bump version to 5.3.3

Cleanup of ThreadStatusUtil structures should use the DB's reference
Summary: instead of thread_local The cleanup path for the rocksdb database might not have the thread_updater_local_cache_ pointer initialized because the thread executing the cleanup is likely not a rocksdb thread. This results in a memory leak detected by Valgrind. The cleanup code path should use the thread_status_updater pointer obtained from the DB object instead of a thread local one. Closes https://github.com/facebook/rocksdb/pull/2059 Differential Revision: D4801611 Pulled By: hermanlee fbshipit-source-id: 407d7de
2026-11-06 11:02:36 -05:00 · 2017-04-12 11:36:12 -07:00 · 2017-04-12 11:34:56 -07:00 · 2017-04-06 17:04:23 -07:00 · 2017-04-06 16:55:23 -07:00 · 2017-04-05 14:13:25 -07:00
388 changed files with 17162 additions and 7971 deletions
@@ -1,39 +1,56 @@
 sudo: false
+dist: trusty
 language: cpp
 os:
  - linux
  - osx
 compiler:
  - clang
+  - gcc
 jdk:
  - oraclejdk7
+cache:
+  - ccache
+  - apt

 addons:
   apt:
      sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
-      packages: ['clang-3.6' , 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'gflags']
+      packages: ['clang-3.6' , 'g++-6', 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'libgflags-dev']
 env:
-  # Run all tests before db_block_cache_test (db_test, db_test2)
-  - JOB_NAME=unittests ROCKSDBTESTS_END=db_block_cache_test
-  # Run all tests starting from db_block_cache_test (db_block_cache_test, db_iter_test, ...)
-  - JOB_NAME=unittests ROCKSDBTESTS_START=db_block_cache_test
+  - TEST_GROUP=platform_dependent1
+  - TEST_GROUP=platform_dependent2
+  - TEST_GROUP=platform_dependent3
+  - TEST_GROUP=1
+  - TEST_GROUP=2
  # Run java tests
  - JOB_NAME=java_test
  # Build ROCKSDB_LITE
  - JOB_NAME=lite_build

-install:
-  # Build gflags
-  # TODO(noetzli): Remove when gflags available through Travis
-  - pushd /tmp/ && curl -L https://github.com/gflags/gflags/archive/v2.1.2.tar.gz -o gflags.tar.gz && tar xfz gflags.tar.gz && cd gflags-2.1.2 && cmake . && make && popd
+matrix:
+  exclude:
+  - os: osx
+    compiler: gcc
+  - os: osx
+    env: TEST_GROUP=1
+  - os: osx
+    env: TEST_GROUP=2

 before_script:
-  - if [[ "${TRAVIS_OS_NAME}" == 'linux' ]]; then CXX=clang++-3.6; fi
+  - if [[ "${TRAVIS_OS_NAME}" == 'linux' && "${CXX}" == 'clang++' ]]; then CXX=clang++-3.6; fi
+  # test one linux g++ build with g++-6
+  - if [[ "${TRAVIS_OS_NAME}" == 'linux' && "${CXX}" == 'g++' && "${JOB_NAME}" == 'unittests' ]]; then CXX=g++-6; fi
  # Limit the maximum number of open file descriptors to 8192
  - ulimit -n 8192 || true

 script:
-  - if [[ "${JOB_NAME}" == 'unittests' ]]; then OPT=-DTRAVIS V=1 make -j4 check_some; fi
+  - ${CXX} --version
+  - if [[ "${TEST_GROUP}" == 'platform_dependent1' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=external_sst_file_basic_test make -j4 check_some; fi
+  - if [[ "${TEST_GROUP}" == 'platform_dependent2' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=external_sst_file_basic_test ROCKSDBTESTS_END=checkpoint_test make -j4 check_some; fi
+  - if [[ "${TEST_GROUP}" == 'platform_dependent3' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=checkpoint_test ROCKSDBTESTS_END=db_block_cache_test make -j4 check_some; fi
+  - if [[ "${TEST_GROUP}" == '1' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=comparator_db_test make -j4 check_some; fi
+  - if [[ "${TEST_GROUP}" == '2' ]]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=comparator_db_test make -j4 check_some; fi
  - if [[ "${JOB_NAME}" == 'java_test' ]]; then OPT=-DTRAVIS V=1 make clean jclean rocksdbjava jtest; fi
  - if [[ "${JOB_NAME}" == 'lite_build' ]]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib; fi
 notifications:
@@ -1,4 +1,5 @@
-# This cmake build is for Windows 64-bit only.
+# Prerequisites for Windows:
+#     This cmake build is for Windows 64-bit only.
 #
 # Prerequisites:
 #     You must have at least Visual Studio 2015 Update 3. Start the Developer Command Prompt window that is a part of Visual Studio installation.
@@ -24,6 +25,12 @@
 # 5. And release mode (/m[:<N>] is also supported)
 #        msbuild rocksdb.sln /p:Configuration=Release
 #
+# Linux:
+#
+# 1. Install a recent toolchain such as devtoolset-3 if you're on a older distro. C++11 required.
+# 2. mkdir build; cd build
+# 3. cmake ..
+# 4. make -j

 cmake_minimum_required(VERSION 2.6)
 project(rocksdb)
@@ -83,6 +90,21 @@ endif()

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

+if (NOT WIN32)
+  execute_process(COMMAND
+      "./build_tools/version.sh" "full"
+      WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
+      OUTPUT_VARIABLE ROCKSDB_VERSION
+  )
+  string(STRIP "${ROCKSDB_VERSION}" ROCKSDB_VERSION)
+  execute_process(COMMAND
+      "./build_tools/version.sh" "major"
+      WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
+      OUTPUT_VARIABLE ROCKSDB_VERSION_MAJOR
+  )
+  string(STRIP "${ROCKSDB_VERSION_MAJOR}" ROCKSDB_VERSION_MAJOR)
+endif()
+
 set(BUILD_VERSION_CC ${CMAKE_BINARY_DIR}/build_version.cc)
 configure_file(util/build_version.cc.in ${BUILD_VERSION_CC} @ONLY)
 add_library(build_version OBJECT ${BUILD_VERSION_CC})
@@ -96,7 +118,14 @@ else()
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wextra -Wall")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wsign-compare -Wshadow -Wno-unused-parameter -Wno-unused-variable -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer -momit-leaf-frame-pointer")
+  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
+    include(CheckCXXCompilerFlag)
+    CHECK_CXX_COMPILER_FLAG("-momit-leaf-frame-pointer" HAVE_OMIT_LEAF_FRAME_POINTER)
+    if(HAVE_OMIT_LEAF_FRAME_POINTER)
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -momit-leaf-frame-pointer")
+    endif()
+  endif()
 endif()

 option(FAIL_ON_WARNINGS "Treat compile warnings as errors" ON)
@@ -229,19 +258,9 @@ endif()
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
 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(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

@@ -295,7 +314,6 @@ set(SOURCES
        db/write_batch_base.cc
        db/write_controller.cc
        db/write_thread.cc
-        db/xfunc_test_points.cc
        memtable/hash_cuckoo_rep.cc
        memtable/hash_linklist_rep.cc
        memtable/hash_skiplist_rep.cc
@@ -318,10 +336,12 @@ set(SOURCES
        table/format.cc
        table/full_filter_block.cc
        table/get_context.cc
+        table/index_builder.cc
        table/iterator.cc
-        table/merger.cc
+        table/merging_iterator.cc
        table/sst_file_writer.cc
        table/meta_blocks.cc
+        table/partitioned_filter_block.cc
        table/plain_table_builder.cc
        table/plain_table_factory.cc
        table/plain_table_index.cc
@@ -358,6 +378,7 @@ set(SOURCES
        util/histogram_windowing.cc
        util/instrumented_mutex.cc
        util/iostats_context.cc
+        
        util/lru_cache.cc
        tools/ldb_cmd.cc
        tools/ldb_tool.cc
@@ -388,7 +409,6 @@ set(SOURCES
        util/thread_status_util.cc
        util/thread_status_util_debug.cc
        util/transaction_test_util.cc
-        util/xfunc.cc
        util/xxhash.cc
        utilities/backupable/backupable_db.cc
        utilities/blob_db/blob_db.cc
@@ -399,7 +419,6 @@ set(SOURCES
        utilities/document/json_document.cc
        utilities/document/json_document_builder.cc
        utilities/env_mirror.cc
-        utilities/env_registry.cc
        utilities/geodb/geodb_impl.cc
        utilities/leveldb_options/leveldb_options.cc
        utilities/lua/rocks_lua_compaction_filter.cc
@@ -443,6 +462,7 @@ if(WIN32)
    port/win/env_default.cc
    port/win/port_win.cc
    port/win/win_logger.cc
+    port/win/win_thread.cc
    port/win/xpress_win.cc)
 else()
  list(APPEND SOURCES
@@ -451,6 +471,29 @@ else()
    util/io_posix.cc)
 endif()

+if(WIN32)
+  set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)
+  set(ROCKSDB_STATIC_LIB rocksdblib${ARTIFACT_SUFFIX})
+  set(ROCKSDB_IMPORT_LIB rocksdb${ARTIFACT_SUFFIX})
+  set(LIBS ${ROCKSDB_STATIC_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
+else()
+  set(SYSTEM_LIBS ${CMAKE_THREAD_LIBS_INIT} rt)
+  set(ROCKSDB_STATIC_LIB rocksdb${ARTIFACT_SUFFIX})
+  set(ROCKSDB_SHARED_LIB rocksdb-shared)
+  set(ROCKSDB_IMPORT_LIB ${ROCKSDB_SHARED_LIB})
+  set(LIBS ${ROCKSDB_SHARED_LIB} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
+
+  add_library(${ROCKSDB_SHARED_LIB} SHARED ${SOURCES})
+  target_link_libraries(${ROCKSDB_SHARED_LIB}
+    ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
+  set_target_properties(${ROCKSDB_SHARED_LIB} PROPERTIES
+                        LINKER_LANGUAGE CXX
+                        VERSION ${ROCKSDB_VERSION}
+                        SOVERSION ${ROCKSDB_VERSION_MAJOR}
+                        CXX_STANDARD 11
+                        OUTPUT_NAME "rocksdb")
+endif()
+
 option(WITH_LIBRADOS "Build with librados" OFF)
 if(WITH_LIBRADOS)
  list(APPEND SOURCES
@@ -458,22 +501,22 @@ if(WITH_LIBRADOS)
  list(APPEND THIRDPARTY_LIBS rados)
 endif()

-add_library(rocksdblib${ARTIFACT_SUFFIX} STATIC ${SOURCES})
-target_link_libraries(rocksdblib${ARTIFACT_SUFFIX}
+add_library(${ROCKSDB_STATIC_LIB} STATIC ${SOURCES})
+target_link_libraries(${ROCKSDB_STATIC_LIB}
  ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})

 if(WIN32)
-  set_target_properties(rocksdblib${ARTIFACT_SUFFIX} PROPERTIES
-    COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/rocksdblib${ARTIFACT_SUFFIX}.pdb")
+  set_target_properties(${ROCKSDB_STATIC_LIB} PROPERTIES
+    COMPILE_FLAGS "/Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_STATIC_LIB}.pdb")
 endif()

-add_library(rocksdb${ARTIFACT_SUFFIX} SHARED ${SOURCES})
-target_link_libraries(rocksdb${ARTIFACT_SUFFIX}
-  ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
-
 if(WIN32)
-  set_target_properties(rocksdb${ARTIFACT_SUFFIX} PROPERTIES
-    COMPILE_FLAGS "-DROCKSDB_DLL -DROCKSDB_LIBRARY_EXPORTS /Fd${CMAKE_CFG_INTDIR}/rocksdb${ARTIFACT_SUFFIX}.pdb")
+  add_library(${ROCKSDB_IMPORT_LIB} SHARED ${SOURCES})
+  target_link_libraries(${ROCKSDB_IMPORT_LIB}
+    ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
+  set_target_properties(${ROCKSDB_IMPORT_LIB} PROPERTIES
+    COMPILE_FLAGS "-DROCKSDB_DLL -DROCKSDB_LIBRARY_EXPORTS /Fd${CMAKE_CFG_INTDIR}/${ROCKSDB_IMPORT_LIB}.pdb")
+else()
 endif()

 option(WITH_JNI "build with JNI" OFF)
@@ -495,6 +538,7 @@ set(TESTS
        db/comparator_db_test.cc
        db/corruption_test.cc
        db/cuckoo_table_db_test.cc
+        db/db_basic_test.cc
        db/db_compaction_filter_test.cc
        db/db_compaction_test.cc
        db/db_dynamic_level_test.cc
@@ -502,6 +546,7 @@ set(TESTS
        db/db_iter_test.cc
        db/db_log_iter_test.cc
        db/db_memtable_test.cc
+        db/db_merge_operator_test.cc
        db/db_options_test.cc
        db/db_properties_test.cc
        db/db_table_properties_test.cc
@@ -514,6 +559,7 @@ set(TESTS
        db/db_iterator_test.cc
        db/db_range_del_test.cc
        db/db_sst_test.cc
+        db/external_sst_file_basic_test.cc
        db/external_sst_file_test.cc
        db/db_universal_compaction_test.cc
        db/db_wal_test.cc
@@ -586,7 +632,7 @@ set(TESTS
        utilities/date_tiered/date_tiered_test.cc
        utilities/document/document_db_test.cc
        utilities/document/json_document_test.cc
-        utilities/env_registry_test.cc
+        utilities/object_registry_test.cc
        utilities/geodb/geodb_test.cc
        utilities/memory/memory_test.cc
        utilities/merge_operators/string_append/stringappend_test.cc
@@ -620,7 +666,7 @@ 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} ${LIBS} gtest)
+  target_link_libraries(${exename}${ARTIFACT_SUFFIX} gtest ${LIBS})
 endforeach(sourcefile ${BENCHMARKS})

 # For test util library that is build only in DEBUG mode
@@ -658,7 +704,7 @@ foreach(sourcefile ${TEST_EXES})
      EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
      EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
      )
-    target_link_libraries(${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} ${LIBS} gtest)
+    target_link_libraries(${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} gtest ${LIBS})
    if(NOT "${exename}" MATCHES "db_sanity_test")
      add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
      add_dependencies(check ${exename}${ARTIFACT_SUFFIX})
@@ -678,7 +724,18 @@ foreach(sourcefile ${C_TEST_EXES})
      EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
      EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
      )
-    target_link_libraries(${exename}${ARTIFACT_SUFFIX} rocksdb${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX})
+    target_link_libraries(${exename}${ARTIFACT_SUFFIX} ${ROCKSDB_IMPORT_LIB} testutillib${ARTIFACT_SUFFIX})
    add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
    add_dependencies(check ${exename}${ARTIFACT_SUFFIX})
 endforeach(sourcefile ${C_TEST_EXES})
+add_subdirectory(tools)
+
+# Installation and packaging for Linux
+if (NOT WIN32)
+install(TARGETS ${ROCKSDB_STATIC_LIB} COMPONENT devel ARCHIVE DESTINATION lib64)
+install(TARGETS ${ROCKSDB_SHARED_LIB} COMPONENT runtime DESTINATION lib64)
+install(DIRECTORY "${PROJECT_SOURCE_DIR}/include/rocksdb/"
+        COMPONENT devel
+        DESTINATION include/rocksdb)
+set(CMAKE_INSTALL_PREFIX /usr)
+endif()
@@ -1,4 +1,11 @@
 # RocksDB default options change log
+## Unreleased
+* Change the default of delayed slowdown value to 16MB/s and further increase the L0 stop condition to 36 files.
+
+## 5.0 (11/17/2016)
+* Options::allow_concurrent_memtable_write and Options::enable_write_thread_adaptive_yield are now true by default
+* Options.level0_stop_writes_trigger default value changes from 24 to 32.
+
 ## 4.8.0 (5/2/2016)
 * options.max_open_files changes from 5000 to -1. It improves performance, but users need to set file descriptor limit to be large enough and watch memory usage for index and bloom filters.
 * options.base_background_compactions changes from max_background_compactions to 1. When users set higher max_background_compactions but the write throughput is not high, the writes are less spiky to disks.
@@ -1,9 +1,43 @@
 # Rocksdb Change Log
-## Unreleased
+## 5.3.0 (03/08/2017)
 ### Public API Change
-* Options.level0_stop_writes_trigger default value changes from 24 to 32.
-* New compaction filter API: CompactionFilter::FilterV2(). Allows to drop ranges of keys.
-* Removed flashcache support.
+* Remove disableDataSync option.
+* Remove timeout_hint_us option from WriteOptions. The option has been deprecated and has no effect since 3.13.0.
+* Remove option min_partial_merge_operands. Partial merge operands will always be merged in flush or compaction if there are more than one.
+* Remove option verify_checksums_in_compaction. Compaction will always verify checksum.
+
+### New Features
+* Memtable flush can be avoided during checkpoint creation if total log file size is smaller than a threshold specified by the user.
+
+### Bug Fixes
+* Fix the bug that iterator may skip keys
+
+## 5.2.0 (02/08/2017)
+### Public API Change
+* NewLRUCache() will determine number of shard bits automatically based on capacity, if the user doesn't pass one. This also impacts the default block cache when the user doesn't explict provide one.
+* Change the default of delayed slowdown value to 16MB/s and further increase the L0 stop condition to 36 files.
+* Options::use_direct_writes and Options::use_direct_reads are now ready to use.
+* (Experimental) Two-level indexing that partition the index and creates a 2nd level index on the partitions. The feature can be enabled by setting kTwoLevelIndexSearch as IndexType and configuring index_per_partition.
+
+### New Features
+* Added new overloaded function GetApproximateSizes that allows to specify if memtable stats should be computed only without computing SST files' stats approximations.
+* Added new function GetApproximateMemTableStats that approximates both number of records and size of memtables.
+* Add Direct I/O mode for SST file I/O
+
+### Bug Fixes
+* RangeSync() should work if ROCKSDB_FALLOCATE_PRESENT is not set
+* Fix wrong results in a data race case in Get()
+* Some fixes related to 2PC.
+* Fix bugs of data corruption in direct I/O
+
+## 5.1.0 (01/13/2017)
+* Support dynamically change `delete_obsolete_files_period_micros` option via SetDBOptions().
+* Added EventListener::OnExternalFileIngested which will be called when IngestExternalFile() add a file successfully.
+* BackupEngine::Open and BackupEngineReadOnly::Open now always return error statuses matching those of the backup Env.
+
+### Bug Fixes
+* Fix the bug that if 2PC is enabled, checkpoints may loss some recent transactions.
+* When file copying is needed when creating checkpoints or bulk loading files, fsync the file after the file copying.

 ## 5.0.0 (11/17/2016)
 ### Public API Change
@@ -13,6 +47,10 @@
 * Support dynamically change `delayed_write_rate` option via SetDBOptions().
 * Options::allow_concurrent_memtable_write and Options::enable_write_thread_adaptive_yield are now true by default.
 * Remove Tickers::SEQUENCE_NUMBER to avoid confusion if statistics object is shared among RocksDB instance. Alternatively DB::GetLatestSequenceNumber() can be used to get the same value.
+* Options.level0_stop_writes_trigger default value changes from 24 to 32.
+* New compaction filter API: CompactionFilter::FilterV2(). Allows to drop ranges of keys.
+* Removed flashcache support.
+* DB::AddFile() is deprecated and is replaced with DB::IngestExternalFile(). DB::IngestExternalFile() remove all the restrictions that existed for DB::AddFile.

 ### New Features
 * Add avoid_flush_during_shutdown option, which speeds up DB shutdown by not flushing unpersisted data (i.e. with disableWAL = true). Unpersisted data will be lost. The options is dynamically changeable via SetDBOptions().
@@ -29,8 +29,10 @@ your make commands, like this: `PORTABLE=1 make static_lib`
 * You can link RocksDB with following compression libraries:
  - [zlib](http://www.zlib.net/) - a library for data compression.
  - [bzip2](http://www.bzip.org/) - a library for data compression.
-  - [snappy](https://code.google.com/p/snappy/) - a library for fast
+  - [snappy](http://google.github.io/snappy/) - a library for fast
      data compression.
+  - [zstandard](http://www.zstd.net) - Fast real-time compression
+      algorithm.

 * All our tools depend on:
  - [gflags](https://gflags.github.io/gflags/) - a library that handles
@@ -48,21 +50,23 @@ your make commands, like this: `PORTABLE=1 make static_lib`
      `sudo apt-get install libsnappy-dev`.
    * Install zlib. Try: `sudo apt-get install zlib1g-dev`.
    * Install bzip2: `sudo apt-get install libbz2-dev`.
-* **Linux - CentOS**
+    * Install zstandard: `sudo apt-get install libzstd-dev`.
+
+* **Linux - CentOS / RHEL**
    * Upgrade your gcc to version at least 4.8 to get C++11 support:
      `yum install gcc48-c++`
    * Install gflags:

-              wget https://gflags.googlecode.com/files/gflags-2.0-no-svn-files.tar.gz
-              tar -xzvf gflags-2.0-no-svn-files.tar.gz
-              cd gflags-2.0
+              git clone https://github.com/gflags/gflags.git
+              git checkout v2.0
+              cd gflags
              ./configure && make && sudo make install

    * Install snappy:

-              wget https://snappy.googlecode.com/files/snappy-1.1.1.tar.gz
-              tar -xzvf snappy-1.1.1.tar.gz
-              cd snappy-1.1.1
+              wget https://github.com/google/snappy/releases/download/1.1.4/snappy-1.1.4.tar.gz
+              tar -xzvf snappy-1.1.4.tar.gz
+              cd snappy-1.1.4
              ./configure && make && sudo make install

    * Install zlib:
@@ -75,6 +79,14 @@ your make commands, like this: `PORTABLE=1 make static_lib`
              sudo yum install bzip2
              sudo yum install bzip2-devel

+    * Install zstandard:
+
+             wget https://github.com/facebook/zstd/archive/v1.1.3.tar.gz
+             mv v1.1.3.tar.gz zstandard-1.1.3.tar.gz
+             tar zxvf zstd-1.1.3.tar.gz
+             cd zstd-1.1.3
+             make && sudo make install
+
 * **OS X**:
    * Install latest C++ compiler that supports C++ 11:
        * Update XCode:  run `xcode-select --install` (or install it from XCode App's settting).
@@ -10,3 +10,4 @@ This is the list of all known third-party language bindings for RocksDB. If some
 * PHP - https://github.com/Photonios/rocksdb-php
 * C# - https://github.com/warrenfalk/rocksdb-sharp
 * Rust - https://github.com/spacejam/rust-rocksdb
+* D programming language - https://github.com/b1naryth1ef/rocksdb
@@ -6,6 +6,9 @@

 #-----------------------------------------------

+BASH_EXISTS := $(shell which bash)
+SHELL := $(shell which bash)
+
 CLEAN_FILES = # deliberately empty, so we can append below.
 CFLAGS += ${EXTRA_CFLAGS}
 CXXFLAGS += ${EXTRA_CXXFLAGS}
@@ -137,7 +140,7 @@ endif

 AM_LINK = $(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 # detect what platform we're building on
-dummy := $(shell (export ROCKSDB_ROOT="$(CURDIR)"; "$(CURDIR)/build_tools/build_detect_platform" "$(CURDIR)/make_config.mk"))
+dummy := $(shell (export ROCKSDB_ROOT="$(CURDIR)"; export PORTABLE="$(PORTABLE)"; "$(CURDIR)/build_tools/build_detect_platform" "$(CURDIR)/make_config.mk"))
 # this file is generated by the previous line to set build flags and sources
 include make_config.mk
 CLEAN_FILES += make_config.mk
@@ -296,9 +299,26 @@ BENCHTOOLOBJECTS = $(BENCH_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
 EXPOBJECTS = $(EXP_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)

 TESTS = \
-	db_test \
+	db_basic_test \
 	db_test2 \
+	external_sst_file_basic_test \
+	auto_roll_logger_test \
+	bloom_test \
+	dynamic_bloom_test \
+	c_test \
+	checkpoint_test \
+	crc32c_test \
+	coding_test \
+	inlineskiplist_test \
+	env_basic_test \
+	env_test \
+	thread_local_test \
+	rate_limiter_test \
+	perf_context_test \
+	iostats_context_test \
+	db_wal_test \
 	db_block_cache_test \
+	db_test \
 	db_bloom_filter_test \
 	db_iter_test \
 	db_log_iter_test \
@@ -309,34 +329,25 @@ TESTS = \
 	db_inplace_update_test \
 	db_iterator_test \
 	db_memtable_test \
+	db_merge_operator_test \
 	db_options_test \
 	db_range_del_test \
 	db_sst_test \
-	external_sst_file_test \
 	db_tailing_iter_test \
 	db_universal_compaction_test \
-	db_wal_test \
 	db_io_failure_test \
 	db_properties_test \
 	db_table_properties_test \
 	autovector_test \
+	cleanable_test \
 	column_family_test \
 	table_properties_collector_test \
 	arena_test \
-	auto_roll_logger_test \
 	block_test \
-	bloom_test \
-	dynamic_bloom_test \
-	c_test \
 	cache_test \
-	checkpoint_test \
-	coding_test \
 	corruption_test \
-	crc32c_test \
 	slice_transform_test \
 	dbformat_test \
-	env_basic_test \
-	env_test \
 	fault_injection_test \
 	filelock_test \
 	filename_test \
@@ -345,7 +356,6 @@ TESTS = \
 	full_filter_block_test \
 	hash_table_test \
 	histogram_test \
-	inlineskiplist_test \
 	log_test \
 	manual_compaction_test \
 	mock_env_test \
@@ -360,6 +370,7 @@ TESTS = \
 	reduce_levels_test \
 	plain_table_db_test \
 	comparator_db_test \
+	external_sst_file_test \
 	prefix_test \
 	skiplist_test \
 	stringappend_test \
@@ -381,9 +392,7 @@ TESTS = \
 	write_controller_test\
 	deletefile_test \
 	table_test \
-	thread_local_test \
 	geodb_test \
-	rate_limiter_test \
 	delete_scheduler_test \
 	options_test \
 	options_settable_test \
@@ -401,7 +410,6 @@ TESTS = \
 	sst_dump_test \
 	column_aware_encoding_test \
 	compact_files_test \
-	perf_context_test \
 	optimistic_transaction_test \
 	write_callback_test \
 	heap_test \
@@ -410,11 +418,13 @@ TESTS = \
 	option_change_migration_test \
 	transaction_test \
 	ldb_cmd_test \
-	iostats_context_test \
 	persistent_cache_test \
 	statistics_test \
 	lua_test \
+	range_del_aggregator_test \
 	lru_cache_test \
+	object_registry_test \
+	repair_test \

 PARALLEL_TEST = \
 	backupable_db_test \
@@ -662,6 +672,8 @@ check_0:
 	  | grep -E '$(tests-regexp)'					\
 	  | build_tools/gnu_parallel -j$(J) --plain --joblog=LOG $$eta --gnu '{} >& t/log-{/}'

+valgrind-blacklist-regexp = InlineSkipTest.ConcurrentInsert|TransactionTest.DeadlockStress|DBCompactionTest.SuggestCompactRangeNoTwoLevel0Compactions|BackupableDBTest.RateLimiting|DBTest.CloseSpeedup|DBTest.ThreadStatusFlush|DBTest.RateLimitingTest|DBTest.EncodeDecompressedBlockSizeTest|FaultInjectionTest.UninstalledCompaction|HarnessTest.Randomized|ExternalSSTFileTest.CompactDuringAddFileRandom|ExternalSSTFileTest.IngestFileWithGlobalSeqnoRandomized
+
 .PHONY: valgrind_check_0
 valgrind_check_0:
 	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD);				\
@@ -675,6 +687,7 @@ valgrind_check_0:
 	}								\
 	  | $(prioritize_long_running_tests)				\
 	  | grep -E '$(tests-regexp)'					\
+	  | grep -E -v '$(valgrind-blacklist-regexp)'					\
 	  | build_tools/gnu_parallel -j$(J) --plain --joblog=LOG $$eta --gnu \
 	  '(if [[ "{}" == "./"* ]] ; then $(DRIVER) {}; else {}; fi) ' \
 	  '>& t/valgrind_log-{/}'
@@ -960,6 +973,9 @@ crc32c_test: util/crc32c_test.o $(LIBOBJECTS) $(TESTHARNESS)
 slice_transform_test: util/slice_transform_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_basic_test: db/db_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_test: db/db_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -996,6 +1012,9 @@ db_iterator_test: db/db_iterator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHA
 db_memtable_test: db/db_memtable_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

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

@@ -1005,6 +1024,9 @@ db_range_del_test: db/db_range_del_test.o db/db_test_util.o $(LIBOBJECTS) $(TEST
 db_sst_test: db/db_sst_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+external_sst_file_basic_test: db/external_sst_file_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 external_sst_file_test: db/external_sst_file_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1076,7 +1098,7 @@ env_librados_test: utilities/env_librados_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 endif

-env_registry_test: utilities/env_registry_test.o $(LIBOBJECTS) $(TESTHARNESS)
+object_registry_test: utilities/object_registry_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
@@ -1139,6 +1161,9 @@ full_filter_block_test: table/full_filter_block_test.o $(LIBOBJECTS) $(TESTHARNE
 log_test: db/log_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+cleanable_test: table/cleanable_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 table_test: table/table_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1301,6 +1326,9 @@ lru_cache_test: util/lru_cache_test.o $(LIBOBJECTS) $(TESTHARNESS)
 lua_test: utilities/lua/rocks_lua_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+range_del_aggregator_test: db/range_del_aggregator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1353,10 +1381,21 @@ ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linu
 ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
 ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-javadoc.jar
 ROCKSDB_SOURCES_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar
+SHA256_CMD = sha256sum
+
+ZLIB_VER = 1.2.11
+ZLIB_SHA256 = c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
+BZIP2_VER = 1.0.6
+BZIP2_SHA256 = a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
+SNAPPY_VER = 1.1.4
+SNAPPY_SHA256 = 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
+LZ4_VER = 1.7.5
+LZ4_SHA256 = 0190cacd63022ccb86f44fa5041dc6c3804407ad61550ca21c382827319e7e7e

 ifeq ($(PLATFORM), OS_MACOSX)
 	ROCKSDBJNILIB = librocksdbjni-osx.jnilib
 	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar
+	SHA256_CMD = openssl sha256 -r
 ifneq ("$(wildcard $(JAVA_HOME)/include/darwin)","")
 	JAVA_INCLUDE = -I$(JAVA_HOME)/include -I $(JAVA_HOME)/include/darwin
 else
@@ -1372,44 +1411,65 @@ ifeq ($(PLATFORM), OS_SOLARIS)
 	ROCKSDBJNILIB = librocksdbjni-solaris$(ARCH).so
 	ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-solaris$(ARCH).jar
 	JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/solaris
+	SHA256_CMD = digest -a sha256
 endif

 libz.a:
-	-rm -rf zlib-1.2.8
-	curl -O -L http://zlib.net/zlib-1.2.8.tar.gz
-	tar xvzf zlib-1.2.8.tar.gz
-	cd zlib-1.2.8 && CFLAGS='-fPIC' ./configure --static && make
-	cp zlib-1.2.8/libz.a .
+	-rm -rf zlib-$(ZLIB_VER)
+	curl -O -L http://zlib.net/zlib-$(ZLIB_VER).tar.gz
+	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
+	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
+		echo zlib-$(ZLIB_VER).tar.gz checksum mismatch, expected=\"$(ZLIB_SHA256)\" actual=\"$$ZLIB_SHA256_ACTUAL\"; \
+		exit 1; \
+	fi
+	tar xvzf zlib-$(ZLIB_VER).tar.gz
+	cd zlib-$(ZLIB_VER) && CFLAGS='-fPIC' ./configure --static && make
+	cp zlib-$(ZLIB_VER)/libz.a .

 libbz2.a:
-	-rm -rf bzip2-1.0.6
-	curl -O -L http://www.bzip.org/1.0.6/bzip2-1.0.6.tar.gz
-	tar xvzf bzip2-1.0.6.tar.gz
-	cd bzip2-1.0.6 && make CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64'
-	cp bzip2-1.0.6/libbz2.a .
+	-rm -rf bzip2-$(BZIP2_VER)
+	curl -O -L http://www.bzip.org/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
+	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
+	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
+		echo bzip2-$(BZIP2_VER).tar.gz checksum mismatch, expected=\"$(BZIP2_SHA256)\" actual=\"$$BZIP2_SHA256_ACTUAL\"; \
+		exit 1; \
+	fi
+	tar xvzf bzip2-$(BZIP2_VER).tar.gz
+	cd bzip2-$(BZIP2_VER) && make CFLAGS='-fPIC -O2 -g -D_FILE_OFFSET_BITS=64'
+	cp bzip2-$(BZIP2_VER)/libbz2.a .

 libsnappy.a:
-	-rm -rf snappy-1.1.3
-	curl -O -L https://github.com/google/snappy/releases/download/1.1.3/snappy-1.1.3.tar.gz
-	tar xvzf snappy-1.1.3.tar.gz
-	cd snappy-1.1.3 && ./configure --with-pic --enable-static
-	cd snappy-1.1.3 && make
-	cp snappy-1.1.3/.libs/libsnappy.a .
+	-rm -rf snappy-$(SNAPPY_VER)
+	curl -O -L https://github.com/google/snappy/releases/download/$(SNAPPY_VER)/snappy-$(SNAPPY_VER).tar.gz
+	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
+	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
+		echo snappy-$(SNAPPY_VER).tar.gz checksum mismatch, expected=\"$(SNAPPY_SHA256)\" actual=\"$$SNAPPY_SHA256_ACTUAL\"; \
+		exit 1; \
+	fi
+	tar xvzf snappy-$(SNAPPY_VER).tar.gz
+	cd snappy-$(SNAPPY_VER) && ./configure --with-pic --enable-static
+	cd snappy-$(SNAPPY_VER) && make
+	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .

 liblz4.a:
-	   -rm -rf lz4-r127
-	   curl -O -L https://codeload.github.com/Cyan4973/lz4/tar.gz/r127
-	   mv r127 lz4-r127.tar.gz
-	   tar xvzf lz4-r127.tar.gz
-	   cd lz4-r127/lib && make CFLAGS='-fPIC' all
-	   cp lz4-r127/lib/liblz4.a .
+	-rm -rf lz4-$(LZ4_VER)
+	curl -O -L https://github.com/lz4/lz4/archive/v$(LZ4_VER).tar.gz
+	mv v$(LZ4_VER).tar.gz lz4-$(LZ4_VER).tar.gz
+	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
+	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
+		echo lz4-$(LZ4_VER).tar.gz checksum mismatch, expected=\"$(LZ4_SHA256)\" actual=\"$$LZ4_SHA256_ACTUAL\"; \
+		exit 1; \
+	fi
+	tar xvzf lz4-$(LZ4_VER).tar.gz
+	cd lz4-$(LZ4_VER)/lib && make CFLAGS='-fPIC' all
+	cp lz4-$(LZ4_VER)/lib/liblz4.a .

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

 JAVA_STATIC_FLAGS = -DZLIB -DBZIP2 -DSNAPPY -DLZ4
-JAVA_STATIC_INCLUDES = -I./zlib-1.2.8 -I./bzip2-1.0.6 -I./snappy-1.1.3 -I./lz4-r127/lib
+JAVA_STATIC_INCLUDES = -I./zlib-$(ZLIB_VER) -I./bzip2-$(BZIP2_VER) -I./snappy-$(SNAPPY_VER) -I./lz4-$(LZ4_VER)/lib

 $(java_static_libobjects): jls/%.o: %.cc libz.a libbz2.a libsnappy.a liblz4.a
 	$(AM_V_CC)mkdir -p $(@D) && $(CXX) $(CXXFLAGS) $(JAVA_STATIC_FLAGS) $(JAVA_STATIC_INCLUDES) -fPIC -c $< -o $@ $(COVERAGEFLAGS)
@@ -1480,13 +1540,6 @@ commit_prereq: build_tools/rocksdb-lego-determinator \
 	J=$(J) build_tools/precommit_checker.py unit unit_481 clang_unit release release_481 clang_release tsan asan ubsan lite unit_non_shm
 	$(MAKE) clean && $(MAKE) jclean && $(MAKE) rocksdbjava;

-xfunc:
-	for xftest in $(XFUNC_TESTS); do \
-		echo "===== Running xftest $$xftest"; \
-		make check ROCKSDB_XFUNC_TEST="$$xftest" tests-regexp="DBTest" ;\
-	done
-
-
 # ---------------------------------------------------------------------------
 #  	Platform-specific compilation
 # ---------------------------------------------------------------------------
@@ -7,7 +7,7 @@ At Facebook, we use RocksDB as storage engines in multiple data management servi
 2. MongoRocks -- https://github.com/mongodb-partners/mongo-rocks
 3. ZippyDB --  Facebook's distributed key-value store with Paxos-style replication, built on top of RocksDB.[*] https://www.youtube.com/watch?v=DfiN7pG0D0khtt
 4. Laser -- Laser is a high query throughput, low (millisecond) latency, key-value storage service built on top of RocksDB.[*]
-4. Dragan -- a distributed graph query engine. https://code.facebook.com/posts/1737605303120405/dragon-a-distributed-graph-query-engine/
+4. Dragon -- a distributed graph query engine. https://code.facebook.com/posts/1737605303120405/dragon-a-distributed-graph-query-engine/
 5. Stylus -- a low-level stream processing framework writtenin C++.[*]

 [*] https://research.facebook.com/publications/realtime-data-processing-at-facebook/
@@ -64,7 +64,7 @@ Pinterest's Object Retrieval System uses RocksDB for storage: https://www.youtub
 quasardb uses a heavily tuned RocksDB as its persistence layer.

 ## Netflix
-[Netflix](http://techblog.netflix.com/2016/05/application-data-caching-using-ssds.html) uses RocksDB on spinning disks to cache application data.
+[Netflix](http://techblog.netflix.com/2016/05/application-data-caching-using-ssds.html) Netflix uses RocksDB on AWS EC2 instances with local SSD drives to cache application data.

 ## TiKV
 [TiKV](https://github.com/pingcap/tikv) is a GEO-replicated, high-performance, distributed, transactional key-value database. TiKV is powered by Rust and Raft. TiKV uses RocksDB as its persistence layer.
@@ -74,3 +74,9 @@ quasardb uses a heavily tuned RocksDB as its persistence layer.

 ## 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.
+
+## Uber
+[Uber](http://eng.uber.com/cherami/) uses RocksDB as a durable and and scalable task queue.
+
+## 360 Pika
+[360](http://www.360.cn/) [Pika](https://github.com/Qihoo360/pika) is a nosql compatible with redis. With the huge amount of data stored, redis may suffer for a capacity bottleneck, and pika was born for solving it. It has widely been widely used in many company
@@ -11,7 +11,8 @@ build:
  verbosity: minimal
 test:
 test_script:
- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test2 -Concurrency 8
- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test -Exclude DBTest.GroupCommitTest -Concurrency 10
- ps: build_tools\run_ci_db_test.ps1 -Run env_test -Concurrency 1
+- ps: build_tools\run_ci_db_test.ps1 -Run db_basic_test -Concurrency 8
+- ps: build_tools\run_ci_db_test.ps1 -Run db_test2 -Concurrency 8
+- ps: build_tools\run_ci_db_test.ps1 -Run db_test -Concurrency 8
+- ps: build_tools\run_ci_db_test.ps1 -Run env_test,env_basic_test -Concurrency 2

@@ -9,6 +9,13 @@ require('RocksDBCommonHelper.php');
 define("DIFF_COMMAND", "diff");

 class FacebookArcanistConfiguration extends ArcanistConfiguration {
+  public function getCustomArgumentsForCommand($command) {
+    if ($command == "land") {
+      return array(
+          'async' => array('help' => 'Just to make tools happy'));
+    }
+    return array();
+  }

  public function didRunWorkflow($command,
                                 ArcanistWorkflow $workflow,
@@ -9,6 +9,13 @@ require('RocksDBCommonHelper.php');
 define("DIFF_COMMAND", "diff");

 class FacebookArcanistConfiguration extends ArcanistConfiguration {
+  public function getCustomArgumentsForCommand($command) {
+    if ($command == "land") {
+      return array(
+          'async' => array('help' => 'Just to make tools happy'));
+    }
+    return array();
+  }

  public function didRunWorkflow($command,
                                 ArcanistBaseWorkflow $workflow,
@@ -21,7 +21,7 @@ function postURL($diffID, $url) {
  assert(is_numeric($diffID));
  assert(strlen($url) > 0);

-  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
+  $cmd = 'echo \'{"diff_id": ' . $diffID . ', '
         . '"name":"click here for sandcastle tests for D' . $diffID . '", '
         . '"link":"' . $url . '"}\' | '
         . 'arc call-conduit '
@@ -35,7 +35,7 @@ function buildUpdateTestStatusCmd($diffID, $test, $status) {
  assert(strlen($test) > 0);
  assert(strlen($status) > 0);

-  $cmd = 'echo \'{"diff_id": "' . $diffID . '", '
+  $cmd = 'echo \'{"diff_id": ' . $diffID . ', '
         . '"name":"' . $test . '", '
         . '"result":"' . $status . '"}\' | '
         . 'arc call-conduit '
@@ -8,7 +8,7 @@
 # for a certain interval.
 #

-SRC_GIT_REPO=/data/git/rocksdb
+SRC_GIT_REPO=/data/git/rocksdb-public
 error=0

 function log {
@@ -11,7 +11,8 @@ if ! which $CLANG_FORMAT_DIFF &> /dev/null
 then
  echo "You didn't have clang-format-diff.py available in your computer!"
  echo "You can download it by running: "
-  echo "    curl http://goo.gl/iUW1u2"
+  echo "    curl --location http://goo.gl/iUW1u2 -o ${CLANG_FORMAT_DIFF}"
+  echo "and move ${CLANG_FORMAT_DIFF} to some directory within PATH=${PATH}"
  exit 128
 fi

@@ -50,14 +51,15 @@ fi
 set -e

 uncommitted_code=`git diff HEAD`
+LAST_MASTER=`git merge-base master HEAD`

 # If there's no uncommitted changes, we assume user are doing post-commit
-# format check, in which case we'll check the modified lines from latest commit.
-# Otherwise, we'll check format of the uncommitted code only.
+# format check, in which case we'll check the modified lines since last commit
+# from master. Otherwise, we'll check format of the uncommitted code only.
 if [ -z "$uncommitted_code" ]
 then
  # Check the format of last commit
-  diffs=$(git diff -U0 HEAD^ | $CLANG_FORMAT_DIFF -p 1)
+  diffs=$(git diff -U0 $LAST_MASTER^ | $CLANG_FORMAT_DIFF -p 1)
 else
  # Check the format of uncommitted lines,
  diffs=$(git diff -U0 HEAD | $CLANG_FORMAT_DIFF -p 1)
@@ -97,7 +99,12 @@ then
 fi

 # Do in-place format adjustment.
-git diff -U0 HEAD^ | $CLANG_FORMAT_DIFF -i -p 1
+if [ -z "$uncommitted_code" ]
+then
+  git diff -U0 $LAST_MASTER^ | $CLANG_FORMAT_DIFF -i -p 1
+else
+  git diff -U0 HEAD^ | $CLANG_FORMAT_DIFF -i -p 1
+fi
 echo "Files reformatted!"

 # Amend to last commit if user do the post-commit format check
@@ -1,4 +1,4 @@
-#!/usr/local/fbcode/gcc-4.8.1-glibc-2.17-fb/bin/python2.7
+#!/usr/local/fbcode/gcc-4.9-glibc-2.20-fb/bin/python2.7

 from __future__ import absolute_import
 from __future__ import division
@@ -55,7 +55,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0  > ${STAT_FILE}.fillseq

@@ -73,7 +72,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=8 > ${STAT_FILE}.overwrite
@@ -92,7 +90,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=1 > /dev/null
@@ -111,7 +108,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandom
@@ -131,7 +127,6 @@ make release
    --use_tailing_iterator=1 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandomtailing
@@ -150,7 +145,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandomsmallblockcache
@@ -171,7 +165,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandom_mem_sst
@@ -191,7 +184,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=1 > /dev/null
@@ -210,7 +202,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > /dev/null
@@ -230,7 +221,6 @@ make release
    --disable_auto_compactions=1 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandom_filluniquerandom
@@ -251,7 +241,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readwhilewriting
@@ -270,7 +259,6 @@ make release
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --value_size=10 \
@@ -295,7 +283,6 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --disable_wal=0 \
    --wal_dir=/dev/shm/rocksdb \
    --sync=0 \
-    --disable_data_sync=1 \
    --verify_checksum=1 \
    --delete_obsolete_files_period_micros=314572800 \
    --max_grandparent_overlap_factor=10 \
@@ -356,7 +343,6 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0  > ${STAT_FILE}.fillseq_lots_column_families

@@ -372,7 +358,6 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --open_files=55000 \
    --statistics=1 \
    --histogram=1 \
-    --disable_data_sync=1 \
    --disable_wal=1 \
    --sync=0 \
    --threads=8 > ${STAT_FILE}.overwrite_lots_column_families
@@ -89,7 +89,7 @@ TSAN="COMPILE_WITH_TSAN=1"
 UBSAN="COMPILE_WITH_UBSAN=1"
 DISABLE_JEMALLOC="DISABLE_JEMALLOC=1"
 HTTP_PROXY="https_proxy=http://fwdproxy.29.prn1:8080 http_proxy=http://fwdproxy.29.prn1:8080 ftp_proxy=http://fwdproxy.29.prn1:8080"
-SETUP_JAVA_ENV="export $HTTP_PROXY; export JAVA_HOME=/usr/local/jdk-7u10-64/; export PATH=\$PATH:\$JAVA_HOME/bin"
+SETUP_JAVA_ENV="export $HTTP_PROXY; export JAVA_HOME=/usr/local/jdk-8u60-64/; export PATH=\$JAVA_HOME/bin:\$PATH"
 PARSER="'parser':'python build_tools/error_filter.py $1'"

 CONTRUN_NAME="ROCKSDB_CONTRUN_NAME"
@@ -50,7 +50,8 @@ Write-Output "Root: $RootFolder, WorkFolder: $WorkFolder"

 # Use JEMALLOC executables
 if($Run -ceq "db_test" -or
-   $Run -ceq "db_test2" ) {
+   $Run -ceq "db_test2" -or
+   $Run -ceq "db_basic_test") {

   $file_name = $Run

@@ -150,7 +151,7 @@ function MakeAndAdd([string]$token, $HashTable) {
 # Test executable name -> Log file
 function Discover-TestBinaries([string]$Pattern, $HashTable) {

-    $Exclusions = @("db_test*", "db_sanity_test*")
+    $Exclusions = @("db_test*", "db_sanity_test*", "db_basic_test*")

    $p = -join ($BinariesFolder, $pattern)

@@ -164,7 +165,8 @@ function Discover-TestBinaries([string]$Pattern, $HashTable) {
 $TestsToRun = [ordered]@{}

 if($Run -ceq "db_test" -or
-   $Run -ceq "db_test2") {
+   $Run -ceq "db_test2" -or
+   $Run -ceq "db_basic_test") {
    Normalize-DbTests -HashTable $TestsToRun
 } elseif($Run -ceq "tests") {
    if($EnableJE) {
@@ -237,7 +239,8 @@ function RunJobs($TestToLog, [int]$ConcurrencyVal, [bool]$AddForRerun)
            $log_path = ($TestToLog.$k)

            if($Run -ceq "db_test" -or
-               $Run -ceq "db_test2") {
+               $Run -ceq "db_test2" -or
+               $Run -ceq "db_basic_test") {
              $job = Start-Job -Name $k -ScriptBlock $InvokeTestCase -ArgumentList @($db_test,$k,$log_path)
            } else {
              [string]$Exe =  -Join ($BinariesFolder, $k)
@@ -1,8 +1,9 @@
 #!/bin/sh
 if [ "$#" = "0" ]; then
-  echo "Usage: $0 major|minor|patch"
+  echo "Usage: $0 major|minor|patch|full"
  exit 1
 fi
+
 if [ "$1" = "major" ]; then
  cat include/rocksdb/version.h  | grep MAJOR | head -n1 | awk '{print $3}'
 fi
@@ -12,3 +13,10 @@ fi
 if [ "$1" = "patch" ]; then
  cat include/rocksdb/version.h  | grep PATCH | head -n1 | awk '{print $3}'
 fi
+if [ "$1" = "full" ]; then
+  awk '/#define ROCKSDB/ { env[$2] = $3 }
+       END { printf "%s.%s.%s\n", env["ROCKSDB_MAJOR"],
+                                  env["ROCKSDB_MINOR"],
+                                  env["ROCKSDB_PATCH"] }'  \
+      include/rocksdb/version.h
+fi
@@ -8,6 +8,7 @@

 namespace rocksdb {

+#ifndef ROCKSDB_LITE
 // -- AutoRollLogger
 Status AutoRollLogger::ResetLogger() {
  TEST_SYNC_POINT("AutoRollLogger::ResetLogger:BeforeNewLogger");
@@ -130,6 +131,7 @@ bool AutoRollLogger::LogExpired() {
  ++cached_now_access_count;
  return cached_now >= ctime_ + kLogFileTimeToRoll;
 }
+#endif  // !ROCKSDB_LITE

 Status CreateLoggerFromOptions(const std::string& dbname,
                               const DBOptions& options,
@@ -147,6 +149,7 @@ Status CreateLoggerFromOptions(const std::string& dbname,

  env->CreateDirIfMissing(dbname);  // In case it does not exist
  // Currently we only support roll by time-to-roll and log size
+#ifndef ROCKSDB_LITE
  if (options.log_file_time_to_roll > 0 || options.max_log_file_size > 0) {
    AutoRollLogger* result = new AutoRollLogger(
        env, dbname, options.db_log_dir, options.max_log_file_size,
@@ -158,17 +161,17 @@ Status CreateLoggerFromOptions(const std::string& dbname,
      logger->reset(result);
    }
    return s;
-  } else {
-    // Open a log file in the same directory as the db
-    env->RenameFile(
-        fname, OldInfoLogFileName(dbname, env->NowMicros(), db_absolute_path,
-                                  options.db_log_dir));
-    auto s = env->NewLogger(fname, logger);
-    if (logger->get() != nullptr) {
-      (*logger)->SetInfoLogLevel(options.info_log_level);
-    }
-    return s;
  }
+#endif  // !ROCKSDB_LITE
+  // Open a log file in the same directory as the db
+  env->RenameFile(fname,
+                  OldInfoLogFileName(dbname, env->NowMicros(), db_absolute_path,
+                                     options.db_log_dir));
+  auto s = env->NewLogger(fname, logger);
+  if (logger->get() != nullptr) {
+    (*logger)->SetInfoLogLevel(options.info_log_level);
+  }
+  return s;
 }

 }  // namespace rocksdb
@@ -18,6 +18,7 @@

 namespace rocksdb {

+#ifndef ROCKSDB_LITE
 // Rolls the log file by size and/or time
 class AutoRollLogger : public Logger {
 public:
@@ -120,6 +121,7 @@ class AutoRollLogger : public Logger {
  uint64_t call_NowMicros_every_N_records_;
  mutable port::Mutex mutex_;
 };
+#endif  // !ROCKSDB_LITE

 // Facade to craete logger automatically
 Status CreateLoggerFromOptions(const std::string& dbname,
@@ -3,6 +3,9 @@
 //  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.
 //
+
+#ifndef ROCKSDB_LITE
+
 #include <string>
 #include <thread>
 #include <vector>
@@ -271,7 +274,7 @@ TEST_F(AutoRollLoggerTest, LogFlushWhileRolling) {
  AutoRollLogger* auto_roll_logger =
      dynamic_cast<AutoRollLogger*>(logger.get());
  ASSERT_TRUE(auto_roll_logger);
-  std::thread flush_thread;
+  rocksdb::port::Thread flush_thread;

  // Notes:
  // (1) Need to pin the old logger before beginning the roll, as rolling grabs
@@ -293,7 +296,7 @@ TEST_F(AutoRollLoggerTest, LogFlushWhileRolling) {
       {"AutoRollLogger::Flush:PinnedLogger", "PosixLogger::Flush:Begin2"}});
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  flush_thread = std::thread([&]() { auto_roll_logger->Flush(); });
+  flush_thread = port::Thread ([&]() { auto_roll_logger->Flush(); });
  TEST_SYNC_POINT(
      "AutoRollLoggerTest::LogFlushWhileRolling:PreRollAndPostThreadInit");
  RollLogFileBySizeTest(auto_roll_logger, options.max_log_file_size,
@@ -469,3 +472,14 @@ int main(int argc, char** argv) {
  ::testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
 }
+
+#else
+#include <stdio.h>
+
+int main(int argc, char** argv) {
+  fprintf(stderr,
+          "SKIPPED as AutoRollLogger is not supported in ROCKSDB_LITE\n");
+  return 0;
+}
+
+#endif  // !ROCKSDB_LITE
@@ -112,7 +112,8 @@ Status BuildTable(
      }
      file->SetIOPriority(io_priority);

-      file_writer.reset(new WritableFileWriter(std::move(file), env_options));
+      file_writer.reset(new WritableFileWriter(std::move(file), env_options,
+                                               ioptions.statistics));

      builder = NewTableBuilder(
          ioptions, internal_comparator, int_tbl_prop_collector_factories,
@@ -122,7 +123,6 @@ Status BuildTable(

    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());

@@ -170,7 +170,7 @@ Status BuildTable(
    delete builder;

    // Finish and check for file errors
-    if (s.ok() && !empty && !ioptions.disable_data_sync) {
+    if (s.ok() && !empty) {
      StopWatch sw(env, ioptions.statistics, TABLE_SYNC_MICROS);
      file_writer->Sync(ioptions.use_fsync);
    }
@@ -104,6 +104,9 @@ struct rocksdb_readoptions_t {
 };
 struct rocksdb_writeoptions_t    { WriteOptions      rep; };
 struct rocksdb_options_t         { Options           rep; };
+struct rocksdb_compactoptions_t {
+  CompactRangeOptions rep;
+};
 struct rocksdb_block_based_table_options_t  { BlockBasedTableOptions rep; };
 struct rocksdb_cuckoo_table_options_t  { CuckooTableOptions rep; };
 struct rocksdb_seqfile_t         { SequentialFile*   rep; };
@@ -445,7 +448,11 @@ rocksdb_backup_engine_t* rocksdb_backup_engine_open(
    const rocksdb_options_t* options, const char* path, char** errptr) {
  BackupEngine* be;
  if (SaveError(errptr, BackupEngine::Open(options->rep.env,
-                                           BackupableDBOptions(path), &be))) {
+                                           BackupableDBOptions(path,
+                                                               nullptr,
+                                                               true,
+                                                               options->rep.info_log.get()),
+                                           &be))) {
    return nullptr;
  }
  rocksdb_backup_engine_t* result = new rocksdb_backup_engine_t;
@@ -894,6 +901,17 @@ char* rocksdb_property_value(
  }
 }

+int rocksdb_property_int(
+    rocksdb_t* db,
+    const char* propname,
+    uint64_t *out_val) {
+  if (db->rep->GetIntProperty(Slice(propname), out_val)) {
+    return 0;
+  } else {
+    return -1;
+  }
+}
+
 char* rocksdb_property_value_cf(
    rocksdb_t* db,
    rocksdb_column_family_handle_t* column_family,
@@ -976,6 +994,30 @@ void rocksdb_compact_range_cf(
      (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr));
 }

+void rocksdb_compact_range_opt(rocksdb_t* db, rocksdb_compactoptions_t* opt,
+                               const char* start_key, size_t start_key_len,
+                               const char* limit_key, size_t limit_key_len) {
+  Slice a, b;
+  db->rep->CompactRange(
+      opt->rep,
+      // Pass nullptr Slice if corresponding "const char*" is nullptr
+      (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+      (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr));
+}
+
+void rocksdb_compact_range_cf_opt(rocksdb_t* db,
+                                  rocksdb_column_family_handle_t* column_family,
+                                  rocksdb_compactoptions_t* opt,
+                                  const char* start_key, size_t start_key_len,
+                                  const char* limit_key, size_t limit_key_len) {
+  Slice a, b;
+  db->rep->CompactRange(
+      opt->rep, column_family->rep,
+      // Pass nullptr Slice if corresponding "const char*" is nullptr
+      (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+      (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr));
+}
+
 void rocksdb_flush(
    rocksdb_t* db,
    const rocksdb_flushoptions_t* options,
@@ -1223,6 +1265,53 @@ void rocksdb_writebatch_deletev_cf(
  b->rep.Delete(column_family->rep, SliceParts(key_slices.data(), num_keys));
 }

+void rocksdb_writebatch_delete_range(rocksdb_writebatch_t* b,
+                                     const char* start_key,
+                                     size_t start_key_len, const char* end_key,
+                                     size_t end_key_len) {
+  b->rep.DeleteRange(Slice(start_key, start_key_len),
+                     Slice(end_key, end_key_len));
+}
+
+void rocksdb_writebatch_delete_range_cf(
+    rocksdb_writebatch_t* b, rocksdb_column_family_handle_t* column_family,
+    const char* start_key, size_t start_key_len, const char* end_key,
+    size_t end_key_len) {
+  b->rep.DeleteRange(column_family->rep, Slice(start_key, start_key_len),
+                     Slice(end_key, end_key_len));
+}
+
+void rocksdb_writebatch_delete_rangev(rocksdb_writebatch_t* b, int num_keys,
+                                      const char* const* start_keys_list,
+                                      const size_t* start_keys_list_sizes,
+                                      const char* const* end_keys_list,
+                                      const size_t* end_keys_list_sizes) {
+  std::vector<Slice> start_key_slices(num_keys);
+  std::vector<Slice> end_key_slices(num_keys);
+  for (int i = 0; i < num_keys; i++) {
+    start_key_slices[i] = Slice(start_keys_list[i], start_keys_list_sizes[i]);
+    end_key_slices[i] = Slice(end_keys_list[i], end_keys_list_sizes[i]);
+  }
+  b->rep.DeleteRange(SliceParts(start_key_slices.data(), num_keys),
+                     SliceParts(end_key_slices.data(), num_keys));
+}
+
+void rocksdb_writebatch_delete_rangev_cf(
+    rocksdb_writebatch_t* b, rocksdb_column_family_handle_t* column_family,
+    int num_keys, const char* const* start_keys_list,
+    const size_t* start_keys_list_sizes, const char* const* end_keys_list,
+    const size_t* end_keys_list_sizes) {
+  std::vector<Slice> start_key_slices(num_keys);
+  std::vector<Slice> end_key_slices(num_keys);
+  for (int i = 0; i < num_keys; i++) {
+    start_key_slices[i] = Slice(start_keys_list[i], start_keys_list_sizes[i]);
+    end_key_slices[i] = Slice(end_keys_list[i], end_keys_list_sizes[i]);
+  }
+  b->rep.DeleteRange(column_family->rep,
+                     SliceParts(start_key_slices.data(), num_keys),
+                     SliceParts(end_key_slices.data(), num_keys));
+}
+
 void rocksdb_writebatch_put_log_data(
    rocksdb_writebatch_t* b,
    const char* blob, size_t len) {
@@ -1258,6 +1347,15 @@ const char* rocksdb_writebatch_data(rocksdb_writebatch_t* b, size_t* size) {
  return b->rep.Data().c_str();
 }

+void rocksdb_writebatch_set_save_point(rocksdb_writebatch_t* b) {
+  b->rep.SetSavePoint();
+}
+
+void rocksdb_writebatch_rollback_to_save_point(rocksdb_writebatch_t* b,
+                                               char** errptr) {
+  SaveError(errptr, b->rep.RollbackToSavePoint());
+}
+
 rocksdb_block_based_table_options_t*
 rocksdb_block_based_options_create() {
  return new rocksdb_block_based_table_options_t;
@@ -1341,11 +1439,6 @@ void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
  options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
 }

-void rocksdb_block_based_options_set_skip_table_builder_flush(
-    rocksdb_block_based_table_options_t* options, unsigned char v) {
-  options->rep.skip_table_builder_flush = v;
-}
-
 void rocksdb_options_set_block_based_table_factory(
    rocksdb_options_t *opt,
    rocksdb_block_based_table_options_t* table_options) {
@@ -1400,6 +1493,14 @@ void rocksdb_options_set_cuckoo_table_factory(
  }
 }

+void rocksdb_set_options(
+    rocksdb_t* db, int count, const char* const keys[], const char* const values[], char** errptr) {
+        std::unordered_map<std::string, std::string> options_map;
+        for (int i=0; i<count; i++)
+            options_map[keys[i]] = values[i];
+        SaveError(errptr,
+            db->rep->SetOptions(options_map));
+    }

 rocksdb_options_t* rocksdb_options_create() {
  return new rocksdb_options_t;
@@ -1607,11 +1708,6 @@ void rocksdb_options_set_prefix_extractor(
  opt->rep.prefix_extractor.reset(prefix_extractor);
 }

-void rocksdb_options_set_disable_data_sync(
-    rocksdb_options_t* opt, int disable_data_sync) {
-  opt->rep.disableDataSync = disable_data_sync;
-}
-
 void rocksdb_options_set_use_fsync(
    rocksdb_options_t* opt, int use_fsync) {
  opt->rep.use_fsync = use_fsync;
@@ -1645,9 +1741,14 @@ void rocksdb_options_set_manifest_preallocation_size(
 void rocksdb_options_set_purge_redundant_kvs_while_flush(rocksdb_options_t* opt,
                                                         unsigned char v) {}

-void rocksdb_options_set_allow_os_buffer(rocksdb_options_t* opt,
-                                         unsigned char v) {
-  opt->rep.allow_os_buffer = v;
+void rocksdb_options_set_use_direct_reads(rocksdb_options_t* opt,
+                                          unsigned char v) {
+  opt->rep.use_direct_reads = v;
+}
+
+void rocksdb_options_set_use_direct_writes(rocksdb_options_t* opt,
+                                           unsigned char v) {
+  opt->rep.use_direct_writes = v;
 }

 void rocksdb_options_set_allow_mmap_reads(
@@ -1718,11 +1819,6 @@ void rocksdb_options_set_enable_write_thread_adaptive_yield(
  opt->rep.enable_write_thread_adaptive_yield = v;
 }

-void rocksdb_options_set_verify_checksums_in_compaction(
-    rocksdb_options_t* opt, unsigned char v) {
-  opt->rep.verify_checksums_in_compaction = v;
-}
-
 void rocksdb_options_set_max_sequential_skip_in_iterations(
    rocksdb_options_t* opt, uint64_t v) {
  opt->rep.max_sequential_skip_in_iterations = v;
@@ -1779,6 +1875,14 @@ void rocksdb_options_set_hard_rate_limit(rocksdb_options_t* opt, double v) {
  opt->rep.hard_rate_limit = v;
 }

+void rocksdb_options_set_soft_pending_compaction_bytes_limit(rocksdb_options_t* opt, size_t v) {
+  opt->rep.soft_pending_compaction_bytes_limit = v;
+}
+
+void rocksdb_options_set_hard_pending_compaction_bytes_limit(rocksdb_options_t* opt, size_t v) {
+  opt->rep.hard_pending_compaction_bytes_limit = v;
+}
+
 void rocksdb_options_set_rate_limit_delay_max_milliseconds(
    rocksdb_options_t* opt, unsigned int v) {
  opt->rep.rate_limit_delay_max_milliseconds = v;
@@ -1808,6 +1912,10 @@ void rocksdb_options_set_disable_auto_compactions(rocksdb_options_t* opt, int di
  opt->rep.disable_auto_compactions = disable;
 }

+void rocksdb_options_set_optimize_filters_for_hits(rocksdb_options_t* opt, int v) {
+  opt->rep.optimize_filters_for_hits = v;
+}
+
 void rocksdb_options_set_delete_obsolete_files_period_micros(
    rocksdb_options_t* opt, uint64_t v) {
  opt->rep.delete_obsolete_files_period_micros = v;
@@ -1862,11 +1970,6 @@ void rocksdb_options_set_max_successive_merges(
  opt->rep.max_successive_merges = v;
 }

-void rocksdb_options_set_min_partial_merge_operands(
-    rocksdb_options_t* opt, uint32_t v) {
-  opt->rep.min_partial_merge_operands = v;
-}
-
 void rocksdb_options_set_bloom_locality(
    rocksdb_options_t* opt, uint32_t v) {
  opt->rep.bloom_locality = v;
@@ -1926,7 +2029,7 @@ rocksdb_ratelimiter_t* rocksdb_ratelimiter_create(

 void rocksdb_ratelimiter_destroy(rocksdb_ratelimiter_t *limiter) {
  if (limiter->rep) {
-	delete limiter->rep;
+    delete limiter->rep;
  }
  delete limiter;
 }
@@ -2168,6 +2271,16 @@ void rocksdb_readoptions_set_readahead_size(
  opt->rep.readahead_size = v;
 }

+void rocksdb_readoptions_set_pin_data(rocksdb_readoptions_t* opt,
+                                      unsigned char v) {
+  opt->rep.pin_data = v;
+}
+
+void rocksdb_readoptions_set_total_order_seek(rocksdb_readoptions_t* opt,
+                                              unsigned char v) {
+  opt->rep.total_order_seek = v;
+}
+
 rocksdb_writeoptions_t* rocksdb_writeoptions_create() {
  return new rocksdb_writeoptions_t;
 }
@@ -2185,6 +2298,28 @@ void rocksdb_writeoptions_disable_WAL(rocksdb_writeoptions_t* opt, int disable)
  opt->rep.disableWAL = disable;
 }

+rocksdb_compactoptions_t* rocksdb_compactoptions_create() {
+  return new rocksdb_compactoptions_t;
+}
+
+void rocksdb_compactoptions_destroy(rocksdb_compactoptions_t* opt) {
+  delete opt;
+}
+
+void rocksdb_compactoptions_set_exclusive_manual_compaction(
+    rocksdb_compactoptions_t* opt, unsigned char v) {
+  opt->rep.exclusive_manual_compaction = v;
+}
+
+void rocksdb_compactoptions_set_change_level(rocksdb_compactoptions_t* opt,
+                                             unsigned char v) {
+  opt->rep.change_level = v;
+}
+
+void rocksdb_compactoptions_set_target_level(rocksdb_compactoptions_t* opt,
+                                             int n) {
+  opt->rep.target_level = n;
+}

 rocksdb_flushoptions_t* rocksdb_flushoptions_create() {
  return new rocksdb_flushoptions_t;
@@ -2213,6 +2348,14 @@ void rocksdb_cache_set_capacity(rocksdb_cache_t* cache, size_t capacity) {
  cache->rep->SetCapacity(capacity);
 }

+size_t rocksdb_cache_get_usage(rocksdb_cache_t* cache) {
+  return cache->rep->GetUsage();
+}
+
+size_t rocksdb_cache_get_pinned_usage(rocksdb_cache_t* cache) {
+  return cache->rep->GetPinnedUsage();
+}
+
 rocksdb_env_t* rocksdb_create_default_env() {
  rocksdb_env_t* result = new rocksdb_env_t;
  result->rep = Env::Default();
@@ -288,6 +288,7 @@ int main(int argc, char** argv) {
  rocksdb_cache_t* cache;
  rocksdb_env_t* env;
  rocksdb_options_t* options;
+  rocksdb_compactoptions_t* coptions;
  rocksdb_block_based_table_options_t* table_options;
  rocksdb_readoptions_t* roptions;
  rocksdb_writeoptions_t* woptions;
@@ -339,11 +340,14 @@ int main(int argc, char** argv) {

  roptions = rocksdb_readoptions_create();
  rocksdb_readoptions_set_verify_checksums(roptions, 1);
-  rocksdb_readoptions_set_fill_cache(roptions, 0);
+  rocksdb_readoptions_set_fill_cache(roptions, 1);

  woptions = rocksdb_writeoptions_create();
  rocksdb_writeoptions_set_sync(woptions, 1);

+  coptions = rocksdb_compactoptions_create();
+  rocksdb_compactoptions_set_exclusive_manual_compaction(coptions, 1);
+
  StartPhase("destroy");
  rocksdb_destroy_db(options, dbname, &err);
  Free(&err);
@@ -425,6 +429,95 @@ int main(int argc, char** argv) {
  rocksdb_compact_range(db, "a", 1, "z", 1);
  CheckGet(db, roptions, "foo", "hello");

+  StartPhase("compactallopt");
+  rocksdb_compact_range_opt(db, coptions, NULL, 0, NULL, 0);
+  CheckGet(db, roptions, "foo", "hello");
+
+  StartPhase("compactrangeopt");
+  rocksdb_compact_range_opt(db, coptions, "a", 1, "z", 1);
+  CheckGet(db, roptions, "foo", "hello");
+
+  // Simple check cache usage
+  StartPhase("cache_usage");
+  {
+    rocksdb_readoptions_set_pin_data(roptions, 1);
+    rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
+    rocksdb_iter_seek(iter, "foo", 3);
+
+    size_t usage = rocksdb_cache_get_usage(cache);
+    CheckCondition(usage > 0);
+
+    size_t pin_usage = rocksdb_cache_get_pinned_usage(cache);
+    CheckCondition(pin_usage > 0);
+
+    rocksdb_iter_next(iter);
+    rocksdb_iter_destroy(iter);
+    rocksdb_readoptions_set_pin_data(roptions, 0);
+  }
+
+  StartPhase("addfile");
+  {
+    rocksdb_envoptions_t* env_opt = rocksdb_envoptions_create();
+    rocksdb_options_t* io_options = rocksdb_options_create();
+    rocksdb_sstfilewriter_t* writer =
+        rocksdb_sstfilewriter_create(env_opt, io_options);
+
+    unlink(sstfilename);
+    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk1", 5, "v1", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v2", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v3", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_finish(writer, &err);
+    CheckNoError(err);
+
+    rocksdb_ingestexternalfileoptions_t* ing_opt =
+        rocksdb_ingestexternalfileoptions_create();
+    const char* file_list[1] = {sstfilename};
+    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "sstk1", "v1");
+    CheckGet(db, roptions, "sstk2", "v2");
+    CheckGet(db, roptions, "sstk3", "v3");
+
+    unlink(sstfilename);
+    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v4", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk22", 6, "v5", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v6", 2, &err);
+    CheckNoError(err);
+    rocksdb_sstfilewriter_finish(writer, &err);
+    CheckNoError(err);
+
+    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "sstk1", "v1");
+    CheckGet(db, roptions, "sstk2", "v4");
+    CheckGet(db, roptions, "sstk22", "v5");
+    CheckGet(db, roptions, "sstk3", "v6");
+
+    rocksdb_ingestexternalfileoptions_destroy(ing_opt);
+    rocksdb_sstfilewriter_destroy(writer);
+    rocksdb_options_destroy(io_options);
+    rocksdb_envoptions_destroy(env_opt);
+
+    // Delete all keys we just ingested
+    rocksdb_delete(db, woptions, "sstk1", 5, &err);
+    CheckNoError(err);
+    rocksdb_delete(db, woptions, "sstk2", 5, &err);
+    CheckNoError(err);
+    rocksdb_delete(db, woptions, "sstk22", 6, &err);
+    CheckNoError(err);
+    rocksdb_delete(db, woptions, "sstk3", 5, &err);
+    CheckNoError(err);
+  }
+
  StartPhase("writebatch");
  {
    rocksdb_writebatch_t* wb = rocksdb_writebatch_create();
@@ -441,6 +534,24 @@ int main(int argc, char** argv) {
    int pos = 0;
    rocksdb_writebatch_iterate(wb, &pos, CheckPut, CheckDel);
    CheckCondition(pos == 3);
+    rocksdb_writebatch_clear(wb);
+    rocksdb_writebatch_put(wb, "bar", 3, "b", 1);
+    rocksdb_writebatch_put(wb, "bay", 3, "d", 1);
+    rocksdb_writebatch_delete_range(wb, "bar", 3, "bay", 3);
+    rocksdb_write(db, woptions, wb, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "bar", NULL);
+    CheckGet(db, roptions, "bay", "d");
+    rocksdb_writebatch_clear(wb);
+    const char* start_list[1] = {"bay"};
+    const size_t start_sizes[1] = {3};
+    const char* end_list[1] = {"baz"};
+    const size_t end_sizes[1] = {3};
+    rocksdb_writebatch_delete_rangev(wb, 1, start_list, start_sizes, end_list,
+                                     end_sizes);
+    rocksdb_write(db, woptions, wb, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "bay", NULL);
    rocksdb_writebatch_destroy(wb);
  }

@@ -462,6 +573,23 @@ int main(int argc, char** argv) {
    rocksdb_writebatch_destroy(wb);
  }

+  StartPhase("writebatch_savepoint");
+  {
+    rocksdb_writebatch_t* wb = rocksdb_writebatch_create();
+    rocksdb_writebatch_set_save_point(wb);
+    const char* k_list[2] = {"z", "ap"};
+    const size_t k_sizes[2] = {1, 2};
+    const char* v_list[3] = {"x", "y", "z"};
+    const size_t v_sizes[3] = {1, 1, 1};
+    rocksdb_writebatch_putv(wb, 2, k_list, k_sizes, 3, v_list, v_sizes);
+    rocksdb_writebatch_rollback_to_save_point(wb, &err);
+    CheckNoError(err);
+    rocksdb_write(db, woptions, wb, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "zap", NULL);
+    rocksdb_writebatch_destroy(wb);
+  }
+
  StartPhase("writebatch_rep");
  {
    rocksdb_writebatch_t* wb1 = rocksdb_writebatch_create();
@@ -579,59 +707,6 @@ int main(int argc, char** argv) {
    rocksdb_release_snapshot(db, snap);
  }

-  StartPhase("addfile");
-  {
-    rocksdb_envoptions_t* env_opt = rocksdb_envoptions_create();
-    rocksdb_options_t* io_options = rocksdb_options_create();
-    rocksdb_sstfilewriter_t* writer =
-        rocksdb_sstfilewriter_create(env_opt, io_options);
-
-    unlink(sstfilename);
-    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk1", 5, "v1", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v2", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v3", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_finish(writer, &err);
-    CheckNoError(err);
-
-    rocksdb_ingestexternalfileoptions_t* ing_opt =
-        rocksdb_ingestexternalfileoptions_create();
-    const char* file_list[1] = {sstfilename};
-    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
-    CheckNoError(err);
-    CheckGet(db, roptions, "sstk1", "v1");
-    CheckGet(db, roptions, "sstk2", "v2");
-    CheckGet(db, roptions, "sstk3", "v3");
-
-    unlink(sstfilename);
-    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk2", 5, "v4", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk22", 6, "v5", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_add(writer, "sstk3", 5, "v6", 2, &err);
-    CheckNoError(err);
-    rocksdb_sstfilewriter_finish(writer, &err);
-    CheckNoError(err);
-
-    rocksdb_ingest_external_file(db, file_list, 1, ing_opt, &err);
-    CheckNoError(err);
-    CheckGet(db, roptions, "sstk1", "v1");
-    CheckGet(db, roptions, "sstk2", "v4");
-    CheckGet(db, roptions, "sstk22", "v5");
-    CheckGet(db, roptions, "sstk3", "v6");
-
-    rocksdb_ingestexternalfileoptions_destroy(ing_opt);
-    rocksdb_sstfilewriter_destroy(writer);
-    rocksdb_options_destroy(io_options);
-    rocksdb_envoptions_destroy(env_opt);
-  }
-
  StartPhase("repair");
  {
    // If we do not compact here, then the lazy deletion of
@@ -765,9 +840,9 @@ int main(int argc, char** argv) {
  {
    rocksdb_close(db);
    rocksdb_destroy_db(options, dbname, &err);
-    CheckNoError(err)
+    CheckNoError(err);

-        rocksdb_options_t* db_options = rocksdb_options_create();
+    rocksdb_options_t* db_options = rocksdb_options_create();
    rocksdb_options_set_create_if_missing(db_options, 1);
    db = rocksdb_open(db_options, dbname, &err);
    CheckNoError(err)
@@ -931,6 +1006,19 @@ int main(int argc, char** argv) {
    CheckNoError(err);
    rocksdb_iter_destroy(iter);

+    rocksdb_readoptions_set_total_order_seek(roptions, 1);
+    iter = rocksdb_create_iterator(db, roptions);
+    CheckCondition(!rocksdb_iter_valid(iter));
+
+    rocksdb_iter_seek(iter, "ba", 2);
+    rocksdb_iter_get_error(iter, &err);
+    CheckNoError(err);
+    CheckCondition(rocksdb_iter_valid(iter));
+    CheckIter(iter, "bar1", "bar");
+
+    rocksdb_iter_destroy(iter);
+    rocksdb_readoptions_set_total_order_seek(roptions, 0);
+
    rocksdb_close(db);
    rocksdb_destroy_db(options, dbname, &err);
  }
@@ -1040,6 +1128,7 @@ int main(int argc, char** argv) {
  rocksdb_block_based_options_destroy(table_options);
  rocksdb_readoptions_destroy(roptions);
  rocksdb_writeoptions_destroy(woptions);
+  rocksdb_compactoptions_destroy(coptions);
  rocksdb_cache_destroy(cache);
  rocksdb_comparator_destroy(cmp);
  rocksdb_env_destroy(env);
@@ -32,7 +32,6 @@
 #include "util/compression.h"
 #include "util/options_helper.h"
 #include "util/thread_status_util.h"
-#include "util/xfunc.h"

 namespace rocksdb {

@@ -182,14 +181,6 @@ ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
  } else if (result.memtable_prefix_bloom_size_ratio < 0) {
    result.memtable_prefix_bloom_size_ratio = 0;
  }
-  XFUNC_TEST("memtablelist_history", "transaction_xftest_SanitizeOptions",
-             xf_transaction_set_memtable_history1,
-             xf_transaction_set_memtable_history,
-             &result.max_write_buffer_number_to_maintain);
-  XFUNC_TEST("memtablelist_history_clear", "transaction_xftest_SanitizeOptions",
-             xf_transaction_clear_memtable_history1,
-             xf_transaction_clear_memtable_history,
-             &result.max_write_buffer_number_to_maintain);

  if (!result.prefix_extractor) {
    assert(result.memtable_factory);
@@ -370,7 +361,8 @@ ColumnFamilyData::ColumnFamilyData(
      column_family_set_(column_family_set),
      pending_flush_(false),
      pending_compaction_(false),
-      prev_compaction_needed_bytes_(0) {
+      prev_compaction_needed_bytes_(0),
+      allow_2pc_(db_options.allow_2pc) {
  Ref();

  // Convert user defined table properties collector factories to internal ones.
@@ -492,6 +484,25 @@ ColumnFamilyOptions ColumnFamilyData::GetLatestCFOptions() const {
  return BuildColumnFamilyOptions(initial_cf_options_, mutable_cf_options_);
 }

+uint64_t ColumnFamilyData::OldestLogToKeep() {
+  auto current_log = GetLogNumber();
+
+  if (allow_2pc_) {
+    auto imm_prep_log = imm()->GetMinLogContainingPrepSection();
+    auto mem_prep_log = mem()->GetMinLogContainingPrepSection();
+
+    if (imm_prep_log > 0 && imm_prep_log < current_log) {
+      current_log = imm_prep_log;
+    }
+
+    if (mem_prep_log > 0 && mem_prep_log < current_log) {
+      current_log = mem_prep_log;
+    }
+  }
+
+  return current_log;
+}
+
 const double kIncSlowdownRatio = 0.8;
 const double kDecSlowdownRatio = 1 / kIncSlowdownRatio;
 const double kNearStopSlowdownRatio = 0.6;
@@ -239,6 +239,9 @@ class ColumnFamilyData {
  uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
  void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }

+  // calculate the oldest log needed for the durability of this column family
+  uint64_t OldestLogToKeep();
+
  // See Memtable constructor for explanation of earliest_seq param.
  MemTable* ConstructNewMemtable(const MutableCFOptions& mutable_cf_options,
                                 SequenceNumber earliest_seq);
@@ -404,6 +407,9 @@ class ColumnFamilyData {
  bool pending_compaction_;

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

 // ColumnFamilySet has interesting thread-safety requirements
@@ -15,6 +15,7 @@

 #include "db/db_impl.h"
 #include "db/db_test_util.h"
+#include "port/port.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -212,6 +213,30 @@ class ColumnFamilyTest : public testing::Test {
 #endif
  }

+  bool IsDbWriteStopped() {
+#ifndef ROCKSDB_LITE
+    uint64_t v;
+    EXPECT_TRUE(dbfull()->GetIntProperty("rocksdb.is-write-stopped", &v));
+    return (v == 1);
+#else
+    return dbfull()->TEST_write_controler().IsStopped();
+#endif  // !ROCKSDB_LITE
+  }
+
+  uint64_t GetDbDelayedWriteRate() {
+#ifndef ROCKSDB_LITE
+    uint64_t v;
+    EXPECT_TRUE(
+        dbfull()->GetIntProperty("rocksdb.actual-delayed-write-rate", &v));
+    return v;
+#else
+    if (!dbfull()->TEST_write_controler().NeedsDelay()) {
+      return 0;
+    }
+    return dbfull()->TEST_write_controler().delayed_write_rate();
+#endif  // !ROCKSDB_LITE
+  }
+
  void Destroy() {
    Close();
    ASSERT_OK(DestroyDB(dbname_, Options(db_options_, column_family_options_)));
@@ -1147,7 +1172,6 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;

  default_cf.compaction_style = kCompactionStyleLevel;
  default_cf.num_levels = 3;
@@ -1220,7 +1244,6 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;

  default_cf.compaction_style = kCompactionStyleLevel;
@@ -1268,7 +1291,7 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::vector<std::thread> threads;
+  std::vector<port::Thread> threads;
  threads.emplace_back([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
@@ -1318,7 +1341,6 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1376,7 +1398,7 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
    WaitForFlush(2);
    AssertFilesPerLevel(ToString(i + 1), 2);
  }
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1411,7 +1433,6 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1464,7 +1485,7 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1507,7 +1528,6 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1557,7 +1577,7 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = true;
    ASSERT_OK(
@@ -1577,7 +1597,7 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
                        1);
  }

-  std::thread threads1([&] {
+  rocksdb::port::Thread threads1([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1606,7 +1626,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1655,7 +1674,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1696,7 +1715,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1747,7 +1765,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1793,7 +1811,6 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -1858,7 +1875,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    CompactRangeOptions compact_options;
    compact_options.exclusive_manual_compaction = false;
    ASSERT_OK(
@@ -1913,7 +1930,6 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
  db_options_.max_open_files = 20;  // only 10 files in file cache
-  db_options_.disableDataSync = true;
  db_options_.max_background_compactions = 3;
  db_options_.base_background_compactions = 3;

@@ -2323,7 +2339,7 @@ TEST_F(ColumnFamilyTest, FlushAndDropRaceCondition) {
  int kKeysNum = 10000;
  PutRandomData(1, kKeysNum, 100);

-  std::vector<std::thread> threads;
+  std::vector<port::Thread> threads;
  threads.emplace_back([&] { ASSERT_OK(db_->DropColumnFamily(handles_[1])); });

  sleeping_task.WakeUp();
@@ -2424,7 +2440,7 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {

  // Start a thread that will drop the first column family
  // and its comparator
-  std::thread drop_cf_thread(DropSingleColumnFamily, this, 1, &comparators);
+  rocksdb::port::Thread drop_cf_thread(DropSingleColumnFamily, this, 1, &comparators);

  DropColumnFamilies({2});

@@ -2461,173 +2477,165 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {

  vstorage->TEST_set_estimated_compaction_needed_bytes(50);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage->TEST_set_estimated_compaction_needed_bytes(201);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

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

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

  vstorage->TEST_set_estimated_compaction_needed_bytes(205);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(202);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(201);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(198);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage->TEST_set_estimated_compaction_needed_bytes(399);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(599);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.25,
-            dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());

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

  vstorage->TEST_set_estimated_compaction_needed_bytes(3001);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage->TEST_set_estimated_compaction_needed_bytes(390);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.25,
-            dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(100);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage->set_l0_delay_trigger_count(100);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
  ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());

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

  vstorage->set_l0_delay_trigger_count(0);
  vstorage->TEST_set_estimated_compaction_needed_bytes(300);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.25 / 1.25,
-            dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());

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

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

  vstorage->set_l0_delay_trigger_count(0);
  vstorage->TEST_set_estimated_compaction_needed_bytes(0);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  mutable_cf_options.disable_auto_compactions = true;
  dbfull()->TEST_write_controler().set_delayed_write_rate(kBaseRate);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage->set_l0_delay_trigger_count(50);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
+  ASSERT_EQ(0, GetDbDelayedWriteRate());
  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());

  vstorage->set_l0_delay_trigger_count(60);
  vstorage->TEST_set_estimated_compaction_needed_bytes(300);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
+  ASSERT_EQ(0, GetDbDelayedWriteRate());
  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());

  mutable_cf_options.disable_auto_compactions = false;
  vstorage->set_l0_delay_trigger_count(70);
  vstorage->TEST_set_estimated_compaction_needed_bytes(500);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->set_l0_delay_trigger_count(71);
  vstorage->TEST_set_estimated_compaction_needed_bytes(501);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.25,
-            dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 }

 TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
@@ -2719,60 +2727,55 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {

  vstorage->TEST_set_estimated_compaction_needed_bytes(50);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(201);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(600);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage->TEST_set_estimated_compaction_needed_bytes(70);
  cfd->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate, dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());

  vstorage1->TEST_set_estimated_compaction_needed_bytes(800);
  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
-  ASSERT_TRUE(!dbfull()->TEST_write_controler().IsStopped());
+  ASSERT_TRUE(!IsDbWriteStopped());
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
-  ASSERT_EQ(kBaseRate / 1.25,
-            dbfull()->TEST_write_controler().delayed_write_rate());
+  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());

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

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

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

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

 TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
@@ -3078,7 +3081,7 @@ TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  std::thread thread([&] { db_->SyncWAL(); });
+  rocksdb::port::Thread thread([&] { db_->SyncWAL(); });

  TEST_SYNC_POINT("ColumnFamilyTest::LogSyncConflictFlush:1");
  Flush(1);
@@ -11,6 +11,7 @@
 #include <vector>

 #include "db/db_impl.h"
+#include "port/port.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "util/string_util.h"
@@ -231,7 +232,7 @@ TEST_F(CompactFilesTest, CapturingPendingFiles) {
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  // Start compacting files.
-  std::thread compaction_thread(
+  rocksdb::port::Thread compaction_thread(
      [&] { EXPECT_OK(db->CompactFiles(CompactionOptions(), l0_files, 1)); });

  // In the meantime flush another file.
@@ -253,6 +254,61 @@ TEST_F(CompactFilesTest, CapturingPendingFiles) {
  delete db;
 }

+TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
+  class FilterWithGet : public CompactionFilter {
+   public:
+    virtual bool Filter(int level, const Slice& key, const Slice& value,
+                        std::string* new_value,
+                        bool* value_changed) const override {
+      if (db_ == nullptr) {
+        return true;
+      }
+      std::string res;
+      db_->Get(ReadOptions(), "", &res);
+      return true;
+    }
+
+    void SetDB(DB* db) {
+      db_ = db;
+    }
+
+    virtual const char* Name() const override { return "FilterWithGet"; }
+
+   private:
+    DB* db_;
+  };
+
+
+  std::shared_ptr<FilterWithGet> cf(new FilterWithGet());
+
+  Options options;
+  options.create_if_missing = true;
+  options.compaction_filter = cf.get();
+
+  DB* db = nullptr;
+  DestroyDB(db_name_, options);
+  Status s = DB::Open(options, db_name_, &db);
+  ASSERT_OK(s);
+
+  cf->SetDB(db);
+
+  // Write one L0 file
+  db->Put(WriteOptions(), "K1", "V1");
+  db->Flush(FlushOptions());
+
+  // Compact all L0 files using CompactFiles
+  rocksdb::ColumnFamilyMetaData meta;
+  db->GetColumnFamilyMetaData(&meta);
+  for (auto& file : meta.levels[0].files) {
+    std::string fname = file.db_path + "/" + file.name;
+    ASSERT_OK(
+        db->CompactFiles(rocksdb::CompactionOptions(), {fname}, 0));
+  }
+
+
+  delete db;
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -76,6 +76,13 @@ class CompactedDBImpl : public DBImpl {
                       ColumnFamilyHandle* column_family) override {
    return Status::NotSupported("Not supported in compacted db mode.");
  }
+  using DB::IngestExternalFile;
+  virtual Status IngestExternalFile(
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& external_files,
+      const IngestExternalFileOptions& ingestion_options) override {
+    return Status::NotSupported("Not supported in compacted db mode.");
+  }

 private:
  friend class DB;
@@ -150,7 +150,8 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
                       bool _manual_compaction, double _score,
                       bool _deletion_compaction,
                       CompactionReason _compaction_reason)
-    : start_level_(_inputs[0].level),
+    : input_vstorage_(vstorage),
+      start_level_(_inputs[0].level),
      output_level_(_output_level),
      max_output_file_size_(_target_file_size),
      max_compaction_bytes_(_max_compaction_bytes),
@@ -204,11 +205,10 @@ Compaction::~Compaction() {
 }

 bool Compaction::InputCompressionMatchesOutput() const {
-  VersionStorageInfo* vstorage = input_version_->storage_info();
-  int base_level = vstorage->base_level();
-  bool matches =
-      (GetCompressionType(*cfd_->ioptions(), vstorage, mutable_cf_options_,
-                          start_level_, base_level) == output_compression_);
+  int base_level = input_vstorage_->base_level();
+  bool matches = (GetCompressionType(immutable_cf_options_, input_vstorage_,
+                                     mutable_cf_options_, start_level_,
+                                     base_level) == output_compression_);
  if (matches) {
    TEST_SYNC_POINT("Compaction::InputCompressionMatchesOutput:Matches");
    return true;
@@ -225,15 +225,14 @@ bool Compaction::IsTrivialMove() const {
  // filter to be applied to that level, and thus cannot be a trivial move.

  // Check if start level have files with overlapping ranges
-  if (start_level_ == 0 &&
-      input_version_->storage_info()->level0_non_overlapping() == false) {
+  if (start_level_ == 0 && input_vstorage_->level0_non_overlapping() == false) {
    // We cannot move files from L0 to L1 if the files are overlapping
    return false;
  }

  if (is_manual_compaction_ &&
-      (cfd_->ioptions()->compaction_filter != nullptr ||
-       cfd_->ioptions()->compaction_filter_factory != nullptr)) {
+      (immutable_cf_options_.compaction_filter != nullptr ||
+       immutable_cf_options_.compaction_filter_factory != nullptr)) {
    // This is a manual compaction and we have a compaction filter that should
    // be executed, we cannot do a trivial move
    return false;
@@ -241,15 +240,34 @@ bool Compaction::IsTrivialMove() const {

  // Used in universal compaction, where trivial move can be done if the
  // input files are non overlapping
-  if ((cfd_->ioptions()->compaction_options_universal.allow_trivial_move) &&
+  if ((immutable_cf_options_.compaction_options_universal.allow_trivial_move) &&
      (output_level_ != 0)) {
    return is_trivial_move_;
  }

-  return (start_level_ != output_level_ && num_input_levels() == 1 &&
+  if (!(start_level_ != output_level_ && num_input_levels() == 1 &&
          input(0, 0)->fd.GetPathId() == output_path_id() &&
-          InputCompressionMatchesOutput() &&
-          TotalFileSize(grandparents_) <= max_compaction_bytes_);
+          InputCompressionMatchesOutput())) {
+    return false;
+  }
+
+  // assert inputs_.size() == 1
+
+  for (const auto& file : inputs_.front().files) {
+    std::vector<FileMetaData*> file_grand_parents;
+    if (output_level_ + 1 >= number_levels_) {
+      continue;
+    }
+    input_vstorage_->GetOverlappingInputs(output_level_ + 1, &file->smallest,
+                                          &file->largest, &file_grand_parents);
+    const auto compaction_size =
+        file->fd.GetFileSize() + TotalFileSize(file_grand_parents);
+    if (compaction_size > max_compaction_bytes_) {
+      return false;
+    }
+  }
+
+  return true;
 }

 void Compaction::AddInputDeletions(VersionEdit* out_edit) {
@@ -272,8 +290,7 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
  // Maybe use binary search to find right entry instead of linear search?
  const Comparator* user_cmp = cfd_->user_comparator();
  for (int lvl = output_level_ + 1; lvl < number_levels_; lvl++) {
-    const std::vector<FileMetaData*>& files =
-        input_version_->storage_info()->LevelFiles(lvl);
+    const std::vector<FileMetaData*>& files = input_vstorage_->LevelFiles(lvl);
    for (; level_ptrs->at(lvl) < files.size(); level_ptrs->at(lvl)++) {
      auto* f = files[level_ptrs->at(lvl)];
      if (user_cmp->Compare(user_key, f->largest.user_key()) <= 0) {
@@ -345,7 +362,7 @@ void Compaction::ReleaseCompactionFiles(Status status) {

 void Compaction::ResetNextCompactionIndex() {
  assert(input_version_ != nullptr);
-  input_version_->storage_info()->ResetNextCompactionIndex(start_level_);
+  input_vstorage_->ResetNextCompactionIndex(start_level_);
 }

 namespace {
@@ -258,6 +258,8 @@ class Compaction {
  static bool IsFullCompaction(VersionStorageInfo* vstorage,
                               const std::vector<CompactionInputFiles>& inputs);

+  VersionStorageInfo* input_vstorage_;
+
  const int start_level_;    // the lowest level to be compacted
  const int output_level_;  // levels to which output files are stored
  uint64_t max_output_file_size_;
@@ -16,14 +16,14 @@ CompactionIterator::CompactionIterator(
    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)
+    const std::atomic<bool>* shutting_down)
    : CompactionIterator(
          input, cmp, merge_helper, last_sequence, snapshots,
          earliest_write_conflict_snapshot, env, expect_valid_internal_key,
          range_del_agg,
          std::unique_ptr<CompactionProxy>(
              compaction ? new CompactionProxy(compaction) : nullptr),
-          compaction_filter, log_buffer) {}
+          compaction_filter, shutting_down) {}

 CompactionIterator::CompactionIterator(
    InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
@@ -31,7 +31,8 @@ CompactionIterator::CompactionIterator(
    SequenceNumber earliest_write_conflict_snapshot, Env* env,
    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
    std::unique_ptr<CompactionProxy> compaction,
-    const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
+    const CompactionFilter* compaction_filter,
+    const std::atomic<bool>* shutting_down)
    : input_(input),
      cmp_(cmp),
      merge_helper_(merge_helper),
@@ -42,7 +43,7 @@ CompactionIterator::CompactionIterator(
      range_del_agg_(range_del_agg),
      compaction_(std::move(compaction)),
      compaction_filter_(compaction_filter),
-      log_buffer_(log_buffer),
+      shutting_down_(shutting_down),
      merge_out_iter_(merge_helper_) {
  assert(compaction_filter_ == nullptr || compaction_ != nullptr);
  bottommost_level_ =
@@ -136,7 +137,7 @@ void CompactionIterator::NextFromInput() {
  at_next_ = false;
  valid_ = false;

-  while (!valid_ && input_->Valid()) {
+  while (!valid_ && input_->Valid() && !IsShuttingDown()) {
    key_ = input_->key();
    value_ = input_->value();
    iter_stats_.num_input_records++;
@@ -217,7 +218,8 @@ void CompactionIterator::NextFromInput() {
        }

        if (filter == CompactionFilter::Decision::kRemove) {
-          // convert the current key to a delete
+          // convert the current key to a delete; key_ is pointing into
+          // current_key_ at this point, so updating current_key_ updates key()
          ikey_.type = kTypeDeletion;
          current_key_.UpdateInternalKey(ikey_.sequence, kTypeDeletion);
          // no value associated with delete
@@ -422,7 +424,6 @@ void CompactionIterator::NextFromInput() {
      input_->Next();
    } else if (ikey_.type == kTypeMerge) {
      if (!merge_helper_->HasOperator()) {
-        LogToBuffer(log_buffer_, "Options::merge_operator is null.");
        status_ = Status::InvalidArgument(
            "merge_operator is not properly initialized.");
        return;
@@ -433,12 +434,13 @@ 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_, range_del_agg_, prev_snapshot,
-                                bottommost_level_);
+      Status s = merge_helper_->MergeUntil(input_, range_del_agg_,
+                                           prev_snapshot, bottommost_level_);
      merge_out_iter_.SeekToFirst();

-      if (merge_helper_->FilteredUntil(&skip_until)) {
-        need_skip = true;
+      if (!s.ok() && !s.IsMergeInProgress()) {
+        status_ = s;
+        return;
      } else if (merge_out_iter_.Valid()) {
        // NOTE: key, value, and ikey_ refer to old entries.
        //       These will be correctly set below.
@@ -460,11 +462,16 @@ void CompactionIterator::NextFromInput() {
        // coming after the merges
        has_current_user_key_ = false;
        pinned_iters_mgr_.ReleasePinnedData();
+
+        if (merge_helper_->FilteredUntil(&skip_until)) {
+          need_skip = true;
+        }
      }
    } else {
      // 1. new user key -OR-
      // 2. different snapshot stripe
-      bool should_delete = range_del_agg_->ShouldDelete(key_);
+      bool should_delete = range_del_agg_->ShouldDelete(
+          key_, RangeDelAggregator::RangePositioningMode::kForwardTraversal);
      if (should_delete) {
        ++iter_stats_.num_record_drop_hidden;
        ++iter_stats_.num_record_drop_range_del;
@@ -478,6 +485,10 @@ void CompactionIterator::NextFromInput() {
      input_->Seek(skip_until);
    }
  }
+
+  if (!valid_ && IsShuttingDown()) {
+    status_ = Status::ShutdownInProgress();
+  }
 }

 void CompactionIterator::PrepareOutput() {
@@ -17,7 +17,6 @@
 #include "db/pinned_iterators_manager.h"
 #include "db/range_del_aggregator.h"
 #include "rocksdb/compaction_filter.h"
-#include "util/log_buffer.h"

 namespace rocksdb {

@@ -61,7 +60,7 @@ class CompactionIterator {
                     RangeDelAggregator* range_del_agg,
                     const Compaction* compaction = nullptr,
                     const CompactionFilter* compaction_filter = nullptr,
-                     LogBuffer* log_buffer = nullptr);
+                     const std::atomic<bool>* shutting_down = nullptr);

  // Constructor with custom CompactionProxy, used for tests.
  CompactionIterator(InternalIterator* input, const Comparator* cmp,
@@ -72,7 +71,7 @@ class CompactionIterator {
                     RangeDelAggregator* range_del_agg,
                     std::unique_ptr<CompactionProxy> compaction,
                     const CompactionFilter* compaction_filter = nullptr,
-                     LogBuffer* log_buffer = nullptr);
+                     const std::atomic<bool>* shutting_down = nullptr);

  ~CompactionIterator();

@@ -125,7 +124,7 @@ class CompactionIterator {
  RangeDelAggregator* range_del_agg_;
  std::unique_ptr<CompactionProxy> compaction_;
  const CompactionFilter* compaction_filter_;
-  LogBuffer* log_buffer_;
+  const std::atomic<bool>* shutting_down_;
  bool bottommost_level_;
  bool valid_ = false;
  bool visible_at_tip_;
@@ -180,5 +179,10 @@ class CompactionIterator {
  // is in or beyond the last file checked during the previous call
  std::vector<size_t> level_ptrs_;
  CompactionIterationStats iter_stats_;
+
+  bool IsShuttingDown() {
+    // This is a best-effort facility, so memory_order_relaxed is sufficient.
+    return shutting_down_ && shutting_down_->load(std::memory_order_relaxed);
+  }
 };
 }  // namespace rocksdb
@@ -8,11 +8,70 @@
 #include <string>
 #include <vector>

+#include "port/port.h"
 #include "util/testharness.h"
 #include "util/testutil.h"

 namespace rocksdb {

+// Expects no merging attempts.
+class NoMergingMergeOp : public MergeOperator {
+ public:
+  bool FullMergeV2(const MergeOperationInput& merge_in,
+                   MergeOperationOutput* merge_out) const override {
+    ADD_FAILURE();
+    return false;
+  }
+  bool PartialMergeMulti(const Slice& key,
+                         const std::deque<Slice>& operand_list,
+                         std::string* new_value,
+                         Logger* logger) const override {
+    ADD_FAILURE();
+    return false;
+  }
+  const char* Name() const override {
+    return "CompactionIteratorTest NoMergingMergeOp";
+  }
+};
+
+// Compaction filter that gets stuck when it sees a particular key,
+// then gets unstuck when told to.
+// Always returns Decition::kRemove.
+class StallingFilter : public CompactionFilter {
+ public:
+  virtual Decision FilterV2(int level, const Slice& key, ValueType t,
+                            const Slice& existing_value, std::string* new_value,
+                            std::string* skip_until) const override {
+    int k = std::atoi(key.ToString().c_str());
+    last_seen.store(k);
+    while (k >= stall_at.load()) {
+      std::this_thread::yield();
+    }
+    return Decision::kRemove;
+  }
+
+  const char* Name() const override {
+    return "CompactionIteratorTest StallingFilter";
+  }
+
+  // Wait until the filter sees a key >= k and stalls at that key.
+  // If `exact`, asserts that the seen key is equal to k.
+  void WaitForStall(int k, bool exact = true) {
+    stall_at.store(k);
+    while (last_seen.load() < k) {
+      std::this_thread::yield();
+    }
+    if (exact) {
+      EXPECT_EQ(k, last_seen.load());
+    }
+  }
+
+  // Filter will stall on key >= stall_at. Advance stall_at to unstall.
+  mutable std::atomic<int> stall_at{0};
+  // Last key the filter was called with.
+  mutable std::atomic<int> last_seen{0};
+};
+
 class LoggingForwardVectorIterator : public InternalIterator {
 public:
  struct Action {
@@ -88,13 +147,15 @@ class FakeCompaction : public CompactionIterator::CompactionProxy {
  virtual int level(size_t compaction_input_level) const { return 0; }
  virtual bool KeyNotExistsBeyondOutputLevel(
      const Slice& user_key, std::vector<size_t>* level_ptrs) const {
-    return false;
+    return key_not_exists_beyond_output_level;
  }
  virtual bool bottommost_level() const { return false; }
  virtual int number_levels() const { return 1; }
  virtual Slice GetLargestUserKey() const {
    return "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
  }
+
+  bool key_not_exists_beyond_output_level = false;
 };

 class CompactionIteratorTest : public testing::Test {
@@ -116,17 +177,19 @@ class CompactionIteratorTest : public testing::Test {

    std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
    if (filter) {
-      compaction.reset(new FakeCompaction());
+      compaction_proxy_ = new FakeCompaction();
+      compaction.reset(compaction_proxy_);
    }

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

  void AddSnapshot(SequenceNumber snapshot) { snapshots_.push_back(snapshot); }
@@ -138,6 +201,8 @@ class CompactionIteratorTest : public testing::Test {
  std::unique_ptr<LoggingForwardVectorIterator> iter_;
  std::unique_ptr<CompactionIterator> c_iter_;
  std::unique_ptr<RangeDelAggregator> range_del_agg_;
+  std::atomic<bool> shutting_down_{false};
+  FakeCompaction* compaction_proxy_;
 };

 // It is possible that the output of the compaction iterator is empty even if
@@ -209,26 +274,6 @@ TEST_F(CompactionIteratorTest, RangeDeletionWithSnapshots) {
 }

 TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
-  // Expect no merging attempts.
-  class MergeOp : public MergeOperator {
-   public:
-    bool FullMergeV2(const MergeOperationInput& merge_in,
-                     MergeOperationOutput* merge_out) const override {
-      ADD_FAILURE();
-      return false;
-    }
-    bool PartialMergeMulti(const Slice& key,
-                           const std::deque<Slice>& operand_list,
-                           std::string* new_value,
-                           Logger* logger) const override {
-      ADD_FAILURE();
-      return false;
-    }
-    const char* Name() const override {
-      return "CompactionIteratorTest.CompactionFilterSkipUntil::MergeOp";
-    }
-  };
-
  class Filter : public CompactionFilter {
    virtual Decision FilterV2(int level, const Slice& key, ValueType t,
                              const Slice& existing_value,
@@ -272,8 +317,8 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
        return Decision::kKeep;
      }
      if (k == "i") {
-        EXPECT_EQ(ValueType::kValue, t);
-        EXPECT_EQ("iv95", v);
+        EXPECT_EQ(ValueType::kMergeOperand, t);
+        EXPECT_EQ("im95", v);
        *skip_until = "z";
        return Decision::kRemoveAndSkipUntil;
      }
@@ -286,7 +331,7 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
    }
  };

-  MergeOp merge_op;
+  NoMergingMergeOp merge_op;
  Filter filter;
  InitIterators(
      {test::KeyStr("a", 50, kTypeValue),  // keep
@@ -299,10 +344,10 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
       test::KeyStr("f", 30, kTypeMerge),  // skip to "g+"
       test::KeyStr("f", 25, kTypeValue), test::KeyStr("g", 90, kTypeValue),
       test::KeyStr("h", 91, kTypeValue),  // keep
-       test::KeyStr("i", 95, kTypeValue),  // skip to "z"
+       test::KeyStr("i", 95, kTypeMerge),  // skip to "z"
       test::KeyStr("j", 99, kTypeValue)},
      {"av50", "am45", "bv60", "bv40", "cv35", "dm70", "em71", "fm65", "fm30",
-       "fv25", "gv90", "hv91", "iv95", "jv99"},
+       "fv25", "gv90", "hv91", "im95", "jv99"},
      {}, {}, kMaxSequenceNumber, &merge_op, &filter);

  // Compaction should output just "a", "e" and "h" keys.
@@ -338,6 +383,77 @@ TEST_F(CompactionIteratorTest, CompactionFilterSkipUntil) {
  ASSERT_EQ(expected_actions, iter_->log);
 }

+TEST_F(CompactionIteratorTest, ShuttingDownInFilter) {
+  NoMergingMergeOp merge_op;
+  StallingFilter filter;
+  InitIterators(
+      {test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeValue),
+       test::KeyStr("3", 3, kTypeValue), test::KeyStr("4", 4, kTypeValue)},
+      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+  // Don't leave tombstones (kTypeDeletion) for filtered keys.
+  compaction_proxy_->key_not_exists_beyond_output_level = true;
+
+  std::atomic<bool> seek_done{false};
+  rocksdb::port::Thread compaction_thread([&] {
+    c_iter_->SeekToFirst();
+    EXPECT_FALSE(c_iter_->Valid());
+    EXPECT_TRUE(c_iter_->status().IsShutdownInProgress());
+    seek_done.store(true);
+  });
+
+  // Let key 1 through.
+  filter.WaitForStall(1);
+
+  // Shutdown during compaction filter call for key 2.
+  filter.WaitForStall(2);
+  shutting_down_.store(true);
+  EXPECT_FALSE(seek_done.load());
+
+  // Unstall filter and wait for SeekToFirst() to return.
+  filter.stall_at.store(3);
+  compaction_thread.join();
+  assert(seek_done.load());
+
+  // Check that filter was never called again.
+  EXPECT_EQ(2, filter.last_seen.load());
+}
+
+// Same as ShuttingDownInFilter, but shutdown happens during filter call for
+// a merge operand, not for a value.
+TEST_F(CompactionIteratorTest, ShuttingDownInMerge) {
+  NoMergingMergeOp merge_op;
+  StallingFilter filter;
+  InitIterators(
+      {test::KeyStr("1", 1, kTypeValue), test::KeyStr("2", 2, kTypeMerge),
+       test::KeyStr("3", 3, kTypeMerge), test::KeyStr("4", 4, kTypeValue)},
+      {"v1", "v2", "v3", "v4"}, {}, {}, kMaxSequenceNumber, &merge_op, &filter);
+  compaction_proxy_->key_not_exists_beyond_output_level = true;
+
+  std::atomic<bool> seek_done{false};
+  rocksdb::port::Thread compaction_thread([&] {
+    c_iter_->SeekToFirst();
+    ASSERT_FALSE(c_iter_->Valid());
+    ASSERT_TRUE(c_iter_->status().IsShutdownInProgress());
+    seek_done.store(true);
+  });
+
+  // Let key 1 through.
+  filter.WaitForStall(1);
+
+  // Shutdown during compaction filter call for key 2.
+  filter.WaitForStall(2);
+  shutting_down_.store(true);
+  EXPECT_FALSE(seek_done.load());
+
+  // Unstall filter and wait for SeekToFirst() to return.
+  filter.stall_at.store(3);
+  compaction_thread.join();
+  assert(seek_done.load());
+
+  // Check that filter was never called again.
+  EXPECT_EQ(2, filter.last_seen.load());
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -45,7 +45,7 @@
 #include "rocksdb/table.h"
 #include "table/block.h"
 #include "table/block_based_table_factory.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "table/table_builder.h"
 #include "util/coding.h"
 #include "util/file_reader_writer.h"
@@ -264,7 +264,7 @@ void CompactionJob::AggregateStatistics() {
 CompactionJob::CompactionJob(
    int job_id, Compaction* compaction, const ImmutableDBOptions& db_options,
    const EnvOptions& env_options, VersionSet* versions,
-    std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
+    const std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
    Directory* db_directory, Directory* output_directory, Statistics* stats,
    InstrumentedMutex* db_mutex, Status* db_bg_error,
    std::vector<SequenceNumber> existing_snapshots,
@@ -522,7 +522,7 @@ Status CompactionJob::Run() {
  const uint64_t start_micros = env_->NowMicros();

  // Launch a thread for each of subcompactions 1...num_threads-1
-  std::vector<std::thread> thread_pool;
+  std::vector<port::Thread> thread_pool;
  thread_pool.reserve(num_threads - 1);
  for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
    thread_pool.emplace_back(&CompactionJob::ProcessKeyValueCompaction, this,
@@ -538,7 +538,7 @@ Status CompactionJob::Run() {
    thread.join();
  }

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

@@ -721,10 +721,10 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  MergeHelper merge(
      env_, cfd->user_comparator(), cfd->ioptions()->merge_operator,
      compaction_filter, db_options_.info_log.get(),
-      mutable_cf_options->min_partial_merge_operands,
      false /* internal key corruption is expected */,
      existing_snapshots_.empty() ? 0 : existing_snapshots_.back(),
-      compact_->compaction->level(), db_options_.statistics.get());
+      compact_->compaction->level(), db_options_.statistics.get(),
+      shutting_down_);

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

@@ -742,9 +742,18 @@ 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,
-      range_del_agg.get(), sub_compact->compaction, compaction_filter));
+      range_del_agg.get(), sub_compact->compaction, compaction_filter,
+      shutting_down_));
  auto c_iter = sub_compact->c_iter.get();
  c_iter->SeekToFirst();
+  if (c_iter->Valid() &&
+      sub_compact->compaction->output_level() != 0) {
+    // ShouldStopBefore() maintains state based on keys processed so far. The
+    // compaction loop always calls it on the "next" key, thus won't tell it the
+    // first key. So we do that here.
+    sub_compact->ShouldStopBefore(
+      c_iter->key(), sub_compact->current_output_file_size);
+  }
  const auto& c_iter_stats = c_iter->iter_stats();
  auto sample_begin_offset_iter = sample_begin_offsets.cbegin();
  // data_begin_offset and compression_dict are only valid while generating
@@ -753,10 +762,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  std::string compression_dict;
  compression_dict.reserve(cfd->ioptions()->compression_opts.max_dict_bytes);

-  // TODO(noetzli): check whether we could check !shutting_down_->... only
-  // only occasionally (see diff D42687)
-  while (status.ok() && !shutting_down_->load(std::memory_order_acquire) &&
-         !cfd->IsDropped() && c_iter->Valid()) {
+  while (status.ok() && !cfd->IsDropped() && c_iter->Valid()) {
    // Invariant: c_iter.status() is guaranteed to be OK if c_iter->Valid()
    // returns true.
    const Slice& key = c_iter->key();
@@ -767,21 +773,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    if (end != nullptr &&
        cfd->user_comparator()->Compare(c_iter->user_key(), *end) >= 0) {
      break;
-    } else if (sub_compact->compaction->output_level() != 0 &&
-               sub_compact->ShouldStopBefore(
-                   key, sub_compact->current_output_file_size) &&
-               sub_compact->builder != nullptr) {
-      CompactionIterationStats range_del_out_stats;
-      status =
-          FinishCompactionOutputFile(input->status(), sub_compact,
-                                     range_del_agg.get(), &range_del_out_stats);
-      RecordDroppedKeys(range_del_out_stats,
-                        &sub_compact->compaction_job_stats);
-      if (!status.ok()) {
-        break;
-      }
    }
-
    if (c_iter_stats.num_input_records % kRecordStatsEvery ==
        kRecordStatsEvery - 1) {
      RecordDroppedKeys(c_iter_stats, &sub_compact->compaction_job_stats);
@@ -853,17 +845,37 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      }
    }

-    Status input_status = input->status();
-    c_iter->Next();
-
-    // Close output file if it is big enough
+    // Close output file if it is big enough. Two possibilities determine it's
+    // time to close it: (1) the current key should be this file's last key, (2)
+    // the next key should not be in this file.
+    //
    // 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)
+    bool output_file_ended = false;
+    Status input_status;
    if (sub_compact->compaction->output_level() != 0 &&
        sub_compact->current_output_file_size >=
            sub_compact->compaction->max_output_file_size()) {
+      // (1) this key terminates the file. For historical reasons, the iterator
+      // status before advancing will be given to FinishCompactionOutputFile().
+      input_status = input->status();
+      output_file_ended = true;
+    }
+    c_iter->Next();
+    if (!output_file_ended && c_iter->Valid() &&
+        sub_compact->compaction->output_level() != 0 &&
+        sub_compact->ShouldStopBefore(
+          c_iter->key(), sub_compact->current_output_file_size) &&
+        sub_compact->builder != nullptr) {
+      // (2) this key belongs to the next file. For historical reasons, the
+      // iterator status after advancing will be given to
+      // FinishCompactionOutputFile().
+      input_status = input->status();
+      output_file_ended = true;
+    }
+    if (output_file_ended) {
      const Slice* next_key = nullptr;
      if (c_iter->Valid()) {
        next_key = &c_iter->key();
@@ -901,27 +913,36 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  RecordDroppedKeys(c_iter_stats, &sub_compact->compaction_job_stats);
  RecordCompactionIOStats();

-  if (status.ok() &&
-      (shutting_down_->load(std::memory_order_acquire) || cfd->IsDropped())) {
+  if (status.ok() && (shutting_down_->load(std::memory_order_relaxed) ||
+                      cfd->IsDropped())) {
    status = Status::ShutdownInProgress(
        "Database shutdown or Column family drop during compaction");
  }
+  if (status.ok()) {
+    status = input->status();
+  }
+  if (status.ok()) {
+    status = c_iter->status();
+  }
+
  if (status.ok() && sub_compact->builder == nullptr &&
      sub_compact->outputs.size() == 0 &&
      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
    // handle subcompaction containing only range deletions
    status = OpenCompactionOutputFile(sub_compact);
  }
-  if (status.ok() && sub_compact->builder != nullptr) {
+
+  // Call FinishCompactionOutputFile() even if status is not ok: it needs to
+  // close the output file.
+  if (sub_compact->builder != nullptr) {
    CompactionIterationStats range_del_out_stats;
-    status =
-        FinishCompactionOutputFile(input->status(), sub_compact,
-                                   range_del_agg.get(), &range_del_out_stats);
+    Status s = FinishCompactionOutputFile(
+        status, sub_compact, range_del_agg.get(), &range_del_out_stats);
+    if (status.ok()) {
+      status = s;
+    }
    RecordDroppedKeys(range_del_out_stats, &sub_compact->compaction_job_stats);
  }
-  if (status.ok()) {
-    status = input->status();
-  }

  if (measure_io_stats_) {
    sub_compact->compaction_job_stats.file_write_nanos +=
@@ -1037,7 +1058,7 @@ Status CompactionJob::FinishCompactionOutputFile(
  sub_compact->total_bytes += current_bytes;

  // Finish and check for file errors
-  if (s.ok() && !db_options_.disable_data_sync) {
+  if (s.ok()) {
    StopWatch sw(env_, stats_, COMPACTION_OUTFILE_SYNC_MICROS);
    s = sub_compact->outfile->Sync(db_options_.use_fsync);
  }
@@ -1084,6 +1105,7 @@ Status CompactionJob::FinishCompactionOutputFile(
      event_logger_, cfd->ioptions()->listeners, dbname_, cfd->GetName(), fname,
      job_id_, meta->fd, tp, TableFileCreationReason::kCompaction, s);

+#ifndef ROCKSDB_LITE
  // Report new file to SstFileManagerImpl
  auto sfm =
      static_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
@@ -1101,6 +1123,7 @@ Status CompactionJob::FinishCompactionOutputFile(
      }
    }
  }
+#endif

  sub_compact->builder.reset();
  sub_compact->current_output_file_size = 0;
@@ -1199,8 +1222,8 @@ Status CompactionJob::OpenCompactionOutputFile(
  writable_file->SetIOPriority(Env::IO_LOW);
  writable_file->SetPreallocationBlockSize(static_cast<size_t>(
      sub_compact->compaction->OutputFilePreallocationSize()));
-  sub_compact->outfile.reset(
-      new WritableFileWriter(std::move(writable_file), env_options_));
+  sub_compact->outfile.reset(new WritableFileWriter(
+      std::move(writable_file), env_options_, db_options_.statistics.get()));

  // If the Column family flag is to only optimize filters for hits,
  // we can skip creating filters if this is the bottommost_level where
@@ -57,7 +57,7 @@ class CompactionJob {
  CompactionJob(int job_id, Compaction* compaction,
                const ImmutableDBOptions& db_options,
                const EnvOptions& env_options, VersionSet* versions,
-                std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
+                const std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
                Directory* db_directory, Directory* output_directory,
                Statistics* stats, InstrumentedMutex* db_mutex,
                Status* db_bg_error,
@@ -131,7 +131,7 @@ class CompactionJob {

  Env* env_;
  VersionSet* versions_;
-  std::atomic<bool>* shutting_down_;
+  const std::atomic<bool>* shutting_down_;
  LogBuffer* log_buffer_;
  Directory* db_directory_;
  Directory* output_directory_;
@@ -61,7 +61,6 @@
 #include "util/testharness.h"
 #include "util/testutil.h"
 #include "util/thread_status_util.h"
-#include "util/xfunc.h"
 #include "utilities/merge_operators.h"

 #if !defined(IOS_CROSS_COMPILE)
@@ -18,7 +18,6 @@
 #include <queue>
 #include <string>
 #include <utility>
-
 #include "db/column_family.h"
 #include "db/filename.h"
 #include "util/log_buffer.h"
@@ -466,49 +465,66 @@ bool CompactionPicker::SetupOtherInputs(
  // user key, while excluding other entries for the same user key. This
  // can happen when one user key spans multiple files.
  if (!output_level_inputs->empty()) {
-    CompactionInputFiles expanded0;
-    expanded0.level = input_level;
-    // Get entire range covered by compaction
+    const uint64_t limit = mutable_cf_options.max_compaction_bytes;
+    const uint64_t output_level_inputs_size =
+        TotalCompensatedFileSize(output_level_inputs->files);
+    const uint64_t inputs_size = TotalCompensatedFileSize(inputs->files);
+    bool expand_inputs = false;
+
+    CompactionInputFiles expanded_inputs;
+    expanded_inputs.level = input_level;
+    // Get closed interval of output level
    InternalKey all_start, all_limit;
    GetRange(*inputs, *output_level_inputs, &all_start, &all_limit);
-
+    bool try_overlapping_inputs = true;
    vstorage->GetOverlappingInputs(input_level, &all_start, &all_limit,
-                                   &expanded0.files, base_index, nullptr);
-    const uint64_t inputs0_size = TotalCompensatedFileSize(inputs->files);
-    const uint64_t inputs1_size =
-        TotalCompensatedFileSize(output_level_inputs->files);
-    const uint64_t expanded0_size = TotalCompensatedFileSize(expanded0.files);
-    uint64_t limit = mutable_cf_options.max_compaction_bytes;
-    if (expanded0.size() > inputs->size() &&
-        inputs1_size + expanded0_size < limit &&
-        !FilesInCompaction(expanded0.files) &&
-        !vstorage->HasOverlappingUserKey(&expanded0.files, input_level)) {
+                                   &expanded_inputs.files, base_index, nullptr);
+    uint64_t expanded_inputs_size =
+        TotalCompensatedFileSize(expanded_inputs.files);
+    if (!ExpandWhileOverlapping(cf_name, vstorage, &expanded_inputs)) {
+      try_overlapping_inputs = false;
+    }
+    if (try_overlapping_inputs && expanded_inputs.size() > inputs->size() &&
+        output_level_inputs_size + expanded_inputs_size < limit &&
+        !FilesInCompaction(expanded_inputs.files)) {
      InternalKey new_start, new_limit;
-      GetRange(expanded0, &new_start, &new_limit);
-      CompactionInputFiles expanded1;
-      expanded1.level = output_level;
+      GetRange(expanded_inputs, &new_start, &new_limit);
+      CompactionInputFiles expanded_output_level_inputs;
+      expanded_output_level_inputs.level = output_level;
      vstorage->GetOverlappingInputs(output_level, &new_start, &new_limit,
-                                     &expanded1.files, *parent_index,
-                                     parent_index);
-      assert(!expanded1.empty());
-      if (!FilesInCompaction(expanded1.files) &&
-          ExpandWhileOverlapping(cf_name, vstorage, &expanded1) &&
-          expanded1.size() == output_level_inputs->size()) {
-        Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
-            "[%s] Expanding@%d %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt "(%" PRIu64
-            "+%" PRIu64 " bytes) to %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt
-            " (%" PRIu64 "+%" PRIu64 "bytes)\n",
-            cf_name.c_str(), input_level, inputs->size(),
-            output_level_inputs->size(), inputs0_size, inputs1_size,
-            expanded0.size(), expanded1.size(), expanded0_size, inputs1_size);
-        smallest = new_start;
-        largest = new_limit;
-        inputs->files = expanded0.files;
-        output_level_inputs->files = expanded1.files;
+                                     &expanded_output_level_inputs.files,
+                                     *parent_index, parent_index);
+      assert(!expanded_output_level_inputs.empty());
+      if (!FilesInCompaction(expanded_output_level_inputs.files) &&
+          ExpandWhileOverlapping(cf_name, vstorage,
+                                 &expanded_output_level_inputs) &&
+          expanded_output_level_inputs.size() == output_level_inputs->size()) {
+        expand_inputs = true;
      }
    }
+    if (!expand_inputs) {
+      vstorage->GetCleanInputsWithinInterval(input_level, &all_start,
+                                             &all_limit, &expanded_inputs.files,
+                                             base_index, nullptr);
+      expanded_inputs_size = TotalCompensatedFileSize(expanded_inputs.files);
+      if (expanded_inputs.size() > inputs->size() &&
+          output_level_inputs_size + expanded_inputs_size < limit &&
+          !FilesInCompaction(expanded_inputs.files)) {
+        expand_inputs = true;
+      }
+    }
+    if (expand_inputs) {
+      Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
+          "[%s] Expanding@%d %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt "(%" PRIu64
+          "+%" PRIu64 " bytes) to %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt
+          " (%" PRIu64 "+%" PRIu64 "bytes)\n",
+          cf_name.c_str(), input_level, inputs->size(),
+          output_level_inputs->size(), inputs_size, output_level_inputs_size,
+          expanded_inputs.size(), output_level_inputs->size(),
+          expanded_inputs_size, output_level_inputs_size);
+      inputs->files = expanded_inputs.files;
+    }
  }
-
  return true;
 }

@@ -119,9 +119,6 @@ class CompactionPicker {
                                  const Slice& largest_user_key,
                                  int level) const;

- protected:
-  int NumberLevels() const { return ioptions_.num_levels; }
-
  // Stores the minimal range that covers all entries in inputs in
  // *smallest, *largest.
  // REQUIRES: inputs is not empty
@@ -141,6 +138,9 @@ class CompactionPicker {
  void GetRange(const std::vector<CompactionInputFiles>& inputs,
                InternalKey* smallest, InternalKey* largest) const;

+ protected:
+  int NumberLevels() const { return ioptions_.num_levels; }
+
  // 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
  // version in a "level". Otherwise, any Get() will search "level" first,
@@ -758,6 +758,70 @@ TEST_F(CompactionPickerTest, OverlappingUserKeys7) {
  ASSERT_EQ(5U, compaction->inputs(1)->back()->fd.GetNumber());
 }

+TEST_F(CompactionPickerTest, OverlappingUserKeys8) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.max_compaction_bytes = 100000000000u;
+  // grow the number of inputs in "level" without
+  // changing the number of "level+1" files we pick up
+  // Expand input level as much as possible
+  // no overlapping case
+  Add(1, 1U, "101", "150", 1U);
+  Add(1, 2U, "151", "200", 1U);
+  Add(1, 3U, "201", "300", 1000000000U);
+  Add(1, 4U, "301", "400", 1U);
+  Add(1, 5U, "401", "500", 1U);
+  Add(2, 6U, "150", "200", 1U);
+  Add(2, 7U, "200", "450", 1U, 0, 0);
+  Add(2, 8U, "500", "600", 1U);
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_EQ(3U, compaction->num_input_files(0));
+  ASSERT_EQ(2U, compaction->num_input_files(1));
+  ASSERT_EQ(2U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(3U, compaction->input(0, 1)->fd.GetNumber());
+  ASSERT_EQ(4U, compaction->input(0, 2)->fd.GetNumber());
+  ASSERT_EQ(6U, compaction->input(1, 0)->fd.GetNumber());
+  ASSERT_EQ(7U, compaction->input(1, 1)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, OverlappingUserKeys9) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.max_compaction_bytes = 100000000000u;
+  // grow the number of inputs in "level" without
+  // changing the number of "level+1" files we pick up
+  // Expand input level as much as possible
+  // overlapping case
+  Add(1, 1U, "121", "150", 1U);
+  Add(1, 2U, "151", "200", 1U);
+  Add(1, 3U, "201", "300", 1000000000U);
+  Add(1, 4U, "301", "400", 1U);
+  Add(1, 5U, "401", "500", 1U);
+  Add(2, 6U, "100", "120", 1U);
+  Add(2, 7U, "150", "200", 1U);
+  Add(2, 8U, "200", "450", 1U, 0, 0);
+  Add(2, 9U, "501", "600", 1U);
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_EQ(5U, compaction->num_input_files(0));
+  ASSERT_EQ(2U, compaction->num_input_files(1));
+  ASSERT_EQ(1U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(2U, compaction->input(0, 1)->fd.GetNumber());
+  ASSERT_EQ(3U, compaction->input(0, 2)->fd.GetNumber());
+  ASSERT_EQ(4U, compaction->input(0, 3)->fd.GetNumber());
+  ASSERT_EQ(7U, compaction->input(1, 0)->fd.GetNumber());
+  ASSERT_EQ(8U, compaction->input(1, 1)->fd.GetNumber());
+}
+
 TEST_F(CompactionPickerTest, NotScheduleL1IfL0WithHigherPri1) {
  NewVersionStorage(6, kCompactionStyleLevel);
  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
@@ -1132,13 +1196,13 @@ TEST_F(CompactionPickerTest, MaxCompactionBytesHit) {
  ioptions_.level_compaction_dynamic_level_bytes = false;
  NewVersionStorage(6, kCompactionStyleLevel);
  // A compaction should be triggered and pick file 2 and 5.
-  // It cannot expand because adding file 1 and 3, the compaction size will
+  // It can expand because adding file 1 and 3, the compaction size will
  // exceed mutable_cf_options_.max_bytes_for_level_base.
  Add(1, 1U, "100", "150", 300000U);
  Add(1, 2U, "151", "200", 300001U, 0, 0);
  Add(1, 3U, "201", "250", 300000U, 0, 0);
  Add(1, 4U, "251", "300", 300000U, 0, 0);
-  Add(2, 5U, "160", "256", 1U);
+  Add(2, 5U, "100", "256", 1U);
  UpdateVersionStorageInfo();

  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
@@ -1152,7 +1216,7 @@ TEST_F(CompactionPickerTest, MaxCompactionBytesHit) {
 }

 TEST_F(CompactionPickerTest, MaxCompactionBytesNotHit) {
-  mutable_cf_options_.max_bytes_for_level_base = 1000000u;
+  mutable_cf_options_.max_bytes_for_level_base = 800000u;
  mutable_cf_options_.max_compaction_bytes = 1000000u;
  ioptions_.level_compaction_dynamic_level_bytes = false;
  NewVersionStorage(6, kCompactionStyleLevel);
@@ -1162,7 +1226,7 @@ TEST_F(CompactionPickerTest, MaxCompactionBytesNotHit) {
  Add(1, 2U, "151", "200", 300001U, 0, 0);
  Add(1, 3U, "201", "250", 300000U, 0, 0);
  Add(1, 4U, "251", "300", 300000U, 0, 0);
-  Add(2, 5U, "000", "233", 1U);
+  Add(2, 5U, "000", "251", 1U);
  UpdateVersionStorageInfo();

  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
@@ -1177,6 +1241,50 @@ TEST_F(CompactionPickerTest, MaxCompactionBytesNotHit) {
  ASSERT_EQ(5U, compaction->input(1, 0)->fd.GetNumber());
 }

+TEST_F(CompactionPickerTest, IsTrivialMoveOn) {
+  mutable_cf_options_.max_bytes_for_level_base = 10000u;
+  mutable_cf_options_.max_compaction_bytes = 10001u;
+  ioptions_.level_compaction_dynamic_level_bytes = false;
+  NewVersionStorage(6, kCompactionStyleLevel);
+  // A compaction should be triggered and pick file 2
+  Add(1, 1U, "100", "150", 3000U);
+  Add(1, 2U, "151", "200", 3001U);
+  Add(1, 3U, "201", "250", 3000U);
+  Add(1, 4U, "251", "300", 3000U);
+
+  Add(3, 5U, "120", "130", 7000U);
+  Add(3, 6U, "170", "180", 7000U);
+  Add(3, 5U, "220", "230", 7000U);
+  Add(3, 5U, "270", "280", 7000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+    cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_TRUE(compaction->IsTrivialMove());
+}
+
+TEST_F(CompactionPickerTest, IsTrivialMoveOff) {
+  mutable_cf_options_.max_bytes_for_level_base = 1000000u;
+  mutable_cf_options_.max_compaction_bytes = 10000u;
+  ioptions_.level_compaction_dynamic_level_bytes = false;
+  NewVersionStorage(6, kCompactionStyleLevel);
+  // A compaction should be triggered and pick all files from level 1
+  Add(1, 1U, "100", "150", 300000U, 0, 0);
+  Add(1, 2U, "150", "200", 300000U, 0, 0);
+  Add(1, 3U, "200", "250", 300000U, 0, 0);
+  Add(1, 4U, "250", "300", 300000U, 0, 0);
+
+  Add(3, 5U, "120", "130", 6000U);
+  Add(3, 6U, "140", "150", 6000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+    cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_FALSE(compaction->IsTrivialMove());
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -41,7 +41,10 @@ class CorruptionTest : public testing::Test {
  DB* db_;

  CorruptionTest() {
-    tiny_cache_ = NewLRUCache(100);
+    // If LRU cache shard bit is smaller than 2 (or -1 which will automatically
+    // set it to 0), test SequenceNumberRecovery will fail, likely because of a
+    // bug in recovery code. Keep it 4 for now to make the test passes.
+    tiny_cache_ = NewLRUCache(100, 4);
    options_.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
    options_.env = &env_;
    dbname_ = test::TmpDir() + "/corruption_test";
@@ -0,0 +1,852 @@
+//  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.
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+#include "rocksdb/perf_context.h"
+#if !defined(ROCKSDB_LITE)
+#include "util/sync_point.h"
+#endif
+
+namespace rocksdb {
+
+class DBBasicTest : public DBTestBase {
+ public:
+  DBBasicTest() : DBTestBase("/db_basic_test") {}
+};
+
+TEST_F(DBBasicTest, OpenWhenOpen) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  rocksdb::DB* db2 = nullptr;
+  rocksdb::Status s = DB::Open(options, dbname_, &db2);
+
+  ASSERT_EQ(Status::Code::kIOError, s.code());
+  ASSERT_EQ(Status::SubCode::kNone, s.subcode());
+  ASSERT_TRUE(strstr(s.getState(), "lock ") != nullptr);
+
+  delete db2;
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBBasicTest, ReadOnlyDB) {
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_OK(Put("bar", "v2"));
+  ASSERT_OK(Put("foo", "v3"));
+  Close();
+
+  auto options = CurrentOptions();
+  assert(options.env = env_);
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ("v3", Get("foo"));
+  ASSERT_EQ("v2", Get("bar"));
+  Iterator* iter = db_->NewIterator(ReadOptions());
+  int count = 0;
+  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+    ASSERT_OK(iter->status());
+    ++count;
+  }
+  ASSERT_EQ(count, 2);
+  delete iter;
+  Close();
+
+  // Reopen and flush memtable.
+  Reopen(options);
+  Flush();
+  Close();
+  // Now check keys in read only mode.
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ("v3", Get("foo"));
+  ASSERT_EQ("v2", Get("bar"));
+  ASSERT_TRUE(db_->SyncWAL().IsNotSupported());
+}
+
+TEST_F(DBBasicTest, CompactedDB) {
+  const uint64_t kFileSize = 1 << 20;
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.write_buffer_size = kFileSize;
+  options.target_file_size_base = kFileSize;
+  options.max_bytes_for_level_base = 1 << 30;
+  options.compression = kNoCompression;
+  Reopen(options);
+  // 1 L0 file, use CompactedDB if max_open_files = -1
+  ASSERT_OK(Put("aaa", DummyString(kFileSize / 2, '1')));
+  Flush();
+  Close();
+  ASSERT_OK(ReadOnlyReopen(options));
+  Status s = Put("new", "value");
+  ASSERT_EQ(s.ToString(),
+            "Not implemented: Not supported operation in read only mode.");
+  ASSERT_EQ(DummyString(kFileSize / 2, '1'), Get("aaa"));
+  Close();
+  options.max_open_files = -1;
+  ASSERT_OK(ReadOnlyReopen(options));
+  s = Put("new", "value");
+  ASSERT_EQ(s.ToString(),
+            "Not implemented: Not supported in compacted db mode.");
+  ASSERT_EQ(DummyString(kFileSize / 2, '1'), Get("aaa"));
+  Close();
+  Reopen(options);
+  // Add more L0 files
+  ASSERT_OK(Put("bbb", DummyString(kFileSize / 2, '2')));
+  Flush();
+  ASSERT_OK(Put("aaa", DummyString(kFileSize / 2, 'a')));
+  Flush();
+  ASSERT_OK(Put("bbb", DummyString(kFileSize / 2, 'b')));
+  ASSERT_OK(Put("eee", DummyString(kFileSize / 2, 'e')));
+  Flush();
+  Close();
+
+  ASSERT_OK(ReadOnlyReopen(options));
+  // Fallback to read-only DB
+  s = Put("new", "value");
+  ASSERT_EQ(s.ToString(),
+            "Not implemented: Not supported operation in read only mode.");
+  Close();
+
+  // Full compaction
+  Reopen(options);
+  // Add more keys
+  ASSERT_OK(Put("fff", DummyString(kFileSize / 2, 'f')));
+  ASSERT_OK(Put("hhh", DummyString(kFileSize / 2, 'h')));
+  ASSERT_OK(Put("iii", DummyString(kFileSize / 2, 'i')));
+  ASSERT_OK(Put("jjj", DummyString(kFileSize / 2, 'j')));
+  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  ASSERT_EQ(3, NumTableFilesAtLevel(1));
+  Close();
+
+  // CompactedDB
+  ASSERT_OK(ReadOnlyReopen(options));
+  s = Put("new", "value");
+  ASSERT_EQ(s.ToString(),
+            "Not implemented: Not supported in compacted db mode.");
+  ASSERT_EQ("NOT_FOUND", Get("abc"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'a'), Get("aaa"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'b'), Get("bbb"));
+  ASSERT_EQ("NOT_FOUND", Get("ccc"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'e'), Get("eee"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'f'), Get("fff"));
+  ASSERT_EQ("NOT_FOUND", Get("ggg"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'h'), Get("hhh"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'i'), Get("iii"));
+  ASSERT_EQ(DummyString(kFileSize / 2, 'j'), Get("jjj"));
+  ASSERT_EQ("NOT_FOUND", Get("kkk"));
+
+  // MultiGet
+  std::vector<std::string> values;
+  std::vector<Status> status_list = dbfull()->MultiGet(
+      ReadOptions(),
+      std::vector<Slice>({Slice("aaa"), Slice("ccc"), Slice("eee"),
+                          Slice("ggg"), Slice("iii"), Slice("kkk")}),
+      &values);
+  ASSERT_EQ(status_list.size(), static_cast<uint64_t>(6));
+  ASSERT_EQ(values.size(), static_cast<uint64_t>(6));
+  ASSERT_OK(status_list[0]);
+  ASSERT_EQ(DummyString(kFileSize / 2, 'a'), values[0]);
+  ASSERT_TRUE(status_list[1].IsNotFound());
+  ASSERT_OK(status_list[2]);
+  ASSERT_EQ(DummyString(kFileSize / 2, 'e'), values[2]);
+  ASSERT_TRUE(status_list[3].IsNotFound());
+  ASSERT_OK(status_list[4]);
+  ASSERT_EQ(DummyString(kFileSize / 2, 'i'), values[4]);
+  ASSERT_TRUE(status_list[5].IsNotFound());
+
+  Reopen(options);
+  // Add a key
+  ASSERT_OK(Put("fff", DummyString(kFileSize / 2, 'f')));
+  Close();
+  ASSERT_OK(ReadOnlyReopen(options));
+  s = Put("new", "value");
+  ASSERT_EQ(s.ToString(),
+            "Not implemented: Not supported operation in read only mode.");
+}
+
+TEST_F(DBBasicTest, LevelLimitReopen) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  const std::string value(1024 * 1024, ' ');
+  int i = 0;
+  while (NumTableFilesAtLevel(2, 1) == 0) {
+    ASSERT_OK(Put(1, Key(i++), value));
+  }
+
+  options.num_levels = 1;
+  options.max_bytes_for_level_multiplier_additional.resize(1, 1);
+  Status s = TryReopenWithColumnFamilies({"default", "pikachu"}, options);
+  ASSERT_EQ(s.IsInvalidArgument(), true);
+  ASSERT_EQ(s.ToString(),
+            "Invalid argument: db has more levels than options.num_levels");
+
+  options.num_levels = 10;
+  options.max_bytes_for_level_multiplier_additional.resize(10, 1);
+  ASSERT_OK(TryReopenWithColumnFamilies({"default", "pikachu"}, options));
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBBasicTest, PutDeleteGet) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_OK(Put(1, "foo", "v2"));
+    ASSERT_EQ("v2", Get(1, "foo"));
+    ASSERT_OK(Delete(1, "foo"));
+    ASSERT_EQ("NOT_FOUND", Get(1, "foo"));
+  } while (ChangeOptions());
+}
+
+TEST_F(DBBasicTest, PutSingleDeleteGet) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_OK(Put(1, "foo2", "v2"));
+    ASSERT_EQ("v2", Get(1, "foo2"));
+    ASSERT_OK(SingleDelete(1, "foo"));
+    ASSERT_EQ("NOT_FOUND", Get(1, "foo"));
+    // Skip HashCuckooRep as it does not support single delete. FIFO and
+    // universal compaction do not apply to the test case. Skip MergePut
+    // because single delete does not get removed when it encounters a merge.
+  } while (ChangeOptions(kSkipHashCuckoo | kSkipFIFOCompaction |
+                         kSkipUniversalCompaction | kSkipMergePut));
+}
+
+TEST_F(DBBasicTest, EmptyFlush) {
+  // It is possible to produce empty flushes when using single deletes. Tests
+  // whether empty flushes cause issues.
+  do {
+    Random rnd(301);
+
+    Options options = CurrentOptions();
+    options.disable_auto_compactions = true;
+    CreateAndReopenWithCF({"pikachu"}, options);
+
+    Put(1, "a", Slice());
+    SingleDelete(1, "a");
+    ASSERT_OK(Flush(1));
+
+    ASSERT_EQ("[ ]", AllEntriesFor("a", 1));
+    // Skip HashCuckooRep as it does not support single delete. FIFO and
+    // universal compaction do not apply to the test case. Skip MergePut
+    // because merges cannot be combined with single deletions.
+  } while (ChangeOptions(kSkipHashCuckoo | kSkipFIFOCompaction |
+                         kSkipUniversalCompaction | kSkipMergePut));
+}
+
+TEST_F(DBBasicTest, GetFromVersions) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("NOT_FOUND", Get(0, "foo"));
+  } while (ChangeOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBBasicTest, GetSnapshot) {
+  anon::OptionsOverride options_override;
+  options_override.skip_policy = kSkipNoSnapshot;
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions(options_override));
+    // Try with both a short key and a long key
+    for (int i = 0; i < 2; i++) {
+      std::string key = (i == 0) ? std::string("foo") : std::string(200, 'x');
+      ASSERT_OK(Put(1, key, "v1"));
+      const Snapshot* s1 = db_->GetSnapshot();
+      if (option_config_ == kHashCuckoo) {
+        // Unsupported case.
+        ASSERT_TRUE(s1 == nullptr);
+        break;
+      }
+      ASSERT_OK(Put(1, key, "v2"));
+      ASSERT_EQ("v2", Get(1, key));
+      ASSERT_EQ("v1", Get(1, key, s1));
+      ASSERT_OK(Flush(1));
+      ASSERT_EQ("v2", Get(1, key));
+      ASSERT_EQ("v1", Get(1, key, s1));
+      db_->ReleaseSnapshot(s1);
+    }
+  } while (ChangeOptions());
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBBasicTest, CheckLock) {
+  do {
+    DB* localdb;
+    Options options = CurrentOptions();
+    ASSERT_OK(TryReopen(options));
+
+    // second open should fail
+    ASSERT_TRUE(!(DB::Open(options, dbname_, &localdb)).ok());
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, FlushMultipleMemtable) {
+  do {
+    Options options = CurrentOptions();
+    WriteOptions writeOpt = WriteOptions();
+    writeOpt.disableWAL = true;
+    options.max_write_buffer_number = 4;
+    options.min_write_buffer_number_to_merge = 3;
+    options.max_write_buffer_number_to_maintain = -1;
+    CreateAndReopenWithCF({"pikachu"}, options);
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "foo", "v1"));
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v1"));
+
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("v1", Get(1, "bar"));
+    ASSERT_OK(Flush(1));
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, FlushEmptyColumnFamily) {
+  // Block flush thread and disable compaction thread
+  env_->SetBackgroundThreads(1, Env::HIGH);
+  env_->SetBackgroundThreads(1, Env::LOW);
+  test::SleepingBackgroundTask sleeping_task_low;
+  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
+                 Env::Priority::LOW);
+  test::SleepingBackgroundTask sleeping_task_high;
+  env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                 &sleeping_task_high, Env::Priority::HIGH);
+
+  Options options = CurrentOptions();
+  // disable compaction
+  options.disable_auto_compactions = true;
+  WriteOptions writeOpt = WriteOptions();
+  writeOpt.disableWAL = true;
+  options.max_write_buffer_number = 2;
+  options.min_write_buffer_number_to_merge = 1;
+  options.max_write_buffer_number_to_maintain = 1;
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  // Compaction can still go through even if no thread can flush the
+  // mem table.
+  ASSERT_OK(Flush(0));
+  ASSERT_OK(Flush(1));
+
+  // Insert can go through
+  ASSERT_OK(dbfull()->Put(writeOpt, handles_[0], "foo", "v1"));
+  ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v1"));
+
+  ASSERT_EQ("v1", Get(0, "foo"));
+  ASSERT_EQ("v1", Get(1, "bar"));
+
+  sleeping_task_high.WakeUp();
+  sleeping_task_high.WaitUntilDone();
+
+  // Flush can still go through.
+  ASSERT_OK(Flush(0));
+  ASSERT_OK(Flush(1));
+
+  sleeping_task_low.WakeUp();
+  sleeping_task_low.WaitUntilDone();
+}
+
+TEST_F(DBBasicTest, FLUSH) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    WriteOptions writeOpt = WriteOptions();
+    writeOpt.disableWAL = true;
+    SetPerfLevel(kEnableTime);
+    ;
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "foo", "v1"));
+    // this will now also flush the last 2 writes
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v1"));
+
+    perf_context.Reset();
+    Get(1, "foo");
+    ASSERT_TRUE((int)perf_context.get_from_output_files_time > 0);
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("v1", Get(1, "bar"));
+
+    writeOpt.disableWAL = true;
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v2"));
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "foo", "v2"));
+    ASSERT_OK(Flush(1));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v2", Get(1, "bar"));
+    perf_context.Reset();
+    ASSERT_EQ("v2", Get(1, "foo"));
+    ASSERT_TRUE((int)perf_context.get_from_output_files_time > 0);
+
+    writeOpt.disableWAL = false;
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "bar", "v3"));
+    ASSERT_OK(dbfull()->Put(writeOpt, handles_[1], "foo", "v3"));
+    ASSERT_OK(Flush(1));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    // 'foo' should be there because its put
+    // has WAL enabled.
+    ASSERT_EQ("v3", Get(1, "foo"));
+    ASSERT_EQ("v3", Get(1, "bar"));
+
+    SetPerfLevel(kDisable);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, ManifestRollOver) {
+  do {
+    Options options;
+    options.max_manifest_file_size = 10;  // 10 bytes
+    options = CurrentOptions(options);
+    CreateAndReopenWithCF({"pikachu"}, options);
+    {
+      ASSERT_OK(Put(1, "manifest_key1", std::string(1000, '1')));
+      ASSERT_OK(Put(1, "manifest_key2", std::string(1000, '2')));
+      ASSERT_OK(Put(1, "manifest_key3", std::string(1000, '3')));
+      uint64_t manifest_before_flush = dbfull()->TEST_Current_Manifest_FileNo();
+      ASSERT_OK(Flush(1));  // This should trigger LogAndApply.
+      uint64_t manifest_after_flush = dbfull()->TEST_Current_Manifest_FileNo();
+      ASSERT_GT(manifest_after_flush, manifest_before_flush);
+      ReopenWithColumnFamilies({"default", "pikachu"}, options);
+      ASSERT_GT(dbfull()->TEST_Current_Manifest_FileNo(), manifest_after_flush);
+      // check if a new manifest file got inserted or not.
+      ASSERT_EQ(std::string(1000, '1'), Get(1, "manifest_key1"));
+      ASSERT_EQ(std::string(1000, '2'), Get(1, "manifest_key2"));
+      ASSERT_EQ(std::string(1000, '3'), Get(1, "manifest_key3"));
+    }
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, IdentityAcrossRestarts) {
+  do {
+    std::string id1;
+    ASSERT_OK(db_->GetDbIdentity(id1));
+
+    Options options = CurrentOptions();
+    Reopen(options);
+    std::string id2;
+    ASSERT_OK(db_->GetDbIdentity(id2));
+    // id1 should match id2 because identity was not regenerated
+    ASSERT_EQ(id1.compare(id2), 0);
+
+    std::string idfilename = IdentityFileName(dbname_);
+    ASSERT_OK(env_->DeleteFile(idfilename));
+    Reopen(options);
+    std::string id3;
+    ASSERT_OK(db_->GetDbIdentity(id3));
+    // id1 should NOT match id3 because identity was regenerated
+    ASSERT_NE(id1.compare(id3), 0);
+  } while (ChangeCompactOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBBasicTest, Snapshot) {
+  anon::OptionsOverride options_override;
+  options_override.skip_policy = kSkipNoSnapshot;
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions(options_override));
+    Put(0, "foo", "0v1");
+    Put(1, "foo", "1v1");
+
+    const Snapshot* s1 = db_->GetSnapshot();
+    ASSERT_EQ(1U, GetNumSnapshots());
+    uint64_t time_snap1 = GetTimeOldestSnapshots();
+    ASSERT_GT(time_snap1, 0U);
+    Put(0, "foo", "0v2");
+    Put(1, "foo", "1v2");
+
+    env_->addon_time_.fetch_add(1);
+
+    const Snapshot* s2 = db_->GetSnapshot();
+    ASSERT_EQ(2U, GetNumSnapshots());
+    ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    Put(0, "foo", "0v3");
+    Put(1, "foo", "1v3");
+
+    {
+      ManagedSnapshot s3(db_);
+      ASSERT_EQ(3U, GetNumSnapshots());
+      ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+
+      Put(0, "foo", "0v4");
+      Put(1, "foo", "1v4");
+      ASSERT_EQ("0v1", Get(0, "foo", s1));
+      ASSERT_EQ("1v1", Get(1, "foo", s1));
+      ASSERT_EQ("0v2", Get(0, "foo", s2));
+      ASSERT_EQ("1v2", Get(1, "foo", s2));
+      ASSERT_EQ("0v3", Get(0, "foo", s3.snapshot()));
+      ASSERT_EQ("1v3", Get(1, "foo", s3.snapshot()));
+      ASSERT_EQ("0v4", Get(0, "foo"));
+      ASSERT_EQ("1v4", Get(1, "foo"));
+    }
+
+    ASSERT_EQ(2U, GetNumSnapshots());
+    ASSERT_EQ(time_snap1, GetTimeOldestSnapshots());
+    ASSERT_EQ("0v1", Get(0, "foo", s1));
+    ASSERT_EQ("1v1", Get(1, "foo", s1));
+    ASSERT_EQ("0v2", Get(0, "foo", s2));
+    ASSERT_EQ("1v2", Get(1, "foo", s2));
+    ASSERT_EQ("0v4", Get(0, "foo"));
+    ASSERT_EQ("1v4", Get(1, "foo"));
+
+    db_->ReleaseSnapshot(s1);
+    ASSERT_EQ("0v2", Get(0, "foo", s2));
+    ASSERT_EQ("1v2", Get(1, "foo", s2));
+    ASSERT_EQ("0v4", Get(0, "foo"));
+    ASSERT_EQ("1v4", Get(1, "foo"));
+    ASSERT_EQ(1U, GetNumSnapshots());
+    ASSERT_LT(time_snap1, GetTimeOldestSnapshots());
+
+    db_->ReleaseSnapshot(s2);
+    ASSERT_EQ(0U, GetNumSnapshots());
+    ASSERT_EQ("0v4", Get(0, "foo"));
+    ASSERT_EQ("1v4", Get(1, "foo"));
+  } while (ChangeOptions(kSkipHashCuckoo));
+}
+
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBBasicTest, CompactBetweenSnapshots) {
+  anon::OptionsOverride options_override;
+  options_override.skip_policy = kSkipNoSnapshot;
+  do {
+    Options options = CurrentOptions(options_override);
+    options.disable_auto_compactions = true;
+    CreateAndReopenWithCF({"pikachu"}, options);
+    Random rnd(301);
+    FillLevels("a", "z", 1);
+
+    Put(1, "foo", "first");
+    const Snapshot* snapshot1 = db_->GetSnapshot();
+    Put(1, "foo", "second");
+    Put(1, "foo", "third");
+    Put(1, "foo", "fourth");
+    const Snapshot* snapshot2 = db_->GetSnapshot();
+    Put(1, "foo", "fifth");
+    Put(1, "foo", "sixth");
+
+    // All entries (including duplicates) exist
+    // before any compaction or flush is triggered.
+    ASSERT_EQ(AllEntriesFor("foo", 1),
+              "[ sixth, fifth, fourth, third, second, first ]");
+    ASSERT_EQ("sixth", Get(1, "foo"));
+    ASSERT_EQ("fourth", Get(1, "foo", snapshot2));
+    ASSERT_EQ("first", Get(1, "foo", snapshot1));
+
+    // After a flush, "second", "third" and "fifth" should
+    // be removed
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ sixth, fourth, first ]");
+
+    // after we release the snapshot1, only two values left
+    db_->ReleaseSnapshot(snapshot1);
+    FillLevels("a", "z", 1);
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+
+    // We have only one valid snapshot snapshot2. Since snapshot1 is
+    // not valid anymore, "first" should be removed by a compaction.
+    ASSERT_EQ("sixth", Get(1, "foo"));
+    ASSERT_EQ("fourth", Get(1, "foo", snapshot2));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ sixth, fourth ]");
+
+    // after we release the snapshot2, only one value should be left
+    db_->ReleaseSnapshot(snapshot2);
+    FillLevels("a", "z", 1);
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ("sixth", Get(1, "foo"));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ sixth ]");
+    // skip HashCuckooRep as it does not support snapshot
+  } while (ChangeOptions(kSkipHashCuckoo | kSkipFIFOCompaction));
+}
+
+TEST_F(DBBasicTest, DBOpen_Options) {
+  Options options = CurrentOptions();
+  std::string dbname = test::TmpDir(env_) + "/db_options_test";
+  ASSERT_OK(DestroyDB(dbname, options));
+
+  // Does not exist, and create_if_missing == false: error
+  DB* db = nullptr;
+  options.create_if_missing = false;
+  Status s = DB::Open(options, dbname, &db);
+  ASSERT_TRUE(strstr(s.ToString().c_str(), "does not exist") != nullptr);
+  ASSERT_TRUE(db == nullptr);
+
+  // Does not exist, and create_if_missing == true: OK
+  options.create_if_missing = true;
+  s = DB::Open(options, dbname, &db);
+  ASSERT_OK(s);
+  ASSERT_TRUE(db != nullptr);
+
+  delete db;
+  db = nullptr;
+
+  // Does exist, and error_if_exists == true: error
+  options.create_if_missing = false;
+  options.error_if_exists = true;
+  s = DB::Open(options, dbname, &db);
+  ASSERT_TRUE(strstr(s.ToString().c_str(), "exists") != nullptr);
+  ASSERT_TRUE(db == nullptr);
+
+  // Does exist, and error_if_exists == false: OK
+  options.create_if_missing = true;
+  options.error_if_exists = false;
+  s = DB::Open(options, dbname, &db);
+  ASSERT_OK(s);
+  ASSERT_TRUE(db != nullptr);
+
+  delete db;
+  db = nullptr;
+}
+
+TEST_F(DBBasicTest, CompactOnFlush) {
+  anon::OptionsOverride options_override;
+  options_override.skip_policy = kSkipNoSnapshot;
+  do {
+    Options options = CurrentOptions(options_override);
+    options.disable_auto_compactions = true;
+    CreateAndReopenWithCF({"pikachu"}, options);
+
+    Put(1, "foo", "v1");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v1 ]");
+
+    // Write two new keys
+    Put(1, "a", "begin");
+    Put(1, "z", "end");
+    Flush(1);
+
+    // Case1: Delete followed by a put
+    Delete(1, "foo");
+    Put(1, "foo", "v2");
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v2, DEL, v1 ]");
+
+    // After the current memtable is flushed, the DEL should
+    // have been removed
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v2, v1 ]");
+
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v2 ]");
+
+    // Case 2: Delete followed by another delete
+    Delete(1, "foo");
+    Delete(1, "foo");
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ DEL, DEL, v2 ]");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ DEL, v2 ]");
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ ]");
+
+    // Case 3: Put followed by a delete
+    Put(1, "foo", "v3");
+    Delete(1, "foo");
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ DEL, v3 ]");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ DEL ]");
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ ]");
+
+    // Case 4: Put followed by another Put
+    Put(1, "foo", "v4");
+    Put(1, "foo", "v5");
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v5, v4 ]");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v5 ]");
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v5 ]");
+
+    // clear database
+    Delete(1, "foo");
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ ]");
+
+    // Case 5: Put followed by snapshot followed by another Put
+    // Both puts should remain.
+    Put(1, "foo", "v6");
+    const Snapshot* snapshot = db_->GetSnapshot();
+    Put(1, "foo", "v7");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v7, v6 ]");
+    db_->ReleaseSnapshot(snapshot);
+
+    // clear database
+    Delete(1, "foo");
+    dbfull()->CompactRange(CompactRangeOptions(), handles_[1], nullptr,
+                           nullptr);
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ ]");
+
+    // Case 5: snapshot followed by a put followed by another Put
+    // Only the last put should remain.
+    const Snapshot* snapshot1 = db_->GetSnapshot();
+    Put(1, "foo", "v8");
+    Put(1, "foo", "v9");
+    ASSERT_OK(Flush(1));
+    ASSERT_EQ(AllEntriesFor("foo", 1), "[ v9 ]");
+    db_->ReleaseSnapshot(snapshot1);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, FlushOneColumnFamily) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  ASSERT_OK(Put(0, "Default", "Default"));
+  ASSERT_OK(Put(1, "pikachu", "pikachu"));
+  ASSERT_OK(Put(2, "ilya", "ilya"));
+  ASSERT_OK(Put(3, "muromec", "muromec"));
+  ASSERT_OK(Put(4, "dobrynia", "dobrynia"));
+  ASSERT_OK(Put(5, "nikitich", "nikitich"));
+  ASSERT_OK(Put(6, "alyosha", "alyosha"));
+  ASSERT_OK(Put(7, "popovich", "popovich"));
+
+  for (int i = 0; i < 8; ++i) {
+    Flush(i);
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), i + 1U);
+  }
+}
+
+TEST_F(DBBasicTest, MultiGetSimple) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "k1", "v1"));
+    ASSERT_OK(Put(1, "k2", "v2"));
+    ASSERT_OK(Put(1, "k3", "v3"));
+    ASSERT_OK(Put(1, "k4", "v4"));
+    ASSERT_OK(Delete(1, "k4"));
+    ASSERT_OK(Put(1, "k5", "v5"));
+    ASSERT_OK(Delete(1, "no_key"));
+
+    std::vector<Slice> keys({"k1", "k2", "k3", "k4", "k5", "no_key"});
+
+    std::vector<std::string> values(20, "Temporary data to be overwritten");
+    std::vector<ColumnFamilyHandle*> cfs(keys.size(), handles_[1]);
+
+    std::vector<Status> s = db_->MultiGet(ReadOptions(), cfs, keys, &values);
+    ASSERT_EQ(values.size(), keys.size());
+    ASSERT_EQ(values[0], "v1");
+    ASSERT_EQ(values[1], "v2");
+    ASSERT_EQ(values[2], "v3");
+    ASSERT_EQ(values[4], "v5");
+
+    ASSERT_OK(s[0]);
+    ASSERT_OK(s[1]);
+    ASSERT_OK(s[2]);
+    ASSERT_TRUE(s[3].IsNotFound());
+    ASSERT_OK(s[4]);
+    ASSERT_TRUE(s[5].IsNotFound());
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, MultiGetEmpty) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    // Empty Key Set
+    std::vector<Slice> keys;
+    std::vector<std::string> values;
+    std::vector<ColumnFamilyHandle*> cfs;
+    std::vector<Status> s = db_->MultiGet(ReadOptions(), cfs, keys, &values);
+    ASSERT_EQ(s.size(), 0U);
+
+    // Empty Database, Empty Key Set
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    DestroyAndReopen(options);
+    CreateAndReopenWithCF({"pikachu"}, options);
+    s = db_->MultiGet(ReadOptions(), cfs, keys, &values);
+    ASSERT_EQ(s.size(), 0U);
+
+    // Empty Database, Search for Keys
+    keys.resize(2);
+    keys[0] = "a";
+    keys[1] = "b";
+    cfs.push_back(handles_[0]);
+    cfs.push_back(handles_[1]);
+    s = db_->MultiGet(ReadOptions(), cfs, keys, &values);
+    ASSERT_EQ(static_cast<int>(s.size()), 2);
+    ASSERT_TRUE(s[0].IsNotFound() && s[1].IsNotFound());
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, ChecksumTest) {
+  BlockBasedTableOptions table_options;
+  Options options = CurrentOptions();
+
+  table_options.checksum = kCRC32c;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_OK(Put("a", "b"));
+  ASSERT_OK(Put("c", "d"));
+  ASSERT_OK(Flush());  // table with crc checksum
+
+  table_options.checksum = kxxHash;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_OK(Put("e", "f"));
+  ASSERT_OK(Put("g", "h"));
+  ASSERT_OK(Flush());  // table with xxhash checksum
+
+  table_options.checksum = kCRC32c;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_EQ("b", Get("a"));
+  ASSERT_EQ("d", Get("c"));
+  ASSERT_EQ("f", Get("e"));
+  ASSERT_EQ("h", Get("g"));
+
+  table_options.checksum = kCRC32c;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  ASSERT_EQ("b", Get("a"));
+  ASSERT_EQ("d", Get("c"));
+  ASSERT_EQ("f", Get("e"));
+  ASSERT_EQ("h", Get("g"));
+}
+
+// On Windows you can have either memory mapped file or a file
+// with unbuffered access. So this asserts and does not make
+// sense to run
+#ifndef OS_WIN
+TEST_F(DBBasicTest, MmapAndBufferOptions) {
+  Options options = CurrentOptions();
+
+  options.use_direct_reads = true;
+  options.allow_mmap_reads = true;
+  ASSERT_NOK(TryReopen(options));
+
+  // All other combinations are acceptable
+  options.use_direct_reads = false;
+  ASSERT_OK(TryReopen(options));
+
+  if (IsDirectIOSupported()) {
+    options.use_direct_reads = true;
+    options.allow_mmap_reads = false;
+    ASSERT_OK(TryReopen(options));
+  }
+
+  options.use_direct_reads = false;
+  ASSERT_OK(TryReopen(options));
+}
+#endif
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -788,8 +788,6 @@ void PrefixScanInit(DBBloomFilterTest* dbtest) {
 }  // namespace

 TEST_F(DBBloomFilterTest, PrefixScan) {
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip,
-             kSkipNoPrefix);
  while (ChangeFilterOptions()) {
    int count;
    Slice prefix;
@@ -836,7 +834,6 @@ TEST_F(DBBloomFilterTest, PrefixScan) {
    ASSERT_EQ(env_->random_read_counter_.Read(), 2);
    Close();
  }  // end of while
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip, 0);
 }

 TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
@@ -9,6 +9,7 @@

 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "port/port.h"
 #include "rocksdb/experimental.h"
 #include "rocksdb/utilities/convenience.h"
 #include "util/sync_point.h"
@@ -1101,7 +1102,7 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
  ASSERT_EQ(trivial_move, 6);
  ASSERT_EQ(non_trivial_move, 0);

-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    compact_options.change_level = false;
    compact_options.exclusive_manual_compaction = false;
    std::string begin_string = Key(0);
@@ -1233,7 +1234,7 @@ TEST_F(DBCompactionTest, ManualPartialFill) {
  ASSERT_EQ(trivial_move, 2);
  ASSERT_EQ(non_trivial_move, 0);

-  std::thread threads([&] {
+  rocksdb::port::Thread threads([&] {
    compact_options.change_level = false;
    compact_options.exclusive_manual_compaction = false;
    std::string begin_string = Key(0);
@@ -13,6 +13,7 @@
 #if !defined(ROCKSDB_LITE)

 #include "db/db_test_util.h"
+#include "port/port.h"
 #include "port/stack_trace.h"

 namespace rocksdb {
@@ -251,7 +252,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
  });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  std::thread thread([this] {
+  rocksdb::port::Thread thread([this] {
    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_start");
    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_finish");
@@ -406,7 +407,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBaseInc) {
  env_->SetBackgroundThreads(1, Env::HIGH);
 }

-TEST_F(DBTestDynamicLevel, MigrateToDynamicLevelMaxBytesBase) {
+TEST_F(DBTestDynamicLevel, DISABLED_MigrateToDynamicLevelMaxBytesBase) {
  Random rnd(301);
  const int kMaxKey = 2000;

@@ -462,7 +463,7 @@ TEST_F(DBTestDynamicLevel, MigrateToDynamicLevelMaxBytesBase) {
  compaction_finished = false;
  // Issue manual compaction in one thread and still verify DB state
  // in main thread.
-  std::thread t([&]() {
+  rocksdb::port::Thread t([&]() {
    CompactRangeOptions compact_options;
    compact_options.change_level = true;
    compact_options.target_level = options.num_levels - 1;
@@ -27,7 +27,6 @@
 #include "util/file_util.h"
 #include "util/mutexlock.h"
 #include "util/sync_point.h"
-#include "util/testharness.h"

 namespace rocksdb {

@@ -9,6 +9,7 @@

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

 namespace rocksdb {
@@ -47,6 +48,40 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
 #endif  // ROCKSDB_LITE
 }

+TEST_F(DBFlushTest, SyncFail) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options;
+  options.disable_auto_compactions = true;
+  options.env = fault_injection_env.get();
+
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBFlushTest::SyncFail:1", "DBImpl::SyncClosedLogs:Start"},
+       {"DBImpl::SyncClosedLogs:Failed", "DBFlushTest::SyncFail:2"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  Reopen(options);
+  Put("key", "value");
+  auto* cfd =
+      reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
+          ->cfd();
+  int refs_before = cfd->current()->TEST_refs();
+  FlushOptions flush_options;
+  flush_options.wait = false;
+  ASSERT_OK(dbfull()->Flush(flush_options));
+  fault_injection_env->SetFilesystemActive(false);
+  TEST_SYNC_POINT("DBFlushTest::SyncFail:1");
+  TEST_SYNC_POINT("DBFlushTest::SyncFail:2");
+  fault_injection_env->SetFilesystemActive(true);
+  dbfull()->TEST_WaitForFlushMemTable();
+#ifndef ROCKSDB_LITE
+  ASSERT_EQ("", FilesPerLevel());  // flush failed.
+#endif                             // ROCKSDB_LITE
+  // Flush job should release ref count to current version.
+  ASSERT_EQ(refs_before, cfd->current()->TEST_refs());
+  Destroy(options);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -58,7 +58,6 @@
 #include "db/version_set.h"
 #include "db/write_batch_internal.h"
 #include "db/write_callback.h"
-#include "db/xfunc_test_points.h"
 #include "memtable/hash_linklist_rep.h"
 #include "memtable/hash_skiplist_rep.h"
 #include "port/likely.h"
@@ -76,7 +75,7 @@
 #include "rocksdb/write_buffer_manager.h"
 #include "table/block.h"
 #include "table/block_based_table_factory.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "table/table_builder.h"
 #include "table/two_level_iterator.h"
 #include "util/autovector.h"
@@ -100,7 +99,6 @@
 #include "util/sync_point.h"
 #include "util/thread_status_updater.h"
 #include "util/thread_status_util.h"
-#include "util/xfunc.h"

 namespace rocksdb {

@@ -150,6 +148,7 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
      result.info_log = nullptr;
    }
  }
+
  if (!result.write_buffer_manager) {
    result.write_buffer_manager.reset(
        new WriteBufferManager(result.db_write_buffer_size));
@@ -198,6 +197,10 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
    result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
  }

+  if (result.use_direct_reads && result.compaction_readahead_size == 0) {
+    result.compaction_readahead_size = 1024 * 1024 * 2;
+  }
+
  if (result.compaction_readahead_size > 0) {
    result.new_table_reader_for_compaction_inputs = true;
  }
@@ -255,11 +258,21 @@ static Status ValidateOptions(
        "More than four DB paths are not supported yet. ");
  }

-  if (db_options.allow_mmap_reads && !db_options.allow_os_buffer) {
+  if (db_options.allow_mmap_reads && db_options.use_direct_reads) {
    // Protect against assert in PosixMMapReadableFile constructor
    return Status::NotSupported(
        "If memory mapped reads (allow_mmap_reads) are enabled "
-        "then os caching (allow_os_buffer) must also be enabled. ");
+        "then direct I/O reads (use_direct_reads) must be disabled. ");
+  }
+
+  if (db_options.allow_mmap_writes && db_options.use_direct_writes) {
+    return Status::NotSupported(
+        "If memory mapped writes (allow_mmap_writes) are enabled "
+        "then direct I/O writes (use_direct_writes) must be disabled. ");
+  }
+
+  if (db_options.keep_log_file_num == 0) {
+    return Status::InvalidArgument("keep_log_file_num must be greater than 0");
  }

  return Status::OK();
@@ -332,12 +345,11 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
      num_running_flushes_(0),
      bg_purge_scheduled_(0),
      disable_delete_obsolete_files_(0),
-      delete_obsolete_files_next_run_(
-          env_->NowMicros() +
-          immutable_db_options_.delete_obsolete_files_period_micros),
+      delete_obsolete_files_last_run_(env_->NowMicros()),
      last_stats_dump_time_microsec_(0),
      next_job_id_(1),
      has_unpersisted_data_(false),
+      unable_to_flush_oldest_log_(false),
      env_options_(BuildDBOptions(immutable_db_options_, mutable_db_options_)),
      num_running_ingest_file_(0),
 #ifndef ROCKSDB_LITE
@@ -375,8 +387,11 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
 void DBImpl::CancelAllBackgroundWork(bool wait) {
  InstrumentedMutexLock l(&mutex_);

+  Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
+      "Shutdown: canceling all background work");
+
  if (!shutting_down_.load(std::memory_order_acquire) &&
-      has_unpersisted_data_ &&
+      has_unpersisted_data_.load(std::memory_order_relaxed) &&
      !mutable_db_options_.avoid_flush_during_shutdown) {
    for (auto cfd : *versions_->GetColumnFamilySet()) {
      if (!cfd->IsDropped() && !cfd->mem()->IsEmpty()) {
@@ -499,6 +514,8 @@ DBImpl::~DBImpl() {
    env_->UnlockFile(db_lock_);
  }

+  Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
+      "Shutdown complete");
  LogFlush(immutable_db_options_.info_log);
 }

@@ -592,7 +609,7 @@ static void DumpMallocStats(std::string* stats) {
 #ifdef ROCKSDB_JEMALLOC
  MallocStatus mstat;
  const uint kMallocStatusLen = 1000000;
-  std::unique_ptr<char> buf{new char[kMallocStatusLen + 1]};
+  std::unique_ptr<char[]> buf{new char[kMallocStatusLen + 1]};
  mstat.cur = buf.get();
  mstat.end = buf.get() + kMallocStatusLen;
  malloc_stats_print(GetJemallocStatus, &mstat, "");
@@ -654,6 +671,10 @@ void DBImpl::MaybeDumpStats() {
 }

 uint64_t DBImpl::FindMinPrepLogReferencedByMemTable() {
+  if (!allow_2pc()) {
+    return 0;
+  }
+
  uint64_t min_log = 0;

  // we must look through the memtables for two phase transactions
@@ -698,6 +719,11 @@ void DBImpl::MarkLogAsContainingPrepSection(uint64_t log) {
 }

 uint64_t DBImpl::FindMinLogContainingOutstandingPrep() {
+
+  if (!allow_2pc()) {
+    return 0;
+  }
+
  std::lock_guard<std::mutex> lock(prep_heap_mutex_);
  uint64_t min_log = 0;

@@ -736,6 +762,34 @@ void DBImpl::ScheduleBgLogWriterClose(JobContext* job_context) {
  }
 }

+uint64_t DBImpl::MinLogNumberToKeep() {
+  uint64_t log_number = versions_->MinLogNumber();
+
+  if (allow_2pc()) {
+    // if are 2pc we must consider logs containing prepared
+    // sections of outstanding transactions.
+    //
+    // We must check min logs with outstanding prep before we check
+    // logs referneces by memtables because a log referenced by the
+    // first data structure could transition to the second under us.
+    //
+    // TODO(horuff): iterating over all column families under db mutex.
+    // should find more optimial solution
+    auto min_log_in_prep_heap = FindMinLogContainingOutstandingPrep();
+
+    if (min_log_in_prep_heap != 0 && min_log_in_prep_heap < log_number) {
+      log_number = min_log_in_prep_heap;
+    }
+
+    auto min_log_refed_by_mem = FindMinPrepLogReferencedByMemTable();
+
+    if (min_log_refed_by_mem != 0 && min_log_refed_by_mem < log_number) {
+      log_number = min_log_refed_by_mem;
+    }
+  }
+  return log_number;
+}
+
 // * Returns the list of live files in 'sst_live'
 // If it's doing full scan:
 // * Returns the list of all files in the filesystem in
@@ -743,7 +797,7 @@ void DBImpl::ScheduleBgLogWriterClose(JobContext* job_context) {
 // Otherwise, gets obsolete files from VersionSet.
 // no_full_scan = true -- never do the full scan using GetChildren()
 // force = false -- don't force the full scan, except every
-//  immutable_db_options_.delete_obsolete_files_period_micros
+//  mutable_db_options_.delete_obsolete_files_period_micros
 // force = true -- force the full scan
 void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
                               bool no_full_scan) {
@@ -760,15 +814,15 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  if (no_full_scan) {
    doing_the_full_scan = false;
  } else if (force ||
-             immutable_db_options_.delete_obsolete_files_period_micros == 0) {
+             mutable_db_options_.delete_obsolete_files_period_micros == 0) {
    doing_the_full_scan = true;
  } else {
    const uint64_t now_micros = env_->NowMicros();
-    if (delete_obsolete_files_next_run_ < now_micros) {
+    if ((delete_obsolete_files_last_run_ +
+         mutable_db_options_.delete_obsolete_files_period_micros) <
+        now_micros) {
      doing_the_full_scan = true;
-      delete_obsolete_files_next_run_ =
-          now_micros +
-          immutable_db_options_.delete_obsolete_files_period_micros;
+      delete_obsolete_files_last_run_ = now_micros;
    }
  }

@@ -794,32 +848,7 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  job_context->manifest_file_number = versions_->manifest_file_number();
  job_context->pending_manifest_file_number =
      versions_->pending_manifest_file_number();
-  job_context->log_number = versions_->MinLogNumber();
-
-  if (allow_2pc()) {
-    // if are 2pc we must consider logs containing prepared
-    // sections of outstanding transactions.
-    //
-    // We must check min logs with outstanding prep before we check
-    // logs referneces by memtables because a log referenced by the
-    // first data structure could transition to the second under us.
-    //
-    // TODO(horuff): iterating over all column families under db mutex.
-    // should find more optimial solution
-    auto min_log_in_prep_heap = FindMinLogContainingOutstandingPrep();
-
-    if (min_log_in_prep_heap != 0 &&
-        min_log_in_prep_heap < job_context->log_number) {
-      job_context->log_number = min_log_in_prep_heap;
-    }
-
-    auto min_log_refed_by_mem = FindMinPrepLogReferencedByMemTable();
-
-    if (min_log_refed_by_mem != 0 &&
-        min_log_refed_by_mem < job_context->log_number) {
-      job_context->log_number = min_log_refed_by_mem;
-    }
-  }
+  job_context->log_number = MinLogNumberToKeep();

  job_context->prev_log_number = versions_->prev_log_number();

@@ -1820,6 +1849,7 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
 }

 Status DBImpl::SyncClosedLogs(JobContext* job_context) {
+  TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
  mutex_.AssertHeld();
  autovector<log::Writer*, 1> logs_to_sync;
  uint64_t current_log_number = logfile_number_;
@@ -1856,6 +1886,7 @@ Status DBImpl::SyncClosedLogs(JobContext* job_context) {
    MarkLogsSynced(current_log_number - 1, true, s);
    if (!s.ok()) {
      bg_error_ = s;
+      TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Failed");
      return s;
    }
  }
@@ -1887,8 +1918,7 @@ Status DBImpl::FlushMemTableToOutputFile(

  Status s;
  if (logfile_number_ > 0 &&
-      versions_->GetColumnFamilySet()->NumberOfColumnFamilies() > 0 &&
-      !immutable_db_options_.disable_data_sync) {
+      versions_->GetColumnFamilySet()->NumberOfColumnFamilies() > 0) {
    // If there are more than one column families, we need to make sure that
    // all the log files except the most recent one are synced. Otherwise if
    // the host crashes after flushing and before WAL is persistent, the
@@ -1906,6 +1936,8 @@ Status DBImpl::FlushMemTableToOutputFile(
  // is unlocked by the current thread.
  if (s.ok()) {
    s = flush_job.Run(&file_meta);
+  } else {
+    flush_job.Cancel();
  }

  if (s.ok()) {
@@ -1930,7 +1962,6 @@ Status DBImpl::FlushMemTableToOutputFile(
    // may temporarily unlock and lock the mutex.
    NotifyOnFlushCompleted(cfd, &file_meta, mutable_cf_options,
                           job_context->job_id, flush_job.GetTableProperties());
-#endif  // ROCKSDB_LITE
    auto sfm = static_cast<SstFileManagerImpl*>(
        immutable_db_options_.sst_file_manager.get());
    if (sfm) {
@@ -1940,10 +1971,12 @@ Status DBImpl::FlushMemTableToOutputFile(
      sfm->OnAddFile(file_path);
      if (sfm->IsMaxAllowedSpaceReached() && bg_error_.ok()) {
        bg_error_ = Status::IOError("Max allowed space was reached");
-        TEST_SYNC_POINT(
-            "DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached");
+        TEST_SYNC_POINT_CALLBACK(
+            "DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
+            &bg_error_);
      }
    }
+#endif  // ROCKSDB_LITE
  }
  return s;
 }
@@ -2123,7 +2156,7 @@ Status DBImpl::CompactFiles(
                       immutable_db_options_.info_log.get());

  // Perform CompactFiles
-  SuperVersion* sv = GetAndRefSuperVersion(cfd);
+  SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
  {
    InstrumentedMutexLock l(&mutex_);

@@ -2135,7 +2168,12 @@ Status DBImpl::CompactFiles(
                         input_file_names, output_level,
                         output_path_id, &job_context, &log_buffer);
  }
-  ReturnAndCleanupSuperVersion(cfd, sv);
+  if (sv->Unref()) {
+    mutex_.Lock();
+    sv->Cleanup();
+    mutex_.Unlock();
+    delete sv;
+  }

  // Find and delete obsolete files
  {
@@ -2493,7 +2531,7 @@ Status DBImpl::SetDBOptions(
      mutable_db_options_ = new_options;

      if (total_log_size_ > GetMaxTotalWalSize()) {
-        FlushColumnFamilies();
+        MaybeFlushColumnFamilies();
      }

      persist_options_status = PersistOptions();
@@ -2740,7 +2778,7 @@ void DBImpl::MarkLogsSynced(
      ++it;
    }
  }
-  assert(logs_.empty() || logs_[0].number > up_to ||
+  assert(!status.ok() || logs_.empty() || logs_[0].number > up_to ||
         (logs_.size() == 1 && !logs_[0].getting_synced));
  log_sync_cv_.SignalAll();
 }
@@ -2939,6 +2977,12 @@ Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd) {
    if (shutting_down_.load(std::memory_order_acquire)) {
      return Status::ShutdownInProgress();
    }
+    if (cfd->IsDropped()) {
+      // FlushJob cannot flush a dropped CF, if we did not break here
+      // we will loop forever since cfd->imm()->NumNotFlushed() will never
+      // drop to zero
+      return Status::InvalidArgument("Cannot flush a dropped CF");
+    }
    bg_cv_.Wait();
  }
  if (!bg_error_.ok()) {
@@ -3875,8 +3919,10 @@ InternalIterator* DBImpl::NewInternalIterator(
  }
  if (s.ok()) {
    // Collect iterators for files in L0 - Ln
-    super_version->current->AddIterators(read_options, env_options_,
-                                         &merge_iter_builder, range_del_agg);
+    if (read_options.read_tier != kMemtableTier) {
+      super_version->current->AddIterators(read_options, env_options_,
+                                           &merge_iter_builder, range_del_agg);
+    }
    internal_iter = merge_iter_builder.Finish();
    IterState* cleanup =
        new IterState(this, &mutex_, super_version,
@@ -3959,15 +4005,32 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();

+  // Acquire SuperVersion
+  SuperVersion* sv = GetAndRefSuperVersion(cfd);
+
+  TEST_SYNC_POINT("DBImpl::GetImpl:1");
+  TEST_SYNC_POINT("DBImpl::GetImpl:2");
+
  SequenceNumber snapshot;
  if (read_options.snapshot != nullptr) {
    snapshot = reinterpret_cast<const SnapshotImpl*>(
        read_options.snapshot)->number_;
  } else {
+    // Since we get and reference the super version before getting
+    // the snapshot number, without a mutex protection, it is possible
+    // that a memtable switch happened in the middle and not all the
+    // data for this snapshot is available. But it will contain all
+    // the data available in the super version we have, which is also
+    // a valid snapshot to read from.
+    // We shouldn't get snapshot before finding and referencing the
+    // super versipon because a flush happening in between may compact
+    // away data for the snapshot, but the snapshot is earlier than the
+    // data overwriting it, so users may see wrong results.
    snapshot = versions_->LastSequence();
  }
-  // Acquire SuperVersion
-  SuperVersion* sv = GetAndRefSuperVersion(cfd);
+  TEST_SYNC_POINT("DBImpl::GetImpl:3");
+  TEST_SYNC_POINT("DBImpl::GetImpl:4");
+
  // Prepare to store a list of merge operations if merge occurs.
  MergeContext merge_context;
  RangeDelAggregator range_del_agg(cfd->internal_comparator(), snapshot);
@@ -3979,8 +4042,8 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
  LookupKey lkey(key, snapshot);
  PERF_TIMER_STOP(get_snapshot_time);

-  bool skip_memtable =
-      (read_options.read_tier == kPersistedTier && has_unpersisted_data_);
+  bool skip_memtable = (read_options.read_tier == kPersistedTier &&
+                        has_unpersisted_data_.load(std::memory_order_relaxed));
  bool done = false;
  if (!skip_memtable) {
    if (sv->mem->Get(lkey, value, &s, &merge_context, &range_del_agg,
@@ -4085,7 +4148,8 @@ std::vector<Status> DBImpl::MultiGet(
    auto mgd = mgd_iter->second;
    auto super_version = mgd->super_version;
    bool skip_memtable =
-        (read_options.read_tier == kPersistedTier && has_unpersisted_data_);
+        (read_options.read_tier == kPersistedTier &&
+         has_unpersisted_data_.load(std::memory_order_relaxed));
    bool done = false;
    if (!skip_memtable) {
      if (super_version->mem->Get(lkey, value, &s, &merge_context,
@@ -4335,10 +4399,6 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
  }
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();
-
-  XFUNC_TEST("", "managed_new", managed_new1, xf_manage_new,
-             reinterpret_cast<DBImpl*>(this),
-             const_cast<ReadOptions*>(&read_options), is_snapshot_supported_);
  if (read_options.managed) {
 #ifdef ROCKSDB_LITE
    // not supported in lite version
@@ -4447,9 +4507,6 @@ Status DBImpl::NewIterators(
  }
  iterators->clear();
  iterators->reserve(column_families.size());
-  XFUNC_TEST("", "managed_new", managed_new1, xf_manage_new,
-             reinterpret_cast<DBImpl*>(this),
-             const_cast<ReadOptions*>(&read_options), is_snapshot_supported_);
  if (read_options.managed) {
 #ifdef ROCKSDB_LITE
    return Status::InvalidArgument(
@@ -4589,22 +4646,9 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  if (my_batch == nullptr) {
    return Status::Corruption("Batch is nullptr!");
  }
-  if (write_options.timeout_hint_us != 0) {
-    return Status::InvalidArgument("timeout_hint_us is deprecated");
-  }

  Status status;

-  bool xfunc_attempted_write = false;
-  XFUNC_TEST("transaction", "transaction_xftest_write_impl",
-             xf_transaction_write1, xf_transaction_write, write_options,
-             immutable_db_options_, my_batch, callback, this, &status,
-             &xfunc_attempted_write);
-  if (xfunc_attempted_write) {
-    // Test already did the write
-    return status;
-  }
-
  PERF_TIMER_GUARD(write_pre_and_post_process_time);
  WriteThread::Writer w;
  w.batch = my_batch;
@@ -4680,10 +4724,10 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
         versions_->GetColumnFamilySet()->NumberOfColumnFamilies() == 1);

  if (UNLIKELY(!single_column_family_mode_ &&
-               !alive_log_files_.begin()->getting_flushed &&
               total_log_size_ > GetMaxTotalWalSize())) {
-    FlushColumnFamilies();
-  } else if (UNLIKELY(write_buffer_manager_->ShouldFlush())) {
+    MaybeFlushColumnFamilies();
+  }
+  if (UNLIKELY(write_buffer_manager_->ShouldFlush())) {
    // Before a new memtable is added in SwitchMemtable(),
    // write_buffer_manager_->ShouldFlush() will keep returning true. If another
    // thread is writing to another DB with the same write buffer, they may also
@@ -4767,6 +4811,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    // and protects against concurrent loggers and concurrent writes
    // into memtables
  }
+  log::Writer* cur_log_writer = logs_.back().writer;

  mutex_.Unlock();

@@ -4818,7 +4863,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
    PERF_TIMER_STOP(write_pre_and_post_process_time);

    if (write_options.disableWAL) {
-      has_unpersisted_data_ = true;
+      has_unpersisted_data_.store(true, std::memory_order_relaxed);
    }

    uint64_t log_size = 0;
@@ -4854,7 +4899,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
      WriteBatchInternal::SetSequence(merged_batch, current_sequence);

      Slice log_entry = WriteBatchInternal::Contents(merged_batch);
-      status = logs_.back().writer->AddRecord(log_entry);
+      status = cur_log_writer->AddRecord(log_entry);
      total_log_size_ += log_entry.size();
      alive_log_files_.back().AddSize(log_entry.size());
      log_empty_ = false;
@@ -5000,28 +5045,51 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  return status;
 }

-void DBImpl::FlushColumnFamilies() {
+void DBImpl::MaybeFlushColumnFamilies() {
  mutex_.AssertHeld();

-  WriteContext context;
-
  if (alive_log_files_.begin()->getting_flushed) {
    return;
  }

-  uint64_t flush_column_family_if_log_file = alive_log_files_.begin()->number;
-  alive_log_files_.begin()->getting_flushed = true;
+  auto oldest_alive_log = alive_log_files_.begin()->number;
+  auto oldest_log_with_uncommited_prep = FindMinLogContainingOutstandingPrep();
+
+  if (allow_2pc() &&
+      oldest_log_with_uncommited_prep > 0 &&
+      oldest_log_with_uncommited_prep <= oldest_alive_log) {
+    if (unable_to_flush_oldest_log_) {
+        // we already attempted to flush all column families dependent on
+        // the oldest alive log but the log still contained uncommited transactions.
+        // the oldest alive log STILL contains uncommited transaction so there
+        // is still nothing that we can do.
+        return;
+    } else {
+      Log(InfoLogLevel::WARN_LEVEL, immutable_db_options_.info_log,
+          "Unable to release oldest log due to uncommited transaction");
+      unable_to_flush_oldest_log_ = true;
+    }
+  } else {
+    // we only mark this log as getting flushed if we have successfully
+    // flushed all data in this log. If this log contains outstanding prepared
+    // transactions then we cannot flush this log until those transactions are commited.
+    unable_to_flush_oldest_log_ = false;
+    alive_log_files_.begin()->getting_flushed = true;
+  }
+
+  WriteContext context;
+
  Log(InfoLogLevel::INFO_LEVEL, immutable_db_options_.info_log,
      "Flushing all column families with data in WAL number %" PRIu64
      ". Total log size is %" PRIu64 " while max_total_wal_size is %" PRIu64,
-      flush_column_family_if_log_file, total_log_size_, GetMaxTotalWalSize());
+      oldest_alive_log, total_log_size_, GetMaxTotalWalSize());
  // no need to refcount because drop is happening in write thread, so can't
  // happen while we're in the write thread
  for (auto cfd : *versions_->GetColumnFamilySet()) {
    if (cfd->IsDropped()) {
      continue;
    }
-    if (cfd->GetLogNumber() <= flush_column_family_if_log_file) {
+    if (cfd->OldestLogToKeep() <= oldest_alive_log) {
      auto status = SwitchMemtable(cfd, &context);
      if (!status.ok()) {
        break;
@@ -5031,6 +5099,7 @@ void DBImpl::FlushColumnFamilies() {
    }
  }
  MaybeScheduleFlushOrCompaction();
+
 }

 uint64_t DBImpl::GetMaxTotalWalSize() const {
@@ -5426,22 +5495,6 @@ SuperVersion* DBImpl::GetAndRefSuperVersion(uint32_t column_family_id) {
  return GetAndRefSuperVersion(cfd);
 }

-// REQUIRED:  mutex is NOT held
-SuperVersion* DBImpl::GetAndRefSuperVersionUnlocked(uint32_t column_family_id) {
-  ColumnFamilyData* cfd;
-  {
-    InstrumentedMutexLock l(&mutex_);
-    auto column_family_set = versions_->GetColumnFamilySet();
-    cfd = column_family_set->GetColumnFamily(column_family_id);
-  }
-
-  if (!cfd) {
-    return nullptr;
-  }
-
-  return GetAndRefSuperVersion(cfd);
-}
-
 void DBImpl::ReturnAndCleanupSuperVersion(ColumnFamilyData* cfd,
                                          SuperVersion* sv) {
  bool unref_sv = !cfd->ReturnThreadLocalSuperVersion(sv);
@@ -5472,22 +5525,6 @@ void DBImpl::ReturnAndCleanupSuperVersion(uint32_t column_family_id,
  ReturnAndCleanupSuperVersion(cfd, sv);
 }

-// REQUIRED: Mutex should NOT be held.
-void DBImpl::ReturnAndCleanupSuperVersionUnlocked(uint32_t column_family_id,
-                                                  SuperVersion* sv) {
-  ColumnFamilyData* cfd;
-  {
-    InstrumentedMutexLock l(&mutex_);
-    auto column_family_set = versions_->GetColumnFamilySet();
-    cfd = column_family_set->GetColumnFamily(column_family_id);
-  }
-
-  // If SuperVersion is held, and we successfully fetched a cfd using
-  // GetAndRefSuperVersion(), it must still exist.
-  assert(cfd != nullptr);
-  ReturnAndCleanupSuperVersion(cfd, sv);
-}
-
 // REQUIRED: this function should only be called on the write thread or if the
 // mutex is held.
 ColumnFamilyHandle* DBImpl::GetColumnFamilyHandle(uint32_t column_family_id) {
@@ -5500,23 +5537,33 @@ ColumnFamilyHandle* DBImpl::GetColumnFamilyHandle(uint32_t column_family_id) {
  return cf_memtables->GetColumnFamilyHandle();
 }

-// REQUIRED: mutex is NOT held.
-ColumnFamilyHandle* DBImpl::GetColumnFamilyHandleUnlocked(
-    uint32_t column_family_id) {
-  ColumnFamilyMemTables* cf_memtables = column_family_memtables_.get();
+void DBImpl::GetApproximateMemTableStats(ColumnFamilyHandle* column_family,
+                                         const Range& range,
+                                         uint64_t* const count,
+                                         uint64_t* const size) {
+  ColumnFamilyHandleImpl* cfh =
+      reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  ColumnFamilyData* cfd = cfh->cfd();
+  SuperVersion* sv = GetAndRefSuperVersion(cfd);

-  InstrumentedMutexLock l(&mutex_);
+  // Convert user_key into a corresponding internal key.
+  InternalKey k1(range.start, kMaxSequenceNumber, kValueTypeForSeek);
+  InternalKey k2(range.limit, kMaxSequenceNumber, kValueTypeForSeek);
+  MemTable::MemTableStats memStats =
+      sv->mem->ApproximateStats(k1.Encode(), k2.Encode());
+  MemTable::MemTableStats immStats =
+      sv->imm->ApproximateStats(k1.Encode(), k2.Encode());
+  *count = memStats.count + immStats.count;
+  *size = memStats.size + immStats.size;

-  if (!cf_memtables->Seek(column_family_id)) {
-    return nullptr;
-  }
-
-  return cf_memtables->GetColumnFamilyHandle();
+  ReturnAndCleanupSuperVersion(cfd, sv);
 }

 void DBImpl::GetApproximateSizes(ColumnFamilyHandle* column_family,
                                 const Range* range, int n, uint64_t* sizes,
-                                 bool include_memtable) {
+                                 uint8_t include_flags) {
+  assert(include_flags & DB::SizeApproximationFlags::INCLUDE_FILES ||
+         include_flags & DB::SizeApproximationFlags::INCLUDE_MEMTABLES);
  Version* v;
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();
@@ -5527,10 +5574,13 @@ void DBImpl::GetApproximateSizes(ColumnFamilyHandle* column_family,
    // Convert user_key into a corresponding internal key.
    InternalKey k1(range[i].start, kMaxSequenceNumber, kValueTypeForSeek);
    InternalKey k2(range[i].limit, kMaxSequenceNumber, kValueTypeForSeek);
-    sizes[i] = versions_->ApproximateSize(v, k1.Encode(), k2.Encode());
-    if (include_memtable) {
-      sizes[i] += sv->mem->ApproximateSize(k1.Encode(), k2.Encode());
-      sizes[i] += sv->imm->ApproximateSize(k1.Encode(), k2.Encode());
+    sizes[i] = 0;
+    if (include_flags & DB::SizeApproximationFlags::INCLUDE_FILES) {
+      sizes[i] += versions_->ApproximateSize(v, k1.Encode(), k2.Encode());
+    }
+    if (include_flags & DB::SizeApproximationFlags::INCLUDE_MEMTABLES) {
+      sizes[i] += sv->mem->ApproximateStats(k1.Encode(), k2.Encode()).size;
+      sizes[i] += sv->imm->ApproximateStats(k1.Encode(), k2.Encode()).size;
    }
  }

@@ -6050,6 +6100,7 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
  }
  impl->mutex_.Unlock();

+#ifndef ROCKSDB_LITE
  auto sfm = static_cast<SstFileManagerImpl*>(
      impl->immutable_db_options_.sst_file_manager.get());
  if (s.ok() && sfm) {
@@ -6068,6 +6119,7 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
      }
    }
  }
+#endif  // !ROCKSDB_LITE

  if (s.ok()) {
    Log(InfoLogLevel::INFO_LEVEL, impl->immutable_db_options_.info_log,
@@ -6298,7 +6350,7 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
 #endif  // !ROCKSDB_LITE
 }

-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS

 void DBImpl::NewThreadStatusCfInfo(
    ColumnFamilyData* cfd) const {
@@ -6469,10 +6521,15 @@ Status DBImpl::IngestExternalFile(
                                            immutable_db_options_, env_options_,
                                            &snapshots_, ingestion_options);

-  // Make sure that bg cleanup wont delete the files that we are ingesting
  std::list<uint64_t>::iterator pending_output_elem;
  {
    InstrumentedMutexLock l(&mutex_);
+    if (!bg_error_.ok()) {
+      // Don't ingest files when there is a bg_error
+      return bg_error_;
+    }
+
+    // Make sure that bg cleanup wont delete the files that we are ingesting
    pending_output_elem = CaptureCurrentFileNumberInPendingOutputs();
  }

@@ -6493,13 +6550,22 @@ Status DBImpl::IngestExternalFile(

    num_running_ingest_file_++;

+    // We cannot ingest a file into a dropped CF
+    if (cfd->IsDropped()) {
+      status = Status::InvalidArgument(
+          "Cannot ingest an external file into a dropped CF");
+    }
+
    // Figure out if we need to flush the memtable first
-    bool need_flush = false;
-    status = ingestion_job.NeedsFlush(&need_flush);
-    if (status.ok() && need_flush) {
-      mutex_.Unlock();
-      status = FlushMemTable(cfd, FlushOptions(), true /* writes_stopped */);
-      mutex_.Lock();
+    if (status.ok()) {
+      bool need_flush = false;
+      status = ingestion_job.NeedsFlush(&need_flush);
+
+      if (status.ok() && need_flush) {
+        mutex_.Unlock();
+        status = FlushMemTable(cfd, FlushOptions(), true /* writes_stopped */);
+        mutex_.Lock();
+      }
    }

    // Run the ingestion job
@@ -6541,9 +6607,35 @@ Status DBImpl::IngestExternalFile(
  // Cleanup
  ingestion_job.Cleanup(status);

+  if (status.ok()) {
+    NotifyOnExternalFileIngested(cfd, ingestion_job);
+  }
+
  return status;
 }

+void DBImpl::NotifyOnExternalFileIngested(
+    ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job) {
+#ifndef ROCKSDB_LITE
+  if (immutable_db_options_.listeners.empty()) {
+    return;
+  }
+
+  for (const IngestedFileInfo& f : ingestion_job.files_to_ingest()) {
+    ExternalFileIngestionInfo info;
+    info.cf_name = cfd->GetName();
+    info.external_file_path = f.external_file_path;
+    info.internal_file_path = f.internal_file_path;
+    info.global_seqno = f.assigned_seqno;
+    info.table_properties = f.table_properties;
+    for (auto listener : immutable_db_options_.listeners) {
+      listener->OnExternalFileIngested(this, info);
+    }
+  }
+
+#endif
+}
+
 void DBImpl::WaitForIngestFile() {
  mutex_.AssertHeld();
  while (num_running_ingest_file_ > 0) {
@@ -23,6 +23,7 @@
 #include "db/column_family.h"
 #include "db/compaction_job.h"
 #include "db/dbformat.h"
+#include "db/external_sst_file_ingestion_job.h"
 #include "db/flush_job.h"
 #include "db/flush_scheduler.h"
 #include "db/internal_stats.h"
@@ -137,7 +138,13 @@ class DBImpl : public DB {
  using DB::GetApproximateSizes;
  virtual void GetApproximateSizes(ColumnFamilyHandle* column_family,
                                   const Range* range, int n, uint64_t* sizes,
-                                   bool include_memtable = false) override;
+                                   uint8_t include_flags
+                                   = INCLUDE_FILES) override;
+  using DB::GetApproximateMemTableStats;
+  virtual void GetApproximateMemTableStats(ColumnFamilyHandle* column_family,
+                                           const Range& range,
+                                           uint64_t* const count,
+                                           uint64_t* const size) override;
  using DB::CompactRange;
  virtual Status CompactRange(const CompactRangeOptions& options,
                              ColumnFamilyHandle* column_family,
@@ -307,6 +314,16 @@ class DBImpl : public DB {
                           ColumnFamilyHandle* column_family = nullptr,
                           bool disallow_trivial_move = false);

+  void TEST_MaybeFlushColumnFamilies();
+
+  bool TEST_UnableToFlushOldestLog() {
+    return unable_to_flush_oldest_log_;
+  }
+
+  bool TEST_IsLogGettingFlushed() {
+    return alive_log_files_.begin()->getting_flushed;
+  }
+
  // Force current memtable contents to be flushed.
  Status TEST_FlushMemTable(bool wait = true,
                            ColumnFamilyHandle* cfh = nullptr);
@@ -379,6 +396,8 @@ class DBImpl : public DB {
  // schedule a purge
  void ScheduleBgLogWriterClose(JobContext* job_context);

+  uint64_t MinLogNumberToKeep();
+
  // Returns the list of live files in 'live' and the list
  // of all files in the filesystem in 'candidate_files'.
  // If force == false and the last call was less than
@@ -417,9 +436,6 @@ class DBImpl : public DB {
  // mutex is held.
  SuperVersion* GetAndRefSuperVersion(uint32_t column_family_id);

-  // Same as above, should called without mutex held and not on write thread.
-  SuperVersion* GetAndRefSuperVersionUnlocked(uint32_t column_family_id);
-
  // Un-reference the super version and return it to thread local cache if
  // needed. If it is the last reference of the super version. Clean it up
  // after un-referencing it.
@@ -430,18 +446,11 @@ class DBImpl : public DB {
  // REQUIRED: this function should only be called on the write thread.
  void ReturnAndCleanupSuperVersion(uint32_t colun_family_id, SuperVersion* sv);

-  // Same as above, should called without mutex held and not on write thread.
-  void ReturnAndCleanupSuperVersionUnlocked(uint32_t colun_family_id,
-                                            SuperVersion* sv);
-
  // REQUIRED: this function should only be called on the write thread or if the
  // mutex is held.  Return value only valid until next call to this function or
  // mutex is released.
  ColumnFamilyHandle* GetColumnFamilyHandle(uint32_t column_family_id);

-  // Same as above, should called without mutex held and not on write thread.
-  ColumnFamilyHandle* GetColumnFamilyHandleUnlocked(uint32_t column_family_id);
-
  // Returns the number of currently running flushes.
  // REQUIREMENT: mutex_ must be held when calling this function.
  int num_running_flushes() {
@@ -456,6 +465,8 @@ class DBImpl : public DB {
    return num_running_compactions_;
  }

+  const WriteController& write_controller() { return write_controller_; }
+
  // hollow transactions shell used for recovery.
  // these will then be passed to TransactionDB so that
  // locks can be reacquired before writing can resume.
@@ -556,6 +567,11 @@ class DBImpl : public DB {
  void NotifyOnMemTableSealed(ColumnFamilyData* cfd,
                              const MemTableInfo& mem_table_info);

+#ifndef ROCKSDB_LITE
+  void NotifyOnExternalFileIngested(
+      ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job);
+#endif  // !ROCKSDB_LITE
+
  void NewThreadStatusCfInfo(ColumnFamilyData* cfd) const;

  void EraseThreadStatusCfInfo(ColumnFamilyData* cfd) const;
@@ -728,7 +744,7 @@ class DBImpl : public DB {
  // REQUIRES: mutex locked
  Status PersistOptions();

-  void FlushColumnFamilies();
+  void MaybeFlushColumnFamilies();

  uint64_t GetMaxTotalWalSize() const;

@@ -972,8 +988,9 @@ class DBImpl : public DB {
  // without any synchronization
  int disable_delete_obsolete_files_;

-  // next time when we should run DeleteObsoleteFiles with full scan
-  uint64_t delete_obsolete_files_next_run_;
+  // last time when DeleteObsoleteFiles with full scan was executed. Originaly
+  // initialized with startup time.
+  uint64_t delete_obsolete_files_last_run_;

  // last time stats were dumped to LOG
  std::atomic<uint64_t> last_stats_dump_time_microsec_;
@@ -985,7 +1002,15 @@ class DBImpl : public DB {
  // A flag indicating whether the current rocksdb database has any
  // data that is not yet persisted into either WAL or SST file.
  // Used when disableWAL is true.
-  bool has_unpersisted_data_;
+  std::atomic<bool> has_unpersisted_data_;
+
+  // if an attempt was made to flush all column families that
+  // the oldest log depends on but uncommited data in the oldest
+  // log prevents the log from being released.
+  // We must attempt to free the dependent memtables again
+  // at a later time after the transaction in the oldest
+  // log is fully commited.
+  bool unable_to_flush_oldest_log_;

  static const int KEEP_LOG_FILE_NUM = 1000;
  // MSVC version 1800 still does not have constexpr for ::max()
@@ -19,6 +19,11 @@ uint64_t DBImpl::TEST_GetLevel0TotalSize() {
  return default_cf_handle_->cfd()->current()->storage_info()->NumLevelBytes(0);
 }

+void DBImpl::TEST_MaybeFlushColumnFamilies() {
+  InstrumentedMutexLock l(&mutex_);
+  MaybeFlushColumnFamilies();
+}
+
 int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
    ColumnFamilyHandle* column_family) {
  ColumnFamilyData* cfd;
@@ -79,7 +79,6 @@ class DBImplReadOnly : public DBImpl {
    return Status::NotSupported("Not supported operation in read only mode.");
  }

-#ifndef ROCKSDB_LITE
  virtual Status DisableFileDeletions() override {
    return Status::NotSupported("Not supported operation in read only mode.");
  }
@@ -92,7 +91,6 @@ class DBImplReadOnly : public DBImpl {
                              bool flush_memtable = true) override {
    return Status::NotSupported("Not supported operation in read only mode.");
  }
-#endif  // ROCKSDB_LITE

  using DBImpl::Flush;
  virtual Status Flush(const FlushOptions& options,
@@ -105,6 +103,14 @@ class DBImplReadOnly : public DBImpl {
    return Status::NotSupported("Not supported operation in read only mode.");
  }

+  using DB::IngestExternalFile;
+  virtual Status IngestExternalFile(
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& external_files,
+      const IngestExternalFileOptions& ingestion_options) override {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+
 private:
  friend class DB;

@@ -123,7 +123,8 @@ class DBIter: public Iterator {
        prefix_same_as_start_(prefix_same_as_start),
        pin_thru_lifetime_(pin_data),
        total_order_seek_(total_order_seek),
-        range_del_agg_(ioptions.internal_comparator, s) {
+        range_del_agg_(ioptions.internal_comparator, s,
+                       true /* collapse_deletions */) {
    RecordTick(statistics_, NO_ITERATORS);
    prefix_extractor_ = ioptions.prefix_extractor;
    max_skip_ = max_sequential_skip_in_iterations;
@@ -340,6 +341,7 @@ void DBIter::Next() {
 //
 // NOTE: In between, saved_key_ can point to a user key that has
 //       a delete marker or a sequence number higher than sequence_
+//       saved_key_ MUST have a proper user_key before calling this function
 //
 // The prefix_check parameter controls whether we check the iterated
 // keys against the prefix of the seeked key. Set to false when
@@ -411,7 +413,9 @@ void DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
            saved_key_.SetKey(
                ikey.user_key,
                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
-            if (range_del_agg_.ShouldDelete(ikey)) {
+            if (range_del_agg_.ShouldDelete(
+                    ikey, RangeDelAggregator::RangePositioningMode::
+                              kForwardTraversal)) {
              // Arrange to skip all upcoming entries for this key since
              // they are hidden by this deletion.
              skipping = true;
@@ -426,7 +430,9 @@ void DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
            saved_key_.SetKey(
                ikey.user_key,
                !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
-            if (range_del_agg_.ShouldDelete(ikey)) {
+            if (range_del_agg_.ShouldDelete(
+                    ikey, RangeDelAggregator::RangePositioningMode::
+                              kForwardTraversal)) {
              // Arrange to skip all upcoming entries for this key since
              // they are hidden by this deletion.
              skipping = true;
@@ -518,6 +524,7 @@ void DBIter::MergeValuesNewToOld() {
                             iter_->IsValuePinned() /* operand_pinned */);

  ParsedInternalKey ikey;
+  Status s;
  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
    if (!ParseKey(&ikey)) {
      // skip corrupted key
@@ -528,7 +535,9 @@ void DBIter::MergeValuesNewToOld() {
      // hit the next user key, stop right here
      break;
    } else if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type ||
-               range_del_agg_.ShouldDelete(ikey)) {
+               range_del_agg_.ShouldDelete(
+                   ikey, RangeDelAggregator::RangePositioningMode::
+                             kForwardTraversal)) {
      // hit a delete with the same user key, stop right here
      // iter_ is positioned after delete
      iter_->Next();
@@ -538,9 +547,12 @@ void DBIter::MergeValuesNewToOld() {
      // final result in saved_value_. We are done!
      // ignore corruption if there is any.
      const Slice val = iter_->value();
-      MergeHelper::TimedFullMerge(merge_operator_, ikey.user_key, &val,
-                                  merge_context_.GetOperands(), &saved_value_,
-                                  logger_, statistics_, env_, &pinned_value_);
+      s = MergeHelper::TimedFullMerge(
+          merge_operator_, ikey.user_key, &val, merge_context_.GetOperands(),
+          &saved_value_, logger_, statistics_, env_, &pinned_value_);
+      if (!s.ok()) {
+        status_ = s;
+      }
      // iter_ is positioned after put
      iter_->Next();
      return;
@@ -559,9 +571,12 @@ void DBIter::MergeValuesNewToOld() {
  // a deletion marker.
  // feed null as the existing value to the merge operator, such that
  // client can differentiate this scenario and do things accordingly.
-  MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
-                              merge_context_.GetOperands(), &saved_value_,
-                              logger_, statistics_, env_, &pinned_value_);
+  s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
+                                  merge_context_.GetOperands(), &saved_value_,
+                                  logger_, statistics_, env_, &pinned_value_);
+  if (!s.ok()) {
+    status_ = s;
+  }
 }

 void DBIter::Prev() {
@@ -591,6 +606,7 @@ void DBIter::ReverseToForward() {
  direction_ = kForward;
  if (!iter_->Valid()) {
    iter_->SeekToFirst();
+    range_del_agg_.InvalidateTombstoneMapPositions();
  }
 }

@@ -606,6 +622,7 @@ void DBIter::ReverseToBackward() {
    // previous key.
    if (!iter_->Valid()) {
      iter_->SeekToLast();
+      range_del_agg_.InvalidateTombstoneMapPositions();
    }
    ParsedInternalKey ikey;
    FindParseableKey(&ikey, kReverse);
@@ -702,7 +719,9 @@ bool DBIter::FindValueForCurrentKey() {
    last_key_entry_type = ikey.type;
    switch (last_key_entry_type) {
      case kTypeValue:
-        if (range_del_agg_.ShouldDelete(ikey)) {
+        if (range_del_agg_.ShouldDelete(
+                ikey,
+                RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
          last_key_entry_type = kTypeRangeDeletion;
          PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
        } else {
@@ -719,7 +738,9 @@ bool DBIter::FindValueForCurrentKey() {
        PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
        break;
      case kTypeMerge:
-        if (range_del_agg_.ShouldDelete(ikey)) {
+        if (range_del_agg_.ShouldDelete(
+                ikey,
+                RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
          merge_context_.Clear();
          last_key_entry_type = kTypeRangeDeletion;
          last_not_merge_type = last_key_entry_type;
@@ -742,6 +763,7 @@ bool DBIter::FindValueForCurrentKey() {
    FindParseableKey(&ikey, kReverse);
  }

+  Status s;
  switch (last_key_entry_type) {
    case kTypeDeletion:
    case kTypeSingleDeletion:
@@ -753,16 +775,16 @@ bool DBIter::FindValueForCurrentKey() {
      if (last_not_merge_type == kTypeDeletion ||
          last_not_merge_type == kTypeSingleDeletion ||
          last_not_merge_type == kTypeRangeDeletion) {
-        MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
-                                    nullptr, merge_context_.GetOperands(),
-                                    &saved_value_, logger_, statistics_, env_,
-                                    &pinned_value_);
+        s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
+                                        nullptr, merge_context_.GetOperands(),
+                                        &saved_value_, logger_, statistics_,
+                                        env_, &pinned_value_);
      } else {
        assert(last_not_merge_type == kTypeValue);
-        MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
-                                    &pinned_value_,
-                                    merge_context_.GetOperands(), &saved_value_,
-                                    logger_, statistics_, env_, &pinned_value_);
+        s = MergeHelper::TimedFullMerge(
+            merge_operator_, saved_key_.GetKey(), &pinned_value_,
+            merge_context_.GetOperands(), &saved_value_, logger_, statistics_,
+            env_, &pinned_value_);
      }
      break;
    case kTypeValue:
@@ -773,6 +795,9 @@ bool DBIter::FindValueForCurrentKey() {
      break;
  }
  valid_ = true;
+  if (!s.ok()) {
+    status_ = s;
+  }
  return true;
 }

@@ -793,7 +818,8 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  FindParseableKey(&ikey, kForward);

  if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
-      range_del_agg_.ShouldDelete(ikey)) {
+      range_del_agg_.ShouldDelete(
+          ikey, RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
    valid_ = false;
    return false;
  }
@@ -808,9 +834,12 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
  // in operands
  current_entry_is_merged_ = true;
  merge_context_.Clear();
-  while (iter_->Valid() &&
-         user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) &&
-         ikey.type == kTypeMerge && !range_del_agg_.ShouldDelete(ikey)) {
+  while (
+      iter_->Valid() &&
+      user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) &&
+      ikey.type == kTypeMerge &&
+      !range_del_agg_.ShouldDelete(
+          ikey, RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
    merge_context_.PushOperand(iter_->value(),
                               iter_->IsValuePinned() /* operand_pinned */);
    PERF_COUNTER_ADD(internal_merge_count, 1);
@@ -818,13 +847,16 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
    FindParseableKey(&ikey, kForward);
  }

+  Status s;
  if (!iter_->Valid() ||
      !user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) ||
      ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
-      range_del_agg_.ShouldDelete(ikey)) {
-    MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(), nullptr,
-                                merge_context_.GetOperands(), &saved_value_,
-                                logger_, statistics_, env_, &pinned_value_);
+      range_del_agg_.ShouldDelete(
+          ikey, RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+    s = MergeHelper::TimedFullMerge(merge_operator_, saved_key_.GetKey(),
+                                    nullptr, merge_context_.GetOperands(),
+                                    &saved_value_, logger_, statistics_, env_,
+                                    &pinned_value_);
    // Make iter_ valid and point to saved_key_
    if (!iter_->Valid() ||
        !user_comparator_->Equal(ikey.user_key, saved_key_.GetKey())) {
@@ -832,14 +864,20 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
    }
    valid_ = true;
+    if (!s.ok()) {
+      status_ = s;
+    }
    return true;
  }

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

@@ -909,12 +947,12 @@ void DBIter::Seek(const Slice& target) {
  StopWatch sw(env_, statistics_, DB_SEEK);
  ReleaseTempPinnedData();
  saved_key_.Clear();
-  // now savved_key is used to store internal key.
  saved_key_.SetInternalKey(target, sequence_);

  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->Seek(saved_key_.GetKey());
+    range_del_agg_.InvalidateTombstoneMapPositions();
  }
  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
@@ -923,6 +961,9 @@ void DBIter::Seek(const Slice& target) {
    }
    direction_ = kForward;
    ClearSavedValue();
+    // convert the InternalKey to UserKey in saved_key_ before
+    // passed to FindNextUserEntry
+    saved_key_.Reserve(saved_key_.Size() - 8);
    FindNextUserEntry(false /* not skipping */, prefix_same_as_start_);
    if (!valid_) {
      prefix_start_key_.clear();
@@ -954,6 +995,7 @@ void DBIter::SeekForPrev(const Slice& target) {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekForPrev(saved_key_.GetKey());
+    range_del_agg_.InvalidateTombstoneMapPositions();
  }

  RecordTick(statistics_, NUMBER_DB_SEEK);
@@ -995,10 +1037,13 @@ void DBIter::SeekToFirst() {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekToFirst();
+    range_del_agg_.InvalidateTombstoneMapPositions();
  }

  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
+    saved_key_.SetKey(ExtractUserKey(iter_->key()),
+                      !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
    FindNextUserEntry(false /* not skipping */, false /* no prefix check */);
    if (statistics_ != nullptr) {
      if (valid_) {
@@ -1028,12 +1073,14 @@ void DBIter::SeekToLast() {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekToLast();
+    range_del_agg_.InvalidateTombstoneMapPositions();
  }
  // When the iterate_upper_bound is set to a value,
  // it will seek to the last key before the
  // ReadOptions.iterate_upper_bound
  if (iter_->Valid() && iterate_upper_bound_ != nullptr) {
    SeekForPrev(*iterate_upper_bound_);
+    range_del_agg_.InvalidateTombstoneMapPositions();
    if (!Valid()) {
      return;
    } else if (user_comparator_->Equal(*iterate_upper_bound_, key())) {
@@ -16,7 +16,7 @@
 #include "rocksdb/slice.h"
 #include "rocksdb/statistics.h"
 #include "table/iterator_wrapper.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "util/string_util.h"
 #include "util/sync_point.h"
 #include "util/testharness.h"
@@ -1897,6 +1897,66 @@ TEST_F(DBIteratorTest, DBIterator12) {
  ASSERT_FALSE(db_iter->Valid());
 }

+TEST_F(DBIteratorTest, DBIterator13) {
+  Options options;
+  options.merge_operator = nullptr;
+
+  std::string key;
+  key.resize(9);
+  key.assign(9, static_cast<char>(0));
+  key[0] = 'b';
+
+  TestIterator* internal_iter = new TestIterator(BytewiseComparator());
+  internal_iter->AddPut(key, "0");
+  internal_iter->AddPut(key, "1");
+  internal_iter->AddPut(key, "2");
+  internal_iter->AddPut(key, "3");
+  internal_iter->AddPut(key, "4");
+  internal_iter->AddPut(key, "5");
+  internal_iter->AddPut(key, "6");
+  internal_iter->AddPut(key, "7");
+  internal_iter->AddPut(key, "8");
+  internal_iter->Finish();
+
+  std::unique_ptr<Iterator> db_iter(
+      NewDBIterator(env_, ImmutableCFOptions(options), BytewiseComparator(),
+                    internal_iter, 2, 3, 0));
+  db_iter->Seek("b");
+  ASSERT_TRUE(db_iter->Valid());
+  ASSERT_EQ(db_iter->key().ToString(), key);
+  ASSERT_EQ(db_iter->value().ToString(), "2");
+}
+
+TEST_F(DBIteratorTest, DBIterator14) {
+  Options options;
+  options.merge_operator = nullptr;
+
+  std::string key("b");
+  TestIterator* internal_iter = new TestIterator(BytewiseComparator());
+  internal_iter->AddPut("b", "0");
+  internal_iter->AddPut("b", "1");
+  internal_iter->AddPut("b", "2");
+  internal_iter->AddPut("b", "3");
+  internal_iter->AddPut("a", "4");
+  internal_iter->AddPut("a", "5");
+  internal_iter->AddPut("a", "6");
+  internal_iter->AddPut("c", "7");
+  internal_iter->AddPut("c", "8");
+  internal_iter->AddPut("c", "9");
+  internal_iter->Finish();
+
+  std::unique_ptr<Iterator> db_iter(
+      NewDBIterator(env_, ImmutableCFOptions(options), BytewiseComparator(),
+                    internal_iter, 4, 1, 0));
+  db_iter->Seek("b");
+  ASSERT_TRUE(db_iter->Valid());
+  ASSERT_EQ(db_iter->key().ToString(), "b");
+  ASSERT_EQ(db_iter->value().ToString(), "3");
+  db_iter->SeekToFirst();
+  ASSERT_EQ(db_iter->key().ToString(), "a");
+  ASSERT_EQ(db_iter->value().ToString(), "4");
+}
+
 class DBIterWithMergeIterTest : public testing::Test {
 public:
  DBIterWithMergeIterTest()
@@ -7,6 +7,8 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

+#include <functional>
+
 #include "db/db_test_util.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
@@ -1642,7 +1644,7 @@ TEST_F(DBIteratorTest, IteratorWithLocalStatistics) {
  std::atomic<uint64_t> total_prev_found(0);
  std::atomic<uint64_t> total_bytes(0);

-  std::vector<std::thread> threads;
+  std::vector<port::Thread> threads;
  std::function<void()> reader_func_next = [&]() {
    Iterator* iter = db_->NewIterator(ReadOptions());

@@ -0,0 +1,98 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+#include <string>
+#include <vector>
+
+#include "db/db_test_util.h"
+#include "db/forward_iterator.h"
+#include "port/stack_trace.h"
+#include "utilities/merge_operators.h"
+
+namespace rocksdb {
+
+// Test merge operator functionality.
+class DBMergeOperatorTest : public DBTestBase {
+ public:
+  DBMergeOperatorTest() : DBTestBase("/db_merge_operator_test") {}
+};
+
+// A test merge operator mimics put but also fails if one of merge operands is
+// "corrupted".
+class TestPutOperator : public MergeOperator {
+ public:
+  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
+                           MergeOperationOutput* merge_out) const override {
+    if (merge_in.existing_value != nullptr &&
+        *(merge_in.existing_value) == "corrupted") {
+      return false;
+    }
+    for (auto value : merge_in.operand_list) {
+      if (value == "corrupted") {
+        return false;
+      }
+    }
+    merge_out->existing_operand = merge_in.operand_list.back();
+    return true;
+  }
+
+  virtual const char* Name() const override { return "TestPutOperator"; }
+};
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnRead) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+  Reopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "corrupted"));
+  std::string value;
+  ASSERT_TRUE(db_->Get(ReadOptions(), "k1", &value).IsCorruption());
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v1"}});
+}
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnWrite) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+  options.max_successive_merges = 3;
+  Reopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "v2"));
+  // Will trigger a merge when hitting max_successive_merges and the merge
+  // will fail. The delta will be inserted nevertheless.
+  ASSERT_OK(Merge("k1", "corrupted"));
+  // Data should stay unmerged after the error.
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v2"}, {"k1", "v1"}});
+}
+
+TEST_F(DBMergeOperatorTest, MergeErrorOnIteration) {
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator.reset(new TestPutOperator());
+
+  DestroyAndReopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Merge("k1", "corrupted"));
+  ASSERT_OK(Put("k2", "v2"));
+  VerifyDBFromMap({{"k1", ""}, {"k2", "v2"}}, nullptr, false,
+                  {{"k1", Status::Corruption()}});
+  VerifyDBInternal({{"k1", "corrupted"}, {"k1", "v1"}, {"k2", "v2"}});
+
+  DestroyAndReopen(options);
+  ASSERT_OK(Merge("k1", "v1"));
+  ASSERT_OK(Put("k2", "v2"));
+  ASSERT_OK(Merge("k2", "corrupted"));
+  VerifyDBFromMap({{"k1", "v1"}, {"k2", ""}}, nullptr, false,
+                  {{"k2", Status::Corruption()}});
+  VerifyDBInternal({{"k1", "v1"}, {"k2", "corrupted"}, {"k2", "v2"}});
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -316,6 +316,47 @@ TEST_F(DBOptionsTest, MaxTotalWalSizeChange) {
  }
 }

+static void assert_candidate_files_empty(DBImpl* dbfull, const bool empty) {
+  dbfull->TEST_LockMutex();
+  JobContext job_context(0);
+  dbfull->FindObsoleteFiles(&job_context, false);
+  ASSERT_EQ(empty, job_context.full_scan_candidate_files.empty());
+  job_context.Clean();
+  dbfull->TEST_UnlockMutex();
+}
+
+TEST_F(DBOptionsTest, DeleteObsoleteFilesPeriodChange) {
+  SpecialEnv env(Env::Default());
+  env.time_elapse_only_sleep_ = true;
+  Options options;
+  options.env = &env;
+  options.create_if_missing = true;
+  ASSERT_OK(TryReopen(options));
+
+  // Verify that candidate files set is empty when no full scan requested.
+  assert_candidate_files_empty(dbfull(), true);
+
+  ASSERT_OK(
+      dbfull()->SetDBOptions({{"delete_obsolete_files_period_micros", "0"}}));
+
+  // After delete_obsolete_files_period_micros updated to 0, the next call
+  // to FindObsoleteFiles should make a full scan
+  assert_candidate_files_empty(dbfull(), false);
+
+  ASSERT_OK(
+      dbfull()->SetDBOptions({{"delete_obsolete_files_period_micros", "20"}}));
+
+  assert_candidate_files_empty(dbfull(), true);
+
+  env.addon_time_.store(20);
+  assert_candidate_files_empty(dbfull(), true);
+
+  env.addon_time_.store(21);
+  assert_candidate_files_empty(dbfull(), false);
+
+  Close();
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -41,7 +41,7 @@ TEST_F(DBRangeDelTest, NonBlockBasedTableNotSupported) {

 TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
-                             "dr1"));
+                             "dr2"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
 }
@@ -112,6 +112,81 @@ TEST_F(DBRangeDelTest, CompactionOutputFilesExactlyFilled) {
  db_->ReleaseSnapshot(snapshot);
 }

+TEST_F(DBRangeDelTest, MaxCompactionBytesCutsOutputFiles) {
+  // Ensures range deletion spanning multiple compaction output files that are
+  // cut by max_compaction_bytes will have non-overlapping key-ranges.
+  // https://github.com/facebook/rocksdb/issues/1778
+  const int kNumFiles = 2, kNumPerFile = 1 << 8, kBytesPerVal = 1 << 12;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.disable_auto_compactions = true;
+  opts.level0_file_num_compaction_trigger = kNumFiles;
+  opts.max_compaction_bytes = kNumPerFile * kBytesPerVal;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  // Want max_compaction_bytes to trigger the end of compaction output file, not
+  // target_file_size_base, so make the latter much bigger
+  opts.target_file_size_base = 100 * opts.max_compaction_bytes;
+  Reopen(opts);
+
+  // snapshot protects range tombstone from dropping due to becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+
+  // It spans the whole key-range, thus will be included in all output files
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                             GetNumericStr(0),
+                             GetNumericStr(kNumFiles * kNumPerFile - 1)));
+  Random rnd(301);
+  for (int i = 0; i < kNumFiles; ++i) {
+    std::vector<std::string> values;
+    // Write 1MB (256 values, each 4K)
+    for (int j = 0; j < kNumPerFile; j++) {
+      values.push_back(RandomString(&rnd, kBytesPerVal));
+      ASSERT_OK(Put(GetNumericStr(kNumPerFile * i + j), values[j]));
+    }
+    // extra entry to trigger SpecialSkipListFactory's flush
+    ASSERT_OK(Put(GetNumericStr(kNumPerFile), ""));
+    dbfull()->TEST_WaitForFlushMemTable();
+    ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
+  }
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                              true /* disallow_trivial_move */);
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GE(NumTableFilesAtLevel(1), 2);
+
+  std::vector<std::vector<FileMetaData>> files;
+  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
+
+  for (size_t i = 0; i < files[1].size() - 1; ++i) {
+    ASSERT_TRUE(InternalKeyComparator(opts.comparator)
+                    .Compare(files[1][i].largest, files[1][i + 1].smallest) <
+                0);
+  }
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, SentinelsOmittedFromOutputFile) {
+  // Regression test for bug where sentinel range deletions (i.e., ones with
+  // sequence number of zero) were included in output files.
+  // snapshot protects range tombstone from dropping due to becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+
+  // gaps between ranges creates sentinels in our internal representation
+  std::vector<std::pair<std::string, std::string>> range_dels = {{"a", "b"}, {"c", "d"}, {"e", "f"}};
+  for (const auto& range_del : range_dels) {
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                               range_del.first, range_del.second));
+  }
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+
+  std::vector<std::vector<FileMetaData>> files;
+  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
+  ASSERT_GT(files[0][0].smallest_seqno, 0);
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_F(DBRangeDelTest, FlushRangeDelsSameStartKey) {
  db_->Put(WriteOptions(), "b1", "val");
  ASSERT_OK(
@@ -428,6 +503,310 @@ TEST_F(DBRangeDelTest, ObsoleteTombstoneCleanup) {

  db_->ReleaseSnapshot(snapshot);
 }
+
+TEST_F(DBRangeDelTest, TableEvictedDuringScan) {
+  // The RangeDelAggregator holds pointers into range deletion blocks created by
+  // table readers. This test ensures the aggregator can still access those
+  // blocks even if it outlives the table readers that created them.
+  //
+  // DBIter always keeps readers open for L0 files. So, in order to test
+  // aggregator outliving reader, we need to have deletions in L1 files, which
+  // are opened/closed on-demand during the scan. This is accomplished by
+  // setting kNumRanges > level0_stop_writes_trigger, which prevents deletions
+  // from all lingering in L0 (there is at most one range deletion per L0 file).
+  //
+  // The first L1 file will contain a range deletion since its begin key is 0.
+  // SeekToFirst() references that table's reader and adds its range tombstone
+  // to the aggregator. Upon advancing beyond that table's key-range via Next(),
+  // the table reader will be unreferenced by the iterator. Since we manually
+  // call Evict() on all readers before the full scan, this unreference causes
+  // the reader's refcount to drop to zero and thus be destroyed.
+  //
+  // When it is destroyed, we do not remove its range deletions from the
+  // aggregator. So, subsequent calls to Next() must be able to use these
+  // deletions to decide whether a key is covered. This will work as long as
+  // the aggregator properly references the range deletion block.
+  const int kNum = 25, kRangeBegin = 0, kRangeEnd = 7, kNumRanges = 5;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.level0_file_num_compaction_trigger = 4;
+  opts.level0_stop_writes_trigger = 4;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+  opts.num_levels = 2;
+  BlockBasedTableOptions bbto;
+  bbto.cache_index_and_filter_blocks = true;
+  bbto.block_cache = NewLRUCache(8 << 20);
+  opts.table_factory.reset(NewBlockBasedTableFactory(bbto));
+  Reopen(opts);
+
+  // Hold a snapshot so range deletions can't become obsolete during compaction
+  // to bottommost level (i.e., L1).
+  const Snapshot* snapshot = db_->GetSnapshot();
+  for (int i = 0; i < kNum; ++i) {
+    db_->Put(WriteOptions(), GetNumericStr(i), "val");
+    if (i > 0) {
+      dbfull()->TEST_WaitForFlushMemTable();
+    }
+    if (i >= kNum / 2 && i < kNum / 2 + kNumRanges) {
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
+    }
+  }
+  // Must be > 1 so the first L1 file can be closed before scan finishes
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_GT(NumTableFilesAtLevel(1), 1);
+  std::vector<uint64_t> file_numbers = ListTableFiles(env_, dbname_);
+
+  ReadOptions read_opts;
+  auto* iter = db_->NewIterator(read_opts);
+  int expected = kRangeEnd;
+  iter->SeekToFirst();
+  for (auto file_number : file_numbers) {
+    // This puts table caches in the state of being externally referenced only
+    // so they are destroyed immediately upon iterator unreferencing.
+    TableCache::Evict(dbfull()->TEST_table_cache(), file_number);
+  }
+  for (; iter->Valid(); iter->Next()) {
+    ASSERT_EQ(GetNumericStr(expected), iter->key());
+    ++expected;
+    // Keep clearing block cache's LRU so range deletion block can be freed as
+    // soon as its refcount drops to zero.
+    bbto.block_cache->EraseUnRefEntries();
+  }
+  ASSERT_EQ(kNum, expected);
+  delete iter;
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, GetCoveredKeyFromMutableMemtable) {
+  db_->Put(WriteOptions(), "key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+
+  ReadOptions read_opts;
+  std::string value;
+  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+}
+
+TEST_F(DBRangeDelTest, GetCoveredKeyFromImmutableMemtable) {
+  Options opts = CurrentOptions();
+  opts.max_write_buffer_number = 3;
+  opts.min_write_buffer_number_to_merge = 2;
+  // SpecialSkipListFactory lets us specify maximum number of elements the
+  // memtable can hold. It switches the active memtable to immutable (flush is
+  // prevented by the above options) upon inserting an element that would
+  // overflow the memtable.
+  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+  Reopen(opts);
+
+  db_->Put(WriteOptions(), "key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  db_->Put(WriteOptions(), "blah", "val");
+
+  ReadOptions read_opts;
+  std::string value;
+  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+}
+
+TEST_F(DBRangeDelTest, GetCoveredKeyFromSst) {
+  db_->Put(WriteOptions(), "key", "val");
+  // snapshot prevents key from being deleted during flush
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+
+  ReadOptions read_opts;
+  std::string value;
+  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, GetCoveredMergeOperandFromMemtable) {
+  const int kNumMergeOps = 10;
+  Options opts = CurrentOptions();
+  opts.merge_operator = MergeOperators::CreateUInt64AddOperator();
+  Reopen(opts);
+
+  for (int i = 0; i < kNumMergeOps; ++i) {
+    std::string val;
+    PutFixed64(&val, i);
+    db_->Merge(WriteOptions(), "key", val);
+    if (i == kNumMergeOps / 2) {
+      // deletes [0, 5]
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "key",
+                       "key_");
+    }
+  }
+
+  ReadOptions read_opts;
+  std::string expected, actual;
+  ASSERT_OK(db_->Get(read_opts, "key", &actual));
+  PutFixed64(&expected, 30);  // 6+7+8+9
+  ASSERT_EQ(expected, actual);
+
+  expected.clear();
+  read_opts.ignore_range_deletions = true;
+  ASSERT_OK(db_->Get(read_opts, "key", &actual));
+  PutFixed64(&expected, 45);  // 0+1+2+...+9
+  ASSERT_EQ(expected, actual);
+}
+
+TEST_F(DBRangeDelTest, GetIgnoresRangeDeletions) {
+  Options opts = CurrentOptions();
+  opts.max_write_buffer_number = 4;
+  opts.min_write_buffer_number_to_merge = 3;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+  Reopen(opts);
+
+  db_->Put(WriteOptions(), "sst_key", "val");
+  // snapshot prevents key from being deleted during flush
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  db_->Put(WriteOptions(), "imm_key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  db_->Put(WriteOptions(), "mem_key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+
+  ReadOptions read_opts;
+  read_opts.ignore_range_deletions = true;
+  for (std::string key : {"sst_key", "imm_key", "mem_key"}) {
+    std::string value;
+    ASSERT_OK(db_->Get(read_opts, key, &value));
+  }
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, IteratorRemovesCoveredKeys) {
+  const int kNum = 200, kRangeBegin = 50, kRangeEnd = 150, kNumPerFile = 25;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  Reopen(opts);
+
+  // Write half of the keys before the tombstone and half after the tombstone.
+  // Only covered keys (i.e., within the range and older than the tombstone)
+  // should be deleted.
+  for (int i = 0; i < kNum; ++i) {
+    if (i == kNum / 2) {
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
+    }
+    db_->Put(WriteOptions(), GetNumericStr(i), "val");
+  }
+  ReadOptions read_opts;
+  auto* iter = db_->NewIterator(read_opts);
+
+  int expected = 0;
+  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+    ASSERT_EQ(GetNumericStr(expected), iter->key());
+    if (expected == kRangeBegin - 1) {
+      expected = kNum / 2;
+    } else {
+      ++expected;
+    }
+  }
+  ASSERT_EQ(kNum, expected);
+  delete iter;
+}
+
+TEST_F(DBRangeDelTest, IteratorOverUserSnapshot) {
+  const int kNum = 200, kRangeBegin = 50, kRangeEnd = 150, kNumPerFile = 25;
+  Options opts = CurrentOptions();
+  opts.comparator = test::Uint64Comparator();
+  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
+  Reopen(opts);
+
+  const Snapshot* snapshot = nullptr;
+  // Put a snapshot before the range tombstone, verify an iterator using that
+  // snapshot sees all inserted keys.
+  for (int i = 0; i < kNum; ++i) {
+    if (i == kNum / 2) {
+      snapshot = db_->GetSnapshot();
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
+    }
+    db_->Put(WriteOptions(), GetNumericStr(i), "val");
+  }
+  ReadOptions read_opts;
+  read_opts.snapshot = snapshot;
+  auto* iter = db_->NewIterator(read_opts);
+
+  int expected = 0;
+  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+    ASSERT_EQ(GetNumericStr(expected), iter->key());
+    ++expected;
+  }
+  ASSERT_EQ(kNum / 2, expected);
+  delete iter;
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, IteratorIgnoresRangeDeletions) {
+  Options opts = CurrentOptions();
+  opts.max_write_buffer_number = 4;
+  opts.min_write_buffer_number_to_merge = 3;
+  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+  Reopen(opts);
+
+  db_->Put(WriteOptions(), "sst_key", "val");
+  // snapshot prevents key from being deleted during flush
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  db_->Put(WriteOptions(), "imm_key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  db_->Put(WriteOptions(), "mem_key", "val");
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+
+  ReadOptions read_opts;
+  read_opts.ignore_range_deletions = true;
+  auto* iter = db_->NewIterator(read_opts);
+  int i = 0;
+  std::string expected[] = {"imm_key", "mem_key", "sst_key"};
+  for (iter->SeekToFirst(); iter->Valid(); iter->Next(), ++i) {
+    std::string key;
+    ASSERT_EQ(expected[i], iter->key());
+  }
+  ASSERT_EQ(3, i);
+  delete iter;
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, TailingIteratorRangeTombstoneUnsupported) {
+  db_->Put(WriteOptions(), "key", "val");
+  // snapshot prevents key from being deleted during flush
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+
+  // iterations check unsupported in memtable, l0, and then l1
+  for (int i = 0; i < 3; ++i) {
+    ReadOptions read_opts;
+    read_opts.tailing = true;
+    auto* iter = db_->NewIterator(read_opts);
+    if (i == 2) {
+      // For L1+, iterators over files are created on-demand, so need seek
+      iter->SeekToFirst();
+    }
+    ASSERT_TRUE(iter->status().IsNotSupported());
+    delete iter;
+    if (i == 0) {
+      ASSERT_OK(db_->Flush(FlushOptions()));
+    } else if (i == 1) {
+      MoveFilesToLevel(1);
+    }
+  }
+  db_->ReleaseSnapshot(snapshot);
+}
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -221,8 +221,6 @@ TEST_F(DBSSTTest, DeleteObsoleteFilesPendingOutputs) {
  listener->VerifyMatchedCount(1);
 }

-#endif  // ROCKSDB_LITE
-
 TEST_F(DBSSTTest, DBWithSstFileManager) {
  std::shared_ptr<SstFileManager> sst_file_manager(NewSstFileManager(env_));
  auto sfm = static_cast<SstFileManagerImpl*>(sst_file_manager.get());
@@ -287,7 +285,6 @@ TEST_F(DBSSTTest, DBWithSstFileManager) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

-#ifndef ROCKSDB_LITE
 TEST_F(DBSSTTest, RateLimitedDelete) {
  Destroy(last_options_);
  rocksdb::SyncPoint::GetInstance()->LoadDependency({
@@ -331,9 +328,10 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
  std::string trash_dir = test::TmpDir(env_) + "/trash";
  int64_t rate_bytes_per_sec = 1024 * 10;  // 10 Kbs / Sec
  Status s;
-  options.sst_file_manager.reset(NewSstFileManager(
-      env_, nullptr, trash_dir, rate_bytes_per_sec, false, &s));
+  options.sst_file_manager.reset(
+      NewSstFileManager(env_, nullptr, trash_dir, 0, false, &s));
  ASSERT_OK(s);
+  options.sst_file_manager->SetDeleteRateBytesPerSecond(rate_bytes_per_sec);
  auto sfm = static_cast<SstFileManagerImpl*>(options.sst_file_manager.get());

  ASSERT_OK(TryReopen(options));
@@ -482,7 +480,6 @@ TEST_F(DBSSTTest, DestroyDBWithRateLimitedDelete) {
  // We have deleted the 4 sst files in the delete_scheduler
  ASSERT_EQ(bg_delete_file, 4);
 }
-#endif  // ROCKSDB_LITE

 TEST_F(DBSSTTest, DBWithMaxSpaceAllowed) {
  std::shared_ptr<SstFileManager> sst_file_manager(NewSstFileManager(env_));
@@ -520,18 +517,25 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
  // than the limit.

  std::vector<int> max_space_limits_mbs = {1, 2, 4, 8, 10};
-
+  decltype(max_space_limits_mbs)::value_type limit_mb_cb;
  bool bg_error_set = false;
  uint64_t total_sst_files_size = 0;

+  std::atomic<int> estimate_multiplier(1);
  int reached_max_space_on_flush = 0;
  int reached_max_space_on_compaction = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
      [&](void* arg) {
+        Status* bg_error = static_cast<Status*>(arg);
        bg_error_set = true;
        GetAllSSTFiles(&total_sst_files_size);
        reached_max_space_on_flush++;
+        // low limit for size calculated using sst files
+        ASSERT_GE(total_sst_files_size, limit_mb_cb * 1024 * 1024);
+        // clear error to ensure compaction callback is called
+        *bg_error = Status::OK();
+        estimate_multiplier++;  // used in the main loop assert
      });

  rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -545,6 +549,8 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
  for (auto limit_mb : max_space_limits_mbs) {
    bg_error_set = false;
    total_sst_files_size = 0;
+    estimate_multiplier = 1;
+    limit_mb_cb = limit_mb;
    rocksdb::SyncPoint::GetInstance()->ClearTrace();
    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
    std::shared_ptr<SstFileManager> sst_file_manager(NewSstFileManager(env_));
@@ -569,7 +575,8 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
      // Check the estimated db size vs the db limit just to make sure we
      // dont run into an infinite loop
      estimated_db_size = keys_written * 60;  // ~60 bytes per key
-      ASSERT_LT(estimated_db_size, limit_mb * 1024 * 1024 * 2);
+      ASSERT_LT(estimated_db_size,
+                estimate_multiplier * limit_mb * 1024 * 1024 * 2);
    }
    ASSERT_TRUE(bg_error_set);
    ASSERT_GE(total_sst_files_size, limit_mb * 1024 * 1024);
@@ -580,7 +587,6 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
  ASSERT_GT(reached_max_space_on_compaction, 0);
 }

-#ifndef ROCKSDB_LITE
 TEST_F(DBSSTTest, OpenDBWithInfiniteMaxOpenFiles) {
  // Open DB with infinite max open files
  //  - First iteration use 1 thread to open files
@@ -618,7 +624,7 @@ TEST_F(DBSSTTest, OpenDBWithInfiniteMaxOpenFiles) {
    }
    Close();

-    // Reopening the DB will load all exisitng files
+    // Reopening the DB will load all existing files
    Reopen(options);
    ASSERT_EQ("12,0,12", FilesPerLevel(0));
    std::vector<std::vector<FileMetaData>> files;
@@ -294,8 +294,6 @@ TEST_F(DBTestTailingIterator, TailingIteratorDeletes) {
 }

 TEST_F(DBTestTailingIterator, TailingIteratorPrefixSeek) {
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip,
-             kSkipNoPrefix);
  ReadOptions read_options;
  read_options.tailing = true;

@@ -326,7 +324,6 @@ TEST_F(DBTestTailingIterator, TailingIteratorPrefixSeek) {

  iter->Next();
  ASSERT_TRUE(!iter->Valid());
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip, 0);
 }

 TEST_F(DBTestTailingIterator, TailingIteratorIncomplete) {
@@ -614,8 +611,6 @@ TEST_F(DBTestTailingIterator, ManagedTailingIteratorDeletes) {
 }

 TEST_F(DBTestTailingIterator, ManagedTailingIteratorPrefixSeek) {
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip,
-             kSkipNoPrefix);
  ReadOptions read_options;
  read_options.tailing = true;
  read_options.managed = true;
@@ -647,7 +642,6 @@ TEST_F(DBTestTailingIterator, ManagedTailingIteratorPrefixSeek) {

  iter->Next();
  ASSERT_TRUE(!iter->Valid());
-  XFUNC_TEST("", "dbtest_prefix", prefix_skip1, XFuncPoint::SetSkip, 0);
 }

 TEST_F(DBTestTailingIterator, ManagedTailingIteratorIncomplete) {
@@ -11,6 +11,7 @@
 #include <functional>

 #include "db/db_test_util.h"
+#include "port/port.h"
 #include "port/stack_trace.h"
 #include "rocksdb/persistent_cache.h"
 #include "rocksdb/wal_filter.h"
@@ -1505,6 +1506,10 @@ class MockPersistentCache : public PersistentCache {

  bool IsCompressed() override { return is_compressed_; }

+  std::string GetPrintableOptions() const override {
+    return "MockPersistentCache";
+  }
+
  port::Mutex lock_;
  std::map<std::string, std::string> data_;
  const bool is_compressed_ = true;
@@ -1612,8 +1617,8 @@ TEST_F(DBTest2, SyncPointMarker) {
    CountSyncPoint();
  };

-  auto thread1 = std::thread(func1);
-  auto thread2 = std::thread(func2);
+  auto thread1 = port::Thread(func1);
+  auto thread2 = port::Thread(func2);
  thread1.join();
  thread2.join();

@@ -1902,8 +1907,8 @@ TEST_P(MergeOperatorPinningTest, TailingIterator) {
    delete iter;
  };

-  std::thread writer_thread(writer_func);
-  std::thread reader_thread(reader_func);
+  rocksdb::port::Thread writer_thread(writer_func);
+  rocksdb::port::Thread reader_thread(reader_func);

  writer_thread.join();
  reader_thread.join();
@@ -2174,7 +2179,7 @@ TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {
    cro.exclusive_manual_compaction = false;
    ASSERT_OK(db_->CompactRange(cro, &k1s, &k2s));
  };
-  std::thread bg_thread;
+  rocksdb::port::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
@@ -2195,7 +2200,7 @@ TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {
        ASSERT_OK(Flush());

        // Start a non-exclusive manual compaction in a bg thread
-        bg_thread = std::thread(bg_manual_compact);
+        bg_thread = port::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);
@@ -2212,8 +2217,151 @@ TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
+
+TEST_F(DBTest2, OptimizeForPointLookup) {
+  Options options = CurrentOptions();
+  Close();
+  options.OptimizeForPointLookup(2);
+  ASSERT_OK(DB::Open(options, dbname_, &db_));
+
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_EQ("v1", Get("foo"));
+  Flush();
+  ASSERT_EQ("v1", Get("foo"));
+}
+
 #endif  // ROCKSDB_LITE

+TEST_F(DBTest2, GetRaceFlush1) {
+  ASSERT_OK(Put("foo", "v1"));
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::GetImpl:1", "DBTest2::GetRaceFlush:1"},
+       {"DBTest2::GetRaceFlush:2", "DBImpl::GetImpl:2"}});
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  rocksdb::port::Thread t1([&] {
+    TEST_SYNC_POINT("DBTest2::GetRaceFlush:1");
+    ASSERT_OK(Put("foo", "v2"));
+    Flush();
+    TEST_SYNC_POINT("DBTest2::GetRaceFlush:2");
+  });
+
+  // Get() is issued after the first Put(), so it should see either
+  // "v1" or "v2".
+  ASSERT_NE("NOT_FOUND", Get("foo"));
+  t1.join();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBTest2, GetRaceFlush2) {
+  ASSERT_OK(Put("foo", "v1"));
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::GetImpl:3", "DBTest2::GetRaceFlush:1"},
+       {"DBTest2::GetRaceFlush:2", "DBImpl::GetImpl:4"}});
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  port::Thread t1([&] {
+    TEST_SYNC_POINT("DBTest2::GetRaceFlush:1");
+    ASSERT_OK(Put("foo", "v2"));
+    Flush();
+    TEST_SYNC_POINT("DBTest2::GetRaceFlush:2");
+  });
+
+  // Get() is issued after the first Put(), so it should see either
+  // "v1" or "v2".
+  ASSERT_NE("NOT_FOUND", Get("foo"));
+  t1.join();
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBTest2, DirectIO) {
+  if (!IsDirectIOSupported()) {
+    return;
+  }
+  Options options = CurrentOptions();
+  options.use_direct_reads = options.use_direct_writes = true;
+  options.allow_mmap_reads = options.allow_mmap_writes = false;
+  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));
+  Reopen(options);
+}
+
+TEST_F(DBTest2, MemtableOnlyIterator) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "foo", "first"));
+  ASSERT_OK(Put(1, "bar", "second"));
+
+  ReadOptions ropt;
+  ropt.read_tier = kMemtableTier;
+  std::string value;
+  Iterator* it = nullptr;
+
+  // Before flushing
+  // point lookups
+  ASSERT_OK(db_->Get(ropt, handles_[1], "foo", &value));
+  ASSERT_EQ("first", value);
+  ASSERT_OK(db_->Get(ropt, handles_[1], "bar", &value));
+  ASSERT_EQ("second", value);
+
+  // Memtable-only iterator (read_tier=kMemtableTier); data not flushed yet.
+  it = db_->NewIterator(ropt, handles_[1]);
+  int count = 0;
+  for (it->SeekToFirst(); it->Valid(); it->Next()) {
+    ASSERT_TRUE(it->Valid());
+    count++;
+  }
+  ASSERT_TRUE(!it->Valid());
+  ASSERT_EQ(2, count);
+  delete it;
+
+  Flush(1);
+
+  // After flushing
+  // point lookups
+  ASSERT_OK(db_->Get(ropt, handles_[1], "foo", &value));
+  ASSERT_EQ("first", value);
+  ASSERT_OK(db_->Get(ropt, handles_[1], "bar", &value));
+  ASSERT_EQ("second", value);
+  // nothing should be returned using memtable-only iterator after flushing.
+  it = db_->NewIterator(ropt, handles_[1]);
+  count = 0;
+  for (it->SeekToFirst(); it->Valid(); it->Next()) {
+    ASSERT_TRUE(it->Valid());
+    count++;
+  }
+  ASSERT_TRUE(!it->Valid());
+  ASSERT_EQ(0, count);
+  delete it;
+
+  // Add a key to memtable
+  ASSERT_OK(Put(1, "foobar", "third"));
+  it = db_->NewIterator(ropt, handles_[1]);
+  count = 0;
+  for (it->SeekToFirst(); it->Valid(); it->Next()) {
+    ASSERT_TRUE(it->Valid());
+    ASSERT_EQ("foobar", it->key().ToString());
+    ASSERT_EQ("third", it->value().ToString());
+    count++;
+  }
+  ASSERT_TRUE(!it->Valid());
+  ASSERT_EQ(1, count);
+  delete it;
+}
 }  // namespace rocksdb

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

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

 namespace rocksdb {

@@ -71,7 +72,12 @@ DBTestBase::~DBTestBase() {
  options.db_paths.emplace_back(dbname_ + "_2", 0);
  options.db_paths.emplace_back(dbname_ + "_3", 0);
  options.db_paths.emplace_back(dbname_ + "_4", 0);
-  EXPECT_OK(DestroyDB(dbname_, options));
+
+  if (getenv("KEEP_DB")) {
+    printf("DB is still at %s\n", dbname_.c_str());
+  } else {
+    EXPECT_OK(DestroyDB(dbname_, options));
+  }
  delete env_;
 }

@@ -185,6 +191,36 @@ bool DBTestBase::ChangeCompactOptions() {
  }
 }

+// Switch between different WAL settings
+bool DBTestBase::ChangeWalOptions() {
+  if (option_config_ == kDefault) {
+    option_config_ = kDBLogDir;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    Destroy(options);
+    options.create_if_missing = true;
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kDBLogDir) {
+    option_config_ = kWalDirAndMmapReads;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    Destroy(options);
+    options.create_if_missing = true;
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kWalDirAndMmapReads) {
+    option_config_ = kRecycleLogFiles;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    Destroy(options);
+    TryReopen(options);
+    return true;
+  } else {
+    return false;
+  }
+}
+
 // Switch between different filter policy
 // Jump from kDefault to kFilter to kFullFilter
 bool DBTestBase::ChangeFilterOptions() {
@@ -223,9 +259,6 @@ Options DBTestBase::CurrentOptions(
    const anon::OptionsOverride& options_override) {
  // this redundant copy is to minimize code change w/o having lint error.
  Options options = defaultOptions;
-  XFUNC_TEST("", "dbtest_options", inplace_options1, GetXFTestOptions,
-             reinterpret_cast<Options*>(&options),
-             options_override.skip_policy);
  BlockBasedTableOptions table_options;
  bool set_block_based_table_factory = true;
  switch (option_config_) {
@@ -307,6 +340,7 @@ Options DBTestBase::CurrentOptions(
      break;
    case kManifestFileSize:
      options.max_manifest_file_size = 50;  // 50 bytes
+      break;
    case kPerfOptions:
      options.soft_rate_limit = 2.0;
      options.delayed_write_rate = 8 * 1024 * 1024;
@@ -346,6 +380,11 @@ Options DBTestBase::CurrentOptions(
      options.prefix_extractor.reset(NewNoopTransform());
      break;
    }
+    case kBlockBasedTableWithPartitionedIndex: {
+      table_options.index_type = BlockBasedTableOptions::kTwoLevelIndexSearch;
+      options.prefix_extractor.reset(NewNoopTransform());
+      break;
+    }
    case kBlockBasedTableWithIndexRestartInterval: {
      table_options.index_block_restart_interval = 8;
      break;
@@ -472,10 +511,33 @@ Status DBTestBase::ReadOnlyReopen(const Options& options) {

 Status DBTestBase::TryReopen(const Options& options) {
  Close();
+  last_options_.table_factory.reset();
+  // Note: operator= is an unsafe approach here since it destructs shared_ptr in
+  // the same order of their creation, in contrast to destructors which
+  // destructs them in the opposite order of creation. One particular problme is
+  // that the cache destructor might invoke callback functions that use Option
+  // members such as statistics. To work around this problem, we manually call
+  // destructor of table_facotry which eventually clears the block cache.
  last_options_ = options;
  return DB::Open(options, dbname_, &db_);
 }

+bool DBTestBase::IsDirectIOSupported() {
+  EnvOptions env_options;
+  env_options.use_mmap_writes = false;
+  env_options.use_direct_writes = true;
+  std::string tmp = TempFileName(dbname_, 999);
+  Status s;
+  {
+    unique_ptr<WritableFile> file;
+    s = env_->NewWritableFile(tmp, &file, env_options);
+  }
+  if (s.ok()) {
+    s = env_->DeleteFile(tmp);
+  }
+  return s.ok();
+}
+
 Status DBTestBase::Flush(int cf) {
  if (cf == 0) {
    return db_->Flush(FlushOptions());
@@ -501,6 +563,15 @@ Status DBTestBase::Put(int cf, const Slice& k, const Slice& v,
  }
 }

+Status DBTestBase::Merge(const Slice& k, const Slice& v, WriteOptions wo) {
+  return db_->Merge(wo, k, v);
+}
+
+Status DBTestBase::Merge(int cf, const Slice& k, const Slice& v,
+                         WriteOptions wo) {
+  return db_->Merge(wo, handles_[cf], k, v);
+}
+
 Status DBTestBase::Delete(const std::string& k) {
  return db_->Delete(WriteOptions(), k);
 }
@@ -1089,11 +1160,18 @@ std::vector<std::uint64_t> DBTestBase::ListTableFiles(Env* env,
 }

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

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

@@ -1106,21 +1184,40 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
    // Verify Iterator::Next()
    iter_cnt = 0;
    auto data_iter = true_data.begin();
+    Status s;
    for (iter->SeekToFirst(); iter->Valid(); iter->Next(), data_iter++) {
      ASSERT_EQ(iter->key().ToString(), data_iter->first);
-      ASSERT_EQ(iter->value().ToString(), data_iter->second);
+      Status current_status = status[data_iter->first];
+      if (!current_status.ok()) {
+        s = current_status;
+      }
+      ASSERT_EQ(iter->status(), s);
+      if (current_status.ok()) {
+        ASSERT_EQ(iter->value().ToString(), data_iter->second);
+      }
      iter_cnt++;
      total_reads++;
    }
    ASSERT_EQ(data_iter, true_data.end()) << iter_cnt << " / "
                                          << true_data.size();
+    delete iter;

    // Verify Iterator::Prev()
+    // Use a new iterator to make sure its status is clean.
+    iter = db_->NewIterator(ro);
    iter_cnt = 0;
+    s = Status::OK();
    auto data_rev = true_data.rbegin();
    for (iter->SeekToLast(); iter->Valid(); iter->Prev(), data_rev++) {
      ASSERT_EQ(iter->key().ToString(), data_rev->first);
-      ASSERT_EQ(iter->value().ToString(), data_rev->second);
+      Status current_status = status[data_rev->first];
+      if (!current_status.ok()) {
+        s = current_status;
+      }
+      ASSERT_EQ(iter->status(), s);
+      if (current_status.ok()) {
+        ASSERT_EQ(iter->value().ToString(), data_rev->second);
+      }
      iter_cnt++;
      total_reads++;
    }
@@ -1134,7 +1231,6 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
      ASSERT_EQ(kv.second, iter->value().ToString());
      total_reads++;
    }
-
    delete iter;
  }

@@ -1176,6 +1272,25 @@ void DBTestBase::VerifyDBFromMap(std::map<std::string, std::string> true_data,
  }
 }

+void DBTestBase::VerifyDBInternal(
+    std::vector<std::pair<std::string, std::string>> true_data) {
+  Arena arena;
+  InternalKeyComparator icmp(last_options_.comparator);
+  RangeDelAggregator range_del_agg(icmp, {});
+  auto iter = dbfull()->NewInternalIterator(&arena, &range_del_agg);
+  iter->SeekToFirst();
+  for (auto p : true_data) {
+    ASSERT_TRUE(iter->Valid());
+    ParsedInternalKey ikey;
+    ASSERT_TRUE(ParseInternalKey(iter->key(), &ikey));
+    ASSERT_EQ(p.first, ikey.user_key);
+    ASSERT_EQ(p.second, iter->value());
+    iter->Next();
+  };
+  ASSERT_FALSE(iter->Valid());
+  iter->~InternalIterator();
+}
+
 #ifndef ROCKSDB_LITE

 uint64_t DBTestBase::GetNumberOfSstFilesForColumnFamily(
@@ -58,7 +58,6 @@
 #endif  // !(defined NDEBUG) || !defined(OS_WIN)
 #include "util/testharness.h"
 #include "util/testutil.h"
-#include "util/xfunc.h"
 #include "utilities/merge_operators.h"

 namespace rocksdb {
@@ -120,6 +119,8 @@ struct OptionsOverride {

 }  // namespace anon

+enum SkipPolicy { kSkipNone = 0, kSkipNoSnapshot = 1, kSkipNoPrefix = 2 };
+
 // A hacky skip list mem table that triggers flush after number of entries.
 class SpecialMemTableRep : public MemTableRep {
 public:
@@ -231,6 +232,20 @@ class SpecialEnv : public EnvWrapper {
          return base_->Append(data);
        }
      }
+      Status PositionedAppend(const Slice& data, uint64_t offset) override {
+        if (env_->table_write_callback_) {
+          (*env_->table_write_callback_)();
+        }
+        if (env_->drop_writes_.load(std::memory_order_acquire)) {
+          // Drop writes on the floor
+          return Status::OK();
+        } else if (env_->no_space_.load(std::memory_order_acquire)) {
+          return Status::NoSpace("No space left on device");
+        } else {
+          env_->bytes_written_ += data.size();
+          return base_->PositionedAppend(data, offset);
+        }
+      }
      Status Truncate(uint64_t size) override { return base_->Truncate(size); }
      Status Close() override {
 // SyncPoint is not supported in Released Windows Mode.
@@ -257,6 +272,9 @@ class SpecialEnv : public EnvWrapper {
      Env::IOPriority GetIOPriority() override {
        return base_->GetIOPriority();
      }
+      bool use_direct_io() const override {
+        return base_->use_direct_io();
+      }
    };
    class ManifestFile : public WritableFile {
     public:
@@ -358,7 +376,14 @@ class SpecialEnv : public EnvWrapper {
      return Status::IOError("simulated write error");
    }

-    Status s = target()->NewWritableFile(f, r, soptions);
+    EnvOptions optimized = soptions;
+    if (strstr(f.c_str(), "MANIFEST") != nullptr ||
+        strstr(f.c_str(), "log") != nullptr) {
+      optimized.use_mmap_writes = false;
+      optimized.use_direct_writes = false;
+    }
+
+    Status s = target()->NewWritableFile(f, r, optimized);
    if (s.ok()) {
      if (strstr(f.c_str(), ".sst") != nullptr) {
        r->reset(new SSTableFile(this, std::move(*r)));
@@ -594,6 +619,7 @@ class DBTestBase : public testing::Test {
    kLevelSubcompactions = 31,
    kUniversalSubcompactions = 32,
    kBlockBasedTableWithIndexRestartInterval = 33,
+    kBlockBasedTableWithPartitionedIndex = 34,
  };
  int option_config_;

@@ -645,9 +671,12 @@ class DBTestBase : public testing::Test {
  // test.  Return false if there are no more configurations to test.
  bool ChangeOptions(int skip_mask = kNoSkip);

-  // Switch between different compaction styles (we have only 2 now).
+  // Switch between different compaction styles.
  bool ChangeCompactOptions();

+  // Switch between different WAL-realted options.
+  bool ChangeWalOptions();
+
  // Switch between different filter policy
  // Jump from kDefault to kFilter to kFullFilter
  bool ChangeFilterOptions();
@@ -692,6 +721,8 @@ class DBTestBase : public testing::Test {

  Status TryReopen(const Options& options);

+  bool IsDirectIOSupported();
+
  Status Flush(int cf = 0);

  Status Put(const Slice& k, const Slice& v, WriteOptions wo = WriteOptions());
@@ -699,6 +730,12 @@ class DBTestBase : public testing::Test {
  Status Put(int cf, const Slice& k, const Slice& v,
             WriteOptions wo = WriteOptions());

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

  Status Delete(int cf, const std::string& k);
@@ -825,9 +862,13 @@ 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,
-                       bool tailing_iter = false);
+  void VerifyDBFromMap(
+      std::map<std::string, std::string> true_data,
+      size_t* total_reads_res = nullptr, bool tailing_iter = false,
+      std::map<std::string, Status> status = std::map<std::string, Status>());
+
+  void VerifyDBInternal(
+      std::vector<std::pair<std::string, std::string>> true_data);

 #ifndef ROCKSDB_LITE
  uint64_t GetNumberOfSstFilesForColumnFamily(DB* db,
@@ -1028,13 +1028,15 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
 // Test that checks trivial move in universal compaction
 TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
  int32_t trivial_move = 0;
-  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "DBImpl::BackgroundCompaction:NonTrivial",
-      [&](void* arg) { non_trivial_move++; });
+      "DBImpl::BackgroundCompaction:NonTrivial", [&](void* arg) {
+        ASSERT_TRUE(arg != nullptr);
+        int output_level = *(static_cast<int*>(arg));
+        ASSERT_EQ(output_level, 0);
+      });

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

@@ -1044,7 +1046,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
  options.num_levels = 15;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 8;
-  options.max_background_compactions = 1;
+  options.max_background_compactions = 2;
  options.target_file_size_base = 64 * 1024;
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);
@@ -1065,7 +1067,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
  dbfull()->TEST_WaitForCompact();

  ASSERT_GT(trivial_move, 0);
-  ASSERT_EQ(non_trivial_move, 0);

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include "db/db_test_util.h"
+#include "port/port.h"
 #include "port/stack_trace.h"
 #include "util/fault_injection_test_env.h"
 #include "util/options_helper.h"
@@ -50,7 +51,7 @@ TEST_F(DBWALTest, WAL) {
    // again both values should be present.
    ASSERT_EQ("v3", Get(1, "foo"));
    ASSERT_EQ("v3", Get(1, "bar"));
-  } while (ChangeCompactOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, RollLog) {
@@ -67,7 +68,7 @@ TEST_F(DBWALTest, RollLog) {
    for (int i = 0; i < 10; i++) {
      ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    }
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, SyncWALNotBlockWrite) {
@@ -86,7 +87,7 @@ TEST_F(DBWALTest, SyncWALNotBlockWrite) {
  });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  std::thread thread([&]() { ASSERT_OK(db_->SyncWAL()); });
+  rocksdb::port::Thread thread([&]() { ASSERT_OK(db_->SyncWAL()); });

  TEST_SYNC_POINT("DBWALTest::SyncWALNotBlockWrite:1");
  ASSERT_OK(Put("foo2", "bar2"));
@@ -118,7 +119,7 @@ TEST_F(DBWALTest, SyncWALNotWaitWrite) {
  });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

-  std::thread thread([&]() { ASSERT_OK(Put("foo2", "bar2")); });
+  rocksdb::port::Thread thread([&]() { ASSERT_OK(Put("foo2", "bar2")); });
  TEST_SYNC_POINT("DBWALTest::SyncWALNotWaitWrite:1");
  ASSERT_OK(db_->SyncWAL());
  TEST_SYNC_POINT("DBWALTest::SyncWALNotWaitWrite:2");
@@ -150,7 +151,7 @@ TEST_F(DBWALTest, Recover) {
    ASSERT_EQ("v4", Get(1, "foo"));
    ASSERT_EQ("v2", Get(1, "bar"));
    ASSERT_EQ("v5", Get(1, "baz"));
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, RecoverWithTableHandle) {
@@ -187,23 +188,22 @@ TEST_F(DBWALTest, RecoverWithTableHandle) {
        }
      }
    }
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, IgnoreRecoveredLog) {
  std::string backup_logs = dbname_ + "/backup_logs";

-  // delete old files in backup_logs directory
-  env_->CreateDirIfMissing(backup_logs);
-  std::vector<std::string> old_files;
-  env_->GetChildren(backup_logs, &old_files);
-  for (auto& file : old_files) {
-    if (file != "." && file != "..") {
-      env_->DeleteFile(backup_logs + "/" + file);
-    }
-  }
-
  do {
+    // delete old files in backup_logs directory
+    env_->CreateDirIfMissing(backup_logs);
+    std::vector<std::string> old_files;
+    env_->GetChildren(backup_logs, &old_files);
+    for (auto& file : old_files) {
+      if (file != "." && file != "..") {
+        env_->DeleteFile(backup_logs + "/" + file);
+      }
+    }
    Options options = CurrentOptions();
    options.create_if_missing = true;
    options.merge_operator = MergeOperators::CreateUInt64AddOperator();
@@ -276,7 +276,8 @@ TEST_F(DBWALTest, IgnoreRecoveredLog) {
    }
    Status s = TryReopen(options);
    ASSERT_TRUE(!s.ok());
-  } while (ChangeOptions(kSkipHashCuckoo));
+    Destroy(options);
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, RecoveryWithEmptyLog) {
@@ -289,7 +290,7 @@ TEST_F(DBWALTest, RecoveryWithEmptyLog) {
    ASSERT_OK(Put(1, "foo", "v3"));
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_EQ("v3", Get(1, "foo"));
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 #if !(defined NDEBUG) || !defined(OS_WIN)
@@ -428,7 +429,7 @@ TEST_F(DBWALTest, GetSortedWalFiles) {
    ASSERT_OK(Put(1, "foo", "v1"));
    ASSERT_OK(dbfull()->GetSortedWalFiles(log_files));
    ASSERT_EQ(1, log_files.size());
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, RecoveryWithLogDataForSomeCFs) {
@@ -453,7 +454,7 @@ TEST_F(DBWALTest, RecoveryWithLogDataForSomeCFs) {
    }
    // Check at least the first WAL was cleaned up during the recovery.
    ASSERT_LT(earliest_log_nums[0], earliest_log_nums[1]);
-  } while (ChangeOptions());
+  } while (ChangeWalOptions());
 }

 TEST_F(DBWALTest, RecoverWithLargeLog) {
@@ -480,7 +481,7 @@ TEST_F(DBWALTest, RecoverWithLargeLog) {
    ASSERT_EQ(std::string(10, '3'), Get(1, "small3"));
    ASSERT_EQ(std::string(10, '4'), Get(1, "small4"));
    ASSERT_GT(NumTableFilesAtLevel(0, 1), 1);
-  } while (ChangeCompactOptions());
+  } while (ChangeWalOptions());
 }

 // In https://reviews.facebook.net/D20661 we change
@@ -678,9 +679,9 @@ class RecoveryTestHelper {
  // Starting number for the WAL file name like 00010.log
  static const int kWALFileOffset = 10;
  // Keys to be written per WAL file
-  static const int kKeysPerWALFile = 1024;
+  static const int kKeysPerWALFile = 133;
  // Size of the value
-  static const int kValueSize = 10;
+  static const int kValueSize = 96;

  // Create WAL files with values filled in
  static void FillData(DBWALTest* test, const Options& options,
@@ -689,7 +690,7 @@ class RecoveryTestHelper {

    *count = 0;

-    shared_ptr<Cache> table_cache = NewLRUCache(50000, 16);
+    shared_ptr<Cache> table_cache = NewLRUCache(50, 0);
    EnvOptions env_options;
    WriteBufferManager write_buffer_manager(db_options.db_write_buffer_size);

@@ -716,12 +717,13 @@ class RecoveryTestHelper {
          new log::Writer(std::move(file_writer), current_log_number,
                          db_options.recycle_log_file_num > 0));

+      WriteBatch batch;
      for (int i = 0; i < kKeysPerWALFile; i++) {
        std::string key = "key" + ToString((*count)++);
        std::string value = test->DummyString(kValueSize);
        assert(current_log_writer.get() != nullptr);
        uint64_t seq = versions->LastSequence() + 1;
-        WriteBatch batch;
+        batch.Clear();
        batch.Put(key, value);
        WriteBatchInternal::SetSequence(&batch, seq);
        current_log_writer->AddRecord(WriteBatchInternal::Contents(&batch));
@@ -772,7 +774,7 @@ class RecoveryTestHelper {
    if (trunc) {
      ASSERT_EQ(0, truncate(fname.c_str(), static_cast<int64_t>(size * off)));
    } else {
-      InduceCorruption(fname, static_cast<size_t>(size * off),
+      InduceCorruption(fname, static_cast<size_t>(size * off + 8),
                       static_cast<size_t>(size * len));
    }
  }
@@ -791,7 +793,7 @@ class RecoveryTestHelper {
    ASSERT_EQ(offset, lseek(fd, static_cast<long>(offset), SEEK_SET));

    void* buf = alloca(len);
-    memset(buf, 'a', len);
+    memset(buf, 'b', len);
    ASSERT_EQ(len, write(fd, buf, static_cast<unsigned int>(len)));

    close(fd);
@@ -807,7 +809,7 @@ TEST_F(DBWALTest, kTolerateCorruptedTailRecords) {
  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;

  for (auto trunc : {true, false}) {        /* Corruption style */
-    for (int i = 0; i < 4; i++) {           /* Corruption offset position */
+    for (int i = 0; i < 3; i++) {           /* Corruption offset position */
      for (int j = jstart; j < jend; j++) { /* WAL file */
        // Fill data for testing
        Options options = CurrentOptions();
@@ -528,10 +528,11 @@ struct RangeTombstone {
  Slice start_key_;
  Slice end_key_;
  SequenceNumber seq_;
-  explicit RangeTombstone(Slice sk, Slice ek, SequenceNumber sn)
+  RangeTombstone() = default;
+  RangeTombstone(Slice sk, Slice ek, SequenceNumber sn)
      : start_key_(sk), end_key_(ek), seq_(sn) {}

-  explicit RangeTombstone(ParsedInternalKey parsed_key, Slice value) {
+  RangeTombstone(ParsedInternalKey parsed_key, Slice value) {
    start_key_ = parsed_key.user_key;
    seq_ = parsed_key.sequence;
    end_key_ = value;
@@ -128,7 +128,7 @@ void EventHelpers::LogAndNotifyTableFileDeletion(
  info.job_id = job_id;
  info.file_path = file_path;
  info.status = status;
-  for (auto listener : listeners) {
+  for (auto& listener : listeners) {
    listener->OnTableFileDeleted(info);
  }
 #endif  // !ROCKSDB_LITE
@@ -0,0 +1,323 @@
+//  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 <functional>
+
+#include "db/db_test_util.h"
+#include "port/port.h"
+#include "port/stack_trace.h"
+#include "rocksdb/sst_file_writer.h"
+#include "util/testutil.h"
+
+namespace rocksdb {
+
+#ifndef ROCKSDB_LITE
+class ExternalSSTFileBasicTest : public DBTestBase {
+ public:
+  ExternalSSTFileBasicTest() : DBTestBase("/external_sst_file_test") {
+    sst_files_dir_ = dbname_ + "/sst_files/";
+    DestroyAndRecreateExternalSSTFilesDir();
+  }
+
+  void DestroyAndRecreateExternalSSTFilesDir() {
+    test::DestroyDir(env_, sst_files_dir_);
+    env_->CreateDir(sst_files_dir_);
+  }
+
+  Status DeprecatedAddFile(const std::vector<std::string>& files,
+                           bool move_files = false,
+                           bool skip_snapshot_check = false) {
+    IngestExternalFileOptions opts;
+    opts.move_files = move_files;
+    opts.snapshot_consistency = !skip_snapshot_check;
+    opts.allow_global_seqno = false;
+    opts.allow_blocking_flush = false;
+    return db_->IngestExternalFile(files, opts);
+  }
+
+  Status GenerateAndAddExternalFile(
+      const Options options, std::vector<int> keys, int file_id,
+      std::map<std::string, std::string>* true_data) {
+    std::string file_path = sst_files_dir_ + ToString(file_id);
+    SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator,
+                                  nullptr);
+
+    Status s = sst_file_writer.Open(file_path);
+    if (!s.ok()) {
+      return s;
+    }
+    for (int k : keys) {
+      std::string key = Key(k);
+      std::string value = Key(k) + ToString(file_id);
+      s = sst_file_writer.Add(key, value);
+      (*true_data)[key] = value;
+      if (!s.ok()) {
+        sst_file_writer.Finish();
+        return s;
+      }
+    }
+    s = sst_file_writer.Finish();
+
+    if (s.ok()) {
+      IngestExternalFileOptions ifo;
+      ifo.allow_global_seqno = true;
+      s = db_->IngestExternalFile({file_path}, ifo);
+    }
+
+    return s;
+  }
+
+  ~ExternalSSTFileBasicTest() { test::DestroyDir(env_, sst_files_dir_); }
+
+ protected:
+  std::string sst_files_dir_;
+};
+
+TEST_F(ExternalSSTFileBasicTest, Basic) {
+  Options options = CurrentOptions();
+
+  SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
+
+  // Current file size should be 0 after sst_file_writer init and before open a
+  // file.
+  ASSERT_EQ(sst_file_writer.FileSize(), 0);
+
+  // file1.sst (0 => 99)
+  std::string file1 = sst_files_dir_ + "file1.sst";
+  ASSERT_OK(sst_file_writer.Open(file1));
+  for (int k = 0; k < 100; k++) {
+    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+  }
+  ExternalSstFileInfo file1_info;
+  Status s = sst_file_writer.Finish(&file1_info);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+
+  // Current file size should be non-zero after success write.
+  ASSERT_GT(sst_file_writer.FileSize(), 0);
+
+  ASSERT_EQ(file1_info.file_path, file1);
+  ASSERT_EQ(file1_info.num_entries, 100);
+  ASSERT_EQ(file1_info.smallest_key, Key(0));
+  ASSERT_EQ(file1_info.largest_key, Key(99));
+  // sst_file_writer already finished, cannot add this value
+  s = sst_file_writer.Add(Key(100), "bad_val");
+  ASSERT_FALSE(s.ok()) << s.ToString();
+
+  DestroyAndReopen(options);
+  // Add file using file path
+  s = DeprecatedAddFile({file1});
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_EQ(db_->GetLatestSequenceNumber(), 0U);
+  for (int k = 0; k < 100; k++) {
+    ASSERT_EQ(Get(Key(k)), Key(k) + "_val");
+  }
+
+  DestroyAndRecreateExternalSSTFilesDir();
+}
+
+TEST_F(ExternalSSTFileBasicTest, NoCopy) {
+  Options options = CurrentOptions();
+  const ImmutableCFOptions ioptions(options);
+
+  SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
+
+  // file1.sst (0 => 99)
+  std::string file1 = sst_files_dir_ + "file1.sst";
+  ASSERT_OK(sst_file_writer.Open(file1));
+  for (int k = 0; k < 100; k++) {
+    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+  }
+  ExternalSstFileInfo file1_info;
+  Status s = sst_file_writer.Finish(&file1_info);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_EQ(file1_info.file_path, file1);
+  ASSERT_EQ(file1_info.num_entries, 100);
+  ASSERT_EQ(file1_info.smallest_key, Key(0));
+  ASSERT_EQ(file1_info.largest_key, Key(99));
+
+  // file2.sst (100 => 299)
+  std::string file2 = sst_files_dir_ + "file2.sst";
+  ASSERT_OK(sst_file_writer.Open(file2));
+  for (int k = 100; k < 300; k++) {
+    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
+  }
+  ExternalSstFileInfo file2_info;
+  s = sst_file_writer.Finish(&file2_info);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_EQ(file2_info.file_path, file2);
+  ASSERT_EQ(file2_info.num_entries, 200);
+  ASSERT_EQ(file2_info.smallest_key, Key(100));
+  ASSERT_EQ(file2_info.largest_key, Key(299));
+
+  // file3.sst (110 => 124) .. overlap with file2.sst
+  std::string file3 = sst_files_dir_ + "file3.sst";
+  ASSERT_OK(sst_file_writer.Open(file3));
+  for (int k = 110; k < 125; k++) {
+    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
+  }
+  ExternalSstFileInfo file3_info;
+  s = sst_file_writer.Finish(&file3_info);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_EQ(file3_info.file_path, file3);
+  ASSERT_EQ(file3_info.num_entries, 15);
+  ASSERT_EQ(file3_info.smallest_key, Key(110));
+  ASSERT_EQ(file3_info.largest_key, Key(124));
+  s = DeprecatedAddFile({file1}, true /* move file */);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_EQ(Status::NotFound(), env_->FileExists(file1));
+
+  s = DeprecatedAddFile({file2}, false /* copy file */);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  ASSERT_OK(env_->FileExists(file2));
+
+  // This file have overlapping values with the existing data
+  s = DeprecatedAddFile({file2}, true /* move file */);
+  ASSERT_FALSE(s.ok()) << s.ToString();
+  ASSERT_OK(env_->FileExists(file3));
+
+  for (int k = 0; k < 300; k++) {
+    ASSERT_EQ(Get(Key(k)), Key(k) + "_val");
+  }
+}
+
+TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  std::map<std::string, std::string> true_data;
+
+  int file_id = 1;
+
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6}, file_id++,
+                                       &true_data));
+  // File dont overwrite any keys, No seqno needed
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+  ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13}, file_id++,
+                                       &true_data));
+  // File dont overwrite any keys, No seqno needed
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {1, 4, 6}, file_id++, &true_data));
+  // File overwrite some keys, a seqno will be assigned
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {11, 15, 19}, file_id++, &true_data));
+  // File overwrite some keys, a seqno will be assigned
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {120, 130}, file_id++, &true_data));
+  // File dont overwrite any keys, No seqno needed
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {1, 130}, file_id++, &true_data));
+  // File overwrite some keys, a seqno will be assigned
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+
+  // Write some keys through normal write path
+  for (int i = 0; i < 50; i++) {
+    ASSERT_OK(Put(Key(i), "memtable"));
+    true_data[Key(i)] = "memtable";
+  }
+  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {60, 61, 62}, file_id++, &true_data));
+  // File dont overwrite any keys, No seqno needed
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {40, 41, 42}, file_id++, &true_data));
+  // File overwrite some keys, a seqno will be assigned
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {20, 30, 40}, file_id++, &true_data));
+  // File overwrite some keys, a seqno will be assigned
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
+
+  const Snapshot* snapshot = db_->GetSnapshot();
+
+  // We will need a seqno for the file regardless if the file overwrite
+  // keys in the DB or not because we have a snapshot
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {1000, 1002}, file_id++, &true_data));
+  // A global seqno will be assigned anyway because of the snapshot
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {2000, 3002}, file_id++, &true_data));
+  // A global seqno will be assigned anyway because of the snapshot
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
+
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
+                                       file_id++, &true_data));
+  // A global seqno will be assigned anyway because of the snapshot
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+  db_->ReleaseSnapshot(snapshot);
+
+  ASSERT_OK(
+      GenerateAndAddExternalFile(options, {5000, 5001}, file_id++, &true_data));
+  // No snapshot anymore, no need to assign a seqno
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
+
+  size_t kcnt = 0;
+  VerifyDBFromMap(true_data, &kcnt, false);
+}
+
+TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
+  Options options = CurrentOptions();
+  const int kNumKeys = 10000;
+
+  size_t total_fadvised_bytes = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "SstFileWriter::InvalidatePageCache", [&](void* arg) {
+        size_t fadvise_size = *(reinterpret_cast<size_t*>(arg));
+        total_fadvised_bytes += fadvise_size;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::unique_ptr<SstFileWriter> sst_file_writer;
+
+  std::string sst_file_path = sst_files_dir_ + "file_fadvise_disable.sst";
+  sst_file_writer.reset(new SstFileWriter(EnvOptions(), options,
+                                          options.comparator, nullptr, false));
+  ASSERT_OK(sst_file_writer->Open(sst_file_path));
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(sst_file_writer->Add(Key(i), Key(i)));
+  }
+  ASSERT_OK(sst_file_writer->Finish());
+  // fadvise disabled
+  ASSERT_EQ(total_fadvised_bytes, 0);
+
+
+  sst_file_path = sst_files_dir_ + "file_fadvise_enable.sst";
+  sst_file_writer.reset(new SstFileWriter(EnvOptions(), options,
+                                          options.comparator, nullptr, true));
+  ASSERT_OK(sst_file_writer->Open(sst_file_path));
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(sst_file_writer->Add(Key(i), Key(i)));
+  }
+  ASSERT_OK(sst_file_writer->Finish());
+  // fadvise enabled
+  ASSERT_EQ(total_fadvised_bytes, sst_file_writer->FileSize());
+  ASSERT_GT(total_fadvised_bytes, 0);
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+#endif  // ROCKSDB_LITE
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -3,6 +3,8 @@
 //  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.

+#ifndef ROCKSDB_LITE
+
 #include "db/external_sst_file_ingestion_job.h"

 #define __STDC_FORMAT_MACROS
@@ -13,7 +15,7 @@
 #include <vector>

 #include "db/version_edit.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "table/scoped_arena_iterator.h"
 #include "table/sst_file_writer_collectors.h"
 #include "table/table_builder.h"
@@ -38,6 +40,15 @@ Status ExternalSstFileIngestionJob::Prepare(
    files_to_ingest_.push_back(file_to_ingest);
  }

+  for (const IngestedFileInfo& f : files_to_ingest_) {
+    if (f.cf_id !=
+            TablePropertiesCollectorFactory::Context::kUnknownColumnFamily &&
+        f.cf_id != cfd_->GetID()) {
+      return Status::InvalidArgument(
+          "External file column family id dont match");
+    }
+  }
+
  const Comparator* ucmp = cfd_->internal_comparator().user_comparator();
  auto num_files = files_to_ingest_.size();
  if (num_files == 0) {
@@ -87,10 +98,12 @@ Status ExternalSstFileIngestionJob::Prepare(
      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);
+        status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+                          db_options_.use_fsync);
      }
    } else {
-      status = CopyFile(env_, path_outside_db, path_inside_db, 0);
+      status = CopyFile(env_, path_outside_db, path_inside_db, 0,
+                        db_options_.use_fsync);
    }
    TEST_SYNC_POINT("DBImpl::AddFile:FileCopied");
    if (!status.ok()) {
@@ -295,15 +308,30 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
    if (file_to_ingest->global_seqno_offset == 0) {
      return Status::Corruption("Was not able to find file global seqno field");
    }
+  } else if (file_to_ingest->version == 1) {
+    // SST file V1 should not have global seqno field
+    assert(seqno_iter == uprops.end());
+    file_to_ingest->original_seqno = 0;
+    if (ingestion_options_.allow_blocking_flush ||
+            ingestion_options_.allow_global_seqno) {
+      return Status::InvalidArgument(
+            "External SST file V1 does not support global seqno");
+    }
  } else {
-    return Status::InvalidArgument("external file version is not supported");
+    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()));
+  ReadOptions ro;
+  // During reading the external file we can cache blocks that we read into
+  // the block cache, if we later change the global seqno of this file, we will
+  // have block in cache that will include keys with wrong seqno.
+  // We need to disable fill_cache so that we read from the file without
+  // updating the block cache.
+  ro.fill_cache = false;
+  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(ro));

  // Get first (smallest) key from file
  iter->SeekToFirst();
@@ -325,6 +353,10 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
  }
  file_to_ingest->largest_user_key = key.user_key.ToString();

+  file_to_ingest->cf_id = static_cast<uint32_t>(props->column_family_id);
+
+  file_to_ingest->table_properties = *props;
+
  return status;
 }

@@ -495,3 +527,5 @@ bool ExternalSstFileIngestionJob::IngestedFileFitInLevel(
 }

 }  // namespace rocksdb
+
+#endif  // !ROCKSDB_LITE
@@ -36,6 +36,10 @@ struct IngestedFileInfo {
  uint64_t file_size;
  // total number of keys in external file
  uint64_t num_entries;
+  // Id of column family this file shoule be ingested into
+  uint32_t cf_id;
+  // TableProperties read from external file
+  TableProperties table_properties;
  // Version of external file
  int version;

@@ -96,6 +100,10 @@ class ExternalSstFileIngestionJob {

  VersionEdit* edit() { return &edit_; }

+  const autovector<IngestedFileInfo>& files_to_ingest() const {
+    return files_to_ingest_;
+  }
+
 private:
  // Open the external file and populate `file_to_ingest` with all the
  // external information we need to ingest this file.
@@ -3,6 +3,9 @@
 //  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.

+#ifndef ROCKSDB_LITE
+
+#include <functional>
 #include "db/db_test_util.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
@@ -11,11 +14,10 @@

 namespace rocksdb {

-#ifndef ROCKSDB_LITE
 class ExternalSSTFileTest : public DBTestBase {
 public:
  ExternalSSTFileTest() : DBTestBase("/external_sst_file_test") {
-    sst_files_dir_ = test::TmpDir(env_) + "/sst_files/";
+    sst_files_dir_ = dbname_ + "/sst_files/";
    DestroyAndRecreateExternalSSTFilesDir();
  }

@@ -28,7 +30,8 @@ class ExternalSSTFileTest : public DBTestBase {
      const Options options,
      std::vector<std::pair<std::string, std::string>> data, int file_id = -1,
      bool allow_global_seqno = false, bool sort_data = false,
-      std::map<std::string, std::string>* true_data = nullptr) {
+      std::map<std::string, std::string>* true_data = nullptr,
+      ColumnFamilyHandle* cfh = nullptr) {
    // Generate a file id if not provided
    if (file_id == -1) {
      file_id = last_file_id_ + 1;
@@ -51,7 +54,8 @@ class ExternalSSTFileTest : public DBTestBase {
      data.resize(uniq_iter - data.begin());
    }
    std::string file_path = sst_files_dir_ + ToString(file_id);
-    SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
+    SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator,
+                                  cfh);

    Status s = sst_file_writer.Open(file_path);
    if (!s.ok()) {
@@ -69,7 +73,11 @@ class ExternalSSTFileTest : public DBTestBase {
    if (s.ok()) {
      IngestExternalFileOptions ifo;
      ifo.allow_global_seqno = allow_global_seqno;
-      s = db_->IngestExternalFile({file_path}, ifo);
+      if (cfh) {
+        s = db_->IngestExternalFile(cfh, {file_path}, ifo);
+      } else {
+        s = db_->IngestExternalFile({file_path}, ifo);
+      }
    }

    if (s.ok() && true_data) {
@@ -84,25 +92,29 @@ class ExternalSSTFileTest : public DBTestBase {
  Status GenerateAndAddExternalFile(
      const Options options, std::vector<std::pair<int, std::string>> data,
      int file_id = -1, bool allow_global_seqno = false, bool sort_data = false,
-      std::map<std::string, std::string>* true_data = nullptr) {
+      std::map<std::string, std::string>* true_data = nullptr,
+      ColumnFamilyHandle* cfh = nullptr) {
    std::vector<std::pair<std::string, std::string>> file_data;
    for (auto& entry : data) {
      file_data.emplace_back(Key(entry.first), entry.second);
    }
    return GenerateAndAddExternalFile(options, file_data, file_id,
-                                      allow_global_seqno, sort_data, true_data);
+                                      allow_global_seqno, sort_data, true_data,
+                                      cfh);
  }

  Status GenerateAndAddExternalFile(
      const Options options, std::vector<int> keys, int file_id = -1,
      bool allow_global_seqno = false, bool sort_data = false,
-      std::map<std::string, std::string>* true_data = nullptr) {
+      std::map<std::string, std::string>* true_data = nullptr,
+      ColumnFamilyHandle* cfh = nullptr) {
    std::vector<std::pair<std::string, std::string>> file_data;
    for (auto& k : keys) {
      file_data.emplace_back(Key(k), Key(k) + ToString(file_id));
    }
    return GenerateAndAddExternalFile(options, file_data, file_id,
-                                      allow_global_seqno, sort_data, true_data);
+                                      allow_global_seqno, sort_data, true_data,
+                                      cfh);
  }

  Status DeprecatedAddFile(const std::vector<std::string>& files,
@@ -129,6 +141,9 @@ TEST_F(ExternalSSTFileTest, Basic) {

    SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);

+    // Current file size should be 0 after sst_file_writer init and before open a file.
+    ASSERT_EQ(sst_file_writer.FileSize(), 0);
+
    // file1.sst (0 => 99)
    std::string file1 = sst_files_dir_ + "file1.sst";
    ASSERT_OK(sst_file_writer.Open(file1));
@@ -138,6 +153,10 @@ TEST_F(ExternalSSTFileTest, Basic) {
    ExternalSstFileInfo file1_info;
    Status s = sst_file_writer.Finish(&file1_info);
    ASSERT_TRUE(s.ok()) << s.ToString();
+
+    // Current file size should be non-zero after success write.
+    ASSERT_GT(sst_file_writer.FileSize(), 0);
+
    ASSERT_EQ(file1_info.file_path, file1);
    ASSERT_EQ(file1_info.num_entries, 100);
    ASSERT_EQ(file1_info.smallest_key, Key(0));
@@ -172,7 +191,10 @@ TEST_F(ExternalSSTFileTest, Basic) {
    }
    ExternalSstFileInfo file3_info;
    s = sst_file_writer.Finish(&file3_info);
+
    ASSERT_TRUE(s.ok()) << s.ToString();
+    // Current file size should be non-zero after success finish.
+    ASSERT_GT(sst_file_writer.FileSize(), 0);
    ASSERT_EQ(file3_info.file_path, file3);
    ASSERT_EQ(file3_info.num_entries, 105);
    ASSERT_EQ(file3_info.smallest_key, Key(195));
@@ -236,11 +258,11 @@ TEST_F(ExternalSSTFileTest, Basic) {
      ASSERT_EQ(Get(Key(k)), Key(k) + "_val");
    }

-    // This file has overlapping values with the exisitng data
+    // This file has overlapping values with the existing data
    s = DeprecatedAddFile({file3});
    ASSERT_FALSE(s.ok()) << s.ToString();

-    // This file has overlapping values with the exisitng data
+    // This file has overlapping values with the existing data
    s = DeprecatedAddFile({file4});
    ASSERT_FALSE(s.ok()) << s.ToString();

@@ -488,7 +510,7 @@ TEST_F(ExternalSSTFileTest, AddList) {
      ASSERT_EQ(user_props["xyz_Count"], "100");
    }

-    // This file list has overlapping values with the exisitng data
+    // This file list has overlapping values with the existing data
    s = DeprecatedAddFile(file_list3);
    ASSERT_FALSE(s.ok()) << s.ToString();

@@ -628,71 +650,6 @@ TEST_F(ExternalSSTFileTest, PurgeObsoleteFilesBug) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

-TEST_F(ExternalSSTFileTest, NoCopy) {
-  Options options = CurrentOptions();
-  const ImmutableCFOptions ioptions(options);
-
-  SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
-
-  // file1.sst (0 => 99)
-  std::string file1 = sst_files_dir_ + "file1.sst";
-  ASSERT_OK(sst_file_writer.Open(file1));
-  for (int k = 0; k < 100; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
-  }
-  ExternalSstFileInfo file1_info;
-  Status s = sst_file_writer.Finish(&file1_info);
-  ASSERT_TRUE(s.ok()) << s.ToString();
-  ASSERT_EQ(file1_info.file_path, file1);
-  ASSERT_EQ(file1_info.num_entries, 100);
-  ASSERT_EQ(file1_info.smallest_key, Key(0));
-  ASSERT_EQ(file1_info.largest_key, Key(99));
-
-  // file2.sst (100 => 299)
-  std::string file2 = sst_files_dir_ + "file2.sst";
-  ASSERT_OK(sst_file_writer.Open(file2));
-  for (int k = 100; k < 300; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val"));
-  }
-  ExternalSstFileInfo file2_info;
-  s = sst_file_writer.Finish(&file2_info);
-  ASSERT_TRUE(s.ok()) << s.ToString();
-  ASSERT_EQ(file2_info.file_path, file2);
-  ASSERT_EQ(file2_info.num_entries, 200);
-  ASSERT_EQ(file2_info.smallest_key, Key(100));
-  ASSERT_EQ(file2_info.largest_key, Key(299));
-
-  // file3.sst (110 => 124) .. overlap with file2.sst
-  std::string file3 = sst_files_dir_ + "file3.sst";
-  ASSERT_OK(sst_file_writer.Open(file3));
-  for (int k = 110; k < 125; k++) {
-    ASSERT_OK(sst_file_writer.Add(Key(k), Key(k) + "_val_overlap"));
-  }
-  ExternalSstFileInfo file3_info;
-  s = sst_file_writer.Finish(&file3_info);
-  ASSERT_TRUE(s.ok()) << s.ToString();
-  ASSERT_EQ(file3_info.file_path, file3);
-  ASSERT_EQ(file3_info.num_entries, 15);
-  ASSERT_EQ(file3_info.smallest_key, Key(110));
-  ASSERT_EQ(file3_info.largest_key, Key(124));
-  s = DeprecatedAddFile({file1}, true /* move file */);
-  ASSERT_TRUE(s.ok()) << s.ToString();
-  ASSERT_EQ(Status::NotFound(), env_->FileExists(file1));
-
-  s = DeprecatedAddFile({file2}, false /* copy file */);
-  ASSERT_TRUE(s.ok()) << s.ToString();
-  ASSERT_OK(env_->FileExists(file2));
-
-  // This file have overlapping values with the exisitng data
-  s = DeprecatedAddFile({file2}, true /* move file */);
-  ASSERT_FALSE(s.ok()) << s.ToString();
-  ASSERT_OK(env_->FileExists(file3));
-
-  for (int k = 0; k < 300; k++) {
-    ASSERT_EQ(Get(Key(k)), Key(k) + "_val");
-  }
-}
-
 TEST_F(ExternalSSTFileTest, SkipSnapshot) {
  Options options = CurrentOptions();

@@ -799,7 +756,7 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {
      ASSERT_TRUE(s.ok()) << s.ToString();
    };
    // Write num_files files in parallel
-    std::vector<std::thread> sst_writer_threads;
+    std::vector<port::Thread> sst_writer_threads;
    for (int i = 0; i < num_files; ++i) {
      sst_writer_threads.emplace_back(write_file_func);
    }
@@ -842,7 +799,7 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {
    };

    // Bulk load num_files files in parallel
-    std::vector<std::thread> add_file_threads;
+    std::vector<port::Thread> add_file_threads;
    DestroyAndReopen(options);
    for (int i = 0; i < num_files; ++i) {
      add_file_threads.emplace_back(load_file_func);
@@ -1086,13 +1043,13 @@ TEST_F(ExternalSSTFileTest, PickedLevelBug) {
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  // While writing the MANIFEST start a thread that will ask for compaction
-  std::thread bg_compact([&]() {
+  rocksdb::port::Thread bg_compact([&]() {
    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
  });
  TEST_SYNC_POINT("ExternalSSTFileTest::PickedLevelBug:2");

  // Start a thread that will ingest a new file
-  std::thread bg_addfile([&]() {
+  rocksdb::port::Thread bg_addfile([&]() {
    file_keys = {1, 2, 3};
    ASSERT_OK(GenerateAndAddExternalFile(options, file_keys, 1));
  });
@@ -1147,7 +1104,7 @@ TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
    ASSERT_OK(GenerateAndAddExternalFile(options, file_keys, range_id));
  };

-  std::vector<std::thread> threads;
+  std::vector<port::Thread> threads;
  while (range_id < 5000) {
    int range_start = range_id * 10;
    int range_end = range_start + 10;
@@ -1515,93 +1472,6 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
  VerifyDBFromMap(true_data, &kcnt, false);
 }

-TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoPickedSeqno) {
-  Options options = CurrentOptions();
-  DestroyAndReopen(options);
-  std::map<std::string, std::string> true_data;
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6}, -1, true,
-                                       false, &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13}, -1, true,
-                                       false, &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 4, 6}, -1, true, false,
-                                       &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {11, 15, 19}, -1, true, false,
-                                       &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {120, 130}, -1, true, false,
-                                       &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 130}, -1, true, false,
-                                       &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
-
-  // Write some keys through normal write path
-  for (int i = 0; i < 50; i++) {
-    ASSERT_OK(Put(Key(i), "memtable"));
-    true_data[Key(i)] = "memtable";
-  }
-  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {60, 61, 62}, -1, true, false,
-                                       &true_data));
-  // File dont overwrite any keys, No seqno needed
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {40, 41, 42}, -1, true, false,
-                                       &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {20, 30, 40}, -1, true, false,
-                                       &true_data));
-  // File overwrite some keys, a seqno will be assigned
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
-
-  const Snapshot* snapshot = db_->GetSnapshot();
-
-  // We will need a seqno for the file regardless if the file overwrite
-  // keys in the DB or not because we have a snapshot
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1000, 1002}, -1, true, false,
-                                       &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {2000, 3002}, -1, true, false,
-                                       &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150}, -1, true,
-                                       false, &true_data));
-  // A global seqno will be assigned anyway because of the snapshot
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
-
-  db_->ReleaseSnapshot(snapshot);
-
-  ASSERT_OK(GenerateAndAddExternalFile(options, {5000, 5001}, -1, true, false,
-                                       &true_data));
-  // No snapshot anymore, no need to assign a seqno
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
-
-  size_t kcnt = 0;
-  VerifyDBFromMap(true_data, &kcnt, false);
-}
-
 TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
  Options options = CurrentOptions();
  DestroyAndReopen(options);
@@ -1706,7 +1576,7 @@ TEST_F(ExternalSSTFileTest, CompactionDeadlock) {
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  // Start ingesting and extrnal file in the background
-  std::thread bg_ingest_file([&]() {
+  rocksdb::port::Thread bg_ingest_file([&]() {
    running_threads += 1;
    ASSERT_OK(GenerateAndAddExternalFile(options, {5, 6}));
    running_threads -= 1;
@@ -1726,7 +1596,7 @@ TEST_F(ExternalSSTFileTest, CompactionDeadlock) {

  // This thread will try to insert into the memtable but since we have 4 L0
  // files this thread will be blocked and hold the writer thread
-  std::thread bg_block_put([&]() {
+  rocksdb::port::Thread bg_block_put([&]() {
    running_threads += 1;
    ASSERT_OK(Put(Key(10), "memtable"));
    running_threads -= 1;
@@ -1738,8 +1608,6 @@ TEST_F(ExternalSSTFileTest, CompactionDeadlock) {
  // `DBImpl::AddFile:Start` will wait until we be here
  TEST_SYNC_POINT("ExternalSSTFileTest::DeadLock:1");

-  ASSERT_EQ(running_threads.load(), 2);
-
  // Wait for IngestExternalFile() to start and aquire mutex
  TEST_SYNC_POINT("ExternalSSTFileTest::DeadLock:2");

@@ -1780,8 +1648,150 @@ TEST_F(ExternalSSTFileTest, DirtyExit) {
  ASSERT_NOK(sst_file_writer->Finish());
 }

-#endif  // ROCKSDB_LITE
+TEST_F(ExternalSSTFileTest, FileWithCFInfo) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"koko", "toto"}, options);

+  SstFileWriter sfw_default(EnvOptions(), options, options.comparator,
+                            handles_[0]);
+  SstFileWriter sfw_cf1(EnvOptions(), options, options.comparator, handles_[1]);
+  SstFileWriter sfw_cf2(EnvOptions(), options, options.comparator, handles_[2]);
+  SstFileWriter sfw_unknown(EnvOptions(), options, options.comparator);
+
+  // default_cf.sst
+  const std::string cf_default_sst = sst_files_dir_ + "/default_cf.sst";
+  ASSERT_OK(sfw_default.Open(cf_default_sst));
+  ASSERT_OK(sfw_default.Add("K1", "V1"));
+  ASSERT_OK(sfw_default.Add("K2", "V2"));
+  ASSERT_OK(sfw_default.Finish());
+
+  // cf1.sst
+  const std::string cf1_sst = sst_files_dir_ + "/cf1.sst";
+  ASSERT_OK(sfw_cf1.Open(cf1_sst));
+  ASSERT_OK(sfw_cf1.Add("K3", "V1"));
+  ASSERT_OK(sfw_cf1.Add("K4", "V2"));
+  ASSERT_OK(sfw_cf1.Finish());
+
+  // cf_unknown.sst
+  const std::string unknown_sst = sst_files_dir_ + "/cf_unknown.sst";
+  ASSERT_OK(sfw_unknown.Open(unknown_sst));
+  ASSERT_OK(sfw_unknown.Add("K5", "V1"));
+  ASSERT_OK(sfw_unknown.Add("K6", "V2"));
+  ASSERT_OK(sfw_unknown.Finish());
+
+  IngestExternalFileOptions ifo;
+
+  // SST CF dont match
+  ASSERT_NOK(db_->IngestExternalFile(handles_[0], {cf1_sst}, ifo));
+  // SST CF dont match
+  ASSERT_NOK(db_->IngestExternalFile(handles_[2], {cf1_sst}, ifo));
+  // SST CF match
+  ASSERT_OK(db_->IngestExternalFile(handles_[1], {cf1_sst}, ifo));
+
+  // SST CF dont match
+  ASSERT_NOK(db_->IngestExternalFile(handles_[1], {cf_default_sst}, ifo));
+  // SST CF dont match
+  ASSERT_NOK(db_->IngestExternalFile(handles_[2], {cf_default_sst}, ifo));
+  // SST CF match
+  ASSERT_OK(db_->IngestExternalFile(handles_[0], {cf_default_sst}, ifo));
+
+  // SST CF unknown
+  ASSERT_OK(db_->IngestExternalFile(handles_[1], {unknown_sst}, ifo));
+  // SST CF unknown
+  ASSERT_OK(db_->IngestExternalFile(handles_[2], {unknown_sst}, ifo));
+  // SST CF unknown
+  ASSERT_OK(db_->IngestExternalFile(handles_[0], {unknown_sst}, ifo));
+
+  // Cannot ingest a file into a dropped CF
+  ASSERT_OK(db_->DropColumnFamily(handles_[1]));
+  ASSERT_NOK(db_->IngestExternalFile(handles_[1], {unknown_sst}, ifo));
+
+  // CF was not dropped, ok to Ingest
+  ASSERT_OK(db_->IngestExternalFile(handles_[2], {unknown_sst}, ifo));
+}
+
+class TestIngestExternalFileListener : public EventListener {
+ public:
+  void OnExternalFileIngested(DB* db,
+                              const ExternalFileIngestionInfo& info) override {
+    ingested_files.push_back(info);
+  }
+
+  std::vector<ExternalFileIngestionInfo> ingested_files;
+};
+
+TEST_F(ExternalSSTFileTest, IngestionListener) {
+  Options options = CurrentOptions();
+  TestIngestExternalFileListener* listener =
+      new TestIngestExternalFileListener();
+  options.listeners.emplace_back(listener);
+  CreateAndReopenWithCF({"koko", "toto"}, options);
+
+  // Ingest into default cf
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
+                                       handles_[0]));
+  ASSERT_EQ(listener->ingested_files.size(), 1);
+  ASSERT_EQ(listener->ingested_files.back().cf_name, "default");
+  ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_id,
+            0);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_name,
+            "default");
+
+  // Ingest into cf1
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
+                                       handles_[1]));
+  ASSERT_EQ(listener->ingested_files.size(), 2);
+  ASSERT_EQ(listener->ingested_files.back().cf_name, "koko");
+  ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_id,
+            1);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_name,
+            "koko");
+
+  // Ingest into cf2
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
+                                       handles_[2]));
+  ASSERT_EQ(listener->ingested_files.size(), 3);
+  ASSERT_EQ(listener->ingested_files.back().cf_name, "toto");
+  ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_id,
+            2);
+  ASSERT_EQ(listener->ingested_files.back().table_properties.column_family_name,
+            "toto");
+}
+
+TEST_F(ExternalSSTFileTest, SnapshotInconsistencyBug) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  const int kNumKeys = 10000;
+
+  // Insert keys using normal path and take a snapshot
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(Put(Key(i), Key(i) + "_V1"));
+  }
+  const Snapshot* snap = db_->GetSnapshot();
+
+  // Overwrite all keys using IngestExternalFile
+  std::string sst_file_path = sst_files_dir_ + "file1.sst";
+  SstFileWriter sst_file_writer(EnvOptions(), options, options.comparator);
+  ASSERT_OK(sst_file_writer.Open(sst_file_path));
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_OK(sst_file_writer.Add(Key(i), Key(i) + "_V2"));
+  }
+  ASSERT_OK(sst_file_writer.Finish());
+
+  IngestExternalFileOptions ifo;
+  ifo.move_files = true;
+  ASSERT_OK(db_->IngestExternalFile({sst_file_path}, ifo));
+
+  for (int i = 0; i < kNumKeys; i++) {
+    ASSERT_EQ(Get(Key(i), snap), Key(i) + "_V1");
+    ASSERT_EQ(Get(Key(i)), Key(i) + "_V2");
+  }
+
+  db_->ReleaseSnapshot(snap);
+}
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -1789,3 +1799,15 @@ int main(int argc, char** argv) {
  ::testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
 }
+
+#else
+#include <stdio.h>
+
+int main(int argc, char** argv) {
+  fprintf(stderr,
+          "SKIPPED as External SST File Writer and Ingestion are not supported "
+          "in ROCKSDB_LITE\n");
+  return 0;
+}
+
+#endif  // !ROCKSDB_LITE
@@ -9,8 +9,9 @@

 #include "db/file_indexer.h"
 #include <algorithm>
-#include "rocksdb/comparator.h"
+#include <functional>
 #include "db/version_edit.h"
+#include "rocksdb/comparator.h"

 namespace rocksdb {

@@ -389,12 +389,8 @@ Status SetIdentityFile(Env* env, const std::string& dbname) {
 Status SyncManifest(Env* env, const ImmutableDBOptions* db_options,
                    WritableFileWriter* file) {
  TEST_KILL_RANDOM("SyncManifest:0", rocksdb_kill_odds * REDUCE_ODDS2);
-  if (db_options->disable_data_sync) {
-    return Status::OK();
-  } else {
-    StopWatch sw(env, db_options->statistics.get(), MANIFEST_FILE_SYNC_MICROS);
-    return file->Sync(db_options->use_fsync);
-  }
+  StopWatch sw(env, db_options->statistics.get(), MANIFEST_FILE_SYNC_MICROS);
+  return file->Sync(db_options->use_fsync);
 }

 }  // namespace rocksdb
@@ -38,7 +38,7 @@
 #include "rocksdb/table.h"
 #include "table/block.h"
 #include "table/block_based_table_factory.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "table/table_builder.h"
 #include "table/two_level_iterator.h"
 #include "util/coding.h"
@@ -230,6 +230,12 @@ Status FlushJob::Run(FileMetaData* file_meta) {
  return s;
 }

+void FlushJob::Cancel() {
+  db_mutex_->AssertHeld();
+  assert(base_ != nullptr);
+  base_->Unref();
+}
+
 Status FlushJob::WriteLevel0Table() {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_FLUSH_WRITE_L0);
@@ -277,7 +283,8 @@ Status FlushJob::WriteLevel0Table() {
          NewMergingIterator(&cfd_->internal_comparator(), &memtables[0],
                             static_cast<int>(memtables.size()), &arena));
      std::unique_ptr<InternalIterator> range_del_iter(NewMergingIterator(
-          &cfd_->internal_comparator(), &range_del_iters[0],
+          &cfd_->internal_comparator(),
+          range_del_iters.empty() ? nullptr : &range_del_iters[0],
          static_cast<int>(range_del_iters.size())));
      Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
          "[%s] [JOB %d] Level-0 flush table #%" PRIu64 ": started",
@@ -306,7 +313,7 @@ Status FlushJob::WriteLevel0Table() {
        meta_.fd.GetFileSize(), s.ToString().c_str(),
        meta_.marked_for_compaction ? " (needs compaction)" : "");

-    if (!db_options_.disable_data_sync && output_file_directory_ != nullptr) {
+    if (output_file_directory_ != nullptr) {
      output_file_directory_->Fsync();
    }
    TEST_SYNC_POINT("FlushJob::WriteLevel0Table");
@@ -67,9 +67,11 @@ class FlushJob {

  ~FlushJob();

-  // Require db_mutex held
+  // Require db_mutex held.
+  // Once PickMemTable() is called, either Run() or Cancel() has to be call.
  void PickMemTable();
  Status Run(FileMetaData* file_meta = nullptr);
+  void Cancel();
  TableProperties GetTableProperties() const { return table_properties_; }

 private:
@@ -18,7 +18,7 @@
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/slice_transform.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "util/string_util.h"
 #include "util/sync_point.h"

@@ -69,12 +69,18 @@ class LevelIterator : public InternalIterator {
      delete file_iter_;
    }

+    RangeDelAggregator range_del_agg(
+        cfd_->internal_comparator(), {} /* snapshots */);
    file_iter_ = cfd_->table_cache()->NewIterator(
        read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
-        files_[file_index_]->fd, nullptr /* range_del_agg */,
+        files_[file_index_]->fd, read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
        nullptr /* table_reader_ptr */, nullptr, false);
-
    file_iter_->SetPinnedItersMgr(pinned_iters_mgr_);
+    if (!range_del_agg.IsEmpty()) {
+      status_ = Status::NotSupported(
+          "Range tombstones unsupported with ForwardIterator");
+      valid_ = false;
+    }
  }
  void SeekToLast() override {
    status_ = Status::NotSupported("LevelIterator::SeekToLast()");
@@ -558,8 +564,17 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
    // New
    sv_ = cfd_->GetReferencedSuperVersion(&(db_->mutex_));
  }
+  RangeDelAggregator range_del_agg(
+      InternalKeyComparator(cfd_->internal_comparator()), {} /* snapshots */);
  mutable_iter_ = sv_->mem->NewIterator(read_options_, &arena_);
  sv_->imm->AddIterators(read_options_, &imm_iters_, &arena_);
+  if (!read_options_.ignore_range_deletions) {
+    std::unique_ptr<InternalIterator> range_del_iter(
+        sv_->mem->NewRangeTombstoneIterator(read_options_));
+    range_del_agg.AddTombstones(std::move(range_del_iter));
+    sv_->imm->AddRangeTombstoneIterators(read_options_, &arena_,
+                                         &range_del_agg);
+  }
  has_iter_trimmed_for_upper_bound_ = false;

  const auto* vstorage = sv_->current->storage_info();
@@ -575,13 +590,18 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
    }
    l0_iters_.push_back(cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), l0->fd,
-        nullptr /* range_del_agg */));
+        read_options_.ignore_range_deletions ? nullptr : &range_del_agg));
  }
  BuildLevelIterators(vstorage);
  current_ = nullptr;
  is_prev_set_ = false;

  UpdateChildrenPinnedItersMgr();
+  if (!range_del_agg.IsEmpty()) {
+    status_ = Status::NotSupported(
+        "Range tombstones unsupported with ForwardIterator");
+    valid_ = false;
+  }
 }

 void ForwardIterator::RenewIterators() {
@@ -599,6 +619,15 @@ void ForwardIterator::RenewIterators() {

  mutable_iter_ = svnew->mem->NewIterator(read_options_, &arena_);
  svnew->imm->AddIterators(read_options_, &imm_iters_, &arena_);
+  RangeDelAggregator range_del_agg(
+      InternalKeyComparator(cfd_->internal_comparator()), {} /* snapshots */);
+  if (!read_options_.ignore_range_deletions) {
+    std::unique_ptr<InternalIterator> range_del_iter(
+        svnew->mem->NewRangeTombstoneIterator(read_options_));
+    range_del_agg.AddTombstones(std::move(range_del_iter));
+    sv_->imm->AddRangeTombstoneIterators(read_options_, &arena_,
+                                         &range_del_agg);
+  }

  const auto* vstorage = sv_->current->storage_info();
  const auto& l0_files = vstorage->LevelFiles(0);
@@ -630,7 +659,8 @@ void ForwardIterator::RenewIterators() {
    }
    l0_iters_new.push_back(cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files_new[inew]->fd, nullptr /* range_del_agg */));
+        l0_files_new[inew]->fd,
+        read_options_.ignore_range_deletions ? nullptr : &range_del_agg));
  }

  for (auto* f : l0_iters_) {
@@ -650,6 +680,11 @@ void ForwardIterator::RenewIterators() {
  sv_ = svnew;

  UpdateChildrenPinnedItersMgr();
+  if (!range_del_agg.IsEmpty()) {
+    status_ = Status::NotSupported(
+        "Range tombstones unsupported with ForwardIterator");
+    valid_ = false;
+  }
 }

 void ForwardIterator::BuildLevelIterators(const VersionStorageInfo* vstorage) {
@@ -34,6 +34,7 @@ int main() { return 0; }
 #include "rocksdb/db.h"
 #include "rocksdb/status.h"
 #include "rocksdb/table.h"
+#include "port/port.h"
 #include "util/testharness.h"

 const int MAX_SHARDS = 100000;
@@ -94,7 +95,7 @@ struct Reader {
  explicit Reader(std::vector<ShardState>* shard_states, rocksdb::DB* db)
      : shard_states_(shard_states), db_(db) {
    sem_init(&sem_, 0, 0);
-    thread_ = std::thread(&Reader::run, this);
+    thread_ = port::Thread(&Reader::run, this);
  }

  void run() {
@@ -193,7 +194,7 @@ struct Reader {
  char pad1[128] __attribute__((__unused__));
  std::vector<ShardState>* shard_states_;
  rocksdb::DB* db_;
-  std::thread thread_;
+  rocksdb::port::Thread thread_;
  sem_t sem_;
  std::mutex queue_mutex_;
  std::bitset<MAX_SHARDS + 1> shards_pending_set_;
@@ -206,7 +207,7 @@ struct Writer {
  explicit Writer(std::vector<ShardState>* shard_states, rocksdb::DB* db)
      : shard_states_(shard_states), db_(db) {}

-  void start() { thread_ = std::thread(&Writer::run, this); }
+  void start() { thread_ = port::Thread(&Writer::run, this); }

  void run() {
    std::queue<std::chrono::steady_clock::time_point> workq;
@@ -263,7 +264,7 @@ struct Writer {
  char pad1[128] __attribute__((__unused__));
  std::vector<ShardState>* shard_states_;
  rocksdb::DB* db_;
-  std::thread thread_;
+  rocksdb::port::Thread thread_;
  char pad2[128] __attribute__((__unused__));
 };

@@ -313,7 +314,7 @@ struct StatsThread {
  rocksdb::DB* db_;
  std::mutex cvm_;
  std::condition_variable cv_;
-  std::thread thread_;
+  rocksdb::port::Thread thread_;
  std::atomic<bool> done_{false};
 };

@@ -330,7 +331,7 @@ int main(int argc, char** argv) {
  options.compaction_style = rocksdb::CompactionStyle::kCompactionStyleNone;
  options.level0_slowdown_writes_trigger = 99999;
  options.level0_stop_writes_trigger = 99999;
-  options.allow_os_buffer = false;
+  options.use_direct_writes = true;
  options.write_buffer_size = FLAGS_memtable_size;
  rocksdb::BlockBasedTableOptions table_options;
  table_options.block_cache = rocksdb::NewLRUCache(FLAGS_block_cache_size);
@@ -221,6 +221,7 @@ static const std::string num_live_versions = "num-live-versions";
 static const std::string current_version_number =
    "current-super-version-number";
 static const std::string estimate_live_data_size = "estimate-live-data-size";
+static const std::string min_log_number_to_keep = "min-log-number-to-keep";
 static const std::string base_level = "base-level";
 static const std::string total_sst_files_size = "total-sst-files-size";
 static const std::string estimate_pending_comp_bytes =
@@ -231,6 +232,9 @@ static const std::string aggregated_table_properties_at_level =
    aggregated_table_properties + "-at-level";
 static const std::string num_running_compactions = "num-running-compactions";
 static const std::string num_running_flushes = "num-running-flushes";
+static const std::string actual_delayed_write_rate =
+    "actual-delayed-write-rate";
+static const std::string is_write_stopped = "is-write-stopped";

 const std::string DB::Properties::kNumFilesAtLevelPrefix =
                      rocksdb_prefix + num_files_at_level_prefix;
@@ -285,6 +289,8 @@ const std::string DB::Properties::kCurrentSuperVersionNumber =
    rocksdb_prefix + current_version_number;
 const std::string DB::Properties::kEstimateLiveDataSize =
                      rocksdb_prefix + estimate_live_data_size;
+const std::string DB::Properties::kMinLogNumberToKeep =
+    rocksdb_prefix + min_log_number_to_keep;
 const std::string DB::Properties::kTotalSstFilesSize =
                      rocksdb_prefix + total_sst_files_size;
 const std::string DB::Properties::kBaseLevel = rocksdb_prefix + base_level;
@@ -294,6 +300,10 @@ const std::string DB::Properties::kAggregatedTableProperties =
    rocksdb_prefix + aggregated_table_properties;
 const std::string DB::Properties::kAggregatedTablePropertiesAtLevel =
    rocksdb_prefix + aggregated_table_properties_at_level;
+const std::string DB::Properties::kActualDelayedWriteRate =
+    rocksdb_prefix + actual_delayed_write_rate;
+const std::string DB::Properties::kIsWriteStopped =
+    rocksdb_prefix + is_write_stopped;

 const std::unordered_map<std::string, DBPropertyInfo>
    InternalStats::ppt_name_to_info = {
@@ -368,6 +378,8 @@ const std::unordered_map<std::string, DBPropertyInfo>
          nullptr}},
        {DB::Properties::kEstimateLiveDataSize,
         {true, nullptr, &InternalStats::HandleEstimateLiveDataSize, nullptr}},
+        {DB::Properties::kMinLogNumberToKeep,
+         {false, nullptr, &InternalStats::HandleMinLogNumberToKeep, nullptr}},
        {DB::Properties::kBaseLevel,
         {false, nullptr, &InternalStats::HandleBaseLevel, nullptr}},
        {DB::Properties::kTotalSstFilesSize,
@@ -380,6 +392,11 @@ const std::unordered_map<std::string, DBPropertyInfo>
        {DB::Properties::kNumRunningCompactions,
         {false, nullptr, &InternalStats::HandleNumRunningCompactions,
          nullptr}},
+        {DB::Properties::kActualDelayedWriteRate,
+         {false, nullptr, &InternalStats::HandleActualDelayedWriteRate,
+          nullptr}},
+        {DB::Properties::kIsWriteStopped,
+         {false, nullptr, &InternalStats::HandleIsWriteStopped, nullptr}},
 };

 const DBPropertyInfo* GetPropertyInfo(const Slice& property) {
@@ -705,6 +722,29 @@ bool InternalStats::HandleEstimateLiveDataSize(uint64_t* value, DBImpl* db,
  return true;
 }

+bool InternalStats::HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db,
+                                             Version* version) {
+  *value = db->MinLogNumberToKeep();
+  return true;
+}
+
+bool InternalStats::HandleActualDelayedWriteRate(uint64_t* value, DBImpl* db,
+                                                 Version* version) {
+  const WriteController& wc = db->write_controller();
+  if (!wc.NeedsDelay()) {
+    *value = 0;
+  } else {
+    *value = wc.delayed_write_rate();
+  }
+  return true;
+}
+
+bool InternalStats::HandleIsWriteStopped(uint64_t* value, DBImpl* db,
+                                         Version* version) {
+  *value = db->write_controller().IsStopped() ? 1 : 0;
+  return true;
+}
+
 void InternalStats::DumpDBStats(std::string* value) {
  char buf[1000];
  // DB-level stats, only available from default column family
@@ -854,7 +894,7 @@ void InternalStats::DumpCFMapStats(std::map<std::string, double>* cf_stats) {
  }
 }

-int InternalStats::DumpCFMapStats(
+void InternalStats::DumpCFMapStats(
    std::map<int, std::map<LevelStatType, double>>* levels_stats,
    CompactionStats* compaction_stats_sum) {
  const VersionStorageInfo* vstorage = cfd_->current()->storage_info();
@@ -914,11 +954,10 @@ int InternalStats::DumpCFMapStats(
  PrepareLevelStats(&sum_stats, total_files, total_files_being_compacted,
                    total_file_size, 0, w_amp, *compaction_stats_sum);
  (*levels_stats)[-1] = sum_stats;  //  -1 is for the Sum level
-  return num_levels_to_check;
 }

 void InternalStats::DumpCFStats(std::string* value) {
-  char buf[1000];
+  char buf[2000];
  // Per-ColumnFamily stats
  PrintLevelStatsHeader(buf, sizeof(buf), cfd_->GetName());
  value->append(buf);
@@ -926,8 +965,8 @@ void InternalStats::DumpCFStats(std::string* value) {
  // Print stats for each level
  std::map<int, std::map<LevelStatType, double>> levels_stats;
  CompactionStats compaction_stats_sum(0);
-  int levels = DumpCFMapStats(&levels_stats, &compaction_stats_sum);
-  for (int l = 0; l < levels; ++l) {
+  DumpCFMapStats(&levels_stats, &compaction_stats_sum);
+  for (int l = 0; l < number_levels_; ++l) {
    if (levels_stats.find(l) != levels_stats.end()) {
      PrintLevelStats(buf, sizeof(buf), "L" + ToString(l), levels_stats[l]);
      value->append(buf);
@@ -273,7 +273,7 @@ class InternalStats {
 private:
  void DumpDBStats(std::string* value);
  void DumpCFMapStats(std::map<std::string, double>* cf_stats);
-  int DumpCFMapStats(
+  void DumpCFMapStats(
      std::map<int, std::map<LevelStatType, double>>* level_stats,
      CompactionStats* compaction_stats_sum);
  void DumpCFStats(std::string* value);
@@ -401,6 +401,10 @@ class InternalStats {
                                     Version* version);
  bool HandleEstimateLiveDataSize(uint64_t* value, DBImpl* db,
                                  Version* version);
+  bool HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db, Version* version);
+  bool HandleActualDelayedWriteRate(uint64_t* value, DBImpl* db,
+                                    Version* version);
+  bool HandleIsWriteStopped(uint64_t* value, DBImpl* db, Version* version);

  // Total number of background errors encountered. Every time a flush task
  // or compaction task fails, this counter is incremented. The failure can
@@ -119,7 +119,7 @@ TEST_F(EventListenerTest, OnSingleDBCompactionTest) {
  options.max_bytes_for_level_base = options.target_file_size_base * 2;
  options.max_bytes_for_level_multiplier = 2;
  options.compression = kNoCompression;
-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
  options.enable_thread_tracking = true;
 #endif  // ROCKSDB_USING_THREAD_STATUS
  options.level0_file_num_compaction_trigger = kNumL0Files;
@@ -176,7 +176,7 @@ class TestFlushListener : public EventListener {
    ASSERT_GT(info.table_properties.num_data_blocks, 0U);
    ASSERT_GT(info.table_properties.num_entries, 0U);

-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
    // Verify the id of the current thread that created this table
    // file matches the id of any active flush or compaction thread.
    uint64_t thread_id = env_->GetThreadID();
@@ -232,7 +232,7 @@ class TestFlushListener : public EventListener {
 TEST_F(EventListenerTest, OnSingleDBFlushTest) {
  Options options;
  options.write_buffer_size = k110KB;
-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
  options.enable_thread_tracking = true;
 #endif  // ROCKSDB_USING_THREAD_STATUS
  TestFlushListener* listener = new TestFlushListener(options.env);
@@ -268,7 +268,7 @@ TEST_F(EventListenerTest, OnSingleDBFlushTest) {
 TEST_F(EventListenerTest, MultiCF) {
  Options options;
  options.write_buffer_size = k110KB;
-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
  options.enable_thread_tracking = true;
 #endif  // ROCKSDB_USING_THREAD_STATUS
  TestFlushListener* listener = new TestFlushListener(options.env);
@@ -302,7 +302,7 @@ TEST_F(EventListenerTest, MultiCF) {

 TEST_F(EventListenerTest, MultiDBMultiListeners) {
  Options options;
-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
  options.enable_thread_tracking = true;
 #endif  // ROCKSDB_USING_THREAD_STATUS
  options.table_properties_collector_factories.push_back(
@@ -384,7 +384,7 @@ TEST_F(EventListenerTest, MultiDBMultiListeners) {

 TEST_F(EventListenerTest, DisableBGCompaction) {
  Options options;
-#if ROCKSDB_USING_THREAD_STATUS
+#ifdef ROCKSDB_USING_THREAD_STATUS
  options.enable_thread_tracking = true;
 #endif  // ROCKSDB_USING_THREAD_STATUS
  TestFlushListener* listener = new TestFlushListener(options.env);
@@ -15,12 +15,10 @@
 #include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/dbformat.h"
-#include "db/xfunc_test_points.h"
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/slice_transform.h"
-#include "table/merger.h"
-#include "util/xfunc.h"
+#include "table/merging_iterator.h"

 namespace rocksdb {

@@ -42,8 +40,6 @@ class MILock {
  }
  ~MILock() {
    this->mu_->unlock();
-    XFUNC_TEST("managed_xftest_release", "managed_unlock", managed_unlock1,
-               xf_manage_release, mi_);
  }
  ManagedIterator* GetManagedIterator() { return mi_; }

@@ -84,8 +80,6 @@ ManagedIterator::ManagedIterator(DBImpl* db, const ReadOptions& read_options,
  }
  cfh_.SetCFD(cfd);
  mutable_iter_ = unique_ptr<Iterator>(db->NewIterator(read_options_, &cfh_));
-  XFUNC_TEST("managed_xftest_dropold", "managed_create", xf_managed_create1,
-             xf_manage_create, this);
 }

 ManagedIterator::~ManagedIterator() {
@@ -261,8 +255,6 @@ bool ManagedIterator::TryLock() { return in_use_.try_lock(); }

 void ManagedIterator::UnLock() {
  in_use_.unlock();
-  XFUNC_TEST("managed_xftest_release", "managed_unlock", managed_unlock1,
-             xf_manage_release, this);
 }

 }  // namespace rocksdb
@@ -26,7 +26,7 @@
 #include "rocksdb/write_buffer_manager.h"
 #include "table/internal_iterator.h"
 #include "table/iterator_wrapper.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "util/arena.h"
 #include "util/autovector.h"
 #include "util/coding.h"
@@ -390,16 +390,16 @@ port::RWMutex* MemTable::GetLock(const Slice& key) {
  return &locks_[hash(key) % locks_.size()];
 }

-uint64_t MemTable::ApproximateSize(const Slice& start_ikey,
-                                   const Slice& end_ikey) {
+MemTable::MemTableStats MemTable::ApproximateStats(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;
+    return {0, 0};
  }
  uint64_t n = num_entries_.load(std::memory_order_relaxed);
  if (n == 0) {
-    return 0;
+    return {0, 0};
  }
  if (entry_count > n) {
    // (range_del_)table_->ApproximateNumEntries() is just an estimate so it can
@@ -408,7 +408,7 @@ uint64_t MemTable::ApproximateSize(const Slice& start_ikey,
    entry_count = n;
  }
  uint64_t data_size = data_size_.load(std::memory_order_relaxed);
-  return entry_count * (data_size / n);
+  return {entry_count * (data_size / n), entry_count};
 }

 void MemTable::Add(SequenceNumber s, ValueType type,
@@ -560,7 +560,7 @@ static bool SaveValue(void* arg, const char* entry) {
    ValueType type;
    UnPackSequenceAndType(tag, &s->seq, &type);

-    if ((type == kTypeValue || type == kTypeDeletion) &&
+    if ((type == kTypeValue || type == kTypeMerge) &&
        range_del_agg->ShouldDelete(Slice(key_ptr, key_length))) {
      type = kTypeRangeDeletion;
    }
@@ -717,29 +717,22 @@ void MemTable::Update(SequenceNumber seq,
      ValueType type;
      SequenceNumber unused;
      UnPackSequenceAndType(tag, &unused, &type);
-      switch (type) {
-        case kTypeValue: {
-          Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
-          uint32_t prev_size = static_cast<uint32_t>(prev_value.size());
-          uint32_t new_size = static_cast<uint32_t>(value.size());
+      if (type == kTypeValue) {
+        Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
+        uint32_t prev_size = static_cast<uint32_t>(prev_value.size());
+        uint32_t new_size = static_cast<uint32_t>(value.size());

-          // Update value, if new value size  <= previous value size
-          if (new_size <= prev_size ) {
-            char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
-                                     new_size);
-            WriteLock wl(GetLock(lkey.user_key()));
-            memcpy(p, value.data(), value.size());
-            assert((unsigned)((p + value.size()) - entry) ==
-                   (unsigned)(VarintLength(key_length) + key_length +
-                              VarintLength(value.size()) + value.size()));
-            return;
-          }
+        // Update value, if new value size  <= previous value size
+        if (new_size <= prev_size ) {
+          char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
+                                   new_size);
+          WriteLock wl(GetLock(lkey.user_key()));
+          memcpy(p, value.data(), value.size());
+          assert((unsigned)((p + value.size()) - entry) ==
+                 (unsigned)(VarintLength(key_length) + key_length +
+                            VarintLength(value.size()) + value.size()));
+          return;
        }
-        default:
-          // If the latest value is kTypeDeletion, kTypeMerge or kTypeLogData
-          // we don't have enough space for update inplace
-            Add(seq, kTypeValue, key, value);
-            return;
      }
    }
  }
@@ -326,7 +326,13 @@ class MemTable {
    return table_->IsSnapshotSupported() && !moptions_.inplace_update_support;
  }

-  uint64_t ApproximateSize(const Slice& start_ikey, const Slice& end_ikey);
+  struct MemTableStats {
+    uint64_t size;
+    uint64_t count;
+  };
+
+  MemTableStats ApproximateStats(const Slice& start_ikey,
+                                 const Slice& end_ikey);

  // Get the lock associated for the key
  port::RWMutex* GetLock(const Slice& key);
@@ -16,7 +16,7 @@
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "util/coding.h"
 #include "util/log_buffer.h"
 #include "util/sync_point.h"
@@ -190,13 +190,15 @@ uint64_t MemTableListVersion::GetTotalNumEntries() const {
  return total_num;
 }

-uint64_t MemTableListVersion::ApproximateSize(const Slice& start_ikey,
-                                              const Slice& end_ikey) {
-  uint64_t total_size = 0;
+MemTable::MemTableStats MemTableListVersion::ApproximateStats(
+    const Slice& start_ikey, const Slice& end_ikey) {
+  MemTable::MemTableStats total_stats = {0, 0};
  for (auto& m : memlist_) {
-    total_size += m->ApproximateSize(start_ikey, end_ikey);
+    auto mStats = m->ApproximateStats(start_ikey, end_ikey);
+    total_stats.size += mStats.size;
+    total_stats.count += mStats.count;
  }
-  return total_size;
+  return total_stats;
 }

 uint64_t MemTableListVersion::GetTotalNumDeletes() const {
@@ -95,7 +95,8 @@ class MemTableListVersion {

  uint64_t GetTotalNumDeletes() const;

-  uint64_t ApproximateSize(const Slice& start_ikey, const Slice& end_ikey);
+  MemTable::MemTableStats ApproximateStats(const Slice& start_ikey,
+                                           const Slice& end_ikey);

  // Returns the value of MemTable::GetEarliestSequenceNumber() on the most
  // recent MemTable in this list or kMaxSequenceNumber if the list is empty.
@@ -221,6 +222,8 @@ class MemTableList {
  // PickMemtablesToFlush() is called.
  void FlushRequested() { flush_requested_ = true; }

+  bool HasFlushRequested() { return flush_requested_; }
+
  // Copying allowed
  // MemTableList(const MemTableList&);
  // void operator=(const MemTableList&);
@@ -426,7 +426,7 @@ class Benchmark {
  virtual ~Benchmark() {}
  virtual void Run() {
    std::cout << "Number of threads: " << num_threads_ << std::endl;
-    std::vector<std::thread> threads;
+    std::vector<port::Thread> threads;
    uint64_t bytes_written = 0;
    uint64_t bytes_read = 0;
    uint64_t read_hits = 0;
@@ -457,7 +457,7 @@ class Benchmark {
    }
  }

-  virtual void RunThreads(std::vector<std::thread>* threads,
+  virtual void RunThreads(std::vector<port::Thread>* threads,
                          uint64_t* bytes_written, uint64_t* bytes_read,
                          bool write, uint64_t* read_hits) = 0;

@@ -478,7 +478,7 @@ class FillBenchmark : public Benchmark {
    num_write_ops_per_thread_ = FLAGS_num_operations;
  }

-  void RunThreads(std::vector<std::thread>* threads, uint64_t* bytes_written,
+  void RunThreads(std::vector<port::Thread>* threads, uint64_t* bytes_written,
                  uint64_t* bytes_read, bool write,
                  uint64_t* read_hits) override {
    FillBenchmarkThread(table_, key_gen_, bytes_written, bytes_read, sequence_,
@@ -494,7 +494,7 @@ class ReadBenchmark : public Benchmark {
    num_read_ops_per_thread_ = FLAGS_num_operations / FLAGS_num_threads;
  }

-  void RunThreads(std::vector<std::thread>* threads, uint64_t* bytes_written,
+  void RunThreads(std::vector<port::Thread>* threads, uint64_t* bytes_written,
                  uint64_t* bytes_read, bool write,
                  uint64_t* read_hits) override {
    for (int i = 0; i < FLAGS_num_threads; ++i) {
@@ -518,7 +518,7 @@ class SeqReadBenchmark : public Benchmark {
    num_read_ops_per_thread_ = FLAGS_num_scans;
  }

-  void RunThreads(std::vector<std::thread>* threads, uint64_t* bytes_written,
+  void RunThreads(std::vector<port::Thread>* threads, uint64_t* bytes_written,
                  uint64_t* bytes_read, bool write,
                  uint64_t* read_hits) override {
    for (int i = 0; i < FLAGS_num_threads; ++i) {
@@ -545,7 +545,7 @@ class ReadWriteBenchmark : public Benchmark {
    num_write_ops_per_thread_ = FLAGS_num_operations;
  }

-  void RunThreads(std::vector<std::thread>* threads, uint64_t* bytes_written,
+  void RunThreads(std::vector<port::Thread>* threads, uint64_t* bytes_written,
                  uint64_t* bytes_read, bool write,
                  uint64_t* read_hits) override {
    std::atomic_int threads_done;
@@ -104,6 +104,10 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
  Status s;
  bool hit_the_next_user_key = false;
  for (; iter->Valid(); iter->Next(), original_key_is_iter = false) {
+    if (IsShuttingDown()) {
+      return Status::ShutdownInProgress();
+    }
+
    ParsedInternalKey ikey;
    assert(keys_.size() == merge_context_.GetNumOperands());

@@ -192,7 +196,10 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
              ? CompactionFilter::Decision::kKeep
              : FilterMerge(orig_ikey.user_key, value_slice);
      if (range_del_agg != nullptr &&
-          range_del_agg->ShouldDelete(iter->key()) &&
+
+          range_del_agg->ShouldDelete(
+              iter->key(),
+              RangeDelAggregator::RangePositioningMode::kForwardTraversal) &&
          filter != CompactionFilter::Decision::kRemoveAndSkipUntil) {
        filter = CompactionFilter::Decision::kRemove;
      }
@@ -275,13 +282,8 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
    // We haven't seen the beginning of the key nor a Put/Delete.
    // Attempt to use the user's associative merge function to
    // merge the stacked merge operands into a single operand.
-    //
-    // TODO(noetzli) The docblock of MergeUntil suggests that a successful
-    // partial merge returns Status::OK(). Should we change the status code
-    // after a successful partial merge?
    s = Status::MergeInProgress();
-    if (merge_context_.GetNumOperands() >= 2 &&
-        merge_context_.GetNumOperands() >= min_partial_merge_operands_) {
+    if (merge_context_.GetNumOperands() >= 2) {
      bool merge_success = false;
      std::string merge_result;
      {
@@ -32,15 +32,15 @@ class MergeHelper {
  MergeHelper(Env* env, const Comparator* user_comparator,
              const MergeOperator* user_merge_operator,
              const CompactionFilter* compaction_filter, Logger* logger,
-              unsigned min_partial_merge_operands,
              bool assert_valid_internal_key, SequenceNumber latest_snapshot,
-              int level = 0, Statistics* stats = nullptr)
+              int level = 0, Statistics* stats = nullptr,
+              const std::atomic<bool>* shutting_down = nullptr)
      : env_(env),
        user_comparator_(user_comparator),
        user_merge_operator_(user_merge_operator),
        compaction_filter_(compaction_filter),
+        shutting_down_(shutting_down),
        logger_(logger),
-        min_partial_merge_operands_(min_partial_merge_operands),
        assert_valid_internal_key_(assert_valid_internal_key),
        latest_snapshot_(latest_snapshot),
        level_(level),
@@ -81,10 +81,12 @@ class MergeHelper {
  //
  // Returns one of the following statuses:
  // - OK: Entries were successfully merged.
-  // - MergeInProgress: Put/Delete not encountered and unable to merge operands.
+  // - MergeInProgress: Put/Delete not encountered, and didn't reach the start
+  //   of key's history. Output consists of merge operands only.
  // - Corruption: Merge operator reported unsuccessful merge or a corrupted
  //   key has been encountered and not expected (applies only when compiling
  //   with asserts removed).
+  // - ShutdownInProgress: interrupted by shutdown (*shutting_down == true).
  //
  // REQUIRED: The first key in the input is not corrupted.
  Status MergeUntil(InternalIterator* iter,
@@ -150,8 +152,8 @@ class MergeHelper {
  const Comparator* user_comparator_;
  const MergeOperator* user_merge_operator_;
  const CompactionFilter* compaction_filter_;
+  const std::atomic<bool>* shutting_down_;
  Logger* logger_;
-  unsigned min_partial_merge_operands_;
  bool assert_valid_internal_key_; // enforce no internal key corruption?
  SequenceNumber latest_snapshot_;
  int level_;
@@ -171,6 +173,11 @@ class MergeHelper {
  bool has_compaction_filter_skip_until_ = false;
  std::string compaction_filter_value_;
  InternalKey compaction_filter_skip_until_;
+
+  bool IsShuttingDown() {
+    // This is a best-effort facility, so memory_order_relaxed is sufficient.
+    return shutting_down_ && shutting_down_->load(std::memory_order_relaxed);
+  }
 };

 // MergeOutputIterator can be used to iterate over the result of a merge.
@@ -28,7 +28,7 @@ class MergeHelperTest : public testing::Test {
    iter_->SeekToFirst();
    merge_helper_.reset(new MergeHelper(env_, BytewiseComparator(),
                                        merge_op_.get(), filter_.get(), nullptr,
-                                        2U, false, latest_snapshot));
+                                        false, latest_snapshot));
    return merge_helper_->MergeUntil(iter_.get(), nullptr /* range_del_agg */,
                                     stop_before, at_bottom);
  }
@@ -76,14 +76,12 @@ class CountMergeOperator : public AssociativeMergeOperator {

 namespace {
 std::shared_ptr<DB> OpenDb(const std::string& dbname, const bool ttl = false,
-                           const size_t max_successive_merges = 0,
-                           const uint32_t min_partial_merge_operands = 2) {
+                           const size_t max_successive_merges = 0) {
  DB* db;
  Options options;
  options.create_if_missing = true;
  options.merge_operator = std::make_shared<CountMergeOperator>();
  options.max_successive_merges = max_successive_merges;
-  options.min_partial_merge_operands = min_partial_merge_operands;
  Status s;
  DestroyDB(dbname, Options());
 // DBWithTTL is not supported in ROCKSDB_LITE
@@ -448,20 +446,20 @@ void runTest(int argc, const std::string& dbname, const bool use_ttl = false) {
  {
    std::cout << "Test Partial-Merge\n";
    size_t max_merge = 100;
-    for (uint32_t min_merge = 5; min_merge < 25; min_merge += 5) {
-      for (uint32_t count = min_merge - 1; count <= min_merge + 1; count++) {
-        auto db = OpenDb(dbname, use_ttl, max_merge, min_merge);
-        MergeBasedCounters counters(db, 0);
-        testPartialMerge(&counters, db.get(), max_merge, min_merge, count);
-        DestroyDB(dbname, Options());
-      }
-      {
-        auto db = OpenDb(dbname, use_ttl, max_merge, min_merge);
-        MergeBasedCounters counters(db, 0);
-        testPartialMerge(&counters, db.get(), max_merge, min_merge,
-                         min_merge * 10);
-        DestroyDB(dbname, Options());
-      }
+    // Min merge is hard-coded to 2.
+    uint32_t min_merge = 2;
+    for (uint32_t count = min_merge - 1; count <= min_merge + 1; count++) {
+      auto db = OpenDb(dbname, use_ttl, max_merge);
+      MergeBasedCounters counters(db, 0);
+      testPartialMerge(&counters, db.get(), max_merge, min_merge, count);
+      DestroyDB(dbname, Options());
+    }
+    {
+      auto db = OpenDb(dbname, use_ttl, max_merge);
+      MergeBasedCounters counters(db, 0);
+      testPartialMerge(&counters, db.get(), max_merge, min_merge,
+                       min_merge * 10);
+      DestroyDB(dbname, Options());
    }
  }

@@ -12,6 +12,7 @@
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/perf_context.h"
 #include "rocksdb/slice_transform.h"
+#include "port/port.h"
 #include "util/histogram.h"
 #include "util/instrumented_mutex.h"
 #include "util/stop_watch.h"
@@ -561,7 +562,7 @@ TEST_F(PerfContextTest, DBMutexLockCounter) {
    for (int c = 0; c < 2; ++c) {
    InstrumentedMutex mutex(nullptr, Env::Default(), stats_code[c]);
    mutex.Lock();
-    std::thread child_thread([&] {
+    rocksdb::port::Thread child_thread([&] {
      SetPerfLevel(perf_level);
      perf_context.Reset();
      ASSERT_EQ(perf_context.db_mutex_lock_nanos, 0);
@@ -16,7 +16,7 @@ namespace rocksdb {
 // PinnedIteratorsManager will be notified whenever we need to pin an Iterator
 // and it will be responsible for deleting pinned Iterators when they are
 // not needed anymore.
-class PinnedIteratorsManager {
+class PinnedIteratorsManager : public Cleanable {
 public:
  PinnedIteratorsManager() : pinning_enabled(false) {}
  ~PinnedIteratorsManager() {
@@ -59,16 +59,16 @@ class PinnedIteratorsManager {

    // Remove duplicate pointers
    std::sort(pinned_ptrs_.begin(), pinned_ptrs_.end());
-    std::unique(pinned_ptrs_.begin(), pinned_ptrs_.end());
+    auto unique_end = std::unique(pinned_ptrs_.begin(), pinned_ptrs_.end());

-    for (size_t i = 0; i < pinned_ptrs_.size(); i++) {
-      assert(i == 0 || pinned_ptrs_[i].first != pinned_ptrs_[i - 1].first);
-
-      void* ptr = pinned_ptrs_[i].first;
-      ReleaseFunction release_func = pinned_ptrs_[i].second;
+    for (auto i = pinned_ptrs_.begin(); i != unique_end; ++i) {
+      void* ptr = i->first;
+      ReleaseFunction release_func = i->second;
      release_func(ptr);
    }
    pinned_ptrs_.clear();
+    // Also do cleanups from the base Cleanable
+    Cleanable::Reset();
  }

 private:
@@ -11,29 +11,40 @@ namespace rocksdb {

 RangeDelAggregator::RangeDelAggregator(
    const InternalKeyComparator& icmp,
-    const std::vector<SequenceNumber>& snapshots)
-    : upper_bound_(kMaxSequenceNumber), icmp_(icmp) {
+    const std::vector<SequenceNumber>& snapshots,
+    bool collapse_deletions /* = true */)
+    : upper_bound_(kMaxSequenceNumber),
+      icmp_(icmp),
+      collapse_deletions_(collapse_deletions) {
  InitRep(snapshots);
 }

 RangeDelAggregator::RangeDelAggregator(const InternalKeyComparator& icmp,
-                                       SequenceNumber snapshot)
-    : upper_bound_(snapshot), icmp_(icmp) {}
+                                       SequenceNumber snapshot,
+                                       bool collapse_deletions /* = false */)
+    : upper_bound_(snapshot),
+      icmp_(icmp),
+      collapse_deletions_(collapse_deletions) {}

 void RangeDelAggregator::InitRep(const std::vector<SequenceNumber>& snapshots) {
  assert(rep_ == nullptr);
  rep_.reset(new Rep());
  for (auto snapshot : snapshots) {
    rep_->stripe_map_.emplace(
-        snapshot, TombstoneMap(stl_wrappers::LessOfComparator(&icmp_)));
+        snapshot,
+        PositionalTombstoneMap(TombstoneMap(
+            stl_wrappers::LessOfComparator(icmp_.user_comparator()))));
  }
  // Data newer than any snapshot falls in this catch-all stripe
  rep_->stripe_map_.emplace(
-      kMaxSequenceNumber, TombstoneMap(stl_wrappers::LessOfComparator(&icmp_)));
+      kMaxSequenceNumber,
+      PositionalTombstoneMap(TombstoneMap(
+          stl_wrappers::LessOfComparator(icmp_.user_comparator()))));
  rep_->pinned_iters_mgr_.StartPinning();
 }

-bool RangeDelAggregator::ShouldDelete(const Slice& internal_key) {
+bool RangeDelAggregator::ShouldDelete(
+    const Slice& internal_key, RangeDelAggregator::RangePositioningMode mode) {
  if (rep_ == nullptr) {
    return false;
  }
@@ -41,32 +52,103 @@ bool RangeDelAggregator::ShouldDelete(const Slice& internal_key) {
  if (!ParseInternalKey(internal_key, &parsed)) {
    assert(false);
  }
-  return ShouldDelete(parsed);
+  return ShouldDelete(parsed, mode);
 }

-bool RangeDelAggregator::ShouldDelete(const ParsedInternalKey& parsed) {
+bool RangeDelAggregator::ShouldDelete(
+    const ParsedInternalKey& parsed,
+    RangeDelAggregator::RangePositioningMode mode) {
  assert(IsValueType(parsed.type));
  if (rep_ == nullptr) {
    return false;
  }
-  const auto& tombstone_map = GetTombstoneMap(parsed.sequence);
-  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;
-    }
+  auto& positional_tombstone_map = GetPositionalTombstoneMap(parsed.sequence);
+  const auto& tombstone_map = positional_tombstone_map.raw_map;
+  if (tombstone_map.empty()) {
+    return false;
  }
-  return false;
+  auto& tombstone_map_iter = positional_tombstone_map.iter;
+  if (tombstone_map_iter == tombstone_map.end() &&
+      (mode == kForwardTraversal || mode == kBackwardTraversal)) {
+    // invalid (e.g., if AddTombstones() changed the deletions), so need to
+    // reseek
+    mode = kBinarySearch;
+  }
+  switch (mode) {
+    case kFullScan:
+      assert(!collapse_deletions_);
+      // The maintained state (PositionalTombstoneMap::iter) isn't useful when
+      // we linear scan from the beginning each time, but we maintain it anyways
+      // for consistency.
+      tombstone_map_iter = tombstone_map.begin();
+      while (tombstone_map_iter != tombstone_map.end()) {
+        const auto& tombstone = tombstone_map_iter->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;
+        }
+        ++tombstone_map_iter;
+      }
+      return false;
+    case kForwardTraversal:
+      assert(collapse_deletions_ && tombstone_map_iter != tombstone_map.end());
+      if (tombstone_map_iter == tombstone_map.begin() &&
+          icmp_.user_comparator()->Compare(parsed.user_key,
+                                           tombstone_map_iter->first) < 0) {
+        // before start of deletion intervals
+        return false;
+      }
+      while (std::next(tombstone_map_iter) != tombstone_map.end() &&
+             icmp_.user_comparator()->Compare(
+                 std::next(tombstone_map_iter)->first, parsed.user_key) <= 0) {
+        ++tombstone_map_iter;
+      }
+      break;
+    case kBackwardTraversal:
+      assert(collapse_deletions_ && tombstone_map_iter != tombstone_map.end());
+      while (tombstone_map_iter != tombstone_map.begin() &&
+             icmp_.user_comparator()->Compare(parsed.user_key,
+                                              tombstone_map_iter->first) < 0) {
+        --tombstone_map_iter;
+      }
+      if (tombstone_map_iter == tombstone_map.begin() &&
+          icmp_.user_comparator()->Compare(parsed.user_key,
+                                           tombstone_map_iter->first) < 0) {
+        // before start of deletion intervals
+        return false;
+      }
+      break;
+    case kBinarySearch:
+      assert(collapse_deletions_);
+      tombstone_map_iter =
+          tombstone_map.upper_bound(parsed.user_key);
+      if (tombstone_map_iter == tombstone_map.begin()) {
+        // before start of deletion intervals
+        return false;
+      }
+      --tombstone_map_iter;
+      break;
+  }
+  assert(mode != kFullScan);
+  assert(tombstone_map_iter != tombstone_map.end() &&
+         icmp_.user_comparator()->Compare(tombstone_map_iter->first,
+                                          parsed.user_key) <= 0);
+  assert(std::next(tombstone_map_iter) == tombstone_map.end() ||
+         icmp_.user_comparator()->Compare(
+             parsed.user_key, std::next(tombstone_map_iter)->first) < 0);
+  return parsed.sequence < tombstone_map_iter->second.seq_;
 }

 bool RangeDelAggregator::ShouldAddTombstones(
    bool bottommost_level /* = false */) {
+  // TODO(andrewkr): can we just open a file and throw it away if it ends up
+  // empty after AddToBuilder()? This function doesn't take into subcompaction
+  // boundaries so isn't completely accurate.
  if (rep_ == nullptr) {
    return false;
  }
@@ -78,7 +160,7 @@ bool RangeDelAggregator::ShouldAddTombstones(
    ++stripe_map_iter;
  }
  while (stripe_map_iter != rep_->stripe_map_.end()) {
-    if (!stripe_map_iter->second.empty()) {
+    if (!stripe_map_iter->second.raw_map.empty()) {
      return true;
    }
    ++stripe_map_iter;
@@ -97,6 +179,8 @@ Status RangeDelAggregator::AddTombstones(
    if (first_iter) {
      if (rep_ == nullptr) {
        InitRep({upper_bound_});
+      } else {
+        InvalidateTombstoneMapPositions();
      }
      first_iter = false;
    }
@@ -105,8 +189,7 @@ Status RangeDelAggregator::AddTombstones(
      return Status::Corruption("Unable to parse range tombstone InternalKey");
    }
    RangeTombstone tombstone(parsed_key, input->value());
-    auto& tombstone_map = GetTombstoneMap(tombstone.seq_);
-    tombstone_map.emplace(input->key(), std::move(tombstone));
+    AddTombstone(std::move(tombstone));
    input->Next();
  }
  if (!first_iter) {
@@ -115,8 +198,172 @@ Status RangeDelAggregator::AddTombstones(
  return Status::OK();
 }

-RangeDelAggregator::TombstoneMap& RangeDelAggregator::GetTombstoneMap(
-    SequenceNumber seq) {
+void RangeDelAggregator::InvalidateTombstoneMapPositions() {
+  if (rep_ == nullptr) {
+    return;
+  }
+  for (auto stripe_map_iter = rep_->stripe_map_.begin();
+       stripe_map_iter != rep_->stripe_map_.end(); ++stripe_map_iter) {
+    stripe_map_iter->second.iter = stripe_map_iter->second.raw_map.end();
+  }
+}
+
+Status RangeDelAggregator::AddTombstone(RangeTombstone tombstone) {
+  auto& positional_tombstone_map = GetPositionalTombstoneMap(tombstone.seq_);
+  auto& tombstone_map = positional_tombstone_map.raw_map;
+  if (collapse_deletions_) {
+    // In collapsed mode, we only fill the seq_ field in the TombstoneMap's
+    // values. The end_key is unneeded because we assume the tombstone extends
+    // until the next tombstone starts. For gaps between real tombstones and
+    // for the last real tombstone, we denote end keys by inserting fake
+    // tombstones with sequence number zero.
+    std::vector<RangeTombstone> new_range_dels{
+        tombstone, RangeTombstone(tombstone.end_key_, Slice(), 0)};
+    auto new_range_dels_iter = new_range_dels.begin();
+    // Position at the first overlapping existing tombstone; if none exists,
+    // insert until we find an existing one overlapping a new point
+    const Slice* tombstone_map_begin = nullptr;
+    if (!tombstone_map.empty()) {
+      tombstone_map_begin = &tombstone_map.begin()->first;
+    }
+    auto last_range_dels_iter = new_range_dels_iter;
+    while (new_range_dels_iter != new_range_dels.end() &&
+           (tombstone_map_begin == nullptr ||
+            icmp_.user_comparator()->Compare(new_range_dels_iter->start_key_,
+                                             *tombstone_map_begin) < 0)) {
+      tombstone_map.emplace(
+          new_range_dels_iter->start_key_,
+          RangeTombstone(Slice(), Slice(), new_range_dels_iter->seq_));
+      last_range_dels_iter = new_range_dels_iter;
+      ++new_range_dels_iter;
+    }
+    if (new_range_dels_iter == new_range_dels.end()) {
+      return Status::OK();
+    }
+    // above loop advances one too far
+    new_range_dels_iter = last_range_dels_iter;
+    auto tombstone_map_iter =
+        tombstone_map.upper_bound(new_range_dels_iter->start_key_);
+    // if nothing overlapped we would've already inserted all the new points
+    // and returned early
+    assert(tombstone_map_iter != tombstone_map.begin());
+    tombstone_map_iter--;
+
+    // untermed_seq is non-kMaxSequenceNumber when we covered an existing point
+    // but haven't seen its corresponding endpoint. It's used for (1) deciding
+    // whether to forcibly insert the new interval's endpoint; and (2) possibly
+    // raising the seqnum for the to-be-inserted element (we insert the max
+    // seqnum between the next new interval and the unterminated interval).
+    SequenceNumber untermed_seq = kMaxSequenceNumber;
+    while (tombstone_map_iter != tombstone_map.end() &&
+           new_range_dels_iter != new_range_dels.end()) {
+      const Slice *tombstone_map_iter_end = nullptr,
+                  *new_range_dels_iter_end = nullptr;
+      if (tombstone_map_iter != tombstone_map.end()) {
+        auto next_tombstone_map_iter = std::next(tombstone_map_iter);
+        if (next_tombstone_map_iter != tombstone_map.end()) {
+          tombstone_map_iter_end = &next_tombstone_map_iter->first;
+        }
+      }
+      if (new_range_dels_iter != new_range_dels.end()) {
+        auto next_new_range_dels_iter = std::next(new_range_dels_iter);
+        if (next_new_range_dels_iter != new_range_dels.end()) {
+          new_range_dels_iter_end = &next_new_range_dels_iter->start_key_;
+        }
+      }
+
+      // our positions in existing/new tombstone collections should always
+      // overlap. The non-overlapping cases are handled above and below this
+      // loop.
+      assert(new_range_dels_iter_end == nullptr ||
+             icmp_.user_comparator()->Compare(tombstone_map_iter->first,
+                                              *new_range_dels_iter_end) < 0);
+      assert(tombstone_map_iter_end == nullptr ||
+             icmp_.user_comparator()->Compare(new_range_dels_iter->start_key_,
+                                              *tombstone_map_iter_end) < 0);
+
+      int new_to_old_start_cmp = icmp_.user_comparator()->Compare(
+          new_range_dels_iter->start_key_, tombstone_map_iter->first);
+      // nullptr end means extends infinitely rightwards, set new_to_old_end_cmp
+      // accordingly so we can use common code paths later.
+      int new_to_old_end_cmp;
+      if (new_range_dels_iter_end == nullptr &&
+          tombstone_map_iter_end == nullptr) {
+        new_to_old_end_cmp = 0;
+      } else if (new_range_dels_iter_end == nullptr) {
+        new_to_old_end_cmp = 1;
+      } else if (tombstone_map_iter_end == nullptr) {
+        new_to_old_end_cmp = -1;
+      } else {
+        new_to_old_end_cmp = icmp_.user_comparator()->Compare(
+            *new_range_dels_iter_end, *tombstone_map_iter_end);
+      }
+
+      if (new_to_old_start_cmp < 0) {
+        // the existing one's left endpoint comes after, so raise/delete it if
+        // it's covered.
+        if (tombstone_map_iter->second.seq_ < new_range_dels_iter->seq_) {
+          untermed_seq = tombstone_map_iter->second.seq_;
+          if (tombstone_map_iter != tombstone_map.begin() &&
+              std::prev(tombstone_map_iter)->second.seq_ ==
+                  new_range_dels_iter->seq_) {
+            tombstone_map_iter = tombstone_map.erase(tombstone_map_iter);
+            --tombstone_map_iter;
+          } else {
+            tombstone_map_iter->second.seq_ = new_range_dels_iter->seq_;
+          }
+        }
+      } else if (new_to_old_start_cmp > 0) {
+        if (untermed_seq != kMaxSequenceNumber ||
+            tombstone_map_iter->second.seq_ < new_range_dels_iter->seq_) {
+          auto seq = tombstone_map_iter->second.seq_;
+          // need to adjust this element if not intended to span beyond the new
+          // element (i.e., was_tombstone_map_iter_raised == true), or if it
+          // can be raised
+          tombstone_map_iter = tombstone_map.emplace(
+              new_range_dels_iter->start_key_,
+              RangeTombstone(
+                  Slice(), Slice(),
+                  std::max(
+                      untermed_seq == kMaxSequenceNumber ? 0 : untermed_seq,
+                      new_range_dels_iter->seq_)));
+          untermed_seq = seq;
+        }
+      } else {
+        // their left endpoints coincide, so raise the existing one if needed
+        if (tombstone_map_iter->second.seq_ < new_range_dels_iter->seq_) {
+          untermed_seq = tombstone_map_iter->second.seq_;
+          tombstone_map_iter->second.seq_ = new_range_dels_iter->seq_;
+        }
+      }
+
+      // advance whichever one ends earlier, or both if their right endpoints
+      // coincide
+      if (new_to_old_end_cmp < 0) {
+        ++new_range_dels_iter;
+      } else if (new_to_old_end_cmp > 0) {
+        ++tombstone_map_iter;
+        untermed_seq = kMaxSequenceNumber;
+      } else {
+        ++new_range_dels_iter;
+        ++tombstone_map_iter;
+        untermed_seq = kMaxSequenceNumber;
+      }
+    }
+    while (new_range_dels_iter != new_range_dels.end()) {
+      tombstone_map.emplace(
+          new_range_dels_iter->start_key_,
+          RangeTombstone(Slice(), Slice(), new_range_dels_iter->seq_));
+      ++new_range_dels_iter;
+    }
+  } else {
+    tombstone_map.emplace(tombstone.start_key_, std::move(tombstone));
+  }
+  return Status::OK();
+}
+
+RangeDelAggregator::PositionalTombstoneMap&
+RangeDelAggregator::GetPositionalTombstoneMap(SequenceNumber seq) {
  assert(rep_ != nullptr);
  // The stripe includes seqnum for the snapshot above and excludes seqnum for
  // the snapshot below.
@@ -148,10 +395,16 @@ void RangeDelAggregator::AddToBuilder(
  auto stripe_map_iter = rep_->stripe_map_.begin();
  assert(stripe_map_iter != rep_->stripe_map_.end());
  if (bottommost_level) {
-    range_del_out_stats->num_range_del_drop_obsolete +=
-        static_cast<int64_t>(stripe_map_iter->second.size());
-    range_del_out_stats->num_record_drop_obsolete +=
-        static_cast<int64_t>(stripe_map_iter->second.size());
+    // TODO(andrewkr): these are counted for each compaction output file, so
+    // lots of double-counting.
+    if (!stripe_map_iter->second.raw_map.empty()) {
+      range_del_out_stats->num_range_del_drop_obsolete +=
+          static_cast<int64_t>(stripe_map_iter->second.raw_map.size()) -
+          (collapse_deletions_ ? 1 : 0);
+      range_del_out_stats->num_record_drop_obsolete +=
+          static_cast<int64_t>(stripe_map_iter->second.raw_map.size()) -
+          (collapse_deletions_ ? 1 : 0);
+    }
    // 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;
@@ -161,8 +414,23 @@ void RangeDelAggregator::AddToBuilder(
  // insert them into a std::map on the read path.
  bool first_added = false;
  while (stripe_map_iter != rep_->stripe_map_.end()) {
-    for (const auto& start_key_and_tombstone : stripe_map_iter->second) {
-      const auto& tombstone = start_key_and_tombstone.second;
+    for (auto tombstone_map_iter = stripe_map_iter->second.raw_map.begin();
+         tombstone_map_iter != stripe_map_iter->second.raw_map.end();
+         ++tombstone_map_iter) {
+      RangeTombstone tombstone;
+      if (collapse_deletions_) {
+        auto next_tombstone_map_iter = std::next(tombstone_map_iter);
+        if (next_tombstone_map_iter == stripe_map_iter->second.raw_map.end() ||
+            tombstone_map_iter->second.seq_ == 0) {
+          // it's a sentinel tombstone
+          continue;
+        }
+        tombstone.start_key_ = tombstone_map_iter->first;
+        tombstone.end_key_ = next_tombstone_map_iter->first;
+        tombstone.seq_ = tombstone_map_iter->second.seq_;
+      } else {
+        tombstone = tombstone_map_iter->second;
+      }
      if (upper_bound != nullptr &&
          icmp_.user_comparator()->Compare(*upper_bound,
                                           tombstone.start_key_) <= 0) {
@@ -241,7 +509,7 @@ bool RangeDelAggregator::IsEmpty() {
  }
  for (auto stripe_map_iter = rep_->stripe_map_.begin();
       stripe_map_iter != rep_->stripe_map_.end(); ++stripe_map_iter) {
-    if (!stripe_map_iter->second.empty()) {
+    if (!stripe_map_iter->second.raw_map.empty()) {
      return false;
    }
  }
@@ -32,27 +32,51 @@ class RangeDelAggregator {
  //    stripes, which is the seqnum range between consecutive snapshots,
  //    including the higher snapshot and excluding the lower one. Currently,
  //    this is used by ShouldDelete() to prevent deletion of keys that are
-  //    covered by range tombstones in other snapshot stripes. In case of writes
-  //    (flush/compaction), all DB snapshots are provided such that no keys are
-  //    removed that are uncovered according to any DB snapshot. In case of read
-  //    (get/iterator), only the user snapshot is provided such that the seqnum
-  //    space is divided into two stripes, where only tombstones in the older
-  //    stripe are considered by ShouldDelete().
+  //    covered by range tombstones in other snapshot stripes. This constructor
+  //    is used for writes (flush/compaction). All DB snapshots are provided
+  //    such that no keys are removed that are uncovered according to any DB
+  //    snapshot.
  // Note this overload does not lazily initialize Rep.
  RangeDelAggregator(const InternalKeyComparator& icmp,
-                     const std::vector<SequenceNumber>& snapshots);
+                     const std::vector<SequenceNumber>& snapshots,
+                     bool collapse_deletions = true);

  // @param upper_bound Similar to snapshots above, except with a single
  //    snapshot, which allows us to store the snapshot on the stack and defer
  //    initialization of heap-allocating members (in Rep) until the first range
-  //    deletion is encountered.
+  //    deletion is encountered. This constructor is used in case of reads (get/
+  //    iterator), for which only the user snapshot (upper_bound) is provided
+  //    such that the seqnum space is divided into two stripes. Only the older
+  //    stripe will be used by ShouldDelete().
  RangeDelAggregator(const InternalKeyComparator& icmp,
-                     SequenceNumber upper_bound);
+                     SequenceNumber upper_bound,
+                     bool collapse_deletions = false);
+
+  // We maintain position in the tombstone map across calls to ShouldDelete. The
+  // caller may wish to specify a mode to optimize positioning the iterator
+  // during the next call to ShouldDelete. The non-kFullScan modes are only
+  // available when deletion collapsing is enabled.
+  //
+  // For example, if we invoke Next() on an iterator, kForwardTraversal should
+  // be specified to advance one-by-one through deletions until one is found
+  // with its interval containing the key. This will typically be faster than
+  // doing a full binary search (kBinarySearch).
+  enum RangePositioningMode {
+    kFullScan,  // used iff collapse_deletions_ == false
+    kForwardTraversal,
+    kBackwardTraversal,
+    kBinarySearch,
+  };

  // Returns whether the key should be deleted, which is the case when it is
  // covered by a range tombstone residing in the same snapshot stripe.
-  bool ShouldDelete(const ParsedInternalKey& parsed);
-  bool ShouldDelete(const Slice& internal_key);
+  // @param mode If collapse_deletions_ is true, this dictates how we will find
+  //             the deletion whose interval contains this key. Otherwise, its
+  //             value must be kFullScan indicating linear scan from beginning..
+  bool ShouldDelete(const ParsedInternalKey& parsed,
+                    RangePositioningMode mode = kFullScan);
+  bool ShouldDelete(const Slice& internal_key,
+                    RangePositioningMode mode = kFullScan);
  bool ShouldAddTombstones(bool bottommost_level = false);

  // Adds tombstones to the tombstone aggregation structure maintained by this
@@ -60,6 +84,13 @@ class RangeDelAggregator {
  // @return non-OK status if any of the tombstone keys are corrupted.
  Status AddTombstones(std::unique_ptr<InternalIterator> input);

+  // Resets iterators maintained across calls to ShouldDelete(). This may be
+  // called when the tombstones change, or the owner may call explicitly, e.g.,
+  // if it's an iterator that just seeked to an arbitrary position. The effect
+  // of invalidation is that the following call to ShouldDelete() will binary
+  // search for its tombstone.
+  void InvalidateTombstoneMapPositions();
+
  // Writes tombstones covering a range to a table builder.
  // @param extend_before_min_key If true, the range of tombstones to be added
  //    to the TableBuilder starts from the beginning of the key-range;
@@ -87,12 +118,26 @@ class RangeDelAggregator {
  bool IsEmpty();

 private:
-  // Maps tombstone internal start key -> tombstone object
-  typedef std::map<Slice, RangeTombstone, stl_wrappers::LessOfComparator>
+  // Maps tombstone user start key -> tombstone object
+  typedef std::multimap<Slice, RangeTombstone, stl_wrappers::LessOfComparator>
      TombstoneMap;
+  // Also maintains position in TombstoneMap last seen by ShouldDelete(). The
+  // end iterator indicates invalidation (e.g., if AddTombstones() changes the
+  // underlying map). End iterator cannot be invalidated.
+  struct PositionalTombstoneMap {
+    explicit PositionalTombstoneMap(TombstoneMap _raw_map)
+        : raw_map(std::move(_raw_map)), iter(raw_map.end()) {}
+    PositionalTombstoneMap(const PositionalTombstoneMap&) = delete;
+    PositionalTombstoneMap(PositionalTombstoneMap&& other)
+        : raw_map(std::move(other.raw_map)), iter(raw_map.end()) {}
+
+    TombstoneMap raw_map;
+    TombstoneMap::const_iterator iter;
+  };
+
  // 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;
+  typedef std::map<SequenceNumber, PositionalTombstoneMap> StripeMap;

  struct Rep {
    StripeMap stripe_map_;
@@ -103,11 +148,14 @@ class RangeDelAggregator {
  // once the first range deletion is encountered.
  void InitRep(const std::vector<SequenceNumber>& snapshots);

-  TombstoneMap& GetTombstoneMap(SequenceNumber seq);
+  PositionalTombstoneMap& GetPositionalTombstoneMap(SequenceNumber seq);
+  Status AddTombstone(RangeTombstone tombstone);

  SequenceNumber upper_bound_;
  std::unique_ptr<Rep> rep_;
  const InternalKeyComparator& icmp_;
+  // collapse range deletions so they're binary searchable
+  const bool collapse_deletions_;
 };

 }  // namespace rocksdb
@@ -0,0 +1,157 @@
+//  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 <algorithm>
+
+#include "db/db_test_util.h"
+#include "db/range_del_aggregator.h"
+#include "rocksdb/comparator.h"
+#include "util/testutil.h"
+
+namespace rocksdb {
+
+class RangeDelAggregatorTest : public testing::Test {};
+
+namespace {
+
+struct ExpectedPoint {
+  Slice begin;
+  SequenceNumber seq;
+};
+
+enum Direction {
+  kForward,
+  kReverse,
+};
+
+void VerifyRangeDels(const std::vector<RangeTombstone>& range_dels,
+                     const std::vector<ExpectedPoint>& expected_points) {
+  // Test same result regardless of which order the range deletions are added.
+  for (Direction dir : {kForward, kReverse}) {
+    auto icmp = InternalKeyComparator(BytewiseComparator());
+    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */, true);
+    std::vector<std::string> keys, values;
+    for (const auto& range_del : range_dels) {
+      auto key_and_value = range_del.Serialize();
+      keys.push_back(key_and_value.first.Encode().ToString());
+      values.push_back(key_and_value.second.ToString());
+    }
+    if (dir == kReverse) {
+      std::reverse(keys.begin(), keys.end());
+      std::reverse(values.begin(), values.end());
+    }
+    std::unique_ptr<test::VectorIterator> range_del_iter(
+        new test::VectorIterator(keys, values));
+    range_del_agg.AddTombstones(std::move(range_del_iter));
+
+    for (const auto expected_point : expected_points) {
+      ParsedInternalKey parsed_key;
+      parsed_key.user_key = expected_point.begin;
+      parsed_key.sequence = expected_point.seq;
+      parsed_key.type = kTypeValue;
+      ASSERT_FALSE(range_del_agg.ShouldDelete(
+          parsed_key,
+          RangeDelAggregator::RangePositioningMode::kForwardTraversal));
+      if (parsed_key.sequence > 0) {
+        --parsed_key.sequence;
+        ASSERT_TRUE(range_del_agg.ShouldDelete(
+            parsed_key,
+            RangeDelAggregator::RangePositioningMode::kForwardTraversal));
+      }
+    }
+  }
+}
+
+}  // anonymous namespace
+
+TEST_F(RangeDelAggregatorTest, Empty) { VerifyRangeDels({}, {{"a", 0}}); }
+
+TEST_F(RangeDelAggregatorTest, SameStartAndEnd) {
+  VerifyRangeDels({{"a", "a", 5}}, {{" ", 0}, {"a", 0}, {"b", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, Single) {
+  VerifyRangeDels({{"a", "b", 10}}, {{" ", 0}, {"a", 10}, {"b", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, OverlapAboveLeft) {
+  VerifyRangeDels({{"a", "c", 10}, {"b", "d", 5}},
+                  {{" ", 0}, {"a", 10}, {"c", 5}, {"d", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, OverlapAboveRight) {
+  VerifyRangeDels({{"a", "c", 5}, {"b", "d", 10}},
+                  {{" ", 0}, {"a", 5}, {"b", 10}, {"d", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, OverlapAboveMiddle) {
+  VerifyRangeDels({{"a", "d", 5}, {"b", "c", 10}},
+                  {{" ", 0}, {"a", 5}, {"b", 10}, {"c", 5}, {"d", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, OverlapFully) {
+  VerifyRangeDels({{"a", "d", 10}, {"b", "c", 5}},
+                  {{" ", 0}, {"a", 10}, {"d", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, OverlapPoint) {
+  VerifyRangeDels({{"a", "b", 5}, {"b", "c", 10}},
+                  {{" ", 0}, {"a", 5}, {"b", 10}, {"c", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, SameStartKey) {
+  VerifyRangeDels({{"a", "c", 5}, {"a", "b", 10}},
+                  {{" ", 0}, {"a", 10}, {"b", 5}, {"c", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, SameEndKey) {
+  VerifyRangeDels({{"a", "d", 5}, {"b", "d", 10}},
+                  {{" ", 0}, {"a", 5}, {"b", 10}, {"d", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, GapsBetweenRanges) {
+  VerifyRangeDels(
+      {{"a", "b", 5}, {"c", "d", 10}, {"e", "f", 15}},
+      {{" ", 0}, {"a", 5}, {"b", 0}, {"c", 10}, {"d", 0}, {"e", 15}, {"f", 0}});
+}
+
+// Note the Cover* tests also test cases where tombstones are inserted under a
+// larger one when VerifyRangeDels() runs them in reverse
+TEST_F(RangeDelAggregatorTest, CoverMultipleFromLeft) {
+  VerifyRangeDels(
+      {{"b", "d", 5}, {"c", "f", 10}, {"e", "g", 15}, {"a", "f", 20}},
+      {{" ", 0}, {"a", 20}, {"f", 15}, {"g", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, CoverMultipleFromRight) {
+  VerifyRangeDels(
+      {{"b", "d", 5}, {"c", "f", 10}, {"e", "g", 15}, {"c", "h", 20}},
+      {{" ", 0}, {"b", 5}, {"c", 20}, {"h", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, CoverMultipleFully) {
+  VerifyRangeDels(
+      {{"b", "d", 5}, {"c", "f", 10}, {"e", "g", 15}, {"a", "h", 20}},
+      {{" ", 0}, {"a", 20}, {"h", 0}});
+}
+
+TEST_F(RangeDelAggregatorTest, AlternateMultipleAboveBelow) {
+  VerifyRangeDels(
+      {{"b", "d", 15}, {"c", "f", 10}, {"e", "g", 20}, {"a", "h", 5}},
+      {{" ", 0},
+       {"a", 5},
+       {"b", 15},
+       {"d", 10},
+       {"e", 20},
+       {"g", 5},
+       {"h", 0}});
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -486,6 +486,7 @@ class Repairer {
        if (empty) {
          empty = false;
          t->meta.smallest.DecodeFrom(key);
+          t->min_sequence = parsed.sequence;
        }
        t->meta.largest.DecodeFrom(key);
        if (parsed.sequence < t->min_sequence) {
@@ -3,6 +3,8 @@
 //  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.

+#ifndef ROCKSDB_LITE
+
 #include <algorithm>
 #include <string>
 #include <vector>
@@ -17,6 +19,7 @@

 namespace rocksdb {

+#ifndef ROCKSDB_LITE
 class RepairTest : public DBTestBase {
 public:
  RepairTest() : DBTestBase("/repair_test") {}
@@ -253,7 +256,10 @@ TEST_F(RepairTest, RepairColumnFamilyOptions) {
  db_->GetPropertiesOfAllTables(handles_[1], &fname_to_props);
  ASSERT_EQ(fname_to_props.size(), 2U);
  for (const auto& fname_and_props : fname_to_props) {
-    ASSERT_EQ(InternalKeyComparator(rev_opts.comparator).Name(),
+    std::string comparator_name (
+      InternalKeyComparator(rev_opts.comparator).Name());
+    comparator_name = comparator_name.substr(comparator_name.find(':') + 1);
+    ASSERT_EQ(comparator_name,
              fname_and_props.second->comparator_name);
  }

@@ -268,9 +274,21 @@ TEST_F(RepairTest, RepairColumnFamilyOptions) {
    }
  }
 }
+
+#endif  // ROCKSDB_LITE
 }  // namespace rocksdb

 int main(int argc, char** argv) {
  ::testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
 }
+
+#else
+#include <stdio.h>
+
+int main(int argc, char** argv) {
+  fprintf(stderr, "SKIPPED as RepairDB is not supported in ROCKSDB_LITE\n");
+  return 0;
+}
+
+#endif  // ROCKSDB_LITE
@@ -16,6 +16,7 @@
 #include <inttypes.h>
 #include <algorithm>
 #include <atomic>
+#include <functional>
 #include <set>
 #include <thread>
 #include <unordered_map>
@@ -27,6 +28,7 @@
 #include "db/internal_stats.h"
 #include "db/table_cache.h"
 #include "db/version_set.h"
+#include "port/port.h"
 #include "table/table_reader.h"

 namespace rocksdb {
@@ -358,7 +360,7 @@ class VersionBuilder::Rep {
    if (max_threads <= 1) {
      load_handlers_func();
    } else {
-      std::vector<std::thread> threads;
+      std::vector<port::Thread> threads;
      for (int i = 0; i < max_threads; i++) {
        threads.emplace_back(load_handlers_func);
      }
@@ -22,7 +22,6 @@
 #include <unordered_map>
 #include <vector>
 #include <string>
-
 #include "db/compaction.h"
 #include "db/filename.h"
 #include "db/internal_stats.h"
@@ -40,7 +39,7 @@
 #include "table/format.h"
 #include "table/get_context.h"
 #include "table/internal_iterator.h"
-#include "table/merger.h"
+#include "table/merging_iterator.h"
 #include "table/meta_blocks.h"
 #include "table/plain_table_factory.h"
 #include "table/table_reader.h"
@@ -1640,10 +1639,11 @@ void VersionStorageInfo::GetOverlappingInputs(
  }
  const Comparator* user_cmp = user_comparator_;
  if (begin != nullptr && end != nullptr && level > 0) {
-    GetOverlappingInputsBinarySearch(level, user_begin, user_end, inputs,
-      hint_index, file_index);
+    GetOverlappingInputsRangeBinarySearch(level, user_begin, user_end, inputs,
+                                          hint_index, file_index);
    return;
  }
+
  for (size_t i = 0; i < level_files_brief_[level].num_files; ) {
    FdWithKeyRange* f = &(level_files_brief_[level].files[i++]);
    const Slice file_start = ExtractUserKey(f->smallest_key);
@@ -1674,13 +1674,49 @@ void VersionStorageInfo::GetOverlappingInputs(
  }
 }

+// Store in "*inputs" files in "level" that within range [begin,end]
+// Guarantee a "clean cut" boundary between the files in inputs
+// and the surrounding files and the maxinum number of files.
+// This will ensure that no parts of a key are lost during compaction.
+// If hint_index is specified, then it points to a file in the range.
+// The file_index returns a pointer to any file in an overlapping range.
+void VersionStorageInfo::GetCleanInputsWithinInterval(
+    int level, const InternalKey* begin, const InternalKey* end,
+    std::vector<FileMetaData*>* inputs, int hint_index, int* file_index) const {
+  if (level >= num_non_empty_levels_) {
+    // this level is empty, no inputs within range
+    return;
+  }
+
+  inputs->clear();
+  Slice user_begin, user_end;
+  if (begin != nullptr) {
+    user_begin = begin->user_key();
+  }
+  if (end != nullptr) {
+    user_end = end->user_key();
+  }
+  if (file_index) {
+    *file_index = -1;
+  }
+  if (begin != nullptr && end != nullptr && level > 0) {
+    GetOverlappingInputsRangeBinarySearch(level, user_begin, user_end, inputs,
+                                          hint_index, file_index,
+                                          true /* within_interval */);
+  }
+}
+
 // Store in "*inputs" all files in "level" that overlap [begin,end]
 // Employ binary search to find at least one file that overlaps the
 // specified range. From that file, iterate backwards and
 // forwards to find all overlapping files.
-void VersionStorageInfo::GetOverlappingInputsBinarySearch(
+// if within_range is set, then only store the maximum clean inputs
+// within range [begin, end]. "clean" means there is a boudnary
+// between the files in "*inputs" and the surrounding files
+void VersionStorageInfo::GetOverlappingInputsRangeBinarySearch(
    int level, const Slice& user_begin, const Slice& user_end,
-    std::vector<FileMetaData*>* inputs, int hint_index, int* file_index) const {
+    std::vector<FileMetaData*>* inputs, int hint_index, int* file_index,
+    bool within_interval) const {
  assert(level > 0);
  int min = 0;
  int mid = 0;
@@ -1700,9 +1736,13 @@ void VersionStorageInfo::GetOverlappingInputsBinarySearch(
    FdWithKeyRange* f = &(level_files_brief_[level].files[mid]);
    const Slice file_start = ExtractUserKey(f->smallest_key);
    const Slice file_limit = ExtractUserKey(f->largest_key);
-    if (user_cmp->Compare(file_limit, user_begin) < 0) {
+    if ((!within_interval && user_cmp->Compare(file_limit, user_begin) < 0) ||
+        (within_interval && user_cmp->Compare(file_start, user_begin) < 0)) {
      min = mid + 1;
-    } else if (user_cmp->Compare(user_end, file_start) < 0) {
+    } else if ((!within_interval &&
+                user_cmp->Compare(user_end, file_start) < 0) ||
+               (within_interval &&
+                user_cmp->Compare(user_end, file_limit) < 0)) {
      max = mid - 1;
    } else {
      foundOverlap = true;
@@ -1718,24 +1758,38 @@ void VersionStorageInfo::GetOverlappingInputsBinarySearch(
  if (file_index) {
    *file_index = mid;
  }
-  ExtendOverlappingInputs(level, user_begin, user_end, inputs, mid);
+
+  int start_index, end_index;
+  if (within_interval) {
+    ExtendFileRangeWithinInterval(level, user_begin, user_end, mid, &start_index,
+                                  &end_index);
+  } else {
+    ExtendFileRangeOverlappingInterval(level, user_begin, user_end, mid,
+                                       &start_index, &end_index);
+  }
+  assert(end_index >= start_index);
+  // insert overlapping files into vector
+  for (int i = start_index; i <= end_index; i++) {
+    inputs->push_back(files_[level][i]);
+  }
 }

-// Store in "*inputs" all files in "level" that overlap [begin,end]
-// The midIndex specifies the index of at least one file that
+// Store in *start_index and *end_index the range of all files in
+// "level" that overlap [begin,end]
+// The mid_index specifies the index of at least one file that
 // overlaps the specified range. From that file, iterate backward
 // and forward to find all overlapping files.
 // Use FileLevel in searching, make it faster
-void VersionStorageInfo::ExtendOverlappingInputs(
+void VersionStorageInfo::ExtendFileRangeOverlappingInterval(
    int level, const Slice& user_begin, const Slice& user_end,
-    std::vector<FileMetaData*>* inputs, unsigned int midIndex) const {
+    unsigned int mid_index, int* start_index, int* end_index) const {
  const Comparator* user_cmp = user_comparator_;
  const FdWithKeyRange* files = level_files_brief_[level].files;
 #ifndef NDEBUG
  {
-    // assert that the file at midIndex overlaps with the range
-    assert(midIndex < level_files_brief_[level].num_files);
-    const FdWithKeyRange* f = &files[midIndex];
+    // assert that the file at mid_index overlaps with the range
+    assert(mid_index < level_files_brief_[level].num_files);
+    const FdWithKeyRange* f = &files[mid_index];
    const Slice fstart = ExtractUserKey(f->smallest_key);
    const Slice flimit = ExtractUserKey(f->largest_key);
    if (user_cmp->Compare(fstart, user_begin) >= 0) {
@@ -1745,91 +1799,105 @@ void VersionStorageInfo::ExtendOverlappingInputs(
    }
  }
 #endif
-  int startIndex = midIndex + 1;
-  int endIndex = midIndex;
+  *start_index = mid_index + 1;
+  *end_index = mid_index;
  int count __attribute__((unused)) = 0;

  // check backwards from 'mid' to lower indices
-  for (int i = midIndex; i >= 0 ; i--) {
+  for (int i = mid_index; i >= 0 ; i--) {
    const FdWithKeyRange* f = &files[i];
    const Slice file_limit = ExtractUserKey(f->largest_key);
    if (user_cmp->Compare(file_limit, user_begin) >= 0) {
-      startIndex = i;
+      *start_index = i;
      assert((count++, true));
    } else {
      break;
    }
  }
  // check forward from 'mid+1' to higher indices
-  for (unsigned int i = midIndex+1;
+  for (unsigned int i = mid_index+1;
       i < level_files_brief_[level].num_files; i++) {
    const FdWithKeyRange* f = &files[i];
    const Slice file_start = ExtractUserKey(f->smallest_key);
    if (user_cmp->Compare(file_start, user_end) <= 0) {
      assert((count++, true));
-      endIndex = i;
+      *end_index = i;
    } else {
      break;
    }
  }
-  assert(count == endIndex - startIndex + 1);
-
-  // insert overlapping files into vector
-  for (int i = startIndex; i <= endIndex; i++) {
-    FileMetaData* f = files_[level][i];
-    inputs->push_back(f);
-  }
+  assert(count == *end_index - *start_index + 1);
 }

-// Returns true iff the first or last file in inputs contains
-// an overlapping user key to the file "just outside" of it (i.e.
-// just after the last file, or just before the first file)
-// REQUIRES: "*inputs" is a sorted list of non-overlapping files
-bool VersionStorageInfo::HasOverlappingUserKey(
-    const std::vector<FileMetaData*>* inputs, int level) {
-
-  // If inputs empty, there is no overlap.
-  // If level == 0, it is assumed that all needed files were already included.
-  if (inputs->empty() || level == 0){
-    return false;
-  }
-
+// Store in *start_index and *end_index the clean range of all files in
+// "level" within [begin,end]
+// The mid_index specifies the index of at least one file within
+// the specified range. From that file, iterate backward
+// and forward to find all overlapping files and then "shrink" to
+// the clean range required.
+// Use FileLevel in searching, make it faster
+void VersionStorageInfo::ExtendFileRangeWithinInterval(
+    int level, const Slice& user_begin, const Slice& user_end,
+    unsigned int mid_index, int* start_index, int* end_index) const {
+  assert(level != 0);
  const Comparator* user_cmp = user_comparator_;
-  const rocksdb::LevelFilesBrief& file_level = level_files_brief_[level];
  const FdWithKeyRange* files = level_files_brief_[level].files;
-  const size_t kNumFiles = file_level.num_files;
-
-  // Check the last file in inputs against the file after it
-  size_t last_file = FindFile(*internal_comparator_, file_level,
-                              inputs->back()->largest.Encode());
-  assert(last_file < kNumFiles);  // File should exist!
-  if (last_file < kNumFiles-1) {                    // If not the last file
-    const Slice last_key_in_input = ExtractUserKey(
-        files[last_file].largest_key);
-    const Slice first_key_after = ExtractUserKey(
-        files[last_file+1].smallest_key);
-    if (user_cmp->Equal(last_key_in_input, first_key_after)) {
-      // The last user key in input overlaps with the next file's first key
-      return true;
+#ifndef NDEBUG
+  {
+    // assert that the file at mid_index is within the range
+    assert(mid_index < level_files_brief_[level].num_files);
+    const FdWithKeyRange* f = &files[mid_index];
+    const Slice fstart = ExtractUserKey(f->smallest_key);
+    const Slice flimit = ExtractUserKey(f->largest_key);
+    assert(user_cmp->Compare(fstart, user_begin) >= 0 &&
+           user_cmp->Compare(flimit, user_end) <= 0);
+  }
+#endif
+  ExtendFileRangeOverlappingInterval(level, user_begin, user_end, mid_index,
+                                     start_index, end_index);
+  int left = *start_index;
+  int right = *end_index;
+  // shrink from left to right
+  while (left <= right) {
+    const Slice& first_key_in_range = ExtractUserKey(files[left].smallest_key);
+    if (user_cmp->Compare(first_key_in_range, user_begin) < 0) {
+      left++;
+      continue;
    }
+    if (left > 0) {  // If not first file
+      const Slice& last_key_before =
+          ExtractUserKey(files[left - 1].largest_key);
+      if (user_cmp->Equal(first_key_in_range, last_key_before)) {
+        // The first user key in range overlaps with the previous file's last
+        // key
+        left++;
+        continue;
+      }
+    }
+    break;
+  }
+  // shrink from right to left
+  while (left <= right) {
+    const Slice last_key_in_range = ExtractUserKey(files[right].largest_key);
+    if (user_cmp->Compare(last_key_in_range, user_end) > 0) {
+      right--;
+      continue;
+    }
+    if (right < static_cast<int>(level_files_brief_[level].num_files) -
+                    1) {  // If not the last file
+      const Slice first_key_after =
+          ExtractUserKey(files[right + 1].smallest_key);
+      if (user_cmp->Equal(last_key_in_range, first_key_after)) {
+        // The last user key in range overlaps with the next file's first key
+        right--;
+        continue;
+      }
+    }
+    break;
  }

-  // Check the first file in inputs against the file just before it
-  size_t first_file = FindFile(*internal_comparator_, file_level,
-                               inputs->front()->smallest.Encode());
-  assert(first_file <= last_file);   // File should exist!
-  if (first_file > 0) {                                 // If not first file
-    const Slice& first_key_in_input = ExtractUserKey(
-        files[first_file].smallest_key);
-    const Slice& last_key_before = ExtractUserKey(
-        files[first_file-1].largest_key);
-    if (user_cmp->Equal(first_key_in_input, last_key_before)) {
-      // The first user key in input overlaps with the previous file's last key
-      return true;
-    }
-  }
-
-  return false;
+  *start_index = left;
+  *end_index = right;
 }

 uint64_t VersionStorageInfo::NumLevelBytes(int level) const {
@@ -2361,9 +2429,8 @@ 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_->disable_data_sync ? nullptr : db_directory);
+      s = SetCurrentFile(env_, dbname_, pending_manifest_file_number_,
+                         db_directory);
    }

    if (s.ok()) {
@@ -3364,8 +3431,7 @@ InternalIterator* VersionSet::MakeInputIterator(
    const Compaction* c, RangeDelAggregator* range_del_agg) {
  auto cfd = c->column_family_data();
  ReadOptions read_options;
-  read_options.verify_checksums =
-    c->mutable_cf_options()->verify_checksums_in_compaction;
+  read_options.verify_checksums = true;
  read_options.fill_cache = false;
  if (c->ShouldFormSubcompactions()) {
    read_options.total_order_seek = true;