptions().
+* Options::allow_concurrent_memtable_write and Options::enable_write_thread_adaptive_yield are now true by default.
+* Remove Tickers::SEQUENCE_NUMBER to avoid confusion if statistics object is shared among RocksDB instance. Alternatively DB::GetLatestSequenceNumber() can be used to get the same value.
+* Options.level0_stop_writes_trigger default value changes from 24 to 32.
+* New compaction filter API: CompactionFilter::FilterV2(). Allows to drop ranges of keys.
+* Removed flashcache support.
+* DB::AddFile() is deprecated and is replaced with DB::IngestExternalFile(). DB::IngestExternalFile() remove all the restrictions that existed for DB::AddFile.
+
+### New Features
+* Add avoid_flush_during_shutdown option, which speeds up DB shutdown by not flushing unpersisted data (i.e. with disableWAL = true). Unpersisted data will be lost. The options is dynamically changeable via SetDBOptions().
+* Add memtable_insert_with_hint_prefix_extractor option. The option is mean to reduce CPU usage for inserting keys into memtable, if keys can be group by prefix and insert for each prefix are sequential or almost sequential. See include/rocksdb/options.h for more details.
+* Add LuaCompactionFilter in utilities.  This allows developers to write compaction filters in Lua.  To use this feature, LUA_PATH needs to be set to the root directory of Lua.
+* No longer populate "LATEST_BACKUP" file in backup directory, which formerly contained the number of the latest backup. The latest backup can be determined by finding the highest numbered file in the "meta/" subdirectory.
+
+## 4.13.0 (10/18/2016)
+### Public API Change
+* DB::GetOptions() reflect dynamic changed options (i.e. through DB::SetOptions()) and return copy of options instead of reference.
+* Added Statistics::getAndResetTickerCount().
+
+### New Features
+* Add DB::SetDBOptions() to dynamic change base_background_compactions and max_background_compactions.
+* Added Iterator::SeekForPrev(). This new API will seek to the last key that less than or equal to the target key.
+
+## 4.12.0 (9/12/2016)
 ### Public API Change
 * CancelAllBackgroundWork() flushes all memtables for databases containing writes that have bypassed the WAL (writes issued with WriteOptions::disableWAL=true) before shutting down background threads.
 * Merge options source_compaction_factor, max_grandparent_overlap_bytes and expanded_compaction_factor into max_compaction_bytes.
@@ -9,6 +45,7 @@
 ### New Features
 * Introduce NewClockCache, which is based on CLOCK algorithm with better concurrent performance in some cases. It can be used to replace the default LRU-based block cache and table cache. To use it, RocksDB need to be linked with TBB lib.
 * Change ticker/histogram statistics implementations to accumulate data in thread-local storage, which improves CPU performance by reducing cache coherency costs. Callers of CreateDBStatistics do not need to change anything to use this feature.
+* Block cache mid-point insertion, where index and filter block are inserted into LRU block cache with higher priority. The feature can be enabled by setting BlockBasedTableOptions::cache_index_and_filter_blocks_with_high_priority to true and high_pri_pool_ratio > 0 when creating NewLRUCache.

 ## 4.11.0 (8/1/2016)
 ### Public API Change
@@ -40,7 +40,7 @@ your make commands, like this: `PORTABLE=1 make static_lib`
 ## Supported platforms

 * **Linux - Ubuntu**
-    * Upgrade your gcc to version at least 4.7 to get C++11 support.
+    * Upgrade your gcc to version at least 4.8 to get C++11 support.
    * Install gflags. First, try: `sudo apt-get install libgflags-dev`
      If this doesn't work and you're using Ubuntu, here's a nice tutorial:
      (http://askubuntu.com/questions/312173/installing-gflags-12-04)
@@ -49,8 +49,8 @@ your make commands, like this: `PORTABLE=1 make static_lib`
    * Install zlib. Try: `sudo apt-get install zlib1g-dev`.
    * Install bzip2: `sudo apt-get install libbz2-dev`.
 * **Linux - CentOS**
-    * Upgrade your gcc to version at least 4.7 to get C++11 support:
-      `yum install gcc47-c++`
+    * Upgrade your gcc to version at least 4.8 to get C++11 support:
+      `yum install gcc48-c++`
    * Install gflags:

              wget https://gflags.googlecode.com/files/gflags-2.0-no-svn-files.tar.gz
@@ -80,7 +80,7 @@ your make commands, like this: `PORTABLE=1 make static_lib`
        * Update XCode:  run `xcode-select --install` (or install it from XCode App's settting).
        * Install via [homebrew](http://brew.sh/).
            * If you're first time developer in MacOS, you still need to run: `xcode-select --install` in your command line.
-            * run `brew tap homebrew/versions; brew install gcc47 --use-llvm` to install gcc 4.7 (or higher).
+            * run `brew tap homebrew/versions; brew install gcc48 --use-llvm` to install gcc 4.8 (or higher).
    * run `brew install rocksdb`

 * **iOS**:
@@ -166,6 +166,8 @@ ifneq ($(filter -DROCKSDB_LITE,$(OPT)),)
 	# found
 	CFLAGS += -fno-exceptions
 	CXXFLAGS += -fno-exceptions
+	# LUA is not supported under ROCKSDB_LITE
+	LUA_PATH =
 endif

 # ASAN doesn't work well with jemalloc. If we're compiling with ASAN, we should use regular malloc.
@@ -185,6 +187,8 @@ ifdef COMPILE_WITH_TSAN
        # Turn off -pg when enabling TSAN testing, because that induces
        # a link failure.  TODO: find the root cause
 	PROFILING_FLAGS =
+	# LUA is not supported under TSAN
+	LUA_PATH =
 endif

 # USAN doesn't work well with jemalloc. If we're compiling with USAN, we should use regular malloc.
@@ -216,10 +220,34 @@ default: all
 WARNING_FLAGS = -W -Wextra -Wall -Wsign-compare -Wshadow \
  -Wno-unused-parameter

-ifndef DISABLE_WARNING_AS_ERROR
-	WARNING_FLAGS += -Werror
+
+ifdef LUA_PATH
+
+ifndef LUA_INCLUDE
+LUA_INCLUDE=$(LUA_PATH)/include
 endif

+LUA_INCLUDE_FILE=$(LUA_INCLUDE)/lualib.h
+
+ifeq ("$(wildcard $(LUA_INCLUDE_FILE))", "")
+# LUA_INCLUDE_FILE does not exist
+$(error Cannot find lualib.h under $(LUA_INCLUDE).  Try to specify both LUA_PATH and LUA_INCLUDE manually)
+endif
+LUA_FLAGS = -I$(LUA_INCLUDE) -DLUA -DLUA_COMPAT_ALL
+CFLAGS += $(LUA_FLAGS)
+CXXFLAGS += $(LUA_FLAGS)
+
+ifndef LUA_LIB
+LUA_LIB = $(LUA_PATH)/lib/liblua.a
+endif
+ifeq ("$(wildcard $(LUA_LIB))", "") # LUA_LIB does not exist
+$(error $(LUA_LIB) does not exist.  Try to specify both LUA_PATH and LUA_LIB manually)
+endif
+LDFLAGS += $(LUA_LIB)
+
+endif
+
+
 CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers

@@ -231,14 +259,7 @@ ifdef FORCE_GIT_SHA
 else
 	git_sha := $(shell git rev-parse HEAD 2>/dev/null)
 endif
-gen_build_version =							\
-  printf '%s\n'								\
-    '\#include "build_version.h"'					\
-    'const char* rocksdb_build_git_sha =				\
-      "rocksdb_build_git_sha:$(git_sha)";'			\
-    'const char* rocksdb_build_git_date =				\
-      "rocksdb_build_git_date:$(date)";'				\
-    'const char* rocksdb_build_compile_date = __DATE__;'
+gen_build_version = sed -e s/@@GIT_SHA@@/$(git_sha)/ -e s/@@GIT_DATE_TIME@@/$(date)/ util/build_version.cc.in

 # Record the version of the source that we are compiling.
 # We keep a record of the git revision in this file.  It is then built
@@ -283,7 +304,10 @@ TESTS = \
 	db_flush_test \
 	db_inplace_update_test \
 	db_iterator_test \
+	db_memtable_test \
+	db_merge_operator_test \
 	db_options_test \
+	db_range_del_test \
 	db_sst_test \
 	external_sst_file_test \
 	db_tailing_iter_test \
@@ -386,6 +410,7 @@ TESTS = \
 	iostats_context_test \
 	persistent_cache_test \
 	statistics_test \
+	lua_test \
 	lru_cache_test \

 PARALLEL_TEST = \
@@ -393,13 +418,16 @@ PARALLEL_TEST = \
 	db_compaction_filter_test \
 	db_compaction_test \
 	db_sst_test \
-	external_sst_file_test \
 	db_test \
 	db_universal_compaction_test \
+	db_wal_test \
+	external_sst_file_test \
 	fault_injection_test \
 	inlineskiplist_test \
 	manual_compaction_test \
-	table_test
+	persistent_cache_test \
+	table_test \
+	transaction_test

 SUBSET := $(TESTS)
 ifdef ROCKSDBTESTS_START
@@ -599,7 +627,7 @@ gen_parallel_tests:
 # 107.816 PASS t/DBTest.EncodeDecompressedBlockSizeTest
 #
 slow_test_regexp = \
-	^t/run-table_test-HarnessTest.Randomized$$|^t/run-db_test-.*(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$
+	^t/run-table_test-HarnessTest.Randomized$$|^t/run-db_test-.*(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$|^.*RecoverFromCorruptedWALWithoutFlush$$
 prioritize_long_running_tests =						\
  perl -pe 's,($(slow_test_regexp)),100 $$1,'				\
    | sort -k1,1gr							\
@@ -679,7 +707,8 @@ ifeq ($(filter -DROCKSDB_LITE,$(OPT)),)
 	sh tools/rocksdb_dump_test.sh
 endif

-check_some: $(SUBSET) ldb_tests
+# TODO add ldb_tests
+check_some: $(SUBSET)
 	for t in $(SUBSET); do echo "===== Running $$t"; ./$$t || exit 1; done

 .PHONY: ldb_tests
@@ -722,6 +751,9 @@ ubsan_crash_test:
 	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test
 	$(MAKE) clean

+valgrind_test:
+	DISABLE_JEMALLOC=1 $(MAKE) valgrind_check
+
 valgrind_check: $(TESTS)
 	$(MAKE) DRIVER="$(VALGRIND_VER) $(VALGRIND_OPTS)" gen_parallel_tests
 	$(AM_V_GEN)if test "$(J)" != 1                                  \
@@ -820,6 +852,7 @@ clean:
 	find . -name "*.[oda]" -exec rm -f {} \;
 	find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
 	rm -rf bzip2* snappy* zlib* lz4*
+	cd java; $(MAKE) clean

 tags:
 	ctags * -R
@@ -957,9 +990,18 @@ db_inplace_update_test: db/db_inplace_update_test.o db/db_test_util.o $(LIBOBJEC
 db_iterator_test: db/db_iterator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_memtable_test: db/db_memtable_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_merge_operator_test: db/db_merge_operator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_options_test: db/db_options_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

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

@@ -1256,6 +1298,8 @@ statistics_test: util/statistics_test.o $(LIBOBJECTS) $(TESTHARNESS)
 lru_cache_test: util/lru_cache_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+lua_test: utilities/lua/rocks_lua_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)

 #-------------------------------------------------
 # make install related stuff
@@ -1319,6 +1363,11 @@ else
 	JAVA_INCLUDE = -I/System/Library/Frameworks/JavaVM.framework/Headers/
 endif
 endif
+ifeq ($(PLATFORM), OS_FREEBSD)
+	JAVA_INCLUDE += -I$(JAVA_HOME)/include/freebsd
+	ROCKSDBJNILIB = librocksdbjni-freebsd$(ARCH).so
+	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-freebsd$(ARCH).jar
+endif
 ifeq ($(PLATFORM), OS_SOLARIS)
 	ROCKSDBJNILIB = librocksdbjni-solaris$(ARCH).so
 	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-solaris$(ARCH).jar
@@ -1385,12 +1434,15 @@ rocksdbjavastaticrelease: rocksdbjavastatic
 	cd java/target;jar -uf $(ROCKSDB_JAR_ALL) librocksdbjni-*.so librocksdbjni-*.jnilib
 	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR_ALL) org/rocksdb/*.class org/rocksdb/util/*.class

-rocksdbjavastaticpublish: rocksdbjavastaticrelease
+rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral
+
+rocksdbjavastaticpublishcentral:
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-javadoc.jar -Dclassifier=javadoc
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar -Dclassifier=sources
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux64.jar -Dclassifier=linux64
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux32.jar -Dclassifier=linux32
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar -Dclassifier=osx
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-win64.jar -Dclassifier=win64
 	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/target/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar

 # A version of each $(LIBOBJECTS) compiled with -fPIC
@@ -71,3 +71,6 @@ quasardb uses a heavily tuned RocksDB as its persistence layer.

 ## Apache Flink
 [Apache Flink](https://flink.apache.org/news/2016/03/08/release-1.0.0.html) uses RocksDB to store state locally on a machine.
+
+## Dgraph
+[Dgraph](https://github.com/dgraph-io/dgraph) is an open-source, scalable, distributed, low latency, high throughput Graph database .They use RocksDB to store state locally on a machine.
@@ -1,3 +1,4 @@
+# Vagrant file
 Vagrant.configure("2") do |config|

  config.vm.provider "virtualbox" do |v|
@@ -12,7 +12,6 @@ define("ENV_HTTPS_APP_VALUE", "HTTPS_APP_VALUE");
 define("ENV_HTTPS_TOKEN_VALUE", "HTTPS_TOKEN_VALUE");

 define("PRIMARY_TOKEN_FILE", '/home/krad/.sandcastle');
-define("SECONDARY_TOKEN_FILE", '$HOME/.sandcastle');
 define("CONT_RUN_ALIAS", "leveldb");

 //////////////////////////////////////////////////////////////////////
@@ -25,8 +24,7 @@ function postURL($diffID, $url) {
  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
         . '"name":"click here for sandcastle tests for D' . $diffID . '", '
         . '"link":"' . $url . '"}\' | '
-         . 'http_proxy=fwdproxy.any.facebook.com:8080 '
-         . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
+         . 'arc call-conduit '
         . 'differential.updateunitresults';
  shell_exec($cmd);
 }
@@ -40,8 +38,7 @@ function buildUpdateTestStatusCmd($diffID, $test, $status) {
  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
         . '"name":"' . $test . '", '
         . '"result":"' . $status . '"}\' | '
-         . 'http_proxy=fwdproxy.any.facebook.com:8080 '
-         . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
+         . 'arc call-conduit '
         . 'differential.updateunitresults';
  return $cmd;
 }
@@ -106,7 +103,7 @@ function getSteps($applyDiff, $diffID, $username, $test) {
    // Patch the code (keep your fingures crossed).
    $patch = array(
      "name" => "Patch " . $diffID,
-      "shell" => "HTTPS_PROXY=fwdproxy:8080 arc --arcrc-file ~/.arcrc "
+      "shell" => "arc --arcrc-file ~/.arcrc "
                  . "patch --nocommit --diff " . $diffID,
      "user" => "root"
    );
@@ -175,6 +172,8 @@ function getSteps($applyDiff, $diffID, $username, $test) {
 function getSandcastleConfig() {
  $sandcastle_config = array();

+  $cwd = getcwd();
+  $cwd_token_file = "{$cwd}/.sandcastle";
  // This is a case when we're executed from a continuous run. Fetch the values
  // from the environment.
  if (getenv(ENV_POST_RECEIVE_HOOK)) {
@@ -183,14 +182,22 @@ function getSandcastleConfig() {
  } else {
    // This is a typical `[p]arc diff` case. Fetch the values from the specific
    // configuration files.
+    for ($i = 0; $i < 50; $i++) {
+      if (file_exists(PRIMARY_TOKEN_FILE) ||
+          file_exists($cwd_token_file)) {
+        break;
+      }
+      // If we failed to fetch the tokens, sleep for 0.2 second and try again
+      usleep(200000);
+    }
    assert(file_exists(PRIMARY_TOKEN_FILE) ||
-           file_exists(SECONDARY_TOKEN_FILE));
+           file_exists($cwd_token_file));

    // Try the primary location first, followed by a secondary.
    if (file_exists(PRIMARY_TOKEN_FILE)) {
      $cmd = 'cat ' . PRIMARY_TOKEN_FILE;
    } else {
-      $cmd = 'cat ' . SECONDARY_TOKEN_FILE;
+      $cmd = 'cat ' . $cwd_token_file;
    }

    assert(strlen($cmd) > 0);
@@ -292,11 +299,21 @@ function getSandcastleConfig() {
  );

  // Submit to Sandcastle.
-  $url = 'https://interngraph.intern.facebook.com/sandcastle/generate?'
-          .'command=SandcastleUniversalCommand'
-          .'&vcs=rocksdb-git&revision=origin%2Fmaster&type=lego'
-          .'&user=' . $username . '&alias=rocksdb-precommit'
-          .'&command-args=' . urlencode(json_encode($command));
+  $url = 'https://interngraph.intern.facebook.com/sandcastle/create';
+
+  $job = array(
+    'command' => 'SandcastleUniversalCommand',
+    'args' => $command,
+    'capabilities' => array(
+      'vcs' => 'rocksdb-int-git',
+      'type' => 'lego',
+    ),
+    'hash' => 'origin/master',
+    'user' => $username,
+    'alias' => 'rocksdb-precommit',
+    'tags' => array('rocksdb'),
+    'description' => 'Rocksdb precommit job',
+  );

  // Fetch the configuration necessary to submit a successful HTTPS request.
  $sandcastle_config = getSandcastleConfig();
@@ -304,8 +321,9 @@ function getSandcastleConfig() {
  $app = $sandcastle_config[0];
  $token = $sandcastle_config[1];

-  $cmd = 'https_proxy= HTTPS_PROXY= curl -s -k -F app=' . $app . ' '
-          . '-F token=' . $token . ' "' . $url . '"';
+  $cmd = 'curl -s -k -F app=' . $app . ' '
+          . '-F token=' . $token . ' -F job=\'' . json_encode($job)
+          .'\' "' . $url . '"';

  $output = shell_exec($cmd);
  assert(strlen($output) > 0);
@@ -51,12 +51,7 @@ if [ -z "$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
-      source "$PWD/build_tools/fbcode_config.sh"
-    else
-      # we need this to build with MySQL. Don't use for other purposes.
-      source "$PWD/build_tools/fbcode_config4.8.1.sh"
-    fi
+    source "$PWD/build_tools/fbcode_config.sh"
 fi

 # Delete existing output, if it exists
@@ -466,3 +461,4 @@ echo "PROFILING_FLAGS=$PROFILING_FLAGS" >> "$OUTPUT"
 if test -n "$JEMALLOC"; then
  echo "JEMALLOC=1" >> "$OUTPUT"
 fi
+echo "LUA_PATH=$LUA_PATH" >> "$OUTPUT"
@@ -1,17 +1,19 @@
-GCC_BASE=/mnt/gvfs/third-party2/gcc/cf7d14c625ce30bae1a4661c2319c5a283e4dd22/4.9.x/centos6-native/108cf83
-CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/8598c375b0e94e1448182eb3df034704144a838d/stable/centos6-native/3f16ddd
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/d6e0a7da6faba45f5e5b1638f9edd7afc2f34e7d/4.9.x/gcc-4.9-glibc-2.20/024dbc3
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/d282e6e8f3d20f4e40a516834847bdc038e07973/2.20/gcc-4.9-glibc-2.20/500e281
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-4.9-glibc-2.20/e9936bf
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/0882df3713c7a84f15abe368dc004581f20b39d7/1.2.8/gcc-5-glibc-2.23/9bc6787
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/740325875f6729f42d28deaa2147b0854f3a347e/1.0.6/gcc-5-glibc-2.23/9bc6787
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/0e790b441e2d9acd68d51e1d2e028f88c6a79ddf/r131/gcc-5-glibc-2.23/9bc6787
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/9455f75ff7f4831dc9fda02a6a0f8c68922fad8f/1.0.0/gcc-5-glibc-2.23/9bc6787
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/f001a51b2854957676d07306ef3abf67186b5c8b/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/fc8a13ca1fffa4d0765c716c5a0b49f0c107518f/master/gcc-5-glibc-2.23/1c32b4b
-NUMA_BASE=/mnt/gvfs/third-party2/numa/17c514c4d102a25ca15f4558be564eeed76f4b6a/2.0.8/gcc-5-glibc-2.23/9bc6787
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/ad576de2a1ea560c4d3434304f0fc4e079bede42/trunk/gcc-5-glibc-2.23/b1847cb
-TBB_BASE=/mnt/gvfs/third-party2/tbb/9d9a554877d0c5bef330fe818ab7178806dd316a/4.0_update2/gcc-4.9-glibc-2.20/e9936bf
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/7c111ff27e0c466235163f00f280a9d617c3d2ec/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/b7fd454c4b10c6a81015d4524ed06cdeab558490/2.26/centos6-native/da39a3e
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d7f4d4d86674a57668e3a96f76f0e17dd0eb8765/3.10.0/gcc-4.9-glibc-2.20/e9936bf
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+GCC_BASE=/mnt/gvfs/third-party2/gcc/7331085db891a2ef4a88a48a751d834e8d68f4cb/7.x/centos7-native/b2ef2b6
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/963d9aeda70cc4779885b1277484fe7544a04e3e/9.0.0/platform007/9e92d53/
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/6ace84e956873d53638c738b6f65f3f469cca74c/7.x/platform007/5620abc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/192b0f42d63dcf6210d6ceae387b49af049e6e0c/2.26/platform007/f259413
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2d9f0b9a4274cc21f61272a9e89bdb859bce8f1f/1.2.8/platform007/ca4da3d
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0f607f8fc442ea7d6b876931b1898bb573d5e5da/1.9.1/platform007/ca4da3d
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/ca22bc441a4eb709e9e0b1f9fec9750fed7b31c5/1.4.x/platform007/15a3614
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/c26f08f47ac35fc31da2633b7da92d6b863246eb/master/platform007/c26c002
+NUMA_BASE=/mnt/gvfs/third-party2/numa/3f3fb57a5ccc5fd21c66416c0b83e0aa76a05376/2.0.11/platform007/ca4da3d
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/40c73d874898b386a71847f1b99115d93822d11f/1.4/platform007/6f3e0a9
+TBB_BASE=/mnt/gvfs/third-party2/tbb/4ce8e8dba77cdbd81b75d6f0c32fd7a1b76a11ec/2018_U5/platform007/ca4da3d
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/fb251ecd2f5ae16f8671f7014c246e52a748fe0b/fb/platform007/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/ab9f09bba370e7066cafd4eb59752db93f2e8312/2.29.1/platform007/15a3614
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d42d152a15636529b0861ec493927200ebebca8e/3.15.0/platform007/ca4da3d
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832
@@ -15,3 +15,4 @@ TBB_BASE=/mnt/gvfs/third-party2/tbb/9d9a554877d0c5bef330fe818ab7178806dd316a/4.0
 KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/7c111ff27e0c466235163f00f280a9d617c3d2ec/4.0.9-36_fbk5_2933_gd092e3f/gcc-4.8.1-glibc-2.17/da39a3e
 BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/b7fd454c4b10c6a81015d4524ed06cdeab558490/2.26/centos6-native/da39a3e
 VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d7f4d4d86674a57668e3a96f76f0e17dd0eb8765/3.8.1/gcc-4.8.1-glibc-2.17/c3f970a
+LUA_BASE=/mnt/gvfs/third-party2/lua/61e4abf5813bbc39bc4f548757ccfcadde175a48/5.2.3/centos6-native/730f94e
@@ -13,7 +13,7 @@ source "$BASEDIR/dependencies.sh"
 CFLAGS=""

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

 # glibc
@@ -43,12 +43,16 @@ if test -z $PIC_BUILD; then
  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
  CFLAGS+=" -DLZ4"
-
-  ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
-  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
-  CFLAGS+=" -DZSTD"
 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
@@ -56,7 +60,7 @@ if test -z $PIC_BUILD; then
 else
  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
 fi
-CFLAGS+=" -DGFLAGS=google"
+CFLAGS+=" -DGFLAGS=gflags"

 # location of jemalloc
 JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
@@ -104,8 +108,8 @@ if [ -z "$USE_CLANG" ]; then
  CXX="$GCC_BASE/bin/g++"

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

  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/4.9.x "
-  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/4.9.x/x86_64-facebook-linux "
+  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"
@@ -128,13 +132,14 @@ else
 fi

 CFLAGS+=" $DEPS_INCLUDE"
-CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
 CXXFLAGS+=" $CFLAGS"

 EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
-EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/gcc-4.9-glibc-2.20/lib/ld.so"
+EXEC_LDFLAGS+=" -B$BINUTILS/gold"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
-EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/gcc-4.9-glibc-2.20/lib"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
 # required by libtbb
 EXEC_LDFLAGS+=" -ldl"

@@ -144,4 +149,4 @@ EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GF

 VALGRIND_VER="$VALGRIND_BASE/bin/"

-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD 
@@ -110,4 +110,6 @@ EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP2_LIBS $LZ4_LIBS $ZSTD_LIBS $G

 VALGRIND_VER="$VALGRIND_BASE/bin/"

-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE
+LUA_PATH="$LUA_BASE"
+
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE LUA_PATH
@@ -65,8 +65,8 @@ function main() {

  if [[ -d /vagrant ]]; then
    if [[ $OS = "ubuntu" ]]; then
-      package g++-4.7
-      export CXX=g++-4.7
+      package g++-4.8
+      export CXX=g++-4.8

      # the deb would depend on libgflags2, but the static lib is the only thing
      # installed by make install
@@ -100,7 +100,7 @@ CONTRUN_NAME="ROCKSDB_CONTRUN_NAME"
 # the value of $ONCALL. If it's a diff then just call `false` to make sure
 # that errors will be properly propagated to the caller.
 if [ ! -z $ONCALL ]; then
-  TASK_CREATION_TOOL="/usr/local/bin/mysql_mtr_filter --rocksdb"
+  TASK_CREATION_TOOL="/usr/local/bin/mysql_mtr_filter --rocksdb --oncall $ONCALL"
 else
  TASK_CREATION_TOOL="false"
 fi
@@ -548,7 +548,7 @@ VALGRIND_TEST_COMMANDS="[
            {
                'name':'Run RocksDB debug unit tests',
                'timeout': 86400,
-                'shell':'$DISABLE_JEMALLOC $SHM $DEBUG make $PARALLELISM valgrind_check || $CONTRUN_NAME=valgrind_check $TASK_CREATION_TOOL',
+                'shell':'$SHM $DEBUG make $PARALLELISM valgrind_test || $CONTRUN_NAME=valgrind_check $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            },
@@ -87,6 +87,7 @@ get_lib_base tbb        4.0_update2 gcc-4.9-glibc-2.20
 get_lib_base kernel-headers LATEST 
 get_lib_base binutils   LATEST centos6-native 
 get_lib_base valgrind   3.10.0 gcc-4.9-glibc-2.20
+get_lib_base lua        5.2.3 gcc-4.9-glibc-2.20

 git diff $OUTPUT

@@ -125,5 +126,6 @@ get_lib_base tbb        4.0_update2 gcc-4.8.1-glibc-2.17
 get_lib_base kernel-headers LATEST gcc-4.8.1-glibc-2.17 
 get_lib_base binutils   LATEST centos6-native 
 get_lib_base valgrind   3.8.1  gcc-4.8.1-glibc-2.17
+get_lib_base lua        5.2.3 centos6-native

 git diff $OUTPUT
@@ -0,0 +1,22 @@
+# - Find JeMalloc library
+# Find the native JeMalloc includes and library
+# This module defines
+#  JEMALLOC_INCLUDE_DIRS, where to find jemalloc.h, Set when
+#                        JEMALLOC_INCLUDE_DIR is found.
+#  JEMALLOC_LIBRARIES, libraries to link against to use JeMalloc.
+#  JEMALLOC_FOUND, If false, do not try to use JeMalloc.
+#
+find_path(JEMALLOC_INCLUDE_DIR
+  jemalloc/jemalloc.h)
+
+find_library(JEMALLOC_LIBRARIES
+  jemalloc)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(JeMalloc DEFAULT_MSG
+    JEMALLOC_LIBRARIES JEMALLOC_INCLUDE_DIR)
+
+MARK_AS_ADVANCED(
+  JEMALLOC_INCLUDE_DIR
+  JEMALLOC_LIBRARIES
+)
@@ -0,0 +1,23 @@
+# - Find Snappy
+# Find the snappy compression library and includes
+#
+# SNAPPY_INCLUDE_DIR - where to find snappy.h, etc.
+# SNAPPY_LIBRARIES - List of libraries when using snappy.
+# SNAPPY_FOUND - True if snappy found.
+
+find_path(SNAPPY_INCLUDE_DIR
+  NAMES snappy.h
+  HINTS ${SNAPPY_ROOT_DIR}/include)
+
+find_library(SNAPPY_LIBRARIES
+  NAMES snappy
+  HINTS ${SNAPPY_ROOT_DIR}/lib)
+
+# handle the QUIETLY and REQUIRED arguments and set UUID_FOUND to TRUE if
+# all listed variables are TRUE
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(snappy DEFAULT_MSG SNAPPY_LIBRARIES SNAPPY_INCLUDE_DIR)
+
+mark_as_advanced(
+  SNAPPY_LIBRARIES
+  SNAPPY_INCLUDE_DIR)
@@ -45,6 +45,7 @@ TableBuilder* NewTableBuilder(
    uint32_t column_family_id, const std::string& column_family_name,
    WritableFileWriter* file, const CompressionType compression_type,
    const CompressionOptions& compression_opts,
+    int level,
    const std::string* compression_dict, const bool skip_filters) {
  assert((column_family_id ==
          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
@@ -53,14 +54,15 @@ TableBuilder* NewTableBuilder(
      TableBuilderOptions(ioptions, internal_comparator,
                          int_tbl_prop_collector_factories, compression_type,
                          compression_opts, compression_dict, skip_filters,
-                          column_family_name),
+                          column_family_name, level),
      column_family_id, file);
 }

 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, FileMetaData* meta,
+    TableCache* table_cache, InternalIterator* iter,
+    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
@@ -80,6 +82,13 @@ Status BuildTable(
  Status s;
  meta->fd.file_size = 0;
  iter->SeekToFirst();
+  std::unique_ptr<RangeDelAggregator> range_del_agg(
+      new RangeDelAggregator(internal_comparator, snapshots));
+  s = range_del_agg->AddTombstones(std::move(range_del_iter));
+  if (!s.ok()) {
+    // may be non-ok if a range tombstone key is unparsable
+    return s;
+  }

  std::string fname = TableFileName(ioptions.db_paths, meta->fd.GetNumber(),
                                    meta->fd.GetPathId());
@@ -89,7 +98,7 @@ Status BuildTable(
 #endif  // !ROCKSDB_LITE
  TableProperties tp;

-  if (iter->Valid()) {
+  if (iter->Valid() || range_del_agg->ShouldAddTombstones()) {
    TableBuilder* builder;
    unique_ptr<WritableFileWriter> file_writer;
    {
@@ -108,7 +117,7 @@ Status BuildTable(
      builder = NewTableBuilder(
          ioptions, internal_comparator, int_tbl_prop_collector_factories,
          column_family_id, column_family_name, file_writer.get(), compression,
-          compression_opts);
+          compression_opts, level);
    }

    MergeHelper merge(env, internal_comparator.user_comparator(),
@@ -117,10 +126,10 @@ Status BuildTable(
                      true /* internal key corruption is not ok */,
                      snapshots.empty() ? 0 : snapshots.back());

-    CompactionIterator c_iter(iter, internal_comparator.user_comparator(),
-                              &merge, kMaxSequenceNumber, &snapshots,
-                              earliest_write_conflict_snapshot, env,
-                              true /* internal key corruption is not ok */);
+    CompactionIterator c_iter(
+        iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
+        &snapshots, earliest_write_conflict_snapshot, env,
+        true /* internal key corruption is not ok */, range_del_agg.get());
    c_iter.SeekToFirst();
    for (; c_iter.Valid(); c_iter.Next()) {
      const Slice& key = c_iter.key();
@@ -135,6 +144,9 @@ Status BuildTable(
            ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
      }
    }
+    // nullptr for table_{min,max} so all range tombstones will be flushed
+    range_del_agg->AddToBuilder(builder, nullptr /* lower_bound */,
+                                nullptr /* upper_bound */, meta);

    // Finish and check for builder errors
    bool empty = builder->NumEntries() == 0;
@@ -169,7 +181,8 @@ Status BuildTable(
    if (s.ok() && !empty) {
      // Verify that the table is usable
      std::unique_ptr<InternalIterator> it(table_cache->NewIterator(
-          ReadOptions(), env_options, internal_comparator, meta->fd, nullptr,
+          ReadOptions(), env_options, internal_comparator, meta->fd,
+          nullptr /* range_del_agg */, nullptr,
          (internal_stats == nullptr) ? nullptr
                                      : internal_stats->GetFileReadHist(0),
          false /* for_compaction */, nullptr /* arena */,
@@ -17,6 +17,7 @@
 #include "rocksdb/status.h"
 #include "rocksdb/table_properties.h"
 #include "rocksdb/types.h"
+#include "table/scoped_arena_iterator.h"
 #include "util/cf_options.h"
 #include "util/event_logger.h"

@@ -48,6 +49,7 @@ TableBuilder* NewTableBuilder(
    uint32_t column_family_id, const std::string& column_family_name,
    WritableFileWriter* file, const CompressionType compression_type,
    const CompressionOptions& compression_opts,
+    int level,
    const std::string* compression_dict = nullptr,
    const bool skip_filters = false);

@@ -62,7 +64,8 @@ TableBuilder* NewTableBuilder(
 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, FileMetaData* meta,
+    TableCache* table_cache, InternalIterator* iter,
+    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
    const InternalKeyComparator& internal_comparator,
    const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories,
@@ -30,6 +30,7 @@
 #include "rocksdb/statistics.h"
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/table.h"
+#include "rocksdb/rate_limiter.h"
 #include "rocksdb/utilities/backupable_db.h"
 #include "utilities/merge_operators.h"

@@ -47,10 +48,12 @@ using rocksdb::WALRecoveryMode;
 using rocksdb::DB;
 using rocksdb::DBOptions;
 using rocksdb::Env;
+using rocksdb::EnvOptions;
 using rocksdb::InfoLogLevel;
 using rocksdb::FileLock;
 using rocksdb::FilterPolicy;
 using rocksdb::FlushOptions;
+using rocksdb::IngestExternalFileOptions;
 using rocksdb::Iterator;
 using rocksdb::Logger;
 using rocksdb::MergeOperator;
@@ -68,6 +71,7 @@ using rocksdb::Slice;
 using rocksdb::SliceParts;
 using rocksdb::SliceTransform;
 using rocksdb::Snapshot;
+using rocksdb::SstFileWriter;
 using rocksdb::Status;
 using rocksdb::WritableFile;
 using rocksdb::WriteBatch;
@@ -78,6 +82,8 @@ using rocksdb::BackupableDBOptions;
 using rocksdb::BackupInfo;
 using rocksdb::RestoreOptions;
 using rocksdb::CompactRangeOptions;
+using rocksdb::RateLimiter;
+using rocksdb::NewGenericRateLimiter;

 using std::shared_ptr;

@@ -108,6 +114,10 @@ struct rocksdb_logger_t          { shared_ptr<Logger>  rep; };
 struct rocksdb_cache_t           { shared_ptr<Cache>   rep; };
 struct rocksdb_livefiles_t       { std::vector<LiveFileMetaData> rep; };
 struct rocksdb_column_family_handle_t  { ColumnFamilyHandle* rep; };
+struct rocksdb_envoptions_t      { EnvOptions        rep; };
+struct rocksdb_ingestexternalfileoptions_t  { IngestExternalFileOptions rep; };
+struct rocksdb_sstfilewriter_t   { SstFileWriter*    rep; };
+struct rocksdb_ratelimiter_t     { RateLimiter*      rep; };

 struct rocksdb_compactionfiltercontext_t {
  CompactionFilter::Context rep;
@@ -1021,6 +1031,11 @@ void rocksdb_iter_seek(rocksdb_iterator_t* iter, const char* k, size_t klen) {
  iter->rep->Seek(Slice(k, klen));
 }

+void rocksdb_iter_seek_for_prev(rocksdb_iterator_t* iter, const char* k,
+                                size_t klen) {
+  iter->rep->SeekForPrev(Slice(k, klen));
+}
+
 void rocksdb_iter_next(rocksdb_iterator_t* iter) {
  iter->rep->Next();
 }
@@ -1511,8 +1526,13 @@ void rocksdb_options_set_max_bytes_for_level_base(
  opt->rep.max_bytes_for_level_base = n;
 }

-void rocksdb_options_set_max_bytes_for_level_multiplier(
-    rocksdb_options_t* opt, int n) {
+void rocksdb_options_set_level_compaction_dynamic_level_bytes(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.level_compaction_dynamic_level_bytes = v;
+}
+
+void rocksdb_options_set_max_bytes_for_level_multiplier(rocksdb_options_t* opt,
+                                                        double n) {
  opt->rep.max_bytes_for_level_multiplier = n;
 }

@@ -1688,6 +1708,16 @@ void rocksdb_options_set_bytes_per_sync(
  opt->rep.bytes_per_sync = v;
 }

+void rocksdb_options_set_allow_concurrent_memtable_write(rocksdb_options_t* opt,
+                                                         unsigned char v) {
+  opt->rep.allow_concurrent_memtable_write = v;
+}
+
+void rocksdb_options_set_enable_write_thread_adaptive_yield(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.enable_write_thread_adaptive_yield = v;
+}
+
 void rocksdb_options_set_verify_checksums_in_compaction(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.verify_checksums_in_compaction = v;
@@ -1715,6 +1745,11 @@ void rocksdb_options_set_max_background_compactions(rocksdb_options_t* opt, int
  opt->rep.max_background_compactions = n;
 }

+void rocksdb_options_set_base_background_compactions(rocksdb_options_t* opt,
+                                                     int n) {
+  opt->rep.base_background_compactions = n;
+}
+
 void rocksdb_options_set_max_background_flushes(rocksdb_options_t* opt, int n) {
  opt->rep.max_background_flushes = n;
 }
@@ -1783,11 +1818,7 @@ void rocksdb_options_prepare_for_bulk_load(rocksdb_options_t* opt) {
 }

 void rocksdb_options_set_memtable_vector_rep(rocksdb_options_t *opt) {
-  static rocksdb::VectorRepFactory* factory = 0;
-  if (!factory) {
-    factory = new rocksdb::VectorRepFactory;
-  }
-  opt->rep.memtable_factory.reset(factory);
+  opt->rep.memtable_factory.reset(new rocksdb::VectorRepFactory);
 }

 void rocksdb_options_set_memtable_prefix_bloom_size_ratio(
@@ -1803,36 +1834,26 @@ void rocksdb_options_set_memtable_huge_page_size(rocksdb_options_t* opt,
 void rocksdb_options_set_hash_skip_list_rep(
    rocksdb_options_t *opt, size_t bucket_count,
    int32_t skiplist_height, int32_t skiplist_branching_factor) {
-  static rocksdb::MemTableRepFactory* factory = 0;
-  if (!factory) {
-    factory = rocksdb::NewHashSkipListRepFactory(
-        bucket_count, skiplist_height, skiplist_branching_factor);
-  }
+  rocksdb::MemTableRepFactory* factory = rocksdb::NewHashSkipListRepFactory(
+      bucket_count, skiplist_height, skiplist_branching_factor);
  opt->rep.memtable_factory.reset(factory);
 }

 void rocksdb_options_set_hash_link_list_rep(
    rocksdb_options_t *opt, size_t bucket_count) {
-  static rocksdb::MemTableRepFactory* factory = 0;
-  if (!factory) {
-    factory = rocksdb::NewHashLinkListRepFactory(bucket_count);
-  }
-  opt->rep.memtable_factory.reset(factory);
+  opt->rep.memtable_factory.reset(rocksdb::NewHashLinkListRepFactory(bucket_count));
 }

 void rocksdb_options_set_plain_table_factory(
    rocksdb_options_t *opt, uint32_t user_key_len, int bloom_bits_per_key,
    double hash_table_ratio, size_t index_sparseness) {
-  static rocksdb::TableFactory* factory = 0;
-  if (!factory) {
-    rocksdb::PlainTableOptions options;
-    options.user_key_len = user_key_len;
-    options.bloom_bits_per_key = bloom_bits_per_key;
-    options.hash_table_ratio = hash_table_ratio;
-    options.index_sparseness = index_sparseness;
+  rocksdb::PlainTableOptions options;
+  options.user_key_len = user_key_len;
+  options.bloom_bits_per_key = bloom_bits_per_key;
+  options.hash_table_ratio = hash_table_ratio;
+  options.index_sparseness = index_sparseness;

-    factory = rocksdb::NewPlainTableFactory(options);
-  }
+  rocksdb::TableFactory* factory = rocksdb::NewPlainTableFactory(options);
  opt->rep.table_factory.reset(factory);
 }

@@ -1888,6 +1909,28 @@ char *rocksdb_options_statistics_get_string(rocksdb_options_t *opt) {
  return nullptr;
 }

+void rocksdb_options_set_ratelimiter(rocksdb_options_t *opt, rocksdb_ratelimiter_t *limiter) {
+  opt->rep.rate_limiter.reset(limiter->rep);
+  limiter->rep = nullptr;
+}
+
+rocksdb_ratelimiter_t* rocksdb_ratelimiter_create(
+    int64_t rate_bytes_per_sec,
+    int64_t refill_period_us,
+    int32_t fairness) {
+  rocksdb_ratelimiter_t* rate_limiter = new rocksdb_ratelimiter_t;
+  rate_limiter->rep = NewGenericRateLimiter(rate_bytes_per_sec,
+                                            refill_period_us, fairness);
+  return rate_limiter;
+}
+
+void rocksdb_ratelimiter_destroy(rocksdb_ratelimiter_t *limiter) {
+  if (limiter->rep) {
+	delete limiter->rep;
+  }
+  delete limiter;
+}
+
 /*
 TODO:
 DB::OpenForReadOnly
@@ -2201,6 +2244,107 @@ void rocksdb_env_destroy(rocksdb_env_t* env) {
  delete env;
 }

+rocksdb_envoptions_t* rocksdb_envoptions_create() {
+  rocksdb_envoptions_t* opt = new rocksdb_envoptions_t;
+  return opt;
+}
+
+void rocksdb_envoptions_destroy(rocksdb_envoptions_t* opt) { delete opt; }
+
+rocksdb_sstfilewriter_t* rocksdb_sstfilewriter_create(
+    const rocksdb_envoptions_t* env, const rocksdb_options_t* io_options) {
+  rocksdb_sstfilewriter_t* writer = new rocksdb_sstfilewriter_t;
+  writer->rep =
+      new SstFileWriter(env->rep, io_options->rep, io_options->rep.comparator);
+  return writer;
+}
+
+rocksdb_sstfilewriter_t* rocksdb_sstfilewriter_create_with_comparator(
+    const rocksdb_envoptions_t* env, const rocksdb_options_t* io_options,
+    const rocksdb_comparator_t* comparator) {
+  rocksdb_sstfilewriter_t* writer = new rocksdb_sstfilewriter_t;
+  writer->rep =
+      new SstFileWriter(env->rep, io_options->rep, io_options->rep.comparator);
+  return writer;
+}
+
+void rocksdb_sstfilewriter_open(rocksdb_sstfilewriter_t* writer,
+                                const char* name, char** errptr) {
+  SaveError(errptr, writer->rep->Open(std::string(name)));
+}
+
+void rocksdb_sstfilewriter_add(rocksdb_sstfilewriter_t* writer, const char* key,
+                               size_t keylen, const char* val, size_t vallen,
+                               char** errptr) {
+  SaveError(errptr, writer->rep->Add(Slice(key, keylen), Slice(val, vallen)));
+}
+
+void rocksdb_sstfilewriter_finish(rocksdb_sstfilewriter_t* writer,
+                                  char** errptr) {
+  SaveError(errptr, writer->rep->Finish(NULL));
+}
+
+void rocksdb_sstfilewriter_destroy(rocksdb_sstfilewriter_t* writer) {
+  delete writer->rep;
+  delete writer;
+}
+
+rocksdb_ingestexternalfileoptions_t*
+rocksdb_ingestexternalfileoptions_create() {
+  rocksdb_ingestexternalfileoptions_t* opt =
+      new rocksdb_ingestexternalfileoptions_t;
+  return opt;
+}
+
+void rocksdb_ingestexternalfileoptions_set_move_files(
+    rocksdb_ingestexternalfileoptions_t* opt, unsigned char move_files) {
+  opt->rep.move_files = move_files;
+}
+
+void rocksdb_ingestexternalfileoptions_set_snapshot_consistency(
+    rocksdb_ingestexternalfileoptions_t* opt,
+    unsigned char snapshot_consistency) {
+  opt->rep.snapshot_consistency = snapshot_consistency;
+}
+
+void rocksdb_ingestexternalfileoptions_set_allow_global_seqno(
+    rocksdb_ingestexternalfileoptions_t* opt,
+    unsigned char allow_global_seqno) {
+  opt->rep.allow_global_seqno = allow_global_seqno;
+}
+
+void rocksdb_ingestexternalfileoptions_set_allow_blocking_flush(
+    rocksdb_ingestexternalfileoptions_t* opt,
+    unsigned char allow_blocking_flush) {
+  opt->rep.allow_blocking_flush = allow_blocking_flush;
+}
+
+void rocksdb_ingestexternalfileoptions_destroy(
+    rocksdb_ingestexternalfileoptions_t* opt) {
+  delete opt;
+}
+
+void rocksdb_ingest_external_file(
+    rocksdb_t* db, const char* const* file_list, const size_t list_len,
+    const rocksdb_ingestexternalfileoptions_t* opt, char** errptr) {
+  std::vector<std::string> files(list_len);
+  for (size_t i = 0; i < list_len; ++i) {
+    files[i] = std::string(file_list[i]);
+  }
+  SaveError(errptr, db->rep->IngestExternalFile(files, opt->rep));
+}
+
+void rocksdb_ingest_external_file_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* handle,
+    const char* const* file_list, const size_t list_len,
+    const rocksdb_ingestexternalfileoptions_t* opt, char** errptr) {
+  std::vector<std::string> files(list_len);
+  for (size_t i = 0; i < list_len; ++i) {
+    files[i] = std::string(file_list[i]);
+  }
+  SaveError(errptr, db->rep->IngestExternalFile(handle->rep, files, opt->rep));
+}
+
 rocksdb_slicetransform_t* rocksdb_slicetransform_create(
    void* state,
    void (*destructor)(void*),
@@ -2,17 +2,18 @@
   Use of this source code is governed by a BSD-style license that can be
   found in the LICENSE file. See the AUTHORS file for names of contributors. */

+#include <stdio.h>
+
 #ifndef ROCKSDB_LITE  // Lite does not support C API

 #include "rocksdb/c.h"

 #include <stddef.h>
-#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/types.h>
 #ifndef OS_WIN
-#  include <unistd.h>
+#include <unistd.h>
 #endif
 #include <inttypes.h>

@@ -21,11 +22,10 @@

 #include <Windows.h>

-# define snprintf _snprintf
+#define snprintf _snprintf

 // Ok for uniqueness
 int geteuid() {
-
  int result = 0;

  result = ((int)GetCurrentProcessId() << 16);
@@ -38,13 +38,13 @@ int geteuid() {

 const char* phase = "";
 static char dbname[200];
+static char sstfilename[200];
 static char dbbackupname[200];

 static void StartPhase(const char* name) {
  fprintf(stderr, "=== Test %s\n", name);
  phase = name;
 }
-
 static const char* GetTempDir(void) {
    const char* ret = getenv("TEST_TMPDIR");
    if (ret == NULL || ret[0] == '\0')
@@ -291,6 +291,7 @@ int main(int argc, char** argv) {
  rocksdb_block_based_table_options_t* table_options;
  rocksdb_readoptions_t* roptions;
  rocksdb_writeoptions_t* woptions;
+  rocksdb_ratelimiter_t* rate_limiter;
  char* err = NULL;
  int run = -1;

@@ -304,6 +305,11 @@ int main(int argc, char** argv) {
           GetTempDir(),
           ((int) geteuid()));

+  snprintf(sstfilename, sizeof(sstfilename),
+           "%s/rocksdb_c_test-%d-sst",
+           GetTempDir(),
+           ((int)geteuid()));
+
  StartPhase("create_objects");
  cmp = rocksdb_comparator_create(NULL, CmpDestroy, CmpCompare, CmpName);
  env = rocksdb_create_default_env();
@@ -317,6 +323,7 @@ int main(int argc, char** argv) {
  rocksdb_options_set_write_buffer_size(options, 100000);
  rocksdb_options_set_paranoid_checks(options, 1);
  rocksdb_options_set_max_open_files(options, 10);
+  rocksdb_options_set_base_background_compactions(options, 1);
  table_options = rocksdb_block_based_options_create();
  rocksdb_block_based_options_set_block_cache(table_options, cache);
  rocksdb_options_set_block_based_table_factory(options, table_options);
@@ -326,6 +333,9 @@ int main(int argc, char** argv) {
  int compression_levels[] = {rocksdb_no_compression, rocksdb_no_compression,
                              rocksdb_no_compression, rocksdb_no_compression};
  rocksdb_options_set_compression_per_level(options, compression_levels, 4);
+  rate_limiter = rocksdb_ratelimiter_create(1000 * 1024 * 1024, 100 * 1000, 10);
+  rocksdb_options_set_ratelimiter(options, rate_limiter);
+  rocksdb_ratelimiter_destroy(rate_limiter);

  roptions = rocksdb_readoptions_create();
  rocksdb_readoptions_set_verify_checksums(roptions, 1);
@@ -487,6 +497,10 @@ int main(int argc, char** argv) {
    CheckIter(iter, "foo", "hello");
    rocksdb_iter_seek(iter, "b", 1);
    CheckIter(iter, "box", "c");
+    rocksdb_iter_seek_for_prev(iter, "g", 1);
+    CheckIter(iter, "foo", "hello");
+    rocksdb_iter_seek_for_prev(iter, "box", 3);
+    CheckIter(iter, "box", "c");
    rocksdb_iter_get_error(iter, &err);
    CheckNoError(err);
    rocksdb_iter_destroy(iter);
@@ -565,6 +579,59 @@ int main(int argc, char** argv) {
    rocksdb_release_snapshot(db, snap);
  }

+  StartPhase("addfile");
+  {
+    rocksdb_envoptions_t* env_opt = rocksdb_envoptions_create();
+    rocksdb_options_t* io_options = rocksdb_options_create();
+    rocksdb_sstfilewriter_t* writer =
+        rocksdb_sstfilewriter_create(env_opt, io_options);
+
+    unlink(sstfilename);
+    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk1", 5, "v1", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v2", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v3", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_finish(writer, &err);
+    CheckNoError(err);
+
+    rocksdb_ingestexternalfileoptions_t* ing_opt =
+        rocksdb_ingestexternalfileoptions_create();
+    const char* file_list[1] = {sstfilename};
+    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "sstk1", "v1");
+    CheckGet(db, roptions, "sstk2", "v2");
+    CheckGet(db, roptions, "sstk3", "v3");
+
+    unlink(sstfilename);
+    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v4", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk22", 6, "v5", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v6", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_finish(writer, &err);
+    CheckNoError(err);
+
+    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "sstk1", "v1");
+    CheckGet(db, roptions, "sstk2", "v4");
+    CheckGet(db, roptions, "sstk22", "v5");
+    CheckGet(db, roptions, "sstk3", "v6");
+
+    rocksdb_ingestexternalfileoptions_destroy(ing_opt);
+    rocksdb_sstfilewriter_destroy(writer);
+    rocksdb_options_destroy(io_options);
+    rocksdb_envoptions_destroy(env_opt);
+  }
+
  StartPhase("repair");
  {
    // If we do not compact here, then the lazy deletion of
@@ -700,7 +767,7 @@ int main(int argc, char** argv) {
    rocksdb_destroy_db(options, dbname, &err);
    CheckNoError(err)

-    rocksdb_options_t* db_options = rocksdb_options_create();
+        rocksdb_options_t* db_options = rocksdb_options_create();
    rocksdb_options_set_create_if_missing(db_options, 1);
    db = rocksdb_open(db_options, dbname, &err);
    CheckNoError(err)
@@ -829,6 +896,7 @@ int main(int argc, char** argv) {
    rocksdb_options_set_prefix_extractor(options, rocksdb_slicetransform_create_fixed_prefix(3));
    rocksdb_options_set_hash_skip_list_rep(options, 5000, 4, 4);
    rocksdb_options_set_plain_table_factory(options, 4, 10, 0.75, 16);
+    rocksdb_options_set_allow_concurrent_memtable_write(options, 0);

    db = rocksdb_open(options, dbname, &err);
    CheckNoError(err);
@@ -944,6 +1012,28 @@ int main(int argc, char** argv) {
    }
  }

+  // Simple sanity check that setting memtable rep works.
+  StartPhase("memtable_reps");
+  {
+    // Create database with vector memtable.
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+    CheckNoError(err);
+
+    rocksdb_options_set_memtable_vector_rep(options);
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+
+    // Create database with hash skiplist memtable.
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+    CheckNoError(err);
+
+    rocksdb_options_set_hash_skip_list_rep(options, 5000, 4, 4);
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+  }
+
  StartPhase("cleanup");
  rocksdb_close(db);
  rocksdb_options_destroy(options);
@@ -959,7 +1049,6 @@ int main(int argc, char** argv) {
 }

 #else
-#include <stdio.h>

 int main() {
  fprintf(stderr, "SKIPPED\n");
@@ -27,6 +27,7 @@
 #include "db/version_set.h"
 #include "db/write_controller.h"
 #include "memtable/hash_skiplist_rep.h"
+#include "table/block_based_table_factory.h"
 #include "util/autovector.h"
 #include "util/compression.h"
 #include "util/options_helper.h"
@@ -45,6 +46,11 @@ ColumnFamilyHandleImpl::ColumnFamilyHandleImpl(

 ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
  if (cfd_ != nullptr) {
+#ifndef ROCKSDB_LITE
+    for (auto& listener : cfd_->ioptions()->listeners) {
+      listener->OnColumnFamilyHandleDeletionStarted(this);
+    }
+#endif  // ROCKSDB_LITE
    // Job id == 0 means that this is not our background process, but rather
    // user thread
    JobContext job_context(0);
@@ -71,10 +77,7 @@ Status ColumnFamilyHandleImpl::GetDescriptor(ColumnFamilyDescriptor* desc) {
 #ifndef ROCKSDB_LITE
  // accessing mutable cf-options requires db mutex.
  InstrumentedMutexLock l(mutex_);
-  *desc = ColumnFamilyDescriptor(
-      cfd()->GetName(),
-      BuildColumnFamilyOptions(*cfd()->options(),
-                               *cfd()->GetLatestMutableCFOptions()));
+  *desc = ColumnFamilyDescriptor(cfd()->GetName(), cfd()->GetLatestCFOptions());
  return Status::OK();
 #else
  return Status::NotSupported();
@@ -86,11 +89,11 @@ const Comparator* ColumnFamilyHandleImpl::GetComparator() const {
 }

 void GetIntTblPropCollectorFactory(
-    const ColumnFamilyOptions& cf_options,
+    const ImmutableCFOptions& ioptions,
    std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories) {
-  auto& collector_factories = cf_options.table_properties_collector_factories;
-  for (size_t i = 0; i < cf_options.table_properties_collector_factories.size();
+  auto& collector_factories = ioptions.table_properties_collector_factories;
+  for (size_t i = 0; i < ioptions.table_properties_collector_factories.size();
       ++i) {
    assert(collector_factories[i]);
    int_tbl_prop_collector_factories->emplace_back(
@@ -136,11 +139,9 @@ Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
  return Status::OK();
 }

-ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
-                                    const InternalKeyComparator* icmp,
+ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                    const ColumnFamilyOptions& src) {
  ColumnFamilyOptions result = src;
-  result.comparator = icmp;
  size_t clamp_max = std::conditional<
      sizeof(size_t) == 4, std::integral_constant<size_t, 0xffffffff>,
      std::integral_constant<uint64_t, 64ull << 30>>::type::value;
@@ -158,6 +159,10 @@ ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
  result.min_write_buffer_number_to_merge =
      std::min(result.min_write_buffer_number_to_merge,
               result.max_write_buffer_number - 1);
+  if (result.min_write_buffer_number_to_merge < 1) {
+    result.min_write_buffer_number_to_merge = 1;
+  }
+
  if (result.num_levels < 1) {
    result.num_levels = 1;
  }
@@ -338,7 +343,7 @@ void SuperVersionUnrefHandle(void* ptr) {
 ColumnFamilyData::ColumnFamilyData(
    uint32_t id, const std::string& name, Version* _dummy_versions,
    Cache* _table_cache, WriteBufferManager* write_buffer_manager,
-    const ColumnFamilyOptions& cf_options, const DBOptions* db_options,
+    const ColumnFamilyOptions& cf_options, const ImmutableDBOptions& db_options,
    const EnvOptions& env_options, ColumnFamilySet* column_family_set)
    : id_(id),
      name_(name),
@@ -347,14 +352,15 @@ ColumnFamilyData::ColumnFamilyData(
      refs_(0),
      dropped_(false),
      internal_comparator_(cf_options.comparator),
-      options_(*db_options,
-               SanitizeOptions(*db_options, &internal_comparator_, cf_options)),
-      ioptions_(options_),
-      mutable_cf_options_(options_, ioptions_),
+      initial_cf_options_(SanitizeOptions(db_options, cf_options)),
+      ioptions_(db_options, initial_cf_options_),
+      mutable_cf_options_(initial_cf_options_),
+      is_delete_range_supported_(strcmp(cf_options.table_factory->Name(),
+                                        BlockBasedTableFactory().Name()) == 0),
      write_buffer_manager_(write_buffer_manager),
      mem_(nullptr),
-      imm_(options_.min_write_buffer_number_to_merge,
-           options_.max_write_buffer_number_to_maintain),
+      imm_(ioptions_.min_write_buffer_number_to_merge,
+           ioptions_.max_write_buffer_number_to_maintain),
      super_version_(nullptr),
      super_version_number_(0),
      local_sv_(new ThreadLocalPtr(&SuperVersionUnrefHandle)),
@@ -364,16 +370,17 @@ ColumnFamilyData::ColumnFamilyData(
      column_family_set_(column_family_set),
      pending_flush_(false),
      pending_compaction_(false),
-      prev_compaction_needed_bytes_(0) {
+      prev_compaction_needed_bytes_(0),
+      allow_2pc_(db_options.allow_2pc) {
  Ref();

  // Convert user defined table properties collector factories to internal ones.
-  GetIntTblPropCollectorFactory(options_, &int_tbl_prop_collector_factories_);
+  GetIntTblPropCollectorFactory(ioptions_, &int_tbl_prop_collector_factories_);

  // if _dummy_versions is nullptr, then this is a dummy column family.
  if (_dummy_versions != nullptr) {
    internal_stats_.reset(
-        new InternalStats(ioptions_.num_levels, db_options->env, this));
+        new InternalStats(ioptions_.num_levels, db_options.env, this));
    table_cache_.reset(new TableCache(ioptions_, env_options, _table_cache));
    if (ioptions_.compaction_style == kCompactionStyleLevel) {
      compaction_picker_.reset(
@@ -405,7 +412,7 @@ ColumnFamilyData::ColumnFamilyData(
    if (column_family_set_->NumberOfColumnFamilies() < 10) {
      Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
          "--------------- Options for column family [%s]:\n", name.c_str());
-      options_.DumpCFOptions(ioptions_.info_log);
+      initial_cf_options_.Dump(ioptions_.info_log);
    } else {
      Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
          "\t(skipping printing options)\n");
@@ -482,15 +489,43 @@ void ColumnFamilyData::SetDropped() {
  column_family_set_->RemoveColumnFamily(this);
 }

-const double kSlowdownRatio = 1.2;
+ColumnFamilyOptions ColumnFamilyData::GetLatestCFOptions() const {
+  return BuildColumnFamilyOptions(initial_cf_options_, mutable_cf_options_);
+}
+
+uint64_t ColumnFamilyData::OldestLogToKeep() {
+  auto current_log = GetLogNumber();
+
+  if (allow_2pc_) {
+    auto imm_prep_log = imm()->GetMinLogContainingPrepSection();
+    auto mem_prep_log = mem()->GetMinLogContainingPrepSection();
+
+    if (imm_prep_log > 0 && imm_prep_log < current_log) {
+      current_log = imm_prep_log;
+    }
+
+    if (mem_prep_log > 0 && mem_prep_log < current_log) {
+      current_log = mem_prep_log;
+    }
+  }
+
+  return current_log;
+}
+
+const double kIncSlowdownRatio = 0.8;
+const double kDecSlowdownRatio = 1 / kIncSlowdownRatio;
+const double kNearStopSlowdownRatio = 0.6;
+const double kDelayRecoverSlowdownRatio = 1.4;

 namespace {
+// If penalize_stop is true, we further reduce slowdown rate.
 std::unique_ptr<WriteControllerToken> SetupDelay(
-    uint64_t max_write_rate, WriteController* write_controller,
-    uint64_t compaction_needed_bytes, uint64_t prev_compaction_neeed_bytes,
+    WriteController* write_controller, uint64_t compaction_needed_bytes,
+    uint64_t prev_compaction_need_bytes, bool penalize_stop,
    bool auto_comapctions_disabled) {
-  const uint64_t kMinWriteRate = 1024u;  // Minimum write rate 1KB/s.
+  const uint64_t kMinWriteRate = 16 * 1024u;  // Minimum write rate 16KB/s.

+  uint64_t max_write_rate = write_controller->max_delayed_write_rate();
  uint64_t write_rate = write_controller->delayed_write_rate();

  if (auto_comapctions_disabled) {
@@ -513,19 +548,32 @@ std::unique_ptr<WriteControllerToken> SetupDelay(
    // insert to mem tables, so we need to actively slow down before we get
    // feedback signal from compaction and flushes to avoid the full stop
    // because of hitting the max write buffer number.
-    if (prev_compaction_neeed_bytes > 0 &&
-        prev_compaction_neeed_bytes <= compaction_needed_bytes) {
-      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) /
-                                         kSlowdownRatio);
+    //
+    // If DB just falled into the stop condition, we need to further reduce
+    // the write rate to avoid the stop condition.
+    if (penalize_stop) {
+      // Penalize the near stop or stop condition by more agressive slowdown.
+      // This is to provide the long term slowdown increase signal.
+      // The penalty is more than the reward of recovering to the normal
+      // condition.
+      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) *
+                                         kNearStopSlowdownRatio);
      if (write_rate < kMinWriteRate) {
        write_rate = kMinWriteRate;
      }
-    } else if (prev_compaction_neeed_bytes > compaction_needed_bytes) {
+    } else if (prev_compaction_need_bytes > 0 &&
+               prev_compaction_need_bytes <= compaction_needed_bytes) {
+      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) *
+                                         kIncSlowdownRatio);
+      if (write_rate < kMinWriteRate) {
+        write_rate = kMinWriteRate;
+      }
+    } else if (prev_compaction_need_bytes > compaction_needed_bytes) {
      // We are speeding up by ratio of kSlowdownRatio when we have paid
      // compaction debt. But we'll never speed up to faster than the write rate
      // given by users.
      write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) *
-                                         kSlowdownRatio);
+                                         kDecSlowdownRatio);
      if (write_rate > max_write_rate) {
        write_rate = max_write_rate;
      }
@@ -539,14 +587,31 @@ int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
  // SanitizeOptions() ensures it.
  assert(level0_file_num_compaction_trigger <= level0_slowdown_writes_trigger);

+  if (level0_file_num_compaction_trigger < 0) {
+    return std::numeric_limits<int>::max();
+  }
+
+  const int64_t twice_level0_trigger =
+      static_cast<int64_t>(level0_file_num_compaction_trigger) * 2;
+
+  const int64_t one_fourth_trigger_slowdown =
+      static_cast<int64_t>(level0_file_num_compaction_trigger) +
+      ((level0_slowdown_writes_trigger - level0_file_num_compaction_trigger) /
+       4);
+
+  assert(twice_level0_trigger >= 0);
+  assert(one_fourth_trigger_slowdown >= 0);
+
  // 1/4 of the way between L0 compaction trigger threshold and slowdown
  // condition.
  // Or twice as compaction trigger, if it is smaller.
-  return std::min(level0_file_num_compaction_trigger * 2,
-                  level0_file_num_compaction_trigger +
-                      (level0_slowdown_writes_trigger -
-                       level0_file_num_compaction_trigger) /
-                          4);
+  int64_t res = std::min(twice_level0_trigger, one_fourth_trigger_slowdown);
+  if (res >= port::kMaxInt32) {
+    return port::kMaxInt32;
+  } else {
+    // res fits in int
+    return static_cast<int>(res);
+  }
 }
 }  // namespace

@@ -558,6 +623,9 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
    uint64_t compaction_needed_bytes =
        vstorage->estimated_compaction_needed_bytes();

+    bool was_stopped = write_controller->IsStopped();
+    bool needed_delay = write_controller->NeedsDelay();
+
    if (imm()->NumNotFlushed() >= mutable_cf_options.max_write_buffer_number) {
      write_controller_token_ = write_controller->GetStopToken();
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_COMPACTION, 1);
@@ -593,8 +661,8 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
               imm()->NumNotFlushed() >=
                   mutable_cf_options.max_write_buffer_number - 1) {
      write_controller_token_ =
-          SetupDelay(ioptions_.delayed_write_rate, write_controller,
-                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+          SetupDelay(write_controller, compaction_needed_bytes,
+                     prev_compaction_needed_bytes_, was_stopped,
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::MEMTABLE_SLOWDOWN, 1);
      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
@@ -608,9 +676,12 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
               mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
               vstorage->l0_delay_trigger_count() >=
                   mutable_cf_options.level0_slowdown_writes_trigger) {
+      // L0 is the last two files from stopping.
+      bool near_stop = vstorage->l0_delay_trigger_count() >=
+                       mutable_cf_options.level0_stop_writes_trigger - 2;
      write_controller_token_ =
-          SetupDelay(ioptions_.delayed_write_rate, write_controller,
-                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+          SetupDelay(write_controller, compaction_needed_bytes,
+                     prev_compaction_needed_bytes_, was_stopped || near_stop,
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(InternalStats::LEVEL0_SLOWDOWN_TOTAL, 1);
      if (compaction_picker_->IsLevel0CompactionInProgress()) {
@@ -626,9 +697,20 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
               mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
               vstorage->estimated_compaction_needed_bytes() >=
                   mutable_cf_options.soft_pending_compaction_bytes_limit) {
+      // If the distance to hard limit is less than 1/4 of the gap between soft
+      // and
+      // hard bytes limit, we think it is near stop and speed up the slowdown.
+      bool near_stop =
+          mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
+          (compaction_needed_bytes -
+           mutable_cf_options.soft_pending_compaction_bytes_limit) >
+              3 * (mutable_cf_options.hard_pending_compaction_bytes_limit -
+                   mutable_cf_options.soft_pending_compaction_bytes_limit) /
+                  4;
+
      write_controller_token_ =
-          SetupDelay(ioptions_.delayed_write_rate, write_controller,
-                     compaction_needed_bytes, prev_compaction_needed_bytes_,
+          SetupDelay(write_controller, compaction_needed_bytes,
+                     prev_compaction_needed_bytes_, was_stopped || near_stop,
                     mutable_cf_options.disable_auto_compactions);
      internal_stats_->AddCFStats(
          InternalStats::SOFT_PENDING_COMPACTION_BYTES_LIMIT, 1);
@@ -637,31 +719,43 @@ void ColumnFamilyData::RecalculateWriteStallConditions(
          "bytes %" PRIu64 " rate %" PRIu64,
          name_.c_str(), vstorage->estimated_compaction_needed_bytes(),
          write_controller->delayed_write_rate());
-    } else if (vstorage->l0_delay_trigger_count() >=
-               GetL0ThresholdSpeedupCompaction(
-                   mutable_cf_options.level0_file_num_compaction_trigger,
-                   mutable_cf_options.level0_slowdown_writes_trigger)) {
-      write_controller_token_ = write_controller->GetCompactionPressureToken();
-      Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
-          "[%s] Increasing compaction threads because we have %d level-0 "
-          "files ",
-          name_.c_str(), vstorage->l0_delay_trigger_count());
-    } else if (vstorage->estimated_compaction_needed_bytes() >=
-               mutable_cf_options.soft_pending_compaction_bytes_limit / 4) {
-      // Increase compaction threads if bytes needed for compaction exceeds
-      // 1/4 of threshold for slowing down.
-      // If soft pending compaction byte limit is not set, always speed up
-      // compaction.
-      write_controller_token_ = write_controller->GetCompactionPressureToken();
-      if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0) {
-        Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
-            "[%s] Increasing compaction threads because of estimated pending "
-            "compaction "
-            "bytes %" PRIu64,
-            name_.c_str(), vstorage->estimated_compaction_needed_bytes());
-      }
    } else {
-      write_controller_token_.reset();
+      if (vstorage->l0_delay_trigger_count() >=
+          GetL0ThresholdSpeedupCompaction(
+              mutable_cf_options.level0_file_num_compaction_trigger,
+              mutable_cf_options.level0_slowdown_writes_trigger)) {
+        write_controller_token_ =
+            write_controller->GetCompactionPressureToken();
+        Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
+            "[%s] Increasing compaction threads because we have %d level-0 "
+            "files ",
+            name_.c_str(), vstorage->l0_delay_trigger_count());
+      } else if (vstorage->estimated_compaction_needed_bytes() >=
+                 mutable_cf_options.soft_pending_compaction_bytes_limit / 4) {
+        // Increase compaction threads if bytes needed for compaction exceeds
+        // 1/4 of threshold for slowing down.
+        // If soft pending compaction byte limit is not set, always speed up
+        // compaction.
+        write_controller_token_ =
+            write_controller->GetCompactionPressureToken();
+        if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0) {
+          Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
+              "[%s] Increasing compaction threads because of estimated pending "
+              "compaction "
+              "bytes %" PRIu64,
+              name_.c_str(), vstorage->estimated_compaction_needed_bytes());
+        }
+      } else {
+        write_controller_token_.reset();
+      }
+      // If the DB recovers from delay conditions, we reward with reducing
+      // double the slowdown ratio. This is to balance the long term slowdown
+      // increase signal.
+      if (needed_delay) {
+        uint64_t write_rate = write_controller->delayed_write_rate();
+        write_controller->set_delayed_write_rate(static_cast<uint64_t>(
+            static_cast<double>(write_rate) * kDelayRecoverSlowdownRatio));
+      }
    }
    prev_compaction_needed_bytes_ = compaction_needed_bytes;
  }
@@ -713,6 +807,13 @@ Compaction* ColumnFamilyData::PickCompaction(
  return result;
 }

+bool ColumnFamilyData::RangeOverlapWithCompaction(
+    const Slice& smallest_user_key, const Slice& largest_user_key,
+    int level) const {
+  return compaction_picker_->RangeOverlapWithCompaction(
+      smallest_user_key, largest_user_key, level);
+}
+
 const int ColumnFamilyData::kCompactAllLevels = -1;
 const int ColumnFamilyData::kCompactToBaseLevel = -2;

@@ -863,14 +964,14 @@ Status ColumnFamilyData::SetOptions(
 #endif  // ROCKSDB_LITE

 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
-                                 const DBOptions* db_options,
+                                 const ImmutableDBOptions* db_options,
                                 const EnvOptions& env_options,
                                 Cache* table_cache,
                                 WriteBufferManager* write_buffer_manager,
                                 WriteController* write_controller)
    : max_column_family_(0),
      dummy_cfd_(new ColumnFamilyData(0, "", nullptr, nullptr, nullptr,
-                                      ColumnFamilyOptions(), db_options,
+                                      ColumnFamilyOptions(), *db_options,
                                      env_options, nullptr)),
      default_cfd_cache_(nullptr),
      db_name_(dbname),
@@ -942,7 +1043,7 @@ ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
  assert(column_families_.find(name) == column_families_.end());
  ColumnFamilyData* new_cfd = new ColumnFamilyData(
      id, name, dummy_versions, table_cache_, write_buffer_manager_, options,
-      db_options_, env_options_, this);
+      *db_options_, env_options_, this);
  column_families_.insert({name, id});
  column_family_data_.insert({id, new_cfd});
  max_column_family_ = std::max(max_column_family_, id);
@@ -42,7 +42,7 @@ class LogBuffer;
 class InstrumentedMutex;
 class InstrumentedMutexLock;

-extern const double kSlowdownRatio;
+extern const double kIncSlowdownRatio;

 // ColumnFamilyHandleImpl is the class that clients use to access different
 // column families. It has non-trivial destructor, which gets called when client
@@ -136,14 +136,13 @@ extern Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options);
 extern Status CheckConcurrentWritesSupported(
    const ColumnFamilyOptions& cf_options);

-extern ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
-                                           const InternalKeyComparator* icmp,
+extern ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                           const ColumnFamilyOptions& src);
 // Wrap user defined table proproties collector factories `from cf_options`
 // into internal ones in int_tbl_prop_collector_factories. Add a system internal
 // one too.
 extern void GetIntTblPropCollectorFactory(
-    const ColumnFamilyOptions& cf_options,
+    const ImmutableCFOptions& ioptions,
    std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
        int_tbl_prop_collector_factories);

@@ -201,9 +200,6 @@ class ColumnFamilyData {
  void SetLogNumber(uint64_t log_number) { log_number_ = log_number; }
  uint64_t GetLogNumber() const { return log_number_; }

-  // !!! To be deprecated! Please don't not use this function anymore!
-  const Options* options() const { return &options_; }
-
  // thread-safe
  const EnvOptions* soptions() const;
  const ImmutableCFOptions* ioptions() const { return &ioptions_; }
@@ -218,6 +214,14 @@ class ColumnFamilyData {
  const MutableCFOptions* GetLatestMutableCFOptions() const {
    return &mutable_cf_options_;
  }
+
+  // REQUIRES: DB mutex held
+  // Build ColumnFamiliesOptions with immutable options and latest mutable
+  // options.
+  ColumnFamilyOptions GetLatestCFOptions() const;
+
+  bool is_delete_range_supported() { return is_delete_range_supported_; }
+
 #ifndef ROCKSDB_LITE
  // REQUIRES: DB mutex held
  Status SetOptions(
@@ -235,6 +239,9 @@ class ColumnFamilyData {
  uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
  void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }

+  // calculate the oldest log needed for the durability of this column family
+  uint64_t OldestLogToKeep();
+
  // See Memtable constructor for explanation of earliest_seq param.
  MemTable* ConstructNewMemtable(const MutableCFOptions& mutable_cf_options,
                                 SequenceNumber earliest_seq);
@@ -249,6 +256,13 @@ class ColumnFamilyData {
  // REQUIRES: DB mutex held
  Compaction* PickCompaction(const MutableCFOptions& mutable_options,
                             LogBuffer* log_buffer);
+
+  // Check if the passed range overlap with any running compactions.
+  // REQUIRES: DB mutex held
+  bool RangeOverlapWithCompaction(const Slice& smallest_user_key,
+                                  const Slice& largest_user_key,
+                                  int level) const;
+
  // A flag to tell a manual compaction is to compact all levels together
  // instad of for specific level.
  static const int kCompactAllLevels;
@@ -325,7 +339,8 @@ class ColumnFamilyData {
                   Version* dummy_versions, Cache* table_cache,
                   WriteBufferManager* write_buffer_manager,
                   const ColumnFamilyOptions& options,
-                   const DBOptions* db_options, const EnvOptions& env_options,
+                   const ImmutableDBOptions& db_options,
+                   const EnvOptions& env_options,
                   ColumnFamilySet* column_family_set);

  uint32_t id_;
@@ -340,10 +355,12 @@ class ColumnFamilyData {
  std::vector<std::unique_ptr<IntTblPropCollectorFactory>>
      int_tbl_prop_collector_factories_;

-  const Options options_;
+  const ColumnFamilyOptions initial_cf_options_;
  const ImmutableCFOptions ioptions_;
  MutableCFOptions mutable_cf_options_;

+  const bool is_delete_range_supported_;
+
  std::unique_ptr<TableCache> table_cache_;

  std::unique_ptr<InternalStats> internal_stats_;
@@ -390,6 +407,9 @@ class ColumnFamilyData {
  bool pending_compaction_;

  uint64_t prev_compaction_needed_bytes_;
+
+  // if the database was opened with 2pc enabled
+  bool allow_2pc_;
 };

 // ColumnFamilySet has interesting thread-safety requirements
@@ -436,7 +456,8 @@ class ColumnFamilySet {
    ColumnFamilyData* current_;
  };

-  ColumnFamilySet(const std::string& dbname, const DBOptions* db_options,
+  ColumnFamilySet(const std::string& dbname,
+                  const ImmutableDBOptions* db_options,
                  const EnvOptions& env_options, Cache* table_cache,
                  WriteBufferManager* write_buffer_manager,
                  WriteController* write_controller);
@@ -493,7 +514,7 @@ class ColumnFamilySet {
  ColumnFamilyData* default_cfd_cache_;

  const std::string db_name_;
-  const DBOptions* const db_options_;
+  const ImmutableDBOptions* const db_options_;
  const EnvOptions env_options_;
  Cache* table_cache_;
  WriteBufferManager* write_buffer_manager_;
@@ -280,29 +280,23 @@ class ColumnFamilyTest : public testing::Test {
    }
  }

+#ifndef ROCKSDB_LITE  // TEST functions in DB are not supported in lite
  void WaitForFlush(int cf) {
-#ifndef ROCKSDB_LITE  // TEST functions are not supported in lite
    ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[cf]));
-#endif  // !ROCKSDB_LITE
  }

  void WaitForCompaction() {
-#ifndef ROCKSDB_LITE  // TEST functions are not supported in lite
    ASSERT_OK(dbfull()->TEST_WaitForCompact());
-#endif  // !ROCKSDB_LITE
  }

  uint64_t MaxTotalInMemoryState() {
-#ifndef ROCKSDB_LITE
    return dbfull()->TEST_MaxTotalInMemoryState();
-#else
-    return 0;
-#endif  // !ROCKSDB_LITE
  }

  void AssertMaxTotalInMemoryState(uint64_t value) {
    ASSERT_EQ(value, MaxTotalInMemoryState());
  }
+#endif  // !ROCKSDB_LITE

  Status Put(int cf, const std::string& key, const std::string& value) {
    return db_->Put(WriteOptions(), handles_[cf], Slice(key), Slice(value));
@@ -798,6 +792,7 @@ TEST_F(ColumnFamilyTest, IgnoreRecoveredLog) {
  }
 }

+#ifndef ROCKSDB_LITE  // TEST functions used are not supported
 TEST_F(ColumnFamilyTest, FlushTest) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});
@@ -912,6 +907,7 @@ TEST_F(ColumnFamilyTest, LogDeletionTest) {
  AssertCountLiveLogFiles(4);
  Close();
 }
+#endif  // !ROCKSDB_LITE

 TEST_F(ColumnFamilyTest, CrashAfterFlush) {
  std::unique_ptr<FaultInjectionTestEnv> fault_env(
@@ -945,6 +941,13 @@ TEST_F(ColumnFamilyTest, CrashAfterFlush) {
  db_options_.env = env_;
 }

+TEST_F(ColumnFamilyTest, OpenNonexistentColumnFamily) {
+  ASSERT_OK(TryOpen({"default"}));
+  Close();
+  ASSERT_TRUE(TryOpen({"default", "dne"}).IsInvalidArgument());
+}
+
+#ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 // Makes sure that obsolete log files get deleted
 TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
  // disable flushing stale column families
@@ -1049,9 +1052,11 @@ TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
  AssertCountLiveLogFiles(7);
  Close();
 }
+#endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // Cuckoo is not supported in lite
 TEST_F(ColumnFamilyTest, MemtableNotSupportSnapshot) {
+  db_options_.allow_concurrent_memtable_write = false;
  Open();
  auto* s1 = dbfull()->GetSnapshot();
  ASSERT_TRUE(s1 != nullptr);
@@ -1136,6 +1141,7 @@ TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
  Close();
 }

+#ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
  Open();
  CreateColumnFamilies({"one", "two"});
@@ -1204,6 +1210,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {

  Close();
 }
+#endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE
 // Sync points not supported in RocksDB Lite
@@ -2097,6 +2104,7 @@ TEST_F(ColumnFamilyTest, ReadOnlyDBTest) {
 }
 #endif  // !ROCKSDB_LITE

+#ifndef ROCKSDB_LITE  //  WaitForFlush() is not supported in lite
 TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
@@ -2118,7 +2126,9 @@ TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
  ASSERT_EQ(static_cast<size_t>(total_new_writable_files), handles_.size() + 1);
  Close();
 }
+#endif  // !ROCKSDB_LITE

+#ifndef ROCKSDB_LITE  //  WaitForCompaction() is not supported in lite
 TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
  Open();
  CreateColumnFamilies({"one", "two"});
@@ -2152,6 +2162,7 @@ TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
  ASSERT_EQ(0, dbfull()->TEST_total_log_size());
  Close();
 }
+#endif  // !ROCKSDB_LITE

 TEST_F(ColumnFamilyTest, CreateMissingColumnFamilies) {
  Status s = TryOpen({"one", "two"});
@@ -2179,7 +2190,7 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
                l * 4 * 1024 * 1024 + i * 1024 * 1024 + j * 1024 + k;

            ColumnFamilyOptions result =
-                SanitizeOptions(db_options, nullptr, original);
+                SanitizeOptions(ImmutableDBOptions(db_options), original);
            ASSERT_TRUE(result.level0_stop_writes_trigger >=
                        result.level0_slowdown_writes_trigger);
            ASSERT_TRUE(result.level0_slowdown_writes_trigger >=
@@ -2429,7 +2440,7 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
 #endif  // !ROCKSDB_LITE

 TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
-  const uint64_t kBaseRate = 810000u;
+  const uint64_t kBaseRate = 800000u;
  db_options_.delayed_write_rate = kBaseRate;
  db_options_.base_background_compactions = 2;
  db_options_.max_background_compactions = 6;
@@ -2440,9 +2451,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {

  VersionStorageInfo* vstorage = cfd->current()->storage_info();

-  MutableCFOptions mutable_cf_options(
-      Options(db_options_, column_family_options_),
-      ImmutableCFOptions(Options(db_options_, column_family_options_)));
+  MutableCFOptions mutable_cf_options(column_family_options_);

  mutable_cf_options.level0_slowdown_writes_trigger = 20;
  mutable_cf_options.level0_stop_writes_trigger = 10000;
@@ -2460,28 +2469,28 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(400);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(500);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(450);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(205);
@@ -2517,14 +2526,14 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(2001);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(3001);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2535,7 +2544,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(100);
@@ -2547,15 +2556,14 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
-            dbfull()->TEST_write_controler().delayed_write_rate());
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(101);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->set_l0_delay_trigger_count(0);
@@ -2563,21 +2571,21 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->set_l0_delay_trigger_count(101);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2 / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(200);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->set_l0_delay_trigger_count(0);
@@ -2618,7 +2626,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());
 }

@@ -2631,9 +2639,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

  VersionStorageInfo* vstorage = cfd->current()->storage_info();

-  MutableCFOptions mutable_cf_options(
-      Options(db_options_, column_family_options_),
-      ImmutableCFOptions(Options(db_options_, column_family_options_)));
+  MutableCFOptions mutable_cf_options(column_family_options_);

  // Speed up threshold = min(4 * 2, 4 + (36 - 4)/4) = 8
  mutable_cf_options.level0_file_num_compaction_trigger = 4;
@@ -2645,31 +2651,31 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

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

  vstorage->TEST_set_estimated_compaction_needed_bytes(50);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(300);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

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

  vstorage->set_l0_delay_trigger_count(9);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

  // Speed up threshold = min(4 * 2, 4 + (12 - 4)/4) = 6
  mutable_cf_options.level0_file_num_compaction_trigger = 4;
@@ -2678,15 +2684,15 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {

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

  vstorage->set_l0_delay_trigger_count(7);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

 TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
@@ -2702,9 +2708,7 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
      static_cast<ColumnFamilyHandleImpl*>(handles_[1])->cfd();
  VersionStorageInfo* vstorage1 = cfd1->current()->storage_info();

-  MutableCFOptions mutable_cf_options(
-      Options(db_options_, column_family_options_),
-      ImmutableCFOptions(Options(db_options_, column_family_options_)));
+  MutableCFOptions mutable_cf_options(column_family_options_);
  mutable_cf_options.level0_slowdown_writes_trigger = 20;
  mutable_cf_options.level0_stop_writes_trigger = 10000;
  mutable_cf_options.soft_pending_compaction_bytes_limit = 200;
@@ -2739,35 +2743,35 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(300);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(700);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(500);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2 / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(600);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.2,
+  ASSERT_EQ(kBaseRate / 1.25,
            dbfull()->TEST_write_controler().delayed_write_rate());
 }

@@ -2786,9 +2790,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
      static_cast<ColumnFamilyHandleImpl*>(handles_[1])->cfd();
  VersionStorageInfo* vstorage1 = cfd1->current()->storage_info();

-  MutableCFOptions mutable_cf_options(
-      Options(db_options_, column_family_options_),
-      ImmutableCFOptions(Options(db_options_, column_family_options_)));
+  MutableCFOptions mutable_cf_options(column_family_options_);
  // Speed up threshold = min(4 * 2, 4 + (36 - 4)/4) = 8
  mutable_cf_options.level0_file_num_compaction_trigger = 4;
  mutable_cf_options.level0_slowdown_writes_trigger = 36;
@@ -2802,41 +2804,41 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {

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

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

  vstorage1->TEST_set_estimated_compaction_needed_bytes(30);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(70);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(20);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

  vstorage->set_l0_delay_trigger_count(9);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage1->set_l0_delay_trigger_count(2);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_EQ(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

 #ifndef ROCKSDB_LITE
@@ -2881,6 +2883,7 @@ TEST_F(ColumnFamilyTest, FlushCloseWALFiles) {
 }
 #endif  // !ROCKSDB_LITE

+#ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 TEST_F(ColumnFamilyTest, IteratorCloseWALFile1) {
  SpecialEnv env(Env::Default());
  db_options_.env = &env;
@@ -2980,6 +2983,7 @@ TEST_F(ColumnFamilyTest, IteratorCloseWALFile2) {
  db_options_.env = env_;
  Close();
 }
+#endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // TEST functions are not supported in lite
 TEST_F(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
@@ -50,6 +50,10 @@ class FlushedFileCollector : public EventListener {
    }
    return result;
  }
+  void ClearFlushedFiles() {
+    std::lock_guard<std::mutex> lock(mutex_);
+    flushed_files_.clear();
+  }

 private:
  std::vector<std::string> flushed_files_;
@@ -116,13 +120,12 @@ TEST_F(CompactFilesTest, L0ConflictsFiles) {
 TEST_F(CompactFilesTest, ObsoleteFiles) {
  Options options;
  // to trigger compaction more easily
-  const int kWriteBufferSize = 10000;
+  const int kWriteBufferSize = 65536;
  options.create_if_missing = true;
  // Disable RocksDB background compaction.
  options.compaction_style = kCompactionStyleNone;
-  // Small slowdown and stop trigger for experimental purpose.
-  options.level0_slowdown_writes_trigger = 20;
-  options.level0_stop_writes_trigger = 20;
+  options.level0_slowdown_writes_trigger = (1 << 30);
+  options.level0_stop_writes_trigger = (1 << 30);
  options.write_buffer_size = kWriteBufferSize;
  options.max_write_buffer_number = 2;
  options.compression = kNoCompression;
@@ -154,6 +157,46 @@ TEST_F(CompactFilesTest, ObsoleteFiles) {
  delete db;
 }

+TEST_F(CompactFilesTest, NotCutOutputOnLevel0) {
+  Options options;
+  options.create_if_missing = true;
+  // Disable RocksDB background compaction.
+  options.compaction_style = kCompactionStyleNone;
+  options.level0_slowdown_writes_trigger = 1000;
+  options.level0_stop_writes_trigger = 1000;
+  options.write_buffer_size = 65536;
+  options.max_write_buffer_number = 2;
+  options.compression = kNoCompression;
+  options.max_compaction_bytes = 5000;
+
+  // Add listener
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+
+  DB* db = nullptr;
+  DestroyDB(db_name_, options);
+  Status s = DB::Open(options, db_name_, &db);
+  assert(s.ok());
+  assert(db);
+
+  // create couple files
+  for (int i = 0; i < 500; ++i) {
+    db->Put(WriteOptions(), ToString(i), std::string(1000, 'a' + (i % 26)));
+  }
+  reinterpret_cast<DBImpl*>(db)->TEST_WaitForFlushMemTable();
+  auto l0_files_1 = collector->GetFlushedFiles();
+  collector->ClearFlushedFiles();
+  for (int i = 0; i < 500; ++i) {
+    db->Put(WriteOptions(), ToString(i), std::string(1000, 'a' + (i % 26)));
+  }
+  reinterpret_cast<DBImpl*>(db)->TEST_WaitForFlushMemTable();
+  auto l0_files_2 = collector->GetFlushedFiles();
+  ASSERT_OK(db->CompactFiles(CompactionOptions(), l0_files_1, 0));
+  ASSERT_OK(db->CompactFiles(CompactionOptions(), l0_files_2, 0));
+  // no assertion failure
+  delete db;
+}
+
 TEST_F(CompactFilesTest, CapturingPendingFiles) {
  Options options;
  options.create_if_missing = true;
@@ -46,7 +46,7 @@ Status CompactedDBImpl::Get(const ReadOptions& options,
     ColumnFamilyHandle*, const Slice& key, std::string* value) {
  GetContext get_context(user_comparator_, nullptr, nullptr, nullptr,
                         GetContext::kNotFound, key, value, nullptr, nullptr,
-                         nullptr);
+                         nullptr, nullptr);
  LookupKey lkey(key, kMaxSequenceNumber);
  files_.files[FindFile(key)].fd.table_reader->Get(
      options, lkey.internal_key(), &get_context);
@@ -77,7 +77,7 @@ std::vector<Status> CompactedDBImpl::MultiGet(const ReadOptions& options,
    if (r != nullptr) {
      GetContext get_context(user_comparator_, nullptr, nullptr, nullptr,
                             GetContext::kNotFound, keys[idx], &(*values)[idx],
-                             nullptr, nullptr, nullptr);
+                             nullptr, nullptr, nullptr, nullptr);
      LookupKey lkey(keys[idx], kMaxSequenceNumber);
      r->Get(options, lkey.internal_key(), &get_context);
      if (get_context.State() == GetContext::kFound) {
@@ -151,9 +151,9 @@ Status CompactedDBImpl::Open(const Options& options,
  std::unique_ptr<CompactedDBImpl> db(new CompactedDBImpl(db_options, dbname));
  Status s = db->Init(options);
  if (s.ok()) {
-    Log(INFO_LEVEL, db->db_options_.info_log,
+    Log(INFO_LEVEL, db->immutable_db_options_.info_log,
        "Opened the db as fully compacted mode");
-    LogFlush(db->db_options_.info_log);
+    LogFlush(db->immutable_db_options_.info_log);
    *dbptr = db.release();
  }
  return s;
@@ -140,6 +140,7 @@ bool Compaction::IsFullCompaction(
 }

 Compaction::Compaction(VersionStorageInfo* vstorage,
+                       const ImmutableCFOptions& _immutable_cf_options,
                       const MutableCFOptions& _mutable_cf_options,
                       std::vector<CompactionInputFiles> _inputs,
                       int _output_level, uint64_t _target_file_size,
@@ -153,6 +154,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
      output_level_(_output_level),
      max_output_file_size_(_target_file_size),
      max_compaction_bytes_(_max_compaction_bytes),
+      immutable_cf_options_(_immutable_cf_options),
      mutable_cf_options_(_mutable_cf_options),
      input_version_(nullptr),
      number_levels_(vstorage->num_levels()),
@@ -427,7 +429,7 @@ bool Compaction::IsOutputLevelEmpty() const {
 }

 bool Compaction::ShouldFormSubcompactions() const {
-  if (mutable_cf_options_.max_subcompactions <= 1 || cfd_ == nullptr) {
+  if (immutable_cf_options_.max_subcompactions <= 1 || cfd_ == nullptr) {
    return false;
  }
  if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel) {
@@ -35,6 +35,7 @@ class CompactionFilter;
 class Compaction {
 public:
  Compaction(VersionStorageInfo* input_version,
+             const ImmutableCFOptions& immutable_cf_options,
             const MutableCFOptions& mutable_cf_options,
             std::vector<CompactionInputFiles> inputs, int output_level,
             uint64_t target_file_size, uint64_t max_compaction_bytes,
@@ -103,6 +104,8 @@ class Compaction {
    return &inputs_[compaction_input_level].files;
  }

+  const std::vector<CompactionInputFiles>* inputs() { return &inputs_; }
+
  // Returns the LevelFilesBrief of the specified compaction input level.
  const LevelFilesBrief* input_levels(size_t compaction_input_level) const {
    return &input_levels_[compaction_input_level];
@@ -171,6 +174,12 @@ class Compaction {
  // How many total levels are there?
  int number_levels() const { return number_levels_; }

+  // Return the ImmutableCFOptions that should be used throughout the compaction
+  // procedure
+  const ImmutableCFOptions* immutable_cf_options() const {
+    return &immutable_cf_options_;
+  }
+
  // Return the MutableCFOptions that should be used throughout the compaction
  // procedure
  const MutableCFOptions* mutable_cf_options() const {
@@ -253,7 +262,8 @@ class Compaction {
  const int output_level_;  // levels to which output files are stored
  uint64_t max_output_file_size_;
  uint64_t max_compaction_bytes_;
-  MutableCFOptions mutable_cf_options_;
+  const ImmutableCFOptions immutable_cf_options_;
+  const MutableCFOptions mutable_cf_options_;
  Version* input_version_;
  VersionEdit edit_;
  const int number_levels_;
@@ -0,0 +1,33 @@
+//  Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#pragma once
+
+struct CompactionIterationStats {
+  // Compaction statistics
+
+  // Doesn't include records skipped because of
+  // CompactionFilter::Decision::kRemoveAndSkipUntil.
+  int64_t num_record_drop_user = 0;
+
+  int64_t num_record_drop_hidden = 0;
+  int64_t num_record_drop_obsolete = 0;
+  int64_t num_record_drop_range_del = 0;
+  int64_t num_range_del_drop_obsolete = 0;
+  uint64_t total_filter_time = 0;
+
+  // Input statistics
+  // TODO(noetzli): The stats are incomplete. They are lacking everything
+  // consumed by MergeHelper.
+  uint64_t num_input_records = 0;
+  uint64_t num_input_deletion_records = 0;
+  uint64_t num_input_corrupt_records = 0;
+  uint64_t total_input_raw_key_bytes = 0;
+  uint64_t total_input_raw_value_bytes = 0;
+
+  // Single-Delete diagnostics for exceptional situations
+  uint64_t num_single_del_fallthru = 0;
+  uint64_t num_single_del_mismatch = 0;
+};
@@ -14,7 +14,23 @@ CompactionIterator::CompactionIterator(
    InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
    SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
    SequenceNumber earliest_write_conflict_snapshot, Env* env,
-    bool expect_valid_internal_key, const Compaction* compaction,
+    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    const Compaction* compaction, const CompactionFilter* compaction_filter,
+    LogBuffer* log_buffer)
+    : CompactionIterator(
+          input, cmp, merge_helper, last_sequence, snapshots,
+          earliest_write_conflict_snapshot, env, expect_valid_internal_key,
+          range_del_agg,
+          std::unique_ptr<CompactionProxy>(
+              compaction ? new CompactionProxy(compaction) : nullptr),
+          compaction_filter, log_buffer) {}
+
+CompactionIterator::CompactionIterator(
+    InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
+    SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
+    SequenceNumber earliest_write_conflict_snapshot, Env* env,
+    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    std::unique_ptr<CompactionProxy> compaction,
    const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
    : input_(input),
      cmp_(cmp),
@@ -23,7 +39,8 @@ CompactionIterator::CompactionIterator(
      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      env_(env),
      expect_valid_internal_key_(expect_valid_internal_key),
-      compaction_(compaction),
+      range_del_agg_(range_del_agg),
+      compaction_(std::move(compaction)),
      compaction_filter_(compaction_filter),
      log_buffer_(log_buffer),
      merge_out_iter_(merge_helper_) {
@@ -36,11 +53,11 @@ CompactionIterator::CompactionIterator(

  if (snapshots_->size() == 0) {
    // optimize for fast path if there are no snapshots
-    visible_at_tip_ = last_sequence;
-    earliest_snapshot_ = visible_at_tip_;
+    visible_at_tip_ = true;
+    earliest_snapshot_ = last_sequence;
    latest_snapshot_ = 0;
  } else {
-    visible_at_tip_ = 0;
+    visible_at_tip_ = false;
    earliest_snapshot_ = snapshots_->at(0);
    latest_snapshot_ = snapshots_->back();
  }
@@ -61,6 +78,8 @@ void CompactionIterator::ResetRecordCounts() {
  iter_stats_.num_record_drop_user = 0;
  iter_stats_.num_record_drop_hidden = 0;
  iter_stats_.num_record_drop_obsolete = 0;
+  iter_stats_.num_record_drop_range_del = 0;
+  iter_stats_.num_range_del_drop_obsolete = 0;
 }

 void CompactionIterator::SeekToFirst() {
@@ -106,7 +125,7 @@ void CompactionIterator::Next() {
  }

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

@@ -147,12 +166,20 @@ void CompactionIterator::NextFromInput() {
    iter_stats_.total_input_raw_key_bytes += key_.size();
    iter_stats_.total_input_raw_value_bytes += value_.size();

+    // If need_skip is true, we should seek the input iterator
+    // to internal key skip_until and continue from there.
+    bool need_skip = false;
+    // Points either into compaction_filter_skip_until_ or into
+    // merge_helper_->compaction_filter_skip_until_.
+    Slice skip_until;
+
    // Check whether the user key changed. After this if statement current_key_
    // is a copy of the current input key (maybe converted to a delete by the
    // compaction filter). ikey_.user_key is pointing to the copy.
    if (!has_current_user_key_ ||
        !cmp_->Equal(ikey_.user_key, current_user_key_)) {
      // First occurrence of this user key
+      // Copy key for output
      key_ = current_key_.SetKey(key_, &ikey_);
      current_user_key_ = ikey_.user_key;
      has_current_user_key_ = true;
@@ -168,26 +195,41 @@ void CompactionIterator::NextFromInput() {
        // number is greater than any external snapshot, then invoke the
        // filter. If the return value of the compaction filter is true,
        // replace the entry with a deletion marker.
-        bool value_changed = false;
-        bool to_delete = false;
+        CompactionFilter::Decision filter;
        compaction_filter_value_.clear();
+        compaction_filter_skip_until_.Clear();
        {
          StopWatchNano timer(env_, true);
-          to_delete = compaction_filter_->Filter(
-              compaction_->level(), ikey_.user_key, value_,
-              &compaction_filter_value_, &value_changed);
+          filter = compaction_filter_->FilterV2(
+              compaction_->level(), ikey_.user_key,
+              CompactionFilter::ValueType::kValue, value_,
+              &compaction_filter_value_, compaction_filter_skip_until_.rep());
          iter_stats_.total_filter_time +=
              env_ != nullptr ? timer.ElapsedNanos() : 0;
        }
-        if (to_delete) {
+
+        if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil &&
+            cmp_->Compare(*compaction_filter_skip_until_.rep(),
+                          ikey_.user_key) <= 0) {
+          // Can't skip to a key smaller than the current one.
+          // Keep the key as per FilterV2 documentation.
+          filter = CompactionFilter::Decision::kKeep;
+        }
+
+        if (filter == CompactionFilter::Decision::kRemove) {
          // convert the current key to a delete
          ikey_.type = kTypeDeletion;
          current_key_.UpdateInternalKey(ikey_.sequence, kTypeDeletion);
          // no value associated with delete
          value_.clear();
          iter_stats_.num_record_drop_user++;
-        } else if (value_changed) {
+        } else if (filter == CompactionFilter::Decision::kChangeValue) {
          value_ = compaction_filter_value_;
+        } else if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil) {
+          need_skip = true;
+          compaction_filter_skip_until_.ConvertFromUserKey(kMaxSequenceNumber,
+                                                           kValueTypeForSeek);
+          skip_until = compaction_filter_skip_until_.Encode();
        }
      }
    } else {
@@ -210,10 +252,13 @@ void CompactionIterator::NextFromInput() {
    SequenceNumber last_snapshot = current_user_key_snapshot_;
    SequenceNumber prev_snapshot = 0;  // 0 means no previous snapshot
    current_user_key_snapshot_ =
-        visible_at_tip_ ? visible_at_tip_ : findEarliestVisibleSnapshot(
-                                                ikey_.sequence, &prev_snapshot);
+        visible_at_tip_
+            ? earliest_snapshot_
+            : findEarliestVisibleSnapshot(ikey_.sequence, &prev_snapshot);

-    if (clear_and_output_next_key_) {
+    if (need_skip) {
+      // This case is handled below.
+    } else if (clear_and_output_next_key_) {
      // In the previous iteration we encountered a single delete that we could
      // not compact out.  We will keep this Put, but can drop it's data.
      // (See Optimization 3, below.)
@@ -304,7 +349,6 @@ void CompactionIterator::NextFromInput() {
            // to know that a write happened in this snapshot (Rule 2 above).
            // Clear the value and output the SingleDelete. (The value will be
            // outputted on the next iteration.)
-            ++iter_stats_.num_record_drop_hidden;

            // Setting valid_ to true will output the current SingleDelete
            valid_ = true;
@@ -320,7 +364,7 @@ void CompactionIterator::NextFromInput() {
        }
      } else {
        // We are at the end of the input, could not parse the next key, or hit
-        // the next key. The iterator returns the single delete if the key
+        // a different key. The iterator returns the single delete if the key
        // possibly exists beyond the current output level.  We set
        // has_current_user_key to false so that if the iterator is at the next
        // key, we do not compare it again against the previous key at the next
@@ -389,7 +433,8 @@ void CompactionIterator::NextFromInput() {
      // have hit (A)
      // We encapsulate the merge related state machine in a different
      // object to minimize change to the existing flow.
-      merge_helper_->MergeUntil(input_, prev_snapshot, bottommost_level_);
+      merge_helper_->MergeUntil(input_, range_del_agg_, prev_snapshot,
+                                bottommost_level_);
      merge_out_iter_.SeekToFirst();

      if (merge_out_iter_.Valid()) {
@@ -413,9 +458,26 @@ void CompactionIterator::NextFromInput() {
        // coming after the merges
        has_current_user_key_ = false;
        pinned_iters_mgr_.ReleasePinnedData();
+
+        if (merge_helper_->FilteredUntil(&skip_until)) {
+          need_skip = true;
+        }
      }
    } else {
-      valid_ = true;
+      // 1. new user key -OR-
+      // 2. different snapshot stripe
+      bool should_delete = range_del_agg_->ShouldDelete(key_);
+      if (should_delete) {
+        ++iter_stats_.num_record_drop_hidden;
+        ++iter_stats_.num_record_drop_range_del;
+        input_->Next();
+      } else {
+        valid_ = true;
+      }
+    }
+
+    if (need_skip) {
+      input_->Seek(skip_until);
    }
  }
 }
@@ -12,45 +12,68 @@
 #include <vector>

 #include "db/compaction.h"
+#include "db/compaction_iteration_stats.h"
 #include "db/merge_helper.h"
 #include "db/pinned_iterators_manager.h"
+#include "db/range_del_aggregator.h"
 #include "rocksdb/compaction_filter.h"
 #include "util/log_buffer.h"

 namespace rocksdb {

-struct CompactionIteratorStats {
-  // Compaction statistics
-  int64_t num_record_drop_user = 0;
-  int64_t num_record_drop_hidden = 0;
-  int64_t num_record_drop_obsolete = 0;
-  uint64_t total_filter_time = 0;
-
-  // Input statistics
-  // TODO(noetzli): The stats are incomplete. They are lacking everything
-  // consumed by MergeHelper.
-  uint64_t num_input_records = 0;
-  uint64_t num_input_deletion_records = 0;
-  uint64_t num_input_corrupt_records = 0;
-  uint64_t total_input_raw_key_bytes = 0;
-  uint64_t total_input_raw_value_bytes = 0;
-
-  // Single-Delete diagnostics for exceptional situations
-  uint64_t num_single_del_fallthru = 0;
-  uint64_t num_single_del_mismatch = 0;
-};
-
 class CompactionIterator {
 public:
+  // A wrapper around Compaction. Has a much smaller interface, only what
+  // CompactionIterator uses. Tests can override it.
+  class CompactionProxy {
+   public:
+    explicit CompactionProxy(const Compaction* compaction)
+        : compaction_(compaction) {}
+
+    virtual ~CompactionProxy() = default;
+    virtual int level(size_t compaction_input_level = 0) const {
+      return compaction_->level();
+    }
+    virtual bool KeyNotExistsBeyondOutputLevel(
+        const Slice& user_key, std::vector<size_t>* level_ptrs) const {
+      return compaction_->KeyNotExistsBeyondOutputLevel(user_key, level_ptrs);
+    }
+    virtual bool bottommost_level() const {
+      return compaction_->bottommost_level();
+    }
+    virtual int number_levels() const { return compaction_->number_levels(); }
+    virtual Slice GetLargestUserKey() const {
+      return compaction_->GetLargestUserKey();
+    }
+
+   protected:
+    CompactionProxy() = default;
+
+   private:
+    const Compaction* compaction_;
+  };
+
  CompactionIterator(InternalIterator* input, const Comparator* cmp,
                     MergeHelper* merge_helper, SequenceNumber last_sequence,
                     std::vector<SequenceNumber>* snapshots,
                     SequenceNumber earliest_write_conflict_snapshot, Env* env,
                     bool expect_valid_internal_key,
+                     RangeDelAggregator* range_del_agg,
                     const Compaction* compaction = nullptr,
                     const CompactionFilter* compaction_filter = nullptr,
                     LogBuffer* log_buffer = nullptr);

+  // Constructor with custom CompactionProxy, used for tests.
+  CompactionIterator(InternalIterator* input, const Comparator* cmp,
+                     MergeHelper* merge_helper, SequenceNumber last_sequence,
+                     std::vector<SequenceNumber>* snapshots,
+                     SequenceNumber earliest_write_conflict_snapshot, Env* env,
+                     bool expect_valid_internal_key,
+                     RangeDelAggregator* range_del_agg,
+                     std::unique_ptr<CompactionProxy> compaction,
+                     const CompactionFilter* compaction_filter = nullptr,
+                     LogBuffer* log_buffer = nullptr);
+
  ~CompactionIterator();

  void ResetRecordCounts();
@@ -72,7 +95,7 @@ class CompactionIterator {
  const ParsedInternalKey& ikey() const { return ikey_; }
  bool Valid() const { return valid_; }
  const Slice& user_key() const { return current_user_key_; }
-  const CompactionIteratorStats& iter_stats() const { return iter_stats_; }
+  const CompactionIterationStats& iter_stats() const { return iter_stats_; }

 private:
  // Processes the input stream to find the next output
@@ -99,12 +122,13 @@ class CompactionIterator {
  const SequenceNumber earliest_write_conflict_snapshot_;
  Env* env_;
  bool expect_valid_internal_key_;
-  const Compaction* compaction_;
+  RangeDelAggregator* range_del_agg_;
+  std::unique_ptr<CompactionProxy> compaction_;
  const CompactionFilter* compaction_filter_;
  LogBuffer* log_buffer_;
  bool bottommost_level_;
  bool valid_ = false;
-  SequenceNumber visible_at_tip_;
+  bool visible_at_tip_;
  SequenceNumber earliest_snapshot_;
  SequenceNumber latest_snapshot_;
  bool ignore_snapshots_;
@@ -147,6 +171,7 @@ class CompactionIterator {
  // merge operands and then releasing them after consuming them.
  PinnedIteratorsManager pinned_iters_mgr_;
  std::string compaction_filter_value_;
+  InternalKey compaction_filter_skip_until_;
  // "level_ptrs" holds indices that remember which file of an associated
  // level we were last checking during the last call to compaction->
  // KeyNotExistsBeyondOutputLevel(). This allows future calls to the function
@@ -154,6 +179,6 @@ class CompactionIterator {
  // increasing so a later call to the function must be looking for a key that
  // is in or beyond the last file checked during the previous call
  std::vector<size_t> level_ptrs_;
-  CompactionIteratorStats iter_stats_;
+  CompactionIterationStats iter_stats_;
 };
 }  // namespace rocksdb
@@ -4,40 +4,148 @@
 //  of patent rights can be found in the PATENTS file in the same directory.

 #include "db/compaction_iterator.h"
+
+#include <string>
+#include <vector>
+
 #include "util/testharness.h"
 #include "util/testutil.h"

 namespace rocksdb {

+class LoggingForwardVectorIterator : public InternalIterator {
+ public:
+  struct Action {
+    enum class Type {
+      SEEK_TO_FIRST,
+      SEEK,
+      NEXT,
+    };
+
+    Type type;
+    std::string arg;
+
+    explicit Action(Type _type, std::string _arg = "")
+        : type(_type), arg(_arg) {}
+
+    bool operator==(const Action& rhs) const {
+      return std::tie(type, arg) == std::tie(rhs.type, rhs.arg);
+    }
+  };
+
+  LoggingForwardVectorIterator(const std::vector<std::string>& keys,
+                               const std::vector<std::string>& values)
+      : keys_(keys), values_(values), current_(keys.size()) {
+    assert(keys_.size() == values_.size());
+  }
+
+  virtual bool Valid() const override { return current_ < keys_.size(); }
+
+  virtual void SeekToFirst() override {
+    log.emplace_back(Action::Type::SEEK_TO_FIRST);
+    current_ = 0;
+  }
+  virtual void SeekToLast() override { assert(false); }
+
+  virtual void Seek(const Slice& target) override {
+    log.emplace_back(Action::Type::SEEK, target.ToString());
+    current_ = std::lower_bound(keys_.begin(), keys_.end(), target.ToString()) -
+               keys_.begin();
+  }
+
+  virtual void SeekForPrev(const Slice& target) override { assert(false); }
+
+  virtual void Next() override {
+    assert(Valid());
+    log.emplace_back(Action::Type::NEXT);
+    current_++;
+  }
+  virtual void Prev() override { assert(false); }
+
+  virtual Slice key() const override {
+    assert(Valid());
+    return Slice(keys_[current_]);
+  }
+  virtual Slice value() const override {
+    assert(Valid());
+    return Slice(values_[current_]);
+  }
+
+  virtual Status status() const override { return Status::OK(); }
+
+  std::vector<Action> log;
+
+ private:
+  std::vector<std::string> keys_;
+  std::vector<std::string> values_;
+  size_t current_;
+};
+
+class FakeCompaction : public CompactionIterator::CompactionProxy {
+ public:
+  FakeCompaction() = default;
+
+  virtual int level(size_t compaction_input_level) const { return 0; }
+  virtual bool KeyNotExistsBeyondOutputLevel(
+      const Slice& user_key, std::vector<size_t>* level_ptrs) const {
+    return false;
+  }
+  virtual bool bottommost_level() const { return false; }
+  virtual int number_levels() const { return 1; }
+  virtual Slice GetLargestUserKey() const {
+    return "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
+  }
+};
+
 class CompactionIteratorTest : public testing::Test {
 public:
-  CompactionIteratorTest() : cmp_(BytewiseComparator()), snapshots_({}) {}
+  CompactionIteratorTest()
+      : cmp_(BytewiseComparator()), icmp_(cmp_), snapshots_({}) {}

-  void InitIterator(const std::vector<std::string>& ks,
-                    const std::vector<std::string>& vs,
-                    SequenceNumber last_sequence) {
-    merge_helper_.reset(new MergeHelper(Env::Default(), cmp_, nullptr, nullptr,
+  void InitIterators(const std::vector<std::string>& ks,
+                     const std::vector<std::string>& vs,
+                     const std::vector<std::string>& range_del_ks,
+                     const std::vector<std::string>& range_del_vs,
+                     SequenceNumber last_sequence,
+                     MergeOperator* merge_op = nullptr,
+                     CompactionFilter* filter = nullptr) {
+    std::unique_ptr<InternalIterator> range_del_iter(
+        new test::VectorIterator(range_del_ks, range_del_vs));
+    range_del_agg_.reset(new RangeDelAggregator(icmp_, snapshots_));
+    ASSERT_OK(range_del_agg_->AddTombstones(std::move(range_del_iter)));
+
+    std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
+    if (filter) {
+      compaction.reset(new FakeCompaction());
+    }
+
+    merge_helper_.reset(new MergeHelper(Env::Default(), cmp_, merge_op, filter,
                                        nullptr, 0U, false, 0));
-    iter_.reset(new test::VectorIterator(ks, vs));
+    iter_.reset(new LoggingForwardVectorIterator(ks, vs));
    iter_->SeekToFirst();
    c_iter_.reset(new CompactionIterator(
        iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
-        kMaxSequenceNumber, Env::Default(), false));
+        kMaxSequenceNumber, Env::Default(), false, range_del_agg_.get(),
+        std::move(compaction), filter));
  }

+  void AddSnapshot(SequenceNumber snapshot) { snapshots_.push_back(snapshot); }
+
  const Comparator* cmp_;
+  const InternalKeyComparator icmp_;
  std::vector<SequenceNumber> snapshots_;
  std::unique_ptr<MergeHelper> merge_helper_;
-  std::unique_ptr<test::VectorIterator> iter_;
+  std::unique_ptr<LoggingForwardVectorIterator> iter_;
  std::unique_ptr<CompactionIterator> c_iter_;
+  std::unique_ptr<RangeDelAggregator> range_del_agg_;
 };

 // It is possible that the output of the compaction iterator is empty even if
 // the input is not.
 TEST_F(CompactionIteratorTest, EmptyResult) {
-  InitIterator({test::KeyStr("a", 5, kTypeSingleDeletion),
-                test::KeyStr("a", 3, kTypeValue)},
-               {"", "val"}, 5);
+  InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
+                 test::KeyStr("a", 3, kTypeValue)},
+                {"", "val"}, {}, {}, 5);
  c_iter_->SeekToFirst();
  ASSERT_FALSE(c_iter_->Valid());
 }
@@ -45,10 +153,10 @@ TEST_F(CompactionIteratorTest, EmptyResult) {
 // If there is a corruption after a single deletion, the corrupted key should
 // be preserved.
 TEST_F(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
-  InitIterator({test::KeyStr("a", 5, kTypeSingleDeletion),
-                test::KeyStr("a", 3, kTypeValue, true),
-                test::KeyStr("b", 10, kTypeValue)},
-               {"", "val", "val2"}, 10);
+  InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
+                 test::KeyStr("a", 3, kTypeValue, true),
+                 test::KeyStr("b", 10, kTypeValue)},
+                {"", "val", "val2"}, {}, {}, 10);
  c_iter_->SeekToFirst();
  ASSERT_TRUE(c_iter_->Valid());
  ASSERT_EQ(test::KeyStr("a", 5, kTypeSingleDeletion),
@@ -63,6 +171,173 @@ TEST_F(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
  ASSERT_FALSE(c_iter_->Valid());
 }

+TEST_F(CompactionIteratorTest, SimpleRangeDeletion) {
+  InitIterators({test::KeyStr("morning", 5, kTypeValue),
+                 test::KeyStr("morning", 2, kTypeValue),
+                 test::KeyStr("night", 3, kTypeValue)},
+                {"zao", "zao", "wan"},
+                {test::KeyStr("ma", 4, kTypeRangeDeletion)}, {"mz"}, 5);
+  c_iter_->SeekToFirst();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("morning", 5, kTypeValue), c_iter_->key().ToString());
+  c_iter_->Next();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("night", 3, kTypeValue), c_iter_->key().ToString());
+  c_iter_->Next();
+  ASSERT_FALSE(c_iter_->Valid());
+}
+
+TEST_F(CompactionIteratorTest, RangeDeletionWithSnapshots) {
+  AddSnapshot(10);
+  std::vector<std::string> ks1;
+  ks1.push_back(test::KeyStr("ma", 28, kTypeRangeDeletion));
+  std::vector<std::string> vs1{"mz"};
+  std::vector<std::string> ks2{test::KeyStr("morning", 15, kTypeValue),
+                               test::KeyStr("morning", 5, kTypeValue),
+                               test::KeyStr("night", 40, kTypeValue),
+                               test::KeyStr("night", 20, kTypeValue)};
+  std::vector<std::string> vs2{"zao 15", "zao 5", "wan 40", "wan 20"};
+  InitIterators(ks2, vs2, ks1, vs1, 40);
+  c_iter_->SeekToFirst();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("morning", 5, kTypeValue), c_iter_->key().ToString());
+  c_iter_->Next();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("night", 40, kTypeValue), c_iter_->key().ToString());
+  c_iter_->Next();
+  ASSERT_FALSE(c_iter_->Valid());
+}
+
+TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
+  // Expect no merging attempts.
+  class MergeOp : public MergeOperator {
+   public:
+    bool FullMergeV2(const MergeOperationInput& merge_in,
+                     MergeOperationOutput* merge_out) const override {
+      ADD_FAILURE();
+      return false;
+    }
+    bool PartialMergeMulti(const Slice& key,
+                           const std::deque<Slice>& operand_list,
+                           std::string* new_value,
+                           Logger* logger) const override {
+      ADD_FAILURE();
+      return false;
+    }
+    const char* Name() const override {
+      return "CompactionIteratorTest.CompactionFilterSkipUntil::MergeOp";
+    }
+  };
+
+  class Filter : public CompactionFilter {
+    virtual Decision FilterV2(int level, const Slice& key, ValueType t,
+                              const Slice& existing_value,
+                              std::string* new_value,
+                              std::string* skip_until) const override {
+      std::string k = key.ToString();
+      std::string v = existing_value.ToString();
+      // See InitIterators() call below for the sequence of keys and their
+      // filtering decisions. Here we closely assert that compaction filter is
+      // called with the expected keys and only them, and with the right values.
+      if (k == "a") {
+        EXPECT_EQ(ValueType::kValue, t);
+        EXPECT_EQ("av50", v);
+        return Decision::kKeep;
+      }
+      if (k == "b") {
+        EXPECT_EQ(ValueType::kValue, t);
+        EXPECT_EQ("bv60", v);
+        *skip_until = "d+";
+        return Decision::kRemoveAndSkipUntil;
+      }
+      if (k == "e") {
+        EXPECT_EQ(ValueType::kMergeOperand, t);
+        EXPECT_EQ("em71", v);
+        return Decision::kKeep;
+      }
+      if (k == "f") {
+        if (v == "fm65") {
+          EXPECT_EQ(ValueType::kMergeOperand, t);
+          *skip_until = "f";
+        } else {
+          EXPECT_EQ("fm30", v);
+          EXPECT_EQ(ValueType::kMergeOperand, t);
+          *skip_until = "g+";
+        }
+        return Decision::kRemoveAndSkipUntil;
+      }
+      if (k == "h") {
+        EXPECT_EQ(ValueType::kValue, t);
+        EXPECT_EQ("hv91", v);
+        return Decision::kKeep;
+      }
+      if (k == "i") {
+        EXPECT_EQ(ValueType::kMergeOperand, t);
+        EXPECT_EQ("im95", v);
+        *skip_until = "z";
+        return Decision::kRemoveAndSkipUntil;
+      }
+      ADD_FAILURE();
+      return Decision::kKeep;
+    }
+
+    const char* Name() const override {
+      return "CompactionIteratorTest.CompactionFilterSkipUntil::Filter";
+    }
+  };
+
+  MergeOp merge_op;
+  Filter filter;
+  InitIterators(
+      {test::KeyStr("a", 50, kTypeValue),  // keep
+       test::KeyStr("a", 45, kTypeMerge),
+       test::KeyStr("b", 60, kTypeValue),  // skip to "d+"
+       test::KeyStr("b", 40, kTypeValue), test::KeyStr("c", 35, kTypeValue),
+       test::KeyStr("d", 70, kTypeMerge),
+       test::KeyStr("e", 71, kTypeMerge),  // keep
+       test::KeyStr("f", 65, kTypeMerge),  // skip to "f", aka keep
+       test::KeyStr("f", 30, kTypeMerge),  // skip to "g+"
+       test::KeyStr("f", 25, kTypeValue), test::KeyStr("g", 90, kTypeValue),
+       test::KeyStr("h", 91, kTypeValue),  // keep
+       test::KeyStr("i", 95, kTypeMerge),  // skip to "z"
+       test::KeyStr("j", 99, kTypeValue)},
+      {"av50", "am45", "bv60", "bv40", "cv35", "dm70", "em71", "fm65", "fm30",
+       "fv25", "gv90", "hv91", "im95", "jv99"},
+      {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+
+  // Compaction should output just "a", "e" and "h" keys.
+  c_iter_->SeekToFirst();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("a", 50, kTypeValue), c_iter_->key().ToString());
+  ASSERT_EQ("av50", c_iter_->value().ToString());
+  c_iter_->Next();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("e", 71, kTypeMerge), c_iter_->key().ToString());
+  ASSERT_EQ("em71", c_iter_->value().ToString());
+  c_iter_->Next();
+  ASSERT_TRUE(c_iter_->Valid());
+  ASSERT_EQ(test::KeyStr("h", 91, kTypeValue), c_iter_->key().ToString());
+  ASSERT_EQ("hv91", c_iter_->value().ToString());
+  c_iter_->Next();
+  ASSERT_FALSE(c_iter_->Valid());
+
+  // Check that the compaction iterator did the correct sequence of calls on
+  // the underlying iterator.
+  using A = LoggingForwardVectorIterator::Action;
+  using T = A::Type;
+  std::vector<A> expected_actions = {
+      A(T::SEEK_TO_FIRST),
+      A(T::NEXT),
+      A(T::NEXT),
+      A(T::SEEK, test::KeyStr("d+", kMaxSequenceNumber, kValueTypeForSeek)),
+      A(T::NEXT),
+      A(T::NEXT),
+      A(T::SEEK, test::KeyStr("g+", kMaxSequenceNumber, kValueTypeForSeek)),
+      A(T::NEXT),
+      A(T::SEEK, test::KeyStr("z", kMaxSequenceNumber, kValueTypeForSeek))};
+  ASSERT_EQ(expected_actions, iter_->log);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -262,7 +262,7 @@ void CompactionJob::AggregateStatistics() {
 }

 CompactionJob::CompactionJob(
-    int job_id, Compaction* compaction, const DBOptions& db_options,
+    int job_id, Compaction* compaction, const ImmutableDBOptions& db_options,
    const EnvOptions& env_options, VersionSet* versions,
    std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
    Directory* db_directory, Directory* output_directory, Statistics* stats,
@@ -297,7 +297,7 @@ CompactionJob::CompactionJob(
  assert(log_buffer_ != nullptr);
  const auto* cfd = compact_->compaction->column_family_data();
  ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
-                                    cfd->options()->enable_thread_tracking);
+                                    db_options_.enable_thread_tracking);
  ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_COMPACTION);
  ReportStartedCompaction(compaction);
 }
@@ -311,7 +311,7 @@ void CompactionJob::ReportStartedCompaction(
    Compaction* compaction) {
  const auto* cfd = compact_->compaction->column_family_data();
  ThreadStatusUtil::SetColumnFamily(cfd, cfd->ioptions()->env,
-                                    cfd->options()->enable_thread_tracking);
+                                    db_options_.enable_thread_tracking);

  ThreadStatusUtil::SetThreadOperationProperty(
      ThreadStatus::COMPACTION_JOB_ID,
@@ -484,9 +484,8 @@ void CompactionJob::GenSubcompactionBoundaries() {
                static_cast<uint64_t>(db_options_.max_subcompactions),
                max_output_files});

-  double mean = sum * 1.0 / subcompactions;
-
  if (subcompactions > 1) {
+    double mean = sum * 1.0 / subcompactions;
    // Greedily add ranges to the subcompaction until the sum of the ranges'
    // sizes becomes >= the expected mean size of a subcompaction
    sum = 0;
@@ -539,7 +538,7 @@ Status CompactionJob::Run() {
    thread.join();
  }

-  if (output_directory_ && !db_options_.disableDataSync) {
+  if (output_directory_ && !db_options_.disable_data_sync) {
    output_directory_->Fsync();
  }

@@ -591,6 +590,16 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
  VersionStorageInfo::LevelSummaryStorage tmp;
  auto vstorage = cfd->current()->storage_info();
  const auto& stats = compaction_stats_;
+
+  double read_write_amp = 0.0;
+  double write_amp = 0.0;
+  if (stats.bytes_read_non_output_levels > 0) {
+    read_write_amp = (stats.bytes_written + stats.bytes_read_output_level +
+                      stats.bytes_read_non_output_levels) /
+                     static_cast<double>(stats.bytes_read_non_output_levels);
+    write_amp = stats.bytes_written /
+                static_cast<double>(stats.bytes_read_non_output_levels);
+  }
  LogToBuffer(
      log_buffer_,
      "[%s] compacted to: %s, MB/sec: %.1f rd, %.1f wr, level %d, "
@@ -603,16 +612,10 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
      stats.bytes_written / static_cast<double>(stats.micros),
      compact_->compaction->output_level(),
      stats.num_input_files_in_non_output_levels,
-      stats.num_input_files_in_output_level,
-      stats.num_output_files,
+      stats.num_input_files_in_output_level, stats.num_output_files,
      stats.bytes_read_non_output_levels / 1048576.0,
      stats.bytes_read_output_level / 1048576.0,
-      stats.bytes_written / 1048576.0,
-      (stats.bytes_written + stats.bytes_read_output_level +
-       stats.bytes_read_non_output_levels) /
-          static_cast<double>(stats.bytes_read_non_output_levels),
-      stats.bytes_written /
-          static_cast<double>(stats.bytes_read_non_output_levels),
+      stats.bytes_written / 1048576.0, read_write_amp, write_amp,
      status.ToString().c_str(), stats.num_input_records,
      stats.num_dropped_records);

@@ -658,8 +661,11 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {

 void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  assert(sub_compact != nullptr);
-  std::unique_ptr<InternalIterator> input(
-      versions_->MakeInputIterator(sub_compact->compaction));
+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  std::unique_ptr<RangeDelAggregator> range_del_agg(
+      new RangeDelAggregator(cfd->internal_comparator(), existing_snapshots_));
+  std::unique_ptr<InternalIterator> input(versions_->MakeInputIterator(
+      sub_compact->compaction, range_del_agg.get()));

  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_PROCESS_KV);
@@ -680,7 +686,6 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    prev_prepare_write_nanos = IOSTATS(prepare_write_nanos);
  }

-  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
  const MutableCFOptions* mutable_cf_options =
      sub_compact->compaction->mutable_cf_options();

@@ -737,7 +742,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  sub_compact->c_iter.reset(new CompactionIterator(
      input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
      &existing_snapshots_, earliest_write_conflict_snapshot_, env_, false,
-      sub_compact->compaction, compaction_filter));
+      range_del_agg.get(), sub_compact->compaction, compaction_filter));
  auto c_iter = sub_compact->c_iter.get();
  c_iter->SeekToFirst();
  const auto& c_iter_stats = c_iter->iter_stats();
@@ -762,10 +767,16 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    if (end != nullptr &&
        cfd->user_comparator()->Compare(c_iter->user_key(), *end) >= 0) {
      break;
-    } else if (sub_compact->ShouldStopBefore(
+    } else if (sub_compact->compaction->output_level() != 0 &&
+               sub_compact->ShouldStopBefore(
                   key, sub_compact->current_output_file_size) &&
               sub_compact->builder != nullptr) {
-      status = FinishCompactionOutputFile(input->status(), sub_compact);
+      CompactionIterationStats range_del_out_stats;
+      status = FinishCompactionOutputFile(input->status(), sub_compact,
+                                          range_del_agg.get(),
+                                          &range_del_out_stats, &key);
+      RecordDroppedKeys(range_del_out_stats,
+                        &sub_compact->compaction_job_stats);
      if (!status.ok()) {
        break;
      }
@@ -847,16 +858,30 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    // during subcompactions (i.e. if output size, estimated by input size, is
    // going to be 1.2MB and max_output_file_size = 1MB, prefer to have 0.6MB
    // and 0.6MB instead of 1MB and 0.2MB)
-    if (sub_compact->current_output_file_size >=
-        sub_compact->compaction->max_output_file_size()) {
-      status = FinishCompactionOutputFile(input->status(), sub_compact);
+    if (sub_compact->compaction->output_level() != 0 &&
+        sub_compact->current_output_file_size >=
+            sub_compact->compaction->max_output_file_size()) {
+      Status input_status = input->status();
+      c_iter->Next();
+
+      const Slice* next_key = nullptr;
+      if (c_iter->Valid()) {
+        next_key = &c_iter->key();
+      }
+      CompactionIterationStats range_del_out_stats;
+      status = FinishCompactionOutputFile(input_status, sub_compact,
+                                          range_del_agg.get(),
+                                          &range_del_out_stats, next_key);
+      RecordDroppedKeys(range_del_out_stats,
+                        &sub_compact->compaction_job_stats);
      if (sub_compact->outputs.size() == 1) {
        // Use dictionary from first output file for compression of subsequent
        // files.
        sub_compact->compression_dict = std::move(compression_dict);
      }
+    } else {
+      c_iter->Next();
    }
-    c_iter->Next();
  }

  sub_compact->num_input_records = c_iter_stats.num_input_records;
@@ -883,8 +908,18 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    status = Status::ShutdownInProgress(
        "Database shutdown or Column family drop during compaction");
  }
+  if (status.ok() && sub_compact->builder == nullptr &&
+      sub_compact->outputs.size() == 0 &&
+      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
+    // handle subcompaction containing only range deletions
+    status = OpenCompactionOutputFile(sub_compact);
+  }
  if (status.ok() && sub_compact->builder != nullptr) {
-    status = FinishCompactionOutputFile(input->status(), sub_compact);
+    CompactionIterationStats range_del_out_stats;
+    status =
+        FinishCompactionOutputFile(input->status(), sub_compact,
+                                   range_del_agg.get(), &range_del_out_stats);
+    RecordDroppedKeys(range_del_out_stats, &sub_compact->compaction_job_stats);
  }
  if (status.ok()) {
    status = input->status();
@@ -910,7 +945,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
 }

 void CompactionJob::RecordDroppedKeys(
-    const CompactionIteratorStats& c_iter_stats,
+    const CompactionIterationStats& c_iter_stats,
    CompactionJobStats* compaction_job_stats) {
  if (c_iter_stats.num_record_drop_user > 0) {
    RecordTick(stats_, COMPACTION_KEY_DROP_USER,
@@ -932,10 +967,21 @@ void CompactionJob::RecordDroppedKeys(
          c_iter_stats.num_record_drop_obsolete;
    }
  }
+  if (c_iter_stats.num_record_drop_range_del > 0) {
+    RecordTick(stats_, COMPACTION_KEY_DROP_RANGE_DEL,
+               c_iter_stats.num_record_drop_range_del);
+  }
+  if (c_iter_stats.num_range_del_drop_obsolete > 0) {
+    RecordTick(stats_, COMPACTION_RANGE_DEL_DROP_OBSOLETE,
+               c_iter_stats.num_range_del_drop_obsolete);
+  }
 }

 Status CompactionJob::FinishCompactionOutputFile(
-    const Status& input_status, SubcompactionState* sub_compact) {
+    const Status& input_status, SubcompactionState* sub_compact,
+    RangeDelAggregator* range_del_agg,
+    CompactionIterationStats* range_del_out_stats,
+    const Slice* next_table_min_key /* = nullptr */) {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_SYNC_FILE);
  assert(sub_compact != nullptr);
@@ -950,6 +996,36 @@ Status CompactionJob::FinishCompactionOutputFile(
  // Check for iterator errors
  Status s = input_status;
  auto meta = &sub_compact->current_output()->meta;
+  if (s.ok()) {
+    Slice lower_bound_guard, upper_bound_guard;
+    const Slice *lower_bound, *upper_bound;
+    if (sub_compact->outputs.size() == 1) {
+      // For the first output table, include range tombstones before the min key
+      // but after the subcompaction boundary.
+      lower_bound = sub_compact->start;
+    } else if (meta->smallest.size() > 0) {
+      // For subsequent output tables, only include range tombstones from min
+      // key onwards since the previous file was extended to contain range
+      // tombstones falling before min key.
+      lower_bound_guard = meta->smallest.user_key();
+      lower_bound = &lower_bound_guard;
+    } else {
+      lower_bound = nullptr;
+    }
+    if (next_table_min_key != nullptr) {
+      // This isn't the last file in the subcompaction, so extend until the next
+      // file starts.
+      upper_bound_guard = ExtractUserKey(*next_table_min_key);
+      upper_bound = &upper_bound_guard;
+    } else {
+      // This is the last file in the subcompaction, so extend until the
+      // subcompaction ends.
+      upper_bound = sub_compact->end;
+    }
+    range_del_agg->AddToBuilder(sub_compact->builder.get(), lower_bound,
+                                upper_bound, meta, range_del_out_stats,
+                                bottommost_level_);
+  }
  const uint64_t current_entries = sub_compact->builder->NumEntries();
  meta->marked_for_compaction = sub_compact->builder->NeedCompact();
  if (s.ok()) {
@@ -963,7 +1039,7 @@ Status CompactionJob::FinishCompactionOutputFile(
  sub_compact->total_bytes += current_bytes;

  // Finish and check for file errors
-  if (s.ok() && !db_options_.disableDataSync) {
+  if (s.ok() && !db_options_.disable_data_sync) {
    StopWatch sw(env_, stats_, COMPACTION_OUTFILE_SYNC_MICROS);
    s = sub_compact->outfile->Sync(db_options_.use_fsync);
  }
@@ -978,8 +1054,9 @@ Status CompactionJob::FinishCompactionOutputFile(
    // Verify that the table is usable
    InternalIterator* iter = cfd->table_cache()->NewIterator(
        ReadOptions(), env_options_, cfd->internal_comparator(), meta->fd,
-        nullptr, cfd->internal_stats()->GetFileReadHist(
-                     compact_->compaction->output_level()),
+        nullptr /* range_del_agg */, nullptr,
+        cfd->internal_stats()->GetFileReadHist(
+            compact_->compaction->output_level()),
        false);
    s = iter->status();

@@ -1136,7 +1213,9 @@ Status CompactionJob::OpenCompactionOutputFile(
      *cfd->ioptions(), cfd->internal_comparator(),
      cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
      sub_compact->outfile.get(), sub_compact->compaction->output_compression(),
-      cfd->ioptions()->compression_opts, &sub_compact->compression_dict,
+      cfd->ioptions()->compression_opts,
+      sub_compact->compaction->output_level(),
+      &sub_compact->compression_dict,
      skip_filters));
  LogFlush(db_options_.info_log);
  return s;
@@ -1184,8 +1263,7 @@ void CompactionJob::UpdateCompactionStats() {
  for (int input_level = 0;
       input_level < static_cast<int>(compaction->num_input_levels());
       ++input_level) {
-    if (compaction->start_level() + input_level
-        != compaction->output_level()) {
+    if (compaction->level(input_level) != compaction->output_level()) {
      UpdateCompactionInputStatsHelper(
          &compaction_stats_.num_input_files_in_non_output_levels,
          &compaction_stats_.bytes_read_non_output_levels,
@@ -25,6 +25,7 @@
 #include "db/job_context.h"
 #include "db/log_writer.h"
 #include "db/memtable_list.h"
+#include "db/range_del_aggregator.h"
 #include "db/version_edit.h"
 #include "db/write_controller.h"
 #include "db/write_thread.h"
@@ -37,6 +38,7 @@
 #include "rocksdb/transaction_log.h"
 #include "table/scoped_arena_iterator.h"
 #include "util/autovector.h"
+#include "util/db_options.h"
 #include "util/event_logger.h"
 #include "util/stop_watch.h"
 #include "util/thread_local.h"
@@ -52,7 +54,8 @@ class Arena;

 class CompactionJob {
 public:
-  CompactionJob(int job_id, Compaction* compaction, const DBOptions& db_options,
+  CompactionJob(int job_id, Compaction* compaction,
+                const ImmutableDBOptions& db_options,
                const EnvOptions& env_options, VersionSet* versions,
                std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
                Directory* db_directory, Directory* output_directory,
@@ -93,15 +96,18 @@ class CompactionJob {
  // kv-pairs
  void ProcessKeyValueCompaction(SubcompactionState* sub_compact);

-  Status FinishCompactionOutputFile(const Status& input_status,
-                                    SubcompactionState* sub_compact);
+  Status FinishCompactionOutputFile(
+      const Status& input_status, SubcompactionState* sub_compact,
+      RangeDelAggregator* range_del_agg,
+      CompactionIterationStats* range_del_out_stats,
+      const Slice* next_table_min_key = nullptr);
  Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options);
  void RecordCompactionIOStats();
  Status OpenCompactionOutputFile(SubcompactionState* sub_compact);
  void CleanupCompaction();
  void UpdateCompactionJobStats(
    const InternalStats::CompactionStats& stats) const;
-  void RecordDroppedKeys(const CompactionIteratorStats& c_iter_stats,
+  void RecordDroppedKeys(const CompactionIterationStats& c_iter_stats,
                         CompactionJobStats* compaction_job_stats = nullptr);

  void UpdateCompactionStats();
@@ -120,7 +126,7 @@ class CompactionJob {

  // DBImpl state
  const std::string& dbname_;
-  const DBOptions& db_options_;
+  const ImmutableDBOptions& db_options_;
  const EnvOptions& env_options_;

  Env* env_;
@@ -68,7 +68,8 @@ class CompactionJobTest : public testing::Test {
  CompactionJobTest()
      : env_(Env::Default()),
        dbname_(test::TmpDir() + "/compaction_job_test"),
-        mutable_cf_options_(Options(), ImmutableCFOptions(Options())),
+        db_options_(),
+        mutable_cf_options_(cf_options_),
        table_cache_(NewLRUCache(50000, 16)),
        write_buffer_manager_(db_options_.db_write_buffer_size),
        versions_(new VersionSet(dbname_, &db_options_, env_options_,
@@ -239,7 +240,7 @@ class CompactionJobTest : public testing::Test {
      num_input_files += level_files.size();
    }

-    Compaction compaction(cfd->current()->storage_info(),
+    Compaction compaction(cfd->current()->storage_info(), *cfd->ioptions(),
                          *cfd->GetLatestMutableCFOptions(),
                          compaction_input_files, 1, 1024 * 1024,
                          10 * 1024 * 1024, 0, kNoCompression, {}, true);
@@ -280,11 +281,11 @@ class CompactionJobTest : public testing::Test {
  Env* env_;
  std::string dbname_;
  EnvOptions env_options_;
+  ImmutableDBOptions db_options_;
+  ColumnFamilyOptions cf_options_;
  MutableCFOptions mutable_cf_options_;
  std::shared_ptr<Cache> table_cache_;
  WriteController write_controller_;
-  DBOptions db_options_;
-  ColumnFamilyOptions cf_options_;
  WriteBufferManager write_buffer_manager_;
  std::unique_ptr<VersionSet> versions_;
  InstrumentedMutex mutex_;
@@ -151,17 +151,15 @@ CompactionPicker::~CompactionPicker() {}

 // Delete this compaction from the list of running compactions.
 void CompactionPicker::ReleaseCompactionFiles(Compaction* c, Status status) {
-  if (c->start_level() == 0 ||
-      ioptions_.compaction_style == kCompactionStyleUniversal) {
-    level0_compactions_in_progress_.erase(c);
-  }
+  UnregisterCompaction(c);
  if (!status.ok()) {
    c->ResetNextCompactionIndex();
  }
 }

 void CompactionPicker::GetRange(const CompactionInputFiles& inputs,
-                                InternalKey* smallest, InternalKey* largest) {
+                                InternalKey* smallest,
+                                InternalKey* largest) const {
  const int level = inputs.level;
  assert(!inputs.empty());
  smallest->Clear();
@@ -190,7 +188,8 @@ void CompactionPicker::GetRange(const CompactionInputFiles& inputs,

 void CompactionPicker::GetRange(const CompactionInputFiles& inputs1,
                                const CompactionInputFiles& inputs2,
-                                InternalKey* smallest, InternalKey* largest) {
+                                InternalKey* smallest,
+                                InternalKey* largest) const {
  assert(!inputs1.empty() || !inputs2.empty());
  if (inputs1.empty()) {
    GetRange(inputs2, smallest, largest);
@@ -206,6 +205,33 @@ void CompactionPicker::GetRange(const CompactionInputFiles& inputs1,
  }
 }

+void CompactionPicker::GetRange(const std::vector<CompactionInputFiles>& inputs,
+                                InternalKey* smallest,
+                                InternalKey* largest) const {
+  InternalKey current_smallest;
+  InternalKey current_largest;
+  bool initialized = false;
+  for (const auto& in : inputs) {
+    if (in.empty()) {
+      continue;
+    }
+    GetRange(in, &current_smallest, &current_largest);
+    if (!initialized) {
+      *smallest = current_smallest;
+      *largest = current_largest;
+      initialized = true;
+    } else {
+      if (icmp_->Compare(current_smallest, *smallest) < 0) {
+        *smallest = current_smallest;
+      }
+      if (icmp_->Compare(current_largest, *largest) > 0) {
+        *largest = current_largest;
+      }
+    }
+  }
+  assert(initialized);
+}
+
 bool CompactionPicker::ExpandWhileOverlapping(const std::string& cf_name,
                                              VersionStorageInfo* vstorage,
                                              CompactionInputFiles* inputs) {
@@ -250,6 +276,42 @@ bool CompactionPicker::ExpandWhileOverlapping(const std::string& cf_name,
  return true;
 }

+bool CompactionPicker::RangeOverlapWithCompaction(
+    const Slice& smallest_user_key, const Slice& largest_user_key,
+    int level) const {
+  const Comparator* ucmp = icmp_->user_comparator();
+  for (Compaction* c : compactions_in_progress_) {
+    if (c->output_level() == level &&
+        ucmp->Compare(smallest_user_key, c->GetLargestUserKey()) <= 0 &&
+        ucmp->Compare(largest_user_key, c->GetSmallestUserKey()) >= 0) {
+      // Overlap
+      return true;
+    }
+  }
+  // Did not overlap with any running compaction in level `level`
+  return false;
+}
+
+bool CompactionPicker::FilesRangeOverlapWithCompaction(
+    const std::vector<CompactionInputFiles>& inputs, int level) const {
+  bool is_empty = true;
+  for (auto& in : inputs) {
+    if (!in.empty()) {
+      is_empty = false;
+      break;
+    }
+  }
+  if (is_empty) {
+    // No files in inputs
+    return false;
+  }
+
+  InternalKey smallest, largest;
+  GetRange(inputs, &smallest, &largest);
+  return RangeOverlapWithCompaction(smallest.user_key(), largest.user_key(),
+                                    level);
+}
+
 // Returns true if any one of specified files are being compacted
 bool CompactionPicker::FilesInCompaction(
    const std::vector<FileMetaData*>& files) {
@@ -274,17 +336,19 @@ Compaction* CompactionPicker::FormCompaction(
  if ((input_files[0].level == 0) && !level0_compactions_in_progress_.empty()) {
    return nullptr;
  }
+  // This compaction output could overlap with a running compaction
+  if (FilesRangeOverlapWithCompaction(input_files, output_level)) {
+    return nullptr;
+  }
  auto c =
-      new Compaction(vstorage, mutable_cf_options, input_files, output_level,
-                     compact_options.output_file_size_limit,
+      new Compaction(vstorage, ioptions_, mutable_cf_options, input_files,
+                     output_level, compact_options.output_file_size_limit,
                     mutable_cf_options.max_compaction_bytes, output_path_id,
                     compact_options.compression, /* grandparents */ {}, true);

  // If it's level 0 compaction, make sure we don't execute any other level 0
  // compactions in parallel
-  if ((c != nullptr) && (input_files[0].level == 0)) {
-    level0_compactions_in_progress_.insert(c);
-  }
+  RegisterCompaction(c);
  return c;
 }

@@ -387,13 +451,14 @@ bool CompactionPicker::SetupOtherInputs(
  vstorage->GetOverlappingInputs(output_level, &smallest, &largest,
                                 &output_level_inputs->files, *parent_index,
                                 parent_index);
-  if (!output_level_inputs->empty()) {
-    ExpandWhileOverlapping(cf_name, vstorage, output_level_inputs);
-  }
-
  if (FilesInCompaction(output_level_inputs->files)) {
    return false;
  }
+  if (!output_level_inputs->empty()) {
+    if (!ExpandWhileOverlapping(cf_name, vstorage, output_level_inputs)) {
+      return false;
+    }
+  }

  // See if we can further grow the number of inputs in "level" without
  // changing the number of "level+1" files we pick up. We also choose NOT
@@ -420,11 +485,15 @@ bool CompactionPicker::SetupOtherInputs(
        !vstorage->HasOverlappingUserKey(&expanded0.files, input_level)) {
      InternalKey new_start, new_limit;
      GetRange(expanded0, &new_start, &new_limit);
-      std::vector<FileMetaData*> expanded1;
+      CompactionInputFiles expanded1;
+      expanded1.level = output_level;
      vstorage->GetOverlappingInputs(output_level, &new_start, &new_limit,
-                                     &expanded1, *parent_index, parent_index);
-      if (expanded1.size() == output_level_inputs->size() &&
-          !FilesInCompaction(expanded1)) {
+                                     &expanded1.files, *parent_index,
+                                     parent_index);
+      assert(!expanded1.empty());
+      if (!FilesInCompaction(expanded1.files) &&
+          ExpandWhileOverlapping(cf_name, vstorage, &expanded1) &&
+          expanded1.size() == output_level_inputs->size()) {
        Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
            "[%s] Expanding@%d %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt "(%" PRIu64
            "+%" PRIu64 " bytes) to %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt
@@ -435,7 +504,7 @@ bool CompactionPicker::SetupOtherInputs(
        smallest = new_start;
        largest = new_limit;
        inputs->files = expanded0.files;
-        output_level_inputs->files = expanded1;
+        output_level_inputs->files = expanded1.files;
      }
    }
  }
@@ -506,16 +575,24 @@ Compaction* CompactionPicker::CompactRange(
        return nullptr;
      }
    }
+
+    // 2 non-exclusive manual compactions could run at the same time producing
+    // overlaping outputs in the same level.
+    if (FilesRangeOverlapWithCompaction(inputs, output_level)) {
+      // This compaction output could potentially conflict with the output
+      // of a currently running compaction, we cannot run it.
+      *manual_conflict = true;
+      return nullptr;
+    }
+
    Compaction* c = new Compaction(
-        vstorage, mutable_cf_options, std::move(inputs), output_level,
-        mutable_cf_options.MaxFileSizeForLevel(output_level),
+        vstorage, ioptions_, mutable_cf_options, std::move(inputs),
+        output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
        /* max_compaction_bytes */ LLONG_MAX, output_path_id,
        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
                           output_level, 1),
        /* grandparents */ {}, /* is manual */ true);
-    if (start_level == 0) {
-      level0_compactions_in_progress_.insert(c);
-    }
+    RegisterCompaction(c);
    return c;
  }

@@ -604,26 +681,33 @@ Compaction* CompactionPicker::CompactRange(
    }
  }

+  // 2 non-exclusive manual compactions could run at the same time producing
+  // overlaping outputs in the same level.
+  if (FilesRangeOverlapWithCompaction(compaction_inputs, output_level)) {
+    // This compaction output could potentially conflict with the output
+    // of a currently running compaction, we cannot run it.
+    *manual_conflict = true;
+    return nullptr;
+  }
+
  std::vector<FileMetaData*> grandparents;
  GetGrandparents(vstorage, inputs, output_level_inputs, &grandparents);
  Compaction* compaction = new Compaction(
-      vstorage, mutable_cf_options, std::move(compaction_inputs), output_level,
-      mutable_cf_options.MaxFileSizeForLevel(output_level),
+      vstorage, ioptions_, mutable_cf_options, std::move(compaction_inputs),
+      output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
      mutable_cf_options.max_compaction_bytes, output_path_id,
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
                         vstorage->base_level()),
      std::move(grandparents), /* is manual compaction */ true);

  TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
-  if (input_level == 0) {
-    level0_compactions_in_progress_.insert(compaction);
-  }
+  RegisterCompaction(compaction);

  // Creating a compaction influences the compaction score because the score
  // takes running compactions into account (by skipping files that are already
  // being compacted). Since we just changed compaction score, we recalculate it
  // here
-  vstorage->ComputeCompactionScore(mutable_cf_options);
+  vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options);

  return compaction;
 }
@@ -853,6 +937,30 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
 }
 #endif  // !ROCKSDB_LITE

+void CompactionPicker::RegisterCompaction(Compaction* c) {
+  if (c == nullptr) {
+    return;
+  }
+  assert(ioptions_.compaction_style != kCompactionStyleLevel ||
+         !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level()));
+  if (c->start_level() == 0 ||
+      ioptions_.compaction_style == kCompactionStyleUniversal) {
+    level0_compactions_in_progress_.insert(c);
+  }
+  compactions_in_progress_.insert(c);
+}
+
+void CompactionPicker::UnregisterCompaction(Compaction* c) {
+  if (c == nullptr) {
+    return;
+  }
+  if (c->start_level() == 0 ||
+      ioptions_.compaction_style == kCompactionStyleUniversal) {
+    level0_compactions_in_progress_.erase(c);
+  }
+  compactions_in_progress_.erase(c);
+}
+
 bool LevelCompactionPicker::NeedsCompaction(
    const VersionStorageInfo* vstorage) const {
  if (!vstorage->FilesMarkedForCompaction().empty()) {
@@ -936,7 +1044,8 @@ Compaction* LevelCompactionPicker::PickCompaction(
      output_level = (level == 0) ? vstorage->base_level() : level + 1;
      if (PickCompactionBySize(vstorage, level, output_level, &inputs,
                               &parent_index, &base_index) &&
-          ExpandWhileOverlapping(cf_name, vstorage, &inputs)) {
+          ExpandWhileOverlapping(cf_name, vstorage, &inputs) &&
+          !FilesRangeOverlapWithCompaction({inputs}, output_level)) {
        // found the compaction!
        if (level == 0) {
          // L0 score = `num L0 files` / `level0_file_num_compaction_trigger`
@@ -1009,11 +1118,23 @@ Compaction* LevelCompactionPicker::PickCompaction(
    compaction_inputs.push_back(output_level_inputs);
  }

+  // In some edge cases we could pick a compaction that will be compacting
+  // a key range that overlap with another running compaction, and both
+  // of them have the same output leve. This could happen if
+  // (1) we are running a non-exclusive manual compaction
+  // (2) AddFile ingest a new file into the LSM tree
+  // We need to disallow this from happening.
+  if (FilesRangeOverlapWithCompaction(compaction_inputs, output_level)) {
+    // This compaction output could potentially conflict with the output
+    // of a currently running compaction, we cannot run it.
+    return nullptr;
+  }
+
  std::vector<FileMetaData*> grandparents;
  GetGrandparents(vstorage, inputs, output_level_inputs, &grandparents);
  auto c = new Compaction(
-      vstorage, mutable_cf_options, std::move(compaction_inputs), output_level,
-      mutable_cf_options.MaxFileSizeForLevel(output_level),
+      vstorage, ioptions_, mutable_cf_options, std::move(compaction_inputs),
+      output_level, mutable_cf_options.MaxFileSizeForLevel(output_level),
      mutable_cf_options.max_compaction_bytes,
      GetPathId(ioptions_, mutable_cf_options, output_level),
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
@@ -1023,15 +1144,13 @@ Compaction* LevelCompactionPicker::PickCompaction(

  // If it's level 0 compaction, make sure we don't execute any other level 0
  // compactions in parallel
-  if (level == 0) {
-    level0_compactions_in_progress_.insert(c);
-  }
+  RegisterCompaction(c);

  // Creating a compaction influences the compaction score because the score
  // takes running compactions into account (by skipping files that are already
  // being compacted). Since we just changed compaction score, we recalculate it
  // here
-  vstorage->ComputeCompactionScore(mutable_cf_options);
+  vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options);

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

@@ -1065,7 +1184,8 @@ uint32_t LevelCompactionPicker::GetPathId(
        return p;
      } else {
        current_path_size -= level_size;
-        level_size *= mutable_cf_options.max_bytes_for_level_multiplier;
+        level_size = static_cast<uint64_t>(
+            level_size * mutable_cf_options.max_bytes_for_level_multiplier);
        cur_level++;
        continue;
      }
@@ -1421,7 +1541,7 @@ Compaction* UniversalCompactionPicker::PickCompaction(
  MeasureTime(ioptions_.statistics, NUM_FILES_IN_SINGLE_COMPACTION,
              c->inputs(0)->size());

-  level0_compactions_in_progress_.insert(c);
+  RegisterCompaction(c);

  TEST_SYNC_POINT_CALLBACK("UniversalCompactionPicker::PickCompaction:Return",
                           c);
@@ -1643,7 +1763,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    compaction_reason = CompactionReason::kUniversalSizeRatio;
  }
  return new Compaction(
-      vstorage, mutable_cf_options, std::move(inputs), output_level,
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
      mutable_cf_options.MaxFileSizeForLevel(output_level), LLONG_MAX, path_id,
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, start_level,
                         1, enable_compression),
@@ -1767,7 +1887,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
  }

  return new Compaction(
-      vstorage, mutable_cf_options, std::move(inputs),
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs),
      vstorage->num_levels() - 1,
      mutable_cf_options.MaxFileSizeForLevel(vstorage->num_levels() - 1),
      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
@@ -1832,10 +1952,10 @@ Compaction* FIFOCompactionPicker::PickCompaction(
    }
  }
  Compaction* c = new Compaction(
-      vstorage, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
      kNoCompression, {}, /* is manual */ false, vstorage->CompactionScore(0),
      /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
-  level0_compactions_in_progress_.insert(c);
+  RegisterCompaction(c);
  return c;
 }

@@ -113,6 +113,12 @@ class CompactionPicker {
    return !level0_compactions_in_progress_.empty();
  }

+  // Return true if the passed key range overlap with a compaction output
+  // that is currently running.
+  bool RangeOverlapWithCompaction(const Slice& smallest_user_key,
+                                  const Slice& largest_user_key,
+                                  int level) const;
+
 protected:
  int NumberLevels() const { return ioptions_.num_levels; }

@@ -120,14 +126,20 @@ class CompactionPicker {
  // *smallest, *largest.
  // REQUIRES: inputs is not empty
  void GetRange(const CompactionInputFiles& inputs, InternalKey* smallest,
-                InternalKey* largest);
+                InternalKey* largest) const;

  // Stores the minimal range that covers all entries in inputs1 and inputs2
  // in *smallest, *largest.
  // REQUIRES: inputs is not empty
  void GetRange(const CompactionInputFiles& inputs1,
                const CompactionInputFiles& inputs2, InternalKey* smallest,
-                InternalKey* largest);
+                InternalKey* largest) const;
+
+  // Stores the minimal range that covers all entries in inputs
+  // in *smallest, *largest.
+  // REQUIRES: inputs is not empty (at least on entry have one file)
+  void GetRange(const std::vector<CompactionInputFiles>& inputs,
+                InternalKey* smallest, InternalKey* largest) const;

  // Add more files to the inputs on "level" to make sure that
  // no newer version of a key is compacted to "level+1" while leaving an older
@@ -148,6 +160,11 @@ class CompactionPicker {
                         const InternalKey* smallest,
                         const InternalKey* largest, int level, int* index);

+  // Returns true if the key range that `inputs` files cover overlap with the
+  // key range of a currently running compaction.
+  bool FilesRangeOverlapWithCompaction(
+      const std::vector<CompactionInputFiles>& inputs, int level) const;
+
  bool SetupOtherInputs(const std::string& cf_name,
                        const MutableCFOptions& mutable_cf_options,
                        VersionStorageInfo* vstorage,
@@ -170,10 +187,20 @@ class CompactionPicker {
      const ColumnFamilyMetaData& cf_meta, const int output_level) const;
 #endif  // ROCKSDB_LITE

+  // Register this compaction in the set of running compactions
+  void RegisterCompaction(Compaction* c);
+
+  // Remove this compaction from the set of running compactions
+  void UnregisterCompaction(Compaction* c);
+
  // Keeps track of all compactions that are running on Level0.
-  // It is protected by DB mutex
+  // Protected by DB mutex
  std::set<Compaction*> level0_compactions_in_progress_;

+  // Keeps track of all compactions that are running.
+  // Protected by DB mutex
+  std::unordered_set<Compaction*> compactions_in_progress_;
+
  const InternalKeyComparator* const icmp_;
 };

@@ -48,7 +48,7 @@ class CompactionPickerTest : public testing::Test {
      : ucmp_(BytewiseComparator()),
        icmp_(ucmp_),
        ioptions_(options_),
-        mutable_cf_options_(options_, ioptions_),
+        mutable_cf_options_(options_),
        level_compaction_picker(ioptions_, &icmp_),
        cf_name_("dummy"),
        log_buffer_(InfoLogLevel::INFO_LEVEL, &logger_),
@@ -66,8 +66,8 @@ class CompactionPickerTest : public testing::Test {
  void NewVersionStorage(int num_levels, CompactionStyle style) {
    DeleteVersionStorage();
    options_.num_levels = num_levels;
-    vstorage_.reset(new VersionStorageInfo(
-        &icmp_, ucmp_, options_.num_levels, style, nullptr));
+    vstorage_.reset(new VersionStorageInfo(&icmp_, ucmp_, options_.num_levels,
+                                           style, nullptr, false));
    vstorage_->CalculateBaseBytes(ioptions_, mutable_cf_options_);
  }

@@ -115,12 +115,13 @@ class CompactionPickerTest : public testing::Test {

  void UpdateVersionStorageInfo() {
    vstorage_->CalculateBaseBytes(ioptions_, mutable_cf_options_);
-    vstorage_->UpdateFilesByCompactionPri(mutable_cf_options_);
+    vstorage_->UpdateFilesByCompactionPri(ioptions_.compaction_pri);
    vstorage_->UpdateNumNonEmptyLevels();
    vstorage_->GenerateFileIndexer();
    vstorage_->GenerateLevelFilesBrief();
-    vstorage_->ComputeCompactionScore(mutable_cf_options_);
+    vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_);
    vstorage_->GenerateLevel0NonOverlapping();
+    vstorage_->ComputeFilesMarkedForCompaction();
    vstorage_->SetFinalized();
  }
 };
@@ -350,10 +351,10 @@ TEST_F(CompactionPickerTest, Level0TriggerDynamic4) {
 TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
  int num_levels = ioptions_.num_levels;
  ioptions_.level_compaction_dynamic_level_bytes = true;
+  ioptions_.compaction_pri = kMinOverlappingRatio;
  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
  mutable_cf_options_.max_bytes_for_level_base = 200;
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
-  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
  NewVersionStorage(num_levels, kCompactionStyleLevel);
  Add(0, 1U, "150", "200");
  Add(num_levels - 1, 3U, "200", "250", 300U);
@@ -489,10 +490,10 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
  mutable_cf_options_.target_file_size_base = 10000000;
  mutable_cf_options_.target_file_size_multiplier = 10;
  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
-  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;

  Add(2, 6U, "150", "179", 50000000U);
  Add(2, 7U, "180", "220", 50000000U);
@@ -516,10 +517,10 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
  mutable_cf_options_.target_file_size_base = 10000000;
  mutable_cf_options_.target_file_size_multiplier = 10;
  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
-  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;

  Add(2, 6U, "150", "175",
      60000000U);  // Overlaps with file 26, 27, total size 521M
@@ -546,9 +547,9 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping3) {
  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
  mutable_cf_options_.max_bytes_for_level_base = 10000000;
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
-  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;

  // file 7 and 8 over lap with the same file, but file 8 is smaller so
  // it will be picked.
@@ -598,7 +599,7 @@ TEST_F(CompactionPickerTest, ParentIndexResetBug) {
 // ranges (with different sequence numbers) in the input files.
 TEST_F(CompactionPickerTest, OverlappingUserKeys) {
  NewVersionStorage(6, kCompactionStyleLevel);
-  mutable_cf_options_.compaction_pri = kByCompensatedSize;
+  ioptions_.compaction_pri = kByCompensatedSize;

  Add(1, 1U, "100", "150", 1U);
  // Overlapping user keys
@@ -693,6 +694,70 @@ TEST_F(CompactionPickerTest, OverlappingUserKeys4) {
  ASSERT_EQ(7U, compaction->input(1, 0)->fd.GetNumber());
 }

+TEST_F(CompactionPickerTest, OverlappingUserKeys5) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  // Overlapping user keys on same level and output level
+  Add(1, 1U, "200", "400", 1000000000U);
+  Add(1, 2U, "400", "500", 1U, 0, 0);
+  Add(2, 3U, "000", "100", 1U);
+  Add(2, 4U, "100", "600", 1U, 0, 0);
+  Add(2, 5U, "600", "700", 1U, 0, 0);
+
+  vstorage_->LevelFiles(2)[2]->being_compacted = true;
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() == nullptr);
+}
+
+TEST_F(CompactionPickerTest, OverlappingUserKeys6) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  // Overlapping user keys on same level and output level
+  Add(1, 1U, "200", "400", 1U, 0, 0);
+  Add(1, 2U, "401", "500", 1U, 0, 0);
+  Add(2, 3U, "000", "100", 1U);
+  Add(2, 4U, "100", "300", 1U, 0, 0);
+  Add(2, 5U, "305", "450", 1U, 0, 0);
+  Add(2, 6U, "460", "600", 1U, 0, 0);
+  Add(2, 7U, "600", "700", 1U, 0, 0);
+
+  vstorage_->LevelFiles(1)[0]->marked_for_compaction = true;
+  vstorage_->LevelFiles(1)[1]->marked_for_compaction = true;
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  ASSERT_EQ(3U, compaction->num_input_files(1));
+}
+
+TEST_F(CompactionPickerTest, OverlappingUserKeys7) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.max_compaction_bytes = 100000000000u;
+  // Overlapping user keys on same level and output level
+  Add(1, 1U, "200", "400", 1U, 0, 0);
+  Add(1, 2U, "401", "500", 1000000000U, 0, 0);
+  Add(2, 3U, "100", "250", 1U);
+  Add(2, 4U, "300", "600", 1U, 0, 0);
+  Add(2, 5U, "600", "800", 1U, 0, 0);
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_GE(1U, compaction->num_input_files(0));
+  ASSERT_GE(2U, compaction->num_input_files(1));
+  // File 5 has to be included in the compaction
+  ASSERT_EQ(5U, compaction->inputs(1)->back()->fd.GetNumber());
+}
+
 TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri1) {
  NewVersionStorage(6, kCompactionStyleLevel);
  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
@@ -40,6 +40,10 @@ class KVIter : public Iterator {
  virtual void Seek(const Slice& k) override {
    iter_ = map_->lower_bound(k.ToString());
  }
+  virtual void SeekForPrev(const Slice& k) override {
+    iter_ = map_->upper_bound(k.ToString());
+    Prev();
+  }
  virtual void Next() override { ++iter_; }
  virtual void Prev() override {
    if (iter_ == map_->begin()) {
@@ -41,6 +41,7 @@ class CuckooTableDBTest : public testing::Test {
    options.memtable_factory.reset(NewHashLinkListRepFactory(4, 0, 3, true));
    options.allow_mmap_reads = true;
    options.create_if_missing = true;
+    options.allow_concurrent_memtable_write = false;
    return options;
  }

@@ -810,6 +810,7 @@ TEST_F(DBBloomFilterTest, PrefixScan) {
    options.max_background_compactions = 2;
    options.create_if_missing = true;
    options.memtable_factory.reset(NewHashSkipListRepFactory(16));
+    options.allow_concurrent_memtable_write = false;

    BlockBasedTableOptions table_options;
    table_options.no_block_cache = true;
@@ -13,6 +13,7 @@
 namespace rocksdb {

 static int cfilter_count = 0;
+static int cfilter_skips = 0;

 // This is a static filter used for filtering
 // kvs during the compaction process.
@@ -65,6 +66,30 @@ class DeleteISFilter : public CompactionFilter {
  virtual const char* Name() const override { return "DeleteFilter"; }
 };

+// Skip x if floor(x/10) is even, use range skips. Requires that keys are
+// zero-padded to length 10.
+class SkipEvenFilter : public CompactionFilter {
+ public:
+  virtual Decision FilterV2(int level, const Slice& key, ValueType value_type,
+                            const Slice& existing_value, std::string* new_value,
+                            std::string* skip_until) const override {
+    cfilter_count++;
+    int i = std::stoi(key.ToString());
+    if (i / 10 % 2 == 0) {
+      char key_str[100];
+      snprintf(key_str, sizeof(key), "%010d", i / 10 * 10 + 10);
+      *skip_until = key_str;
+      ++cfilter_skips;
+      return Decision::kRemoveAndSkipUntil;
+    }
+    return Decision::kKeep;
+  }
+
+  virtual bool IgnoreSnapshots() const override { return true; }
+
+  virtual const char* Name() const override { return "DeleteFilter"; }
+};
+
 class DelayFilter : public CompactionFilter {
 public:
  explicit DelayFilter(DBTestBase* d) : db_test(d) {}
@@ -174,6 +199,20 @@ class DeleteISFilterFactory : public CompactionFilterFactory {
  virtual const char* Name() const override { return "DeleteFilterFactory"; }
 };

+class SkipEvenFilterFactory : public CompactionFilterFactory {
+ public:
+  virtual std::unique_ptr<CompactionFilter> CreateCompactionFilter(
+      const CompactionFilter::Context& context) override {
+    if (context.is_manual_compaction) {
+      return std::unique_ptr<CompactionFilter>(new SkipEvenFilter());
+    } else {
+      return std::unique_ptr<CompactionFilter>(nullptr);
+    }
+  }
+
+  virtual const char* Name() const override { return "SkipEvenFilterFactory"; }
+};
+
 class DelayFilterFactory : public CompactionFilterFactory {
 public:
  explicit DelayFilterFactory(DBTestBase* d) : db_test(d) {}
@@ -261,8 +300,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
  int total = 0;
  Arena arena;
  {
+    InternalKeyComparator icmp(options.comparator);
+    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
    ScopedArenaIterator iter(
-        dbfull()->NewInternalIterator(&arena, handles_[1]));
+        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[1]));
    iter->SeekToFirst();
    ASSERT_OK(iter->status());
    while (iter->Valid()) {
@@ -349,8 +390,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
  // level Lmax because this record is at the tip
  count = 0;
  {
+    InternalKeyComparator icmp(options.comparator);
+    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
    ScopedArenaIterator iter(
-        dbfull()->NewInternalIterator(&arena, handles_[1]));
+        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[1]));
    iter->SeekToFirst();
    ASSERT_OK(iter->status());
    while (iter->Valid()) {
@@ -566,7 +609,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextManual) {
    int count = 0;
    int total = 0;
    Arena arena;
-    ScopedArenaIterator iter(dbfull()->NewInternalIterator(&arena));
+    InternalKeyComparator icmp(options.comparator);
+    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
+    ScopedArenaIterator iter(
+        dbfull()->NewInternalIterator(&arena, &range_del_agg));
    iter->SeekToFirst();
    ASSERT_OK(iter->status());
    while (iter->Valid()) {
@@ -712,6 +758,47 @@ TEST_F(DBTestCompactionFilter, CompactionFilterIgnoreSnapshot) {
 }
 #endif  // ROCKSDB_LITE

+TEST_F(DBTestCompactionFilter, SkipUntil) {
+  Options options = CurrentOptions();
+  options.compaction_filter_factory = std::make_shared<SkipEvenFilterFactory>();
+  options.disable_auto_compactions = true;
+  options.create_if_missing = true;
+  DestroyAndReopen(options);
+
+  // Write 100K keys, these are written to a few files in L0.
+  for (int table = 0; table < 4; ++table) {
+    // Key ranges in tables are [0, 38], [106, 149], [212, 260], [318, 371].
+    for (int i = table * 6; i < 39 + table * 11; ++i) {
+      char key[100];
+      snprintf(key, sizeof(key), "%010d", table * 100 + i);
+      Put(key, std::to_string(table * 1000 + i));
+    }
+    Flush();
+  }
+
+  cfilter_skips = 0;
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  // Numberof skips in tables: 2, 3, 3, 3.
+  ASSERT_EQ(11, cfilter_skips);
+
+  for (int table = 0; table < 4; ++table) {
+    for (int i = table * 6; i < 39 + table * 11; ++i) {
+      int k = table * 100 + i;
+      char key[100];
+      snprintf(key, sizeof(key), "%010d", table * 100 + i);
+      auto expected = std::to_string(table * 1000 + i);
+      std::string val;
+      Status s = db_->Get(ReadOptions(), key, &val);
+      if (k / 10 % 2 == 0) {
+        ASSERT_TRUE(s.IsNotFound());
+      } else {
+        ASSERT_OK(s);
+        ASSERT_EQ(expected, val);
+      }
+    }
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -305,7 +305,8 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  dbfull()->TEST_WaitForCompact();
  // Preloading iterator issues one table cache lookup and creates
  // a new table reader. One file is created for flush and one for compaction.
-  // Compaction inputs make no table cache look-up.
+  // Compaction inputs make no table cache look-up for data/range deletion
+  // iterators
  ASSERT_EQ(num_table_cache_lookup, 2);
  // Create new iterator for:
  // (1) 1 for verifying flush results
@@ -326,7 +327,8 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
  cro.target_level = 2;
  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
  db_->CompactRange(cro, nullptr, nullptr);
-  // Only verifying compaction outputs issues one table cache lookup.
+  // Only verifying compaction outputs issues one table cache lookup
+  // for both data block and range deletion block).
  ASSERT_EQ(num_table_cache_lookup, 1);
  // One for compaction input, one for verifying compaction results.
  ASSERT_EQ(num_new_table_reader, 2);
@@ -1258,7 +1260,8 @@ TEST_F(DBCompactionTest, ManualPartialFill) {
  uint64_t target_size = 4 * options.max_bytes_for_level_base;
  for (int32_t i = 1; i < options.num_levels; i++) {
    ASSERT_LE(SizeAtLevel(i), target_size);
-    target_size *= options.max_bytes_for_level_multiplier;
+    target_size = static_cast<uint64_t>(target_size *
+                                        options.max_bytes_for_level_multiplier);
  }

  TEST_SYNC_POINT("DBCompaction::PartialFill:2");
@@ -1335,7 +1338,8 @@ TEST_F(DBCompactionTest, DeleteFileRange) {
  uint64_t target_size = 4 * options.max_bytes_for_level_base;
  for (int32_t i = 1; i < options.num_levels; i++) {
    ASSERT_LE(SizeAtLevel(i), target_size);
-    target_size *= options.max_bytes_for_level_multiplier;
+    target_size = static_cast<uint64_t>(target_size *
+                                        options.max_bytes_for_level_multiplier);
  }

  size_t old_num_files = CountFiles();
@@ -35,10 +35,10 @@ Status DBImpl::DisableFileDeletions() {
  InstrumentedMutexLock l(&mutex_);
  ++disable_delete_obsolete_files_;
  if (disable_delete_obsolete_files_ == 1) {
-    Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+    Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
        "File Deletions Disabled");
  } else {
-    Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
+    Log(InfoLogLevel::WARN_LEVEL, immutable_db_options_.info_log,
        "File Deletions Disabled, but already disabled. Counter: %d",
        disable_delete_obsolete_files_);
  }
@@ -59,12 +59,12 @@ Status DBImpl::EnableFileDeletions(bool force) {
      --disable_delete_obsolete_files_;
    }
    if (disable_delete_obsolete_files_ == 0)  {
-      Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+      Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
          "File Deletions Enabled");
      should_purge_files = true;
      FindObsoleteFiles(&job_context, true);
    } else {
-      Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
+      Log(InfoLogLevel::WARN_LEVEL, immutable_db_options_.info_log,
          "File Deletions Enable, but not really enabled. Counter: %d",
          disable_delete_obsolete_files_);
    }
@@ -73,7 +73,7 @@ Status DBImpl::EnableFileDeletions(bool force) {
    PurgeObsoleteFiles(job_context);
  }
  job_context.Clean();
-  LogFlush(db_options_.info_log);
+  LogFlush(immutable_db_options_.info_log);
  return Status::OK();
 }

@@ -110,7 +110,7 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,

    if (!status.ok()) {
      mutex_.Unlock();
-      Log(InfoLogLevel::ERROR_LEVEL, db_options_.info_log,
+      Log(InfoLogLevel::ERROR_LEVEL, immutable_db_options_.info_log,
          "Cannot Flush data %s\n", status.ToString().c_str());
      return status;
    }
@@ -13,6 +13,7 @@
 #include <functional>
 #include <limits>
 #include <list>
+#include <map>
 #include <queue>
 #include <set>
 #include <string>
@@ -22,6 +23,7 @@
 #include "db/column_family.h"
 #include "db/compaction_job.h"
 #include "db/dbformat.h"
+#include "db/external_sst_file_ingestion_job.h"
 #include "db/flush_job.h"
 #include "db/flush_scheduler.h"
 #include "db/internal_stats.h"
@@ -40,6 +42,7 @@
 #include "rocksdb/write_buffer_manager.h"
 #include "table/scoped_arena_iterator.h"
 #include "util/autovector.h"
+#include "util/db_options.h"
 #include "util/event_logger.h"
 #include "util/hash.h"
 #include "util/instrumented_mutex.h"
@@ -122,6 +125,10 @@ class DBImpl : public DB {
  using DB::GetProperty;
  virtual bool GetProperty(ColumnFamilyHandle* column_family,
                           const Slice& property, std::string* value) override;
+  using DB::GetMapProperty;
+  virtual bool GetMapProperty(ColumnFamilyHandle* column_family,
+                              const Slice& property,
+                              std::map<std::string, double>* value) override;
  using DB::GetIntProperty;
  virtual bool GetIntProperty(ColumnFamilyHandle* column_family,
                              const Slice& property, uint64_t* value) override;
@@ -155,6 +162,9 @@ class DBImpl : public DB {
      ColumnFamilyHandle* column_family,
      const std::unordered_map<std::string, std::string>& options_map) override;

+  virtual Status SetDBOptions(
+      const std::unordered_map<std::string, std::string>& options_map) override;
+
  using DB::NumberLevels;
  virtual int NumberLevels(ColumnFamilyHandle* column_family) override;
  using DB::MaxMemCompactionLevel;
@@ -165,10 +175,9 @@ class DBImpl : public DB {
  virtual const std::string& GetName() const override;
  virtual Env* GetEnv() const override;
  using DB::GetOptions;
-  virtual const Options& GetOptions(
-      ColumnFamilyHandle* column_family) const override;
+  virtual Options GetOptions(ColumnFamilyHandle* column_family) const override;
  using DB::GetDBOptions;
-  virtual const DBOptions& GetDBOptions() const override;
+  virtual DBOptions GetDBOptions() const override;
  using DB::Flush;
  virtual Status Flush(const FlushOptions& options,
                       ColumnFamilyHandle* column_family) override;
@@ -257,13 +266,11 @@ class DBImpl : public DB {
                                 bool cache_only, SequenceNumber* seq,
                                 bool* found_record_for_key);

-  using DB::AddFile;
-  virtual Status AddFile(ColumnFamilyHandle* column_family,
-                         const std::vector<ExternalSstFileInfo>& file_info_list,
-                         bool move_file, bool skip_snapshot_check) override;
-  virtual Status AddFile(ColumnFamilyHandle* column_family,
-                         const std::vector<std::string>& file_path_list,
-                         bool move_file, bool skip_snapshot_check) override;
+  using DB::IngestExternalFile;
+  virtual Status IngestExternalFile(
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& external_files,
+      const IngestExternalFileOptions& ingestion_options) override;

 #endif  // ROCKSDB_LITE

@@ -289,7 +296,8 @@ 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, ColumnFamilyHandle* column_family = nullptr);
+      Arena* arena, RangeDelAggregator* range_del_agg,
+      ColumnFamilyHandle* column_family = nullptr);

 #ifndef NDEBUG
  // Extra methods (for testing) that are not in the public DB interface
@@ -300,6 +308,16 @@ class DBImpl : public DB {
                           ColumnFamilyHandle* column_family = nullptr,
                           bool disallow_trivial_move = false);

+  void TEST_MaybeFlushColumnFamilies();
+
+  bool TEST_UnableToFlushOldestLog() {
+    return unable_to_flush_oldest_log_;
+  }
+
+  bool TEST_IsLogGettingFlushed() {
+    return alive_log_files_.begin()->getting_flushed;
+  }
+
  // Force current memtable contents to be flushed.
  Status TEST_FlushMemTable(bool wait = true,
                            ColumnFamilyHandle* cfh = nullptr);
@@ -360,6 +378,8 @@ class DBImpl : public DB {
  uint64_t TEST_FindMinLogContainingOutstandingPrep();
  uint64_t TEST_FindMinPrepLogReferencedByMemTable();

+  int TEST_BGCompactionsAllowed() const;
+
 #endif  // NDEBUG

  // Return maximum background compaction allowed to be scheduled based on
@@ -370,6 +390,8 @@ class DBImpl : public DB {
  // schedule a purge
  void ScheduleBgLogWriterClose(JobContext* job_context);

+  uint64_t MinLogNumberToKeep();
+
  // Returns the list of live files in 'live' and the list
  // of all files in the filesystem in 'candidate_files'.
  // If force == false and the last call was less than
@@ -391,6 +413,10 @@ class DBImpl : public DB {

  const SnapshotList& snapshots() const { return snapshots_; }

+  const ImmutableDBOptions& immutable_db_options() const {
+    return immutable_db_options_;
+  }
+
  void CancelAllBackgroundWork(bool wait);

  // Find Super version and reference it. Based on options, it might return
@@ -457,7 +483,7 @@ class DBImpl : public DB {
    ~RecoveredTransaction() { delete batch_; }
  };

-  bool allow_2pc() const { return db_options_.allow_2pc; }
+  bool allow_2pc() const { return immutable_db_options_.allow_2pc; }

  std::unordered_map<std::string, RecoveredTransaction*>
  recovered_transactions() {
@@ -508,7 +534,9 @@ class DBImpl : public DB {
  Env* const env_;
  const std::string dbname_;
  unique_ptr<VersionSet> versions_;
-  const DBOptions db_options_;
+  const DBOptions initial_db_options_;
+  const ImmutableDBOptions immutable_db_options_;
+  MutableDBOptions mutable_db_options_;
  Statistics* stats_;
  std::unordered_map<std::string, RecoveredTransaction*>
      recovered_transactions_;
@@ -516,7 +544,8 @@ class DBImpl : public DB {
  InternalIterator* NewInternalIterator(const ReadOptions&,
                                        ColumnFamilyData* cfd,
                                        SuperVersion* super_version,
-                                        Arena* arena);
+                                        Arena* arena,
+                                        RangeDelAggregator* range_del_agg);

  // Except in DB::Open(), WriteOptionsFile can only be called when:
  // 1. WriteThread::Writer::EnterUnbatched() is used.
@@ -540,6 +569,9 @@ class DBImpl : public DB {
  void NotifyOnMemTableSealed(ColumnFamilyData* cfd,
                              const MemTableInfo& mem_table_info);

+  void NotifyOnExternalFileIngested(
+      ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job);
+
  void NewThreadStatusCfInfo(ColumnFamilyData* cfd) const;

  void EraseThreadStatusCfInfo(ColumnFamilyData* cfd) const;
@@ -620,7 +652,7 @@ class DBImpl : public DB {

  // REQUIRES: log_numbers are sorted in ascending order
  Status RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
-                         SequenceNumber* max_sequence, bool read_only);
+                         SequenceNumber* next_sequence, bool read_only);

  // The following two methods are used to flush a memtable to
  // storage. The first one is used at database RecoveryTime (when the
@@ -632,33 +664,35 @@ class DBImpl : public DB {

  // num_bytes: for slowdown case, delay time is calculated based on
  //            `num_bytes` going through.
-  Status DelayWrite(uint64_t num_bytes);
+  Status DelayWrite(uint64_t num_bytes, const WriteOptions& write_options);

  Status ScheduleFlushes(WriteContext* context);

  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context);

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

  // Wait for memtable flushed
  Status WaitForFlushMemTable(ColumnFamilyData* cfd);

 #ifndef ROCKSDB_LITE
-  // Finds the lowest level in the DB that the ingested file can be added to
+
+  Status CompactFilesImpl(const CompactionOptions& compact_options,
+                          ColumnFamilyData* cfd, Version* version,
+                          const std::vector<std::string>& input_file_names,
+                          const int output_level, int output_path_id,
+                          JobContext* job_context, LogBuffer* log_buffer);
+
+  // Wait for current IngestExternalFile() calls to finish.
  // REQUIRES: mutex_ held
-  int PickLevelForIngestedFile(ColumnFamilyData* cfd,
-                               const ExternalSstFileInfo& file_info);
-
-  Status CompactFilesImpl(
-      const CompactionOptions& compact_options, ColumnFamilyData* cfd,
-      Version* version, const std::vector<std::string>& input_file_names,
-      const int output_level, int output_path_id, JobContext* job_context,
-      LogBuffer* log_buffer);
-  Status ReadExternalSstFileInfo(ColumnFamilyHandle* column_family,
-                                 const std::string& file_path,
-                                 ExternalSstFileInfo* file_info);
+  void WaitForIngestFile();

+#else
+  // IngestExternalFile is not supported in ROCKSDB_LITE so this function
+  // will be no-op
+  void WaitForIngestFile() {}
 #endif  // ROCKSDB_LITE

  ColumnFamilyData* GetColumnFamilyDataByName(const std::string& cf_name);
@@ -706,6 +740,14 @@ class DBImpl : public DB {

  const Snapshot* GetSnapshotImpl(bool is_write_conflict_boundary);

+  // Persist RocksDB options under the single write thread
+  // REQUIRES: mutex locked
+  Status PersistOptions();
+
+  void MaybeFlushColumnFamilies();
+
+  uint64_t GetMaxTotalWalSize() const;
+
  // table_cache_ provides its own synchronization
  std::shared_ptr<Cache> table_cache_;

@@ -717,7 +759,7 @@ class DBImpl : public DB {
  //       same time.
  InstrumentedMutex options_files_mutex_;
  // State below is protected by mutex_
-  InstrumentedMutex mutex_;
+  mutable InstrumentedMutex mutex_;

  std::atomic<bool> shutting_down_;
  // This condition variable is signaled on these conditions:
@@ -728,6 +770,7 @@ class DBImpl : public DB {
  // * whenever bg_flush_scheduled_ or bg_purge_scheduled_ value decreases
  // (i.e. whenever a flush is done, even if it didn't make any progress)
  // * whenever there is an error in background purge, flush or compaction
+  // * whenever num_running_ingest_file_ goes to 0.
  InstrumentedCondVar bg_cv_;
  uint64_t logfile_number_;
  std::deque<uint64_t>
@@ -960,6 +1003,15 @@ class DBImpl : public DB {
  // Used when disableWAL is true.
  bool has_unpersisted_data_;

+
+  // if an attempt was made to flush all column families that
+  // the oldest log depends on but uncommited data in the oldest
+  // log prevents the log from being released.
+  // We must attempt to free the dependent memtables again
+  // at a later time after the transaction in the oldest
+  // log is fully commited.
+  bool unable_to_flush_oldest_log_;
+
  static const int KEEP_LOG_FILE_NUM = 1000;
  // MSVC version 1800 still does not have constexpr for ::max()
  static const uint64_t kNoTimeOut = port::kMaxUint64;
@@ -969,9 +1021,9 @@ class DBImpl : public DB {
  // The options to access storage files
  const EnvOptions env_options_;

-  // A set of compactions that are running right now
+  // Number of running IngestExternalFile() calls.
  // REQUIRES: mutex held
-  std::unordered_set<Compaction*> running_compactions_;
+  int num_running_ingest_file_;

 #ifndef ROCKSDB_LITE
  WalManager wal_manager_;
@@ -1068,13 +1120,13 @@ class DBImpl : public DB {
  bool ShouldntRunManualCompaction(ManualCompaction* m);
  bool HaveManualCompaction(ColumnFamilyData* cfd);
  bool MCOverlap(ManualCompaction* m, ManualCompaction* m1);
+
+  size_t GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
 };

-// Sanitize db options.  The caller should delete result.info_log if
-// it is not equal to src.info_log.
 extern Options SanitizeOptions(const std::string& db,
-                               const InternalKeyComparator* icmp,
                               const Options& src);
+
 extern DBOptions SanitizeOptions(const std::string& db, const DBOptions& src);

 // Fix user-supplied options to be reasonable
@@ -1,400 +0,0 @@
-//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
-//  This source code is licensed under the BSD-style license found in the
-//  LICENSE file in the root directory of this source tree. An additional grant
-//  of patent rights can be found in the PATENTS file in the same directory.
-
-#include "db/db_impl.h"
-
-#ifndef __STDC_FORMAT_MACROS
-#define __STDC_FORMAT_MACROS
-#endif
-
-#include <inttypes.h>
-
-#include "db/builder.h"
-#include "rocksdb/db.h"
-#include "rocksdb/env.h"
-#include "rocksdb/sst_file_writer.h"
-#include "table/table_builder.h"
-#include "util/file_reader_writer.h"
-#include "util/file_util.h"
-#include "util/sync_point.h"
-
-namespace rocksdb {
-
-#ifndef ROCKSDB_LITE
-
-Status DBImpl::ReadExternalSstFileInfo(ColumnFamilyHandle* column_family,
-                                       const std::string& file_path,
-                                       ExternalSstFileInfo* file_info) {
-  Status status;
-  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-  auto cfd = cfh->cfd();
-
-  file_info->file_path = file_path;
-  status = env_->GetFileSize(file_path, &file_info->file_size);
-  if (!status.ok()) {
-    return status;
-  }
-
-  // Access the file using TableReader to extract
-  // version, number of entries, smallest user key, largest user key
-  std::unique_ptr<RandomAccessFile> sst_file;
-  status = env_->NewRandomAccessFile(file_path, &sst_file, env_options_);
-  if (!status.ok()) {
-    return status;
-  }
-  std::unique_ptr<RandomAccessFileReader> sst_file_reader;
-  sst_file_reader.reset(new RandomAccessFileReader(std::move(sst_file)));
-
-  std::unique_ptr<TableReader> table_reader;
-  status = cfd->ioptions()->table_factory->NewTableReader(
-      TableReaderOptions(*cfd->ioptions(), env_options_,
-                         cfd->internal_comparator()),
-      std::move(sst_file_reader), file_info->file_size, &table_reader);
-  if (!status.ok()) {
-    return status;
-  }
-
-  // Get the external sst file version from table properties
-  const UserCollectedProperties& user_collected_properties =
-      table_reader->GetTableProperties()->user_collected_properties;
-  UserCollectedProperties::const_iterator external_sst_file_version_iter =
-      user_collected_properties.find(ExternalSstFilePropertyNames::kVersion);
-  if (external_sst_file_version_iter == user_collected_properties.end()) {
-    return Status::InvalidArgument("Generated table version not found");
-  }
-
-  file_info->version =
-      DecodeFixed32(external_sst_file_version_iter->second.c_str());
-  if (file_info->version == 1) {
-    // version 1 imply that all sequence numbers in table equal 0
-    file_info->sequence_number = 0;
-  } else {
-    return Status::InvalidArgument("Generated table version is not supported");
-  }
-  // Get number of entries in table
-  file_info->num_entries = table_reader->GetTableProperties()->num_entries;
-
-  ParsedInternalKey key;
-  std::unique_ptr<InternalIterator> iter(
-      table_reader->NewIterator(ReadOptions()));
-
-  // Get first (smallest) key from file
-  iter->SeekToFirst();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("Generated table have corrupted keys");
-  }
-  if (key.sequence != 0) {
-    return Status::Corruption("Generated table have non zero sequence number");
-  }
-  file_info->smallest_key = key.user_key.ToString();
-
-  // Get last (largest) key from file
-  iter->SeekToLast();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("Generated table have corrupted keys");
-  }
-  if (key.sequence != 0) {
-    return Status::Corruption("Generated table have non zero sequence number");
-  }
-  file_info->largest_key = key.user_key.ToString();
-
-  return Status::OK();
-}
-
-Status DBImpl::AddFile(ColumnFamilyHandle* column_family,
-                       const std::vector<std::string>& file_path_list,
-                       bool move_file, bool skip_snapshot_check) {
-  Status status;
-  auto num_files = file_path_list.size();
-  if (num_files == 0) {
-    return Status::InvalidArgument("The list of files is empty");
-  }
-
-  std::vector<ExternalSstFileInfo> file_info_list(num_files);
-  for (size_t i = 0; i < num_files; i++) {
-    status = ReadExternalSstFileInfo(column_family, file_path_list[i],
-                                     &file_info_list[i]);
-    if (!status.ok()) {
-      return status;
-    }
-  }
-  return AddFile(column_family, file_info_list, move_file, skip_snapshot_check);
-}
-
-Status DBImpl::AddFile(ColumnFamilyHandle* column_family,
-                       const std::vector<ExternalSstFileInfo>& file_info_list,
-                       bool move_file, bool skip_snapshot_check) {
-  Status status;
-  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-  ColumnFamilyData* cfd = cfh->cfd();
-  const Comparator* user_cmp = cfd->internal_comparator().user_comparator();
-
-  auto num_files = file_info_list.size();
-  if (num_files == 0) {
-    return Status::InvalidArgument("The list of files is empty");
-  }
-
-  // Verify that passed files dont have overlapping ranges
-  if (num_files > 1) {
-    std::vector<const ExternalSstFileInfo*> sorted_file_info_list(num_files);
-    for (size_t i = 0; i < num_files; i++) {
-      sorted_file_info_list[i] = &file_info_list[i];
-    }
-
-    std::sort(sorted_file_info_list.begin(), sorted_file_info_list.end(),
-              [&user_cmp, &file_info_list](const ExternalSstFileInfo* info1,
-                                           const ExternalSstFileInfo* info2) {
-                return user_cmp->Compare(info1->smallest_key,
-                                         info2->smallest_key) < 0;
-              });
-
-    for (size_t i = 0; i < num_files - 1; i++) {
-      if (user_cmp->Compare(sorted_file_info_list[i]->largest_key,
-                            sorted_file_info_list[i + 1]->smallest_key) >= 0) {
-        return Status::NotSupported("Files have overlapping ranges");
-      }
-    }
-  }
-
-  std::vector<uint64_t> micro_list(num_files, 0);
-  std::vector<FileMetaData> meta_list(num_files);
-  for (size_t i = 0; i < num_files; i++) {
-    StopWatch sw(env_, nullptr, 0, &micro_list[i], false);
-    if (file_info_list[i].num_entries == 0) {
-      return Status::InvalidArgument("File contain no entries");
-    }
-    if (file_info_list[i].version != 1) {
-      return Status::InvalidArgument(
-          "Generated table version is not supported");
-    }
-    // version 1 imply that file have only Put Operations with Sequence Number =
-    // 0
-
-    meta_list[i].smallest =
-        InternalKey(file_info_list[i].smallest_key,
-                    file_info_list[i].sequence_number, ValueType::kTypeValue);
-    meta_list[i].largest =
-        InternalKey(file_info_list[i].largest_key,
-                    file_info_list[i].sequence_number, ValueType::kTypeValue);
-    if (!meta_list[i].smallest.Valid() || !meta_list[i].largest.Valid()) {
-      return Status::Corruption("Generated table have corrupted keys");
-    }
-    meta_list[i].smallest_seqno = file_info_list[i].sequence_number;
-    meta_list[i].largest_seqno = file_info_list[i].sequence_number;
-    if (meta_list[i].smallest_seqno != 0 || meta_list[i].largest_seqno != 0) {
-      return Status::InvalidArgument(
-          "Non zero sequence numbers are not supported");
-    }
-  }
-
-  std::vector<std::list<uint64_t>::iterator> pending_outputs_inserted_elem_list(
-      num_files);
-  // Generate locations for the new tables
-  {
-    InstrumentedMutexLock l(&mutex_);
-    for (size_t i = 0; i < num_files; i++) {
-      StopWatch sw(env_, nullptr, 0, &micro_list[i], false);
-      pending_outputs_inserted_elem_list[i] =
-          CaptureCurrentFileNumberInPendingOutputs();
-      meta_list[i].fd = FileDescriptor(versions_->NewFileNumber(), 0,
-                                       file_info_list[i].file_size);
-    }
-  }
-
-  // Copy/Move external files into DB
-  std::vector<std::string> db_fname_list(num_files);
-  size_t j = 0;
-  for (; j < num_files; j++) {
-    StopWatch sw(env_, nullptr, 0, &micro_list[j], false);
-    db_fname_list[j] =
-        TableFileName(db_options_.db_paths, meta_list[j].fd.GetNumber(),
-                      meta_list[j].fd.GetPathId());
-    if (move_file) {
-      status = env_->LinkFile(file_info_list[j].file_path, db_fname_list[j]);
-      if (status.IsNotSupported()) {
-        // Original file is on a different FS, use copy instead of hard linking
-        status =
-            CopyFile(env_, file_info_list[j].file_path, db_fname_list[j], 0);
-      }
-    } else {
-      status = CopyFile(env_, file_info_list[j].file_path, db_fname_list[j], 0);
-    }
-    TEST_SYNC_POINT("DBImpl::AddFile:FileCopied");
-    if (!status.ok()) {
-      for (size_t i = 0; i < j; i++) {
-        Status s = env_->DeleteFile(db_fname_list[i]);
-        if (!s.ok()) {
-          Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
-              "AddFile() clean up for file %s failed : %s",
-              db_fname_list[i].c_str(), s.ToString().c_str());
-        }
-      }
-      return status;
-    }
-  }
-
-  {
-    InstrumentedMutexLock l(&mutex_);
-    const MutableCFOptions mutable_cf_options =
-        *cfd->GetLatestMutableCFOptions();
-
-    WriteThread::Writer w;
-    write_thread_.EnterUnbatched(&w, &mutex_);
-
-    if (!skip_snapshot_check && !snapshots_.empty()) {
-      // Check that no snapshots are being held
-      status =
-          Status::NotSupported("Cannot add a file while holding snapshots");
-    }
-
-    if (status.ok()) {
-      // Verify that added file key range dont overlap with any keys in DB
-      SuperVersion* sv = cfd->GetSuperVersion()->Ref();
-      Arena arena;
-      ReadOptions ro;
-      ro.total_order_seek = true;
-      ScopedArenaIterator iter(NewInternalIterator(ro, cfd, sv, &arena));
-
-      for (size_t i = 0; i < num_files; i++) {
-        StopWatch sw(env_, nullptr, 0, &micro_list[i], false);
-        InternalKey range_start(file_info_list[i].smallest_key,
-                                kMaxSequenceNumber, kTypeValue);
-        iter->Seek(range_start.Encode());
-        status = iter->status();
-
-        if (status.ok() && iter->Valid()) {
-          ParsedInternalKey seek_result;
-          if (ParseInternalKey(iter->key(), &seek_result)) {
-            if (user_cmp->Compare(seek_result.user_key,
-                                  file_info_list[i].largest_key) <= 0) {
-              status = Status::NotSupported("Cannot add overlapping range");
-              break;
-            }
-          } else {
-            status = Status::Corruption("DB have corrupted keys");
-            break;
-          }
-        }
-      }
-    }
-
-    // The levels that the files will be ingested into
-    std::vector<int> target_level_list(num_files, 0);
-    if (status.ok()) {
-      VersionEdit edit;
-      edit.SetColumnFamily(cfd->GetID());
-      for (size_t i = 0; i < num_files; i++) {
-        StopWatch sw(env_, nullptr, 0, &micro_list[i], false);
-        // Add file to the lowest possible level
-        target_level_list[i] = PickLevelForIngestedFile(cfd, file_info_list[i]);
-        edit.AddFile(target_level_list[i], meta_list[i].fd.GetNumber(),
-                     meta_list[i].fd.GetPathId(), meta_list[i].fd.GetFileSize(),
-                     meta_list[i].smallest, meta_list[i].largest,
-                     meta_list[i].smallest_seqno, meta_list[i].largest_seqno,
-                     meta_list[i].marked_for_compaction);
-      }
-      status = versions_->LogAndApply(cfd, mutable_cf_options, &edit, &mutex_,
-                                      directories_.GetDbDir());
-    }
-    write_thread_.ExitUnbatched(&w);
-
-    if (status.ok()) {
-      delete InstallSuperVersionAndScheduleWork(cfd, nullptr,
-                                                mutable_cf_options);
-    }
-    for (size_t i = 0; i < num_files; i++) {
-      // Update internal stats
-      InternalStats::CompactionStats stats(1);
-      stats.micros = micro_list[i];
-      stats.bytes_written = meta_list[i].fd.GetFileSize();
-      stats.num_output_files = 1;
-      cfd->internal_stats()->AddCompactionStats(target_level_list[i], stats);
-      cfd->internal_stats()->AddCFStats(InternalStats::BYTES_INGESTED_ADD_FILE,
-                                        meta_list[i].fd.GetFileSize());
-      ReleaseFileNumberFromPendingOutputs(
-          pending_outputs_inserted_elem_list[i]);
-    }
-  }
-
-  if (!status.ok()) {
-    // We failed to add the files to the database
-    for (size_t i = 0; i < num_files; i++) {
-      Status s = env_->DeleteFile(db_fname_list[i]);
-      if (!s.ok()) {
-        Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
-            "AddFile() clean up for file %s failed : %s",
-            db_fname_list[i].c_str(), s.ToString().c_str());
-      }
-    }
-  } else if (status.ok() && move_file) {
-    // The files were moved and added successfully, remove original file links
-    for (size_t i = 0; i < num_files; i++) {
-      Status s = env_->DeleteFile(file_info_list[i].file_path);
-      if (!s.ok()) {
-        Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
-            "%s was added to DB successfully but failed to remove original "
-            "file "
-            "link : %s",
-            file_info_list[i].file_path.c_str(), s.ToString().c_str());
-      }
-    }
-  }
-  return status;
-}
-
-// Finds the lowest level in the DB that the ingested file can be added to
-int DBImpl::PickLevelForIngestedFile(ColumnFamilyData* cfd,
-                                     const ExternalSstFileInfo& file_info) {
-  mutex_.AssertHeld();
-
-  int target_level = 0;
-  auto* vstorage = cfd->current()->storage_info();
-  auto* ucmp = vstorage->InternalComparator()->user_comparator();
-
-  Slice file_smallest_user_key(file_info.smallest_key);
-  Slice file_largest_user_key(file_info.largest_key);
-
-  for (int lvl = cfd->NumberLevels() - 1; lvl >= vstorage->base_level();
-       lvl--) {
-    if (vstorage->OverlapInLevel(lvl, &file_smallest_user_key,
-                                 &file_largest_user_key) == false) {
-      // Make sure that the file dont overlap with the output of any
-      // compaction running right now
-      Slice compaction_smallest_user_key;
-      Slice compaction_largest_user_key;
-      bool overlap_with_compaction_output = false;
-      for (Compaction* c : running_compactions_) {
-        if (c->column_family_data()->GetID() != cfd->GetID() ||
-            c->output_level() != lvl) {
-          continue;
-        }
-
-        compaction_smallest_user_key = c->GetSmallestUserKey();
-        compaction_largest_user_key = c->GetLargestUserKey();
-
-        if (ucmp->Compare(file_smallest_user_key,
-                          compaction_largest_user_key) <= 0 &&
-            ucmp->Compare(file_largest_user_key,
-                          compaction_smallest_user_key) >= 0) {
-          overlap_with_compaction_output = true;
-          break;
-        }
-      }
-
-      if (overlap_with_compaction_output == false) {
-        // Level lvl is the lowest level that dont have any files with key
-        // range overlapping with our file key range and no compactions
-        // planning to add overlapping files in it.
-        target_level = lvl;
-        break;
-      }
-    }
-  }
-
-  return target_level;
-}
-#endif  // ROCKSDB_LITE
-
-}  // namespace rocksdb
@@ -19,6 +19,11 @@ uint64_t DBImpl::TEST_GetLevel0TotalSize() {
  return default_cf_handle_->cfd()->current()->storage_info()->NumLevelBytes(0);
 }

+void DBImpl::TEST_MaybeFlushColumnFamilies() {
+  InstrumentedMutexLock l(&mutex_);
+  MaybeFlushColumnFamilies();
+}
+
 int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
    ColumnFamilyHandle* column_family) {
  ColumnFamilyData* cfd;
@@ -178,5 +183,10 @@ Status DBImpl::TEST_GetLatestMutableCFOptions(
  return Status::OK();
 }

+int DBImpl::TEST_BGCompactionsAllowed() const {
+  InstrumentedMutexLock l(&mutex_);
+  return BGCompactionsAllowed();
+}
+
 }  // namespace rocksdb
 #endif  // NDEBUG
@@ -49,7 +49,8 @@ Status DBImpl::SuggestCompactRange(ColumnFamilyHandle* column_family,
    }
    // Since we have some more files to compact, we should also recompute
    // compaction score
-    vstorage->ComputeCompactionScore(*cfd->GetLatestMutableCFOptions());
+    vstorage->ComputeCompactionScore(*cfd->ioptions(),
+                                     *cfd->GetLatestMutableCFOptions());
    SchedulePendingCompaction(cfd);
    MaybeScheduleFlushOrCompaction();
  }
@@ -60,7 +61,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
  assert(column_family);

  if (target_level < 1) {
-    Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+    Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
        "PromoteL0 FAILED. Invalid target level %d\n", target_level);
    return Status::InvalidArgument("Invalid target level");
  }
@@ -74,7 +75,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
    const auto* vstorage = cfd->current()->storage_info();

    if (target_level >= vstorage->num_levels()) {
-      Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+      Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
          "PromoteL0 FAILED. Target level %d does not exist\n", target_level);
      job_context.Clean();
      return Status::InvalidArgument("Target level does not exist");
@@ -93,7 +94,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
    for (size_t i = 0; i < l0_files.size(); ++i) {
      auto f = l0_files[i];
      if (f->being_compacted) {
-        Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+        Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
            "PromoteL0 FAILED. File %" PRIu64 " being compacted\n",
            f->fd.GetNumber());
        job_context.Clean();
@@ -103,7 +104,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
      if (i == 0) continue;
      auto prev_f = l0_files[i - 1];
      if (icmp->Compare(prev_f->largest, f->smallest) >= 0) {
-        Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+        Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
            "PromoteL0 FAILED. Files %" PRIu64 " and %" PRIu64
            " have overlapping ranges\n",
            prev_f->fd.GetNumber(), f->fd.GetNumber());
@@ -115,7 +116,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
    // Check that all levels up to target_level are empty.
    for (int level = 1; level <= target_level; ++level) {
      if (vstorage->NumLevelFiles(level) > 0) {
-        Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+        Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
            "PromoteL0 FAILED. Level %d not empty\n", level);
        job_context.Clean();
        return Status::InvalidArgument(
@@ -140,7 +141,7 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
          cfd, &job_context, *cfd->GetLatestMutableCFOptions());
    }
  }  // lock released here
-  LogFlush(db_options_.info_log);
+  LogFlush(immutable_db_options_.info_log);
  job_context.Clean();

  return status;
@@ -8,8 +8,9 @@

 #include "db/compacted_db_impl.h"
 #include "db/db_impl.h"
-#include "db/merge_context.h"
 #include "db/db_iter.h"
+#include "db/merge_context.h"
+#include "db/range_del_aggregator.h"
 #include "util/perf_context_imp.h"

 namespace rocksdb {
@@ -19,8 +20,9 @@ namespace rocksdb {
 DBImplReadOnly::DBImplReadOnly(const DBOptions& db_options,
                               const std::string& dbname)
    : DBImpl(db_options, dbname) {
-  Log(INFO_LEVEL, db_options_.info_log, "Opening the db in read only mode");
-  LogFlush(db_options_.info_log);
+  Log(INFO_LEVEL, immutable_db_options_.info_log,
+      "Opening the db in read only mode");
+  LogFlush(immutable_db_options_.info_log);
 }

 DBImplReadOnly::~DBImplReadOnly() {
@@ -36,11 +38,14 @@ Status DBImplReadOnly::Get(const ReadOptions& read_options,
  auto cfd = cfh->cfd();
  SuperVersion* super_version = cfd->GetSuperVersion();
  MergeContext merge_context;
+  RangeDelAggregator range_del_agg(cfd->internal_comparator(), snapshot);
  LookupKey lkey(key, snapshot);
-  if (super_version->mem->Get(lkey, value, &s, &merge_context)) {
+  if (super_version->mem->Get(lkey, value, &s, &merge_context, &range_del_agg,
+                              read_options)) {
  } else {
    PERF_TIMER_GUARD(get_from_output_files_time);
-    super_version->current->Get(read_options, lkey, value, &s, &merge_context);
+    super_version->current->Get(read_options, lkey, value, &s, &merge_context,
+                                &range_del_agg);
  }
  return s;
 }
@@ -59,8 +64,9 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
           : latest_snapshot),
      super_version->mutable_cf_options.max_sequential_skip_in_iterations,
      super_version->version_number);
-  auto internal_iter = NewInternalIterator(
-      read_options, cfd, super_version, db_iter->GetArena());
+  auto internal_iter =
+      NewInternalIterator(read_options, cfd, super_version, db_iter->GetArena(),
+                          db_iter->GetRangeDelAggregator());
  db_iter->SetIterUnderDBIter(internal_iter);
  return db_iter;
 }
@@ -87,8 +93,9 @@ Status DBImplReadOnly::NewIterators(
             : latest_snapshot),
        sv->mutable_cf_options.max_sequential_skip_in_iterations,
        sv->version_number);
-    auto* internal_iter = NewInternalIterator(
-        read_options, cfd, sv, db_iter->GetArena());
+    auto* internal_iter =
+        NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
+                            db_iter->GetRangeDelAggregator());
    db_iter->SetIterUnderDBIter(internal_iter);
    iterators->push_back(db_iter);
  }
@@ -16,12 +16,12 @@
 #include <vector>

 #include "db/filename.h"
-#include "rocksdb/options.h"
 #include "rocksdb/env.h"

 namespace rocksdb {

-void DumpDBFileSummary(const DBOptions& options, const std::string& dbname) {
+void DumpDBFileSummary(const ImmutableDBOptions& options,
+                       const std::string& dbname) {
  if (options.info_log == nullptr) {
    return;
  }
@@ -6,8 +6,9 @@

 #include <string>

-#include "rocksdb/options.h"
+#include "util/db_options.h"

 namespace rocksdb {
-void DumpDBFileSummary(const DBOptions& options, const std::string& dbname);
+void DumpDBFileSummary(const ImmutableDBOptions& options,
+                       const std::string& dbname);
 }  // namespace rocksdb
@@ -23,6 +23,8 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdate) {
    options.inplace_update_support = true;
    options.env = env_;
    options.write_buffer_size = 100000;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller size
@@ -45,6 +47,8 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateLargeNewValue) {
    options.inplace_update_support = true;
    options.env = env_;
    options.write_buffer_size = 100000;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of larger size
@@ -70,6 +74,8 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerSize;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller size
@@ -97,6 +103,8 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerVarintSize;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of smaller varint size
@@ -124,6 +132,8 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
    options.write_buffer_size = 100000;
    options.inplace_callback =
      rocksdb::DBTestInPlaceUpdate::updateInPlaceLargerSize;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Update key with values of larger size
@@ -148,7 +158,9 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackNoAction) {
    options.env = env_;
    options.write_buffer_size = 100000;
    options.inplace_callback =
-      rocksdb::DBTestInPlaceUpdate::updateInPlaceNoAction;
+        rocksdb::DBTestInPlaceUpdate::updateInPlaceNoAction;
+    options.allow_concurrent_memtable_write = false;
+    Reopen(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Callback function requests no actions from db
@@ -33,7 +33,7 @@ TEST_F(DBIOFailureTest, DropWrites) {
    // Force out-of-space errors
    env_->drop_writes_.store(true, std::memory_order_release);
    env_->sleep_counter_.Reset();
-    env_->no_sleep_ = true;
+    env_->no_slowdown_ = true;
    for (int i = 0; i < 5; i++) {
      if (option_config_ != kUniversalCompactionMultiLevel &&
          option_config_ != kUniversalSubcompactions) {
@@ -105,7 +105,8 @@ class DBIter: public Iterator {
         InternalIterator* iter, SequenceNumber s, bool arena_mode,
         uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
         const Slice* iterate_upper_bound = nullptr,
-         bool prefix_same_as_start = false, bool pin_data = false)
+         bool prefix_same_as_start = false, bool pin_data = false,
+         bool total_order_seek = false)
      : arena_mode_(arena_mode),
        env_(env),
        logger_(ioptions.info_log),
@@ -120,7 +121,9 @@ class DBIter: public Iterator {
        version_number_(version_number),
        iterate_upper_bound_(iterate_upper_bound),
        prefix_same_as_start_(prefix_same_as_start),
-        pin_thru_lifetime_(pin_data) {
+        pin_thru_lifetime_(pin_data),
+        total_order_seek_(total_order_seek),
+        range_del_agg_(ioptions.internal_comparator, s) {
    RecordTick(statistics_, NO_ITERATORS);
    prefix_extractor_ = ioptions.prefix_extractor;
    max_skip_ = max_sequential_skip_in_iterations;
@@ -149,6 +152,10 @@ class DBIter: public Iterator {
    iter_ = iter;
    iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
  }
+  virtual RangeDelAggregator* GetRangeDelAggregator() {
+    return &range_del_agg_;
+  }
+
  virtual bool Valid() const override { return valid_; }
  virtual Slice key() const override {
    assert(valid_);
@@ -199,10 +206,12 @@ class DBIter: public Iterator {
  virtual void Next() override;
  virtual void Prev() override;
  virtual void Seek(const Slice& target) override;
+  virtual void SeekForPrev(const Slice& target) override;
  virtual void SeekToFirst() override;
  virtual void SeekToLast() override;

 private:
+  void ReverseToForward();
  void ReverseToBackward();
  void PrevInternal();
  void FindParseableKey(ParsedInternalKey* ikey, Direction direction);
@@ -255,6 +264,7 @@ class DBIter: public Iterator {
  Direction direction_;
  bool valid_;
  bool current_entry_is_merged_;
+  // for prefix seek mode to support prev()
  Statistics* statistics_;
  uint64_t max_skip_;
  uint64_t version_number_;
@@ -265,8 +275,10 @@ class DBIter: public Iterator {
  // Means that we will pin all data blocks we read as long the Iterator
  // is not deleted, will be true if ReadOptions::pin_data is true
  const bool pin_thru_lifetime_;
+  const bool total_order_seek_;
  // List of operands for merge operator.
  MergeContext merge_context_;
+  RangeDelAggregator range_del_agg_;
  LocalStatistics local_stats_;
  PinnedIteratorsManager pinned_iters_mgr_;

@@ -293,11 +305,7 @@ void DBIter::Next() {
  // Release temporarily pinned blocks from last operation
  ReleaseTempPinnedData();
  if (direction_ == kReverse) {
-    FindNextUserKey();
-    direction_ = kForward;
-    if (!iter_->Valid()) {
-      iter_->SeekToFirst();
-    }
+    ReverseToForward();
  } else if (iter_->Valid() && !current_entry_is_merged_) {
    // If the current value is not a merge, the iter position is the
    // current key, which is already returned. We can safely issue a
@@ -331,7 +339,7 @@ void DBIter::Next() {
 //           saved_value_             => the merged value
 //
 // NOTE: In between, saved_key_ can point to a user key that has
-//       a delete marker
+//       a delete marker or a sequence number higher than sequence_
 //
 // The prefix_check parameter controls whether we check the iterated
 // keys against the prefix of the seeked key. Set to false when
@@ -348,71 +356,136 @@ void DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
  assert(iter_->Valid());
  assert(direction_ == kForward);
  current_entry_is_merged_ = false;
+
+  // How many times in a row we have skipped an entry with user key less than
+  // or equal to saved_key_. We could skip these entries either because
+  // sequence numbers were too high or because skipping = true.
+  // What saved_key_ contains throughout this method:
+  //  - if skipping        : saved_key_ contains the key that we need to skip,
+  //                         and we haven't seen any keys greater than that,
+  //  - if num_skipped > 0 : saved_key_ contains the key that we have skipped
+  //                         num_skipped times, and we haven't seen any keys
+  //                         greater than that,
+  //  - none of the above  : saved_key_ can contain anything, it doesn't matter.
  uint64_t num_skipped = 0;
+
  do {
    ParsedInternalKey ikey;

-    if (ParseKey(&ikey)) {
-      if (iterate_upper_bound_ != nullptr &&
-          user_comparator_->Compare(ikey.user_key, *iterate_upper_bound_) >= 0) {
-        break;
-      }
+    if (!ParseKey(&ikey)) {
+      // Skip corrupted keys.
+      iter_->Next();
+      continue;
+    }

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

-      if (ikey.sequence <= sequence_) {
-        if (skipping &&
-           user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
-          num_skipped++;  // skip this entry
-          PERF_COUNTER_ADD(internal_key_skipped_count, 1);
-        } else {
-          switch (ikey.type) {
-            case kTypeDeletion:
-            case kTypeSingleDeletion:
+    if (prefix_extractor_ && prefix_check &&
+        prefix_extractor_->Transform(ikey.user_key)
+          .compare(prefix_start_key_) != 0) {
+      break;
+    }
+
+    if (ikey.sequence <= sequence_) {
+      if (skipping &&
+          user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
+        num_skipped++;  // skip this entry
+        PERF_COUNTER_ADD(internal_key_skipped_count, 1);
+      } else {
+        num_skipped = 0;
+        switch (ikey.type) {
+          case kTypeDeletion:
+          case kTypeSingleDeletion:
+            // Arrange to skip all upcoming entries for this key since
+            // they are hidden by this deletion.
+            saved_key_.SetKey(
+                ikey.user_key,
+                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
+            skipping = true;
+            PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+            break;
+          case kTypeValue:
+            saved_key_.SetKey(
+                ikey.user_key,
+                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
+            if (range_del_agg_.ShouldDelete(ikey)) {
              // Arrange to skip all upcoming entries for this key since
              // they are hidden by this deletion.
-              saved_key_.SetKey(
-                  ikey.user_key,
-                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
              skipping = true;
              num_skipped = 0;
              PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
-              break;
-            case kTypeValue:
+            } else {
              valid_ = true;
-              saved_key_.SetKey(
-                  ikey.user_key,
-                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
              return;
-            case kTypeMerge:
-              // By now, we are sure the current ikey is going to yield a value
-              saved_key_.SetKey(
-                  ikey.user_key,
-                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
+            }
+            break;
+          case kTypeMerge:
+            saved_key_.SetKey(
+                ikey.user_key,
+                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
+            if (range_del_agg_.ShouldDelete(ikey)) {
+              // Arrange to skip all upcoming entries for this key since
+              // they are hidden by this deletion.
+              skipping = true;
+              num_skipped = 0;
+              PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+            } else {
+              // By now, we are sure the current ikey is going to yield a
+              // value
              current_entry_is_merged_ = true;
              valid_ = true;
              MergeValuesNewToOld();  // Go to a different state machine
              return;
-            default:
-              assert(false);
-              break;
-          }
+            }
+            break;
+          default:
+            assert(false);
+            break;
        }
      }
+    } else {
+      // This key was inserted after our snapshot was taken.
+      PERF_COUNTER_ADD(internal_recent_skipped_count, 1);
+
+      // Here saved_key_ may contain some old key, or the default empty key, or
+      // key assigned by some random other method. We don't care.
+      if (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
+        num_skipped++;
+      } else {
+        saved_key_.SetKey(
+                ikey.user_key,
+                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
+        skipping = false;
+        num_skipped = 0;
+      }
    }
-    // If we have sequentially iterated via numerous keys and still not
-    // found the next user-key, then it is better to seek so that we can
-    // avoid too many key comparisons. We seek to the last occurrence of
-    // our current key by looking for sequence number 0 and type deletion
-    // (the smallest type).
-    if (skipping && num_skipped > max_skip_) {
+
+    // If we have sequentially iterated via numerous equal keys, then it's
+    // better to seek so that we can avoid too many key comparisons.
+    if (num_skipped > max_skip_) {
      num_skipped = 0;
      std::string last_key;
-      AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(), 0,
-                                                     kTypeDeletion));
+      if (skipping) {
+        // We're looking for the next user-key but all we see are the same
+        // user-key with decreasing sequence numbers. Fast forward to
+        // sequence number 0 and type deletion (the smallest type).
+        AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(), 0,
+                                                       kTypeDeletion));
+        // Don't set skipping = false because we may still see more user-keys
+        // equal to saved_key_.
+      } else {
+        // We saw multiple entries with this user key and sequence numbers
+        // higher than sequence_. Fast forward to sequence_.
+        // Note that this only covers a case when a higher key was overwritten
+        // many times since our snapshot was taken, not the case when a lot of
+        // different keys were inserted after our snapshot was taken.
+        AppendInternalKey(&last_key,
+                          ParsedInternalKey(saved_key_.GetKey(), sequence_,
+                                            kValueTypeForSeek));
+      }
      iter_->Seek(last_key);
      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
    } else {
@@ -445,6 +518,7 @@ void DBIter::MergeValuesNewToOld() {
                             iter_->IsValuePinned() /* operand_pinned */);

  ParsedInternalKey ikey;
+  Status s;
  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
    if (!ParseKey(&ikey)) {
      // skip corrupted key
@@ -454,7 +528,8 @@ void DBIter::MergeValuesNewToOld() {
    if (!user_comparator_->Equal(ikey.user_key, saved_key_.GetKey())) {
      // hit the next user key, stop right here
      break;
-    } else if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type) {
+    } else if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type ||
+               range_del_agg_.ShouldDelete(ikey)) {
      // hit a delete with the same user key, stop right here
      // iter_ is positioned after delete
      iter_->Next();
@@ -464,9 +539,12 @@ void DBIter::MergeValuesNewToOld() {
      // final result in saved_value_. We are done!
      // ignore corruption if there is any.
      const Slice val = iter_->value();
-      MergeHelper::TimedFullMerge(merge_operator_, ikey.user_key, &val,
-                                  merge_context_.GetOperands(), &saved_value_,
-                                  logger_, statistics_, env_, &pinned_value_);
+      s = MergeHelper::TimedFullMerge(
+          merge_operator_, ikey.user_key, &val, merge_context_.GetOperands(),
+          &saved_value_, logger_, statistics_, env_, &pinned_value_);
+      if (!s.ok()) {
+        status_ = s;
+      }
      // iter_ is positioned after put
      iter_->Next();
      return;
@@ -475,6 +553,7 @@ void DBIter::MergeValuesNewToOld() {
      // when complete, add result to operands and continue.
      merge_context_.PushOperand(iter_->value(),
                                 iter_->IsValuePinned() /* operand_pinned */);
+      PERF_COUNTER_ADD(internal_merge_count, 1);
    } else {
      assert(false);
    }
@@ -484,9 +563,12 @@ void DBIter::MergeValuesNewToOld() {
  // a deletion marker.
  // feed null as the existing value to the merge operator, such that
  // client can differentiate this scenario and do things accordingly.
-  MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
-                              merge_context_.GetOperands(), &saved_value_,
-                              logger_, statistics_, env_, &pinned_value_);
+  s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
+                                  merge_context_.GetOperands(), &saved_value_,
+                                  logger_, statistics_, env_, &pinned_value_);
+  if (!s.ok()) {
+    status_ = s;
+  }
 }

 void DBIter::Prev() {
@@ -503,14 +585,29 @@ void DBIter::Prev() {
      local_stats_.bytes_read_ += (key().size() + value().size());
    }
  }
-  if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
-      prefix_extractor_->Transform(saved_key_.GetKey())
-              .compare(prefix_start_key_) != 0) {
-    valid_ = false;
+}
+
+void DBIter::ReverseToForward() {
+  if (prefix_extractor_ != nullptr && !total_order_seek_) {
+    IterKey last_key;
+    last_key.SetInternalKey(ParsedInternalKey(
+        saved_key_.GetKey(), kMaxSequenceNumber, kValueTypeForSeek));
+    iter_->Seek(last_key.GetKey());
+  }
+  FindNextUserKey();
+  direction_ = kForward;
+  if (!iter_->Valid()) {
+    iter_->SeekToFirst();
  }
 }

 void DBIter::ReverseToBackward() {
+  if (prefix_extractor_ != nullptr && !total_order_seek_) {
+    IterKey last_key;
+    last_key.SetInternalKey(
+        ParsedInternalKey(saved_key_.GetKey(), 0, kValueTypeForSeekForPrev));
+    iter_->SeekForPrev(last_key.GetKey());
+  }
  if (current_entry_is_merged_) {
    // Not placed in the same key. Need to call Prev() until finding the
    // previous key.
@@ -521,6 +618,11 @@ void DBIter::ReverseToBackward() {
    FindParseableKey(&ikey, kReverse);
    while (iter_->Valid() &&
           user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) > 0) {
+      if (ikey.sequence > sequence_) {
+        PERF_COUNTER_ADD(internal_recent_skipped_count, 1);
+      } else {
+        PERF_COUNTER_ADD(internal_key_skipped_count, 1);
+      }
      iter_->Prev();
      FindParseableKey(&ikey, kReverse);
    }
@@ -557,6 +659,11 @@ void DBIter::PrevInternal() {
      if (user_comparator_->Equal(ikey.user_key, saved_key_.GetKey())) {
        FindPrevUserKey();
      }
+      if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
+          prefix_extractor_->Transform(saved_key_.GetKey())
+                  .compare(prefix_start_key_) != 0) {
+        valid_ = false;
+      }
      return;
    }
    if (!iter_->Valid()) {
@@ -568,6 +675,7 @@ void DBIter::PrevInternal() {
    }
  }
  // We haven't found any key - iterator is not valid
+  // Or the prefix is different than start prefix
  assert(!iter_->Valid());
  valid_ = false;
 }
@@ -601,10 +709,15 @@ bool DBIter::FindValueForCurrentKey() {
    last_key_entry_type = ikey.type;
    switch (last_key_entry_type) {
      case kTypeValue:
+        if (range_del_agg_.ShouldDelete(ikey)) {
+          last_key_entry_type = kTypeRangeDeletion;
+          PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+        } else {
+          assert(iter_->IsValuePinned());
+          pinned_value_ = iter_->value();
+        }
        merge_context_.Clear();
-        assert(iter_->IsValuePinned());
-        pinned_value_ = iter_->value();
-        last_not_merge_type = kTypeValue;
+        last_not_merge_type = last_key_entry_type;
        break;
      case kTypeDeletion:
      case kTypeSingleDeletion:
@@ -613,9 +726,17 @@ bool DBIter::FindValueForCurrentKey() {
        PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
        break;
      case kTypeMerge:
-        assert(merge_operator_ != nullptr);
-        merge_context_.PushOperandBack(
-            iter_->value(), iter_->IsValuePinned() /* operand_pinned */);
+        if (range_del_agg_.ShouldDelete(ikey)) {
+          merge_context_.Clear();
+          last_key_entry_type = kTypeRangeDeletion;
+          last_not_merge_type = last_key_entry_type;
+          PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+        } else {
+          assert(merge_operator_ != nullptr);
+          merge_context_.PushOperandBack(
+              iter_->value(), iter_->IsValuePinned() /* operand_pinned */);
+          PERF_COUNTER_ADD(internal_merge_count, 1);
+        }
        break;
      default:
        assert(false);
@@ -628,24 +749,28 @@ bool DBIter::FindValueForCurrentKey() {
    FindParseableKey(&ikey, kReverse);
  }

+  Status s;
  switch (last_key_entry_type) {
    case kTypeDeletion:
    case kTypeSingleDeletion:
+    case kTypeRangeDeletion:
      valid_ = false;
      return false;
    case kTypeMerge:
      current_entry_is_merged_ = true;
-      if (last_not_merge_type == kTypeDeletion) {
-        MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
-                                    nullptr, merge_context_.GetOperands(),
-                                    &saved_value_, logger_, statistics_, env_,
-                                    &pinned_value_);
+      if (last_not_merge_type == kTypeDeletion ||
+          last_not_merge_type == kTypeSingleDeletion ||
+          last_not_merge_type == kTypeRangeDeletion) {
+        s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
+                                        nullptr, merge_context_.GetOperands(),
+                                        &saved_value_, logger_, statistics_,
+                                        env_, &pinned_value_);
      } else {
        assert(last_not_merge_type == kTypeValue);
-        MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
-                                    &pinned_value_,
-                                    merge_context_.GetOperands(), &saved_value_,
-                                    logger_, statistics_, env_, &pinned_value_);
+        s = MergeHelper::TimedFullMerge(
+            merge_operator_, saved_key_.GetKey(), &pinned_value_,
+            merge_context_.GetOperands(), &saved_value_, logger_, statistics_,
+            env_, &pinned_value_);
      }
      break;
    case kTypeValue:
@@ -656,6 +781,9 @@ bool DBIter::FindValueForCurrentKey() {
      break;
  }
  valid_ = true;
+  if (!s.ok()) {
+    status_ = s;
+  }
  return true;
 }

@@ -675,17 +803,17 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  ParsedInternalKey ikey;
  FindParseableKey(&ikey, kForward);

-  if (ikey.type == kTypeValue || ikey.type == kTypeDeletion ||
-      ikey.type == kTypeSingleDeletion) {
-    if (ikey.type == kTypeValue) {
-      assert(iter_->IsValuePinned());
-      pinned_value_ = iter_->value();
-      valid_ = true;
-      return true;
-    }
+  if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
+      range_del_agg_.ShouldDelete(ikey)) {
    valid_ = false;
    return false;
  }
+  if (ikey.type == kTypeValue) {
+    assert(iter_->IsValuePinned());
+    pinned_value_ = iter_->value();
+    valid_ = true;
+    return true;
+  }

  // kTypeMerge. We need to collect all kTypeMerge values and save them
  // in operands
@@ -693,19 +821,23 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  merge_context_.Clear();
  while (iter_->Valid() &&
         user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) &&
-         ikey.type == kTypeMerge) {
+         ikey.type == kTypeMerge && !range_del_agg_.ShouldDelete(ikey)) {
    merge_context_.PushOperand(iter_->value(),
                               iter_->IsValuePinned() /* operand_pinned */);
+    PERF_COUNTER_ADD(internal_merge_count, 1);
    iter_->Next();
    FindParseableKey(&ikey, kForward);
  }

+  Status s;
  if (!iter_->Valid() ||
      !user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) ||
-      ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion) {
-    MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
-                                merge_context_.GetOperands(), &saved_value_,
-                                logger_, statistics_, env_, &pinned_value_);
+      ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
+      range_del_agg_.ShouldDelete(ikey)) {
+    s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
+                                    nullptr, merge_context_.GetOperands(),
+                                    &saved_value_, logger_, statistics_, env_,
+                                    &pinned_value_);
    // Make iter_ valid and point to saved_key_
    if (!iter_->Valid() ||
        !user_comparator_->Equal(ikey.user_key, saved_key_.GetKey())) {
@@ -713,14 +845,20 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
    }
    valid_ = true;
+    if (!s.ok()) {
+      status_ = s;
+    }
    return true;
  }

  const Slice& val = iter_->value();
-  MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), &val,
-                              merge_context_.GetOperands(), &saved_value_,
-                              logger_, statistics_, env_, &pinned_value_);
+  s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), &val,
+                                  merge_context_.GetOperands(), &saved_value_,
+                                  logger_, statistics_, env_, &pinned_value_);
  valid_ = true;
+  if (!s.ok()) {
+    status_ = s;
+  }
  return true;
 }

@@ -765,6 +903,11 @@ void DBIter::FindPrevUserKey() {
        ++num_skipped;
      }
    }
+    if (ikey.sequence > sequence_) {
+      PERF_COUNTER_ADD(internal_recent_skipped_count, 1);
+    } else {
+      PERF_COUNTER_ADD(internal_key_skipped_count, 1);
+    }
    iter_->Prev();
    FindParseableKey(&ikey, kReverse);
  }
@@ -792,7 +935,6 @@ void DBIter::Seek(const Slice& target) {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->Seek(saved_key_.GetKey());
  }
-
  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
    if (prefix_extractor_ && prefix_same_as_start_) {
@@ -813,6 +955,46 @@ void DBIter::Seek(const Slice& target) {
  } else {
    valid_ = false;
  }
+
+  if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
+    prefix_start_buf_.SetKey(prefix_start_key_);
+    prefix_start_key_ = prefix_start_buf_.GetKey();
+  }
+}
+
+void DBIter::SeekForPrev(const Slice& target) {
+  StopWatch sw(env_, statistics_, DB_SEEK);
+  ReleaseTempPinnedData();
+  saved_key_.Clear();
+  // now saved_key is used to store internal key.
+  saved_key_.SetInternalKey(target, 0 /* sequence_number */,
+                            kValueTypeForSeekForPrev);
+
+  {
+    PERF_TIMER_GUARD(seek_internal_seek_time);
+    iter_->SeekForPrev(saved_key_.GetKey());
+  }
+
+  RecordTick(statistics_, NUMBER_DB_SEEK);
+  if (iter_->Valid()) {
+    if (prefix_extractor_ && prefix_same_as_start_) {
+      prefix_start_key_ = prefix_extractor_->Transform(target);
+    }
+    direction_ = kReverse;
+    ClearSavedValue();
+    PrevInternal();
+    if (!valid_) {
+      prefix_start_key_.clear();
+    }
+    if (statistics_ != nullptr) {
+      if (valid_) {
+        RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
+        RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
+      }
+    }
+  } else {
+    valid_ = false;
+  }
  if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
    prefix_start_buf_.SetKey(prefix_start_key_);
    prefix_start_key_ = prefix_start_buf_.GetKey();
@@ -870,25 +1052,15 @@ void DBIter::SeekToLast() {
  // it will seek to the last key before the
  // ReadOptions.iterate_upper_bound
  if (iter_->Valid() && iterate_upper_bound_ != nullptr) {
-    saved_key_.SetKey(*iterate_upper_bound_, false /* copy */);
-    std::string last_key;
-    AppendInternalKey(&last_key,
-                      ParsedInternalKey(saved_key_.GetKey(), kMaxSequenceNumber,
-                                        kValueTypeForSeek));
-
-    iter_->Seek(last_key);
-
-    if (!iter_->Valid()) {
-      iter_->SeekToLast();
-    } else {
-      iter_->Prev();
-      if (!iter_->Valid()) {
-        valid_ = false;
-        return;
-      }
+    SeekForPrev(*iterate_upper_bound_);
+    if (!Valid()) {
+      return;
+    } else if (user_comparator_->Equal(*iterate_upper_bound_, key())) {
+      Prev();
    }
+  } else {
+    PrevInternal();
  }
-  PrevInternal();
  if (statistics_ != nullptr) {
    RecordTick(statistics_, NUMBER_DB_SEEK);
    if (valid_) {
@@ -902,18 +1074,16 @@ void DBIter::SeekToLast() {
  }
 }

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

@@ -921,6 +1091,10 @@ ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }

 void ArenaWrappedDBIter::SetDBIter(DBIter* iter) { db_iter_ = iter; }

+RangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
+  return db_iter_->GetRangeDelAggregator();
+}
+
 void ArenaWrappedDBIter::SetIterUnderDBIter(InternalIterator* iter) {
  static_cast<DBIter*>(db_iter_)->SetIter(iter);
 }
@@ -931,6 +1105,9 @@ inline void ArenaWrappedDBIter::SeekToLast() { db_iter_->SeekToLast(); }
 inline void ArenaWrappedDBIter::Seek(const Slice& target) {
  db_iter_->Seek(target);
 }
+inline void ArenaWrappedDBIter::SeekForPrev(const Slice& target) {
+  db_iter_->SeekForPrev(target);
+}
 inline void ArenaWrappedDBIter::Next() { db_iter_->Next(); }
 inline void ArenaWrappedDBIter::Prev() { db_iter_->Prev(); }
 inline Slice ArenaWrappedDBIter::key() const { return db_iter_->key(); }
@@ -949,15 +1126,15 @@ ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    Env* env, const ImmutableCFOptions& ioptions,
    const Comparator* user_key_comparator, const SequenceNumber& sequence,
    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
-    const Slice* iterate_upper_bound, bool prefix_same_as_start,
-    bool pin_data) {
+    const Slice* iterate_upper_bound, bool prefix_same_as_start, bool pin_data,
+    bool total_order_seek) {
  ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
  Arena* arena = iter->GetArena();
  auto mem = arena->AllocateAligned(sizeof(DBIter));
-  DBIter* db_iter =
-      new (mem) DBIter(env, ioptions, user_key_comparator, nullptr, sequence,
-                       true, max_sequential_skip_in_iterations, version_number,
-                       iterate_upper_bound, prefix_same_as_start, pin_data);
+  DBIter* db_iter = new (mem) DBIter(
+      env, ioptions, user_key_comparator, nullptr, sequence, true,
+      max_sequential_skip_in_iterations, version_number, iterate_upper_bound,
+      prefix_same_as_start, pin_data, total_order_seek);

  iter->SetDBIter(db_iter);

@@ -11,6 +11,7 @@
 #include <stdint.h>
 #include <string>
 #include "db/dbformat.h"
+#include "db/range_del_aggregator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/iterator.h"
 #include "util/arena.h"
@@ -31,7 +32,8 @@ extern Iterator* NewDBIterator(
    const Comparator* user_key_comparator, InternalIterator* internal_iter,
    const SequenceNumber& sequence, uint64_t max_sequential_skip_in_iterations,
    uint64_t version_number, const Slice* iterate_upper_bound = nullptr,
-    bool prefix_same_as_start = false, bool pin_data = false);
+    bool prefix_same_as_start = false, bool pin_data = false,
+    bool total_order_seek = false);

 // A wrapper iterator which wraps DB Iterator and the arena, with which the DB
 // iterator is supposed be allocated. This class is used as an entry point of
@@ -45,6 +47,7 @@ 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 RangeDelAggregator* GetRangeDelAggregator();

  // Set the DB Iterator to be wrapped

@@ -57,6 +60,7 @@ class ArenaWrappedDBIter : public Iterator {
  virtual void SeekToFirst() override;
  virtual void SeekToLast() override;
  virtual void Seek(const Slice& target) override;
+  virtual void SeekForPrev(const Slice& target) override;
  virtual void Next() override;
  virtual void Prev() override;
  virtual Slice key() const override;
@@ -77,6 +81,7 @@ extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
    const Comparator* user_key_comparator, const SequenceNumber& sequence,
    uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
    const Slice* iterate_upper_bound = nullptr,
-    bool prefix_same_as_start = false, bool pin_data = false);
+    bool prefix_same_as_start = false, bool pin_data = false,
+    bool total_order_seek = false);

 }  // namespace rocksdb
@@ -117,6 +117,11 @@ class TestIterator : public InternalIterator {
    }
  }

+  virtual void SeekForPrev(const Slice& target) override {
+    assert(initialized_);
+    SeekForPrevImpl(target, &cmp);
+  }
+
  virtual void Next() override {
    assert(initialized_);
    if (data_.empty() || (iter_ == data_.size() - 1)) {
@@ -352,7 +357,7 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    db_iter->SeekToLast();

    ASSERT_TRUE(db_iter->Valid());
-    ASSERT_EQ(static_cast<int>(perf_context.internal_key_skipped_count), 1);
+    ASSERT_EQ(static_cast<int>(perf_context.internal_key_skipped_count), 7);
    ASSERT_EQ(db_iter->key().ToString(), "b");

    SetPerfLevel(kDisable);
@@ -475,7 +480,7 @@ TEST_F(DBIteratorTest, DBIteratorPrevNext) {
    db_iter->SeekToLast();

    ASSERT_TRUE(db_iter->Valid());
-    ASSERT_EQ(static_cast<int>(perf_context.internal_delete_skipped_count), 0);
+    ASSERT_EQ(static_cast<int>(perf_context.internal_delete_skipped_count), 1);
    ASSERT_EQ(db_iter->key().ToString(), "b");

    SetPerfLevel(kDisable);
@@ -1124,6 +1129,26 @@ TEST_F(DBIteratorTest, DBIterator5) {
    db_iter->Prev();
    ASSERT_TRUE(!db_iter->Valid());
  }
+
+  {
+    // put, singledelete, merge
+    TestIterator* internal_iter = new TestIterator(BytewiseComparator());
+    internal_iter->AddPut("a", "val_a");
+    internal_iter->AddSingleDeletion("a");
+    internal_iter->AddMerge("a", "merge_1");
+    internal_iter->AddMerge("a", "merge_2");
+    internal_iter->AddPut("b", "val_b");
+    internal_iter->Finish();
+    std::unique_ptr<Iterator> db_iter(NewDBIterator(
+        env_, ImmutableCFOptions(options), BytewiseComparator(), internal_iter,
+        10, options.max_sequential_skip_in_iterations, 0));
+    db_iter->Seek("b");
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "b");
+    db_iter->Prev();
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "a");
+  }
 }

 TEST_F(DBIteratorTest, DBIterator6) {
@@ -1726,6 +1751,15 @@ TEST_F(DBIteratorTest, DBIterator9) {
    ASSERT_EQ(db_iter->key().ToString(), "a");
    ASSERT_EQ(db_iter->value().ToString(), "merge_1,merge_2");

+    db_iter->SeekForPrev("b");
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "b");
+    ASSERT_EQ(db_iter->value().ToString(), "merge_3,merge_4");
+    db_iter->Next();
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "d");
+    ASSERT_EQ(db_iter->value().ToString(), "merge_5,merge_6");
+
    db_iter->Seek("c");
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "d");
@@ -1734,6 +1768,15 @@ TEST_F(DBIteratorTest, DBIterator9) {
    ASSERT_TRUE(db_iter->Valid());
    ASSERT_EQ(db_iter->key().ToString(), "b");
    ASSERT_EQ(db_iter->value().ToString(), "merge_3,merge_4");
+
+    db_iter->SeekForPrev("c");
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "b");
+    ASSERT_EQ(db_iter->value().ToString(), "merge_3,merge_4");
+    db_iter->Next();
+    ASSERT_TRUE(db_iter->Valid());
+    ASSERT_EQ(db_iter->key().ToString(), "d");
+    ASSERT_EQ(db_iter->value().ToString(), "merge_5,merge_6");
  }
 }

@@ -1764,6 +1807,18 @@ TEST_F(DBIteratorTest, DBIterator10) {
  ASSERT_TRUE(db_iter->Valid());
  ASSERT_EQ(db_iter->key().ToString(), "c");
  ASSERT_EQ(db_iter->value().ToString(), "3");
+
+  db_iter->SeekForPrev("c");
+  ASSERT_TRUE(db_iter->Valid());
+  db_iter->Next();
+  ASSERT_TRUE(db_iter->Valid());
+  ASSERT_EQ(db_iter->key().ToString(), "d");
+  ASSERT_EQ(db_iter->value().ToString(), "4");
+
+  db_iter->Prev();
+  ASSERT_TRUE(db_iter->Valid());
+  ASSERT_EQ(db_iter->key().ToString(), "c");
+  ASSERT_EQ(db_iter->value().ToString(), "3");
 }

 TEST_F(DBIteratorTest, SeekToLastOccurrenceSeq0) {
@@ -1914,7 +1969,7 @@ TEST_F(DBIterWithMergeIterTest, InnerMergeIterator1) {

 TEST_F(DBIterWithMergeIterTest, InnerMergeIterator2) {
  // Test Prev() when one child iterator is at its end.
-  db_iter_->Seek("g");
+  db_iter_->SeekForPrev("g");
  ASSERT_TRUE(db_iter_->Valid());
  ASSERT_EQ(db_iter_->key().ToString(), "g");
  ASSERT_EQ(db_iter_->value().ToString(), "3");
@@ -8,10 +8,10 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include "db/db_test_util.h"
+#include "port/port.h"
 #include "port/stack_trace.h"
 #include "rocksdb/iostats_context.h"
 #include "rocksdb/perf_context.h"
-#include "port/port.h"

 namespace rocksdb {

@@ -204,6 +204,22 @@ TEST_F(DBIteratorTest, IterSeekBeforePrev) {
  delete iter;
 }

+TEST_F(DBIteratorTest, IterSeekForPrevBeforeNext) {
+  ASSERT_OK(Put("a", "b"));
+  ASSERT_OK(Put("c", "d"));
+  dbfull()->Flush(FlushOptions());
+  ASSERT_OK(Put("0", "f"));
+  ASSERT_OK(Put("1", "h"));
+  dbfull()->Flush(FlushOptions());
+  ASSERT_OK(Put("2", "j"));
+  auto iter = db_->NewIterator(ReadOptions());
+  iter->SeekForPrev(Slice("0"));
+  iter->Next();
+  iter->SeekForPrev(Slice("1"));
+  iter->Next();
+  delete iter;
+}
+
 namespace {
 std::string MakeLongKey(size_t length, char c) {
  return std::string(length, c);
@@ -231,6 +247,13 @@ TEST_F(DBIteratorTest, IterLongKeys) {
  ASSERT_EQ(IterStatus(iter), MakeLongKey(127, 3) + "->3");
  iter->Next();
  ASSERT_EQ(IterStatus(iter), MakeLongKey(64, 4) + "->4");
+
+  iter->SeekForPrev(MakeLongKey(127, 3));
+  ASSERT_EQ(IterStatus(iter), MakeLongKey(127, 3) + "->3");
+  iter->Prev();
+  ASSERT_EQ(IterStatus(iter), MakeLongKey(32, 2) + "->2");
+  iter->Prev();
+  ASSERT_EQ(IterStatus(iter), MakeLongKey(50, 1) + "->1");
  delete iter;

  iter = db_->NewIterator(ReadOptions());
@@ -261,6 +284,11 @@ TEST_F(DBIteratorTest, IterNextWithNewerSeq) {
  ASSERT_EQ(IterStatus(iter), "a->b");
  iter->Next();
  ASSERT_EQ(IterStatus(iter), "c->d");
+  iter->SeekForPrev(Slice("b"));
+  ASSERT_EQ(IterStatus(iter), "a->b");
+  iter->Next();
+  ASSERT_EQ(IterStatus(iter), "c->d");
+
  delete iter;
 }

@@ -284,7 +312,13 @@ TEST_F(DBIteratorTest, IterPrevWithNewerSeq) {
  ASSERT_EQ(IterStatus(iter), "c->d");
  iter->Prev();
  ASSERT_EQ(IterStatus(iter), "a->b");
-
+  iter->Prev();
+  iter->SeekForPrev(Slice("d"));
+  ASSERT_EQ(IterStatus(iter), "d->e");
+  iter->Prev();
+  ASSERT_EQ(IterStatus(iter), "c->d");
+  iter->Prev();
+  ASSERT_EQ(IterStatus(iter), "a->b");
  iter->Prev();
  delete iter;
 }
@@ -294,10 +328,13 @@ TEST_F(DBIteratorTest, IterPrevWithNewerSeq2) {
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
-  ASSERT_OK(Put("d", "e"));
+  ASSERT_OK(Put("e", "f"));
  auto iter = db_->NewIterator(ReadOptions());
+  auto iter2 = db_->NewIterator(ReadOptions());
  iter->Seek(Slice("c"));
+  iter2->SeekForPrev(Slice("d"));
  ASSERT_EQ(IterStatus(iter), "c->d");
+  ASSERT_EQ(IterStatus(iter2), "c->d");

  // Create a key that needs to be skipped for Seq too new
  for (uint64_t i = 0; i < last_options_.max_sequential_skip_in_iterations + 1;
@@ -307,9 +344,12 @@ TEST_F(DBIteratorTest, IterPrevWithNewerSeq2) {

  iter->Prev();
  ASSERT_EQ(IterStatus(iter), "a->b");
-
  iter->Prev();
+  iter2->Prev();
+  ASSERT_EQ(IterStatus(iter2), "a->b");
+  iter2->Prev();
  delete iter;
+  delete iter2;
 }

 TEST_F(DBIteratorTest, IterEmpty) {
@@ -326,6 +366,9 @@ TEST_F(DBIteratorTest, IterEmpty) {
    iter->Seek("foo");
    ASSERT_EQ(IterStatus(iter), "(invalid)");

+    iter->SeekForPrev("foo");
+    ASSERT_EQ(IterStatus(iter), "(invalid)");
+
    delete iter;
  } while (ChangeCompactOptions());
 }
@@ -358,14 +401,24 @@ TEST_F(DBIteratorTest, IterSingle) {
    ASSERT_EQ(IterStatus(iter), "a->va");
    iter->Next();
    ASSERT_EQ(IterStatus(iter), "(invalid)");
+    iter->SeekForPrev("");
+    ASSERT_EQ(IterStatus(iter), "(invalid)");

    iter->Seek("a");
    ASSERT_EQ(IterStatus(iter), "a->va");
    iter->Next();
    ASSERT_EQ(IterStatus(iter), "(invalid)");
+    iter->SeekForPrev("a");
+    ASSERT_EQ(IterStatus(iter), "a->va");
+    iter->Prev();
+    ASSERT_EQ(IterStatus(iter), "(invalid)");

    iter->Seek("b");
    ASSERT_EQ(IterStatus(iter), "(invalid)");
+    iter->SeekForPrev("b");
+    ASSERT_EQ(IterStatus(iter), "a->va");
+    iter->Prev();
+    ASSERT_EQ(IterStatus(iter), "(invalid)");

    delete iter;
  } while (ChangeCompactOptions());
@@ -411,11 +464,21 @@ TEST_F(DBIteratorTest, IterMulti) {
    ASSERT_EQ(IterStatus(iter), "a->va");
    iter->Seek("ax");
    ASSERT_EQ(IterStatus(iter), "b->vb");
+    iter->SeekForPrev("d");
+    ASSERT_EQ(IterStatus(iter), "c->vc");
+    iter->SeekForPrev("c");
+    ASSERT_EQ(IterStatus(iter), "c->vc");
+    iter->SeekForPrev("bx");
+    ASSERT_EQ(IterStatus(iter), "b->vb");

    iter->Seek("b");
    ASSERT_EQ(IterStatus(iter), "b->vb");
    iter->Seek("z");
    ASSERT_EQ(IterStatus(iter), "(invalid)");
+    iter->SeekForPrev("b");
+    ASSERT_EQ(IterStatus(iter), "b->vb");
+    iter->SeekForPrev("");
+    ASSERT_EQ(IterStatus(iter), "(invalid)");

    // Switch from reverse to forward
    iter->SeekToLast();
@@ -603,7 +666,8 @@ TEST_F(DBIteratorTest, IterMultiWithDelete) {
      // TODO: merge operator does not support backward iteration yet
      if (kPlainTableAllBytesPrefix != option_config_ &&
          kBlockBasedTableWithWholeKeyHashIndex != option_config_ &&
-          kHashLinkList != option_config_) {
+          kHashLinkList != option_config_ &&
+          kHashSkipList != option_config_) {  // doesn't support SeekToLast
        iter->Prev();
        ASSERT_EQ(IterStatus(iter), "ka->va");
      }
@@ -658,6 +722,7 @@ TEST_F(DBIteratorTest, IterWithSnapshot) {
    options.snapshot = snapshot;
    Iterator* iter = db_->NewIterator(options, handles_[1]);

+    ASSERT_OK(Put(1, "key0", "val0"));
    // Put more values after the snapshot
    ASSERT_OK(Put(1, "key100", "val100"));
    ASSERT_OK(Put(1, "key101", "val101"));
@@ -668,7 +733,7 @@ TEST_F(DBIteratorTest, IterWithSnapshot) {
      // TODO: merge operator does not support backward iteration yet
      if (kPlainTableAllBytesPrefix != option_config_ &&
          kBlockBasedTableWithWholeKeyHashIndex != option_config_ &&
-          kHashLinkList != option_config_) {
+          kHashLinkList != option_config_ && kHashSkipList != option_config_) {
        iter->Prev();
        ASSERT_EQ(IterStatus(iter), "key4->val4");
        iter->Prev();
@@ -682,6 +747,26 @@ TEST_F(DBIteratorTest, IterWithSnapshot) {
      iter->Next();
      ASSERT_TRUE(!iter->Valid());
    }
+
+    if (!CurrentOptions().merge_operator) {
+      // TODO(gzh): merge operator does not support backward iteration yet
+      if (kPlainTableAllBytesPrefix != option_config_ &&
+          kBlockBasedTableWithWholeKeyHashIndex != option_config_ &&
+          kHashLinkList != option_config_ && kHashSkipList != option_config_) {
+        iter->SeekForPrev("key1");
+        ASSERT_EQ(IterStatus(iter), "key1->val1");
+        iter->Next();
+        ASSERT_EQ(IterStatus(iter), "key2->val2");
+        iter->Next();
+        ASSERT_EQ(IterStatus(iter), "key3->val3");
+        iter->Prev();
+        ASSERT_EQ(IterStatus(iter), "key2->val2");
+        iter->Prev();
+        ASSERT_EQ(IterStatus(iter), "key1->val1");
+        iter->Prev();
+        ASSERT_TRUE(!iter->Valid());
+      }
+    }
    db_->ReleaseSnapshot(snapshot);
    delete iter;
    // skip as HashCuckooRep does not support snapshot
@@ -745,6 +830,19 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {
    iter->Next();
    ASSERT_TRUE(iter->Valid());
    ASSERT_EQ(iter->key().compare(Slice("g1")), 0);
+
+    iter->SeekForPrev("g1");
+
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("g1")), 0);
+
+    iter->Prev();
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo1")), 0);
+
+    iter->Prev();
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo")), 0);
  }

  // testing iterate_upper_bound and forward iterator
@@ -877,7 +975,7 @@ TEST_F(DBIteratorTest, DBIteratorBoundTest) {

 // TODO(3.13): fix the issue of Seek() + Prev() which might not necessary
 //             return the biggest key which is smaller than the seek key.
-TEST_F(DBIteratorTest, PrevAfterMerge) {
+TEST_F(DBIteratorTest, PrevAfterAndNextAfterMerge) {
  Options options;
  options.create_if_missing = true;
  options.merge_operator = MergeOperators::CreatePutOperator();
@@ -898,6 +996,14 @@ TEST_F(DBIteratorTest, PrevAfterMerge) {
  it->Prev();
  ASSERT_TRUE(it->Valid());
  ASSERT_EQ("1", it->key().ToString());
+
+  it->SeekForPrev("1");
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("1", it->key().ToString());
+
+  it->Next();
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("2", it->key().ToString());
 }

 TEST_F(DBIteratorTest, PinnedDataIteratorRandomized) {
@@ -980,7 +1086,6 @@ TEST_F(DBIteratorTest, PinnedDataIteratorRandomized) {

    {
      // Test Seek to random keys
-      printf("Testing seek on %" ROCKSDB_PRIszt " keys\n", random_keys.size());
      std::vector<Slice> keys_slices;
      std::vector<std::string> true_keys;
      for (auto& k : random_keys) {
@@ -1002,9 +1107,31 @@ TEST_F(DBIteratorTest, PinnedDataIteratorRandomized) {
      }
    }

+    {
+      // Test SeekForPrev to random keys
+      std::vector<Slice> keys_slices;
+      std::vector<std::string> true_keys;
+      for (auto& k : random_keys) {
+        iter->SeekForPrev(k);
+        if (!iter->Valid()) {
+          ASSERT_EQ(true_data.upper_bound(k), true_data.begin());
+          continue;
+        }
+        std::string prop_value;
+        ASSERT_OK(
+            iter->GetProperty("rocksdb.iterator.is-key-pinned", &prop_value));
+        ASSERT_EQ("1", prop_value);
+        keys_slices.push_back(iter->key());
+        true_keys.push_back((--true_data.upper_bound(k))->first);
+      }
+
+      for (size_t i = 0; i < keys_slices.size(); i++) {
+        ASSERT_EQ(keys_slices[i].ToString(), true_keys[i]);
+      }
+    }
+
    {
      // Test iterating all data forward
-      printf("Testing iterating forward on all keys\n");
      std::vector<Slice> all_keys;
      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
        std::string prop_value;
@@ -1025,7 +1152,6 @@ TEST_F(DBIteratorTest, PinnedDataIteratorRandomized) {

    {
      // Test iterating all data backward
-      printf("Testing iterating backward on all keys\n");
      std::vector<Slice> all_keys;
      for (iter->SeekToLast(); iter->Valid(); iter->Prev()) {
        std::string prop_value;
@@ -1230,6 +1356,60 @@ TEST_F(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
  delete iter;
 }

+TEST_F(DBIteratorTest, IterSeekForPrevCrossingFiles) {
+  Options options = CurrentOptions();
+  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+  options.disable_auto_compactions = true;
+  // Enable prefix bloom for SST files
+  BlockBasedTableOptions table_options;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put("a1", "va1"));
+  ASSERT_OK(Put("a2", "va2"));
+  ASSERT_OK(Put("a3", "va3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b1", "vb1"));
+  ASSERT_OK(Put("b2", "vb2"));
+  ASSERT_OK(Put("b3", "vb3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b4", "vb4"));
+  ASSERT_OK(Put("d1", "vd1"));
+  ASSERT_OK(Put("d2", "vd2"));
+  ASSERT_OK(Put("d4", "vd4"));
+  ASSERT_OK(Flush());
+
+  MoveFilesToLevel(1);
+  {
+    ReadOptions ro;
+    Iterator* iter = db_->NewIterator(ro);
+
+    iter->SeekForPrev("a4");
+    ASSERT_EQ(iter->key().ToString(), "a3");
+    ASSERT_EQ(iter->value().ToString(), "va3");
+
+    iter->SeekForPrev("c2");
+    ASSERT_EQ(iter->key().ToString(), "b3");
+    iter->SeekForPrev("d3");
+    ASSERT_EQ(iter->key().ToString(), "d2");
+    iter->SeekForPrev("b5");
+    ASSERT_EQ(iter->key().ToString(), "b4");
+    delete iter;
+  }
+
+  {
+    ReadOptions ro;
+    ro.prefix_same_as_start = true;
+    Iterator* iter = db_->NewIterator(ro);
+    iter->SeekForPrev("c2");
+    ASSERT_TRUE(!iter->Valid());
+    delete iter;
+  }
+}
+
 TEST_F(DBIteratorTest, IterPrevKeyCrossingBlocks) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
@@ -1518,13 +1698,14 @@ TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {
    t.join();
  }

-  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_NEXT), total_next);
+  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_NEXT), (uint64_t)total_next);
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_NEXT_FOUND),
-            total_next_found);
-  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_PREV), total_prev);
+            (uint64_t)total_next_found);
+  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_PREV), (uint64_t)total_prev);
  ASSERT_EQ(TestGetTickerCount(options, NUMBER_DB_PREV_FOUND),
-            total_prev_found);
-  ASSERT_EQ(TestGetTickerCount(options, ITER_BYTES_READ), total_bytes);
+            (uint64_t)total_prev_found);
+  ASSERT_EQ(TestGetTickerCount(options, ITER_BYTES_READ), (uint64_t)total_bytes);
+
 }

 TEST_F(DBIteratorTest, ReadAhead) {
@@ -1597,6 +1778,62 @@ TEST_F(DBIteratorTest, ReadAhead) {
  delete iter;
 }

+// Insert a key, create a snapshot iterator, overwrite key lots of times,
+// seek to a smaller key. Expect DBIter to fall back to a seek instead of
+// going through all the overwrites linearly.
+TEST_F(DBIteratorTest, DBIteratorSkipRecentDuplicatesTest) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  options.max_sequential_skip_in_iterations = 3;
+  options.prefix_extractor = nullptr;
+  options.write_buffer_size = 1 << 27;  // big enough to avoid flush
+  options.statistics = rocksdb::CreateDBStatistics();
+  DestroyAndReopen(options);
+
+  // Insert.
+  ASSERT_OK(Put("b", "0"));
+
+  // Create iterator.
+  ReadOptions ro;
+  std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
+
+  // Insert a lot.
+  for (int i = 0; i < 100; ++i) {
+    ASSERT_OK(Put("b", std::to_string(i + 1).c_str()));
+  }
+
+#ifndef ROCKSDB_LITE
+  // Check that memtable wasn't flushed.
+  std::string val;
+  ASSERT_TRUE(db_->GetProperty("rocksdb.num-files-at-level0", &val));
+  EXPECT_EQ("0", val);
+#endif
+
+  // Seek iterator to a smaller key.
+  perf_context.Reset();
+  iter->Seek("a");
+  ASSERT_TRUE(iter->Valid());
+  EXPECT_EQ("b", iter->key().ToString());
+  EXPECT_EQ("0", iter->value().ToString());
+
+  // Check that the seek didn't do too much work.
+  // Checks are not tight, just make sure that everything is well below 100.
+  EXPECT_LT(perf_context.internal_key_skipped_count, 4);
+  EXPECT_LT(perf_context.internal_recent_skipped_count, 8);
+  EXPECT_LT(perf_context.seek_on_memtable_count, 10);
+  EXPECT_LT(perf_context.next_on_memtable_count, 10);
+  EXPECT_LT(perf_context.prev_on_memtable_count, 10);
+
+  // Check that iterator did something like what we expect.
+  EXPECT_EQ(perf_context.internal_delete_skipped_count, 0);
+  EXPECT_EQ(perf_context.internal_merge_count, 0);
+  EXPECT_GE(perf_context.internal_recent_skipped_count, 2);
+  EXPECT_GE(perf_context.seek_on_memtable_count, 2);
+  EXPECT_EQ(1, options.statistics->getTickerCount(
+                 NUMBER_OF_RESEEKS_IN_ITERATION));
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -0,0 +1,163 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#include <memory>
+#include <string>
+
+#include "db/db_test_util.h"
+#include "db/memtable.h"
+#include "port/stack_trace.h"
+#include "rocksdb/memtablerep.h"
+#include "rocksdb/slice_transform.h"
+
+namespace rocksdb {
+
+class DBMemTableTest : public DBTestBase {
+ public:
+  DBMemTableTest() : DBTestBase("/db_memtable_test") {}
+};
+
+class MockMemTableRep : public MemTableRep {
+ public:
+  explicit MockMemTableRep(MemTableAllocator* allocator, MemTableRep* rep)
+      : MemTableRep(allocator), rep_(rep), num_insert_with_hint_(0) {}
+
+  virtual KeyHandle Allocate(const size_t len, char** buf) override {
+    return rep_->Allocate(len, buf);
+  }
+
+  virtual void Insert(KeyHandle handle) override {
+    return rep_->Insert(handle);
+  }
+
+  virtual void InsertWithHint(KeyHandle handle, void** hint) override {
+    num_insert_with_hint_++;
+    ASSERT_NE(nullptr, hint);
+    last_hint_in_ = *hint;
+    rep_->InsertWithHint(handle, hint);
+    last_hint_out_ = *hint;
+  }
+
+  virtual bool Contains(const char* key) const override {
+    return rep_->Contains(key);
+  }
+
+  virtual void Get(const LookupKey& k, void* callback_args,
+                   bool (*callback_func)(void* arg,
+                                         const char* entry)) override {
+    rep_->Get(k, callback_args, callback_func);
+  }
+
+  virtual size_t ApproximateMemoryUsage() override {
+    return rep_->ApproximateMemoryUsage();
+  }
+
+  virtual Iterator* GetIterator(Arena* arena) override {
+    return rep_->GetIterator(arena);
+  }
+
+  void* last_hint_in() { return last_hint_in_; }
+  void* last_hint_out() { return last_hint_out_; }
+  int num_insert_with_hint() { return num_insert_with_hint_; }
+
+ private:
+  std::unique_ptr<MemTableRep> rep_;
+  void* last_hint_in_;
+  void* last_hint_out_;
+  int num_insert_with_hint_;
+};
+
+class MockMemTableRepFactory : public MemTableRepFactory {
+ public:
+  virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator& cmp,
+                                         MemTableAllocator* allocator,
+                                         const SliceTransform* transform,
+                                         Logger* logger) override {
+    SkipListFactory factory;
+    MemTableRep* skiplist_rep =
+        factory.CreateMemTableRep(cmp, allocator, transform, logger);
+    mock_rep_ = new MockMemTableRep(allocator, skiplist_rep);
+    return mock_rep_;
+  }
+
+  virtual const char* Name() const override { return "MockMemTableRepFactory"; }
+
+  MockMemTableRep* rep() { return mock_rep_; }
+
+  bool IsInsertConcurrentlySupported() const override { return false; }
+
+ private:
+  MockMemTableRep* mock_rep_;
+};
+
+class TestPrefixExtractor : public SliceTransform {
+ public:
+  virtual const char* Name() const override { return "TestPrefixExtractor"; }
+
+  virtual Slice Transform(const Slice& key) const override {
+    const char* p = separator(key);
+    if (p == nullptr) {
+      return Slice();
+    }
+    return Slice(key.data(), p - key.data() + 1);
+  }
+
+  virtual bool InDomain(const Slice& key) const override {
+    return separator(key) != nullptr;
+  }
+
+  virtual bool InRange(const Slice& key) const override { return false; }
+
+ private:
+  const char* separator(const Slice& key) const {
+    return reinterpret_cast<const char*>(memchr(key.data(), '_', key.size()));
+  }
+};
+
+TEST_F(DBMemTableTest, InsertWithHint) {
+  Options options;
+  options.allow_concurrent_memtable_write = false;
+  options.create_if_missing = true;
+  options.memtable_factory.reset(new MockMemTableRepFactory());
+  options.memtable_insert_with_hint_prefix_extractor.reset(
+      new TestPrefixExtractor());
+  Reopen(options);
+  MockMemTableRep* rep =
+      reinterpret_cast<MockMemTableRepFactory*>(options.memtable_factory.get())
+          ->rep();
+  ASSERT_OK(Put("foo_k1", "foo_v1"));
+  ASSERT_EQ(nullptr, rep->last_hint_in());
+  void* hint_foo = rep->last_hint_out();
+  ASSERT_OK(Put("foo_k2", "foo_v2"));
+  ASSERT_EQ(hint_foo, rep->last_hint_in());
+  ASSERT_EQ(hint_foo, rep->last_hint_out());
+  ASSERT_OK(Put("foo_k3", "foo_v3"));
+  ASSERT_EQ(hint_foo, rep->last_hint_in());
+  ASSERT_EQ(hint_foo, rep->last_hint_out());
+  ASSERT_OK(Put("bar_k1", "bar_v1"));
+  ASSERT_EQ(nullptr, rep->last_hint_in());
+  void* hint_bar = rep->last_hint_out();
+  ASSERT_NE(hint_foo, hint_bar);
+  ASSERT_OK(Put("bar_k2", "bar_v2"));
+  ASSERT_EQ(hint_bar, rep->last_hint_in());
+  ASSERT_EQ(hint_bar, rep->last_hint_out());
+  ASSERT_EQ(5, rep->num_insert_with_hint());
+  ASSERT_OK(Put("whitelisted", "vvv"));
+  ASSERT_EQ(5, rep->num_insert_with_hint());
+  ASSERT_EQ("foo_v1", Get("foo_k1"));
+  ASSERT_EQ("foo_v2", Get("foo_k2"));
+  ASSERT_EQ("foo_v3", Get("foo_k3"));
+  ASSERT_EQ("bar_v1", Get("bar_k1"));
+  ASSERT_EQ("bar_v2", Get("bar_k2"));
+  ASSERT_EQ("vvv", Get("whitelisted"));
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -0,0 +1,98 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+#include <string>
+#include <vector>
+
+#include "db/db_test_util.h"
+#include "db/forward_iterator.h"
+#include "port/stack_trace.h"
+#include "utilities/merge_operators.h"
+
+namespace rocksdb {
+
+// Test merge operator functionality.
+class DBMergeOperatorTest : public DBTestBase {
+ public:
+  DBMergeOperatorTest() : DBTestBase("/db_merge_operator_test") {}
+};
+
+// A test merge operator mimics put but also fails if one of merge operands is
+// "corrupted".
+class TestPutOperator : public MergeOperator {
+ public:
+  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
+                           MergeOperationOutput* merge_out) const override {
+    if (merge_in.existing_value != nullptr &&
+        *(merge_in.existing_value) == "corrupted") {
+      return false;
+    }
+    for (auto value : merge_in.operand_list) {
+      if (value == "corrupted") {
+        return false;
+      }
+    }
+    merge_out->existing_operand = merge_in.operand_list.back();
+    return true;
+  }
+
+  virtual const char* Name() const override { return "TestPutOperator"; }
+};
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnRead) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+  Reopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "corrupted"));
+  std::string value;
+  ASSERT_TRUE(db_->Get(ReadOptions(), "k1", &value).IsCorruption());
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v1"}});
+}
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnWrite) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+  options.max_successive_merges = 3;
+  Reopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "v2"));
+  // Will trigger a merge when hitting max_successive_merges and the merge
+  // will fail. The delta will be inserted nevertheless.
+  ASSERT_OK(Merge("k1", "corrupted"));
+  // Data should stay unmerged after the error.
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v2"}, {"k1", "v1"}});
+}
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnIteration) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+
+  DestroyAndReopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "corrupted"));
+  ASSERT_OK(Put("k2", "v2"));
+  VerifyDBFromMap({{"k1", ""}, {"k2", "v2"}}, nullptr, false,
+                  {{"k1", Status::Corruption()}});
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v1"}, {"k2", "v2"}});
+
+  DestroyAndReopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Put("k2", "v2"));
+  ASSERT_OK(Merge("k2", "corrupted"));
+  VerifyDBFromMap({{"k1", "v1"}, {"k2", ""}}, nullptr, false,
+                  {{"k2", Status::Corruption()}});
+  VerifyDBInternal({{"k1", "v1"}, {"k2", "corrupted"}, {"k2", "v2"}});
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -8,21 +8,119 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 #include <limits>
 #include <string>
+#include <unordered_map>

+#include "db/column_family.h"
+#include "db/db_impl.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "rocksdb/convenience.h"
+#include "util/options_helper.h"
+#include "util/random.h"
 #include "util/sync_point.h"
+#include "util/testutil.h"

 namespace rocksdb {

 class DBOptionsTest : public DBTestBase {
 public:
  DBOptionsTest() : DBTestBase("/db_options_test") {}
+
+#ifndef ROCKSDB_LITE
+  std::unordered_map<std::string, std::string> GetMutableDBOptionsMap(
+      const DBOptions& options) {
+    std::string options_str;
+    GetStringFromDBOptions(&options_str, options);
+    std::unordered_map<std::string, std::string> options_map;
+    StringToMap(options_str, &options_map);
+    std::unordered_map<std::string, std::string> mutable_map;
+    for (const auto opt : db_options_type_info) {
+      if (opt.second.is_mutable &&
+          opt.second.verification != OptionVerificationType::kDeprecated) {
+        mutable_map[opt.first] = options_map[opt.first];
+      }
+    }
+    return mutable_map;
+  }
+
+  std::unordered_map<std::string, std::string> GetMutableCFOptionsMap(
+      const ColumnFamilyOptions& options) {
+    std::string options_str;
+    GetStringFromColumnFamilyOptions(&options_str, options);
+    std::unordered_map<std::string, std::string> options_map;
+    StringToMap(options_str, &options_map);
+    std::unordered_map<std::string, std::string> mutable_map;
+    for (const auto opt : cf_options_type_info) {
+      if (opt.second.is_mutable &&
+          opt.second.verification != OptionVerificationType::kDeprecated) {
+        mutable_map[opt.first] = options_map[opt.first];
+      }
+    }
+    return mutable_map;
+  }
+
+  std::unordered_map<std::string, std::string> GetRandomizedMutableCFOptionsMap(
+      Random* rnd) {
+    Options options;
+    ImmutableDBOptions db_options(options);
+    test::RandomInitCFOptions(&options, rnd);
+    auto sanitized_options = SanitizeOptions(db_options, options);
+    auto opt_map = GetMutableCFOptionsMap(sanitized_options);
+    delete options.compaction_filter;
+    return opt_map;
+  }
+
+  std::unordered_map<std::string, std::string> GetRandomizedMutableDBOptionsMap(
+      Random* rnd) {
+    DBOptions db_options;
+    test::RandomInitDBOptions(&db_options, rnd);
+    auto sanitized_options = SanitizeOptions(dbname_, db_options);
+    return GetMutableDBOptionsMap(sanitized_options);
+  }
+#endif  // ROCKSDB_LITE
 };

 // RocksDB lite don't support dynamic options.
 #ifndef ROCKSDB_LITE

+TEST_F(DBOptionsTest, GetLatestDBOptions) {
+  // GetOptions should be able to get latest option changed by SetOptions.
+  Options options;
+  options.create_if_missing = true;
+  Random rnd(228);
+  Reopen(options);
+  auto new_options = GetRandomizedMutableDBOptionsMap(&rnd);
+  ASSERT_OK(dbfull()->SetDBOptions(new_options));
+  ASSERT_EQ(new_options, GetMutableDBOptionsMap(dbfull()->GetDBOptions()));
+}
+
+TEST_F(DBOptionsTest, GetLatestCFOptions) {
+  // GetOptions should be able to get latest option changed by SetOptions.
+  Options options;
+  options.create_if_missing = true;
+  Random rnd(228);
+  Reopen(options);
+  CreateColumnFamilies({"foo"}, options);
+  ReopenWithColumnFamilies({"default", "foo"}, options);
+  auto options_default = GetRandomizedMutableCFOptionsMap(&rnd);
+  auto options_foo = GetRandomizedMutableCFOptionsMap(&rnd);
+  ASSERT_OK(dbfull()->SetOptions(handles_[0], options_default));
+  ASSERT_OK(dbfull()->SetOptions(handles_[1], options_foo));
+  ASSERT_EQ(options_default,
+            GetMutableCFOptionsMap(dbfull()->GetOptions(handles_[0])));
+  ASSERT_EQ(options_foo,
+            GetMutableCFOptionsMap(dbfull()->GetOptions(handles_[1])));
+}
+
+TEST_F(DBOptionsTest, SetOptionsAndReopen) {
+  Random rnd(1044);
+  auto rand_opts = GetRandomizedMutableCFOptionsMap(&rnd);
+  ASSERT_OK(dbfull()->SetOptions(rand_opts));
+  // Verify if DB can be reopen after setting options.
+  Options options;
+  ASSERT_OK(TryReopen(options));
+}
+
 TEST_F(DBOptionsTest, EnableAutoCompactionAndTriggerStall) {
  const std::string kValue(1024, 'v');
  for (int method_type = 0; method_type < 2; method_type++) {
@@ -126,6 +224,98 @@ TEST_F(DBOptionsTest, EnableAutoCompactionAndTriggerStall) {
  }
 }

+TEST_F(DBOptionsTest, SetOptionsMayTriggerCompaction) {
+  Options options;
+  options.create_if_missing = true;
+  options.level0_file_num_compaction_trigger = 1000;
+  Reopen(options);
+  for (int i = 0; i < 3; i++) {
+    // Need to insert two keys to avoid trivial move.
+    ASSERT_OK(Put("foo", ToString(i)));
+    ASSERT_OK(Put("bar", ToString(i)));
+    Flush();
+  }
+  ASSERT_EQ("3", FilesPerLevel());
+  ASSERT_OK(
+      dbfull()->SetOptions({{"level0_file_num_compaction_trigger", "3"}}));
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ("0,1", FilesPerLevel());
+}
+
+TEST_F(DBOptionsTest, SetBackgroundCompactionThreads) {
+  Options options;
+  options.create_if_missing = true;
+  options.base_background_compactions = 1;  // default value
+  options.max_background_compactions = 1;   // default value
+  Reopen(options);
+  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
+  ASSERT_OK(dbfull()->SetDBOptions({{"base_background_compactions", "2"},
+                                    {"max_background_compactions", "3"}}));
+  ASSERT_EQ(2, dbfull()->TEST_BGCompactionsAllowed());
+  auto stop_token = dbfull()->TEST_write_controler().GetStopToken();
+  ASSERT_EQ(3, dbfull()->TEST_BGCompactionsAllowed());
+}
+
+TEST_F(DBOptionsTest, AvoidFlushDuringShutdown) {
+  Options options;
+  options.create_if_missing = true;
+  options.disable_auto_compactions = true;
+  WriteOptions write_without_wal;
+  write_without_wal.disableWAL = true;
+
+  ASSERT_FALSE(options.avoid_flush_during_shutdown);
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "v1", write_without_wal));
+  Reopen(options);
+  ASSERT_EQ("v1", Get("foo"));
+  ASSERT_EQ("1", FilesPerLevel());
+
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "v2", write_without_wal));
+  ASSERT_OK(dbfull()->SetDBOptions({{"avoid_flush_during_shutdown", "true"}}));
+  Reopen(options);
+  ASSERT_EQ("NOT_FOUND", Get("foo"));
+  ASSERT_EQ("", FilesPerLevel());
+}
+
+TEST_F(DBOptionsTest, SetDelayedWriteRateOption) {
+  Options options;
+  options.create_if_missing = true;
+  options.delayed_write_rate = 2 * 1024U * 1024U;
+  Reopen(options);
+  ASSERT_EQ(2 * 1024U * 1024U, dbfull()->TEST_write_controler().max_delayed_write_rate());
+
+  ASSERT_OK(dbfull()->SetDBOptions({{"delayed_write_rate", "20000"}}));
+  ASSERT_EQ(20000, dbfull()->TEST_write_controler().max_delayed_write_rate());
+}
+
+TEST_F(DBOptionsTest, MaxTotalWalSizeChange) {
+  Random rnd(1044);
+  const auto value_size = size_t(1024);
+  std::string value;
+  test::RandomString(&rnd, value_size, &value);
+
+  Options options;
+  options.create_if_missing = true;
+  CreateColumnFamilies({"1", "2", "3"}, options);
+  ReopenWithColumnFamilies({"default", "1", "2", "3"}, options);
+
+  WriteOptions write_options;
+
+  const int key_count = 100;
+  for (int i = 0; i < key_count; ++i) {
+    for (size_t cf = 0; cf < handles_.size(); ++cf) {
+      ASSERT_OK(Put(static_cast<int>(cf), Key(i), value));
+    }
+  }
+  ASSERT_OK(dbfull()->SetDBOptions({{"max_total_wal_size", "10"}}));
+
+  for (size_t cf = 0; cf < handles_.size(); ++cf) {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+    ASSERT_EQ("1", FilesPerLevel(static_cast<int>(cf)));
+  }
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -34,6 +34,7 @@ TEST_F(DBPropertiesTest, Empty) {
    Options options;
    options.env = env_;
    options.write_buffer_size = 100000;  // Small write buffer
+    options.allow_concurrent_memtable_write = false;
    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

@@ -491,6 +492,7 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {

    std::string big_value(1000000 * 2, 'x');
    std::string num;
+    uint64_t value;
    SetPerfLevel(kEnableTime);
    ASSERT_TRUE(GetPerfLevel() == kEnableTime);

@@ -555,11 +557,11 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {
    ASSERT_TRUE(dbfull()->GetProperty(
        handles_[1], DB::Properties::kNumImmutableMemTableFlushed, &num));
    ASSERT_EQ(num, "3");
-    ASSERT_TRUE(dbfull()->GetProperty(
-        handles_[1], "rocksdb.cur-size-active-mem-table", &num));
-    // "192" is the size of the metadata of an empty skiplist, this would
+    ASSERT_TRUE(dbfull()->GetIntProperty(
+        handles_[1], "rocksdb.cur-size-active-mem-table", &value));
+    // "192" is the size of the metadata of two empty skiplists, this would
    // break if we change the default skiplist implementation
-    ASSERT_EQ(num, "192");
+    ASSERT_GE(value, 192);

    uint64_t int_num;
    uint64_t base_total_size;
@@ -0,0 +1,492 @@
+//  Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+#include "util/testutil.h"
+#include "utilities/merge_operators.h"
+
+namespace rocksdb {
+
+class DBRangeDelTest : public DBTestBase {
+ public:
+  DBRangeDelTest() : DBTestBase("/db_range_del_test") {}
+
+  std::string GetNumericStr(int key) {
+    uint64_t uint64_key = static_cast<uint64_t>(key);
+    std::string str;
+    str.resize(8);
+    memcpy(&str[0], static_cast<void*>(&uint64_key), 8);
+    return str;
+  }
+};
+
+// PlainTableFactory and NumTableFilesAtLevel() are not supported in
+// ROCKSDB_LITE
+#ifndef ROCKSDB_LITE
+TEST_F(DBRangeDelTest, NonBlockBasedTableNotSupported) {
+  Options opts = CurrentOptions();
+  opts.table_factory.reset(new PlainTableFactory());
+  opts.prefix_extractor.reset(NewNoopTransform());
+  opts.allow_mmap_reads = true;
+  opts.max_sequential_skip_in_iterations = 999999;
+  Reopen(opts);
+
+  ASSERT_TRUE(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1", "dr1")
+          .IsNotSupported());
+}
+
+TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
+                             "dr1"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+}
+
+TEST_F(DBRangeDelTest, CompactionOutputHasOnlyRangeTombstone) {
+  Options opts = CurrentOptions();
+  opts.disable_auto_compactions = true;
+  opts.statistics = CreateDBStatistics();
+  Reopen(opts);
+
+  // snapshot protects range tombstone from dropping due to becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
+  db_->Flush(FlushOptions());
+
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(1));
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                              true /* disallow_trivial_move */);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(1));
+  ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, CompactionOutputFilesExactlyFilled) {
+  // regression test for exactly filled compaction output files. Previously
+  // another file would be generated containing all range deletions, which
+  // could invalidate the non-overlapping file boundary invariant.
+  const int kNumPerFile = 4, kNumFiles = 2, kFileBytes = 9 << 10;
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.level0_file_num_compaction_trigger = kNumFiles;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  options.num_levels = 2;
+  options.target_file_size_base = kFileBytes;
+  BlockBasedTableOptions table_options;
+  table_options.block_size_deviation = 50;  // each block holds two keys
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+
+  // snapshot protects range tombstone from dropping due to becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0), Key(1));
+
+  Random rnd(301);
+  for (int i = 0; i < kNumFiles; ++i) {
+    std::vector<std::string> values;
+    // Write 12K (4 values, each 3K)
+    for (int j = 0; j < kNumPerFile; j++) {
+      values.push_back(RandomString(&rnd, 3 << 10));
+      ASSERT_OK(Put(Key(i * kNumPerFile + j), values[j]));
+      if (j == 0 && i > 0) {
+        dbfull()->TEST_WaitForFlushMemTable();
+      }
+    }
+  }
+  // put extra key to trigger final flush
+  ASSERT_OK(Put("", ""));
+  dbfull()->TEST_WaitForFlushMemTable();
+  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(1));
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                              true /* disallow_trivial_move */);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(2, NumTableFilesAtLevel(1));
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, MaxCompactionBytesCutsOutputFiles) {
+  // Ensures range deletion spanning multiple compaction output files that are
+  // cut by max_compaction_bytes will have non-overlapping key-ranges.
+  // https://github.com/facebook/rocksdb/issues/1778
+  const int kNumFiles = 2, kNumPerFile = 1 << 8, kBytesPerVal = 1 << 12;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.disable_auto_compactions = true;
+  opts.level0_file_num_compaction_trigger = kNumFiles;
+  opts.max_compaction_bytes = kNumPerFile * kBytesPerVal;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  // Want max_compaction_bytes to trigger the end of compaction output file, not
+  // target_file_size_base, so make the latter much bigger
+  opts.target_file_size_base = 100 * opts.max_compaction_bytes;
+  Reopen(opts);
+
+  // snapshot protects range tombstone from dropping due to becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+
+  // It spans the whole key-range, thus will be included in all output files
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                             GetNumericStr(0),
+                             GetNumericStr(kNumFiles * kNumPerFile - 1)));
+  Random rnd(301);
+  for (int i = 0; i < kNumFiles; ++i) {
+    std::vector<std::string> values;
+    // Write 1MB (256 values, each 4K)
+    for (int j = 0; j < kNumPerFile; j++) {
+      values.push_back(RandomString(&rnd, kBytesPerVal));
+      ASSERT_OK(Put(GetNumericStr(kNumPerFile * i + j), values[j]));
+    }
+    // extra entry to trigger SpecialSkipListFactory's flush
+    ASSERT_OK(Put(GetNumericStr(kNumPerFile), ""));
+    dbfull()->TEST_WaitForFlushMemTable();
+    ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
+  }
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                              true /* disallow_trivial_move */);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GE(NumTableFilesAtLevel(1), 2);
+
+  std::vector<std::vector<FileMetaData>> files;
+  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
+
+  for (size_t i = 0; i < files[1].size() - 1; ++i) {
+    ASSERT_TRUE(InternalKeyComparator(opts.comparator)
+                    .Compare(files[1][i].largest, files[1][i + 1].smallest) <
+                0);
+  }
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, FlushRangeDelsSameStartKey) {
+  db_->Put(WriteOptions(), "b1", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "c"));
+  db_->Put(WriteOptions(), "b2", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
+  // first iteration verifies query correctness in memtable, second verifies
+  // query correctness for a single SST file
+  for (int i = 0; i < 2; ++i) {
+    if (i > 0) {
+      ASSERT_OK(db_->Flush(FlushOptions()));
+      ASSERT_EQ(1, NumTableFilesAtLevel(0));
+    }
+    std::string value;
+    ASSERT_TRUE(db_->Get(ReadOptions(), "b1", &value).IsNotFound());
+    ASSERT_OK(db_->Get(ReadOptions(), "b2", &value));
+  }
+}
+
+TEST_F(DBRangeDelTest, CompactRangeDelsSameStartKey) {
+  db_->Put(WriteOptions(), "unused", "val");  // prevents empty after compaction
+  db_->Put(WriteOptions(), "b1", "val");
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "c"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  ASSERT_EQ(3, NumTableFilesAtLevel(0));
+
+  for (int i = 0; i < 2; ++i) {
+    if (i > 0) {
+      dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                                  true /* disallow_trivial_move */);
+      ASSERT_EQ(0, NumTableFilesAtLevel(0));
+      ASSERT_EQ(1, NumTableFilesAtLevel(1));
+    }
+    std::string value;
+    ASSERT_TRUE(db_->Get(ReadOptions(), "b1", &value).IsNotFound());
+  }
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBRangeDelTest, FlushRemovesCoveredKeys) {
+  const int kNum = 300, kRangeBegin = 50, kRangeEnd = 250;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  Reopen(opts);
+
+  // Write a third before snapshot, a third between snapshot and tombstone, and
+  // a third after the tombstone. Keys older than snapshot or newer than the
+  // tombstone should be preserved.
+  const Snapshot* snapshot = nullptr;
+  for (int i = 0; i < kNum; ++i) {
+    if (i == kNum / 3) {
+      snapshot = db_->GetSnapshot();
+    } else if (i == 2 * kNum / 3) {
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
+    }
+    db_->Put(WriteOptions(), GetNumericStr(i), "val");
+  }
+  db_->Flush(FlushOptions());
+
+  for (int i = 0; i < kNum; ++i) {
+    ReadOptions read_opts;
+    read_opts.ignore_range_deletions = true;
+    std::string value;
+    if (i < kRangeBegin || i > kRangeEnd || i < kNum / 3 || i >= 2 * kNum / 3) {
+      ASSERT_OK(db_->Get(read_opts, GetNumericStr(i), &value));
+    } else {
+      ASSERT_TRUE(db_->Get(read_opts, GetNumericStr(i), &value).IsNotFound());
+    }
+  }
+  db_->ReleaseSnapshot(snapshot);
+}
+
+// NumTableFilesAtLevel() is not supported in ROCKSDB_LITE
+#ifndef ROCKSDB_LITE
+TEST_F(DBRangeDelTest, CompactionRemovesCoveredKeys) {
+  const int kNumPerFile = 100, kNumFiles = 4;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.disable_auto_compactions = true;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  opts.num_levels = 2;
+  opts.statistics = CreateDBStatistics();
+  Reopen(opts);
+
+  for (int i = 0; i < kNumFiles; ++i) {
+    if (i > 0) {
+      // range tombstone covers first half of the previous file
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                       GetNumericStr((i - 1) * kNumPerFile),
+                       GetNumericStr((i - 1) * kNumPerFile + kNumPerFile / 2));
+    }
+    // Make sure a given key appears in each file so compaction won't be able to
+    // use trivial move, which would happen if the ranges were non-overlapping.
+    // Also, we need an extra element since flush is only triggered when the
+    // number of keys is one greater than SpecialSkipListFactory's limit.
+    // We choose a key outside the key-range used by the test to avoid conflict.
+    db_->Put(WriteOptions(), GetNumericStr(kNumPerFile * kNumFiles), "val");
+
+    for (int j = 0; j < kNumPerFile; ++j) {
+      db_->Put(WriteOptions(), GetNumericStr(i * kNumPerFile + j), "val");
+    }
+    dbfull()->TEST_WaitForFlushMemTable();
+    ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
+  }
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(1), 0);
+  ASSERT_EQ((kNumFiles - 1) * kNumPerFile / 2,
+            TestGetTickerCount(opts, COMPACTION_KEY_DROP_RANGE_DEL));
+
+  for (int i = 0; i < kNumFiles; ++i) {
+    for (int j = 0; j < kNumPerFile; ++j) {
+      ReadOptions read_opts;
+      read_opts.ignore_range_deletions = true;
+      std::string value;
+      if (i == kNumFiles - 1 || j >= kNumPerFile / 2) {
+        ASSERT_OK(
+            db_->Get(read_opts, GetNumericStr(i * kNumPerFile + j), &value));
+      } else {
+        ASSERT_TRUE(
+            db_->Get(read_opts, GetNumericStr(i * kNumPerFile + j), &value)
+                .IsNotFound());
+      }
+    }
+  }
+}
+
+TEST_F(DBRangeDelTest, ValidLevelSubcompactionBoundaries) {
+  const int kNumPerFile = 100, kNumFiles = 4, kFileBytes = 100 << 10;
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = kNumFiles;
+  options.max_bytes_for_level_base = 2 * kFileBytes;
+  options.max_subcompactions = 4;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  options.num_levels = 3;
+  options.target_file_size_base = kFileBytes;
+  options.target_file_size_multiplier = 1;
+  Reopen(options);
+
+  Random rnd(301);
+  for (int i = 0; i < 2; ++i) {
+    for (int j = 0; j < kNumFiles; ++j) {
+      if (i > 0) {
+        // delete [95,105) in two files, [295,305) in next two
+        int mid = (j + (1 - j % 2)) * kNumPerFile;
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                         Key(mid - 5), Key(mid + 5));
+      }
+      std::vector<std::string> values;
+      // Write 100KB (100 values, each 1K)
+      for (int k = 0; k < kNumPerFile; k++) {
+        values.push_back(RandomString(&rnd, 990));
+        ASSERT_OK(Put(Key(j * kNumPerFile + k), values[k]));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put("", ""));
+      dbfull()->TEST_WaitForFlushMemTable();
+      if (j < kNumFiles - 1) {
+        // background compaction may happen early for kNumFiles'th file
+        ASSERT_EQ(NumTableFilesAtLevel(0), j + 1);
+      }
+      if (j == options.level0_file_num_compaction_trigger - 1) {
+        // When i == 1, compaction will output some files to L1, at which point
+        // L1 is not bottommost so range deletions cannot be compacted away. The
+        // new L1 files must be generated with non-overlapping key ranges even
+        // though multiple subcompactions see the same ranges deleted, else an
+        // assertion will fail.
+        dbfull()->TEST_WaitForCompact();
+        ASSERT_EQ(NumTableFilesAtLevel(0), 0);
+        ASSERT_GT(NumTableFilesAtLevel(1), 0);
+        ASSERT_GT(NumTableFilesAtLevel(2), 0);
+      }
+    }
+  }
+}
+
+TEST_F(DBRangeDelTest, ValidUniversalSubcompactionBoundaries) {
+  const int kNumPerFile = 100, kFilesPerLevel = 4, kNumLevels = 4;
+  Options options = CurrentOptions();
+  options.compaction_options_universal.min_merge_width = kFilesPerLevel;
+  options.compaction_options_universal.max_merge_width = kFilesPerLevel;
+  options.compaction_options_universal.size_ratio = 10;
+  options.compaction_style = kCompactionStyleUniversal;
+  options.level0_file_num_compaction_trigger = kFilesPerLevel;
+  options.max_subcompactions = 4;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  options.num_levels = kNumLevels;
+  options.target_file_size_base = kNumPerFile << 10;
+  options.target_file_size_multiplier = 1;
+  Reopen(options);
+
+  Random rnd(301);
+  for (int i = 0; i < kNumLevels - 1; ++i) {
+    for (int j = 0; j < kFilesPerLevel; ++j) {
+      if (i == kNumLevels - 2) {
+        // insert range deletions [95,105) in two files, [295,305) in next two
+        // to prepare L1 for later manual compaction.
+        int mid = (j + (1 - j % 2)) * kNumPerFile;
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                         Key(mid - 5), Key(mid + 5));
+      }
+      std::vector<std::string> values;
+      // Write 100KB (100 values, each 1K)
+      for (int k = 0; k < kNumPerFile; k++) {
+        values.push_back(RandomString(&rnd, 990));
+        ASSERT_OK(Put(Key(j * kNumPerFile + k), values[k]));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put("", ""));
+      dbfull()->TEST_WaitForFlushMemTable();
+      if (j < kFilesPerLevel - 1) {
+        // background compaction may happen early for kFilesPerLevel'th file
+        ASSERT_EQ(NumTableFilesAtLevel(0), j + 1);
+      }
+    }
+    dbfull()->TEST_WaitForCompact();
+    ASSERT_EQ(NumTableFilesAtLevel(0), 0);
+    ASSERT_GT(NumTableFilesAtLevel(kNumLevels - 1 - i), kFilesPerLevel - 1);
+  }
+  // Now L1-L3 are full, when we compact L1->L2 we should see (1) subcompactions
+  // happen since input level > 0; (2) range deletions are not dropped since
+  // output level is not bottommost. If no file boundary assertion fails, that
+  // probably means universal compaction + subcompaction + range deletion are
+  // compatible.
+  ASSERT_OK(dbfull()->RunManualCompaction(
+      reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
+          ->cfd(),
+      1 /* input_level */, 2 /* output_level */, 0 /* output_path_id */,
+      nullptr /* begin */, nullptr /* end */, true /* exclusive */,
+      true /* disallow_trivial_move */));
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBRangeDelTest, CompactionRemovesCoveredMergeOperands) {
+  const int kNumPerFile = 3, kNumFiles = 3;
+  Options opts = CurrentOptions();
+  opts.disable_auto_compactions = true;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(2 * kNumPerFile));
+  opts.merge_operator = MergeOperators::CreateUInt64AddOperator();
+  opts.num_levels = 2;
+  Reopen(opts);
+
+  // Iterates kNumFiles * kNumPerFile + 1 times since flushing the last file
+  // requires an extra entry.
+  for (int i = 0; i <= kNumFiles * kNumPerFile; ++i) {
+    if (i % kNumPerFile == 0 && i / kNumPerFile == kNumFiles - 1) {
+      // Delete merge operands from all but the last file
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "key",
+                       "key_");
+    }
+    std::string val;
+    PutFixed64(&val, i);
+    db_->Merge(WriteOptions(), "key", val);
+    // we need to prevent trivial move using Puts so compaction will actually
+    // process the merge operands.
+    db_->Put(WriteOptions(), "prevent_trivial_move", "");
+    if (i > 0 && i % kNumPerFile == 0) {
+      dbfull()->TEST_WaitForFlushMemTable();
+    }
+  }
+
+  ReadOptions read_opts;
+  read_opts.ignore_range_deletions = true;
+  std::string expected, actual;
+  ASSERT_OK(db_->Get(read_opts, "key", &actual));
+  PutFixed64(&expected, 45);  // 1+2+...+9
+  ASSERT_EQ(expected, actual);
+
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  expected.clear();
+  ASSERT_OK(db_->Get(read_opts, "key", &actual));
+  uint64_t tmp;
+  Slice tmp2(actual);
+  GetFixed64(&tmp2, &tmp);
+  PutFixed64(&expected, 30);  // 6+7+8+9 (earlier operands covered by tombstone)
+  ASSERT_EQ(expected, actual);
+}
+
+// NumTableFilesAtLevel() is not supported in ROCKSDB_LITE
+#ifndef ROCKSDB_LITE
+TEST_F(DBRangeDelTest, ObsoleteTombstoneCleanup) {
+  // During compaction to bottommost level, verify range tombstones older than
+  // the oldest snapshot are removed, while others are preserved.
+  Options opts = CurrentOptions();
+  opts.disable_auto_compactions = true;
+  opts.num_levels = 2;
+  opts.statistics = CreateDBStatistics();
+  Reopen(opts);
+
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
+                   "dr1");  // obsolete after compaction
+  db_->Put(WriteOptions(), "key", "val");
+  db_->Flush(FlushOptions());
+  const Snapshot* snapshot = db_->GetSnapshot();
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr2",
+                   "dr2");  // protected by snapshot
+  db_->Put(WriteOptions(), "key", "val");
+  db_->Flush(FlushOptions());
+
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(1));
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(1));
+  ASSERT_EQ(1, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
+
+  db_->ReleaseSnapshot(snapshot);
+}
+#endif  // ROCKSDB_LITE
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -299,8 +299,31 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DeleteScheduler::BackgroundEmptyTrash:Wait",
      [&](void* arg) { penalties.push_back(*(static_cast<int*>(arg))); });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "InstrumentedCondVar::TimedWaitInternal", [&](void* arg) {
+        // Turn timed wait into a simulated sleep
+        uint64_t* abs_time_us = static_cast<uint64_t*>(arg);
+        int64_t cur_time = 0;
+        env_->GetCurrentTime(&cur_time);
+        if (*abs_time_us > static_cast<uint64_t>(cur_time)) {
+          env_->addon_time_.fetch_add(*abs_time_us -
+                                      static_cast<uint64_t>(cur_time));
+        }
+
+        // Randomly sleep shortly
+        env_->addon_time_.fetch_add(
+            static_cast<uint64_t>(Random::GetTLSInstance()->Uniform(10)));
+
+        // Set wait until time to before current to force not to sleep.
+        int64_t real_cur_time = 0;
+        Env::Default()->GetCurrentTime(&real_cur_time);
+        *abs_time_us = static_cast<uint64_t>(real_cur_time);
+      });
+
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

+  env_->no_slowdown_ = true;
+  env_->time_elapse_only_sleep_ = true;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.env = env_;
@@ -348,6 +371,7 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
    ASSERT_EQ(expected_penlty, penalties[i]);
  }
  ASSERT_GT(time_spent_deleting, expected_penlty * 0.9);
+  ASSERT_LT(time_spent_deleting, expected_penlty * 1.1);

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
@@ -305,6 +305,7 @@ TEST_F(DBTestTailingIterator, TailingIteratorPrefixSeek) {
  options.disable_auto_compactions = true;
  options.prefix_extractor.reset(NewFixedPrefixTransform(2));
  options.memtable_factory.reset(NewHashSkipListRepFactory(16));
+  options.allow_concurrent_memtable_write = false;
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -428,6 +429,59 @@ TEST_F(DBTestTailingIterator, TailingIteratorUpperBound) {
  ASSERT_EQ(0, immutable_seeks);
 }

+TEST_F(DBTestTailingIterator, TailingIteratorGap) {
+  // level 1:            [20, 25]  [35, 40]
+  // level 2:  [10 - 15]                    [45 - 50]
+  // level 3:            [20,    30,    40]
+  // Previously there is a bug in tailing_iterator that if there is a gap in
+  // lower level, the key will be skipped if it is within the range between
+  // the largest key of index n file and the smallest key of index n+1 file
+  // if both file fit in that gap. In this example, 25 < key < 35
+  // https://github.com/facebook/rocksdb/issues/1372
+  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+
+  ReadOptions read_options;
+  read_options.tailing = true;
+
+  ASSERT_OK(Put(1, "20", "20"));
+  ASSERT_OK(Put(1, "30", "30"));
+  ASSERT_OK(Put(1, "40", "40"));
+  ASSERT_OK(Flush(1));
+  MoveFilesToLevel(3, 1);
+
+  ASSERT_OK(Put(1, "10", "10"));
+  ASSERT_OK(Put(1, "15", "15"));
+  ASSERT_OK(Flush(1));
+  ASSERT_OK(Put(1, "45", "45"));
+  ASSERT_OK(Put(1, "50", "50"));
+  ASSERT_OK(Flush(1));
+  MoveFilesToLevel(2, 1);
+
+  ASSERT_OK(Put(1, "20", "20"));
+  ASSERT_OK(Put(1, "25", "25"));
+  ASSERT_OK(Flush(1));
+  ASSERT_OK(Put(1, "35", "35"));
+  ASSERT_OK(Put(1, "40", "40"));
+  ASSERT_OK(Flush(1));
+  MoveFilesToLevel(1, 1);
+
+  ColumnFamilyMetaData meta;
+  db_->GetColumnFamilyMetaData(handles_[1], &meta);
+
+  std::unique_ptr<Iterator> it(db_->NewIterator(read_options, handles_[1]));
+  it->Seek("30");
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("30", it->key().ToString());
+
+  it->Next();
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("35", it->key().ToString());
+
+  it->Next();
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("40", it->key().ToString());
+}
+
 TEST_F(DBTestTailingIterator, ManagedTailingIteratorSingle) {
  ReadOptions read_options;
  read_options.tailing = true;
@@ -572,6 +626,7 @@ TEST_F(DBTestTailingIterator, ManagedTailingIteratorPrefixSeek) {
  options.disable_auto_compactions = true;
  options.prefix_extractor.reset(NewFixedPrefixTransform(2));
  options.memtable_factory.reset(NewHashSkipListRepFactory(16));
+  options.allow_concurrent_memtable_write = false;
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -699,6 +754,58 @@ TEST_F(DBTestTailingIterator, ForwardIteratorVersionProperty) {
  }
  ASSERT_EQ(v3, v4);
 }
+
+TEST_F(DBTestTailingIterator, SeekWithUpperBoundBug) {
+  ReadOptions read_options;
+  read_options.tailing = true;
+  const Slice upper_bound("cc", 3);
+  read_options.iterate_upper_bound = &upper_bound;
+
+
+  // 1st L0 file
+  ASSERT_OK(db_->Put(WriteOptions(), "aa", "SEEN"));
+  ASSERT_OK(Flush());
+
+  // 2nd L0 file
+  ASSERT_OK(db_->Put(WriteOptions(), "zz", "NOT-SEEN"));
+  ASSERT_OK(Flush());
+
+  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+
+  iter->Seek("aa");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().ToString(), "aa");
+}
+
+TEST_F(DBTestTailingIterator, SeekToFirstWithUpperBoundBug) {
+  ReadOptions read_options;
+  read_options.tailing = true;
+  const Slice upper_bound("cc", 3);
+  read_options.iterate_upper_bound = &upper_bound;
+
+
+  // 1st L0 file
+  ASSERT_OK(db_->Put(WriteOptions(), "aa", "SEEN"));
+  ASSERT_OK(Flush());
+
+  // 2nd L0 file
+  ASSERT_OK(db_->Put(WriteOptions(), "zz", "NOT-SEEN"));
+  ASSERT_OK(Flush());
+
+  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+
+  iter->SeekToFirst();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().ToString(), "aa");
+
+  iter->Next();
+  ASSERT_FALSE(iter->Valid());
+
+  iter->SeekToFirst();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().ToString(), "aa");
+}
+
 }  // namespace rocksdb

 #endif  // !defined(ROCKSDB_LITE)
@@ -215,6 +215,50 @@ TEST_F(DBTest, WriteEmptyBatch) {
  ASSERT_EQ("bar", Get(1, "foo"));
 }

+TEST_F(DBTest, SkipDelay) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.write_buffer_size = 100000;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  for (bool sync : {true, false}) {
+    for (bool disableWAL : {true, false}) {
+      // Use a small number to ensure a large delay that is still effective
+      // when we do Put
+      // TODO(myabandeh): this is time dependent and could potentially make
+      // the test flaky
+      auto token = dbfull()->TEST_write_controler().GetDelayToken(1);
+      std::atomic<int> sleep_count(0);
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "DBImpl::DelayWrite:Sleep",
+          [&](void* arg) { sleep_count.fetch_add(1); });
+      std::atomic<int> wait_count(0);
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "DBImpl::DelayWrite:Wait",
+          [&](void* arg) { wait_count.fetch_add(1); });
+      rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+      WriteOptions wo;
+      wo.sync = sync;
+      wo.disableWAL = disableWAL;
+      wo.no_slowdown = true;
+      dbfull()->Put(wo, "foo", "bar");
+      // We need the 2nd write to trigger delay. This is because delay is
+      // estimated based on the last write size which is 0 for the first write.
+      ASSERT_NOK(dbfull()->Put(wo, "foo2", "bar2"));
+      ASSERT_GE(sleep_count.load(), 0);
+      ASSERT_GE(wait_count.load(), 0);
+      token.reset();
+
+      token = dbfull()->TEST_write_controler().GetDelayToken(1000000000);
+      wo.no_slowdown = false;
+      ASSERT_OK(dbfull()->Put(wo, "foo3", "bar3"));
+      ASSERT_GE(sleep_count.load(), 1);
+      token.reset();
+    }
+  }
+}
+
 #ifndef ROCKSDB_LITE
 TEST_F(DBTest, ReadOnlyDB) {
  ASSERT_OK(Put("foo", "v1"));
@@ -1252,7 +1296,9 @@ TEST_F(DBTest, MinLevelToCompress2) {
  MinLevelHelper(this, options);
 }

-TEST_F(DBTest, RepeatedWritesToSameKey) {
+// This test may fail because of a legit case that multiple L0 files
+// are trivial moved to L1.
+TEST_F(DBTest, DISABLED_RepeatedWritesToSameKey) {
  do {
    Options options = CurrentOptions();
    options.env = env_;
@@ -2467,11 +2513,13 @@ class MultiThreadedDBTest : public DBTest,
 TEST_P(MultiThreadedDBTest, MultiThreaded) {
  anon::OptionsOverride options_override;
  options_override.skip_policy = kSkipNoSnapshot;
+  Options options = CurrentOptions(options_override);
  std::vector<std::string> cfs;
  for (int i = 1; i < kColumnFamilies; ++i) {
    cfs.push_back(ToString(i));
  }
-  CreateAndReopenWithCF(cfs, CurrentOptions(options_override));
+  Reopen(options);
+  CreateAndReopenWithCF(cfs, options);
  // Initialize state
  MTState mt;
  mt.test = this;
@@ -2646,15 +2694,11 @@ class ModelDB : public DB {
  }

 #ifndef ROCKSDB_LITE
-  using DB::AddFile;
-  virtual Status AddFile(ColumnFamilyHandle* column_family,
-                         const std::vector<ExternalSstFileInfo>& file_info_list,
-                         bool move_file, bool skip_snapshot_check) override {
-    return Status::NotSupported("Not implemented.");
-  }
-  virtual Status AddFile(ColumnFamilyHandle* column_family,
-                         const std::vector<std::string>& file_path_list,
-                         bool move_file, bool skip_snapshot_check) override {
+  using DB::IngestExternalFile;
+  virtual Status IngestExternalFile(
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& external_files,
+      const IngestExternalFileOptions& options) override {
    return Status::NotSupported("Not implemented.");
  }

@@ -2742,6 +2786,12 @@ class ModelDB : public DB {
                              const Slice& property, uint64_t* value) override {
    return false;
  }
+  using DB::GetMapProperty;
+  virtual bool GetMapProperty(ColumnFamilyHandle* column_family,
+                              const Slice& property,
+                              std::map<std::string, double>* value) override {
+    return false;
+  }
  using DB::GetAggregatedIntProperty;
  virtual bool GetAggregatedIntProperty(const Slice& property,
                                        uint64_t* value) override {
@@ -2762,6 +2812,12 @@ class ModelDB : public DB {
    return Status::NotSupported("Not supported operation.");
  }

+  virtual Status SetDBOptions(
+      const std::unordered_map<std::string, std::string>& new_options)
+      override {
+    return Status::NotSupported("Not supported operation.");
+  }
+
  using DB::CompactFiles;
  virtual Status CompactFiles(const CompactionOptions& compact_options,
                              ColumnFamilyHandle* column_family,
@@ -2806,13 +2862,12 @@ class ModelDB : public DB {
  virtual Env* GetEnv() const override { return nullptr; }

  using DB::GetOptions;
-  virtual const Options& GetOptions(
-      ColumnFamilyHandle* column_family) const override {
+  virtual Options GetOptions(ColumnFamilyHandle* column_family) const override {
    return options_;
  }

  using DB::GetDBOptions;
-  virtual const DBOptions& GetDBOptions() const override { return options_; }
+  virtual DBOptions GetDBOptions() const override { return options_; }

  using DB::Flush;
  virtual Status Flush(const rocksdb::FlushOptions& options,
@@ -2882,6 +2937,10 @@ class ModelDB : public DB {
    virtual void Seek(const Slice& k) override {
      iter_ = map_->lower_bound(k.ToString());
    }
+    virtual void SeekForPrev(const Slice& k) override {
+      iter_ = map_->upper_bound(k.ToString());
+      Prev();
+    }
    virtual void Next() override { ++iter_; }
    virtual void Prev() override {
      if (iter_ == map_->begin()) {
@@ -3639,14 +3698,24 @@ TEST_F(DBTest, GetThreadStatus) {
      env_->SetBackgroundThreads(kHighPriCounts[test], Env::HIGH);
      env_->SetBackgroundThreads(kLowPriCounts[test], Env::LOW);
      // Wait to ensure the all threads has been registered
-      env_->SleepForMicroseconds(100000);
-      s = env_->GetThreadList(&thread_list);
-      ASSERT_OK(s);
      unsigned int thread_type_counts[ThreadStatus::NUM_THREAD_TYPES];
-      memset(thread_type_counts, 0, sizeof(thread_type_counts));
-      for (auto thread : thread_list) {
-        ASSERT_LT(thread.thread_type, ThreadStatus::NUM_THREAD_TYPES);
-        thread_type_counts[thread.thread_type]++;
+      // Try up to 60 seconds.
+      for (int num_try = 0; num_try < 60000; num_try++) {
+        env_->SleepForMicroseconds(1000);
+        thread_list.clear();
+        s = env_->GetThreadList(&thread_list);
+        ASSERT_OK(s);
+        memset(thread_type_counts, 0, sizeof(thread_type_counts));
+        for (auto thread : thread_list) {
+          ASSERT_LT(thread.thread_type, ThreadStatus::NUM_THREAD_TYPES);
+          thread_type_counts[thread.thread_type]++;
+        }
+        if (thread_type_counts[ThreadStatus::HIGH_PRIORITY] ==
+                kHighPriCounts[test] &&
+            thread_type_counts[ThreadStatus::LOW_PRIORITY] ==
+                kLowPriCounts[test]) {
+          break;
+        }
      }
      // Verify the total number of threades
      ASSERT_EQ(thread_type_counts[ThreadStatus::HIGH_PRIORITY] +
@@ -4869,7 +4938,7 @@ TEST_F(DBTest, MergeTestTime) {
  SetPerfLevel(kEnableTime);
  this->env_->addon_time_.store(0);
  this->env_->time_elapse_only_sleep_ = true;
-  this->env_->no_sleep_ = true;
+  this->env_->no_slowdown_ = true;
  Options options = CurrentOptions();
  options.statistics = rocksdb::CreateDBStatistics();
  options.merge_operator.reset(new DelayedMergeOperator(this));
@@ -5309,12 +5378,12 @@ TEST_F(DBTest, FlushesInParallelWithCompactRange) {

 TEST_F(DBTest, DelayedWriteRate) {
  const int kEntriesPerMemTable = 100;
-  const int kTotalFlushes = 20;
+  const int kTotalFlushes = 12;

  Options options = CurrentOptions();
  env_->SetBackgroundThreads(1, Env::LOW);
  options.env = env_;
-  env_->no_sleep_ = true;
+  env_->no_slowdown_ = true;
  options.write_buffer_size = 100000000;
  options.max_write_buffer_number = 256;
  options.max_background_compactions = 1;
@@ -5359,8 +5428,8 @@ TEST_F(DBTest, DelayedWriteRate) {
    dbfull()->TEST_WaitForFlushMemTable();
    estimated_sleep_time += size_memtable * 1000000u / cur_rate;
    // Slow down twice. One for memtable switch and one for flush finishes.
-    cur_rate = static_cast<uint64_t>(static_cast<double>(cur_rate) /
-                                     kSlowdownRatio / kSlowdownRatio);
+    cur_rate = static_cast<uint64_t>(static_cast<double>(cur_rate) *
+                                     kIncSlowdownRatio * kIncSlowdownRatio);
  }
  // Estimate the total sleep time fall into the rough range.
  ASSERT_GT(env_->addon_time_.load(),
@@ -5368,7 +5437,7 @@ TEST_F(DBTest, DelayedWriteRate) {
  ASSERT_LT(env_->addon_time_.load(),
            static_cast<int64_t>(estimated_sleep_time * 2));

-  env_->no_sleep_ = false;
+  env_->no_slowdown_ = false;
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
  sleeping_task_low.WakeUp();
  sleeping_task_low.WaitUntilDone();
@@ -145,6 +145,22 @@ TEST_F(DBTest2, CacheIndexAndFilterWithDBRestart) {
  value = Get(1, "a");
 }

+TEST_F(DBTest2, MaxSuccessiveMergesChangeWithDBRecovery) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.max_successive_merges = 3;
+  options.merge_operator = MergeOperators::CreatePutOperator();
+  options.disable_auto_compactions = true;
+  DestroyAndReopen(options);
+  Put("poi", "Finch");
+  db_->Merge(WriteOptions(), "poi", "Reese");
+  db_->Merge(WriteOptions(), "poi", "Shaw");
+  db_->Merge(WriteOptions(), "poi", "Root");
+  options.max_successive_merges = 2;
+  Reopen(options);
+}
+
 #ifndef ROCKSDB_LITE
 class DBTestSharedWriteBufferAcrossCFs
    : public DBTestBase,
@@ -940,6 +956,7 @@ TEST_F(DBTest2, PresetCompressionDict) {
  const int kNumL0Files = 5;

  Options options;
+  options.allow_concurrent_memtable_write = false;
  options.arena_block_size = kBlockSizeBytes;
  options.compaction_style = kCompactionStyleUniversal;
  options.create_if_missing = true;
@@ -1453,6 +1470,10 @@ class MockPersistentCache : public PersistentCache {

  virtual ~MockPersistentCache() {}

+  PersistentCache::StatsType Stats() override {
+    return PersistentCache::StatsType();
+  }
+
  Status Insert(const Slice& page_key, const char* data,
                const size_t size) override {
    MutexLock _(&lock_);
@@ -1889,23 +1910,6 @@ TEST_P(MergeOperatorPinningTest, TailingIterator) {
 }
 #endif  // ROCKSDB_LITE

-TEST_F(DBTest2, MaxSuccessiveMergesInRecovery) {
-  Options options;
-  options = CurrentOptions(options);
-  options.merge_operator = MergeOperators::CreatePutOperator();
-  DestroyAndReopen(options);
-
-  db_->Put(WriteOptions(), "foo", "bar");
-  ASSERT_OK(db_->Merge(WriteOptions(), "foo", "bar"));
-  ASSERT_OK(db_->Merge(WriteOptions(), "foo", "bar"));
-  ASSERT_OK(db_->Merge(WriteOptions(), "foo", "bar"));
-  ASSERT_OK(db_->Merge(WriteOptions(), "foo", "bar"));
-  ASSERT_OK(db_->Merge(WriteOptions(), "foo", "bar"));
-
-  options.max_successive_merges = 3;
-  Reopen(options);
-}
-
 size_t GetEncodedEntrySize(size_t key_size, size_t value_size) {
  std::string buffer;

@@ -2066,6 +2070,162 @@ TEST_F(DBTest2, ReadAmpBitmapLiveInCacheAfterDBClose) {
            total_loaded_bytes_iter1 + total_loaded_bytes_iter2);
 }

+#ifndef ROCKSDB_LITE
+TEST_F(DBTest2, AutomaticCompactionOverlapManualCompaction) {
+  Options options = CurrentOptions();
+  options.num_levels = 3;
+  options.IncreaseParallelism(20);
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put(Key(0), "a"));
+  ASSERT_OK(Put(Key(5), "a"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(Key(10), "a"));
+  ASSERT_OK(Put(Key(15), "a"));
+  ASSERT_OK(Flush());
+
+  CompactRangeOptions cro;
+  cro.change_level = true;
+  cro.target_level = 2;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  auto get_stat = [](std::string level_str, LevelStatType type,
+                     std::map<std::string, double> props) {
+    auto prop_str =
+        level_str + "." +
+        InternalStats::compaction_level_stats.at(type).property_name.c_str();
+    auto prop_item = props.find(prop_str);
+    return prop_item == props.end() ? 0 : prop_item->second;
+  };
+
+  // Trivial move 2 files to L2
+  ASSERT_EQ("0,0,2", FilesPerLevel());
+  // Also test that the stats GetMapProperty API reporting the same result
+  {
+    std::map<std::string, double> prop;
+    ASSERT_TRUE(dbfull()->GetMapProperty("rocksdb.cfstats", &prop));
+    ASSERT_EQ(0, get_stat("L0", LevelStatType::NUM_FILES, prop));
+    ASSERT_EQ(0, get_stat("L1", LevelStatType::NUM_FILES, prop));
+    ASSERT_EQ(2, get_stat("L2", LevelStatType::NUM_FILES, prop));
+    ASSERT_EQ(2, get_stat("Sum", LevelStatType::NUM_FILES, prop));
+  }
+
+  // While the compaction is running, we will create 2 new files that
+  // can fit in L2, these 2 files will be moved to L2 and overlap with
+  // the running compaction and break the LSM consistency.
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::Run():Start", [&](void* arg) {
+        ASSERT_OK(
+            dbfull()->SetOptions({{"level0_file_num_compaction_trigger", "2"},
+                                  {"max_bytes_for_level_base", "1"}}));
+        ASSERT_OK(Put(Key(6), "a"));
+        ASSERT_OK(Put(Key(7), "a"));
+        ASSERT_OK(Flush());
+
+        ASSERT_OK(Put(Key(8), "a"));
+        ASSERT_OK(Put(Key(9), "a"));
+        ASSERT_OK(Flush());
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Run a manual compaction that will compact the 2 files in L2
+  // into 1 file in L2
+  cro.exclusive_manual_compaction = false;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  // Test that the stats GetMapProperty API reporting 1 file in L2
+  {
+    std::map<std::string, double> prop;
+    ASSERT_TRUE(dbfull()->GetMapProperty("rocksdb.cfstats", &prop));
+    ASSERT_EQ(1, get_stat("L2", LevelStatType::NUM_FILES, prop));
+  }
+}
+
+TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {
+  Options options = CurrentOptions();
+  options.num_levels = 2;
+  options.IncreaseParallelism(20);
+  options.disable_auto_compactions = true;
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put(Key(0), "a"));
+  ASSERT_OK(Put(Key(5), "a"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(Key(10), "a"));
+  ASSERT_OK(Put(Key(15), "a"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  // Trivial move 2 files to L1
+  ASSERT_EQ("0,2", FilesPerLevel());
+
+  std::function<void()> bg_manual_compact = [&]() {
+    std::string k1 = Key(6);
+    std::string k2 = Key(9);
+    Slice k1s(k1);
+    Slice k2s(k2);
+    CompactRangeOptions cro;
+    cro.exclusive_manual_compaction = false;
+    ASSERT_OK(db_->CompactRange(cro, &k1s, &k2s));
+  };
+  std::thread bg_thread;
+
+  // While the compaction is running, we will create 2 new files that
+  // can fit in L1, these 2 files will be moved to L1 and overlap with
+  // the running compaction and break the LSM consistency.
+  std::atomic<bool> flag(false);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::Run():Start", [&](void* arg) {
+        if (flag.exchange(true)) {
+          // We want to make sure to call this callback only once
+          return;
+        }
+        ASSERT_OK(Put(Key(6), "a"));
+        ASSERT_OK(Put(Key(7), "a"));
+        ASSERT_OK(Flush());
+
+        ASSERT_OK(Put(Key(8), "a"));
+        ASSERT_OK(Put(Key(9), "a"));
+        ASSERT_OK(Flush());
+
+        // Start a non-exclusive manual compaction in a bg thread
+        bg_thread = std::thread(bg_manual_compact);
+        // This manual compaction conflict with the other manual compaction
+        // so it should wait until the first compaction finish
+        env_->SleepForMicroseconds(1000000);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Run a manual compaction that will compact the 2 files in L1
+  // into 1 file in L1
+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = false;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  bg_thread.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBTest2, OptimizeForPointLookup) {
+  Options options = CurrentOptions();
+  Close();
+  options.OptimizeForPointLookup(2);
+  ASSERT_OK(DB::Open(options, dbname_, &db_));
+
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_EQ("v1", Get("foo"));
+  Flush();
+  ASSERT_EQ("v1", Get("foo"));
+}
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include "db/db_test_util.h"
+#include "db/forward_iterator.h"

 namespace rocksdb {

@@ -19,7 +20,7 @@ SpecialEnv::SpecialEnv(Env* base)
      sleep_counter_(this),
      addon_time_(0),
      time_elapse_only_sleep_(false),
-      no_sleep_(false) {
+      no_slowdown_(false) {
  delay_sstable_sync_.store(false, std::memory_order_release);
  drop_writes_.store(false, std::memory_order_release);
  no_space_.store(false, std::memory_order_release);
@@ -233,6 +234,7 @@ Options DBTestBase::CurrentOptions(
    case kHashSkipList:
      options.prefix_extractor.reset(NewFixedPrefixTransform(1));
      options.memtable_factory.reset(NewHashSkipListRepFactory(16));
+      options.allow_concurrent_memtable_write = false;
      break;
    case kPlainTableFirstBytePrefix:
      options.table_factory.reset(new PlainTableFactory());
@@ -264,15 +266,18 @@ Options DBTestBase::CurrentOptions(
      break;
    case kVectorRep:
      options.memtable_factory.reset(new VectorRepFactory(100));
+      options.allow_concurrent_memtable_write = false;
      break;
    case kHashLinkList:
      options.prefix_extractor.reset(NewFixedPrefixTransform(1));
      options.memtable_factory.reset(
          NewHashLinkListRepFactory(4, 0, 3, true, 4));
+      options.allow_concurrent_memtable_write = false;
      break;
    case kHashCuckoo:
      options.memtable_factory.reset(
          NewHashCuckooRepFactory(options.write_buffer_size));
+      options.allow_concurrent_memtable_write = false;
      break;
 #endif  // ROCKSDB_LITE
    case kMergePut:
@@ -497,6 +502,15 @@ Status DBTestBase::Put(int cf, const Slice& k, const Slice& v,
  }
 }

+Status DBTestBase::Merge(const Slice& k, const Slice& v, WriteOptions wo) {
+  return db_->Merge(wo, k, v);
+}
+
+Status DBTestBase::Merge(int cf, const Slice& k, const Slice& v,
+                         WriteOptions wo) {
+  return db_->Merge(wo, handles_[cf], k, v);
+}
+
 Status DBTestBase::Delete(const std::string& k) {
  return db_->Delete(WriteOptions(), k);
 }
@@ -585,11 +599,15 @@ std::string DBTestBase::Contents(int cf) {

 std::string DBTestBase::AllEntriesFor(const Slice& user_key, int cf) {
  Arena arena;
+  auto options = CurrentOptions();
+  InternalKeyComparator icmp(options.comparator);
+  RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
  ScopedArenaIterator iter;
  if (cf == 0) {
-    iter.set(dbfull()->NewInternalIterator(&arena));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg));
  } else {
-    iter.set(dbfull()->NewInternalIterator(&arena, handles_[cf]));
+    iter.set(
+        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[cf]));
  }
  InternalKey target(user_key, kMaxSequenceNumber, kTypeValue);
  iter->Seek(target.Encode());
@@ -990,10 +1008,14 @@ UpdateStatus DBTestBase::updateInPlaceNoAction(char* prevValue,
 void DBTestBase::validateNumberOfEntries(int numValues, int cf) {
  ScopedArenaIterator iter;
  Arena arena;
+  auto options = CurrentOptions();
+  InternalKeyComparator icmp(options.comparator);
+  RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
  if (cf != 0) {
-    iter.set(dbfull()->NewInternalIterator(&arena, handles_[cf]));
+    iter.set(
+        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[cf]));
  } else {
-    iter.set(dbfull()->NewInternalIterator(&arena));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg));
  }
  iter->SeekToFirst();
  ASSERT_EQ(iter->status().ok(), true);
@@ -1077,11 +1099,18 @@ std::vector<std::uint64_t> DBTestBase::ListTableFiles(Env* env,
 }

 void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
-                                 size_t* total_reads_res) {
+                                 size_t* total_reads_res, bool tailing_iter,
+                                 std::map<std::string, Status> status) {
  size_t total_reads = 0;

  for (auto& kv : true_data) {
-    ASSERT_EQ(Get(kv.first), kv.second);
+    Status s = status[kv.first];
+    if (s.ok()) {
+      ASSERT_EQ(Get(kv.first), kv.second);
+    } else {
+      std::string value;
+      ASSERT_EQ(s, db_->Get(ReadOptions(), kv.first, &value));
+    }
    total_reads++;
  }

@@ -1094,21 +1123,40 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
    // Verify Iterator::Next()
    iter_cnt = 0;
    auto data_iter = true_data.begin();
+    Status s;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next(), data_iter++) {
      ASSERT_EQ(iter->key().ToString(), data_iter->first);
-      ASSERT_EQ(iter->value().ToString(), data_iter->second);
+      Status current_status = status[data_iter->first];
+      if (!current_status.ok()) {
+        s = current_status;
+      }
+      ASSERT_EQ(iter->status(), s);
+      if (current_status.ok()) {
+        ASSERT_EQ(iter->value().ToString(), data_iter->second);
+      }
      iter_cnt++;
      total_reads++;
    }
    ASSERT_EQ(data_iter, true_data.end()) << iter_cnt << " / "
                                          << true_data.size();
+    delete iter;

    // Verify Iterator::Prev()
+    // Use a new iterator to make sure its status is clean.
+    iter = db_->NewIterator(ro);
    iter_cnt = 0;
+    s = Status::OK();
    auto data_rev = true_data.rbegin();
    for (iter->SeekToLast(); iter->Valid(); iter->Prev(), data_rev++) {
      ASSERT_EQ(iter->key().ToString(), data_rev->first);
-      ASSERT_EQ(iter->value().ToString(), data_rev->second);
+      Status current_status = status[data_rev->first];
+      if (!current_status.ok()) {
+        s = current_status;
+      }
+      ASSERT_EQ(iter->status(), s);
+      if (current_status.ok()) {
+        ASSERT_EQ(iter->value().ToString(), data_rev->second);
+      }
      iter_cnt++;
      total_reads++;
    }
@@ -1122,76 +1170,68 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
      ASSERT_EQ(kv.second, iter->value().ToString());
      total_reads++;
    }
-
    delete iter;
  }

+  if (tailing_iter) {
 #ifndef ROCKSDB_LITE
-  // Tailing iterator
-  int iter_cnt = 0;
-  ReadOptions ro;
-  ro.tailing = true;
-  ro.total_order_seek = true;
-  Iterator* iter = db_->NewIterator(ro);
+    // Tailing iterator
+    int iter_cnt = 0;
+    ReadOptions ro;
+    ro.tailing = true;
+    ro.total_order_seek = true;
+    Iterator* iter = db_->NewIterator(ro);

-  // Verify ForwardIterator::Next()
-  iter_cnt = 0;
-  auto data_iter = true_data.begin();
-  for (iter->SeekToFirst(); iter->Valid(); iter->Next(), data_iter++) {
-    ASSERT_EQ(iter->key().ToString(), data_iter->first);
-    ASSERT_EQ(iter->value().ToString(), data_iter->second);
-    iter_cnt++;
-    total_reads++;
-  }
-  ASSERT_EQ(data_iter, true_data.end()) << iter_cnt << " / "
-                                        << true_data.size();
+    // Verify ForwardIterator::Next()
+    iter_cnt = 0;
+    auto data_iter = true_data.begin();
+    for (iter->SeekToFirst(); iter->Valid(); iter->Next(), data_iter++) {
+      ASSERT_EQ(iter->key().ToString(), data_iter->first);
+      ASSERT_EQ(iter->value().ToString(), data_iter->second);
+      iter_cnt++;
+      total_reads++;
+    }
+    ASSERT_EQ(data_iter, true_data.end()) << iter_cnt << " / "
+                                          << true_data.size();

-  // Verify ForwardIterator::Seek()
-  for (auto kv : true_data) {
-    iter->Seek(kv.first);
-    ASSERT_EQ(kv.first, iter->key().ToString());
-    ASSERT_EQ(kv.second, iter->value().ToString());
-    total_reads++;
-  }
+    // Verify ForwardIterator::Seek()
+    for (auto kv : true_data) {
+      iter->Seek(kv.first);
+      ASSERT_EQ(kv.first, iter->key().ToString());
+      ASSERT_EQ(kv.second, iter->value().ToString());
+      total_reads++;
+    }

-  delete iter;
+    delete iter;
 #endif  // ROCKSDB_LITE
+  }

  if (total_reads_res) {
    *total_reads_res = total_reads;
  }
 }

-#ifndef ROCKSDB_LITE
-
-Status DBTestBase::GenerateAndAddExternalFile(const Options options,
-                                              std::vector<int> keys,
-                                              size_t file_id) {
-  std::string file_path =
-      test::TmpDir(env_) + "/sst_files/" + ToString(file_id);
-  SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
-
-  Status s = sst_file_writer.Open(file_path);
-  if (!s.ok()) {
-    return s;
-  }
-  for (auto& entry : keys) {
-    std::string k = Key(entry);
-    std::string v = k + ToString(file_id);
-    s = sst_file_writer.Add(k, v);
-    if (!s.ok()) {
-      return s;
-    }
-  }
-  s = sst_file_writer.Finish();
-
-  if (s.ok()) {
-    s = db_->AddFile(std::vector<std::string>(1, file_path));
-  }
-
-  return s;
+void DBTestBase::VerifyDBInternal(
+    std::vector<std::pair<std::string, std::string>> true_data) {
+  Arena arena;
+  InternalKeyComparator icmp(last_options_.comparator);
+  RangeDelAggregator range_del_agg(icmp, {});
+  auto iter = dbfull()->NewInternalIterator(&arena, &range_del_agg);
+  iter->SeekToFirst();
+  for (auto p : true_data) {
+    ASSERT_TRUE(iter->Valid());
+    ParsedInternalKey ikey;
+    ASSERT_TRUE(ParseInternalKey(iter->key(), &ikey));
+    ASSERT_EQ(p.first, ikey.user_key);
+    ASSERT_EQ(p.second, iter->value());
+    iter->Next();
+  };
+  ASSERT_FALSE(iter->Valid());
+  iter->~InternalIterator();
 }

+#ifndef ROCKSDB_LITE
+
 uint64_t DBTestBase::GetNumberOfSstFilesForColumnFamily(
    DB* db, std::string column_family_name) {
  std::vector<LiveFileMetaData> metadata;
@@ -193,6 +193,10 @@ class SpecialSkipListFactory : public MemTableRepFactory {
  }
  virtual const char* Name() const override { return "SkipListFactory"; }

+  bool IsInsertConcurrentlySupported() const override {
+    return factory_.IsInsertConcurrentlySupported();
+  }
+
 private:
  SkipListFactory factory_;
  int num_entries_flush_;
@@ -310,7 +314,18 @@ class SpecialEnv : public EnvWrapper {
        return s;
      }
      Status Truncate(uint64_t size) override { return base_->Truncate(size); }
-      Status Close() override { return base_->Close(); }
+      Status Close() override {
+// SyncPoint is not supported in Released Windows Mode.
+#if !(defined NDEBUG) || !defined(OS_WIN)
+        // Check preallocation size
+        // preallocation size is never passed to base file.
+        size_t preallocation_size = preallocation_block_size();
+        TEST_SYNC_POINT_CALLBACK("DBTestWalFile.GetPreallocationStatus",
+                                 &preallocation_size);
+#endif  // !(defined NDEBUG) || !defined(OS_WIN)
+
+        return base_->Close();
+      }
      Status Flush() override { return base_->Flush(); }
      Status Sync() override {
        ++env_->sync_counter_;
@@ -417,10 +432,10 @@ class SpecialEnv : public EnvWrapper {

  virtual void SleepForMicroseconds(int micros) override {
    sleep_counter_.Increment();
-    if (no_sleep_ || time_elapse_only_sleep_) {
+    if (no_slowdown_ || time_elapse_only_sleep_) {
      addon_time_.fetch_add(micros);
    }
-    if (!no_sleep_) {
+    if (!no_slowdown_) {
      target()->SleepForMicroseconds(micros);
    }
  }
@@ -509,7 +524,7 @@ class SpecialEnv : public EnvWrapper {

  bool time_elapse_only_sleep_;

-  bool no_sleep_;
+  bool no_slowdown_;

  std::atomic<bool> is_wal_sync_thread_safe_{true};
 };
@@ -684,6 +699,12 @@ class DBTestBase : public testing::Test {
  Status Put(int cf, const Slice& k, const Slice& v,
             WriteOptions wo = WriteOptions());

+  Status Merge(const Slice& k, const Slice& v,
+               WriteOptions wo = WriteOptions());
+
+  Status Merge(int cf, const Slice& k, const Slice& v,
+               WriteOptions wo = WriteOptions());
+
  Status Delete(const std::string& k);

  Status Delete(int cf, const std::string& k);
@@ -811,12 +832,14 @@ class DBTestBase : public testing::Test {
  std::vector<std::uint64_t> ListTableFiles(Env* env, const std::string& path);

  void VerifyDBFromMap(std::map<std::string, std::string> true_data,
-                       size_t* total_reads_res = nullptr);
+                       size_t* total_reads_res = nullptr,
+                       bool tailing_iter = false,
+                       std::map<std::string, Status> status = {});
+
+  void VerifyDBInternal(
+      std::vector<std::pair<std::string, std::string>> true_data);

 #ifndef ROCKSDB_LITE
-  Status GenerateAndAddExternalFile(const Options options,
-                                    std::vector<int> keys, size_t file_id);
-
  uint64_t GetNumberOfSstFilesForColumnFamily(DB* db,
                                              std::string column_family_name);
 #endif  // ROCKSDB_LITE
@@ -1044,7 +1044,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
  options.num_levels = 15;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 8;
-  options.max_background_compactions = 4;
+  options.max_background_compactions = 1;
  options.target_file_size_base = 64 * 1024;
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);
@@ -9,6 +9,7 @@

 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "util/fault_injection_test_env.h"
 #include "util/options_helper.h"
 #include "util/sync_point.h"

@@ -291,7 +292,170 @@ TEST_F(DBWALTest, RecoveryWithEmptyLog) {
  } while (ChangeOptions());
 }

+#if !(defined NDEBUG) || !defined(OS_WIN)
+TEST_F(DBWALTest, PreallocateBlock) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 10 * 1000 * 1000;
+  options.max_total_wal_size = 0;
+
+  size_t expected_preallocation_size = static_cast<size_t>(
+      options.write_buffer_size + options.write_buffer_size / 10);
+
+  DestroyAndReopen(options);
+
+  std::atomic<int> called(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBTestWalFile.GetPreallocationStatus", [&](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        size_t preallocation_size = *(static_cast<size_t*>(arg));
+        ASSERT_EQ(expected_preallocation_size, preallocation_size);
+        called.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  Put("", "");
+  Flush();
+  Put("", "");
+  Close();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  ASSERT_EQ(2, called.load());
+
+  options.max_total_wal_size = 1000 * 1000;
+  expected_preallocation_size = static_cast<size_t>(options.max_total_wal_size);
+  Reopen(options);
+  called.store(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBTestWalFile.GetPreallocationStatus", [&](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        size_t preallocation_size = *(static_cast<size_t*>(arg));
+        ASSERT_EQ(expected_preallocation_size, preallocation_size);
+        called.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  Put("", "");
+  Flush();
+  Put("", "");
+  Close();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  ASSERT_EQ(2, called.load());
+
+  options.db_write_buffer_size = 800 * 1000;
+  expected_preallocation_size =
+      static_cast<size_t>(options.db_write_buffer_size);
+  Reopen(options);
+  called.store(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBTestWalFile.GetPreallocationStatus", [&](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        size_t preallocation_size = *(static_cast<size_t*>(arg));
+        ASSERT_EQ(expected_preallocation_size, preallocation_size);
+        called.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  Put("", "");
+  Flush();
+  Put("", "");
+  Close();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  ASSERT_EQ(2, called.load());
+
+  expected_preallocation_size = 700 * 1000;
+  std::shared_ptr<WriteBufferManager> write_buffer_manager =
+      std::make_shared<WriteBufferManager>(static_cast<uint64_t>(700 * 1000));
+  options.write_buffer_manager = write_buffer_manager;
+  Reopen(options);
+  called.store(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBTestWalFile.GetPreallocationStatus", [&](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        size_t preallocation_size = *(static_cast<size_t*>(arg));
+        ASSERT_EQ(expected_preallocation_size, preallocation_size);
+        called.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  Put("", "");
+  Flush();
+  Put("", "");
+  Close();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  ASSERT_EQ(2, called.load());
+}
+#endif  // !(defined NDEBUG) || !defined(OS_WIN)
+
 #ifndef ROCKSDB_LITE
+TEST_F(DBWALTest, FullPurgePreservesRecycledLog) {
+  // For github issue #1303
+  for (int i = 0; i < 2; ++i) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.recycle_log_file_num = 2;
+    if (i != 0) {
+      options.wal_dir = alternative_wal_dir_;
+    }
+
+    DestroyAndReopen(options);
+    ASSERT_OK(Put("foo", "v1"));
+    VectorLogPtr log_files;
+    ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
+    ASSERT_GT(log_files.size(), 0);
+    ASSERT_OK(Flush());
+
+    // Now the original WAL is in log_files[0] and should be marked for
+    // recycling.
+    // Verify full purge cannot remove this file.
+    JobContext job_context(0);
+    dbfull()->TEST_LockMutex();
+    dbfull()->FindObsoleteFiles(&job_context, true /* force */);
+    dbfull()->TEST_UnlockMutex();
+    dbfull()->PurgeObsoleteFiles(job_context);
+
+    if (i == 0) {
+      ASSERT_OK(
+          env_->FileExists(LogFileName(dbname_, log_files[0]->LogNumber())));
+    } else {
+      ASSERT_OK(env_->FileExists(
+          LogFileName(alternative_wal_dir_, log_files[0]->LogNumber())));
+    }
+  }
+}
+
+TEST_F(DBWALTest, GetSortedWalFiles) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    VectorLogPtr log_files;
+    ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
+    ASSERT_EQ(0, log_files.size());
+
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
+    ASSERT_EQ(1, log_files.size());
+  } while (ChangeOptions());
+}
+
+TEST_F(DBWALTest, RecoveryWithLogDataForSomeCFs) {
+  // Test for regression of WAL cleanup missing files that don't contain data
+  // for every column family.
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "foo", "v2"));
+    uint64_t earliest_log_nums[2];
+    for (int i = 0; i < 2; ++i) {
+      if (i > 0) {
+        ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+      }
+      VectorLogPtr log_files;
+      ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
+      if (log_files.size() > 0) {
+        earliest_log_nums[i] = log_files[0]->LogNumber();
+      } else {
+        earliest_log_nums[i] = port::kMaxUint64;
+      }
+    }
+    // Check at least the first WAL was cleaned up during the recovery.
+    ASSERT_LT(earliest_log_nums[0], earliest_log_nums[1]);
+  } while (ChangeOptions());
+}
+
 TEST_F(DBWALTest, RecoverWithLargeLog) {
  do {
    {
@@ -471,6 +635,39 @@ TEST_F(DBWALTest, SyncMultipleLogs) {
  ASSERT_OK(dbfull()->SyncWAL());
 }

+// Github issue 1339. Prior the fix we read sequence id from the first log to
+// a local variable, then keep increase the variable as we replay logs,
+// ignoring actual sequence id of the records. This is incorrect if some writes
+// come with WAL disabled.
+TEST_F(DBWALTest, PartOfWritesWithWALDisabled) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(env_));
+  Options options = CurrentOptions();
+  options.env = fault_env.get();
+  options.disable_auto_compactions = true;
+  WriteOptions wal_on, wal_off;
+  wal_on.sync = true;
+  wal_on.disableWAL = false;
+  wal_off.disableWAL = true;
+  CreateAndReopenWithCF({"dummy"}, options);
+  ASSERT_OK(Put(1, "dummy", "d1", wal_on));  // seq id 1
+  ASSERT_OK(Put(1, "dummy", "d2", wal_off));
+  ASSERT_OK(Put(1, "dummy", "d3", wal_off));
+  ASSERT_OK(Put(0, "key", "v4", wal_on));  // seq id 4
+  ASSERT_OK(Flush(0));
+  ASSERT_OK(Put(0, "key", "v5", wal_on));  // seq id 5
+  ASSERT_EQ("v5", Get(0, "key"));
+  // Simulate a crash.
+  fault_env->SetFilesystemActive(false);
+  Close();
+  fault_env->ResetState();
+  ReopenWithColumnFamilies({"default", "dummy"}, options);
+  // Prior to the fix, we may incorrectly recover "v5" with sequence id = 3.
+  ASSERT_EQ("v5", Get(0, "key"));
+  // Destroy DB before destruct fault_env.
+  Destroy(options);
+}
+
 //
 // Test WAL recovery for the various modes available
 //
@@ -488,7 +685,7 @@ class RecoveryTestHelper {
  // Create WAL files with values filled in
  static void FillData(DBWALTest* test, const Options& options,
                       const size_t wal_count, size_t* count) {
-    const DBOptions& db_options = options;
+    const ImmutableDBOptions db_options(options);

    *count = 0;

@@ -807,6 +1004,32 @@ TEST_F(DBWALTest, AvoidFlushDuringRecovery) {
  ASSERT_EQ(2, TotalTableFiles());
 }

+TEST_F(DBWALTest, WalCleanupAfterAvoidFlushDuringRecovery) {
+  // Verifies WAL files that were present during recovery, but not flushed due
+  // to avoid_flush_during_recovery, will be considered for deletion at a later
+  // stage. We check at least one such file is deleted during Flush().
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.avoid_flush_during_recovery = true;
+  Reopen(options);
+
+  ASSERT_OK(Put("foo", "v1"));
+  Reopen(options);
+  for (int i = 0; i < 2; ++i) {
+    if (i > 0) {
+      // Flush() triggers deletion of obsolete tracked files
+      Flush();
+    }
+    VectorLogPtr log_files;
+    ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
+    if (i == 0) {
+      ASSERT_GT(log_files.size(), 0);
+    } else {
+      ASSERT_EQ(0, log_files.size());
+    }
+  }
+}
+
 TEST_F(DBWALTest, RecoverWithoutFlush) {
  Options options = CurrentOptions();
  options.avoid_flush_during_recovery = true;
@@ -967,6 +1190,29 @@ TEST_F(DBWALTest, RecoverFromCorruptedWALWithoutFlush) {

 #endif  // ROCKSDB_LITE

+TEST_F(DBWALTest, WalTermTest) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+
+  WriteOptions wo;
+  wo.sync = true;
+  wo.disableWAL = false;
+
+  WriteBatch batch;
+  batch.Put("foo", "bar");
+  batch.MarkWalTerminationPoint();
+  batch.Put("foo2", "bar2");
+
+  ASSERT_OK(dbfull()->Write(wo, &batch));
+
+  // make sure we can re-open it.
+  ASSERT_OK(TryReopenWithColumnFamilies({"default", "pikachu"}, options));
+  ASSERT_EQ("bar", Get(1, "foo"));
+  ASSERT_EQ("NOT_FOUND", Get(1, "foo2"));
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -20,6 +20,15 @@

 namespace rocksdb {

+// kValueTypeForSeek defines the ValueType that should be passed when
+// constructing a ParsedInternalKey object for seeking to a particular
+// sequence number (since we sort sequence numbers in decreasing order
+// and the value type is embedded as the low 8 bits in the sequence
+// number in internal keys, we need to use the highest-numbered
+// ValueType, not the lowest).
+const ValueType kValueTypeForSeek = kTypeSingleDeletion;
+const ValueType kValueTypeForSeekForPrev = kTypeDeletion;
+
 uint64_t PackSequenceAndType(uint64_t seq, ValueType t) {
  assert(seq <= kMaxSequenceNumber);
  assert(IsExtendedValueType(t));
@@ -39,6 +48,11 @@ void AppendInternalKey(std::string* result, const ParsedInternalKey& key) {
  PutFixed64(result, PackSequenceAndType(key.sequence, key.type));
 }

+void AppendInternalKeyFooter(std::string* result, SequenceNumber s,
+                             ValueType t) {
+  PutFixed64(result, PackSequenceAndType(s, t));
+}
+
 std::string ParsedInternalKey::DebugString(bool hex) const {
  char buf[50];
  snprintf(buf, sizeof(buf), "' @ %" PRIu64 ": %d", sequence,
@@ -50,13 +50,9 @@ enum ValueType : unsigned char {
  kMaxValue = 0x7F                        // Not used for storing records.
 };

-// kValueTypeForSeek defines the ValueType that should be passed when
-// constructing a ParsedInternalKey object for seeking to a particular
-// sequence number (since we sort sequence numbers in decreasing order
-// and the value type is embedded as the low 8 bits in the sequence
-// number in internal keys, we need to use the highest-numbered
-// ValueType, not the lowest).
-static const ValueType kValueTypeForSeek = kTypeSingleDeletion;
+// Defined in dbformat.cc
+extern const ValueType kValueTypeForSeek;
+extern const ValueType kValueTypeForSeekForPrev;

 // Checks whether a type is an inline value type
 // (i.e. a type used in memtable skiplist and sst file datablock).
@@ -75,6 +71,8 @@ inline bool IsExtendedValueType(ValueType t) {
 static const SequenceNumber kMaxSequenceNumber =
    ((0x1ull << 56) - 1);

+static const SequenceNumber kDisableGlobalSequenceNumber = port::kMaxUint64;
+
 struct ParsedInternalKey {
  Slice user_key;
  SequenceNumber sequence;
@@ -101,6 +99,11 @@ extern void UnPackSequenceAndType(uint64_t packed, uint64_t* seq, ValueType* t);
 // Append the serialization of "key" to *result.
 extern void AppendInternalKey(std::string* result,
                              const ParsedInternalKey& key);
+// Serialized internal key consists of user key followed by footer.
+// This function appends the footer to *result, assuming that *result already
+// contains the user key at the end.
+extern void AppendInternalKeyFooter(std::string* result, SequenceNumber s,
+                                    ValueType t);

 // Attempt to parse an internal key from "internal_key".  On success,
 // stores the parsed data in "*result", and returns true.
@@ -199,6 +202,16 @@ class InternalKey {

  void Clear() { rep_.clear(); }

+  // The underlying representation.
+  // Intended only to be used together with ConvertFromUserKey().
+  std::string* rep() { return &rep_; }
+
+  // Assuming that *rep() contains a user key, this method makes internal key
+  // out of it in-place. This saves a memcpy compared to Set()/SetFrom().
+  void ConvertFromUserKey(SequenceNumber s, ValueType t) {
+    AppendInternalKeyFooter(&rep_, s, t);
+  }
+
  std::string DebugString(bool hex = false) const;
 };

@@ -361,6 +374,15 @@ class IterKey {
    return Slice(key_, key_n);
  }

+  // Copy the key into IterKey own buf_
+  void OwnKey() {
+    assert(IsKeyPinned() == true);
+
+    Reserve(key_size_);
+    memcpy(buf_, key_, key_size_);
+    key_ = buf_;
+  }
+
  // Update the sequence number in the internal key.  Guarantees not to
  // invalidate slices to the key (and the user key).
  void UpdateInternalKey(uint64_t seq, ValueType t) {
@@ -509,25 +531,28 @@ struct RangeTombstone {
  explicit RangeTombstone(Slice sk, Slice ek, SequenceNumber sn)
      : start_key_(sk), end_key_(ek), seq_(sn) {}

-  explicit RangeTombstone(Slice internal_key, Slice value) {
-    ParsedInternalKey parsed_key;
-    if (ParseInternalKey(internal_key, &parsed_key)) {
-      start_key_ = parsed_key.user_key;
-      seq_ = parsed_key.sequence;
-      end_key_ = value;
-    }
+  explicit RangeTombstone(ParsedInternalKey parsed_key, Slice value) {
+    start_key_ = parsed_key.user_key;
+    seq_ = parsed_key.sequence;
+    end_key_ = value;
  }

-  // be careful to use Serialize(); InternalKey() allocates new memory
-  std::pair<InternalKey, Slice> Serialize() {
+  // be careful to use Serialize(), allocates new memory
+  std::pair<InternalKey, Slice> Serialize() const {
    auto key = InternalKey(start_key_, seq_, kTypeRangeDeletion);
    Slice value = end_key_;
    return std::make_pair(std::move(key), std::move(value));
  }

-  InternalKey SerializeKey() {
+  // be careful to use SerializeKey(), allocates new memory
+  InternalKey SerializeKey() const {
    return InternalKey(start_key_, seq_, kTypeRangeDeletion);
  }
+
+  // be careful to use SerializeEndKey(), allocates new memory
+  InternalKey SerializeEndKey() const {
+    return InternalKey(end_key_, seq_, kTypeRangeDeletion);
+  }
 };

 }  // namespace rocksdb
@@ -0,0 +1,518 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#include "db/external_sst_file_ingestion_job.h"
+
+#define __STDC_FORMAT_MACROS
+
+#include <inttypes.h>
+#include <algorithm>
+#include <string>
+#include <vector>
+
+#include "db/version_edit.h"
+#include "table/merger.h"
+#include "table/scoped_arena_iterator.h"
+#include "table/sst_file_writer_collectors.h"
+#include "table/table_builder.h"
+#include "util/file_reader_writer.h"
+#include "util/file_util.h"
+#include "util/stop_watch.h"
+#include "util/sync_point.h"
+
+namespace rocksdb {
+
+Status ExternalSstFileIngestionJob::Prepare(
+    const std::vector<std::string>& external_files_paths) {
+  Status status;
+
+  // Read the information of files we are ingesting
+  for (const std::string& file_path : external_files_paths) {
+    IngestedFileInfo file_to_ingest;
+    status = GetIngestedFileInfo(file_path, &file_to_ingest);
+    if (!status.ok()) {
+      return status;
+    }
+    files_to_ingest_.push_back(file_to_ingest);
+  }
+
+  for (const IngestedFileInfo& f : files_to_ingest_) {
+    if (f.cf_id !=
+            TablePropertiesCollectorFactory::Context::kUnknownColumnFamily &&
+        f.cf_id != cfd_->GetID()) {
+      return Status::InvalidArgument(
+          "External file column family id dont match");
+    }
+  }
+
+  const Comparator* ucmp = cfd_->internal_comparator().user_comparator();
+  auto num_files = files_to_ingest_.size();
+  if (num_files == 0) {
+    return Status::InvalidArgument("The list of files is empty");
+  } else if (num_files > 1) {
+    // Verify that passed files dont have overlapping ranges
+    autovector<const IngestedFileInfo*> sorted_files;
+    for (size_t i = 0; i < num_files; i++) {
+      sorted_files.push_back(&files_to_ingest_[i]);
+    }
+
+    std::sort(
+        sorted_files.begin(), sorted_files.end(),
+        [&ucmp](const IngestedFileInfo* info1, const IngestedFileInfo* info2) {
+          return ucmp->Compare(info1->smallest_user_key,
+                               info2->smallest_user_key) < 0;
+        });
+
+    for (size_t i = 0; i < num_files - 1; i++) {
+      if (ucmp->Compare(sorted_files[i]->largest_user_key,
+                        sorted_files[i + 1]->smallest_user_key) >= 0) {
+        return Status::NotSupported("Files have overlapping ranges");
+      }
+    }
+  }
+
+  for (IngestedFileInfo& f : files_to_ingest_) {
+    if (f.num_entries == 0) {
+      return Status::InvalidArgument("File contain no entries");
+    }
+
+    if (!f.smallest_internal_key().Valid() ||
+        !f.largest_internal_key().Valid()) {
+      return Status::Corruption("Generated table have corrupted keys");
+    }
+  }
+
+  // Copy/Move external files into DB
+  for (IngestedFileInfo& f : files_to_ingest_) {
+    f.fd = FileDescriptor(versions_->NewFileNumber(), 0, f.file_size);
+
+    const std::string path_outside_db = f.external_file_path;
+    const std::string path_inside_db =
+        TableFileName(db_options_.db_paths, f.fd.GetNumber(), f.fd.GetPathId());
+
+    if (ingestion_options_.move_files) {
+      status = env_->LinkFile(path_outside_db, path_inside_db);
+      if (status.IsNotSupported()) {
+        // Original file is on a different FS, use copy instead of hard linking
+        status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+                          db_options_.use_fsync);
+      }
+    } else {
+      status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+                        db_options_.use_fsync);
+    }
+    TEST_SYNC_POINT("DBImpl::AddFile:FileCopied");
+    if (!status.ok()) {
+      break;
+    }
+    f.internal_file_path = path_inside_db;
+  }
+
+  if (!status.ok()) {
+    // We failed, remove all files that we copied into the db
+    for (IngestedFileInfo& f : files_to_ingest_) {
+      if (f.internal_file_path == "") {
+        break;
+      }
+      Status s = env_->DeleteFile(f.internal_file_path);
+      if (!s.ok()) {
+        Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
+            "AddFile() clean up for file %s failed : %s",
+            f.internal_file_path.c_str(), s.ToString().c_str());
+      }
+    }
+  }
+
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::NeedsFlush(bool* flush_needed) {
+  SuperVersion* super_version = cfd_->GetSuperVersion();
+  Status status =
+      IngestedFilesOverlapWithMemtables(super_version, flush_needed);
+
+  if (status.ok() && *flush_needed &&
+      !ingestion_options_.allow_blocking_flush) {
+    status = Status::InvalidArgument("External file requires flush");
+  }
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::Run() {
+  Status status;
+#ifndef NDEBUG
+  // We should never run the job with a memtable that is overlapping
+  // with the files we are ingesting
+  bool need_flush = false;
+  status = NeedsFlush(&need_flush);
+  assert(status.ok() && need_flush == false);
+#endif
+
+  bool consumed_seqno = false;
+  bool force_global_seqno = false;
+  const SequenceNumber last_seqno = versions_->LastSequence();
+  if (ingestion_options_.snapshot_consistency && !db_snapshots_->empty()) {
+    // We need to assign a global sequence number to all the files even
+    // if the dont overlap with any ranges since we have snapshots
+    force_global_seqno = true;
+  }
+
+  SuperVersion* super_version = cfd_->GetSuperVersion();
+  edit_.SetColumnFamily(cfd_->GetID());
+  // The levels that the files will be ingested into
+  for (IngestedFileInfo& f : files_to_ingest_) {
+    bool overlap_with_db = false;
+    status = AssignLevelForIngestedFile(super_version, &f, &overlap_with_db);
+    if (!status.ok()) {
+      return status;
+    }
+
+    if (overlap_with_db || force_global_seqno) {
+      status = AssignGlobalSeqnoForIngestedFile(&f, last_seqno + 1);
+      consumed_seqno = true;
+    } else {
+      status = AssignGlobalSeqnoForIngestedFile(&f, 0);
+    }
+
+    if (!status.ok()) {
+      return status;
+    }
+
+    edit_.AddFile(f.picked_level, f.fd.GetNumber(), f.fd.GetPathId(),
+                  f.fd.GetFileSize(), f.smallest_internal_key(),
+                  f.largest_internal_key(), f.assigned_seqno, f.assigned_seqno,
+                  false);
+  }
+
+  if (consumed_seqno) {
+    versions_->SetLastSequence(last_seqno + 1);
+  }
+
+  return status;
+}
+
+void ExternalSstFileIngestionJob::UpdateStats() {
+  // Update internal stats for new ingested files
+  uint64_t total_keys = 0;
+  uint64_t total_l0_files = 0;
+  uint64_t total_time = env_->NowMicros() - job_start_time_;
+  for (IngestedFileInfo& f : files_to_ingest_) {
+    InternalStats::CompactionStats stats(1);
+    stats.micros = total_time;
+    stats.bytes_written = f.fd.GetFileSize();
+    stats.num_output_files = 1;
+    cfd_->internal_stats()->AddCompactionStats(f.picked_level, stats);
+    cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_INGESTED_ADD_FILE,
+                                       f.fd.GetFileSize());
+    total_keys += f.num_entries;
+    if (f.picked_level == 0) {
+      total_l0_files += 1;
+    }
+    Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+        "[AddFile] External SST file %s was ingested in L%d with path %s "
+        "(global_seqno=%" PRIu64 ")\n",
+        f.external_file_path.c_str(), f.picked_level,
+        f.internal_file_path.c_str(), f.assigned_seqno);
+  }
+  cfd_->internal_stats()->AddCFStats(InternalStats::INGESTED_NUM_KEYS_TOTAL,
+                                     total_keys);
+  cfd_->internal_stats()->AddCFStats(InternalStats::INGESTED_NUM_FILES_TOTAL,
+                                     files_to_ingest_.size());
+  cfd_->internal_stats()->AddCFStats(
+      InternalStats::INGESTED_LEVEL0_NUM_FILES_TOTAL, total_l0_files);
+}
+
+void ExternalSstFileIngestionJob::Cleanup(const Status& status) {
+  if (!status.ok()) {
+    // We failed to add the files to the database
+    // remove all the files we copied
+    for (IngestedFileInfo& f : files_to_ingest_) {
+      Status s = env_->DeleteFile(f.internal_file_path);
+      if (!s.ok()) {
+        Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
+            "AddFile() clean up for file %s failed : %s",
+            f.internal_file_path.c_str(), s.ToString().c_str());
+      }
+    }
+  } else if (status.ok() && ingestion_options_.move_files) {
+    // The files were moved and added successfully, remove original file links
+    for (IngestedFileInfo& f : files_to_ingest_) {
+      Status s = env_->DeleteFile(f.external_file_path);
+      if (!s.ok()) {
+        Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
+            "%s was added to DB successfully but failed to remove original "
+            "file link : %s",
+            f.external_file_path.c_str(), s.ToString().c_str());
+      }
+    }
+  }
+}
+
+Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
+    const std::string& external_file, IngestedFileInfo* file_to_ingest) {
+  file_to_ingest->external_file_path = external_file;
+
+  // Get external file size
+  Status status = env_->GetFileSize(external_file, &file_to_ingest->file_size);
+  if (!status.ok()) {
+    return status;
+  }
+
+  // Create TableReader for external file
+  std::unique_ptr<TableReader> table_reader;
+  std::unique_ptr<RandomAccessFile> sst_file;
+  std::unique_ptr<RandomAccessFileReader> sst_file_reader;
+
+  status = env_->NewRandomAccessFile(external_file, &sst_file, env_options_);
+  if (!status.ok()) {
+    return status;
+  }
+  sst_file_reader.reset(new RandomAccessFileReader(std::move(sst_file)));
+
+  status = cfd_->ioptions()->table_factory->NewTableReader(
+      TableReaderOptions(*cfd_->ioptions(), env_options_,
+                         cfd_->internal_comparator()),
+      std::move(sst_file_reader), file_to_ingest->file_size, &table_reader);
+  if (!status.ok()) {
+    return status;
+  }
+
+  // Get the external file properties
+  auto props = table_reader->GetTableProperties();
+  const auto& uprops = props->user_collected_properties;
+
+  // Get table version
+  auto version_iter = uprops.find(ExternalSstFilePropertyNames::kVersion);
+  if (version_iter == uprops.end()) {
+    return Status::Corruption("External file version not found");
+  }
+  file_to_ingest->version = DecodeFixed32(version_iter->second.c_str());
+
+  auto seqno_iter = uprops.find(ExternalSstFilePropertyNames::kGlobalSeqno);
+  if (file_to_ingest->version == 2) {
+    // version 2 imply that we have global sequence number
+    if (seqno_iter == uprops.end()) {
+      return Status::Corruption(
+          "External file global sequence number not found");
+    }
+
+    // Set the global sequence number
+    file_to_ingest->original_seqno = DecodeFixed64(seqno_iter->second.c_str());
+    file_to_ingest->global_seqno_offset = props->properties_offsets.at(
+        ExternalSstFilePropertyNames::kGlobalSeqno);
+
+    if (file_to_ingest->global_seqno_offset == 0) {
+      return Status::Corruption("Was not able to find file global seqno field");
+    }
+  } else {
+    return Status::InvalidArgument("external file version is not supported");
+  }
+  // Get number of entries in table
+  file_to_ingest->num_entries = props->num_entries;
+
+  ParsedInternalKey key;
+  ReadOptions ro;
+  // During reading the external file we can cache blocks that we read into
+  // the block cache, if we later change the global seqno of this file, we will
+  // have block in cache that will include keys with wrong seqno.
+  // We need to disable fill_cache so that we read from the file without
+  // updating the block cache.
+  ro.fill_cache = false;
+  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(ro));
+
+  // Get first (smallest) key from file
+  iter->SeekToFirst();
+  if (!ParseInternalKey(iter->key(), &key)) {
+    return Status::Corruption("external file have corrupted keys");
+  }
+  if (key.sequence != 0) {
+    return Status::Corruption("external file have non zero sequence number");
+  }
+  file_to_ingest->smallest_user_key = key.user_key.ToString();
+
+  // Get last (largest) key from file
+  iter->SeekToLast();
+  if (!ParseInternalKey(iter->key(), &key)) {
+    return Status::Corruption("external file have corrupted keys");
+  }
+  if (key.sequence != 0) {
+    return Status::Corruption("external file have non zero sequence number");
+  }
+  file_to_ingest->largest_user_key = key.user_key.ToString();
+
+  file_to_ingest->cf_id = static_cast<uint32_t>(props->column_family_id);
+
+  file_to_ingest->table_properties = *props;
+
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::IngestedFilesOverlapWithMemtables(
+    SuperVersion* sv, bool* overlap) {
+  // Create an InternalIterator over all memtables
+  Arena arena;
+  ReadOptions ro;
+  ro.total_order_seek = true;
+  MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(), &arena);
+  merge_iter_builder.AddIterator(sv->mem->NewIterator(ro, &arena));
+  sv->imm->AddIterators(ro, &merge_iter_builder);
+  ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
+
+  Status status;
+  *overlap = false;
+  for (IngestedFileInfo& f : files_to_ingest_) {
+    status =
+        IngestedFileOverlapWithIteratorRange(&f, memtable_iter.get(), overlap);
+    if (!status.ok() || *overlap == true) {
+      break;
+    }
+  }
+
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::AssignLevelForIngestedFile(
+    SuperVersion* sv, IngestedFileInfo* file_to_ingest, bool* overlap_with_db) {
+  *overlap_with_db = false;
+
+  Arena arena;
+  ReadOptions ro;
+  ro.total_order_seek = true;
+
+  Status status;
+  int target_level = 0;
+  auto* vstorage = cfd_->current()->storage_info();
+  for (int lvl = 0; lvl < cfd_->NumberLevels(); lvl++) {
+    if (lvl > 0 && lvl < vstorage->base_level()) {
+      continue;
+    }
+
+    if (vstorage->NumLevelFiles(lvl) > 0) {
+      bool overlap_with_level = false;
+      MergeIteratorBuilder merge_iter_builder(&cfd_->internal_comparator(),
+                                              &arena);
+      RangeDelAggregator range_del_agg(cfd_->internal_comparator(),
+                                       {} /* snapshots */);
+      sv->current->AddIteratorsForLevel(ro, env_options_, &merge_iter_builder,
+                                        lvl, &range_del_agg);
+      if (!range_del_agg.IsEmpty()) {
+        return Status::NotSupported(
+            "file ingestion with range tombstones is currently unsupported");
+      }
+      ScopedArenaIterator level_iter(merge_iter_builder.Finish());
+
+      status = IngestedFileOverlapWithIteratorRange(
+          file_to_ingest, level_iter.get(), &overlap_with_level);
+      if (!status.ok()) {
+        return status;
+      }
+
+      if (overlap_with_level) {
+        // We must use L0 or any level higher than `lvl` to be able to overwrite
+        // the keys that we overlap with in this level, We also need to assign
+        // this file a seqno to overwrite the existing keys in level `lvl`
+        *overlap_with_db = true;
+        break;
+      }
+    }
+
+    // We dont overlap with any keys in this level, but we still need to check
+    // if our file can fit in it
+
+    if (IngestedFileFitInLevel(file_to_ingest, lvl)) {
+      target_level = lvl;
+    }
+  }
+  file_to_ingest->picked_level = target_level;
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
+    IngestedFileInfo* file_to_ingest, SequenceNumber seqno) {
+  if (file_to_ingest->original_seqno == seqno) {
+    // This file already have the correct global seqno
+    return Status::OK();
+  } else if (!ingestion_options_.allow_global_seqno) {
+    return Status::InvalidArgument("Global seqno is required, but disabled");
+  } else if (file_to_ingest->global_seqno_offset == 0) {
+    return Status::InvalidArgument(
+        "Trying to set global seqno for a file that dont have a global seqno "
+        "field");
+  }
+
+  std::unique_ptr<RandomRWFile> rwfile;
+  Status status = env_->NewRandomRWFile(file_to_ingest->internal_file_path,
+                                        &rwfile, env_options_);
+  if (!status.ok()) {
+    return status;
+  }
+
+  // Write the new seqno in the global sequence number field in the file
+  std::string seqno_val;
+  PutFixed64(&seqno_val, seqno);
+  status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
+  if (status.ok()) {
+    file_to_ingest->assigned_seqno = seqno;
+  }
+  return status;
+}
+
+Status ExternalSstFileIngestionJob::IngestedFileOverlapWithIteratorRange(
+    const IngestedFileInfo* file_to_ingest, InternalIterator* iter,
+    bool* overlap) {
+  auto* vstorage = cfd_->current()->storage_info();
+  auto* ucmp = vstorage->InternalComparator()->user_comparator();
+  InternalKey range_start(file_to_ingest->smallest_user_key, kMaxSequenceNumber,
+                          kValueTypeForSeek);
+  iter->Seek(range_start.Encode());
+  if (!iter->status().ok()) {
+    return iter->status();
+  }
+
+  *overlap = false;
+  if (iter->Valid()) {
+    ParsedInternalKey seek_result;
+    if (!ParseInternalKey(iter->key(), &seek_result)) {
+      return Status::Corruption("DB have corrupted keys");
+    }
+
+    if (ucmp->Compare(seek_result.user_key, file_to_ingest->largest_user_key) <=
+        0) {
+      *overlap = true;
+    }
+  }
+
+  return iter->status();
+}
+
+bool ExternalSstFileIngestionJob::IngestedFileFitInLevel(
+    const IngestedFileInfo* file_to_ingest, int level) {
+  if (level == 0) {
+    // Files can always fit in L0
+    return true;
+  }
+
+  auto* vstorage = cfd_->current()->storage_info();
+  Slice file_smallest_user_key(file_to_ingest->smallest_user_key);
+  Slice file_largest_user_key(file_to_ingest->largest_user_key);
+
+  if (vstorage->OverlapInLevel(level, &file_smallest_user_key,
+                               &file_largest_user_key)) {
+    // File overlap with another files in this level, we cannot
+    // add it to this level
+    return false;
+  }
+  if (cfd_->RangeOverlapWithCompaction(file_smallest_user_key,
+                                       file_largest_user_key, level)) {
+    // File overlap with a running compaction output that will be stored
+    // in this level, we cannot add this file to this level
+    return false;
+  }
+
+  // File did not overlap with level files, our compaction output
+  return true;
+}
+
+}  // namespace rocksdb
@@ -0,0 +1,151 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#pragma once
+#include <string>
+#include <unordered_set>
+#include <vector>
+
+#include "db/column_family.h"
+#include "db/dbformat.h"
+#include "db/internal_stats.h"
+#include "db/snapshot_impl.h"
+#include "rocksdb/db.h"
+#include "rocksdb/env.h"
+#include "rocksdb/sst_file_writer.h"
+#include "util/autovector.h"
+#include "util/db_options.h"
+
+namespace rocksdb {
+
+struct IngestedFileInfo {
+  // External file path
+  std::string external_file_path;
+  // Smallest user key in external file
+  std::string smallest_user_key;
+  // Largest user key in external file
+  std::string largest_user_key;
+  // Sequence number for keys in external file
+  SequenceNumber original_seqno;
+  // Offset of the global sequence number field in the file, will
+  // be zero if version is 1 (global seqno is not supported)
+  size_t global_seqno_offset;
+  // External file size
+  uint64_t file_size;
+  // total number of keys in external file
+  uint64_t num_entries;
+  // Id of column family this file shoule be ingested into
+  uint32_t cf_id;
+  // TableProperties read from external file
+  TableProperties table_properties;
+  // Version of external file
+  int version;
+
+  // FileDescriptor for the file inside the DB
+  FileDescriptor fd;
+  // file path that we picked for file inside the DB
+  std::string internal_file_path = "";
+  // Global sequence number that we picked for the file inside the DB
+  SequenceNumber assigned_seqno = 0;
+  // Level inside the DB we picked for the external file.
+  int picked_level = 0;
+
+  InternalKey smallest_internal_key() const {
+    return InternalKey(smallest_user_key, assigned_seqno,
+                       ValueType::kTypeValue);
+  }
+
+  InternalKey largest_internal_key() const {
+    return InternalKey(largest_user_key, assigned_seqno, ValueType::kTypeValue);
+  }
+};
+
+class ExternalSstFileIngestionJob {
+ public:
+  ExternalSstFileIngestionJob(
+      Env* env, VersionSet* versions, ColumnFamilyData* cfd,
+      const ImmutableDBOptions& db_options, const EnvOptions& env_options,
+      SnapshotList* db_snapshots,
+      const IngestExternalFileOptions& ingestion_options)
+      : env_(env),
+        versions_(versions),
+        cfd_(cfd),
+        db_options_(db_options),
+        env_options_(env_options),
+        db_snapshots_(db_snapshots),
+        ingestion_options_(ingestion_options),
+        job_start_time_(env_->NowMicros()) {}
+
+  // Prepare the job by copying external files into the DB.
+  Status Prepare(const std::vector<std::string>& external_files_paths);
+
+  // Check if we need to flush the memtable before running the ingestion job
+  // This will be true if the files we are ingesting are overlapping with any
+  // key range in the memtable.
+  // REQUIRES: Mutex held
+  Status NeedsFlush(bool* flush_needed);
+
+  // Will execute the ingestion job and prepare edit() to be applied.
+  // REQUIRES: Mutex held
+  Status Run();
+
+  // Update column family stats.
+  // REQUIRES: Mutex held
+  void UpdateStats();
+
+  // Cleanup after successfull/failed job
+  void Cleanup(const Status& status);
+
+  VersionEdit* edit() { return &edit_; }
+
+  const autovector<IngestedFileInfo>& files_to_ingest() const {
+    return files_to_ingest_;
+  }
+
+ private:
+  // Open the external file and populate `file_to_ingest` with all the
+  // external information we need to ingest this file.
+  Status GetIngestedFileInfo(const std::string& external_file,
+                             IngestedFileInfo* file_to_ingest);
+
+  // Check if the files we are ingesting overlap with any memtable.
+  // REQUIRES: Mutex held
+  Status IngestedFilesOverlapWithMemtables(SuperVersion* sv, bool* overlap);
+
+  // Assign `file_to_ingest` the lowest possible level that it can
+  // be ingested to.
+  // REQUIRES: Mutex held
+  Status AssignLevelForIngestedFile(SuperVersion* sv,
+                                    IngestedFileInfo* file_to_ingest,
+                                    bool* overlap_with_db);
+
+  // Set the file global sequence number to `seqno`
+  Status AssignGlobalSeqnoForIngestedFile(IngestedFileInfo* file_to_ingest,
+                                          SequenceNumber seqno);
+
+  // Check if `file_to_ingest` key range overlap with the range `iter` represent
+  // REQUIRES: Mutex held
+  Status IngestedFileOverlapWithIteratorRange(
+      const IngestedFileInfo* file_to_ingest, InternalIterator* iter,
+      bool* overlap);
+
+  // Check if `file_to_ingest` can fit in level `level`
+  // REQUIRES: Mutex held
+  bool IngestedFileFitInLevel(const IngestedFileInfo* file_to_ingest,
+                              int level);
+
+  Env* env_;
+  VersionSet* versions_;
+  ColumnFamilyData* cfd_;
+  const ImmutableDBOptions& db_options_;
+  const EnvOptions& env_options_;
+  SnapshotList* db_snapshots_;
+  autovector<IngestedFileInfo> files_to_ingest_;
+  const IngestExternalFileOptions& ingestion_options_;
+  VersionEdit edit_;
+  uint64_t job_start_time_;
+};
+
+}  // namespace rocksdb
@@ -228,6 +228,11 @@ class FaultInjectionTest : public testing::Test,
    return Status::OK();
  }

+#if __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 9)
+__attribute__((__no_sanitize__("undefined")))
+#elif __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9)
+__attribute__((__no_sanitize_undefined__))
+#endif
  // Return the ith key
  Slice Key(int i, std::string* storage) const {
    int num = i;
@@ -503,6 +508,30 @@ TEST_P(FaultInjectionTest, ManualLogSyncTest) {
  ASSERT_EQ(value_space, val);
 }

+TEST_P(FaultInjectionTest, WriteBatchWalTerminationTest) {
+  ReadOptions ro;
+  Options options = CurrentOptions();
+  options.env = env_;
+
+  WriteOptions wo;
+  wo.sync = true;
+  wo.disableWAL = false;
+  WriteBatch batch;
+  batch.Put("cats", "dogs");
+  batch.MarkWalTerminationPoint();
+  batch.Put("boys", "girls");
+  ASSERT_OK(db_->Write(wo, &batch));
+
+  env_->SetFilesystemActive(false);
+  NoWriteTestReopenWithFault(kResetDropAndDeleteUnsynced);
+  ASSERT_OK(OpenDB());
+
+  std::string val;
+  ASSERT_OK(db_->Get(ro, "cats", &val));
+  ASSERT_EQ("dogs", val);
+  ASSERT_EQ(db_->Get(ro, "boys", &val), Status::NotFound());
+}
+
 INSTANTIATE_TEST_CASE_P(FaultTest, FaultInjectionTest, ::testing::Bool());

 }  // namespace rocksdb
@@ -386,10 +386,10 @@ Status SetIdentityFile(Env* env, const std::string& dbname) {
  return s;
 }

-Status SyncManifest(Env* env, const DBOptions* db_options,
+Status SyncManifest(Env* env, const ImmutableDBOptions* db_options,
                    WritableFileWriter* file) {
  TEST_KILL_RANDOM("SyncManifest:0", rocksdb_kill_odds * REDUCE_ODDS2);
-  if (db_options->disableDataSync) {
+  if (db_options->disable_data_sync) {
    return Status::OK();
  } else {
    StopWatch sw(env, db_options->statistics.get(), MANIFEST_FILE_SYNC_MICROS);
@@ -20,6 +20,7 @@
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
 #include "rocksdb/transaction_log.h"
+#include "util/db_options.h"

 namespace rocksdb {

@@ -159,7 +160,7 @@ extern Status SetCurrentFile(Env* env, const std::string& dbname,
 extern Status SetIdentityFile(Env* env, const std::string& dbname);

 // Sync manifest file `file`.
-extern Status SyncManifest(Env* env, const DBOptions* db_options,
+extern Status SyncManifest(Env* env, const ImmutableDBOptions* db_options,
                           WritableFileWriter* file);

 }  // namespace rocksdb
@@ -56,7 +56,7 @@
 namespace rocksdb {

 FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
-                   const DBOptions& db_options,
+                   const ImmutableDBOptions& db_options,
                   const MutableCFOptions& mutable_cf_options,
                   const EnvOptions& env_options, VersionSet* versions,
                   InstrumentedMutex* db_mutex,
@@ -97,7 +97,7 @@ FlushJob::~FlushJob() {

 void FlushJob::ReportStartedFlush() {
  ThreadStatusUtil::SetColumnFamily(cfd_, cfd_->ioptions()->env,
-                                    cfd_->options()->enable_thread_tracking);
+                                    db_options_.enable_thread_tracking);
  ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_FLUSH);
  ThreadStatusUtil::SetThreadOperationProperty(
      ThreadStatus::COMPACTION_JOB_ID,
@@ -241,7 +241,11 @@ Status FlushJob::WriteLevel0Table() {
    if (log_buffer_) {
      log_buffer_->FlushBufferToLog();
    }
+    // memtables and range_del_iters store internal iterators over each data
+    // memtable and its associated range deletion memtable, respectively, at
+    // corresponding indexes.
    std::vector<InternalIterator*> memtables;
+    std::vector<InternalIterator*> range_del_iters;
    ReadOptions ro;
    ro.total_order_seek = true;
    Arena arena;
@@ -252,6 +256,10 @@ Status FlushJob::WriteLevel0Table() {
          "[%s] [JOB %d] Flushing memtable with next log file: %" PRIu64 "\n",
          cfd_->GetName().c_str(), job_context_->job_id, m->GetNextLogNumber());
      memtables.push_back(m->NewIterator(ro, &arena));
+      auto* range_del_iter = m->NewRangeTombstoneIterator(ro);
+      if (range_del_iter != nullptr) {
+        range_del_iters.push_back(range_del_iter);
+      }
      total_num_entries += m->num_entries();
      total_num_deletes += m->num_deletes();
      total_memory_usage += m->ApproximateMemoryUsage();
@@ -268,6 +276,10 @@ 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())));
      Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
          "[%s] [JOB %d] Level-0 flush table #%" PRIu64 ": started",
          cfd_->GetName().c_str(), job_context_->job_id, meta_.fd.GetNumber());
@@ -276,9 +288,10 @@ Status FlushJob::WriteLevel0Table() {
                               &output_compression_);
      s = BuildTable(
          dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
-          env_options_, cfd_->table_cache(), iter.get(), &meta_,
-          cfd_->internal_comparator(), cfd_->int_tbl_prop_collector_factories(),
-          cfd_->GetID(), cfd_->GetName(), existing_snapshots_,
+          env_options_, cfd_->table_cache(), iter.get(),
+          std::move(range_del_iter), &meta_, cfd_->internal_comparator(),
+          cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
+          cfd_->GetName(), existing_snapshots_,
          earliest_write_conflict_snapshot_, output_compression_,
          cfd_->ioptions()->compression_opts,
          mutable_cf_options_.paranoid_file_checks, cfd_->internal_stats(),
@@ -294,7 +307,7 @@ Status FlushJob::WriteLevel0Table() {
        meta_.fd.GetFileSize(), s.ToString().c_str(),
        meta_.marked_for_compaction ? " (needs compaction)" : "");

-    if (!db_options_.disableDataSync && output_file_directory_ != nullptr) {
+    if (!db_options_.disable_data_sync && output_file_directory_ != nullptr) {
      output_file_directory_->Fsync();
    }
    TEST_SYNC_POINT("FlushJob::WriteLevel0Table");
@@ -34,6 +34,7 @@
 #include "rocksdb/transaction_log.h"
 #include "table/scoped_arena_iterator.h"
 #include "util/autovector.h"
+#include "util/db_options.h"
 #include "util/event_logger.h"
 #include "util/instrumented_mutex.h"
 #include "util/stop_watch.h"
@@ -53,7 +54,7 @@ class FlushJob {
  // TODO(icanadi) make effort to reduce number of parameters here
  // IMPORTANT: mutable_cf_options needs to be alive while FlushJob is alive
  FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
-           const DBOptions& db_options,
+           const ImmutableDBOptions& db_options,
           const MutableCFOptions& mutable_cf_options,
           const EnvOptions& env_options, VersionSet* versions,
           InstrumentedMutex* db_mutex, std::atomic<bool>* shutting_down,
@@ -78,7 +79,7 @@ class FlushJob {
  Status WriteLevel0Table();
  const std::string& dbname_;
  ColumnFamilyData* cfd_;
-  const DBOptions& db_options_;
+  const ImmutableDBOptions& db_options_;
  const MutableCFOptions& mutable_cf_options_;
  const EnvOptions& env_options_;
  VersionSet* versions_;
@@ -28,6 +28,8 @@ class FlushJobTest : public testing::Test {
  FlushJobTest()
      : env_(Env::Default()),
        dbname_(test::TmpDir() + "/flush_job_test"),
+        options_(),
+        db_options_(options_),
        table_cache_(NewLRUCache(50000, 16)),
        write_buffer_manager_(db_options_.db_write_buffer_size),
        versions_(new VersionSet(dbname_, &db_options_, env_options_,
@@ -74,9 +76,10 @@ class FlushJobTest : public testing::Test {
  Env* env_;
  std::string dbname_;
  EnvOptions env_options_;
+  Options options_;
+  ImmutableDBOptions db_options_;
  std::shared_ptr<Cache> table_cache_;
  WriteController write_controller_;
-  DBOptions db_options_;
  WriteBufferManager write_buffer_manager_;
  ColumnFamilyOptions cf_options_;
  std::unique_ptr<VersionSet> versions_;
@@ -112,18 +115,19 @@ TEST_F(FlushJobTest, NonEmpty) {
  // Test data:
  //   seqno [    1,    2 ... 8998, 8999, 9000, 9001, 9002 ... 9999 ]
  //   key   [ 1001, 1002 ... 9998, 9999,    0,    1,    2 ...  999 ]
-  // Expected:
-  //   smallest_key   = "0"
-  //   largest_key    = "9999"
-  //   smallest_seqno = 1
-  //   smallest_seqno = 9999
+  //   range-delete "9995" -> "9999" at seqno 10000
  for (int i = 1; i < 10000; ++i) {
    std::string key(ToString((i + 1000) % 10000));
    std::string value("value" + key);
    new_mem->Add(SequenceNumber(i), kTypeValue, key, value);
-    InternalKey internal_key(key, SequenceNumber(i), kTypeValue);
-    inserted_keys.insert({internal_key.Encode().ToString(), value});
+    if ((i + 1000) % 10000 < 9995) {
+      InternalKey internal_key(key, SequenceNumber(i), kTypeValue);
+      inserted_keys.insert({internal_key.Encode().ToString(), value});
+    }
  }
+  new_mem->Add(SequenceNumber(10000), kTypeRangeDeletion, "9995", "9999a");
+  InternalKey internal_key("9995", SequenceNumber(10000), kTypeRangeDeletion);
+  inserted_keys.insert({internal_key.Encode().ToString(), "9999a"});

  autovector<MemTable*> to_delete;
  cfd->imm()->Add(new_mem, &to_delete);
@@ -143,9 +147,10 @@ TEST_F(FlushJobTest, NonEmpty) {
  ASSERT_OK(flush_job.Run(&fd));
  mutex_.Unlock();
  ASSERT_EQ(ToString(0), fd.smallest.user_key().ToString());
-  ASSERT_EQ(ToString(9999), fd.largest.user_key().ToString());
+  ASSERT_EQ("9999a",
+            fd.largest.user_key().ToString());  // range tombstone end key
  ASSERT_EQ(1, fd.smallest_seqno);
-  ASSERT_EQ(9999, fd.largest_seqno);
+  ASSERT_EQ(10000, fd.largest_seqno);  // range tombstone seqnum 10000
  mock_table_factory_->AssertSingleFile(inserted_keys);
  job_context.Clean();
 }
@@ -71,8 +71,8 @@ class LevelIterator : public InternalIterator {

    file_iter_ = cfd_->table_cache()->NewIterator(
        read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
-        files_[file_index_]->fd, nullptr /* table_reader_ptr */, nullptr,
-        false);
+        files_[file_index_]->fd, nullptr /* range_del_agg */,
+        nullptr /* table_reader_ptr */, nullptr, false);

    file_iter_->SetPinnedItersMgr(pinned_iters_mgr_);
  }
@@ -97,6 +97,10 @@ class LevelIterator : public InternalIterator {
    file_iter_->Seek(internal_key);
    valid_ = file_iter_->Valid();
  }
+  void SeekForPrev(const Slice& internal_key) override {
+    status_ = Status::NotSupported("LevelIterator::SeekForPrev()");
+    valid_ = false;
+  }
  void Next() override {
    assert(valid_);
    file_iter_->Next();
@@ -292,9 +296,15 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,
  // an option to turn it off.
  if (seek_to_first || NeedToSeekImmutable(internal_key)) {
    immutable_status_ = Status::OK();
-    if ((has_iter_trimmed_for_upper_bound_) &&
-        (cfd_->internal_comparator().InternalKeyComparator::Compare(
-             prev_key_.GetKey(), internal_key) > 0)) {
+    if (has_iter_trimmed_for_upper_bound_ &&
+        (
+            // prev_ is not set yet
+            is_prev_set_ == false ||
+            // We are doing SeekToFirst() and internal_key.size() = 0
+            seek_to_first ||
+            // prev_key_ > internal_key
+            cfd_->internal_comparator().InternalKeyComparator::Compare(
+                prev_key_.GetKey(), internal_key) > 0)) {
      // Some iterators are trimmed. Need to rebuild.
      RebuildIterators(true);
      // Already seeked mutable iter, so seek again
@@ -355,61 +365,19 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,
      }
    }

-    int32_t search_left_bound = 0;
-    int32_t search_right_bound = FileIndexer::kLevelMaxIndex;
    for (int32_t level = 1; level < vstorage->num_levels(); ++level) {
      const std::vector<FileMetaData*>& level_files =
          vstorage->LevelFiles(level);
      if (level_files.empty()) {
-        search_left_bound = 0;
-        search_right_bound = FileIndexer::kLevelMaxIndex;
        continue;
      }
      if (level_iters_[level - 1] == nullptr) {
        continue;
      }
      uint32_t f_idx = 0;
-      const auto& indexer = vstorage->file_indexer();
      if (!seek_to_first) {
-        if (search_left_bound == search_right_bound) {
-          f_idx = search_left_bound;
-        } else if (search_left_bound < search_right_bound) {
-          f_idx =
-              FindFileInRange(level_files, internal_key, search_left_bound,
-                              search_right_bound == FileIndexer::kLevelMaxIndex
-                                  ? static_cast<uint32_t>(level_files.size())
-                                  : search_right_bound);
-        } else {
-          // search_left_bound > search_right_bound
-          // There are only 2 cases this can happen:
-          // (1) target key is smaller than left most file
-          // (2) target key is larger than right most file
-          assert(search_left_bound == (int32_t)level_files.size() ||
-                 search_right_bound == -1);
-          if (search_right_bound == -1) {
-            assert(search_left_bound == 0);
-            f_idx = 0;
-          } else {
-            indexer.GetNextLevelIndex(
-                level, level_files.size() - 1,
-                1, 1, &search_left_bound, &search_right_bound);
-            continue;
-          }
-        }
-
-        // Prepare hints for the next level
-        if (f_idx < level_files.size()) {
-          int cmp_smallest = user_comparator_->Compare(
-              user_key, level_files[f_idx]->smallest.user_key());
-          assert(user_comparator_->Compare(
-                     user_key, level_files[f_idx]->largest.user_key()) <= 0);
-          indexer.GetNextLevelIndex(level, f_idx, cmp_smallest, -1,
-                                    &search_left_bound, &search_right_bound);
-        } else {
-          indexer.GetNextLevelIndex(
-              level, level_files.size() - 1,
-              1, 1, &search_left_bound, &search_right_bound);
-        }
+        f_idx = FindFileInRange(level_files, internal_key, 0,
+                                static_cast<uint32_t>(level_files.size()));
      }

      // Seek
@@ -606,7 +574,8 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
      continue;
    }
    l0_iters_.push_back(cfd_->table_cache()->NewIterator(
-        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), l0->fd));
+        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), l0->fd,
+        nullptr /* range_del_agg */));
  }
  BuildLevelIterators(vstorage);
  current_ = nullptr;
@@ -661,7 +630,7 @@ void ForwardIterator::RenewIterators() {
    }
    l0_iters_new.push_back(cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files_new[inew]->fd));
+        l0_files_new[inew]->fd, nullptr /* range_del_agg */));
  }

  for (auto* f : l0_iters_) {
@@ -713,7 +682,7 @@ void ForwardIterator::ResetIncompleteIterators() {
    DeleteIterator(l0_iters_[i]);
    l0_iters_[i] = cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files[i]->fd);
+        l0_files[i]->fd, nullptr /* range_del_agg */);
    l0_iters_[i]->SetPinnedItersMgr(pinned_iters_mgr_);
  }

@@ -55,6 +55,10 @@ class ForwardIterator : public InternalIterator {
                  ColumnFamilyData* cfd, SuperVersion* current_sv = nullptr);
  virtual ~ForwardIterator();

+  void SeekForPrev(const Slice& target) override {
+    status_ = Status::NotSupported("ForwardIterator::SeekForPrev()");
+    valid_ = false;
+  }
  void SeekToLast() override {
    status_ = Status::NotSupported("ForwardIterator::SeekToLast()");
    valid_ = false;
@@ -44,6 +44,7 @@
 #pragma once
 #include <assert.h>
 #include <stdlib.h>
+#include <algorithm>
 #include <atomic>
 #include "port/port.h"
 #include "util/allocator.h"
@@ -55,8 +56,11 @@ template <class Comparator>
 class InlineSkipList {
 private:
  struct Node;
+  struct Splice;

 public:
+  static const uint16_t kMaxPossibleHeight = 32;
+
  // Create a new InlineSkipList object that will use "cmp" for comparing
  // keys, and will allocate memory using "*allocator".  Objects allocated
  // in the allocator must remain allocated for the lifetime of the
@@ -70,22 +74,58 @@ class InlineSkipList {
  // is thread-safe.
  char* AllocateKey(size_t key_size);

+  // Allocate a splice using allocator.
+  Splice* AllocateSplice();
+
  // Inserts a key allocated by AllocateKey, after the actual key value
  // has been filled in.
  //
  // REQUIRES: nothing that compares equal to key is currently in the list.
-  // REQUIRES: no concurrent calls to INSERT
+  // REQUIRES: no concurrent calls to any of inserts.
  void Insert(const char* key);

+  // Inserts a key allocated by AllocateKey with a hint of last insert
+  // position in the skip-list. If hint points to nullptr, a new hint will be
+  // populated, which can be used in subsequent calls.
+  //
+  // It can be used to optimize the workload where there are multiple groups
+  // of keys, and each key is likely to insert to a location close to the last
+  // inserted key in the same group. One example is sequential inserts.
+  //
+  // REQUIRES: nothing that compares equal to key is currently in the list.
+  // REQUIRES: no concurrent calls to any of inserts.
+  void InsertWithHint(const char* key, void** hint);
+
  // Like Insert, but external synchronization is not required.
  void InsertConcurrently(const char* key);

+  // Inserts a node into the skip list.  key must have been allocated by
+  // AllocateKey and then filled in by the caller.  If UseCAS is true,
+  // then external synchronization is not required, otherwise this method
+  // may not be called concurrently with any other insertions.
+  //
+  // Regardless of whether UseCAS is true, the splice must be owned
+  // exclusively by the current thread.  If allow_partial_splice_fix is
+  // true, then the cost of insertion is amortized O(log D), where D is
+  // the distance from the splice to the inserted key (measured as the
+  // number of intervening nodes).  Note that this bound is very good for
+  // sequential insertions!  If allow_partial_splice_fix is false then
+  // the existing splice will be ignored unless the current key is being
+  // inserted immediately after the splice.  allow_partial_splice_fix ==
+  // false has worse running time for the non-sequential case O(log N),
+  // but a better constant factor.
+  template <bool UseCAS>
+  void Insert(const char* key, Splice* splice, bool allow_partial_splice_fix);
+
  // Returns true iff an entry that compares equal to key is in the list.
  bool Contains(const char* key) const;

  // Return estimated number of entries smaller than `key`.
  uint64_t EstimateCount(const char* key) const;

+  // Validate correctness of the skip-list.
+  void TEST_Validate() const;
+
  // Iteration over the contents of a skip list
  class Iterator {
   public:
@@ -116,6 +156,9 @@ class InlineSkipList {
    // Advance to the first entry with a key >= target
    void Seek(const char* target);

+    // Retreat to the last entry with a key <= target
+    void SeekForPrev(const char* target);
+
    // Position at the first entry in list.
    // Final state of iterator is Valid() iff list is not empty.
    void SeekToFirst();
@@ -131,8 +174,6 @@ class InlineSkipList {
  };

 private:
-  enum MaxPossibleHeightEnum : uint16_t { kMaxPossibleHeight = 32 };
-
  const uint16_t kMaxHeight_;
  const uint16_t kBranching_;
  const uint32_t kScaledInverseBranching_;
@@ -147,13 +188,10 @@ class InlineSkipList {
  // values are ok.
  std::atomic<int> max_height_;  // Height of the entire list

-  // Used for optimizing sequential insert patterns.  Tricky.  prev_height_
-  // of zero means prev_ is undefined.  Otherwise: prev_[i] for i up
-  // to max_height_ - 1 (inclusive) is the predecessor of prev_[0], and
-  // prev_height_ is the height of prev_[0].  prev_[0] can only be equal
-  // to head when max_height_ and prev_height_ are both 1.
-  Node** prev_;
-  std::atomic<int32_t> prev_height_;
+  // seq_splice_ is a Splice used for insertions in the non-concurrent
+  // case.  It caches the prev and next found during the most recent
+  // non-concurrent insertion.
+  Splice* seq_splice_;

  inline int GetMaxHeight() const {
    return max_height_.load(std::memory_order_relaxed);
@@ -167,8 +205,12 @@ class InlineSkipList {
    return (compare_(a, b) == 0);
  }

+  bool LessThan(const char* a, const char* b) const {
+    return (compare_(a, b) < 0);
+  }
+
  // Return true if key is greater than the data stored in "n".  Null n
-  // is considered infinite.
+  // is considered infinite.  n should not be head_.
  bool KeyIsAfterNode(const char* key, Node* n) const;

  // Returns the earliest node with a key >= key.
@@ -181,6 +223,13 @@ class InlineSkipList {
  // level in [0..max_height_-1], if prev is non-null.
  Node* FindLessThan(const char* key, Node** prev = nullptr) const;

+  // Return the latest node with a key < key on bottom_level. Start searching
+  // from root node on the level below top_level.
+  // Fills prev[level] with pointer to previous node at "level" for every
+  // level in [bottom_level..top_level-1], if prev is non-null.
+  Node* FindLessThan(const char* key, Node** prev, Node* root, int top_level,
+                     int bottom_level) const;
+
  // Return the last node in the list.
  // Return head_ if list is empty.
  Node* FindLast() const;
@@ -191,8 +240,13 @@ class InlineSkipList {
  // point to a node that is before the key, and after should point to
  // a node that is after the key.  after should be nullptr if a good after
  // node isn't conveniently available.
-  void FindLevelSplice(const char* key, Node* before, Node* after, int level,
-                       Node** out_prev, Node** out_next);
+  void FindSpliceForLevel(const char* key, Node* before, Node* after, int level,
+                          Node** out_prev, Node** out_next);
+
+  // Recomputes Splice levels from highest_level (inclusive) down to
+  // lowest_level (inclusive).
+  void RecomputeSpliceLevels(const char* key, Splice* splice,
+                             int recompute_level);

  // No copying allowed
  InlineSkipList(const InlineSkipList&);
@@ -201,6 +255,19 @@ class InlineSkipList {

 // Implementation details follow

+template <class Comparator>
+struct InlineSkipList<Comparator>::Splice {
+  // The invariant of a Splice is that prev_[i+1].key <= prev_[i].key <
+  // next_[i].key <= next_[i+1].key for all i.  That means that if a
+  // key is bracketed by prev_[i] and next_[i] then it is bracketed by
+  // all higher levels.  It is _not_ required that prev_[i]->Next(i) ==
+  // next_[i] (it probably did at some point in the past, but intervening
+  // or concurrent operations might have inserted nodes in between).
+  int height_ = 0;
+  Node** prev_;
+  Node** next_;
+};
+
 // The Node data type is more of a pointer into custom-managed memory than
 // a traditional C++ struct.  The key is stored in the bytes immediately
 // after the struct, and the next_ pointers for nodes with height > 1 are
@@ -258,6 +325,14 @@ struct InlineSkipList<Comparator>::Node {
    next_[-n].store(x, std::memory_order_relaxed);
  }

+  // Insert node after prev on specific level.
+  void InsertAfter(Node* prev, int level) {
+    // NoBarrier_SetNext() suffices since we will add a barrier when
+    // we publish a pointer to "this" in prev.
+    NoBarrier_SetNext(level, prev->NoBarrier_Next(level));
+    prev->SetNext(level, this);
+  }
+
 private:
  // next_[0] is the lowest level link (level 0).  Higher levels are
  // stored _earlier_, so level 1 is at next_[-1].
@@ -310,6 +385,18 @@ inline void InlineSkipList<Comparator>::Iterator::Seek(const char* target) {
  node_ = list_->FindGreaterOrEqual(target);
 }

+template <class Comparator>
+inline void InlineSkipList<Comparator>::Iterator::SeekForPrev(
+    const char* target) {
+  Seek(target);
+  if (!Valid()) {
+    SeekToLast();
+  }
+  while (Valid() && list_->LessThan(target, key())) {
+    Prev();
+  }
+}
+
 template <class Comparator>
 inline void InlineSkipList<Comparator>::Iterator::SeekToFirst() {
  node_ = list_->head_->Next(0);
@@ -343,6 +430,7 @@ template <class Comparator>
 bool InlineSkipList<Comparator>::KeyIsAfterNode(const char* key,
                                                Node* n) const {
  // nullptr n is considered infinite
+  assert(n != head_);
  return (n != nullptr) && (compare_(n->Key(), key) < 0);
 }

@@ -382,8 +470,17 @@ InlineSkipList<Comparator>::FindGreaterOrEqual(const char* key) const {
 template <class Comparator>
 typename InlineSkipList<Comparator>::Node*
 InlineSkipList<Comparator>::FindLessThan(const char* key, Node** prev) const {
-  Node* x = head_;
-  int level = GetMaxHeight() - 1;
+  return FindLessThan(key, prev, head_, GetMaxHeight(), 0);
+}
+
+template <class Comparator>
+typename InlineSkipList<Comparator>::Node*
+InlineSkipList<Comparator>::FindLessThan(const char* key, Node** prev,
+                                         Node* root, int top_level,
+                                         int bottom_level) const {
+  assert(top_level > bottom_level);
+  int level = top_level - 1;
+  Node* x = root;
  // KeyIsAfter(key, last_not_after) is definitely false
  Node* last_not_after = nullptr;
  while (true) {
@@ -397,10 +494,10 @@ InlineSkipList<Comparator>::FindLessThan(const char* key, Node** prev) const {
      if (prev != nullptr) {
        prev[level] = x;
      }
-      if (level == 0) {
+      if (level == bottom_level) {
        return x;
      } else {
-        // Switch to next list, reuse KeyIUsAfterNode() result
+        // Switch to next list, reuse KeyIsAfterNode() result
        last_not_after = next;
        level--;
      }
@@ -464,19 +561,14 @@ InlineSkipList<Comparator>::InlineSkipList(const Comparator cmp,
      allocator_(allocator),
      head_(AllocateNode(0, max_height)),
      max_height_(1),
-      prev_height_(1) {
+      seq_splice_(AllocateSplice()) {
  assert(max_height > 0 && kMaxHeight_ == static_cast<uint32_t>(max_height));
  assert(branching_factor > 1 &&
         kBranching_ == static_cast<uint32_t>(branching_factor));
  assert(kScaledInverseBranching_ > 0);
-  // Allocate the prev_ Node* array, directly from the passed-in allocator.
-  // prev_ does not need to be freed, as its life cycle is tied up with
-  // the allocator as a whole.
-  prev_ = reinterpret_cast<Node**>(
-      allocator_->AllocateAligned(sizeof(Node*) * kMaxHeight_));
-  for (int i = 0; i < kMaxHeight_; i++) {
+
+  for (int i = 0; i < kMaxHeight_; ++i) {
    head_->SetNext(i, nullptr);
-    prev_[i] = head_;
  }
 }

@@ -510,70 +602,49 @@ InlineSkipList<Comparator>::AllocateNode(size_t key_size, int height) {
 }

 template <class Comparator>
-void InlineSkipList<Comparator>::Insert(const char* key) {
-  // InsertConcurrently often can't maintain the prev_ invariants, so
-  // it just sets prev_height_ to zero, letting us know that we should
-  // ignore it.  A relaxed load suffices here because write thread
-  // synchronization separates Insert calls from InsertConcurrently calls.
-  auto prev_height = prev_height_.load(std::memory_order_relaxed);
-
-  // fast path for sequential insertion
-  if (prev_height > 0 && !KeyIsAfterNode(key, prev_[0]->NoBarrier_Next(0)) &&
-      (prev_[0] == head_ || KeyIsAfterNode(key, prev_[0]))) {
-    assert(prev_[0] != head_ || (prev_height == 1 && GetMaxHeight() == 1));
-
-    // Outside of this method prev_[1..max_height_] is the predecessor
-    // of prev_[0], and prev_height_ refers to prev_[0].  Inside Insert
-    // prev_[0..max_height - 1] is the predecessor of key.  Switch from
-    // the external state to the internal
-    for (int i = 1; i < prev_height; i++) {
-      prev_[i] = prev_[0];
-    }
-  } else {
-    // TODO(opt): we could use a NoBarrier predecessor search as an
-    // optimization for architectures where memory_order_acquire needs
-    // a synchronization instruction.  Doesn't matter on x86
-    FindLessThan(key, prev_);
-  }
-
-  // Our data structure does not allow duplicate insertion
-  assert(prev_[0]->Next(0) == nullptr || !Equal(key, prev_[0]->Next(0)->Key()));
-
-  // Find the Node that we placed before the key in AllocateKey
-  Node* x = reinterpret_cast<Node*>(const_cast<char*>(key)) - 1;
-  int height = x->UnstashHeight();
-  assert(height >= 1 && height <= kMaxHeight_);
-
-  if (height > GetMaxHeight()) {
-    for (int i = GetMaxHeight(); i < height; i++) {
-      prev_[i] = head_;
-    }
-
-    // It is ok to mutate max_height_ without any synchronization
-    // with concurrent readers.  A concurrent reader that observes
-    // the new value of max_height_ will see either the old value of
-    // new level pointers from head_ (nullptr), or a new value set in
-    // the loop below.  In the former case the reader will
-    // immediately drop to the next level since nullptr sorts after all
-    // keys.  In the latter case the reader will use the new node.
-    max_height_.store(height, std::memory_order_relaxed);
-  }
-
-  for (int i = 0; i < height; i++) {
-    // NoBarrier_SetNext() suffices since we will add a barrier when
-    // we publish a pointer to "x" in prev[i].
-    x->NoBarrier_SetNext(i, prev_[i]->NoBarrier_Next(i));
-    prev_[i]->SetNext(i, x);
-  }
-  prev_[0] = x;
-  prev_height_.store(height, std::memory_order_relaxed);
+typename InlineSkipList<Comparator>::Splice*
+InlineSkipList<Comparator>::AllocateSplice() {
+  // size of prev_ and next_
+  size_t array_size = sizeof(Node*) * (kMaxHeight_ + 1);
+  char* raw = allocator_->AllocateAligned(sizeof(Splice) + array_size * 2);
+  Splice* splice = reinterpret_cast<Splice*>(raw);
+  splice->height_ = 0;
+  splice->prev_ = reinterpret_cast<Node**>(raw + sizeof(Splice));
+  splice->next_ = reinterpret_cast<Node**>(raw + sizeof(Splice) + array_size);
+  return splice;
 }

 template <class Comparator>
-void InlineSkipList<Comparator>::FindLevelSplice(const char* key, Node* before,
-                                                 Node* after, int level,
-                                                 Node** out_prev,
-                                                 Node** out_next) {
+void InlineSkipList<Comparator>::Insert(const char* key) {
+  Insert<false>(key, seq_splice_, false);
+}
+
+template <class Comparator>
+void InlineSkipList<Comparator>::InsertConcurrently(const char* key) {
+  Node* prev[kMaxPossibleHeight];
+  Node* next[kMaxPossibleHeight];
+  Splice splice;
+  splice.prev_ = prev;
+  splice.next_ = next;
+  Insert<true>(key, &splice, false);
+}
+
+template <class Comparator>
+void InlineSkipList<Comparator>::InsertWithHint(const char* key, void** hint) {
+  assert(hint != nullptr);
+  Splice* splice = reinterpret_cast<Splice*>(*hint);
+  if (splice == nullptr) {
+    splice = AllocateSplice();
+    *hint = reinterpret_cast<void*>(splice);
+  }
+  Insert<false>(key, splice, true);
+}
+
+template <class Comparator>
+void InlineSkipList<Comparator>::FindSpliceForLevel(const char* key,
+                                                    Node* before, Node* after,
+                                                    int level, Node** out_prev,
+                                                    Node** out_next) {
  while (true) {
    Node* next = before->Next(level);
    assert(before == head_ || next == nullptr ||
@@ -590,27 +661,28 @@ void InlineSkipList<Comparator>::FindLevelSplice(const char* key, Node* before,
 }

 template <class Comparator>
-void InlineSkipList<Comparator>::InsertConcurrently(const char* key) {
+void InlineSkipList<Comparator>::RecomputeSpliceLevels(const char* key,
+                                                       Splice* splice,
+                                                       int recompute_level) {
+  assert(recompute_level > 0);
+  assert(recompute_level <= splice->height_);
+  for (int i = recompute_level - 1; i >= 0; --i) {
+    FindSpliceForLevel(key, splice->prev_[i + 1], splice->next_[i + 1], i,
+                       &splice->prev_[i], &splice->next_[i]);
+  }
+}
+
+template <class Comparator>
+template <bool UseCAS>
+void InlineSkipList<Comparator>::Insert(const char* key, Splice* splice,
+                                        bool allow_partial_splice_fix) {
  Node* x = reinterpret_cast<Node*>(const_cast<char*>(key)) - 1;
  int height = x->UnstashHeight();
  assert(height >= 1 && height <= kMaxHeight_);

-  // We don't have a lock-free algorithm for updating prev_, but we do have
-  // the option of invalidating the entire sequential-insertion cache.
-  // prev_'s invariant is that prev_[i] (i > 0) is the predecessor of
-  // prev_[0] at that level.  We're only going to violate that if height
-  // > 1 and key lands after prev_[height - 1] but before prev_[0].
-  // Comparisons are pretty expensive, so an easier version is to just
-  // clear the cache if height > 1.  We only write to prev_height_ if the
-  // nobody else has, to avoid invalidating the root of the skip list in
-  // all of the other CPU caches.
-  if (height > 1 && prev_height_.load(std::memory_order_relaxed) != 0) {
-    prev_height_.store(0, std::memory_order_relaxed);
-  }
-
  int max_height = max_height_.load(std::memory_order_relaxed);
  while (height > max_height) {
-    if (max_height_.compare_exchange_strong(max_height, height)) {
+    if (max_height_.compare_exchange_weak(max_height, height)) {
      // successfully updated it
      max_height = height;
      break;
@@ -620,28 +692,159 @@ void InlineSkipList<Comparator>::InsertConcurrently(const char* key) {
  }
  assert(max_height <= kMaxPossibleHeight);

-  Node* prev[kMaxPossibleHeight + 1];
-  Node* next[kMaxPossibleHeight + 1];
-  prev[max_height] = head_;
-  next[max_height] = nullptr;
-  for (int i = max_height - 1; i >= 0; --i) {
-    FindLevelSplice(key, prev[i + 1], next[i + 1], i, &prev[i], &next[i]);
-  }
-  for (int i = 0; i < height; ++i) {
-    while (true) {
-      x->NoBarrier_SetNext(i, next[i]);
-      if (prev[i]->CASNext(i, next[i], x)) {
-        // success
+  int recompute_height = 0;
+  if (splice->height_ < max_height) {
+    // Either splice has never been used or max_height has grown since
+    // last use.  We could potentially fix it in the latter case, but
+    // that is tricky.
+    splice->prev_[max_height] = head_;
+    splice->next_[max_height] = nullptr;
+    splice->height_ = max_height;
+    recompute_height = max_height;
+  } else {
+    // Splice is a valid proper-height splice that brackets some
+    // key, but does it bracket this one?  We need to validate it and
+    // recompute a portion of the splice (levels 0..recompute_height-1)
+    // that is a superset of all levels that don't bracket the new key.
+    // Several choices are reasonable, because we have to balance the work
+    // saved against the extra comparisons required to validate the Splice.
+    //
+    // One strategy is just to recompute all of orig_splice_height if the
+    // bottom level isn't bracketing.  This pessimistically assumes that
+    // we will either get a perfect Splice hit (increasing sequential
+    // inserts) or have no locality.
+    //
+    // Another strategy is to walk up the Splice's levels until we find
+    // a level that brackets the key.  This strategy lets the Splice
+    // hint help for other cases: it turns insertion from O(log N) into
+    // O(log D), where D is the number of nodes in between the key that
+    // produced the Splice and the current insert (insertion is aided
+    // whether the new key is before or after the splice).  If you have
+    // a way of using a prefix of the key to map directly to the closest
+    // Splice out of O(sqrt(N)) Splices and we make it so that splices
+    // can also be used as hints during read, then we end up with Oshman's
+    // and Shavit's SkipTrie, which has O(log log N) lookup and insertion
+    // (compare to O(log N) for skip list).
+    //
+    // We control the pessimistic strategy with allow_partial_splice_fix.
+    // A good strategy is probably to be pessimistic for seq_splice_,
+    // optimistic if the caller actually went to the work of providing
+    // a Splice.
+    while (recompute_height < max_height) {
+      if (splice->prev_[recompute_height]->Next(recompute_height) !=
+          splice->next_[recompute_height]) {
+        // splice isn't tight at this level, there must have been some inserts
+        // to this
+        // location that didn't update the splice.  We might only be a little
+        // stale, but if
+        // the splice is very stale it would be O(N) to fix it.  We haven't used
+        // up any of
+        // our budget of comparisons, so always move up even if we are
+        // pessimistic about
+        // our chances of success.
+        ++recompute_height;
+      } else if (splice->prev_[recompute_height] != head_ &&
+                 !KeyIsAfterNode(key, splice->prev_[recompute_height])) {
+        // key is from before splice
+        if (allow_partial_splice_fix) {
+          // skip all levels with the same node without more comparisons
+          Node* bad = splice->prev_[recompute_height];
+          while (splice->prev_[recompute_height] == bad) {
+            ++recompute_height;
+          }
+        } else {
+          // we're pessimistic, recompute everything
+          recompute_height = max_height;
+        }
+      } else if (KeyIsAfterNode(key, splice->next_[recompute_height])) {
+        // key is from after splice
+        if (allow_partial_splice_fix) {
+          Node* bad = splice->next_[recompute_height];
+          while (splice->next_[recompute_height] == bad) {
+            ++recompute_height;
+          }
+        } else {
+          recompute_height = max_height;
+        }
+      } else {
+        // this level brackets the key, we won!
        break;
      }
-      // CAS failed, we need to recompute prev and next. It is unlikely
-      // to be helpful to try to use a different level as we redo the
-      // search, because it should be unlikely that lots of nodes have
-      // been inserted between prev[i] and next[i]. No point in using
-      // next[i] as the after hint, because we know it is stale.
-      FindLevelSplice(key, prev[i], nullptr, i, &prev[i], &next[i]);
    }
  }
+  assert(recompute_height <= max_height);
+  if (recompute_height > 0) {
+    RecomputeSpliceLevels(key, splice, recompute_height);
+  }
+
+  bool splice_is_valid = true;
+  if (UseCAS) {
+    for (int i = 0; i < height; ++i) {
+      while (true) {
+        assert(splice->next_[i] == nullptr ||
+               compare_(x->Key(), splice->next_[i]->Key()) < 0);
+        assert(splice->prev_[i] == head_ ||
+               compare_(splice->prev_[i]->Key(), x->Key()) < 0);
+        x->NoBarrier_SetNext(i, splice->next_[i]);
+        if (splice->prev_[i]->CASNext(i, splice->next_[i], x)) {
+          // success
+          break;
+        }
+        // CAS failed, we need to recompute prev and next. It is unlikely
+        // to be helpful to try to use a different level as we redo the
+        // search, because it should be unlikely that lots of nodes have
+        // been inserted between prev[i] and next[i]. No point in using
+        // next[i] as the after hint, because we know it is stale.
+        FindSpliceForLevel(key, splice->prev_[i], nullptr, i, &splice->prev_[i],
+                           &splice->next_[i]);
+
+        // Since we've narrowed the bracket for level i, we might have
+        // violated the Splice constraint between i and i-1.  Make sure
+        // we recompute the whole thing next time.
+        if (i > 0) {
+          splice_is_valid = false;
+        }
+      }
+    }
+  } else {
+    for (int i = 0; i < height; ++i) {
+      if (i >= recompute_height &&
+          splice->prev_[i]->Next(i) != splice->next_[i]) {
+        FindSpliceForLevel(key, splice->prev_[i], nullptr, i, &splice->prev_[i],
+                           &splice->next_[i]);
+      }
+      assert(splice->next_[i] == nullptr ||
+             compare_(x->Key(), splice->next_[i]->Key()) < 0);
+      assert(splice->prev_[i] == head_ ||
+             compare_(splice->prev_[i]->Key(), x->Key()) < 0);
+      assert(splice->prev_[i]->Next(i) == splice->next_[i]);
+      x->NoBarrier_SetNext(i, splice->next_[i]);
+      splice->prev_[i]->SetNext(i, x);
+    }
+  }
+  if (splice_is_valid) {
+    for (int i = 0; i < height; ++i) {
+      splice->prev_[i] = x;
+    }
+    assert(splice->prev_[splice->height_] == head_);
+    assert(splice->next_[splice->height_] == nullptr);
+    for (int i = 0; i < splice->height_; ++i) {
+      assert(splice->next_[i] == nullptr ||
+             compare_(key, splice->next_[i]->Key()) < 0);
+      assert(splice->prev_[i] == head_ ||
+             compare_(splice->prev_[i]->Key(), key) <= 0);
+      assert(splice->prev_[i + 1] == splice->prev_[i] ||
+             splice->prev_[i + 1] == head_ ||
+             compare_(splice->prev_[i + 1]->Key(), splice->prev_[i]->Key()) <
+                 0);
+      assert(splice->next_[i + 1] == splice->next_[i] ||
+             splice->next_[i + 1] == nullptr ||
+             compare_(splice->next_[i]->Key(), splice->next_[i + 1]->Key()) <
+                 0);
+    }
+  } else {
+    splice->height_ = 0;
+  }
 }

 template <class Comparator>
@@ -654,4 +857,44 @@ bool InlineSkipList<Comparator>::Contains(const char* key) const {
  }
 }

+template <class Comparator>
+void InlineSkipList<Comparator>::TEST_Validate() const {
+  // Interate over all levels at the same time, and verify nodes appear in
+  // the right order, and nodes appear in upper level also appear in lower
+  // levels.
+  Node* nodes[kMaxPossibleHeight];
+  int max_height = GetMaxHeight();
+  for (int i = 0; i < max_height; i++) {
+    nodes[i] = head_;
+  }
+  while (nodes[0] != nullptr) {
+    Node* l0_next = nodes[0]->Next(0);
+    if (l0_next == nullptr) {
+      break;
+    }
+    assert(nodes[0] == head_ || compare_(nodes[0]->Key(), l0_next->Key()) < 0);
+    nodes[0] = l0_next;
+
+    int i = 1;
+    while (i < max_height) {
+      Node* next = nodes[i]->Next(i);
+      if (next == nullptr) {
+        break;
+      }
+      auto cmp = compare_(nodes[0]->Key(), next->Key());
+      assert(cmp <= 0);
+      if (cmp == 0) {
+        assert(next == nodes[0]);
+        nodes[i] = next;
+      } else {
+        break;
+      }
+      i++;
+    }
+  }
+  for (int i = 1; i < max_height; i++) {
+    assert(nodes[i]->Next(i) == nullptr);
+  }
+}
+
 }  // namespace rocksdb
@@ -9,6 +9,7 @@

 #include "db/inlineskiplist.h"
 #include <set>
+#include <unordered_set>
 #include "rocksdb/env.h"
 #include "util/concurrent_arena.h"
 #include "util/hash.h"
@@ -42,7 +43,48 @@ struct TestComparator {
  }
 };

-class InlineSkipTest : public testing::Test {};
+typedef InlineSkipList<TestComparator> TestInlineSkipList;
+
+class InlineSkipTest : public testing::Test {
+ public:
+  void Insert(TestInlineSkipList* list, Key key) {
+    char* buf = list->AllocateKey(sizeof(Key));
+    memcpy(buf, &key, sizeof(Key));
+    list->Insert(buf);
+    keys_.insert(key);
+  }
+
+  void InsertWithHint(TestInlineSkipList* list, Key key, void** hint) {
+    char* buf = list->AllocateKey(sizeof(Key));
+    memcpy(buf, &key, sizeof(Key));
+    list->InsertWithHint(buf, hint);
+    keys_.insert(key);
+  }
+
+  void Validate(TestInlineSkipList* list) {
+    // Check keys exist.
+    for (Key key : keys_) {
+      ASSERT_TRUE(list->Contains(Encode(&key)));
+    }
+    // Iterate over the list, make sure keys appears in order and no extra
+    // keys exist.
+    TestInlineSkipList::Iterator iter(list);
+    ASSERT_FALSE(iter.Valid());
+    Key zero = 0;
+    iter.Seek(Encode(&zero));
+    for (Key key : keys_) {
+      ASSERT_TRUE(iter.Valid());
+      ASSERT_EQ(key, Decode(iter.key()));
+      iter.Next();
+    }
+    ASSERT_FALSE(iter.Valid());
+    // Validate the list is well-formed.
+    list->TEST_Validate();
+  }
+
+ private:
+  std::set<Key> keys_;
+};

 TEST_F(InlineSkipTest, Empty) {
  Arena arena;
@@ -58,6 +100,8 @@ TEST_F(InlineSkipTest, Empty) {
  key = 100;
  iter.Seek(Encode(&key));
  ASSERT_TRUE(!iter.Valid());
+  iter.SeekForPrev(Encode(&key));
+  ASSERT_TRUE(!iter.Valid());
  iter.SeekToLast();
  ASSERT_TRUE(!iter.Valid());
 }
@@ -97,6 +141,11 @@ TEST_F(InlineSkipTest, InsertAndLookup) {
    ASSERT_TRUE(iter.Valid());
    ASSERT_EQ(*(keys.begin()), Decode(iter.key()));

+    uint64_t max_key = R - 1;
+    iter.SeekForPrev(Encode(&max_key));
+    ASSERT_TRUE(iter.Valid());
+    ASSERT_EQ(*(keys.rbegin()), Decode(iter.key()));
+
    iter.SeekToFirst();
    ASSERT_TRUE(iter.Valid());
    ASSERT_EQ(*(keys.begin()), Decode(iter.key()));
@@ -127,21 +176,122 @@ TEST_F(InlineSkipTest, InsertAndLookup) {
  }

  // Backward iteration test
-  {
+  for (Key i = 0; i < R; i++) {
    InlineSkipList<TestComparator>::Iterator iter(&list);
-    iter.SeekToLast();
+    iter.SeekForPrev(Encode(&i));

    // Compare against model iterator
-    for (std::set<Key>::reverse_iterator model_iter = keys.rbegin();
-         model_iter != keys.rend(); ++model_iter) {
-      ASSERT_TRUE(iter.Valid());
-      ASSERT_EQ(*model_iter, Decode(iter.key()));
-      iter.Prev();
+    std::set<Key>::iterator model_iter = keys.upper_bound(i);
+    for (int j = 0; j < 3; j++) {
+      if (model_iter == keys.begin()) {
+        ASSERT_TRUE(!iter.Valid());
+        break;
+      } else {
+        ASSERT_TRUE(iter.Valid());
+        ASSERT_EQ(*--model_iter, Decode(iter.key()));
+        iter.Prev();
+      }
    }
-    ASSERT_TRUE(!iter.Valid());
  }
 }

+TEST_F(InlineSkipTest, InsertWithHint_Sequential) {
+  const int N = 100000;
+  Arena arena;
+  TestComparator cmp;
+  TestInlineSkipList list(cmp, &arena);
+  void* hint = nullptr;
+  for (int i = 0; i < N; i++) {
+    Key key = i;
+    InsertWithHint(&list, key, &hint);
+  }
+  Validate(&list);
+}
+
+TEST_F(InlineSkipTest, InsertWithHint_MultipleHints) {
+  const int N = 100000;
+  const int S = 100;
+  Random rnd(534);
+  Arena arena;
+  TestComparator cmp;
+  TestInlineSkipList list(cmp, &arena);
+  void* hints[S];
+  Key last_key[S];
+  for (int i = 0; i < S; i++) {
+    hints[i] = nullptr;
+    last_key[i] = 0;
+  }
+  for (int i = 0; i < N; i++) {
+    Key s = rnd.Uniform(S);
+    Key key = (s << 32) + (++last_key[s]);
+    InsertWithHint(&list, key, &hints[s]);
+  }
+  Validate(&list);
+}
+
+TEST_F(InlineSkipTest, InsertWithHint_MultipleHintsRandom) {
+  const int N = 100000;
+  const int S = 100;
+  Random rnd(534);
+  Arena arena;
+  TestComparator cmp;
+  TestInlineSkipList list(cmp, &arena);
+  void* hints[S];
+  for (int i = 0; i < S; i++) {
+    hints[i] = nullptr;
+  }
+  for (int i = 0; i < N; i++) {
+    Key s = rnd.Uniform(S);
+    Key key = (s << 32) + rnd.Next();
+    InsertWithHint(&list, key, &hints[s]);
+  }
+  Validate(&list);
+}
+
+TEST_F(InlineSkipTest, InsertWithHint_CompatibleWithInsertWithoutHint) {
+  const int N = 100000;
+  const int S1 = 100;
+  const int S2 = 100;
+  Random rnd(534);
+  Arena arena;
+  TestComparator cmp;
+  TestInlineSkipList list(cmp, &arena);
+  std::unordered_set<Key> used;
+  Key with_hint[S1];
+  Key without_hint[S2];
+  void* hints[S1];
+  for (int i = 0; i < S1; i++) {
+    hints[i] = nullptr;
+    while (true) {
+      Key s = rnd.Next();
+      if (used.insert(s).second) {
+        with_hint[i] = s;
+        break;
+      }
+    }
+  }
+  for (int i = 0; i < S2; i++) {
+    while (true) {
+      Key s = rnd.Next();
+      if (used.insert(s).second) {
+        without_hint[i] = s;
+        break;
+      }
+    }
+  }
+  for (int i = 0; i < N; i++) {
+    Key s = rnd.Uniform(S1 + S2);
+    if (s < S1) {
+      Key key = (with_hint[s] << 32) + rnd.Next();
+      InsertWithHint(&list, key, &hints[s]);
+    } else {
+      Key key = (without_hint[s - S1] << 32) + rnd.Next();
+      Insert(&list, key);
+    }
+  }
+  Validate(&list);
+}
+
 // We want to make sure that with a single writer and multiple
 // concurrent readers (with no synchronization other than when a
 // reader's iterator is created), the reader always observes all the
@@ -25,62 +25,146 @@
 namespace rocksdb {

 #ifndef ROCKSDB_LITE
+
+const std::map<LevelStatType, LevelStat> InternalStats::compaction_level_stats =
+    {
+        {LevelStatType::NUM_FILES, LevelStat{"NumFiles", "Files"}},
+        {LevelStatType::COMPACTED_FILES,
+         LevelStat{"CompactedFiles", "CompactedFiles"}},
+        {LevelStatType::SIZE_MB, LevelStat{"SizeMB", "Size(MB}"}},
+        {LevelStatType::SCORE, LevelStat{"Score", "Score"}},
+        {LevelStatType::READ_GB, LevelStat{"ReadGB", "Read(GB}"}},
+        {LevelStatType::RN_GB, LevelStat{"RnGB", "Rn(GB}"}},
+        {LevelStatType::RNP1_GB, LevelStat{"Rnp1GB", "Rnp1(GB}"}},
+        {LevelStatType::WRITE_GB, LevelStat{"WriteGB", "Write(GB}"}},
+        {LevelStatType::W_NEW_GB, LevelStat{"WnewGB", "Wnew(GB}"}},
+        {LevelStatType::MOVED_GB, LevelStat{"MovedGB", "Moved(GB}"}},
+        {LevelStatType::WRITE_AMP, LevelStat{"WriteAmp", "W-Amp"}},
+        {LevelStatType::READ_MBPS, LevelStat{"ReadMBps", "Rd(MB/s}"}},
+        {LevelStatType::WRITE_MBPS, LevelStat{"WriteMBps", "Wr(MB/s}"}},
+        {LevelStatType::COMP_SEC, LevelStat{"CompSec", "Comp(sec}"}},
+        {LevelStatType::COMP_COUNT, LevelStat{"CompCount", "Comp(cnt}"}},
+        {LevelStatType::AVG_SEC, LevelStat{"AvgSec", "Avg(sec}"}},
+        {LevelStatType::KEY_IN, LevelStat{"KeyIn", "KeyIn"}},
+        {LevelStatType::KEY_DROP, LevelStat{"KeyDrop", "KeyDrop"}},
+};
+
 namespace {
 const double kMB = 1048576.0;
 const double kGB = kMB * 1024;
 const double kMicrosInSec = 1000000.0;

 void PrintLevelStatsHeader(char* buf, size_t len, const std::string& cf_name) {
-  snprintf(
-      buf, len,
-      "\n** Compaction Stats [%s] **\n"
-      "Level    Files   Size(MB) Score Read(GB)  Rn(GB) Rnp1(GB) "
-      "Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) "
-      "Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop\n"
-      "--------------------------------------------------------------------"
-      "-----------------------------------------------------------"
-      "--------------------------------------\n",
-      cf_name.c_str());
+  int written_size =
+      snprintf(buf, len, "\n** Compaction Stats [%s] **\n", cf_name.c_str());
+  auto hdr = [](LevelStatType t) {
+    return InternalStats::compaction_level_stats.at(t).header_name.c_str();
+  };
+  int line_size = snprintf(
+      buf + written_size, len - written_size,
+      "Level    %s   %s %s %s  %s %s %s %s %s %s %s %s %s %s %s %s %s\n",
+      // Note that we skip COMPACTED_FILES and merge it with Files column
+      hdr(LevelStatType::NUM_FILES), hdr(LevelStatType::SIZE_MB),
+      hdr(LevelStatType::SCORE), hdr(LevelStatType::READ_GB),
+      hdr(LevelStatType::RN_GB), hdr(LevelStatType::RNP1_GB),
+      hdr(LevelStatType::WRITE_GB), hdr(LevelStatType::W_NEW_GB),
+      hdr(LevelStatType::MOVED_GB), hdr(LevelStatType::WRITE_AMP),
+      hdr(LevelStatType::READ_MBPS), hdr(LevelStatType::WRITE_MBPS),
+      hdr(LevelStatType::COMP_SEC), hdr(LevelStatType::COMP_COUNT),
+      hdr(LevelStatType::AVG_SEC), hdr(LevelStatType::KEY_IN),
+      hdr(LevelStatType::KEY_DROP));
+
+  written_size += line_size;
+  snprintf(buf + written_size, len - written_size, "%s\n",
+           std::string(line_size, '-').c_str());
 }

-void PrintLevelStats(char* buf, size_t len, const std::string& name,
-  int num_files, int being_compacted, double total_file_size,
-  double score, double w_amp,
-  const InternalStats::CompactionStats& stats) {
+void PrepareLevelStats(std::map<LevelStatType, double>* level_stats,
+                       int num_files, int being_compacted,
+                       double total_file_size, double score, double w_amp,
+                       const InternalStats::CompactionStats& stats) {
  uint64_t bytes_read =
      stats.bytes_read_non_output_levels + stats.bytes_read_output_level;
  int64_t bytes_new =
      stats.bytes_written - stats.bytes_read_output_level;
  double elapsed = (stats.micros + 1) / kMicrosInSec;
-  std::string num_input_records = NumberToHumanString(stats.num_input_records);
-  std::string num_dropped_records =
-      NumberToHumanString(stats.num_dropped_records);

+  (*level_stats)[LevelStatType::NUM_FILES] = num_files;
+  (*level_stats)[LevelStatType::COMPACTED_FILES] = being_compacted;
+  (*level_stats)[LevelStatType::SIZE_MB] = total_file_size / kMB;
+  (*level_stats)[LevelStatType::SCORE] = score;
+  (*level_stats)[LevelStatType::READ_GB] = bytes_read / kGB;
+  (*level_stats)[LevelStatType::RN_GB] =
+      stats.bytes_read_non_output_levels / kGB;
+  (*level_stats)[LevelStatType::RNP1_GB] = stats.bytes_read_output_level / kGB;
+  (*level_stats)[LevelStatType::WRITE_GB] = stats.bytes_written / kGB;
+  (*level_stats)[LevelStatType::W_NEW_GB] = bytes_new / kGB;
+  (*level_stats)[LevelStatType::MOVED_GB] = stats.bytes_moved / kGB;
+  (*level_stats)[LevelStatType::WRITE_AMP] = w_amp;
+  (*level_stats)[LevelStatType::READ_MBPS] = bytes_read / kMB / elapsed;
+  (*level_stats)[LevelStatType::WRITE_MBPS] =
+      stats.bytes_written / kMB / elapsed;
+  (*level_stats)[LevelStatType::COMP_SEC] = stats.micros / kMicrosInSec;
+  (*level_stats)[LevelStatType::COMP_COUNT] = stats.count;
+  (*level_stats)[LevelStatType::AVG_SEC] =
+      stats.count == 0 ? 0 : stats.micros / kMicrosInSec / stats.count;
+  (*level_stats)[LevelStatType::KEY_IN] =
+      static_cast<double>(stats.num_input_records);
+  (*level_stats)[LevelStatType::KEY_DROP] =
+      static_cast<double>(stats.num_dropped_records);
+}
+
+void PrintLevelStats(char* buf, size_t len, const std::string& name,
+                     const std::map<LevelStatType, double>& stat_value) {
  snprintf(buf, len,
-           "%4s %6d/%-3d %8.2f %5.1f " /* Level, Files, Size(MB), Score */
-           "%8.1f "                    /* Read(GB) */
-           "%7.1f "                    /* Rn(GB) */
-           "%8.1f "                    /* Rnp1(GB) */
-           "%9.1f "                    /* Write(GB) */
-           "%8.1f "                    /* Wnew(GB) */
-           "%9.1f "                    /* Moved(GB) */
-           "%5.1f "                    /* W-Amp */
-           "%8.1f "                    /* Rd(MB/s) */
-           "%8.1f "                    /* Wr(MB/s) */
-           "%9.0f "                    /* Comp(sec) */
-           "%9d "                      /* Comp(cnt) */
-           "%8.3f "                    /* Avg(sec) */
-           "%7s "   /* KeyIn */
-           "%6s\n", /* KeyDrop */
+           "%4s %6d/%-3d %8.2f %5.1f " /*  Level, Files, Size(MB), Score */
+           "%8.1f "                    /*  Read(GB) */
+           "%7.1f "                    /*  Rn(GB) */
+           "%8.1f "                    /*  Rnp1(GB) */
+           "%9.1f "                    /*  Write(GB) */
+           "%8.1f "                    /*  Wnew(GB) */
+           "%9.1f "                    /*  Moved(GB) */
+           "%5.1f "                    /*  W-Amp */
+           "%8.1f "                    /*  Rd(MB/s) */
+           "%8.1f "                    /*  Wr(MB/s) */
+           "%9.0f "                    /*  Comp(sec) */
+           "%9d "                      /*  Comp(cnt) */
+           "%8.3f "                    /*  Avg(sec) */
+           "%7s "                      /*  KeyIn */
+           "%6s\n",                    /*  KeyDrop */
           name.c_str(),
-           num_files, being_compacted, total_file_size / kMB, score,
-           bytes_read / kGB, stats.bytes_read_non_output_levels / kGB,
-           stats.bytes_read_output_level / kGB, stats.bytes_written / kGB,
-           bytes_new / kGB, stats.bytes_moved / kGB, w_amp,
-           bytes_read / kMB / elapsed, stats.bytes_written / kMB / elapsed,
-           stats.micros / kMicrosInSec, stats.count,
-           stats.count == 0 ? 0 : stats.micros / kMicrosInSec / stats.count,
-           num_input_records.c_str(), num_dropped_records.c_str());
+           static_cast<int>(stat_value.at(LevelStatType::NUM_FILES)),
+           static_cast<int>(stat_value.at(LevelStatType::COMPACTED_FILES)),
+           stat_value.at(LevelStatType::SIZE_MB),
+           stat_value.at(LevelStatType::SCORE),
+           stat_value.at(LevelStatType::READ_GB),
+           stat_value.at(LevelStatType::RN_GB),
+           stat_value.at(LevelStatType::RNP1_GB),
+           stat_value.at(LevelStatType::WRITE_GB),
+           stat_value.at(LevelStatType::W_NEW_GB),
+           stat_value.at(LevelStatType::MOVED_GB),
+           stat_value.at(LevelStatType::WRITE_AMP),
+           stat_value.at(LevelStatType::READ_MBPS),
+           stat_value.at(LevelStatType::WRITE_MBPS),
+           stat_value.at(LevelStatType::COMP_SEC),
+           static_cast<int>(stat_value.at(LevelStatType::COMP_COUNT)),
+           stat_value.at(LevelStatType::AVG_SEC),
+           NumberToHumanString(
+               static_cast<std::int64_t>(stat_value.at(LevelStatType::KEY_IN)))
+               .c_str(),
+           NumberToHumanString(static_cast<std::int64_t>(
+                                   stat_value.at(LevelStatType::KEY_DROP)))
+               .c_str());
+}
+
+void PrintLevelStats(char* buf, size_t len, const std::string& name,
+                     int num_files, int being_compacted, double total_file_size,
+                     double score, double w_amp,
+                     const InternalStats::CompactionStats& stats) {
+  std::map<LevelStatType, double> level_stats;
+  PrepareLevelStats(&level_stats, num_files, being_compacted, total_file_size,
+                    score, w_amp, stats);
+  PrintLevelStats(buf, len, name, level_stats);
 }

 // Assumes that trailing numbers represent an optional argument. This requires
@@ -137,6 +221,7 @@ static const std::string num_live_versions = "num-live-versions";
 static const std::string current_version_number =
    "current-super-version-number";
 static const std::string estimate_live_data_size = "estimate-live-data-size";
+static const std::string min_log_number_to_keep = "min-log-number-to-keep";
 static const std::string base_level = "base-level";
 static const std::string total_sst_files_size = "total-sst-files-size";
 static const std::string estimate_pending_comp_bytes =
@@ -201,6 +286,8 @@ const std::string DB::Properties::kCurrentSuperVersionNumber =
    rocksdb_prefix + current_version_number;
 const std::string DB::Properties::kEstimateLiveDataSize =
                      rocksdb_prefix + estimate_live_data_size;
+const std::string DB::Properties::kMinLogNumberToKeep =
+    rocksdb_prefix + min_log_number_to_keep;
 const std::string DB::Properties::kTotalSstFilesSize =
                      rocksdb_prefix + total_sst_files_size;
 const std::string DB::Properties::kBaseLevel = rocksdb_prefix + base_level;
@@ -211,73 +298,93 @@ const std::string DB::Properties::kAggregatedTableProperties =
 const std::string DB::Properties::kAggregatedTablePropertiesAtLevel =
    rocksdb_prefix + aggregated_table_properties_at_level;

-const std::unordered_map<std::string,
-                         DBPropertyInfo> InternalStats::ppt_name_to_info = {
-    {DB::Properties::kNumFilesAtLevelPrefix,
-     {false, &InternalStats::HandleNumFilesAtLevel, nullptr}},
-    {DB::Properties::kCompressionRatioAtLevelPrefix,
-     {false, &InternalStats::HandleCompressionRatioAtLevelPrefix, nullptr}},
-    {DB::Properties::kLevelStats,
-     {false, &InternalStats::HandleLevelStats, nullptr}},
-    {DB::Properties::kStats, {false, &InternalStats::HandleStats, nullptr}},
-    {DB::Properties::kCFStats, {false, &InternalStats::HandleCFStats, nullptr}},
-    {DB::Properties::kDBStats, {false, &InternalStats::HandleDBStats, nullptr}},
-    {DB::Properties::kSSTables,
-     {false, &InternalStats::HandleSsTables, nullptr}},
-    {DB::Properties::kAggregatedTableProperties,
-     {false, &InternalStats::HandleAggregatedTableProperties, nullptr}},
-    {DB::Properties::kAggregatedTablePropertiesAtLevel,
-     {false, &InternalStats::HandleAggregatedTablePropertiesAtLevel, nullptr}},
-    {DB::Properties::kNumImmutableMemTable,
-     {false, nullptr, &InternalStats::HandleNumImmutableMemTable}},
-    {DB::Properties::kNumImmutableMemTableFlushed,
-     {false, nullptr, &InternalStats::HandleNumImmutableMemTableFlushed}},
-    {DB::Properties::kMemTableFlushPending,
-     {false, nullptr, &InternalStats::HandleMemTableFlushPending}},
-    {DB::Properties::kCompactionPending,
-     {false, nullptr, &InternalStats::HandleCompactionPending}},
-    {DB::Properties::kBackgroundErrors,
-     {false, nullptr, &InternalStats::HandleBackgroundErrors}},
-    {DB::Properties::kCurSizeActiveMemTable,
-     {false, nullptr, &InternalStats::HandleCurSizeActiveMemTable}},
-    {DB::Properties::kCurSizeAllMemTables,
-     {false, nullptr, &InternalStats::HandleCurSizeAllMemTables}},
-    {DB::Properties::kSizeAllMemTables,
-     {false, nullptr, &InternalStats::HandleSizeAllMemTables}},
-    {DB::Properties::kNumEntriesActiveMemTable,
-     {false, nullptr, &InternalStats::HandleNumEntriesActiveMemTable}},
-    {DB::Properties::kNumEntriesImmMemTables,
-     {false, nullptr, &InternalStats::HandleNumEntriesImmMemTables}},
-    {DB::Properties::kNumDeletesActiveMemTable,
-     {false, nullptr, &InternalStats::HandleNumDeletesActiveMemTable}},
-    {DB::Properties::kNumDeletesImmMemTables,
-     {false, nullptr, &InternalStats::HandleNumDeletesImmMemTables}},
-    {DB::Properties::kEstimateNumKeys,
-     {false, nullptr, &InternalStats::HandleEstimateNumKeys}},
-    {DB::Properties::kEstimateTableReadersMem,
-     {true, nullptr, &InternalStats::HandleEstimateTableReadersMem}},
-    {DB::Properties::kIsFileDeletionsEnabled,
-     {false, nullptr, &InternalStats::HandleIsFileDeletionsEnabled}},
-    {DB::Properties::kNumSnapshots,
-     {false, nullptr, &InternalStats::HandleNumSnapshots}},
-    {DB::Properties::kOldestSnapshotTime,
-     {false, nullptr, &InternalStats::HandleOldestSnapshotTime}},
-    {DB::Properties::kNumLiveVersions,
-     {false, nullptr, &InternalStats::HandleNumLiveVersions}},
-    {DB::Properties::kCurrentSuperVersionNumber,
-     {false, nullptr, &InternalStats::HandleCurrentSuperVersionNumber}},
-    {DB::Properties::kEstimateLiveDataSize,
-     {true, nullptr, &InternalStats::HandleEstimateLiveDataSize}},
-    {DB::Properties::kBaseLevel,
-     {false, nullptr, &InternalStats::HandleBaseLevel}},
-    {DB::Properties::kTotalSstFilesSize,
-     {false, nullptr, &InternalStats::HandleTotalSstFilesSize}},
-    {DB::Properties::kEstimatePendingCompactionBytes,
-     {false, nullptr, &InternalStats::HandleEstimatePendingCompactionBytes}},
-    {DB::Properties::kNumRunningFlushes,
-     {false, nullptr, &InternalStats::HandleNumRunningFlushes}},
-    {DB::Properties::kNumRunningCompactions,
-     {false, nullptr, &InternalStats::HandleNumRunningCompactions}},
+const std::unordered_map<std::string, DBPropertyInfo>
+    InternalStats::ppt_name_to_info = {
+        {DB::Properties::kNumFilesAtLevelPrefix,
+         {false, &InternalStats::HandleNumFilesAtLevel, nullptr, nullptr}},
+        {DB::Properties::kCompressionRatioAtLevelPrefix,
+         {false, &InternalStats::HandleCompressionRatioAtLevelPrefix, nullptr,
+          nullptr}},
+        {DB::Properties::kLevelStats,
+         {false, &InternalStats::HandleLevelStats, nullptr, nullptr}},
+        {DB::Properties::kStats,
+         {false, &InternalStats::HandleStats, nullptr, nullptr}},
+        {DB::Properties::kCFStats,
+         {false, &InternalStats::HandleCFStats, nullptr,
+          &InternalStats::HandleCFMapStats}},
+        {DB::Properties::kDBStats,
+         {false, &InternalStats::HandleDBStats, nullptr, nullptr}},
+        {DB::Properties::kSSTables,
+         {false, &InternalStats::HandleSsTables, nullptr, nullptr}},
+        {DB::Properties::kAggregatedTableProperties,
+         {false, &InternalStats::HandleAggregatedTableProperties, nullptr,
+          nullptr}},
+        {DB::Properties::kAggregatedTablePropertiesAtLevel,
+         {false, &InternalStats::HandleAggregatedTablePropertiesAtLevel,
+          nullptr, nullptr}},
+        {DB::Properties::kNumImmutableMemTable,
+         {false, nullptr, &InternalStats::HandleNumImmutableMemTable, nullptr}},
+        {DB::Properties::kNumImmutableMemTableFlushed,
+         {false, nullptr, &InternalStats::HandleNumImmutableMemTableFlushed,
+          nullptr}},
+        {DB::Properties::kMemTableFlushPending,
+         {false, nullptr, &InternalStats::HandleMemTableFlushPending, nullptr}},
+        {DB::Properties::kCompactionPending,
+         {false, nullptr, &InternalStats::HandleCompactionPending, nullptr}},
+        {DB::Properties::kBackgroundErrors,
+         {false, nullptr, &InternalStats::HandleBackgroundErrors, nullptr}},
+        {DB::Properties::kCurSizeActiveMemTable,
+         {false, nullptr, &InternalStats::HandleCurSizeActiveMemTable,
+          nullptr}},
+        {DB::Properties::kCurSizeAllMemTables,
+         {false, nullptr, &InternalStats::HandleCurSizeAllMemTables, nullptr}},
+        {DB::Properties::kSizeAllMemTables,
+         {false, nullptr, &InternalStats::HandleSizeAllMemTables, nullptr}},
+        {DB::Properties::kNumEntriesActiveMemTable,
+         {false, nullptr, &InternalStats::HandleNumEntriesActiveMemTable,
+          nullptr}},
+        {DB::Properties::kNumEntriesImmMemTables,
+         {false, nullptr, &InternalStats::HandleNumEntriesImmMemTables,
+          nullptr}},
+        {DB::Properties::kNumDeletesActiveMemTable,
+         {false, nullptr, &InternalStats::HandleNumDeletesActiveMemTable,
+          nullptr}},
+        {DB::Properties::kNumDeletesImmMemTables,
+         {false, nullptr, &InternalStats::HandleNumDeletesImmMemTables,
+          nullptr}},
+        {DB::Properties::kEstimateNumKeys,
+         {false, nullptr, &InternalStats::HandleEstimateNumKeys, nullptr}},
+        {DB::Properties::kEstimateTableReadersMem,
+         {true, nullptr, &InternalStats::HandleEstimateTableReadersMem,
+          nullptr}},
+        {DB::Properties::kIsFileDeletionsEnabled,
+         {false, nullptr, &InternalStats::HandleIsFileDeletionsEnabled,
+          nullptr}},
+        {DB::Properties::kNumSnapshots,
+         {false, nullptr, &InternalStats::HandleNumSnapshots, nullptr}},
+        {DB::Properties::kOldestSnapshotTime,
+         {false, nullptr, &InternalStats::HandleOldestSnapshotTime, nullptr}},
+        {DB::Properties::kNumLiveVersions,
+         {false, nullptr, &InternalStats::HandleNumLiveVersions, nullptr}},
+        {DB::Properties::kCurrentSuperVersionNumber,
+         {false, nullptr, &InternalStats::HandleCurrentSuperVersionNumber,
+          nullptr}},
+        {DB::Properties::kEstimateLiveDataSize,
+         {true, nullptr, &InternalStats::HandleEstimateLiveDataSize, nullptr}},
+        {DB::Properties::kMinLogNumberToKeep,
+         {false, nullptr, &InternalStats::HandleMinLogNumberToKeep, nullptr}},
+        {DB::Properties::kBaseLevel,
+         {false, nullptr, &InternalStats::HandleBaseLevel, nullptr}},
+        {DB::Properties::kTotalSstFilesSize,
+         {false, nullptr, &InternalStats::HandleTotalSstFilesSize, nullptr}},
+        {DB::Properties::kEstimatePendingCompactionBytes,
+         {false, nullptr, &InternalStats::HandleEstimatePendingCompactionBytes,
+          nullptr}},
+        {DB::Properties::kNumRunningFlushes,
+         {false, nullptr, &InternalStats::HandleNumRunningFlushes, nullptr}},
+        {DB::Properties::kNumRunningCompactions,
+         {false, nullptr, &InternalStats::HandleNumRunningCompactions,
+          nullptr}},
 };

 const DBPropertyInfo* GetPropertyInfo(const Slice& property) {
@@ -298,6 +405,14 @@ bool InternalStats::GetStringProperty(const DBPropertyInfo& property_info,
  return (this->*(property_info.handle_string))(value, arg);
 }

+bool InternalStats::GetMapProperty(const DBPropertyInfo& property_info,
+                                   const Slice& property,
+                                   std::map<std::string, double>* value) {
+  assert(value != nullptr);
+  assert(property_info.handle_map != nullptr);
+  return (this->*(property_info.handle_map))(value);
+}
+
 bool InternalStats::GetIntProperty(const DBPropertyInfo& property_info,
                                   uint64_t* value, DBImpl* db) {
  assert(value != nullptr);
@@ -370,6 +485,11 @@ bool InternalStats::HandleStats(std::string* value, Slice suffix) {
  return true;
 }

+bool InternalStats::HandleCFMapStats(std::map<std::string, double>* cf_stats) {
+  DumpCFMapStats(cf_stats);
+  return true;
+}
+
 bool InternalStats::HandleCFStats(std::string* value, Slice suffix) {
  DumpCFStats(value);
  return true;
@@ -590,6 +710,12 @@ bool InternalStats::HandleEstimateLiveDataSize(uint64_t* value, DBImpl* db,
  return true;
 }

+bool InternalStats::HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db,
+                                             Version* version) {
+  *value = db->MinLogNumberToKeep();
+  return true;
+}
+
 void InternalStats::DumpDBStats(std::string* value) {
  char buf[1000];
  // DB-level stats, only available from default column family
@@ -716,7 +842,32 @@ void InternalStats::DumpDBStats(std::string* value) {
  db_stats_snapshot_.write_stall_micros = write_stall_micros;
 }

-void InternalStats::DumpCFStats(std::string* value) {
+/**
+ * Dump Compaction Level stats to a map of stat name to value in double.
+ * The level in stat name is represented with a prefix "Lx" where "x"
+ * is the level number. A special level "Sum" represents the sum of a stat
+ * for all levels.
+ */
+void InternalStats::DumpCFMapStats(std::map<std::string, double>* cf_stats) {
+  CompactionStats compaction_stats_sum(0);
+  std::map<int, std::map<LevelStatType, double>> levels_stats;
+  DumpCFMapStats(&levels_stats, &compaction_stats_sum);
+  for (auto const& level_ent : levels_stats) {
+    auto level_str =
+        level_ent.first == -1 ? "Sum" : "L" + ToString(level_ent.first);
+    for (auto const& stat_ent : level_ent.second) {
+      auto stat_type = stat_ent.first;
+      auto key_str =
+          level_str + "." +
+          InternalStats::compaction_level_stats.at(stat_type).property_name;
+      (*cf_stats)[key_str] = stat_ent.second;
+    }
+  }
+}
+
+int InternalStats::DumpCFMapStats(
+    std::map<int, std::map<LevelStatType, double>>* levels_stats,
+    CompactionStats* compaction_stats_sum) {
  const VersionStorageInfo* vstorage = cfd_->current()->storage_info();

  int num_levels_to_check =
@@ -724,7 +875,7 @@ void InternalStats::DumpCFStats(std::string* value) {
          ? vstorage->num_levels() - 1
          : 1;

-  // Compaction scores are sorted base on its value. Restore them to the
+  // Compaction scores are sorted based on its value. Restore them to the
  // level order
  std::vector<double> compaction_score(number_levels_, 0);
  for (int i = 0; i < num_levels_to_check; ++i) {
@@ -741,12 +892,6 @@ void InternalStats::DumpCFStats(std::string* value) {
    }
  }

-  char buf[1000];
-  // Per-ColumnFamily stats
-  PrintLevelStatsHeader(buf, sizeof(buf), cfd_->GetName());
-  value->append(buf);
-
-  CompactionStats stats_sum(0);
  int total_files = 0;
  int total_files_being_compacted = 0;
  double total_file_size = 0;
@@ -755,47 +900,78 @@ void InternalStats::DumpCFStats(std::string* value) {
    total_files += files;
    total_files_being_compacted += files_being_compacted[level];
    if (comp_stats_[level].micros > 0 || files > 0) {
-      stats_sum.Add(comp_stats_[level]);
+      compaction_stats_sum->Add(comp_stats_[level]);
      total_file_size += vstorage->NumLevelBytes(level);
      double w_amp =
-          (comp_stats_[level].bytes_read_non_output_levels == 0) ? 0.0
-          : static_cast<double>(comp_stats_[level].bytes_written) /
-            comp_stats_[level].bytes_read_non_output_levels;
-      PrintLevelStats(buf, sizeof(buf), "L" + ToString(level), files,
-                      files_being_compacted[level],
-                      static_cast<double>(vstorage->NumLevelBytes(level)),
-                      compaction_score[level],
-                      w_amp, comp_stats_[level]);
-      value->append(buf);
+          (comp_stats_[level].bytes_read_non_output_levels == 0)
+              ? 0.0
+              : static_cast<double>(comp_stats_[level].bytes_written) /
+                    comp_stats_[level].bytes_read_non_output_levels;
+      std::map<LevelStatType, double> level_stats;
+      PrepareLevelStats(&level_stats, files, files_being_compacted[level],
+                        static_cast<double>(vstorage->NumLevelBytes(level)),
+                        compaction_score[level], w_amp, comp_stats_[level]);
+      (*levels_stats)[level] = level_stats;
    }
  }
-
  uint64_t flush_ingest = cf_stats_value_[BYTES_FLUSHED];
  uint64_t add_file_ingest = cf_stats_value_[BYTES_INGESTED_ADD_FILE];
  uint64_t curr_ingest = flush_ingest + add_file_ingest;
  // Cumulative summary
-  double w_amp = stats_sum.bytes_written / static_cast<double>(curr_ingest + 1);
+  double w_amp = compaction_stats_sum->bytes_written /
+                 static_cast<double>(curr_ingest + 1);
+  // Stats summary across levels
+  std::map<LevelStatType, double> sum_stats;
+  PrepareLevelStats(&sum_stats, total_files, total_files_being_compacted,
+                    total_file_size, 0, w_amp, *compaction_stats_sum);
+  (*levels_stats)[-1] = sum_stats;  //  -1 is for the Sum level
+  return num_levels_to_check;
+}
+
+void InternalStats::DumpCFStats(std::string* value) {
+  char buf[1000];
+  // Per-ColumnFamily stats
+  PrintLevelStatsHeader(buf, sizeof(buf), cfd_->GetName());
+  value->append(buf);
+
+  // Print stats for each level
+  std::map<int, std::map<LevelStatType, double>> levels_stats;
+  CompactionStats compaction_stats_sum(0);
+  int levels = DumpCFMapStats(&levels_stats, &compaction_stats_sum);
+  for (int l = 0; l < levels; ++l) {
+    if (levels_stats.find(l) != levels_stats.end()) {
+      PrintLevelStats(buf, sizeof(buf), "L" + ToString(l), levels_stats[l]);
+      value->append(buf);
+    }
+  }
+  // Print sum of level stats
+  PrintLevelStats(buf, sizeof(buf), "Sum", levels_stats[-1]);
+  value->append(buf);
+
+  uint64_t flush_ingest = cf_stats_value_[BYTES_FLUSHED];
+  uint64_t add_file_ingest = cf_stats_value_[BYTES_INGESTED_ADD_FILE];
+  uint64_t ingest_files_addfile = cf_stats_value_[INGESTED_NUM_FILES_TOTAL];
+  uint64_t ingest_l0_files_addfile =
+      cf_stats_value_[INGESTED_LEVEL0_NUM_FILES_TOTAL];
+  uint64_t ingest_keys_addfile = cf_stats_value_[INGESTED_NUM_KEYS_TOTAL];
+  // Cumulative summary
  uint64_t total_stall_count =
      cf_stats_count_[LEVEL0_SLOWDOWN_TOTAL] +
      cf_stats_count_[LEVEL0_NUM_FILES_TOTAL] +
      cf_stats_count_[SOFT_PENDING_COMPACTION_BYTES_LIMIT] +
      cf_stats_count_[HARD_PENDING_COMPACTION_BYTES_LIMIT] +
      cf_stats_count_[MEMTABLE_COMPACTION] + cf_stats_count_[MEMTABLE_SLOWDOWN];
-  // Stats summary across levels
-  PrintLevelStats(buf, sizeof(buf), "Sum", total_files,
-                  total_files_being_compacted, total_file_size, 0, w_amp,
-                  stats_sum);
-  value->append(buf);
  // Interval summary
  uint64_t interval_flush_ingest =
      flush_ingest - cf_stats_snapshot_.ingest_bytes_flush;
  uint64_t interval_add_file_inget =
-      add_file_ingest - cf_stats_snapshot_.ingest_bytes_add_file;
+      add_file_ingest - cf_stats_snapshot_.ingest_bytes_addfile;
  uint64_t interval_ingest =
      interval_flush_ingest + interval_add_file_inget + 1;
-  CompactionStats interval_stats(stats_sum);
+  CompactionStats interval_stats(compaction_stats_sum);
  interval_stats.Subtract(cf_stats_snapshot_.comp_stats);
-  w_amp = interval_stats.bytes_written / static_cast<double>(interval_ingest);
+  double w_amp =
+      interval_stats.bytes_written / static_cast<double>(interval_ingest);
  PrintLevelStats(buf, sizeof(buf), "Int", 0, 0, 0, 0, w_amp, interval_stats);
  value->append(buf);

@@ -804,13 +980,34 @@ void InternalStats::DumpCFStats(std::string* value) {
  snprintf(buf, sizeof(buf), "Uptime(secs): %.1f total, %.1f interval\n",
           seconds_up, interval_seconds_up);
  value->append(buf);
-
  snprintf(buf, sizeof(buf), "Flush(GB): cumulative %.3f, interval %.3f\n",
           flush_ingest / kGB, interval_flush_ingest / kGB);
+  value->append(buf);
  snprintf(buf, sizeof(buf), "AddFile(GB): cumulative %.3f, interval %.3f\n",
           add_file_ingest / kGB, interval_add_file_inget / kGB);
  value->append(buf);

+  uint64_t interval_ingest_files_addfile =
+      ingest_files_addfile - cf_stats_snapshot_.ingest_files_addfile;
+  snprintf(buf, sizeof(buf), "AddFile(Total Files): cumulative %" PRIu64
+                             ", interval %" PRIu64 "\n",
+           ingest_files_addfile, interval_ingest_files_addfile);
+  value->append(buf);
+
+  uint64_t interval_ingest_l0_files_addfile =
+      ingest_l0_files_addfile - cf_stats_snapshot_.ingest_l0_files_addfile;
+  snprintf(buf, sizeof(buf),
+           "AddFile(L0 Files): cumulative %" PRIu64 ", interval %" PRIu64 "\n",
+           ingest_l0_files_addfile, interval_ingest_l0_files_addfile);
+  value->append(buf);
+
+  uint64_t interval_ingest_keys_addfile =
+      ingest_keys_addfile - cf_stats_snapshot_.ingest_keys_addfile;
+  snprintf(buf, sizeof(buf),
+           "AddFile(Keys): cumulative %" PRIu64 ", interval %" PRIu64 "\n",
+           ingest_keys_addfile, interval_ingest_keys_addfile);
+  value->append(buf);
+
  // Compact
  uint64_t compact_bytes_read = 0;
  uint64_t compact_bytes_write = 0;
@@ -881,8 +1078,11 @@ void InternalStats::DumpCFStats(std::string* value) {
  value->append(buf);

  cf_stats_snapshot_.ingest_bytes_flush = flush_ingest;
-  cf_stats_snapshot_.ingest_bytes_add_file = add_file_ingest;
-  cf_stats_snapshot_.comp_stats = stats_sum;
+  cf_stats_snapshot_.ingest_bytes_addfile = add_file_ingest;
+  cf_stats_snapshot_.ingest_files_addfile = ingest_files_addfile;
+  cf_stats_snapshot_.ingest_l0_files_addfile = ingest_l0_files_addfile;
+  cf_stats_snapshot_.ingest_keys_addfile = ingest_keys_addfile;
+  cf_stats_snapshot_.comp_stats = compaction_stats_sum;
  cf_stats_snapshot_.stall_count = total_stall_count;
 }

@@ -9,10 +9,11 @@
 //

 #pragma once
-#include "db/version_set.h"
-
-#include <vector>
+#include <map>
 #include <string>
+#include <vector>
+
+#include "db/version_set.h"

 class ColumnFamilyData;

@@ -41,13 +42,47 @@ struct DBPropertyInfo {
  //      holding db mutex, which is only supported for int properties.
  bool (InternalStats::*handle_int)(uint64_t* value, DBImpl* db,
                                    Version* version);
+  bool (InternalStats::*handle_map)(
+      std::map<std::string, double>* compaction_stats);
 };

 extern const DBPropertyInfo* GetPropertyInfo(const Slice& property);

 #ifndef ROCKSDB_LITE
+enum class LevelStatType {
+  INVALID = 0,
+  NUM_FILES,
+  COMPACTED_FILES,
+  SIZE_MB,
+  SCORE,
+  READ_GB,
+  RN_GB,
+  RNP1_GB,
+  WRITE_GB,
+  W_NEW_GB,
+  MOVED_GB,
+  WRITE_AMP,
+  READ_MBPS,
+  WRITE_MBPS,
+  COMP_SEC,
+  COMP_COUNT,
+  AVG_SEC,
+  KEY_IN,
+  KEY_DROP,
+  TOTAL  // total number of types
+};
+
+struct LevelStat {
+  // This what will be L?.property_name in the flat map returned to the user
+  std::string property_name;
+  // This will be what we will print in the header in the cli
+  std::string header_name;
+};
+
 class InternalStats {
 public:
+  static const std::map<LevelStatType, LevelStat> compaction_level_stats;
+
  enum InternalCFStatsType {
    LEVEL0_SLOWDOWN_TOTAL,
    LEVEL0_SLOWDOWN_WITH_COMPACTION,
@@ -60,6 +95,9 @@ class InternalStats {
    WRITE_STALLS_ENUM_MAX,
    BYTES_FLUSHED,
    BYTES_INGESTED_ADD_FILE,
+    INGESTED_NUM_FILES_TOTAL,
+    INGESTED_LEVEL0_NUM_FILES_TOTAL,
+    INGESTED_NUM_KEYS_TOTAL,
    INTERNAL_CF_STATS_ENUM_MAX,
  };

@@ -218,6 +256,10 @@ class InternalStats {
  bool GetStringProperty(const DBPropertyInfo& property_info,
                         const Slice& property, std::string* value);

+  bool GetMapProperty(const DBPropertyInfo& property_info,
+                      const Slice& property,
+                      std::map<std::string, double>* value);
+
  bool GetIntProperty(const DBPropertyInfo& property_info, uint64_t* value,
                      DBImpl* db);

@@ -230,6 +272,10 @@ class InternalStats {

 private:
  void DumpDBStats(std::string* value);
+  void DumpCFMapStats(std::map<std::string, double>* cf_stats);
+  int DumpCFMapStats(
+      std::map<int, std::map<LevelStatType, double>>* level_stats,
+      CompactionStats* compaction_stats_sum);
  void DumpCFStats(std::string* value);

  // Per-DB stats
@@ -246,7 +292,6 @@ class InternalStats {
    // ColumnFamily-level stats
    CompactionStats comp_stats;
    uint64_t ingest_bytes_flush;      // Bytes written to L0 (Flush)
-    uint64_t ingest_bytes_add_file;   // Bytes written to L0 (AddFile)
    uint64_t stall_count;             // Stall count
    // Stats from compaction jobs - bytes written, bytes read, duration.
    uint64_t compact_bytes_write;
@@ -254,15 +299,24 @@ class InternalStats {
    uint64_t compact_micros;
    double seconds_up;

+    // AddFile specific stats
+    uint64_t ingest_bytes_addfile;     // Total Bytes ingested
+    uint64_t ingest_files_addfile;     // Total number of files ingested
+    uint64_t ingest_l0_files_addfile;  // Total number of files ingested to L0
+    uint64_t ingest_keys_addfile;      // Total number of keys ingested
+
    CFStatsSnapshot()
        : comp_stats(0),
          ingest_bytes_flush(0),
-          ingest_bytes_add_file(0),
          stall_count(0),
          compact_bytes_write(0),
          compact_bytes_read(0),
          compact_micros(0),
-          seconds_up(0) {}
+          seconds_up(0),
+          ingest_bytes_addfile(0),
+          ingest_files_addfile(0),
+          ingest_l0_files_addfile(0),
+          ingest_keys_addfile(0) {}
  } cf_stats_snapshot_;

  struct DBStatsSnapshot {
@@ -302,6 +356,7 @@ class InternalStats {
  bool HandleCompressionRatioAtLevelPrefix(std::string* value, Slice suffix);
  bool HandleLevelStats(std::string* value, Slice suffix);
  bool HandleStats(std::string* value, Slice suffix);
+  bool HandleCFMapStats(std::map<std::string, double>* compaction_stats);
  bool HandleCFStats(std::string* value, Slice suffix);
  bool HandleDBStats(std::string* value, Slice suffix);
  bool HandleSsTables(std::string* value, Slice suffix);
@@ -346,6 +401,7 @@ class InternalStats {
                                     Version* version);
  bool HandleEstimateLiveDataSize(uint64_t* value, DBImpl* db,
                                  Version* version);
+  bool HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db, Version* version);

  // Total number of background errors encountered. Every time a flush task
  // or compaction task fails, this counter is incremented. The failure can
@@ -376,6 +432,9 @@ class InternalStats {
    WRITE_STALLS_ENUM_MAX,
    BYTES_FLUSHED,
    BYTES_INGESTED_ADD_FILE,
+    INGESTED_NUM_FILES_TOTAL,
+    INGESTED_LEVEL0_NUM_FILES_TOTAL,
+    INGESTED_NUM_KEYS_TOTAL,
    INTERNAL_CF_STATS_ENUM_MAX,
  };

@@ -434,6 +493,12 @@ class InternalStats {
    return false;
  }

+  bool GetMapProperty(const DBPropertyInfo& property_info,
+                      const Slice& property,
+                      std::map<std::string, double>* value) {
+    return false;
+  }
+
  bool GetIntProperty(const DBPropertyInfo& property_info, uint64_t* value,
                      DBImpl* db) const {
    return false;
@@ -55,6 +55,10 @@ struct JobContext {
  // a list of log files that we need to delete
  std::vector<uint64_t> log_delete_files;

+  // a list of log files that we need to preserve during full purge since they
+  // will be reused later
+  std::vector<uint64_t> log_recycle_files;
+
  // a list of manifest files that we need to delete
  std::vector<std::string> manifest_delete_files;

@@ -754,8 +754,43 @@ TEST_F(EventListenerTest, MemTableSealedListenerTest) {
    ASSERT_OK(Flush());
  }
 }
-} // namespace rocksdb

+class ColumnFamilyHandleDeletionStartedListener : public EventListener {
+ private:
+  std::vector<std::string> cfs_;
+  int counter;
+
+ public:
+  explicit ColumnFamilyHandleDeletionStartedListener(
+      const std::vector<std::string>& cfs)
+      : cfs_(cfs), counter(0) {
+    cfs_.insert(cfs_.begin(), kDefaultColumnFamilyName);
+  }
+  void OnColumnFamilyHandleDeletionStarted(
+      ColumnFamilyHandle* handle) override {
+    ASSERT_EQ(cfs_[handle->GetID()], handle->GetName());
+    counter++;
+  }
+  int getCounter() { return counter; }
+};
+
+TEST_F(EventListenerTest, ColumnFamilyHandleDeletionStartedListenerTest) {
+  std::vector<std::string> cfs{"pikachu", "eevee", "Mewtwo"};
+  auto listener =
+      std::make_shared<ColumnFamilyHandleDeletionStartedListener>(cfs);
+  Options options;
+  options.create_if_missing = true;
+  options.listeners.push_back(listener);
+  CreateAndReopenWithCF(cfs, options);
+  ASSERT_EQ(handles_.size(), 4);
+  delete handles_[3];
+  delete handles_[2];
+  delete handles_[1];
+  handles_.resize(1);
+  ASSERT_EQ(listener->getCounter(), 3);
+}
+
+}  // namespace rocksdb

 #endif  // ROCKSDB_LITE

@@ -122,6 +122,16 @@ void ManagedIterator::Seek(const Slice& user_key) {
  SeekInternal(user_key, false);
 }

+void ManagedIterator::SeekForPrev(const Slice& user_key) {
+  MILock l(&in_use_, this);
+  if (NeedToRebuild()) {
+    RebuildIterator();
+  }
+  assert(mutable_iter_ != nullptr);
+  mutable_iter_->SeekForPrev(user_key);
+  UpdateCurrent();
+}
+
 void ManagedIterator::SeekInternal(const Slice& user_key, bool seek_to_first) {
  if (NeedToRebuild()) {
    RebuildIterator();
@@ -43,6 +43,7 @@ class ManagedIterator : public Iterator {
  virtual bool Valid() const override;
  void SeekToFirst() override;
  virtual void Seek(const Slice& target) override;
+  virtual void SeekForPrev(const Slice& target) override;
  virtual void Next() override;
  virtual Slice key() const override;
  virtual Slice value() const override;
@@ -17,6 +17,7 @@
 #include "db/merge_context.h"
 #include "db/merge_helper.h"
 #include "db/pinned_iterators_manager.h"
+#include "port/port.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -24,9 +25,12 @@
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/write_buffer_manager.h"
 #include "table/internal_iterator.h"
+#include "table/iterator_wrapper.h"
 #include "table/merger.h"
 #include "util/arena.h"
+#include "util/autovector.h"
 #include "util/coding.h"
+#include "util/memory_usage.h"
 #include "util/murmurhash.h"
 #include "util/mutexlock.h"
 #include "util/perf_context_imp.h"
@@ -68,6 +72,10 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
      table_(ioptions.memtable_factory->CreateMemTableRep(
          comparator_, &allocator_, ioptions.prefix_extractor,
          ioptions.info_log)),
+      range_del_table_(SkipListFactory().CreateMemTableRep(
+          comparator_, &allocator_, nullptr /* transform */,
+          ioptions.info_log)),
+      is_range_del_table_empty_(true),
      data_size_(0),
      num_entries_(0),
      num_deletes_(0),
@@ -83,7 +91,9 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
                 : 0),
      prefix_extractor_(ioptions.prefix_extractor),
      flush_state_(FLUSH_NOT_REQUESTED),
-      env_(ioptions.env) {
+      env_(ioptions.env),
+      insert_with_hint_prefix_extractor_(
+          ioptions.memtable_insert_with_hint_prefix_extractor) {
  UpdateFlushState();
  // something went wrong if we need to flush before inserting anything
  assert(!ShouldScheduleFlush());
@@ -99,16 +109,21 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
 MemTable::~MemTable() { assert(refs_ == 0); }

 size_t MemTable::ApproximateMemoryUsage() {
-  size_t arena_usage = arena_.ApproximateMemoryUsage();
-  size_t table_usage = table_->ApproximateMemoryUsage();
-  // let MAX_USAGE =  std::numeric_limits<size_t>::max()
-  // then if arena_usage + total_usage >= MAX_USAGE, return MAX_USAGE.
-  // the following variation is to avoid numeric overflow.
-  if (arena_usage >= std::numeric_limits<size_t>::max() - table_usage) {
-    return std::numeric_limits<size_t>::max();
+  autovector<size_t> usages = {arena_.ApproximateMemoryUsage(),
+                               table_->ApproximateMemoryUsage(),
+                               range_del_table_->ApproximateMemoryUsage(),
+                               rocksdb::ApproximateMemoryUsage(insert_hints_)};
+  size_t total_usage = 0;
+  for (size_t usage : usages) {
+    // If usage + total_usage >= kMaxSizet, return kMaxSizet.
+    // the following variation is to avoid numeric overflow.
+    if (usage >= port::kMaxSizet - total_usage) {
+      return port::kMaxSizet;
+    }
+    total_usage += usage;
  }
  // otherwise, return the actual usage
-  return arena_usage + table_usage;
+  return total_usage;
 }

 bool MemTable::ShouldFlushNow() const {
@@ -122,8 +137,9 @@ bool MemTable::ShouldFlushNow() const {

  // If arena still have room for new block allocation, we can safely say it
  // shouldn't flush.
-  auto allocated_memory =
-      table_->ApproximateMemoryUsage() + arena_.MemoryAllocatedBytes();
+  auto allocated_memory = table_->ApproximateMemoryUsage() +
+                          range_del_table_->ApproximateMemoryUsage() +
+                          arena_.MemoryAllocatedBytes();

  // if we can still allocate one more block without exceeding the
  // over-allocation ratio, then we should not flush.
@@ -219,13 +235,16 @@ const char* EncodeKey(std::string* scratch, const Slice& target) {
 class MemTableIterator : public InternalIterator {
 public:
  MemTableIterator(const MemTable& mem, const ReadOptions& read_options,
-                   Arena* arena)
+                   Arena* arena, bool use_range_del_table = false)
      : bloom_(nullptr),
        prefix_extractor_(mem.prefix_extractor_),
+        comparator_(mem.comparator_),
        valid_(false),
        arena_mode_(arena != nullptr),
        value_pinned_(!mem.GetMemTableOptions()->inplace_update_support) {
-    if (prefix_extractor_ != nullptr && !read_options.total_order_seek) {
+    if (use_range_del_table) {
+      iter_ = mem.range_del_table_->GetIterator(arena);
+    } else if (prefix_extractor_ != nullptr && !read_options.total_order_seek) {
      bloom_ = mem.prefix_bloom_.get();
      iter_ = mem.table_->GetDynamicPrefixIterator(arena);
    } else {
@@ -272,6 +291,28 @@ class MemTableIterator : public InternalIterator {
    iter_->Seek(k, nullptr);
    valid_ = iter_->Valid();
  }
+  virtual void SeekForPrev(const Slice& k) override {
+    PERF_TIMER_GUARD(seek_on_memtable_time);
+    PERF_COUNTER_ADD(seek_on_memtable_count, 1);
+    if (bloom_ != nullptr) {
+      if (!bloom_->MayContain(
+              prefix_extractor_->Transform(ExtractUserKey(k)))) {
+        PERF_COUNTER_ADD(bloom_memtable_miss_count, 1);
+        valid_ = false;
+        return;
+      } else {
+        PERF_COUNTER_ADD(bloom_memtable_hit_count, 1);
+      }
+    }
+    iter_->Seek(k, nullptr);
+    valid_ = iter_->Valid();
+    if (!Valid()) {
+      SeekToLast();
+    }
+    while (Valid() && comparator_.comparator.Compare(k, key()) < 0) {
+      Prev();
+    }
+  }
  virtual void SeekToFirst() override {
    iter_->SeekToFirst();
    valid_ = iter_->Valid();
@@ -281,11 +322,13 @@ class MemTableIterator : public InternalIterator {
    valid_ = iter_->Valid();
  }
  virtual void Next() override {
+    PERF_COUNTER_ADD(next_on_memtable_count, 1);
    assert(Valid());
    iter_->Next();
    valid_ = iter_->Valid();
  }
  virtual void Prev() override {
+    PERF_COUNTER_ADD(prev_on_memtable_count, 1);
    assert(Valid());
    iter_->Prev();
    valid_ = iter_->Valid();
@@ -315,6 +358,7 @@ class MemTableIterator : public InternalIterator {
 private:
  DynamicBloom* bloom_;
  const SliceTransform* const prefix_extractor_;
+  const MemTable::KeyComparator comparator_;
  MemTableRep::Iterator* iter_;
  bool valid_;
  bool arena_mode_;
@@ -332,6 +376,15 @@ InternalIterator* MemTable::NewIterator(const ReadOptions& read_options,
  return new (mem) MemTableIterator(*this, read_options, arena);
 }

+InternalIterator* MemTable::NewRangeTombstoneIterator(
+    const ReadOptions& read_options) {
+  if (read_options.ignore_range_deletions || is_range_del_table_empty_) {
+    return nullptr;
+  }
+  return new MemTableIterator(*this, read_options, nullptr /* arena */,
+                              true /* use_range_del_table */);
+}
+
 port::RWMutex* MemTable::GetLock(const Slice& key) {
  static murmur_hash hash;
  return &locks_[hash(key) % locks_.size()];
@@ -340,6 +393,7 @@ port::RWMutex* MemTable::GetLock(const Slice& key) {
 uint64_t MemTable::ApproximateSize(const Slice& start_ikey,
                                   const Slice& end_ikey) {
  uint64_t entry_count = table_->ApproximateNumEntries(start_ikey, end_ikey);
+  entry_count += range_del_table_->ApproximateNumEntries(start_ikey, end_ikey);
  if (entry_count == 0) {
    return 0;
  }
@@ -348,9 +402,9 @@ uint64_t MemTable::ApproximateSize(const Slice& start_ikey,
    return 0;
  }
  if (entry_count > n) {
-    // table_->ApproximateNumEntries() is just an estimate so it can be larger
-    // than actual entries we have. Cap it to entries we have to limit the
-    // inaccuracy.
+    // (range_del_)table_->ApproximateNumEntries() is just an estimate so it can
+    // be larger than actual entries we have. Cap it to entries we have to limit
+    // the inaccuracy.
    entry_count = n;
  }
  uint64_t data_size = data_size_.load(std::memory_order_relaxed);
@@ -373,10 +427,13 @@ void MemTable::Add(SequenceNumber s, ValueType type,
                               internal_key_size + VarintLength(val_size) +
                               val_size;
  char* buf = nullptr;
-  KeyHandle handle = table_->Allocate(encoded_len, &buf);
+  std::unique_ptr<MemTableRep>& table =
+      type == kTypeRangeDeletion ? range_del_table_ : table_;
+  KeyHandle handle = table->Allocate(encoded_len, &buf);

  char* p = EncodeVarint32(buf, internal_key_size);
  memcpy(p, key.data(), key_size);
+  Slice key_slice(p, key_size);
  p += key_size;
  uint64_t packed = PackSequenceAndType(s, type);
  EncodeFixed64(p, packed);
@@ -385,7 +442,14 @@ void MemTable::Add(SequenceNumber s, ValueType type,
  memcpy(p, value.data(), val_size);
  assert((unsigned)(p + val_size - buf) == (unsigned)encoded_len);
  if (!allow_concurrent) {
-    table_->Insert(handle);
+    // Extract prefix for insert with hint.
+    if (insert_with_hint_prefix_extractor_ != nullptr &&
+        insert_with_hint_prefix_extractor_->InDomain(key_slice)) {
+      Slice prefix = insert_with_hint_prefix_extractor_->Transform(key_slice);
+      table->InsertWithHint(handle, &insert_hints_[prefix]);
+    } else {
+      table->Insert(handle);
+    }

    // this is a bit ugly, but is the way to avoid locked instructions
    // when incrementing an atomic
@@ -417,7 +481,7 @@ void MemTable::Add(SequenceNumber s, ValueType type,
    assert(post_process_info == nullptr);
    UpdateFlushState();
  } else {
-    table_->InsertConcurrently(handle);
+    table->InsertConcurrently(handle);

    assert(post_process_info != nullptr);
    post_process_info->num_entries++;
@@ -443,6 +507,9 @@ void MemTable::Add(SequenceNumber s, ValueType type,
        !first_seqno_.compare_exchange_weak(cur_earliest_seqno, s)) {
    }
  }
+  if (is_range_del_table_empty_ && type == kTypeRangeDeletion) {
+    is_range_del_table_empty_ = false;
+  }
 }

 // Callback from MemTable::Get()
@@ -458,6 +525,7 @@ struct Saver {
  const MergeOperator* merge_operator;
  // the merge operations encountered;
  MergeContext* merge_context;
+  RangeDelAggregator* range_del_agg;
  MemTable* mem;
  Logger* logger;
  Statistics* statistics;
@@ -469,9 +537,10 @@ struct Saver {
 static bool SaveValue(void* arg, const char* entry) {
  Saver* s = reinterpret_cast<Saver*>(arg);
  MergeContext* merge_context = s->merge_context;
+  RangeDelAggregator* range_del_agg = s->range_del_agg;
  const MergeOperator* merge_operator = s->merge_operator;

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

  // entry format is:
  //    klength  varint32
@@ -491,6 +560,10 @@ static bool SaveValue(void* arg, const char* entry) {
    ValueType type;
    UnPackSequenceAndType(tag, &s->seq, &type);

+    if ((type == kTypeValue || type == kTypeDeletion) &&
+        range_del_agg->ShouldDelete(Slice(key_ptr, key_length))) {
+      type = kTypeRangeDeletion;
+    }
    switch (type) {
      case kTypeValue: {
        if (s->inplace_update_support) {
@@ -513,7 +586,8 @@ static bool SaveValue(void* arg, const char* entry) {
        return false;
      }
      case kTypeDeletion:
-      case kTypeSingleDeletion: {
+      case kTypeSingleDeletion:
+      case kTypeRangeDeletion: {
        if (*(s->merge_in_progress)) {
          *(s->status) = MergeHelper::TimedFullMerge(
              merge_operator, s->key->user_key(), nullptr,
@@ -553,7 +627,9 @@ static bool SaveValue(void* arg, const char* entry) {
 }

 bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
-                   MergeContext* merge_context, SequenceNumber* seq) {
+                   MergeContext* merge_context,
+                   RangeDelAggregator* range_del_agg, SequenceNumber* seq,
+                   const ReadOptions& read_opts) {
  // The sequence number is updated synchronously in version_set.h
  if (IsEmpty()) {
    // Avoiding recording stats for speed.
@@ -576,6 +652,13 @@ bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
    if (prefix_bloom_) {
      PERF_COUNTER_ADD(bloom_memtable_hit_count, 1);
    }
+    std::unique_ptr<InternalIterator> range_del_iter(
+        NewRangeTombstoneIterator(read_opts));
+    Status status = range_del_agg->AddTombstones(std::move(range_del_iter));
+    if (!status.ok()) {
+      *s = status;
+      return false;
+    }
    Saver saver;
    saver.status = s;
    saver.found_final_value = &found_final_value;
@@ -585,6 +668,7 @@ bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
    saver.seq = kMaxSequenceNumber;
    saver.mem = this;
    saver.merge_context = merge_context;
+    saver.range_del_agg = range_del_agg;
    saver.merge_operator = moptions_.merge_operator;
    saver.logger = moptions_.info_log;
    saver.inplace_update_support = moptions_.inplace_update_support;
@@ -13,9 +13,11 @@
 #include <functional>
 #include <memory>
 #include <string>
+#include <unordered_map>
 #include <vector>
 #include "db/dbformat.h"
 #include "db/memtable_allocator.h"
+#include "db/range_del_aggregator.h"
 #include "db/skiplist.h"
 #include "db/version_edit.h"
 #include "rocksdb/db.h"
@@ -24,6 +26,7 @@
 #include "util/cf_options.h"
 #include "util/concurrent_arena.h"
 #include "util/dynamic_bloom.h"
+#include "util/hash.h"
 #include "util/instrumented_mutex.h"

 namespace rocksdb {
@@ -158,6 +161,8 @@ class MemTable {
  //        those allocated in arena.
  InternalIterator* NewIterator(const ReadOptions& read_options, Arena* arena);

+  InternalIterator* NewRangeTombstoneIterator(const ReadOptions& read_options);
+
  // Add an entry into memtable that maps key to value at the
  // specified sequence number and with the specified type.
  // Typically value will be empty if type==kTypeDeletion.
@@ -182,12 +187,14 @@ class MemTable {
  // On success, *s may be set to OK, NotFound, or MergeInProgress.  Any other
  // status returned indicates a corruption or other unexpected error.
  bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext* merge_context, SequenceNumber* seq);
+           MergeContext* merge_context, RangeDelAggregator* range_del_agg,
+           SequenceNumber* seq, const ReadOptions& read_opts);

  bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext* merge_context) {
+           MergeContext* merge_context, RangeDelAggregator* range_del_agg,
+           const ReadOptions& read_opts) {
    SequenceNumber seq;
-    return Get(key, value, s, merge_context, &seq);
+    return Get(key, value, s, merge_context, range_del_agg, &seq, read_opts);
  }

  // Attempts to update the new_value inplace, else does normal Add
@@ -344,6 +351,8 @@ class MemTable {
  ConcurrentArena arena_;
  MemTableAllocator allocator_;
  unique_ptr<MemTableRep> table_;
+  unique_ptr<MemTableRep> range_del_table_;
+  bool is_range_del_table_empty_;

  // Total data size of all data inserted
  std::atomic<uint64_t> data_size_;
@@ -383,6 +392,12 @@ class MemTable {

  Env* env_;

+  // Extract sequential insert prefixes.
+  const SliceTransform* insert_with_hint_prefix_extractor_;
+
+  // Insert hints for each prefix.
+  std::unordered_map<Slice, void*, SliceHasher> insert_hints_;
+
  // Returns a heuristic flush decision
  bool ShouldFlushNow() const;

@@ -102,27 +102,36 @@ int MemTableList::NumFlushed() const {
 // Operands stores the list of merge operations to apply, so far.
 bool MemTableListVersion::Get(const LookupKey& key, std::string* value,
                              Status* s, MergeContext* merge_context,
-                              SequenceNumber* seq) {
-  return GetFromList(&memlist_, key, value, s, merge_context, seq);
+                              RangeDelAggregator* range_del_agg,
+                              SequenceNumber* seq,
+                              const ReadOptions& read_opts) {
+  return GetFromList(&memlist_, key, value, s, merge_context, range_del_agg,
+                     seq, read_opts);
 }

 bool MemTableListVersion::GetFromHistory(const LookupKey& key,
                                         std::string* value, Status* s,
                                         MergeContext* merge_context,
-                                         SequenceNumber* seq) {
-  return GetFromList(&memlist_history_, key, value, s, merge_context, seq);
+                                         RangeDelAggregator* range_del_agg,
+                                         SequenceNumber* seq,
+                                         const ReadOptions& read_opts) {
+  return GetFromList(&memlist_history_, key, value, s, merge_context,
+                     range_del_agg, seq, read_opts);
 }

 bool MemTableListVersion::GetFromList(std::list<MemTable*>* list,
                                      const LookupKey& key, std::string* value,
                                      Status* s, MergeContext* merge_context,
-                                      SequenceNumber* seq) {
+                                      RangeDelAggregator* range_del_agg,
+                                      SequenceNumber* seq,
+                                      const ReadOptions& read_opts) {
  *seq = kMaxSequenceNumber;

  for (auto& memtable : *list) {
    SequenceNumber current_seq = kMaxSequenceNumber;

-    bool done = memtable->Get(key, value, s, merge_context, &current_seq);
+    bool done = memtable->Get(key, value, s, merge_context, range_del_agg,
+                              &current_seq, read_opts);
    if (*seq == kMaxSequenceNumber) {
      // Store the most recent sequence number of any operation on this key.
      // Since we only care about the most recent change, we only need to
@@ -135,10 +144,28 @@ bool MemTableListVersion::GetFromList(std::list<MemTable*>* list,
      assert(*seq != kMaxSequenceNumber);
      return true;
    }
+    if (!done && !s->ok() && !s->IsMergeInProgress() && !s->IsNotFound()) {
+      return false;
+    }
  }
  return false;
 }

+Status MemTableListVersion::AddRangeTombstoneIterators(
+    const ReadOptions& read_opts, Arena* arena,
+    RangeDelAggregator* range_del_agg) {
+  assert(range_del_agg != nullptr);
+  for (auto& m : memlist_) {
+    std::unique_ptr<InternalIterator> range_del_iter(
+        m->NewRangeTombstoneIterator(read_opts));
+    Status s = range_del_agg->AddTombstones(std::move(range_del_iter));
+    if (!s.ok()) {
+      return s;
+    }
+  }
+  return Status::OK();
+}
+
 void MemTableListVersion::AddIterators(
    const ReadOptions& options, std::vector<InternalIterator*>* iterator_list,
    Arena* arena) {
@@ -14,6 +14,7 @@
 #include "db/dbformat.h"
 #include "db/filename.h"
 #include "db/memtable.h"
+#include "db/range_del_aggregator.h"
 #include "db/skiplist.h"
 #include "rocksdb/db.h"
 #include "rocksdb/iterator.h"
@@ -53,12 +54,14 @@ class MemTableListVersion {
  // will be stored in *seq on success (regardless of whether true/false is
  // returned).  Otherwise, *seq will be set to kMaxSequenceNumber.
  bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext* merge_context, SequenceNumber* seq);
+           MergeContext* merge_context, RangeDelAggregator* range_del_agg,
+           SequenceNumber* seq, const ReadOptions& read_opts);

  bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext* merge_context) {
+           MergeContext* merge_context, RangeDelAggregator* range_del_agg,
+           const ReadOptions& read_opts) {
    SequenceNumber seq;
-    return Get(key, value, s, merge_context, &seq);
+    return Get(key, value, s, merge_context, range_del_agg, &seq, read_opts);
  }

  // Similar to Get(), but searches the Memtable history of memtables that
@@ -66,13 +69,21 @@ class MemTableListVersion {
  // queries (such as Transaction validation) as the history may contain
  // writes that are also present in the SST files.
  bool GetFromHistory(const LookupKey& key, std::string* value, Status* s,
-                      MergeContext* merge_context, SequenceNumber* seq);
+                      MergeContext* merge_context,
+                      RangeDelAggregator* range_del_agg, SequenceNumber* seq,
+                      const ReadOptions& read_opts);
  bool GetFromHistory(const LookupKey& key, std::string* value, Status* s,
-                      MergeContext* merge_context) {
+                      MergeContext* merge_context,
+                      RangeDelAggregator* range_del_agg,
+                      const ReadOptions& read_opts) {
    SequenceNumber seq;
-    return GetFromHistory(key, value, s, merge_context, &seq);
+    return GetFromHistory(key, value, s, merge_context, range_del_agg, &seq,
+                          read_opts);
  }

+  Status AddRangeTombstoneIterators(const ReadOptions& read_opts, Arena* arena,
+                                    RangeDelAggregator* range_del_agg);
+
  void AddIterators(const ReadOptions& options,
                    std::vector<InternalIterator*>* iterator_list,
                    Arena* arena);
@@ -102,7 +113,8 @@ class MemTableListVersion {

  bool GetFromList(std::list<MemTable*>* list, const LookupKey& key,
                   std::string* value, Status* s, MergeContext* merge_context,
-                   SequenceNumber* seq);
+                   RangeDelAggregator* range_del_agg, SequenceNumber* seq,
+                   const ReadOptions& read_opts);

  void AddMemTable(MemTable* m);

@@ -209,6 +221,8 @@ class MemTableList {
  // PickMemtablesToFlush() is called.
  void FlushRequested() { flush_requested_ = true; }

+  bool HasFlushRequested() { return flush_requested_; }
+
  // Copying allowed
  // MemTableList(const MemTableList&);
  // void operator=(const MemTableList&);