Fix deadlock when calling getMergedHistogram

Summary: When calling StatisticsImpl::HistogramInfo::getMergedHistogram(), if there is a dying thread, which is calling ThreadLocalPtr::StaticMeta::OnThreadExit() to merge its thread values to HistogramInfo, deadlock will occur. Because the former try to hold merge_lock then ThreadMeta::mutex_, but the later try to hold ThreadMeta::mutex_ then merge_lock. In short, the locking order isn't the same. This patch addressed this issue by releasing merge_lock before folding thread values. Closes https://github.com/facebook/rocksdb/pull/1552 Differential Revision: D4211942 Pulled By: ajkr fbshipit-source-id: ef89bcb
Remove Ticker::SEQUENCE_NUMBER
2026-11-06 19:12:36 -05:00 · 2016-12-09 12:59:51 -08:00 · 2016-12-09 12:55:55 -08:00 · 2016-12-08 19:06:47 -08:00 · 2016-12-08 19:03:02 -08:00 · 2016-12-01 13:29:14 -08:00
318 changed files with 13663 additions and 4967 deletions
@@ -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 +28,108 @@
 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()
+  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)
+file(GENERATE OUTPUT ${BUILD_VERSION_CC} CONTENT
+  "#include \"build_version.h\"
+const char* rocksdb_build_git_sha = \"rocksdb_build_git_sha:${GIT_SHA}\";
+const char* rocksdb_build_git_datetime = \"rocksdb_build_git_datetime:${GIT_DATE_TIME}\";
+const char* rocksdb_build_compile_date = __DATE__;
+")
+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 /WX /wd4127 /wd4800 /wd4996 /wd4351")
+else()
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wextra -Wall -Werror")
+  set(CMAKE_CXX_FLAGS "${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} -march=native -fno-omit-frame-pointer -momit-leaf-frame-pointer")
+endif()

-add_custom_target(GenerateBuildVersion DEPENDS ${BUILD_VERSION_CC})
+option(WITH_ASAN "build with ASAN" OFF)
+if(WITH_ASAN)
+  add_definitions(-DROCKSDB_TSAN_RUN)
+  set(CMAKE_EXE_LINKER_FLAGS "${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 "${CMAKE_CXX_FLAGS} /Zi /nologo  /EHsc /GS /Gd /GR /GF /fp:precise /Zc:wchar_t /Zc:forScope /errorReport:queue")
+option(WITH_TSAN "build with TSAN" OFF)
+if(WITH_TSAN)
+  set(CMAKE_EXE_LINKER_FLAGS "${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()

-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W3 /WX /wd4127 /wd4800 /wd4996 /wd4351")
+option(WITH_UBSAN "build with UBSAN" OFF)
+if(WITH_UBSAN)
+  set(CMAKE_EXE_LINKER_FLAGS "${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 +140,102 @@ 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 +252,11 @@ 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 +272,7 @@ 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 +293,6 @@ 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 +326,6 @@ 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 +333,7 @@ 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 +371,6 @@ 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
@@ -258,7 +391,6 @@ 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/memory/memory_util.cc
@@ -292,56 +424,56 @@ 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
@@ -408,7 +540,6 @@ 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
@@ -460,20 +591,42 @@ set(TESTS
        utilities/write_batch_with_index/write_batch_with_index_test.cc
        utilities/column_aware_encoding_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 +637,23 @@ 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 +666,6 @@ 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})
@@ -1,5 +1,18 @@
 # Rocksdb Change Log
-## Unreleased
+## 4.13.4
+### Public API Change
+* Removed flashcache support.
+
+## 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 +22,7 @@
 ### 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
@@ -393,9 +393,10 @@ 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 \
@@ -599,7 +600,7 @@ 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							\
@@ -820,6 +821,7 @@ 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
@@ -1385,12 +1387,15 @@ 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.
@@ -25,6 +25,7 @@ function postURL($diffID, $url) {
  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
         . '"name":"click here for sandcastle tests for D' . $diffID . '", '
         . '"link":"' . $url . '"}\' | '
+         . 'no_proxy=facebook.com,tfbnw.net,fb.com '
         . 'http_proxy=fwdproxy.any.facebook.com:8080 '
         . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
         . 'differential.updateunitresults';
@@ -40,6 +41,7 @@ function buildUpdateTestStatusCmd($diffID, $test, $status) {
  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
         . '"name":"' . $test . '", '
         . '"result":"' . $status . '"}\' | '
+         . 'no_proxy=facebook.com,tfbnw.net,fb.com '
         . 'http_proxy=fwdproxy.any.facebook.com:8080 '
         . 'https_proxy=fwdproxy.any.facebook.com:8080 arc call-conduit '
         . 'differential.updateunitresults';
@@ -106,7 +108,8 @@ function getSteps($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" => "no_proxy=facebook.com,tfbnw.net,fb.com "
+                  ."HTTPS_PROXY=fwdproxy:8080 arc --arcrc-file ~/.arcrc "
                  . "patch --nocommit --diff " . $diffID,
      "user" => "root"
    );
@@ -183,6 +186,14 @@ function getSandcastleConfig() {
  } else {
    // This is a typical `[p]arc diff` case. Fetch the values from the specific
    // configuration files.
+    for ($i = 0; $i < 50; $i++) {
+      if (file_exists(PRIMARY_TOKEN_FILE) ||
+          file_exists(SECONDARY_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));

@@ -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) ==
@@ -53,7 +54,7 @@ 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);
 }

@@ -108,19 +109,21 @@ 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);
    }

+    std::unique_ptr<RangeDelAggregator> range_del_agg;
+    range_del_agg.reset(new RangeDelAggregator(internal_comparator, snapshots));
    MergeHelper merge(env, internal_comparator.user_comparator(),
                      ioptions.merge_operator, nullptr, ioptions.info_log,
                      mutable_cf_options.min_partial_merge_operands,
                      true /* internal key corruption is not ok */,
                      snapshots.empty() ? 0 : snapshots.back());

-    CompactionIterator c_iter(iter, internal_comparator.user_comparator(),
-                              &merge, kMaxSequenceNumber, &snapshots,
-                              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();
@@ -48,6 +48,7 @@ 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);

@@ -1783,11 +1783,7 @@ 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 +1799,26 @@ 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);
 }

@@ -944,6 +944,28 @@ 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);
@@ -45,6 +45,11 @@ 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 +76,7 @@ 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 +88,11 @@ 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(
@@ -136,7 +138,7 @@ Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
  return Status::OK();
 }

-ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
+ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                    const InternalKeyComparator* icmp,
                                    const ColumnFamilyOptions& src) {
  ColumnFamilyOptions result = src;
@@ -158,6 +160,10 @@ 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 +344,7 @@ 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 +353,14 @@ 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, &internal_comparator_, cf_options)),
+      ioptions_(db_options, initial_cf_options_),
+      mutable_cf_options_(initial_cf_options_),
      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)),
@@ -368,12 +374,12 @@ ColumnFamilyData::ColumnFamilyData(
  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 +411,7 @@ 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");
@@ -482,6 +488,10 @@ void ColumnFamilyData::SetDropped() {
  column_family_set_->RemoveColumnFamily(this);
 }

+ColumnFamilyOptions ColumnFamilyData::GetLatestCFOptions() const {
+  return BuildColumnFamilyOptions(initial_cf_options_, mutable_cf_options_);
+}
+
 const double kSlowdownRatio = 1.2;

 namespace {
@@ -539,14 +549,31 @@ 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

@@ -713,6 +740,13 @@ 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 +897,14 @@ 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 +976,7 @@ 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);
@@ -136,14 +136,14 @@ extern Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options);
 extern Status CheckConcurrentWritesSupported(
    const ColumnFamilyOptions& cf_options);

-extern ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
+extern ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                           const InternalKeyComparator* icmp,
                                           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 +201,6 @@ 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 +215,12 @@ 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;
+
 #ifndef ROCKSDB_LITE
  // REQUIRES: DB mutex held
  Status SetOptions(
@@ -249,6 +252,13 @@ 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 +335,8 @@ 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_;
@@ -340,7 +351,7 @@ 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_;

@@ -436,7 +447,8 @@ 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 +505,7 @@ 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
@@ -1049,6 +1052,7 @@ TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
  AssertCountLiveLogFiles(7);
  Close();
 }
+#endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // Cuckoo is not supported in lite
 TEST_F(ColumnFamilyTest, MemtableNotSupportSnapshot) {
@@ -1136,6 +1140,7 @@ TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
  Close();
 }

+#ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
  Open();
  CreateColumnFamilies({"one", "two"});
@@ -1204,6 +1209,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {

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

 #ifndef ROCKSDB_LITE
 // Sync points not supported in RocksDB Lite
@@ -2097,6 +2103,7 @@ 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 +2125,9 @@ 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 +2161,7 @@ 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"});
@@ -2178,8 +2188,8 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
            original.write_buffer_size =
                l * 4 * 1024 * 1024 + i * 1024 * 1024 + j * 1024 + k;

-            ColumnFamilyOptions result =
-                SanitizeOptions(db_options, nullptr, original);
+            ColumnFamilyOptions result = SanitizeOptions(
+                ImmutableDBOptions(db_options), nullptr, original);
            ASSERT_TRUE(result.level0_stop_writes_trigger >=
                        result.level0_slowdown_writes_trigger);
            ASSERT_TRUE(result.level0_slowdown_writes_trigger >=
@@ -2440,9 +2450,7 @@ 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;
@@ -2460,7 +2468,7 @@ 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);
@@ -2468,7 +2476,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  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(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(500);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2524,7 +2532,7 @@ 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(6, dbfull()->BGCompactionsAllowed());
+  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->TEST_set_estimated_compaction_needed_bytes(3001);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -2549,7 +2557,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  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(6, dbfull()->TEST_BGCompactionsAllowed());

  vstorage->set_l0_delay_trigger_count(101);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
@@ -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;
@@ -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;
@@ -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_;
@@ -14,8 +14,9 @@ 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,
-    const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
+    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    const Compaction* compaction, const CompactionFilter* compaction_filter,
+    LogBuffer* log_buffer)
    : input_(input),
      cmp_(cmp),
      merge_helper_(merge_helper),
@@ -23,6 +24,7 @@ CompactionIterator::CompactionIterator(
      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      env_(env),
      expect_valid_internal_key_(expect_valid_internal_key),
+      range_del_agg_(range_del_agg),
      compaction_(compaction),
      compaction_filter_(compaction_filter),
      log_buffer_(log_buffer),
@@ -36,11 +38,11 @@ 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();
  }
@@ -153,6 +155,7 @@ void CompactionIterator::NextFromInput() {
    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;
@@ -210,8 +213,9 @@ 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_) {
      // In the previous iteration we encountered a single delete that we could
@@ -320,7 +324,7 @@ 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 +393,8 @@ 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()) {
@@ -415,7 +420,15 @@ void CompactionIterator::NextFromInput() {
        pinned_iters_mgr_.ReleasePinnedData();
      }
    } else {
-      valid_ = true;
+      // 1. new user key -OR-
+      // 2. different snapshot stripe
+      bool should_delete =
+          range_del_agg_->ShouldDelete(key_, true /* for_compaction */);
+      if (should_delete) {
+        input_->Next();
+      } else {
+        valid_ = true;
+      }
    }
  }
 }
@@ -14,6 +14,7 @@
 #include "db/compaction.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"

@@ -47,6 +48,7 @@ class CompactionIterator {
                     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);
@@ -99,12 +101,13 @@ class CompactionIterator {
  const SequenceNumber earliest_write_conflict_snapshot_;
  Env* env_;
  bool expect_valid_internal_key_;
+  RangeDelAggregator* range_del_agg_;
  const Compaction* 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_;
@@ -4,6 +4,10 @@
 //  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"

@@ -11,33 +15,45 @@ namespace rocksdb {

 class CompactionIteratorTest : public testing::Test {
 public:
-  CompactionIteratorTest() : cmp_(BytewiseComparator()), snapshots_({}) {}
+  CompactionIteratorTest()
+      : cmp_(BytewiseComparator()), icmp_(cmp_), snapshots_({}) {}
+
+  void InitIterators(const std::vector<std::string>& ks,
+                     const std::vector<std::string>& vs,
+                     const std::vector<std::string>& range_del_ks,
+                     const std::vector<std::string>& range_del_vs,
+                     SequenceNumber last_sequence) {
+    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)));

-  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,
                                        nullptr, 0U, false, 0));
    iter_.reset(new test::VectorIterator(ks, vs));
    iter_->SeekToFirst();
    c_iter_.reset(new CompactionIterator(
        iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
-        kMaxSequenceNumber, Env::Default(), false));
+        kMaxSequenceNumber, Env::Default(), false, range_del_agg_.get()));
  }

+  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<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 +61,10 @@ 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 +79,43 @@ 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());
+}
+
 }  // 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 +767,12 @@ 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);
+      status = FinishCompactionOutputFile(input->status(), sub_compact,
+                                          range_del_agg.get());
      if (!status.ok()) {
        break;
      }
@@ -842,21 +849,29 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      }
    }

+    Status input_status = input->status();
+    c_iter->Next();
+
    // Close output file if it is big enough
    // TODO(aekmekji): determine if file should be closed earlier than this
    // during subcompactions (i.e. if output size, estimated by input size, is
    // 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()) {
+      const Slice* next_key = nullptr;
+      if (c_iter->Valid()) {
+        next_key = &c_iter->key();
+      }
+      status = FinishCompactionOutputFile(input_status, sub_compact,
+                                          range_del_agg.get(), next_key);
      if (sub_compact->outputs.size() == 1) {
        // Use dictionary from first output file for compression of subsequent
        // files.
        sub_compact->compression_dict = std::move(compression_dict);
      }
    }
-    c_iter->Next();
  }

  sub_compact->num_input_records = c_iter_stats.num_input_records;
@@ -883,8 +898,13 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    status = Status::ShutdownInProgress(
        "Database shutdown or Column family drop during compaction");
  }
+  if (status.ok() && sub_compact->builder == nullptr &&
+      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
+    status = OpenCompactionOutputFile(sub_compact);
+  }
  if (status.ok() && sub_compact->builder != nullptr) {
-    status = FinishCompactionOutputFile(input->status(), sub_compact);
+    status = FinishCompactionOutputFile(input->status(), sub_compact,
+                                        range_del_agg.get());
  }
  if (status.ok()) {
    status = input->status();
@@ -935,7 +955,9 @@ void CompactionJob::RecordDroppedKeys(
 }

 Status CompactionJob::FinishCompactionOutputFile(
-    const Status& input_status, SubcompactionState* sub_compact) {
+    const Status& input_status, SubcompactionState* sub_compact,
+    RangeDelAggregator* range_del_agg,
+    const Slice* next_table_min_key /* = nullptr */) {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_SYNC_FILE);
  assert(sub_compact != nullptr);
@@ -950,6 +972,16 @@ Status CompactionJob::FinishCompactionOutputFile(
  // Check for iterator errors
  Status s = input_status;
  auto meta = &sub_compact->current_output()->meta;
+  if (s.ok()) {
+    // For the first output table, include range tombstones before the min key
+    // boundary. For subsequent output tables, this is unnecessary because we
+    // extend each file's max key boundary up until the next file's min key when
+    // range tombstones fall in the gap.
+    range_del_agg->AddToBuilder(
+        sub_compact->builder.get(),
+        sub_compact->outputs.size() == 1 /* extend_before_min_key */,
+        next_table_min_key, meta, bottommost_level_);
+  }
  const uint64_t current_entries = sub_compact->builder->NumEntries();
  meta->marked_for_compaction = sub_compact->builder->NeedCompact();
  if (s.ok()) {
@@ -963,7 +995,7 @@ 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);
  }
@@ -1136,7 +1168,9 @@ 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;
@@ -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,
@@ -94,7 +97,9 @@ class CompactionJob {
  void ProcessKeyValueCompaction(SubcompactionState* sub_compact);

  Status FinishCompactionOutputFile(const Status& input_status,
-                                    SubcompactionState* sub_compact);
+                                    SubcompactionState* sub_compact,
+                                    RangeDelAggregator* range_del_agg = nullptr,
+                                    const Slice* next_table_min_key = nullptr);
  Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options);
  void RecordCompactionIOStats();
  Status OpenCompactionOutputFile(SubcompactionState* sub_compact);
@@ -120,7 +125,7 @@ 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;
 }

@@ -506,16 +570,24 @@ 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 +676,33 @@ 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 +932,30 @@ 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 +1039,8 @@ 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 +1113,23 @@ 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 +1139,13 @@ 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);

@@ -1421,7 +1535,7 @@ 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 +1757,7 @@ 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 +1881,7 @@ 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 +1946,10 @@ 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_);
  }

@@ -115,11 +115,11 @@ 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_->SetFinalized();
  }
@@ -350,10 +350,10 @@ 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 +489,10 @@ 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 +516,10 @@ 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 +546,9 @@ 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 +598,7 @@ 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
@@ -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()) {
@@ -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 iterators or
+  // range tombstone iterators since they're already cached.
  ASSERT_EQ(num_table_cache_lookup, 2);
  // Create new iterator for:
  // (1) 1 for verifying flush results
@@ -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;
    }
@@ -40,6 +40,7 @@
 #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"
@@ -155,6 +156,9 @@ 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 +169,9 @@ 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 +260,11 @@ 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

@@ -360,6 +361,8 @@ 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
@@ -391,6 +394,10 @@ 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 +464,7 @@ 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 +515,9 @@ 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_;
@@ -620,7 +629,7 @@ 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
@@ -639,26 +648,28 @@ class DBImpl : public DB {
  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 +717,10 @@ 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();
+
  // table_cache_ provides its own synchronization
  std::shared_ptr<Cache> table_cache_;

@@ -717,7 +732,7 @@ 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 +743,7 @@ 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>
@@ -969,9 +985,9 @@ 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 +1084,14 @@ 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
@@ -178,5 +178,10 @@ 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;
@@ -19,8 +19,9 @@ 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() {
@@ -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
@@ -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 +121,8 @@ 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) {
    RecordTick(statistics_, NO_ITERATORS);
    prefix_extractor_ = ioptions.prefix_extractor;
    max_skip_ = max_sequential_skip_in_iterations;
@@ -199,10 +201,12 @@ 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 +259,7 @@ 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_;
@@ -265,6 +270,7 @@ 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_;
  LocalStatistics local_stats_;
@@ -293,11 +299,7 @@ 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
@@ -503,14 +505,29 @@ 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.
@@ -557,6 +574,11 @@ 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 +590,7 @@ 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;
 }
@@ -635,7 +658,8 @@ bool DBIter::FindValueForCurrentKey() {
      return false;
    case kTypeMerge:
      current_entry_is_merged_ = true;
-      if (last_not_merge_type == kTypeDeletion) {
+      if (last_not_merge_type == kTypeDeletion ||
+          last_not_merge_type == kTypeSingleDeletion) {
        MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
                                    nullptr, merge_context_.GetOperands(),
                                    &saved_value_, logger_, statistics_, env_,
@@ -792,7 +816,6 @@ 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 +836,46 @@ 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 +933,15 @@ 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 +955,16 @@ 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;
 }

@@ -931,6 +982,9 @@ 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 +1003,15 @@ 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);

@@ -31,7 +31,8 @@ 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
@@ -57,6 +58,7 @@ 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 +79,7 @@ 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");
@@ -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;
@@ -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, nullptr, 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 +224,38 @@ 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());
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -557,9 +557,9 @@ TEST_F(DBPropertiesTest, NumImmutableMemTable) {
    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
+    // "384" 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_EQ(num, "384");

    uint64_t int_num;
    uint64_t base_total_size;
@@ -699,6 +699,58 @@ 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)
@@ -2646,15 +2646,11 @@ 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.");
  }

@@ -2762,6 +2758,12 @@ 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 +2808,12 @@ 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 +2883,10 @@ 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()) {
@@ -2066,6 +2066,125 @@ 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));
+
+  // Trivial move 2 files to L2
+  ASSERT_EQ("0,0,2", FilesPerLevel());
+
+  // 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_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();
+}
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1077,7 +1077,7 @@ 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) {
  size_t total_reads = 0;

  for (auto& kv : true_data) {
@@ -1126,36 +1126,38 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
    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;
@@ -1164,34 +1166,6 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,

 #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;
-}
-
 uint64_t DBTestBase::GetNumberOfSstFilesForColumnFamily(
    DB* db, std::string column_family_name) {
  std::vector<LiveFileMetaData> metadata;
@@ -310,7 +310,18 @@ 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_;
@@ -811,12 +822,10 @@ 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);

 #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
@@ -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 +635,41 @@ 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;
+  // TODO(yiwu): fix for 2PC.
+  options.allow_2pc = false;
+  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 +687,7 @@ 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 +1006,32 @@ 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 +1192,29 @@ 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));
@@ -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;
@@ -361,6 +359,15 @@ 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 +516,28 @@ 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 +1,491 @@
+//  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);
+  }
+
+  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);
+      }
+    } else {
+      status = CopyFile(env_, path_outside_db, path_inside_db, 0);
+    }
+    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;
+  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("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();
+
+  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);
+      sv->current->AddIteratorsForLevel(ro, env_options_, &merge_iter_builder,
+                                        lvl);
+      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 +1,143 @@
+//  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;
+  // 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_; }
+
+ 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 defined(__clang__)
+__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,
@@ -294,7 +294,7 @@ 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_;
@@ -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
@@ -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;
@@ -116,6 +116,9 @@ 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();
@@ -167,6 +170,10 @@ 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.
  bool KeyIsAfterNode(const char* key, Node* n) const;
@@ -310,6 +317,18 @@ 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);
@@ -58,6 +58,8 @@ 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 +99,11 @@ 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()));
@@ -127,18 +134,22 @@ 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());
  }
 }

@@ -769,10 +769,13 @@ void InternalStats::DumpCFStats(std::string* value) {
      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 curr_ingest = flush_ingest + add_file_ingest;
+  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
  double w_amp = stats_sum.bytes_written / static_cast<double>(curr_ingest + 1);
  uint64_t total_stall_count =
@@ -790,7 +793,7 @@ void InternalStats::DumpCFStats(std::string* value) {
  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);
@@ -804,13 +807,34 @@ 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;
@@ -881,7 +905,10 @@ 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_.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 = stats_sum;
  cf_stats_snapshot_.stall_count = total_stall_count;
 }
@@ -60,6 +60,9 @@ 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,
  };

@@ -246,7 +249,6 @@ 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 +256,24 @@ 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 {
@@ -376,6 +387,9 @@ 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,
  };

@@ -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;
@@ -68,6 +68,9 @@ 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)),
      data_size_(0),
      num_entries_(0),
      num_deletes_(0),
@@ -101,6 +104,7 @@ MemTable::~MemTable() { assert(refs_ == 0); }
 size_t MemTable::ApproximateMemoryUsage() {
  size_t arena_usage = arena_.ApproximateMemoryUsage();
  size_t table_usage = table_->ApproximateMemoryUsage();
+  table_usage += range_del_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.
@@ -122,8 +126,9 @@ 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 +224,16 @@ 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 +280,28 @@ 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();
@@ -315,6 +345,7 @@ 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 +363,14 @@ InternalIterator* MemTable::NewIterator(const ReadOptions& read_options,
  return new (mem) MemTableIterator(*this, read_options, arena);
 }

+InternalIterator* MemTable::NewRangeTombstoneIterator(
+    const ReadOptions& read_options, Arena* arena) {
+  assert(arena != nullptr);
+  auto mem = arena->AllocateAligned(sizeof(MemTableIterator));
+  return new (mem) MemTableIterator(*this, read_options, 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 +379,7 @@ 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 +388,9 @@ 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);
@@ -373,7 +413,9 @@ 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);
@@ -385,7 +427,7 @@ 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);
+    table->Insert(handle);

    // this is a bit ugly, but is the way to avoid locked instructions
    // when incrementing an atomic
@@ -417,7 +459,7 @@ 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++;
@@ -158,6 +158,9 @@ class MemTable {
  //        those allocated in arena.
  InternalIterator* NewIterator(const ReadOptions& read_options, Arena* arena);

+  InternalIterator* NewRangeTombstoneIterator(const ReadOptions& read_options,
+                                              Arena* arena);
+
  // 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.
@@ -344,6 +347,7 @@ class MemTable {
  ConcurrentArena arena_;
  MemTableAllocator allocator_;
  unique_ptr<MemTableRep> table_;
+  unique_ptr<MemTableRep> range_del_table_;

  // Total data size of all data inserted
  std::atomic<uint64_t> data_size_;
@@ -57,14 +57,16 @@ class MemTableListTest : public testing::Test {

    // Create a mock VersionSet
    DBOptions db_options;
+    ImmutableDBOptions immutable_db_options(db_options);
    EnvOptions env_options;
    shared_ptr<Cache> table_cache(NewLRUCache(50000, 16));
    WriteBufferManager write_buffer_manager(db_options.db_write_buffer_size);
    WriteController write_controller(10000000u);

    CreateDB();
-    VersionSet versions(dbname, &db_options, env_options, table_cache.get(),
-                        &write_buffer_manager, &write_controller);
+    VersionSet versions(dbname, &immutable_db_options, env_options,
+                        table_cache.get(), &write_buffer_manager,
+                        &write_controller);

    // Create mock default ColumnFamilyData
    ColumnFamilyOptions cf_options;
@@ -127,9 +129,8 @@ TEST_F(MemTableListTest, GetTest) {
  ImmutableCFOptions ioptions(options);

  WriteBufferManager wb(options.db_write_buffer_size);
-  MemTable* mem =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb,
-                   kMaxSequenceNumber);
+  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                               kMaxSequenceNumber);
  mem->Ref();

  // Write some keys to this memtable.
@@ -164,9 +165,8 @@ TEST_F(MemTableListTest, GetTest) {

  // Create another memtable and write some keys to it
  WriteBufferManager wb2(options.db_write_buffer_size);
-  MemTable* mem2 =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb2,
-                   kMaxSequenceNumber);
+  MemTable* mem2 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb2,
+                                kMaxSequenceNumber);
  mem2->Ref();

  mem2->Add(++seq, kTypeDeletion, "key1", "");
@@ -229,9 +229,8 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {
  ImmutableCFOptions ioptions(options);

  WriteBufferManager wb(options.db_write_buffer_size);
-  MemTable* mem =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb,
-                   kMaxSequenceNumber);
+  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                               kMaxSequenceNumber);
  mem->Ref();

  // Write some keys to this memtable.
@@ -272,8 +271,8 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {
  list.PickMemtablesToFlush(&to_flush);
  ASSERT_EQ(1, to_flush.size());

-  s = Mock_InstallMemtableFlushResults(
-      &list, MutableCFOptions(options, ioptions), to_flush, &to_delete);
+  s = Mock_InstallMemtableFlushResults(&list, MutableCFOptions(options),
+                                       to_flush, &to_delete);
  ASSERT_OK(s);
  ASSERT_EQ(0, list.NumNotFlushed());
  ASSERT_EQ(1, list.NumFlushed());
@@ -304,9 +303,8 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {

  // Create another memtable and write some keys to it
  WriteBufferManager wb2(options.db_write_buffer_size);
-  MemTable* mem2 =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb2,
-                   kMaxSequenceNumber);
+  MemTable* mem2 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb2,
+                                kMaxSequenceNumber);
  mem2->Ref();

  mem2->Add(++seq, kTypeDeletion, "key1", "");
@@ -321,8 +319,8 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {
  ASSERT_EQ(1, to_flush.size());

  // Flush second memtable
-  s = Mock_InstallMemtableFlushResults(
-      &list, MutableCFOptions(options, ioptions), to_flush, &to_delete);
+  s = Mock_InstallMemtableFlushResults(&list, MutableCFOptions(options),
+                                       to_flush, &to_delete);
  ASSERT_OK(s);
  ASSERT_EQ(0, list.NumNotFlushed());
  ASSERT_EQ(2, list.NumFlushed());
@@ -330,9 +328,8 @@ TEST_F(MemTableListTest, GetFromHistoryTest) {

  // Add a third memtable to push the first memtable out of the history
  WriteBufferManager wb3(options.db_write_buffer_size);
-  MemTable* mem3 =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb3,
-                   kMaxSequenceNumber);
+  MemTable* mem3 = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb3,
+                                kMaxSequenceNumber);
  mem3->Ref();
  list.Add(mem3, &to_delete);
  ASSERT_EQ(1, list.NumNotFlushed());
@@ -401,7 +398,7 @@ TEST_F(MemTableListTest, FlushPendingTest) {

  // Create some MemTables
  std::vector<MemTable*> tables;
-  MutableCFOptions mutable_cf_options(options, ioptions);
+  MutableCFOptions mutable_cf_options(options);
  for (int i = 0; i < num_tables; i++) {
    MemTable* mem = new MemTable(cmp, ioptions, mutable_cf_options, &wb,
                                 kMaxSequenceNumber);
@@ -539,8 +536,8 @@ TEST_F(MemTableListTest, FlushPendingTest) {
  ASSERT_FALSE(list.imm_flush_needed.load(std::memory_order_acquire));

  // Flush the 4 memtables that were picked in to_flush
-  s = Mock_InstallMemtableFlushResults(
-      &list, MutableCFOptions(options, ioptions), to_flush, &to_delete);
+  s = Mock_InstallMemtableFlushResults(&list, MutableCFOptions(options),
+                                       to_flush, &to_delete);
  ASSERT_OK(s);

  // Note:  now to_flush contains tables[0,1,2,4].  to_flush2 contains
@@ -560,7 +557,7 @@ TEST_F(MemTableListTest, FlushPendingTest) {

  // Flush the 1 memtable that was picked in to_flush2
  s = MemTableListTest::Mock_InstallMemtableFlushResults(
-      &list, MutableCFOptions(options, ioptions), to_flush2, &to_delete);
+      &list, MutableCFOptions(options), to_flush2, &to_delete);
  ASSERT_OK(s);

  // This will actually install 2 tables.  The 1 we told it to flush, and also
@@ -75,6 +75,7 @@ Status MergeHelper::TimedFullMerge(const MergeOperator* merge_operator,
 //       operands_ stores the list of merge operands encountered while merging.
 //       keys_[i] corresponds to operands_[i] for each i.
 Status MergeHelper::MergeUntil(InternalIterator* iter,
+                               RangeDelAggregator* range_del_agg,
                               const SequenceNumber stop_before,
                               const bool at_bottom) {
  // Get a copy of the internal key, before it's invalidated by iter->Next()
@@ -128,17 +129,8 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,

    assert(IsValueType(ikey.type));
    if (ikey.type != kTypeMerge) {
-      if (ikey.type != kTypeValue && ikey.type != kTypeDeletion) {
-        // Merges operands can only be used with puts and deletions, single
-        // deletions are not supported.
-        assert(false);
-        // release build doesn't have asserts, so we return error status
-        return Status::InvalidArgument(
-            " Merges operands can only be used with puts and deletions, single "
-            "deletions are not supported.");
-      }

-      // hit a put/delete
+      // hit a put/delete/single delete
      //   => merge the put value or a nullptr with operands_
      //   => store result in operands_.back() (and update keys_.back())
      //   => change the entry type to kTypeValue for keys_.back()
@@ -180,6 +172,7 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
    } else {
      // hit a merge
      //   => if there is a compaction filter, apply it.
+      //   => check for range tombstones covering the operand
      //   => merge the operand into the front of the operands_ list
      //      if not filtered
      //   => then continue because we haven't yet seen a Put/Delete.
@@ -192,8 +185,10 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
      // 1) it's included in one of the snapshots. in that case we *must* write
      // it out, no matter what compaction filter says
      // 2) it's not filtered by a compaction filter
-      if (ikey.sequence <= latest_snapshot_ ||
-          !FilterMerge(orig_ikey.user_key, value_slice)) {
+      if ((ikey.sequence <= latest_snapshot_ ||
+           !FilterMerge(orig_ikey.user_key, value_slice)) &&
+          (range_del_agg == nullptr ||
+           !range_del_agg->ShouldDelete(iter->key()))) {
        if (original_key_is_iter) {
          // this is just an optimization that saves us one memcpy
          keys_.push_front(std::move(original_key));
@@ -12,6 +12,7 @@

 #include "db/dbformat.h"
 #include "db/merge_context.h"
+#include "db/range_del_aggregator.h"
 #include "rocksdb/compaction_filter.h"
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
@@ -71,6 +72,7 @@ class MergeHelper {
  //  or - the end of iteration
  // iter: (IN)  points to the first merge type entry
  //       (OUT) points to the first entry not included in the merge process
+  // range_del_agg: (IN) filters merge operands covered by range tombstones.
  // stop_before: (IN) a sequence number that merge should not cross.
  //                   0 means no restriction
  // at_bottom:   (IN) true if the iterator covers the bottem level, which means
@@ -85,6 +87,7 @@ class MergeHelper {
  //
  // REQUIRED: The first key in the input is not corrupted.
  Status MergeUntil(InternalIterator* iter,
+                    RangeDelAggregator* range_del_agg = nullptr,
                    const SequenceNumber stop_before = 0,
                    const bool at_bottom = false);

@@ -29,7 +29,8 @@ class MergeHelperTest : public testing::Test {
    merge_helper_.reset(new MergeHelper(env_, BytewiseComparator(),
                                        merge_op_.get(), filter_.get(), nullptr,
                                        2U, false, latest_snapshot));
-    return merge_helper_->MergeUntil(iter_.get(), stop_before, at_bottom);
+    return merge_helper_->MergeUntil(iter_.get(), nullptr /* range_del_agg */,
+                                     stop_before, at_bottom);
  }

  void AddKeyVal(const std::string& user_key, const SequenceNumber& seq,
@@ -386,10 +386,6 @@ TEST_P(PlainTableDBTest, Flush) {
      for (int total_order = 0; total_order <= 1; total_order++) {
        for (int store_index_in_file = 0; store_index_in_file <= 1;
             ++store_index_in_file) {
-          if (!bloom_bits && store_index_in_file) {
-            continue;
-          }
-
          Options options = CurrentOptions();
          options.create_if_missing = true;
          // Set only one bucket to force bucket conflict.
@@ -18,6 +18,7 @@ int main() {
 #include <vector>

 #include <gflags/gflags.h>
+#include "db/db_impl.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/filter_policy.h"
@@ -26,9 +27,11 @@ int main() {
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/table.h"
 #include "util/histogram.h"
+#include "util/random.h"
 #include "util/stop_watch.h"
 #include "util/string_util.h"
 #include "util/testharness.h"
+#include "utilities/merge_operators.h"

 using GFLAGS::ParseCommandLineFlags;

@@ -46,6 +49,7 @@ DEFINE_int32(skiplist_height, 4, "");
 DEFINE_double(memtable_prefix_bloom_size_ratio, 0.1, "");
 DEFINE_int32(memtable_huge_page_size, 2 * 1024 * 1024, "");
 DEFINE_int32(value_size, 40, "");
+DEFINE_bool(enable_print, false, "Print options generated to console.");

 // Path to the database on file system
 const std::string kDbName = rocksdb::test::TmpDir() + "/prefix_test";
@@ -106,6 +110,10 @@ class TestKeyComparator : public Comparator {
    return 0;
  }

+  bool operator()(const TestKey& a, const TestKey& b) const {
+    return Compare(TestKeyToSlice(a), TestKeyToSlice(b)) < 0;
+  }
+
  virtual const char* Name() const override {
    return "TestKeyComparator";
  }
@@ -124,6 +132,23 @@ void PutKey(DB* db, WriteOptions write_options, uint64_t prefix,
  ASSERT_OK(db->Put(write_options, key, value));
 }

+void PutKey(DB* db, WriteOptions write_options, const TestKey& test_key,
+            const Slice& value) {
+  Slice key = TestKeyToSlice(test_key);
+  ASSERT_OK(db->Put(write_options, key, value));
+}
+
+void MergeKey(DB* db, WriteOptions write_options, const TestKey& test_key,
+              const Slice& value) {
+  Slice key = TestKeyToSlice(test_key);
+  ASSERT_OK(db->Merge(write_options, key, value));
+}
+
+void DeleteKey(DB* db, WriteOptions write_options, const TestKey& test_key) {
+  Slice key = TestKeyToSlice(test_key);
+  ASSERT_OK(db->Delete(write_options, key));
+}
+
 void SeekIterator(Iterator* iter, uint64_t prefix, uint64_t suffix) {
  TestKey test_key(prefix, suffix);
  Slice key = TestKeyToSlice(test_key);
@@ -267,6 +292,7 @@ TEST(SamePrefixTest, InDomainTest) {
  WriteOptions write_options;
  ReadOptions read_options;
  {
+    ASSERT_OK(DestroyDB(kDbName, Options()));
    ASSERT_OK(DB::Open(options, kDbName, &db));
    ASSERT_OK(db->Put(write_options, "HHKB pro2", "Mar 24, 2006"));
    ASSERT_OK(db->Put(write_options, "HHKB pro2 Type-S", "June 29, 2011"));
@@ -629,8 +655,206 @@ TEST_F(PrefixTest, DynamicPrefixIterator) {
  }
 }

+TEST_F(PrefixTest, PrefixSeekModePrev) {
+  // Only for SkipListFactory
+  options.memtable_factory.reset(new SkipListFactory);
+  options.merge_operator = MergeOperators::CreatePutOperator();
+  options.write_buffer_size = 1024 * 1024;
+  Random rnd(1);
+  for (size_t m = 1; m < 100; m++) {
+    std::cout << "[" + std::to_string(m) + "]" + "*** Mem table: "
+              << options.memtable_factory->Name() << std::endl;
+    DestroyDB(kDbName, Options());
+    auto db = OpenDb();
+    WriteOptions write_options;
+    ReadOptions read_options;
+    std::map<TestKey, std::string, TestKeyComparator> entry_maps[3], whole_map;
+    for (uint64_t i = 0; i < 10; i++) {
+      int div = i % 3 + 1;
+      for (uint64_t j = 0; j < 10; j++) {
+        whole_map[TestKey(i, j)] = entry_maps[rnd.Uniform(div)][TestKey(i, j)] =
+            'v' + std::to_string(i) + std::to_string(j);
+      }
+    }
+
+    std::map<TestKey, std::string, TestKeyComparator> type_map;
+    for (size_t i = 0; i < 3; i++) {
+      for (auto& kv : entry_maps[i]) {
+        if (rnd.OneIn(3)) {
+          PutKey(db.get(), write_options, kv.first, kv.second);
+          type_map[kv.first] = "value";
+        } else {
+          MergeKey(db.get(), write_options, kv.first, kv.second);
+          type_map[kv.first] = "merge";
+        }
+      }
+      if (i < 2) {
+        db->Flush(FlushOptions());
+      }
+    }
+
+    for (size_t i = 0; i < 2; i++) {
+      for (auto& kv : entry_maps[i]) {
+        if (rnd.OneIn(10)) {
+          whole_map.erase(kv.first);
+          DeleteKey(db.get(), write_options, kv.first);
+          entry_maps[2][kv.first] = "delete";
+        }
+      }
+    }
+
+    if (FLAGS_enable_print) {
+      for (size_t i = 0; i < 3; i++) {
+        for (auto& kv : entry_maps[i]) {
+          std::cout << "[" << i << "]" << kv.first.prefix << kv.first.sorted
+                    << " " << kv.second + " " + type_map[kv.first] << std::endl;
+        }
+      }
+    }
+
+    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+    for (uint64_t prefix = 0; prefix < 10; prefix++) {
+      uint64_t start_suffix = rnd.Uniform(9);
+      SeekIterator(iter.get(), prefix, start_suffix);
+      auto it = whole_map.find(TestKey(prefix, start_suffix));
+      if (it == whole_map.end()) {
+        continue;
+      }
+      ASSERT_NE(it, whole_map.end());
+      ASSERT_TRUE(iter->Valid());
+      if (FLAGS_enable_print) {
+        std::cout << "round " << prefix
+                  << " iter: " << SliceToTestKey(iter->key())->prefix
+                  << SliceToTestKey(iter->key())->sorted
+                  << " | map: " << it->first.prefix << it->first.sorted << " | "
+                  << iter->value().ToString() << " " << it->second << std::endl;
+      }
+      ASSERT_EQ(iter->value(), it->second);
+      uint64_t stored_prefix = prefix;
+      for (size_t k = 0; k < 9; k++) {
+        if (rnd.OneIn(2) || it == whole_map.begin()) {
+          iter->Next();
+          it++;
+          if (FLAGS_enable_print) {
+            std::cout << "Next >> ";
+          }
+        } else {
+          iter->Prev();
+          it--;
+          if (FLAGS_enable_print) {
+            std::cout << "Prev >> ";
+          }
+        }
+        if (!iter->Valid() ||
+            SliceToTestKey(iter->key())->prefix != stored_prefix) {
+          break;
+        }
+        stored_prefix = SliceToTestKey(iter->key())->prefix;
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_NE(it, whole_map.end());
+        ASSERT_EQ(iter->value(), it->second);
+        if (FLAGS_enable_print) {
+          std::cout << "iter: " << SliceToTestKey(iter->key())->prefix
+                    << SliceToTestKey(iter->key())->sorted
+                    << " | map: " << it->first.prefix << it->first.sorted
+                    << " | " << iter->value().ToString() << " " << it->second
+                    << std::endl;
+        }
+      }
+    }
+  }
 }

+TEST_F(PrefixTest, PrefixSeekModePrev2) {
+  // Only for SkipListFactory
+  // test the case
+  //        iter1                iter2
+  // | prefix | suffix |  | prefix | suffix |
+  // |   1    |   1    |  |   1    |   2    |
+  // |   1    |   3    |  |   1    |   4    |
+  // |   2    |   1    |  |   3    |   3    |
+  // |   2    |   2    |  |   3    |   4    |
+  // after seek(15), iter1 will be at 21 and iter2 will be 33.
+  // Then if call Prev() in prefix mode where SeekForPrev(21) gets called,
+  // iter2 should turn to invalid state because of bloom filter.
+  options.memtable_factory.reset(new SkipListFactory);
+  options.write_buffer_size = 1024 * 1024;
+  std::string v13("v13");
+  DestroyDB(kDbName, Options());
+  auto db = OpenDb();
+  WriteOptions write_options;
+  ReadOptions read_options;
+  PutKey(db.get(), write_options, TestKey(1, 2), "v12");
+  PutKey(db.get(), write_options, TestKey(1, 4), "v14");
+  PutKey(db.get(), write_options, TestKey(3, 3), "v33");
+  PutKey(db.get(), write_options, TestKey(3, 4), "v34");
+  db->Flush(FlushOptions());
+  reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+  PutKey(db.get(), write_options, TestKey(1, 1), "v11");
+  PutKey(db.get(), write_options, TestKey(1, 3), "v13");
+  PutKey(db.get(), write_options, TestKey(2, 1), "v21");
+  PutKey(db.get(), write_options, TestKey(2, 2), "v22");
+  db->Flush(FlushOptions());
+  reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+  std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+  SeekIterator(iter.get(), 1, 5);
+  iter->Prev();
+  ASSERT_EQ(iter->value(), v13);
+}
+
+TEST_F(PrefixTest, PrefixSeekModePrev3) {
+  // Only for SkipListFactory
+  // test SeekToLast() with iterate_upper_bound_ in prefix_seek_mode
+  options.memtable_factory.reset(new SkipListFactory);
+  options.write_buffer_size = 1024 * 1024;
+  std::string v14("v14");
+  TestKey upper_bound_key = TestKey(1, 5);
+  Slice upper_bound = TestKeyToSlice(upper_bound_key);
+
+  {
+    DestroyDB(kDbName, Options());
+    auto db = OpenDb();
+    WriteOptions write_options;
+    ReadOptions read_options;
+    read_options.iterate_upper_bound = &upper_bound;
+    PutKey(db.get(), write_options, TestKey(1, 2), "v12");
+    PutKey(db.get(), write_options, TestKey(1, 4), "v14");
+    db->Flush(FlushOptions());
+    reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+    PutKey(db.get(), write_options, TestKey(1, 1), "v11");
+    PutKey(db.get(), write_options, TestKey(1, 3), "v13");
+    PutKey(db.get(), write_options, TestKey(2, 1), "v21");
+    PutKey(db.get(), write_options, TestKey(2, 2), "v22");
+    db->Flush(FlushOptions());
+    reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+    iter->SeekToLast();
+    ASSERT_EQ(iter->value(), v14);
+  }
+  {
+    DestroyDB(kDbName, Options());
+    auto db = OpenDb();
+    WriteOptions write_options;
+    ReadOptions read_options;
+    read_options.iterate_upper_bound = &upper_bound;
+    PutKey(db.get(), write_options, TestKey(1, 2), "v12");
+    PutKey(db.get(), write_options, TestKey(1, 4), "v14");
+    PutKey(db.get(), write_options, TestKey(3, 3), "v33");
+    PutKey(db.get(), write_options, TestKey(3, 4), "v34");
+    db->Flush(FlushOptions());
+    reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+    PutKey(db.get(), write_options, TestKey(1, 1), "v11");
+    PutKey(db.get(), write_options, TestKey(1, 3), "v13");
+    db->Flush(FlushOptions());
+    reinterpret_cast<DBImpl*>(db.get())->TEST_WaitForFlushMemTable();
+    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+    iter->SeekToLast();
+    ASSERT_EQ(iter->value(), v14);
+  }
+}
+
+}  // end namespace rocksdb
+
 int main(int argc, char** argv) {
  ::testing::InitGoogleTest(&argc, argv);
  ParseCommandLineFlags(&argc, &argv, true);
@@ -0,0 +1,200 @@
+//  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/range_del_aggregator.h"
+
+#include <algorithm>
+
+namespace rocksdb {
+
+RangeDelAggregator::RangeDelAggregator(
+    const InternalKeyComparator& icmp,
+    const std::vector<SequenceNumber>& snapshots)
+    : icmp_(icmp) {
+  pinned_iters_mgr_.StartPinning();
+  for (auto snapshot : snapshots) {
+    stripe_map_.emplace(
+        snapshot,
+        TombstoneMap(stl_wrappers::LessOfComparator(icmp_.user_comparator())));
+  }
+  // Data newer than any snapshot falls in this catch-all stripe
+  stripe_map_.emplace(kMaxSequenceNumber, TombstoneMap());
+}
+
+bool RangeDelAggregator::ShouldDelete(const Slice& internal_key,
+                                      bool for_compaction /* = false */) {
+  ParsedInternalKey parsed;
+  if (!ParseInternalKey(internal_key, &parsed)) {
+    assert(false);
+  }
+  assert(IsValueType(parsed.type));
+
+  // Starting point is the snapshot stripe in which the key lives, then need to
+  // search all earlier stripes too, unless it's for compaction.
+  for (auto stripe_map_iter = GetStripeMapIter(parsed.sequence);
+       stripe_map_iter != stripe_map_.end(); ++stripe_map_iter) {
+    const auto& tombstone_map = stripe_map_iter->second;
+    for (const auto& start_key_and_tombstone : tombstone_map) {
+      const auto& tombstone = start_key_and_tombstone.second;
+      if (icmp_.user_comparator()->Compare(parsed.user_key,
+                                           tombstone.start_key_) < 0) {
+        break;
+      }
+      if (parsed.sequence < tombstone.seq_ &&
+          icmp_.user_comparator()->Compare(parsed.user_key,
+                                           tombstone.end_key_) <= 0) {
+        return true;
+      }
+    }
+    if (for_compaction) {
+      break;
+    }
+  }
+  return false;
+}
+
+bool RangeDelAggregator::ShouldAddTombstones(
+    bool bottommost_level /* = false */) {
+  auto stripe_map_iter = stripe_map_.begin();
+  assert(stripe_map_iter != stripe_map_.end());
+  if (bottommost_level) {
+    // For the bottommost level, keys covered by tombstones in the first
+    // (oldest) stripe have been compacted away, so the tombstones are obsolete.
+    ++stripe_map_iter;
+  }
+  while (stripe_map_iter != stripe_map_.end()) {
+    if (!stripe_map_iter->second.empty()) {
+      return true;
+    }
+    ++stripe_map_iter;
+  }
+  return false;
+}
+
+Status RangeDelAggregator::AddTombstones(ScopedArenaIterator input) {
+  return AddTombstones(input.release(), true /* arena */);
+}
+
+Status RangeDelAggregator::AddTombstones(
+    std::unique_ptr<InternalIterator> input) {
+  return AddTombstones(input.release(), false /* arena */);
+}
+
+Status RangeDelAggregator::AddTombstones(InternalIterator* input, bool arena) {
+  pinned_iters_mgr_.PinIterator(input, arena);
+  input->SeekToFirst();
+  while (input->Valid()) {
+    ParsedInternalKey parsed_key;
+    if (!ParseInternalKey(input->key(), &parsed_key)) {
+      return Status::Corruption("Unable to parse range tombstone InternalKey");
+    }
+    RangeTombstone tombstone(parsed_key, input->value());
+    auto& tombstone_map = GetStripeMapIter(tombstone.seq_)->second;
+    tombstone_map.emplace(tombstone.start_key_.ToString(),
+                          std::move(tombstone));
+    input->Next();
+  }
+  return Status::OK();
+}
+
+RangeDelAggregator::StripeMap::iterator RangeDelAggregator::GetStripeMapIter(
+    SequenceNumber seq) {
+  // The stripe includes seqnum for the snapshot above and excludes seqnum for
+  // the snapshot below.
+  StripeMap::iterator iter;
+  if (seq > 0) {
+    // upper_bound() checks strict inequality so need to subtract one
+    iter = stripe_map_.upper_bound(seq - 1);
+  } else {
+    iter = stripe_map_.begin();
+  }
+  // catch-all stripe justifies this assertion in either of above cases
+  assert(iter != stripe_map_.end());
+  return iter;
+}
+
+// TODO(andrewkr): We should implement an iterator over range tombstones in our
+// map. It'd enable compaction to open tables on-demand, i.e., only once range
+// tombstones are known to be available, without the code duplication we have
+// in ShouldAddTombstones(). It'll also allow us to move the table-modifying
+// code into more coherent places: CompactionJob and BuildTable().
+void RangeDelAggregator::AddToBuilder(TableBuilder* builder,
+                                      bool extend_before_min_key,
+                                      const Slice* next_table_min_key,
+                                      FileMetaData* meta,
+                                      bool bottommost_level /* = false */) {
+  auto stripe_map_iter = stripe_map_.begin();
+  assert(stripe_map_iter != stripe_map_.end());
+  if (bottommost_level) {
+    // For the bottommost level, keys covered by tombstones in the first
+    // (oldest) stripe have been compacted away, so the tombstones are obsolete.
+    ++stripe_map_iter;
+  }
+
+  // Note the order in which tombstones are stored is insignificant since we
+  // insert them into a std::map on the read path.
+  bool first_added = false;
+  while (stripe_map_iter != stripe_map_.end()) {
+    for (const auto& start_key_and_tombstone : stripe_map_iter->second) {
+      const auto& tombstone = start_key_and_tombstone.second;
+      if (next_table_min_key != nullptr &&
+          icmp_.user_comparator()->Compare(*next_table_min_key,
+                                           tombstone.start_key_) < 0) {
+        // Tombstones starting after next_table_min_key only need to be included
+        // in the next table.
+        break;
+      }
+      if (!extend_before_min_key && meta->smallest.size() != 0 &&
+          icmp_.user_comparator()->Compare(tombstone.end_key_,
+                                           meta->smallest.user_key()) < 0) {
+        // Tombstones ending before this table's smallest key can conditionally
+        // be excluded, e.g., when this table is a non-first compaction output,
+        // we know such tombstones are included in the previous table. In that
+        // case extend_before_min_key would be false.
+        continue;
+      }
+
+      auto ikey_and_end_key = tombstone.Serialize();
+      builder->Add(ikey_and_end_key.first.Encode(), ikey_and_end_key.second);
+      if (!first_added) {
+        first_added = true;
+        if (extend_before_min_key &&
+            (meta->smallest.size() == 0 ||
+             icmp_.Compare(ikey_and_end_key.first, meta->smallest) < 0)) {
+          meta->smallest = ikey_and_end_key.first;
+        }
+      }
+      auto end_ikey = tombstone.SerializeEndKey();
+      if (meta->largest.size() == 0 ||
+          icmp_.Compare(meta->largest, end_ikey) < 0) {
+        if (next_table_min_key != nullptr &&
+            icmp_.Compare(*next_table_min_key, end_ikey.Encode()) < 0) {
+          // Pretend the largest key has the same user key as the min key in the
+          // following table in order for files to appear key-space partitioned.
+          // Choose highest seqnum so this file's largest comes before the next
+          // file's smallest. The fake seqnum is OK because the read path's
+          // file-picking code only considers the user key portion.
+          //
+          // Note Seek() also creates InternalKey with (user_key,
+          // kMaxSequenceNumber), but with kTypeDeletion (0x7) instead of
+          // kTypeRangeDeletion (0xF), so the range tombstone comes before the
+          // Seek() key in InternalKey's ordering. So Seek() will look in the
+          // next file for the user key.
+          ParsedInternalKey parsed;
+          ParseInternalKey(*next_table_min_key, &parsed);
+          meta->largest = InternalKey(parsed.user_key, kMaxSequenceNumber,
+                                      kTypeRangeDeletion);
+        } else {
+          meta->largest = std::move(end_ikey);
+        }
+      }
+      meta->smallest_seqno = std::min(meta->smallest_seqno, tombstone.seq_);
+      meta->largest_seqno = std::max(meta->largest_seqno, tombstone.seq_);
+    }
+    ++stripe_map_iter;
+  }
+}
+
+}  // namespace rocksdb
@@ -0,0 +1,53 @@
+//  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
+
+#include <map>
+#include <string>
+#include <vector>
+
+#include "db/dbformat.h"
+#include "db/pinned_iterators_manager.h"
+#include "db/version_edit.h"
+#include "include/rocksdb/comparator.h"
+#include "include/rocksdb/types.h"
+#include "table/internal_iterator.h"
+#include "table/scoped_arena_iterator.h"
+#include "table/table_builder.h"
+#include "util/kv_map.h"
+
+namespace rocksdb {
+class RangeDelAggregator {
+ public:
+  RangeDelAggregator(const InternalKeyComparator& icmp,
+                     const std::vector<SequenceNumber>& snapshots);
+
+  bool ShouldDelete(const Slice& internal_key, bool for_compaction = false);
+  bool ShouldAddTombstones(bool bottommost_level = false);
+  Status AddTombstones(ScopedArenaIterator input);
+  Status AddTombstones(std::unique_ptr<InternalIterator> input);
+  // write tombstones covering a range to a table builder
+  // usually don't add to a max-level table builder
+  void AddToBuilder(TableBuilder* builder, bool extend_before_min_key,
+                    const Slice* next_table_min_key, FileMetaData* meta,
+                    bool bottommost_level = false);
+
+ private:
+  // Maps tombstone start key -> tombstone object
+  typedef std::map<std::string, RangeTombstone, stl_wrappers::LessOfComparator>
+      TombstoneMap;
+  // Maps snapshot seqnum -> map of tombstones that fall in that stripe, i.e.,
+  // their seqnums are greater than the next smaller snapshot's seqnum.
+  typedef std::map<SequenceNumber, TombstoneMap> StripeMap;
+
+  Status AddTombstones(InternalIterator* input, bool arena);
+  StripeMap::iterator GetStripeMapIter(SequenceNumber seq);
+
+  PinnedIteratorsManager pinned_iters_mgr_;
+  StripeMap stripe_map_;
+  const InternalKeyComparator icmp_;
+};
+}  // namespace rocksdb
@@ -99,10 +99,11 @@ class Repairer {
        env_(db_options.env),
        env_options_(),
        db_options_(SanitizeOptions(dbname_, db_options)),
+        immutable_db_options_(db_options_),
        icmp_(default_cf_opts.comparator),
        default_cf_opts_(default_cf_opts),
        default_cf_iopts_(
-            ImmutableCFOptions(Options(db_options_, default_cf_opts))),
+            ImmutableCFOptions(immutable_db_options_, default_cf_opts)),
        unknown_cf_opts_(unknown_cf_opts),
        create_unknown_cfs_(create_unknown_cfs),
        raw_table_cache_(
@@ -113,8 +114,8 @@ class Repairer {
                                    raw_table_cache_.get())),
        wb_(db_options_.db_write_buffer_size),
        wc_(db_options_.delayed_write_rate),
-        vset_(dbname_, &db_options_, env_options_, raw_table_cache_.get(), &wb_,
-              &wc_),
+        vset_(dbname_, &immutable_db_options_, env_options_,
+              raw_table_cache_.get(), &wb_, &wc_),
        next_file_number_(1) {
    for (const auto& cfd : column_families) {
      cf_name_to_opts_[cfd.name] = cfd.options;
@@ -141,7 +142,7 @@ class Repairer {
                                cf_name + ", id=" + ToString(cf_id));
    }
    Options opts(db_options_, *cf_opts);
-    MutableCFOptions mut_cf_opts(opts, ImmutableCFOptions(opts));
+    MutableCFOptions mut_cf_opts(opts);

    VersionEdit edit;
    edit.SetComparatorName(opts.comparator->Name());
@@ -222,6 +223,7 @@ class Repairer {
  Env* const env_;
  const EnvOptions env_options_;
  const DBOptions db_options_;
+  const ImmutableDBOptions immutable_db_options_;
  const InternalKeyComparator icmp_;
  const ColumnFamilyOptions default_cf_opts_;
  const ImmutableCFOptions default_cf_iopts_;  // table_cache_ holds reference
@@ -92,6 +92,9 @@ class SkipList {
    // Advance to the first entry with a key >= target
    void Seek(const Key& target);

+    // Retreat to the last entry with a key <= target
+    void SeekForPrev(const Key& target);
+
    // Position at the first entry in list.
    // Final state of iterator is Valid() iff list is not empty.
    void SeekToFirst();
@@ -135,6 +138,9 @@ class SkipList {
  Node* NewNode(const Key& key, int height);
  int RandomHeight();
  bool Equal(const Key& a, const Key& b) const { return (compare_(a, b) == 0); }
+  bool LessThan(const Key& a, const Key& b) const {
+    return (compare_(a, b) < 0);
+  }

  // Return true if key is greater than the data stored in "n"
  bool KeyIsAfterNode(const Key& key, Node* n) const;
@@ -247,7 +253,19 @@ inline void SkipList<Key, Comparator>::Iterator::Seek(const Key& target) {
  node_ = list_->FindGreaterOrEqual(target);
 }

-template<typename Key, class Comparator>
+template <typename Key, class Comparator>
+inline void SkipList<Key, Comparator>::Iterator::SeekForPrev(
+    const Key& target) {
+  Seek(target);
+  if (!Valid()) {
+    SeekToLast();
+  }
+  while (Valid() && list_->LessThan(target, key())) {
+    Prev();
+  }
+}
+
+template <typename Key, class Comparator>
 inline void SkipList<Key, Comparator>::Iterator::SeekToFirst() {
  node_ = list_->head_->Next(0);
 }
@@ -45,6 +45,8 @@ TEST_F(SkipTest, Empty) {
  ASSERT_TRUE(!iter.Valid());
  iter.Seek(100);
  ASSERT_TRUE(!iter.Valid());
+  iter.SeekForPrev(100);
+  ASSERT_TRUE(!iter.Valid());
  iter.SeekToLast();
  ASSERT_TRUE(!iter.Valid());
 }
@@ -81,6 +83,10 @@ TEST_F(SkipTest, InsertAndLookup) {
    ASSERT_TRUE(iter.Valid());
    ASSERT_EQ(*(keys.begin()), iter.key());

+    iter.SeekForPrev(R - 1);
+    ASSERT_TRUE(iter.Valid());
+    ASSERT_EQ(*(keys.rbegin()), iter.key());
+
    iter.SeekToFirst();
    ASSERT_TRUE(iter.Valid());
    ASSERT_EQ(*(keys.begin()), iter.key());
@@ -111,19 +117,22 @@ TEST_F(SkipTest, InsertAndLookup) {
  }

  // Backward iteration test
-  {
+  for (int i = 0; i < R; i++) {
    SkipList<Key, TestComparator>::Iterator iter(&list);
-    iter.SeekToLast();
+    iter.SeekForPrev(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, 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, iter.key());
+        iter.Prev();
+      }
    }
-    ASSERT_TRUE(!iter.Valid());
  }
 }

@@ -170,7 +170,10 @@ InternalIterator* TableCache::NewIterator(
    const ReadOptions& options, const EnvOptions& env_options,
    const InternalKeyComparator& icomparator, const FileDescriptor& fd,
    TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
-    bool for_compaction, Arena* arena, bool skip_filters, int level) {
+    bool for_compaction, Arena* arena, bool skip_filters, int level,
+    RangeDelAggregator* range_del_agg /* = nullptr */,
+    bool is_range_del_only /* = false */) {
+  assert(!is_range_del_only || range_del_agg != nullptr);
  PERF_TIMER_GUARD(new_table_iterator_nanos);

  if (table_reader_ptr != nullptr) {
@@ -182,14 +185,18 @@ InternalIterator* TableCache::NewIterator(

  size_t readahead = 0;
  bool create_new_table_reader = false;
-  if (for_compaction) {
-    if (ioptions_.new_table_reader_for_compaction_inputs) {
-      readahead = ioptions_.compaction_readahead_size;
-      create_new_table_reader = true;
+  // pointless to create a new table reader for range tombstones only since the
+  // reader isn't reused
+  if (!is_range_del_only) {
+    if (for_compaction) {
+      if (ioptions_.new_table_reader_for_compaction_inputs) {
+        readahead = ioptions_.compaction_readahead_size;
+        create_new_table_reader = true;
+      }
+    } else {
+      readahead = options.readahead_size;
+      create_new_table_reader = readahead > 0;
    }
-  } else {
-    readahead = options.readahead_size;
-    create_new_table_reader = readahead > 0;
  }

  if (create_new_table_reader) {
@@ -216,23 +223,39 @@ InternalIterator* TableCache::NewIterator(
    }
  }

-  InternalIterator* result =
-      table_reader->NewIterator(options, arena, skip_filters);
-
-  if (create_new_table_reader) {
-    assert(handle == nullptr);
-    result->RegisterCleanup(&DeleteTableReader, table_reader, nullptr);
-  } else if (handle != nullptr) {
-    result->RegisterCleanup(&UnrefEntry, cache_, handle);
+  if (range_del_agg != nullptr) {
+    std::unique_ptr<InternalIterator> iter(
+        table_reader->NewRangeTombstoneIterator(options));
+    Status s = range_del_agg->AddTombstones(std::move(iter));
+    if (!s.ok()) {
+      return NewErrorInternalIterator(s, arena);
+    }
  }

-  if (for_compaction) {
-    table_reader->SetupForCompaction();
-  }
-  if (table_reader_ptr != nullptr) {
-    *table_reader_ptr = table_reader;
-  }
+  InternalIterator* result = nullptr;
+  if (!is_range_del_only) {
+    result = table_reader->NewIterator(options, arena, skip_filters);
+    if (create_new_table_reader) {
+      assert(handle == nullptr);
+      result->RegisterCleanup(&DeleteTableReader, table_reader, nullptr);
+    } else if (handle != nullptr) {
+      result->RegisterCleanup(&UnrefEntry, cache_, handle);
+    }

+    if (for_compaction) {
+      table_reader->SetupForCompaction();
+    }
+    if (table_reader_ptr != nullptr) {
+      *table_reader_ptr = table_reader;
+    }
+  } else {
+    assert(!create_new_table_reader);
+    // don't need the table reader at all since the iterator over the meta-block
+    // doesn't require it
+    if (handle != nullptr) {
+      UnrefEntry(cache_, handle);
+    }
+  }
  return result;
 }

@@ -15,6 +15,7 @@
 #include <stdint.h>

 #include "db/dbformat.h"
+#include "db/range_del_aggregator.h"
 #include "port/port.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/env.h"
@@ -47,12 +48,18 @@ class TableCache {
  // returned iterator is live.
  // @param skip_filters Disables loading/accessing the filter block
  // @param level The level this table is at, -1 for "not set / don't know"
+  // @param range_del_agg When non-nullptr, creates a range tombstone iterator
+  //    over this file's meta-block and gives it to this object
+  // @param is_range_del_only When set, this function only gives a range
+  //    tombstone iterator to range_del_agg and then returns nullptr
  InternalIterator* NewIterator(
      const ReadOptions& options, const EnvOptions& toptions,
      const InternalKeyComparator& internal_comparator,
      const FileDescriptor& file_fd, TableReader** table_reader_ptr = nullptr,
      HistogramImpl* file_read_hist = nullptr, bool for_compaction = false,
-      Arena* arena = nullptr, bool skip_filters = false, int level = -1);
+      Arena* arena = nullptr, bool skip_filters = false, int level = -1,
+      RangeDelAggregator* range_del_agg = nullptr,
+      bool is_range_del_only = false);

  // If a seek to internal key "k" in specified file finds an entry,
  // call (*handle_result)(arg, found_key, found_value) repeatedly until
@@ -46,11 +46,12 @@ void MakeBuilder(const Options& options, const ImmutableCFOptions& ioptions,
                 std::unique_ptr<TableBuilder>* builder) {
  unique_ptr<WritableFile> wf(new test::StringSink);
  writable->reset(new WritableFileWriter(std::move(wf), EnvOptions()));
-
+  int unknown_level = -1;
  builder->reset(NewTableBuilder(
      ioptions, internal_comparator, int_tbl_prop_collector_factories,
      kTestColumnFamilyId, kTestColumnFamilyName,
-      writable->get(), options.compression, options.compression_opts));
+      writable->get(), options.compression, options.compression_opts,
+      unknown_level));
 }
 }  // namespace

@@ -255,7 +256,7 @@ void TestCustomizedTablePropertiesCollector(
    int_tbl_prop_collector_factories.emplace_back(
        new RegularKeysStartWithAFactory(backward_mode));
  } else {
-    GetIntTblPropCollectorFactory(options, &int_tbl_prop_collector_factories);
+    GetIntTblPropCollectorFactory(ioptions, &int_tbl_prop_collector_factories);
  }
  MakeBuilder(options, ioptions, internal_comparator,
              &int_tbl_prop_collector_factories, &writer, &builder);
@@ -392,7 +393,8 @@ void TestInternalKeyPropertiesCollector(
    options = SanitizeOptions("db",            // just a place holder
                              &pikc,
                              options);
-    GetIntTblPropCollectorFactory(options, &int_tbl_prop_collector_factories);
+    ImmutableCFOptions ioptions(options);
+    GetIntTblPropCollectorFactory(ioptions, &int_tbl_prop_collector_factories);
    options.comparator = comparator;
  } else {
    int_tbl_prop_collector_factories.emplace_back(
@@ -16,7 +16,7 @@
 namespace rocksdb {

 TransactionLogIteratorImpl::TransactionLogIteratorImpl(
-    const std::string& dir, const DBOptions* options,
+    const std::string& dir, const ImmutableDBOptions* options,
    const TransactionLogIterator::ReadOptions& read_options,
    const EnvOptions& soptions, const SequenceNumber seq,
    std::unique_ptr<VectorLogPtr> files, VersionSet const* const versions)
@@ -2,19 +2,20 @@
 //  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

 #ifndef ROCKSDB_LITE
-#pragma once
 #include <vector>

+#include "db/filename.h"
+#include "db/log_reader.h"
+#include "db/version_set.h"
+#include "port/port.h"
 #include "rocksdb/env.h"
 #include "rocksdb/options.h"
-#include "rocksdb/types.h"
 #include "rocksdb/transaction_log.h"
-#include "db/version_set.h"
-#include "db/log_reader.h"
-#include "db/filename.h"
-#include "port/port.h"
+#include "rocksdb/types.h"
+#include "util/db_options.h"

 namespace rocksdb {

@@ -58,7 +59,7 @@ class LogFileImpl : public LogFile {
 class TransactionLogIteratorImpl : public TransactionLogIterator {
 public:
  TransactionLogIteratorImpl(
-      const std::string& dir, const DBOptions* options,
+      const std::string& dir, const ImmutableDBOptions* options,
      const TransactionLogIterator::ReadOptions& read_options,
      const EnvOptions& soptions, const SequenceNumber seqNum,
      std::unique_ptr<VectorLogPtr> files, VersionSet const* const versions);
@@ -73,7 +74,7 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {

 private:
  const std::string& dir_;
-  const DBOptions* options_;
+  const ImmutableDBOptions* options_;
  const TransactionLogIterator::ReadOptions read_options_;
  const EnvOptions& soptions_;
  SequenceNumber startingSequenceNumber_;
@@ -32,12 +32,12 @@
 namespace rocksdb {

 bool NewestFirstBySeqNo(FileMetaData* a, FileMetaData* b) {
-  if (a->smallest_seqno != b->smallest_seqno) {
-    return a->smallest_seqno > b->smallest_seqno;
-  }
  if (a->largest_seqno != b->largest_seqno) {
    return a->largest_seqno > b->largest_seqno;
  }
+  if (a->smallest_seqno != b->smallest_seqno) {
+    return a->smallest_seqno > b->smallest_seqno;
+  }
  // Break ties by file number
  return a->fd.GetNumber() > b->fd.GetNumber();
 }
@@ -127,7 +127,13 @@ class VersionBuilder::Rep {
  }

  void CheckConsistency(VersionStorageInfo* vstorage) {
-#ifndef NDEBUG
+#ifdef NDEBUG
+    if (!vstorage->force_consistency_checks()) {
+      // Dont run consistency checks in release mode except if
+      // explicitly asked to
+      return;
+    }
+#endif
    // make sure the files are sorted correctly
    for (int level = 0; level < vstorage->num_levels(); level++) {
      auto& level_files = vstorage->LevelFiles(level);
@@ -135,31 +141,57 @@ class VersionBuilder::Rep {
        auto f1 = level_files[i - 1];
        auto f2 = level_files[i];
        if (level == 0) {
-          assert(level_zero_cmp_(f1, f2));
-          assert(f1->largest_seqno > f2->largest_seqno ||
-                 // We can have multiple files with seqno = 0 as a result of
-                 // using DB::AddFile()
-                 (f1->largest_seqno == 0 && f2->largest_seqno == 0));
+          if (!level_zero_cmp_(f1, f2)) {
+            fprintf(stderr, "L0 files are not sorted properly");
+            abort();
+          }
+
+          if (f2->smallest_seqno == f2->largest_seqno) {
+            // This is an external file that we ingested
+            SequenceNumber external_file_seqno = f2->smallest_seqno;
+            if (!(external_file_seqno < f1->largest_seqno ||
+                  external_file_seqno == 0)) {
+              fprintf(stderr, "L0 file with seqno %" PRIu64 " %" PRIu64
+                              " vs. file with global_seqno %" PRIu64 "\n",
+                      f1->smallest_seqno, f1->largest_seqno,
+                      external_file_seqno);
+              abort();
+            }
+          } else if (f1->smallest_seqno <= f2->smallest_seqno) {
+            fprintf(stderr, "L0 files seqno %" PRIu64 " %" PRIu64
+                            " vs. %" PRIu64 " %" PRIu64 "\n",
+                    f1->smallest_seqno, f1->largest_seqno, f2->smallest_seqno,
+                    f2->largest_seqno);
+            abort();
+          }
        } else {
-          assert(level_nonzero_cmp_(f1, f2));
+          if (!level_nonzero_cmp_(f1, f2)) {
+            fprintf(stderr, "L%d files are not sorted properly", level);
+            abort();
+          }

          // Make sure there is no overlap in levels > 0
          if (vstorage->InternalComparator()->Compare(f1->largest,
                                                      f2->smallest) >= 0) {
-            fprintf(stderr, "overlapping ranges in same level %s vs. %s\n",
-                    (f1->largest).DebugString().c_str(),
-                    (f2->smallest).DebugString().c_str());
+            fprintf(stderr, "L%d have overlapping ranges %s vs. %s\n", level,
+                    (f1->largest).DebugString(true).c_str(),
+                    (f2->smallest).DebugString(true).c_str());
            abort();
          }
        }
      }
    }
-#endif
  }

  void CheckConsistencyForDeletes(VersionEdit* edit, uint64_t number,
                                  int level) {
-#ifndef NDEBUG
+#ifdef NDEBUG
+    if (!base_vstorage_->force_consistency_checks()) {
+      // Dont run consistency checks in release mode except if
+      // explicitly asked to
+      return;
+    }
+#endif
    // a file to be deleted better exist in the previous version
    bool found = false;
    for (int l = 0; !found && l < base_vstorage_->num_levels(); l++) {
@@ -195,9 +227,8 @@ class VersionBuilder::Rep {
    }
    if (!found) {
      fprintf(stderr, "not found %" PRIu64 "\n", number);
+      abort();
    }
-    assert(found);
-#endif
  }

  // Apply all of the edits in *edit to the current state.
@@ -29,9 +29,9 @@ class VersionBuilderTest : public testing::Test {
      : ucmp_(BytewiseComparator()),
        icmp_(ucmp_),
        ioptions_(options_),
-        mutable_cf_options_(options_, ioptions_),
+        mutable_cf_options_(options_),
        vstorage_(&icmp_, ucmp_, options_.num_levels, kCompactionStyleLevel,
-                 nullptr),
+                  nullptr, false),
        file_num_(1) {
    mutable_cf_options_.RefreshDerivedOptions(ioptions_);
    size_being_compacted_.resize(options_.num_levels);
@@ -77,7 +77,7 @@ class VersionBuilderTest : public testing::Test {
  }

  void UpdateVersionStorageInfo() {
-    vstorage_.UpdateFilesByCompactionPri(mutable_cf_options_);
+    vstorage_.UpdateFilesByCompactionPri(ioptions_.compaction_pri);
    vstorage_.UpdateNumNonEmptyLevels();
    vstorage_.GenerateFileIndexer();
    vstorage_.GenerateLevelFilesBrief();
@@ -123,7 +123,7 @@ TEST_F(VersionBuilderTest, ApplyAndSaveTo) {
  VersionBuilder version_builder(env_options, nullptr, &vstorage_);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
-                                  kCompactionStyleLevel, nullptr);
+                                  kCompactionStyleLevel, nullptr, false);
  version_builder.Apply(&version_edit);
  version_builder.SaveTo(&new_vstorage);

@@ -158,7 +158,7 @@ TEST_F(VersionBuilderTest, ApplyAndSaveToDynamic) {
  VersionBuilder version_builder(env_options, nullptr, &vstorage_);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
-                                  kCompactionStyleLevel, nullptr);
+                                  kCompactionStyleLevel, nullptr, false);
  version_builder.Apply(&version_edit);
  version_builder.SaveTo(&new_vstorage);

@@ -198,7 +198,7 @@ TEST_F(VersionBuilderTest, ApplyAndSaveToDynamic2) {
  VersionBuilder version_builder(env_options, nullptr, &vstorage_);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
-                                  kCompactionStyleLevel, nullptr);
+                                  kCompactionStyleLevel, nullptr, false);
  version_builder.Apply(&version_edit);
  version_builder.SaveTo(&new_vstorage);

@@ -229,7 +229,7 @@ TEST_F(VersionBuilderTest, ApplyMultipleAndSaveTo) {
  VersionBuilder version_builder(env_options, nullptr, &vstorage_);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
-                                  kCompactionStyleLevel, nullptr);
+                                  kCompactionStyleLevel, nullptr, false);
  version_builder.Apply(&version_edit);
  version_builder.SaveTo(&new_vstorage);

@@ -244,7 +244,7 @@ TEST_F(VersionBuilderTest, ApplyDeleteAndSaveTo) {
  EnvOptions env_options;
  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
-                                  kCompactionStyleLevel, nullptr);
+                                  kCompactionStyleLevel, nullptr, false);

  VersionEdit version_edit;
  version_edit.AddFile(2, 666, 0, 100U, GetInternalKey("301"),
@@ -446,6 +446,10 @@ class LevelFileNumIterator : public InternalIterator {
  virtual void Seek(const Slice& target) override {
    index_ = FindFile(icmp_, *flevel_, target);
  }
+  virtual void SeekForPrev(const Slice& target) override {
+    SeekForPrevImpl(target, &icmp_);
+  }
+
  virtual void SeekToFirst() override { index_ = 0; }
  virtual void SeekToLast() override {
    index_ = (flevel_->num_files == 0)
@@ -493,7 +497,8 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
                         const EnvOptions& env_options,
                         const InternalKeyComparator& icomparator,
                         HistogramImpl* file_read_hist, bool for_compaction,
-                         bool prefix_enabled, bool skip_filters, int level)
+                         bool prefix_enabled, bool skip_filters, int level,
+                         RangeDelAggregator* range_del_agg)
      : TwoLevelIteratorState(prefix_enabled),
        table_cache_(table_cache),
        read_options_(read_options),
@@ -502,7 +507,8 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
        file_read_hist_(file_read_hist),
        for_compaction_(for_compaction),
        skip_filters_(skip_filters),
-        level_(level) {}
+        level_(level),
+        range_del_agg_(range_del_agg) {}

  InternalIterator* NewSecondaryIterator(const Slice& meta_handle) override {
    if (meta_handle.size() != sizeof(FileDescriptor)) {
@@ -514,7 +520,8 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
      return table_cache_->NewIterator(
          read_options_, env_options_, icomparator_, *fd,
          nullptr /* don't need reference to table*/, file_read_hist_,
-          for_compaction_, nullptr /* arena */, skip_filters_, level_);
+          for_compaction_, nullptr /* arena */, skip_filters_, level_,
+          range_del_agg_, false /* is_range_del_only */);
    }
  }

@@ -531,6 +538,7 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
  bool for_compaction_;
  bool skip_filters_;
  int level_;
+  RangeDelAggregator* range_del_agg_;
 };

 // A wrapper of version builder which references the current version in
@@ -800,48 +808,59 @@ void Version::AddIterators(const ReadOptions& read_options,
                           MergeIteratorBuilder* merge_iter_builder) {
  assert(storage_info_.finalized_);

-  if (storage_info_.num_non_empty_levels() == 0) {
-    // No file in the Version.
+  for (int level = 0; level < storage_info_.num_non_empty_levels(); level++) {
+    AddIteratorsForLevel(read_options, soptions, merge_iter_builder, level);
+  }
+}
+
+void Version::AddIteratorsForLevel(const ReadOptions& read_options,
+                                   const EnvOptions& soptions,
+                                   MergeIteratorBuilder* merge_iter_builder,
+                                   int level) {
+  assert(storage_info_.finalized_);
+  if (level >= storage_info_.num_non_empty_levels()) {
+    // This is an empty level
+    return;
+  } else if (storage_info_.LevelFilesBrief(level).num_files == 0) {
+    // No files in this level
    return;
  }

  auto* arena = merge_iter_builder->GetArena();
-
-  // Merge all level zero files together since they may overlap
-  for (size_t i = 0; i < storage_info_.LevelFilesBrief(0).num_files; i++) {
-    const auto& file = storage_info_.LevelFilesBrief(0).files[i];
-    merge_iter_builder->AddIterator(cfd_->table_cache()->NewIterator(
-        read_options, soptions, cfd_->internal_comparator(), file.fd, nullptr,
-        cfd_->internal_stats()->GetFileReadHist(0), false, arena,
-        false /* skip_filters */, 0 /* level */));
-  }
-
-  // For levels > 0, we can use a concatenating iterator that sequentially
-  // walks through the non-overlapping files in the level, opening them
-  // lazily.
-  for (int level = 1; level < storage_info_.num_non_empty_levels(); level++) {
-    if (storage_info_.LevelFilesBrief(level).num_files != 0) {
-      auto* mem = arena->AllocateAligned(sizeof(LevelFileIteratorState));
-      auto* state = new (mem)
-          LevelFileIteratorState(cfd_->table_cache(), read_options, soptions,
-                                 cfd_->internal_comparator(),
-                                 cfd_->internal_stats()->GetFileReadHist(level),
-                                 false /* for_compaction */,
-                                 cfd_->ioptions()->prefix_extractor != nullptr,
-                                 IsFilterSkipped(level), level);
-      mem = arena->AllocateAligned(sizeof(LevelFileNumIterator));
-      auto* first_level_iter = new (mem) LevelFileNumIterator(
-          cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level));
-      merge_iter_builder->AddIterator(
-          NewTwoLevelIterator(state, first_level_iter, arena, false));
+  if (level == 0) {
+    // Merge all level zero files together since they may overlap
+    for (size_t i = 0; i < storage_info_.LevelFilesBrief(0).num_files; i++) {
+      const auto& file = storage_info_.LevelFilesBrief(0).files[i];
+      merge_iter_builder->AddIterator(cfd_->table_cache()->NewIterator(
+          read_options, soptions, cfd_->internal_comparator(), file.fd, nullptr,
+          cfd_->internal_stats()->GetFileReadHist(0), false, arena,
+          false /* skip_filters */, 0 /* level */));
    }
+  } else {
+    // For levels > 0, we can use a concatenating iterator that sequentially
+    // walks through the non-overlapping files in the level, opening them
+    // lazily.
+    auto* mem = arena->AllocateAligned(sizeof(LevelFileIteratorState));
+    auto* state = new (mem) LevelFileIteratorState(
+        cfd_->table_cache(), read_options, soptions,
+        cfd_->internal_comparator(),
+        cfd_->internal_stats()->GetFileReadHist(level),
+        false /* for_compaction */,
+        cfd_->ioptions()->prefix_extractor != nullptr, IsFilterSkipped(level),
+        level, nullptr /* range_del_agg */);
+    mem = arena->AllocateAligned(sizeof(LevelFileNumIterator));
+    auto* first_level_iter = new (mem) LevelFileNumIterator(
+        cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level));
+    merge_iter_builder->AddIterator(
+        NewTwoLevelIterator(state, first_level_iter, arena, false));
  }
 }

 VersionStorageInfo::VersionStorageInfo(
    const InternalKeyComparator* internal_comparator,
    const Comparator* user_comparator, int levels,
-    CompactionStyle compaction_style, VersionStorageInfo* ref_vstorage)
+    CompactionStyle compaction_style, VersionStorageInfo* ref_vstorage,
+    bool _force_consistency_checks)
    : internal_comparator_(internal_comparator),
      user_comparator_(user_comparator),
      // cfd is nullptr if Version is dummy
@@ -866,7 +885,8 @@ VersionStorageInfo::VersionStorageInfo(
      current_num_deletions_(0),
      current_num_samples_(0),
      estimated_compaction_needed_bytes_(0),
-      finalized_(false) {
+      finalized_(false),
+      force_consistency_checks_(_force_consistency_checks) {
  if (ref_vstorage != nullptr) {
    accumulated_file_size_ = ref_vstorage->accumulated_file_size_;
    accumulated_raw_key_size_ = ref_vstorage->accumulated_raw_key_size_;
@@ -890,14 +910,16 @@ Version::Version(ColumnFamilyData* column_family_data, VersionSet* vset,
      table_cache_((cfd_ == nullptr) ? nullptr : cfd_->table_cache()),
      merge_operator_((cfd_ == nullptr) ? nullptr
                                        : cfd_->ioptions()->merge_operator),
-      storage_info_((cfd_ == nullptr) ? nullptr : &cfd_->internal_comparator(),
-                    (cfd_ == nullptr) ? nullptr : cfd_->user_comparator(),
-                    cfd_ == nullptr ? 0 : cfd_->NumberLevels(),
-                    cfd_ == nullptr ? kCompactionStyleLevel
-                                    : cfd_->ioptions()->compaction_style,
-                    (cfd_ == nullptr || cfd_->current() == nullptr)
-                        ? nullptr
-                        : cfd_->current()->storage_info()),
+      storage_info_(
+          (cfd_ == nullptr) ? nullptr : &cfd_->internal_comparator(),
+          (cfd_ == nullptr) ? nullptr : cfd_->user_comparator(),
+          cfd_ == nullptr ? 0 : cfd_->NumberLevels(),
+          cfd_ == nullptr ? kCompactionStyleLevel
+                          : cfd_->ioptions()->compaction_style,
+          (cfd_ == nullptr || cfd_->current() == nullptr)
+              ? nullptr
+              : cfd_->current()->storage_info(),
+          cfd_ == nullptr ? false : cfd_->ioptions()->force_consistency_checks),
      vset_(vset),
      next_(this),
      prev_(this),
@@ -1014,7 +1036,7 @@ void Version::PrepareApply(
  UpdateAccumulatedStats(update_stats);
  storage_info_.UpdateNumNonEmptyLevels();
  storage_info_.CalculateBaseBytes(*cfd_->ioptions(), mutable_cf_options);
-  storage_info_.UpdateFilesByCompactionPri(mutable_cf_options);
+  storage_info_.UpdateFilesByCompactionPri(cfd_->ioptions()->compaction_pri);
  storage_info_.GenerateFileIndexer();
  storage_info_.GenerateLevelFilesBrief();
  storage_info_.GenerateLevel0NonOverlapping();
@@ -1240,6 +1262,7 @@ void VersionStorageInfo::EstimateCompactionBytesNeeded(
 }

 void VersionStorageInfo::ComputeCompactionScore(
+    const ImmutableCFOptions& immutable_cf_options,
    const MutableCFOptions& mutable_cf_options) {
  for (int level = 0; level <= MaxInputLevel(); level++) {
    double score;
@@ -1275,8 +1298,9 @@ void VersionStorageInfo::ComputeCompactionScore(
      }

      if (compaction_style_ == kCompactionStyleFIFO) {
-        score = static_cast<double>(total_size) /
-                mutable_cf_options.compaction_options_fifo.max_table_files_size;
+        score =
+            static_cast<double>(total_size) /
+            immutable_cf_options.compaction_options_fifo.max_table_files_size;
      } else {
        score = static_cast<double>(num_sorted_runs) /
                mutable_cf_options.level0_file_num_compaction_trigger;
@@ -1476,7 +1500,7 @@ void SortFileByOverlappingRatio(
 }  // namespace

 void VersionStorageInfo::UpdateFilesByCompactionPri(
-    const MutableCFOptions& mutable_cf_options) {
+    CompactionPri compaction_pri) {
  if (compaction_style_ == kCompactionStyleFIFO ||
      compaction_style_ == kCompactionStyleUniversal) {
    // don't need this
@@ -1500,7 +1524,7 @@ void VersionStorageInfo::UpdateFilesByCompactionPri(
    if (num > temp.size()) {
      num = temp.size();
    }
-    switch (mutable_cf_options.compaction_pri) {
+    switch (compaction_pri) {
      case kByCompensatedSize:
        std::partial_sort(temp.begin(), temp.begin() + num, temp.end(),
                          CompareCompensatedSizeDescending);
@@ -2091,7 +2115,8 @@ struct VersionSet::ManifestWriter {
      : done(false), cv(mu), cfd(_cfd), edit_list(e) {}
 };

-VersionSet::VersionSet(const std::string& dbname, const DBOptions* db_options,
+VersionSet::VersionSet(const std::string& dbname,
+                       const ImmutableDBOptions* db_options,
                       const EnvOptions& storage_options, Cache* table_cache,
                       WriteBufferManager* write_buffer_manager,
                       WriteController* write_controller)
@@ -2132,6 +2157,7 @@ void VersionSet::AppendVersion(ColumnFamilyData* column_family_data,
                               Version* v) {
  // compute new compaction score
  v->storage_info()->ComputeCompactionScore(
+      *column_family_data->ioptions(),
      *column_family_data->GetLatestMutableCFOptions());

  // Mark v finalized
@@ -2332,8 +2358,9 @@ Status VersionSet::LogAndApply(ColumnFamilyData* column_family_data,
    // If we just created a new descriptor file, install it by writing a
    // new CURRENT file that points to it.
    if (s.ok() && new_descriptor_log) {
-      s = SetCurrentFile(env_, dbname_, pending_manifest_file_number_,
-                         db_options_->disableDataSync ? nullptr : db_directory);
+      s = SetCurrentFile(
+          env_, dbname_, pending_manifest_file_number_,
+          db_options_->disable_data_sync ? nullptr : db_directory);
    }

    if (s.ok()) {
@@ -2844,12 +2871,13 @@ Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
        "Number of levels needs to be bigger than 1");
  }

+  ImmutableDBOptions db_options(*options);
  ColumnFamilyOptions cf_options(*options);
  std::shared_ptr<Cache> tc(NewLRUCache(options->max_open_files - 10,
                                        options->table_cache_numshardbits));
  WriteController wc(options->delayed_write_rate);
  WriteBufferManager wb(options->db_write_buffer_size);
-  VersionSet versions(dbname, options, env_options, tc.get(), &wb, &wc);
+  VersionSet versions(dbname, &db_options, env_options, tc.get(), &wb, &wc);
  Status status;

  std::vector<ColumnFamilyDescriptor> dummy;
@@ -2909,7 +2937,7 @@ Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
  vstorage->files_ = new_files_list;
  vstorage->num_levels_ = new_levels;

-  MutableCFOptions mutable_cf_options(*options, ImmutableCFOptions(*options));
+  MutableCFOptions mutable_cf_options(*options);
  VersionEdit ve;
  InstrumentedMutex dummy_mutex;
  InstrumentedMutexLock l(&dummy_mutex);
@@ -3329,7 +3357,8 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
  }
 }

-InternalIterator* VersionSet::MakeInputIterator(const Compaction* c) {
+InternalIterator* VersionSet::MakeInputIterator(
+    const Compaction* c, RangeDelAggregator* range_del_agg) {
  auto cfd = c->column_family_data();
  ReadOptions read_options;
  read_options.verify_checksums =
@@ -3357,7 +3386,7 @@ InternalIterator* VersionSet::MakeInputIterator(const Compaction* c) {
              cfd->internal_comparator(), flevel->files[i].fd, nullptr,
              nullptr, /* no per level latency histogram*/
              true /* for_compaction */, nullptr /* arena */,
-              false /* skip_filters */, (int)which /* level */);
+              false /* skip_filters */, (int)which /* level */, range_del_agg);
        }
      } else {
        // Create concatenating iterator for the files from this level
@@ -3367,7 +3396,8 @@ InternalIterator* VersionSet::MakeInputIterator(const Compaction* c) {
                cfd->internal_comparator(),
                nullptr /* no per level latency histogram */,
                true /* for_compaction */, false /* prefix enabled */,
-                false /* skip_filters */, (int)which /* level */),
+                false /* skip_filters */, (int)which /* level */,
+                range_del_agg),
            new LevelFileNumIterator(cfd->internal_comparator(),
                                     c->input_levels(which)));
      }
@@ -28,18 +28,19 @@
 #include <utility>
 #include <vector>

-#include "db/dbformat.h"
-#include "db/version_builder.h"
-#include "db/version_edit.h"
-#include "port/port.h"
-#include "db/table_cache.h"
+#include "db/column_family.h"
 #include "db/compaction.h"
 #include "db/compaction_picker.h"
-#include "db/column_family.h"
-#include "db/log_reader.h"
+#include "db/dbformat.h"
 #include "db/file_indexer.h"
+#include "db/log_reader.h"
+#include "db/table_cache.h"
+#include "db/version_builder.h"
+#include "db/version_edit.h"
 #include "db/write_controller.h"
+#include "port/port.h"
 #include "rocksdb/env.h"
+#include "util/db_options.h"
 #include "util/instrumented_mutex.h"

 namespace rocksdb {
@@ -92,7 +93,8 @@ class VersionStorageInfo {
  VersionStorageInfo(const InternalKeyComparator* internal_comparator,
                     const Comparator* user_comparator, int num_levels,
                     CompactionStyle compaction_style,
-                     VersionStorageInfo* src_vstorage);
+                     VersionStorageInfo* src_vstorage,
+                     bool _force_consistency_checks);
  ~VersionStorageInfo();

  void Reserve(int level, size_t size) { files_[level].reserve(size); }
@@ -120,7 +122,8 @@ class VersionStorageInfo {
  // We use compaction scores to figure out which compaction to do next
  // REQUIRES: db_mutex held!!
  // TODO find a better way to pass compaction_options_fifo.
-  void ComputeCompactionScore(const MutableCFOptions& mutable_cf_options);
+  void ComputeCompactionScore(const ImmutableCFOptions& immutable_cf_options,
+                              const MutableCFOptions& mutable_cf_options);

  // Estimate est_comp_needed_bytes_
  void EstimateCompactionBytesNeeded(
@@ -135,7 +138,7 @@ class VersionStorageInfo {
  // Sort all files for this version based on their file size and
  // record results in files_by_compaction_pri_. The largest files are listed
  // first.
-  void UpdateFilesByCompactionPri(const MutableCFOptions& mutable_cf_options);
+  void UpdateFilesByCompactionPri(CompactionPri compaction_pri);

  void GenerateLevel0NonOverlapping();
  bool level0_non_overlapping() const {
@@ -329,6 +332,8 @@ class VersionStorageInfo {
    estimated_compaction_needed_bytes_ = v;
  }

+  bool force_consistency_checks() const { return force_consistency_checks_; }
+
 private:
  const InternalKeyComparator* internal_comparator_;
  const Comparator* user_comparator_;
@@ -411,6 +416,10 @@ class VersionStorageInfo {

  bool finalized_;

+  // If set to true, we will run consistency checks even if RocksDB
+  // is compiled in release mode
+  bool force_consistency_checks_;
+
  friend class Version;
  friend class VersionSet;
  // No copying allowed
@@ -426,6 +435,10 @@ class Version {
  void AddIterators(const ReadOptions&, const EnvOptions& soptions,
                    MergeIteratorBuilder* merger_iter_builder);

+  void AddIteratorsForLevel(const ReadOptions&, const EnvOptions& soptions,
+                            MergeIteratorBuilder* merger_iter_builder,
+                            int level);
+
  // Lookup the value for key.  If found, store it in *val and
  // return OK.  Else return a non-OK status.
  // Uses *operands to store merge_operator operations to apply later.
@@ -573,7 +586,7 @@ class Version {

 class VersionSet {
 public:
-  VersionSet(const std::string& dbname, const DBOptions* db_options,
+  VersionSet(const std::string& dbname, const ImmutableDBOptions* db_options,
             const EnvOptions& env_options, Cache* table_cache,
             WriteBufferManager* write_buffer_manager,
             WriteController* write_controller);
@@ -686,7 +699,8 @@ class VersionSet {

  // Create an iterator that reads over the compaction inputs for "*c".
  // The caller should delete the iterator when no longer needed.
-  InternalIterator* MakeInputIterator(const Compaction* c);
+  InternalIterator* MakeInputIterator(const Compaction* c,
+                                      RangeDelAggregator* range_del_agg);

  // Add all files listed in any live version to *live.
  void AddLiveFiles(std::vector<FileDescriptor>* live_list);
@@ -755,7 +769,7 @@ class VersionSet {

  Env* const env_;
  const std::string dbname_;
-  const DBOptions* const db_options_;
+  const ImmutableDBOptions* const db_options_;
  std::atomic<uint64_t> next_file_number_;
  uint64_t manifest_file_number_;
  uint64_t options_file_number_;
@@ -109,8 +109,8 @@ class VersionStorageInfoTest : public testing::Test {
        logger_(new CountingLogger()),
        options_(GetOptionsWithNumLevels(6, logger_)),
        ioptions_(options_),
-        mutable_cf_options_(options_, ioptions_),
-        vstorage_(&icmp_, ucmp_, 6, kCompactionStyleLevel, nullptr) {}
+        mutable_cf_options_(options_),
+        vstorage_(&icmp_, ucmp_, 6, kCompactionStyleLevel, nullptr, false) {}

  ~VersionStorageInfoTest() {
    for (int i = 0; i < vstorage_.num_levels(); i++) {
@@ -128,8 +128,8 @@ Status WalManager::GetUpdatesSince(
 //    b. get sorted non-empty archived logs
 //    c. delete what should be deleted
 void WalManager::PurgeObsoleteWALFiles() {
-  bool const ttl_enabled = db_options_.WAL_ttl_seconds > 0;
-  bool const size_limit_enabled = db_options_.WAL_size_limit_MB > 0;
+  bool const ttl_enabled = db_options_.wal_ttl_seconds > 0;
+  bool const size_limit_enabled = db_options_.wal_size_limit_mb > 0;
  if (!ttl_enabled && !size_limit_enabled) {
    return;
  }
@@ -144,7 +144,7 @@ void WalManager::PurgeObsoleteWALFiles() {
  }
  uint64_t const now_seconds = static_cast<uint64_t>(current_time);
  uint64_t const time_to_check = (ttl_enabled && !size_limit_enabled)
-                                     ? db_options_.WAL_ttl_seconds / 2
+                                     ? db_options_.wal_ttl_seconds / 2
                                     : kDefaultIntervalToDeleteObsoleteWAL;

  if (purge_wal_files_last_run_ + time_to_check > now_seconds) {
@@ -180,7 +180,7 @@ void WalManager::PurgeObsoleteWALFiles() {
              file_path.c_str(), s.ToString().c_str());
          continue;
        }
-        if (now_seconds - file_m_time > db_options_.WAL_ttl_seconds) {
+        if (now_seconds - file_m_time > db_options_.wal_ttl_seconds) {
          s = env_->DeleteFile(file_path);
          if (!s.ok()) {
            Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log,
@@ -229,7 +229,7 @@ void WalManager::PurgeObsoleteWALFiles() {
  }

  size_t const files_keep_num =
-      db_options_.WAL_size_limit_MB * 1024 * 1024 / log_file_size;
+      db_options_.wal_size_limit_mb * 1024 * 1024 / log_file_size;
  if (log_files_num <= files_keep_num) {
    return;
  }
@@ -383,7 +383,7 @@ Status WalManager::ReadFirstRecord(const WalFileType type,
  Status s;
  if (type == kAliveLogFile) {
    std::string fname = LogFileName(db_options_.wal_dir, number);
-    s = ReadFirstLine(fname, sequence);
+    s = ReadFirstLine(fname, number, sequence);
    if (env_->FileExists(fname).ok() && !s.ok()) {
      // return any error that is not caused by non-existing file
      return s;
@@ -394,7 +394,7 @@ Status WalManager::ReadFirstRecord(const WalFileType type,
    //  check if the file got moved to archive.
    std::string archived_file =
        ArchivedLogFileName(db_options_.wal_dir, number);
-    s = ReadFirstLine(archived_file, sequence);
+    s = ReadFirstLine(archived_file, number, sequence);
    // maybe the file was deleted from archive dir. If that's the case, return
    // Status::OK(). The caller with identify this as empty file because
    // *sequence == 0
@@ -413,6 +413,7 @@ Status WalManager::ReadFirstRecord(const WalFileType type,
 // the function returns status.ok() and sequence == 0 if the file exists, but is
 // empty
 Status WalManager::ReadFirstLine(const std::string& fname,
+                                 const uint64_t number,
                                 SequenceNumber* sequence) {
  struct LogReporter : public log::Reader::Reporter {
    Env* env;
@@ -449,7 +450,7 @@ Status WalManager::ReadFirstLine(const std::string& fname,
  reporter.status = &status;
  reporter.ignore_error = !db_options_.paranoid_checks;
  log::Reader reader(db_options_.info_log, std::move(file_reader), &reporter,
-                     true /*checksum*/, 0 /*initial_offset*/, *sequence);
+                     true /*checksum*/, 0 /*initial_offset*/, number);
  std::string scratch;
  Slice record;

@@ -17,22 +17,21 @@
 #include <string>
 #include <memory>

-#include "port/port.h"
-
-#include "rocksdb/env.h"
-#include "rocksdb/options.h"
-#include "rocksdb/types.h"
-#include "rocksdb/transaction_log.h"
-#include "rocksdb/status.h"
-
 #include "db/version_set.h"
+#include "port/port.h"
+#include "rocksdb/env.h"
+#include "rocksdb/status.h"
+#include "rocksdb/transaction_log.h"
+#include "rocksdb/types.h"
+#include "util/db_options.h"

 namespace rocksdb {

 #ifndef ROCKSDB_LITE
 class WalManager {
 public:
-  WalManager(const DBOptions& db_options, const EnvOptions& env_options)
+  WalManager(const ImmutableDBOptions& db_options,
+             const EnvOptions& env_options)
      : db_options_(db_options),
        env_options_(env_options),
        env_(db_options.env),
@@ -54,9 +53,9 @@ class WalManager {
    return ReadFirstRecord(type, number, sequence);
  }

-  Status TEST_ReadFirstLine(const std::string& fname,
+  Status TEST_ReadFirstLine(const std::string& fname, const uint64_t number,
                            SequenceNumber* sequence) {
-    return ReadFirstLine(fname, sequence);
+    return ReadFirstLine(fname, number, sequence);
  }

 private:
@@ -71,10 +70,11 @@ class WalManager {
  Status ReadFirstRecord(const WalFileType type, const uint64_t number,
                         SequenceNumber* sequence);

-  Status ReadFirstLine(const std::string& fname, SequenceNumber* sequence);
+  Status ReadFirstLine(const std::string& fname, const uint64_t number,
+                       SequenceNumber* sequence);

  // ------- state from DBImpl ------
-  const DBOptions& db_options_;
+  const ImmutableDBOptions& db_options_;
  const EnvOptions& env_options_;
  Env* env_;

@@ -33,6 +33,7 @@ class WalManagerTest : public testing::Test {
  WalManagerTest()
      : env_(new MockEnv(Env::Default())),
        dbname_(test::TmpDir() + "/wal_manager_test"),
+        db_options_(),
        table_cache_(NewLRUCache(50000, 16)),
        write_buffer_manager_(db_options_.db_write_buffer_size),
        current_log_number_(0) {
@@ -100,10 +101,10 @@ class WalManagerTest : public testing::Test {

  std::unique_ptr<MockEnv> env_;
  std::string dbname_;
+  ImmutableDBOptions db_options_;
  WriteController write_controller_;
  EnvOptions env_options_;
  std::shared_ptr<Cache> table_cache_;
-  DBOptions db_options_;
  WriteBufferManager write_buffer_manager_;
  std::unique_ptr<VersionSet> versions_;
  std::unique_ptr<WalManager> wal_manager_;
@@ -119,10 +120,11 @@ TEST_F(WalManagerTest, ReadFirstRecordCache) {
  ASSERT_OK(env_->NewWritableFile(path, &file, EnvOptions()));

  SequenceNumber s;
-  ASSERT_OK(wal_manager_->TEST_ReadFirstLine(path, &s));
+  ASSERT_OK(wal_manager_->TEST_ReadFirstLine(path, 1 /* number */, &s));
  ASSERT_EQ(s, 0U);

-  ASSERT_OK(wal_manager_->TEST_ReadFirstRecord(kAliveLogFile, 1, &s));
+  ASSERT_OK(
+      wal_manager_->TEST_ReadFirstRecord(kAliveLogFile, 1 /* number */, &s));
  ASSERT_EQ(s, 0U);

  unique_ptr<WritableFileWriter> file_writer(
@@ -204,8 +206,8 @@ int CountRecords(TransactionLogIterator* iter) {
 }  // namespace

 TEST_F(WalManagerTest, WALArchivalSizeLimit) {
-  db_options_.WAL_ttl_seconds = 0;
-  db_options_.WAL_size_limit_MB = 1000;
+  db_options_.wal_ttl_seconds = 0;
+  db_options_.wal_size_limit_mb = 1000;
  Init();

  // TEST : Create WalManager with huge size limit and no ttl.
@@ -213,7 +215,7 @@ TEST_F(WalManagerTest, WALArchivalSizeLimit) {
  // Count the archived log files that survived.
  // Assert that all of them did.
  // Change size limit. Re-open WalManager.
-  // Assert that archive is not greater than WAL_size_limit_MB after
+  // Assert that archive is not greater than wal_size_limit_mb after
  // PurgeObsoleteWALFiles()
  // Set ttl and time_to_check_ to small values. Re-open db.
  // Assert that there are no archived logs left.
@@ -225,14 +227,14 @@ TEST_F(WalManagerTest, WALArchivalSizeLimit) {
      ListSpecificFiles(env_.get(), archive_dir, kLogFile);
  ASSERT_EQ(log_files.size(), 20U);

-  db_options_.WAL_size_limit_MB = 8;
+  db_options_.wal_size_limit_mb = 8;
  Reopen();
  wal_manager_->PurgeObsoleteWALFiles();

  uint64_t archive_size = GetLogDirSize(archive_dir, env_.get());
-  ASSERT_TRUE(archive_size <= db_options_.WAL_size_limit_MB * 1024 * 1024);
+  ASSERT_TRUE(archive_size <= db_options_.wal_size_limit_mb * 1024 * 1024);

-  db_options_.WAL_ttl_seconds = 1;
+  db_options_.wal_ttl_seconds = 1;
  env_->FakeSleepForMicroseconds(2 * 1000 * 1000);
  Reopen();
  wal_manager_->PurgeObsoleteWALFiles();
@@ -242,7 +244,7 @@ TEST_F(WalManagerTest, WALArchivalSizeLimit) {
 }

 TEST_F(WalManagerTest, WALArchivalTtl) {
-  db_options_.WAL_ttl_seconds = 1000;
+  db_options_.wal_ttl_seconds = 1000;
  Init();

  // TEST : Create WalManager with a ttl and no size limit.
@@ -258,7 +260,7 @@ TEST_F(WalManagerTest, WALArchivalTtl) {
      ListSpecificFiles(env_.get(), archive_dir, kLogFile);
  ASSERT_GT(log_files.size(), 0U);

-  db_options_.WAL_ttl_seconds = 1;
+  db_options_.wal_ttl_seconds = 1;
  env_->FakeSleepForMicroseconds(3 * 1000 * 1000);
  Reopen();
  wal_manager_->PurgeObsoleteWALFiles();
@@ -118,13 +118,6 @@ struct BatchContentClassifier : public WriteBatch::Handler {

 }  // anon namespace

-
-struct SavePoint {
-  size_t size;  // size of rep_
-  int count;    // count of elements in rep_
-  uint32_t content_flags;
-};
-
 struct SavePoints {
  std::stack<SavePoint> stack;
 };
@@ -143,11 +136,13 @@ WriteBatch::WriteBatch(const std::string& rep)

 WriteBatch::WriteBatch(const WriteBatch& src)
    : save_points_(src.save_points_),
+      wal_term_point_(src.wal_term_point_),
      content_flags_(src.content_flags_.load(std::memory_order_relaxed)),
      rep_(src.rep_) {}

 WriteBatch::WriteBatch(WriteBatch&& src)
    : save_points_(std::move(src.save_points_)),
+      wal_term_point_(std::move(src.wal_term_point_)),
      content_flags_(src.content_flags_.load(std::memory_order_relaxed)),
      rep_(std::move(src.rep_)) {}

@@ -191,6 +186,8 @@ void WriteBatch::Clear() {
      save_points_->stack.pop();
    }
  }
+
+  wal_term_point_.clear();
 }

 int WriteBatch::Count() const {
@@ -213,6 +210,12 @@ uint32_t WriteBatch::ComputeContentFlags() const {
  return rv;
 }

+void WriteBatch::MarkWalTerminationPoint() {
+  wal_term_point_.size = GetDataSize();
+  wal_term_point_.count = Count();
+  wal_term_point_.content_flags = content_flags_;
+}
+
 bool WriteBatch::HasPut() const {
  return (ComputeContentFlags() & ContentFlags::HAS_PUT) != 0;
 }
@@ -729,8 +732,8 @@ void WriteBatch::SetSavePoint() {
    save_points_ = new SavePoints();
  }
  // Record length and count of current batch of writes.
-  save_points_->stack.push(SavePoint{
-      GetDataSize(), Count(), content_flags_.load(std::memory_order_relaxed)});
+  save_points_->stack.push(SavePoint(
+      GetDataSize(), Count(), content_flags_.load(std::memory_order_relaxed)));
 }

 Status WriteBatch::RollbackToSavePoint() {
@@ -1244,14 +1247,29 @@ void WriteBatchInternal::SetContents(WriteBatch* b, const Slice& contents) {
  b->content_flags_.store(ContentFlags::DEFERRED, std::memory_order_relaxed);
 }

-void WriteBatchInternal::Append(WriteBatch* dst, const WriteBatch* src) {
-  SetCount(dst, Count(dst) + Count(src));
+void WriteBatchInternal::Append(WriteBatch* dst, const WriteBatch* src,
+                                const bool wal_only) {
+  size_t src_len;
+  int src_count;
+  uint32_t src_flags;
+
+  const SavePoint& batch_end = src->GetWalTerminationPoint();
+
+  if (wal_only && !batch_end.is_cleared()) {
+    src_len = batch_end.size - WriteBatchInternal::kHeader;
+    src_count = batch_end.count;
+    src_flags = batch_end.content_flags;
+  } else {
+    src_len = src->rep_.size() - WriteBatchInternal::kHeader;
+    src_count = Count(src);
+    src_flags = src->content_flags_.load(std::memory_order_relaxed);
+  }
+
+  SetCount(dst, Count(dst) + src_count);
  assert(src->rep_.size() >= WriteBatchInternal::kHeader);
-  dst->rep_.append(src->rep_.data() + WriteBatchInternal::kHeader,
-    src->rep_.size() - WriteBatchInternal::kHeader);
+  dst->rep_.append(src->rep_.data() + WriteBatchInternal::kHeader, src_len);
  dst->content_flags_.store(
-      dst->content_flags_.load(std::memory_order_relaxed) |
-          src->content_flags_.load(std::memory_order_relaxed),
+      dst->content_flags_.load(std::memory_order_relaxed) | src_flags,
      std::memory_order_relaxed);
 }

@@ -177,7 +177,8 @@ class WriteBatchInternal {
                           uint64_t log_number = 0, DB* db = nullptr,
                           bool concurrent_memtable_writes = false);

-  static void Append(WriteBatch* dst, const WriteBatch* src);
+  static void Append(WriteBatch* dst, const WriteBatch* src,
+                     const bool WAL_only = false);

  // Returns the byte size of appending a WriteBatch with ByteSize
  // leftByteSize and a WriteBatch with ByteSize rightByteSize
@@ -31,9 +31,8 @@ static std::string PrintContents(WriteBatch* b) {
  options.memtable_factory = factory;
  ImmutableCFOptions ioptions(options);
  WriteBufferManager wb(options.db_write_buffer_size);
-  MemTable* mem =
-      new MemTable(cmp, ioptions, MutableCFOptions(options, ioptions), &wb,
-                   kMaxSequenceNumber);
+  MemTable* mem = new MemTable(cmp, ioptions, MutableCFOptions(options), &wb,
+                               kMaxSequenceNumber);
  mem->Ref();
  std::string state;
  ColumnFamilyMemTablesDefault cf_mems_default(mem);
@@ -44,60 +43,65 @@ static std::string PrintContents(WriteBatch* b) {
  int single_delete_count = 0;
  int delete_range_count = 0;
  int merge_count = 0;
-  Arena arena;
-  ScopedArenaIterator iter(mem->NewIterator(ReadOptions(), &arena));
-  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-    ParsedInternalKey ikey;
-    memset((void *)&ikey, 0, sizeof(ikey));
-    EXPECT_TRUE(ParseInternalKey(iter->key(), &ikey));
-    switch (ikey.type) {
-      case kTypeValue:
-        state.append("Put(");
-        state.append(ikey.user_key.ToString());
-        state.append(", ");
-        state.append(iter->value().ToString());
-        state.append(")");
-        count++;
-        put_count++;
-        break;
-      case kTypeDeletion:
-        state.append("Delete(");
-        state.append(ikey.user_key.ToString());
-        state.append(")");
-        count++;
-        delete_count++;
-        break;
-      case kTypeSingleDeletion:
-        state.append("SingleDelete(");
-        state.append(ikey.user_key.ToString());
-        state.append(")");
-        count++;
-        single_delete_count++;
-        break;
-      case kTypeRangeDeletion:
-        state.append("DeleteRange(");
-        state.append(ikey.user_key.ToString());
-        state.append(", ");
-        state.append(iter->value().ToString());
-        state.append(")");
-        count++;
-        delete_range_count++;
-        break;
-      case kTypeMerge:
-        state.append("Merge(");
-        state.append(ikey.user_key.ToString());
-        state.append(", ");
-        state.append(iter->value().ToString());
-        state.append(")");
-        count++;
-        merge_count++;
-        break;
-      default:
-        assert(false);
-        break;
+  for (int i = 0; i < 2; ++i) {
+    Arena arena;
+    auto iter =
+        i == 0 ? ScopedArenaIterator(mem->NewIterator(ReadOptions(), &arena))
+               : ScopedArenaIterator(
+                     mem->NewRangeTombstoneIterator(ReadOptions(), &arena));
+    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+      ParsedInternalKey ikey;
+      memset((void*)&ikey, 0, sizeof(ikey));
+      EXPECT_TRUE(ParseInternalKey(iter->key(), &ikey));
+      switch (ikey.type) {
+        case kTypeValue:
+          state.append("Put(");
+          state.append(ikey.user_key.ToString());
+          state.append(", ");
+          state.append(iter->value().ToString());
+          state.append(")");
+          count++;
+          put_count++;
+          break;
+        case kTypeDeletion:
+          state.append("Delete(");
+          state.append(ikey.user_key.ToString());
+          state.append(")");
+          count++;
+          delete_count++;
+          break;
+        case kTypeSingleDeletion:
+          state.append("SingleDelete(");
+          state.append(ikey.user_key.ToString());
+          state.append(")");
+          count++;
+          single_delete_count++;
+          break;
+        case kTypeRangeDeletion:
+          state.append("DeleteRange(");
+          state.append(ikey.user_key.ToString());
+          state.append(", ");
+          state.append(iter->value().ToString());
+          state.append(")");
+          count++;
+          delete_range_count++;
+          break;
+        case kTypeMerge:
+          state.append("Merge(");
+          state.append(ikey.user_key.ToString());
+          state.append(", ");
+          state.append(iter->value().ToString());
+          state.append(")");
+          count++;
+          merge_count++;
+          break;
+        default:
+          assert(false);
+          break;
+      }
+      state.append("@");
+      state.append(NumberToString(ikey.sequence));
    }
-    state.append("@");
-    state.append(NumberToString(ikey.sequence));
  }
  EXPECT_EQ(b->HasPut(), put_count > 0);
  EXPECT_EQ(b->HasDelete(), delete_count > 0);
@@ -132,10 +136,10 @@ TEST_F(WriteBatchTest, Multiple) {
  ASSERT_EQ(100U, WriteBatchInternal::Sequence(&batch));
  ASSERT_EQ(4, WriteBatchInternal::Count(&batch));
  ASSERT_EQ(
-      "DeleteRange(bar, foo)@102"
      "Put(baz, boo)@103"
      "Delete(box)@101"
-      "Put(foo, bar)@100",
+      "Put(foo, bar)@100"
+      "DeleteRange(bar, foo)@102",
      PrintContents(&batch));
  ASSERT_EQ(4, batch.Count());
 }
@@ -181,6 +185,27 @@ TEST_F(WriteBatchTest, Append) {
            "Delete(foo)@203",
            PrintContents(&b1));
  ASSERT_EQ(4, b1.Count());
+  b2.Clear();
+  b2.Put("c", "cc");
+  b2.Put("d", "dd");
+  b2.MarkWalTerminationPoint();
+  b2.Put("e", "ee");
+  WriteBatchInternal::Append(&b1, &b2, /*wal only*/ true);
+  ASSERT_EQ(
+      "Put(a, va)@200"
+      "Put(b, vb)@202"
+      "Put(b, vb)@201"
+      "Put(c, cc)@204"
+      "Put(d, dd)@205"
+      "Delete(foo)@203",
+      PrintContents(&b1));
+  ASSERT_EQ(6, b1.Count());
+  ASSERT_EQ(
+      "Put(c, cc)@0"
+      "Put(d, dd)@1"
+      "Put(e, ee)@2",
+      PrintContents(&b2));
+  ASSERT_EQ(3, b2.Count());
 }

 TEST_F(WriteBatchTest, SingleDeletion) {