Add skip_memtable_flush in IngestExternalFileOptions

Summary: Set to true if you are sure although the key ranges may overlap, the memtable and the file being ingested do not have the same keys, or not care about that. The memtable flushing may block for hundreds of milliseconds. Test Plan: make check
Fix some defects reported by Coverity Scan (#6933 )
2026-07-08 07:05:23 +08:00 · 2020-06-14 14:49:06 +08:00 · 2020-06-04 15:46:27 -07:00 · 2020-06-04 15:35:08 -07:00 · 2020-06-04 14:21:15 -07:00 · 2020-06-04 14:17:11 -07:00
369 changed files with 18823 additions and 5118 deletions
@@ -6,7 +6,7 @@ orbs:
 executors:
  windows-2xlarge:
    machine:
-      image: 'windows-server-2019-vs2019:stable'
+      image: 'windows-server-2019-vs2019:201908-06'
      resource_class: windows.2xlarge
      shell: bash.exe

@@ -0,0 +1,41 @@
+name: Check buck targets and code format
+on: [push, pull_request]
+jobs:
+  check:
+    name: Check TARGETS file and code format
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout feature branch
+      uses: actions/checkout@v2
+      with:
+        fetch-depth: 0
+
+    - name: Fetch from upstream
+      run: |
+        git remote add upstream https://github.com/facebook/rocksdb.git && git fetch upstream
+
+    - name: Where am I
+      run: |
+        echo git status && git status
+        echo "git remote -v" && git remote -v
+        echo git branch && git branch
+
+    - name: Setup Python
+      uses: actions/setup-python@v1
+
+    - name: Install Dependencies
+      run: python -m pip install --upgrade pip
+
+    - name: Install argparse
+      run: pip install argparse
+
+    - name: Download clang-format-diff.py
+      uses: wei/wget@v1
+      with:
+        args: https://raw.githubusercontent.com/llvm-mirror/clang/master/tools/clang-format/clang-format-diff.py
+
+    - name: Check format
+      run: VERBOSE_CHECK=1 make check-format
+
+    - name: Compare buckify output
+      run: make check-buck-targets
@@ -85,3 +85,5 @@ buckifier/*.pyc
 buckifier/__pycache__

 compile_commands.json
+clang-format-diff.py
+.py3/
@@ -50,14 +50,24 @@ env:
  - JOB_NAME=examples # 5-7 minutes
  - JOB_NAME=cmake # 3-5 minutes
  - JOB_NAME=cmake-gcc8 # 3-5 minutes
+  - JOB_NAME=cmake-gcc9 # 3-5 minutes
+  - JOB_NAME=cmake-gcc9-c++20 # 3-5 minutes
  - JOB_NAME=cmake-mingw # 3 minutes
+  - JOB_NAME=make-gcc4.8
+  - JOB_NAME=status_checked

 matrix:
  exclude:
  - os: osx
    env: JOB_NAME=cmake-gcc8
+  - os: osx
+    env: JOB_NAME=cmake-gcc9
+  - os: osx
+    env: JOB_NAME=cmake-gcc9-c++20
  - os: osx
    env: JOB_NAME=cmake-mingw
+  - os: osx
+    env: JOB_NAME=make-gcc4.8
  - os: osx
    arch: ppc64le
  - os: osx
@@ -65,12 +75,36 @@ matrix:
  - os : linux
    arch: arm64
    env: JOB_NAME=cmake-mingw
+  - os : linux
+    arch: arm64
+    env: JOB_NAME=make-gcc4.8
  - os: linux
    arch: ppc64le
    env: JOB_NAME=cmake-mingw
+  - os: linux
+    arch: ppc64le
+    env: JOB_NAME=make-gcc4.8
  - os: linux
    compiler: clang
+  # Exclude all but most unique cmake variants for pull requests, but build all in branches
+  - if: type = pull_request
+    os : linux
+    arch: amd64
+    env: JOB_NAME=cmake
+  - if: type = pull_request
+    os : linux
+    arch: amd64
+    env: JOB_NAME=cmake-gcc8
+  - if: type = pull_request
+    os : linux
+    arch: amd64
+    env: JOB_NAME=cmake-gcc9
  # Exclude most osx, arm64 and ppc64le tests for pull requests, but build in branches
+  # Temporarily disable ppc64le unit tests in PRs until Travis gets its act together (#6653)
+  - if: type = pull_request
+    os: linux
+    arch: ppc64le
+    env: TEST_GROUP=platform_dependent
  # NB: the cmake build is a partial java test
  - if: type = pull_request
    os: osx
@@ -116,14 +150,14 @@ matrix:
    os: linux
    arch: ppc64le
    env: TEST_GROUP=4
-  - if: type = pull_request
+  - if: type = pull_request AND commit_message !~ /java/
    os : osx
    env: JOB_NAME=java_test
-  - if: type = pull_request
+  - if: type = pull_request AND commit_message !~ /java/
    os : linux
    arch: arm64
    env: JOB_NAME=java_test
-  - if: type = pull_request
+  - if: type = pull_request AND commit_message !~ /java/
    os: linux
    arch: ppc64le
    env: JOB_NAME=java_test
@@ -157,6 +191,33 @@ matrix:
    os: linux
    arch: ppc64le
    env: JOB_NAME=cmake-gcc8
+  - if: type = pull_request
+    os : linux
+    arch: arm64
+    env: JOB_NAME=cmake-gcc9
+  - if: type = pull_request
+    os: linux
+    arch: ppc64le
+    env: JOB_NAME=cmake-gcc9
+  - if: type = pull_request
+    os : linux
+    arch: arm64
+    env: JOB_NAME=cmake-gcc9-c++20
+  - if: type = pull_request
+    os: linux
+    arch: ppc64le
+    env: JOB_NAME=cmake-gcc9-c++20
+  - if: type = pull_request
+    os : osx
+    env: JOB_NAME=status_checked
+  - if: type = pull_request
+    os : linux
+    arch: arm64
+    env: JOB_NAME=status_checked
+  - if: type = pull_request
+    os: linux
+    arch: ppc64le
+    env: JOB_NAME=status_checked

 install:
  - if [ "${TRAVIS_OS_NAME}" == osx ]; then
@@ -166,9 +227,17 @@ install:
      sudo apt-get install -y g++-8;
      CC=gcc-8 && CXX=g++-8;
    fi
+  - if [ "${JOB_NAME}" == cmake-gcc9 ] || [ "${JOB_NAME}" == cmake-gcc9-c++20 ]; then
+      sudo apt-get install -y g++-9;
+      CC=gcc-9 && CXX=g++-9;
+    fi
  - if [ "${JOB_NAME}" == cmake-mingw ]; then
      sudo apt-get install -y mingw-w64 ;
    fi
+  - if [ "${JOB_NAME}" == make-gcc4.8 ]; then
+      sudo apt-get install -y g++-4.8 ;
+      CC=gcc-4.8 && CXX=g++-4.8;
+    fi
  - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
      CMAKE_DIST_URL="https://rocksdb-deps.s3-us-west-2.amazonaws.com/cmake/cmake-3.14.5-Linux-$(uname -m).tar.bz2";
      TAR_OPT="--strip-components=1 -xj";
@@ -201,7 +270,7 @@ before_script:
  - ulimit -n 8192

 script:
-  - ${CXX} --version
+  - date; ${CXX} --version
  - if [ `command -v ccache` ]; then ccache -C; fi
  - case $TEST_GROUP in
    platform_dependent)
@@ -225,7 +294,7 @@ script:
      OPT=-DTRAVIS V=1 make rocksdbjava jtest
      ;;
    lite_build)
-      OPT='-DTRAVIS -DROCKSDB_LITE' V=1 make -j4 static_lib tools
+      OPT='-DTRAVIS -DROCKSDB_LITE' V=1 make -j4 all
      ;;
    examples)
      OPT=-DTRAVIS V=1 make -j4 static_lib && cd examples && make -j4
@@ -235,7 +304,19 @@ script:
      mkdir build && cd build && cmake -DJNI=1 -DWITH_GFLAGS=OFF .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb rocksdbjni
      ;;
    cmake*)
-      mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_BUILD_TYPE=Release && make -j4 rocksdb rocksdbjni
+      case $JOB_NAME in
+        *-c++20)
+          OPT=-DCMAKE_CXX_STANDARD=20
+          ;;
+      esac
+
+      mkdir build && cd build && cmake -DCMAKE_BUILD_TYPE=Release -DWITH_TESTS=0 -DWITH_GFLAGS=0 -DWITH_BENCHMARK_TOOLS=0 -DWITH_TOOLS=0 -DWITH_CORE_TOOLS=1 .. && make -j4 && cd  .. && rm -rf build && mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_BUILD_TYPE=Release $OPT && make -j4 rocksdb rocksdbjni
+      ;;
+    make-gcc4.8)
+      OPT=-DTRAVIS V=1 SKIP_LINK=1 make -j4 all && [ "Linking broken because libgflags compiled with newer ABI" ]
+      ;;
+    status_checked)
+      OPT=-DTRAVIS V=1 ASSERT_STATUS_CHECKED=1 make -j4 check_some
      ;;
    esac
 notifications:
@@ -83,6 +83,10 @@ else()
  option(WITH_FOLLY_DISTRIBUTED_MUTEX "build with folly::DistributedMutex" OFF)
 endif()

+if( NOT DEFINED CMAKE_CXX_STANDARD )
+  set(CMAKE_CXX_STANDARD 11)
+endif()
+
 include(CMakeDependentOption)
 CMAKE_DEPENDENT_OPTION(WITH_GFLAGS "build with GFlags" ON
  "NOT MSVC;NOT MINGW" OFF)
@@ -103,18 +107,32 @@ else()
    endif()
  endif()

-  # No config file for this
  if(WITH_GFLAGS)
-    find_package(gflags REQUIRED)
+    # Config with namespace available since gflags 2.2.2
+    option(GFLAGS_USE_TARGET_NAMESPACE "Use gflags import target with namespace." ON)
+    find_package(gflags CONFIG)
+    if(gflags_FOUND)
+      if(TARGET ${GFLAGS_TARGET})
+        # Config with GFLAGS_TARGET available since gflags 2.2.0
+        list(APPEND THIRDPARTY_LIBS ${GFLAGS_TARGET})
+      else()
+        # Config with GFLAGS_LIBRARIES available since gflags 2.1.0
+        list(APPEND THIRDPARTY_LIBS ${GFLAGS_LIBRARIES})
+      endif()
+    else()
+      find_package(gflags REQUIRED)
+      list(APPEND THIRDPARTY_LIBS gflags::gflags)
+    endif()
    add_definitions(-DGFLAGS=1)
-    include_directories(${gflags_INCLUDE_DIR})
-    list(APPEND THIRDPARTY_LIBS gflags::gflags)
  endif()

  if(WITH_SNAPPY)
-    find_package(snappy REQUIRED)
+    find_package(Snappy CONFIG)
+    if(NOT Snappy_FOUND)
+      find_package(Snappy REQUIRED)
+    endif()
    add_definitions(-DSNAPPY)
-    list(APPEND THIRDPARTY_LIBS snappy::snappy)
+    list(APPEND THIRDPARTY_LIBS Snappy::snappy)
  endif()

  if(WITH_ZLIB)
@@ -191,7 +209,6 @@ else()
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-format -fno-asynchronous-unwind-tables")
    add_definitions(-D_POSIX_C_SOURCE=1)
  endif()
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
    include(CheckCXXCompilerFlag)
@@ -254,6 +271,7 @@ include(CheckCXXSourceCompiles)
 if(NOT MSVC)
  set(CMAKE_REQUIRED_FLAGS "-msse4.2 -mpclmul")
 endif()
+
 CHECK_CXX_SOURCE_COMPILES("
 #include <cstdint>
 #include <nmmintrin.h>
@@ -386,7 +404,15 @@ if(MSVC)
    message(STATUS "Debug optimization is enabled")
    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt")
  else()
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm")
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1")
+
+    # Minimal Build is deprecated after MSVC 2015
+    if( MSVC_VERSION GREATER 1900 )
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Gm-")
+    else()
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Gm")
+    endif()
+
  endif()
  if(WITH_RUNTIME_DEBUG)
    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /${RUNTIME_LIBRARY}d")
@@ -497,7 +523,6 @@ endif()

 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
-include_directories(SYSTEM ${PROJECT_SOURCE_DIR}/third-party/gtest-1.8.1/fused-src)
 if(WITH_FOLLY_DISTRIBUTED_MUTEX)
  include_directories(${PROJECT_SOURCE_DIR}/third-party/folly)
 endif()
@@ -506,6 +531,7 @@ find_package(Threads REQUIRED)
 # Main library source code

 set(SOURCES
+        cache/cache.cc
        cache/clock_cache.cc
        cache/lru_cache.cc
        cache/sharded_cache.cc
@@ -595,6 +621,7 @@ set(SOURCES
        memory/arena.cc
        memory/concurrent_arena.cc
        memory/jemalloc_nodump_allocator.cc
+        memory/memkind_kmem_allocator.cc
        memtable/alloc_tracker.cc
        memtable/hash_linklist_rep.cc
        memtable/hash_skiplist_rep.cc
@@ -619,7 +646,6 @@ set(SOURCES
        options/options.cc
        options/options_helper.cc
        options/options_parser.cc
-        options/options_sanity_check.cc
        port/stack_trace.cc
        table/adaptive/adaptive_table_factory.cc
        table/block_based/binary_search_index_reader.cc
@@ -777,9 +803,13 @@ if(WIN32)
    port/win/env_win.cc
    port/win/env_default.cc
    port/win/port_win.cc
-    port/win/win_logger.cc
-    port/win/win_thread.cc)
-
+    port/win/win_logger.cc)
+  if(NOT MINGW)
+    # Mingw only supports std::thread when using
+    # posix threads.
+    list(APPEND SOURCES
+      port/win/win_thread.cc)
+  endif()
 if(WITH_XPRESS)
  list(APPEND SOURCES
    port/win/xpress_win.cc)
@@ -826,12 +856,12 @@ else()
 endif()

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

 if(ROCKSDB_BUILD_SHARED)
  add_library(${ROCKSDB_SHARED_LIB} SHARED ${SOURCES})
-  target_link_libraries(${ROCKSDB_SHARED_LIB}
+  target_link_libraries(${ROCKSDB_SHARED_LIB} PRIVATE
    ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})

  if(WIN32)
@@ -848,7 +878,6 @@ if(ROCKSDB_BUILD_SHARED)
                          LINKER_LANGUAGE CXX
                          VERSION ${rocksdb_VERSION}
                          SOVERSION ${rocksdb_VERSION_MAJOR}
-                          CXX_STANDARD 11
                          OUTPUT_NAME "rocksdb")
  endif()
 endif()
@@ -860,6 +889,16 @@ else()
 endif()

 option(WITH_JNI "build with JNI" OFF)
+# Tests are excluded from Release builds
+CMAKE_DEPENDENT_OPTION(WITH_TESTS "build with tests" ON
+  "CMAKE_BUILD_TYPE STREQUAL Debug" OFF)
+option(WITH_BENCHMARK_TOOLS "build with benchmarks" ON)
+option(WITH_CORE_TOOLS "build with ldb and sst_dump" ON)
+option(WITH_TOOLS "build with tools" ON)
+
+if(WITH_TESTS OR WITH_BENCHMARK_TOOLS OR WITH_TOOLS OR WITH_JNI OR JNI)
+  include_directories(SYSTEM ${PROJECT_SOURCE_DIR}/third-party/gtest-1.8.1/fused-src)
+endif()
 if(WITH_JNI OR JNI)
  message(STATUS "JNI library is enabled")
  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/java)
@@ -897,6 +936,8 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)

  install(DIRECTORY include/rocksdb COMPONENT devel DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}")

+  install(DIRECTORY "${PROJECT_SOURCE_DIR}/cmake/modules" COMPONENT devel DESTINATION ${package_config_destination})
+
  install(
    TARGETS ${ROCKSDB_STATIC_LIB}
    EXPORT RocksDBTargets
@@ -933,15 +974,11 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
  )
 endif()

-# Tests are excluded from Release builds
-CMAKE_DEPENDENT_OPTION(WITH_TESTS "build with tests" ON
-  "CMAKE_BUILD_TYPE STREQUAL Debug" OFF)
 if(WITH_TESTS)
  add_subdirectory(third-party/gtest-1.8.1/fused-src/gtest)
  add_library(testharness STATIC
  test_util/testharness.cc)
  target_link_libraries(testharness gtest)
-
  set(TESTS
        cache/cache_test.cc
        cache/lru_cache_test.cc
@@ -987,6 +1024,7 @@ if(WITH_TESTS)
        db/db_logical_block_size_cache_test.cc
        db/db_universal_compaction_test.cc
        db/db_wal_test.cc
+        db/db_with_timestamp_compaction_test.cc
        db/db_write_test.cc
        db/dbformat_test.cc
        db/deletefile_test.cc
@@ -1029,6 +1067,7 @@ if(WITH_TESTS)
        logging/env_logger_test.cc
        logging/event_logger_test.cc
        memory/arena_test.cc
+        memory/memkind_kmem_allocator_test.cc
        memtable/inlineskiplist_test.cc
        memtable/skiplist_test.cc
        memtable/write_buffer_manager_test.cc
@@ -1039,6 +1078,7 @@ if(WITH_TESTS)
        options/options_settable_test.cc
        options/options_test.cc
        table/block_based/block_based_filter_block_test.cc
+        table/block_based/block_based_table_reader_test.cc
        table/block_based/block_test.cc
        table/block_based/data_block_hash_index_test.cc
        table/block_based/full_filter_block_test.cc
@@ -1049,6 +1089,7 @@ if(WITH_TESTS)
        table/merger_test.cc
        table/sst_file_reader_test.cc
        table/table_test.cc
+        table/block_fetcher_test.cc
        tools/block_cache_analyzer/block_cache_trace_analyzer_test.cc
        tools/ldb_cmd_test.cc
        tools/reduce_levels_test.cc
@@ -1070,8 +1111,10 @@ if(WITH_TESTS)
        util/slice_test.cc
        util/slice_transform_test.cc
        util/timer_queue_test.cc
+        util/timer_test.cc
        util/thread_list_test.cc
        util/thread_local_test.cc
+        util/work_queue_test.cc
        utilities/backupable/backupable_db_test.cc
        utilities/blob_db/blob_db_test.cc
        utilities/cassandra/cassandra_functional_test.cc
@@ -1091,6 +1134,7 @@ if(WITH_TESTS)
        utilities/table_properties_collectors/compact_on_deletion_collector_test.cc
        utilities/transactions/optimistic_transaction_test.cc
        utilities/transactions/transaction_test.cc
+        utilities/transactions/transaction_lock_mgr_test.cc
        utilities/transactions/write_prepared_transaction_test.cc
        utilities/transactions/write_unprepared_transaction_test.cc
        utilities/ttl/ttl_test.cc
@@ -1124,7 +1168,7 @@ if(WITH_TESTS)
        PROPERTIES EXCLUDE_FROM_DEFAULT_BUILD_RELEASE 1
        EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
        EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
-        )
+  )

  foreach(sourcefile ${TESTS})
      get_filename_component(exename ${sourcefile} NAME_WE)
@@ -1134,12 +1178,16 @@ if(WITH_TESTS)
        EXCLUDE_FROM_DEFAULT_BUILD_MINRELEASE 1
        EXCLUDE_FROM_DEFAULT_BUILD_RELWITHDEBINFO 1
        OUTPUT_NAME ${exename}${ARTIFACT_SUFFIX}
-        )
+      )
      target_link_libraries(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} testutillib${ARTIFACT_SUFFIX} testharness gtest ${ROCKSDB_LIB})
      if(NOT "${exename}" MATCHES "db_sanity_test")
        add_test(NAME ${exename} COMMAND ${exename}${ARTIFACT_SUFFIX})
        add_dependencies(check ${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX})
      endif()
+      if("${exename}" MATCHES "env_librados_test")
+        # env_librados_test.cc uses librados directly
+        target_link_libraries(${CMAKE_PROJECT_NAME}_${exename}${ARTIFACT_SUFFIX} rados)
+      endif()
  endforeach(sourcefile ${TESTS})

  if(WIN32)
@@ -1163,7 +1211,6 @@ if(WITH_TESTS)
  endif()
 endif()

-option(WITH_BENCHMARK_TOOLS "build with benchmarks" ON)
 if(WITH_BENCHMARK_TOOLS)
  add_executable(db_bench
    tools/db_bench.cc
@@ -1202,8 +1249,6 @@ if(WITH_BENCHMARK_TOOLS)
    ${ROCKSDB_LIB})
 endif()

-option(WITH_CORE_TOOLS "build with ldb and sst_dump" ON)
-option(WITH_TOOLS "build with tools" ON)
 if(WITH_CORE_TOOLS OR WITH_TOOLS)
  add_subdirectory(tools)
  add_custom_target(core_tools
@@ -1,17 +1,62 @@
 # Rocksdb Change Log
 ## Unreleased
+### Behavior Changes
+* Disable delete triggered compaction (NewCompactOnDeletionCollectorFactory) in universal compaction mode and num_levels = 1 in order to avoid a corruption bug.
+
+### Bug Fixes
+* Fix consistency checking error swallowing in some cases when options.force_consistency_checks = true.
+* Fix possible false NotFound status from batched MultiGet using index type kHashSearch.
+* Fix corruption caused by enabling delete triggered compaction (NewCompactOnDeletionCollectorFactory) in universal compaction mode, along with parallel compactions. The bug can result in two parallel compactions picking the same input files, resulting in the DB resurrecting older and deleted versions of some keys.
+* Fix a use-after-free bug in best-efforts recovery. column_family_memtables_ needs to point to valid ColumnFamilySet.
+* Let best-efforts recovery ignore corrupted files during table loading.
+* Fix corrupt key read from ingested file when iterator direction switches from reverse to forward at a key that is a prefix of another key in the same file. It is only possible in files with a non-zero global seqno.
+* Fix abnormally large estimate from GetApproximateSizes when a range starts near the end of one SST file and near the beginning of another. Now GetApproximateSizes consistently and fairly includes the size of SST metadata in addition to data blocks, attributing metadata proportionally among the data blocks based on their size.
+* Fix potential file descriptor leakage in PosixEnv's IsDirectory() and NewRandomAccessFile().

 ### Public API Change
-* Add NewFileChecksumGenCrc32cFactory to the file checksum public API, such that the builtin Crc32c based file checksum generator factory can be used by applications.
+* Flush(..., column_family) may return Status::ColumnFamilyDropped() instead of Status::InvalidArgument() if column_family is dropped while processing the flush request.
+* BlobDB now explicitly disallows using the default column family's storage directories as blob directory.
+* DeleteRange now returns `Status::InvalidArgument` if the range's end key comes before its start key according to the user comparator. Previously the behavior was undefined.
+* ldb now uses options.force_consistency_checks = true by default and "--disable_consistency_checks" is added to disable it.
+* DB::OpenForReadOnly no longer creates files or directories if the named DB does not exist, unless create_if_missing is set to true.
+* The consistency checks that validate LSM state changes (table file additions/deletions during flushes and compactions) are now stricter, more efficient, and no longer optional, i.e. they are performed even if `force_consistency_checks` is `false`.

-### New Features
-* Log CompactOnDeletionCollectorFactory window_size and deletion_trigger parameters in the info log file for troubleshooting purposes.
-* Report kFilesMarkedForCompaction for delete triggered compactions in case of Universal and Levelled Compactions.
+### New Feature
+* sst_dump to add a new --readahead_size argument. Users can specify read size when scanning the data. Sst_dump also tries to prefetch tail part of the SST files so usually some number of I/Os are saved there too.
+* Generate file checksum in SstFileWriter if Options.file_checksum_gen_factory is set. The checksum and checksum function name are stored in ExternalSstFileInfo after the sst file write is finished.
+* Add a value_size_soft_limit in read options which limits the cumulative value size of keys read in batches in MultiGet. Once the cumulative value size of found keys exceeds read_options.value_size_soft_limit, all the remaining keys are returned with status Abort without further finding their values. By default the value_size_soft_limit is std::numeric_limits<uint64_t>::max().

-## 6.9.1 (04/09/2020)
+## 6.10 (5/2/2020)
 ### Bug Fixes
 * Fix wrong result being read from ingested file. May happen when a key in the file happen to be prefix of another key also in the file. The issue can further cause more data corruption. The issue exists with rocksdb >= 5.0.0 since DB::IngestExternalFile() was introduced.
+* Finish implementation of BlockBasedTableOptions::IndexType::kBinarySearchWithFirstKey. It's now ready for use. Significantly reduces read amplification in some setups, especially for iterator seeks.
+* Fix a bug by updating CURRENT file so that it points to the correct MANIFEST file after best-efforts recovery.
+* Fixed a bug where ColumnFamilyHandle objects were not cleaned up in case an error happened during BlobDB's open after the base DB had been opened.
+* Fix a potential undefined behavior caused by trying to dereference nullable pointer (timestamp argument) in DB::MultiGet.
+* Fix a bug caused by not including user timestamp in MultiGet LookupKey construction. This can lead to wrong query result since the trailing bytes of a user key, if not shorter than timestamp, will be mistaken for user timestamp.
+* Fix a bug caused by using wrong compare function when sorting the input keys of MultiGet with timestamps.
+* Upgraded version of bzip library (1.0.6 -> 1.0.8) used with RocksJava to address potential vulnerabilities if an attacker can manipulate compressed data saved and loaded by RocksDB (not normal). See issue #6703.
+
+### Public API Change
+* Add a ConfigOptions argument to the APIs dealing with converting options to and from strings and files.  The ConfigOptions is meant to replace some of the options (such as input_strings_escaped and ignore_unknown_options) and allow for more parameters to be passed in the future without changing the function signature.
+* Add NewFileChecksumGenCrc32cFactory to the file checksum public API, such that the builtin Crc32c based file checksum generator factory can be used by applications.
+* Add IsDirectory to Env and FS to indicate if a path is a directory.
+
+### New Features
+* Added support for pipelined & parallel compression optimization for `BlockBasedTableBuilder`. This optimization makes block building, block compression and block appending a pipeline, and uses multiple threads to accelerate block compression. Users can set `CompressionOptions::parallel_threads` greater than 1 to enable compression parallelism. This feature is experimental for now.
+* Provide an allocator for memkind to be used with block cache. This is to work with memory technologies (Intel DCPMM is one such technology currently available) that require different libraries for allocation and management (such as PMDK and memkind). The high capacities available make it possible to provision large caches (up to several TBs in size) beyond what is achievable with DRAM.
+* Option `max_background_flushes` can be set dynamically using DB::SetDBOptions().
+* Added functionality in sst_dump tool to check the compressed file size for different compression levels and print the time spent on compressing files with each compression type. Added arguments `--compression_level_from` and `--compression_level_to` to report size of all compression levels and one compression_type must be specified with it so that it will report compressed sizes of one compression type with different levels.
+* Added statistics for redundant insertions into block cache: rocksdb.block.cache.*add.redundant. (There is currently no coordination to ensure that only one thread loads a table block when many threads are trying to access that same table block.)
+
+### Bug Fixes
 * Fix a bug when making options.bottommost_compression, options.compression_opts and options.bottommost_compression_opts dynamically changeable: the modified values are not written to option files or returned back to users when being queried.
+* Fix a bug where index key comparisons were unaccounted in `PerfContext::user_key_comparison_count` for lookups in files written with `format_version >= 3`.
+* Fix many bloom.filter statistics not being updated in batch MultiGet.
+
+### Performance Improvements
+* Improve performance of batch MultiGet with partitioned filters, by sharing block cache lookups to applicable filter blocks.
+* Reduced memory copies when fetching and uncompressing compressed blocks from sst files.

 ## 6.9.0 (03/29/2020)
 ### Behavior changes
@@ -8,7 +8,9 @@

 BASH_EXISTS := $(shell which bash)
 SHELL := $(shell which bash)
-PYTHON?=$(shell which python)
+# Default to python3. Some distros like CentOS 8 do not have `python`.
+PYTHON?=$(shell which python3 || which python || echo python3)
+export PYTHON

 CLEAN_FILES = # deliberately empty, so we can append below.
 CFLAGS += ${EXTRA_CFLAGS}
@@ -166,6 +168,12 @@ else
 	CXXFLAGS += -fno-rtti
 endif

+ifdef ASSERT_STATUS_CHECKED
+ifeq ($(filter -DROCKSDB_ASSERT_STATUS_CHECKED,$(OPT)),)
+	OPT += -DROCKSDB_ASSERT_STATUS_CHECKED
+endif
+endif
+
 $(warning Warning: Compiling in debug mode. Don't use the resulting binary in production)
 endif

@@ -187,12 +195,16 @@ AM_V_CC = $(am__v_CC_$(V))
 am__v_CC_ = $(am__v_CC_$(AM_DEFAULT_VERBOSITY))
 am__v_CC_0 = @echo "  CC      " $@;
 am__v_CC_1 =
-CCLD = $(CC)
-LINK = $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o $@
+
 AM_V_CCLD = $(am__v_CCLD_$(V))
 am__v_CCLD_ = $(am__v_CCLD_$(AM_DEFAULT_VERBOSITY))
+ifneq ($(SKIP_LINK), 1)
 am__v_CCLD_0 = @echo "  CCLD    " $@;
 am__v_CCLD_1 =
+else
+am__v_CCLD_0 = @echo "  !CCLD   " $@; true skip
+am__v_CCLD_1 = true skip
+endif
 AM_V_AR = $(am__v_AR_$(V))
 am__v_AR_ = $(am__v_AR_$(AM_DEFAULT_VERBOSITY))
 am__v_AR_0 = @echo "  AR      " $@;
@@ -204,8 +216,18 @@ LDFLAGS += -lrados
 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)"; export PORTABLE="$(PORTABLE)"; "$(CURDIR)/build_tools/build_detect_platform" "$(CURDIR)/make_config.mk"))
+
+# Detect what platform we're building on.
+# Export some common variables that might have been passed as Make variables
+# instead of environment variables.
+dummy := $(shell (export ROCKSDB_ROOT="$(CURDIR)"; \
+                  export COMPILE_WITH_ASAN="$(COMPILE_WITH_ASAN)"; \
+                  export COMPILE_WITH_TSAN="$(COMPILE_WITH_TSAN)"; \
+                  export COMPILE_WITH_UBSAN="$(COMPILE_WITH_UBSAN)"; \
+                  export PORTABLE="$(PORTABLE)"; \
+                  export ROCKSDB_NO_FBCODE="$(ROCKSDB_NO_FBCODE)"; \
+                  export USE_CLANG="$(USE_CLANG)"; \
+                  "$(CURDIR)/build_tools/build_detect_platform" "$(CURDIR)/make_config.mk"))
 # this file is generated by the previous line to set build flags and sources
 include make_config.mk
 export JAVAC_ARGS
@@ -312,7 +334,7 @@ endif

 export GTEST_THROW_ON_FAILURE=1
 export GTEST_HAS_EXCEPTIONS=1
-GTEST_DIR = ./third-party/gtest-1.8.1/fused-src
+GTEST_DIR = third-party/gtest-1.8.1/fused-src
 # AIX: pre-defined system headers are surrounded by an extern "C" block
 ifeq ($(PLATFORM), OS_AIX)
 	PLATFORM_CCFLAGS += -I$(GTEST_DIR)
@@ -441,7 +463,12 @@ BENCHTOOLOBJECTS = $(BENCH_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)

 ANALYZETOOLOBJECTS = $(ANALYZER_LIB_SOURCES:.cc=.o)

+ifeq ($(DEBUG_LEVEL),0)
 STRESSTOOLOBJECTS = $(STRESS_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
+else
+STRESSTOOLOBJECTS = $(STRESS_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL) \
+	$(TESTHARNESS)
+endif

 EXPOBJECTS = $(LIBOBJECTS) $(TESTUTIL)

@@ -466,6 +493,7 @@ TESTS = \
 	hash_test \
 	random_test \
 	thread_local_test \
+	work_queue_test \
 	rate_limiter_test \
 	perf_context_test \
 	iostats_context_test \
@@ -504,6 +532,7 @@ TESTS = \
 	column_family_test \
 	table_properties_collector_test \
 	arena_test \
+	memkind_kmem_allocator_test \
 	block_test \
 	data_block_hash_index_test \
 	cache_test \
@@ -517,6 +546,7 @@ TESTS = \
 	random_access_file_reader_test \
 	file_reader_writer_test \
 	block_based_filter_block_test \
+	block_based_table_reader_test \
 	full_filter_block_test \
 	partitioned_filter_block_test \
 	hash_table_test \
@@ -559,6 +589,7 @@ TESTS = \
 	deletefile_test \
 	obsolete_files_test \
 	table_test \
+	block_fetcher_test \
 	delete_scheduler_test \
 	options_test \
 	options_settable_test \
@@ -583,6 +614,7 @@ TESTS = \
 	compaction_job_stats_test \
 	option_change_migration_test \
 	transaction_test \
+	transaction_lock_mgr_test \
 	ldb_cmd_test \
 	persistent_cache_test \
 	statistics_test \
@@ -605,10 +637,8 @@ TESTS = \
 	defer_test \
 	blob_file_addition_test \
 	blob_file_garbage_test \
-
-ifeq ($(USE_FOLLY_DISTRIBUTED_MUTEX),1)
-	TESTS += folly_synchronization_distributed_mutex_test
-endif
+	timer_test \
+	db_with_timestamp_compaction_test \

 PARALLEL_TEST = \
 	backupable_db_test \
@@ -629,13 +659,71 @@ PARALLEL_TEST = \
 	persistent_cache_test \
 	table_test \
 	transaction_test \
+	transaction_lock_mgr_test \
 	write_prepared_transaction_test \
 	write_unprepared_transaction_test \

+ifeq ($(USE_FOLLY_DISTRIBUTED_MUTEX),1)
+	TESTS += folly_synchronization_distributed_mutex_test
+	PARALLEL_TEST += folly_synchronization_distributed_mutex_test
+endif
+
 # options_settable_test doesn't pass with UBSAN as we use hack in the test
 ifdef COMPILE_WITH_UBSAN
        TESTS := $(shell echo $(TESTS) | sed 's/\boptions_settable_test\b//g')
 endif
+ifdef ASSERT_STATUS_CHECKED
+	# This is a new check for which we will add support incrementally. The
+	# whitelist can be removed once support is fully added.
+	TESTS_WHITELIST = \
+		arena_test \
+		autovector_test \
+		blob_file_addition_test \
+		blob_file_garbage_test \
+		bloom_test \
+		cassandra_format_test \
+		cassandra_row_merge_test \
+		cassandra_serialize_test \
+		cleanable_test \
+		coding_test \
+		crc32c_test \
+		dbformat_test \
+		defer_test \
+		dynamic_bloom_test \
+		event_logger_test \
+		file_indexer_test \
+		folly_synchronization_distributed_mutex_test \
+		hash_table_test \
+		hash_test \
+		heap_test \
+		histogram_test \
+		inlineskiplist_test \
+		io_posix_test \
+		iostats_context_test \
+		memkind_kmem_allocator_test \
+		merger_test \
+		mock_env_test \
+		object_registry_test \
+		options_settable_test \
+		options_test \
+		random_test \
+		range_del_aggregator_test \
+		range_tombstone_fragmenter_test \
+		repeatable_thread_test \
+		skiplist_test \
+		slice_test \
+		statistics_test \
+		thread_local_test \
+		timer_queue_test \
+		timer_test \
+		util_merge_operators_test \
+		version_edit_test \
+		work_queue_test \
+		write_controller_test \
+
+	TESTS := $(filter $(TESTS_WHITELIST),$(TESTS))
+	PARALLEL_TEST := $(filter $(TESTS_WHITELIST),$(PARALLEL_TEST))
+endif
 SUBSET := $(TESTS)
 ifdef ROCKSDBTESTS_START
        SUBSET := $(shell echo $(SUBSET) | sed 's/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/')
@@ -934,6 +1022,9 @@ CLEAN_FILES += t LOG $(TMPD)
 watch-log:
 	$(WATCH) --interval=0 'sort -k7,7nr -k4,4gr LOG|$(quoted_perl_command)'

+dump-log:
+	bash -c '$(quoted_perl_command)' < LOG
+
 # If J != 1 and GNU parallel is installed, run the tests in parallel,
 # via the check_0 rule above.  Otherwise, run them sequentially.
 check: all
@@ -945,20 +1036,24 @@ check: all
 	    $(MAKE) T="$$t" TMPD=$(TMPD) check_0;                       \
 	else                                                            \
 	    for t in $(TESTS); do                                       \
-	      echo "===== Running $$t"; ./$$t || exit 1; done;          \
+	      echo "===== Running $$t (`date`)"; ./$$t || exit 1; done;          \
 	fi
 	rm -rf $(TMPD)
 ifneq ($(PLATFORM), OS_AIX)
 	$(PYTHON) tools/check_all_python.py
 ifeq ($(filter -DROCKSDB_LITE,$(OPT)),)
+ifndef ASSERT_STATUS_CHECKED # not yet working with these tests
 	$(PYTHON) tools/ldb_test.py
 	sh tools/rocksdb_dump_test.sh
 endif
 endif
+endif
+	$(MAKE) check-format
+	$(MAKE) check-buck-targets

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

 .PHONY: ldb_tests
 ldb_tests: ldb
@@ -1064,7 +1159,7 @@ ifneq ($(PAR_TEST),)
 parloop:
 	ret_bad=0;							\
 	for t in $(PAR_TEST); do		\
-		echo "===== Running $$t in parallel $(NUM_PAR)";\
+		echo "===== Running $$t in parallel $(NUM_PAR) (`date`)";\
 		if [ $(db_test) -eq 1 ]; then \
 			seq $(J) | v="$$t" build_tools/gnu_parallel --gnu --plain 's=$(TMPD)/rdb-{};  export TEST_TMPDIR=$$s;' \
 				'timeout 2m ./db_test --gtest_filter=$$v >> $$s/log-{} 2>1'; \
@@ -1175,6 +1270,12 @@ tags0:
 format:
 	build_tools/format-diff.sh

+check-format:
+	build_tools/format-diff.sh -c
+
+check-buck-targets:
+	buckifier/check_buck_targets.sh
+
 package:
 	bash build_tools/make_package.sh $(SHARED_MAJOR).$(SHARED_MINOR)

@@ -1238,6 +1339,9 @@ db_repl_stress: tools/db_repl_stress.o $(LIBOBJECTS) $(TESTUTIL)
 arena_test: memory/arena_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+memkind_kmem_allocator_test: memory/memkind_kmem_allocator_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 autovector_test: util/autovector_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1295,6 +1399,9 @@ histogram_test: monitoring/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS)
 thread_local_test: util/thread_local_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+work_queue_test: util/work_queue_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 corruption_test: db/corruption_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1313,6 +1420,9 @@ db_basic_test: db/db_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 db_with_timestamp_basic_test: db/db_with_timestamp_basic_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+db_with_timestamp_compaction_test: db/db_with_timestamp_compaction_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_encryption_test: db/db_encryption_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1516,6 +1626,9 @@ file_reader_writer_test: util/file_reader_writer_test.o $(LIBOBJECTS) $(TESTHARN
 block_based_filter_block_test: table/block_based/block_based_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+block_based_table_reader_test: table/block_based/block_based_table_reader_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 full_filter_block_test: table/block_based/full_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1531,6 +1644,9 @@ cleanable_test: table/cleanable_test.o $(LIBOBJECTS) $(TESTHARNESS)
 table_test: table/table_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+block_fetcher_test: table/block_fetcher_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 block_test: table/block_based/block_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1669,6 +1785,9 @@ write_callback_test: db/write_callback_test.o $(LIBOBJECTS) $(TESTHARNESS)
 heap_test: util/heap_test.o $(GTEST)
 	$(AM_LINK)

+transaction_lock_mgr_test: utilities/transactions/transaction_lock_mgr_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1744,6 +1863,9 @@ blob_file_addition_test: db/blob/blob_file_addition_test.o $(LIBOBJECTS) $(TESTH
 blob_file_garbage_test: db/blob/blob_file_garbage_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+timer_test: util/timer_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1823,9 +1945,9 @@ SHA256_CMD = sha256sum
 ZLIB_VER ?= 1.2.11
 ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
-BZIP2_VER ?= 1.0.6
-BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-BZIP2_DOWNLOAD_BASE ?= https://downloads.sourceforge.net/project/bzip2
+BZIP2_VER ?= 1.0.8
+BZIP2_SHA256 ?= ab5a03176ee106d3f0fa90e381da478ddae405918153cca248e682cd0c4a2269
+BZIP2_DOWNLOAD_BASE ?= https://sourceware.org/pub/bzip2
 SNAPPY_VER ?= 1.1.8
 SNAPPY_SHA256 ?= 16b677f07832a612b0836178db7f374e414f94657c138e6993cbfc5dcc58651f
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
@@ -2152,6 +2274,7 @@ endif
 #  	Source files dependencies detection
 # ---------------------------------------------------------------------------

+# FIXME: nothing checks that entries in MAIN_SOURCES actually exist
 all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(ANALYZER_LIB_SOURCES) $(STRESS_LIB_SOURCES)
 DEPFILES = $(all_sources:.cc=.cc.d)

@@ -24,7 +24,7 @@ The public interface is in `include/`.  Callers should not include or
 rely on the details of any other header files in this package.  Those
 internal APIs may be changed without warning.

-Design discussions are conducted in https://www.facebook.com/groups/rocksdb.dev/
+Design discussions are conducted in https://www.facebook.com/groups/rocksdb.dev/ and https://rocksdb.slack.com/

 ## License

@@ -1,4 +1,4 @@
-# This file @generated by `python buckifier/buckify_rocksdb.py`
+# This file @generated by `python3 buckifier/buckify_rocksdb.py`
 # --> DO NOT EDIT MANUALLY <--
 # This file is a Facebook-specific integration for buck builds, so can
 # only be validated by Facebook employees.
@@ -26,7 +26,6 @@ ROCKSDB_EXTERNAL_DEPS = [
    ("lz4", None, "lz4"),
    ("zstd", None),
    ("tbb", None),
-    ("googletest", None, "gtest"),
 ]

 ROCKSDB_OS_DEPS = [
@@ -79,6 +78,7 @@ ROCKSDB_PREPROCESSOR_FLAGS = [
    # Directories with files for #include
    "-I" + REPO_PATH + "include/",
    "-I" + REPO_PATH,
+    "-I" + REPO_PATH + "third-party/gtest-1.8.1/fused-src/",
 ]

 ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
@@ -109,9 +109,15 @@ ROCKSDB_OS_DEPS += ([(
    ["third-party//jemalloc:headers"],
 )] if sanitizer == "" else [])

+ROCKSDB_LIB_DEPS = [
+    ":rocksdb_lib",
+    ":rocksdb_test_lib",
+] if not is_opt_mode else [":rocksdb_lib"]
+
 cpp_library(
    name = "rocksdb_lib",
    srcs = [
+        "cache/cache.cc",
        "cache/clock_cache.cc",
        "cache/lru_cache.cc",
        "cache/sharded_cache.cc",
@@ -204,6 +210,7 @@ cpp_library(
        "memory/arena.cc",
        "memory/concurrent_arena.cc",
        "memory/jemalloc_nodump_allocator.cc",
+        "memory/memkind_kmem_allocator.cc",
        "memtable/alloc_tracker.cc",
        "memtable/hash_linklist_rep.cc",
        "memtable/hash_skiplist_rep.cc",
@@ -229,7 +236,6 @@ cpp_library(
        "options/options.cc",
        "options/options_helper.cc",
        "options/options_parser.cc",
-        "options/options_sanity_check.cc",
        "port/port_posix.cc",
        "port/stack_trace.cc",
        "table/adaptive/adaptive_table_factory.cc",
@@ -392,7 +398,10 @@ cpp_library(
    os_deps = ROCKSDB_OS_DEPS,
    os_preprocessor_flags = ROCKSDB_OS_PREPROCESSOR_FLAGS,
    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
-    deps = [":rocksdb_lib"],
+    deps = [
+        ":rocksdb_lib",
+        ":rocksdb_third_party_gtest",
+    ],
    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

@@ -436,7 +445,20 @@ cpp_library(
    os_deps = ROCKSDB_OS_DEPS,
    os_preprocessor_flags = ROCKSDB_OS_PREPROCESSOR_FLAGS,
    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
-    deps = [":rocksdb_lib"],
+    deps = ROCKSDB_LIB_DEPS,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
+)
+
+cpp_library(
+    name = "rocksdb_third_party_gtest",
+    srcs = ["third-party/gtest-1.8.1/fused-src/gtest/gtest-all.cc"],
+    headers = ["third-party/gtest-1.8.1/fused-src/gtest/gtest.h"],
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    os_deps = ROCKSDB_OS_DEPS,
+    os_preprocessor_flags = ROCKSDB_OS_PREPROCESSOR_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+    deps = [],
    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )

@@ -511,6 +533,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "block_based_table_reader_test",
+        "table/block_based/block_based_table_reader_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "block_cache_trace_analyzer_test",
        "tools/block_cache_analyzer/block_cache_trace_analyzer_test.cc",
@@ -525,6 +554,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "block_fetcher_test",
+        "table/block_fetcher_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "block_test",
        "table/block_based/block_test.cc",
@@ -931,6 +967,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "db_with_timestamp_compaction_test",
+        "db/db_with_timestamp_compaction_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "db_write_test",
        "db/db_write_test.cc",
@@ -1169,6 +1212,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "memkind_kmem_allocator_test",
+        "memory/memkind_kmem_allocator_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "memory_test",
        "utilities/memory/memory_test.cc",
@@ -1456,6 +1506,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "timer_test",
+        "util/timer_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "trace_analyzer_test",
        "tools/trace_analyzer_test.cc",
@@ -1463,6 +1520,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "transaction_lock_mgr_test",
+        "utilities/transactions/transaction_lock_mgr_test.cc",
+        "parallel",
+        [],
+        [],
+    ],
    [
        "transaction_test",
        "utilities/transactions/transaction_test.cc",
@@ -1512,6 +1576,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "work_queue_test",
+        "util/work_queue_test.cc",
+        "serial",
+        [],
+        [],
+    ],
    [
        "write_batch_test",
        "db/write_batch_test.cc",
@@ -106,3 +106,6 @@ LzLabs is using RocksDB as a storage engine in their multi-database distributed
 ## Nebula Graph

 [Nebula Graph](https://github.com/vesoft-inc/nebula) is a distributed, scalable, lightning-fast, open source graph database capable of hosting super large scale graphs with dozens of billions of vertices (nodes) and trillions of edges, with milliseconds of latency.
+
+## YugabyteDB
+[YugabyteDB](https://www.yugabyte.com/) is an open source, high performance, distributed SQL database that uses RocksDB as its storage layer. For more information, please see https://github.com/yugabyte/yugabyte-db/.
@@ -1,6 +1,6 @@
 version: 1.0.{build}

-image: Visual Studio 2017
+image: Visual Studio 2019

 environment:
  JAVA_HOME: C:\Program Files\Java\jdk1.8.0
@@ -24,6 +24,15 @@ environment:
    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
      CMAKE_GENERATOR: Visual Studio 15 Win64
      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\devenv.com
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2019
+      CMAKE_GENERATOR: Visual Studio 16
+      CMAKE_PLATEFORM_NAME: x64
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\devenv.com
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2019
+      CMAKE_GENERATOR: Visual Studio 16
+      CMAKE_PLATEFORM_NAME: x64
+      CMAKE_OPT: -DCMAKE_CXX_STANDARD=20
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\devenv.com

 install:
  - md %THIRDPARTY_HOME%
@@ -34,7 +43,8 @@ install:
  - cd snappy-1.1.7
  - mkdir build
  - cd build
-  - cmake -G "%CMAKE_GENERATOR%" ..
+  - if DEFINED CMAKE_PLATEFORM_NAME (set "PLATEFORM_OPT=-A %CMAKE_PLATEFORM_NAME%")
+  - cmake .. -G "%CMAKE_GENERATOR%" %PLATEFORM_OPT%
  - msbuild Snappy.sln /p:Configuration=Debug /p:Platform=x64
  - msbuild Snappy.sln /p:Configuration=Release /p:Platform=x64
  - echo "Building LZ4 dependency..."
@@ -57,7 +67,8 @@ install:
 before_build:
  - md %APPVEYOR_BUILD_FOLDER%\build
  - cd %APPVEYOR_BUILD_FOLDER%\build
-  - cmake -G "%CMAKE_GENERATOR%" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DLZ4=1 -DZSTD=1 -DXPRESS=1 -DJNI=1 ..
+  - if DEFINED CMAKE_PLATEFORM_NAME (set "PLATEFORM_OPT=-A %CMAKE_PLATEFORM_NAME%")
+  - cmake .. -G "%CMAKE_GENERATOR%" %PLATEFORM_OPT% %CMAKE_OPT% -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DLZ4=1 -DZSTD=1 -DXPRESS=1 -DJNI=1
  - cd ..

 build:
@@ -20,10 +20,10 @@ from util import ColorString
 # User can pass extra dependencies as a JSON object via command line, and this
 # script can include these dependencies in the generate TARGETS file.
 # Usage:
-# $python buckifier/buckify_rocksdb.py
+# $python3 buckifier/buckify_rocksdb.py
 # (This generates a TARGET file without user-specified dependency for unit
 # tests.)
-# $python buckifier/buckify_rocksdb.py \
+# $python3 buckifier/buckify_rocksdb.py \
 #        '{"fake": { \
 #                      "extra_deps": [":test_dep", "//fakes/module:mock1"],  \
 #                      "extra_compiler_flags": ["-DROCKSDB_LITE", "-Os"], \
@@ -109,6 +109,15 @@ def get_tests(repo_path):
    return tests


+# Get gtest dir from Makefile
+def get_gtest_dir(repo_path):
+    for line in open(repo_path + "/Makefile"):
+        if line.strip().startswith("GTEST_DIR ="):
+            return line.split("=")[1].strip()
+    # if not found
+    exit_with_error("Unable to find GTEST_DIR in Makefile")
+
+
 # Parse extra dependencies passed by user from command line
 def get_dependencies():
    deps_map = {
@@ -142,11 +151,14 @@ def generate_targets(repo_path, deps_map):
    cc_files = get_cc_files(repo_path)
    # get tests from Makefile
    tests = get_tests(repo_path)
+    # get gtest dir
+    gtest_dir = get_gtest_dir(repo_path) + "/"

    if src_mk is None or cc_files is None or tests is None:
        return False

-    TARGETS = TARGETSBuilder("%s/TARGETS" % repo_path)
+    TARGETS = TARGETSBuilder("%s/TARGETS" % repo_path, gtest_dir)
+
    # rocksdb_lib
    TARGETS.add_library(
        "rocksdb_lib",
@@ -159,7 +171,7 @@ def generate_targets(repo_path, deps_map):
        src_mk.get("TEST_LIB_SOURCES", []) +
        src_mk.get("EXP_LIB_SOURCES", []) +
        src_mk.get("ANALYZER_LIB_SOURCES", []),
-        [":rocksdb_lib"])
+        [":rocksdb_lib", ":rocksdb_third_party_gtest"])
    # rocksdb_tools_lib
    TARGETS.add_library(
        "rocksdb_tools_lib",
@@ -168,14 +180,19 @@ def generate_targets(repo_path, deps_map):
        ["test_util/testutil.cc"],
        [":rocksdb_lib"])
    # rocksdb_stress_lib
-    TARGETS.add_library(
+    TARGETS.add_rocksdb_library(
        "rocksdb_stress_lib",
        src_mk.get("ANALYZER_LIB_SOURCES", [])
        + src_mk.get('STRESS_LIB_SOURCES', [])
-        + ["test_util/testutil.cc"],
-        [":rocksdb_lib"])
+        + ["test_util/testutil.cc"])
+    # rocksdb_third_party_gtest
+    TARGETS.add_library(
+        "rocksdb_third_party_gtest",
+        [gtest_dir + "gtest/gtest-all.cc"],
+        [],
+        [gtest_dir + "gtest/gtest.h"])

-    print("Extra dependencies:\n{0}".format(str(deps_map)))
+    print("Extra dependencies:\n{0}".format(json.dumps(deps_map)))
    # test for every test we found in the Makefile
    for target_alias, deps in deps_map.items():
        for test in sorted(tests):
@@ -196,8 +213,8 @@ def generate_targets(repo_path, deps_map):
                test_target_name,
                match_src[0],
                is_parallel,
-                deps['extra_deps'],
-                deps['extra_compiler_flags'])
+                json.dumps(deps['extra_deps']),
+                json.dumps(deps['extra_compiler_flags']))

            if test in _EXPORTED_TEST_LIBS:
                test_library = "%s_lib" % test_target_name
@@ -220,6 +237,7 @@ def get_rocksdb_path():

    return rocksdb_path

+
 def exit_with_error(msg):
    print(ColorString.error(msg))
    sys.exit(1)
@@ -0,0 +1,32 @@
+#!/usr/bin/env bash
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# If clang_format_diff.py command is not specfied, we assume we are able to
+# access directly without any path.
+
+TGT_DIFF=`git diff TARGETS | head -n 1`
+
+if [ ! -z "$TGT_DIFF" ]
+then
+  echo "TARGETS file has uncommitted changes. Skip this check."
+  exit 0
+fi
+
+echo Backup original TARGETS file.
+
+cp TARGETS TARGETS.bkp
+
+${PYTHON:-python3} buckifier/buckify_rocksdb.py
+
+TGT_DIFF=`git diff TARGETS | head -n 1`
+
+if [ -z "$TGT_DIFF" ]
+then
+  mv TARGETS.bkp TARGETS
+  exit 0
+else
+  echo "Please run '${PYTHON:-python3} buckifier/buckify_rocksdb.py' to update TARGETS file."
+  echo "Do not manually update TARGETS file."
+  ${PYTHON:-python3} --version
+  mv TARGETS.bkp TARGETS
+  exit 1
+fi
@@ -25,10 +25,12 @@ def pretty_list(lst, indent=8):


 class TARGETSBuilder(object):
-    def __init__(self, path):
+    def __init__(self, path, gtest_dir):
        self.path = path
        self.targets_file = open(path, 'w')
-        self.targets_file.write(targets_cfg.rocksdb_target_header)
+        header = targets_cfg.rocksdb_target_header_template.format(
+            gtest_dir=gtest_dir)
+        self.targets_file.write(header)
        self.total_lib = 0
        self.total_bin = 0
        self.total_test = 0
@@ -42,6 +44,8 @@ class TARGETSBuilder(object):
        if headers is None:
            headers_attr_prefix = "auto_"
            headers = "AutoHeaders.RECURSIVE_GLOB"
+        else:
+            headers = "[" + pretty_list(headers) + "]"
        self.targets_file.write(targets_cfg.library_template.format(
            name=name,
            srcs=pretty_list(srcs),
@@ -50,6 +54,20 @@ class TARGETSBuilder(object):
            deps=pretty_list(deps)))
        self.total_lib = self.total_lib + 1

+    def add_rocksdb_library(self, name, srcs, headers=None):
+        headers_attr_prefix = ""
+        if headers is None:
+            headers_attr_prefix = "auto_"
+            headers = "AutoHeaders.RECURSIVE_GLOB"
+        else:
+            headers = "[" + pretty_list(headers) + "]"
+        self.targets_file.write(targets_cfg.rocksdb_library_template.format(
+            name=name,
+            srcs=pretty_list(srcs),
+            headers_attr_prefix=headers_attr_prefix,
+            headers=headers))
+        self.total_lib = self.total_lib + 1
+
    def add_binary(self, name, srcs, deps=None):
        self.targets_file.write(targets_cfg.binary_template % (
            name,
@@ -4,7 +4,8 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals

-rocksdb_target_header = """# This file \100generated by `python buckifier/buckify_rocksdb.py`
+rocksdb_target_header_template = \
+    """# This file \100generated by `python3 buckifier/buckify_rocksdb.py`
 # --> DO NOT EDIT MANUALLY <--
 # This file is a Facebook-specific integration for buck builds, so can
 # only be validated by Facebook employees.
@@ -32,7 +33,6 @@ ROCKSDB_EXTERNAL_DEPS = [
    ("lz4", None, "lz4"),
    ("zstd", None),
    ("tbb", None),
-    ("googletest", None, "gtest"),
 ]

 ROCKSDB_OS_DEPS = [
@@ -85,13 +85,14 @@ ROCKSDB_PREPROCESSOR_FLAGS = [
    # Directories with files for #include
    "-I" + REPO_PATH + "include/",
    "-I" + REPO_PATH,
+    "-I" + REPO_PATH + "{gtest_dir}",
 ]

-ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
+ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {{
    "x86_64": [
        "-DHAVE_PCLMUL",
    ],
-}
+}}

 build_mode = read_config("fbcode", "build_mode")

@@ -114,6 +115,11 @@ ROCKSDB_OS_DEPS += ([(
    "linux",
    ["third-party//jemalloc:headers"],
 )] if sanitizer == "" else [])
+
+ROCKSDB_LIB_DEPS = [
+    ":rocksdb_lib",
+    ":rocksdb_test_lib",
+] if not is_opt_mode else [":rocksdb_lib"]
 """


@@ -132,6 +138,21 @@ cpp_library(
 )
 """

+rocksdb_library_template = """
+cpp_library(
+    name = "{name}",
+    srcs = [{srcs}],
+    {headers_attr_prefix}headers = {headers},
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    os_deps = ROCKSDB_OS_DEPS,
+    os_preprocessor_flags = ROCKSDB_OS_PREPROCESSOR_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+    deps = ROCKSDB_LIB_DEPS,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
+)
+"""
+
 binary_template = """
 cpp_binary(
    name = "%s",
@@ -28,6 +28,7 @@
 #       -DZSTD                      if the ZSTD library is present
 #       -DNUMA                      if the NUMA library is present
 #       -DTBB                       if the TBB library is present
+#       -DMEMKIND                   if the memkind library is present
 #
 # Using gflags in rocksdb:
 # Our project depends on gflags, which requires users to take some extra steps
@@ -245,6 +246,10 @@ JAVAC_ARGS="-source 7"
 if [ "$CROSS_COMPILE" = "true" -o "$FBCODE_BUILD" = "true" ]; then
    # Cross-compiling; do not try any compilation tests.
    # Also don't need any compilation tests if compiling on fbcode
+    if [ "$FBCODE_BUILD" = "true" ]; then
+      # Enable backtrace on fbcode since the necessary libraries are present
+      COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+    fi
    true
 else
    if ! test $ROCKSDB_DISABLE_FALLOCATE; then
@@ -425,7 +430,23 @@ EOF
            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
        fi
    fi
-
+    
+    if ! test $ROCKSDB_DISABLE_MEMKIND; then
+        # Test whether memkind library is installed
+        $CXX $CFLAGS $COMMON_FLAGS -lmemkind -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <memkind.h>
+          int main() {
+            memkind_malloc(MEMKIND_DAX_KMEM, 1024);
+            return 0;
+          }
+EOF
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DMEMKIND"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lmemkind"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lmemkind"
+        fi
+    fi
+    
    if ! test $ROCKSDB_DISABLE_PTHREAD_MUTEX_ADAPTIVE_NP; then
        # Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
        $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
@@ -684,6 +705,7 @@ EOF
  fi
 fi

+
 if [ "$FBCODE_BUILD" != "true" -a "$PLATFORM" = OS_LINUX ]; then
  $CXX $COMMON_FLAGS $PLATFORM_SHARED_CFLAGS -x c++ -c - -o test_dl.o 2>/dev/null <<EOF
  void dummy_func() {}
@@ -2,41 +2,99 @@
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
 # If clang_format_diff.py command is not specfied, we assume we are able to
 # access directly without any path.
-if [ -z $CLANG_FORMAT_DIFF ]
-then
-CLANG_FORMAT_DIFF="clang-format-diff.py"
-fi

-# Check clang-format-diff.py
-if ! which $CLANG_FORMAT_DIFF &> /dev/null
-then
-  echo "You didn't have clang-format-diff.py and/or clang-format available in your computer!"
-  echo "You can download clang-format-diff.py by running: "
-  echo "    curl --location http://goo.gl/iUW1u2 -o ${CLANG_FORMAT_DIFF}"
-  echo "You can download clang-format by running:"
-  echo "    brew install clang-format"
-  echo "  Or"
-  echo "    apt install clang-format"
-  echo "  This might work too:"
-  echo "    yum install git-clang-format"
-  echo "Then, move both files (i.e. ${CLANG_FORMAT_DIFF} and clang-format) to some directory within PATH=${PATH}"
-  echo "and make sure ${CLANG_FORMAT_DIFF} is executable."
-  exit 128
-fi
+print_usage () {
+  echo "Usage:"
+  echo "format-diff.sh [OPTIONS]"
+  echo "-c: check only."
+  echo "-h: print this message."
+}

-# Check argparse, a library that clang-format-diff.py requires.
-python 2>/dev/null << EOF
-import argparse
-EOF
+while getopts ':ch' OPTION; do
+  case "$OPTION" in
+    c)
+      CHECK_ONLY=1
+      ;;
+    h)
+      print_usage
+      exit 1
+      ;;
+    ?)
+      print_usage
+      exit 1
+      ;;
+  esac
+done

-if [ "$?" != 0 ]
-then
-  echo "To run clang-format-diff.py, we'll need the library "argparse" to be"
-  echo "installed. You can try either of the follow ways to install it:"
-  echo "  1. Manually download argparse: https://pypi.python.org/pypi/argparse"
-  echo "  2. easy_install argparse (if you have easy_install)"
-  echo "  3. pip install argparse (if you have pip)"
-  exit 129
+REPO_ROOT="$(git rev-parse --show-toplevel)"
+
+if [ "$CLANG_FORMAT_DIFF" ]; then
+  echo "Note: CLANG_FORMAT_DIFF='$CLANG_FORMAT_DIFF'"
+  # Dry run to confirm dependencies like argparse
+  if $CLANG_FORMAT_DIFF --help >/dev/null < /dev/null; then
+    true #Good
+  else
+    exit 128
+  fi
+else
+  # First try directly executing the possibilities
+  if clang-format-diff.py --help &> /dev/null < /dev/null; then
+    CLANG_FORMAT_DIFF=clang-format-diff.py
+  elif $REPO_ROOT/clang-format-diff.py --help &> /dev/null < /dev/null; then
+    CLANG_FORMAT_DIFF=$REPO_ROOT/clang-format-diff.py
+  else
+    # This probably means we need to directly invoke the interpreter.
+    # But first find clang-format-diff.py
+    if [ -f "$REPO_ROOT/clang-format-diff.py" ]; then
+      CFD_PATH="$REPO_ROOT/clang-format-diff.py"
+    elif which clang-format-diff.py &> /dev/null; then
+      CFD_PATH="$(which clang-format-diff.py)"
+    else
+      echo "You didn't have clang-format-diff.py and/or clang-format available in your computer!"
+      echo "You can download clang-format-diff.py by running: "
+      echo "    curl --location http://goo.gl/iUW1u2 -o ${CLANG_FORMAT_DIFF}"
+      echo "You can download clang-format by running:"
+      echo "    brew install clang-format"
+      echo "  Or"
+      echo "    apt install clang-format"
+      echo "  This might work too:"
+      echo "    yum install git-clang-format"
+      echo "Then, move both files (i.e. ${CLANG_FORMAT_DIFF} and clang-format) to some directory within PATH=${PATH}"
+      echo "and make sure ${CLANG_FORMAT_DIFF} is executable."
+      exit 128
+    fi
+    # Check argparse pre-req on interpreter, or it will fail
+    if echo import argparse | ${PYTHON:-python3}; then
+      true # Good
+    else
+      echo "To run clang-format-diff.py, we'll need the library "argparse" to be"
+      echo "installed. You can try either of the follow ways to install it:"
+      echo "  1. Manually download argparse: https://pypi.python.org/pypi/argparse"
+      echo "  2. easy_install argparse (if you have easy_install)"
+      echo "  3. pip install argparse (if you have pip)"
+      exit 129
+    fi
+    # Unfortunately, some machines have a Python2 clang-format-diff.py
+    # installed but only a Python3 interpreter installed. Rather than trying
+    # different Python versions that might be installed, we can try migrating
+    # the code to Python3 if it looks like Python2
+    if grep -q "print '" "$CFD_PATH" && \
+       ${PYTHON:-python3} --version | grep -q 'ython 3'; then
+      if [ ! -f "$REPO_ROOT/.py3/clang-format-diff.py" ]; then
+        echo "Migrating $CFD_PATH to Python3 in a hidden file"
+        mkdir -p "$REPO_ROOT/.py3"
+        ${PYTHON:-python3} -m lib2to3 -w -n -o "$REPO_ROOT/.py3" "$CFD_PATH" > /dev/null || exit 128
+      fi
+      CFD_PATH="$REPO_ROOT/.py3/clang-format-diff.py"
+    fi
+    CLANG_FORMAT_DIFF="${PYTHON:-python3} $CFD_PATH"
+    # This had better work after all those checks
+    if $CLANG_FORMAT_DIFF --help >/dev/null < /dev/null; then
+      true #Good
+    else
+      exit 128
+    fi
+  fi
 fi

 # TODO(kailiu) following work is not complete since we still need to figure
@@ -87,6 +145,14 @@ if [ -z "$diffs" ]
 then
  echo "Nothing needs to be reformatted!"
  exit 0
+elif [ $CHECK_ONLY ]
+then
+  echo "Your change has unformatted code. Please run make format!"
+  if [ $VERBOSE_CHECK ]; then
+    clang-format --version
+    echo "$diffs"
+  fi
+  exit 1
 fi

 # Highlight the insertion/deletion from the clang-format-diff.py's output
@@ -121,7 +187,7 @@ if [ -z "$uncommitted_code" ]
 then
  git diff -U0 "$FORMAT_UPSTREAM_MERGE_BASE" | $CLANG_FORMAT_DIFF -i -p 1
 else
-  git diff -U0 HEAD^ | $CLANG_FORMAT_DIFF -i -p 1
+  git diff -U0 HEAD | $CLANG_FORMAT_DIFF -i -p 1
 fi
 echo "Files reformatted!"

@@ -818,6 +818,8 @@ run_format_compatible()
  rm -rf /dev/shm/rocksdb
  mkdir /dev/shm/rocksdb

+  export https_proxy="fwdproxy:8080"
+  
  tools/check_format_compatible.sh
 }

@@ -1,9 +1,9 @@
 #!/bin/bash
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-set -e
+set -ex

-ROCKSDB_VERSION="5.10.3"
-ZSTD_VERSION="1.1.3"
+ROCKSDB_VERSION="6.7.3"
+ZSTD_VERSION="1.4.4"

 echo "This script configures CentOS with everything needed to build and run RocksDB"

@@ -40,5 +40,6 @@ cd /usr/local/rocksdb
 chown -R vagrant:vagrant /usr/local/rocksdb/
 sudo -u vagrant make static_lib
 cd examples/
-sudo -u vagrant make all
-sudo -u vagrant ./c_simple_example
+sudo -u vagrant LD_LIBRARY_PATH=/usr/local/lib/ make all
+sudo -u vagrant LD_LIBRARY_PATH=/usr/local/lib/ ./c_simple_example
+
@@ -0,0 +1,66 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// 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 "rocksdb/cache.h"
+
+#include "cache/lru_cache.h"
+#include "options/options_helper.h"
+#include "util/string_util.h"
+
+namespace ROCKSDB_NAMESPACE {
+#ifndef ROCKSDB_LITE
+static std::unordered_map<std::string, OptionTypeInfo>
+    lru_cache_options_type_info = {
+        {"capacity",
+         {offsetof(struct LRUCacheOptions, capacity), OptionType::kSizeT,
+          OptionVerificationType::kNormal, OptionTypeFlags::kMutable,
+          offsetof(struct LRUCacheOptions, capacity)}},
+        {"num_shard_bits",
+         {offsetof(struct LRUCacheOptions, num_shard_bits), OptionType::kInt,
+          OptionVerificationType::kNormal, OptionTypeFlags::kMutable,
+          offsetof(struct LRUCacheOptions, num_shard_bits)}},
+        {"strict_capacity_limit",
+         {offsetof(struct LRUCacheOptions, strict_capacity_limit),
+          OptionType::kBoolean, OptionVerificationType::kNormal,
+          OptionTypeFlags::kMutable,
+          offsetof(struct LRUCacheOptions, strict_capacity_limit)}},
+        {"high_pri_pool_ratio",
+         {offsetof(struct LRUCacheOptions, high_pri_pool_ratio),
+          OptionType::kDouble, OptionVerificationType::kNormal,
+          OptionTypeFlags::kMutable,
+          offsetof(struct LRUCacheOptions, high_pri_pool_ratio)}}};
+#endif  // ROCKSDB_LITE
+
+Status Cache::CreateFromString(const ConfigOptions& config_options,
+                               const std::string& value,
+                               std::shared_ptr<Cache>* result) {
+  Status status;
+  std::shared_ptr<Cache> cache;
+  if (value.find('=') == std::string::npos) {
+    cache = NewLRUCache(ParseSizeT(value));
+  } else {
+#ifndef ROCKSDB_LITE
+    LRUCacheOptions cache_opts;
+    status = OptionTypeInfo::ParseStruct(
+        config_options, "", &lru_cache_options_type_info, "", value,
+        reinterpret_cast<char*>(&cache_opts));
+    if (status.ok()) {
+      cache = NewLRUCache(cache_opts);
+    }
+#else
+    (void)config_options;
+    status = Status::NotSupported("Cannot load cache in LITE mode ", value);
+#endif  //! ROCKSDB_LITE
+  }
+  if (status.ok()) {
+    result->swap(cache);
+  }
+  return status;
+}
+}  // namespace ROCKSDB_NAMESPACE
@@ -14,34 +14,47 @@ int main() {
 #include <stdio.h>
 #include <sys/types.h>
 #include <cinttypes>
+#include <limits>

 #include "port/port.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
+#include "util/coding.h"
 #include "util/gflags_compat.h"
+#include "util/hash.h"
 #include "util/mutexlock.h"
 #include "util/random.h"

 using GFLAGS_NAMESPACE::ParseCommandLineFlags;

-static const uint32_t KB = 1024;
+static constexpr uint32_t KiB = uint32_t{1} << 10;
+static constexpr uint32_t MiB = KiB << 10;
+static constexpr uint64_t GiB = MiB << 10;

-DEFINE_int32(threads, 16, "Number of concurrent threads to run.");
-DEFINE_int64(cache_size, 8 * KB * KB,
-             "Number of bytes to use as a cache of uncompressed data.");
-DEFINE_int32(num_shard_bits, 4, "shard_bits.");
+DEFINE_uint32(threads, 16, "Number of concurrent threads to run.");
+DEFINE_uint64(cache_size, 1 * GiB,
+              "Number of bytes to use as a cache of uncompressed data.");
+DEFINE_uint32(num_shard_bits, 6, "shard_bits.");

-DEFINE_int64(max_key, 1 * KB * KB * KB, "Max number of key to place in cache");
-DEFINE_uint64(ops_per_thread, 1200000, "Number of operations per thread.");
+DEFINE_double(resident_ratio, 0.25,
+              "Ratio of keys fitting in cache to keyspace.");
+DEFINE_uint64(ops_per_thread, 0,
+              "Number of operations per thread. (Default: 5 * keyspace size)");
+DEFINE_uint32(value_bytes, 8 * KiB, "Size of each value added.");

-DEFINE_bool(populate_cache, false, "Populate cache before operations");
-DEFINE_int32(insert_percent, 40,
-             "Ratio of insert to total workload (expressed as a percentage)");
-DEFINE_int32(lookup_percent, 50,
-             "Ratio of lookup to total workload (expressed as a percentage)");
-DEFINE_int32(erase_percent, 10,
-             "Ratio of erase to total workload (expressed as a percentage)");
+DEFINE_uint32(skew, 5, "Degree of skew in key selection");
+DEFINE_bool(populate_cache, true, "Populate cache before operations");
+
+DEFINE_uint32(lookup_insert_percent, 87,
+              "Ratio of lookup (+ insert on not found) to total workload "
+              "(expressed as a percentage)");
+DEFINE_uint32(insert_percent, 2,
+              "Ratio of insert to total workload (expressed as a percentage)");
+DEFINE_uint32(lookup_percent, 10,
+              "Ratio of lookup to total workload (expressed as a percentage)");
+DEFINE_uint32(erase_percent, 1,
+              "Ratio of erase to total workload (expressed as a percentage)");

 DEFINE_bool(use_clock_cache, false, "");

@@ -49,21 +62,15 @@ namespace ROCKSDB_NAMESPACE {

 class CacheBench;
 namespace {
-void deleter(const Slice& /*key*/, void* value) {
-    delete reinterpret_cast<char *>(value);
-}
-
 // State shared by all concurrent executions of the same benchmark.
 class SharedState {
 public:
  explicit SharedState(CacheBench* cache_bench)
      : cv_(&mu_),
-        num_threads_(FLAGS_threads),
        num_initialized_(0),
        start_(false),
        num_done_(0),
-        cache_bench_(cache_bench) {
-  }
+        cache_bench_(cache_bench) {}

  ~SharedState() {}

@@ -87,13 +94,9 @@ class SharedState {
    num_done_++;
  }

-  bool AllInitialized() const {
-    return num_initialized_ >= num_threads_;
-  }
+  bool AllInitialized() const { return num_initialized_ >= FLAGS_threads; }

-  bool AllDone() const {
-    return num_done_ >= num_threads_;
-  }
+  bool AllDone() const { return num_done_ >= FLAGS_threads; }

  void SetStart() {
    start_ = true;
@@ -107,7 +110,6 @@ class SharedState {
  port::Mutex mu_;
  port::CondVar cv_;

-  const uint64_t num_threads_;
  uint64_t num_initialized_;
  bool start_;
  uint64_t num_done_;
@@ -118,17 +120,69 @@ class SharedState {
 // Per-thread state for concurrent executions of the same benchmark.
 struct ThreadState {
  uint32_t tid;
-  Random rnd;
+  Random64 rnd;
  SharedState* shared;

  ThreadState(uint32_t index, SharedState* _shared)
      : tid(index), rnd(1000 + index), shared(_shared) {}
 };
+
+struct KeyGen {
+  char key_data[27];
+
+  Slice GetRand(Random64& rnd, uint64_t max_key) {
+    uint64_t raw = rnd.Next();
+    // Skew according to setting
+    for (uint32_t i = 0; i < FLAGS_skew; ++i) {
+      raw = std::min(raw, rnd.Next());
+    }
+    uint64_t key = fastrange64(raw, max_key);
+    // Variable size and alignment
+    size_t off = key % 8;
+    key_data[0] = char{42};
+    EncodeFixed64(key_data + 1, key);
+    key_data[9] = char{11};
+    EncodeFixed64(key_data + 10, key);
+    key_data[18] = char{4};
+    EncodeFixed64(key_data + 19, key);
+    return Slice(&key_data[off], sizeof(key_data) - off);
+  }
+};
+
+char* createValue(Random64& rnd) {
+  char* rv = new char[FLAGS_value_bytes];
+  // Fill with some filler data, and take some CPU time
+  for (uint32_t i = 0; i < FLAGS_value_bytes; i += 8) {
+    EncodeFixed64(rv + i, rnd.Next());
+  }
+  return rv;
+}
+
+void deleter(const Slice& /*key*/, void* value) {
+  delete[] static_cast<char*>(value);
+}
 }  // namespace

 class CacheBench {
+  static constexpr uint64_t kHundredthUint64 =
+      std::numeric_limits<uint64_t>::max() / 100U;
+
 public:
-  CacheBench() : num_threads_(FLAGS_threads) {
+  CacheBench()
+      : max_key_(static_cast<uint64_t>(FLAGS_cache_size / FLAGS_resident_ratio /
+                                       FLAGS_value_bytes)),
+        lookup_insert_threshold_(kHundredthUint64 *
+                                 FLAGS_lookup_insert_percent),
+        insert_threshold_(lookup_insert_threshold_ +
+                          kHundredthUint64 * FLAGS_insert_percent),
+        lookup_threshold_(insert_threshold_ +
+                          kHundredthUint64 * FLAGS_lookup_percent),
+        erase_threshold_(lookup_threshold_ +
+                         kHundredthUint64 * FLAGS_erase_percent) {
+    if (erase_threshold_ != 100U * kHundredthUint64) {
+      fprintf(stderr, "Percentages must add to 100.\n");
+      exit(1);
+    }
    if (FLAGS_use_clock_cache) {
      cache_ = NewClockCache(FLAGS_cache_size, FLAGS_num_shard_bits);
      if (!cache_) {
@@ -138,18 +192,19 @@ class CacheBench {
    } else {
      cache_ = NewLRUCache(FLAGS_cache_size, FLAGS_num_shard_bits);
    }
+    if (FLAGS_ops_per_thread == 0) {
+      FLAGS_ops_per_thread = 5 * max_key_;
+    }
  }

  ~CacheBench() {}

  void PopulateCache() {
-    Random rnd(1);
-    for (int64_t i = 0; i < FLAGS_cache_size; i++) {
-      uint64_t rand_key = rnd.Next() % FLAGS_max_key;
-      // Cast uint64* to be char*, data would be copied to cache
-      Slice key(reinterpret_cast<char*>(&rand_key), 8);
-      // do insert
-      cache_->Insert(key, new char[10], 1, &deleter);
+    Random64 rnd(1);
+    KeyGen keygen;
+    for (uint64_t i = 0; i < 2 * FLAGS_cache_size; i += FLAGS_value_bytes) {
+      cache_->Insert(keygen.GetRand(rnd, max_key_), createValue(rnd),
+                     FLAGS_value_bytes, &deleter);
    }
  }

@@ -158,10 +213,10 @@ class CacheBench {

    PrintEnv();
    SharedState shared(this);
-    std::vector<ThreadState*> threads(num_threads_);
-    for (uint32_t i = 0; i < num_threads_; i++) {
-      threads[i] = new ThreadState(i, &shared);
-      env->StartThread(ThreadBody, threads[i]);
+    std::vector<std::unique_ptr<ThreadState> > threads(FLAGS_threads);
+    for (uint32_t i = 0; i < FLAGS_threads; i++) {
+      threads[i].reset(new ThreadState(i, &shared));
+      env->StartThread(ThreadBody, threads[i].get());
    }
    {
      MutexLock l(shared.GetMutex());
@@ -192,10 +247,15 @@ class CacheBench {

 private:
  std::shared_ptr<Cache> cache_;
-  uint32_t num_threads_;
+  const uint64_t max_key_;
+  // Cumulative thresholds in the space of a random uint64_t
+  const uint64_t lookup_insert_threshold_;
+  const uint64_t insert_threshold_;
+  const uint64_t lookup_threshold_;
+  const uint64_t erase_threshold_;

  static void ThreadBody(void* v) {
-    ThreadState* thread = reinterpret_cast<ThreadState*>(v);
+    ThreadState* thread = static_cast<ThreadState*>(v);
    SharedState* shared = thread->shared;

    {
@@ -220,40 +280,78 @@ class CacheBench {
  }

  void OperateCache(ThreadState* thread) {
+    // To use looked-up values
+    uint64_t result = 0;
+    // To hold handles for a non-trivial amount of time
+    Cache::Handle* handle = nullptr;
+    KeyGen gen;
    for (uint64_t i = 0; i < FLAGS_ops_per_thread; i++) {
-      uint64_t rand_key = thread->rnd.Next() % FLAGS_max_key;
-      // Cast uint64* to be char*, data would be copied to cache
-      Slice key(reinterpret_cast<char*>(&rand_key), 8);
-      int32_t prob_op = thread->rnd.Uniform(100);
-      if (prob_op >= 0 && prob_op < FLAGS_insert_percent) {
-        // do insert
-        cache_->Insert(key, new char[10], 1, &deleter);
-      } else if (prob_op -= FLAGS_insert_percent &&
-                 prob_op < FLAGS_lookup_percent) {
-        // do lookup
-        auto handle = cache_->Lookup(key);
+      Slice key = gen.GetRand(thread->rnd, max_key_);
+      uint64_t random_op = thread->rnd.Next();
+      if (random_op < lookup_insert_threshold_) {
        if (handle) {
          cache_->Release(handle);
+          handle = nullptr;
        }
-      } else if (prob_op -= FLAGS_lookup_percent &&
-                 prob_op < FLAGS_erase_percent) {
+        // do lookup
+        handle = cache_->Lookup(key);
+        if (handle) {
+          // do something with the data
+          result += NPHash64(static_cast<char*>(cache_->Value(handle)),
+                             FLAGS_value_bytes);
+        } else {
+          // do insert
+          cache_->Insert(key, createValue(thread->rnd), FLAGS_value_bytes,
+                         &deleter, &handle);
+        }
+      } else if (random_op < insert_threshold_) {
+        if (handle) {
+          cache_->Release(handle);
+          handle = nullptr;
+        }
+        // do insert
+        cache_->Insert(key, createValue(thread->rnd), FLAGS_value_bytes,
+                       &deleter, &handle);
+      } else if (random_op < lookup_threshold_) {
+        if (handle) {
+          cache_->Release(handle);
+          handle = nullptr;
+        }
+        // do lookup
+        handle = cache_->Lookup(key);
+        if (handle) {
+          // do something with the data
+          result += NPHash64(static_cast<char*>(cache_->Value(handle)),
+                             FLAGS_value_bytes);
+        }
+      } else if (random_op < erase_threshold_) {
        // do erase
        cache_->Erase(key);
+      } else {
+        // Should be extremely unlikely (noop)
+        assert(random_op >= kHundredthUint64 * 100U);
      }
    }
+    if (handle) {
+      cache_->Release(handle);
+      handle = nullptr;
+    }
  }

  void PrintEnv() const {
    printf("RocksDB version     : %d.%d\n", kMajorVersion, kMinorVersion);
-    printf("Number of threads   : %d\n", FLAGS_threads);
+    printf("Number of threads   : %u\n", FLAGS_threads);
    printf("Ops per thread      : %" PRIu64 "\n", FLAGS_ops_per_thread);
    printf("Cache size          : %" PRIu64 "\n", FLAGS_cache_size);
-    printf("Num shard bits      : %d\n", FLAGS_num_shard_bits);
-    printf("Max key             : %" PRIu64 "\n", FLAGS_max_key);
-    printf("Populate cache      : %d\n", FLAGS_populate_cache);
-    printf("Insert percentage   : %d%%\n", FLAGS_insert_percent);
-    printf("Lookup percentage   : %d%%\n", FLAGS_lookup_percent);
-    printf("Erase percentage    : %d%%\n", FLAGS_erase_percent);
+    printf("Num shard bits      : %u\n", FLAGS_num_shard_bits);
+    printf("Max key             : %" PRIu64 "\n", max_key_);
+    printf("Resident ratio      : %g\n", FLAGS_resident_ratio);
+    printf("Skew degree         : %u\n", FLAGS_skew);
+    printf("Populate cache      : %d\n", int{FLAGS_populate_cache});
+    printf("Lookup+Insert pct   : %u%%\n", FLAGS_lookup_insert_percent);
+    printf("Insert percentage   : %u%%\n", FLAGS_insert_percent);
+    printf("Lookup percentage   : %u%%\n", FLAGS_lookup_percent);
+    printf("Erase percentage    : %u%%\n", FLAGS_erase_percent);
    printf("----------------------------\n");
  }
 };
@@ -270,6 +368,8 @@ int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::CacheBench bench;
  if (FLAGS_populate_cache) {
    bench.PopulateCache();
+    printf("Population complete\n");
+    printf("----------------------------\n");
  }
  if (bench.Run()) {
    return 0;
@@ -182,7 +182,7 @@ struct CacheHandle {
  void (*deleter)(const Slice&, void* value);

  // Flags and counters associated with the cache handle:
-  //   lowest bit: n-cache bit
+  //   lowest bit: in-cache bit
  //   second lowest bit: usage bit
  //   the rest bits: reference count
  // The handle is unused when flags equals to 0. The thread decreases the count
@@ -341,7 +341,8 @@ class ClockCacheShard final : public CacheShard {
  CacheHandle* Insert(const Slice& key, uint32_t hash, void* value,
                      size_t change,
                      void (*deleter)(const Slice& key, void* value),
-                      bool hold_reference, CleanupContext* context);
+                      bool hold_reference, CleanupContext* context,
+                      bool* overwritten);

  // Guards list_, head_, and recycle_. In addition, updating table_ also has
  // to hold the mutex, to avoid the cache being in inconsistent state.
@@ -564,7 +565,8 @@ void ClockCacheShard::SetStrictCapacityLimit(bool strict_capacity_limit) {
 CacheHandle* ClockCacheShard::Insert(
    const Slice& key, uint32_t hash, void* value, size_t charge,
    void (*deleter)(const Slice& key, void* value), bool hold_reference,
-    CleanupContext* context) {
+    CleanupContext* context, bool* overwritten) {
+  assert(overwritten != nullptr && *overwritten == false);
  size_t total_charge =
      CacheHandle::CalcTotalCharge(key, charge, metadata_charge_policy_);
  MutexLock l(&mutex_);
@@ -597,6 +599,7 @@ CacheHandle* ClockCacheShard::Insert(
  handle->flags.store(flags, std::memory_order_relaxed);
  HashTable::accessor accessor;
  if (table_.find(accessor, CacheKey(key, hash))) {
+    *overwritten = true;
    CacheHandle* existing_handle = accessor->second;
    table_.erase(accessor);
    UnsetInCache(existing_handle, context);
@@ -619,8 +622,9 @@ Status ClockCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
  char* key_data = new char[key.size()];
  memcpy(key_data, key.data(), key.size());
  Slice key_copy(key_data, key.size());
+  bool overwritten = false;
  CacheHandle* handle = Insert(key_copy, hash, value, charge, deleter,
-                               out_handle != nullptr, &context);
+                               out_handle != nullptr, &context, &overwritten);
  Status s;
  if (out_handle != nullptr) {
    if (handle == nullptr) {
@@ -629,6 +633,10 @@ Status ClockCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
      *out_handle = reinterpret_cast<Cache::Handle*>(handle);
    }
  }
+  if (overwritten) {
+    assert(s.ok());
+    s = Status::OkOverwritten();
+  }
  Cleanup(context);
  return s;
 }
@@ -386,6 +386,7 @@ Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
      LRUHandle* old = table_.Insert(e);
      usage_ += total_charge;
      if (old != nullptr) {
+        s = Status::OkOverwritten();
        assert(old->InCache());
        old->SetInCache(false);
        if (!old->HasRefs()) {
@@ -1,3 +1,54 @@
@PACKAGE_INIT@
+
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/modules")
+
+include(CMakeFindDependencyMacro)
+
+set(GFLAGS_USE_TARGET_NAMESPACE @GFLAGS_USE_TARGET_NAMESPACE@)
+
+if(@WITH_JEMALLOC@)
+  find_dependency(JeMalloc)
+endif()
+
+if(@WITH_GFLAGS@)
+  find_dependency(gflags CONFIG)
+  if(NOT gflags_FOUND)
+    find_dependency(gflags)
+  endif()
+endif()
+
+if(@WITH_SNAPPY@)
+  find_dependency(Snappy CONFIG)
+  if(NOT Snappy_FOUND)
+    find_dependency(Snappy)
+  endif()
+endif()
+
+if(@WITH_ZLIB@)
+  find_dependency(ZLIB)
+endif()
+
+if(@WITH_BZ2@)
+  find_dependency(BZip2)
+endif()
+
+if(@WITH_LZ4@)
+  find_dependency(lz4)
+endif()
+
+if(@WITH_ZSTD@)
+  find_dependency(zstd)
+endif()
+
+if(@WITH_NUMA@)
+  find_dependency(NUMA)
+endif()
+
+if(@WITH_TBB@)
+  find_dependency(TBB)
+endif()
+
+find_dependency(Threads)
+
 include("${CMAKE_CURRENT_LIST_DIR}/RocksDBTargets.cmake")
 check_required_components(RocksDB)
@@ -0,0 +1,7 @@
+macro(get_cxx_std_flags FLAGS_VARIABLE)
+  if( CMAKE_CXX_STANDARD_REQUIRED )
+    set(${FLAGS_VARIABLE} ${CMAKE_CXX${CMAKE_CXX_STANDARD}_STANDARD_COMPILE_OPTION})
+  else()
+    set(${FLAGS_VARIABLE} ${CMAKE_CXX${CMAKE_CXX_STANDARD}_EXTENSION_COMPILE_OPTION})
+  endif()
+endmacro()
@@ -0,0 +1,29 @@
+# - Find Snappy
+# Find the snappy compression library and includes
+#
+# Snappy_INCLUDE_DIRS - where to find snappy.h, etc.
+# Snappy_LIBRARIES - List of libraries when using snappy.
+# Snappy_FOUND - True if snappy found.
+
+find_path(Snappy_INCLUDE_DIRS
+  NAMES snappy.h
+  HINTS ${snappy_ROOT_DIR}/include)
+
+find_library(Snappy_LIBRARIES
+  NAMES snappy
+  HINTS ${snappy_ROOT_DIR}/lib)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(Snappy DEFAULT_MSG Snappy_LIBRARIES Snappy_INCLUDE_DIRS)
+
+mark_as_advanced(
+  Snappy_LIBRARIES
+  Snappy_INCLUDE_DIRS)
+
+if(Snappy_FOUND AND NOT (TARGET Snappy::snappy))
+  add_library (Snappy::snappy UNKNOWN IMPORTED)
+  set_target_properties(Snappy::snappy
+    PROPERTIES
+      IMPORTED_LOCATION ${Snappy_LIBRARIES}
+      INTERFACE_INCLUDE_DIRECTORIES ${Snappy_INCLUDE_DIRS})
+endif()
@@ -1,8 +1,8 @@
 # - Find gflags library
 # Find the gflags includes and library
 #
-# gflags_INCLUDE_DIR - where to find gflags.h.
-# gflags_LIBRARIES - List of libraries when using gflags.
+# GFLAGS_INCLUDE_DIR - where to find gflags.h.
+# GFLAGS_LIBRARIES - List of libraries when using gflags.
 # gflags_FOUND - True if gflags found.

 find_path(GFLAGS_INCLUDE_DIR
@@ -1,29 +0,0 @@
-# - Find Snappy
-# Find the snappy compression library and includes
-#
-# snappy_INCLUDE_DIRS - where to find snappy.h, etc.
-# snappy_LIBRARIES - List of libraries when using snappy.
-# snappy_FOUND - True if snappy found.
-
-find_path(snappy_INCLUDE_DIRS
-  NAMES snappy.h
-  HINTS ${snappy_ROOT_DIR}/include)
-
-find_library(snappy_LIBRARIES
-  NAMES snappy
-  HINTS ${snappy_ROOT_DIR}/lib)
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(snappy DEFAULT_MSG snappy_LIBRARIES snappy_INCLUDE_DIRS)
-
-mark_as_advanced(
-  snappy_LIBRARIES
-  snappy_INCLUDE_DIRS)
-
-if(snappy_FOUND AND NOT (TARGET snappy::snappy))
-  add_library (snappy::snappy UNKNOWN IMPORTED)
-  set_target_properties(snappy::snappy
-    PROPERTIES
-      IMPORTED_LOCATION ${snappy_LIBRARIES}
-      INTERFACE_INCLUDE_DIRECTORIES ${snappy_INCLUDE_DIRS})
-endif()
@@ -12,7 +12,7 @@ fi
 ROOT=".."
 # Fetch right version of gcov
 if [ -d /mnt/gvfs/third-party -a -z "$CXX" ]; then
-  source $ROOT/build_tools/fbcode_config.sh
+  source $ROOT/build_tools/fbcode_config_platform007.sh
  GCOV=$GCC_BASE/bin/gcov
 else
  GCOV=$(which gcov)
@@ -75,7 +75,7 @@ Status ArenaWrappedDBIter::Refresh() {

    InternalIterator* internal_iter = db_impl_->NewInternalIterator(
        read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator(),
-        latest_seq);
+        latest_seq, /* allow_unprepared_value */ true);
    SetIterUnderDBIter(internal_iter);
  } else {
    db_iter_->set_sequence(latest_seq);
@@ -10,10 +10,6 @@

 namespace ROCKSDB_NAMESPACE {

-SharedBlobFileMetaData::~SharedBlobFileMetaData() {
-  // TODO: add the blob file to the list of obsolete files here
-}
-
 std::string SharedBlobFileMetaData::DebugString() const {
  std::ostringstream oss;
  oss << (*this);
@@ -23,6 +23,42 @@ namespace ROCKSDB_NAMESPACE {

 class SharedBlobFileMetaData {
 public:
+  static std::shared_ptr<SharedBlobFileMetaData> Create(
+      uint64_t blob_file_number, uint64_t total_blob_count,
+      uint64_t total_blob_bytes, std::string checksum_method,
+      std::string checksum_value) {
+    return std::shared_ptr<SharedBlobFileMetaData>(new SharedBlobFileMetaData(
+        blob_file_number, total_blob_count, total_blob_bytes,
+        std::move(checksum_method), std::move(checksum_value)));
+  }
+
+  template <typename Deleter>
+  static std::shared_ptr<SharedBlobFileMetaData> Create(
+      uint64_t blob_file_number, uint64_t total_blob_count,
+      uint64_t total_blob_bytes, std::string checksum_method,
+      std::string checksum_value, Deleter deleter) {
+    return std::shared_ptr<SharedBlobFileMetaData>(
+        new SharedBlobFileMetaData(blob_file_number, total_blob_count,
+                                   total_blob_bytes, std::move(checksum_method),
+                                   std::move(checksum_value)),
+        deleter);
+  }
+
+  SharedBlobFileMetaData(const SharedBlobFileMetaData&) = delete;
+  SharedBlobFileMetaData& operator=(const SharedBlobFileMetaData&) = delete;
+
+  SharedBlobFileMetaData(SharedBlobFileMetaData&&) = delete;
+  SharedBlobFileMetaData& operator=(SharedBlobFileMetaData&&) = delete;
+
+  uint64_t GetBlobFileNumber() const { return blob_file_number_; }
+  uint64_t GetTotalBlobCount() const { return total_blob_count_; }
+  uint64_t GetTotalBlobBytes() const { return total_blob_bytes_; }
+  const std::string& GetChecksumMethod() const { return checksum_method_; }
+  const std::string& GetChecksumValue() const { return checksum_value_; }
+
+  std::string DebugString() const;
+
+ private:
  SharedBlobFileMetaData(uint64_t blob_file_number, uint64_t total_blob_count,
                         uint64_t total_blob_bytes, std::string checksum_method,
                         std::string checksum_value)
@@ -34,20 +70,6 @@ class SharedBlobFileMetaData {
    assert(checksum_method_.empty() == checksum_value_.empty());
  }

-  ~SharedBlobFileMetaData();
-
-  SharedBlobFileMetaData(const SharedBlobFileMetaData&) = delete;
-  SharedBlobFileMetaData& operator=(const SharedBlobFileMetaData&) = delete;
-
-  uint64_t GetBlobFileNumber() const { return blob_file_number_; }
-  uint64_t GetTotalBlobCount() const { return total_blob_count_; }
-  uint64_t GetTotalBlobBytes() const { return total_blob_bytes_; }
-  const std::string& GetChecksumMethod() const { return checksum_method_; }
-  const std::string& GetChecksumValue() const { return checksum_value_; }
-
-  std::string DebugString() const;
-
- private:
  uint64_t blob_file_number_;
  uint64_t total_blob_count_;
  uint64_t total_blob_bytes_;
@@ -68,21 +90,19 @@ std::ostream& operator<<(std::ostream& os,

 class BlobFileMetaData {
 public:
-  BlobFileMetaData(std::shared_ptr<SharedBlobFileMetaData> shared_meta,
-                   uint64_t garbage_blob_count, uint64_t garbage_blob_bytes)
-      : shared_meta_(std::move(shared_meta)),
-        garbage_blob_count_(garbage_blob_count),
-        garbage_blob_bytes_(garbage_blob_bytes) {
-    assert(shared_meta_);
-    assert(garbage_blob_count_ <= shared_meta_->GetTotalBlobCount());
-    assert(garbage_blob_bytes_ <= shared_meta_->GetTotalBlobBytes());
+  static std::shared_ptr<BlobFileMetaData> Create(
+      std::shared_ptr<SharedBlobFileMetaData> shared_meta,
+      uint64_t garbage_blob_count, uint64_t garbage_blob_bytes) {
+    return std::shared_ptr<BlobFileMetaData>(new BlobFileMetaData(
+        std::move(shared_meta), garbage_blob_count, garbage_blob_bytes));
  }

-  ~BlobFileMetaData() = default;
-
  BlobFileMetaData(const BlobFileMetaData&) = delete;
  BlobFileMetaData& operator=(const BlobFileMetaData&) = delete;

+  BlobFileMetaData(BlobFileMetaData&&) = delete;
+  BlobFileMetaData& operator=(BlobFileMetaData&&) = delete;
+
  const std::shared_ptr<SharedBlobFileMetaData>& GetSharedMeta() const {
    return shared_meta_;
  }
@@ -114,6 +134,16 @@ class BlobFileMetaData {
  std::string DebugString() const;

 private:
+  BlobFileMetaData(std::shared_ptr<SharedBlobFileMetaData> shared_meta,
+                   uint64_t garbage_blob_count, uint64_t garbage_blob_bytes)
+      : shared_meta_(std::move(shared_meta)),
+        garbage_blob_count_(garbage_blob_count),
+        garbage_blob_bytes_(garbage_blob_bytes) {
+    assert(shared_meta_);
+    assert(garbage_blob_count_ <= shared_meta_->GetTotalBlobCount());
+    assert(garbage_blob_bytes_ <= shared_meta_->GetTotalBlobBytes());
+  }
+
  std::shared_ptr<SharedBlobFileMetaData> shared_meta_;
  uint64_t garbage_blob_count_;
  uint64_t garbage_blob_bytes_;
@@ -183,8 +183,7 @@ Status BuildTable(
    }

    // Finish and check for builder errors
-    tp = builder->GetTableProperties();
-    bool empty = builder->NumEntries() == 0 && tp.num_range_deletions == 0;
+    bool empty = builder->IsEmpty();
    s = c_iter.status();
    TEST_SYNC_POINT("BuildTable:BeforeFinishBuildTable");
    if (!s.ok() || empty) {
@@ -241,7 +240,8 @@ Status BuildTable(
                                      : internal_stats->GetFileReadHist(0),
          TableReaderCaller::kFlush, /*arena=*/nullptr,
          /*skip_filter=*/false, level, /*smallest_compaction_key=*/nullptr,
-          /*largest_compaction_key*/ nullptr));
+          /*largest_compaction_key*/ nullptr,
+          /*allow_unprepared_value*/ false));
      s = it->status();
      if (s.ok() && paranoid_file_checks) {
        for (it->SeekToFirst(); it->Valid(); it->Next()) {
@@ -1466,6 +1466,11 @@ void rocksdb_writebatch_delete(
  b->rep.Delete(Slice(key, klen));
 }

+void rocksdb_writebatch_singledelete(rocksdb_writebatch_t* b, const char* key,
+                                     size_t klen) {
+  b->rep.SingleDelete(Slice(key, klen));
+}
+
 void rocksdb_writebatch_delete_cf(
    rocksdb_writebatch_t* b,
    rocksdb_column_family_handle_t* column_family,
@@ -1473,6 +1478,12 @@ void rocksdb_writebatch_delete_cf(
  b->rep.Delete(column_family->rep, Slice(key, klen));
 }

+void rocksdb_writebatch_singledelete_cf(
+    rocksdb_writebatch_t* b, rocksdb_column_family_handle_t* column_family,
+    const char* key, size_t klen) {
+  b->rep.SingleDelete(column_family->rep, Slice(key, klen));
+}
+
 void rocksdb_writebatch_deletev(
    rocksdb_writebatch_t* b,
    int num_keys, const char* const* keys_list,
@@ -1723,6 +1734,11 @@ void rocksdb_writebatch_wi_delete(
  b->rep->Delete(Slice(key, klen));
 }

+void rocksdb_writebatch_wi_singledelete(rocksdb_writebatch_wi_t* b,
+                                        const char* key, size_t klen) {
+  b->rep->SingleDelete(Slice(key, klen));
+}
+
 void rocksdb_writebatch_wi_delete_cf(
    rocksdb_writebatch_wi_t* b,
    rocksdb_column_family_handle_t* column_family,
@@ -1730,6 +1746,12 @@ void rocksdb_writebatch_wi_delete_cf(
  b->rep->Delete(column_family->rep, Slice(key, klen));
 }

+void rocksdb_writebatch_wi_singledelete_cf(
+    rocksdb_writebatch_wi_t* b, rocksdb_column_family_handle_t* column_family,
+    const char* key, size_t klen) {
+  b->rep->SingleDelete(column_family->rep, Slice(key, klen));
+}
+
 void rocksdb_writebatch_wi_deletev(
    rocksdb_writebatch_wi_t* b,
    int num_keys, const char* const* keys_list,
@@ -2154,6 +2176,10 @@ void rocksdb_options_destroy(rocksdb_options_t* options) {
  delete options;
 }

+rocksdb_options_t* rocksdb_options_create_copy(rocksdb_options_t* options) {
+  return new rocksdb_options_t(*options);
+}
+
 void rocksdb_options_increase_parallelism(
    rocksdb_options_t* opt, int total_threads) {
  opt->rep.IncreaseParallelism(total_threads);
@@ -2179,6 +2205,10 @@ void rocksdb_options_set_allow_ingest_behind(
  opt->rep.allow_ingest_behind = v;
 }

+unsigned char rocksdb_options_get_allow_ingest_behind(rocksdb_options_t* opt) {
+  return opt->rep.allow_ingest_behind;
+}
+
 void rocksdb_options_set_compaction_filter(
    rocksdb_options_t* opt,
    rocksdb_compactionfilter_t* filter) {
@@ -2196,6 +2226,10 @@ void rocksdb_options_compaction_readahead_size(
  opt->rep.compaction_readahead_size = s;
 }

+size_t rocksdb_options_get_compaction_readahead_size(rocksdb_options_t* opt) {
+  return opt->rep.compaction_readahead_size;
+}
+
 void rocksdb_options_set_comparator(
    rocksdb_options_t* opt,
    rocksdb_comparator_t* cmp) {
@@ -2208,27 +2242,43 @@ void rocksdb_options_set_merge_operator(
  opt->rep.merge_operator = std::shared_ptr<MergeOperator>(merge_operator);
 }

-
 void rocksdb_options_set_create_if_missing(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.create_if_missing = v;
 }

+unsigned char rocksdb_options_get_create_if_missing(rocksdb_options_t* opt) {
+  return opt->rep.create_if_missing;
+}
+
 void rocksdb_options_set_create_missing_column_families(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.create_missing_column_families = v;
 }

+unsigned char rocksdb_options_get_create_missing_column_families(
+    rocksdb_options_t* opt) {
+  return opt->rep.create_missing_column_families;
+}
+
 void rocksdb_options_set_error_if_exists(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.error_if_exists = v;
 }

+unsigned char rocksdb_options_get_error_if_exists(rocksdb_options_t* opt) {
+  return opt->rep.error_if_exists;
+}
+
 void rocksdb_options_set_paranoid_checks(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.paranoid_checks = v;
 }

+unsigned char rocksdb_options_get_paranoid_checks(rocksdb_options_t* opt) {
+  return opt->rep.paranoid_checks;
+}
+
 void rocksdb_options_set_db_paths(rocksdb_options_t* opt,
                                  const rocksdb_dbpath_t** dbpath_values,
                                  size_t num_paths) {
@@ -2254,52 +2304,98 @@ void rocksdb_options_set_info_log_level(
  opt->rep.info_log_level = static_cast<InfoLogLevel>(v);
 }

+int rocksdb_options_get_info_log_level(rocksdb_options_t* opt) {
+  return static_cast<int>(opt->rep.info_log_level);
+}
+
 void rocksdb_options_set_db_write_buffer_size(rocksdb_options_t* opt,
                                              size_t s) {
  opt->rep.db_write_buffer_size = s;
 }

+size_t rocksdb_options_get_db_write_buffer_size(rocksdb_options_t* opt) {
+  return opt->rep.db_write_buffer_size;
+}
+
 void rocksdb_options_set_write_buffer_size(rocksdb_options_t* opt, size_t s) {
  opt->rep.write_buffer_size = s;
 }

+size_t rocksdb_options_get_write_buffer_size(rocksdb_options_t* opt) {
+  return opt->rep.write_buffer_size;
+}
+
 void rocksdb_options_set_max_open_files(rocksdb_options_t* opt, int n) {
  opt->rep.max_open_files = n;
 }

+int rocksdb_options_get_max_open_files(rocksdb_options_t* opt) {
+  return opt->rep.max_open_files;
+}
+
 void rocksdb_options_set_max_file_opening_threads(rocksdb_options_t* opt, int n) {
  opt->rep.max_file_opening_threads = n;
 }

+int rocksdb_options_get_max_file_opening_threads(rocksdb_options_t* opt) {
+  return opt->rep.max_file_opening_threads;
+}
+
 void rocksdb_options_set_max_total_wal_size(rocksdb_options_t* opt, uint64_t n) {
  opt->rep.max_total_wal_size = n;
 }

+uint64_t rocksdb_options_get_max_total_wal_size(rocksdb_options_t* opt) {
+  return opt->rep.max_total_wal_size;
+}
+
 void rocksdb_options_set_target_file_size_base(
    rocksdb_options_t* opt, uint64_t n) {
  opt->rep.target_file_size_base = n;
 }

+uint64_t rocksdb_options_get_target_file_size_base(rocksdb_options_t* opt) {
+  return opt->rep.target_file_size_base;
+}
+
 void rocksdb_options_set_target_file_size_multiplier(
    rocksdb_options_t* opt, int n) {
  opt->rep.target_file_size_multiplier = n;
 }

+int rocksdb_options_get_target_file_size_multiplier(rocksdb_options_t* opt) {
+  return opt->rep.target_file_size_multiplier;
+}
+
 void rocksdb_options_set_max_bytes_for_level_base(
    rocksdb_options_t* opt, uint64_t n) {
  opt->rep.max_bytes_for_level_base = n;
 }

+uint64_t rocksdb_options_get_max_bytes_for_level_base(rocksdb_options_t* opt) {
+  return opt->rep.max_bytes_for_level_base;
+}
+
 void rocksdb_options_set_level_compaction_dynamic_level_bytes(
    rocksdb_options_t* opt, unsigned char v) {
  opt->rep.level_compaction_dynamic_level_bytes = v;
 }

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

+double rocksdb_options_get_max_bytes_for_level_multiplier(
+    rocksdb_options_t* opt) {
+  return opt->rep.max_bytes_for_level_multiplier;
+}
+
 void rocksdb_options_set_max_compaction_bytes(rocksdb_options_t* opt,
                                              uint64_t n) {
  opt->rep.max_compaction_bytes = n;
@@ -2322,30 +2418,57 @@ void rocksdb_options_set_skip_stats_update_on_db_open(rocksdb_options_t* opt,
  opt->rep.skip_stats_update_on_db_open = val;
 }

+unsigned char rocksdb_options_get_skip_stats_update_on_db_open(
+    rocksdb_options_t* opt) {
+  return opt->rep.skip_stats_update_on_db_open;
+}
+
 void rocksdb_options_set_skip_checking_sst_file_sizes_on_db_open(
    rocksdb_options_t* opt, unsigned char val) {
  opt->rep.skip_checking_sst_file_sizes_on_db_open = val;
 }

+unsigned char rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(
+    rocksdb_options_t* opt) {
+  return opt->rep.skip_checking_sst_file_sizes_on_db_open;
+}
+
 void rocksdb_options_set_num_levels(rocksdb_options_t* opt, int n) {
  opt->rep.num_levels = n;
 }

+int rocksdb_options_get_num_levels(rocksdb_options_t* opt) {
+  return opt->rep.num_levels;
+}
+
 void rocksdb_options_set_level0_file_num_compaction_trigger(
    rocksdb_options_t* opt, int n) {
  opt->rep.level0_file_num_compaction_trigger = n;
 }

+int rocksdb_options_get_level0_file_num_compaction_trigger(
+    rocksdb_options_t* opt) {
+  return opt->rep.level0_file_num_compaction_trigger;
+}
+
 void rocksdb_options_set_level0_slowdown_writes_trigger(
    rocksdb_options_t* opt, int n) {
  opt->rep.level0_slowdown_writes_trigger = n;
 }

+int rocksdb_options_get_level0_slowdown_writes_trigger(rocksdb_options_t* opt) {
+  return opt->rep.level0_slowdown_writes_trigger;
+}
+
 void rocksdb_options_set_level0_stop_writes_trigger(
    rocksdb_options_t* opt, int n) {
  opt->rep.level0_stop_writes_trigger = n;
 }

+int rocksdb_options_get_level0_stop_writes_trigger(rocksdb_options_t* opt) {
+  return opt->rep.level0_stop_writes_trigger;
+}
+
 void rocksdb_options_set_max_mem_compaction_level(rocksdb_options_t* /*opt*/,
                                                  int /*n*/) {}

@@ -2357,6 +2480,10 @@ void rocksdb_options_set_compression(rocksdb_options_t* opt, int t) {
  opt->rep.compression = static_cast<CompressionType>(t);
 }

+void rocksdb_options_set_bottommost_compression(rocksdb_options_t* opt, int t) {
+  opt->rep.bottommost_compression = static_cast<CompressionType>(t);
+}
+
 void rocksdb_options_set_compression_per_level(rocksdb_options_t* opt,
                                               int* level_values,
                                               size_t num_levels) {
@@ -2371,7 +2498,7 @@ void rocksdb_options_set_bottommost_compression_options(rocksdb_options_t* opt,
                                                        int w_bits, int level,
                                                        int strategy,
                                                        int max_dict_bytes,
-                                                        bool enabled) {
+                                                        unsigned char enabled) {
  opt->rep.bottommost_compression_opts.window_bits = w_bits;
  opt->rep.bottommost_compression_opts.level = level;
  opt->rep.bottommost_compression_opts.strategy = strategy;
@@ -2379,6 +2506,13 @@ void rocksdb_options_set_bottommost_compression_options(rocksdb_options_t* opt,
  opt->rep.bottommost_compression_opts.enabled = enabled;
 }

+void rocksdb_options_set_bottommost_compression_options_zstd_max_train_bytes(
+    rocksdb_options_t* opt, int zstd_max_train_bytes, unsigned char enabled) {
+  opt->rep.bottommost_compression_opts.zstd_max_train_bytes =
+      zstd_max_train_bytes;
+  opt->rep.bottommost_compression_opts.enabled = enabled;
+}
+
 void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
                                             int level, int strategy,
                                             int max_dict_bytes) {
@@ -2388,6 +2522,11 @@ void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
  opt->rep.compression_opts.max_dict_bytes = max_dict_bytes;
 }

+void rocksdb_options_set_compression_options_zstd_max_train_bytes(
+    rocksdb_options_t* opt, int zstd_max_train_bytes) {
+  opt->rep.compression_opts.zstd_max_train_bytes = zstd_max_train_bytes;
+}
+
 void rocksdb_options_set_prefix_extractor(
    rocksdb_options_t* opt, rocksdb_slicetransform_t* prefix_extractor) {
  opt->rep.prefix_extractor.reset(prefix_extractor);
@@ -2527,35 +2666,67 @@ void rocksdb_options_set_max_write_buffer_number(rocksdb_options_t* opt, int n)
  opt->rep.max_write_buffer_number = n;
 }

+int rocksdb_options_get_max_write_buffer_number(rocksdb_options_t* opt) {
+  return opt->rep.max_write_buffer_number;
+}
+
 void rocksdb_options_set_min_write_buffer_number_to_merge(rocksdb_options_t* opt, int n) {
  opt->rep.min_write_buffer_number_to_merge = n;
 }

+int rocksdb_options_get_min_write_buffer_number_to_merge(
+    rocksdb_options_t* opt) {
+  return opt->rep.min_write_buffer_number_to_merge;
+}
+
 void rocksdb_options_set_max_write_buffer_number_to_maintain(
    rocksdb_options_t* opt, int n) {
  opt->rep.max_write_buffer_number_to_maintain = n;
 }

+int rocksdb_options_get_max_write_buffer_number_to_maintain(
+    rocksdb_options_t* opt) {
+  return opt->rep.max_write_buffer_number_to_maintain;
+}
+
 void rocksdb_options_set_max_write_buffer_size_to_maintain(
    rocksdb_options_t* opt, int64_t n) {
  opt->rep.max_write_buffer_size_to_maintain = n;
 }

+int64_t rocksdb_options_get_max_write_buffer_size_to_maintain(
+    rocksdb_options_t* opt) {
+  return opt->rep.max_write_buffer_size_to_maintain;
+}
+
 void rocksdb_options_set_enable_pipelined_write(rocksdb_options_t* opt,
                                                unsigned char v) {
  opt->rep.enable_pipelined_write = v;
 }

+unsigned char rocksdb_options_get_enable_pipelined_write(
+    rocksdb_options_t* opt) {
+  return opt->rep.enable_pipelined_write;
+}
+
 void rocksdb_options_set_unordered_write(rocksdb_options_t* opt,
                                         unsigned char v) {
  opt->rep.unordered_write = v;
 }

+unsigned char rocksdb_options_get_unordered_write(rocksdb_options_t* opt) {
+  return opt->rep.unordered_write;
+}
+
 void rocksdb_options_set_max_subcompactions(rocksdb_options_t* opt,
                                            uint32_t n) {
  opt->rep.max_subcompactions = n;
 }

+uint32_t rocksdb_options_get_max_subcompactions(rocksdb_options_t* opt) {
+  return opt->rep.max_subcompactions;
+}
+
 void rocksdb_options_set_max_background_jobs(rocksdb_options_t* opt, int n) {
  opt->rep.max_background_jobs = n;
 }
@@ -4441,11 +4612,25 @@ uint64_t rocksdb_approximate_memory_usage_get_cache_total(
  return memory_usage->cache_total;
 }

+void rocksdb_options_set_dump_malloc_stats(rocksdb_options_t* opt,
+                                           unsigned char val) {
+  opt->rep.dump_malloc_stats = val;
+}
+
+void rocksdb_options_set_memtable_whole_key_filtering(rocksdb_options_t* opt,
+                                                      unsigned char val) {
+  opt->rep.memtable_whole_key_filtering = val;
+}
+
 // deletes container with memory usage estimates
 void rocksdb_approximate_memory_usage_destroy(rocksdb_memory_usage_t* usage) {
  delete usage;
 }

+void rocksdb_cancel_all_background_work(rocksdb_t* db, unsigned char wait) {
+  CancelAllBackgroundWork(db->rep, wait);
+}
+
 }  // end extern "C"

 #endif  // !ROCKSDB_LITE
@@ -1461,6 +1461,294 @@ int main(int argc, char** argv) {
    rocksdb_cuckoo_options_destroy(cuckoo_options);
  }

+  StartPhase("options");
+  {
+    rocksdb_options_t* o;
+    o = rocksdb_options_create();
+
+    // Set and check options.
+    rocksdb_options_set_allow_ingest_behind(o, 1);
+    CheckCondition(1 == rocksdb_options_get_allow_ingest_behind(o));
+
+    rocksdb_options_compaction_readahead_size(o, 10);
+    CheckCondition(10 == rocksdb_options_get_compaction_readahead_size(o));
+
+    rocksdb_options_set_create_if_missing(o, 1);
+    CheckCondition(1 == rocksdb_options_get_create_if_missing(o));
+
+    rocksdb_options_set_create_missing_column_families(o, 1);
+    CheckCondition(1 == rocksdb_options_get_create_missing_column_families(o));
+
+    rocksdb_options_set_error_if_exists(o, 1);
+    CheckCondition(1 == rocksdb_options_get_error_if_exists(o));
+
+    rocksdb_options_set_paranoid_checks(o, 1);
+    CheckCondition(1 == rocksdb_options_get_paranoid_checks(o));
+
+    rocksdb_options_set_info_log_level(o, 3);
+    CheckCondition(3 == rocksdb_options_get_info_log_level(o));
+
+    rocksdb_options_set_write_buffer_size(o, 100);
+    CheckCondition(100 == rocksdb_options_get_write_buffer_size(o));
+
+    rocksdb_options_set_db_write_buffer_size(o, 1000);
+    CheckCondition(1000 == rocksdb_options_get_db_write_buffer_size(o));
+
+    rocksdb_options_set_max_open_files(o, 21);
+    CheckCondition(21 == rocksdb_options_get_max_open_files(o));
+
+    rocksdb_options_set_max_file_opening_threads(o, 5);
+    CheckCondition(5 == rocksdb_options_get_max_file_opening_threads(o));
+
+    rocksdb_options_set_max_total_wal_size(o, 400);
+    CheckCondition(400 == rocksdb_options_get_max_total_wal_size(o));
+
+    rocksdb_options_set_num_levels(o, 7);
+    CheckCondition(7 == rocksdb_options_get_num_levels(o));
+
+    rocksdb_options_set_level0_file_num_compaction_trigger(o, 4);
+    CheckCondition(4 ==
+                   rocksdb_options_get_level0_file_num_compaction_trigger(o));
+
+    rocksdb_options_set_level0_slowdown_writes_trigger(o, 6);
+    CheckCondition(6 == rocksdb_options_get_level0_slowdown_writes_trigger(o));
+
+    rocksdb_options_set_level0_stop_writes_trigger(o, 8);
+    CheckCondition(8 == rocksdb_options_get_level0_stop_writes_trigger(o));
+
+    rocksdb_options_set_target_file_size_base(o, 256);
+    CheckCondition(256 == rocksdb_options_get_target_file_size_base(o));
+
+    rocksdb_options_set_target_file_size_multiplier(o, 3);
+    CheckCondition(3 == rocksdb_options_get_target_file_size_multiplier(o));
+
+    rocksdb_options_set_max_bytes_for_level_base(o, 1024);
+    CheckCondition(1024 == rocksdb_options_get_max_bytes_for_level_base(o));
+
+    rocksdb_options_set_level_compaction_dynamic_level_bytes(o, 1);
+    CheckCondition(1 ==
+                   rocksdb_options_get_level_compaction_dynamic_level_bytes(o));
+
+    rocksdb_options_set_max_bytes_for_level_multiplier(o, 2.0);
+    CheckCondition(2.0 ==
+                   rocksdb_options_get_max_bytes_for_level_multiplier(o));
+
+    rocksdb_options_set_skip_stats_update_on_db_open(o, 1);
+    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(o));
+
+    rocksdb_options_set_skip_checking_sst_file_sizes_on_db_open(o, 1);
+    CheckCondition(
+        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(o));
+
+    rocksdb_options_set_max_write_buffer_number(o, 97);
+    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(o));
+
+    rocksdb_options_set_min_write_buffer_number_to_merge(o, 23);
+    CheckCondition(23 ==
+                   rocksdb_options_get_min_write_buffer_number_to_merge(o));
+
+    rocksdb_options_set_max_write_buffer_number_to_maintain(o, 64);
+    CheckCondition(64 ==
+                   rocksdb_options_get_max_write_buffer_number_to_maintain(o));
+
+    rocksdb_options_set_max_write_buffer_size_to_maintain(o, 50000);
+    CheckCondition(50000 ==
+                   rocksdb_options_get_max_write_buffer_size_to_maintain(o));
+
+    rocksdb_options_set_enable_pipelined_write(o, 1);
+    CheckCondition(1 == rocksdb_options_get_enable_pipelined_write(o));
+
+    rocksdb_options_set_unordered_write(o, 1);
+    CheckCondition(1 == rocksdb_options_get_unordered_write(o));
+
+    rocksdb_options_set_max_subcompactions(o, 123456);
+    CheckCondition(123456 == rocksdb_options_get_max_subcompactions(o));
+
+    // Create a copy that should be equal to the original.
+    rocksdb_options_t* copy;
+    copy = rocksdb_options_create_copy(o);
+
+    CheckCondition(1 == rocksdb_options_get_allow_ingest_behind(copy));
+    CheckCondition(10 == rocksdb_options_get_compaction_readahead_size(copy));
+    CheckCondition(1 == rocksdb_options_get_create_if_missing(copy));
+    CheckCondition(1 ==
+                   rocksdb_options_get_create_missing_column_families(copy));
+    CheckCondition(1 == rocksdb_options_get_error_if_exists(copy));
+    CheckCondition(1 == rocksdb_options_get_paranoid_checks(copy));
+    CheckCondition(3 == rocksdb_options_get_info_log_level(copy));
+    CheckCondition(100 == rocksdb_options_get_write_buffer_size(copy));
+    CheckCondition(1000 == rocksdb_options_get_db_write_buffer_size(copy));
+    CheckCondition(21 == rocksdb_options_get_max_open_files(copy));
+    CheckCondition(5 == rocksdb_options_get_max_file_opening_threads(copy));
+    CheckCondition(400 == rocksdb_options_get_max_total_wal_size(copy));
+    CheckCondition(7 == rocksdb_options_get_num_levels(copy));
+    CheckCondition(
+        4 == rocksdb_options_get_level0_file_num_compaction_trigger(copy));
+    CheckCondition(6 ==
+                   rocksdb_options_get_level0_slowdown_writes_trigger(copy));
+    CheckCondition(8 == rocksdb_options_get_level0_stop_writes_trigger(copy));
+    CheckCondition(256 == rocksdb_options_get_target_file_size_base(copy));
+    CheckCondition(3 == rocksdb_options_get_target_file_size_multiplier(copy));
+    CheckCondition(1024 == rocksdb_options_get_max_bytes_for_level_base(copy));
+    CheckCondition(
+        1 == rocksdb_options_get_level_compaction_dynamic_level_bytes(copy));
+    CheckCondition(2.0 ==
+                   rocksdb_options_get_max_bytes_for_level_multiplier(copy));
+    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(copy));
+    CheckCondition(
+        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(copy));
+    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(copy));
+    CheckCondition(23 ==
+                   rocksdb_options_get_min_write_buffer_number_to_merge(copy));
+    CheckCondition(
+        64 == rocksdb_options_get_max_write_buffer_number_to_maintain(copy));
+    CheckCondition(50000 ==
+                   rocksdb_options_get_max_write_buffer_size_to_maintain(copy));
+    CheckCondition(1 == rocksdb_options_get_enable_pipelined_write(copy));
+    CheckCondition(1 == rocksdb_options_get_unordered_write(copy));
+    CheckCondition(123456 == rocksdb_options_get_max_subcompactions(copy));
+
+    // Copies should be independent.
+    rocksdb_options_set_allow_ingest_behind(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_allow_ingest_behind(copy));
+    CheckCondition(1 == rocksdb_options_get_allow_ingest_behind(o));
+
+    rocksdb_options_compaction_readahead_size(copy, 20);
+    CheckCondition(20 == rocksdb_options_get_compaction_readahead_size(copy));
+    CheckCondition(10 == rocksdb_options_get_compaction_readahead_size(o));
+
+    rocksdb_options_set_create_if_missing(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_create_if_missing(copy));
+    CheckCondition(1 == rocksdb_options_get_create_if_missing(o));
+
+    rocksdb_options_set_create_missing_column_families(copy, 0);
+    CheckCondition(0 ==
+                   rocksdb_options_get_create_missing_column_families(copy));
+    CheckCondition(1 == rocksdb_options_get_create_missing_column_families(o));
+
+    rocksdb_options_set_error_if_exists(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_error_if_exists(copy));
+    CheckCondition(1 == rocksdb_options_get_error_if_exists(o));
+
+    rocksdb_options_set_paranoid_checks(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_paranoid_checks(copy));
+    CheckCondition(1 == rocksdb_options_get_paranoid_checks(o));
+
+    rocksdb_options_set_info_log_level(copy, 2);
+    CheckCondition(2 == rocksdb_options_get_info_log_level(copy));
+    CheckCondition(3 == rocksdb_options_get_info_log_level(o));
+
+    rocksdb_options_set_write_buffer_size(copy, 200);
+    CheckCondition(200 == rocksdb_options_get_write_buffer_size(copy));
+    CheckCondition(100 == rocksdb_options_get_write_buffer_size(o));
+
+    rocksdb_options_set_db_write_buffer_size(copy, 2000);
+    CheckCondition(2000 == rocksdb_options_get_db_write_buffer_size(copy));
+    CheckCondition(1000 == rocksdb_options_get_db_write_buffer_size(o));
+
+    rocksdb_options_set_max_open_files(copy, 42);
+    CheckCondition(42 == rocksdb_options_get_max_open_files(copy));
+    CheckCondition(21 == rocksdb_options_get_max_open_files(o));
+
+    rocksdb_options_set_max_file_opening_threads(copy, 3);
+    CheckCondition(3 == rocksdb_options_get_max_file_opening_threads(copy));
+    CheckCondition(5 == rocksdb_options_get_max_file_opening_threads(o));
+
+    rocksdb_options_set_max_total_wal_size(copy, 4000);
+    CheckCondition(4000 == rocksdb_options_get_max_total_wal_size(copy));
+    CheckCondition(400 == rocksdb_options_get_max_total_wal_size(o));
+
+    rocksdb_options_set_num_levels(copy, 6);
+    CheckCondition(6 == rocksdb_options_get_num_levels(copy));
+    CheckCondition(7 == rocksdb_options_get_num_levels(o));
+
+    rocksdb_options_set_level0_file_num_compaction_trigger(copy, 14);
+    CheckCondition(
+        14 == rocksdb_options_get_level0_file_num_compaction_trigger(copy));
+    CheckCondition(4 ==
+                   rocksdb_options_get_level0_file_num_compaction_trigger(o));
+
+    rocksdb_options_set_level0_slowdown_writes_trigger(copy, 61);
+    CheckCondition(61 ==
+                   rocksdb_options_get_level0_slowdown_writes_trigger(copy));
+    CheckCondition(6 == rocksdb_options_get_level0_slowdown_writes_trigger(o));
+
+    rocksdb_options_set_level0_stop_writes_trigger(copy, 17);
+    CheckCondition(17 == rocksdb_options_get_level0_stop_writes_trigger(copy));
+    CheckCondition(8 == rocksdb_options_get_level0_stop_writes_trigger(o));
+
+    rocksdb_options_set_target_file_size_base(copy, 128);
+    CheckCondition(128 == rocksdb_options_get_target_file_size_base(copy));
+    CheckCondition(256 == rocksdb_options_get_target_file_size_base(o));
+
+    rocksdb_options_set_target_file_size_multiplier(copy, 13);
+    CheckCondition(13 == rocksdb_options_get_target_file_size_multiplier(copy));
+    CheckCondition(3 == rocksdb_options_get_target_file_size_multiplier(o));
+
+    rocksdb_options_set_max_bytes_for_level_base(copy, 900);
+    CheckCondition(900 == rocksdb_options_get_max_bytes_for_level_base(copy));
+    CheckCondition(1024 == rocksdb_options_get_max_bytes_for_level_base(o));
+
+    rocksdb_options_set_level_compaction_dynamic_level_bytes(copy, 0);
+    CheckCondition(
+        0 == rocksdb_options_get_level_compaction_dynamic_level_bytes(copy));
+    CheckCondition(1 ==
+                   rocksdb_options_get_level_compaction_dynamic_level_bytes(o));
+
+    rocksdb_options_set_max_bytes_for_level_multiplier(copy, 8.0);
+    CheckCondition(8.0 ==
+                   rocksdb_options_get_max_bytes_for_level_multiplier(copy));
+    CheckCondition(2.0 ==
+                   rocksdb_options_get_max_bytes_for_level_multiplier(o));
+
+    rocksdb_options_set_skip_stats_update_on_db_open(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_skip_stats_update_on_db_open(copy));
+    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(o));
+
+    rocksdb_options_set_skip_checking_sst_file_sizes_on_db_open(copy, 0);
+    CheckCondition(
+        0 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(copy));
+    CheckCondition(
+        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(o));
+
+    rocksdb_options_set_max_write_buffer_number(copy, 2000);
+    CheckCondition(2000 == rocksdb_options_get_max_write_buffer_number(copy));
+    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(o));
+
+    rocksdb_options_set_min_write_buffer_number_to_merge(copy, 146);
+    CheckCondition(146 ==
+                   rocksdb_options_get_min_write_buffer_number_to_merge(copy));
+    CheckCondition(23 ==
+                   rocksdb_options_get_min_write_buffer_number_to_merge(o));
+
+    rocksdb_options_set_max_write_buffer_number_to_maintain(copy, 128);
+    CheckCondition(
+        128 == rocksdb_options_get_max_write_buffer_number_to_maintain(copy));
+    CheckCondition(64 ==
+                   rocksdb_options_get_max_write_buffer_number_to_maintain(o));
+
+    rocksdb_options_set_max_write_buffer_size_to_maintain(copy, 9000);
+    CheckCondition(9000 ==
+                   rocksdb_options_get_max_write_buffer_size_to_maintain(copy));
+    CheckCondition(50000 ==
+                   rocksdb_options_get_max_write_buffer_size_to_maintain(o));
+
+    rocksdb_options_set_enable_pipelined_write(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_enable_pipelined_write(copy));
+    CheckCondition(1 == rocksdb_options_get_enable_pipelined_write(o));
+
+    rocksdb_options_set_unordered_write(copy, 0);
+    CheckCondition(0 == rocksdb_options_get_unordered_write(copy));
+    CheckCondition(1 == rocksdb_options_get_unordered_write(o));
+
+    rocksdb_options_set_max_subcompactions(copy, 90001);
+    CheckCondition(90001 == rocksdb_options_get_max_subcompactions(copy));
+    CheckCondition(123456 == rocksdb_options_get_max_subcompactions(o));
+
+    rocksdb_options_destroy(copy);
+    rocksdb_options_destroy(o);
+  }
+
  StartPhase("iterate_upper_bound");
  {
    // Create new empty database
@@ -1840,6 +2128,9 @@ int main(int argc, char** argv) {
    CheckNoError(err);
  }

+  StartPhase("cancel_all_background_work");
+  rocksdb_cancel_all_background_work(db, 1);
+
  StartPhase("cleanup");
  rocksdb_close(db);
  rocksdb_options_destroy(options);
@@ -8,16 +8,16 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include <algorithm>
-#include <vector>
 #include <string>
 #include <thread>
+#include <vector>

 #include "db/db_impl/db_impl.h"
 #include "db/db_test_util.h"
-#include "memtable/hash_skiplist_rep.h"
 #include "options/options_parser.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
+#include "rocksdb/convenience.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -287,7 +287,9 @@ class ColumnFamilyTestBase : public testing::Test {
      // Verify the CF options of the returned CF handle.
      ColumnFamilyDescriptor desc;
      ASSERT_OK(handles_[cfi]->GetDescriptor(&desc));
-      RocksDBOptionsParser::VerifyCFOptions(desc.options, current_cf_opt);
+      RocksDBOptionsParser::VerifyCFOptions(ConfigOptions(), desc.options,
+                                            current_cf_opt);
+
 #endif  // !ROCKSDB_LITE
      cfi++;
    }
@@ -23,7 +23,7 @@ const uint64_t kRangeTombstoneSentinel =

 int sstableKeyCompare(const Comparator* user_cmp, const InternalKey& a,
                      const InternalKey& b) {
-  auto c = user_cmp->Compare(a.user_key(), b.user_key());
+  auto c = user_cmp->CompareWithoutTimestamp(a.user_key(), b.user_key());
  if (c != 0) {
    return c;
  }
@@ -542,7 +542,7 @@ void CompactionJob::GenSubcompactionBoundaries() {
    // Greedily add ranges to the subcompaction until the sum of the ranges'
    // sizes becomes >= the expected mean size of a subcompaction
    sum = 0;
-    for (size_t i = 0; i < ranges.size() - 1; i++) {
+    for (size_t i = 0; i + 1 < ranges.size(); i++) {
      sum += ranges[i].size;
      if (subcompactions == 1) {
        // If there's only one left to schedule then it goes to the end so no
@@ -655,7 +655,8 @@ Status CompactionJob::Run() {
            TableReaderCaller::kCompactionRefill, /*arena=*/nullptr,
            /*skip_filters=*/false, compact_->compaction->output_level(),
            /*smallest_compaction_key=*/nullptr,
-            /*largest_compaction_key=*/nullptr);
+            /*largest_compaction_key=*/nullptr,
+            /*allow_unprepared_value=*/false);
        auto s = iter->status();

        if (s.ok() && paranoid_file_checks_) {
@@ -937,7 +938,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    assert(sub_compact->builder != nullptr);
    assert(sub_compact->current_output() != nullptr);
    sub_compact->builder->Add(key, value);
-    sub_compact->current_output_file_size = sub_compact->builder->FileSize();
+    sub_compact->current_output_file_size =
+        sub_compact->builder->EstimatedFileSize();
    const ParsedInternalKey& ikey = c_iter->ikey();
    sub_compact->current_output()->meta.UpdateBoundaries(
        key, value, ikey.sequence, ikey.type);
@@ -1085,6 +1085,8 @@ void CompactionPicker::PickFilesMarkedForCompaction(
  Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage));
  size_t random_file_index = static_cast<size_t>(rnd.Uniform(
      static_cast<uint64_t>(vstorage->FilesMarkedForCompaction().size())));
+  TEST_SYNC_POINT_CALLBACK("CompactionPicker::PickFilesMarkedForCompaction",
+                           &random_file_index);

  if (continuation(vstorage->FilesMarkedForCompaction()[random_file_index])) {
    // found the compaction!
@@ -71,6 +71,7 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(
  if (current_time > mutable_cf_options.ttl) {
    for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
      FileMetaData* f = *ritr;
+      assert(f);
      if (f->fd.table_reader && f->fd.table_reader->GetTableProperties()) {
        uint64_t creation_time =
            f->fd.table_reader->GetTableProperties()->creation_time;
@@ -96,7 +97,8 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(

  for (const auto& f : inputs[0].files) {
    uint64_t creation_time = 0;
-    if (f && f->fd.table_reader && f->fd.table_reader->GetTableProperties()) {
+    assert(f);
+    if (f->fd.table_reader && f->fd.table_reader->GetTableProperties()) {
      creation_time = f->fd.table_reader->GetTableProperties()->creation_time;
    }
    ROCKS_LOG_BUFFER(log_buffer,
@@ -93,9 +93,13 @@ class LevelCompactionBuilder {
  // otherwise, returns false.
  bool PickIntraL0Compaction();

-  void PickExpiredTtlFiles();
-
-  void PickFilesMarkedForPeriodicCompaction();
+  // Picks a file from level_files to compact.
+  // level_files is a vector of (level, file metadata) in ascending order of
+  // level. If compact_to_next_level is true, compact the file to the next
+  // level, otherwise, compact to the same level as the input file.
+  void PickFileToCompact(
+      const autovector<std::pair<int, FileMetaData*>>& level_files,
+      bool compact_to_next_level);

  const std::string& cf_name_;
  VersionStorageInfo* vstorage_;
@@ -124,72 +128,34 @@ class LevelCompactionBuilder {
  static const int kMinFilesForIntraL0Compaction = 4;
 };

-void LevelCompactionBuilder::PickExpiredTtlFiles() {
-  if (vstorage_->ExpiredTtlFiles().empty()) {
-    return;
-  }
-
-  auto continuation = [&](std::pair<int, FileMetaData*> level_file) {
+void LevelCompactionBuilder::PickFileToCompact(
+    const autovector<std::pair<int, FileMetaData*>>& level_files,
+    bool compact_to_next_level) {
+  for (auto& level_file : level_files) {
    // If it's being compacted it has nothing to do here.
    // If this assert() fails that means that some function marked some
    // files as being_compacted, but didn't call ComputeCompactionScore()
    assert(!level_file.second->being_compacted);
    start_level_ = level_file.first;
-    output_level_ =
-        (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1;
-
-    if ((start_level_ == vstorage_->num_non_empty_levels() - 1) ||
+    if ((compact_to_next_level &&
+         start_level_ == vstorage_->num_non_empty_levels() - 1) ||
        (start_level_ == 0 &&
         !compaction_picker_->level0_compactions_in_progress()->empty())) {
-      return false;
+      continue;
+    }
+    if (compact_to_next_level) {
+      output_level_ =
+          (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1;
+    } else {
+      output_level_ = start_level_;
    }
-
    start_level_inputs_.files = {level_file.second};
    start_level_inputs_.level = start_level_;
-    return compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
-                                                      &start_level_inputs_);
-  };
-
-  for (auto& level_file : vstorage_->ExpiredTtlFiles()) {
-    if (continuation(level_file)) {
-      // found the compaction!
+    if (compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
+                                                   &start_level_inputs_)) {
      return;
    }
  }
-
-  start_level_inputs_.files.clear();
-}
-
-void LevelCompactionBuilder::PickFilesMarkedForPeriodicCompaction() {
-  if (vstorage_->FilesMarkedForPeriodicCompaction().empty()) {
-    return;
-  }
-
-  auto continuation = [&](std::pair<int, FileMetaData*> level_file) {
-    // If it's being compacted it has nothing to do here.
-    // If this assert() fails that means that some function marked some
-    // files as being_compacted, but didn't call ComputeCompactionScore()
-    assert(!level_file.second->being_compacted);
-    output_level_ = start_level_ = level_file.first;
-
-    if (start_level_ == 0 &&
-        !compaction_picker_->level0_compactions_in_progress()->empty()) {
-      return false;
-    }
-
-    start_level_inputs_.files = {level_file.second};
-    start_level_inputs_.level = start_level_;
-    return compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
-                                                      &start_level_inputs_);
-  };
-
-  for (auto& level_file : vstorage_->FilesMarkedForPeriodicCompaction()) {
-    if (continuation(level_file)) {
-      // found the compaction!
-      return;
-    }
-  }
-
  start_level_inputs_.files.clear();
 }

@@ -238,63 +204,46 @@ void LevelCompactionBuilder::SetupInitialFiles() {
          }
        }
      }
+    } else {
+      // Compaction scores are sorted in descending order, no further scores
+      // will be >= 1.
+      break;
    }
  }
+  if (!start_level_inputs_.empty()) {
+    return;
+  }

  // if we didn't find a compaction, check if there are any files marked for
  // compaction
-  if (start_level_inputs_.empty()) {
-    parent_index_ = base_index_ = -1;
+  parent_index_ = base_index_ = -1;

-    compaction_picker_->PickFilesMarkedForCompaction(
-        cf_name_, vstorage_, &start_level_, &output_level_,
-        &start_level_inputs_);
-    if (!start_level_inputs_.empty()) {
-      compaction_reason_ = CompactionReason::kFilesMarkedForCompaction;
-      return;
-    }
+  compaction_picker_->PickFilesMarkedForCompaction(
+      cf_name_, vstorage_, &start_level_, &output_level_, &start_level_inputs_);
+  if (!start_level_inputs_.empty()) {
+    compaction_reason_ = CompactionReason::kFilesMarkedForCompaction;
+    return;
  }

  // Bottommost Files Compaction on deleting tombstones
-  if (start_level_inputs_.empty()) {
-    size_t i;
-    for (i = 0; i < vstorage_->BottommostFilesMarkedForCompaction().size();
-         ++i) {
-      auto& level_and_file = vstorage_->BottommostFilesMarkedForCompaction()[i];
-      assert(!level_and_file.second->being_compacted);
-      start_level_inputs_.level = output_level_ = start_level_ =
-          level_and_file.first;
-      start_level_inputs_.files = {level_and_file.second};
-      if (compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
-                                                     &start_level_inputs_)) {
-        break;
-      }
-    }
-    if (i == vstorage_->BottommostFilesMarkedForCompaction().size()) {
-      start_level_inputs_.clear();
-    } else {
-      assert(!start_level_inputs_.empty());
-      compaction_reason_ = CompactionReason::kBottommostFiles;
-      return;
-    }
+  PickFileToCompact(vstorage_->BottommostFilesMarkedForCompaction(), false);
+  if (!start_level_inputs_.empty()) {
+    compaction_reason_ = CompactionReason::kBottommostFiles;
+    return;
  }

  // TTL Compaction
-  if (start_level_inputs_.empty()) {
-    PickExpiredTtlFiles();
-    if (!start_level_inputs_.empty()) {
-      compaction_reason_ = CompactionReason::kTtl;
-      return;
-    }
+  PickFileToCompact(vstorage_->ExpiredTtlFiles(), true);
+  if (!start_level_inputs_.empty()) {
+    compaction_reason_ = CompactionReason::kTtl;
+    return;
  }

  // Periodic Compaction
-  if (start_level_inputs_.empty()) {
-    PickFilesMarkedForPeriodicCompaction();
-    if (!start_level_inputs_.empty()) {
-      compaction_reason_ = CompactionReason::kPeriodicCompaction;
-      return;
-    }
+  PickFileToCompact(vstorage_->FilesMarkedForPeriodicCompaction(), false);
+  if (!start_level_inputs_.empty()) {
+    compaction_reason_ = CompactionReason::kPeriodicCompaction;
+    return;
  }
 }

@@ -78,8 +78,17 @@ class CompactionPickerTest : public testing::Test {
    vstorage_->CalculateBaseBytes(ioptions_, mutable_cf_options_);
  }

+  // Create a new VersionStorageInfo object so we can add mode files and then
+  // merge it with the existing VersionStorageInfo
+  void AddVersionStorage() {
+    temp_vstorage_.reset(new VersionStorageInfo(
+        &icmp_, ucmp_, options_.num_levels, ioptions_.compaction_style,
+        vstorage_.get(), false));
+  }
+
  void DeleteVersionStorage() {
    vstorage_.reset();
+    temp_vstorage_.reset();
    files_.clear();
    file_map_.clear();
    input_files_.clear();
@@ -88,18 +97,24 @@ class CompactionPickerTest : public testing::Test {
  void Add(int level, uint32_t file_number, const char* smallest,
           const char* largest, uint64_t file_size = 1, uint32_t path_id = 0,
           SequenceNumber smallest_seq = 100, SequenceNumber largest_seq = 100,
-           size_t compensated_file_size = 0) {
-    assert(level < vstorage_->num_levels());
+           size_t compensated_file_size = 0, bool marked_for_compact = false) {
+    VersionStorageInfo* vstorage;
+    if (temp_vstorage_) {
+      vstorage = temp_vstorage_.get();
+    } else {
+      vstorage = vstorage_.get();
+    }
+    assert(level < vstorage->num_levels());
    FileMetaData* f = new FileMetaData(
        file_number, path_id, file_size,
        InternalKey(smallest, smallest_seq, kTypeValue),
        InternalKey(largest, largest_seq, kTypeValue), smallest_seq,
-        largest_seq, /* marked_for_compact */ false, kInvalidBlobFileNumber,
+        largest_seq, marked_for_compact, kInvalidBlobFileNumber,
        kUnknownOldestAncesterTime, kUnknownFileCreationTime,
        kUnknownFileChecksum, kUnknownFileChecksumFuncName);
    f->compensated_file_size =
        (compensated_file_size != 0) ? compensated_file_size : file_size;
-    vstorage_->AddFile(level, f);
+    vstorage->AddFile(level, f);
    files_.emplace_back(f);
    file_map_.insert({file_number, {f, level}});
  }
@@ -122,6 +137,12 @@ class CompactionPickerTest : public testing::Test {
  }

  void UpdateVersionStorageInfo() {
+    if (temp_vstorage_) {
+      VersionBuilder builder(FileOptions(), &ioptions_, nullptr,
+                             vstorage_.get(), nullptr);
+      builder.SaveTo(temp_vstorage_.get());
+      vstorage_ = std::move(temp_vstorage_);
+    }
    vstorage_->CalculateBaseBytes(ioptions_, mutable_cf_options_);
    vstorage_->UpdateFilesByCompactionPri(ioptions_.compaction_pri);
    vstorage_->UpdateNumNonEmptyLevels();
@@ -132,6 +153,28 @@ class CompactionPickerTest : public testing::Test {
    vstorage_->ComputeFilesMarkedForCompaction();
    vstorage_->SetFinalized();
  }
+  void AddFileToVersionStorage(int level, uint32_t file_number,
+                               const char* smallest, const char* largest,
+                               uint64_t file_size = 1, uint32_t path_id = 0,
+                               SequenceNumber smallest_seq = 100,
+                               SequenceNumber largest_seq = 100,
+                               size_t compensated_file_size = 0,
+                               bool marked_for_compact = false) {
+    VersionStorageInfo* base_vstorage = vstorage_.release();
+    vstorage_.reset(new VersionStorageInfo(&icmp_, ucmp_, options_.num_levels,
+                                           kCompactionStyleUniversal,
+                                           base_vstorage, false));
+    Add(level, file_number, smallest, largest, file_size, path_id, smallest_seq,
+        largest_seq, compensated_file_size, marked_for_compact);
+
+    VersionBuilder builder(FileOptions(), &ioptions_, nullptr, base_vstorage,
+                           nullptr);
+    builder.SaveTo(vstorage_.get());
+    UpdateVersionStorageInfo();
+  }
+
+ private:
+  std::unique_ptr<VersionStorageInfo> temp_vstorage_;
 };

 TEST_F(CompactionPickerTest, Empty) {
@@ -371,8 +414,8 @@ TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
  NewVersionStorage(num_levels, kCompactionStyleLevel);
  Add(0, 1U, "150", "200");
-  Add(num_levels - 1, 3U, "200", "250", 300U);
-  Add(num_levels - 1, 4U, "300", "350", 3000U);
+  Add(num_levels - 1, 2U, "200", "250", 300U);
+  Add(num_levels - 1, 3U, "300", "350", 3000U);
  Add(num_levels - 1, 4U, "400", "450", 3U);
  Add(num_levels - 2, 5U, "150", "180", 300U);
  Add(num_levels - 2, 6U, "181", "350", 500U);
@@ -739,7 +782,7 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
  Add(2, 8U, "201", "300",
      60000000U);  // Overlaps with file 28, 29, total size 521M

-  Add(3, 26U, "100", "110", 261000000U);
+  Add(3, 25U, "100", "110", 261000000U);
  Add(3, 26U, "150", "170", 261000000U);
  Add(3, 27U, "171", "179", 260000000U);
  Add(3, 28U, "191", "220", 260000000U);
@@ -1255,7 +1298,7 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded1) {
  // Size ratio L4/L3 is 9.9
  // After merge from L3, L4 size is 1000900
  Add(4, 11U, "400", "500", 999900);
-  Add(5, 11U, "400", "500", 8007200);
+  Add(5, 12U, "400", "500", 8007200);

  UpdateVersionStorageInfo();

@@ -1568,8 +1611,8 @@ TEST_F(CompactionPickerTest, IsTrivialMoveOn) {

  Add(3, 5U, "120", "130", 7000U);
  Add(3, 6U, "170", "180", 7000U);
-  Add(3, 5U, "220", "230", 7000U);
-  Add(3, 5U, "270", "280", 7000U);
+  Add(3, 7U, "220", "230", 7000U);
+  Add(3, 8U, "270", "280", 7000U);
  UpdateVersionStorageInfo();

  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
@@ -1733,6 +1776,165 @@ TEST_F(CompactionPickerTest, IntraL0ForEarliestSeqno) {
  ASSERT_EQ(0, compaction->output_level());
 }

+#ifndef ROCKSDB_LITE
+TEST_F(CompactionPickerTest, UniversalMarkedCompactionFullOverlap) {
+  const uint64_t kFileSize = 100000;
+
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+
+  // This test covers the case where a "regular" universal compaction is
+  // scheduled first, followed by a delete triggered compaction. The latter
+  // should fail
+  NewVersionStorage(5, kCompactionStyleUniversal);
+
+  Add(0, 1U, "150", "200", kFileSize, 0, 500, 550);
+  Add(0, 2U, "201", "250", 2 * kFileSize, 0, 401, 450);
+  Add(0, 4U, "260", "300", 4 * kFileSize, 0, 260, 300);
+  Add(3, 5U, "010", "080", 8 * kFileSize, 0, 200, 251);
+  Add(4, 3U, "301", "350", 8 * kFileSize, 0, 101, 150);
+  Add(4, 6U, "501", "750", 8 * kFileSize, 0, 101, 150);
+
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(
+      universal_compaction_picker.PickCompaction(
+          cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+
+  ASSERT_TRUE(compaction);
+  // Validate that its a compaction to reduce sorted runs
+  ASSERT_EQ(CompactionReason::kUniversalSortedRunNum,
+            compaction->compaction_reason());
+  ASSERT_EQ(0, compaction->output_level());
+  ASSERT_EQ(0, compaction->start_level());
+  ASSERT_EQ(2U, compaction->num_input_files(0));
+
+  AddVersionStorage();
+  // Simulate a flush and mark the file for compaction
+  Add(0, 7U, "150", "200", kFileSize, 0, 551, 600, 0, true);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction2(
+      universal_compaction_picker.PickCompaction(
+          cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_FALSE(compaction2);
+}
+
+TEST_F(CompactionPickerTest, UniversalMarkedCompactionFullOverlap2) {
+  const uint64_t kFileSize = 100000;
+
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+
+  // This test covers the case where a delete triggered compaction is
+  // scheduled first, followed by a "regular" compaction. The latter
+  // should fail
+  NewVersionStorage(5, kCompactionStyleUniversal);
+
+  // Mark file number 4 for compaction
+  Add(0, 4U, "260", "300", 4 * kFileSize, 0, 260, 300, 0, true);
+  Add(3, 5U, "240", "290", 8 * kFileSize, 0, 201, 250);
+  Add(4, 3U, "301", "350", 8 * kFileSize, 0, 101, 150);
+  Add(4, 6U, "501", "750", 8 * kFileSize, 0, 101, 150);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(
+      universal_compaction_picker.PickCompaction(
+          cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+
+  ASSERT_TRUE(compaction);
+  // Validate that its a delete triggered compaction
+  ASSERT_EQ(CompactionReason::kFilesMarkedForCompaction,
+            compaction->compaction_reason());
+  ASSERT_EQ(3, compaction->output_level());
+  ASSERT_EQ(0, compaction->start_level());
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  ASSERT_EQ(1U, compaction->num_input_files(1));
+
+  AddVersionStorage();
+  Add(0, 1U, "150", "200", kFileSize, 0, 500, 550);
+  Add(0, 2U, "201", "250", 2 * kFileSize, 0, 401, 450);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction2(
+      universal_compaction_picker.PickCompaction(
+          cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_FALSE(compaction2);
+}
+
+TEST_F(CompactionPickerTest, UniversalMarkedCompactionStartOutputOverlap) {
+  // The case where universal periodic compaction can be picked
+  // with some newer files being compacted.
+  const uint64_t kFileSize = 100000;
+
+  ioptions_.compaction_style = kCompactionStyleUniversal;
+
+  bool input_level_overlap = false;
+  bool output_level_overlap = false;
+  // Let's mark 2 files in 2 different levels for compaction. The
+  // compaction picker will randomly pick one, so use the sync point to
+  // ensure a deterministic order. Loop until both cases are covered
+  size_t random_index = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "CompactionPicker::PickFilesMarkedForCompaction", [&](void* arg) {
+        size_t* index = static_cast<size_t*>(arg);
+        *index = random_index;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  while (!input_level_overlap || !output_level_overlap) {
+    // Ensure that the L0 file gets picked first
+    random_index = !input_level_overlap ? 0 : 1;
+    UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+    NewVersionStorage(5, kCompactionStyleUniversal);
+
+    Add(0, 1U, "260", "300", 4 * kFileSize, 0, 260, 300, 0, true);
+    Add(3, 2U, "010", "020", 2 * kFileSize, 0, 201, 248);
+    Add(3, 3U, "250", "270", 2 * kFileSize, 0, 202, 249);
+    Add(3, 4U, "290", "310", 2 * kFileSize, 0, 203, 250);
+    Add(3, 5U, "310", "320", 2 * kFileSize, 0, 204, 251, 0, true);
+    Add(4, 6U, "301", "350", 8 * kFileSize, 0, 101, 150);
+    Add(4, 7U, "501", "750", 8 * kFileSize, 0, 101, 150);
+    UpdateVersionStorageInfo();
+
+    std::unique_ptr<Compaction> compaction(
+        universal_compaction_picker.PickCompaction(
+            cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+
+    ASSERT_TRUE(compaction);
+    // Validate that its a delete triggered compaction
+    ASSERT_EQ(CompactionReason::kFilesMarkedForCompaction,
+              compaction->compaction_reason());
+    ASSERT_TRUE(compaction->start_level() == 0 ||
+                compaction->start_level() == 3);
+    if (compaction->start_level() == 0) {
+      // The L0 file was picked. The next compaction will detect an
+      // overlap on its input level
+      input_level_overlap = true;
+      ASSERT_EQ(3, compaction->output_level());
+      ASSERT_EQ(1U, compaction->num_input_files(0));
+      ASSERT_EQ(3U, compaction->num_input_files(1));
+    } else {
+      // The level 3 file was picked. The next compaction will pick
+      // the L0 file and will detect overlap when adding output
+      // level inputs
+      output_level_overlap = true;
+      ASSERT_EQ(4, compaction->output_level());
+      ASSERT_EQ(2U, compaction->num_input_files(0));
+      ASSERT_EQ(1U, compaction->num_input_files(1));
+    }
+
+    vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_);
+    // After recomputing the compaction score, only one marked file will remain
+    random_index = 0;
+    std::unique_ptr<Compaction> compaction2(
+        universal_compaction_picker.PickCompaction(
+            cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+    ASSERT_FALSE(compaction2);
+    DeleteVersionStorage();
+  }
+}
+#endif  // ROCKSDB_LITE
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -120,8 +120,7 @@ class UniversalCompactionBuilder {
  LogBuffer* log_buffer_;

  static std::vector<SortedRun> CalculateSortedRuns(
-      const VersionStorageInfo& vstorage, const ImmutableCFOptions& ioptions,
-      const MutableCFOptions& mutable_cf_options);
+      const VersionStorageInfo& vstorage);

  // Pick a path ID to place a newly generated file, with its estimated file
  // size.
@@ -325,8 +324,7 @@ void UniversalCompactionBuilder::SortedRun::DumpSizeInfo(

 std::vector<UniversalCompactionBuilder::SortedRun>
 UniversalCompactionBuilder::CalculateSortedRuns(
-    const VersionStorageInfo& vstorage, const ImmutableCFOptions& /*ioptions*/,
-    const MutableCFOptions& mutable_cf_options) {
+    const VersionStorageInfo& vstorage) {
  std::vector<UniversalCompactionBuilder::SortedRun> ret;
  for (FileMetaData* f : vstorage.LevelFiles(0)) {
    ret.emplace_back(0, f, f->fd.GetFileSize(), f->compensated_file_size,
@@ -336,27 +334,16 @@ UniversalCompactionBuilder::CalculateSortedRuns(
    uint64_t total_compensated_size = 0U;
    uint64_t total_size = 0U;
    bool being_compacted = false;
-    bool is_first = true;
    for (FileMetaData* f : vstorage.LevelFiles(level)) {
      total_compensated_size += f->compensated_file_size;
      total_size += f->fd.GetFileSize();
-      if (mutable_cf_options.compaction_options_universal.allow_trivial_move ==
-          true) {
-        if (f->being_compacted) {
-          being_compacted = f->being_compacted;
-        }
-      } else {
-        // Compaction always includes all files for a non-zero level, so for a
-        // non-zero level, all the files should share the same being_compacted
-        // value.
-        // This assumption is only valid when
-        // mutable_cf_options.compaction_options_universal.allow_trivial_move
-        // is false
-        assert(is_first || f->being_compacted == being_compacted);
-      }
-      if (is_first) {
+      // Size amp, read amp and periodic compactions always include all files
+      // for a non-zero level. However, a delete triggered compaction and
+      // a trivial move might pick a subset of files in a sorted run. So
+      // always check all files in a sorted run and mark the entire run as
+      // being compacted if one or more files are being compacted
+      if (f->being_compacted) {
        being_compacted = f->being_compacted;
-        is_first = false;
      }
    }
    if (total_compensated_size > 0) {
@@ -372,8 +359,7 @@ UniversalCompactionBuilder::CalculateSortedRuns(
 Compaction* UniversalCompactionBuilder::PickCompaction() {
  const int kLevel0 = 0;
  score_ = vstorage_->CompactionScore(kLevel0);
-  sorted_runs_ =
-      CalculateSortedRuns(*vstorage_, ioptions_, mutable_cf_options_);
+  sorted_runs_ = CalculateSortedRuns(*vstorage_);

  if (sorted_runs_.size() == 0 ||
      (vstorage_->FilesMarkedForPeriodicCompaction().empty() &&
@@ -389,7 +375,7 @@ Compaction* UniversalCompactionBuilder::PickCompaction() {
  VersionStorageInfo::LevelSummaryStorage tmp;
  ROCKS_LOG_BUFFER_MAX_SZ(
      log_buffer_, 3072,
-      "[%s] Universal: sorted runs files(%" ROCKSDB_PRIszt "): %s\n",
+      "[%s] Universal: sorted runs: %" ROCKSDB_PRIszt " files: %s\n",
      cf_name_.c_str(), sorted_runs_.size(), vstorage_->LevelSummary(&tmp));

  Compaction* c = nullptr;
@@ -779,7 +765,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSizeAmp() {
  }

  // Skip files that are already being compacted
-  for (size_t loop = 0; loop < sorted_runs_.size() - 1; loop++) {
+  for (size_t loop = 0; loop + 1 < sorted_runs_.size(); loop++) {
    sr = &sorted_runs_[loop];
    if (!sr->being_compacted) {
      start_index = loop;  // Consider this as the first candidate.
@@ -807,7 +793,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSizeAmp() {
  }

  // keep adding up all the remaining files
-  for (size_t loop = start_index; loop < sorted_runs_.size() - 1; loop++) {
+  for (size_t loop = start_index; loop + 1 < sorted_runs_.size(); loop++) {
    sr = &sorted_runs_[loop];
    if (sr->being_compacted) {
      char file_num_buf[kFormatFileNumberBufSize];
@@ -855,6 +841,7 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
  std::vector<CompactionInputFiles> inputs;

  if (vstorage_->num_levels() == 1) {
+#if defined(ENABLE_SINGLE_LEVEL_DTC)
    // This is single level universal. Since we're basically trying to reclaim
    // space by processing files marked for compaction due to high tombstone
    // density, let's do the same thing as compaction to reduce size amp which
@@ -877,6 +864,11 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
      return nullptr;
    }
    inputs.push_back(start_level_inputs);
+#else
+    // Disable due to a known race condition.
+    // TODO: Reenable once the race condition is fixed
+    return nullptr;
+#endif  // ENABLE_SINGLE_LEVEL_DTC
  } else {
    int start_level;

@@ -14,7 +14,7 @@
 namespace ROCKSDB_NAMESPACE {

 void CancelAllBackgroundWork(DB* db, bool wait) {
-  (static_cast_with_check<DBImpl, DB>(db->GetRootDB()))
+  (static_cast_with_check<DBImpl>(db->GetRootDB()))
      ->CancelAllBackgroundWork(wait);
 }

@@ -28,7 +28,7 @@ Status DeleteFilesInRange(DB* db, ColumnFamilyHandle* column_family,
 Status DeleteFilesInRanges(DB* db, ColumnFamilyHandle* column_family,
                           const RangePtr* ranges, size_t n,
                           bool include_end) {
-  return (static_cast_with_check<DBImpl, DB>(db->GetRootDB()))
+  return (static_cast_with_check<DBImpl>(db->GetRootDB()))
      ->DeleteFilesInRanges(column_family, ranges, n, include_end);
 }

@@ -61,7 +61,8 @@ Status VerifySstFileChecksum(const Options& options,
  s = ioptions.table_factory->NewTableReader(
      TableReaderOptions(ioptions, options.prefix_extractor.get(), env_options,
                         internal_comparator, false /* skip_filters */,
-                         !kImmortal, -1 /* level */),
+                         !kImmortal, false /* force_direct_prefetch */,
+                         -1 /* level */),
      std::move(file_reader), file_size, &table_reader,
      false /* prefetch_index_and_filter_in_cache */);
  if (!s.ok()) {
@@ -1295,14 +1295,18 @@ TEST_F(DBBasicTest, MultiGetBatchedSimpleUnsorted) {
  } while (ChangeCompactOptions());
 }

-TEST_F(DBBasicTest, MultiGetBatchedSimpleSorted) {
+TEST_F(DBBasicTest, MultiGetBatchedSortedMultiFile) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    SetPerfLevel(kEnableCount);
+    // To expand the power of this test, generate > 1 table file and
+    // mix with memtable
    ASSERT_OK(Put(1, "k1", "v1"));
    ASSERT_OK(Put(1, "k2", "v2"));
+    Flush(1);
    ASSERT_OK(Put(1, "k3", "v3"));
    ASSERT_OK(Put(1, "k4", "v4"));
+    Flush(1);
    ASSERT_OK(Delete(1, "k4"));
    ASSERT_OK(Put(1, "k5", "v5"));
    ASSERT_OK(Delete(1, "no_key"));
@@ -1333,7 +1337,7 @@ TEST_F(DBBasicTest, MultiGetBatchedSimpleSorted) {
    ASSERT_TRUE(s[5].IsNotFound());

    SetPerfLevel(kDisable);
-  } while (ChangeCompactOptions());
+  } while (ChangeOptions());
 }

 TEST_F(DBBasicTest, MultiGetBatchedMultiLevel) {
@@ -1497,6 +1501,221 @@ TEST_F(DBBasicTest, MultiGetBatchedMultiLevelMerge) {
  }
 }

+TEST_F(DBBasicTest, MultiGetBatchedValueSizeInMemory) {
+  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+  SetPerfLevel(kEnableCount);
+  ASSERT_OK(Put(1, "k1", "v_1"));
+  ASSERT_OK(Put(1, "k2", "v_2"));
+  ASSERT_OK(Put(1, "k3", "v_3"));
+  ASSERT_OK(Put(1, "k4", "v_4"));
+  ASSERT_OK(Put(1, "k5", "v_5"));
+  ASSERT_OK(Put(1, "k6", "v_6"));
+  std::vector<Slice> keys = {"k1", "k2", "k3", "k4", "k5", "k6"};
+  std::vector<PinnableSlice> values(keys.size());
+  std::vector<Status> s(keys.size());
+  std::vector<ColumnFamilyHandle*> cfs(keys.size(), handles_[1]);
+
+  get_perf_context()->Reset();
+  ReadOptions ro;
+  ro.value_size_soft_limit = 11;
+  db_->MultiGet(ro, handles_[1], keys.size(), keys.data(), values.data(),
+                s.data(), false);
+
+  ASSERT_EQ(values.size(), keys.size());
+  for (unsigned int i = 0; i < 4; i++) {
+    ASSERT_EQ(std::string(values[i].data(), values[i].size()),
+              "v_" + std::to_string(i + 1));
+  }
+
+  for (unsigned int i = 4; i < 6; i++) {
+    ASSERT_TRUE(s[i].IsAborted());
+  }
+
+  ASSERT_EQ(12, (int)get_perf_context()->multiget_read_bytes);
+  SetPerfLevel(kDisable);
+}
+
+TEST_F(DBBasicTest, MultiGetBatchedValueSize) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    SetPerfLevel(kEnableCount);
+
+    ASSERT_OK(Put(1, "k6", "v6"));
+    ASSERT_OK(Put(1, "k7", "v7_"));
+    ASSERT_OK(Put(1, "k3", "v3_"));
+    ASSERT_OK(Put(1, "k4", "v4"));
+    Flush(1);
+    ASSERT_OK(Delete(1, "k4"));
+    ASSERT_OK(Put(1, "k11", "v11"));
+    ASSERT_OK(Delete(1, "no_key"));
+    ASSERT_OK(Put(1, "k8", "v8_"));
+    ASSERT_OK(Put(1, "k13", "v13"));
+    ASSERT_OK(Put(1, "k14", "v14"));
+    ASSERT_OK(Put(1, "k15", "v15"));
+    ASSERT_OK(Put(1, "k16", "v16"));
+    ASSERT_OK(Put(1, "k17", "v17"));
+    Flush(1);
+
+    ASSERT_OK(Put(1, "k1", "v1_"));
+    ASSERT_OK(Put(1, "k2", "v2_"));
+    ASSERT_OK(Put(1, "k5", "v5_"));
+    ASSERT_OK(Put(1, "k9", "v9_"));
+    ASSERT_OK(Put(1, "k10", "v10"));
+    ASSERT_OK(Delete(1, "k2"));
+    ASSERT_OK(Delete(1, "k6"));
+
+    get_perf_context()->Reset();
+
+    std::vector<Slice> keys({"k1", "k10", "k11", "k12", "k13", "k14", "k15",
+                             "k16", "k17", "k2", "k3", "k4", "k5", "k6", "k7",
+                             "k8", "k9", "no_key"});
+    std::vector<PinnableSlice> values(keys.size());
+    std::vector<ColumnFamilyHandle*> cfs(keys.size(), handles_[1]);
+    std::vector<Status> s(keys.size());
+
+    ReadOptions ro;
+    ro.value_size_soft_limit = 20;
+    db_->MultiGet(ro, handles_[1], keys.size(), keys.data(), values.data(),
+                  s.data(), false);
+
+    ASSERT_EQ(values.size(), keys.size());
+
+    // In memory keys
+    ASSERT_EQ(std::string(values[0].data(), values[0].size()), "v1_");
+    ASSERT_EQ(std::string(values[1].data(), values[1].size()), "v10");
+    ASSERT_TRUE(s[9].IsNotFound());  // k2
+    ASSERT_EQ(std::string(values[12].data(), values[12].size()), "v5_");
+    ASSERT_TRUE(s[13].IsNotFound());  // k6
+    ASSERT_EQ(std::string(values[16].data(), values[16].size()), "v9_");
+
+    // In sst files
+    ASSERT_EQ(std::string(values[2].data(), values[1].size()), "v11");
+    ASSERT_EQ(std::string(values[4].data(), values[4].size()), "v13");
+    ASSERT_EQ(std::string(values[5].data(), values[5].size()), "v14");
+
+    // Remaining aborted after value_size exceeds.
+    ASSERT_TRUE(s[3].IsAborted());
+    ASSERT_TRUE(s[6].IsAborted());
+    ASSERT_TRUE(s[7].IsAborted());
+    ASSERT_TRUE(s[8].IsAborted());
+    ASSERT_TRUE(s[10].IsAborted());
+    ASSERT_TRUE(s[11].IsAborted());
+    ASSERT_TRUE(s[14].IsAborted());
+    ASSERT_TRUE(s[15].IsAborted());
+    ASSERT_TRUE(s[17].IsAborted());
+
+    // 6 kv pairs * 3 bytes per value (i.e. 18)
+    ASSERT_EQ(21, (int)get_perf_context()->multiget_read_bytes);
+    SetPerfLevel(kDisable);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBBasicTest, MultiGetBatchedValueSizeMultiLevelMerge) {
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  BlockBasedTableOptions bbto;
+  bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+  Reopen(options);
+  int num_keys = 0;
+
+  for (int i = 0; i < 64; ++i) {
+    ASSERT_OK(Put("key_" + std::to_string(i), "val_l2_" + std::to_string(i)));
+    num_keys++;
+    if (num_keys == 8) {
+      Flush();
+      num_keys = 0;
+    }
+  }
+  if (num_keys > 0) {
+    Flush();
+    num_keys = 0;
+  }
+  MoveFilesToLevel(2);
+
+  for (int i = 0; i < 64; i += 3) {
+    ASSERT_OK(Merge("key_" + std::to_string(i), "val_l1_" + std::to_string(i)));
+    num_keys++;
+    if (num_keys == 8) {
+      Flush();
+      num_keys = 0;
+    }
+  }
+  if (num_keys > 0) {
+    Flush();
+    num_keys = 0;
+  }
+  MoveFilesToLevel(1);
+
+  for (int i = 0; i < 64; i += 5) {
+    ASSERT_OK(Merge("key_" + std::to_string(i), "val_l0_" + std::to_string(i)));
+    num_keys++;
+    if (num_keys == 8) {
+      Flush();
+      num_keys = 0;
+    }
+  }
+  if (num_keys > 0) {
+    Flush();
+    num_keys = 0;
+  }
+  ASSERT_EQ(0, num_keys);
+
+  for (int i = 0; i < 64; i += 9) {
+    ASSERT_OK(
+        Merge("key_" + std::to_string(i), "val_mem_" + std::to_string(i)));
+  }
+
+  std::vector<std::string> keys_str;
+  for (int i = 10; i < 50; ++i) {
+    keys_str.push_back("key_" + std::to_string(i));
+  }
+
+  std::vector<Slice> keys(keys_str.size());
+  for (int i = 0; i < 40; i++) {
+    keys[i] = Slice(keys_str[i]);
+  }
+
+  std::vector<PinnableSlice> values(keys_str.size());
+  std::vector<Status> statuses(keys_str.size());
+  ReadOptions read_options;
+  read_options.verify_checksums = true;
+  read_options.value_size_soft_limit = 380;
+  db_->MultiGet(read_options, dbfull()->DefaultColumnFamily(), keys.size(),
+                keys.data(), values.data(), statuses.data());
+
+  ASSERT_EQ(values.size(), keys.size());
+
+  uint64_t curr_value_size = 0;
+  for (unsigned int j = 0; j < 26; ++j) {
+    int key = j + 10;
+    std::string value;
+    value.append("val_l2_" + std::to_string(key));
+    if (key % 3 == 0) {
+      value.append(",");
+      value.append("val_l1_" + std::to_string(key));
+    }
+    if (key % 5 == 0) {
+      value.append(",");
+      value.append("val_l0_" + std::to_string(key));
+    }
+    if (key % 9 == 0) {
+      value.append(",");
+      value.append("val_mem_" + std::to_string(key));
+    }
+    curr_value_size += value.size();
+    ASSERT_EQ(values[j], value);
+    ASSERT_OK(statuses[j]);
+  }
+  // ASSERT_TRUE(curr_value_size <= read_options.value_size_hard_limit);
+
+  // All remaning keys status is set Status::Abort
+  for (unsigned int j = 26; j < 40; j++) {
+    ASSERT_TRUE(statuses[j].IsAborted());
+  }
+}
+
 // Test class for batched MultiGet with prefix extractor
 // Param bool - If true, use partitioned filters
 //              If false, use full filter block
@@ -1681,13 +1900,13 @@ TEST_F(DBBasicTest, GetAllKeyVersions) {
  ASSERT_EQ(kNumInserts + kNumDeletes + kNumUpdates, key_versions.size());

  // Check non-default column family
-  for (size_t i = 0; i != kNumInserts - 1; ++i) {
+  for (size_t i = 0; i + 1 != kNumInserts; ++i) {
    ASSERT_OK(Put(1, std::to_string(i), "value"));
  }
-  for (size_t i = 0; i != kNumUpdates - 1; ++i) {
+  for (size_t i = 0; i + 1 != kNumUpdates; ++i) {
    ASSERT_OK(Put(1, std::to_string(i), "value1"));
  }
-  for (size_t i = 0; i != kNumDeletes - 1; ++i) {
+  for (size_t i = 0; i + 1 != kNumDeletes; ++i) {
    ASSERT_OK(Delete(1, std::to_string(i)));
  }
  ASSERT_OK(ROCKSDB_NAMESPACE::GetAllKeyVersions(
@@ -1774,6 +1993,28 @@ TEST_F(DBBasicTest, IncrementalRecoveryNoCorrupt) {
  }
 }

+TEST_F(DBBasicTest, BestEffortsRecoveryWithVersionBuildingFailure) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "value"));
+  ASSERT_OK(Flush());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
+        ASSERT_NE(nullptr, arg);
+        *(reinterpret_cast<Status*>(arg)) =
+            Status::Corruption("Inject corruption");
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  options.best_efforts_recovery = true;
+  Status s = TryReopen(options);
+  ASSERT_TRUE(s.IsCorruption());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
 #ifndef ROCKSDB_LITE
 namespace {
 class TableFileListener : public EventListener {
@@ -1814,11 +2055,16 @@ TEST_F(DBBasicTest, RecoverWithMissingFiles) {
    ASSERT_OK(Flush(static_cast<int>(cf)));
  }

-  // Delete files
+  // Delete and corrupt files
  for (size_t i = 0; i < all_cf_names.size(); ++i) {
    std::vector<std::string>& files = listener->GetFiles(all_cf_names[i]);
    ASSERT_EQ(3, files.size());
-    for (int j = static_cast<int>(files.size() - 1); j >= static_cast<int>(i);
+    std::string corrupted_data;
+    ASSERT_OK(ReadFileToString(env_, files[files.size() - 1], &corrupted_data));
+    ASSERT_OK(WriteStringToFile(
+        env_, corrupted_data.substr(0, corrupted_data.size() - 2),
+        files[files.size() - 1], /*should_sync=*/true));
+    for (int j = static_cast<int>(files.size() - 2); j >= static_cast<int>(i);
         --j) {
      ASSERT_OK(env_->DeleteFile(files[j]));
    }
@@ -1850,6 +2096,55 @@ TEST_F(DBBasicTest, RecoverWithMissingFiles) {
  }
 }

+TEST_F(DBBasicTest, BestEffortsRecoveryTryMultipleManifests) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "value0"));
+  ASSERT_OK(Flush());
+  Close();
+  {
+    // Hack by adding a new MANIFEST with high file number
+    std::string garbage(10, '\0');
+    ASSERT_OK(WriteStringToFile(env_, garbage, dbname_ + "/MANIFEST-001000",
+                                /*should_sync=*/true));
+  }
+  {
+    // Hack by adding a corrupted SST not referenced by any MANIFEST
+    std::string garbage(10, '\0');
+    ASSERT_OK(WriteStringToFile(env_, garbage, dbname_ + "/001001.sst",
+                                /*should_sync=*/true));
+  }
+
+  options.best_efforts_recovery = true;
+
+  Reopen(options);
+  ASSERT_OK(Put("bar", "value"));
+}
+
+TEST_F(DBBasicTest, RecoverWithNoCurrentFile) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+  options.best_efforts_recovery = true;
+  ReopenWithColumnFamilies({kDefaultColumnFamilyName, "pikachu"}, options);
+  ASSERT_EQ(2, handles_.size());
+  ASSERT_OK(Put("foo", "value"));
+  ASSERT_OK(Put(1, "bar", "value"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Flush(1));
+  Close();
+  ASSERT_OK(env_->DeleteFile(CurrentFileName(dbname_)));
+  ReopenWithColumnFamilies({kDefaultColumnFamilyName, "pikachu"}, options);
+  std::vector<std::string> cf_names;
+  ASSERT_OK(DB::ListColumnFamilies(DBOptions(options), dbname_, &cf_names));
+  ASSERT_EQ(2, cf_names.size());
+  for (const auto& name : cf_names) {
+    ASSERT_TRUE(name == kDefaultColumnFamilyName || name == "pikachu");
+  }
+}
+
 TEST_F(DBBasicTest, SkipWALIfMissingTableFiles) {
  Options options = CurrentOptions();
  DestroyAndReopen(options);
@@ -1890,15 +2185,14 @@ TEST_F(DBBasicTest, SkipWALIfMissingTableFiles) {
 }
 #endif  // !ROCKSDB_LITE

-class DBBasicTestWithParallelIO
-    : public DBTestBase,
-      public testing::WithParamInterface<std::tuple<bool, bool, bool, bool>> {
+class DBBasicTestMultiGet : public DBTestBase {
 public:
-  DBBasicTestWithParallelIO() : DBTestBase("/db_basic_test_with_parallel_io") {
-    bool compressed_cache = std::get<0>(GetParam());
-    bool uncompressed_cache = std::get<1>(GetParam());
-    compression_enabled_ = std::get<2>(GetParam());
-    fill_cache_ = std::get<3>(GetParam());
+  DBBasicTestMultiGet(std::string test_dir, int num_cfs, bool compressed_cache,
+                      bool uncompressed_cache, bool _compression_enabled,
+                      bool _fill_cache, uint32_t compression_parallel_threads)
+      : DBTestBase(test_dir) {
+    compression_enabled_ = _compression_enabled;
+    fill_cache_ = _fill_cache;

    if (compressed_cache) {
      std::shared_ptr<Cache> cache = NewLRUCache(1048576);
@@ -1953,25 +2247,48 @@ class DBBasicTestWithParallelIO
    options.table_factory.reset(new BlockBasedTableFactory(table_options));
    if (!compression_enabled_) {
      options.compression = kNoCompression;
+    } else {
+      options.compression_opts.parallel_threads = compression_parallel_threads;
    }
    Reopen(options);

-    std::string zero_str(128, '\0');
-    for (int i = 0; i < 100; ++i) {
-      // Make the value compressible. A purely random string doesn't compress
-      // and the resultant data block will not be compressed
-      values_.emplace_back(RandomString(&rnd, 128) + zero_str);
-      assert(Put(Key(i), values_[i]) == Status::OK());
+    if (num_cfs > 1) {
+      for (int cf = 0; cf < num_cfs; ++cf) {
+        cf_names_.emplace_back("cf" + std::to_string(cf));
+      }
+      CreateColumnFamilies(cf_names_, options);
+      cf_names_.emplace_back("default");
    }
-    Flush();

-    for (int i = 0; i < 100; ++i) {
-      // block cannot gain space by compression
-      uncompressable_values_.emplace_back(RandomString(&rnd, 256) + '\0');
-      std::string tmp_key = "a" + Key(i);
-      assert(Put(tmp_key, uncompressable_values_[i]) == Status::OK());
+    std::string zero_str(128, '\0');
+    for (int cf = 0; cf < num_cfs; ++cf) {
+      for (int i = 0; i < 100; ++i) {
+        // Make the value compressible. A purely random string doesn't compress
+        // and the resultant data block will not be compressed
+        values_.emplace_back(RandomString(&rnd, 128) + zero_str);
+        assert(((num_cfs == 1) ? Put(Key(i), values_[i])
+                               : Put(cf, Key(i), values_[i])) == Status::OK());
+      }
+      if (num_cfs == 1) {
+        Flush();
+      } else {
+        dbfull()->Flush(FlushOptions(), handles_[cf]);
+      }
+
+      for (int i = 0; i < 100; ++i) {
+        // block cannot gain space by compression
+        uncompressable_values_.emplace_back(RandomString(&rnd, 256) + '\0');
+        std::string tmp_key = "a" + Key(i);
+        assert(((num_cfs == 1) ? Put(tmp_key, uncompressable_values_[i])
+                               : Put(cf, tmp_key, uncompressable_values_[i])) ==
+               Status::OK());
+      }
+      if (num_cfs == 1) {
+        Flush();
+      } else {
+        dbfull()->Flush(FlushOptions(), handles_[cf]);
+      }
    }
-    Flush();
  }

  bool CheckValue(int i, const std::string& value) {
@@ -1988,6 +2305,8 @@ class DBBasicTestWithParallelIO
    return false;
  }

+  const std::vector<std::string>& GetCFNames() const { return cf_names_; }
+
  int num_lookups() { return uncompressed_cache_->num_lookups(); }
  int num_found() { return uncompressed_cache_->num_found(); }
  int num_inserts() { return uncompressed_cache_->num_inserts(); }
@@ -2004,7 +2323,7 @@ class DBBasicTestWithParallelIO
  static void SetUpTestCase() {}
  static void TearDownTestCase() {}

- private:
+ protected:
  class MyFlushBlockPolicyFactory : public FlushBlockPolicyFactory {
   public:
    MyFlushBlockPolicyFactory() {}
@@ -2045,23 +2364,25 @@ class DBBasicTestWithParallelIO
    const BlockBuilder& data_block_builder_;
  };

-  class MyBlockCache : public Cache {
+  class MyBlockCache : public CacheWrapper {
   public:
-    explicit MyBlockCache(std::shared_ptr<Cache>& target)
-        : target_(target), num_lookups_(0), num_found_(0), num_inserts_(0) {}
+    explicit MyBlockCache(std::shared_ptr<Cache> target)
+        : CacheWrapper(target),
+          num_lookups_(0),
+          num_found_(0),
+          num_inserts_(0) {}

-    virtual const char* Name() const override { return "MyBlockCache"; }
+    const char* Name() const override { return "MyBlockCache"; }

-    virtual Status Insert(const Slice& key, void* value, size_t charge,
-                          void (*deleter)(const Slice& key, void* value),
-                          Handle** handle = nullptr,
-                          Priority priority = Priority::LOW) override {
+    Status Insert(const Slice& key, void* value, size_t charge,
+                  void (*deleter)(const Slice& key, void* value),
+                  Handle** handle = nullptr,
+                  Priority priority = Priority::LOW) override {
      num_inserts_++;
      return target_->Insert(key, value, charge, deleter, handle, priority);
    }

-    virtual Handle* Lookup(const Slice& key,
-                           Statistics* stats = nullptr) override {
+    Handle* Lookup(const Slice& key, Statistics* stats = nullptr) override {
      num_lookups_++;
      Handle* handle = target_->Lookup(key, stats);
      if (handle != nullptr) {
@@ -2069,57 +2390,6 @@ class DBBasicTestWithParallelIO
      }
      return handle;
    }
-
-    virtual bool Ref(Handle* handle) override { return target_->Ref(handle); }
-
-    virtual bool Release(Handle* handle, bool force_erase = false) override {
-      return target_->Release(handle, force_erase);
-    }
-
-    virtual void* Value(Handle* handle) override {
-      return target_->Value(handle);
-    }
-
-    virtual void Erase(const Slice& key) override { target_->Erase(key); }
-    virtual uint64_t NewId() override { return target_->NewId(); }
-
-    virtual void SetCapacity(size_t capacity) override {
-      target_->SetCapacity(capacity);
-    }
-
-    virtual void SetStrictCapacityLimit(bool strict_capacity_limit) override {
-      target_->SetStrictCapacityLimit(strict_capacity_limit);
-    }
-
-    virtual bool HasStrictCapacityLimit() const override {
-      return target_->HasStrictCapacityLimit();
-    }
-
-    virtual size_t GetCapacity() const override {
-      return target_->GetCapacity();
-    }
-
-    virtual size_t GetUsage() const override { return target_->GetUsage(); }
-
-    virtual size_t GetUsage(Handle* handle) const override {
-      return target_->GetUsage(handle);
-    }
-
-    virtual size_t GetPinnedUsage() const override {
-      return target_->GetPinnedUsage();
-    }
-
-    virtual size_t GetCharge(Handle* /*handle*/) const override { return 0; }
-
-    virtual void ApplyToAllCacheEntries(void (*callback)(void*, size_t),
-                                        bool thread_safe) override {
-      return target_->ApplyToAllCacheEntries(callback, thread_safe);
-    }
-
-    virtual void EraseUnRefEntries() override {
-      return target_->EraseUnRefEntries();
-    }
-
    int num_lookups() { return num_lookups_; }

    int num_found() { return num_found_; }
@@ -2127,7 +2397,6 @@ class DBBasicTestWithParallelIO
    int num_inserts() { return num_inserts_; }

   private:
-    std::shared_ptr<Cache> target_;
    int num_lookups_;
    int num_found_;
    int num_inserts_;
@@ -2139,6 +2408,19 @@ class DBBasicTestWithParallelIO
  std::vector<std::string> values_;
  std::vector<std::string> uncompressable_values_;
  bool fill_cache_;
+  std::vector<std::string> cf_names_;
+};
+
+class DBBasicTestWithParallelIO
+    : public DBBasicTestMultiGet,
+      public testing::WithParamInterface<
+          std::tuple<bool, bool, bool, bool, uint32_t>> {
+ public:
+  DBBasicTestWithParallelIO()
+      : DBBasicTestMultiGet("/db_basic_test_with_parallel_io", 1,
+                            std::get<0>(GetParam()), std::get<1>(GetParam()),
+                            std::get<2>(GetParam()), std::get<3>(GetParam()),
+                            std::get<4>(GetParam())) {}
 };

 TEST_P(DBBasicTestWithParallelIO, MultiGet) {
@@ -2354,10 +2636,282 @@ INSTANTIATE_TEST_CASE_P(ParallelIO, DBBasicTestWithParallelIO,
                        // Param 1 - Uncompressed cache enabled
                        // Param 2 - Data compression enabled
                        // Param 3 - ReadOptions::fill_cache
+                        // Param 4 - CompressionOptions::parallel_threads
                        ::testing::Combine(::testing::Bool(), ::testing::Bool(),
-                                           ::testing::Bool(),
-                                           ::testing::Bool()));
+                                           ::testing::Bool(), ::testing::Bool(),
+                                           ::testing::Values(1, 4)));

+// A test class for intercepting random reads and injecting artificial
+// delays. Used for testing the deadline/timeout feature
+class DBBasicTestMultiGetDeadline : public DBBasicTestMultiGet {
+ public:
+  DBBasicTestMultiGetDeadline()
+      : DBBasicTestMultiGet("db_basic_test_multiget_deadline" /*Test dir*/,
+                             10 /*# of column families*/,
+                             false /*compressed cache enabled*/,
+                             true /*uncompressed cache enabled*/,
+                             true /*compression enabled*/,
+                             true /*ReadOptions.fill_cache*/,
+                             1 /*# of parallel compression threads*/) {}
+
+  // Forward declaration
+  class DeadlineFS;
+
+  class DeadlineRandomAccessFile : public FSRandomAccessFileWrapper {
+   public:
+    DeadlineRandomAccessFile(DeadlineFS& fs, SpecialEnv* env,
+                             std::unique_ptr<FSRandomAccessFile>& file)
+        : FSRandomAccessFileWrapper(file.get()),
+          fs_(fs),
+          file_(std::move(file)),
+          env_(env) {}
+
+    IOStatus Read(uint64_t offset, size_t len, const IOOptions& opts,
+          Slice* result, char* scratch, IODebugContext* dbg) const override {
+      int delay;
+      const std::chrono::microseconds deadline = fs_.GetDeadline();
+      if (deadline.count()) {
+        AssertDeadline(deadline, opts);
+      }
+      if (fs_.ShouldDelay(&delay)) {
+        env_->SleepForMicroseconds(delay);
+      }
+      return FSRandomAccessFileWrapper::Read(offset, len, opts, result, scratch,
+                                             dbg);
+    }
+
+    IOStatus MultiRead(FSReadRequest* reqs, size_t num_reqs,
+          const IOOptions& options, IODebugContext* dbg) override {
+      int delay;
+      const std::chrono::microseconds deadline = fs_.GetDeadline();
+      if (deadline.count()) {
+        AssertDeadline(deadline, options);
+      }
+      if (fs_.ShouldDelay(&delay)) {
+        env_->SleepForMicroseconds(delay);
+      }
+      return FSRandomAccessFileWrapper::MultiRead(reqs, num_reqs, options, dbg);
+    }
+
+   private:
+    void AssertDeadline(const std::chrono::microseconds deadline,
+                        const IOOptions& opts) const {
+      // Give a leeway of +- 10us as it can take some time for the Get/
+      // MultiGet call to reach here, in order to avoid false alarms
+      std::chrono::microseconds now =
+          std::chrono::microseconds(env_->NowMicros());
+      ASSERT_EQ(deadline - now, opts.timeout);
+    }
+    DeadlineFS& fs_;
+    std::unique_ptr<FSRandomAccessFile> file_;
+    SpecialEnv* env_;
+  };
+
+  class DeadlineFS : public FileSystemWrapper {
+   public:
+    DeadlineFS(SpecialEnv* env)
+        : FileSystemWrapper(FileSystem::Default()),
+          delay_idx_(0),
+          deadline_(std::chrono::microseconds::zero()),
+          env_(env) {}
+    ~DeadlineFS() = default;
+
+    IOStatus NewRandomAccessFile(const std::string& fname,
+                                 const FileOptions& opts,
+                                 std::unique_ptr<FSRandomAccessFile>* result,
+                                 IODebugContext* dbg) override {
+      std::unique_ptr<FSRandomAccessFile> file;
+      IOStatus s;
+
+      s = target()->NewRandomAccessFile(fname, opts, &file, dbg);
+      result->reset(new DeadlineRandomAccessFile(*this, env_, file));
+      return s;
+    }
+
+    // Set a vector of {IO counter, delay in microseconds} pairs that control
+    // when to inject a delay and duration of the delay
+    void SetDelaySequence(const std::chrono::microseconds deadline,
+                          const std::vector<std::pair<int, int>>&& seq) {
+      int total_delay = 0;
+      for (auto& seq_iter : seq) {
+        // Ensure no individual delay is > 500ms
+        ASSERT_LT(seq_iter.second, 500000);
+        total_delay += seq_iter.second;
+      }
+      // ASSERT total delay is < 1s. This is mainly to keep the test from
+      // timing out in CI test frameworks
+      ASSERT_LT(total_delay, 1000000);
+      delay_seq_ = seq;
+      delay_idx_ = 0;
+      io_count_ = 0;
+      deadline_ = deadline;
+    }
+
+    // Increment the IO counter and return a delay in microseconds
+    bool ShouldDelay(int* delay) {
+      if (delay_idx_ < delay_seq_.size() &&
+          delay_seq_[delay_idx_].first == io_count_++) {
+        *delay = delay_seq_[delay_idx_].second;
+        delay_idx_++;
+        return true;
+      }
+      return false;
+    }
+
+    const std::chrono::microseconds GetDeadline() { return deadline_; }
+
+   private:
+    std::vector<std::pair<int, int>> delay_seq_;
+    size_t delay_idx_;
+    int io_count_;
+    std::chrono::microseconds deadline_;
+    SpecialEnv* env_;
+  };
+
+  inline void CheckStatus(std::vector<Status>& statuses, size_t num_ok) {
+    for (size_t i = 0; i < statuses.size(); ++i) {
+      if (i < num_ok) {
+        EXPECT_OK(statuses[i]);
+      } else {
+        EXPECT_EQ(statuses[i], Status::TimedOut());
+      }
+    }
+  }
+};
+
+TEST_F(DBBasicTestMultiGetDeadline, MultiGetDeadlineExceeded) {
+  std::shared_ptr<DBBasicTestMultiGetDeadline::DeadlineFS> fs(
+      new DBBasicTestMultiGetDeadline::DeadlineFS(env_));
+  std::unique_ptr<Env> env(new CompositeEnvWrapper(env_, fs));
+  Options options = CurrentOptions();
+  env_->SetTimeElapseOnlySleep(&options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(1048576);
+  BlockBasedTableOptions table_options;
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  options.env = env.get();
+  ReopenWithColumnFamilies(GetCFNames(), options);
+
+  // Test the non-batched version of MultiGet with multiple column
+  // families
+  std::vector<std::string> key_str;
+  size_t i;
+  for (i = 0; i < 5; ++i) {
+    key_str.emplace_back(Key(static_cast<int>(i)));
+  }
+  std::vector<ColumnFamilyHandle*> cfs(key_str.size());
+  ;
+  std::vector<Slice> keys(key_str.size());
+  std::vector<std::string> values(key_str.size());
+  for (i = 0; i < key_str.size(); ++i) {
+    cfs[i] = handles_[i];
+    keys[i] = Slice(key_str[i].data(), key_str[i].size());
+  }
+
+  ReadOptions ro;
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  // Delay the first IO by 200ms
+  fs->SetDelaySequence(ro.deadline, {{0, 20000}});
+
+  std::vector<Status> statuses = dbfull()->MultiGet(ro, cfs, keys, &values);
+  // The first key is successful because we check after the lookup, but
+  // subsequent keys fail due to deadline exceeded
+  CheckStatus(statuses, 1);
+
+  // Clear the cache
+  cache->SetCapacity(0);
+  cache->SetCapacity(1048576);
+  // Test non-batched Multiget with multiple column families and
+  // introducing an IO delay in one of the middle CFs
+  key_str.clear();
+  for (i = 0; i < 10; ++i) {
+    key_str.emplace_back(Key(static_cast<int>(i)));
+  }
+  cfs.resize(key_str.size());
+  keys.resize(key_str.size());
+  values.resize(key_str.size());
+  for (i = 0; i < key_str.size(); ++i) {
+    // 2 keys per CF
+    cfs[i] = handles_[i / 2];
+    keys[i] = Slice(key_str[i].data(), key_str[i].size());
+  }
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  fs->SetDelaySequence(ro.deadline, {{1, 20000}});
+  statuses = dbfull()->MultiGet(ro, cfs, keys, &values);
+  CheckStatus(statuses, 3);
+
+  // Test batched MultiGet with an IO delay in the first data block read.
+  // Both keys in the first CF should succeed as they're in the same data
+  // block and would form one batch, and we check for deadline between
+  // batches.
+  std::vector<PinnableSlice> pin_values(keys.size());
+  cache->SetCapacity(0);
+  cache->SetCapacity(1048576);
+  statuses.clear();
+  statuses.resize(keys.size());
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  fs->SetDelaySequence(ro.deadline, {{0, 20000}});
+  dbfull()->MultiGet(ro, keys.size(), cfs.data(), keys.data(),
+                     pin_values.data(), statuses.data());
+  CheckStatus(statuses, 2);
+
+  // Similar to the previous one, but an IO delay in the third CF data block
+  // read
+  for (PinnableSlice& value : pin_values) {
+    value.Reset();
+  }
+  cache->SetCapacity(0);
+  cache->SetCapacity(1048576);
+  statuses.clear();
+  statuses.resize(keys.size());
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  fs->SetDelaySequence(ro.deadline, {{2, 20000}});
+  dbfull()->MultiGet(ro, keys.size(), cfs.data(), keys.data(),
+                     pin_values.data(), statuses.data());
+  CheckStatus(statuses, 6);
+
+  // Similar to the previous one, but an IO delay in the last but one CF
+  for (PinnableSlice& value : pin_values) {
+    value.Reset();
+  }
+  cache->SetCapacity(0);
+  cache->SetCapacity(1048576);
+  statuses.clear();
+  statuses.resize(keys.size());
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  fs->SetDelaySequence(ro.deadline, {{3, 20000}});
+  dbfull()->MultiGet(ro, keys.size(), cfs.data(), keys.data(),
+                     pin_values.data(), statuses.data());
+  CheckStatus(statuses, 8);
+
+  // Test batched MultiGet with single CF and lots of keys. Inject delay
+  // into the second batch of keys. As each batch is 32, the first 64 keys,
+  // i.e first two batches, should succeed and the rest should time out
+  for (PinnableSlice& value : pin_values) {
+    value.Reset();
+  }
+  cache->SetCapacity(0);
+  cache->SetCapacity(1048576);
+  key_str.clear();
+  for (i = 0; i < 100; ++i) {
+    key_str.emplace_back(Key(static_cast<int>(i)));
+  }
+  keys.resize(key_str.size());
+  pin_values.clear();
+  pin_values.resize(key_str.size());
+  for (i = 0; i < key_str.size(); ++i) {
+    keys[i] = Slice(key_str[i].data(), key_str[i].size());
+  }
+  statuses.clear();
+  statuses.resize(keys.size());
+  ro.deadline = std::chrono::microseconds{env->NowMicros() + 10000};
+  fs->SetDelaySequence(ro.deadline, {{1, 20000}});
+  dbfull()->MultiGet(ro, handles_[0], keys.size(), keys.data(),
+                     pin_values.data(), statuses.data());
+  CheckStatus(statuses, 64);
+  Close();
+}

 }  // namespace ROCKSDB_NAMESPACE

@@ -187,7 +187,7 @@ TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
  Iterator* iter = nullptr;

  // Load blocks into cache.
-  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+  for (size_t i = 0; i + 1 < kNumBlocks; i++) {
    iter = db_->NewIterator(read_options);
    iter->Seek(ToString(i));
    ASSERT_OK(iter->status());
@@ -209,12 +209,12 @@ TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
  iter = nullptr;

  // Release iterators and access cache again.
-  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+  for (size_t i = 0; i + 1 < kNumBlocks; i++) {
    iterators[i].reset();
    CheckCacheCounters(options, 0, 0, 0, 0);
  }
  ASSERT_EQ(0, cache->GetPinnedUsage());
-  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+  for (size_t i = 0; i + 1 < kNumBlocks; i++) {
    iter = db_->NewIterator(read_options);
    iter->Seek(ToString(i));
    ASSERT_OK(iter->status());
@@ -243,7 +243,7 @@ TEST_F(DBBlockCacheTest, TestWithCompressedBlockCache) {
  Iterator* iter = nullptr;

  // Load blocks into cache.
-  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+  for (size_t i = 0; i + 1 < kNumBlocks; i++) {
    iter = db_->NewIterator(read_options);
    iter->Seek(ToString(i));
    ASSERT_OK(iter->status());
@@ -517,6 +517,139 @@ TEST_F(DBBlockCacheTest, IndexAndFilterBlocksCachePriority) {
  }
 }

+namespace {
+
+// An LRUCache wrapper that can falsely report "not found" on Lookup.
+// This allows us to manipulate BlockBasedTableReader into thinking
+// another thread inserted the data in between Lookup and Insert,
+// while mostly preserving the LRUCache interface/behavior.
+class LookupLiarCache : public CacheWrapper {
+  int nth_lookup_not_found_ = 0;
+
+ public:
+  explicit LookupLiarCache(std::shared_ptr<Cache> target)
+      : CacheWrapper(std::move(target)) {}
+
+  Handle* Lookup(const Slice& key, Statistics* stats) override {
+    if (nth_lookup_not_found_ == 1) {
+      nth_lookup_not_found_ = 0;
+      return nullptr;
+    }
+    if (nth_lookup_not_found_ > 1) {
+      --nth_lookup_not_found_;
+    }
+    return CacheWrapper::Lookup(key, stats);
+  }
+
+  // 1 == next lookup, 2 == after next, etc.
+  void SetNthLookupNotFound(int n) { nth_lookup_not_found_ = n; }
+};
+
+}  // anonymous namespace
+
+TEST_F(DBBlockCacheTest, AddRedundantStats) {
+  const size_t capacity = size_t{1} << 25;
+  const int num_shard_bits = 0;  // 1 shard
+  int iterations_tested = 0;
+  for (std::shared_ptr<Cache> base_cache :
+       {NewLRUCache(capacity, num_shard_bits),
+        NewClockCache(capacity, num_shard_bits)}) {
+    if (!base_cache) {
+      // Skip clock cache when not supported
+      continue;
+    }
+    ++iterations_tested;
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
+
+    std::shared_ptr<LookupLiarCache> cache =
+        std::make_shared<LookupLiarCache>(base_cache);
+
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.block_cache = cache;
+    table_options.filter_policy.reset(NewBloomFilterPolicy(50));
+    options.table_factory.reset(new BlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    // Create a new table.
+    ASSERT_OK(Put("foo", "value"));
+    ASSERT_OK(Put("bar", "value"));
+    ASSERT_OK(Flush());
+    ASSERT_EQ(1, NumTableFilesAtLevel(0));
+
+    // Normal access filter+index+data.
+    ASSERT_EQ("value", Get("foo"));
+
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
+    // --------
+    ASSERT_EQ(3, TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
+    // --------
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
+
+    // Againt access filter+index+data, but force redundant load+insert on index
+    cache->SetNthLookupNotFound(2);
+    ASSERT_EQ("value", Get("bar"));
+
+    ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
+    // --------
+    ASSERT_EQ(4, TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
+    // --------
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
+
+    // Access just filter (with high probability), and force redundant
+    // load+insert
+    cache->SetNthLookupNotFound(1);
+    ASSERT_EQ("NOT_FOUND", Get("this key was not added"));
+
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
+    // --------
+    EXPECT_EQ(5, TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
+    EXPECT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
+    // --------
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
+
+    // Access just data, forcing redundant load+insert
+    ReadOptions read_options;
+    std::unique_ptr<Iterator> iter{db_->NewIterator(read_options)};
+    cache->SetNthLookupNotFound(1);
+    iter->SeekToFirst();
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key(), "bar");
+
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
+    EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
+    // --------
+    EXPECT_EQ(6, TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
+    EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
+    // --------
+    EXPECT_EQ(3, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
+  }
+  EXPECT_GE(iterations_tested, 1);
+}
+
 TEST_F(DBBlockCacheTest, ParanoidFileChecks) {
  Options options = CurrentOptions();
  options.create_if_missing = true;
@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include "db/db_test_util.h"
+#include "options/options_helper.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
 #include "table/block_based/filter_policy_internal.h"
@@ -1029,6 +1030,215 @@ TEST_F(DBBloomFilterTest, MemtablePrefixBloomOutOfDomain) {
  ASSERT_EQ(kKey, iter->key());
 }

+class DBBloomFilterTestVaryPrefixAndFormatVer
+    : public DBTestBase,
+      public testing::WithParamInterface<std::tuple<bool, uint32_t>> {
+ protected:
+  bool use_prefix_;
+  uint32_t format_version_;
+
+ public:
+  DBBloomFilterTestVaryPrefixAndFormatVer()
+      : DBTestBase("/db_bloom_filter_tests") {}
+
+  ~DBBloomFilterTestVaryPrefixAndFormatVer() override {}
+
+  void SetUp() override {
+    use_prefix_ = std::get<0>(GetParam());
+    format_version_ = std::get<1>(GetParam());
+  }
+
+  static std::string UKey(uint32_t i) { return Key(static_cast<int>(i)); }
+};
+
+TEST_P(DBBloomFilterTestVaryPrefixAndFormatVer, PartitionedMultiGet) {
+  Options options = CurrentOptions();
+  if (use_prefix_) {
+    // Entire key from UKey()
+    options.prefix_extractor.reset(NewCappedPrefixTransform(9));
+  }
+  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
+  BlockBasedTableOptions bbto;
+  bbto.filter_policy.reset(NewBloomFilterPolicy(20));
+  bbto.partition_filters = true;
+  bbto.index_type = BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
+  bbto.whole_key_filtering = !use_prefix_;
+  if (use_prefix_) {  // (not related to prefix, just alternating between)
+    // Make sure code appropriately deals with metadata block size setting
+    // that is "too small" (smaller than minimum size for filter builder)
+    bbto.metadata_block_size = 63;
+  } else {
+    // Make sure the test will work even on platforms with large minimum
+    // filter size, due to large cache line size.
+    // (Largest cache line size + 10+% overhead.)
+    bbto.metadata_block_size = 290;
+  }
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+  DestroyAndReopen(options);
+  ReadOptions ropts;
+
+  constexpr uint32_t N = 12000;
+  // Add N/2 evens
+  for (uint32_t i = 0; i < N; i += 2) {
+    ASSERT_OK(Put(UKey(i), UKey(i)));
+  }
+  ASSERT_OK(Flush());
+#ifndef ROCKSDB_LITE
+  ASSERT_EQ(TotalTableFiles(), 1);
+#endif
+
+  constexpr uint32_t Q = 29;
+  // MultiGet In
+  std::array<std::string, Q> keys;
+  std::array<Slice, Q> key_slices;
+  std::array<ColumnFamilyHandle*, Q> column_families;
+  // MultiGet Out
+  std::array<Status, Q> statuses;
+  std::array<PinnableSlice, Q> values;
+
+  TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_HIT);
+  TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
+  TestGetAndResetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL);
+  TestGetAndResetTickerCount(options, BLOOM_FILTER_USEFUL);
+  TestGetAndResetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED);
+  TestGetAndResetTickerCount(options, BLOOM_FILTER_FULL_POSITIVE);
+  TestGetAndResetTickerCount(options, BLOOM_FILTER_FULL_TRUE_POSITIVE);
+
+  // Check that initial clump of keys only loads one partition filter from
+  // block cache.
+  // And that spread out keys load many partition filters.
+  // In both cases, mix present vs. not present keys.
+  for (uint32_t stride : {uint32_t{1}, (N / Q) | 1}) {
+    for (uint32_t i = 0; i < Q; ++i) {
+      keys[i] = UKey(i * stride);
+      key_slices[i] = Slice(keys[i]);
+      column_families[i] = db_->DefaultColumnFamily();
+      statuses[i] = Status();
+      values[i] = PinnableSlice();
+    }
+
+    db_->MultiGet(ropts, Q, &column_families[0], &key_slices[0], &values[0],
+                  /*timestamps=*/nullptr, &statuses[0], true);
+
+    // Confirm correct status results
+    uint32_t number_not_found = 0;
+    for (uint32_t i = 0; i < Q; ++i) {
+      if ((i * stride % 2) == 0) {
+        ASSERT_OK(statuses[i]);
+      } else {
+        ASSERT_TRUE(statuses[i].IsNotFound());
+        ++number_not_found;
+      }
+    }
+
+    // Confirm correct Bloom stats (no FPs)
+    uint64_t filter_useful = TestGetAndResetTickerCount(
+        options,
+        use_prefix_ ? BLOOM_FILTER_PREFIX_USEFUL : BLOOM_FILTER_USEFUL);
+    uint64_t filter_checked =
+        TestGetAndResetTickerCount(options, use_prefix_
+                                                ? BLOOM_FILTER_PREFIX_CHECKED
+                                                : BLOOM_FILTER_FULL_POSITIVE) +
+        (use_prefix_ ? 0 : filter_useful);
+    EXPECT_EQ(filter_useful, number_not_found);
+    EXPECT_EQ(filter_checked, Q);
+    if (!use_prefix_) {
+      EXPECT_EQ(
+          TestGetAndResetTickerCount(options, BLOOM_FILTER_FULL_TRUE_POSITIVE),
+          Q - number_not_found);
+    }
+
+    // Confirm no duplicate loading same filter partition
+    uint64_t filter_accesses =
+        TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_HIT) +
+        TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
+    if (stride == 1) {
+      EXPECT_EQ(filter_accesses, 1);
+    } else {
+      // for large stride
+      EXPECT_GE(filter_accesses, Q / 2 + 1);
+    }
+  }
+
+  // Check that a clump of keys (present and not) works when spanning
+  // two partitions
+  int found_spanning = 0;
+  for (uint32_t start = 0; start < N / 2;) {
+    for (uint32_t i = 0; i < Q; ++i) {
+      keys[i] = UKey(start + i);
+      key_slices[i] = Slice(keys[i]);
+      column_families[i] = db_->DefaultColumnFamily();
+      statuses[i] = Status();
+      values[i] = PinnableSlice();
+    }
+
+    db_->MultiGet(ropts, Q, &column_families[0], &key_slices[0], &values[0],
+                  /*timestamps=*/nullptr, &statuses[0], true);
+
+    // Confirm correct status results
+    uint32_t number_not_found = 0;
+    for (uint32_t i = 0; i < Q; ++i) {
+      if (((start + i) % 2) == 0) {
+        ASSERT_OK(statuses[i]);
+      } else {
+        ASSERT_TRUE(statuses[i].IsNotFound());
+        ++number_not_found;
+      }
+    }
+
+    // Confirm correct Bloom stats (might see some FPs)
+    uint64_t filter_useful = TestGetAndResetTickerCount(
+        options,
+        use_prefix_ ? BLOOM_FILTER_PREFIX_USEFUL : BLOOM_FILTER_USEFUL);
+    uint64_t filter_checked =
+        TestGetAndResetTickerCount(options, use_prefix_
+                                                ? BLOOM_FILTER_PREFIX_CHECKED
+                                                : BLOOM_FILTER_FULL_POSITIVE) +
+        (use_prefix_ ? 0 : filter_useful);
+    EXPECT_GE(filter_useful, number_not_found - 2);  // possible FP
+    EXPECT_EQ(filter_checked, Q);
+    if (!use_prefix_) {
+      EXPECT_EQ(
+          TestGetAndResetTickerCount(options, BLOOM_FILTER_FULL_TRUE_POSITIVE),
+          Q - number_not_found);
+    }
+
+    // Confirm no duplicate loading of same filter partition
+    uint64_t filter_accesses =
+        TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_HIT) +
+        TestGetAndResetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
+    if (filter_accesses == 2) {
+      // Spanned across partitions.
+      ++found_spanning;
+      if (found_spanning >= 2) {
+        break;
+      } else {
+        // Ensure that at least once we have at least one present and
+        // one non-present key on both sides of partition boundary.
+        start += 2;
+      }
+    } else {
+      EXPECT_EQ(filter_accesses, 1);
+      // See explanation at "start += 2"
+      start += Q - 4;
+    }
+  }
+  EXPECT_TRUE(found_spanning >= 2);
+}
+
+INSTANTIATE_TEST_CASE_P(DBBloomFilterTestVaryPrefixAndFormatVer,
+                        DBBloomFilterTestVaryPrefixAndFormatVer,
+                        ::testing::Values(
+                            // (use_prefix, format_version)
+                            std::make_tuple(false, 2),
+                            std::make_tuple(false, 3),
+                            std::make_tuple(false, 4),
+                            std::make_tuple(false, 5),
+                            std::make_tuple(true, 2),
+                            std::make_tuple(true, 3),
+                            std::make_tuple(true, 4),
+                            std::make_tuple(true, 5)));
+
 #ifndef ROCKSDB_LITE
 namespace {
 namespace BFP2 {
@@ -1343,8 +1553,7 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
  for (int i = 0; i < numkeys; i += 2) {
    keys.push_back(i);
  }
-  std::random_shuffle(std::begin(keys), std::end(keys));
-
+  RandomShuffle(std::begin(keys), std::end(keys));
  int num_inserted = 0;
  for (int key : keys) {
    ASSERT_OK(Put(1, Key(key), "val"));
@@ -157,21 +157,21 @@ Options DeletionTriggerOptions(Options options) {
 bool HaveOverlappingKeyRanges(
    const Comparator* c,
    const SstFileMetaData& a, const SstFileMetaData& b) {
-  if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
-    if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
+  if (c->CompareWithoutTimestamp(a.smallestkey, b.smallestkey) >= 0) {
+    if (c->CompareWithoutTimestamp(a.smallestkey, b.largestkey) <= 0) {
      // b.smallestkey <= a.smallestkey <= b.largestkey
      return true;
    }
-  } else if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
+  } else if (c->CompareWithoutTimestamp(a.largestkey, b.smallestkey) >= 0) {
    // a.smallestkey < b.smallestkey <= a.largestkey
    return true;
  }
-  if (c->Compare(a.largestkey, b.largestkey) <= 0) {
-    if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
+  if (c->CompareWithoutTimestamp(a.largestkey, b.largestkey) <= 0) {
+    if (c->CompareWithoutTimestamp(a.largestkey, b.smallestkey) >= 0) {
      // b.smallestkey <= a.largestkey <= b.largestkey
      return true;
    }
-  } else if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
+  } else if (c->CompareWithoutTimestamp(a.smallestkey, b.largestkey) <= 0) {
    // a.smallestkey <= b.largestkey < a.largestkey
    return true;
  }
@@ -4762,7 +4762,7 @@ TEST_P(CompactionPriTest, Test) {
  for (int i = 0; i < kNKeys; i++) {
    keys[i] = i;
  }
-  std::random_shuffle(std::begin(keys), std::end(keys));
+  RandomShuffle(std::begin(keys), std::end(keys), rnd.Next());

  for (int i = 0; i < kNKeys; i++) {
    ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 102)));
@@ -5024,10 +5024,11 @@ TEST_P(DBCompactionTestWithParam, FixFileIngestionCompactionDeadlock) {

 TEST_F(DBCompactionTest, ConsistencyFailTest) {
  Options options = CurrentOptions();
+  options.force_consistency_checks = true;
  DestroyAndReopen(options);

  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-      "VersionBuilder::CheckConsistency", [&](void* arg) {
+      "VersionBuilder::CheckConsistency0", [&](void* arg) {
        auto p =
            reinterpret_cast<std::pair<FileMetaData**, FileMetaData**>*>(arg);
        // just swap the two FileMetaData so that we hit error
@@ -5046,6 +5047,48 @@ TEST_F(DBCompactionTest, ConsistencyFailTest) {

  ASSERT_NOK(Put("foo", "bar"));
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
+TEST_F(DBCompactionTest, ConsistencyFailTest2) {
+  Options options = CurrentOptions();
+  options.force_consistency_checks = true;
+  options.target_file_size_base = 1000;
+  options.level0_file_num_compaction_trigger = 2;
+  BlockBasedTableOptions bbto;
+  bbto.block_size = 400;  // small block size
+  options.table_factory.reset(new BlockBasedTableFactory(bbto));
+  DestroyAndReopen(options);
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistency1", [&](void* arg) {
+        auto p =
+            reinterpret_cast<std::pair<FileMetaData**, FileMetaData**>*>(arg);
+        // just swap the two FileMetaData so that we hit error
+        // in CheckConsistency funcion
+        FileMetaData* temp = *(p->first);
+        *(p->first) = *(p->second);
+        *(p->second) = temp;
+      });
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  Random rnd(301);
+  std::string value = RandomString(&rnd, 1000);
+
+  ASSERT_OK(Put("foo1", value));
+  ASSERT_OK(Put("z", ""));
+  Flush();
+  ASSERT_OK(Put("foo2", value));
+  ASSERT_OK(Put("z", ""));
+  Flush();
+
+  // This probably returns non-OK, but we rely on the next Put()
+  // to determine the DB is frozen.
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_NOK(Put("foo", "bar"));
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
 }

 void IngestOneKeyValue(DBImpl* db, const std::string& key,
@@ -50,7 +50,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase) {
      keys[i] = i;
    }
    if (ordered_insert == 0) {
-      std::random_shuffle(std::begin(keys), std::end(keys));
+      RandomShuffle(std::begin(keys), std::end(keys), rnd.Next());
    }
    for (int max_background_compactions = 1; max_background_compactions < 4;
         max_background_compactions += 2) {
@@ -90,6 +90,9 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
      mutex_.Unlock();
      status = AtomicFlushMemTables(cfds, FlushOptions(),
                                    FlushReason::kGetLiveFiles);
+      if (status.IsColumnFamilyDropped()) {
+        status = Status::OK();
+      }
      mutex_.Lock();
    } else {
      for (auto cfd : *versions_->GetColumnFamilySet()) {
@@ -103,8 +106,10 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
        TEST_SYNC_POINT("DBImpl::GetLiveFiles:2");
        mutex_.Lock();
        cfd->UnrefAndTryDelete();
-        if (!status.ok()) {
+        if (!status.ok() && !status.IsColumnFamilyDropped()) {
          break;
+        } else if (status.IsColumnFamilyDropped()) {
+          status = Status::OK();
        }
      }
    }
@@ -118,27 +123,33 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
    }
  }

-  // Make a set of all of the live *.sst files
-  std::vector<FileDescriptor> live;
+  // Make a set of all of the live table and blob files
+  std::vector<uint64_t> live_table_files;
+  std::vector<uint64_t> live_blob_files;
  for (auto cfd : *versions_->GetColumnFamilySet()) {
    if (cfd->IsDropped()) {
      continue;
    }
-    cfd->current()->AddLiveFiles(&live);
+    cfd->current()->AddLiveFiles(&live_table_files, &live_blob_files);
  }

  ret.clear();
-  ret.reserve(live.size() + 3);  // *.sst + CURRENT + MANIFEST + OPTIONS
+  ret.reserve(live_table_files.size() + live_blob_files.size() +
+              3);  // for CURRENT + MANIFEST + OPTIONS

  // create names of the live files. The names are not absolute
  // paths, instead they are relative to dbname_;
-  for (const auto& live_file : live) {
-    ret.push_back(MakeTableFileName("", live_file.GetNumber()));
+  for (const auto& table_file_number : live_table_files) {
+    ret.emplace_back(MakeTableFileName("", table_file_number));
  }

-  ret.push_back(CurrentFileName(""));
-  ret.push_back(DescriptorFileName("", versions_->manifest_file_number()));
-  ret.push_back(OptionsFileName("", versions_->options_file_number()));
+  for (const auto& blob_file_number : live_blob_files) {
+    ret.emplace_back(BlobFileName("", blob_file_number));
+  }
+
+  ret.emplace_back(CurrentFileName(""));
+  ret.emplace_back(DescriptorFileName("", versions_->manifest_file_number()));
+  ret.emplace_back(OptionsFileName("", versions_->options_file_number()));

  // find length of manifest file while holding the mutex lock
  *manifest_file_size = versions_->manifest_file_size();
@@ -499,13 +499,13 @@ TEST_P(DBAtomicFlushTest, AtomicFlushTriggeredByMemTableFull) {
  TEST_SYNC_POINT(
      "DBAtomicFlushTest::AtomicFlushTriggeredByMemTableFull:BeforeCheck");
  if (options.atomic_flush) {
-    for (size_t i = 0; i != num_cfs - 1; ++i) {
+    for (size_t i = 0; i + 1 != num_cfs; ++i) {
      auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
      ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
      ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
    }
  } else {
-    for (size_t i = 0; i != num_cfs - 1; ++i) {
+    for (size_t i = 0; i + 1 != num_cfs; ++i) {
      auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
      ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
      ASSERT_FALSE(cfh->cfd()->mem()->IsEmpty());
@@ -870,9 +870,7 @@ void DBImpl::DumpStats() {
 Status DBImpl::TablesRangeTombstoneSummary(ColumnFamilyHandle* column_family,
                                           int max_entries_to_print,
                                           std::string* out_str) {
-  auto* cfh =
-      static_cast_with_check<ColumnFamilyHandleImpl, ColumnFamilyHandle>(
-          column_family);
+  auto* cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
  ColumnFamilyData* cfd = cfh->cfd();

  SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
@@ -1010,12 +1008,12 @@ Status DBImpl::SetDBOptions(
    }
    if (s.ok()) {
      const BGJobLimits current_bg_job_limits =
-          GetBGJobLimits(immutable_db_options_.max_background_flushes,
+          GetBGJobLimits(mutable_db_options_.max_background_flushes,
                         mutable_db_options_.max_background_compactions,
                         mutable_db_options_.max_background_jobs,
                         /* parallelize_compactions */ true);
      const BGJobLimits new_bg_job_limits = GetBGJobLimits(
-          immutable_db_options_.max_background_flushes,
+          new_options.max_background_flushes,
          new_options.max_background_compactions,
          new_options.max_background_jobs, /* parallelize_compactions */ true);

@@ -1313,7 +1311,7 @@ bool DBImpl::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) {

 InternalIterator* DBImpl::NewInternalIterator(
    Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
-    ColumnFamilyHandle* column_family) {
+    ColumnFamilyHandle* column_family, bool allow_unprepared_value) {
  ColumnFamilyData* cfd;
  if (column_family == nullptr) {
    cfd = default_cf_handle_->cfd();
@@ -1327,7 +1325,7 @@ InternalIterator* DBImpl::NewInternalIterator(
  mutex_.Unlock();
  ReadOptions roptions;
  return NewInternalIterator(roptions, cfd, super_version, arena, range_del_agg,
-                             sequence);
+                             sequence, allow_unprepared_value);
 }

 void DBImpl::SchedulePurge() {
@@ -1450,7 +1448,8 @@ InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
                                              SuperVersion* super_version,
                                              Arena* arena,
                                              RangeDelAggregator* range_del_agg,
-                                              SequenceNumber sequence) {
+                                              SequenceNumber sequence,
+                                              bool allow_unprepared_value) {
  InternalIterator* internal_iter;
  assert(arena != nullptr);
  assert(range_del_agg != nullptr);
@@ -1482,7 +1481,8 @@ InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
    // Collect iterators for files in L0 - Ln
    if (read_options.read_tier != kMemtableTier) {
      super_version->current->AddIterators(read_options, file_options_,
-                                           &merge_iter_builder, range_del_agg);
+                                           &merge_iter_builder, range_del_agg,
+                                           allow_unprepared_value);
    }
    internal_iter = merge_iter_builder.Finish();
    IterState* cleanup =
@@ -1527,6 +1527,20 @@ Status DBImpl::GetImpl(const ReadOptions& read_options, const Slice& key,
                       GetImplOptions& get_impl_options) {
  assert(get_impl_options.value != nullptr ||
         get_impl_options.merge_operands != nullptr);
+
+#ifndef NDEBUG
+  assert(get_impl_options.column_family);
+  ColumnFamilyHandle* cf = get_impl_options.column_family;
+  const Comparator* const ucmp = cf->GetComparator();
+  assert(ucmp);
+  if (ucmp->timestamp_size() > 0) {
+    assert(read_options.timestamp);
+    assert(read_options.timestamp->size() == ucmp->timestamp_size());
+  } else {
+    assert(!read_options.timestamp);
+  }
+#endif  // NDEBUG
+
  PERF_CPU_TIMER_GUARD(get_cpu_nanos, env_);
  StopWatch sw(env_, stats_, DB_GET);
  PERF_TIMER_GUARD(get_snapshot_time);
@@ -1702,7 +1716,7 @@ std::vector<Status> DBImpl::MultiGet(
    const std::vector<ColumnFamilyHandle*>& column_family,
    const std::vector<Slice>& keys, std::vector<std::string>* values) {
  return MultiGet(read_options, column_family, keys, values,
-                  /*timestamps*/ nullptr);
+                  /*timestamps=*/nullptr);
 }

 std::vector<Status> DBImpl::MultiGet(
@@ -1714,6 +1728,20 @@ std::vector<Status> DBImpl::MultiGet(
  StopWatch sw(env_, stats_, DB_MULTIGET);
  PERF_TIMER_GUARD(get_snapshot_time);

+#ifndef NDEBUG
+  for (const auto* cfh : column_family) {
+    assert(cfh);
+    const Comparator* const ucmp = cfh->GetComparator();
+    assert(ucmp);
+    if (ucmp->timestamp_size() > 0) {
+      assert(read_options.timestamp);
+      assert(ucmp->timestamp_size() == read_options.timestamp->size());
+    } else {
+      assert(!read_options.timestamp);
+    }
+  }
+#endif  // NDEBUG
+
  SequenceNumber consistent_seqnum;

  std::unordered_map<uint32_t, MultiGetColumnFamilyData> multiget_cf_data(
@@ -1758,14 +1786,17 @@ std::vector<Status> DBImpl::MultiGet(
  // s is both in/out. When in, s could either be OK or MergeInProgress.
  // merge_operands will contain the sequence of merges in the latter case.
  size_t num_found = 0;
-  for (size_t i = 0; i < num_keys; ++i) {
+  size_t keys_read;
+  uint64_t curr_value_size = 0;
+  for (keys_read = 0; keys_read < num_keys; ++keys_read) {
    merge_context.Clear();
-    Status& s = stat_list[i];
-    std::string* value = &(*values)[i];
-    std::string* timestamp = timestamps ? &(*timestamps)[i] : nullptr;
+    Status& s = stat_list[keys_read];
+    std::string* value = &(*values)[keys_read];
+    std::string* timestamp = timestamps ? &(*timestamps)[keys_read] : nullptr;

-    LookupKey lkey(keys[i], consistent_seqnum, read_options.timestamp);
-    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family[i]);
+    LookupKey lkey(keys[keys_read], consistent_seqnum, read_options.timestamp);
+    auto cfh =
+        static_cast_with_check<ColumnFamilyHandleImpl>(column_family[keys_read]);
    SequenceNumber max_covering_tombstone_seq = 0;
    auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
    assert(mgd_iter != multiget_cf_data.end());
@@ -1800,6 +1831,29 @@ std::vector<Status> DBImpl::MultiGet(
    if (s.ok()) {
      bytes_read += value->size();
      num_found++;
+      curr_value_size += value->size();
+      if (curr_value_size > read_options.value_size_soft_limit) {
+        while (++keys_read < num_keys) {
+          stat_list[keys_read] = Status::Aborted();
+        }
+        break;
+      }
+    }
+
+    if (read_options.deadline.count() &&
+        env_->NowMicros() >
+            static_cast<uint64_t>(read_options.deadline.count())) {
+      break;
+    }
+  }
+
+  if (keys_read < num_keys) {
+    // The only reason to break out of the loop is when the deadline is
+    // exceeded
+    assert(env_->NowMicros() >
+        static_cast<uint64_t>(read_options.deadline.count()));
+    for (++keys_read; keys_read < num_keys; ++keys_read) {
+      stat_list[keys_read] = Status::TimedOut();
    }
  }

@@ -1955,7 +2009,7 @@ void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
                      PinnableSlice* values, Status* statuses,
                      const bool sorted_input) {
  return MultiGet(read_options, num_keys, column_families, keys, values,
-                  /*timestamps*/ nullptr, statuses, sorted_input);
+                  /*timestamps=*/nullptr, statuses, sorted_input);
 }

 void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
@@ -1965,12 +2019,29 @@ void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
  if (num_keys == 0) {
    return;
  }
+
+#ifndef NDEBUG
+  for (size_t i = 0; i < num_keys; ++i) {
+    ColumnFamilyHandle* cfh = column_families[i];
+    assert(cfh);
+    const Comparator* const ucmp = cfh->GetComparator();
+    assert(ucmp);
+    if (ucmp->timestamp_size() > 0) {
+      assert(read_options.timestamp);
+      assert(read_options.timestamp->size() == ucmp->timestamp_size());
+    } else {
+      assert(!read_options.timestamp);
+    }
+  }
+#endif  // NDEBUG
+
  autovector<KeyContext, MultiGetContext::MAX_BATCH_SIZE> key_context;
  autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE> sorted_keys;
  sorted_keys.resize(num_keys);
  for (size_t i = 0; i < num_keys; ++i) {
    key_context.emplace_back(column_families[i], keys[i], &values[i],
-                             &timestamps[i], &statuses[i]);
+                             timestamps ? &timestamps[i] : nullptr,
+                             &statuses[i]);
  }
  for (size_t i = 0; i < num_keys; ++i) {
    sorted_keys[i] = &key_context[i];
@@ -2010,14 +2081,31 @@ void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
      read_options, nullptr, iter_deref_lambda, &multiget_cf_data,
      &consistent_seqnum);

-  for (auto cf_iter = multiget_cf_data.begin();
-       cf_iter != multiget_cf_data.end(); ++cf_iter) {
-    MultiGetImpl(read_options, cf_iter->start, cf_iter->num_keys, &sorted_keys,
-                 cf_iter->super_version, consistent_seqnum, nullptr, nullptr);
+  Status s;
+  auto cf_iter = multiget_cf_data.begin();
+  for (; cf_iter != multiget_cf_data.end(); ++cf_iter) {
+    s = MultiGetImpl(read_options, cf_iter->start, cf_iter->num_keys,
+                     &sorted_keys, cf_iter->super_version, consistent_seqnum,
+                     nullptr, nullptr);
+    if (!s.ok()) {
+      break;
+    }
+  }
+  if (!s.ok()) {
+    assert(s.IsTimedOut() || s.IsAborted());
+    for (++cf_iter; cf_iter != multiget_cf_data.end(); ++cf_iter) {
+      for (size_t i = cf_iter->start; i < cf_iter->start + cf_iter->num_keys;
+           ++i) {
+        *sorted_keys[i]->s = s;
+      }
+    }
+  }
+
+  for (const auto& iter : multiget_cf_data) {
    if (!unref_only) {
-      ReturnAndCleanupSuperVersion(cf_iter->cfd, cf_iter->super_version);
+      ReturnAndCleanupSuperVersion(iter.cfd, iter.super_version);
    } else {
-      cf_iter->cfd->GetSuperVersion()->Unref();
+      iter.cfd->GetSuperVersion()->Unref();
    }
  }
 }
@@ -2040,7 +2128,8 @@ struct CompareKeyContext {
    }

    // Both keys are from the same column family
-    int cmp = comparator->Compare(*(lhs->key), *(rhs->key));
+    int cmp = comparator->CompareWithoutTimestamp(
+        *(lhs->key), /*a_has_ts=*/false, *(rhs->key), /*b_has_ts=*/false);
    if (cmp < 0) {
      return true;
    }
@@ -2072,7 +2161,8 @@ void DBImpl::PrepareMultiGetKeys(
        }

        // Both keys are from the same column family
-        int cmp = comparator->Compare(*(lhs->key), *(rhs->key));
+        int cmp = comparator->CompareWithoutTimestamp(
+            *(lhs->key), /*a_has_ts=*/false, *(rhs->key), /*b_has_ts=*/false);
        assert(cmp <= 0);
      }
      index++;
@@ -2158,14 +2248,24 @@ void DBImpl::MultiGetWithCallback(
    consistent_seqnum = callback->max_visible_seq();
  }

-  MultiGetImpl(read_options, 0, num_keys, sorted_keys,
-               multiget_cf_data[0].super_version, consistent_seqnum, nullptr,
-               nullptr);
+  Status s = MultiGetImpl(read_options, 0, num_keys, sorted_keys,
+                          multiget_cf_data[0].super_version, consistent_seqnum,
+                          nullptr, nullptr);
+  assert(s.ok() || s.IsTimedOut() || s.IsAborted());
  ReturnAndCleanupSuperVersion(multiget_cf_data[0].cfd,
                               multiget_cf_data[0].super_version);
 }

-void DBImpl::MultiGetImpl(
+// The actual implementation of batched MultiGet. Parameters -
+// start_key - Index in the sorted_keys vector to start processing from
+// num_keys - Number of keys to lookup, starting with sorted_keys[start_key]
+// sorted_keys - The entire batch of sorted keys for this CF
+//
+// The per key status is returned in the KeyContext structures pointed to by
+// sorted_keys. An overall Status is also returned, with the only possible
+// values being Status::OK() and Status::TimedOut(). The latter indicates
+// that the call exceeded read_options.deadline
+Status DBImpl::MultiGetImpl(
    const ReadOptions& read_options, size_t start_key, size_t num_keys,
    autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys,
    SuperVersion* super_version, SequenceNumber snapshot,
@@ -2178,13 +2278,23 @@ void DBImpl::MultiGetImpl(
  // s is both in/out. When in, s could either be OK or MergeInProgress.
  // merge_operands will contain the sequence of merges in the latter case.
  size_t keys_left = num_keys;
+  Status s;
+  uint64_t curr_value_size = 0;
  while (keys_left) {
+    if (read_options.deadline.count() &&
+        env_->NowMicros() >
+            static_cast<uint64_t>(read_options.deadline.count())) {
+      s = Status::TimedOut();
+      break;
+    }
+
    size_t batch_size = (keys_left > MultiGetContext::MAX_BATCH_SIZE)
                            ? MultiGetContext::MAX_BATCH_SIZE
                            : keys_left;
    MultiGetContext ctx(sorted_keys, start_key + num_keys - keys_left,
-                        batch_size, snapshot);
+                        batch_size, snapshot, read_options);
    MultiGetRange range = ctx.GetMultiGetRange();
+    range.AddValueSize(curr_value_size);
    bool lookup_current = false;

    keys_left -= batch_size;
@@ -2215,19 +2325,32 @@ void DBImpl::MultiGetImpl(
      super_version->current->MultiGet(read_options, &range, callback,
                                       is_blob_index);
    }
+    curr_value_size = range.GetValueSize();
+    if (curr_value_size > read_options.value_size_soft_limit) {
+      s = Status::Aborted();
+      break;
+    }
  }

  // Post processing (decrement reference counts and record statistics)
  PERF_TIMER_GUARD(get_post_process_time);
  size_t num_found = 0;
  uint64_t bytes_read = 0;
-  for (size_t i = start_key; i < start_key + num_keys; ++i) {
+  for (size_t i = start_key; i < start_key + num_keys - keys_left; ++i) {
    KeyContext* key = (*sorted_keys)[i];
    if (key->s->ok()) {
      bytes_read += key->value->size();
      num_found++;
    }
  }
+  if (keys_left) {
+    assert(s.IsTimedOut() || s.IsAborted());
+    for (size_t i = start_key + num_keys - keys_left; i < start_key + num_keys;
+         ++i) {
+      KeyContext* key = (*sorted_keys)[i];
+      *key->s = s;
+    }
+  }

  RecordTick(stats_, NUMBER_MULTIGET_CALLS);
  RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
@@ -2236,6 +2359,8 @@ void DBImpl::MultiGetImpl(
  RecordInHistogram(stats_, BYTES_PER_MULTIGET, bytes_read);
  PERF_COUNTER_ADD(multiget_read_bytes, bytes_read);
  PERF_TIMER_STOP(get_post_process_time);
+
+  return s;
 }

 Status DBImpl::CreateColumnFamily(const ColumnFamilyOptions& cf_options,
@@ -2524,6 +2649,12 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
    return NewErrorIterator(
        Status::NotSupported("Managed iterator is not supported anymore."));
  }
+  // We will eventually support deadline for iterators too, but safeguard
+  // for now
+  if (read_options.deadline != std::chrono::microseconds::zero()) {
+    return NewErrorIterator(
+        Status::NotSupported("ReadOptions deadline is not supported"));
+  }
  Iterator* result = nullptr;
  if (read_options.read_tier == kPersistedTier) {
    return NewErrorIterator(Status::NotSupported(
@@ -2548,7 +2679,8 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,

 #else
    SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
-    auto iter = new ForwardIterator(this, read_options, cfd, sv);
+    auto iter = new ForwardIterator(this, read_options, cfd, sv,
+                                    /* allow_unprepared_value */ true);
    result = NewDBIterator(
        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
        cfd->user_comparator(), iter, kMaxSequenceNumber,
@@ -2625,7 +2757,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,

  InternalIterator* internal_iter =
      NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
-                          db_iter->GetRangeDelAggregator(), snapshot);
+                          db_iter->GetRangeDelAggregator(), snapshot,
+                          /* allow_unprepared_value */ true);
  db_iter->SetIterUnderDBIter(internal_iter);

  return db_iter;
@@ -2653,7 +2786,8 @@ Status DBImpl::NewIterators(
    for (auto cfh : column_families) {
      auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
      SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
-      auto iter = new ForwardIterator(this, read_options, cfd, sv);
+      auto iter = new ForwardIterator(this, read_options, cfd, sv,
+                                      /* allow_unprepared_value */ true);
      iterators->push_back(NewDBIterator(
          env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
          cfd->user_comparator(), iter, kMaxSequenceNumber,
@@ -4119,6 +4253,10 @@ Status DBImpl::IngestExternalFiles(
            "cannot ingest an external file into a dropped CF");
        break;
      }
+      if (args[i].options.skip_memtable_flush) {
+          need_flush[i] = false;
+          continue;
+      }
      bool tmp = false;
      status = ingestion_jobs[i].NeedsFlush(&tmp, cfd->GetSuperVersion());
      need_flush[i] = tmp;
@@ -581,9 +581,14 @@ class DBImpl : public DB {
  // Return an internal iterator over the current state of the database.
  // The keys of this iterator are internal keys (see format.h).
  // The returned iterator should be deleted when no longer needed.
+  // If allow_unprepared_value is true, the returned iterator may defer reading
+  // the value and so will require PrepareValue() to be called before value();
+  // allow_unprepared_value = false is convenient when this optimization is not
+  // useful, e.g. when reading the whole column family.
  InternalIterator* NewInternalIterator(
      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
-      ColumnFamilyHandle* column_family = nullptr);
+      ColumnFamilyHandle* column_family = nullptr,
+      bool allow_unprepared_value = false);

  LogsWithPrepTracker* logs_with_prep_tracker() {
    return &logs_with_prep_tracker_;
@@ -709,7 +714,8 @@ class DBImpl : public DB {

  InternalIterator* NewInternalIterator(
      const ReadOptions&, ColumnFamilyData* cfd, SuperVersion* super_version,
-      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence);
+      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
+      bool allow_unprepared_value);

  // hollow transactions shell used for recovery.
  // these will then be passed to TransactionDB so that
@@ -963,6 +969,12 @@ class DBImpl : public DB {
  void TEST_WaitForPersistStatsRun(std::function<void()> callback) const;
  bool TEST_IsPersistentStatsEnabled() const;
  size_t TEST_EstimateInMemoryStatsHistorySize() const;
+
+  VersionSet* TEST_GetVersionSet() const { return versions_.get(); }
+
+  const std::unordered_set<uint64_t>& TEST_GetFilesGrabbedForPurge() const {
+    return files_grabbed_for_purge_;
+  }
 #endif  // NDEBUG

 protected:
@@ -1143,13 +1155,11 @@ class DBImpl : public DB {
  // REQUIRES: db mutex held when calling this function, but the db mutex can
  // be released and re-acquired. Db mutex will be held when the function
  // returns.
-  // Currently, this function should be called only in best-efforts recovery
-  // mode.
  // After best-efforts recovery, there may be SST files in db/cf paths that are
  // not referenced in the MANIFEST. We delete these SST files. In the
  // meantime, we find out the largest file number present in the paths, and
  // bump up the version set's next_file_number_ to be 1 + largest_file_number.
-  Status CleanupFilesAfterRecovery();
+  Status FinishBestEffortsRecovery();

 private:
  friend class DB;
@@ -1763,7 +1773,7 @@ class DBImpl : public DB {
  // to have acquired the SuperVersion and pass in a snapshot sequence number
  // in order to construct the LookupKeys. The start_key and num_keys specify
  // the range of keys in the sorted_keys vector for a single column family.
-  void MultiGetImpl(
+  Status MultiGetImpl(
      const ReadOptions& read_options, size_t start_key, size_t num_keys,
      autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys,
      SuperVersion* sv, SequenceNumber snap_seqnum, ReadCallback* callback,
@@ -1931,7 +1941,7 @@ class DBImpl : public DB {
  std::unordered_map<uint64_t, PurgeFileInfo> purge_files_;

  // A vector to store the file numbers that have been assigned to certain
-  // JobContext. Current implementation tracks ssts only.
+  // JobContext. Current implementation tracks table and blob files only.
  std::unordered_set<uint64_t> files_grabbed_for_purge_;

  // A queue to store log writers to close
@@ -573,7 +573,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
  // Need to undo atomic flush if something went wrong, i.e. s is not OK and
  // it is not because of CF drop.
  if (!s.ok() && !s.IsColumnFamilyDropped()) {
-    if (!io_s.ok() && io_s.IsColumnFamilyDropped()) {
+    if (!io_s.ok() && !io_s.IsColumnFamilyDropped()) {
      error_handler_.SetBGError(io_s, BackgroundErrorReason::kFlush);
    } else {
      Status new_bg_error = s;
@@ -1906,7 +1906,7 @@ Status DBImpl::WaitForFlushMemTables(
      }
    }
    if (1 == num_dropped && 1 == num) {
-      return Status::InvalidArgument("Cannot flush a dropped CF");
+      return Status::ColumnFamilyDropped();
    }
    // Column families involved in this flush request have either been dropped
    // or finished flush. Then it's time to finish waiting.
@@ -2030,7 +2030,7 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {

 DBImpl::BGJobLimits DBImpl::GetBGJobLimits() const {
  mutex_.AssertHeld();
-  return GetBGJobLimits(immutable_db_options_.max_background_flushes,
+  return GetBGJobLimits(mutable_db_options_.max_background_flushes,
                        mutable_db_options_.max_background_compactions,
                        mutable_db_options_.max_background_jobs,
                        write_controller_.NeedSpeedupCompaction());
@@ -2143,8 +2143,7 @@ void DBImpl::BGWorkFlush(void* arg) {

  IOSTATS_SET_THREAD_POOL_ID(fta.thread_pri_);
  TEST_SYNC_POINT("DBImpl::BGWorkFlush");
-  static_cast_with_check<DBImpl, DB>(fta.db_)->BackgroundCallFlush(
-      fta.thread_pri_);
+  static_cast_with_check<DBImpl>(fta.db_)->BackgroundCallFlush(fta.thread_pri_);
  TEST_SYNC_POINT("DBImpl::BGWorkFlush:done");
 }

@@ -2155,7 +2154,7 @@ void DBImpl::BGWorkCompaction(void* arg) {
  TEST_SYNC_POINT("DBImpl::BGWorkCompaction");
  auto prepicked_compaction =
      static_cast<PrepickedCompaction*>(ca.prepicked_compaction);
-  static_cast_with_check<DBImpl, DB>(ca.db)->BackgroundCallCompaction(
+  static_cast_with_check<DBImpl>(ca.db)->BackgroundCallCompaction(
      prepicked_compaction, Env::Priority::LOW);
  delete prepicked_compaction;
 }
@@ -2549,8 +2548,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                       "[%s] Manual compaction from level-%d from %s .. "
                       "%s; nothing to do\n",
                       m->cfd->GetName().c_str(), m->input_level,
-                       (m->begin ? m->begin->DebugString().c_str() : "(begin)"),
-                       (m->end ? m->end->DebugString().c_str() : "(end)"));
+                       (m->begin ? m->begin->DebugString(true).c_str() : "(begin)"),
+                       (m->end ? m->end->DebugString(true).c_str() : "(end)"));
    } else {
      // First check if we have enough room to do the compaction
      bool enough_room = EnoughRoomForCompaction(
@@ -2569,11 +2568,11 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
            "[%s] Manual compaction from level-%d to level-%d from %s .. "
            "%s; will stop at %s\n",
            m->cfd->GetName().c_str(), m->input_level, c->output_level(),
-            (m->begin ? m->begin->DebugString().c_str() : "(begin)"),
-            (m->end ? m->end->DebugString().c_str() : "(end)"),
+            (m->begin ? m->begin->DebugString(true).c_str() : "(begin)"),
+            (m->end ? m->end->DebugString(true).c_str() : "(end)"),
            ((m->done || m->manual_end == nullptr)
                 ? "(end)"
-                 : m->manual_end->DebugString().c_str()));
+                 : m->manual_end->DebugString(true).c_str()));
      }
    }
  } else if (!is_prepicked && !compaction_queue_.empty()) {
@@ -36,7 +36,7 @@ uint64_t DBImpl::MinObsoleteSstNumberToKeep() {
  return std::numeric_limits<uint64_t>::max();
 }

-// * Returns the list of live files in 'sst_live'
+// * Returns the list of live files in 'sst_live' and 'blob_live'.
 // If it's doing full scan:
 // * Returns the list of all files in the filesystem in
 // 'full_scan_candidate_files'.
@@ -77,26 +77,26 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  // Since job_context->min_pending_output is set, until file scan finishes,
  // mutex_ cannot be released. Otherwise, we might see no min_pending_output
  // here but later find newer generated unfinalized files while scanning.
-  if (!pending_outputs_.empty()) {
-    job_context->min_pending_output = *pending_outputs_.begin();
-  } else {
-    // delete all of them
-    job_context->min_pending_output = std::numeric_limits<uint64_t>::max();
-  }
+  job_context->min_pending_output = MinObsoleteSstNumberToKeep();

  // Get obsolete files.  This function will also update the list of
  // pending files in VersionSet().
-  versions_->GetObsoleteFiles(&job_context->sst_delete_files,
-                              &job_context->manifest_delete_files,
-                              job_context->min_pending_output);
+  versions_->GetObsoleteFiles(
+      &job_context->sst_delete_files, &job_context->blob_delete_files,
+      &job_context->manifest_delete_files, job_context->min_pending_output);

-  // Mark the elements in job_context->sst_delete_files as grabbedForPurge
-  // so that other threads calling FindObsoleteFiles with full_scan=true
-  // will not add these files to candidate list for purge.
+  // Mark the elements in job_context->sst_delete_files and
+  // job_context->blob_delete_files as "grabbed for purge" so that other threads
+  // calling FindObsoleteFiles with full_scan=true will not add these files to
+  // candidate list for purge.
  for (const auto& sst_to_del : job_context->sst_delete_files) {
    MarkAsGrabbedForPurge(sst_to_del.metadata->fd.GetNumber());
  }

+  for (const auto& blob_file : job_context->blob_delete_files) {
+    MarkAsGrabbedForPurge(blob_file.GetBlobFileNumber());
+  }
+
  // store the current filenum, lognum, etc
  job_context->manifest_file_number = versions_->manifest_file_number();
  job_context->pending_manifest_file_number =
@@ -104,7 +104,7 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  job_context->log_number = MinLogNumberToKeep();
  job_context->prev_log_number = versions_->prev_log_number();

-  versions_->AddLiveFiles(&job_context->sst_live);
+  versions_->AddLiveFiles(&job_context->sst_live, &job_context->blob_live);
  if (doing_the_full_scan) {
    InfoLogPrefix info_log_prefix(!immutable_db_options_.db_log_dir.empty(),
                                  dbname_);
@@ -257,8 +257,11 @@ bool CompareCandidateFile(const JobContext::CandidateFileInfo& first,
 void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
                                    const std::string& path_to_sync,
                                    FileType type, uint64_t number) {
+  TEST_SYNC_POINT_CALLBACK("DBImpl::DeleteObsoleteFileImpl::BeforeDeletion",
+                           const_cast<std::string*>(&fname));
+
  Status file_deletion_status;
-  if (type == kTableFile || type == kLogFile) {
+  if (type == kTableFile || type == kBlobFile || type == kLogFile) {
    file_deletion_status =
        DeleteDBFile(&immutable_db_options_, fname, path_to_sync,
                     /*force_bg=*/false, /*force_fg=*/!wal_in_db_path_);
@@ -304,19 +307,19 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
  // FindObsoleteFiles() should've populated this so nonzero
  assert(state.manifest_file_number != 0);

-  // Now, convert live list to an unordered map, WITHOUT mutex held;
-  // set is slow.
-  std::unordered_map<uint64_t, const FileDescriptor*> sst_live_map;
-  for (const FileDescriptor& fd : state.sst_live) {
-    sst_live_map[fd.GetNumber()] = &fd;
-  }
+  // Now, convert lists to unordered sets, WITHOUT mutex held; set is slow.
+  std::unordered_set<uint64_t> sst_live_set(state.sst_live.begin(),
+                                            state.sst_live.end());
+  std::unordered_set<uint64_t> blob_live_set(state.blob_live.begin(),
+                                             state.blob_live.end());
  std::unordered_set<uint64_t> log_recycle_files_set(
      state.log_recycle_files.begin(), state.log_recycle_files.end());

  auto candidate_files = state.full_scan_candidate_files;
  candidate_files.reserve(
      candidate_files.size() + state.sst_delete_files.size() +
-      state.log_delete_files.size() + state.manifest_delete_files.size());
+      state.blob_delete_files.size() + state.log_delete_files.size() +
+      state.manifest_delete_files.size());
  // We may ignore the dbname when generating the file names.
  for (auto& file : state.sst_delete_files) {
    candidate_files.emplace_back(
@@ -327,6 +330,11 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
    file.DeleteMetadata();
  }

+  for (const auto& blob_file : state.blob_delete_files) {
+    candidate_files.emplace_back(BlobFileName(blob_file.GetBlobFileNumber()),
+                                 blob_file.GetPath());
+  }
+
  for (auto file_num : state.log_delete_files) {
    if (file_num > 0) {
      candidate_files.emplace_back(LogFileName(file_num),
@@ -413,12 +421,19 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
      case kTableFile:
        // If the second condition is not there, this makes
        // DontDeletePendingOutputs fail
-        keep = (sst_live_map.find(number) != sst_live_map.end()) ||
+        keep = (sst_live_set.find(number) != sst_live_set.end()) ||
               number >= state.min_pending_output;
        if (!keep) {
          files_to_del.insert(number);
        }
        break;
+      case kBlobFile:
+        keep = number >= state.min_pending_output ||
+               (blob_live_set.find(number) != blob_live_set.end());
+        if (!keep) {
+          files_to_del.insert(number);
+        }
+        break;
      case kTempFile:
        // Any temp files that are currently being written to must
        // be recorded in pending_outputs_, which is inserted into "live".
@@ -428,7 +443,8 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
        //
        // TODO(yhchiang): carefully modify the third condition to safely
        //                 remove the temp options files.
-        keep = (sst_live_map.find(number) != sst_live_map.end()) ||
+        keep = (sst_live_set.find(number) != sst_live_set.end()) ||
+               (blob_live_set.find(number) != blob_live_set.end()) ||
               (number == state.pending_manifest_file_number) ||
               (to_delete.find(kOptionsFileNamePrefix) != std::string::npos);
        break;
@@ -451,7 +467,6 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
      case kDBLockFile:
      case kIdentityFile:
      case kMetaDatabase:
-      case kBlobFile:
        keep = true;
        break;
    }
@@ -467,6 +482,9 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
      TableCache::Evict(table_cache_.get(), number);
      fname = MakeTableFileName(candidate_file.file_path, number);
      dir_to_sync = candidate_file.file_path;
+    } else if (type == kBlobFile) {
+      fname = BlobFileName(candidate_file.file_path, number);
+      dir_to_sync = candidate_file.file_path;
    } else {
      dir_to_sync =
          (type == kLogFile) ? immutable_db_options_.wal_dir : dbname_;
@@ -665,16 +683,18 @@ uint64_t PrecomputeMinLogNumberToKeep(
  return min_log_number_to_keep;
 }

-Status DBImpl::CleanupFilesAfterRecovery() {
+Status DBImpl::FinishBestEffortsRecovery() {
  mutex_.AssertHeld();
  std::vector<std::string> paths;
-  paths.push_back(dbname_);
+  paths.push_back(NormalizePath(dbname_ + std::string(1, kFilePathSeparator)));
  for (const auto& db_path : immutable_db_options_.db_paths) {
-    paths.push_back(db_path.path);
+    paths.push_back(
+        NormalizePath(db_path.path + std::string(1, kFilePathSeparator)));
  }
  for (const auto* cfd : *versions_->GetColumnFamilySet()) {
    for (const auto& cf_path : cfd->ioptions()->cf_paths) {
-      paths.push_back(cf_path.path);
+      paths.push_back(
+          NormalizePath(cf_path.path + std::string(1, kFilePathSeparator)));
    }
  }
  // Dedup paths
@@ -693,7 +713,8 @@ Status DBImpl::CleanupFilesAfterRecovery() {
      if (!ParseFileName(fname, &number, &type)) {
        continue;
      }
-      const std::string normalized_fpath = NormalizePath(path + fname);
+      // path ends with '/' or '\\'
+      const std::string normalized_fpath = path + fname;
      largest_file_number = std::max(largest_file_number, number);
      if (type == kTableFile && number >= next_file_number &&
          files_to_delete.find(normalized_fpath) == files_to_delete.end()) {
@@ -704,8 +725,22 @@ Status DBImpl::CleanupFilesAfterRecovery() {
  if (largest_file_number > next_file_number) {
    versions_->next_file_number_.store(largest_file_number + 1);
  }
+
+  VersionEdit edit;
+  edit.SetNextFile(versions_->next_file_number_.load());
+  assert(versions_->GetColumnFamilySet());
+  ColumnFamilyData* default_cfd = versions_->GetColumnFamilySet()->GetDefault();
+  assert(default_cfd);
+  // Even if new_descriptor_log is false, we will still switch to a new
+  // MANIFEST and update CURRENT file, since this is in recovery.
+  Status s = versions_->LogAndApply(
+      default_cfd, *default_cfd->GetLatestMutableCFOptions(), &edit, &mutex_,
+      directories_.GetDbDir(), /*new_descriptor_log*/ false);
+  if (!s.ok()) {
+    return s;
+  }
+
  mutex_.Unlock();
-  Status s;
  for (const auto& fname : files_to_delete) {
    s = env_->DeleteFile(fname);
    if (!s.ok()) {
@@ -425,7 +425,10 @@ Status DBImpl::Recover(
    s = versions_->TryRecover(column_families, read_only, &db_id_,
                              &missing_table_file);
    if (s.ok()) {
-      s = CleanupFilesAfterRecovery();
+      // TryRecover may delete previous column_family_set_.
+      column_family_memtables_.reset(
+          new ColumnFamilyMemTablesImpl(versions_->GetColumnFamilySet()));
+      s = FinishBestEffortsRecovery();
    }
  }
  if (!s.ok()) {
@@ -87,7 +87,8 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
      super_version->version_number, read_callback);
  auto internal_iter =
      NewInternalIterator(read_options, cfd, super_version, db_iter->GetArena(),
-                          db_iter->GetRangeDelAggregator(), read_seq);
+                          db_iter->GetRangeDelAggregator(), read_seq,
+                          /* allow_unprepared_value */ true);
  db_iter->SetIterUnderDBIter(internal_iter);
  return db_iter;
 }
@@ -118,7 +119,8 @@ Status DBImplReadOnly::NewIterators(
        sv->version_number, read_callback);
    auto* internal_iter =
        NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
-                            db_iter->GetRangeDelAggregator(), read_seq);
+                            db_iter->GetRangeDelAggregator(), read_seq,
+                            /* allow_unprepared_value */ true);
    db_iter->SetIterUnderDBIter(internal_iter);
    iterators->push_back(db_iter);
  }
@@ -126,12 +128,38 @@ Status DBImplReadOnly::NewIterators(
  return Status::OK();
 }

+namespace {
+// Return OK if dbname exists in the file system
+// or create_if_missing is false
+Status OpenForReadOnlyCheckExistence(const DBOptions& db_options,
+                                     const std::string& dbname) {
+  Status s;
+  if (!db_options.create_if_missing) {
+    // Attempt to read "CURRENT" file
+    const std::shared_ptr<FileSystem>& fs = db_options.env->GetFileSystem();
+    std::string manifest_path;
+    uint64_t manifest_file_number;
+    s = VersionSet::GetCurrentManifestPath(dbname, fs.get(), &manifest_path,
+                                           &manifest_file_number);
+    if (!s.ok()) {
+      return Status::NotFound(CurrentFileName(dbname), "does not exist");
+    }
+  }
+  return s;
+}
+}  // namespace
+
 Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
                           DB** dbptr, bool /*error_if_log_file_exist*/) {
+  // If dbname does not exist in the file system, should not do anything
+  Status s = OpenForReadOnlyCheckExistence(options, dbname);
+  if (!s.ok()) {
+    return s;
+  }
+
  *dbptr = nullptr;

  // Try to first open DB as fully compacted DB
-  Status s;
  s = CompactedDBImpl::Open(options, dbname, dbptr);
  if (s.ok()) {
    return s;
@@ -144,7 +172,8 @@ Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
      ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
  std::vector<ColumnFamilyHandle*> handles;

-  s = DB::OpenForReadOnly(db_options, dbname, column_families, &handles, dbptr);
+  s = DBImplReadOnly::OpenForReadOnlyWithoutCheck(
+      db_options, dbname, column_families, &handles, dbptr);
  if (s.ok()) {
    assert(handles.size() == 1);
    // i can delete the handle since DBImpl is always holding a
@@ -159,6 +188,22 @@ Status DB::OpenForReadOnly(
    const std::vector<ColumnFamilyDescriptor>& column_families,
    std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
    bool error_if_log_file_exist) {
+  // If dbname does not exist in the file system, should not do anything
+  Status s = OpenForReadOnlyCheckExistence(db_options, dbname);
+  if (!s.ok()) {
+    return s;
+  }
+
+  return DBImplReadOnly::OpenForReadOnlyWithoutCheck(
+      db_options, dbname, column_families, handles, dbptr,
+      error_if_log_file_exist);
+}
+
+Status DBImplReadOnly::OpenForReadOnlyWithoutCheck(
+    const DBOptions& db_options, const std::string& dbname,
+    const std::vector<ColumnFamilyDescriptor>& column_families,
+    std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
+    bool error_if_log_file_exist) {
  *dbptr = nullptr;
  handles->clear();

@@ -130,6 +130,15 @@ class DBImplReadOnly : public DBImpl {
  }

 private:
+  // A "helper" function for DB::OpenForReadOnly without column families
+  // to reduce unnecessary I/O
+  // It has the same functionality as DB::OpenForReadOnly with column families
+  // but does not check the existence of dbname in the file system
+  static Status OpenForReadOnlyWithoutCheck(
+      const DBOptions& db_options, const std::string& dbname,
+      const std::vector<ColumnFamilyDescriptor>& column_families,
+      std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
+      bool error_if_log_file_exist = false);
  friend class DB;
 };
 }  // namespace ROCKSDB_NAMESPACE
@@ -417,7 +417,8 @@ ArenaWrappedDBIter* DBImplSecondary::NewIteratorImpl(
      super_version->version_number, read_callback);
  auto internal_iter =
      NewInternalIterator(read_options, cfd, super_version, db_iter->GetArena(),
-                          db_iter->GetRangeDelAggregator(), snapshot);
+                          db_iter->GetRangeDelAggregator(), snapshot,
+                          /* allow_unprepared_value */ true);
  db_iter->SetIterUnderDBIter(internal_iter);
  return db_iter;
 }
@@ -1820,6 +1820,8 @@ Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,
  const Slice* ts = opt.timestamp;
  assert(nullptr != ts);
  size_t ts_sz = ts->size();
+  assert(column_family->GetComparator());
+  assert(ts_sz == column_family->GetComparator()->timestamp_size());
  WriteBatch batch(key.size() + ts_sz + value.size() + 24, /*max_bytes=*/0,
                   ts_sz);
  Status s = batch.Put(column_family, key, value);
@@ -1835,8 +1837,30 @@ Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,

 Status DB::Delete(const WriteOptions& opt, ColumnFamilyHandle* column_family,
                  const Slice& key) {
-  WriteBatch batch;
-  batch.Delete(column_family, key);
+  if (nullptr == opt.timestamp) {
+    WriteBatch batch;
+    Status s = batch.Delete(column_family, key);
+    if (!s.ok()) {
+      return s;
+    }
+    return Write(opt, &batch);
+  }
+  const Slice* ts = opt.timestamp;
+  assert(ts != nullptr);
+  const size_t ts_sz = ts->size();
+  constexpr size_t kKeyAndValueLenSize = 11;
+  constexpr size_t kWriteBatchOverhead =
+      WriteBatchInternal::kHeader + sizeof(ValueType) + kKeyAndValueLenSize;
+  WriteBatch batch(key.size() + ts_sz + kWriteBatchOverhead, /*max_bytes=*/0,
+                   ts_sz);
+  Status s = batch.Delete(column_family, key);
+  if (!s.ok()) {
+    return s;
+  }
+  s = batch.AssignTimestamp(*ts);
+  if (!s.ok()) {
+    return s;
+  }
  return Write(opt, &batch);
 }

@@ -45,6 +45,8 @@ class DBSecondaryTest : public DBTestBase {

  void OpenSecondary(const Options& options);

+  Status TryOpenSecondary(const Options& options);
+
  void OpenSecondaryWithColumnFamilies(
      const std::vector<std::string>& column_families, const Options& options);

@@ -70,9 +72,13 @@ class DBSecondaryTest : public DBTestBase {
 };

 void DBSecondaryTest::OpenSecondary(const Options& options) {
+  ASSERT_OK(TryOpenSecondary(options));
+}
+
+Status DBSecondaryTest::TryOpenSecondary(const Options& options) {
  Status s =
      DB::OpenAsSecondary(options, dbname_, secondary_path_, &db_secondary_);
-  ASSERT_OK(s);
+  return s;
 }

 void DBSecondaryTest::OpenSecondaryWithColumnFamilies(
@@ -858,6 +864,56 @@ TEST_F(DBSecondaryTest, CheckConsistencyWhenOpen) {
  thread.join();
  ASSERT_TRUE(called);
 }
+
+TEST_F(DBSecondaryTest, StartFromInconsistent) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "value"));
+  ASSERT_OK(Flush());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
+        ASSERT_NE(nullptr, arg);
+        *(reinterpret_cast<Status*>(arg)) =
+            Status::Corruption("Inject corruption");
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  Options options1;
+  Status s = TryOpenSecondary(options1);
+  ASSERT_TRUE(s.IsCorruption());
+}
+
+TEST_F(DBSecondaryTest, InconsistencyDuringCatchUp) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "value"));
+  ASSERT_OK(Flush());
+
+  Options options1;
+  OpenSecondary(options1);
+
+  {
+    std::string value;
+    ASSERT_OK(db_secondary_->Get(ReadOptions(), "foo", &value));
+    ASSERT_EQ("value", value);
+  }
+
+  ASSERT_OK(Put("bar", "value1"));
+  ASSERT_OK(Flush());
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
+        ASSERT_NE(nullptr, arg);
+        *(reinterpret_cast<Status*>(arg)) =
+            Status::Corruption("Inject corruption");
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  Status s = db_secondary_->TryCatchUpWithPrimary();
+  ASSERT_TRUE(s.IsCorruption());
+}
 #endif  //! ROCKSDB_LITE

 }  // namespace ROCKSDB_NAMESPACE
@@ -73,6 +73,7 @@ DBIter::DBIter(Env* _env, const ReadOptions& read_options,
      cfd_(cfd),
      start_seqnum_(read_options.iter_start_seqnum),
      timestamp_ub_(read_options.timestamp),
+      timestamp_lb_(read_options.iter_start_ts),
      timestamp_size_(timestamp_ub_ ? timestamp_ub_->size() : 0) {
  RecordTick(statistics_, NO_ITERATOR_CREATED);
  if (pin_thru_lifetime_) {
@@ -246,27 +247,36 @@ bool DBIter::FindNextUserEntryInternal(bool skipping_saved_key,

    assert(ikey_.user_key.size() >= timestamp_size_);
    Slice ts;
+    bool more_recent = false;
    if (timestamp_size_ > 0) {
      ts = ExtractTimestampFromUserKey(ikey_.user_key, timestamp_size_);
    }
-    if (IsVisible(ikey_.sequence, ts)) {
+    if (IsVisible(ikey_.sequence, ts, &more_recent)) {
      // If the previous entry is of seqnum 0, the current entry will not
      // possibly be skipped. This condition can potentially be relaxed to
      // prev_key.seq <= ikey_.sequence. We are cautious because it will be more
      // prone to bugs causing the same user key with the same sequence number.
-      if (!is_prev_key_seqnum_zero && skipping_saved_key &&
-          user_comparator_.CompareWithoutTimestamp(
-              ikey_.user_key, saved_key_.GetUserKey()) <= 0) {
+      // Note that with current timestamp implementation, the same user key can
+      // have different timestamps and zero sequence number on the bottommost
+      // level. This may change in the future.
+      if ((!is_prev_key_seqnum_zero || timestamp_size_ > 0) &&
+          skipping_saved_key &&
+          CompareKeyForSkip(ikey_.user_key, saved_key_.GetUserKey()) <= 0) {
        num_skipped++;  // skip this entry
        PERF_COUNTER_ADD(internal_key_skipped_count, 1);
      } else {
        assert(!skipping_saved_key ||
-               user_comparator_.CompareWithoutTimestamp(
-                   ikey_.user_key, saved_key_.GetUserKey()) > 0);
+               CompareKeyForSkip(ikey_.user_key, saved_key_.GetUserKey()) > 0);
+        if (!iter_.PrepareValue()) {
+          assert(!iter_.status().ok());
+          valid_ = false;
+          return false;
+        }
        num_skipped = 0;
        reseek_done = false;
        switch (ikey_.type) {
          case kTypeDeletion:
+          case kTypeDeletionWithTimestamp:
          case kTypeSingleDeletion:
            // Arrange to skip all upcoming entries for this key since
            // they are hidden by this deletion.
@@ -363,11 +373,13 @@ bool DBIter::FindNextUserEntryInternal(bool skipping_saved_key,
        }
      }
    } else {
-      PERF_COUNTER_ADD(internal_recent_skipped_count, 1);
+      if (more_recent) {
+        PERF_COUNTER_ADD(internal_recent_skipped_count, 1);
+      }

-      // This key was inserted after our snapshot was taken.
-      // If this happens too many times in a row for the same user key, we want
-      // to seek to the target sequence number.
+      // This key was inserted after our snapshot was taken or skipped by
+      // timestamp range. If this happens too many times in a row for the same
+      // user key, we want to seek to the target sequence number.
      int cmp = user_comparator_.CompareWithoutTimestamp(
          ikey_.user_key, saved_key_.GetUserKey());
      if (cmp == 0 || (skipping_saved_key && cmp < 0)) {
@@ -446,6 +458,7 @@ bool DBIter::FindNextUserEntryInternal(bool skipping_saved_key,
 // Scan from the newer entries to older entries.
 // PRE: iter_.key() points to the first merge type entry
 //      saved_key_ stores the user key
+//      iter_.PrepareValue() has been called
 // POST: saved_value_ has the merged value for the user key
 //       iter_ points to the next entry (or invalid)
 bool DBIter::MergeValuesNewToOld() {
@@ -475,14 +488,21 @@ bool DBIter::MergeValuesNewToOld() {
    if (!user_comparator_.Equal(ikey.user_key, saved_key_.GetUserKey())) {
      // hit the next user key, stop right here
      break;
-    } else if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type ||
+    }
+    if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type ||
               range_del_agg_.ShouldDelete(
                   ikey, RangeDelPositioningMode::kForwardTraversal)) {
      // hit a delete with the same user key, stop right here
      // iter_ is positioned after delete
      iter_.Next();
      break;
-    } else if (kTypeValue == ikey.type) {
+    }
+    if (!iter_.PrepareValue()) {
+      valid_ = false;
+      return false;
+    }
+
+    if (kTypeValue == ikey.type) {
      // hit a put, merge the put value with operands and store the
      // final result in saved_value_. We are done!
      const Slice val = iter_.value();
@@ -754,6 +774,11 @@ bool DBIter::FindValueForCurrentKey() {
      return FindValueForCurrentKeyUsingSeek();
    }

+    if (!iter_.PrepareValue()) {
+      valid_ = false;
+      return false;
+    }
+
    last_key_entry_type = ikey.type;
    switch (last_key_entry_type) {
      case kTypeValue:
@@ -931,6 +956,10 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
    valid_ = false;
    return false;
  }
+  if (!iter_.PrepareValue()) {
+    valid_ = false;
+    return false;
+  }
  if (ikey.type == kTypeValue || ikey.type == kTypeBlobIndex) {
    assert(iter_.iter()->IsValuePinned());
    pinned_value_ = iter_.value();
@@ -962,12 +991,17 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
    if (!user_comparator_.Equal(ikey.user_key, saved_key_.GetUserKey())) {
      break;
    }
-
    if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
        range_del_agg_.ShouldDelete(
            ikey, RangeDelPositioningMode::kForwardTraversal)) {
      break;
-    } else if (ikey.type == kTypeValue) {
+    }
+    if (!iter_.PrepareValue()) {
+      valid_ = false;
+      return false;
+    }
+
+    if (ikey.type == kTypeValue) {
      const Slice val = iter_.value();
      Status s = MergeHelper::TimedFullMerge(
          merge_operator_, saved_key_.GetUserKey(), &val,
@@ -1101,20 +1135,24 @@ bool DBIter::TooManyInternalKeysSkipped(bool increment) {
  return false;
 }

-bool DBIter::IsVisible(SequenceNumber sequence, const Slice& ts) {
+bool DBIter::IsVisible(SequenceNumber sequence, const Slice& ts,
+                       bool* more_recent) {
  // Remember that comparator orders preceding timestamp as larger.
-  int cmp_ts = timestamp_ub_ != nullptr
-                   ? user_comparator_.CompareTimestamp(ts, *timestamp_ub_)
-                   : 0;
-  if (cmp_ts > 0) {
-    return false;
-  }
-  if (read_callback_ == nullptr) {
-    return sequence <= sequence_;
-  } else {
-    // TODO(yanqin): support timestamp in read_callback_.
-    return read_callback_->IsVisible(sequence);
+  // TODO(yanqin): support timestamp in read_callback_.
+  bool visible_by_seq = (read_callback_ == nullptr)
+                            ? sequence <= sequence_
+                            : read_callback_->IsVisible(sequence);
+
+  bool visible_by_ts =
+      (timestamp_ub_ == nullptr ||
+       user_comparator_.CompareTimestamp(ts, *timestamp_ub_) <= 0) &&
+      (timestamp_lb_ == nullptr ||
+       user_comparator_.CompareTimestamp(ts, *timestamp_lb_) >= 0);
+
+  if (more_recent) {
+    *more_recent = !visible_by_seq;
  }
+  return visible_by_seq && visible_by_ts;
 }

 void DBIter::SetSavedKeyToSeekTarget(const Slice& target) {
@@ -231,7 +231,8 @@ class DBIter final : public Iterator {
  // entry can be found within the prefix.
  void PrevInternal(const Slice* prefix);
  bool TooManyInternalKeysSkipped(bool increment = true);
-  bool IsVisible(SequenceNumber sequence, const Slice& ts);
+  bool IsVisible(SequenceNumber sequence, const Slice& ts,
+                 bool* more_recent = nullptr);

  // Temporarily pin the blocks that we encounter until ReleaseTempPinnedData()
  // is called
@@ -270,6 +271,15 @@ class DBIter final : public Iterator {
    return expect_total_order_inner_iter_;
  }

+  // If lower bound of timestamp is given by ReadOptions.iter_start_ts, we need
+  // to return versions of the same key. We cannot just skip if the key value
+  // is the same but timestamps are different but fall in timestamp range.
+  inline int CompareKeyForSkip(const Slice& a, const Slice& b) {
+    return timestamp_lb_ != nullptr
+               ? user_comparator_.Compare(a, b)
+               : user_comparator_.CompareWithoutTimestamp(a, b);
+  }
+
  const SliceTransform* prefix_extractor_;
  Env* const env_;
  Logger* logger_;
@@ -338,6 +348,7 @@ class DBIter final : public Iterator {
  // if this value > 0 iterator will return internal keys
  SequenceNumber start_seqnum_;
  const Slice* const timestamp_ub_;
+  const Slice* const timestamp_lb_;
  const size_t timestamp_size_;
 };

@@ -1167,32 +1167,62 @@ TEST_P(DBIteratorTest, IndexWithFirstKey) {
    ropt.tailing = tailing;
    std::unique_ptr<Iterator> iter(NewIterator(ropt));

+    ropt.read_tier = ReadTier::kBlockCacheTier;
+    std::unique_ptr<Iterator> nonblocking_iter(NewIterator(ropt));
+
    iter->Seek("b10");
    ASSERT_TRUE(iter->Valid());
    EXPECT_EQ("b2", iter->key().ToString());
    EXPECT_EQ("y2", iter->value().ToString());
    EXPECT_EQ(1, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));

+    // The cache-only iterator should succeed too, using the blocks pulled into
+    // the cache by the previous iterator.
+    nonblocking_iter->Seek("b10");
+    ASSERT_TRUE(nonblocking_iter->Valid());
+    EXPECT_EQ("b2", nonblocking_iter->key().ToString());
+    EXPECT_EQ("y2", nonblocking_iter->value().ToString());
+    EXPECT_EQ(1, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+
+    // ... but it shouldn't be able to step forward since the next block is
+    // not in cache yet.
+    nonblocking_iter->Next();
+    ASSERT_FALSE(nonblocking_iter->Valid());
+    ASSERT_TRUE(nonblocking_iter->status().IsIncomplete());
+
+    // ... nor should a seek to the next key succeed.
+    nonblocking_iter->Seek("b20");
+    ASSERT_FALSE(nonblocking_iter->Valid());
+    ASSERT_TRUE(nonblocking_iter->status().IsIncomplete());
+
    iter->Next();
    ASSERT_TRUE(iter->Valid());
    EXPECT_EQ("b3", iter->key().ToString());
    EXPECT_EQ("y3", iter->value().ToString());
-    EXPECT_EQ(2, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
-    EXPECT_EQ(0, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+    EXPECT_EQ(4, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    EXPECT_EQ(1, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+
+    // After the blocking iterator loaded the next block, the nonblocking
+    // iterator's seek should succeed.
+    nonblocking_iter->Seek("b20");
+    ASSERT_TRUE(nonblocking_iter->Valid());
+    EXPECT_EQ("b3", nonblocking_iter->key().ToString());
+    EXPECT_EQ("y3", nonblocking_iter->value().ToString());
+    EXPECT_EQ(2, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));

    iter->Seek("c0");
    ASSERT_TRUE(iter->Valid());
    EXPECT_EQ("c0", iter->key().ToString());
    EXPECT_EQ("z1,z2", iter->value().ToString());
-    EXPECT_EQ(0, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
-    EXPECT_EQ(4, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    EXPECT_EQ(2, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+    EXPECT_EQ(6, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));

    iter->Next();
    ASSERT_TRUE(iter->Valid());
    EXPECT_EQ("c3", iter->key().ToString());
    EXPECT_EQ("z3", iter->value().ToString());
-    EXPECT_EQ(0, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
-    EXPECT_EQ(5, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    EXPECT_EQ(2, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+    EXPECT_EQ(7, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));

    iter.reset();

@@ -1207,13 +1237,13 @@ TEST_P(DBIteratorTest, IndexWithFirstKey) {
    ASSERT_TRUE(iter->Valid());
    EXPECT_EQ("b2", iter->key().ToString());
    EXPECT_EQ("y2", iter->value().ToString());
-    EXPECT_EQ(1, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
-    EXPECT_EQ(5, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    EXPECT_EQ(3, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+    EXPECT_EQ(7, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));

    iter->Next();
    ASSERT_FALSE(iter->Valid());
-    EXPECT_EQ(1, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
-    EXPECT_EQ(5, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    EXPECT_EQ(3, stats->getTickerCount(BLOCK_CACHE_DATA_HIT));
+    EXPECT_EQ(7, stats->getTickerCount(BLOCK_CACHE_DATA_MISS));
  }
 }

@@ -33,13 +33,14 @@ class DBOptionsTest : public DBTestBase {
  std::unordered_map<std::string, std::string> GetMutableDBOptionsMap(
      const DBOptions& options) {
    std::string options_str;
-    GetStringFromDBOptions(&options_str, options);
+    ConfigOptions config_options;
+    config_options.delimiter = "; ";
+    GetStringFromDBOptions(config_options, options, &options_str);
    std::unordered_map<std::string, std::string> options_map;
    StringToMap(options_str, &options_map);
    std::unordered_map<std::string, std::string> mutable_map;
    for (const auto opt : db_options_type_info) {
-      if (opt.second.is_mutable &&
-          opt.second.verification != OptionVerificationType::kDeprecated) {
+      if (opt.second.IsMutable() && opt.second.ShouldSerialize()) {
        mutable_map[opt.first] = options_map[opt.first];
      }
    }
@@ -49,13 +50,15 @@ class DBOptionsTest : public DBTestBase {
  std::unordered_map<std::string, std::string> GetMutableCFOptionsMap(
      const ColumnFamilyOptions& options) {
    std::string options_str;
-    GetStringFromColumnFamilyOptions(&options_str, options);
+    ConfigOptions config_options;
+    config_options.delimiter = "; ";
+
+    GetStringFromColumnFamilyOptions(config_options, options, &options_str);
    std::unordered_map<std::string, std::string> options_map;
    StringToMap(options_str, &options_map);
    std::unordered_map<std::string, std::string> mutable_map;
    for (const auto opt : cf_options_type_info) {
-      if (opt.second.is_mutable &&
-          opt.second.verification != OptionVerificationType::kDeprecated) {
+      if (opt.second.IsMutable() && opt.second.ShouldSerialize()) {
        mutable_map[opt.first] = options_map[opt.first];
      }
    }
@@ -404,6 +407,20 @@ TEST_F(DBOptionsTest, SetBackgroundCompactionThreads) {
  ASSERT_EQ(3, dbfull()->TEST_BGCompactionsAllowed());
 }

+TEST_F(DBOptionsTest, SetBackgroundFlushThreads) {
+  Options options;
+  options.create_if_missing = true;
+  options.max_background_flushes = 1;
+  options.env = env_;
+  Reopen(options);
+  ASSERT_EQ(1, dbfull()->TEST_BGFlushesAllowed());
+  ASSERT_EQ(1, env_->GetBackgroundThreads(Env::Priority::HIGH));
+  ASSERT_OK(dbfull()->SetDBOptions({{"max_background_flushes", "3"}}));
+  ASSERT_EQ(3, env_->GetBackgroundThreads(Env::Priority::HIGH));
+  ASSERT_EQ(3, dbfull()->TEST_BGFlushesAllowed());
+}
+
+
 TEST_F(DBOptionsTest, SetBackgroundJobs) {
  Options options;
  options.create_if_missing = true;
@@ -859,8 +876,6 @@ TEST_F(DBOptionsTest, FIFOTtlBackwardCompatible) {
  ASSERT_EQ(dbfull()->GetOptions().ttl, 191);
 }

-#endif  // ROCKSDB_LITE
-
 TEST_F(DBOptionsTest, ChangeCompression) {
  if (!Snappy_Supported() || !LZ4_Supported()) {
    return;
@@ -872,6 +887,7 @@ TEST_F(DBOptionsTest, ChangeCompression) {
  options.compression = CompressionType::kLZ4Compression;
  options.bottommost_compression = CompressionType::kNoCompression;
  options.bottommost_compression_opts.level = 2;
+  options.bottommost_compression_opts.parallel_threads = 1;

  ASSERT_OK(TryReopen(options));

@@ -897,12 +913,14 @@ TEST_F(DBOptionsTest, ChangeCompression) {
  ASSERT_TRUE(compacted);
  ASSERT_EQ(CompressionType::kNoCompression, compression_used);
  ASSERT_EQ(options.compression_opts.level, compression_opt_used.level);
+  ASSERT_EQ(options.compression_opts.parallel_threads,
+            compression_opt_used.parallel_threads);

  compression_used = CompressionType::kLZ4Compression;
  compacted = false;
  ASSERT_OK(dbfull()->SetOptions(
      {{"bottommost_compression", "kSnappyCompression"},
-       {"bottommost_compression_opts", "0:6:0:0:0:true"}}));
+       {"bottommost_compression_opts", "0:6:0:0:4:true"}}));
  ASSERT_OK(Put("foo", "foofoofoo"));
  ASSERT_OK(Put("bar", "foofoofoo"));
  ASSERT_OK(Flush());
@@ -913,10 +931,13 @@ TEST_F(DBOptionsTest, ChangeCompression) {
  ASSERT_TRUE(compacted);
  ASSERT_EQ(CompressionType::kSnappyCompression, compression_used);
  ASSERT_EQ(6, compression_opt_used.level);
+  // Right now parallel_level is not yet allowed to be changed.

  SyncPoint::GetInstance()->DisableProcessing();
 }

+#endif  // ROCKSDB_LITE
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -47,6 +47,21 @@ TEST_F(DBRangeDelTest, WriteBatchWithIndexNotSupported) {
  ASSERT_TRUE(indexedBatch.DeleteRange("dr1", "dr1").IsNotSupported());
 }

+TEST_F(DBRangeDelTest, EndSameAsStartCoversNothing) {
+  ASSERT_OK(db_->Put(WriteOptions(), "b", "val"));
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "b", "b"));
+  ASSERT_EQ("val", Get("b"));
+}
+
+TEST_F(DBRangeDelTest, EndComesBeforeStartInvalidArgument) {
+  db_->Put(WriteOptions(), "b", "val");
+  ASSERT_TRUE(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "b", "a")
+          .IsInvalidArgument());
+  ASSERT_EQ("val", Get("b"));
+}
+
 TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
  do {
    DestroyAndReopen(CurrentOptions());
@@ -174,7 +189,7 @@ TEST_F(DBRangeDelTest, MaxCompactionBytesCutsOutputFiles) {
  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);

-  for (size_t i = 0; i < files[1].size() - 1; ++i) {
+  for (size_t i = 0; i + 1 < files[1].size(); ++i) {
    ASSERT_TRUE(InternalKeyComparator(opts.comparator)
                    .Compare(files[1][i].largest, files[1][i + 1].smallest) <
                0);
@@ -356,37 +356,25 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
      "InstrumentedCondVar::TimedWaitInternal", [&](void* arg) {
        // Turn timed wait into a simulated sleep
        uint64_t* abs_time_us = static_cast<uint64_t*>(arg);
-        int64_t cur_time = 0;
-        env_->GetCurrentTime(&cur_time);
-        if (*abs_time_us > static_cast<uint64_t>(cur_time)) {
-          env_->addon_time_.fetch_add(*abs_time_us -
-                                      static_cast<uint64_t>(cur_time));
+        uint64_t cur_time = env_->NowMicros();
+        if (*abs_time_us > cur_time) {
+          env_->addon_time_.fetch_add(*abs_time_us - cur_time);
        }

        // Randomly sleep shortly
        env_->addon_time_.fetch_add(
            static_cast<uint64_t>(Random::GetTLSInstance()->Uniform(10)));

-        // Set wait until time to before current to force not to sleep.
-        int64_t real_cur_time = 0;
-        Env::Default()->GetCurrentTime(&real_cur_time);
-        *abs_time_us = static_cast<uint64_t>(real_cur_time);
+        // Set wait until time to before (actual) current time to force not
+        // to sleep
+        *abs_time_us = Env::Default()->NowMicros();
      });

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

-  env_->no_slowdown_ = true;
-  env_->time_elapse_only_sleep_ = true;
  Options options = CurrentOptions();
+  env_->SetTimeElapseOnlySleep(&options);
  options.disable_auto_compactions = true;
-  // Need to disable stats dumping and persisting which also use
-  // RepeatableThread, one of whose member variables is of type
-  // InstrumentedCondVar. The callback for
-  // InstrumentedCondVar::TimedWaitInternal can be triggered by stats dumping
-  // and persisting threads and cause time_spent_deleting measurement to become
-  // incorrect.
-  options.stats_dump_period_sec = 0;
-  options.stats_persist_period_sec = 0;
  options.env = env_;

  int64_t rate_bytes_per_sec = 1024 * 10;  // 10 Kbs / Sec
@@ -1421,7 +1421,7 @@ TEST_F(DBTest, ApproximateSizesMemTable) {
    keys[i * 3 + 1] = i * 5 + 1;
    keys[i * 3 + 2] = i * 5 + 2;
  }
-  std::random_shuffle(std::begin(keys), std::end(keys));
+  RandomShuffle(std::begin(keys), std::end(keys));

  for (int i = 0; i < N * 3; i++) {
    ASSERT_OK(Put(Key(keys[i] + 1000), RandomString(&rnd, 1024)));
@@ -1489,18 +1489,24 @@ TEST_F(DBTest, ApproximateSizesMemTable) {
 }

 TEST_F(DBTest, ApproximateSizesFilesWithErrorMargin) {
+  // Roughly 4 keys per data block, 1000 keys per file,
+  // with filter substantially larger than a data block
+  BlockBasedTableOptions table_options;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(16));
+  table_options.block_size = 100;
  Options options = CurrentOptions();
-  options.write_buffer_size = 1024 * 1024;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  options.write_buffer_size = 24 * 1024;
  options.compression = kNoCompression;
  options.create_if_missing = true;
-  options.target_file_size_base = 1024 * 1024;
+  options.target_file_size_base = 24 * 1024;
  DestroyAndReopen(options);
  const auto default_cf = db_->DefaultColumnFamily();

  const int N = 64000;
  Random rnd(301);
  for (int i = 0; i < N; i++) {
-    ASSERT_OK(Put(Key(i), RandomString(&rnd, 1024)));
+    ASSERT_OK(Put(Key(i), RandomString(&rnd, 24)));
  }
  // Flush everything to files
  Flush();
@@ -1509,7 +1515,7 @@ TEST_F(DBTest, ApproximateSizesFilesWithErrorMargin) {

  // Write more keys
  for (int i = N; i < (N + N / 4); i++) {
-    ASSERT_OK(Put(Key(i), RandomString(&rnd, 1024)));
+    ASSERT_OK(Put(Key(i), RandomString(&rnd, 24)));
  }
  // Flush everything to files again
  Flush();
@@ -1517,27 +1523,45 @@ TEST_F(DBTest, ApproximateSizesFilesWithErrorMargin) {
  // Wait for compaction to finish
  ASSERT_OK(dbfull()->TEST_WaitForCompact());

-  const std::string start = Key(0);
-  const std::string end = Key(2 * N);
-  const Range r(start, end);
+  {
+    const std::string start = Key(0);
+    const std::string end = Key(2 * N);
+    const Range r(start, end);

-  SizeApproximationOptions size_approx_options;
-  size_approx_options.include_memtabtles = false;
-  size_approx_options.include_files = true;
-  size_approx_options.files_size_error_margin = -1.0;  // disabled
+    SizeApproximationOptions size_approx_options;
+    size_approx_options.include_memtabtles = false;
+    size_approx_options.include_files = true;
+    size_approx_options.files_size_error_margin = -1.0;  // disabled

-  // Get the precise size without any approximation heuristic
-  uint64_t size;
-  db_->GetApproximateSizes(size_approx_options, default_cf, &r, 1, &size);
-  ASSERT_NE(size, 0);
+    // Get the precise size without any approximation heuristic
+    uint64_t size;
+    db_->GetApproximateSizes(size_approx_options, default_cf, &r, 1, &size);
+    ASSERT_NE(size, 0);

-  // Get the size with an approximation heuristic
-  uint64_t size2;
-  const double error_margin = 0.2;
-  size_approx_options.files_size_error_margin = error_margin;
-  db_->GetApproximateSizes(size_approx_options, default_cf, &r, 1, &size2);
-  ASSERT_LT(size2, size * (1 + error_margin));
-  ASSERT_GT(size2, size * (1 - error_margin));
+    // Get the size with an approximation heuristic
+    uint64_t size2;
+    const double error_margin = 0.2;
+    size_approx_options.files_size_error_margin = error_margin;
+    db_->GetApproximateSizes(size_approx_options, default_cf, &r, 1, &size2);
+    ASSERT_LT(size2, size * (1 + error_margin));
+    ASSERT_GT(size2, size * (1 - error_margin));
+  }
+
+  {
+    // Ensure that metadata is not falsely attributed only to the last data in
+    // the file. (In some applications, filters can be large portion of data
+    // size.)
+    // Perform many queries over small range, enough to ensure crossing file
+    // boundary, and make sure we never see a spike for large filter.
+    for (int i = 0; i < 3000; i += 10) {
+      const std::string start = Key(i);
+      const std::string end = Key(i + 11);  // overlap by 1 key
+      const Range r(start, end);
+      uint64_t size;
+      db_->GetApproximateSizes(&r, 1, &size);
+      ASSERT_LE(size, 11 * 100);
+    }
+  }
 }

 TEST_F(DBTest, GetApproximateMemTableStats) {
@@ -1675,12 +1699,13 @@ TEST_F(DBTest, ApproximateSizes_MixOfSmallAndLarge) {
      ASSERT_TRUE(Between(Size("", Key(2), 1), 20000, 21000));
      ASSERT_TRUE(Between(Size("", Key(3), 1), 120000, 121000));
      ASSERT_TRUE(Between(Size("", Key(4), 1), 130000, 131000));
-      ASSERT_TRUE(Between(Size("", Key(5), 1), 230000, 231000));
-      ASSERT_TRUE(Between(Size("", Key(6), 1), 240000, 241000));
-      ASSERT_TRUE(Between(Size("", Key(7), 1), 540000, 541000));
-      ASSERT_TRUE(Between(Size("", Key(8), 1), 550000, 560000));
+      ASSERT_TRUE(Between(Size("", Key(5), 1), 230000, 232000));
+      ASSERT_TRUE(Between(Size("", Key(6), 1), 240000, 242000));
+      // Ensure some overhead is accounted for, even without including all
+      ASSERT_TRUE(Between(Size("", Key(7), 1), 540500, 545000));
+      ASSERT_TRUE(Between(Size("", Key(8), 1), 550500, 555000));

-      ASSERT_TRUE(Between(Size(Key(3), Key(5), 1), 110000, 111000));
+      ASSERT_TRUE(Between(Size(Key(3), Key(5), 1), 110100, 111000));

      dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[1]);
    }
@@ -2309,6 +2334,42 @@ TEST_F(DBTest, ReadonlyDBGetLiveManifestSize) {
    Close();
  } while (ChangeCompactOptions());
 }
+
+TEST_F(DBTest, GetLiveBlobFiles) {
+  VersionSet* const versions = dbfull()->TEST_GetVersionSet();
+  assert(versions);
+  assert(versions->GetColumnFamilySet());
+
+  ColumnFamilyData* const cfd = versions->GetColumnFamilySet()->GetDefault();
+  assert(cfd);
+
+  // Add a live blob file.
+  VersionEdit edit;
+
+  constexpr uint64_t blob_file_number = 234;
+  constexpr uint64_t total_blob_count = 555;
+  constexpr uint64_t total_blob_bytes = 66666;
+  constexpr char checksum_method[] = "CRC32";
+  constexpr char checksum_value[] = "3d87ff57";
+
+  edit.AddBlobFile(blob_file_number, total_blob_count, total_blob_bytes,
+                   checksum_method, checksum_value);
+
+  dbfull()->TEST_LockMutex();
+  Status s = versions->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
+                                   &edit, dbfull()->mutex());
+  dbfull()->TEST_UnlockMutex();
+
+  ASSERT_OK(s);
+
+  // Make sure it appears in the results returned by GetLiveFiles.
+  uint64_t manifest_size = 0;
+  std::vector<std::string> files;
+  ASSERT_OK(dbfull()->GetLiveFiles(files, &manifest_size));
+
+  ASSERT_FALSE(files.empty());
+  ASSERT_EQ(files[0], BlobFileName("", blob_file_number));
+}
 #endif

 TEST_F(DBTest, PurgeInfoLogs) {
@@ -4530,7 +4591,7 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel) {
  for (int i = 0; i < kNKeys; i++) {
    keys[i] = i;
  }
-  std::random_shuffle(std::begin(keys), std::end(keys));
+  RandomShuffle(std::begin(keys), std::end(keys));

  Random rnd(301);
  Options options;
@@ -4613,7 +4674,7 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
  for (int i = 0; i < kNKeys; i++) {
    keys[i] = i;
  }
-  std::random_shuffle(std::begin(keys), std::end(keys));
+  RandomShuffle(std::begin(keys), std::end(keys));

  Random rnd(301);
  Options options;
@@ -25,6 +25,67 @@ class DBTest2 : public DBTestBase {
  DBTest2() : DBTestBase("/db_test2") {}
 };

+TEST_F(DBTest2, OpenForReadOnly) {
+  DB* db_ptr = nullptr;
+  std::string dbname = test::PerThreadDBPath("db_readonly");
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  // OpenForReadOnly should fail but will create <dbname> in the file system
+  ASSERT_NOK(DB::OpenForReadOnly(options, dbname, &db_ptr));
+  // Since <dbname> is created, we should be able to delete the dir
+  // We first get the list files under <dbname>
+  // There should not be any subdirectories -- this is not checked here
+  std::vector<std::string> files;
+  ASSERT_OK(env_->GetChildren(dbname, &files));
+  for (auto& f : files) {
+    if (f != "." && f != "..") {
+      ASSERT_OK(env_->DeleteFile(dbname + "/" + f));
+    }
+  }
+  // <dbname> should be empty now and we should be able to delete it
+  ASSERT_OK(env_->DeleteDir(dbname));
+  options.create_if_missing = false;
+  // OpenForReadOnly should fail since <dbname> was successfully deleted
+  ASSERT_NOK(DB::OpenForReadOnly(options, dbname, &db_ptr));
+  // With create_if_missing false, there should not be a dir in the file system
+  ASSERT_NOK(env_->FileExists(dbname));
+}
+
+TEST_F(DBTest2, OpenForReadOnlyWithColumnFamilies) {
+  DB* db_ptr = nullptr;
+  std::string dbname = test::PerThreadDBPath("db_readonly");
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+
+  ColumnFamilyOptions cf_options(options);
+  std::vector<ColumnFamilyDescriptor> column_families;
+  column_families.push_back(
+      ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
+  column_families.push_back(ColumnFamilyDescriptor("goku", cf_options));
+  std::vector<ColumnFamilyHandle*> handles;
+  // OpenForReadOnly should fail but will create <dbname> in the file system
+  ASSERT_NOK(
+      DB::OpenForReadOnly(options, dbname, column_families, &handles, &db_ptr));
+  // Since <dbname> is created, we should be able to delete the dir
+  // We first get the list files under <dbname>
+  // There should not be any subdirectories -- this is not checked here
+  std::vector<std::string> files;
+  ASSERT_OK(env_->GetChildren(dbname, &files));
+  for (auto& f : files) {
+    if (f != "." && f != "..") {
+      ASSERT_OK(env_->DeleteFile(dbname + "/" + f));
+    }
+  }
+  // <dbname> should be empty now and we should be able to delete it
+  ASSERT_OK(env_->DeleteDir(dbname));
+  options.create_if_missing = false;
+  // OpenForReadOnly should fail since <dbname> was successfully deleted
+  ASSERT_NOK(
+      DB::OpenForReadOnly(options, dbname, column_families, &handles, &db_ptr));
+  // With create_if_missing false, there should not be a dir in the file system
+  ASSERT_NOK(env_->FileExists(dbname));
+}
+
 class PrefixFullBloomWithReverseComparator
    : public DBTestBase,
      public ::testing::WithParamInterface<bool> {
@@ -1268,6 +1329,145 @@ class CompactionCompressionListener : public EventListener {
  const Options* db_options_;
 };

+TEST_F(DBTest2, CompressionFailures) {
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = 2;
+  options.max_bytes_for_level_base = 1024;
+  options.max_bytes_for_level_multiplier = 2;
+  options.num_levels = 7;
+  options.max_background_compactions = 1;
+  options.target_file_size_base = 512;
+
+  BlockBasedTableOptions table_options;
+  table_options.block_size = 512;
+  table_options.verify_compression = true;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+
+  enum CompressionFailureType {
+    kTestCompressionFail,
+    kTestDecompressionFail,
+    kTestDecompressionCorruption
+  } curr_compression_failure_type;
+  std::vector<CompressionFailureType> compression_failure_types = {
+      kTestCompressionFail, kTestDecompressionFail,
+      kTestDecompressionCorruption};
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "BlockBasedTableBuilder::CompressBlockInternal:TamperWithReturnValue",
+      [&curr_compression_failure_type](void* arg) {
+        bool* ret = static_cast<bool*>(arg);
+        if (curr_compression_failure_type == kTestCompressionFail) {
+          *ret = false;
+        }
+      });
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "UncompressBlockContentsForCompressionType:TamperWithReturnValue",
+      [&curr_compression_failure_type](void* arg) {
+        Status* ret = static_cast<Status*>(arg);
+        ASSERT_OK(*ret);
+        if (curr_compression_failure_type == kTestDecompressionFail) {
+          *ret = Status::Corruption("kTestDecompressionFail");
+        }
+      });
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "UncompressBlockContentsForCompressionType:"
+      "TamperWithDecompressionOutput",
+      [&curr_compression_failure_type](void* arg) {
+        if (curr_compression_failure_type == kTestDecompressionCorruption) {
+          BlockContents* contents = static_cast<BlockContents*>(arg);
+          // Ensure uncompressed data != original data
+          const size_t len = contents->data.size() + 1;
+          std::unique_ptr<char[]> fake_data(new char[len]());
+          *contents = BlockContents(std::move(fake_data), len);
+        }
+      });
+
+  std::vector<CompressionType> compression_types = GetSupportedCompressions();
+  std::vector<uint32_t> compression_max_dict_bytes = {0, 10};
+  std::vector<uint32_t> compression_parallel_threads = {1, 4};
+
+  std::map<std::string, std::string> key_value_written;
+
+  const int kKeySize = 5;
+  const int kValUnitSize = 16;
+  const int kValSize = 256;
+  Random rnd(405);
+
+  Status s = Status::OK();
+
+  for (auto compression_failure_type : compression_failure_types) {
+    curr_compression_failure_type = compression_failure_type;
+    for (auto compression_type : compression_types) {
+      if (compression_type == kNoCompression) {
+        continue;
+      }
+      for (auto parallel_threads : compression_parallel_threads) {
+        for (auto max_dict_bytes : compression_max_dict_bytes) {
+          options.compression = compression_type;
+          options.compression_opts.parallel_threads = parallel_threads;
+          options.compression_opts.max_dict_bytes = max_dict_bytes;
+          options.bottommost_compression_opts.parallel_threads =
+              parallel_threads;
+          options.bottommost_compression_opts.max_dict_bytes = max_dict_bytes;
+
+          DestroyAndReopen(options);
+          // Write 10 random files
+          for (int i = 0; i < 10; i++) {
+            for (int j = 0; j < 5; j++) {
+              std::string key = RandomString(&rnd, kKeySize);
+              // Ensure good compression ratio
+              std::string valueUnit = RandomString(&rnd, kValUnitSize);
+              std::string value;
+              for (int k = 0; k < kValSize; k += kValUnitSize) {
+                value += valueUnit;
+              }
+              s = Put(key, value);
+              if (compression_failure_type == kTestCompressionFail) {
+                key_value_written[key] = value;
+                ASSERT_OK(s);
+              }
+            }
+            s = Flush();
+            if (compression_failure_type == kTestCompressionFail) {
+              ASSERT_OK(s);
+            }
+            s = dbfull()->TEST_WaitForCompact();
+            if (compression_failure_type == kTestCompressionFail) {
+              ASSERT_OK(s);
+            }
+            if (i == 4) {
+              // Make compression fail at the mid of table building
+              ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+            }
+          }
+          ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+
+          if (compression_failure_type == kTestCompressionFail) {
+            // Should be kNoCompression, check content consistency
+            std::unique_ptr<Iterator> db_iter(db_->NewIterator(ReadOptions()));
+            for (db_iter->SeekToFirst(); db_iter->Valid(); db_iter->Next()) {
+              std::string key = db_iter->key().ToString();
+              std::string value = db_iter->value().ToString();
+              ASSERT_NE(key_value_written.find(key), key_value_written.end());
+              ASSERT_EQ(key_value_written[key], value);
+              key_value_written.erase(key);
+            }
+            ASSERT_EQ(0, key_value_written.size());
+          } else if (compression_failure_type == kTestDecompressionFail) {
+            ASSERT_EQ(std::string(s.getState()),
+                      "Could not decompress: kTestDecompressionFail");
+          } else if (compression_failure_type == kTestDecompressionCorruption) {
+            ASSERT_EQ(std::string(s.getState()),
+                      "Decompressed block did not match raw block");
+          }
+        }
+      }
+    }
+  }
+}
+
 TEST_F(DBTest2, CompressionOptions) {
  if (!Zlib_Supported() || !Snappy_Supported()) {
    return;
@@ -1288,6 +1488,10 @@ TEST_F(DBTest2, CompressionOptions) {
  const int kValSize = 20;
  Random rnd(301);

+  std::vector<uint32_t> compression_parallel_threads = {1, 4};
+
+  std::map<std::string, std::string> key_value_written;
+
  for (int iter = 0; iter <= 2; iter++) {
    listener->max_level_checked = 0;

@@ -1312,19 +1516,37 @@ TEST_F(DBTest2, CompressionOptions) {
      options.bottommost_compression = kDisableCompressionOption;
    }

-    DestroyAndReopen(options);
-    // Write 10 random files
-    for (int i = 0; i < 10; i++) {
-      for (int j = 0; j < 5; j++) {
-        ASSERT_OK(
-            Put(RandomString(&rnd, kKeySize), RandomString(&rnd, kValSize)));
-      }
-      ASSERT_OK(Flush());
-      dbfull()->TEST_WaitForCompact();
-    }
+    for (auto num_threads : compression_parallel_threads) {
+      options.compression_opts.parallel_threads = num_threads;
+      options.bottommost_compression_opts.parallel_threads = num_threads;

-    // Make sure that we wrote enough to check all 7 levels
-    ASSERT_EQ(listener->max_level_checked, 6);
+      DestroyAndReopen(options);
+      // Write 10 random files
+      for (int i = 0; i < 10; i++) {
+        for (int j = 0; j < 5; j++) {
+          std::string key = RandomString(&rnd, kKeySize);
+          std::string value = RandomString(&rnd, kValSize);
+          key_value_written[key] = value;
+          ASSERT_OK(Put(key, value));
+        }
+        ASSERT_OK(Flush());
+        dbfull()->TEST_WaitForCompact();
+      }
+
+      // Make sure that we wrote enough to check all 7 levels
+      ASSERT_EQ(listener->max_level_checked, 6);
+
+      // Make sure database content is the same as key_value_written
+      std::unique_ptr<Iterator> db_iter(db_->NewIterator(ReadOptions()));
+      for (db_iter->SeekToFirst(); db_iter->Valid(); db_iter->Next()) {
+        std::string key = db_iter->key().ToString();
+        std::string value = db_iter->value().ToString();
+        ASSERT_NE(key_value_written.find(key), key_value_written.end());
+        ASSERT_EQ(key_value_written[key], value);
+        key_value_written.erase(key);
+      }
+      ASSERT_EQ(0, key_value_written.size());
+    }
  }
 }

@@ -4306,6 +4528,24 @@ TEST_F(DBTest2, SameSmallestInSameLevel) {
  ASSERT_EQ("2,3,4,5,6,7,8", Get("key"));
 }

+TEST_F(DBTest2, FileConsistencyCheckInOpen) {
+  Put("foo", "bar");
+  Flush();
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
+        Status* ret_s = static_cast<Status*>(arg);
+        *ret_s = Status::Corruption("fcc");
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options = CurrentOptions();
+  options.force_consistency_checks = true;
+  ASSERT_NOK(TryReopen(options));
+
+  SyncPoint::GetInstance()->DisableProcessing();
+}
+
 TEST_F(DBTest2, BlockBasedTablePrefixIndexSeekForPrev) {
  // create a DB with block prefix index
  BlockBasedTableOptions table_options;
@@ -14,10 +14,19 @@

 namespace ROCKSDB_NAMESPACE {

+namespace {
+int64_t MaybeCurrentTime(Env* env) {
+  int64_t time = 1337346000;  // arbitrary fallback default
+  (void)env->GetCurrentTime(&time);
+  return time;
+}
+}  // namespace
+
 // Special Env used to delay background operations

 SpecialEnv::SpecialEnv(Env* base)
    : EnvWrapper(base),
+      maybe_starting_time_(MaybeCurrentTime(base)),
      rnd_(301),
      sleep_counter_(this),
      addon_time_(0),
@@ -399,20 +408,7 @@ Options DBTestBase::GetOptions(
        options.use_direct_reads = true;
        options.use_direct_io_for_flush_and_compaction = true;
        options.compaction_readahead_size = 2 * 1024 * 1024;
-  #if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
-      !defined(OS_AIX) && !defined(OS_OPENBSD)
-        ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-            "NewWritableFile:O_DIRECT", [&](void* arg) {
-              int* val = static_cast<int*>(arg);
-              *val &= ~O_DIRECT;
-            });
-        ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-            "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
-              int* val = static_cast<int*>(arg);
-              *val &= ~O_DIRECT;
-            });
-        ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
-#endif
+        test::SetupSyncPointsToMockDirectIO();
        break;
      }
 #endif  // ROCKSDB_LITE
@@ -502,10 +502,14 @@ class SpecialEnv : public EnvWrapper {

  virtual Status GetCurrentTime(int64_t* unix_time) override {
    Status s;
-    if (!time_elapse_only_sleep_) {
+    if (time_elapse_only_sleep_) {
+      *unix_time = maybe_starting_time_;
+    } else {
      s = target()->GetCurrentTime(unix_time);
    }
    if (s.ok()) {
+      // FIXME: addon_time_ sometimes used to mean seconds (here) and
+      // sometimes microseconds
      *unix_time += addon_time_.load();
    }
    return s;
@@ -531,6 +535,20 @@ class SpecialEnv : public EnvWrapper {
    return target()->DeleteFile(fname);
  }

+  void SetTimeElapseOnlySleep(Options* options) {
+    time_elapse_only_sleep_ = true;
+    no_slowdown_ = true;
+    // Need to disable stats dumping and persisting which also use
+    // RepeatableThread, which uses InstrumentedCondVar::TimedWaitInternal.
+    // With time_elapse_only_sleep_, this can hang on some platforms.
+    // TODO: why? investigate/fix
+    options->stats_dump_period_sec = 0;
+    options->stats_persist_period_sec = 0;
+  }
+
+  // Something to return when mocking current time
+  const int64_t maybe_starting_time_;
+
  Random rnd_;
  port::Mutex rnd_mutex_;  // Lock to pretect rnd_

@@ -647,6 +665,72 @@ class TestPutOperator : public MergeOperator {
  virtual const char* Name() const override { return "TestPutOperator"; }
 };

+// A wrapper around Cache that can easily be extended with instrumentation,
+// etc.
+class CacheWrapper : public Cache {
+ public:
+  explicit CacheWrapper(std::shared_ptr<Cache> target)
+      : target_(std::move(target)) {}
+
+  const char* Name() const override { return target_->Name(); }
+
+  Status Insert(const Slice& key, void* value, size_t charge,
+                void (*deleter)(const Slice& key, void* value),
+                Handle** handle = nullptr,
+                Priority priority = Priority::LOW) override {
+    return target_->Insert(key, value, charge, deleter, handle, priority);
+  }
+
+  Handle* Lookup(const Slice& key, Statistics* stats = nullptr) override {
+    return target_->Lookup(key, stats);
+  }
+
+  bool Ref(Handle* handle) override { return target_->Ref(handle); }
+
+  bool Release(Handle* handle, bool force_erase = false) override {
+    return target_->Release(handle, force_erase);
+  }
+
+  void* Value(Handle* handle) override { return target_->Value(handle); }
+
+  void Erase(const Slice& key) override { target_->Erase(key); }
+  uint64_t NewId() override { return target_->NewId(); }
+
+  void SetCapacity(size_t capacity) override { target_->SetCapacity(capacity); }
+
+  void SetStrictCapacityLimit(bool strict_capacity_limit) override {
+    target_->SetStrictCapacityLimit(strict_capacity_limit);
+  }
+
+  bool HasStrictCapacityLimit() const override {
+    return target_->HasStrictCapacityLimit();
+  }
+
+  size_t GetCapacity() const override { return target_->GetCapacity(); }
+
+  size_t GetUsage() const override { return target_->GetUsage(); }
+
+  size_t GetUsage(Handle* handle) const override {
+    return target_->GetUsage(handle);
+  }
+
+  size_t GetPinnedUsage() const override { return target_->GetPinnedUsage(); }
+
+  size_t GetCharge(Handle* handle) const override {
+    return target_->GetCharge(handle);
+  }
+
+  void ApplyToAllCacheEntries(void (*callback)(void*, size_t),
+                              bool thread_safe) override {
+    target_->ApplyToAllCacheEntries(callback, thread_safe);
+  }
+
+  void EraseUnRefEntries() override { target_->EraseUnRefEntries(); }
+
+ protected:
+  std::shared_ptr<Cache> target_;
+};
+
 class DBTestBase : public testing::Test {
 public:
  // Sequence of option configurations to try
@@ -1953,6 +1953,7 @@ TEST_F(DBTestUniversalCompaction2, BasicL0toL1) {
  ASSERT_GT(NumTableFilesAtLevel(6), 0);
 }

+#if defined(ENABLE_SINGLE_LEVEL_DTC)
 TEST_F(DBTestUniversalCompaction2, SingleLevel) {
  const int kNumKeys = 3000;
  const int kWindowSize = 100;
@@ -1991,6 +1992,7 @@ TEST_F(DBTestUniversalCompaction2, SingleLevel) {
  dbfull()->TEST_WaitForCompact();
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
 }
+#endif  // ENABLE_SINGLE_LEVEL_DTC

 TEST_F(DBTestUniversalCompaction2, MultipleLevels) {
  const int kWindowSize = 100;
@@ -1422,8 +1422,11 @@ TEST_F(DBWALTest, RecoverFromCorruptedWALWithoutFlush) {
    delete iter;
    return data;
  };
-  for (auto& mode : wal_recovery_mode_string_map) {
-    options.wal_recovery_mode = mode.second;
+  for (auto& mode : {WALRecoveryMode::kTolerateCorruptedTailRecords,
+                     WALRecoveryMode::kAbsoluteConsistency,
+                     WALRecoveryMode::kPointInTimeRecovery,
+                     WALRecoveryMode::kSkipAnyCorruptedRecords}) {
+    options.wal_recovery_mode = mode;
    for (auto trunc : {true, false}) {
      for (int i = 0; i < 4; i++) {
        for (int j = jstart; j < jend; j++) {
@@ -26,22 +26,8 @@ class DBBasicTestWithTimestampBase : public DBTestBase {

 protected:
  static std::string Key1(uint64_t k) {
-    uint32_t x = 1;
-    const bool is_little_endian = (*reinterpret_cast<char*>(&x) != 0);
    std::string ret;
-    if (is_little_endian) {
-      ret.assign(reinterpret_cast<char*>(&k), sizeof(k));
-    } else {
-      ret.resize(sizeof(k));
-      ret[0] = k & 0xff;
-      ret[1] = (k >> 8) & 0xff;
-      ret[2] = (k >> 16) & 0xff;
-      ret[3] = (k >> 24) & 0xff;
-      ret[4] = (k >> 32) & 0xff;
-      ret[5] = (k >> 40) & 0xff;
-      ret[6] = (k >> 48) & 0xff;
-      ret[7] = (k >> 56) & 0xff;
-    }
+    PutFixed64(&ret, k);
    std::reverse(ret.begin(), ret.end());
    return ret;
  }
@@ -173,8 +159,6 @@ TEST_F(DBBasicTestWithTimestamp, SimpleForwardIterate) {
  const uint64_t kMaxKey = 1024;
  Options options = CurrentOptions();
  options.env = env_;
-  // TODO(yanqin) re-enable auto compaction
-  options.disable_auto_compactions = true;
  options.create_if_missing = true;
  const size_t kTimestampSize = Timestamp(0, 0).size();
  TestComparator test_cmp(kTimestampSize);
@@ -235,6 +219,56 @@ TEST_F(DBBasicTestWithTimestamp, SimpleForwardIterate) {
  Close();
 }

+TEST_F(DBBasicTestWithTimestamp, SimpleForwardIterateLowerTsBound) {
+  const int kNumKeysPerFile = 128;
+  const uint64_t kMaxKey = 1024;
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  const size_t kTimestampSize = Timestamp(0, 0).size();
+  TestComparator test_cmp(kTimestampSize);
+  options.comparator = &test_cmp;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  DestroyAndReopen(options);
+  const std::vector<std::string> write_timestamps = {Timestamp(1, 0),
+                                                     Timestamp(3, 0)};
+  const std::vector<std::string> read_timestamps = {Timestamp(2, 0),
+                                                    Timestamp(4, 0)};
+  const std::vector<std::string> read_timestamps_lb = {Timestamp(1, 0),
+                                                       Timestamp(1, 0)};
+  for (size_t i = 0; i < write_timestamps.size(); ++i) {
+    WriteOptions write_opts;
+    Slice write_ts = write_timestamps[i];
+    write_opts.timestamp = &write_ts;
+    for (uint64_t key = 0; key <= kMaxKey; ++key) {
+      Status s = db_->Put(write_opts, Key1(key), "value" + std::to_string(i));
+      ASSERT_OK(s);
+    }
+  }
+  for (size_t i = 0; i < read_timestamps.size(); ++i) {
+    ReadOptions read_opts;
+    Slice read_ts = read_timestamps[i];
+    Slice read_ts_lb = read_timestamps_lb[i];
+    read_opts.timestamp = &read_ts;
+    read_opts.iter_start_ts = &read_ts_lb;
+    std::unique_ptr<Iterator> it(db_->NewIterator(read_opts));
+    int count = 0;
+    uint64_t key = 0;
+    for (it->Seek(Key1(0)), key = 0; it->Valid(); it->Next(), ++count, ++key) {
+      CheckIterUserEntry(it.get(), Key1(key), "value" + std::to_string(i),
+                         write_timestamps[i]);
+      if (i > 0) {
+        it->Next();
+        CheckIterUserEntry(it.get(), Key1(key), "value" + std::to_string(i - 1),
+                           write_timestamps[i - 1]);
+      }
+    }
+    size_t expected_count = kMaxKey + 1;
+    ASSERT_EQ(expected_count, count);
+  }
+  Close();
+}
+
 TEST_F(DBBasicTestWithTimestamp, ForwardIterateStartSeqnum) {
  const int kNumKeysPerFile = 128;
  const uint64_t kMaxKey = 0xffffffffffffffff;
@@ -242,7 +276,9 @@ TEST_F(DBBasicTestWithTimestamp, ForwardIterateStartSeqnum) {
  Options options = CurrentOptions();
  options.env = env_;
  options.create_if_missing = true;
-  // TODO(yanqin) re-enable auto compaction
+  // Need to disable compaction to bottommost level when sequence number will be
+  // zeroed out, causing the verification of sequence number to fail in this
+  // test.
  options.disable_auto_compactions = true;
  const size_t kTimestampSize = Timestamp(0, 0).size();
  TestComparator test_cmp(kTimestampSize);
@@ -300,8 +336,6 @@ TEST_F(DBBasicTestWithTimestamp, ReseekToTargetTimestamp) {
  constexpr size_t kNumKeys = 16;
  options.max_sequential_skip_in_iterations = kNumKeys / 2;
  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
-  // TODO(yanqin) re-enable auto compaction
-  options.disable_auto_compactions = true;
  const size_t kTimestampSize = Timestamp(0, 0).size();
  TestComparator test_cmp(kTimestampSize);
  options.comparator = &test_cmp;
@@ -337,8 +371,6 @@ TEST_F(DBBasicTestWithTimestamp, ReseekToNextUserKey) {
  constexpr size_t kNumKeys = 16;
  options.max_sequential_skip_in_iterations = kNumKeys / 2;
  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
-  // TODO(yanqin) re-enable auto compaction
-  options.disable_auto_compactions = true;
  const size_t kTimestampSize = Timestamp(0, 0).size();
  TestComparator test_cmp(kTimestampSize);
  options.comparator = &test_cmp;
@@ -417,10 +449,70 @@ TEST_F(DBBasicTestWithTimestamp, MaxKeysSkipped) {
  Close();
 }

+// Create two L0, and compact them to a new L1. In this test, L1 is L_bottom.
+// Two L0s:
+//       f1                                  f2
+// <a, 1, kTypeValue>    <a, 3, kTypeDeletionWithTimestamp>...<b, 2, kTypeValue>
+// Since f2.smallest < f1.largest < f2.largest
+// f1 and f2 will be the inputs of a real compaction instead of trivial move.
+TEST_F(DBBasicTestWithTimestamp, CompactDeletionWithTimestampMarkerToBottom) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  const size_t kTimestampSize = Timestamp(0, 0).size();
+  TestComparator test_cmp(kTimestampSize);
+  options.comparator = &test_cmp;
+  options.num_levels = 2;
+  options.level0_file_num_compaction_trigger = 2;
+  DestroyAndReopen(options);
+  WriteOptions write_opts;
+  std::string ts_str = Timestamp(1, 0);
+  Slice ts = ts_str;
+  write_opts.timestamp = &ts;
+  ASSERT_OK(db_->Put(write_opts, "a", "value0"));
+  ASSERT_OK(Flush());
+
+  ts_str = Timestamp(2, 0);
+  ts = ts_str;
+  write_opts.timestamp = &ts;
+  ASSERT_OK(db_->Put(write_opts, "b", "value0"));
+  ts_str = Timestamp(3, 0);
+  ts = ts_str;
+  write_opts.timestamp = &ts;
+  ASSERT_OK(db_->Delete(write_opts, "a"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  ReadOptions read_opts;
+  ts_str = Timestamp(1, 0);
+  ts = ts_str;
+  read_opts.timestamp = &ts;
+  std::string value;
+  Status s = db_->Get(read_opts, "a", &value);
+  ASSERT_OK(s);
+  ASSERT_EQ("value0", value);
+
+  ts_str = Timestamp(3, 0);
+  ts = ts_str;
+  read_opts.timestamp = &ts;
+  s = db_->Get(read_opts, "a", &value);
+  ASSERT_TRUE(s.IsNotFound());
+
+  // Time-travel to the past before deletion
+  ts_str = Timestamp(2, 0);
+  ts = ts_str;
+  read_opts.timestamp = &ts;
+  s = db_->Get(read_opts, "a", &value);
+  ASSERT_OK(s);
+  ASSERT_EQ("value0", value);
+  Close();
+}
+
 class DBBasicTestWithTimestampCompressionSettings
    : public DBBasicTestWithTimestampBase,
-      public testing::WithParamInterface<std::tuple<
-          std::shared_ptr<const FilterPolicy>, CompressionType, uint32_t>> {
+      public testing::WithParamInterface<
+          std::tuple<std::shared_ptr<const FilterPolicy>, CompressionType,
+                     uint32_t, uint32_t>> {
 public:
  DBBasicTestWithTimestampCompressionSettings()
      : DBBasicTestWithTimestampBase(
@@ -460,6 +552,7 @@ TEST_P(DBBasicTestWithTimestampCompressionSettings, PutAndGet) {
  if (comp_type == kZSTD) {
    options.compression_opts.zstd_max_train_bytes = std::get<2>(GetParam());
  }
+  options.compression_opts.parallel_threads = std::get<3>(GetParam());
  options.target_file_size_base = 1 << 26;  // 64MB
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);
@@ -502,6 +595,104 @@ TEST_P(DBBasicTestWithTimestampCompressionSettings, PutAndGet) {
  Close();
 }

+TEST_P(DBBasicTestWithTimestampCompressionSettings, PutDeleteGet) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  const size_t kTimestampSize = Timestamp(0, 0).size();
+  TestComparator test_cmp(kTimestampSize);
+  options.comparator = &test_cmp;
+  const int kNumKeysPerFile = 1024;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  BlockBasedTableOptions bbto;
+  bbto.filter_policy = std::get<0>(GetParam());
+  bbto.whole_key_filtering = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+
+  const CompressionType comp_type = std::get<1>(GetParam());
+#if LZ4_VERSION_NUMBER < 10400  // r124+
+  if (comp_type == kLZ4Compression || comp_type == kLZ4HCCompression) {
+    return;
+  }
+#endif  // LZ4_VERSION_NUMBER >= 10400
+  if (!ZSTD_Supported() && comp_type == kZSTD) {
+    return;
+  }
+  if (!Zlib_Supported() && comp_type == kZlibCompression) {
+    return;
+  }
+
+  options.compression = comp_type;
+  options.compression_opts.max_dict_bytes = std::get<2>(GetParam());
+  if (comp_type == kZSTD) {
+    options.compression_opts.zstd_max_train_bytes = std::get<2>(GetParam());
+  }
+  options.compression_opts.parallel_threads = std::get<3>(GetParam());
+  options.target_file_size_base = 1 << 26;  // 64MB
+
+  DestroyAndReopen(options);
+
+  const size_t kNumL0Files =
+      static_cast<size_t>(Options().level0_file_num_compaction_trigger);
+  {
+    // Generate enough L0 files with ts=1 to trigger compaction to L1
+    std::string ts_str = Timestamp(1, 0);
+    Slice ts = ts_str;
+    WriteOptions wopts;
+    wopts.timestamp = &ts;
+    for (size_t i = 0; i != kNumL0Files; ++i) {
+      for (int j = 0; j != kNumKeysPerFile; ++j) {
+        ASSERT_OK(db_->Put(wopts, Key1(j), "value" + std::to_string(i)));
+      }
+      ASSERT_OK(db_->Flush(FlushOptions()));
+    }
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
+    // Generate another L0 at ts=3
+    ts_str = Timestamp(3, 0);
+    ts = ts_str;
+    wopts.timestamp = &ts;
+    for (int i = 0; i != kNumKeysPerFile; ++i) {
+      std::string key_str = Key1(i);
+      Slice key(key_str);
+      if ((i % 3) == 0) {
+        ASSERT_OK(db_->Delete(wopts, key));
+      } else {
+        ASSERT_OK(db_->Put(wopts, key, "new_value"));
+      }
+    }
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    // Populate memtable at ts=5
+    ts_str = Timestamp(5, 0);
+    ts = ts_str;
+    wopts.timestamp = &ts;
+    for (int i = 0; i != kNumKeysPerFile; ++i) {
+      std::string key_str = Key1(i);
+      Slice key(key_str);
+      if ((i % 3) == 1) {
+        ASSERT_OK(db_->Delete(wopts, key));
+      } else if ((i % 3) == 2) {
+        ASSERT_OK(db_->Put(wopts, key, "new_value_2"));
+      }
+    }
+  }
+  {
+    std::string ts_str = Timestamp(6, 0);
+    Slice ts = ts_str;
+    ReadOptions ropts;
+    ropts.timestamp = &ts;
+    for (uint64_t i = 0; i != static_cast<uint64_t>(kNumKeysPerFile); ++i) {
+      std::string value;
+      Status s = db_->Get(ropts, Key1(i), &value);
+      if ((i % 3) == 2) {
+        ASSERT_OK(s);
+        ASSERT_EQ("new_value_2", value);
+      } else {
+        ASSERT_TRUE(s.IsNotFound());
+      }
+    }
+  }
+}
+
 #ifndef ROCKSDB_LITE
 // A class which remembers the name of each flushed file.
 class FlushedFileCollector : public EventListener {
@@ -572,6 +763,7 @@ TEST_P(DBBasicTestWithTimestampCompressionSettings, PutAndGetWithCompaction) {
  if (comp_type == kZSTD) {
    options.compression_opts.zstd_max_train_bytes = std::get<2>(GetParam());
  }
+  options.compression_opts.parallel_threads = std::get<3>(GetParam());
  DestroyAndReopen(options);
  CreateAndReopenWithCF({"pikachu"}, options);

@@ -581,7 +773,7 @@ TEST_P(DBBasicTestWithTimestampCompressionSettings, PutAndGetWithCompaction) {
  std::vector<std::string> read_ts_list;

  const auto& verify_records_func = [&](size_t i, size_t begin, size_t end,
-                                       ColumnFamilyHandle* cfh) {
+                                        ColumnFamilyHandle* cfh) {
    std::string value;
    std::string timestamp;

@@ -619,9 +811,9 @@ TEST_P(DBBasicTestWithTimestampCompressionSettings, PutAndGetWithCompaction) {
          // higherlevel[0].largest.userkey
          ASSERT_OK(Flush(cf));

-          // compact files (2 at each level) to a lower level such that all keys
-          // with the same timestamp is at one level, with newer versions at
-          // higher levels.
+          // compact files (2 at each level) to a lower level such that all
+          // keys with the same timestamp is at one level, with newer versions
+          // at higher levels.
          CompactionOptions compact_opt;
          compact_opt.compression = kNoCompression;
          db_->CompactFiles(compact_opt, handles_[cf],
@@ -739,6 +931,55 @@ TEST_F(DBBasicTestWithTimestamp, BatchWriteAndMultiGet) {
  Close();
 }

+TEST_F(DBBasicTestWithTimestamp, MultiGetNoReturnTs) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  const size_t kTimestampSize = Timestamp(0, 0).size();
+  TestComparator test_cmp(kTimestampSize);
+  options.comparator = &test_cmp;
+  DestroyAndReopen(options);
+  WriteOptions write_opts;
+  std::string ts_str = Timestamp(1, 0);
+  Slice ts = ts_str;
+  write_opts.timestamp = &ts;
+  ASSERT_OK(db_->Put(write_opts, "foo", "value"));
+  ASSERT_OK(db_->Put(write_opts, "bar", "value"));
+  ASSERT_OK(db_->Put(write_opts, "fooxxxxxxxxxxxxxxxx", "value"));
+  ASSERT_OK(db_->Put(write_opts, "barxxxxxxxxxxxxxxxx", "value"));
+  ColumnFamilyHandle* cfh = dbfull()->DefaultColumnFamily();
+  ts_str = Timestamp(2, 0);
+  ts = ts_str;
+  ReadOptions read_opts;
+  read_opts.timestamp = &ts;
+  {
+    ColumnFamilyHandle* column_families[] = {cfh, cfh};
+    Slice keys[] = {"foo", "bar"};
+    PinnableSlice values[] = {PinnableSlice(), PinnableSlice()};
+    Status statuses[] = {Status::OK(), Status::OK()};
+    dbfull()->MultiGet(read_opts, /*num_keys=*/2, &column_families[0], &keys[0],
+                       &values[0], &statuses[0], /*sorted_input=*/false);
+    for (const auto& s : statuses) {
+      ASSERT_OK(s);
+    }
+  }
+  {
+    ColumnFamilyHandle* column_families[] = {cfh, cfh, cfh, cfh};
+    // Make user keys longer than configured timestamp size (16 bytes) to
+    // verify RocksDB does not use the trailing bytes 'x' as timestamp.
+    Slice keys[] = {"fooxxxxxxxxxxxxxxxx", "barxxxxxxxxxxxxxxxx", "foo", "bar"};
+    PinnableSlice values[] = {PinnableSlice(), PinnableSlice(), PinnableSlice(),
+                              PinnableSlice()};
+    Status statuses[] = {Status::OK(), Status::OK(), Status::OK(),
+                         Status::OK()};
+    dbfull()->MultiGet(read_opts, /*num_keys=*/4, &column_families[0], &keys[0],
+                       &values[0], &statuses[0], /*sorted_input=*/false);
+    for (const auto& s : statuses) {
+      ASSERT_OK(s);
+    }
+  }
+  Close();
+}
+
 #endif  // !ROCKSDB_LITE

 INSTANTIATE_TEST_CASE_P(
@@ -749,7 +990,7 @@ INSTANTIATE_TEST_CASE_P(
                              NewBloomFilterPolicy(10, false))),
        ::testing::Values(kNoCompression, kZlibCompression, kLZ4Compression,
                          kLZ4HCCompression, kZSTD),
-        ::testing::Values(0, 1 << 14)));
+        ::testing::Values(0, 1 << 14), ::testing::Values(1, 4)));

 class DBBasicTestWithTimestampPrefixSeek
    : public DBBasicTestWithTimestampBase,
@@ -767,8 +1008,6 @@ TEST_P(DBBasicTestWithTimestampPrefixSeek, ForwardIterateWithPrefix) {
  Options options = CurrentOptions();
  options.env = env_;
  options.create_if_missing = true;
-  // TODO(yanqin): re-enable auto compactions
-  options.disable_auto_compactions = true;
  const size_t kTimestampSize = Timestamp(0, 0).size();
  TestComparator test_cmp(kTimestampSize);
  options.comparator = &test_cmp;
@@ -788,15 +1027,10 @@ TEST_P(DBBasicTestWithTimestampPrefixSeek, ForwardIterateWithPrefix) {
    for (size_t i = 0; i != write_ts_list.size(); ++i) {
      Slice write_ts = write_ts_list[i];
      write_opts.timestamp = &write_ts;
-      uint64_t key = kMinKey;
-      do {
+      for (uint64_t key = kMaxKey; key >= kMinKey; --key) {
        Status s = db_->Put(write_opts, Key1(key), "value" + std::to_string(i));
        ASSERT_OK(s);
-        if (key == kMaxKey) {
-          break;
-        }
-        ++key;
-      } while (true);
+      }
    }
  }
  const std::vector<std::string> read_ts_list = {Timestamp(5, 0xffffffff),
@@ -880,6 +1114,90 @@ INSTANTIATE_TEST_CASE_P(
                                                   false))),
        ::testing::Bool()));

+class DBBasicTestWithTsIterTombstones
+    : public DBBasicTestWithTimestampBase,
+      public testing::WithParamInterface<
+          std::tuple<std::shared_ptr<const SliceTransform>,
+                     std::shared_ptr<const FilterPolicy>, int>> {
+ public:
+  DBBasicTestWithTsIterTombstones()
+      : DBBasicTestWithTimestampBase("/db_basic_ts_iter_tombstones") {}
+};
+
+TEST_P(DBBasicTestWithTsIterTombstones, ForwardIterDelete) {
+  constexpr size_t kNumKeysPerFile = 128;
+  Options options = CurrentOptions();
+  options.env = env_;
+  const size_t kTimestampSize = Timestamp(0, 0).size();
+  TestComparator test_cmp(kTimestampSize);
+  options.comparator = &test_cmp;
+  options.prefix_extractor = std::get<0>(GetParam());
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  BlockBasedTableOptions bbto;
+  bbto.filter_policy = std::get<1>(GetParam());
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+  options.num_levels = std::get<2>(GetParam());
+  DestroyAndReopen(options);
+  std::vector<std::string> write_ts_strs = {Timestamp(2, 0), Timestamp(4, 0)};
+  constexpr uint64_t kMaxKey = 0xffffffffffffffff;
+  constexpr uint64_t kMinKey = 0xfffffffffffff000;
+  // Insert kMinKey...kMaxKey
+  uint64_t key = kMinKey;
+  WriteOptions write_opts;
+  Slice ts = write_ts_strs[0];
+  write_opts.timestamp = &ts;
+  do {
+    Status s = db_->Put(write_opts, Key1(key), "value" + std::to_string(key));
+    ASSERT_OK(s);
+    if (kMaxKey == key) {
+      break;
+    }
+    ++key;
+  } while (true);
+  // Delete them all
+  ts = write_ts_strs[1];
+  write_opts.timestamp = &ts;
+  for (key = kMaxKey; key >= kMinKey; --key) {
+    Status s;
+    if (0 != (key % 2)) {
+      s = db_->Put(write_opts, Key1(key), "value1" + std::to_string(key));
+    } else {
+      s = db_->Delete(write_opts, Key1(key));
+    }
+    ASSERT_OK(s);
+  }
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  {
+    std::string read_ts = Timestamp(4, 0);
+    ts = read_ts;
+    ReadOptions read_opts;
+    read_opts.total_order_seek = true;
+    read_opts.timestamp = &ts;
+    std::unique_ptr<Iterator> iter(db_->NewIterator(read_opts));
+    size_t count = 0;
+    key = kMinKey + 1;
+    for (iter->SeekToFirst(); iter->Valid(); iter->Next(), ++count, key += 2) {
+      ASSERT_EQ(Key1(key), iter->key());
+      ASSERT_EQ("value1" + std::to_string(key), iter->value());
+    }
+    ASSERT_EQ((kMaxKey - kMinKey + 1) / 2, count);
+  }
+  Close();
+}
+
+INSTANTIATE_TEST_CASE_P(
+    Timestamp, DBBasicTestWithTsIterTombstones,
+    ::testing::Combine(
+        ::testing::Values(
+            std::shared_ptr<const SliceTransform>(NewFixedPrefixTransform(7)),
+            std::shared_ptr<const SliceTransform>(NewFixedPrefixTransform(8))),
+        ::testing::Values(std::shared_ptr<const FilterPolicy>(nullptr),
+                          std::shared_ptr<const FilterPolicy>(
+                              NewBloomFilterPolicy(10, false)),
+                          std::shared_ptr<const FilterPolicy>(
+                              NewBloomFilterPolicy(20, false))),
+        ::testing::Values(2, 6)));
+
 }  // namespace ROCKSDB_NAMESPACE

 #ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
@@ -0,0 +1,119 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// 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/compaction/compaction.h"
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace {
+std::string Key1(uint64_t key) {
+  std::string ret;
+  PutFixed64(&ret, key);
+  std::reverse(ret.begin(), ret.end());
+  return ret;
+}
+
+std::string Timestamp(uint64_t ts) {
+  std::string ret;
+  PutFixed64(&ret, ts);
+  return ret;
+}
+}  // anonymous namespace
+
+class TimestampCompatibleCompactionTest : public DBTestBase {
+ public:
+  TimestampCompatibleCompactionTest()
+      : DBTestBase("/ts_compatible_compaction_test") {}
+
+  std::string Get(const std::string& key, uint64_t ts) {
+    ReadOptions read_opts;
+    std::string ts_str = Timestamp(ts);
+    Slice ts_slice = ts_str;
+    read_opts.timestamp = &ts_slice;
+    std::string value;
+    Status s = db_->Get(read_opts, key, &value);
+    if (s.IsNotFound()) {
+      value.assign("NOT_FOUND");
+    } else if (!s.ok()) {
+      value.assign(s.ToString());
+    }
+    return value;
+  }
+};
+
+TEST_F(TimestampCompatibleCompactionTest, UserKeyCrossFileBoundary) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.compaction_style = kCompactionStyleLevel;
+  options.comparator = test::ComparatorWithU64Ts();
+  options.level0_file_num_compaction_trigger = 3;
+  constexpr size_t kNumKeysPerFile = 101;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  DestroyAndReopen(options);
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        const auto* compaction = reinterpret_cast<Compaction*>(arg);
+        ASSERT_NE(nullptr, compaction);
+        ASSERT_EQ(0, compaction->start_level());
+        ASSERT_EQ(1, compaction->num_input_levels());
+        // Check that all 3 L0 ssts are picked for level compaction.
+        ASSERT_EQ(3, compaction->num_input_files(0));
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+  // Write a L0 with keys 0, 1, ..., 99 with ts from 100 to 199.
+  uint64_t ts = 100;
+  uint64_t key = 0;
+  WriteOptions write_opts;
+  for (; key < kNumKeysPerFile - 1; ++key, ++ts) {
+    std::string ts_str = Timestamp(ts);
+    Slice ts_slice = ts_str;
+    write_opts.timestamp = &ts_slice;
+    ASSERT_OK(db_->Put(write_opts, Key1(key), "foo_" + std::to_string(key)));
+  }
+  // Write another L0 with keys 99 with newer ts.
+  ASSERT_OK(Flush());
+  uint64_t saved_read_ts1 = ts++;
+  key = 99;
+  for (int i = 0; i < 4; ++i, ++ts) {
+    std::string ts_str = Timestamp(ts);
+    Slice ts_slice = ts_str;
+    write_opts.timestamp = &ts_slice;
+    ASSERT_OK(db_->Put(write_opts, Key1(key), "bar_" + std::to_string(key)));
+  }
+  ASSERT_OK(Flush());
+  uint64_t saved_read_ts2 = ts++;
+  // Write another L0 with keys 99, 100, 101, ..., 150
+  for (; key <= 150; ++key, ++ts) {
+    std::string ts_str = Timestamp(ts);
+    Slice ts_slice = ts_str;
+    write_opts.timestamp = &ts_slice;
+    ASSERT_OK(db_->Put(write_opts, Key1(key), "foo1_" + std::to_string(key)));
+  }
+  ASSERT_OK(Flush());
+  // Wait for compaction to finish
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  uint64_t read_ts = ts;
+  ASSERT_EQ("foo_99", Get(Key1(99), saved_read_ts1));
+  ASSERT_EQ("bar_99", Get(Key1(99), saved_read_ts2));
+  ASSERT_EQ("foo1_99", Get(Key1(99), read_ts));
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->DisableProcessing();
+}
+
+}  // namespace ROCKSDB_NAMESPACE
+
+int main(int argc, char** argv) {
+  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -23,7 +23,7 @@ namespace ROCKSDB_NAMESPACE {
 // and the value type is embedded as the low 8 bits in the sequence
 // number in internal keys, we need to use the highest-numbered
 // ValueType, not the lowest).
-const ValueType kValueTypeForSeek = kTypeBlobIndex;
+const ValueType kValueTypeForSeek = kTypeDeletionWithTimestamp;
 const ValueType kValueTypeForSeekForPrev = kTypeDeletion;

 uint64_t PackSequenceAndType(uint64_t seq, ValueType t) {
@@ -69,7 +69,8 @@ enum ValueType : unsigned char {
  // generated by WriteUnprepared write policy is not mistakenly read by
  // another.
  kTypeBeginUnprepareXID = 0x13,  // WAL only.
-  kMaxValue = 0x7F                // Not used for storing records.
+  kTypeDeletionWithTimestamp = 0x14,
+  kMaxValue = 0x7F  // Not used for storing records.
 };

 // Defined in dbformat.cc
@@ -79,7 +80,8 @@ extern const ValueType kValueTypeForSeekForPrev;
 // Checks whether a type is an inline value type
 // (i.e. a type used in memtable skiplist and sst file datablock).
 inline bool IsValueType(ValueType t) {
-  return t <= kTypeMerge || t == kTypeSingleDeletion || t == kTypeBlobIndex;
+  return t <= kTypeMerge || t == kTypeSingleDeletion || t == kTypeBlobIndex ||
+         kTypeDeletionWithTimestamp == t;
 }

 // Checks whether a type is from user operation
@@ -498,15 +498,15 @@ TEST_F(DBErrorHandlingFSTest, CompactionManifestWriteRetryableError) {
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
      "VersionSet::LogAndApply:WriteManifest", [&](void*) {
        if (fail_manifest.load()) {
-          fault_fs->SetFilesystemActive(false,error_msg); }
-        });
+          fault_fs->SetFilesystemActive(false, error_msg);
+        }
+      });
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();

  Put(Key(1), "val");
  s = Flush();
  ASSERT_EQ(s, Status::OK());

-
  TEST_SYNC_POINT("CompactionManifestWriteError:0");
  TEST_SYNC_POINT("CompactionManifestWriteError:1");

@@ -6,6 +6,7 @@
 #include <functional>

 #include "db/db_test_util.h"
+#include "db/version_edit.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
 #include "rocksdb/sst_file_writer.h"
@@ -174,6 +175,111 @@ TEST_F(ExternalSSTFileBasicTest, Basic) {
  ASSERT_EQ(file1_info.num_range_del_entries, 0);
  ASSERT_EQ(file1_info.smallest_range_del_key, "");
  ASSERT_EQ(file1_info.largest_range_del_key, "");
+  ASSERT_EQ(file1_info.file_checksum, kUnknownFileChecksum);
+  ASSERT_EQ(file1_info.file_checksum_func_name, kUnknownFileChecksumFuncName);
+  // sst_file_writer already finished, cannot add this value
+  s = sst_file_writer.Put(Key(100), "bad_val");
+  ASSERT_FALSE(s.ok()) << s.ToString();
+  s = sst_file_writer.DeleteRange(Key(100), Key(200));
+  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();
+}
+
+class ChecksumVerifyHelper {
+ private:
+  Options options_;
+
+ public:
+  ChecksumVerifyHelper(Options& options) : options_(options) {}
+  ~ChecksumVerifyHelper() {}
+
+  Status GetSingleFileChecksumAndFuncName(
+      const std::string& file_path, std::string* file_checksum,
+      std::string* file_checksum_func_name) {
+    Status s;
+    EnvOptions soptions;
+    std::unique_ptr<SequentialFile> file_reader;
+    s = options_.env->NewSequentialFile(file_path, &file_reader, soptions);
+    if (!s.ok()) {
+      return s;
+    }
+    std::unique_ptr<char[]> scratch(new char[2048]);
+    Slice result;
+    FileChecksumGenFactory* file_checksum_gen_factory =
+        options_.file_checksum_gen_factory.get();
+    if (file_checksum_gen_factory == nullptr) {
+      *file_checksum = kUnknownFileChecksum;
+      *file_checksum_func_name = kUnknownFileChecksumFuncName;
+      return Status::OK();
+    } else {
+      FileChecksumGenContext gen_context;
+      std::unique_ptr<FileChecksumGenerator> file_checksum_gen =
+          file_checksum_gen_factory->CreateFileChecksumGenerator(gen_context);
+      *file_checksum_func_name = file_checksum_gen->Name();
+      s = file_reader->Read(2048, &result, scratch.get());
+      if (!s.ok()) {
+        return s;
+      }
+      while (result.size() != 0) {
+        file_checksum_gen->Update(scratch.get(), result.size());
+        s = file_reader->Read(2048, &result, scratch.get());
+        if (!s.ok()) {
+          return s;
+        }
+      }
+      file_checksum_gen->Finalize();
+      *file_checksum = file_checksum_gen->GetChecksum();
+    }
+    return Status::OK();
+  }
+};
+
+TEST_F(ExternalSSTFileBasicTest, BasicWithFileChecksumCrc32c) {
+  Options options = CurrentOptions();
+  options.file_checksum_gen_factory = GetFileChecksumGenCrc32cFactory();
+  ChecksumVerifyHelper checksum_helper(options);
+
+  SstFileWriter sst_file_writer(EnvOptions(), options);
+
+  // 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.Put(Key(k), Key(k) + "_val"));
+  }
+  ExternalSstFileInfo file1_info;
+  Status s = sst_file_writer.Finish(&file1_info);
+  ASSERT_TRUE(s.ok()) << s.ToString();
+  std::string file_checksum, file_checksum_func_name;
+  ASSERT_OK(checksum_helper.GetSingleFileChecksumAndFuncName(
+      file1, &file_checksum, &file_checksum_func_name));
+
+  // 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));
+  ASSERT_EQ(file1_info.num_range_del_entries, 0);
+  ASSERT_EQ(file1_info.smallest_range_del_key, "");
+  ASSERT_EQ(file1_info.largest_range_del_key, "");
+  ASSERT_EQ(file1_info.file_checksum, file_checksum);
+  ASSERT_EQ(file1_info.file_checksum_func_name, file_checksum_func_name);
  // sst_file_writer already finished, cannot add this value
  s = sst_file_writer.Put(Key(100), "bad_val");
  ASSERT_FALSE(s.ok()) << s.ToString();
@@ -68,7 +68,7 @@ Status ExternalSstFileIngestionJob::Prepare(
                                   info2->smallest_internal_key) < 0;
        });

-    for (size_t i = 0; i < num_files - 1; i++) {
+    for (size_t i = 0; i + 1 < num_files; i++) {
      if (sstableKeyCompare(ucmp, sorted_files[i]->largest_internal_key,
                            sorted_files[i + 1]->smallest_internal_key) >= 0) {
        files_overlap_ = true;
@@ -2801,7 +2801,7 @@ TEST_P(ExternalSSTFileTest, IngestFilesTriggerFlushingWithTwoWriteQueue) {
  GenerateAndAddExternalFile(options, data);
 }

-TEST_P(ExternalSSTFileTest, DeltaEncodingWhileGlobalSeqnoPresents) {
+TEST_P(ExternalSSTFileTest, DeltaEncodingWhileGlobalSeqnoPresent) {
  Options options = CurrentOptions();
  DestroyAndReopen(options);
  constexpr size_t kValueSize = 8;
@@ -2842,6 +2842,61 @@ TEST_P(ExternalSSTFileTest, DeltaEncodingWhileGlobalSeqnoPresents) {
  ASSERT_EQ(value, Get(key2));
 }

+TEST_P(ExternalSSTFileTest,
+       DeltaEncodingWhileGlobalSeqnoPresentIteratorSwitch) {
+  // Regression test for bug where global seqno corrupted the shared bytes
+  // buffer when switching from reverse iteration to forward iteration.
+  constexpr size_t kValueSize = 8;
+  Options options = CurrentOptions();
+
+  Random rnd(301);
+  std::string value(RandomString(&rnd, kValueSize));
+
+  std::string key0 = "aa";
+  std::string key1 = "ab";
+  // Make the prefix of key2 is same with key1 add zero seqno. The tail of every
+  // key is composed as (seqno << 8 | value_type), and here `1` represents
+  // ValueType::kTypeValue
+  std::string key2 = "ab";
+  PutFixed64(&key2, PackSequenceAndType(0, kTypeValue));
+  key2 += "cdefghijkl";
+  std::string key3 = key2 + "_";
+
+  // Write some key to make global seqno larger than zero
+  ASSERT_OK(Put(key0, value));
+
+  std::string fname = sst_files_dir_ + "test_file";
+  rocksdb::SstFileWriter writer(EnvOptions(), options);
+  ASSERT_OK(writer.Open(fname));
+
+  // key0 is a dummy to ensure the turnaround point (key1) comes from Prev
+  // cache rather than block (restart keys are pinned in block).
+  ASSERT_OK(writer.Put(key0, value));
+  ASSERT_OK(writer.Put(key1, value));
+  ASSERT_OK(writer.Put(key2, value));
+  ASSERT_OK(writer.Put(key3, value));
+
+  ExternalSstFileInfo info;
+  ASSERT_OK(writer.Finish(&info));
+
+  ASSERT_OK(dbfull()->IngestExternalFile({info.file_path},
+                                         IngestExternalFileOptions()));
+  ReadOptions read_opts;
+  // Prevents Seek() when switching directions, which circumvents the bug.
+  read_opts.total_order_seek = true;
+  Iterator* iter = db_->NewIterator(read_opts);
+  // Scan backwards to key2. File iterator will then be positioned at key1.
+  iter->Seek(key3);
+  ASSERT_EQ(key3, iter->key());
+  iter->Prev();
+  ASSERT_EQ(key2, iter->key());
+  // Scan forwards and make sure key3 is present. Previously key3 would be
+  // corrupted by the global seqno from key1.
+  iter->Next();
+  ASSERT_EQ(key3, iter->key());
+  delete iter;
+}
+
 INSTANTIATE_TEST_CASE_P(ExternalSSTFileTest, ExternalSSTFileTest,
                        testing::Values(std::make_tuple(false, false),
                                        std::make_tuple(false, true),
@@ -181,8 +181,8 @@ TEST_F(FlushJobTest, NonEmpty) {
  // Note: the first two blob references will not be considered when resolving
  // the oldest blob file referenced (the first one is inlined TTL, while the
  // second one is TTL and thus points to a TTL blob file).
-  constexpr std::array<uint64_t, 6> blob_file_numbers{
-      kInvalidBlobFileNumber, 5, 103, 17, 102, 101};
+  constexpr std::array<uint64_t, 6> blob_file_numbers{{
+      kInvalidBlobFileNumber, 5, 103, 17, 102, 101}};
  for (size_t i = 0; i < blob_file_numbers.size(); ++i) {
    std::string key(ToString(i + 10001));
    std::string blob_index;
@@ -36,7 +36,8 @@ class ForwardLevelIterator : public InternalIterator {
  ForwardLevelIterator(const ColumnFamilyData* const cfd,
                       const ReadOptions& read_options,
                       const std::vector<FileMetaData*>& files,
-                       const SliceTransform* prefix_extractor)
+                       const SliceTransform* prefix_extractor,
+                       bool allow_unprepared_value)
      : cfd_(cfd),
        read_options_(read_options),
        files_(files),
@@ -44,7 +45,8 @@ class ForwardLevelIterator : public InternalIterator {
        file_index_(std::numeric_limits<uint32_t>::max()),
        file_iter_(nullptr),
        pinned_iters_mgr_(nullptr),
-        prefix_extractor_(prefix_extractor) {}
+        prefix_extractor_(prefix_extractor),
+        allow_unprepared_value_(allow_unprepared_value) {}

  ~ForwardLevelIterator() override {
    // Reset current pointer
@@ -83,7 +85,8 @@ class ForwardLevelIterator : public InternalIterator {
        /*file_read_hist=*/nullptr, TableReaderCaller::kUserIterator,
        /*arena=*/nullptr, /*skip_filters=*/false, /*level=*/-1,
        /*smallest_compaction_key=*/nullptr,
-        /*largest_compaction_key=*/nullptr);
+        /*largest_compaction_key=*/nullptr,
+        allow_unprepared_value_);
    file_iter_->SetPinnedItersMgr(pinned_iters_mgr_);
    valid_ = false;
    if (!range_del_agg.IsEmpty()) {
@@ -171,6 +174,16 @@ class ForwardLevelIterator : public InternalIterator {
    }
    return Status::OK();
  }
+  bool PrepareValue() override {
+    assert(valid_);
+    if (file_iter_->PrepareValue()) {
+      return true;
+    }
+
+    assert(!file_iter_->Valid());
+    valid_ = false;
+    return false;
+  }
  bool IsKeyPinned() const override {
    return pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled() &&
           file_iter_->IsKeyPinned();
@@ -197,16 +210,19 @@ class ForwardLevelIterator : public InternalIterator {
  InternalIterator* file_iter_;
  PinnedIteratorsManager* pinned_iters_mgr_;
  const SliceTransform* prefix_extractor_;
+  const bool allow_unprepared_value_;
 };

 ForwardIterator::ForwardIterator(DBImpl* db, const ReadOptions& read_options,
                                 ColumnFamilyData* cfd,
-                                 SuperVersion* current_sv)
+                                 SuperVersion* current_sv,
+                                 bool allow_unprepared_value)
    : db_(db),
      read_options_(read_options),
      cfd_(cfd),
      prefix_extractor_(current_sv->mutable_cf_options.prefix_extractor.get()),
      user_comparator_(cfd->user_comparator()),
+      allow_unprepared_value_(allow_unprepared_value),
      immutable_min_heap_(MinIterComparator(&cfd_->internal_comparator())),
      sv_(current_sv),
      mutable_iter_(nullptr),
@@ -560,6 +576,22 @@ Status ForwardIterator::status() const {
  return immutable_status_;
 }

+bool ForwardIterator::PrepareValue() {
+  assert(valid_);
+  if (current_->PrepareValue()) {
+    return true;
+  }
+
+  assert(!current_->Valid());
+  assert(!current_->status().ok());
+  assert(current_ != mutable_iter_); // memtable iterator can't fail
+  assert(immutable_status_.ok());
+
+  valid_ = false;
+  immutable_status_ = current_->status();
+  return false;
+}
+
 Status ForwardIterator::GetProperty(std::string prop_name, std::string* prop) {
  assert(prop != nullptr);
  if (prop_name == "rocksdb.iterator.super-version-number") {
@@ -655,7 +687,8 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
        TableReaderCaller::kUserIterator, /*arena=*/nullptr,
        /*skip_filters=*/false, /*level=*/-1,
        /*smallest_compaction_key=*/nullptr,
-        /*largest_compaction_key=*/nullptr));
+        /*largest_compaction_key=*/nullptr,
+        allow_unprepared_value_));
  }
  BuildLevelIterators(vstorage);
  current_ = nullptr;
@@ -732,7 +765,8 @@ void ForwardIterator::RenewIterators() {
        TableReaderCaller::kUserIterator, /*arena=*/nullptr,
        /*skip_filters=*/false, /*level=*/-1,
        /*smallest_compaction_key=*/nullptr,
-        /*largest_compaction_key=*/nullptr));
+        /*largest_compaction_key=*/nullptr,
+        allow_unprepared_value_));
  }

  for (auto* f : l0_iters_) {
@@ -775,7 +809,8 @@ void ForwardIterator::BuildLevelIterators(const VersionStorageInfo* vstorage) {
    } else {
      level_iters_.push_back(new ForwardLevelIterator(
          cfd_, read_options_, level_files,
-          sv_->mutable_cf_options.prefix_extractor.get()));
+          sv_->mutable_cf_options.prefix_extractor.get(),
+          allow_unprepared_value_));
    }
  }
 }
@@ -796,7 +831,8 @@ void ForwardIterator::ResetIncompleteIterators() {
        TableReaderCaller::kUserIterator, /*arena=*/nullptr,
        /*skip_filters=*/false, /*level=*/-1,
        /*smallest_compaction_key=*/nullptr,
-        /*largest_compaction_key=*/nullptr);
+        /*largest_compaction_key=*/nullptr,
+        allow_unprepared_value_);
    l0_iters_[i]->SetPinnedItersMgr(pinned_iters_mgr_);
  }

@@ -52,7 +52,8 @@ typedef std::priority_queue<InternalIterator*, std::vector<InternalIterator*>,
 class ForwardIterator : public InternalIterator {
 public:
  ForwardIterator(DBImpl* db, const ReadOptions& read_options,
-                  ColumnFamilyData* cfd, SuperVersion* current_sv = nullptr);
+                  ColumnFamilyData* cfd, SuperVersion* current_sv = nullptr,
+                  bool allow_unprepared_value = false);
  virtual ~ForwardIterator();

  void SeekForPrev(const Slice& /*target*/) override {
@@ -75,6 +76,7 @@ class ForwardIterator : public InternalIterator {
  virtual Slice key() const override;
  virtual Slice value() const override;
  virtual Status status() const override;
+  virtual bool PrepareValue() override;
  virtual Status GetProperty(std::string prop_name, std::string* prop) override;
  virtual void SetPinnedItersMgr(
      PinnedIteratorsManager* pinned_iters_mgr) override;
@@ -120,6 +122,7 @@ class ForwardIterator : public InternalIterator {
  ColumnFamilyData* const cfd_;
  const SliceTransform* const prefix_extractor_;
  const Comparator* user_comparator_;
+  const bool allow_unprepared_value_;
  MinIterHeap immutable_min_heap_;

  SuperVersion* sv_;
@@ -33,7 +33,6 @@ Status ImportColumnFamilyJob::Prepare(uint64_t next_file_number,
    files_to_import_.push_back(file_to_import);
  }

-  const auto ucmp = cfd_->internal_comparator().user_comparator();
  auto num_files = files_to_import_.size();
  if (num_files == 0) {
    return Status::InvalidArgument("The list of files is empty");
@@ -55,17 +54,18 @@ Status ImportColumnFamilyJob::Prepare(uint64_t next_file_number,
        }
      }

-      std::sort(sorted_files.begin(), sorted_files.end(),
-                [&ucmp](const IngestedFileInfo* info1,
-                        const IngestedFileInfo* info2) {
-                  return sstableKeyCompare(ucmp, info1->smallest_internal_key,
-                                           info2->smallest_internal_key) < 0;
-                });
+      std::sort(
+          sorted_files.begin(), sorted_files.end(),
+          [this](const IngestedFileInfo* info1, const IngestedFileInfo* info2) {
+            return cfd_->internal_comparator().Compare(
+                       info1->smallest_internal_key,
+                       info2->smallest_internal_key) < 0;
+          });

-      for (size_t i = 0; i < sorted_files.size() - 1; i++) {
-        if (sstableKeyCompare(ucmp, sorted_files[i]->largest_internal_key,
-                              sorted_files[i + 1]->smallest_internal_key) >=
-            0) {
+      for (size_t i = 0; i + 1 < sorted_files.size(); i++) {
+        if (cfd_->internal_comparator().Compare(
+                sorted_files[i]->largest_internal_key,
+                sorted_files[i + 1]->smallest_internal_key) >= 0) {
          return Status::InvalidArgument("Files have overlapping ranges");
        }
      }
@@ -430,6 +430,68 @@ TEST_F(ImportColumnFamilyTest, ImportExportedSSTFromAnotherDB) {
  test::DestroyDir(env_, dbname_ + "/db_copy");
 }

+TEST_F(ImportColumnFamilyTest, LevelFilesOverlappingAtEndpoints) {
+  // Imports a column family containing a level where two files overlap at their
+  // endpoints. "Overlap" means the largest user key in one file is the same as
+  // the smallest user key in the second file.
+  const int kFileBytes = 128 << 10;  // 128KB
+  const int kValueBytes = 1 << 10;   // 1KB
+  const int kNumFiles = 4;
+
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.num_levels = 2;
+  CreateAndReopenWithCF({"koko"}, options);
+
+  Random rnd(301);
+  // Every key is snapshot protected to ensure older versions will not be
+  // dropped during compaction.
+  std::vector<const Snapshot*> snapshots;
+  snapshots.reserve(kFileBytes / kValueBytes * kNumFiles);
+  for (int i = 0; i < kNumFiles; ++i) {
+    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
+      auto value = RandomString(&rnd, kValueBytes);
+      ASSERT_OK(Put(1, "key", value));
+      snapshots.push_back(db_->GetSnapshot());
+    }
+    ASSERT_OK(Flush(1));
+  }
+
+  // Compact to create overlapping L1 files.
+  ASSERT_OK(
+      db_->CompactRange(CompactRangeOptions(), handles_[1], nullptr, nullptr));
+  ASSERT_GT(NumTableFilesAtLevel(1, 1), 1);
+
+  Checkpoint* checkpoint;
+  ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
+  ASSERT_OK(checkpoint->ExportColumnFamily(handles_[1], export_files_dir_,
+                                           &metadata_ptr_));
+  ASSERT_NE(metadata_ptr_, nullptr);
+  delete checkpoint;
+
+  // Create a new db and import the files.
+  DB* db_copy;
+  test::DestroyDir(env_, dbname_ + "/db_copy");
+  ASSERT_OK(DB::Open(options, dbname_ + "/db_copy", &db_copy));
+  ColumnFamilyHandle* cfh = nullptr;
+  ASSERT_OK(db_copy->CreateColumnFamilyWithImport(ColumnFamilyOptions(), "yoyo",
+                                                  ImportColumnFamilyOptions(),
+                                                  *metadata_ptr_, &cfh));
+  ASSERT_NE(cfh, nullptr);
+
+  {
+    std::string value;
+    ASSERT_OK(db_copy->Get(ReadOptions(), cfh, "key", &value));
+  }
+  db_copy->DropColumnFamily(cfh);
+  db_copy->DestroyColumnFamilyHandle(cfh);
+  delete db_copy;
+  test::DestroyDir(env_, dbname_ + "/db_copy");
+  for (const Snapshot* snapshot : snapshots) {
+    db_->ReleaseSnapshot(snapshot);
+  }
+}
+
 TEST_F(ImportColumnFamilyTest, ImportColumnFamilyNegativeTest) {
  Options options = CurrentOptions();
  CreateAndReopenWithCF({"koko"}, options);
@@ -102,8 +102,9 @@ struct SuperVersionContext {

 struct JobContext {
  inline bool HaveSomethingToDelete() const {
-    return full_scan_candidate_files.size() || sst_delete_files.size() ||
-           log_delete_files.size() || manifest_delete_files.size();
+    return !(full_scan_candidate_files.empty() && sst_delete_files.empty() &&
+             blob_delete_files.empty() && log_delete_files.empty() &&
+             manifest_delete_files.empty());
  }

  inline bool HaveSomethingToClean() const {
@@ -140,11 +141,17 @@ struct JobContext {
  std::vector<CandidateFileInfo> full_scan_candidate_files;

  // the list of all live sst files that cannot be deleted
-  std::vector<FileDescriptor> sst_live;
+  std::vector<uint64_t> sst_live;

-  // a list of sst files that we need to delete
+  // the list of sst files that we need to delete
  std::vector<ObsoleteFileInfo> sst_delete_files;

+  // the list of all live blob files that cannot be deleted
+  std::vector<uint64_t> blob_live;
+
+  // the list of blob files that we need to delete
+  std::vector<ObsoleteBlobFileInfo> blob_delete_files;
+
  // a list of log files that we need to delete
  std::vector<uint64_t> log_delete_files;

@@ -11,11 +11,11 @@

 #include <stdio.h>
 #include "file/sequence_file_reader.h"
+#include "port/lang.h"
 #include "rocksdb/env.h"
 #include "test_util/sync_point.h"
 #include "util/coding.h"
 #include "util/crc32c.h"
-#include "util/util.h"

 namespace ROCKSDB_NAMESPACE {
 namespace log {
@@ -49,7 +49,6 @@ class Reader {
  //
  // If "checksum" is true, verify checksums if available.
  Reader(std::shared_ptr<Logger> info_log,
-         // @lint-ignore TXT2 T25377293 Grandfathered in
         std::unique_ptr<SequentialFileReader>&& file, Reporter* reporter,
         bool checksum, uint64_t log_num);
  // No copying allowed
@@ -159,7 +158,6 @@ class Reader {
 class FragmentBufferedReader : public Reader {
 public:
  FragmentBufferedReader(std::shared_ptr<Logger> info_log,
-                         // @lint-ignore TXT2 T25377293 Grandfathered in
                         std::unique_ptr<SequentialFileReader>&& _file,
                         Reporter* reporter, bool checksum, uint64_t log_num)
      : Reader(info_log, std::move(_file), reporter, checksum, log_num),
@@ -23,6 +23,7 @@
 #include "memory/memory_usage.h"
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/statistics.h"
+#include "port/lang.h"
 #include "port/port.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
@@ -36,7 +37,6 @@
 #include "util/autovector.h"
 #include "util/coding.h"
 #include "util/mutexlock.h"
-#include "util/util.h"

 namespace ROCKSDB_NAMESPACE {

@@ -724,6 +724,7 @@ static bool SaveValue(void* arg, const char* entry) {
        return false;
      }
      case kTypeDeletion:
+      case kTypeDeletionWithTimestamp:
      case kTypeSingleDeletion:
      case kTypeRangeDeletion: {
        if (*(s->merge_in_progress)) {
@@ -929,8 +930,18 @@ void MemTable::MultiGet(const ReadOptions& read_options, MultiGetRange* range,

    if (found_final_value) {
      iter->value->PinSelf();
+      range->AddValueSize(iter->value->size());
      range->MarkKeyDone(iter);
      RecordTick(moptions_.statistics, MEMTABLE_HIT);
+      if (range->GetValueSize() > read_options.value_size_soft_limit) {
+        // Set all remaining keys in range to Abort
+        for (auto range_iter = range->begin(); range_iter != range->end();
+             ++range_iter) {
+          range->MarkKeyDone(range_iter);
+          *(range_iter->s) = Status::Aborted();
+        }
+        break;
+      }
    }
  }
  PERF_COUNTER_ADD(get_from_memtable_count, 1);
@@ -43,22 +43,20 @@ void MemTableListVersion::UnrefMemTable(autovector<MemTable*>* to_delete,
 }

 MemTableListVersion::MemTableListVersion(
-    size_t* parent_memtable_list_memory_usage, MemTableListVersion* old)
+    size_t* parent_memtable_list_memory_usage, const MemTableListVersion& old)
    : max_write_buffer_number_to_maintain_(
-          old->max_write_buffer_number_to_maintain_),
+          old.max_write_buffer_number_to_maintain_),
      max_write_buffer_size_to_maintain_(
-          old->max_write_buffer_size_to_maintain_),
+          old.max_write_buffer_size_to_maintain_),
      parent_memtable_list_memory_usage_(parent_memtable_list_memory_usage) {
-  if (old != nullptr) {
-    memlist_ = old->memlist_;
-    for (auto& m : memlist_) {
-      m->Ref();
-    }
+  memlist_ = old.memlist_;
+  for (auto& m : memlist_) {
+    m->Ref();
+  }

-    memlist_history_ = old->memlist_history_;
-    for (auto& m : memlist_history_) {
-      m->Ref();
-    }
+  memlist_history_ = old.memlist_history_;
+  for (auto& m : memlist_history_) {
+    m->Ref();
  }
 }

@@ -604,7 +602,7 @@ void MemTableList::InstallNewVersion() {
  } else {
    // somebody else holds the current version, we need to create new one
    MemTableListVersion* version = current_;
-    current_ = new MemTableListVersion(&current_memory_usage_, current_);
+    current_ = new MemTableListVersion(&current_memory_usage_, *version);
    current_->Ref();
    version->Unref();
  }
@@ -44,7 +44,7 @@ struct FlushJobInfo;
 class MemTableListVersion {
 public:
  explicit MemTableListVersion(size_t* parent_memtable_list_memory_usage,
-                               MemTableListVersion* old = nullptr);
+                               const MemTableListVersion& old);
  explicit MemTableListVersion(size_t* parent_memtable_list_memory_usage,
                               int max_write_buffer_number_to_maintain,
                               int64_t max_write_buffer_size_to_maintain);
@@ -10,6 +10,7 @@
 #ifndef ROCKSDB_LITE

 #include <stdlib.h>
+#include <algorithm>
 #include <map>
 #include <string>
 #include <vector>
@@ -193,6 +194,124 @@ TEST_F(ObsoleteFilesTest, DeleteObsoleteOptionsFile) {
  ASSERT_EQ(2, opts_file_count);
 }

+TEST_F(ObsoleteFilesTest, BlobFiles) {
+  VersionSet* const versions = dbfull()->TEST_GetVersionSet();
+  assert(versions);
+  assert(versions->GetColumnFamilySet());
+
+  ColumnFamilyData* const cfd = versions->GetColumnFamilySet()->GetDefault();
+  assert(cfd);
+
+  // Add a blob file that consists of nothing but garbage (and is thus obsolete)
+  // and another one that is live.
+  VersionEdit edit;
+
+  constexpr uint64_t first_blob_file_number = 234;
+  constexpr uint64_t first_total_blob_count = 555;
+  constexpr uint64_t first_total_blob_bytes = 66666;
+  constexpr char first_checksum_method[] = "CRC32";
+  constexpr char first_checksum_value[] = "3d87ff57";
+
+  edit.AddBlobFile(first_blob_file_number, first_total_blob_count,
+                   first_total_blob_bytes, first_checksum_method,
+                   first_checksum_value);
+  edit.AddBlobFileGarbage(first_blob_file_number, first_total_blob_count,
+                          first_total_blob_bytes);
+
+  constexpr uint64_t second_blob_file_number = 456;
+  constexpr uint64_t second_total_blob_count = 100;
+  constexpr uint64_t second_total_blob_bytes = 2000000;
+  constexpr char second_checksum_method[] = "CRC32B";
+  constexpr char second_checksum_value[] = "6dbdf23a";
+  edit.AddBlobFile(second_blob_file_number, second_total_blob_count,
+                   second_total_blob_bytes, second_checksum_method,
+                   second_checksum_value);
+
+  dbfull()->TEST_LockMutex();
+  Status s = versions->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
+                                   &edit, dbfull()->mutex());
+  dbfull()->TEST_UnlockMutex();
+
+  ASSERT_OK(s);
+
+  // Check for obsolete files and make sure the first blob file is picked up
+  // and grabbed for purge. The second blob file should be on the live list.
+  constexpr int job_id = 0;
+  JobContext job_context{job_id};
+
+  dbfull()->TEST_LockMutex();
+  constexpr bool force_full_scan = false;
+  dbfull()->FindObsoleteFiles(&job_context, force_full_scan);
+  dbfull()->TEST_UnlockMutex();
+
+  ASSERT_TRUE(job_context.HaveSomethingToDelete());
+  ASSERT_EQ(job_context.blob_delete_files.size(), 1);
+  ASSERT_EQ(job_context.blob_delete_files[0].GetBlobFileNumber(),
+            first_blob_file_number);
+
+  const auto& files_grabbed_for_purge =
+      dbfull()->TEST_GetFilesGrabbedForPurge();
+  ASSERT_NE(files_grabbed_for_purge.find(first_blob_file_number),
+            files_grabbed_for_purge.end());
+
+  ASSERT_EQ(job_context.blob_live.size(), 1);
+  ASSERT_EQ(job_context.blob_live[0], second_blob_file_number);
+
+  // Hack the job context a bit by adding a few files to the full scan
+  // list and adjusting the pending file number. We add the two files
+  // above as well as two additional ones, where one is old
+  // and should be cleaned up, and the other is still pending.
+  assert(cfd->ioptions());
+  assert(!cfd->ioptions()->cf_paths.empty());
+  const std::string& path = cfd->ioptions()->cf_paths.front().path;
+
+  constexpr uint64_t old_blob_file_number = 123;
+  constexpr uint64_t pending_blob_file_number = 567;
+
+  job_context.full_scan_candidate_files.emplace_back(
+      BlobFileName(old_blob_file_number), path);
+  job_context.full_scan_candidate_files.emplace_back(
+      BlobFileName(first_blob_file_number), path);
+  job_context.full_scan_candidate_files.emplace_back(
+      BlobFileName(second_blob_file_number), path);
+  job_context.full_scan_candidate_files.emplace_back(
+      BlobFileName(pending_blob_file_number), path);
+
+  job_context.min_pending_output = pending_blob_file_number;
+
+  // Purge obsolete files and make sure we purge the old file and the first file
+  // (and keep the second file and the pending file).
+  std::vector<std::string> deleted_files;
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DeleteObsoleteFileImpl::BeforeDeletion", [&](void* arg) {
+        const std::string* file = static_cast<std::string*>(arg);
+        assert(file);
+
+        constexpr char blob_extension[] = ".blob";
+
+        if (file->find(blob_extension) != std::string::npos) {
+          deleted_files.emplace_back(*file);
+        }
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  dbfull()->PurgeObsoleteFiles(job_context);
+  job_context.Clean();
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  ASSERT_EQ(files_grabbed_for_purge.find(first_blob_file_number),
+            files_grabbed_for_purge.end());
+
+  std::sort(deleted_files.begin(), deleted_files.end());
+  const std::vector<std::string> expected_deleted_files{
+      BlobFileName(path, old_blob_file_number),
+      BlobFileName(path, first_blob_file_number)};
+
+  ASSERT_EQ(deleted_files, expected_deleted_files);
+}
+
 }  // namespace ROCKSDB_NAMESPACE

 #ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
@@ -251,7 +251,7 @@ void ProfileQueries(bool enabled_time = false) {
  }

  if (FLAGS_random_key) {
-    std::random_shuffle(keys.begin(), keys.end());
+    RandomShuffle(std::begin(keys), std::end(keys));
  }
 #ifndef NDEBUG
  ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 1U);
@@ -524,7 +524,7 @@ TEST_F(PerfContextTest, SeekKeyComparison) {
  }

  if (FLAGS_random_key) {
-    std::random_shuffle(keys.begin(), keys.end());
+    RandomShuffle(std::begin(keys), std::end(keys));
  }

  HistogramImpl hist_put_time;
@@ -529,7 +529,8 @@ class Repairer {
          /*table_reader_ptr=*/nullptr, /*file_read_hist=*/nullptr,
          TableReaderCaller::kRepair, /*arena=*/nullptr, /*skip_filters=*/false,
          /*level=*/-1, /*smallest_compaction_key=*/nullptr,
-          /*largest_compaction_key=*/nullptr);
+          /*largest_compaction_key=*/nullptr,
+          /*allow_unprepared_value=*/false);
      ParsedInternalKey parsed;
      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
        Slice key = iter->key();
@@ -62,6 +62,8 @@ void AppendVarint64(IterKey* key, uint64_t v) {

 }  // namespace

+const int kLoadConcurency = 128;
+
 TableCache::TableCache(const ImmutableCFOptions& ioptions,
                       const FileOptions& file_options, Cache* const cache,
                       BlockCacheTracer* const block_cache_tracer)
@@ -69,7 +71,8 @@ TableCache::TableCache(const ImmutableCFOptions& ioptions,
      file_options_(file_options),
      cache_(cache),
      immortal_tables_(false),
-      block_cache_tracer_(block_cache_tracer) {
+      block_cache_tracer_(block_cache_tracer),
+      loader_mutex_(kLoadConcurency, GetSliceNPHash64) {
  if (ioptions_.row_cache) {
    // If the same cache is shared by multiple instances, we need to
    // disambiguate its entries.
@@ -120,7 +123,8 @@ Status TableCache::GetTableReader(
    s = ioptions_.table_factory->NewTableReader(
        TableReaderOptions(ioptions_, prefix_extractor, file_options,
                           internal_comparator, skip_filters, immortal_tables_,
-                           level, fd.largest_seqno, block_cache_tracer_),
+                           false /* force_direct_prefetch */, level,
+                           fd.largest_seqno, block_cache_tracer_),
        std::move(file_reader), fd.GetFileSize(), table_reader,
        prefetch_index_and_filter_in_cache);
    TEST_SYNC_POINT("TableCache::GetTableReader:0");
@@ -155,6 +159,13 @@ Status TableCache::FindTable(const FileOptions& file_options,
    if (no_io) {  // Don't do IO and return a not-found status
      return Status::Incomplete("Table not found in table_cache, no_io is set");
    }
+    MutexLock load_lock(loader_mutex_.get(key));
+    // We check the cache again under loading mutex
+    *handle = cache_->Lookup(key);
+    if (*handle != nullptr) {
+      return s;
+    }
+
    std::unique_ptr<TableReader> table_reader;
    s = GetTableReader(file_options, internal_comparator, fd,
                       false /* sequential mode */, record_read_stats,
@@ -184,7 +195,7 @@ InternalIterator* TableCache::NewIterator(
    TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
    TableReaderCaller caller, Arena* arena, bool skip_filters, int level,
    const InternalKey* smallest_compaction_key,
-    const InternalKey* largest_compaction_key) {
+    const InternalKey* largest_compaction_key, bool allow_unprepared_value) {
  PERF_TIMER_GUARD(new_table_iterator_nanos);

  Status s;
@@ -213,7 +224,8 @@ InternalIterator* TableCache::NewIterator(
    } else {
      result = table_reader->NewIterator(options, prefix_extractor, arena,
                                   skip_filters, caller,
-                                   file_options.compaction_readahead_size);
+                                   file_options.compaction_readahead_size,
+                                   allow_unprepared_value);
    }
    if (handle != nullptr) {
      result->RegisterCleanup(&UnrefEntry, cache_, handle);
@@ -303,8 +315,7 @@ void TableCache::CreateRowCacheKeyPrefix(const ReadOptions& options,
  // Maybe we can include the whole file ifsnapshot == fd.largest_seqno.
  if (options.snapshot != nullptr &&
      (get_context->has_callback() ||
-       static_cast_with_check<const SnapshotImpl, const Snapshot>(
-           options.snapshot)
+       static_cast_with_check<const SnapshotImpl>(options.snapshot)
               ->GetSequenceNumber() <= fd.largest_seqno)) {
    // We should consider to use options.snapshot->GetSequenceNumber()
    // instead of GetInternalKeySeqno(k), which will make the code
@@ -73,7 +73,7 @@ class TableCache {
      const SliceTransform* prefix_extractor, TableReader** table_reader_ptr,
      HistogramImpl* file_read_hist, TableReaderCaller caller, Arena* arena,
      bool skip_filters, int level, const InternalKey* smallest_compaction_key,
-      const InternalKey* largest_compaction_key);
+      const InternalKey* largest_compaction_key, bool allow_unprepared_value);

  // If a seek to internal key "k" in specified file finds an entry,
  // call get_context->SaveValue() repeatedly until
@@ -221,6 +221,7 @@ class TableCache {
  std::string row_cache_id_;
  bool immortal_tables_;
  BlockCacheTracer* const block_cache_tracer_;
+  Striped<port::Mutex, Slice> loader_mutex_;
 };

 }  // namespace ROCKSDB_NAMESPACE
@@ -86,36 +86,79 @@ class VersionBuilder::Rep {
    std::unordered_map<uint64_t, FileMetaData*> added_files;
  };

+  class BlobFileMetaDataDelta {
+   public:
+    bool IsEmpty() const {
+      return !shared_meta_ && !additional_garbage_count_ &&
+             !additional_garbage_bytes_;
+    }
+
+    std::shared_ptr<SharedBlobFileMetaData> GetSharedMeta() const {
+      return shared_meta_;
+    }
+
+    uint64_t GetAdditionalGarbageCount() const {
+      return additional_garbage_count_;
+    }
+
+    uint64_t GetAdditionalGarbageBytes() const {
+      return additional_garbage_bytes_;
+    }
+
+    void SetSharedMeta(std::shared_ptr<SharedBlobFileMetaData> shared_meta) {
+      assert(!shared_meta_);
+      assert(shared_meta);
+
+      shared_meta_ = std::move(shared_meta);
+    }
+
+    void AddGarbage(uint64_t count, uint64_t bytes) {
+      additional_garbage_count_ += count;
+      additional_garbage_bytes_ += bytes;
+    }
+
+   private:
+    std::shared_ptr<SharedBlobFileMetaData> shared_meta_;
+    uint64_t additional_garbage_count_ = 0;
+    uint64_t additional_garbage_bytes_ = 0;
+  };
+
  const FileOptions& file_options_;
-  Logger* info_log_;
+  const ImmutableCFOptions* const ioptions_;
  TableCache* table_cache_;
  VersionStorageInfo* base_vstorage_;
+  VersionSet* version_set_;
  int num_levels_;
  LevelState* levels_;
-  // Store states of levels larger than num_levels_. We do this instead of
+  // Store sizes of levels larger than num_levels_. We do this instead of
  // storing them in levels_ to avoid regression in case there are no files
  // on invalid levels. The version is not consistent if in the end the files
  // on invalid levels don't cancel out.
-  std::map<int, std::unordered_set<uint64_t>> invalid_levels_;
+  std::unordered_map<int, size_t> invalid_level_sizes_;
  // Whether there are invalid new files or invalid deletion on levels larger
  // than num_levels_.
  bool has_invalid_levels_;
+  // Current levels of table files affected by additions/deletions.
+  std::unordered_map<uint64_t, int> table_file_levels_;
  FileComparator level_zero_cmp_;
  FileComparator level_nonzero_cmp_;

-  // Metadata for all blob files affected by the series of version edits.
-  std::map<uint64_t, std::shared_ptr<BlobFileMetaData>> changed_blob_files_;
+  // Metadata delta for all blob files affected by the series of version edits.
+  std::map<uint64_t, BlobFileMetaDataDelta> blob_file_meta_deltas_;

 public:
-  Rep(const FileOptions& file_options, Logger* info_log,
-      TableCache* table_cache,
-      VersionStorageInfo* base_vstorage)
+  Rep(const FileOptions& file_options, const ImmutableCFOptions* ioptions,
+      TableCache* table_cache, VersionStorageInfo* base_vstorage,
+      VersionSet* version_set)
      : file_options_(file_options),
-        info_log_(info_log),
+        ioptions_(ioptions),
        table_cache_(table_cache),
        base_vstorage_(base_vstorage),
+        version_set_(version_set),
        num_levels_(base_vstorage->num_levels()),
        has_invalid_levels_(false) {
+    assert(ioptions_);
+
    levels_ = new LevelState[num_levels_];
    level_zero_cmp_.sort_method = FileComparator::kLevel0;
    level_nonzero_cmp_.sort_method = FileComparator::kLevelNon0;
@@ -146,14 +189,12 @@ class VersionBuilder::Rep {
    }
  }

-  std::shared_ptr<BlobFileMetaData> GetBlobFileMetaData(
-      uint64_t blob_file_number) const {
-    auto changed_it = changed_blob_files_.find(blob_file_number);
-    if (changed_it != changed_blob_files_.end()) {
-      const auto& meta = changed_it->second;
-      assert(meta);
-
-      return meta;
+  bool IsBlobFileInVersion(uint64_t blob_file_number) const {
+    auto delta_it = blob_file_meta_deltas_.find(blob_file_number);
+    if (delta_it != blob_file_meta_deltas_.end()) {
+      if (delta_it->second.GetSharedMeta()) {
+        return true;
+      }
    }

    assert(base_vstorage_);
@@ -162,13 +203,13 @@ class VersionBuilder::Rep {

    auto base_it = base_blob_files.find(blob_file_number);
    if (base_it != base_blob_files.end()) {
-      const auto& meta = base_it->second;
-      assert(meta);
+      assert(base_it->second);
+      assert(base_it->second->GetSharedMeta());

-      return meta;
+      return true;
    }

-    return std::shared_ptr<BlobFileMetaData>();
+    return false;
  }

  Status CheckConsistencyOfOldestBlobFileReference(
@@ -185,8 +226,7 @@ class VersionBuilder::Rep {
      return Status::OK();
    }

-    const auto meta = GetBlobFileMetaData(blob_file_number);
-    if (!meta) {
+    if (!IsBlobFileInVersion(blob_file_number)) {
      std::ostringstream oss;
      oss << "Blob file #" << blob_file_number
          << " is not part of this version";
@@ -232,11 +272,11 @@ class VersionBuilder::Rep {

        auto f1 = level_files[i - 1];
        auto f2 = level_files[i];
-#ifndef NDEBUG
-        auto pair = std::make_pair(&f1, &f2);
-        TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistency", &pair);
-#endif
        if (level == 0) {
+#ifndef NDEBUG
+          auto pair = std::make_pair(&f1, &f2);
+          TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistency0", &pair);
+#endif
          if (!level_zero_cmp_(f1, f2)) {
            fprintf(stderr, "L0 files are not sorted properly");
            return Status::Corruption("L0 files are not sorted properly");
@@ -275,6 +315,10 @@ class VersionBuilder::Rep {
                NumberToString(f2->fd.GetNumber()));
          }
        } else {
+#ifndef NDEBUG
+          auto pair = std::make_pair(&f1, &f2);
+          TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistency1", &pair);
+#endif
          if (!level_nonzero_cmp_(f1, f2)) {
            fprintf(stderr, "L%d files are not sorted properly", level);
            return Status::Corruption("L" + NumberToString(level) +
@@ -312,94 +356,64 @@ class VersionBuilder::Rep {
      }
    }

-    return Status::OK();
+    Status ret_s;
+    TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistencyBeforeReturn",
+                             &ret_s);
+    return ret_s;
  }

-  Status CheckConsistencyForDeletes(VersionEdit* /*edit*/, uint64_t number,
-                                    int level) {
-#ifdef NDEBUG
-    if (!base_vstorage_->force_consistency_checks()) {
-      // Dont run consistency checks in release mode except if
-      // explicitly asked to
-      return Status::OK();
-    }
-#endif
-    // a file to be deleted better exist in the previous version
-    bool found = false;
-    for (int l = 0; !found && l < num_levels_; l++) {
-      const std::vector<FileMetaData*>& base_files =
-          base_vstorage_->LevelFiles(l);
-      for (size_t i = 0; i < base_files.size(); i++) {
-        FileMetaData* f = base_files[i];
-        if (f->fd.GetNumber() == number) {
-          found = true;
-          break;
-        }
-      }
-    }
-    // if the file did not exist in the previous version, then it
-    // is possibly moved from lower level to higher level in current
-    // version
-    for (int l = level + 1; !found && l < num_levels_; l++) {
-      auto& level_added = levels_[l].added_files;
-      auto got = level_added.find(number);
-      if (got != level_added.end()) {
-        found = true;
-        break;
-      }
-    }
-
-    // maybe this file was added in a previous edit that was Applied
-    if (!found) {
-      auto& level_added = levels_[level].added_files;
-      auto got = level_added.find(number);
-      if (got != level_added.end()) {
-        found = true;
-      }
-    }
-    if (!found) {
-      fprintf(stderr, "not found %" PRIu64 "\n", number);
-      return Status::Corruption("not found " + NumberToString(number));
-    }
-    return Status::OK();
-  }
-
-  bool CheckConsistencyForNumLevels() {
+  bool CheckConsistencyForNumLevels() const {
    // Make sure there are no files on or beyond num_levels().
    if (has_invalid_levels_) {
      return false;
    }
-    for (auto& level : invalid_levels_) {
-      if (level.second.size() > 0) {
+
+    for (const auto& pair : invalid_level_sizes_) {
+      const size_t level_size = pair.second;
+      if (level_size != 0) {
        return false;
      }
    }
+
    return true;
  }

  Status ApplyBlobFileAddition(const BlobFileAddition& blob_file_addition) {
    const uint64_t blob_file_number = blob_file_addition.GetBlobFileNumber();

-    auto meta = GetBlobFileMetaData(blob_file_number);
-    if (meta) {
+    if (IsBlobFileInVersion(blob_file_number)) {
      std::ostringstream oss;
      oss << "Blob file #" << blob_file_number << " already added";

      return Status::Corruption("VersionBuilder", oss.str());
    }

-    auto shared_meta = std::make_shared<SharedBlobFileMetaData>(
+    // Note: we use C++11 for now but in C++14, this could be done in a more
+    // elegant way using generalized lambda capture.
+    VersionSet* const vs = version_set_;
+    const ImmutableCFOptions* const ioptions = ioptions_;
+
+    auto deleter = [vs, ioptions](SharedBlobFileMetaData* shared_meta) {
+      if (vs) {
+        assert(ioptions);
+        assert(!ioptions->cf_paths.empty());
+        assert(shared_meta);
+
+        vs->AddObsoleteBlobFile(shared_meta->GetBlobFileNumber(),
+                                ioptions->cf_paths.front().path);
+      }
+
+      delete shared_meta;
+    };
+
+    auto shared_meta = SharedBlobFileMetaData::Create(
        blob_file_number, blob_file_addition.GetTotalBlobCount(),
        blob_file_addition.GetTotalBlobBytes(),
        blob_file_addition.GetChecksumMethod(),
-        blob_file_addition.GetChecksumValue());
+        blob_file_addition.GetChecksumValue(), deleter);

-    constexpr uint64_t garbage_blob_count = 0;
-    constexpr uint64_t garbage_blob_bytes = 0;
-    auto new_meta = std::make_shared<BlobFileMetaData>(
-        std::move(shared_meta), garbage_blob_count, garbage_blob_bytes);
-
-    changed_blob_files_.emplace(blob_file_number, std::move(new_meta));
+    blob_file_meta_deltas_[blob_file_number].SetSharedMeta(
+        std::move(shared_meta));

    return Status::OK();
  }
@@ -407,81 +421,162 @@ class VersionBuilder::Rep {
  Status ApplyBlobFileGarbage(const BlobFileGarbage& blob_file_garbage) {
    const uint64_t blob_file_number = blob_file_garbage.GetBlobFileNumber();

-    auto meta = GetBlobFileMetaData(blob_file_number);
-    if (!meta) {
+    if (!IsBlobFileInVersion(blob_file_number)) {
      std::ostringstream oss;
      oss << "Blob file #" << blob_file_number << " not found";

      return Status::Corruption("VersionBuilder", oss.str());
    }

-    assert(meta->GetBlobFileNumber() == blob_file_number);
+    blob_file_meta_deltas_[blob_file_number].AddGarbage(
+        blob_file_garbage.GetGarbageBlobCount(),
+        blob_file_garbage.GetGarbageBlobBytes());

-    auto new_meta = std::make_shared<BlobFileMetaData>(
-        meta->GetSharedMeta(),
-        meta->GetGarbageBlobCount() + blob_file_garbage.GetGarbageBlobCount(),
-        meta->GetGarbageBlobBytes() + blob_file_garbage.GetGarbageBlobBytes());
+    return Status::OK();
+  }

-    changed_blob_files_[blob_file_number] = std::move(new_meta);
+  int GetCurrentLevelForTableFile(uint64_t file_number) const {
+    auto it = table_file_levels_.find(file_number);
+    if (it != table_file_levels_.end()) {
+      return it->second;
+    }
+
+    assert(base_vstorage_);
+    return base_vstorage_->GetFileLocation(file_number).GetLevel();
+  }
+
+  Status ApplyFileDeletion(int level, uint64_t file_number) {
+    assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
+
+    const int current_level = GetCurrentLevelForTableFile(file_number);
+
+    if (level != current_level) {
+      if (level >= num_levels_) {
+        has_invalid_levels_ = true;
+      }
+
+      std::ostringstream oss;
+      oss << "Cannot delete table file #" << file_number << " from level "
+          << level << " since it is ";
+      if (current_level ==
+          VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
+        oss << "not in the LSM tree";
+      } else {
+        oss << "on level " << current_level;
+      }
+
+      return Status::Corruption("VersionBuilder", oss.str());
+    }
+
+    if (level >= num_levels_) {
+      assert(invalid_level_sizes_[level] > 0);
+      --invalid_level_sizes_[level];
+
+      table_file_levels_[file_number] =
+          VersionStorageInfo::FileLocation::Invalid().GetLevel();
+
+      return Status::OK();
+    }
+
+    auto& level_state = levels_[level];
+
+    auto& add_files = level_state.added_files;
+    auto add_it = add_files.find(file_number);
+    if (add_it != add_files.end()) {
+      UnrefFile(add_it->second);
+      add_files.erase(add_it);
+    } else {
+      auto& del_files = level_state.deleted_files;
+      assert(del_files.find(file_number) == del_files.end());
+      del_files.emplace(file_number);
+    }
+
+    table_file_levels_[file_number] =
+        VersionStorageInfo::FileLocation::Invalid().GetLevel();
+
+    return Status::OK();
+  }
+
+  Status ApplyFileAddition(int level, const FileMetaData& meta) {
+    assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
+
+    const uint64_t file_number = meta.fd.GetNumber();
+
+    const int current_level = GetCurrentLevelForTableFile(file_number);
+
+    if (current_level !=
+        VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
+      if (level >= num_levels_) {
+        has_invalid_levels_ = true;
+      }
+
+      std::ostringstream oss;
+      oss << "Cannot add table file #" << file_number << " to level " << level
+          << " since it is already in the LSM tree on level " << current_level;
+      return Status::Corruption("VersionBuilder", oss.str());
+    }
+
+    if (level >= num_levels_) {
+      ++invalid_level_sizes_[level];
+      table_file_levels_[file_number] = level;
+
+      return Status::OK();
+    }
+
+    auto& level_state = levels_[level];
+
+    auto& del_files = level_state.deleted_files;
+    auto del_it = del_files.find(file_number);
+    if (del_it != del_files.end()) {
+      del_files.erase(del_it);
+    } else {
+      FileMetaData* const f = new FileMetaData(meta);
+      f->refs = 1;
+
+      auto& add_files = level_state.added_files;
+      assert(add_files.find(file_number) == add_files.end());
+      add_files.emplace(file_number, f);
+    }
+
+    table_file_levels_[file_number] = level;

    return Status::OK();
  }

  // Apply all of the edits in *edit to the current state.
  Status Apply(VersionEdit* edit) {
-    Status s = CheckConsistency(base_vstorage_);
-    if (!s.ok()) {
-      return s;
+    {
+      const Status s = CheckConsistency(base_vstorage_);
+      if (!s.ok()) {
+        return s;
+      }
    }

    // Delete files
-    const auto& del = edit->GetDeletedFiles();
-    for (const auto& del_file : del) {
-      const auto level = del_file.first;
-      const auto number = del_file.second;
-      if (level < num_levels_) {
-        levels_[level].deleted_files.insert(number);
-        CheckConsistencyForDeletes(edit, number, level);
+    for (const auto& deleted_file : edit->GetDeletedFiles()) {
+      const int level = deleted_file.first;
+      const uint64_t file_number = deleted_file.second;

-        auto exising = levels_[level].added_files.find(number);
-        if (exising != levels_[level].added_files.end()) {
-          UnrefFile(exising->second);
-          levels_[level].added_files.erase(exising);
-        }
-      } else {
-        if (invalid_levels_[level].erase(number) == 0) {
-          // Deleting an non-existing file on invalid level.
-          has_invalid_levels_ = true;
-        }
+      const Status s = ApplyFileDeletion(level, file_number);
+      if (!s.ok()) {
+        return s;
      }
    }

    // Add new files
    for (const auto& new_file : edit->GetNewFiles()) {
      const int level = new_file.first;
-      if (level < num_levels_) {
-        FileMetaData* f = new FileMetaData(new_file.second);
-        f->refs = 1;
+      const FileMetaData& meta = new_file.second;

-        assert(levels_[level].added_files.find(f->fd.GetNumber()) ==
-               levels_[level].added_files.end());
-        levels_[level].deleted_files.erase(f->fd.GetNumber());
-        levels_[level].added_files[f->fd.GetNumber()] = f;
-      } else {
-        uint64_t number = new_file.second.fd.GetNumber();
-        auto& lvls = invalid_levels_[level];
-        if (lvls.count(number) == 0) {
-          lvls.insert(number);
-        } else {
-          // Creating an already existing file on invalid level.
-          has_invalid_levels_ = true;
-        }
+      const Status s = ApplyFileAddition(level, meta);
+      if (!s.ok()) {
+        return s;
      }
    }

    // Add new blob files
    for (const auto& blob_file_addition : edit->GetBlobFileAdditions()) {
-      s = ApplyBlobFileAddition(blob_file_addition);
+      const Status s = ApplyBlobFileAddition(blob_file_addition);
      if (!s.ok()) {
        return s;
      }
@@ -489,13 +584,47 @@ class VersionBuilder::Rep {

    // Increase the amount of garbage for blob files affected by GC
    for (const auto& blob_file_garbage : edit->GetBlobFileGarbages()) {
-      s = ApplyBlobFileGarbage(blob_file_garbage);
+      const Status s = ApplyBlobFileGarbage(blob_file_garbage);
      if (!s.ok()) {
        return s;
      }
    }

-    return s;
+    return Status::OK();
+  }
+
+  static std::shared_ptr<BlobFileMetaData> CreateMetaDataForNewBlobFile(
+      const BlobFileMetaDataDelta& delta) {
+    auto shared_meta = delta.GetSharedMeta();
+    assert(shared_meta);
+
+    auto meta = BlobFileMetaData::Create(std::move(shared_meta),
+                                         delta.GetAdditionalGarbageCount(),
+                                         delta.GetAdditionalGarbageBytes());
+
+    return meta;
+  }
+
+  static std::shared_ptr<BlobFileMetaData>
+  GetOrCreateMetaDataForExistingBlobFile(
+      const std::shared_ptr<BlobFileMetaData>& base_meta,
+      const BlobFileMetaDataDelta& delta) {
+    assert(base_meta);
+    assert(!delta.GetSharedMeta());
+
+    if (delta.IsEmpty()) {
+      return base_meta;
+    }
+
+    auto shared_meta = base_meta->GetSharedMeta();
+    assert(shared_meta);
+
+    auto meta = BlobFileMetaData::Create(
+        std::move(shared_meta),
+        base_meta->GetGarbageBlobCount() + delta.GetAdditionalGarbageCount(),
+        base_meta->GetGarbageBlobBytes() + delta.GetAdditionalGarbageBytes());
+
+    return meta;
  }

  void AddBlobFileIfNeeded(
@@ -519,42 +648,41 @@ class VersionBuilder::Rep {
    auto base_it = base_blob_files.begin();
    const auto base_it_end = base_blob_files.end();

-    auto changed_it = changed_blob_files_.begin();
-    const auto changed_it_end = changed_blob_files_.end();
+    auto delta_it = blob_file_meta_deltas_.begin();
+    const auto delta_it_end = blob_file_meta_deltas_.end();

-    while (base_it != base_it_end && changed_it != changed_it_end) {
+    while (base_it != base_it_end && delta_it != delta_it_end) {
      const uint64_t base_blob_file_number = base_it->first;
-      const uint64_t changed_blob_file_number = changed_it->first;
+      const uint64_t delta_blob_file_number = delta_it->first;

-      const auto& base_meta = base_it->second;
-      const auto& changed_meta = changed_it->second;
-
-      assert(base_meta);
-      assert(changed_meta);
-
-      if (base_blob_file_number < changed_blob_file_number) {
+      if (base_blob_file_number < delta_blob_file_number) {
+        const auto& base_meta = base_it->second;
+        assert(base_meta);
        assert(base_meta->GetGarbageBlobCount() <
               base_meta->GetTotalBlobCount());

        vstorage->AddBlobFile(base_meta);

        ++base_it;
-      } else if (changed_blob_file_number < base_blob_file_number) {
-        AddBlobFileIfNeeded(vstorage, changed_meta);
+      } else if (delta_blob_file_number < base_blob_file_number) {
+        // Note: blob file numbers are strictly increasing over time and
+        // once blob files get marked obsolete, they never reappear. Thus,
+        // this case is not possible.
+        assert(false);

-        ++changed_it;
+        ++delta_it;
      } else {
-        assert(base_blob_file_number == changed_blob_file_number);
-        assert(base_meta->GetSharedMeta() == changed_meta->GetSharedMeta());
-        assert(base_meta->GetGarbageBlobCount() <=
-               changed_meta->GetGarbageBlobCount());
-        assert(base_meta->GetGarbageBlobBytes() <=
-               changed_meta->GetGarbageBlobBytes());
+        assert(base_blob_file_number == delta_blob_file_number);

-        AddBlobFileIfNeeded(vstorage, changed_meta);
+        const auto& base_meta = base_it->second;
+        const auto& delta = delta_it->second;
+
+        auto meta = GetOrCreateMetaDataForExistingBlobFile(base_meta, delta);
+
+        AddBlobFileIfNeeded(vstorage, meta);

        ++base_it;
-        ++changed_it;
+        ++delta_it;
      }
    }

@@ -567,12 +695,14 @@ class VersionBuilder::Rep {
      ++base_it;
    }

-    while (changed_it != changed_it_end) {
-      const auto& changed_meta = changed_it->second;
-      assert(changed_meta);
+    while (delta_it != delta_it_end) {
+      const auto& delta = delta_it->second;

-      AddBlobFileIfNeeded(vstorage, changed_meta);
-      ++changed_it;
+      auto meta = CreateMetaDataForNewBlobFile(delta);
+
+      AddBlobFileIfNeeded(vstorage, meta);
+
+      ++delta_it;
    }
  }

@@ -738,16 +868,19 @@ class VersionBuilder::Rep {
      // f is to-be-deleted table file
      vstorage->RemoveCurrentStats(f);
    } else {
-      vstorage->AddFile(level, f, info_log_);
+      assert(ioptions_);
+      vstorage->AddFile(level, f, ioptions_->info_log);
    }
  }
 };

 VersionBuilder::VersionBuilder(const FileOptions& file_options,
+                               const ImmutableCFOptions* ioptions,
                               TableCache* table_cache,
                               VersionStorageInfo* base_vstorage,
-                               Logger* info_log)
-    : rep_(new Rep(file_options, info_log, table_cache, base_vstorage)) {}
+                               VersionSet* version_set)
+    : rep_(new Rep(file_options, ioptions, table_cache, base_vstorage,
+                   version_set)) {}

 VersionBuilder::~VersionBuilder() = default;

@@ -773,8 +906,9 @@ Status VersionBuilder::LoadTableHandlers(
 BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
    ColumnFamilyData* cfd)
    : version_builder_(new VersionBuilder(
-          cfd->current()->version_set()->file_options(), cfd->table_cache(),
-          cfd->current()->storage_info(), cfd->ioptions()->info_log)),
+          cfd->current()->version_set()->file_options(), cfd->ioptions(),
+          cfd->table_cache(), cfd->current()->storage_info(),
+          cfd->current()->version_set())),
      version_(cfd->current()) {
  version_->Ref();
 }
@@ -782,8 +916,8 @@ BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
 BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
    ColumnFamilyData* cfd, Version* v)
    : version_builder_(new VersionBuilder(
-          cfd->current()->version_set()->file_options(), cfd->table_cache(),
-          v->storage_info(), cfd->ioptions()->info_log)),
+          cfd->current()->version_set()->file_options(), cfd->ioptions(),
+          cfd->table_cache(), v->storage_info(), v->version_set())),
      version_(v) {
  assert(version_ != cfd->current());
 }
@@ -16,12 +16,14 @@

 namespace ROCKSDB_NAMESPACE {

+struct ImmutableCFOptions;
 class TableCache;
 class VersionStorageInfo;
 class VersionEdit;
 struct FileMetaData;
 class InternalStats;
 class Version;
+class VersionSet;
 class ColumnFamilyData;

 // A helper class so we can efficiently apply a whole sequence
@@ -29,8 +31,9 @@ class ColumnFamilyData;
 // Versions that contain full copies of the intermediate state.
 class VersionBuilder {
 public:
-  VersionBuilder(const FileOptions& file_options, TableCache* table_cache,
-                 VersionStorageInfo* base_vstorage, Logger* info_log = nullptr);
+  VersionBuilder(const FileOptions& file_options,
+                 const ImmutableCFOptions* ioptions, TableCache* table_cache,
+                 VersionStorageInfo* base_vstorage, VersionSet* version_set);
  ~VersionBuilder();

  bool CheckConsistencyForNumLevels();
@@ -54,7 +54,7 @@ class VersionBuilderTest : public testing::Test {
    return InternalKey(ukey, smallest_seq, kTypeValue);
  }

-  void Add(int level, uint32_t file_number, const char* smallest,
+  void Add(int level, uint64_t file_number, const char* smallest,
           const char* largest, uint64_t file_size = 0, uint32_t path_id = 0,
           SequenceNumber smallest_seq = 100, SequenceNumber largest_seq = 100,
           uint64_t num_entries = 0, uint64_t num_deletions = 0,
@@ -82,10 +82,10 @@ class VersionBuilderTest : public testing::Test {
               uint64_t total_blob_bytes, std::string checksum_method,
               std::string checksum_value, uint64_t garbage_blob_count,
               uint64_t garbage_blob_bytes) {
-    auto shared_meta = std::make_shared<SharedBlobFileMetaData>(
+    auto shared_meta = SharedBlobFileMetaData::Create(
        blob_file_number, total_blob_count, total_blob_bytes,
        std::move(checksum_method), std::move(checksum_value));
-    auto meta = std::make_shared<BlobFileMetaData>(
+    auto meta = BlobFileMetaData::Create(
        std::move(shared_meta), garbage_blob_count, garbage_blob_bytes);

    vstorage_.AddBlobFile(std::move(meta));
@@ -152,8 +152,11 @@ TEST_F(VersionBuilderTest, ApplyAndSaveTo) {
  version_edit.DeleteFile(3, 27U);

  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;

-  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
                                  kCompactionStyleLevel, nullptr, false);
@@ -190,8 +193,11 @@ TEST_F(VersionBuilderTest, ApplyAndSaveToDynamic) {
  version_edit.DeleteFile(0, 88U);

  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;

-  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
                                  kCompactionStyleLevel, nullptr, false);
@@ -233,8 +239,11 @@ TEST_F(VersionBuilderTest, ApplyAndSaveToDynamic2) {
  version_edit.DeleteFile(4, 8U);

  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;

-  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
                                  kCompactionStyleLevel, nullptr, false);
@@ -279,8 +288,11 @@ TEST_F(VersionBuilderTest, ApplyMultipleAndSaveTo) {
                       kUnknownFileChecksumFuncName);

  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;

-  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);

  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
                                  kCompactionStyleLevel, nullptr, false);
@@ -296,7 +308,12 @@ TEST_F(VersionBuilderTest, ApplyDeleteAndSaveTo) {
  UpdateVersionStorageInfo();

  EnvOptions env_options;
-  VersionBuilder version_builder(env_options, nullptr, &vstorage_);
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);
+
  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
                                  kCompactionStyleLevel, nullptr, false);

@@ -350,10 +367,242 @@ TEST_F(VersionBuilderTest, ApplyDeleteAndSaveTo) {
  UnrefFilesInVersion(&new_vstorage);
 }

+TEST_F(VersionBuilderTest, ApplyFileDeletionIncorrectLevel) {
+  constexpr int level = 1;
+  constexpr uint64_t file_number = 2345;
+  constexpr char smallest[] = "bar";
+  constexpr char largest[] = "foo";
+
+  Add(level, file_number, smallest, largest);
+
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit edit;
+
+  constexpr int incorrect_level = 3;
+
+  edit.DeleteFile(incorrect_level, file_number);
+
+  const Status s = builder.Apply(&edit);
+  ASSERT_TRUE(s.IsCorruption());
+  ASSERT_TRUE(std::strstr(s.getState(),
+                          "Cannot delete table file #2345 from level 3 since "
+                          "it is on level 1"));
+}
+
+TEST_F(VersionBuilderTest, ApplyFileDeletionNotInLSMTree) {
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit edit;
+
+  constexpr int level = 3;
+  constexpr uint64_t file_number = 1234;
+
+  edit.DeleteFile(level, file_number);
+
+  const Status s = builder.Apply(&edit);
+  ASSERT_TRUE(s.IsCorruption());
+  ASSERT_TRUE(std::strstr(s.getState(),
+                          "Cannot delete table file #1234 from level 3 since "
+                          "it is not in the LSM tree"));
+}
+
+TEST_F(VersionBuilderTest, ApplyFileDeletionAndAddition) {
+  constexpr int level = 1;
+  constexpr uint64_t file_number = 2345;
+  constexpr char smallest[] = "bar";
+  constexpr char largest[] = "foo";
+
+  Add(level, file_number, smallest, largest);
+
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit deletion;
+
+  deletion.DeleteFile(level, file_number);
+
+  ASSERT_OK(builder.Apply(&deletion));
+
+  VersionEdit addition;
+
+  constexpr uint32_t path_id = 0;
+  constexpr uint64_t file_size = 10000;
+  constexpr SequenceNumber smallest_seqno = 100;
+  constexpr SequenceNumber largest_seqno = 1000;
+  constexpr bool marked_for_compaction = false;
+
+  addition.AddFile(level, file_number, path_id, file_size,
+                   GetInternalKey(smallest), GetInternalKey(largest),
+                   smallest_seqno, largest_seqno, marked_for_compaction,
+                   kInvalidBlobFileNumber, kUnknownOldestAncesterTime,
+                   kUnknownFileCreationTime, kUnknownFileChecksum,
+                   kUnknownFileChecksumFuncName);
+
+  ASSERT_OK(builder.Apply(&addition));
+
+  constexpr bool force_consistency_checks = false;
+  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
+                                  kCompactionStyleLevel, &vstorage_,
+                                  force_consistency_checks);
+
+  ASSERT_OK(builder.SaveTo(&new_vstorage));
+  ASSERT_EQ(new_vstorage.GetFileLocation(file_number).GetLevel(), level);
+
+  UnrefFilesInVersion(&new_vstorage);
+}
+
+TEST_F(VersionBuilderTest, ApplyFileAdditionAlreadyInBase) {
+  constexpr int level = 1;
+  constexpr uint64_t file_number = 2345;
+  constexpr char smallest[] = "bar";
+  constexpr char largest[] = "foo";
+
+  Add(level, file_number, smallest, largest);
+
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit edit;
+
+  constexpr int new_level = 2;
+  constexpr uint32_t path_id = 0;
+  constexpr uint64_t file_size = 10000;
+  constexpr SequenceNumber smallest_seqno = 100;
+  constexpr SequenceNumber largest_seqno = 1000;
+  constexpr bool marked_for_compaction = false;
+
+  edit.AddFile(new_level, file_number, path_id, file_size,
+               GetInternalKey(smallest), GetInternalKey(largest),
+               smallest_seqno, largest_seqno, marked_for_compaction,
+               kInvalidBlobFileNumber, kUnknownOldestAncesterTime,
+               kUnknownFileCreationTime, kUnknownFileChecksum,
+               kUnknownFileChecksumFuncName);
+
+  const Status s = builder.Apply(&edit);
+  ASSERT_TRUE(s.IsCorruption());
+  ASSERT_TRUE(std::strstr(s.getState(),
+                          "Cannot add table file #2345 to level 2 since it is "
+                          "already in the LSM tree on level 1"));
+}
+
+TEST_F(VersionBuilderTest, ApplyFileAdditionAlreadyApplied) {
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit edit;
+
+  constexpr int level = 3;
+  constexpr uint64_t file_number = 2345;
+  constexpr uint32_t path_id = 0;
+  constexpr uint64_t file_size = 10000;
+  constexpr char smallest[] = "bar";
+  constexpr char largest[] = "foo";
+  constexpr SequenceNumber smallest_seqno = 100;
+  constexpr SequenceNumber largest_seqno = 1000;
+  constexpr bool marked_for_compaction = false;
+
+  edit.AddFile(level, file_number, path_id, file_size, GetInternalKey(smallest),
+               GetInternalKey(largest), smallest_seqno, largest_seqno,
+               marked_for_compaction, kInvalidBlobFileNumber,
+               kUnknownOldestAncesterTime, kUnknownFileCreationTime,
+               kUnknownFileChecksum, kUnknownFileChecksumFuncName);
+
+  ASSERT_OK(builder.Apply(&edit));
+
+  VersionEdit other_edit;
+
+  constexpr int new_level = 2;
+
+  other_edit.AddFile(new_level, file_number, path_id, file_size,
+                     GetInternalKey(smallest), GetInternalKey(largest),
+                     smallest_seqno, largest_seqno, marked_for_compaction,
+                     kInvalidBlobFileNumber, kUnknownOldestAncesterTime,
+                     kUnknownFileCreationTime, kUnknownFileChecksum,
+                     kUnknownFileChecksumFuncName);
+
+  const Status s = builder.Apply(&other_edit);
+  ASSERT_TRUE(s.IsCorruption());
+  ASSERT_TRUE(std::strstr(s.getState(),
+                          "Cannot add table file #2345 to level 2 since it is "
+                          "already in the LSM tree on level 3"));
+}
+
+TEST_F(VersionBuilderTest, ApplyFileAdditionAndDeletion) {
+  constexpr int level = 1;
+  constexpr uint64_t file_number = 2345;
+  constexpr uint32_t path_id = 0;
+  constexpr uint64_t file_size = 10000;
+  constexpr char smallest[] = "bar";
+  constexpr char largest[] = "foo";
+  constexpr SequenceNumber smallest_seqno = 100;
+  constexpr SequenceNumber largest_seqno = 1000;
+  constexpr bool marked_for_compaction = false;
+
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);
+
+  VersionEdit addition;
+
+  addition.AddFile(level, file_number, path_id, file_size,
+                   GetInternalKey(smallest), GetInternalKey(largest),
+                   smallest_seqno, largest_seqno, marked_for_compaction,
+                   kInvalidBlobFileNumber, kUnknownOldestAncesterTime,
+                   kUnknownFileCreationTime, kUnknownFileChecksum,
+                   kUnknownFileChecksumFuncName);
+
+  ASSERT_OK(builder.Apply(&addition));
+
+  VersionEdit deletion;
+
+  deletion.DeleteFile(level, file_number);
+
+  ASSERT_OK(builder.Apply(&deletion));
+
+  constexpr bool force_consistency_checks = false;
+  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
+                                  kCompactionStyleLevel, &vstorage_,
+                                  force_consistency_checks);
+
+  ASSERT_OK(builder.SaveTo(&new_vstorage));
+  ASSERT_FALSE(new_vstorage.GetFileLocation(file_number).IsValid());
+
+  UnrefFilesInVersion(&new_vstorage);
+}
+
 TEST_F(VersionBuilderTest, ApplyBlobFileAddition) {
  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -406,7 +655,10 @@ TEST_F(VersionBuilderTest, ApplyBlobFileAdditionAlreadyInBase) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -423,7 +675,10 @@ TEST_F(VersionBuilderTest, ApplyBlobFileAdditionAlreadyApplied) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -463,7 +718,10 @@ TEST_F(VersionBuilderTest, ApplyBlobFileGarbageFileInBase) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -505,7 +763,10 @@ TEST_F(VersionBuilderTest, ApplyBlobFileGarbageFileAdditionApplied) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit addition;

@@ -558,7 +819,10 @@ TEST_F(VersionBuilderTest, ApplyBlobFileGarbageFileNotFound) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -591,7 +855,10 @@ TEST_F(VersionBuilderTest, SaveBlobFilesTo) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -681,7 +948,10 @@ TEST_F(VersionBuilderTest, CheckConsistencyForBlobFiles) {
  // new table file--blob file pair.
  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  VersionEdit edit;

@@ -739,7 +1009,10 @@ TEST_F(VersionBuilderTest, CheckConsistencyForBlobFilesNotInVersion) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  // Save to a new version in order to trigger consistency checks.
  constexpr bool force_consistency_checks = true;
@@ -776,7 +1049,10 @@ TEST_F(VersionBuilderTest, CheckConsistencyForBlobFilesAllGarbage) {

  EnvOptions env_options;
  constexpr TableCache* table_cache = nullptr;
-  VersionBuilder builder(env_options, table_cache, &vstorage_);
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder builder(env_options, &ioptions_, table_cache, &vstorage_,
+                         version_set);

  // Save to a new version in order to trigger consistency checks.
  constexpr bool force_consistency_checks = true;
@@ -792,6 +1068,36 @@ TEST_F(VersionBuilderTest, CheckConsistencyForBlobFilesAllGarbage) {
  UnrefFilesInVersion(&new_vstorage);
 }

+TEST_F(VersionBuilderTest, CheckConsistencyForFileDeletedTwice) {
+  Add(0, 1U, "150", "200", 100U);
+  UpdateVersionStorageInfo();
+
+  VersionEdit version_edit;
+  version_edit.DeleteFile(0, 1U);
+
+  EnvOptions env_options;
+  constexpr TableCache* table_cache = nullptr;
+  constexpr VersionSet* version_set = nullptr;
+
+  VersionBuilder version_builder(env_options, &ioptions_, table_cache,
+                                 &vstorage_, version_set);
+  VersionStorageInfo new_vstorage(&icmp_, ucmp_, options_.num_levels,
+                                  kCompactionStyleLevel, nullptr,
+                                  true /* force_consistency_checks */);
+  ASSERT_OK(version_builder.Apply(&version_edit));
+  ASSERT_OK(version_builder.SaveTo(&new_vstorage));
+
+  VersionBuilder version_builder2(env_options, &ioptions_, table_cache,
+                                 &new_vstorage, version_set);
+  VersionStorageInfo new_vstorage2(&icmp_, ucmp_, options_.num_levels,
+                                  kCompactionStyleLevel, nullptr,
+                                  true /* force_consistency_checks */);
+  ASSERT_NOK(version_builder2.Apply(&version_edit));
+
+  UnrefFilesInVersion(&new_vstorage);
+  UnrefFilesInVersion(&new_vstorage2);
+}
+
 TEST_F(VersionBuilderTest, EstimatedActiveKeys) {
  const uint32_t kTotalSamples = 20;
  const uint32_t kNumLevels = 5;
--- a/Show More
+++ b/Show More