Fix EnvLibrados and add to CI (#9088 )

Summary: This feature was not part of any common or CI build, so no surprise it broke. Now we can at least ensure compilation. I don't know how to run the test successfully (missing config file) so it is bypassed for now. Fixes https://github.com/facebook/rocksdb/issues/9078 Pull Request resolved: https://github.com/facebook/rocksdb/pull/9088 Test Plan: CI Reviewed By: mrambacher Differential Revision: D32009467 Pulled By: pdillinger fbshipit-source-id: 3e0d1e5fde7f0ece703d48a81479e1cc7392c25c
update HISTORY.md and version.h for 6.25.3
2026-11-07 03:22:34 -05:00 · 2021-10-29 09:21:24 -07:00 · 2021-10-14 10:46:37 -07:00 · 2021-10-14 10:45:19 -07:00 · 2021-10-14 10:44:46 -07:00 · 2021-10-11 16:42:46 -07:00
539 changed files with 32923 additions and 9720 deletions
@@ -5,9 +5,9 @@ orbs:
  slack: circleci/slack@3.4.2

 aliases:
-  - &notify-on-master-failure
+  - &notify-on-main-failure
    fail_only: true
-    only_for_branches: master
+    only_for_branches: main

 commands:
  install-pyenv-on-macos:
@@ -60,7 +60,7 @@ commands:

  post-steps:
    steps:
-      - slack/status: *notify-on-master-failure
+      - slack/status: *notify-on-main-failure
      - store_test_results: # store test result if there's any
          path: /tmp/test-results
      - store_artifacts: # store LOG for debugging if there's any
@@ -91,6 +91,20 @@ commands:
          command: |
            sudo apt-get update -y && sudo apt-get install -y libgflags-dev

+  install-benchmark:
+    steps:
+      - run: # currently doesn't support ubuntu-1604 which doesn't have libbenchmark package, user can still install by building it youself
+          name: Install benchmark
+          command: |
+            sudo apt-get update -y && sudo apt-get install -y libbenchmark-dev
+
+  install-librados:
+    steps:
+      - run:
+          name: Install librados
+          command: |
+            sudo apt-get update -y && sudo apt-get install -y librados-dev
+
  upgrade-cmake:
    steps:
      - run:
@@ -164,14 +178,15 @@ jobs:
      - run: make V=1 J=32 -j32 check | .circleci/cat_ignore_eagain
      - post-steps

-  build-linux-mem-env:
+  build-linux-mem-env-librados:
    machine:
      image: ubuntu-1604:202104-01
    resource_class: 2xlarge
    steps:
      - pre-steps
      - install-gflags
-      - run: MEM_ENV=1 make V=1 J=32 -j32 check | .circleci/cat_ignore_eagain
+      - install-librados
+      - run: MEM_ENV=1 ROCKSDB_USE_LIBRADOS=1 make V=1 J=32 -j32 check | .circleci/cat_ignore_eagain
      - post-steps

  build-linux-encrypted-env:
@@ -304,7 +319,7 @@ jobs:
      image: ubuntu-1604:202104-01
    resource_class: 2xlarge
    steps:
-      - checkout # check out the code in the project directory
+      - pre-steps
      - install-gflags
      - upgrade-cmake
      - run: (mkdir build && cd build && cmake -DWITH_GFLAGS=1 .. && make V=1 -j20 && ctest -j20) | .circleci/cat_ignore_eagain
@@ -315,12 +330,13 @@ jobs:
      image: ubuntu-2004:202104-01
    resource_class: 2xlarge
    steps:
-      - checkout # check out the code in the project directory
+      - pre-steps
      - install-gflags
-      - run: (mkdir build && cd build && cmake -DWITH_GFLAGS=1 .. && make V=1 -j20 && ctest -j20) | .circleci/cat_ignore_eagain
+      - install-benchmark
+      - run: (mkdir build && cd build && cmake -DWITH_GFLAGS=1 -DWITH_BENCHMARK=1 .. && make V=1 -j20 && ctest -j20 && make microbench) | .circleci/cat_ignore_eagain
      - post-steps

-  build-linux-unity:
+  build-linux-unity-and-headers:
    docker: # executor type
      - image: gcc:latest
    resource_class: large
@@ -328,6 +344,7 @@ jobs:
      - checkout # check out the code in the project directory
      - run: apt-get update -y && apt-get install -y libgflags-dev
      - run: TEST_TMPDIR=/dev/shm && make V=1 -j8 unity_test | .circleci/cat_ignore_eagain
+      - run: make V=1 -j8 -k check-headers | .circleci/cat_ignore_eagain # could be moved to a different build
      - post-steps

  build-linux-gcc-4_8-no_test_run:
@@ -370,6 +387,17 @@ jobs:
      - run: CC=gcc-10 CXX=g++-10 V=1 SKIP_LINK=1 ROCKSDB_CXX_STANDARD=c++20 make -j16 all | .circleci/cat_ignore_eagain # Linking broken because libgflags compiled with newer ABI
      - post-steps

+  # This job is only to make sure the microbench tests are able to run, the benchmark result is not meaningful as the CI host is changing.
+  build-linux-microbench:
+    machine:
+      image: ubuntu-2004:202010-01
+    resource_class: xlarge
+    steps:
+      - pre-steps
+      - install-benchmark
+      - run: DEBUG_LEVEL=0 make microbench | .circleci/cat_ignore_eagain
+      - post-steps
+
  build-windows:
    executor: windows-2xlarge
    parameters:
@@ -678,9 +706,9 @@ workflows:
    jobs:
      - build-linux-cmake
      - build-linux-cmake-ubuntu-20
-  build-linux-mem-env:
+  build-linux-mem-env-librados:
    jobs:
-      - build-linux-mem-env
+      - build-linux-mem-env-librados
  build-linux-encrypted-env:
    jobs:
      - build-linux-encrypted-env
@@ -711,9 +739,9 @@ workflows:
  build-linux-clang10-clang-analyze:
    jobs:
      - build-linux-clang10-clang-analyze
-  build-linux-unity:
+  build-linux-unity-and-headers:
    jobs:
-      - build-linux-unity
+      - build-linux-unity-and-headers
  build-windows-vs2019:
    jobs:
      - build-windows:
@@ -729,12 +757,6 @@ workflows:
          name: "build-windows-vs2017"
          vs_year: "2017"
          cmake_generator: "Visual Studio 15 Win64"
-#  build-windows-vs2015:
-#    jobs:
-#      - build-windows:
-#          name: "build-windows-vs2015"
-#          vs_year: "2015"
-#          cmake_generator: "Visual Studio 14 Win64"
  build-java:
    jobs:
      - build-linux-java
@@ -778,6 +800,9 @@ workflows:
  build-linux-arm:
    jobs:
      - build-linux-arm
+  build-microbench:
+    jobs:
+      - build-linux-microbench
  nightly:
    triggers:
      - schedule:
@@ -785,7 +810,7 @@ workflows:
          filters:
            branches:
              only:
-                - master
+                - main
    jobs:
      - build-format-compatible
      - build-linux-arm-test-full
@@ -39,3 +39,6 @@ jobs:

    - name: Compare buckify output
      run: make check-buck-targets
+
+    - name: Simple source code checks
+      run: make check-sources
@@ -624,6 +624,7 @@ find_package(Threads REQUIRED)
 set(SOURCES
        cache/cache.cc
        cache/cache_entry_roles.cc
+        cache/cache_reservation_manager.cc
        cache/clock_cache.cc
        cache/lru_cache.cc
        cache/sharded_cache.cc
@@ -635,6 +636,7 @@ set(SOURCES
        db/blob/blob_file_garbage.cc
        db/blob/blob_file_meta.cc
        db/blob/blob_file_reader.cc
+        db/blob/blob_garbage_meter.cc
        db/blob/blob_log_format.cc
        db/blob/blob_log_sequential_reader.cc
        db/blob/blob_log_writer.cc
@@ -711,6 +713,7 @@ set(SOURCES
        env/file_system_tracer.cc
        env/fs_remap.cc
        env/mock_env.cc
+        env/unique_id.cc
        file/delete_scheduler.cc
        file/file_prefetch_buffer.cc
        file/file_util.cc
@@ -815,9 +818,12 @@ set(SOURCES
        tools/ldb_tool.cc
        tools/sst_dump_tool.cc
        tools/trace_analyzer_tool.cc
-        trace_replay/trace_replay.cc
        trace_replay/block_cache_tracer.cc
        trace_replay/io_tracer.cc
+        trace_replay/trace_record_handler.cc
+        trace_replay/trace_record_result.cc
+        trace_replay/trace_record.cc
+        trace_replay/trace_replay.cc
        util/coding.cc
        util/compaction_job_stats_impl.cc
        util/comparator.cc
@@ -830,6 +836,7 @@ set(SOURCES
        util/random.cc
        util/rate_limiter.cc
        util/ribbon_config.cc
+        util/regex.cc
        util/slice.cc
        util/file_checksum_helper.cc
        util/status.cc
@@ -848,6 +855,7 @@ set(SOURCES
        utilities/cassandra/format.cc
        utilities/cassandra/merge_operator.cc
        utilities/checkpoint/checkpoint_impl.cc
+        utilities/compaction_filters.cc
        utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc
        utilities/debug.cc
        utilities/env_mirror.cc
@@ -856,6 +864,7 @@ set(SOURCES
        utilities/fault_injection_fs.cc
        utilities/leveldb_options/leveldb_options.cc
        utilities/memory/memory_util.cc
+        utilities/merge_operators.cc
        utilities/merge_operators/bytesxor.cc
        utilities/merge_operators/max.cc
        utilities/merge_operators/put.cc
@@ -875,6 +884,7 @@ set(SOURCES
        utilities/simulator_cache/sim_cache.cc
        utilities/table_properties_collectors/compact_on_deletion_collector.cc
        utilities/trace/file_trace_reader_writer.cc
+        utilities/trace/replayer_impl.cc
        utilities/transactions/lock/lock_manager.cc
        utilities/transactions/lock/point/point_lock_tracker.cc
        utilities/transactions/lock/point/point_lock_manager.cc
@@ -933,13 +943,8 @@ if(WIN32)
    port/win/env_win.cc
    port/win/env_default.cc
    port/win/port_win.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()
+    port/win/win_logger.cc
+    port/win/win_thread.cc)
 if(WITH_XPRESS)
  list(APPEND SOURCES
    port/win/xpress_win.cc)
@@ -1121,13 +1126,16 @@ if(WITH_TESTS)
  )
  if(WITH_ALL_TESTS)
    list(APPEND TESTS
+        cache/cache_reservation_manager_test.cc
        cache/cache_test.cc
        cache/lru_cache_test.cc
+        db/blob/blob_counting_iterator_test.cc
        db/blob/blob_file_addition_test.cc
        db/blob/blob_file_builder_test.cc
        db/blob/blob_file_cache_test.cc
        db/blob/blob_file_garbage_test.cc
        db/blob/blob_file_reader_test.cc
+        db/blob/blob_garbage_meter_test.cc
        db/blob/db_blob_basic_test.cc
        db/blob/db_blob_compaction_test.cc
        db/blob/db_blob_corruption_test.cc
@@ -1424,3 +1432,8 @@ option(WITH_EXAMPLES "build with examples" OFF)
 if(WITH_EXAMPLES)
  add_subdirectory(examples)
 endif()
+
+option(WITH_BENCHMARK "build benchmark tests" OFF)
+if(WITH_BENCHMARK)
+  add_subdirectory(${PROJECT_SOURCE_DIR}/microbench/)
+endif()
@@ -1,21 +1,123 @@
 # Rocksdb Change Log
-## 6.22.3 (2021-07-19)
+## 6.25.3 (2021-10-14)
 ### Bug Fixes
-* Fixed block cache entry stat scans not to hold the DB mutex, which was a serious performance bug for tail latencies in TransactionDB and likely elsewhere.
+* Fixed bug in calls to `IngestExternalFiles()` with files for multiple column families. The bug could have introduced a delay in ingested file keys becoming visible after `IngestExternalFiles()` returned. Furthermore, mutations to ingested file keys while they were invisible could have been dropped (not necessarily immediately).
+* Fixed a possible race condition impacting users of `WriteBufferManager` who constructed it with `allow_stall == true`. The race condition led to undefined behavior (in our experience, typically a process crash).
+* Fixed a bug where stalled writes would remain stalled forever after the user calls `WriteBufferManager::SetBufferSize()` with `new_size == 0` to dynamically disable memory limiting.

-## 6.22.2 (2021-07-14)
+## 6.25.2 (2021-10-11)
 ### Bug Fixes
-* Fix continuous logging of an existing background error on every user write
+* Fix `DisableManualCompaction()` to cancel compactions even when they are waiting on automatic compactions to drain due to `CompactRangeOptions::exclusive_manual_compactions == true`.
+* Fix contract of `Env::ReopenWritableFile()` and `FileSystem::ReopenWritableFile()` to specify any existing file must not be deleted or truncated.

-## 6.22.1 (2021-06-25)
+## 6.25.1 (2021-09-28)
 ### Bug Fixes
+* Fixes a bug in directed IO mode when calling MultiGet() for blobs in the same blob file. The bug is caused by not sorting the blob read requests by file offsets.
+
+## 6.25.0 (2021-09-20)
+### Bug Fixes
+* Allow secondary instance to refresh iterator. Assign read seq after referencing SuperVersion.
+* Fixed a bug of secondary instance's last_sequence going backward, and reads on the secondary fail to see recent updates from the primary.
+* Fixed a bug that could lead to duplicate DB ID or DB session ID in POSIX environments without /proc/sys/kernel/random/uuid.
+* Fix a race in DumpStats() with column family destruction due to not taking a Ref on each entry while iterating the ColumnFamilySet.
+* Fix a race in item ref counting in LRUCache when promoting an item from the SecondaryCache.
+* Fix a race in BackupEngine if RateLimiter is reconfigured during concurrent Restore operations.
+* Fix a bug on POSIX in which failure to create a lock file (e.g. out of space) can prevent future LockFile attempts in the same process on the same file from succeeding.
+* Fix a bug that backup_rate_limiter and restore_rate_limiter in BackupEngine could not limit read rates.
+* Fix the implementation of `prepopulate_block_cache = kFlushOnly` to only apply to flushes rather than to all generated files.
+* Fix WAL log data corruption when using DBOptions.manual_wal_flush(true) and WriteOptions.sync(true) together. The sync WAL should work with locked log_write_mutex_.
+* Add checks for validity of the IO uring completion queue entries, and fail the BlockBasedTableReader MultiGet sub-batch if there's an invalid completion
+* Add an interface RocksDbIOUringEnable() that, if defined by the user, will allow them to enable/disable the use of IO uring by RocksDB
+* Fix the bug that when direct I/O is used and MultiRead() returns a short result, RandomAccessFileReader::MultiRead() still returns full size buffer, with returned short value together with some data in original buffer. This bug is unlikely cause incorrect results, because (1) since FileSystem layer is expected to retry on short result, returning short results is only possible when asking more bytes in the end of the file, which RocksDB doesn't do when using MultiRead(); (2) checksum is unlikely to match.
+
+### New Features
+* RemoteCompaction's interface now includes `db_name`, `db_id`, `session_id`, which could help the user uniquely identify compaction job between db instances and sessions.
+* Added a ticker statistic, "rocksdb.verify_checksum.read.bytes", reporting how many bytes were read from file to serve `VerifyChecksum()` and `VerifyFileChecksums()` queries.
+* Added ticker statistics, "rocksdb.backup.read.bytes" and "rocksdb.backup.write.bytes", reporting how many bytes were read and written during backup.
+* Added properties for BlobDB: `rocksdb.num-blob-files`, `rocksdb.blob-stats`, `rocksdb.total-blob-file-size`, and `rocksdb.live-blob-file-size`. The existing property `rocksdb.estimate_live-data-size` was also extended to include live bytes residing in blob files.
+* Added two new RateLimiter IOPriorities: `Env::IO_USER`,`Env::IO_MID`. `Env::IO_USER` will have superior priority over all other RateLimiter IOPriorities without being subject to fair scheduling constraint.
+* `SstFileWriter` now supports `Put`s and `Delete`s with user-defined timestamps. Note that the ingestion logic itself is not timestamp-aware yet.
+* Allow a single write batch to include keys from multiple column families whose timestamps' formats can differ. For example, some column families may disable timestamp, while others enable timestamp.
+* Add compaction priority information in RemoteCompaction, which can be used to schedule high priority job first.
+* Added new callback APIs `OnBlobFileCreationStarted`,`OnBlobFileCreated`and `OnBlobFileDeleted` in `EventListener` class of listener.h. It notifies listeners during creation/deletion of individual blob files in Integrated BlobDB. It also log blob file creation finished event and deletion event in LOG file.
+* Batch blob read requests for `DB::MultiGet` using `MultiRead`.
+* Add support for fallback to local compaction, the user can return `CompactionServiceJobStatus::kUseLocal` to instruct RocksDB to run the compaction locally instead of waiting for the remote compaction result.
+* Add built-in rate limiter's implementation of `RateLimiter::GetTotalPendingRequest(int64_t* total_pending_requests, const Env::IOPriority pri)` for the total number of requests that are pending for bytes in the rate limiter.
+* Charge memory usage during data buffering, from which training samples are gathered for dictionary compression, to block cache. Unbuffering data can now be triggered if the block cache becomes full and `strict_capacity_limit=true` for the block cache, in addition to existing conditions that can trigger unbuffering.
+
+### Public API change
+* Remove obsolete implementation details FullKey and ParseFullKey from public API
+* Change `SstFileMetaData::size` from `size_t` to `uint64_t`.
+* Made Statistics extend the Customizable class and added a CreateFromString method.  Implementations of Statistics need to be registered with the ObjectRegistry and to implement a Name() method in order to be created via this method.
+* Extended `FlushJobInfo` and `CompactionJobInfo` in listener.h to provide information about the blob files generated by a flush/compaction and garbage collected during compaction in Integrated BlobDB. Added struct members `blob_file_addition_infos` and `blob_file_garbage_infos` that contain this information.
+* Extended parameter `output_file_names` of `CompactFiles` API to also include paths of the blob files generated by the compaction in Integrated BlobDB.
+* Most `BackupEngine` functions now return `IOStatus` instead of `Status`. Most existing code should be compatible with this change but some calls might need to be updated.
+
+## 6.24.0 (2021-08-20)
+### Bug Fixes
+* If the primary's CURRENT file is missing or inaccessible, the secondary instance should not hang repeatedly trying to switch to a new MANIFEST. It should instead return the error code encountered while accessing the file.
+* Restoring backups with BackupEngine is now a logically atomic operation, so that if a restore operation is interrupted, DB::Open on it will fail. Using BackupEngineOptions::sync (default) ensures atomicity even in case of power loss or OS crash.
+* Fixed a race related to the destruction of `ColumnFamilyData` objects. The earlier logic unlocked the DB mutex before destroying the thread-local `SuperVersion` pointers, which could result in a process crash if another thread managed to get a reference to the `ColumnFamilyData` object.
+* Removed a call to `RenameFile()` on a non-existent info log file ("LOG") when opening a new DB. Such a call was guaranteed to fail though did not impact applications since we swallowed the error. Now we also stopped swallowing errors in renaming "LOG" file.
+* Fixed an issue where `OnFlushCompleted` was not called for atomic flush.
+* Fixed a bug affecting the batched `MultiGet` API when used with keys spanning multiple column families and `sorted_input == false`.
+* Fixed a potential incorrect result in opt mode and assertion failures caused by releasing snapshot(s) during compaction.
+* Fixed passing of BlobFileCompletionCallback to Compaction job and Atomic flush job which was default paramter (nullptr). BlobFileCompletitionCallback is internal callback that manages addition of blob files to SSTFileManager.
+* Fixed MultiGet not updating the block_read_count and block_read_byte PerfContext counters.
+
+### New Features
+* Made the EventListener extend the Customizable class.
+* EventListeners that have a non-empty Name() and that are registered with the ObjectRegistry can now be serialized to/from the OPTIONS file.
+* Insert warm blocks (data blocks, uncompressed dict blocks, index and filter blocks) in Block cache during flush under option BlockBasedTableOptions.prepopulate_block_cache. Previously it was enabled for only data blocks.
+* BlockBasedTableOptions.prepopulate_block_cache can be dynamically configured using DB::SetOptions.
+* Add CompactionOptionsFIFO.age_for_warm, which allows RocksDB to move old files to warm tier in FIFO compactions. Note that file temperature is still an experimental feature.
+* Add a comment to suggest btrfs user to disable file preallocation by setting `options.allow_fallocate=false`.
+* Fast forward option in Trace replay changed to double type to allow replaying at a lower speed, by settings the value between 0 and 1. This option can be set via `ReplayOptions` in `Replayer::Replay()`, or via `--trace_replay_fast_forward` in db_bench.
+* Add property `LiveSstFilesSizeAtTemperature` to retrieve sst file size at different temperature.
+* Added a stat rocksdb.secondary.cache.hits.
+* Added a PerfContext counter secondary_cache_hit_count.
+* The integrated BlobDB implementation now supports the tickers `BLOB_DB_BLOB_FILE_BYTES_READ`, `BLOB_DB_GC_NUM_KEYS_RELOCATED`, and `BLOB_DB_GC_BYTES_RELOCATED`, as well as the histograms `BLOB_DB_COMPRESSION_MICROS` and `BLOB_DB_DECOMPRESSION_MICROS`.
+* Added hybrid configuration of Ribbon filter and Bloom filter where some LSM levels use Ribbon for memory space efficiency and some use Bloom for speed. See NewRibbonFilterPolicy. This also changes the default behavior of NewRibbonFilterPolicy to use Bloom for flushes under Leveled and Universal compaction and Ribbon otherwise. The C API function `rocksdb_filterpolicy_create_ribbon` is unchanged but adds new `rocksdb_filterpolicy_create_ribbon_hybrid`.
+
+### Public API change
+* Added APIs to decode and replay trace file via Replayer class. Added `DB::NewDefaultReplayer()` to create a default Replayer instance. Added `TraceReader::Reset()` to restart reading a trace file. Created trace_record.h, trace_record_result.h and utilities/replayer.h files to access the decoded Trace records, replay them, and query the actual operation results.
+* Added Configurable::GetOptionsMap to the public API for use in creating new Customizable classes.
+* Generalized bits_per_key parameters in C API from int to double for greater configurability. Although this is a compatible change for existing C source code, anything depending on C API signatures, such as foreign function interfaces, will need to be updated.
+
+### Performance Improvements
+* Try to avoid updating DBOptions if `SetDBOptions()` does not change any option value.
+
+### Behavior Changes
+* `StringAppendOperator` additionally accepts a string as the delimiter.
+* BackupEngineOptions::sync (default true) now applies to restoring backups in addition to creating backups. This could slow down restores, but ensures they are fully persisted before returning OK. (Consider increasing max_background_operations to improve performance.)
+
+## 6.23.0 (2021-07-16)
+### Behavior Changes
+* Obsolete keys in the bottommost level that were preserved for a snapshot will now be cleaned upon snapshot release in all cases. This form of compaction (snapshot release triggered compaction) previously had an artificial limitation that multiple tombstones needed to be present.
+### Bug Fixes
+* Blob file checksums are now printed in hexadecimal format when using the `manifest_dump` `ldb` command.
 * `GetLiveFilesMetaData()` now populates the `temperature`, `oldest_ancester_time`, and `file_creation_time` fields of its `LiveFileMetaData` results when the information is available. Previously these fields always contained zero indicating unknown.
+* Fix mismatches of OnCompaction{Begin,Completed} in case of DisableManualCompaction().
+* Fix continuous logging of an existing background error on every user write
+* Fix a bug that `Get()` return Status::OK() and an empty value for non-existent key when `read_options.read_tier = kBlockCacheTier`.
+* Fix a bug that stat in `get_context` didn't accumulate to statistics when query is failed.
+* Fixed handling of DBOptions::wal_dir with LoadLatestOptions() or ldb --try_load_options on a copied or moved DB. Previously, when the WAL directory is same as DB directory (default), a copied or moved DB would reference the old path of the DB as the WAL directory, potentially corrupting both copies. Under this change, the wal_dir from DB::GetOptions() or LoadLatestOptions() may now be empty, indicating that the current DB directory is used for WALs. This is also a subtle API change.
+
+### New Features
+* ldb has a new feature, `list_live_files_metadata`, that shows the live SST files, as well as their LSM storage level and the column family they belong to.
+* The new BlobDB implementation now tracks the amount of garbage in each blob file in the MANIFEST.
+* Integrated BlobDB now supports Merge with base values (Put/Delete etc.).
+* RemoteCompaction supports sub-compaction, the job_id in the user interface is changed from `int` to `uint64_t` to support sub-compaction id.
+* Expose statistics option in RemoteCompaction worker.
+
+### Public API change
+* Added APIs to the Customizable class to allow developers to create their own Customizable classes.  Created the utilities/customizable_util.h file to contain helper methods for developing new Customizable classes.
+* Change signature of SecondaryCache::Name().  Make SecondaryCache customizable and add SecondaryCache::CreateFromString method.

 ## 6.22.0 (2021-06-18)
 ### Behavior Changes
 * Added two additional tickers, MEMTABLE_PAYLOAD_BYTES_AT_FLUSH and MEMTABLE_GARBAGE_BYTES_AT_FLUSH. These stats can be used to estimate the ratio of "garbage" (outdated) bytes in the memtable that are discarded at flush time.
 * Added API comments clarifying safe usage of Disable/EnableManualCompaction and EventListener callbacks for compaction.
-
 ### Bug Fixes
 * fs_posix.cc GetFreeSpace() always report disk space available to root even when running as non-root.  Linux defaults often have disk mounts with 5 to 10 percent of total space reserved only for root.  Out of space could result for non-root users.
 * Subcompactions are now disabled when user-defined timestamps are used, since the subcompaction boundary picking logic is currently not timestamp-aware, which could lead to incorrect results when different subcompactions process keys that only differ by timestamp.
@@ -29,6 +131,12 @@
 * Add BlockBasedTableOptions.prepopulate_block_cache.  If enabled, it prepopulate warm/hot data blocks which are already in memory into block cache at the time of flush. On a flush, the data block that is in memory (in memtables) get flushed to the device. If using Direct IO, additional IO is incurred to read this data back into memory again, which is avoided by enabling this option and it also helps with Distributed FileSystem. More details in include/rocksdb/table.h.
 * Added a `cancel` field to `CompactRangeOptions`, allowing individual in-process manual range compactions to be cancelled.

+### New Features
+* Added BlobMetaData to the ColumnFamilyMetaData to return information about blob files
+
+### Public API change
+* Added GetAllColumnFamilyMetaData API to retrieve the ColumnFamilyMetaData about all column families.
+
 ## 6.21.0 (2021-05-21)
 ### Bug Fixes
 * Fixed a bug in handling file rename error in distributed/network file systems when the server succeeds but client returns error. The bug can cause CURRENT file to point to non-existing MANIFEST file, thus DB cannot be opened.
@@ -1,6 +1,6 @@
 This is the list of all known third-party language bindings for RocksDB. If something is missing, please open a pull request to add it.

-* Java - https://github.com/facebook/rocksdb/tree/master/java
+* Java - https://github.com/facebook/rocksdb/tree/main/java
 * Python
    * http://python-rocksdb.readthedocs.io/en/latest/
    * http://pyrocksdb.readthedocs.org/en/latest/ (unmaintained)
@@ -222,6 +222,7 @@ am__v_AR_1 =

 ifdef ROCKSDB_USE_LIBRADOS
 LIB_SOURCES += utilities/env_librados.cc
+TEST_MAIN_SOURCES += utilities/env_librados_test.cc
 LDFLAGS += -lrados
 endif

@@ -344,6 +345,12 @@ ifdef ROCKSDB_VALGRIND_RUN
 	PLATFORM_CCFLAGS += -DROCKSDB_VALGRIND_RUN
 	PLATFORM_CXXFLAGS += -DROCKSDB_VALGRIND_RUN
 endif
+ifdef ROCKSDB_FULL_VALGRIND_RUN
+	# Some tests are slow when run under valgrind and are only run when
+	# explicitly requested via the ROCKSDB_FULL_VALGRIND_RUN compiler flag.
+	PLATFORM_CCFLAGS += -DROCKSDB_VALGRIND_RUN -DROCKSDB_FULL_VALGRIND_RUN
+	PLATFORM_CXXFLAGS += -DROCKSDB_VALGRIND_RUN -DROCKSDB_FULL_VALGRIND_RUN
+endif

 ifndef DISABLE_JEMALLOC
 	ifdef JEMALLOC
@@ -488,6 +495,7 @@ VALGRIND_ERROR = 2
 VALGRIND_VER := $(join $(VALGRIND_VER),valgrind)

 VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full
+# Not yet supported: --show-leak-kinds=definite,possible,reachable --errors-for-leak-kinds=definite,possible,reachable

 TEST_OBJECTS = $(patsubst %.cc, $(OBJ_DIR)/%.o, $(TEST_LIB_SOURCES) $(MOCK_LIB_SOURCES)) $(GTEST)
 BENCH_OBJECTS = $(patsubst %.cc, $(OBJ_DIR)/%.o, $(BENCH_LIB_SOURCES))
@@ -509,6 +517,29 @@ ifeq ($(USE_FOLLY_DISTRIBUTED_MUTEX),1)
 	ALL_SOURCES += third-party/folly/folly/synchronization/test/DistributedMutexTest.cc
 endif

+# `make check-headers` to very that each header file includes its own
+# dependencies
+ifneq ($(filter check-headers, $(MAKECMDGOALS)),)
+# TODO: add/support JNI headers
+	DEV_HEADER_DIRS := $(sort include/ hdfs/ $(dir $(ALL_SOURCES)))
+# Some headers like in port/ are platform-specific
+	DEV_HEADERS := $(shell $(FIND) $(DEV_HEADER_DIRS) -type f -name '*.h' | egrep -v 'port/|plugin/|lua/|range_tree/|tools/rdb/db_wrapper.h|include/rocksdb/utilities/env_librados.h')
+else
+	DEV_HEADERS :=
+endif
+HEADER_OK_FILES = $(patsubst %.h, %.h.ok, $(DEV_HEADERS))
+
+AM_V_CCH = $(am__v_CCH_$(V))
+am__v_CCH_ = $(am__v_CCH_$(AM_DEFAULT_VERBOSITY))
+am__v_CCH_0 = @echo "  CC.h    " $<;
+am__v_CCH_1 =
+
+%.h.ok: %.h # .h.ok not actually created, so re-checked on each invocation
+# -DROCKSDB_NAMESPACE=42 ensures the namespace header is included
+	$(AM_V_CCH) echo '#include "$<"' | $(CXX) $(CXXFLAGS) -DROCKSDB_NAMESPACE=42 -x c++ -c - -o /dev/null
+
+check-headers: $(HEADER_OK_FILES)
+
 # 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')
@@ -518,7 +549,6 @@ ifdef ASSERT_STATUS_CHECKED
 	TESTS_FAILING_ASC = \
 		c_test \
 		db_test \
-		db_test2 \
 		env_test \
 		range_locking_test \
 		testutil_test \
@@ -595,6 +625,8 @@ TEST_LIBS = \
 # TODO: add back forward_iterator_bench, after making it build in all environemnts.
 BENCHMARKS = $(patsubst %.cc, %, $(notdir $(BENCH_MAIN_SOURCES)))

+MICROBENCHS = $(patsubst %.cc, %, $(notdir $(MICROBENCH_SOURCES)))
+
 # if user didn't config LIBNAME, set the default
 ifeq ($(LIBNAME),)
  LIBNAME=librocksdb
@@ -708,7 +740,7 @@ endif  # PLATFORM_SHARED_EXT
 .PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests package \
 	release tags tags0 valgrind_check whitebox_crash_test format static_lib shared_lib all \
 	dbg rocksdbjavastatic rocksdbjava gen-pc install install-static install-shared uninstall \
-	analyze tools tools_lib \
+	analyze tools tools_lib check-headers \
 	blackbox_crash_test_with_atomic_flush whitebox_crash_test_with_atomic_flush  \
 	blackbox_crash_test_with_txn whitebox_crash_test_with_txn \
 	blackbox_crash_test_with_best_efforts_recovery \
@@ -733,6 +765,9 @@ test_libs: $(TEST_LIBS)

 benchmarks: $(BENCHMARKS)

+microbench: $(MICROBENCHS)
+	for t in $(MICROBENCHS); do echo "===== Running benchmark $$t (`date`)"; ./$$t || exit 1; done;
+
 dbg: $(LIBRARY) $(BENCHMARKS) tools $(TESTS)

 # creates library and programs
@@ -932,6 +967,7 @@ endif
 ifndef SKIP_FORMAT_BUCK_CHECKS
 	$(MAKE) check-format
 	$(MAKE) check-buck-targets
+	$(MAKE) check-sources
 endif

 # TODO add ldb_tests
@@ -1046,6 +1082,12 @@ ubsan_crash_test_with_best_efforts_recovery: clean
 	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test_with_best_efforts_recovery
 	$(MAKE) clean

+full_valgrind_test:
+	ROCKSDB_FULL_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 $(MAKE) valgrind_check
+
+full_valgrind_test_some:
+	ROCKSDB_FULL_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 $(MAKE) valgrind_check_some
+
 valgrind_test:
 	ROCKSDB_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 $(MAKE) valgrind_check

@@ -1164,7 +1206,7 @@ clean-not-downloaded: clean-ext-libraries-bin clean-rocks clean-not-downloaded-r
 clean-rocks:
 	echo shared=$(ALL_SHARED_LIBS)
 	echo static=$(ALL_STATIC_LIBS)
-	rm -f $(BENCHMARKS) $(TOOLS) $(TESTS) $(PARALLEL_TEST) $(ALL_STATIC_LIBS) $(ALL_SHARED_LIBS)
+	rm -f $(BENCHMARKS) $(TOOLS) $(TESTS) $(PARALLEL_TEST) $(ALL_STATIC_LIBS) $(ALL_SHARED_LIBS) $(MICROBENCHS)
 	rm -rf $(CLEAN_FILES) ios-x86 ios-arm scan_build_report
 	$(FIND) . -name "*.[oda]" -exec rm -f {} \;
 	$(FIND) . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm -f {} \;
@@ -1203,6 +1245,9 @@ check-format:
 check-buck-targets:
 	buckifier/check_buck_targets.sh

+check-sources:
+	build_tools/check-sources.sh
+
 package:
 	bash build_tools/make_package.sh $(SHARED_MAJOR).$(SHARED_MINOR)

@@ -1827,6 +1872,9 @@ block_cache_trace_analyzer_test: $(OBJ_DIR)/tools/block_cache_analyzer/block_cac
 defer_test: $(OBJ_DIR)/util/defer_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

+blob_counting_iterator_test: $(OBJ_DIR)/db/blob/blob_counting_iterator_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
+
 blob_file_addition_test: $(OBJ_DIR)/db/blob/blob_file_addition_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

@@ -1842,6 +1890,9 @@ blob_file_garbage_test: $(OBJ_DIR)/db/blob/blob_file_garbage_test.o $(TEST_LIBRA
 blob_file_reader_test: $(OBJ_DIR)/db/blob/blob_file_reader_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

+blob_garbage_meter_test: $(OBJ_DIR)/db/blob/blob_garbage_meter_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
+
 timer_test: $(OBJ_DIR)/util/timer_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

@@ -1872,6 +1923,11 @@ db_write_buffer_manager_test: $(OBJ_DIR)/db/db_write_buffer_manager_test.o $(TES
 clipping_iterator_test: $(OBJ_DIR)/db/compaction/clipping_iterator_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

+ribbon_bench: $(OBJ_DIR)/microbench/ribbon_bench.o $(LIBRARY)
+	$(AM_LINK)
+
+cache_reservation_manager_test: $(OBJ_DIR)/cache/cache_reservation_manager_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
 #-------------------------------------------------
 # make install related stuff
 PREFIX ?= /usr/local
@@ -2336,7 +2392,7 @@ build_subset_tests: $(ROCKSDBTESTS_SUBSET)

 # Remove the rules for which dependencies should not be generated and see if any are left.
 #If so, include the dependencies; if not, do not include the dependency files
-ROCKS_DEP_RULES=$(filter-out clean format check-format check-buck-targets jclean jtest package analyze tags rocksdbjavastatic% unity.% unity_test, $(MAKECMDGOALS))
+ROCKS_DEP_RULES=$(filter-out clean format check-format check-buck-targets check-headers check-sources jclean jtest package analyze tags rocksdbjavastatic% unity.% unity_test, $(MAKECMDGOALS))
 ifneq ("$(ROCKS_DEP_RULES)", "")
 -include $(DEPFILES)
 endif
@@ -1,8 +1,8 @@
 ## RocksDB: A Persistent Key-Value Store for Flash and RAM Storage

 [![CircleCI Status](https://circleci.com/gh/facebook/rocksdb.svg?style=svg)](https://circleci.com/gh/facebook/rocksdb)
-[![TravisCI Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
-[![Appveyor Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
+[![TravisCI Status](https://api.travis-ci.com/facebook/rocksdb.svg?branch=main)](https://travis-ci.com/github/facebook/rocksdb)
+[![Appveyor Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/main?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/main)
 [![PPC64le Build Status](http://140-211-168-68-openstack.osuosl.org:8080/buildStatus/icon?job=rocksdb&style=plastic)](http://140-211-168-68-openstack.osuosl.org:8080/job/rocksdb)

 RocksDB is developed and maintained by Facebook Database Engineering Team.
@@ -17,7 +17,7 @@ and Space-Amplification-Factor (SAF). It has multi-threaded compactions,
 making it especially suitable for storing multiple terabytes of data in a
 single database.

-Start with example usage here: https://github.com/facebook/rocksdb/tree/master/examples
+Start with example usage here: https://github.com/facebook/rocksdb/tree/main/examples

 See the [github wiki](https://github.com/facebook/rocksdb/wiki) for more explanation.

@@ -133,6 +133,7 @@ cpp_library(
    srcs = [
        "cache/cache.cc",
        "cache/cache_entry_roles.cc",
+        "cache/cache_reservation_manager.cc",
        "cache/clock_cache.cc",
        "cache/lru_cache.cc",
        "cache/sharded_cache.cc",
@@ -144,6 +145,7 @@ cpp_library(
        "db/blob/blob_file_garbage.cc",
        "db/blob/blob_file_meta.cc",
        "db/blob/blob_file_reader.cc",
+        "db/blob/blob_garbage_meter.cc",
        "db/blob/blob_log_format.cc",
        "db/blob/blob_log_sequential_reader.cc",
        "db/blob/blob_log_writer.cc",
@@ -223,6 +225,7 @@ cpp_library(
        "env/fs_remap.cc",
        "env/io_posix.cc",
        "env/mock_env.cc",
+        "env/unique_id.cc",
        "file/delete_scheduler.cc",
        "file/file_prefetch_buffer.cc",
        "file/file_util.cc",
@@ -334,6 +337,9 @@ cpp_library(
        "tools/sst_dump_tool.cc",
        "trace_replay/block_cache_tracer.cc",
        "trace_replay/io_tracer.cc",
+        "trace_replay/trace_record.cc",
+        "trace_replay/trace_record_handler.cc",
+        "trace_replay/trace_record_result.cc",
        "trace_replay/trace_replay.cc",
        "util/build_version.cc",
        "util/coding.cc",
@@ -349,6 +355,7 @@ cpp_library(
        "util/murmurhash.cc",
        "util/random.cc",
        "util/rate_limiter.cc",
+        "util/regex.cc",
        "util/ribbon_config.cc",
        "util/slice.cc",
        "util/status.cc",
@@ -367,6 +374,7 @@ cpp_library(
        "utilities/cassandra/format.cc",
        "utilities/cassandra/merge_operator.cc",
        "utilities/checkpoint/checkpoint_impl.cc",
+        "utilities/compaction_filters.cc",
        "utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc",
        "utilities/convenience/info_log_finder.cc",
        "utilities/debug.cc",
@@ -376,6 +384,7 @@ cpp_library(
        "utilities/fault_injection_fs.cc",
        "utilities/leveldb_options/leveldb_options.cc",
        "utilities/memory/memory_util.cc",
+        "utilities/merge_operators.cc",
        "utilities/merge_operators/bytesxor.cc",
        "utilities/merge_operators/max.cc",
        "utilities/merge_operators/put.cc",
@@ -395,6 +404,7 @@ cpp_library(
        "utilities/simulator_cache/sim_cache.cc",
        "utilities/table_properties_collectors/compact_on_deletion_collector.cc",
        "utilities/trace/file_trace_reader_writer.cc",
+        "utilities/trace/replayer_impl.cc",
        "utilities/transactions/lock/lock_manager.cc",
        "utilities/transactions/lock/point/point_lock_manager.cc",
        "utilities/transactions/lock/point/point_lock_tracker.cc",
@@ -445,6 +455,7 @@ cpp_library(
    srcs = [
        "cache/cache.cc",
        "cache/cache_entry_roles.cc",
+        "cache/cache_reservation_manager.cc",
        "cache/clock_cache.cc",
        "cache/lru_cache.cc",
        "cache/sharded_cache.cc",
@@ -456,6 +467,7 @@ cpp_library(
        "db/blob/blob_file_garbage.cc",
        "db/blob/blob_file_meta.cc",
        "db/blob/blob_file_reader.cc",
+        "db/blob/blob_garbage_meter.cc",
        "db/blob/blob_log_format.cc",
        "db/blob/blob_log_sequential_reader.cc",
        "db/blob/blob_log_writer.cc",
@@ -535,6 +547,7 @@ cpp_library(
        "env/fs_remap.cc",
        "env/io_posix.cc",
        "env/mock_env.cc",
+        "env/unique_id.cc",
        "file/delete_scheduler.cc",
        "file/file_prefetch_buffer.cc",
        "file/file_util.cc",
@@ -646,6 +659,9 @@ cpp_library(
        "tools/sst_dump_tool.cc",
        "trace_replay/block_cache_tracer.cc",
        "trace_replay/io_tracer.cc",
+        "trace_replay/trace_record.cc",
+        "trace_replay/trace_record_handler.cc",
+        "trace_replay/trace_record_result.cc",
        "trace_replay/trace_replay.cc",
        "util/build_version.cc",
        "util/coding.cc",
@@ -661,6 +677,7 @@ cpp_library(
        "util/murmurhash.cc",
        "util/random.cc",
        "util/rate_limiter.cc",
+        "util/regex.cc",
        "util/ribbon_config.cc",
        "util/slice.cc",
        "util/status.cc",
@@ -679,6 +696,7 @@ cpp_library(
        "utilities/cassandra/format.cc",
        "utilities/cassandra/merge_operator.cc",
        "utilities/checkpoint/checkpoint_impl.cc",
+        "utilities/compaction_filters.cc",
        "utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc",
        "utilities/convenience/info_log_finder.cc",
        "utilities/debug.cc",
@@ -688,6 +706,7 @@ cpp_library(
        "utilities/fault_injection_fs.cc",
        "utilities/leveldb_options/leveldb_options.cc",
        "utilities/memory/memory_util.cc",
+        "utilities/merge_operators.cc",
        "utilities/merge_operators/bytesxor.cc",
        "utilities/merge_operators/max.cc",
        "utilities/merge_operators/put.cc",
@@ -707,6 +726,7 @@ cpp_library(
        "utilities/simulator_cache/sim_cache.cc",
        "utilities/table_properties_collectors/compact_on_deletion_collector.cc",
        "utilities/trace/file_trace_reader_writer.cc",
+        "utilities/trace/replayer_impl.cc",
        "utilities/transactions/lock/lock_manager.cc",
        "utilities/transactions/lock/point/point_lock_manager.cc",
        "utilities/transactions/lock/point/point_lock_tracker.cc",
@@ -906,6 +926,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "blob_counting_iterator_test",
+        "db/blob/blob_counting_iterator_test.cc",
+        "parallel",
+        [],
+        [],
+    ],
    [
        "blob_db_test",
        "utilities/blob_db/blob_db_test.cc",
@@ -948,6 +975,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "blob_garbage_meter_test",
+        "db/blob/blob_garbage_meter_test.cc",
+        "parallel",
+        [],
+        [],
+    ],
    [
        "block_based_filter_block_test",
        "table/block_based/block_based_filter_block_test.cc",
@@ -997,6 +1031,13 @@ ROCKS_TESTS = [
        [],
        [],
    ],
+    [
+        "cache_reservation_manager_test",
+        "cache/cache_reservation_manager_test.cc",
+        "parallel",
+        [],
+        [],
+    ],
    [
        "cache_simulator_test",
        "utilities/simulator_cache/cache_simulator_test.cc",
@@ -115,3 +115,12 @@ LzLabs is using RocksDB as a storage engine in their multi-database distributed
 ## ArangoDB
 [ArangoDB](https://www.arangodb.com/) is a native multi-model database with flexible data models for documents, graphs, and key-values, for building high performance applications using a convenient SQL-like query language or JavaScript extensions. It uses RocksDB as its sotrage engine.

+## Milvus
+[Milvus](https://milvus.io/) is an open source vector database for unstructured data. It uses RocksDB not only as one of the supported kv storage engines, but also as a message queue.
+
+## Kafka
+[Kafka](https://kafka.apache.org/) is an open-source distributed event streaming platform, it uses RocksDB to store state in Kafka Streams: https://www.confluent.io/blog/how-to-tune-rocksdb-kafka-streams-state-stores-performance/.
+
+## Others
+More databases using RocksDB can be found at [dbdb.io](https://dbdb.io/browse?embeds=rocksdb).
+
@@ -24,7 +24,7 @@ We strive to achieve the following goals:
 * make all unit test pass both in debug and release builds. 
  * Note: latest introduction of SyncPoint seems to disable running db_test in Release.
 * make performance on par with published benchmarks accounting for HW differences
-* we would like to keep the port code inline with the master branch with no forking
+* we would like to keep the port code inline with the main branch with no forking

 ## Build system
 We have chosen CMake as a widely accepted build system to build the Windows port. It is very fast and convenient. 
@@ -66,7 +66,7 @@ We endeavored to make it functionally on par with posix_env. This means we repli
 Even though Windows provides its own efficient thread-pool implementation we chose to replicate posix logic using `std::thread` primitives. This allows anyone to quickly detect any changes within the posix source code and replicate them within windows env. This has proven to work very well. At the same time for anyone who wishes to replace the built-in thread-pool can do so using RocksDB stackable environments.

 For disk access we implemented all of the functionality present within the posix_env which includes memory mapped files, random access, rate-limiter support etc.
-The `use_os_buffer` flag on Posix platforms currently denotes disabling read-ahead log via `fadvise` mechanism. Windows does not have `fadvise` system call. What is more, it implements disk cache in a way that differs from Linux greatly. It’s not an uncommon practice on Windows to perform un-buffered disk access to gain control of the memory consumption. We think that in our use case this may also be a good configuration option at the expense of disk throughput. To compensate one may increase the configured in-memory cache size instead. Thus we have chosen  `use_os_buffer=false` to disable OS disk buffering for `WinWritableFile` and `WinRandomAccessFile`. The OS imposes restrictions on the alignment of the disk offsets, buffers used and the amount of data that is read/written when accessing files in un-buffered mode. When the option is true, the classes behave in a standard way. This allows to perform writes and reads in cases when un-buffered access does not make sense such as WAL and MANIFEST.
+The `use_os_buffer` flag on Posix platforms currently denotes disabling read-ahead log via `fadvise` mechanism. Windows does not have `fadvise` system call. What is more, it implements disk cache in a way that differs from Linux greatly. It's not an uncommon practice on Windows to perform un-buffered disk access to gain control of the memory consumption. We think that in our use case this may also be a good configuration option at the expense of disk throughput. To compensate one may increase the configured in-memory cache size instead. Thus we have chosen  `use_os_buffer=false` to disable OS disk buffering for `WinWritableFile` and `WinRandomAccessFile`. The OS imposes restrictions on the alignment of the disk offsets, buffers used and the amount of data that is read/written when accessing files in un-buffered mode. When the option is true, the classes behave in a standard way. This allows to perform writes and reads in cases when un-buffered access does not make sense such as WAL and MANIFEST.

 We have replaced `pread/pwrite` with `WriteFile/ReadFile` with `OVERLAPPED` structure so we can atomically seek to the position of the disk operation but still perform the operation synchronously. Thus we able to emulate that functionality of `pread/pwrite` reasonably well. The only difference is that the file pointer is not returned to its original position but that hardly matters given the random nature of access.

@@ -189,9 +189,6 @@ EOF
                COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_IOURING_PRESENT"
            fi
        fi
-        if test -z "$USE_FOLLY_DISTRIBUTED_MUTEX"; then
-          USE_FOLLY_DISTRIBUTED_MUTEX=1
-        fi
        # PORT_FILES=port/linux/linux_specific.cc
        ;;
    SunOS)
@@ -596,6 +593,16 @@ EOF
            PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS -faligned-new -DHAVE_ALIGNED_NEW"
        fi
    fi
+    if ! test $ROCKSDB_DISABLE_BENCHMARK; then
+        # Test whether google benchmark is available
+        $CXX $PLATFORM_CXXFLAGS -x c++ - -o /dev/null -lbenchmark 2>/dev/null  <<EOF
+          #include <benchmark/benchmark.h>
+          int main() {}
+EOF
+        if [ "$?" = 0 ]; then
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbenchmark"
+        fi
+    fi
 fi

 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+#
+# Check for some simple mistakes that should prevent commit or push
+
+BAD=""
+
+git grep 'namespace rocksdb' -- '*.[ch]*'
+if [ "$?" != "1" ]; then
+  echo "^^^^^ Do not hardcode namespace rocksdb. Use ROCKSDB_NAMESPACE"
+  BAD=1
+fi
+
+git grep -i 'nocommit' -- ':!build_tools/check-sources.sh'
+if [ "$?" != "1" ]; then
+  echo "^^^^^ Code was not intended to be committed"
+  BAD=1
+fi
+
+git grep '<rocksdb/' -- ':!build_tools/check-sources.sh'
+if [ "$?" != "1" ]; then
+  echo '^^^^^ Use double-quotes as in #include "rocksdb/something.h"'
+  BAD=1
+fi
+
+if [ "$BAD" ]; then
+  exit 1
+fi
@@ -167,6 +167,4 @@ VALGRIND_VER="$VALGRIND_BASE/bin/"
 LUA_PATH=
 LUA_LIB=

-USE_FOLLY_DISTRIBUTED_MUTEX=1
-
 export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
@@ -102,6 +102,7 @@ export PORTABLE

 BINUTILS="$BINUTILS_BASE/bin"
 AR="$BINUTILS/ar"
+AS="$BINUTILS/as"

 DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE $LIBURING_INCLUDE"

@@ -150,7 +151,6 @@ CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PR
 CXXFLAGS+=" $CFLAGS"

 EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS $LIBURING_LIBS"
-EXEC_LDFLAGS+=" -B$BINUTILS"
 EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform009/lib/ld.so"
 EXEC_LDFLAGS+=" $LIBUNWIND"
 EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform009/lib"
@@ -159,6 +159,7 @@ EXEC_LDFLAGS+=" -Wl,-rpath=$GCC_BASE/lib64"
 EXEC_LDFLAGS+=" -ldl"

 PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
+PLATFORM_LDFLAGS+=" -B$BINUTILS"

 EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS $LIBURING_LIBS"

@@ -168,6 +169,4 @@ VALGRIND_VER="$VALGRIND_BASE/bin/"
 LUA_PATH=
 LUA_LIB=

-USE_FOLLY_DISTRIBUTED_MUTEX=1
-
-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
+export CC CXX AR AS CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
@@ -120,7 +120,7 @@ uncommitted_code=`git diff HEAD`

 # If there's no uncommitted changes, we assume user are doing post-commit
 # format check, in which case we'll try to check the modified lines vs. the
-# facebook/rocksdb.git master branch. Otherwise, we'll check format of the
+# facebook/rocksdb.git main branch. Otherwise, we'll check format of the
 # uncommitted code only.
 if [ -z "$uncommitted_code" ]
 then
@@ -128,8 +128,8 @@ then
  [ "$FORMAT_REMOTE" ] || FORMAT_REMOTE="$(git remote -v | grep 'facebook/rocksdb.git' | head -n 1 | cut -f 1)"
  # Fall back on 'origin' if that fails
  [ "$FORMAT_REMOTE" ] || FORMAT_REMOTE=origin
-  # Use master branch from that remote
-  [ "$FORMAT_UPSTREAM" ] || FORMAT_UPSTREAM="$FORMAT_REMOTE/master"
+  # Use main branch from that remote
+  [ "$FORMAT_UPSTREAM" ] || FORMAT_UPSTREAM="$FORMAT_REMOTE/$(git remote show $FORMAT_REMOTE | sed -n '/HEAD branch/s/.*: //p')"
  # Get the common ancestor with that remote branch. Everything after that
  # common ancestor would be considered the contents of a pull request, so
  # should be relevant for formatting fixes.
@@ -20,26 +20,11 @@ STAT_FILE=${STAT_FILE:-$(mktemp -t -u rocksdb_test_stats_XXXX)}

 function cleanup {
  rm -rf $DATA_DIR
-  rm -f $STAT_FILE.fillseq
-  rm -f $STAT_FILE.readrandom
-  rm -f $STAT_FILE.overwrite
-  rm -f $STAT_FILE.memtablefillreadrandom
+  rm -f $STAT_FILE.*
 }

 trap cleanup EXIT

-if [ -z $GIT_BRANCH ]; then
-  git_br=`git rev-parse --abbrev-ref HEAD`
-else
-  git_br=$(basename $GIT_BRANCH)
-fi
-
-if [ $git_br == "master" ]; then
-  git_br=""
-else
-  git_br="."$git_br
-fi
-
 make release

 # measure fillseq + fill up the DB for overwrite benchmark
@@ -286,12 +271,10 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
    --sync=0 \
    --verify_checksum=1 \
    --delete_obsolete_files_period_micros=314572800 \
-    --max_grandparent_overlap_factor=10 \
    --use_plain_table=1 \
    --open_files=-1 \
    --mmap_read=1 \
    --mmap_write=0 \
-    --memtablerep=prefix_hash \
    --bloom_bits=10 \
    --bloom_locality=1 \
    --perf_level=0"
@@ -378,7 +361,7 @@ function send_to_ods {
    echo >&2 "ERROR: Key $key doesn't have a value."
    return
  fi
-  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
+  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=$key&value=$value" \
    --connect-timeout 60
 }

@@ -113,7 +113,7 @@ NON_TSAN_CRASH="CRASH_TEST_EXT_ARGS=--compression_type=zstd"
 DISABLE_JEMALLOC="DISABLE_JEMALLOC=1"
 HTTP_PROXY="https_proxy=http://fwdproxy.29.prn1:8080 http_proxy=http://fwdproxy.29.prn1:8080 ftp_proxy=http://fwdproxy.29.prn1:8080"
 SETUP_JAVA_ENV="export $HTTP_PROXY; export JAVA_HOME=/usr/local/jdk-8u60-64/; export PATH=\$JAVA_HOME/bin:\$PATH"
-PARSER="'parser':'python build_tools/error_filter.py $1'"
+PARSER="'parser':'/usr/bin/env python3 build_tools/error_filter.py $1'"

 CONTRUN_NAME="ROCKSDB_CONTRUN_NAME"
 SKIP_FORMAT_CHECKS="SKIP_FORMAT_BUCK_CHECKS=1"
@@ -379,23 +379,6 @@ LITE_BUILD_COMMANDS="[
    }
 ]"

-#
-# Report RocksDB lite binary size to scuba
-REPORT_LITE_BINARY_SIZE_COMMANDS="[
-    {
-        'name':'Rocksdb Lite Binary Size',
-        'oncall':'$ONCALL',
-        'executeLocal': 'true',
-        'steps': [
-            $CLEANUP_ENV,
-            {
-                'name':'Report RocksDB Lite binary size to scuba',
-                'shell':'cd $WORKING_DIR; tools/report_lite_binary_size.sh',
-                'user':'root',
-            },
-        ],
-]"
-
 #
 # RocksDB stress/crash test
 #
@@ -588,7 +571,7 @@ WRITE_STRESS_COMMANDS="[
            $CLEANUP_ENV,
            {
                'name':'Build and run RocksDB write stress tests',
-                'shell':'cd $WORKING_DIR; make write_stress && python tools/write_stress_runner.py --runtime_sec=3600 --db=/tmp/rocksdb_write_stress || $CONTRUN_NAME=write_stress $TASK_CREATION_TOOL',
+                'shell':'cd $WORKING_DIR; make write_stress && /usr/bin/env python3 tools/write_stress_runner.py --runtime_sec=3600 --db=/tmp/rocksdb_write_stress || $CONTRUN_NAME=write_stress $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            }
@@ -611,7 +594,7 @@ ASAN_TEST_COMMANDS="[
            $CLEANUP_ENV,
            {
                'name':'Test RocksDB debug under ASAN',
-'shell':'cd $WORKING_DIR; set -o pipefail && ($SHM $ASAN $DEBUG $SKIP_FORMAT_CHECKS make $PARALLELISM asan_check || $CONTRUN_NAME=asan_check $TASK_CREATION_TOOL) |& /usr/facebook/ops/scripts/asan_symbolize.py -d',
+'shell':'cd $WORKING_DIR; set -o pipefail && $SHM $ASAN $DEBUG $SKIP_FORMAT_CHECKS make $PARALLELISM asan_check || $CONTRUN_NAME=asan_check $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            }
@@ -1120,7 +1103,7 @@ NO_COMPRESSION_COMMANDS="[
 #
 run_regression()
 {
-  time -v bash -vx ./build_tools/regression_build_test.sh $(mktemp -d  $WORKSPACE/leveldb.XXXX) $(mktemp leveldb_test_stats.XXXX)
+  time bash -vx ./build_tools/regression_build_test.sh $(mktemp -d  $WORKING_DIR/rocksdb.XXXX) $(mktemp rocksdb_test_stats.XXXX)

  # ======= report size to ODS ========

@@ -1161,6 +1144,7 @@ REGRESSION_COMMANDS="[
    {
        'name':'Rocksdb regression commands',
        'oncall':'$ONCALL',
+        'executeLocal': 'true',
        'steps': [
            $CLEANUP_ENV,
            {
@@ -1195,6 +1179,40 @@ JAVA_BUILD_TEST_COMMANDS="[
    }
 ]"

+#
+# RocksDB fbcode stress/crash test
+#
+FBCODE_STRESS_CRASH_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb Fbcode Stress and Crash Test',
+        'oncall':'$ONCALL',
+        'executeLocal': 'true',
+        'timeout': 86400,
+        'steps': [
+            {
+                'name':'Copy RocksDB code to fbcode repo',
+                'shell':'cd internal_repo_rocksdb/repo && git init -b main && git add * && git commit -a -m \"Make internal_repo_rocksdb/repo a git repo\" && cd ../.. && echo Y | python3 rocks/release_script/release_to_fbcode.py -u internal_repo_rocksdb/repo main || $CONTRUN_NAME=db_stress_fbcode $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            {
+                'name':'Build RocksDB fbcode stress tests',
+                'shell':'cd $WORKING_DIR; buck build @mode/dbg rocks/tools:rocks_db_stress || $CONTRUN_NAME=db_stress_fbcode $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+            {
+                'name':'Run RocksDB whitebox crash tests',
+                'timeout': 86400,
+                'shell':'cd $WORKING_DIR; mkdir /dev/shm/rocksdb_whitebox_crash_test && TEST_TMPDIR=\$(mktemp -d --tmpdir=/dev/shm/rocksdb_whitebox_crash_test) python3 rocksdb/src/tools/db_crashtest.py --stress_cmd=buck-out/dbg/gen/rocks/tools/rocks_db_stress -secondary_cache_uri=\"$SECONDARY_CACHE_URI\" --env_uri=$ENV_URI -checkpoint_one_in=0 -backup_one_in=0 -cache_size=134217728 -cache_numshardbits=4 $EXTRA_DB_STRESS_ARGS -logtostderr=false whitebox || $CONTRUN_NAME=db_stress_fbcode $TASK_CREATION_TOOL',
+                'user':'root',
+                $PARSER
+            },
+        ],
+        $REPORT
+    }
+]"
+

 case $1 in
  unit)
@@ -1230,9 +1248,6 @@ case $1 in
  lite)
    echo $LITE_BUILD_COMMANDS
    ;;
-  report_lite_binary_size)
-    echo $REPORT_LITE_BINARY_SIZE_COMMANDS
-    ;;
  stress_crash)
    echo $STRESS_CRASH_TEST_COMMANDS
    ;;
@@ -1327,11 +1342,18 @@ case $1 in
    echo $REGRESSION_COMMANDS
    ;;
  run_regression)
+    set -e
    run_regression
+    set +e
    ;;
  java_build)
    echo $JAVA_BUILD_TEST_COMMANDS
    ;;
+  fbcode_stress_crash)
+    set -f
+    echo $FBCODE_STRESS_CRASH_TEST_COMMANDS
+    set +f
+    ;;
  *)
    echo "Invalid determinator command"
    exit 1
@@ -10,6 +10,8 @@
 #include "rocksdb/cache.h"

 #include "cache/lru_cache.h"
+#include "rocksdb/secondary_cache.h"
+#include "rocksdb/utilities/customizable_util.h"
 #include "rocksdb/utilities/options_type.h"
 #include "util/string_util.h"

@@ -34,6 +36,13 @@ static std::unordered_map<std::string, OptionTypeInfo>
 };
 #endif  // ROCKSDB_LITE

+Status SecondaryCache::CreateFromString(
+    const ConfigOptions& config_options, const std::string& value,
+    std::shared_ptr<SecondaryCache>* result) {
+  return LoadSharedObject<SecondaryCache>(config_options, value, nullptr,
+                                          result);
+}
+
 Status Cache::CreateFromString(const ConfigOptions& config_options,
                               const std::string& value,
                               std::shared_ptr<Cache>* result) {
@@ -13,7 +13,7 @@ int main() {
  return 1;
 }
 #else
-#include <rocksdb/cache_bench_tool.h>
+#include "rocksdb/cache_bench_tool.h"
 int main(int argc, char** argv) {
  return ROCKSDB_NAMESPACE::cache_bench_tool(argc, argv);
 }
@@ -13,11 +13,11 @@
 #include "monitoring/histogram.h"
 #include "port/port.h"
 #include "rocksdb/cache.h"
+#include "rocksdb/convenience.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/secondary_cache.h"
 #include "rocksdb/system_clock.h"
-#include "rocksdb/utilities/object_registry.h"
 #include "table/block_based/cachable_entry.h"
 #include "util/coding.h"
 #include "util/gflags_compat.h"
@@ -233,9 +233,8 @@ class CacheBench {
      LRUCacheOptions opts(FLAGS_cache_size, FLAGS_num_shard_bits, false, 0.5);
 #ifndef ROCKSDB_LITE
      if (!FLAGS_secondary_cache_uri.empty()) {
-        Status s =
-            ObjectRegistry::NewInstance()->NewSharedObject<SecondaryCache>(
-                FLAGS_secondary_cache_uri, &secondary_cache);
+        Status s = SecondaryCache::CreateFromString(
+            ConfigOptions(), FLAGS_secondary_cache_uri, &secondary_cache);
        if (secondary_cache == nullptr) {
          fprintf(
              stderr,
@@ -19,6 +19,7 @@ std::array<const char*, kNumCacheEntryRoles> kCacheEntryRoleToCamelString{{
    "IndexBlock",
    "OtherBlock",
    "WriteBuffer",
+    "CompressionDictionaryBuildingBuffer",
    "Misc",
 }};

@@ -30,6 +31,7 @@ std::array<const char*, kNumCacheEntryRoles> kCacheEntryRoleToHyphenString{{
    "index-block",
    "other-block",
    "write-buffer",
+    "compression-dictionary-building-buffer",
    "misc",
 }};

@@ -14,6 +14,8 @@
 namespace ROCKSDB_NAMESPACE {

 // Classifications of block cache entries, for reporting statistics
+// Adding new enum to this class requires corresponding updates to
+// kCacheEntryRoleToCamelString and kCacheEntryRoleToHyphenString
 enum class CacheEntryRole {
  // Block-based table data block
  kDataBlock,
@@ -29,6 +31,9 @@ enum class CacheEntryRole {
  kOtherBlock,
  // WriteBufferManager reservations to account for memtable usage
  kWriteBuffer,
+  // BlockBasedTableBuilder reservations to account for
+  // compression dictionary building buffer's memory usage
+  kCompressionDictionaryBuildingBuffer,
  // Default bucket, for miscellaneous cache entries. Do not use for
  // entries that could potentially add up to large usage.
  kMisc,
@@ -15,6 +15,7 @@
 #include "rocksdb/cache.h"
 #include "rocksdb/status.h"
 #include "rocksdb/system_clock.h"
+#include "test_util/sync_point.h"
 #include "util/coding_lean.h"

 namespace ROCKSDB_NAMESPACE {
@@ -0,0 +1,131 @@
+//  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 "cache/cache_reservation_manager.h"
+
+#include <cassert>
+#include <cstddef>
+#include <cstring>
+#include <memory>
+
+#include "cache/cache_entry_roles.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/slice.h"
+#include "rocksdb/status.h"
+#include "table/block_based/block_based_table_reader.h"
+#include "util/coding.h"
+
+namespace ROCKSDB_NAMESPACE {
+CacheReservationManager::CacheReservationManager(std::shared_ptr<Cache> cache,
+                                                 bool delayed_decrease)
+    : delayed_decrease_(delayed_decrease), cache_allocated_size_(0) {
+  assert(cache != nullptr);
+  cache_ = cache;
+  std::memset(cache_key_, 0, kCacheKeyPrefixSize + kMaxVarint64Length);
+  EncodeVarint64(cache_key_, cache_->NewId());
+}
+
+CacheReservationManager::~CacheReservationManager() {
+  for (auto* handle : dummy_handles_) {
+    cache_->Release(handle, true);
+  }
+}
+
+template <CacheEntryRole R>
+Status CacheReservationManager::UpdateCacheReservation(
+    std::size_t new_mem_used) {
+  std::size_t cur_cache_allocated_size =
+      cache_allocated_size_.load(std::memory_order_relaxed);
+  if (new_mem_used == cur_cache_allocated_size) {
+    return Status::OK();
+  } else if (new_mem_used > cur_cache_allocated_size) {
+    Status s = IncreaseCacheReservation<R>(new_mem_used);
+    return s;
+  } else {
+    // In delayed decrease mode, we don't decrease cache reservation
+    // untill the memory usage is less than 3/4 of what we reserve
+    // in the cache.
+    // We do this because
+    // (1) Dummy entry insertion is expensive in block cache
+    // (2) Delayed releasing previously inserted dummy entries can save such
+    // expensive dummy entry insertion on memory increase in the near future,
+    // which is likely to happen when the memory usage is greater than or equal
+    // to 3/4 of what we reserve
+    if (delayed_decrease_ && new_mem_used >= cur_cache_allocated_size / 4 * 3) {
+      return Status::OK();
+    } else {
+      Status s = DecreaseCacheReservation(new_mem_used);
+      return s;
+    }
+  }
+}
+
+// Explicitly instantiate templates for "CacheEntryRole" values we use.
+// This makes it possible to keep the template definitions in the .cc file.
+template Status CacheReservationManager::UpdateCacheReservation<
+    CacheEntryRole::kWriteBuffer>(std::size_t new_mem_used);
+template Status CacheReservationManager::UpdateCacheReservation<
+    CacheEntryRole::kCompressionDictionaryBuildingBuffer>(
+    std::size_t new_mem_used);
+// For cache reservation manager unit tests
+template Status CacheReservationManager::UpdateCacheReservation<
+    CacheEntryRole::kMisc>(std::size_t new_mem_used);
+
+template <CacheEntryRole R>
+Status CacheReservationManager::IncreaseCacheReservation(
+    std::size_t new_mem_used) {
+  Status return_status = Status::OK();
+  while (new_mem_used > cache_allocated_size_.load(std::memory_order_relaxed)) {
+    Cache::Handle* handle = nullptr;
+    return_status = cache_->Insert(GetNextCacheKey(), nullptr, kSizeDummyEntry,
+                                   GetNoopDeleterForRole<R>(), &handle);
+
+    if (return_status != Status::OK()) {
+      return return_status;
+    }
+
+    dummy_handles_.push_back(handle);
+    cache_allocated_size_ += kSizeDummyEntry;
+  }
+  return return_status;
+}
+
+Status CacheReservationManager::DecreaseCacheReservation(
+    std::size_t new_mem_used) {
+  Status return_status = Status::OK();
+
+  // Decrease to the smallest multiple of kSizeDummyEntry that is greater than
+  // or equal to new_mem_used We do addition instead of new_mem_used <=
+  // cache_allocated_size_.load(std::memory_order_relaxed) - kSizeDummyEntry to
+  // avoid underflow of size_t when cache_allocated_size_ = 0
+  while (new_mem_used + kSizeDummyEntry <=
+         cache_allocated_size_.load(std::memory_order_relaxed)) {
+    assert(!dummy_handles_.empty());
+    auto* handle = dummy_handles_.back();
+    cache_->Release(handle, true);
+    dummy_handles_.pop_back();
+    cache_allocated_size_ -= kSizeDummyEntry;
+  }
+  return return_status;
+}
+
+std::size_t CacheReservationManager::GetTotalReservedCacheSize() {
+  return cache_allocated_size_.load(std::memory_order_relaxed);
+}
+
+Slice CacheReservationManager::GetNextCacheKey() {
+  // Calling this function will have the side-effect of changing the
+  // underlying cache_key_ that is shared among other keys generated from this
+  // fucntion. Therefore please make sure the previous keys are saved/copied
+  // before calling this function.
+  std::memset(cache_key_ + kCacheKeyPrefixSize, 0, kMaxVarint64Length);
+  char* end =
+      EncodeVarint64(cache_key_ + kCacheKeyPrefixSize, next_cache_key_id_++);
+  return Slice(cache_key_, static_cast<std::size_t>(end - cache_key_));
+}
+}  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,98 @@
+//  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.
+
+#pragma once
+
+#include <atomic>
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+#include <vector>
+
+#include "cache/cache_entry_roles.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/slice.h"
+#include "rocksdb/status.h"
+#include "table/block_based/block_based_table_reader.h"
+#include "util/coding.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+// CacheReservationManager is for reserving cache space for the memory used
+// through inserting/releasing dummy entries in the cache.
+// This class is not thread-safe.
+class CacheReservationManager {
+ public:
+  // Construct a CacheReservationManager
+  // @param cache The cache where dummy entries are inserted and released for
+  // reserving cache space
+  // @param delayed_decrease If set true, then dummy entries won't be released
+  // immediately when memory usage decreases.
+  //                         Instead, it will be released when the memory usage
+  //                         decreases to 3/4 of what we have reserved so far.
+  //                         This is for saving some future dummy entry
+  //                         insertion when memory usage increases are likely to
+  //                         happen in the near future.
+  explicit CacheReservationManager(std::shared_ptr<Cache> cache,
+                                   bool delayed_decrease = false);
+
+  // no copy constructor, copy assignment, move constructor, move assignment
+  CacheReservationManager(const CacheReservationManager &) = delete;
+  CacheReservationManager &operator=(const CacheReservationManager &) = delete;
+  CacheReservationManager(CacheReservationManager &&) = delete;
+  CacheReservationManager &operator=(CacheReservationManager &&) = delete;
+
+  ~CacheReservationManager();
+
+  template <CacheEntryRole R>
+
+  // Insert and release dummy entries in the cache to
+  // match the size of total dummy entries with the smallest multiple of
+  // kSizeDummyEntry that is greater than or equal to new_mem_used
+  //
+  // Insert dummy entries if new_memory_used > cache_allocated_size_;
+  //
+  // Release dummy entries if new_memory_used < cache_allocated_size_
+  // (and new_memory_used < cache_allocated_size_ * 3/4
+  // when delayed_decrease is set true);
+  //
+  // Keey dummy entries the same if (1) new_memory_used == cache_allocated_size_
+  // or (2) new_memory_used is in the interval of
+  // [cache_allocated_size_ * 3/4, cache_allocated_size) when delayed_decrease
+  // is set true.
+  //
+  // @param new_memory_used The number of bytes used by new memory
+  // @return On inserting dummy entries, it returns Status::OK() if all dummy
+  // entry insertions succeed. Otherwise, it returns the first non-ok status;
+  // On releasing dummy entries, it always returns Status::OK().
+  // On keeping dummy entries the same, it always returns Status::OK().
+  Status UpdateCacheReservation(std::size_t new_memory_used);
+  std::size_t GetTotalReservedCacheSize();
+
+ private:
+  static constexpr std::size_t kSizeDummyEntry = 256 * 1024;
+  // The key will be longer than keys for blocks in SST files so they won't
+  // conflict.
+  static const std::size_t kCacheKeyPrefixSize =
+      BlockBasedTable::kMaxCacheKeyPrefixSize + kMaxVarint64Length;
+
+  Slice GetNextCacheKey();
+  template <CacheEntryRole R>
+  Status IncreaseCacheReservation(std::size_t new_mem_used);
+  Status DecreaseCacheReservation(std::size_t new_mem_used);
+
+  std::shared_ptr<Cache> cache_;
+  bool delayed_decrease_;
+  std::atomic<std::size_t> cache_allocated_size_;
+  std::vector<Cache::Handle *> dummy_handles_;
+  std::uint64_t next_cache_key_id_ = 0;
+  // The non-prefix part will be updated according to the ID to use.
+  char cache_key_[kCacheKeyPrefixSize + kMaxVarint64Length];
+};
+}  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,411 @@
+//  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 "cache/cache_reservation_manager.h"
+
+#include <cstddef>
+#include <cstring>
+#include <memory>
+
+#include "cache/cache_entry_roles.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/slice.h"
+#include "table/block_based/block_based_table_reader.h"
+#include "test_util/testharness.h"
+#include "util/coding.h"
+
+namespace ROCKSDB_NAMESPACE {
+class CacheReservationManagerTest : public ::testing::Test {
+ protected:
+  static constexpr std::size_t kOneGigabyte = 1024 * 1024 * 1024;
+  static constexpr int kNumShardBits = 0;  // 2^0 shard
+
+  static constexpr std::size_t kSizeDummyEntry = 256 * 1024;
+  static const std::size_t kCacheKeyPrefixSize =
+      BlockBasedTable::kMaxCacheKeyPrefixSize + kMaxVarint64Length;
+  static constexpr std::size_t kMetaDataChargeOverhead = 10000;
+
+  std::shared_ptr<Cache> cache = NewLRUCache(kOneGigabyte, kNumShardBits);
+  std::unique_ptr<CacheReservationManager> test_cache_rev_mng;
+
+  CacheReservationManagerTest() {
+    test_cache_rev_mng.reset(new CacheReservationManager(cache));
+  }
+};
+
+TEST_F(CacheReservationManagerTest, GenerateCacheKey) {
+  // The first cache reservation manager owning the cache will have
+  // cache->NewId() = 1
+  constexpr std::size_t kCacheNewId = 1;
+  // The first key generated inside of cache reservation manager will have
+  // next_cache_key_id = 0
+  constexpr std::size_t kCacheKeyId = 0;
+
+  char expected_cache_key[kCacheKeyPrefixSize + kMaxVarint64Length];
+  std::memset(expected_cache_key, 0, kCacheKeyPrefixSize + kMaxVarint64Length);
+
+  EncodeVarint64(expected_cache_key, kCacheNewId);
+  char* end =
+      EncodeVarint64(expected_cache_key + kCacheKeyPrefixSize, kCacheKeyId);
+  Slice expected_cache_key_slice(
+      expected_cache_key, static_cast<std::size_t>(end - expected_cache_key));
+
+  std::size_t new_mem_used = 1 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  ASSERT_EQ(s, Status::OK());
+  ASSERT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry);
+  ASSERT_LT(cache->GetPinnedUsage(),
+            1 * kSizeDummyEntry + kMetaDataChargeOverhead);
+
+  Cache::Handle* handle = cache->Lookup(expected_cache_key_slice);
+  EXPECT_NE(handle, nullptr)
+      << "Failed to generate the cache key for the dummy entry correctly";
+  // Clean up the returned handle from Lookup() to prevent memory leak
+  cache->Release(handle);
+}
+
+TEST_F(CacheReservationManagerTest, KeepCacheReservationTheSame) {
+  std::size_t new_mem_used = 1 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  ASSERT_EQ(s, Status::OK());
+  ASSERT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry);
+  std::size_t initial_pinned_usage = cache->GetPinnedUsage();
+  ASSERT_GE(initial_pinned_usage, 1 * kSizeDummyEntry);
+  ASSERT_LT(initial_pinned_usage,
+            1 * kSizeDummyEntry + kMetaDataChargeOverhead);
+
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to keep cache reservation the same when new_mem_used equals "
+         "to current cache reservation";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly when new_mem_used equals to current "
+         "cache reservation";
+  EXPECT_EQ(cache->GetPinnedUsage(), initial_pinned_usage)
+      << "Failed to keep underlying dummy entries the same when new_mem_used "
+         "equals to current cache reservation";
+}
+
+TEST_F(CacheReservationManagerTest,
+       IncreaseCacheReservationByMultiplesOfDummyEntrySize) {
+  std::size_t new_mem_used = 2 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to increase cache reservation correctly";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            2 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation increase correctly";
+  EXPECT_GE(cache->GetPinnedUsage(), 2 * kSizeDummyEntry)
+      << "Failed to increase underlying dummy entries in cache correctly";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            2 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to increase underlying dummy entries in cache correctly";
+}
+
+TEST_F(CacheReservationManagerTest,
+       IncreaseCacheReservationNotByMultiplesOfDummyEntrySize) {
+  std::size_t new_mem_used = 2 * kSizeDummyEntry + kSizeDummyEntry / 2;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to increase cache reservation correctly";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            3 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation increase correctly";
+  EXPECT_GE(cache->GetPinnedUsage(), 3 * kSizeDummyEntry)
+      << "Failed to increase underlying dummy entries in cache correctly";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            3 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to increase underlying dummy entries in cache correctly";
+}
+
+TEST(CacheReservationManagerIncreaseReservcationOnFullCacheTest,
+     IncreaseCacheReservationOnFullCache) {
+  constexpr std::size_t kOneMegabyte = 1024 * 1024;
+  constexpr std::size_t kOneGigabyte = 1024 * 1024 * 1024;
+  constexpr std::size_t kSizeDummyEntry = 256 * 1024;
+  constexpr std::size_t kMetaDataChargeOverhead = 10000;
+
+  LRUCacheOptions lo;
+  lo.capacity = kOneMegabyte;
+  lo.num_shard_bits = 0;  // 2^0 shard
+  lo.strict_capacity_limit = true;
+  std::shared_ptr<Cache> cache = NewLRUCache(lo);
+  std::unique_ptr<CacheReservationManager> test_cache_rev_mng(
+      new CacheReservationManager(cache));
+
+  std::size_t new_mem_used = kOneMegabyte + 1;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::Incomplete())
+      << "Failed to return status to indicate failure of dummy entry insertion "
+         "during cache reservation on full cache";
+  EXPECT_GE(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly before cache resevation failure happens "
+         "due to full cache";
+  EXPECT_LE(test_cache_rev_mng->GetTotalReservedCacheSize(), kOneMegabyte)
+      << "Failed to bookkeep correctly (i.e, bookkeep only successful dummy "
+         "entry insertions) when encountering cache resevation failure due to "
+         "full cache";
+  EXPECT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry)
+      << "Failed to insert underlying dummy entries correctly when "
+         "encountering cache resevation failure due to full cache";
+  EXPECT_LE(cache->GetPinnedUsage(), kOneMegabyte)
+      << "Failed to insert underlying dummy entries correctly when "
+         "encountering cache resevation failure due to full cache";
+
+  new_mem_used = kOneMegabyte / 2;  // 2 dummy entries
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to decrease cache reservation after encountering cache "
+         "reservation failure due to full cache";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            2 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation decrease correctly after "
+         "encountering cache reservation due to full cache";
+  EXPECT_GE(cache->GetPinnedUsage(), 2 * kSizeDummyEntry)
+      << "Failed to release underlying dummy entries correctly on cache "
+         "reservation decrease after encountering cache resevation failure due "
+         "to full cache";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            2 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to release underlying dummy entries correctly on cache "
+         "reservation decrease after encountering cache resevation failure due "
+         "to full cache";
+
+  // Create cache full again for subsequent tests
+  new_mem_used = kOneMegabyte + 1;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::Incomplete())
+      << "Failed to return status to indicate failure of dummy entry insertion "
+         "during cache reservation on full cache";
+  EXPECT_GE(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly before cache resevation failure happens "
+         "due to full cache";
+  EXPECT_LE(test_cache_rev_mng->GetTotalReservedCacheSize(), kOneMegabyte)
+      << "Failed to bookkeep correctly (i.e, bookkeep only successful dummy "
+         "entry insertions) when encountering cache resevation failure due to "
+         "full cache";
+  EXPECT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry)
+      << "Failed to insert underlying dummy entries correctly when "
+         "encountering cache resevation failure due to full cache";
+  EXPECT_LE(cache->GetPinnedUsage(), kOneMegabyte)
+      << "Failed to insert underlying dummy entries correctly when "
+         "encountering cache resevation failure due to full cache";
+
+  // Increase cache capacity so the previously failed insertion can fully
+  // succeed
+  cache->SetCapacity(kOneGigabyte);
+  new_mem_used = kOneMegabyte + 1;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to increase cache reservation after increasing cache capacity "
+         "and mitigating cache full error";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            5 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation increase correctly after "
+         "increasing cache capacity and mitigating cache full error";
+  EXPECT_GE(cache->GetPinnedUsage(), 5 * kSizeDummyEntry)
+      << "Failed to insert underlying dummy entries correctly after increasing "
+         "cache capacity and mitigating cache full error";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            5 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to insert underlying dummy entries correctly after increasing "
+         "cache capacity and mitigating cache full error";
+}
+
+TEST_F(CacheReservationManagerTest,
+       DecreaseCacheReservationByMultiplesOfDummyEntrySize) {
+  std::size_t new_mem_used = 2 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  ASSERT_EQ(s, Status::OK());
+  ASSERT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            2 * kSizeDummyEntry);
+  ASSERT_GE(cache->GetPinnedUsage(), 2 * kSizeDummyEntry);
+  ASSERT_LT(cache->GetPinnedUsage(),
+            2 * kSizeDummyEntry + kMetaDataChargeOverhead);
+
+  new_mem_used = 1 * kSizeDummyEntry;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to decrease cache reservation correctly";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation decrease correctly";
+  EXPECT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry)
+      << "Failed to decrease underlying dummy entries in cache correctly";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            1 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to decrease underlying dummy entries in cache correctly";
+}
+
+TEST_F(CacheReservationManagerTest,
+       DecreaseCacheReservationNotByMultiplesOfDummyEntrySize) {
+  std::size_t new_mem_used = 2 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  ASSERT_EQ(s, Status::OK());
+  ASSERT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            2 * kSizeDummyEntry);
+  ASSERT_GE(cache->GetPinnedUsage(), 2 * kSizeDummyEntry);
+  ASSERT_LT(cache->GetPinnedUsage(),
+            2 * kSizeDummyEntry + kMetaDataChargeOverhead);
+
+  new_mem_used = kSizeDummyEntry / 2;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to decrease cache reservation correctly";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            1 * kSizeDummyEntry)
+      << "Failed to bookkeep cache reservation decrease correctly";
+  EXPECT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry)
+      << "Failed to decrease underlying dummy entries in cache correctly";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            1 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to decrease underlying dummy entries in cache correctly";
+}
+
+TEST(CacheReservationManagerWithDelayedDecreaseTest,
+     DecreaseCacheReservationWithDelayedDecrease) {
+  constexpr std::size_t kOneGigabyte = 1024 * 1024 * 1024;
+  constexpr std::size_t kSizeDummyEntry = 256 * 1024;
+  constexpr std::size_t kMetaDataChargeOverhead = 10000;
+
+  LRUCacheOptions lo;
+  lo.capacity = kOneGigabyte;
+  lo.num_shard_bits = 0;
+  std::shared_ptr<Cache> cache = NewLRUCache(lo);
+  std::unique_ptr<CacheReservationManager> test_cache_rev_mng(
+      new CacheReservationManager(cache, true /* delayed_decrease */));
+
+  std::size_t new_mem_used = 8 * kSizeDummyEntry;
+  Status s =
+      test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  ASSERT_EQ(s, Status::OK());
+  ASSERT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            8 * kSizeDummyEntry);
+  std::size_t initial_pinned_usage = cache->GetPinnedUsage();
+  ASSERT_GE(initial_pinned_usage, 8 * kSizeDummyEntry);
+  ASSERT_LT(initial_pinned_usage,
+            8 * kSizeDummyEntry + kMetaDataChargeOverhead);
+
+  new_mem_used = 6 * kSizeDummyEntry;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK()) << "Failed to delay decreasing cache reservation";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            8 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly when delaying cache reservation "
+         "decrease";
+  EXPECT_EQ(cache->GetPinnedUsage(), initial_pinned_usage)
+      << "Failed to delay decreasing underlying dummy entries in cache";
+
+  new_mem_used = 7 * kSizeDummyEntry;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK()) << "Failed to delay decreasing cache reservation";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            8 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly when delaying cache reservation "
+         "decrease";
+  EXPECT_EQ(cache->GetPinnedUsage(), initial_pinned_usage)
+      << "Failed to delay decreasing underlying dummy entries in cache";
+
+  new_mem_used = 6 * kSizeDummyEntry - 1;
+  s = test_cache_rev_mng
+          ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+              new_mem_used);
+  EXPECT_EQ(s, Status::OK())
+      << "Failed to decrease cache reservation correctly when new_mem_used < "
+         "GetTotalReservedCacheSize() * 3 / 4 on delayed decrease mode";
+  EXPECT_EQ(test_cache_rev_mng->GetTotalReservedCacheSize(),
+            6 * kSizeDummyEntry)
+      << "Failed to bookkeep correctly when new_mem_used < "
+         "GetTotalReservedCacheSize() * 3 / 4 on delayed decrease mode";
+  EXPECT_GE(cache->GetPinnedUsage(), 6 * kSizeDummyEntry)
+      << "Failed to decrease underlying dummy entries in cache when "
+         "new_mem_used < GetTotalReservedCacheSize() * 3 / 4 on delayed "
+         "decrease mode";
+  EXPECT_LT(cache->GetPinnedUsage(),
+            6 * kSizeDummyEntry + kMetaDataChargeOverhead)
+      << "Failed to decrease underlying dummy entries in cache when "
+         "new_mem_used < GetTotalReservedCacheSize() * 3 / 4 on delayed "
+         "decrease mode";
+}
+
+TEST(CacheReservationManagerDestructorTest,
+     ReleaseRemainingDummyEntriesOnDestruction) {
+  constexpr std::size_t kOneGigabyte = 1024 * 1024 * 1024;
+  constexpr std::size_t kSizeDummyEntry = 256 * 1024;
+  constexpr std::size_t kMetaDataChargeOverhead = 10000;
+
+  LRUCacheOptions lo;
+  lo.capacity = kOneGigabyte;
+  lo.num_shard_bits = 0;
+  std::shared_ptr<Cache> cache = NewLRUCache(lo);
+  {
+    std::unique_ptr<CacheReservationManager> test_cache_rev_mng(
+        new CacheReservationManager(cache));
+    std::size_t new_mem_used = 1 * kSizeDummyEntry;
+    Status s =
+        test_cache_rev_mng
+            ->UpdateCacheReservation<ROCKSDB_NAMESPACE::CacheEntryRole::kMisc>(
+                new_mem_used);
+    ASSERT_EQ(s, Status::OK());
+    ASSERT_GE(cache->GetPinnedUsage(), 1 * kSizeDummyEntry);
+    ASSERT_LT(cache->GetPinnedUsage(),
+              1 * kSizeDummyEntry + kMetaDataChargeOverhead);
+  }
+  EXPECT_EQ(cache->GetPinnedUsage(), 0 * kSizeDummyEntry)
+      << "Failed to release remaining underlying dummy entries in cache in "
+         "CacheReservationManager's destructor";
+}
+}  // namespace ROCKSDB_NAMESPACE
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -260,7 +260,7 @@ struct CleanupContext {
 class ClockCacheShard final : public CacheShard {
 public:
  // Hash map type.
-  typedef tbb::concurrent_hash_map<CacheKey, CacheHandle*, CacheKey> HashTable;
+  using HashTable = tbb::concurrent_hash_map<CacheKey, CacheHandle*, CacheKey>;

  ClockCacheShard();
  ~ClockCacheShard() override;
@@ -280,7 +280,8 @@ class ClockCacheShard final : public CacheShard {
  Cache::Handle* Lookup(const Slice& key, uint32_t hash,
                        const Cache::CacheItemHelper* /*helper*/,
                        const Cache::CreateCallback& /*create_cb*/,
-                        Cache::Priority /*priority*/, bool /*wait*/) override {
+                        Cache::Priority /*priority*/, bool /*wait*/,
+                        Statistics* /*stats*/) override {
    return Lookup(key, hash);
  }
  bool Release(Cache::Handle* handle, bool /*useful*/,
@@ -13,6 +13,8 @@
 #include <cstdint>
 #include <cstdio>

+#include "monitoring/perf_context_imp.h"
+#include "monitoring/statistics.h"
 #include "util/mutexlock.h"

 namespace ROCKSDB_NAMESPACE {
@@ -358,7 +360,10 @@ Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
      if (handle == nullptr) {
        LRU_Insert(e);
      } else {
-        e->Ref();
+        // If caller already holds a ref, no need to take one here
+        if (!e->HasRefs()) {
+          e->Ref();
+        }
        *handle = reinterpret_cast<Cache::Handle*>(e);
      }
    }
@@ -396,22 +401,17 @@ void LRUCacheShard::Promote(LRUHandle* e) {
  if (e->value) {
    Cache::Handle* handle = reinterpret_cast<Cache::Handle*>(e);
    Status s = InsertItem(e, &handle, /*free_handle_on_fail=*/false);
-    if (s.ok()) {
-      // InsertItem would have taken a reference on the item, so decrement it
-      // here as we expect the caller to already hold a reference
-      e->Unref();
-    } else {
+    if (!s.ok()) {
      // Item is in memory, but not accounted against the cache capacity.
      // When the handle is released, the item should get deleted
      assert(!e->InCache());
    }
  } else {
    // Since the secondary cache lookup failed, mark the item as not in cache
-    // and charge the cache only for metadata usage, i.e handle, key etc
+    // Don't charge the cache as its only metadata that'll shortly be released
    MutexLock l(&mutex_);
    e->charge = 0;
    e->SetInCache(false);
-    usage_ += e->CalcTotalCharge(metadata_charge_policy_);
  }
 }

@@ -419,7 +419,7 @@ Cache::Handle* LRUCacheShard::Lookup(
    const Slice& key, uint32_t hash,
    const ShardedCache::CacheItemHelper* helper,
    const ShardedCache::CreateCallback& create_cb, Cache::Priority priority,
-    bool wait) {
+    bool wait, Statistics* stats) {
  LRUHandle* e = nullptr;
  {
    MutexLock l(&mutex_);
@@ -472,11 +472,18 @@ Cache::Handle* LRUCacheShard::Lookup(
          e->Unref();
          e->Free();
          e = nullptr;
+        } else {
+          PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
+          RecordTick(stats, SECONDARY_CACHE_HITS);
        }
      } else {
        // If wait is false, we always return a handle and let the caller
        // release the handle after checking for success or failure
        e->SetIncomplete(true);
+        // This may be slightly inaccurate, if the lookup eventually fails.
+        // But the probability is very low.
+        PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
+        RecordTick(stats, SECONDARY_CACHE_HITS);
      }
    }
  }
@@ -522,7 +529,12 @@ bool LRUCacheShard::Release(Cache::Handle* handle, bool force_erase) {
        last_reference = false;
      }
    }
-    if (last_reference) {
+    // If it was the last reference, and the entry is either not secondary
+    // cache compatible (i.e a dummy entry for accounting), or is secondary
+    // cache compatible and has a non-null value, then decrement the cache
+    // usage. If value is null in the latter case, taht means the lookup
+    // failed and we didn't charge the cache.
+    if (last_reference && (!e->IsSecondaryCacheCompatible() || e->value)) {
      size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
      assert(usage_ >= total_charge);
      usage_ -= total_charge;
@@ -319,10 +319,11 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash,
                                const ShardedCache::CacheItemHelper* helper,
                                const ShardedCache::CreateCallback& create_cb,
-                                ShardedCache::Priority priority,
-                                bool wait) override;
+                                ShardedCache::Priority priority, bool wait,
+                                Statistics* stats) override;
  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash) override {
-    return Lookup(key, hash, nullptr, nullptr, Cache::Priority::LOW, true);
+    return Lookup(key, hash, nullptr, nullptr, Cache::Priority::LOW, true,
+                  nullptr);
  }
  virtual bool Release(Cache::Handle* handle, bool /*useful*/,
                       bool force_erase) override {
@@ -224,7 +224,7 @@ class TestSecondaryCache : public SecondaryCache {
  }
  ~TestSecondaryCache() override { cache_.reset(); }

-  std::string Name() override { return "TestSecondaryCache"; }
+  const char* Name() const override { return "TestSecondaryCache"; }

  void InjectFailure() { inject_failure_ = true; }

@@ -378,7 +378,7 @@ class TestSecondaryCache : public SecondaryCache {
 class DBSecondaryCacheTest : public DBTestBase {
 public:
  DBSecondaryCacheTest()
-      : DBTestBase("/db_secondary_cache_test", /*env_do_fsync=*/true) {
+      : DBTestBase("db_secondary_cache_test", /*env_do_fsync=*/true) {
    fault_fs_.reset(new FaultInjectionTestFS(env_->GetFileSystem()));
    fault_env_.reset(new CompositeEnvWrapper(env_, fault_fs_));
  }
@@ -468,6 +468,7 @@ TEST_F(LRUSecondaryCacheTest, BasicTest) {
      std::make_shared<TestSecondaryCache>(2048);
  opts.secondary_cache = secondary_cache;
  std::shared_ptr<Cache> cache = NewLRUCache(opts);
+  std::shared_ptr<Statistics> stats = CreateDBStatistics();

  Random rnd(301);
  std::string str1 = rnd.RandomString(1020);
@@ -476,22 +477,29 @@ TEST_F(LRUSecondaryCacheTest, BasicTest) {
                          str1.length()));
  std::string str2 = rnd.RandomString(1020);
  TestItem* item2 = new TestItem(str2.data(), str2.length());
-  // k2 should be demoted to NVM
+  // k1 should be demoted to NVM
  ASSERT_OK(cache->Insert("k2", item2, &LRUSecondaryCacheTest::helper_,
                          str2.length()));

+  get_perf_context()->Reset();
  Cache::Handle* handle;
-  handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
-                         test_item_creator, Cache::Priority::LOW, true);
+  handle =
+      cache->Lookup("k2", &LRUSecondaryCacheTest::helper_, test_item_creator,
+                    Cache::Priority::LOW, true, stats.get());
  ASSERT_NE(handle, nullptr);
  cache->Release(handle);
  // This lookup should promote k1 and demote k2
-  handle = cache->Lookup("k1", &LRUSecondaryCacheTest::helper_,
-                         test_item_creator, Cache::Priority::LOW, true);
+  handle =
+      cache->Lookup("k1", &LRUSecondaryCacheTest::helper_, test_item_creator,
+                    Cache::Priority::LOW, true, stats.get());
  ASSERT_NE(handle, nullptr);
  cache->Release(handle);
  ASSERT_EQ(secondary_cache->num_inserts(), 2u);
  ASSERT_EQ(secondary_cache->num_lookups(), 1u);
+  ASSERT_EQ(stats->getTickerCount(SECONDARY_CACHE_HITS),
+            secondary_cache->num_lookups());
+  PerfContext perf_ctx = *get_perf_context();
+  ASSERT_EQ(perf_ctx.secondary_cache_hit_count, secondary_cache->num_lookups());

  cache.reset();
  secondary_cache.reset();
@@ -83,10 +83,10 @@ Cache::Handle* ShardedCache::Lookup(const Slice& key,
                                    const CacheItemHelper* helper,
                                    const CreateCallback& create_cb,
                                    Priority priority, bool wait,
-                                    Statistics* /*stats*/) {
+                                    Statistics* stats) {
  uint32_t hash = HashSlice(key);
  return GetShard(Shard(hash))
-      ->Lookup(key, hash, helper, create_cb, priority, wait);
+      ->Lookup(key, hash, helper, create_cb, priority, wait, stats);
 }

 bool ShardedCache::IsReady(Handle* handle) {
@@ -34,7 +34,8 @@ class CacheShard {
  virtual Cache::Handle* Lookup(const Slice& key, uint32_t hash,
                                const Cache::CacheItemHelper* helper,
                                const Cache::CreateCallback& create_cb,
-                                Cache::Priority priority, bool wait) = 0;
+                                Cache::Priority priority, bool wait,
+                                Statistics* stats) = 0;
  virtual bool Release(Cache::Handle* handle, bool useful,
                       bool force_erase) = 0;
  virtual bool IsReady(Cache::Handle* handle) = 0;
@@ -24,7 +24,7 @@ mkdir -p $COVERAGE_DIR

 # Find all gcno files to generate the coverage report

-PYTHON=${1:-`which python`}
+PYTHON=${1:-`which python3`}
 echo -e "Using $PYTHON"
 GCNO_FILES=`find $ROOT -name "*.gcno"`
 $GCOV --preserve-paths --relative-only --no-output $GCNO_FILES 2>/dev/null |
@@ -34,7 +34,13 @@ class Version;
 // the same as the inner DBIter.
 class ArenaWrappedDBIter : public Iterator {
 public:
-  virtual ~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
+  virtual ~ArenaWrappedDBIter() {
+    if (db_iter_ != nullptr) {
+      db_iter_->~DBIter();
+    } else {
+      assert(false);
+    }
+  }

  // Get the arena to be used to allocate memory for DBIter to be wrapped,
  // as well as child iterators in it.
@@ -90,7 +96,7 @@ class ArenaWrappedDBIter : public Iterator {
  }

 private:
-  DBIter* db_iter_;
+  DBIter* db_iter_ = nullptr;
  Arena arena_;
  uint64_t sv_number_;
  ColumnFamilyData* cfd_ = nullptr;
@@ -0,0 +1,146 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#include <cassert>
+
+#include "db/blob/blob_garbage_meter.h"
+#include "rocksdb/rocksdb_namespace.h"
+#include "rocksdb/status.h"
+#include "table/internal_iterator.h"
+#include "test_util/sync_point.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+// An internal iterator that passes each key-value encountered to
+// BlobGarbageMeter as inflow in order to measure the total number and size of
+// blobs in the compaction input on a per-blob file basis.
+class BlobCountingIterator : public InternalIterator {
+ public:
+  BlobCountingIterator(InternalIterator* iter,
+                       BlobGarbageMeter* blob_garbage_meter)
+      : iter_(iter), blob_garbage_meter_(blob_garbage_meter) {
+    assert(iter_);
+    assert(blob_garbage_meter_);
+
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  bool Valid() const override { return iter_->Valid() && status_.ok(); }
+
+  void SeekToFirst() override {
+    iter_->SeekToFirst();
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  void SeekToLast() override {
+    iter_->SeekToLast();
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  void Seek(const Slice& target) override {
+    iter_->Seek(target);
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  void SeekForPrev(const Slice& target) override {
+    iter_->SeekForPrev(target);
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  void Next() override {
+    assert(Valid());
+
+    iter_->Next();
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  bool NextAndGetResult(IterateResult* result) override {
+    assert(Valid());
+
+    const bool res = iter_->NextAndGetResult(result);
+    UpdateAndCountBlobIfNeeded();
+    return res;
+  }
+
+  void Prev() override {
+    assert(Valid());
+
+    iter_->Prev();
+    UpdateAndCountBlobIfNeeded();
+  }
+
+  Slice key() const override {
+    assert(Valid());
+    return iter_->key();
+  }
+
+  Slice user_key() const override {
+    assert(Valid());
+    return iter_->user_key();
+  }
+
+  Slice value() const override {
+    assert(Valid());
+    return iter_->value();
+  }
+
+  Status status() const override { return status_; }
+
+  bool PrepareValue() override {
+    assert(Valid());
+    return iter_->PrepareValue();
+  }
+
+  bool MayBeOutOfLowerBound() override {
+    assert(Valid());
+    return iter_->MayBeOutOfLowerBound();
+  }
+
+  IterBoundCheck UpperBoundCheckResult() override {
+    assert(Valid());
+    return iter_->UpperBoundCheckResult();
+  }
+
+  void SetPinnedItersMgr(PinnedIteratorsManager* pinned_iters_mgr) override {
+    iter_->SetPinnedItersMgr(pinned_iters_mgr);
+  }
+
+  bool IsKeyPinned() const override {
+    assert(Valid());
+    return iter_->IsKeyPinned();
+  }
+
+  bool IsValuePinned() const override {
+    assert(Valid());
+    return iter_->IsValuePinned();
+  }
+
+  Status GetProperty(std::string prop_name, std::string* prop) override {
+    return iter_->GetProperty(prop_name, prop);
+  }
+
+ private:
+  void UpdateAndCountBlobIfNeeded() {
+    assert(!iter_->Valid() || iter_->status().ok());
+
+    if (!iter_->Valid()) {
+      status_ = iter_->status();
+      return;
+    }
+
+    TEST_SYNC_POINT(
+        "BlobCountingIterator::UpdateAndCountBlobIfNeeded:ProcessInFlow");
+
+    status_ = blob_garbage_meter_->ProcessInFlow(key(), value());
+  }
+
+  InternalIterator* iter_;
+  BlobGarbageMeter* blob_garbage_meter_;
+  Status status_;
+};
+
+}  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,325 @@
+//  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).
+
+#include "db/blob/blob_counting_iterator.h"
+
+#include <string>
+#include <vector>
+
+#include "db/blob/blob_garbage_meter.h"
+#include "db/blob/blob_index.h"
+#include "db/blob/blob_log_format.h"
+#include "db/dbformat.h"
+#include "test_util/testharness.h"
+#include "test_util/testutil.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+void CheckInFlow(const BlobGarbageMeter& blob_garbage_meter,
+                 uint64_t blob_file_number, uint64_t count, uint64_t bytes) {
+  const auto& flows = blob_garbage_meter.flows();
+
+  const auto it = flows.find(blob_file_number);
+  if (it == flows.end()) {
+    ASSERT_EQ(count, 0);
+    ASSERT_EQ(bytes, 0);
+    return;
+  }
+
+  const auto& in = it->second.GetInFlow();
+
+  ASSERT_EQ(in.GetCount(), count);
+  ASSERT_EQ(in.GetBytes(), bytes);
+}
+
+TEST(BlobCountingIteratorTest, CountBlobs) {
+  // Note: the input consists of three key-values: two are blob references to
+  // different blob files, while the third one is a plain value.
+  constexpr char user_key0[] = "key0";
+  constexpr char user_key1[] = "key1";
+  constexpr char user_key2[] = "key2";
+
+  const std::vector<std::string> keys{
+      test::KeyStr(user_key0, 1, kTypeBlobIndex),
+      test::KeyStr(user_key1, 2, kTypeBlobIndex),
+      test::KeyStr(user_key2, 3, kTypeValue)};
+
+  constexpr uint64_t first_blob_file_number = 4;
+  constexpr uint64_t first_offset = 1000;
+  constexpr uint64_t first_size = 2000;
+
+  std::string first_blob_index;
+  BlobIndex::EncodeBlob(&first_blob_index, first_blob_file_number, first_offset,
+                        first_size, kNoCompression);
+
+  constexpr uint64_t second_blob_file_number = 6;
+  constexpr uint64_t second_offset = 2000;
+  constexpr uint64_t second_size = 4000;
+
+  std::string second_blob_index;
+  BlobIndex::EncodeBlob(&second_blob_index, second_blob_file_number,
+                        second_offset, second_size, kNoCompression);
+
+  const std::vector<std::string> values{first_blob_index, second_blob_index,
+                                        "raw_value"};
+
+  assert(keys.size() == values.size());
+
+  test::VectorIterator input(keys, values);
+  BlobGarbageMeter blob_garbage_meter;
+
+  BlobCountingIterator blob_counter(&input, &blob_garbage_meter);
+
+  constexpr uint64_t first_expected_bytes =
+      first_size +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(user_key0) - 1);
+  constexpr uint64_t second_expected_bytes =
+      second_size +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(user_key1) - 1);
+
+  // Call SeekToFirst and iterate forward
+  blob_counter.SeekToFirst();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[0]);
+  ASSERT_EQ(blob_counter.user_key(), user_key0);
+  ASSERT_EQ(blob_counter.value(), values[0]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 1,
+              first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 0, 0);
+
+  blob_counter.Next();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[1]);
+  ASSERT_EQ(blob_counter.user_key(), user_key1);
+  ASSERT_EQ(blob_counter.value(), values[1]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 1,
+              first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 1,
+              second_expected_bytes);
+
+  blob_counter.Next();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[2]);
+  ASSERT_EQ(blob_counter.user_key(), user_key2);
+  ASSERT_EQ(blob_counter.value(), values[2]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 1,
+              first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 1,
+              second_expected_bytes);
+
+  blob_counter.Next();
+  ASSERT_FALSE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 1,
+              first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 1,
+              second_expected_bytes);
+
+  // Do it again using NextAndGetResult
+  blob_counter.SeekToFirst();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[0]);
+  ASSERT_EQ(blob_counter.user_key(), user_key0);
+  ASSERT_EQ(blob_counter.value(), values[0]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+              2 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 1,
+              second_expected_bytes);
+
+  {
+    IterateResult result;
+    ASSERT_TRUE(blob_counter.NextAndGetResult(&result));
+    ASSERT_EQ(result.key, keys[1]);
+    ASSERT_EQ(blob_counter.user_key(), user_key1);
+    ASSERT_TRUE(blob_counter.Valid());
+    ASSERT_OK(blob_counter.status());
+    ASSERT_EQ(blob_counter.key(), keys[1]);
+    ASSERT_EQ(blob_counter.value(), values[1]);
+    CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+                2 * first_expected_bytes);
+    CheckInFlow(blob_garbage_meter, second_blob_file_number, 2,
+                2 * second_expected_bytes);
+  }
+
+  {
+    IterateResult result;
+    ASSERT_TRUE(blob_counter.NextAndGetResult(&result));
+    ASSERT_EQ(result.key, keys[2]);
+    ASSERT_EQ(blob_counter.user_key(), user_key2);
+    ASSERT_TRUE(blob_counter.Valid());
+    ASSERT_OK(blob_counter.status());
+    ASSERT_EQ(blob_counter.key(), keys[2]);
+    ASSERT_EQ(blob_counter.value(), values[2]);
+    CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+                2 * first_expected_bytes);
+    CheckInFlow(blob_garbage_meter, second_blob_file_number, 2,
+                2 * second_expected_bytes);
+  }
+
+  {
+    IterateResult result;
+    ASSERT_FALSE(blob_counter.NextAndGetResult(&result));
+    ASSERT_FALSE(blob_counter.Valid());
+    ASSERT_OK(blob_counter.status());
+    CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+                2 * first_expected_bytes);
+    CheckInFlow(blob_garbage_meter, second_blob_file_number, 2,
+                2 * second_expected_bytes);
+  }
+
+  // Call SeekToLast and iterate backward
+  blob_counter.SeekToLast();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[2]);
+  ASSERT_EQ(blob_counter.user_key(), user_key2);
+  ASSERT_EQ(blob_counter.value(), values[2]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+              2 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 2,
+              2 * second_expected_bytes);
+
+  blob_counter.Prev();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[1]);
+  ASSERT_EQ(blob_counter.user_key(), user_key1);
+  ASSERT_EQ(blob_counter.value(), values[1]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 2,
+              2 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 3,
+              3 * second_expected_bytes);
+
+  blob_counter.Prev();
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[0]);
+  ASSERT_EQ(blob_counter.user_key(), user_key0);
+  ASSERT_EQ(blob_counter.value(), values[0]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 3,
+              3 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 3,
+              3 * second_expected_bytes);
+
+  blob_counter.Prev();
+  ASSERT_FALSE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 3,
+              3 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 3,
+              3 * second_expected_bytes);
+
+  // Call Seek for all keys (plus one that's greater than all of them)
+  blob_counter.Seek(keys[0]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[0]);
+  ASSERT_EQ(blob_counter.user_key(), user_key0);
+  ASSERT_EQ(blob_counter.value(), values[0]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 4,
+              4 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 3,
+              3 * second_expected_bytes);
+
+  blob_counter.Seek(keys[1]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[1]);
+  ASSERT_EQ(blob_counter.user_key(), user_key1);
+  ASSERT_EQ(blob_counter.value(), values[1]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 4,
+              4 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 4,
+              4 * second_expected_bytes);
+
+  blob_counter.Seek(keys[2]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[2]);
+  ASSERT_EQ(blob_counter.user_key(), user_key2);
+  ASSERT_EQ(blob_counter.value(), values[2]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 4,
+              4 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 4,
+              4 * second_expected_bytes);
+
+  blob_counter.Seek("zzz");
+  ASSERT_FALSE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 4,
+              4 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 4,
+              4 * second_expected_bytes);
+
+  // Call SeekForPrev for all keys (plus one that's less than all of them)
+  blob_counter.SeekForPrev("aaa");
+  ASSERT_FALSE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 4,
+              4 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 4,
+              4 * second_expected_bytes);
+
+  blob_counter.SeekForPrev(keys[0]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[0]);
+  ASSERT_EQ(blob_counter.user_key(), user_key0);
+  ASSERT_EQ(blob_counter.value(), values[0]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 5,
+              5 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 4,
+              4 * second_expected_bytes);
+
+  blob_counter.SeekForPrev(keys[1]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[1]);
+  ASSERT_EQ(blob_counter.user_key(), user_key1);
+  ASSERT_EQ(blob_counter.value(), values[1]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 5,
+              5 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 5,
+              5 * second_expected_bytes);
+
+  blob_counter.SeekForPrev(keys[2]);
+  ASSERT_TRUE(blob_counter.Valid());
+  ASSERT_OK(blob_counter.status());
+  ASSERT_EQ(blob_counter.key(), keys[2]);
+  ASSERT_EQ(blob_counter.user_key(), user_key2);
+  ASSERT_EQ(blob_counter.value(), values[2]);
+  CheckInFlow(blob_garbage_meter, first_blob_file_number, 5,
+              5 * first_expected_bytes);
+  CheckInFlow(blob_garbage_meter, second_blob_file_number, 5,
+              5 * second_expected_bytes);
+}
+
+TEST(BlobCountingIteratorTest, CorruptBlobIndex) {
+  const std::vector<std::string> keys{
+      test::KeyStr("user_key", 1, kTypeBlobIndex)};
+  const std::vector<std::string> values{"i_am_not_a_blob_index"};
+
+  assert(keys.size() == values.size());
+
+  test::VectorIterator input(keys, values);
+  BlobGarbageMeter blob_garbage_meter;
+
+  BlobCountingIterator blob_counter(&input, &blob_garbage_meter);
+
+  blob_counter.SeekToFirst();
+  ASSERT_FALSE(blob_counter.Valid());
+  ASSERT_NOK(blob_counter.status());
+}
+
+}  // namespace ROCKSDB_NAMESPACE
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -135,7 +135,8 @@ std::ostream& operator<<(std::ostream& os,
     << " total_blob_count: " << blob_file_addition.GetTotalBlobCount()
     << " total_blob_bytes: " << blob_file_addition.GetTotalBlobBytes()
     << " checksum_method: " << blob_file_addition.GetChecksumMethod()
-     << " checksum_value: " << blob_file_addition.GetChecksumValue();
+     << " checksum_value: "
+     << Slice(blob_file_addition.GetChecksumValue()).ToString(/* hex */ true);

  return os;
 }
@@ -146,7 +147,8 @@ JSONWriter& operator<<(JSONWriter& jw,
     << "TotalBlobCount" << blob_file_addition.GetTotalBlobCount()
     << "TotalBlobBytes" << blob_file_addition.GetTotalBlobBytes()
     << "ChecksumMethod" << blob_file_addition.GetChecksumMethod()
-     << "ChecksumValue" << blob_file_addition.GetChecksumValue();
+     << "ChecksumValue"
+     << Slice(blob_file_addition.GetChecksumValue()).ToString(/* hex */ true);

  return jw;
 }
@@ -46,7 +46,9 @@ TEST_F(BlobFileAdditionTest, NonEmpty) {
  constexpr uint64_t total_blob_count = 2;
  constexpr uint64_t total_blob_bytes = 123456;
  const std::string checksum_method("SHA1");
-  const std::string checksum_value("bdb7f34a59dfa1592ce7f52e99f98c570c525cbd");
+  const std::string checksum_value(
+      "\xbd\xb7\xf3\x4a\x59\xdf\xa1\x59\x2c\xe7\xf5\x2e\x99\xf9\x8c\x57\x0c\x52"
+      "\x5c\xbd");

  BlobFileAddition blob_file_addition(blob_file_number, total_blob_count,
                                      total_blob_bytes, checksum_method,
@@ -113,7 +115,9 @@ TEST_F(BlobFileAdditionTest, DecodeErrors) {
    ASSERT_TRUE(std::strstr(s.getState(), "checksum value"));
  }

-  constexpr char checksum_value[] = "bdb7f34a59dfa1592ce7f52e99f98c570c525cbd";
+  constexpr char checksum_value[] =
+      "\xbd\xb7\xf3\x4a\x59\xdf\xa1\x59\x2c\xe7\xf5\x2e\x99\xf9\x8c\x57\x0c\x52"
+      "\x5c\xbd";
  PutLengthPrefixedSlice(&str, checksum_value);
  slice = str;

@@ -150,7 +154,7 @@ TEST_F(BlobFileAdditionTest, ForwardCompatibleCustomField) {
  constexpr uint64_t total_blob_count = 9999;
  constexpr uint64_t total_blob_bytes = 100000000;
  const std::string checksum_method("CRC32");
-  const std::string checksum_value("3d87ff57");
+  const std::string checksum_value("\x3d\x87\xff\x57");

  BlobFileAddition blob_file_addition(blob_file_number, total_blob_count,
                                      total_blob_bytes, checksum_method,
@@ -178,7 +182,7 @@ TEST_F(BlobFileAdditionTest, ForwardIncompatibleCustomField) {
  constexpr uint64_t total_blob_count = 100;
  constexpr uint64_t total_blob_bytes = 2000000;
  const std::string checksum_method("CRC32B");
-  const std::string checksum_value("6dbdf23a");
+  const std::string checksum_value("\x6d\xbd\xf2\x3a");

  BlobFileAddition blob_file_addition(blob_file_number, total_blob_count,
                                      total_blob_bytes, checksum_method,
@@ -12,6 +12,7 @@
 #include "db/blob/blob_index.h"
 #include "db/blob/blob_log_format.h"
 #include "db/blob/blob_log_writer.h"
+#include "db/event_helpers.h"
 #include "db/version_set.h"
 #include "file/filename.h"
 #include "file/read_write_util.h"
@@ -36,13 +37,14 @@ BlobFileBuilder::BlobFileBuilder(
    Env::WriteLifeTimeHint write_hint,
    const std::shared_ptr<IOTracer>& io_tracer,
    BlobFileCompletionCallback* blob_callback,
+    BlobFileCreationReason creation_reason,
    std::vector<std::string>* blob_file_paths,
    std::vector<BlobFileAddition>* blob_file_additions)
    : BlobFileBuilder([versions]() { return versions->NewFileNumber(); }, fs,
                      immutable_options, mutable_cf_options, file_options,
                      job_id, column_family_id, column_family_name, io_priority,
-                      write_hint, io_tracer, blob_callback, blob_file_paths,
-                      blob_file_additions) {}
+                      write_hint, io_tracer, blob_callback, creation_reason,
+                      blob_file_paths, blob_file_additions) {}

 BlobFileBuilder::BlobFileBuilder(
    std::function<uint64_t()> file_number_generator, FileSystem* fs,
@@ -53,6 +55,7 @@ BlobFileBuilder::BlobFileBuilder(
    Env::WriteLifeTimeHint write_hint,
    const std::shared_ptr<IOTracer>& io_tracer,
    BlobFileCompletionCallback* blob_callback,
+    BlobFileCreationReason creation_reason,
    std::vector<std::string>* blob_file_paths,
    std::vector<BlobFileAddition>* blob_file_additions)
    : file_number_generator_(std::move(file_number_generator)),
@@ -69,6 +72,7 @@ BlobFileBuilder::BlobFileBuilder(
      write_hint_(write_hint),
      io_tracer_(io_tracer),
      blob_callback_(blob_callback),
+      creation_reason_(creation_reason),
      blob_file_paths_(blob_file_paths),
      blob_file_additions_(blob_file_additions),
      blob_count_(0),
@@ -161,6 +165,11 @@ Status BlobFileBuilder::OpenBlobFileIfNeeded() {
  std::string blob_file_path =
      BlobFileName(immutable_options_->cf_paths.front().path, blob_file_number);

+  if (blob_callback_) {
+    blob_callback_->OnBlobFileCreationStarted(
+        blob_file_path, column_family_name_, job_id_, creation_reason_);
+  }
+
  std::unique_ptr<FSWritableFile> file;

  {
@@ -191,7 +200,7 @@ Status BlobFileBuilder::OpenBlobFileIfNeeded() {
      immutable_options_->clock, io_tracer_, statistics,
      immutable_options_->listeners,
      immutable_options_->file_checksum_gen_factory.get(),
-      tmp_set.Contains(FileType::kBlobFile)));
+      tmp_set.Contains(FileType::kBlobFile), false));

  constexpr bool do_flush = false;

@@ -228,6 +237,7 @@ Status BlobFileBuilder::CompressBlobIfNeeded(
  assert(blob);
  assert(compressed_blob);
  assert(compressed_blob->empty());
+  assert(immutable_options_);

  if (blob_compression_type_ == kNoCompression) {
    return Status::OK();
@@ -242,7 +252,16 @@ Status BlobFileBuilder::CompressBlobIfNeeded(

  constexpr uint32_t compression_format_version = 2;

-  if (!CompressData(*blob, info, compression_format_version, compressed_blob)) {
+  bool success = false;
+
+  {
+    StopWatch stop_watch(immutable_options_->clock, immutable_options_->stats,
+                         BLOB_DB_COMPRESSION_MICROS);
+    success =
+        CompressData(*blob, info, compression_format_version, compressed_blob);
+  }
+
+  if (!success) {
    return Status::Corruption("Error compressing blob");
  }

@@ -295,6 +314,13 @@ Status BlobFileBuilder::CloseBlobFile() {

  const uint64_t blob_file_number = writer_->get_log_number();

+  if (blob_callback_) {
+    s = blob_callback_->OnBlobFileCompleted(
+        blob_file_paths_->back(), column_family_name_, job_id_,
+        blob_file_number, creation_reason_, s, checksum_value, checksum_method,
+        blob_count_, blob_bytes_);
+  }
+
  assert(blob_file_additions_);
  blob_file_additions_->emplace_back(blob_file_number, blob_count_, blob_bytes_,
                                     std::move(checksum_method),
@@ -306,9 +332,6 @@ Status BlobFileBuilder::CloseBlobFile() {
                 " total blobs, %" PRIu64 " total bytes",
                 column_family_name_.c_str(), job_id_, blob_file_number,
                 blob_count_, blob_bytes_);
-  if (blob_callback_) {
-    s = blob_callback_->OnBlobFileCompleted(blob_file_paths_->back());
-  }

  writer_.reset();
  blob_count_ = 0;
@@ -330,15 +353,18 @@ Status BlobFileBuilder::CloseBlobFileIfNeeded() {
  return CloseBlobFile();
 }

-void BlobFileBuilder::Abandon() {
+void BlobFileBuilder::Abandon(const Status& s) {
  if (!IsBlobFileOpen()) {
    return;
  }
-
  if (blob_callback_) {
    // BlobFileBuilder::Abandon() is called because of error while writing to
    // Blob files. So we can ignore the below error.
-    blob_callback_->OnBlobFileCompleted(blob_file_paths_->back())
+    blob_callback_
+        ->OnBlobFileCompleted(blob_file_paths_->back(), column_family_name_,
+                              job_id_, writer_->get_log_number(),
+                              creation_reason_, s, "", "", blob_count_,
+                              blob_bytes_)
        .PermitUncheckedError();
  }

@@ -13,6 +13,7 @@
 #include "rocksdb/compression_type.h"
 #include "rocksdb/env.h"
 #include "rocksdb/rocksdb_namespace.h"
+#include "rocksdb/types.h"

 namespace ROCKSDB_NAMESPACE {

@@ -41,6 +42,7 @@ class BlobFileBuilder {
                  Env::WriteLifeTimeHint write_hint,
                  const std::shared_ptr<IOTracer>& io_tracer,
                  BlobFileCompletionCallback* blob_callback,
+                  BlobFileCreationReason creation_reason,
                  std::vector<std::string>* blob_file_paths,
                  std::vector<BlobFileAddition>* blob_file_additions);

@@ -54,6 +56,7 @@ class BlobFileBuilder {
                  Env::WriteLifeTimeHint write_hint,
                  const std::shared_ptr<IOTracer>& io_tracer,
                  BlobFileCompletionCallback* blob_callback,
+                  BlobFileCreationReason creation_reason,
                  std::vector<std::string>* blob_file_paths,
                  std::vector<BlobFileAddition>* blob_file_additions);

@@ -64,7 +67,7 @@ class BlobFileBuilder {

  Status Add(const Slice& key, const Slice& value, std::string* blob_index);
  Status Finish();
-  void Abandon();
+  void Abandon(const Status& s);

 private:
  bool IsBlobFileOpen() const;
@@ -89,6 +92,7 @@ class BlobFileBuilder {
  Env::WriteLifeTimeHint write_hint_;
  std::shared_ptr<IOTracer> io_tracer_;
  BlobFileCompletionCallback* blob_callback_;
+  BlobFileCreationReason creation_reason_;
  std::vector<std::string>* blob_file_paths_;
  std::vector<BlobFileAddition>* blob_file_additions_;
  std::unique_ptr<BlobLogWriter> writer_;
@@ -145,7 +145,7 @@ TEST_F(BlobFileBuilderTest, BuildAndCheckOneFile) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  std::vector<std::pair<std::string, std::string>> expected_key_value_pairs(
      number_of_blobs);
@@ -229,7 +229,7 @@ TEST_F(BlobFileBuilderTest, BuildAndCheckMultipleFiles) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  std::vector<std::pair<std::string, std::string>> expected_key_value_pairs(
      number_of_blobs);
@@ -315,7 +315,7 @@ TEST_F(BlobFileBuilderTest, InlinedValues) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  for (size_t i = 0; i < number_of_blobs; ++i) {
    const std::string key = std::to_string(i);
@@ -368,7 +368,7 @@ TEST_F(BlobFileBuilderTest, Compression) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  const std::string key("1");
  const std::string uncompressed_value(value_size, 'x');
@@ -450,7 +450,7 @@ TEST_F(BlobFileBuilderTest, CompressionError) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  SyncPoint::GetInstance()->SetCallBack("CompressData:TamperWithReturnValue",
                                        [](void* arg) {
@@ -528,7 +528,7 @@ TEST_F(BlobFileBuilderTest, Checksum) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  const std::string key("1");
  const std::string value("deadbeef");
@@ -624,7 +624,7 @@ TEST_P(BlobFileBuilderIOErrorTest, IOError) {
      TestFileNumberGenerator(), fs_, &immutable_options, &mutable_cf_options,
      &file_options_, job_id, column_family_id, column_family_name, io_priority,
      write_hint, nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
-      &blob_file_paths, &blob_file_additions);
+      BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);

  SyncPoint::GetInstance()->SetCallBack(sync_point_, [this](void* arg) {
    Status* const s = static_cast<Status*>(arg);
@@ -9,6 +9,7 @@
 #pragma once

 #include "db/error_handler.h"
+#include "db/event_helpers.h"
 #include "file/sst_file_manager_impl.h"
 #include "rocksdb/status.h"

@@ -16,23 +17,46 @@ namespace ROCKSDB_NAMESPACE {

 class BlobFileCompletionCallback {
 public:
-#ifdef ROCKSDB_LITE
-  BlobFileCompletionCallback(SstFileManager* /*sst_file_manager*/,
-                             InstrumentedMutex* /*mutex*/,
-                             ErrorHandler* /*error_handler*/) {}
-  Status OnBlobFileCompleted(const std::string& /*file_name*/) {
-    return Status::OK();
-  }
-#else
-  BlobFileCompletionCallback(SstFileManager* sst_file_manager,
-                             InstrumentedMutex* mutex,
-                             ErrorHandler* error_handler)
+  BlobFileCompletionCallback(
+      SstFileManager* sst_file_manager, InstrumentedMutex* mutex,
+      ErrorHandler* error_handler, EventLogger* event_logger,
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      const std::string& dbname)
      : sst_file_manager_(sst_file_manager),
        mutex_(mutex),
-        error_handler_(error_handler) {}
+        error_handler_(error_handler),
+        event_logger_(event_logger),
+        listeners_(listeners),
+        dbname_(dbname) {}

-  Status OnBlobFileCompleted(const std::string& file_name) {
+  void OnBlobFileCreationStarted(const std::string& file_name,
+                                 const std::string& column_family_name,
+                                 int job_id,
+                                 BlobFileCreationReason creation_reason) {
+#ifndef ROCKSDB_LITE
+    // Notify the listeners.
+    EventHelpers::NotifyBlobFileCreationStarted(listeners_, dbname_,
+                                                column_family_name, file_name,
+                                                job_id, creation_reason);
+#else
+    (void)file_name;
+    (void)column_family_name;
+    (void)job_id;
+    (void)creation_reason;
+#endif
+  }
+
+  Status OnBlobFileCompleted(const std::string& file_name,
+                             const std::string& column_family_name, int job_id,
+                             uint64_t file_number,
+                             BlobFileCreationReason creation_reason,
+                             const Status& report_status,
+                             const std::string& checksum_value,
+                             const std::string& checksum_method,
+                             uint64_t blob_count, uint64_t blob_bytes) {
    Status s;
+
+#ifndef ROCKSDB_LITE
    auto sfm = static_cast<SstFileManagerImpl*>(sst_file_manager_);
    if (sfm) {
      // Report new blob files to SstFileManagerImpl
@@ -45,6 +69,17 @@ class BlobFileCompletionCallback {
        error_handler_->SetBGError(s, BackgroundErrorReason::kFlush);
      }
    }
+#endif  // !ROCKSDB_LITE
+
+    // Notify the listeners.
+    EventHelpers::LogAndNotifyBlobFileCreationFinished(
+        event_logger_, listeners_, dbname_, column_family_name, file_name,
+        job_id, file_number, creation_reason,
+        (!report_status.ok() ? report_status : s),
+        (checksum_value.empty() ? kUnknownFileChecksum : checksum_value),
+        (checksum_method.empty() ? kUnknownFileChecksumFuncName
+                                 : checksum_method),
+        blob_count, blob_bytes);
    return s;
  }

@@ -52,6 +87,8 @@ class BlobFileCompletionCallback {
  SstFileManager* sst_file_manager_;
  InstrumentedMutex* mutex_;
  ErrorHandler* error_handler_;
-#endif  // ROCKSDB_LITE
+  EventLogger* event_logger_;
+  std::vector<std::shared_ptr<EventListener>> listeners_;
+  std::string dbname_;
 };
 }  // namespace ROCKSDB_NAMESPACE
@@ -8,7 +8,13 @@
 #include <ostream>
 #include <sstream>

+#include "db/blob/blob_log_format.h"
+#include "rocksdb/slice.h"
+
 namespace ROCKSDB_NAMESPACE {
+uint64_t SharedBlobFileMetaData::GetBlobFileSize() const {
+  return BlobLogHeader::kSize + total_blob_bytes_ + BlobLogFooter::kSize;
+}

 std::string SharedBlobFileMetaData::DebugString() const {
  std::ostringstream oss;
@@ -23,7 +29,8 @@ std::ostream& operator<<(std::ostream& os,
     << " total_blob_count: " << shared_meta.GetTotalBlobCount()
     << " total_blob_bytes: " << shared_meta.GetTotalBlobBytes()
     << " checksum_method: " << shared_meta.GetChecksumMethod()
-     << " checksum_value: " << shared_meta.GetChecksumValue();
+     << " checksum_value: "
+     << Slice(shared_meta.GetChecksumValue()).ToString(/* hex */ true);

  return os;
 }
@@ -51,6 +51,7 @@ class SharedBlobFileMetaData {
  SharedBlobFileMetaData(SharedBlobFileMetaData&&) = delete;
  SharedBlobFileMetaData& operator=(SharedBlobFileMetaData&&) = delete;

+  uint64_t GetBlobFileSize() const;
  uint64_t GetBlobFileNumber() const { return blob_file_number_; }
  uint64_t GetTotalBlobCount() const { return total_blob_count_; }
  uint64_t GetTotalBlobBytes() const { return total_blob_bytes_; }
@@ -112,6 +113,11 @@ class BlobFileMetaData {
    return shared_meta_;
  }

+  uint64_t GetBlobFileSize() const {
+    assert(shared_meta_);
+    return shared_meta_->GetBlobFileSize();
+  }
+
  uint64_t GetBlobFileNumber() const {
    assert(shared_meta_);
    return shared_meta_->GetBlobFileNumber();
@@ -10,6 +10,7 @@

 #include "db/blob/blob_log_format.h"
 #include "file/filename.h"
+#include "monitoring/statistics.h"
 #include "options/cf_options.h"
 #include "rocksdb/file_system.h"
 #include "rocksdb/slice.h"
@@ -17,6 +18,7 @@
 #include "test_util/sync_point.h"
 #include "util/compression.h"
 #include "util/crc32c.h"
+#include "util/stop_watch.h"

 namespace ROCKSDB_NAMESPACE {

@@ -42,25 +44,28 @@ Status BlobFileReader::Create(

  assert(file_reader);

+  Statistics* const statistics = immutable_options.stats;
+
  CompressionType compression_type = kNoCompression;

  {
-    const Status s =
-        ReadHeader(file_reader.get(), column_family_id, &compression_type);
+    const Status s = ReadHeader(file_reader.get(), column_family_id, statistics,
+                                &compression_type);
    if (!s.ok()) {
      return s;
    }
  }

  {
-    const Status s = ReadFooter(file_size, file_reader.get());
+    const Status s = ReadFooter(file_reader.get(), file_size, statistics);
    if (!s.ok()) {
      return s;
    }
  }

  blob_file_reader->reset(
-      new BlobFileReader(std::move(file_reader), file_size, compression_type));
+      new BlobFileReader(std::move(file_reader), file_size, compression_type,
+                         immutable_options.clock, statistics));

  return Status::OK();
 }
@@ -127,6 +132,7 @@ Status BlobFileReader::OpenFile(

 Status BlobFileReader::ReadHeader(const RandomAccessFileReader* file_reader,
                                  uint32_t column_family_id,
+                                  Statistics* statistics,
                                  CompressionType* compression_type) {
  assert(file_reader);
  assert(compression_type);
@@ -141,8 +147,9 @@ Status BlobFileReader::ReadHeader(const RandomAccessFileReader* file_reader,
    constexpr uint64_t read_offset = 0;
    constexpr size_t read_size = BlobLogHeader::kSize;

-    const Status s = ReadFromFile(file_reader, read_offset, read_size,
-                                  &header_slice, &buf, &aligned_buf);
+    const Status s =
+        ReadFromFile(file_reader, read_offset, read_size, statistics,
+                     &header_slice, &buf, &aligned_buf);
    if (!s.ok()) {
      return s;
    }
@@ -175,8 +182,8 @@ Status BlobFileReader::ReadHeader(const RandomAccessFileReader* file_reader,
  return Status::OK();
 }

-Status BlobFileReader::ReadFooter(uint64_t file_size,
-                                  const RandomAccessFileReader* file_reader) {
+Status BlobFileReader::ReadFooter(const RandomAccessFileReader* file_reader,
+                                  uint64_t file_size, Statistics* statistics) {
  assert(file_size >= BlobLogHeader::kSize + BlobLogFooter::kSize);
  assert(file_reader);

@@ -190,8 +197,9 @@ Status BlobFileReader::ReadFooter(uint64_t file_size,
    const uint64_t read_offset = file_size - BlobLogFooter::kSize;
    constexpr size_t read_size = BlobLogFooter::kSize;

-    const Status s = ReadFromFile(file_reader, read_offset, read_size,
-                                  &footer_slice, &buf, &aligned_buf);
+    const Status s =
+        ReadFromFile(file_reader, read_offset, read_size, statistics,
+                     &footer_slice, &buf, &aligned_buf);
    if (!s.ok()) {
      return s;
    }
@@ -220,14 +228,16 @@ Status BlobFileReader::ReadFooter(uint64_t file_size,

 Status BlobFileReader::ReadFromFile(const RandomAccessFileReader* file_reader,
                                    uint64_t read_offset, size_t read_size,
-                                    Slice* slice, Buffer* buf,
-                                    AlignedBuf* aligned_buf) {
+                                    Statistics* statistics, Slice* slice,
+                                    Buffer* buf, AlignedBuf* aligned_buf) {
  assert(slice);
  assert(buf);
  assert(aligned_buf);

  assert(file_reader);

+  RecordTick(statistics, BLOB_DB_BLOB_FILE_BYTES_READ, read_size);
+
  Status s;

  if (file_reader->use_direct_io()) {
@@ -256,10 +266,13 @@ Status BlobFileReader::ReadFromFile(const RandomAccessFileReader* file_reader,

 BlobFileReader::BlobFileReader(
    std::unique_ptr<RandomAccessFileReader>&& file_reader, uint64_t file_size,
-    CompressionType compression_type)
+    CompressionType compression_type, SystemClock* clock,
+    Statistics* statistics)
    : file_reader_(std::move(file_reader)),
      file_size_(file_size),
-      compression_type_(compression_type) {
+      compression_type_(compression_type),
+      clock_(clock),
+      statistics_(statistics) {
  assert(file_reader_);
 }

@@ -304,7 +317,7 @@ Status BlobFileReader::GetBlob(const ReadOptions& read_options,
    TEST_SYNC_POINT("BlobFileReader::GetBlob:ReadFromFile");

    const Status s = ReadFromFile(file_reader_.get(), record_offset,
-                                  static_cast<size_t>(record_size),
+                                  static_cast<size_t>(record_size), statistics_,
                                  &record_slice, &buf, &aligned_buf);
    if (!s.ok()) {
      return s;
@@ -324,8 +337,8 @@ Status BlobFileReader::GetBlob(const ReadOptions& read_options,
  const Slice value_slice(record_slice.data() + adjustment, value_size);

  {
-    const Status s =
-        UncompressBlobIfNeeded(value_slice, compression_type, value);
+    const Status s = UncompressBlobIfNeeded(value_slice, compression_type,
+                                            clock_, statistics_, value);
    if (!s.ok()) {
      return s;
    }
@@ -338,6 +351,117 @@ Status BlobFileReader::GetBlob(const ReadOptions& read_options,
  return Status::OK();
 }

+void BlobFileReader::MultiGetBlob(
+    const ReadOptions& read_options,
+    const autovector<std::reference_wrapper<const Slice>>& user_keys,
+    const autovector<uint64_t>& offsets,
+    const autovector<uint64_t>& value_sizes, autovector<Status*>& statuses,
+    autovector<PinnableSlice*>& values, uint64_t* bytes_read) const {
+  const size_t num_blobs = user_keys.size();
+  assert(num_blobs == offsets.size());
+  assert(num_blobs == value_sizes.size());
+  assert(num_blobs == statuses.size());
+  assert(num_blobs == values.size());
+
+  std::vector<FSReadRequest> read_reqs(num_blobs);
+  autovector<uint64_t> adjustments;
+  uint64_t total_len = 0;
+  for (size_t i = 0; i < num_blobs; ++i) {
+    const size_t key_size = user_keys[i].get().size();
+    assert(IsValidBlobOffset(offsets[i], key_size, value_sizes[i], file_size_));
+    const uint64_t adjustment =
+        read_options.verify_checksums
+            ? BlobLogRecord::CalculateAdjustmentForRecordHeader(key_size)
+            : 0;
+    assert(offsets[i] >= adjustment);
+    adjustments.push_back(adjustment);
+    read_reqs[i].offset = offsets[i] - adjustment;
+    read_reqs[i].len = value_sizes[i] + adjustment;
+    total_len += read_reqs[i].len;
+  }
+
+  RecordTick(statistics_, BLOB_DB_BLOB_FILE_BYTES_READ, total_len);
+
+  Buffer buf;
+  AlignedBuf aligned_buf;
+
+  Status s;
+  bool direct_io = file_reader_->use_direct_io();
+  if (direct_io) {
+    for (size_t i = 0; i < read_reqs.size(); ++i) {
+      read_reqs[i].scratch = nullptr;
+    }
+  } else {
+    buf.reset(new char[total_len]);
+    std::ptrdiff_t pos = 0;
+    for (size_t i = 0; i < read_reqs.size(); ++i) {
+      read_reqs[i].scratch = buf.get() + pos;
+      pos += read_reqs[i].len;
+    }
+  }
+  TEST_SYNC_POINT("BlobFileReader::MultiGetBlob:ReadFromFile");
+  s = file_reader_->MultiRead(IOOptions(), read_reqs.data(), read_reqs.size(),
+                              direct_io ? &aligned_buf : nullptr);
+  if (!s.ok()) {
+    for (auto& req : read_reqs) {
+      req.status.PermitUncheckedError();
+    }
+    for (size_t i = 0; i < num_blobs; ++i) {
+      assert(statuses[i]);
+      *statuses[i] = s;
+    }
+    return;
+  }
+
+  assert(s.ok());
+  for (size_t i = 0; i < num_blobs; ++i) {
+    auto& req = read_reqs[i];
+    assert(statuses[i]);
+    if (req.status.ok() && req.result.size() != req.len) {
+      req.status = IOStatus::Corruption("Failed to read data from blob file");
+    }
+    *statuses[i] = req.status;
+  }
+
+  if (read_options.verify_checksums) {
+    for (size_t i = 0; i < num_blobs; ++i) {
+      assert(statuses[i]);
+      if (!statuses[i]->ok()) {
+        continue;
+      }
+      const Slice& record_slice = read_reqs[i].result;
+      s = VerifyBlob(record_slice, user_keys[i], value_sizes[i]);
+      if (!s.ok()) {
+        assert(statuses[i]);
+        *statuses[i] = s;
+      }
+    }
+  }
+
+  for (size_t i = 0; i < num_blobs; ++i) {
+    assert(statuses[i]);
+    if (!statuses[i]->ok()) {
+      continue;
+    }
+    const Slice& record_slice = read_reqs[i].result;
+    const Slice value_slice(record_slice.data() + adjustments[i],
+                            value_sizes[i]);
+    s = UncompressBlobIfNeeded(value_slice, compression_type_, clock_,
+                               statistics_, values[i]);
+    if (!s.ok()) {
+      *statuses[i] = s;
+    }
+  }
+
+  if (bytes_read) {
+    uint64_t total_bytes = 0;
+    for (const auto& req : read_reqs) {
+      total_bytes += req.result.size();
+    }
+    *bytes_read = total_bytes;
+  }
+}
+
 Status BlobFileReader::VerifyBlob(const Slice& record_slice,
                                  const Slice& user_key, uint64_t value_size) {
  BlobLogRecord record;
@@ -382,6 +506,8 @@ Status BlobFileReader::VerifyBlob(const Slice& record_slice,

 Status BlobFileReader::UncompressBlobIfNeeded(const Slice& value_slice,
                                              CompressionType compression_type,
+                                              SystemClock* clock,
+                                              Statistics* statistics,
                                              PinnableSlice* value) {
  assert(value);

@@ -399,9 +525,14 @@ Status BlobFileReader::UncompressBlobIfNeeded(const Slice& value_slice,
  constexpr uint32_t compression_format_version = 2;
  constexpr MemoryAllocator* allocator = nullptr;

-  CacheAllocationPtr output =
-      UncompressData(info, value_slice.data(), value_slice.size(),
-                     &uncompressed_size, compression_format_version, allocator);
+  CacheAllocationPtr output;
+
+  {
+    StopWatch stop_watch(clock, statistics, BLOB_DB_DECOMPRESSION_MICROS);
+    output = UncompressData(info, value_slice.data(), value_slice.size(),
+                            &uncompressed_size, compression_format_version,
+                            allocator);
+  }

  TEST_SYNC_POINT_CALLBACK(
      "BlobFileReader::UncompressBlobIfNeeded:TamperWithResult", &output);
@@ -11,6 +11,7 @@
 #include "file/random_access_file_reader.h"
 #include "rocksdb/compression_type.h"
 #include "rocksdb/rocksdb_namespace.h"
+#include "util/autovector.h"

 namespace ROCKSDB_NAMESPACE {

@@ -21,6 +22,7 @@ class HistogramImpl;
 struct ReadOptions;
 class Slice;
 class PinnableSlice;
+class Statistics;

 class BlobFileReader {
 public:
@@ -42,9 +44,21 @@ class BlobFileReader {
                 CompressionType compression_type, PinnableSlice* value,
                 uint64_t* bytes_read) const;

+  void MultiGetBlob(
+      const ReadOptions& read_options,
+      const autovector<std::reference_wrapper<const Slice>>& user_keys,
+      const autovector<uint64_t>& offsets,
+      const autovector<uint64_t>& value_sizes, autovector<Status*>& statuses,
+      autovector<PinnableSlice*>& values, uint64_t* bytes_read) const;
+
+  CompressionType GetCompressionType() const { return compression_type_; }
+
+  uint64_t GetFileSize() const { return file_size_; }
+
 private:
  BlobFileReader(std::unique_ptr<RandomAccessFileReader>&& file_reader,
-                 uint64_t file_size, CompressionType compression_type);
+                 uint64_t file_size, CompressionType compression_type,
+                 SystemClock* clock, Statistics* statistics);

  static Status OpenFile(const ImmutableOptions& immutable_options,
                         const FileOptions& file_opts,
@@ -55,17 +69,17 @@ class BlobFileReader {
                         std::unique_ptr<RandomAccessFileReader>* file_reader);

  static Status ReadHeader(const RandomAccessFileReader* file_reader,
-                           uint32_t column_family_id,
+                           uint32_t column_family_id, Statistics* statistics,
                           CompressionType* compression_type);

-  static Status ReadFooter(uint64_t file_size,
-                           const RandomAccessFileReader* file_reader);
+  static Status ReadFooter(const RandomAccessFileReader* file_reader,
+                           uint64_t file_size, Statistics* statistics);

  using Buffer = std::unique_ptr<char[]>;

  static Status ReadFromFile(const RandomAccessFileReader* file_reader,
                             uint64_t read_offset, size_t read_size,
-                             Slice* slice, Buffer* buf,
+                             Statistics* statistics, Slice* slice, Buffer* buf,
                             AlignedBuf* aligned_buf);

  static Status VerifyBlob(const Slice& record_slice, const Slice& user_key,
@@ -73,6 +87,8 @@ class BlobFileReader {

  static Status UncompressBlobIfNeeded(const Slice& value_slice,
                                       CompressionType compression_type,
+                                       SystemClock* clock,
+                                       Statistics* statistics,
                                       PinnableSlice* value);

  static void SaveValue(const Slice& src, PinnableSlice* dst);
@@ -80,6 +96,8 @@ class BlobFileReader {
  std::unique_ptr<RandomAccessFileReader> file_reader_;
  uint64_t file_size_;
  CompressionType compression_type_;
+  SystemClock* clock_;
+  Statistics* statistics_;
 };

 }  // namespace ROCKSDB_NAMESPACE
@@ -27,23 +27,23 @@ namespace ROCKSDB_NAMESPACE {

 namespace {

-// Creates a test blob file with a single blob in it. Note: this method
-// makes it possible to test various corner cases by allowing the caller
-// to specify the contents of various blob file header/footer fields.
+// Creates a test blob file with `num` blobs in it.
 void WriteBlobFile(const ImmutableOptions& immutable_options,
                   uint32_t column_family_id, bool has_ttl,
                   const ExpirationRange& expiration_range_header,
                   const ExpirationRange& expiration_range_footer,
-                   uint64_t blob_file_number, const Slice& key,
-                   const Slice& blob, CompressionType compression_type,
-                   uint64_t* blob_offset, uint64_t* blob_size) {
+                   uint64_t blob_file_number, const std::vector<Slice>& keys,
+                   const std::vector<Slice>& blobs, CompressionType compression,
+                   std::vector<uint64_t>& blob_offsets,
+                   std::vector<uint64_t>& blob_sizes) {
  assert(!immutable_options.cf_paths.empty());
-  assert(blob_offset);
-  assert(blob_size);
+  size_t num = keys.size();
+  assert(num == blobs.size());
+  assert(num == blob_offsets.size());
+  assert(num == blob_sizes.size());

  const std::string blob_file_path =
      BlobFileName(immutable_options.cf_paths.front().path, blob_file_number);
-
  std::unique_ptr<FSWritableFile> file;
  ASSERT_OK(NewWritableFile(immutable_options.fs.get(), blob_file_path, &file,
                            FileOptions()));
@@ -59,50 +59,77 @@ void WriteBlobFile(const ImmutableOptions& immutable_options,
                                statistics, blob_file_number, use_fsync,
                                do_flush);

-  BlobLogHeader header(column_family_id, compression_type, has_ttl,
+  BlobLogHeader header(column_family_id, compression, has_ttl,
                       expiration_range_header);

  ASSERT_OK(blob_log_writer.WriteHeader(header));

-  std::string compressed_blob;
-  Slice blob_to_write;
-
-  if (compression_type == kNoCompression) {
-    blob_to_write = blob;
-    *blob_size = blob.size();
+  std::vector<std::string> compressed_blobs(num);
+  std::vector<Slice> blobs_to_write(num);
+  if (kNoCompression == compression) {
+    for (size_t i = 0; i < num; ++i) {
+      blobs_to_write[i] = blobs[i];
+      blob_sizes[i] = blobs[i].size();
+    }
  } else {
    CompressionOptions opts;
-    CompressionContext context(compression_type);
+    CompressionContext context(compression);
    constexpr uint64_t sample_for_compression = 0;
-
    CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
-                         compression_type, sample_for_compression);
+                         compression, sample_for_compression);

    constexpr uint32_t compression_format_version = 2;

-    ASSERT_TRUE(
-        CompressData(blob, info, compression_format_version, &compressed_blob));
-
-    blob_to_write = compressed_blob;
-    *blob_size = compressed_blob.size();
+    for (size_t i = 0; i < num; ++i) {
+      ASSERT_TRUE(CompressData(blobs[i], info, compression_format_version,
+                               &compressed_blobs[i]));
+      blobs_to_write[i] = compressed_blobs[i];
+      blob_sizes[i] = compressed_blobs[i].size();
+    }
  }

-  uint64_t key_offset = 0;
-
-  ASSERT_OK(
-      blob_log_writer.AddRecord(key, blob_to_write, &key_offset, blob_offset));
+  for (size_t i = 0; i < num; ++i) {
+    uint64_t key_offset = 0;
+    ASSERT_OK(blob_log_writer.AddRecord(keys[i], blobs_to_write[i], &key_offset,
+                                        &blob_offsets[i]));
+  }

  BlobLogFooter footer;
-  footer.blob_count = 1;
+  footer.blob_count = num;
  footer.expiration_range = expiration_range_footer;

  std::string checksum_method;
  std::string checksum_value;
-
  ASSERT_OK(
      blob_log_writer.AppendFooter(footer, &checksum_method, &checksum_value));
 }

+// Creates a test blob file with a single blob in it. Note: this method
+// makes it possible to test various corner cases by allowing the caller
+// to specify the contents of various blob file header/footer fields.
+void WriteBlobFile(const ImmutableOptions& immutable_options,
+                   uint32_t column_family_id, bool has_ttl,
+                   const ExpirationRange& expiration_range_header,
+                   const ExpirationRange& expiration_range_footer,
+                   uint64_t blob_file_number, const Slice& key,
+                   const Slice& blob, CompressionType compression,
+                   uint64_t* blob_offset, uint64_t* blob_size) {
+  std::vector<Slice> keys{key};
+  std::vector<Slice> blobs{blob};
+  std::vector<uint64_t> blob_offsets{0};
+  std::vector<uint64_t> blob_sizes{0};
+  WriteBlobFile(immutable_options, column_family_id, has_ttl,
+                expiration_range_header, expiration_range_footer,
+                blob_file_number, keys, blobs, compression, blob_offsets,
+                blob_sizes);
+  if (blob_offset) {
+    *blob_offset = blob_offsets[0];
+  }
+  if (blob_size) {
+    *blob_size = blob_sizes[0];
+  }
+}
+
 }  // anonymous namespace

 class BlobFileReaderTest : public testing::Test {
@@ -127,15 +154,19 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
  constexpr bool has_ttl = false;
  constexpr ExpirationRange expiration_range;
  constexpr uint64_t blob_file_number = 1;
-  constexpr char key[] = "key";
-  constexpr char blob[] = "blob";
+  constexpr size_t num_blobs = 3;
+  const std::vector<std::string> key_strs = {"key1", "key2", "key3"};
+  const std::vector<std::string> blob_strs = {"blob1", "blob2", "blob3"};

-  uint64_t blob_offset = 0;
-  uint64_t blob_size = 0;
+  const std::vector<Slice> keys = {key_strs[0], key_strs[1], key_strs[2]};
+  const std::vector<Slice> blobs = {blob_strs[0], blob_strs[1], blob_strs[2]};
+
+  std::vector<uint64_t> blob_offsets(keys.size());
+  std::vector<uint64_t> blob_sizes(keys.size());

  WriteBlobFile(immutable_options, column_family_id, has_ttl, expiration_range,
-                expiration_range, blob_file_number, key, blob, kNoCompression,
-                &blob_offset, &blob_size);
+                expiration_range, blob_file_number, keys, blobs, kNoCompression,
+                blob_offsets, blob_sizes);

  constexpr HistogramImpl* blob_file_read_hist = nullptr;

@@ -153,10 +184,36 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    PinnableSlice value;
    uint64_t bytes_read = 0;

-    ASSERT_OK(reader->GetBlob(read_options, key, blob_offset, blob_size,
-                              kNoCompression, &value, &bytes_read));
-    ASSERT_EQ(value, blob);
-    ASSERT_EQ(bytes_read, blob_size);
+    ASSERT_OK(reader->GetBlob(read_options, keys[0], blob_offsets[0],
+                              blob_sizes[0], kNoCompression, &value,
+                              &bytes_read));
+    ASSERT_EQ(value, blobs[0]);
+    ASSERT_EQ(bytes_read, blob_sizes[0]);
+
+    // MultiGetBlob
+    bytes_read = 0;
+    size_t total_size = 0;
+    autovector<std::reference_wrapper<const Slice>> key_refs;
+    for (const auto& key_ref : keys) {
+      key_refs.emplace_back(std::cref(key_ref));
+    }
+    autovector<uint64_t> offsets{blob_offsets[0], blob_offsets[1],
+                                 blob_offsets[2]};
+    autovector<uint64_t> sizes{blob_sizes[0], blob_sizes[1], blob_sizes[2]};
+    std::array<Status, num_blobs> statuses_buf;
+    autovector<Status*> statuses{&statuses_buf[0], &statuses_buf[1],
+                                 &statuses_buf[2]};
+    std::array<PinnableSlice, num_blobs> value_buf;
+    autovector<PinnableSlice*> values{&value_buf[0], &value_buf[1],
+                                      &value_buf[2]};
+    reader->MultiGetBlob(read_options, key_refs, offsets, sizes, statuses,
+                         values, &bytes_read);
+    for (size_t i = 0; i < num_blobs; ++i) {
+      ASSERT_OK(statuses_buf[i]);
+      ASSERT_EQ(value_buf[i], blobs[i]);
+      total_size += blob_sizes[i];
+    }
+    ASSERT_EQ(bytes_read, total_size);
  }

  read_options.verify_checksums = true;
@@ -165,14 +222,15 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    PinnableSlice value;
    uint64_t bytes_read = 0;

-    ASSERT_OK(reader->GetBlob(read_options, key, blob_offset, blob_size,
-                              kNoCompression, &value, &bytes_read));
-    ASSERT_EQ(value, blob);
+    ASSERT_OK(reader->GetBlob(read_options, keys[1], blob_offsets[1],
+                              blob_sizes[1], kNoCompression, &value,
+                              &bytes_read));
+    ASSERT_EQ(value, blobs[1]);

-    constexpr uint64_t key_size = sizeof(key) - 1;
+    const uint64_t key_size = keys[1].size();
    ASSERT_EQ(bytes_read,
              BlobLogRecord::CalculateAdjustmentForRecordHeader(key_size) +
-                  blob_size);
+                  blob_sizes[1]);
  }

  // Invalid offset (too close to start of file)
@@ -181,8 +239,9 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    uint64_t bytes_read = 0;

    ASSERT_TRUE(reader
-                    ->GetBlob(read_options, key, blob_offset - 1, blob_size,
-                              kNoCompression, &value, &bytes_read)
+                    ->GetBlob(read_options, keys[0], blob_offsets[0] - 1,
+                              blob_sizes[0], kNoCompression, &value,
+                              &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
  }
@@ -193,8 +252,9 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    uint64_t bytes_read = 0;

    ASSERT_TRUE(reader
-                    ->GetBlob(read_options, key, blob_offset + 1, blob_size,
-                              kNoCompression, &value, &bytes_read)
+                    ->GetBlob(read_options, keys[2], blob_offsets[2] + 1,
+                              blob_sizes[2], kNoCompression, &value,
+                              &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
  }
@@ -205,8 +265,8 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    uint64_t bytes_read = 0;

    ASSERT_TRUE(reader
-                    ->GetBlob(read_options, key, blob_offset, blob_size, kZSTD,
-                              &value, &bytes_read)
+                    ->GetBlob(read_options, keys[0], blob_offsets[0],
+                              blob_sizes[0], kZSTD, &value, &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
  }
@@ -219,23 +279,82 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {

    ASSERT_TRUE(reader
                    ->GetBlob(read_options, shorter_key,
-                              blob_offset - (sizeof(key) - sizeof(shorter_key)),
-                              blob_size, kNoCompression, &value, &bytes_read)
+                              blob_offsets[0] -
+                                  (keys[0].size() - sizeof(shorter_key) + 1),
+                              blob_sizes[0], kNoCompression, &value,
+                              &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
+
+    // MultiGetBlob
+    autovector<std::reference_wrapper<const Slice>> key_refs;
+    for (const auto& key_ref : keys) {
+      key_refs.emplace_back(std::cref(key_ref));
+    }
+    Slice shorter_key_slice(shorter_key, sizeof(shorter_key) - 1);
+    key_refs[1] = std::cref(shorter_key_slice);
+
+    autovector<uint64_t> offsets{
+        blob_offsets[0],
+        blob_offsets[1] - (keys[1].size() - key_refs[1].get().size()),
+        blob_offsets[2]};
+    autovector<uint64_t> sizes{blob_sizes[0], blob_sizes[1], blob_sizes[2]};
+    std::array<Status, num_blobs> statuses_buf;
+    autovector<Status*> statuses{&statuses_buf[0], &statuses_buf[1],
+                                 &statuses_buf[2]};
+    std::array<PinnableSlice, num_blobs> value_buf;
+    autovector<PinnableSlice*> values{&value_buf[0], &value_buf[1],
+                                      &value_buf[2]};
+    reader->MultiGetBlob(read_options, key_refs, offsets, sizes, statuses,
+                         values, &bytes_read);
+    for (size_t i = 0; i < num_blobs; ++i) {
+      if (i == 1) {
+        ASSERT_TRUE(statuses_buf[i].IsCorruption());
+      } else {
+        ASSERT_OK(statuses_buf[i]);
+      }
+    }
  }

  // Incorrect key
  {
-    constexpr char incorrect_key[] = "foo";
+    constexpr char incorrect_key[] = "foo1";
    PinnableSlice value;
    uint64_t bytes_read = 0;

    ASSERT_TRUE(reader
-                    ->GetBlob(read_options, incorrect_key, blob_offset,
-                              blob_size, kNoCompression, &value, &bytes_read)
+                    ->GetBlob(read_options, incorrect_key, blob_offsets[0],
+                              blob_sizes[0], kNoCompression, &value,
+                              &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
+
+    // MultiGetBlob
+    autovector<std::reference_wrapper<const Slice>> key_refs;
+    for (const auto& key_ref : keys) {
+      key_refs.emplace_back(std::cref(key_ref));
+    }
+    Slice wrong_key_slice(incorrect_key, sizeof(incorrect_key) - 1);
+    key_refs[2] = std::cref(wrong_key_slice);
+
+    autovector<uint64_t> offsets{blob_offsets[0], blob_offsets[1],
+                                 blob_offsets[2]};
+    autovector<uint64_t> sizes{blob_sizes[0], blob_sizes[1], blob_sizes[2]};
+    std::array<Status, num_blobs> statuses_buf;
+    autovector<Status*> statuses{&statuses_buf[0], &statuses_buf[1],
+                                 &statuses_buf[2]};
+    std::array<PinnableSlice, num_blobs> value_buf;
+    autovector<PinnableSlice*> values{&value_buf[0], &value_buf[1],
+                                      &value_buf[2]};
+    reader->MultiGetBlob(read_options, key_refs, offsets, sizes, statuses,
+                         values, &bytes_read);
+    for (size_t i = 0; i < num_blobs; ++i) {
+      if (i == num_blobs - 1) {
+        ASSERT_TRUE(statuses_buf[i].IsCorruption());
+      } else {
+        ASSERT_OK(statuses_buf[i]);
+      }
+    }
  }

  // Incorrect value size
@@ -244,10 +363,35 @@ TEST_F(BlobFileReaderTest, CreateReaderAndGetBlob) {
    uint64_t bytes_read = 0;

    ASSERT_TRUE(reader
-                    ->GetBlob(read_options, key, blob_offset, blob_size + 1,
-                              kNoCompression, &value, &bytes_read)
+                    ->GetBlob(read_options, keys[1], blob_offsets[1],
+                              blob_sizes[1] + 1, kNoCompression, &value,
+                              &bytes_read)
                    .IsCorruption());
    ASSERT_EQ(bytes_read, 0);
+
+    // MultiGetBlob
+    autovector<std::reference_wrapper<const Slice>> key_refs;
+    for (const auto& key_ref : keys) {
+      key_refs.emplace_back(std::cref(key_ref));
+    }
+    autovector<uint64_t> offsets{blob_offsets[0], blob_offsets[1],
+                                 blob_offsets[2]};
+    autovector<uint64_t> sizes{blob_sizes[0], blob_sizes[1] + 1, blob_sizes[2]};
+    std::array<Status, num_blobs> statuses_buf;
+    autovector<Status*> statuses{&statuses_buf[0], &statuses_buf[1],
+                                 &statuses_buf[2]};
+    std::array<PinnableSlice, num_blobs> value_buf;
+    autovector<PinnableSlice*> values{&value_buf[0], &value_buf[1],
+                                      &value_buf[2]};
+    reader->MultiGetBlob(read_options, key_refs, offsets, sizes, statuses,
+                         values, &bytes_read);
+    for (size_t i = 0; i < num_blobs; ++i) {
+      if (i != 1) {
+        ASSERT_OK(statuses_buf[i]);
+      } else {
+        ASSERT_TRUE(statuses_buf[i].IsCorruption());
+      }
+    }
  }
 }

@@ -0,0 +1,100 @@
+//  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).
+
+#include "db/blob/blob_garbage_meter.h"
+
+#include "db/blob/blob_index.h"
+#include "db/blob/blob_log_format.h"
+#include "db/dbformat.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+Status BlobGarbageMeter::ProcessInFlow(const Slice& key, const Slice& value) {
+  uint64_t blob_file_number = kInvalidBlobFileNumber;
+  uint64_t bytes = 0;
+
+  const Status s = Parse(key, value, &blob_file_number, &bytes);
+  if (!s.ok()) {
+    return s;
+  }
+
+  if (blob_file_number == kInvalidBlobFileNumber) {
+    return Status::OK();
+  }
+
+  flows_[blob_file_number].AddInFlow(bytes);
+
+  return Status::OK();
+}
+
+Status BlobGarbageMeter::ProcessOutFlow(const Slice& key, const Slice& value) {
+  uint64_t blob_file_number = kInvalidBlobFileNumber;
+  uint64_t bytes = 0;
+
+  const Status s = Parse(key, value, &blob_file_number, &bytes);
+  if (!s.ok()) {
+    return s;
+  }
+
+  if (blob_file_number == kInvalidBlobFileNumber) {
+    return Status::OK();
+  }
+
+  // Note: in order to measure the amount of additional garbage, we only need to
+  // track the outflow for preexisting files, i.e. those that also had inflow.
+  // (Newly written files would only have outflow.)
+  auto it = flows_.find(blob_file_number);
+  if (it == flows_.end()) {
+    return Status::OK();
+  }
+
+  it->second.AddOutFlow(bytes);
+
+  return Status::OK();
+}
+
+Status BlobGarbageMeter::Parse(const Slice& key, const Slice& value,
+                               uint64_t* blob_file_number, uint64_t* bytes) {
+  assert(blob_file_number);
+  assert(*blob_file_number == kInvalidBlobFileNumber);
+  assert(bytes);
+  assert(*bytes == 0);
+
+  ParsedInternalKey ikey;
+
+  {
+    constexpr bool log_err_key = false;
+    const Status s = ParseInternalKey(key, &ikey, log_err_key);
+    if (!s.ok()) {
+      return s;
+    }
+  }
+
+  if (ikey.type != kTypeBlobIndex) {
+    return Status::OK();
+  }
+
+  BlobIndex blob_index;
+
+  {
+    const Status s = blob_index.DecodeFrom(value);
+    if (!s.ok()) {
+      return s;
+    }
+  }
+
+  if (blob_index.IsInlined() || blob_index.HasTTL()) {
+    return Status::Corruption("Unexpected TTL/inlined blob index");
+  }
+
+  *blob_file_number = blob_index.file_number();
+  *bytes =
+      blob_index.size() +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(ikey.user_key.size());
+
+  return Status::OK();
+}
+
+}  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,102 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#include <cassert>
+#include <cstdint>
+#include <unordered_map>
+
+#include "db/blob/blob_constants.h"
+#include "rocksdb/rocksdb_namespace.h"
+#include "rocksdb/status.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+class Slice;
+
+// A class that can be used to compute the amount of additional garbage
+// generated by a compaction. It parses the keys and blob references in the
+// input and output of a compaction, and aggregates the "inflow" and "outflow"
+// on a per-blob file basis. The amount of additional garbage for any given blob
+// file can then be computed by subtracting the outflow from the inflow.
+class BlobGarbageMeter {
+ public:
+  // A class to store the number and total size of blobs on a per-blob file
+  // basis.
+  class BlobStats {
+   public:
+    void Add(uint64_t bytes) {
+      ++count_;
+      bytes_ += bytes;
+    }
+    void Add(uint64_t count, uint64_t bytes) {
+      count_ += count;
+      bytes_ += bytes;
+    }
+
+    uint64_t GetCount() const { return count_; }
+    uint64_t GetBytes() const { return bytes_; }
+
+   private:
+    uint64_t count_ = 0;
+    uint64_t bytes_ = 0;
+  };
+
+  // A class to keep track of the "inflow" and the "outflow" and to compute the
+  // amount of additional garbage for a given blob file.
+  class BlobInOutFlow {
+   public:
+    void AddInFlow(uint64_t bytes) {
+      in_flow_.Add(bytes);
+      assert(IsValid());
+    }
+    void AddOutFlow(uint64_t bytes) {
+      out_flow_.Add(bytes);
+      assert(IsValid());
+    }
+
+    const BlobStats& GetInFlow() const { return in_flow_; }
+    const BlobStats& GetOutFlow() const { return out_flow_; }
+
+    bool IsValid() const {
+      return in_flow_.GetCount() >= out_flow_.GetCount() &&
+             in_flow_.GetBytes() >= out_flow_.GetBytes();
+    }
+    bool HasGarbage() const {
+      assert(IsValid());
+      return in_flow_.GetCount() > out_flow_.GetCount();
+    }
+    uint64_t GetGarbageCount() const {
+      assert(IsValid());
+      assert(HasGarbage());
+      return in_flow_.GetCount() - out_flow_.GetCount();
+    }
+    uint64_t GetGarbageBytes() const {
+      assert(IsValid());
+      assert(HasGarbage());
+      return in_flow_.GetBytes() - out_flow_.GetBytes();
+    }
+
+   private:
+    BlobStats in_flow_;
+    BlobStats out_flow_;
+  };
+
+  Status ProcessInFlow(const Slice& key, const Slice& value);
+  Status ProcessOutFlow(const Slice& key, const Slice& value);
+
+  const std::unordered_map<uint64_t, BlobInOutFlow>& flows() const {
+    return flows_;
+  }
+
+ private:
+  static Status Parse(const Slice& key, const Slice& value,
+                      uint64_t* blob_file_number, uint64_t* bytes);
+
+  std::unordered_map<uint64_t, BlobInOutFlow> flows_;
+};
+
+}  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,196 @@
+//  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).
+
+#include "db/blob/blob_garbage_meter.h"
+
+#include <string>
+#include <vector>
+
+#include "db/blob/blob_index.h"
+#include "db/blob/blob_log_format.h"
+#include "db/dbformat.h"
+#include "test_util/testharness.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+TEST(BlobGarbageMeterTest, MeasureGarbage) {
+  BlobGarbageMeter blob_garbage_meter;
+
+  struct BlobDescriptor {
+    std::string user_key;
+    uint64_t blob_file_number;
+    uint64_t offset;
+    uint64_t size;
+    CompressionType compression_type;
+    bool has_in_flow;
+    bool has_out_flow;
+
+    uint64_t GetExpectedBytes() const {
+      return size +
+             BlobLogRecord::CalculateAdjustmentForRecordHeader(user_key.size());
+    }
+  };
+
+  // Note: blob file 4 has the same inflow and outflow and hence no additional
+  // garbage. Blob file 5 has less outflow than inflow and thus it does have
+  // additional garbage. Blob file 6 is a newly written file (i.e. no inflow,
+  // only outflow) and is thus not tracked by the meter.
+  std::vector<BlobDescriptor> blobs{
+      {"key", 4, 1234, 555, kLZ4Compression, true, true},
+      {"other_key", 4, 6789, 101010, kLZ4Compression, true, true},
+      {"yet_another_key", 5, 22222, 3456, kLZ4Compression, true, true},
+      {"foo_key", 5, 77777, 8888, kLZ4Compression, true, true},
+      {"bar_key", 5, 999999, 1212, kLZ4Compression, true, false},
+      {"baz_key", 5, 1234567, 890, kLZ4Compression, true, false},
+      {"new_key", 6, 7777, 9999, kNoCompression, false, true}};
+
+  for (const auto& blob : blobs) {
+    constexpr SequenceNumber seq = 123;
+    const InternalKey key(blob.user_key, seq, kTypeBlobIndex);
+    const Slice key_slice = key.Encode();
+
+    std::string value;
+    BlobIndex::EncodeBlob(&value, blob.blob_file_number, blob.offset, blob.size,
+                          blob.compression_type);
+    const Slice value_slice(value);
+
+    if (blob.has_in_flow) {
+      ASSERT_OK(blob_garbage_meter.ProcessInFlow(key_slice, value_slice));
+    }
+    if (blob.has_out_flow) {
+      ASSERT_OK(blob_garbage_meter.ProcessOutFlow(key_slice, value_slice));
+    }
+  }
+
+  const auto& flows = blob_garbage_meter.flows();
+  ASSERT_EQ(flows.size(), 2);
+
+  {
+    const auto it = flows.find(4);
+    ASSERT_NE(it, flows.end());
+
+    const auto& flow = it->second;
+
+    constexpr uint64_t expected_count = 2;
+    const uint64_t expected_bytes =
+        blobs[0].GetExpectedBytes() + blobs[1].GetExpectedBytes();
+
+    const auto& in = flow.GetInFlow();
+    ASSERT_EQ(in.GetCount(), expected_count);
+    ASSERT_EQ(in.GetBytes(), expected_bytes);
+
+    const auto& out = flow.GetOutFlow();
+    ASSERT_EQ(out.GetCount(), expected_count);
+    ASSERT_EQ(out.GetBytes(), expected_bytes);
+
+    ASSERT_TRUE(flow.IsValid());
+    ASSERT_FALSE(flow.HasGarbage());
+  }
+
+  {
+    const auto it = flows.find(5);
+    ASSERT_NE(it, flows.end());
+
+    const auto& flow = it->second;
+
+    const auto& in = flow.GetInFlow();
+
+    constexpr uint64_t expected_in_count = 4;
+    const uint64_t expected_in_bytes =
+        blobs[2].GetExpectedBytes() + blobs[3].GetExpectedBytes() +
+        blobs[4].GetExpectedBytes() + blobs[5].GetExpectedBytes();
+
+    ASSERT_EQ(in.GetCount(), expected_in_count);
+    ASSERT_EQ(in.GetBytes(), expected_in_bytes);
+
+    const auto& out = flow.GetOutFlow();
+
+    constexpr uint64_t expected_out_count = 2;
+    const uint64_t expected_out_bytes =
+        blobs[2].GetExpectedBytes() + blobs[3].GetExpectedBytes();
+
+    ASSERT_EQ(out.GetCount(), expected_out_count);
+    ASSERT_EQ(out.GetBytes(), expected_out_bytes);
+
+    ASSERT_TRUE(flow.IsValid());
+    ASSERT_TRUE(flow.HasGarbage());
+    ASSERT_EQ(flow.GetGarbageCount(), expected_in_count - expected_out_count);
+    ASSERT_EQ(flow.GetGarbageBytes(), expected_in_bytes - expected_out_bytes);
+  }
+}
+
+TEST(BlobGarbageMeterTest, PlainValue) {
+  constexpr char user_key[] = "user_key";
+  constexpr SequenceNumber seq = 123;
+
+  const InternalKey key(user_key, seq, kTypeValue);
+  const Slice key_slice = key.Encode();
+
+  constexpr char value[] = "value";
+  const Slice value_slice(value);
+
+  BlobGarbageMeter blob_garbage_meter;
+
+  ASSERT_OK(blob_garbage_meter.ProcessInFlow(key_slice, value_slice));
+  ASSERT_OK(blob_garbage_meter.ProcessOutFlow(key_slice, value_slice));
+  ASSERT_TRUE(blob_garbage_meter.flows().empty());
+}
+
+TEST(BlobGarbageMeterTest, CorruptInternalKey) {
+  constexpr char corrupt_key[] = "i_am_corrupt";
+  const Slice key_slice(corrupt_key);
+
+  constexpr char value[] = "value";
+  const Slice value_slice(value);
+
+  BlobGarbageMeter blob_garbage_meter;
+
+  ASSERT_NOK(blob_garbage_meter.ProcessInFlow(key_slice, value_slice));
+  ASSERT_NOK(blob_garbage_meter.ProcessOutFlow(key_slice, value_slice));
+}
+
+TEST(BlobGarbageMeterTest, CorruptBlobIndex) {
+  constexpr char user_key[] = "user_key";
+  constexpr SequenceNumber seq = 123;
+
+  const InternalKey key(user_key, seq, kTypeBlobIndex);
+  const Slice key_slice = key.Encode();
+
+  constexpr char value[] = "i_am_not_a_blob_index";
+  const Slice value_slice(value);
+
+  BlobGarbageMeter blob_garbage_meter;
+
+  ASSERT_NOK(blob_garbage_meter.ProcessInFlow(key_slice, value_slice));
+  ASSERT_NOK(blob_garbage_meter.ProcessOutFlow(key_slice, value_slice));
+}
+
+TEST(BlobGarbageMeterTest, InlinedTTLBlobIndex) {
+  constexpr char user_key[] = "user_key";
+  constexpr SequenceNumber seq = 123;
+
+  const InternalKey key(user_key, seq, kTypeBlobIndex);
+  const Slice key_slice = key.Encode();
+
+  constexpr uint64_t expiration = 1234567890;
+  constexpr char inlined_value[] = "inlined";
+
+  std::string value;
+  BlobIndex::EncodeInlinedTTL(&value, expiration, inlined_value);
+
+  const Slice value_slice(value);
+
+  BlobGarbageMeter blob_garbage_meter;
+
+  ASSERT_NOK(blob_garbage_meter.ProcessInFlow(key_slice, value_slice));
+  ASSERT_NOK(blob_garbage_meter.ProcessOutFlow(key_slice, value_slice));
+}
+
+}  // namespace ROCKSDB_NAMESPACE
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
@@ -52,6 +52,9 @@ class BlobIndex {

  BlobIndex() : type_(Type::kUnknown) {}

+  BlobIndex(const BlobIndex&) = default;
+  BlobIndex& operator=(const BlobIndex&) = default;
+
  bool IsInlined() const { return type_ == Type::kInlinedTTL; }

  bool HasTTL() const {
@@ -107,7 +107,8 @@ struct BlobLogRecord {
  // Note that the offset field of BlobIndex actually points to the blob value
  // as opposed to the start of the blob record. The following method can
  // be used to calculate the adjustment needed to read the blob record header.
-  static uint64_t CalculateAdjustmentForRecordHeader(uint64_t key_size) {
+  static constexpr uint64_t CalculateAdjustmentForRecordHeader(
+      uint64_t key_size) {
    return key_size + kHeaderSize;
  }

@@ -10,6 +10,8 @@
 #include "db/blob/blob_log_format.h"
 #include "rocksdb/slice.h"

+#define MAX_HEADER_SIZE(a, b, c) (a > b ? (a > c ? a : c) : (b > c ? b : c))
+
 namespace ROCKSDB_NAMESPACE {

 class RandomAccessFileReader;
@@ -69,10 +71,13 @@ class BlobLogSequentialReader {
  Statistics* statistics_;

  Slice buffer_;
-  char header_buf_[BlobLogRecord::kHeaderSize];
+  char header_buf_[MAX_HEADER_SIZE(BlobLogHeader::kSize, BlobLogFooter::kSize,
+                                   BlobLogRecord::kHeaderSize)];

  // which byte to read next
  uint64_t next_byte_;
 };

 }  // namespace ROCKSDB_NAMESPACE
+
+#undef MAX_HEADER_SIZE
@@ -4,8 +4,10 @@
 //  (found in the LICENSE.Apache file in the root directory).

 #include <array>
+#include <sstream>

 #include "db/blob/blob_index.h"
+#include "db/blob/blob_log_format.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "test_util/sync_point.h"
@@ -16,7 +18,7 @@ namespace ROCKSDB_NAMESPACE {
 class DBBlobBasicTest : public DBTestBase {
 protected:
  DBBlobBasicTest()
-      : DBTestBase("/db_blob_basic_test", /* env_do_fsync */ false) {}
+      : DBTestBase("db_blob_basic_test", /* env_do_fsync */ false) {}
 };

 TEST_F(DBBlobBasicTest, GetBlob) {
@@ -125,6 +127,237 @@ TEST_F(DBBlobBasicTest, MultiGetBlobs) {
  }
 }

+#ifndef ROCKSDB_LITE
+TEST_F(DBBlobBasicTest, MultiGetWithDirectIO) {
+  Options options = GetDefaultOptions();
+
+  // First, create an external SST file ["b"].
+  const std::string file_path = dbname_ + "/test.sst";
+  {
+    SstFileWriter sst_file_writer(EnvOptions(), GetDefaultOptions());
+    Status s = sst_file_writer.Open(file_path);
+    ASSERT_OK(s);
+    ASSERT_OK(sst_file_writer.Put("b", "b_value"));
+    ASSERT_OK(sst_file_writer.Finish());
+  }
+
+  options.enable_blob_files = true;
+  options.min_blob_size = 1000;
+  options.use_direct_reads = true;
+  options.allow_ingest_behind = true;
+
+  // Open DB with fixed-prefix sst-partitioner so that compaction will cut
+  // new table file when encountering a new key whose 1-byte prefix changes.
+  constexpr size_t key_len = 1;
+  options.sst_partitioner_factory =
+      NewSstPartitionerFixedPrefixFactory(key_len);
+
+  Status s = TryReopen(options);
+  if (s.IsInvalidArgument()) {
+    ROCKSDB_GTEST_SKIP("This test requires direct IO support");
+    return;
+  }
+  ASSERT_OK(s);
+
+  constexpr size_t num_keys = 3;
+  constexpr size_t blob_size = 3000;
+
+  constexpr char first_key[] = "a";
+  const std::string first_blob(blob_size, 'a');
+  ASSERT_OK(Put(first_key, first_blob));
+
+  constexpr char second_key[] = "b";
+  const std::string second_blob(2 * blob_size, 'b');
+  ASSERT_OK(Put(second_key, second_blob));
+
+  constexpr char third_key[] = "d";
+  const std::string third_blob(blob_size, 'd');
+  ASSERT_OK(Put(third_key, third_blob));
+
+  // first_blob, second_blob and third_blob in the same blob file.
+  //      SST                    Blob file
+  // L0  ["a",    "b",    "d"]   |'aaaa', 'bbbb', 'dddd'|
+  //       |       |       |         ^       ^        ^
+  //       |       |       |         |       |        |
+  //       |       |       +---------|-------|--------+
+  //       |       +-----------------|-------+
+  //       +-------------------------+
+  ASSERT_OK(Flush());
+
+  constexpr char fourth_key[] = "c";
+  const std::string fourth_blob(blob_size, 'c');
+  ASSERT_OK(Put(fourth_key, fourth_blob));
+  // fourth_blob in another blob file.
+  //      SST                    Blob file                 SST     Blob file
+  // L0  ["a",    "b",    "d"]   |'aaaa', 'bbbb', 'dddd'|  ["c"]   |'cccc'|
+  //       |       |       |         ^       ^        ^      |       ^
+  //       |       |       |         |       |        |      |       |
+  //       |       |       +---------|-------|--------+      +-------+
+  //       |       +-----------------|-------+
+  //       +-------------------------+
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                              /*end=*/nullptr));
+
+  // Due to the above sst partitioner, we get 4 L1 files. The blob files are
+  // unchanged.
+  //                             |'aaaa', 'bbbb', 'dddd'|  |'cccc'|
+  //                                 ^       ^     ^         ^
+  //                                 |       |     |         |
+  // L0                              |       |     |         |
+  // L1  ["a"]   ["b"]   ["c"]       |       |   ["d"]       |
+  //       |       |       |         |       |               |
+  //       |       |       +---------|-------|---------------+
+  //       |       +-----------------|-------+
+  //       +-------------------------+
+  ASSERT_EQ(4, NumTableFilesAtLevel(/*level=*/1));
+
+  {
+    // Ingest the external SST file into bottommost level.
+    std::vector<std::string> ext_files{file_path};
+    IngestExternalFileOptions opts;
+    opts.ingest_behind = true;
+    ASSERT_OK(
+        db_->IngestExternalFile(db_->DefaultColumnFamily(), ext_files, opts));
+  }
+
+  // Now the database becomes as follows.
+  //                             |'aaaa', 'bbbb', 'dddd'|  |'cccc'|
+  //                                 ^       ^     ^         ^
+  //                                 |       |     |         |
+  // L0                              |       |     |         |
+  // L1  ["a"]   ["b"]   ["c"]       |       |   ["d"]       |
+  //       |       |       |         |       |               |
+  //       |       |       +---------|-------|---------------+
+  //       |       +-----------------|-------+
+  //       +-------------------------+
+  //
+  // L6          ["b"]
+
+  {
+    // Compact ["b"] to bottommost level.
+    Slice begin = Slice(second_key);
+    Slice end = Slice(second_key);
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    ASSERT_OK(db_->CompactRange(cro, &begin, &end));
+  }
+
+  //                             |'aaaa', 'bbbb', 'dddd'|  |'cccc'|
+  //                                 ^       ^     ^         ^
+  //                                 |       |     |         |
+  // L0                              |       |     |         |
+  // L1  ["a"]           ["c"]       |       |   ["d"]       |
+  //       |               |         |       |               |
+  //       |               +---------|-------|---------------+
+  //       |       +-----------------|-------+
+  //       +-------|-----------------+
+  //               |
+  // L6          ["b"]
+  ASSERT_EQ(3, NumTableFilesAtLevel(/*level=*/1));
+  ASSERT_EQ(1, NumTableFilesAtLevel(/*level=*/6));
+
+  bool called = false;
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "RandomAccessFileReader::MultiRead:AlignedReqs", [&](void* arg) {
+        auto* aligned_reqs = static_cast<std::vector<FSReadRequest>*>(arg);
+        assert(aligned_reqs);
+        ASSERT_EQ(1, aligned_reqs->size());
+        called = true;
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  std::array<Slice, num_keys> keys{{first_key, third_key, second_key}};
+
+  {
+    std::array<PinnableSlice, num_keys> values;
+    std::array<Status, num_keys> statuses;
+
+    // The MultiGet(), when constructing the KeyContexts, will process the keys
+    // in such order: a, d, b. The reason is that ["a"] and ["d"] are in L1,
+    // while ["b"] resides in L6.
+    // Consequently, the original FSReadRequest list prepared by
+    // Version::MultiGetblob() will be for "a", "d" and "b". It is unsorted as
+    // follows:
+    //
+    // ["a", offset=30, len=3033],
+    // ["d", offset=9096, len=3033],
+    // ["b", offset=3063, len=6033]
+    //
+    // If we do not sort them before calling MultiRead() in DirectIO, then the
+    // underlying IO merging logic will yield two requests.
+    //
+    // [offset=0, len=4096] (for "a")
+    // [offset=0, len=12288] (result of merging the request for "d" and "b")
+    //
+    // We need to sort them in Version::MultiGetBlob() so that the underlying
+    // IO merging logic in DirectIO mode works as expected. The correct
+    // behavior will be one aligned request:
+    //
+    // [offset=0, len=12288]
+
+    db_->MultiGet(ReadOptions(), db_->DefaultColumnFamily(), num_keys, &keys[0],
+                  &values[0], &statuses[0]);
+
+    SyncPoint::GetInstance()->DisableProcessing();
+    SyncPoint::GetInstance()->ClearAllCallBacks();
+
+    ASSERT_TRUE(called);
+
+    ASSERT_OK(statuses[0]);
+    ASSERT_EQ(values[0], first_blob);
+
+    ASSERT_OK(statuses[1]);
+    ASSERT_EQ(values[1], third_blob);
+
+    ASSERT_OK(statuses[2]);
+    ASSERT_EQ(values[2], second_blob);
+  }
+}
+#endif  // !ROCKSDB_LITE
+
+TEST_F(DBBlobBasicTest, MultiGetBlobsFromMultipleFiles) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+
+  Reopen(options);
+
+  constexpr size_t kNumBlobFiles = 3;
+  constexpr size_t kNumBlobsPerFile = 3;
+  constexpr size_t kNumKeys = kNumBlobsPerFile * kNumBlobFiles;
+
+  std::vector<std::string> key_strs;
+  std::vector<std::string> value_strs;
+  for (size_t i = 0; i < kNumBlobFiles; ++i) {
+    for (size_t j = 0; j < kNumBlobsPerFile; ++j) {
+      std::string key = "key" + std::to_string(i) + "_" + std::to_string(j);
+      std::string value =
+          "value_as_blob" + std::to_string(i) + "_" + std::to_string(j);
+      ASSERT_OK(Put(key, value));
+      key_strs.push_back(key);
+      value_strs.push_back(value);
+    }
+    ASSERT_OK(Flush());
+  }
+  assert(key_strs.size() == kNumKeys);
+  std::array<Slice, kNumKeys> keys;
+  for (size_t i = 0; i < keys.size(); ++i) {
+    keys[i] = key_strs[i];
+  }
+  std::array<PinnableSlice, kNumKeys> values;
+  std::array<Status, kNumKeys> statuses;
+  db_->MultiGet(ReadOptions(), db_->DefaultColumnFamily(), kNumKeys, &keys[0],
+                &values[0], &statuses[0]);
+
+  for (size_t i = 0; i < kNumKeys; ++i) {
+    ASSERT_OK(statuses[i]);
+    ASSERT_EQ(value_strs[i], values[i]);
+  }
+}
+
 TEST_F(DBBlobBasicTest, GetBlob_CorruptIndex) {
  Options options = GetDefaultOptions();
  options.enable_blob_files = true;
@@ -148,6 +381,83 @@ TEST_F(DBBlobBasicTest, GetBlob_CorruptIndex) {
                  .IsCorruption());
 }

+TEST_F(DBBlobBasicTest, MultiGetBlob_CorruptIndex) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+  options.create_if_missing = true;
+
+  DestroyAndReopen(options);
+
+  constexpr size_t kNumOfKeys = 3;
+  std::array<std::string, kNumOfKeys> key_strs;
+  std::array<std::string, kNumOfKeys> value_strs;
+  std::array<Slice, kNumOfKeys + 1> keys;
+  for (size_t i = 0; i < kNumOfKeys; ++i) {
+    key_strs[i] = "foo" + std::to_string(i);
+    value_strs[i] = "blob_value" + std::to_string(i);
+    ASSERT_OK(Put(key_strs[i], value_strs[i]));
+    keys[i] = key_strs[i];
+  }
+
+  constexpr char key[] = "key";
+  {
+    // Fake a corrupt blob index.
+    const std::string blob_index("foobar");
+    WriteBatch batch;
+    ASSERT_OK(WriteBatchInternal::PutBlobIndex(&batch, 0, key, blob_index));
+    ASSERT_OK(db_->Write(WriteOptions(), &batch));
+    keys[kNumOfKeys] = Slice(static_cast<const char*>(key), sizeof(key) - 1);
+  }
+
+  ASSERT_OK(Flush());
+
+  std::array<PinnableSlice, kNumOfKeys + 1> values;
+  std::array<Status, kNumOfKeys + 1> statuses;
+  db_->MultiGet(ReadOptions(), dbfull()->DefaultColumnFamily(), kNumOfKeys + 1,
+                keys.data(), values.data(), statuses.data(),
+                /*sorted_input=*/false);
+  for (size_t i = 0; i < kNumOfKeys + 1; ++i) {
+    if (i != kNumOfKeys) {
+      ASSERT_OK(statuses[i]);
+      ASSERT_EQ("blob_value" + std::to_string(i), values[i]);
+    } else {
+      ASSERT_TRUE(statuses[i].IsCorruption());
+    }
+  }
+}
+
+TEST_F(DBBlobBasicTest, MultiGetBlob_ExceedSoftLimit) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+
+  Reopen(options);
+
+  constexpr size_t kNumOfKeys = 3;
+  std::array<std::string, kNumOfKeys> key_bufs;
+  std::array<std::string, kNumOfKeys> value_bufs;
+  std::array<Slice, kNumOfKeys> keys;
+  for (size_t i = 0; i < kNumOfKeys; ++i) {
+    key_bufs[i] = "foo" + std::to_string(i);
+    value_bufs[i] = "blob_value" + std::to_string(i);
+    ASSERT_OK(Put(key_bufs[i], value_bufs[i]));
+    keys[i] = key_bufs[i];
+  }
+  ASSERT_OK(Flush());
+
+  std::array<PinnableSlice, kNumOfKeys> values;
+  std::array<Status, kNumOfKeys> statuses;
+  ReadOptions read_opts;
+  read_opts.value_size_soft_limit = 1;
+  db_->MultiGet(read_opts, dbfull()->DefaultColumnFamily(), kNumOfKeys,
+                keys.data(), values.data(), statuses.data(),
+                /*sorted_input=*/true);
+  for (const auto& s : statuses) {
+    ASSERT_TRUE(s.IsAborted());
+  }
+}
+
 TEST_F(DBBlobBasicTest, GetBlob_InlinedTTLIndex) {
  constexpr uint64_t min_blob_size = 10;

@@ -367,6 +677,147 @@ TEST_F(DBBlobBasicTest, MultiGetMergeBlobWithPut) {
  ASSERT_EQ(values[2], "v2_0");
 }

+#ifndef ROCKSDB_LITE
+TEST_F(DBBlobBasicTest, Properties) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+
+  Reopen(options);
+
+  constexpr char key1[] = "key1";
+  constexpr size_t key1_size = sizeof(key1) - 1;
+
+  constexpr char key2[] = "key2";
+  constexpr size_t key2_size = sizeof(key2) - 1;
+
+  constexpr char key3[] = "key3";
+  constexpr size_t key3_size = sizeof(key3) - 1;
+
+  constexpr char blob[] = "0000000000";
+  constexpr size_t blob_size = sizeof(blob) - 1;
+
+  ASSERT_OK(Put(key1, blob));
+  ASSERT_OK(Put(key2, blob));
+  ASSERT_OK(Flush());
+
+  constexpr size_t first_blob_file_expected_size =
+      BlobLogHeader::kSize +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key1_size) + blob_size +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) + blob_size +
+      BlobLogFooter::kSize;
+
+  ASSERT_OK(Put(key3, blob));
+  ASSERT_OK(Flush());
+
+  constexpr size_t second_blob_file_expected_size =
+      BlobLogHeader::kSize +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key3_size) + blob_size +
+      BlobLogFooter::kSize;
+
+  constexpr size_t total_expected_size =
+      first_blob_file_expected_size + second_blob_file_expected_size;
+
+  // Number of blob files
+  uint64_t num_blob_files = 0;
+  ASSERT_TRUE(
+      db_->GetIntProperty(DB::Properties::kNumBlobFiles, &num_blob_files));
+  ASSERT_EQ(num_blob_files, 2);
+
+  // Total size of live blob files
+  uint64_t live_blob_file_size = 0;
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kLiveBlobFileSize,
+                                  &live_blob_file_size));
+  ASSERT_EQ(live_blob_file_size, total_expected_size);
+
+  // Total size of all blob files across all versions
+  // Note: this should be the same as above since we only have one
+  // version at this point.
+  uint64_t total_blob_file_size = 0;
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kTotalBlobFileSize,
+                                  &total_blob_file_size));
+  ASSERT_EQ(total_blob_file_size, total_expected_size);
+
+  // Delete key2 to create some garbage
+  ASSERT_OK(Delete(key2));
+  ASSERT_OK(Flush());
+
+  constexpr Slice* begin = nullptr;
+  constexpr Slice* end = nullptr;
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), begin, end));
+
+  constexpr size_t expected_garbage_size =
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) + blob_size;
+
+  // Blob file stats
+  std::string blob_stats;
+  ASSERT_TRUE(db_->GetProperty(DB::Properties::kBlobStats, &blob_stats));
+
+  std::ostringstream oss;
+  oss << "Number of blob files: 2\nTotal size of blob files: "
+      << total_expected_size
+      << "\nTotal size of garbage in blob files: " << expected_garbage_size
+      << '\n';
+
+  ASSERT_EQ(blob_stats, oss.str());
+}
+
+TEST_F(DBBlobBasicTest, PropertiesMultiVersion) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+
+  Reopen(options);
+
+  constexpr char key1[] = "key1";
+  constexpr char key2[] = "key2";
+  constexpr char key3[] = "key3";
+
+  constexpr size_t key_size = sizeof(key1) - 1;
+  static_assert(sizeof(key2) - 1 == key_size, "unexpected size: key2");
+  static_assert(sizeof(key3) - 1 == key_size, "unexpected size: key3");
+
+  constexpr char blob[] = "0000000000";
+  constexpr size_t blob_size = sizeof(blob) - 1;
+
+  ASSERT_OK(Put(key1, blob));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(key2, blob));
+  ASSERT_OK(Flush());
+
+  // Create an iterator to keep the current version alive
+  std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+  ASSERT_OK(iter->status());
+
+  // Note: the Delete and subsequent compaction results in the first blob file
+  // not making it to the final version. (It is still part of the previous
+  // version kept alive by the iterator though.) On the other hand, the Put
+  // results in a third blob file.
+  ASSERT_OK(Delete(key1));
+  ASSERT_OK(Put(key3, blob));
+  ASSERT_OK(Flush());
+
+  constexpr Slice* begin = nullptr;
+  constexpr Slice* end = nullptr;
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), begin, end));
+
+  // Total size of all blob files across all versions: between the two versions,
+  // we should have three blob files of the same size with one blob each.
+  // The version kept alive by the iterator contains the first and the second
+  // blob file, while the final version contains the second and the third blob
+  // file. (The second blob file is thus shared by the two versions but should
+  // be counted only once.)
+  uint64_t total_blob_file_size = 0;
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kTotalBlobFileSize,
+                                  &total_blob_file_size));
+  ASSERT_EQ(total_blob_file_size,
+            3 * (BlobLogHeader::kSize +
+                 BlobLogRecord::CalculateAdjustmentForRecordHeader(key_size) +
+                 blob_size + BlobLogFooter::kSize));
+}
+#endif  // !ROCKSDB_LITE
+
 class DBBlobBasicIOErrorTest : public DBBlobBasicTest,
                               public testing::WithParamInterface<std::string> {
 protected:
@@ -379,11 +830,21 @@ class DBBlobBasicIOErrorTest : public DBBlobBasicTest,
  std::string sync_point_;
 };

+class DBBlobBasicIOErrorMultiGetTest : public DBBlobBasicIOErrorTest {
+ public:
+  DBBlobBasicIOErrorMultiGetTest() : DBBlobBasicIOErrorTest() {}
+};
+
 INSTANTIATE_TEST_CASE_P(DBBlobBasicTest, DBBlobBasicIOErrorTest,
                        ::testing::ValuesIn(std::vector<std::string>{
                            "BlobFileReader::OpenFile:NewRandomAccessFile",
                            "BlobFileReader::GetBlob:ReadFromFile"}));

+INSTANTIATE_TEST_CASE_P(DBBlobBasicTest, DBBlobBasicIOErrorMultiGetTest,
+                        ::testing::ValuesIn(std::vector<std::string>{
+                            "BlobFileReader::OpenFile:NewRandomAccessFile",
+                            "BlobFileReader::MultiGetBlob:ReadFromFile"}));
+
 TEST_P(DBBlobBasicIOErrorTest, GetBlob_IOError) {
  Options options;
  options.env = fault_injection_env_.get();
@@ -413,7 +874,7 @@ TEST_P(DBBlobBasicIOErrorTest, GetBlob_IOError) {
  SyncPoint::GetInstance()->ClearAllCallBacks();
 }

-TEST_P(DBBlobBasicIOErrorTest, MultiGetBlobs_IOError) {
+TEST_P(DBBlobBasicIOErrorMultiGetTest, MultiGetBlobs_IOError) {
  Options options = GetDefaultOptions();
  options.env = fault_injection_env_.get();
  options.enable_blob_files = true;
@@ -455,6 +916,53 @@ TEST_P(DBBlobBasicIOErrorTest, MultiGetBlobs_IOError) {
  ASSERT_TRUE(statuses[1].IsIOError());
 }

+TEST_P(DBBlobBasicIOErrorMultiGetTest, MultipleBlobFiles) {
+  Options options = GetDefaultOptions();
+  options.env = fault_injection_env_.get();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+
+  Reopen(options);
+
+  constexpr size_t num_keys = 2;
+
+  constexpr char key1[] = "key1";
+  constexpr char value1[] = "blob1";
+
+  ASSERT_OK(Put(key1, value1));
+  ASSERT_OK(Flush());
+
+  constexpr char key2[] = "key2";
+  constexpr char value2[] = "blob2";
+
+  ASSERT_OK(Put(key2, value2));
+  ASSERT_OK(Flush());
+
+  std::array<Slice, num_keys> keys{{key1, key2}};
+  std::array<PinnableSlice, num_keys> values;
+  std::array<Status, num_keys> statuses;
+
+  bool first_blob_file = true;
+  SyncPoint::GetInstance()->SetCallBack(
+      sync_point_, [&first_blob_file, this](void* /* arg */) {
+        if (first_blob_file) {
+          first_blob_file = false;
+          return;
+        }
+        fault_injection_env_->SetFilesystemActive(false,
+                                                  Status::IOError(sync_point_));
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  db_->MultiGet(ReadOptions(), db_->DefaultColumnFamily(), num_keys,
+                keys.data(), values.data(), statuses.data());
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  ASSERT_OK(statuses[0]);
+  ASSERT_EQ(value1, values[0]);
+  ASSERT_TRUE(statuses[1].IsIOError());
+}
+
 namespace {

 class ReadBlobCompactionFilter : public CompactionFilter {
@@ -510,8 +1018,17 @@ TEST_P(DBBlobBasicIOErrorTest, CompactionFilterReadBlob_IOError) {

 }  // namespace ROCKSDB_NAMESPACE

+#ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+extern "C" {
+void RegisterCustomObjects(int argc, char** argv);
+}
+#else
+void RegisterCustomObjects(int /*argc*/, char** /*argv*/) {}
+#endif  // !ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+
 int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
+  RegisterCustomObjects(argc, argv);
  return RUN_ALL_TESTS();
 }
@@ -4,17 +4,17 @@
 //  (found in the LICENSE.Apache file in the root directory).

 #include "db/blob/blob_index.h"
+#include "db/blob/blob_log_format.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "test_util/sync_point.h"
-#include "utilities/fault_injection_env.h"

 namespace ROCKSDB_NAMESPACE {

 class DBBlobCompactionTest : public DBTestBase {
 public:
  explicit DBBlobCompactionTest()
-      : DBTestBase("/db_blob_compaction_test", /*env_do_fsync=*/false) {}
+      : DBTestBase("db_blob_compaction_test", /*env_do_fsync=*/false) {}

 #ifndef ROCKSDB_LITE
  const std::vector<InternalStats::CompactionStats>& GetCompactionStats() {
@@ -152,6 +152,31 @@ class AlwaysKeepFilter : public CompactionFilter {
    return CompactionFilter::Decision::kKeep;
  }
 };
+
+class SkipUntilFilter : public CompactionFilter {
+ public:
+  explicit SkipUntilFilter(std::string skip_until)
+      : skip_until_(std::move(skip_until)) {}
+
+  const char* Name() const override {
+    return "rocksdb.compaction.filter.skip.until";
+  }
+
+  CompactionFilter::Decision FilterV2(int /* level */, const Slice& /* key */,
+                                      ValueType /* value_type */,
+                                      const Slice& /* existing_value */,
+                                      std::string* /* new_value */,
+                                      std::string* skip_until) const override {
+    assert(skip_until);
+    *skip_until = skip_until_;
+
+    return CompactionFilter::Decision::kRemoveAndSkipUntil;
+  }
+
+ private:
+  std::string skip_until_;
+};
+
 }  // anonymous namespace

 class DBBlobBadCompactionFilterTest
@@ -254,6 +279,49 @@ TEST_F(DBBlobCompactionTest, BlindWriteFilter) {
  Close();
 }

+TEST_F(DBBlobCompactionTest, SkipUntilFilter) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+
+  std::unique_ptr<CompactionFilter> compaction_filter_guard(
+      new SkipUntilFilter("z"));
+  options.compaction_filter = compaction_filter_guard.get();
+
+  Reopen(options);
+
+  const std::vector<std::string> keys{"a", "b", "c"};
+  const std::vector<std::string> values{"a_value", "b_value", "c_value"};
+  assert(keys.size() == values.size());
+
+  for (size_t i = 0; i < keys.size(); ++i) {
+    ASSERT_OK(Put(keys[i], values[i]));
+  }
+
+  ASSERT_OK(Flush());
+
+  int process_in_flow_called = 0;
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "BlobCountingIterator::UpdateAndCountBlobIfNeeded:ProcessInFlow",
+      [&process_in_flow_called](void* /* arg */) { ++process_in_flow_called; });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /* begin */ nullptr,
+                              /* end */ nullptr));
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  for (const auto& key : keys) {
+    ASSERT_EQ(Get(key), "NOT_FOUND");
+  }
+
+  // Make sure SkipUntil was performed using iteration rather than Seek
+  ASSERT_EQ(process_in_flow_called, keys.size());
+
+  Close();
+}
+
 TEST_P(DBBlobBadCompactionFilterTest, BadDecisionFromCompactionFilter) {
  Options options = GetDefaultOptions();
  options.enable_blob_files = true;
@@ -390,10 +458,151 @@ TEST_F(DBBlobCompactionTest, CompactionFilterReadBlobAndKeep) {
  Close();
 }

+TEST_F(DBBlobCompactionTest, TrackGarbage) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+
+  Reopen(options);
+
+  // First table+blob file pair: 4 blobs with different keys
+  constexpr char first_key[] = "first_key";
+  constexpr char first_value[] = "first_value";
+  constexpr char second_key[] = "second_key";
+  constexpr char second_value[] = "second_value";
+  constexpr char third_key[] = "third_key";
+  constexpr char third_value[] = "third_value";
+  constexpr char fourth_key[] = "fourth_key";
+  constexpr char fourth_value[] = "fourth_value";
+
+  ASSERT_OK(Put(first_key, first_value));
+  ASSERT_OK(Put(second_key, second_value));
+  ASSERT_OK(Put(third_key, third_value));
+  ASSERT_OK(Put(fourth_key, fourth_value));
+  ASSERT_OK(Flush());
+
+  // Second table+blob file pair: overwrite 2 existing keys
+  constexpr char new_first_value[] = "new_first_value";
+  constexpr char new_second_value[] = "new_second_value";
+
+  ASSERT_OK(Put(first_key, new_first_value));
+  ASSERT_OK(Put(second_key, new_second_value));
+  ASSERT_OK(Flush());
+
+  // Compact them together. The first blob file should have 2 garbage blobs
+  // corresponding to the 2 overwritten keys.
+  constexpr Slice* begin = nullptr;
+  constexpr Slice* end = nullptr;
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), begin, end));
+
+  VersionSet* const versions = dbfull()->TEST_GetVersionSet();
+  assert(versions);
+  assert(versions->GetColumnFamilySet());
+
+  ColumnFamilyData* const cfd = versions->GetColumnFamilySet()->GetDefault();
+  assert(cfd);
+
+  Version* const current = cfd->current();
+  assert(current);
+
+  const VersionStorageInfo* const storage_info = current->storage_info();
+  assert(storage_info);
+
+  const auto& blob_files = storage_info->GetBlobFiles();
+  ASSERT_EQ(blob_files.size(), 2);
+
+  {
+    auto it = blob_files.begin();
+    const auto& meta = it->second;
+    assert(meta);
+
+    constexpr uint64_t first_expected_bytes =
+        sizeof(first_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(first_key) -
+                                                          1);
+    constexpr uint64_t second_expected_bytes =
+        sizeof(second_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(second_key) -
+                                                          1);
+    constexpr uint64_t third_expected_bytes =
+        sizeof(third_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(third_key) -
+                                                          1);
+    constexpr uint64_t fourth_expected_bytes =
+        sizeof(fourth_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(fourth_key) -
+                                                          1);
+
+    ASSERT_EQ(meta->GetTotalBlobCount(), 4);
+    ASSERT_EQ(meta->GetTotalBlobBytes(),
+              first_expected_bytes + second_expected_bytes +
+                  third_expected_bytes + fourth_expected_bytes);
+    ASSERT_EQ(meta->GetGarbageBlobCount(), 2);
+    ASSERT_EQ(meta->GetGarbageBlobBytes(),
+              first_expected_bytes + second_expected_bytes);
+  }
+
+  {
+    auto it = blob_files.rbegin();
+    const auto& meta = it->second;
+    assert(meta);
+
+    constexpr uint64_t new_first_expected_bytes =
+        sizeof(new_first_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(first_key) -
+                                                          1);
+    constexpr uint64_t new_second_expected_bytes =
+        sizeof(new_second_value) - 1 +
+        BlobLogRecord::CalculateAdjustmentForRecordHeader(sizeof(second_key) -
+                                                          1);
+
+    ASSERT_EQ(meta->GetTotalBlobCount(), 2);
+    ASSERT_EQ(meta->GetTotalBlobBytes(),
+              new_first_expected_bytes + new_second_expected_bytes);
+    ASSERT_EQ(meta->GetGarbageBlobCount(), 0);
+    ASSERT_EQ(meta->GetGarbageBlobBytes(), 0);
+  }
+}
+
+TEST_F(DBBlobCompactionTest, MergeBlobWithBase) {
+  Options options = GetDefaultOptions();
+  options.enable_blob_files = true;
+  options.min_blob_size = 0;
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  options.disable_auto_compactions = true;
+
+  Reopen(options);
+  ASSERT_OK(Put("Key1", "v1_1"));
+  ASSERT_OK(Put("Key2", "v2_1"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Merge("Key1", "v1_2"));
+  ASSERT_OK(Merge("Key2", "v2_2"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Merge("Key1", "v1_3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                              /*end=*/nullptr));
+  ASSERT_EQ(Get("Key1"), "v1_1,v1_2,v1_3");
+  ASSERT_EQ(Get("Key2"), "v2_1,v2_2");
+  Close();
+}
+
 }  // namespace ROCKSDB_NAMESPACE

+#ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+extern "C" {
+void RegisterCustomObjects(int argc, char** argv);
+}
+#else
+void RegisterCustomObjects(int /*argc*/, char** /*argv*/) {}
+#endif  // !ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+
 int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
+  RegisterCustomObjects(argc, argv);
  return RUN_ALL_TESTS();
 }
@@ -12,7 +12,7 @@ namespace ROCKSDB_NAMESPACE {
 class DBBlobCorruptionTest : public DBTestBase {
 protected:
  DBBlobCorruptionTest()
-      : DBTestBase("/db_blob_corruption_test", /* env_do_fsync */ false) {}
+      : DBTestBase("db_blob_corruption_test", /* env_do_fsync */ false) {}

  void Corrupt(FileType filetype, int offset, int bytes_to_corrupt) {
    // Pick file to corrupt
@@ -74,8 +74,17 @@ TEST_F(DBBlobCorruptionTest, VerifyWholeBlobFileChecksum) {
 #endif  // !ROCKSDB_LITE
 }  // namespace ROCKSDB_NAMESPACE

+#ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+extern "C" {
+void RegisterCustomObjects(int argc, char** argv);
+}
+#else
+void RegisterCustomObjects(int /*argc*/, char** /*argv*/) {}
+#endif  // !ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+
 int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
+  RegisterCustomObjects(argc, argv);
  return RUN_ALL_TESTS();
 }
@@ -41,8 +41,7 @@ class DBBlobIndexTest : public DBTestBase {
                                       Tier::kImmutableMemtables,
                                       Tier::kL0SstFile, Tier::kLnSstFile};

-  DBBlobIndexTest()
-      : DBTestBase("/db_blob_index_test", /*env_do_fsync=*/true) {}
+  DBBlobIndexTest() : DBTestBase("db_blob_index_test", /*env_do_fsync=*/true) {}

  ColumnFamilyHandle* cfh() { return dbfull()->DefaultColumnFamily(); }

@@ -565,8 +564,17 @@ TEST_F(DBBlobIndexTest, IntegratedBlobIterate) {

 }  // namespace ROCKSDB_NAMESPACE

+#ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+extern "C" {
+void RegisterCustomObjects(int argc, char** argv);
+}
+#else
+void RegisterCustomObjects(int /*argc*/, char** /*argv*/) {}
+#endif  // !ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
+
 int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
+  RegisterCustomObjects(argc, argv);
  return RUN_ALL_TESTS();
 }
@@ -65,7 +65,8 @@ Status BuildTable(
    SequenceNumber earliest_write_conflict_snapshot,
    SnapshotChecker* snapshot_checker, bool paranoid_file_checks,
    InternalStats* internal_stats, IOStatus* io_status,
-    const std::shared_ptr<IOTracer>& io_tracer, EventLogger* event_logger,
+    const std::shared_ptr<IOTracer>& io_tracer,
+    BlobFileCreationReason blob_creation_reason, EventLogger* event_logger,
    int job_id, const Env::IOPriority io_priority,
    TableProperties* table_properties, Env::WriteLifeTimeHint write_hint,
    const std::string* full_history_ts_low,
@@ -165,7 +166,7 @@ Status BuildTable(
          std::move(file), fname, file_options, ioptions.clock, io_tracer,
          ioptions.stats, ioptions.listeners,
          ioptions.file_checksum_gen_factory.get(),
-          tmp_set.Contains(FileType::kTableFile)));
+          tmp_set.Contains(FileType::kTableFile), false));

      builder = NewTableBuilder(tboptions, file_writer.get());
    }
@@ -178,12 +179,12 @@ Status BuildTable(

    std::unique_ptr<BlobFileBuilder> blob_file_builder(
        (mutable_cf_options.enable_blob_files && blob_file_additions)
-            ? new BlobFileBuilder(versions, fs, &ioptions, &mutable_cf_options,
-                                  &file_options, job_id,
-                                  tboptions.column_family_id,
-                                  tboptions.column_family_name, io_priority,
-                                  write_hint, io_tracer, blob_callback,
-                                  &blob_file_paths, blob_file_additions)
+            ? new BlobFileBuilder(
+                  versions, fs, &ioptions, &mutable_cf_options, &file_options,
+                  job_id, tboptions.column_family_id,
+                  tboptions.column_family_name, io_priority, write_hint,
+                  io_tracer, blob_callback, blob_creation_reason,
+                  &blob_file_paths, blob_file_additions)
            : nullptr);

    CompactionIterator c_iter(
@@ -204,6 +205,8 @@ Status BuildTable(
      const Slice& value = c_iter.value();
      const ParsedInternalKey& ikey = c_iter.ikey();
      // Generate a rolling 64-bit hash of the key and values
+      // Note :
+      // Here "key" integrates 'sequence_number'+'kType'+'user key'.
      s = output_validator.Add(key, value);
      if (!s.ok()) {
        break;
@@ -309,7 +312,7 @@ Status BuildTable(
      if (s.ok()) {
        s = blob_file_builder->Finish();
      } else {
-        blob_file_builder->Abandon();
+        blob_file_builder->Abandon(s);
      }
      blob_file_builder.reset();
    }
@@ -60,6 +60,7 @@ extern Status BuildTable(
    SnapshotChecker* snapshot_checker, bool paranoid_file_checks,
    InternalStats* internal_stats, IOStatus* io_status,
    const std::shared_ptr<IOTracer>& io_tracer,
+    BlobFileCreationReason blob_creation_reason,
    EventLogger* event_logger = nullptr, int job_id = 0,
    const Env::IOPriority io_priority = Env::IO_HIGH,
    TableProperties* table_properties = nullptr,
@@ -88,6 +88,7 @@ using ROCKSDB_NAMESPACE::NewBloomFilterPolicy;
 using ROCKSDB_NAMESPACE::NewCompactOnDeletionCollectorFactory;
 using ROCKSDB_NAMESPACE::NewGenericRateLimiter;
 using ROCKSDB_NAMESPACE::NewLRUCache;
+using ROCKSDB_NAMESPACE::NewRibbonFilterPolicy;
 using ROCKSDB_NAMESPACE::OptimisticTransactionDB;
 using ROCKSDB_NAMESPACE::OptimisticTransactionOptions;
 using ROCKSDB_NAMESPACE::Options;
@@ -3028,6 +3029,11 @@ unsigned char rocksdb_options_get_advise_random_on_open(
  return opt->rep.advise_random_on_open;
 }

+void rocksdb_options_set_experimental_mempurge_threshold(rocksdb_options_t* opt,
+                                                         double v) {
+  opt->rep.experimental_mempurge_threshold = v;
+}
+
 void rocksdb_options_set_access_hint_on_compaction_start(
    rocksdb_options_t* opt, int v) {
  switch(v) {
@@ -3834,7 +3840,8 @@ void rocksdb_filterpolicy_destroy(rocksdb_filterpolicy_t* filter) {
  delete filter;
 }

-rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_format(int bits_per_key, bool original_format) {
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_format(
+    double bits_per_key, bool original_format) {
  // Make a rocksdb_filterpolicy_t, but override all of its methods so
  // they delegate to a NewBloomFilterPolicy() instead of user
  // supplied C functions.
@@ -3869,14 +3876,63 @@ rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_format(int bits_per_ke
  return wrapper;
 }

-rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_full(int bits_per_key) {
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom_full(
+    double bits_per_key) {
  return rocksdb_filterpolicy_create_bloom_format(bits_per_key, false);
 }

-rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(int bits_per_key) {
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(double bits_per_key) {
  return rocksdb_filterpolicy_create_bloom_format(bits_per_key, true);
 }

+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_ribbon_format(
+    double bloom_equivalent_bits_per_key, int bloom_before_level) {
+  // Make a rocksdb_filterpolicy_t, but override all of its methods so
+  // they delegate to a NewRibbonFilterPolicy() instead of user
+  // supplied C functions.
+  struct Wrapper : public rocksdb_filterpolicy_t {
+    const FilterPolicy* rep_;
+    ~Wrapper() override { delete rep_; }
+    const char* Name() const override { return rep_->Name(); }
+    void CreateFilter(const Slice* keys, int n,
+                      std::string* dst) const override {
+      return rep_->CreateFilter(keys, n, dst);
+    }
+    bool KeyMayMatch(const Slice& key, const Slice& filter) const override {
+      return rep_->KeyMayMatch(key, filter);
+    }
+    ROCKSDB_NAMESPACE::FilterBitsBuilder* GetBuilderWithContext(
+        const ROCKSDB_NAMESPACE::FilterBuildingContext& context)
+        const override {
+      return rep_->GetBuilderWithContext(context);
+    }
+    ROCKSDB_NAMESPACE::FilterBitsReader* GetFilterBitsReader(
+        const Slice& contents) const override {
+      return rep_->GetFilterBitsReader(contents);
+    }
+    static void DoNothing(void*) {}
+  };
+  Wrapper* wrapper = new Wrapper;
+  wrapper->rep_ =
+      NewRibbonFilterPolicy(bloom_equivalent_bits_per_key, bloom_before_level);
+  wrapper->state_ = nullptr;
+  wrapper->delete_filter_ = nullptr;
+  wrapper->destructor_ = &Wrapper::DoNothing;
+  return wrapper;
+}
+
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_ribbon(
+    double bloom_equivalent_bits_per_key) {
+  return rocksdb_filterpolicy_create_ribbon_format(
+      bloom_equivalent_bits_per_key, /*bloom_before_level = disabled*/ -1);
+}
+
+rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_ribbon_hybrid(
+    double bloom_equivalent_bits_per_key, int bloom_before_level) {
+  return rocksdb_filterpolicy_create_ribbon_format(
+      bloom_equivalent_bits_per_key, bloom_before_level);
+}
+
 rocksdb_mergeoperator_t* rocksdb_mergeoperator_create(
    void* state, void (*destructor)(void*),
    char* (*full_merge)(void*, const char* key, size_t key_length,
@@ -4828,6 +4884,27 @@ void rocksdb_optimistictransaction_options_set_set_snapshot(
  opt->rep.set_snapshot = v;
 }

+char* rocksdb_optimistictransactiondb_property_value(
+    rocksdb_optimistictransactiondb_t* db, const char* propname) {
+  std::string tmp;
+  if (db->rep->GetProperty(Slice(propname), &tmp)) {
+    // We use strdup() since we expect human readable output.
+    return strdup(tmp.c_str());
+  } else {
+    return nullptr;
+  }
+}
+
+int rocksdb_optimistictransactiondb_property_int(
+    rocksdb_optimistictransactiondb_t* db, const char* propname,
+    uint64_t* out_val) {
+  if (db->rep->GetIntProperty(Slice(propname), out_val)) {
+    return 0;
+  } else {
+    return -1;
+  }
+}
+
 rocksdb_column_family_handle_t* rocksdb_transactiondb_create_column_family(
    rocksdb_transactiondb_t* txn_db,
    const rocksdb_options_t* column_family_options,
@@ -4898,6 +4975,27 @@ void rocksdb_transactiondb_release_snapshot(
  delete snapshot;
 }

+char* rocksdb_transactiondb_property_value(rocksdb_transactiondb_t* db,
+                                           const char* propname) {
+  std::string tmp;
+  if (db->rep->GetProperty(Slice(propname), &tmp)) {
+    // We use strdup() since we expect human readable output.
+    return strdup(tmp.c_str());
+  } else {
+    return nullptr;
+  }
+}
+
+int rocksdb_transactiondb_property_int(rocksdb_transactiondb_t* db,
+                                       const char* propname,
+                                       uint64_t* out_val) {
+  if (db->rep->GetIntProperty(Slice(propname), out_val)) {
+    return 0;
+  } else {
+    return -1;
+  }
+}
+
 rocksdb_transaction_t* rocksdb_transaction_begin(
    rocksdb_transactiondb_t* txn_db,
    const rocksdb_writeoptions_t* write_options,
@@ -5300,12 +5398,31 @@ rocksdb_transaction_t* rocksdb_optimistictransaction_begin(
  return old_txn;
 }

+// Write batch into OptimisticTransactionDB
+void rocksdb_optimistictransactiondb_write(
+    rocksdb_optimistictransactiondb_t* otxn_db,
+    const rocksdb_writeoptions_t* options, rocksdb_writebatch_t* batch,
+    char** errptr) {
+  SaveError(errptr, otxn_db->rep->Write(options->rep, &batch->rep));
+}
+
 void rocksdb_optimistictransactiondb_close(
    rocksdb_optimistictransactiondb_t* otxn_db) {
  delete otxn_db->rep;
  delete otxn_db;
 }

+rocksdb_checkpoint_t* rocksdb_optimistictransactiondb_checkpoint_object_create(
+    rocksdb_optimistictransactiondb_t* otxn_db, char** errptr) {
+  Checkpoint* checkpoint;
+  if (SaveError(errptr, Checkpoint::Create(otxn_db->rep, &checkpoint))) {
+    return nullptr;
+  }
+  rocksdb_checkpoint_t* result = new rocksdb_checkpoint_t;
+  result->rep = checkpoint;
+  return result;
+}
+
 void rocksdb_free(void* ptr) { free(ptr); }

 rocksdb_pinnableslice_t* rocksdb_get_pinned(
@@ -1043,19 +1043,25 @@ int main(int argc, char** argv) {
  }

  StartPhase("filter");
-  for (run = 0; run <= 2; run++) {
-    // First run uses custom filter
-    // Second run uses old block-based bloom filter
-    // Third run uses full bloom filter
+  for (run = 0; run <= 4; run++) {
+    // run=0 uses custom filter
+    // run=1 uses old block-based bloom filter
+    // run=2 run uses full bloom filter
+    // run=3 uses Ribbon
+    // run=4 uses Ribbon-Bloom hybrid configuration
    CheckNoError(err);
    rocksdb_filterpolicy_t* policy;
    if (run == 0) {
      policy = rocksdb_filterpolicy_create(NULL, FilterDestroy, FilterCreate,
                                           FilterKeyMatch, NULL, FilterName);
    } else if (run == 1) {
-      policy = rocksdb_filterpolicy_create_bloom(8);
+      policy = rocksdb_filterpolicy_create_bloom(8.0);
+    } else if (run == 2) {
+      policy = rocksdb_filterpolicy_create_bloom_full(8.0);
+    } else if (run == 3) {
+      policy = rocksdb_filterpolicy_create_ribbon(8.0);
    } else {
-      policy = rocksdb_filterpolicy_create_bloom_full(8);
+      policy = rocksdb_filterpolicy_create_ribbon_hybrid(8.0, 1);
    }
    rocksdb_block_based_options_set_filter_policy(table_options, policy);

@@ -1121,9 +1127,12 @@ int main(int argc, char** argv) {
      } else if (run == 1) {
        // Essentially a fingerprint of the block-based Bloom schema
        CheckCondition(hits == 241);
-      } else {
+      } else if (run == 2 || run == 4) {
        // Essentially a fingerprint of full Bloom schema, format_version=5
        CheckCondition(hits == 188);
+      } else {
+        // Essentially a fingerprint of Ribbon schema
+        CheckCondition(hits == 226);
      }
      CheckCondition(
          (keys_to_query - hits) ==
@@ -29,7 +29,6 @@
 #include "db/version_set.h"
 #include "db/write_controller.h"
 #include "file/sst_file_manager_impl.h"
-#include "memtable/hash_skiplist_rep.h"
 #include "monitoring/thread_status_util.h"
 #include "options/options_helper.h"
 #include "port/port.h"
@@ -203,7 +202,7 @@ Status CheckCFPathsSupported(const DBOptions& db_options,
 namespace {
 const uint64_t kDefaultTtl = 0xfffffffffffffffe;
 const uint64_t kDefaultPeriodicCompSecs = 0xfffffffffffffffe;
-};  // namespace
+}  // namespace

 ColumnFamilyOptions SanitizeOptions(const ImmutableDBOptions& db_options,
                                    const ColumnFamilyOptions& src) {
@@ -445,6 +444,9 @@ bool SuperVersion::Unref() {

 void SuperVersion::Cleanup() {
  assert(refs.load(std::memory_order_relaxed) == 0);
+  // Since this SuperVersion object is being deleted,
+  // decrement reference to the immutable MemtableList
+  // this SV object was pointing to.
  imm->Unref(&to_delete);
  MemTable* m = mem->Unref();
  if (m != nullptr) {
@@ -454,7 +456,7 @@ void SuperVersion::Cleanup() {
    to_delete.push_back(m);
  }
  current->Unref();
-  cfd->UnrefAndTryDelete(this);
+  cfd->UnrefAndTryDelete();
 }

 void SuperVersion::Init(ColumnFamilyData* new_cfd, MemTable* new_mem,
@@ -472,10 +474,10 @@ void SuperVersion::Init(ColumnFamilyData* new_cfd, MemTable* new_mem,

 namespace {
 void SuperVersionUnrefHandle(void* ptr) {
-  // UnrefHandle is called when a thread exists or a ThreadLocalPtr gets
-  // destroyed. When former happens, the thread shouldn't see kSVInUse.
-  // When latter happens, we are in ~ColumnFamilyData(), no get should happen as
-  // well.
+  // UnrefHandle is called when a thread exits or a ThreadLocalPtr gets
+  // destroyed. When the former happens, the thread shouldn't see kSVInUse.
+  // When the latter happens, only super_version_ holds a reference
+  // to ColumnFamilyData, so no further queries are possible.
  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
  bool was_last_ref __attribute__((__unused__));
  was_last_ref = sv->Unref();
@@ -502,7 +504,7 @@ ColumnFamilyData::ColumnFamilyData(
    uint32_t id, const std::string& name, Version* _dummy_versions,
    Cache* _table_cache, WriteBufferManager* write_buffer_manager,
    const ColumnFamilyOptions& cf_options, const ImmutableDBOptions& db_options,
-    const FileOptions& file_options, ColumnFamilySet* column_family_set,
+    const FileOptions* file_options, ColumnFamilySet* column_family_set,
    BlockCacheTracer* const block_cache_tracer,
    const std::shared_ptr<IOTracer>& io_tracer,
    const std::string& db_session_id)
@@ -665,7 +667,7 @@ ColumnFamilyData::~ColumnFamilyData() {
  }
 }

-bool ColumnFamilyData::UnrefAndTryDelete(SuperVersion* sv_under_cleanup) {
+bool ColumnFamilyData::UnrefAndTryDelete() {
  int old_refs = refs_.fetch_sub(1);
  assert(old_refs > 0);

@@ -675,22 +677,17 @@ bool ColumnFamilyData::UnrefAndTryDelete(SuperVersion* sv_under_cleanup) {
    return true;
  }

-  // If called under SuperVersion::Cleanup, we should not re-enter Cleanup on
-  // the same SuperVersion. (But while installing a new SuperVersion, this
-  // cfd could be referenced only by two SuperVersions.)
-  if (old_refs == 2 && super_version_ != nullptr &&
-      super_version_ != sv_under_cleanup) {
+  if (old_refs == 2 && super_version_ != nullptr) {
    // Only the super_version_ holds me
    SuperVersion* sv = super_version_;
    super_version_ = nullptr;
-    // Release SuperVersion reference kept in ThreadLocalPtr.
-    // This must be done outside of mutex_ since unref handler can lock mutex.
-    sv->db_mutex->Unlock();
+
+    // Release SuperVersion references kept in ThreadLocalPtr.
    local_sv_.reset();
-    sv->db_mutex->Lock();

    if (sv->Unref()) {
-      // May delete this ColumnFamilyData after calling Cleanup()
+      // Note: sv will delete this ColumnFamilyData during Cleanup()
+      assert(sv->cfd == this);
      sv->Cleanup();
      delete sv;
      return true;
@@ -1051,6 +1048,10 @@ uint64_t ColumnFamilyData::GetTotalSstFilesSize() const {
  return VersionSet::GetTotalSstFilesSize(dummy_versions_);
 }

+uint64_t ColumnFamilyData::GetTotalBlobFileSize() const {
+  return VersionSet::GetTotalBlobFileSize(dummy_versions_);
+}
+
 uint64_t ColumnFamilyData::GetLiveSstFilesSize() const {
  return current_->GetSstFilesSize();
 }
@@ -1255,14 +1256,13 @@ bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
 void ColumnFamilyData::InstallSuperVersion(
    SuperVersionContext* sv_context, InstrumentedMutex* db_mutex) {
  db_mutex->AssertHeld();
-  return InstallSuperVersion(sv_context, db_mutex, mutable_cf_options_);
+  return InstallSuperVersion(sv_context, mutable_cf_options_);
 }

 void ColumnFamilyData::InstallSuperVersion(
-    SuperVersionContext* sv_context, InstrumentedMutex* db_mutex,
+    SuperVersionContext* sv_context,
    const MutableCFOptions& mutable_cf_options) {
  SuperVersion* new_superversion = sv_context->new_superversion.release();
-  new_superversion->db_mutex = db_mutex;
  new_superversion->mutable_cf_options = mutable_cf_options;
  new_superversion->Init(this, mem_, imm_.current(), current_);
  SuperVersion* old_superversion = super_version_;
@@ -1456,14 +1456,14 @@ ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                 const std::shared_ptr<IOTracer>& io_tracer,
                                 const std::string& db_session_id)
    : max_column_family_(0),
+      file_options_(file_options),
      dummy_cfd_(new ColumnFamilyData(
          ColumnFamilyData::kDummyColumnFamilyDataId, "", nullptr, nullptr,
-          nullptr, ColumnFamilyOptions(), *db_options, file_options, nullptr,
+          nullptr, ColumnFamilyOptions(), *db_options, &file_options_, nullptr,
          block_cache_tracer, io_tracer, db_session_id)),
      default_cfd_cache_(nullptr),
      db_name_(dbname),
      db_options_(db_options),
-      file_options_(file_options),
      table_cache_(table_cache),
      write_buffer_manager_(_write_buffer_manager),
      write_controller_(_write_controller),
@@ -1535,7 +1535,7 @@ ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
  assert(column_families_.find(name) == column_families_.end());
  ColumnFamilyData* new_cfd = new ColumnFamilyData(
      id, name, dummy_versions, table_cache_, write_buffer_manager_, options,
-      *db_options_, file_options_, this, block_cache_tracer_, io_tracer_,
+      *db_options_, &file_options_, this, block_cache_tracer_, io_tracer_,
      db_session_id_);
  column_families_.insert({name, id});
  column_family_data_.insert({id, new_cfd});
@@ -208,8 +208,6 @@ struct SuperVersion {
  uint64_t version_number;
  WriteStallCondition write_stall_condition;

-  InstrumentedMutex* db_mutex;
-
  // should be called outside the mutex
  SuperVersion() = default;
  ~SuperVersion();
@@ -281,8 +279,7 @@ class ColumnFamilyData {
  // UnrefAndTryDelete() decreases the reference count and do free if needed,
  // return true if this is freed else false, UnrefAndTryDelete() can only
  // be called while holding a DB mutex, or during single-threaded recovery.
-  // sv_under_cleanup is only provided when called from SuperVersion::Cleanup.
-  bool UnrefAndTryDelete(SuperVersion* sv_under_cleanup = nullptr);
+  bool UnrefAndTryDelete();

  // SetDropped() can only be called under following conditions:
  // 1) Holding a DB mutex,
@@ -360,6 +357,7 @@ class ColumnFamilyData {
  uint64_t GetNumLiveVersions() const;  // REQUIRE: DB mutex held
  uint64_t GetTotalSstFilesSize() const;  // REQUIRE: DB mutex held
  uint64_t GetLiveSstFilesSize() const;   // REQUIRE: DB mutex held
+  uint64_t GetTotalBlobFileSize() const;  // REQUIRE: DB mutex held
  void SetMemtable(MemTable* new_mem) {
    uint64_t memtable_id = last_memtable_id_.fetch_add(1) + 1;
    new_mem->SetID(memtable_id);
@@ -453,7 +451,6 @@ class ColumnFamilyData {
  // the clients to allocate SuperVersion outside of mutex.
  // IMPORTANT: Only call this from DBImpl::InstallSuperVersion()
  void InstallSuperVersion(SuperVersionContext* sv_context,
-                           InstrumentedMutex* db_mutex,
                           const MutableCFOptions& mutable_cf_options);
  void InstallSuperVersion(SuperVersionContext* sv_context,
                           InstrumentedMutex* db_mutex);
@@ -520,6 +517,7 @@ class ColumnFamilyData {
  }

  ThreadLocalPtr* TEST_GetLocalSV() { return local_sv_.get(); }
+  WriteBufferManager* write_buffer_mgr() { return write_buffer_manager_; }

 private:
  friend class ColumnFamilySet;
@@ -529,7 +527,7 @@ class ColumnFamilyData {
                   WriteBufferManager* write_buffer_manager,
                   const ColumnFamilyOptions& options,
                   const ImmutableDBOptions& db_options,
-                   const FileOptions& file_options,
+                   const FileOptions* file_options,
                   ColumnFamilySet* column_family_set,
                   BlockCacheTracer* const block_cache_tracer,
                   const std::shared_ptr<IOTracer>& io_tracer,
@@ -720,6 +718,8 @@ class ColumnFamilySet {
  std::unordered_map<uint32_t, ColumnFamilyData*> column_family_data_;

  uint32_t max_column_family_;
+  const FileOptions file_options_;
+
  ColumnFamilyData* dummy_cfd_;
  // We don't hold the refcount here, since default column family always exists
  // We are also not responsible for cleaning up default_cfd_cache_. This is
@@ -729,7 +729,6 @@ class ColumnFamilySet {

  const std::string db_name_;
  const ImmutableDBOptions* const db_options_;
-  const FileOptions file_options_;
  Cache* table_cache_;
  WriteBufferManager* write_buffer_manager_;
  WriteController* write_controller_;
@@ -2975,7 +2975,8 @@ TEST_P(ColumnFamilyTest, FlushCloseWALFiles) {
  SpecialEnv env(Env::Default());
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
-  column_family_options_.memtable_factory.reset(new SpecialSkipListFactory(2));
+  column_family_options_.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(2));
  Open();
  CreateColumnFamilies({"one"});
  ASSERT_OK(Put(1, "fodor", "mirko"));
@@ -3020,7 +3021,8 @@ TEST_P(ColumnFamilyTest, IteratorCloseWALFile1) {
  SpecialEnv env(Env::Default());
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
-  column_family_options_.memtable_factory.reset(new SpecialSkipListFactory(2));
+  column_family_options_.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(2));
  Open();
  CreateColumnFamilies({"one"});
  ASSERT_OK(Put(1, "fodor", "mirko"));
@@ -3071,7 +3073,8 @@ TEST_P(ColumnFamilyTest, IteratorCloseWALFile2) {
  env.SetBackgroundThreads(2, Env::HIGH);
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
-  column_family_options_.memtable_factory.reset(new SpecialSkipListFactory(2));
+  column_family_options_.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(2));
  Open();
  CreateColumnFamilies({"one"});
  ASSERT_OK(Put(1, "fodor", "mirko"));
@@ -3129,7 +3132,8 @@ TEST_P(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
  env.SetBackgroundThreads(2, Env::HIGH);
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
-  column_family_options_.memtable_factory.reset(new SpecialSkipListFactory(3));
+  column_family_options_.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(3));
  column_family_options_.level0_file_num_compaction_trigger = 2;
  Open();
  CreateColumnFamilies({"one"});
@@ -211,9 +211,9 @@ Compaction::Compaction(
    std::vector<CompactionInputFiles> _inputs, int _output_level,
    uint64_t _target_file_size, uint64_t _max_compaction_bytes,
    uint32_t _output_path_id, CompressionType _compression,
-    CompressionOptions _compression_opts, uint32_t _max_subcompactions,
-    std::vector<FileMetaData*> _grandparents, bool _manual_compaction,
-    double _score, bool _deletion_compaction,
+    CompressionOptions _compression_opts, Temperature _output_temperature,
+    uint32_t _max_subcompactions, std::vector<FileMetaData*> _grandparents,
+    bool _manual_compaction, double _score, bool _deletion_compaction,
    CompactionReason _compaction_reason)
    : input_vstorage_(vstorage),
      start_level_(_inputs[0].level),
@@ -229,6 +229,7 @@ Compaction::Compaction(
      output_path_id_(_output_path_id),
      output_compression_(_compression),
      output_compression_opts_(_compression_opts),
+      output_temperature_(_output_temperature),
      deletion_compaction_(_deletion_compaction),
      inputs_(PopulateWithAtomicBoundaries(vstorage, std::move(_inputs))),
      grandparents_(std::move(_grandparents)),
@@ -237,7 +238,8 @@ Compaction::Compaction(
      is_full_compaction_(IsFullCompaction(vstorage, inputs_)),
      is_manual_compaction_(_manual_compaction),
      is_trivial_move_(false),
-      compaction_reason_(_compaction_reason) {
+      compaction_reason_(_compaction_reason),
+      notify_on_compaction_completion_(false) {
  MarkFilesBeingCompacted(true);
  if (is_manual_compaction_) {
    compaction_reason_ = CompactionReason::kManualCompaction;
@@ -307,6 +309,12 @@ bool Compaction::IsTrivialMove() const {
    return false;
  }

+  if (start_level_ == output_level_) {
+    // It doesn't make sense if compaction picker picks files just to trivial
+    // move to the same level.
+    return false;
+  }
+
  // Used in universal compaction, where trivial move can be done if the
  // input files are non overlapping
  if ((mutable_cf_options_.compaction_options_universal.allow_trivial_move) &&
@@ -581,6 +589,29 @@ bool Compaction::ShouldFormSubcompactions() const {
  }
 }

+bool Compaction::DoesInputReferenceBlobFiles() const {
+  assert(input_version_);
+
+  const VersionStorageInfo* storage_info = input_version_->storage_info();
+  assert(storage_info);
+
+  if (storage_info->GetBlobFiles().empty()) {
+    return false;
+  }
+
+  for (size_t i = 0; i < inputs_.size(); ++i) {
+    for (const FileMetaData* meta : inputs_[i].files) {
+      assert(meta);
+
+      if (meta->oldest_blob_file_number != kInvalidBlobFileNumber) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
 uint64_t Compaction::MinInputFileOldestAncesterTime() const {
  uint64_t min_oldest_ancester_time = port::kMaxUint64;
  for (const auto& level_files : inputs_) {
@@ -76,7 +76,8 @@ class Compaction {
             std::vector<CompactionInputFiles> inputs, int output_level,
             uint64_t target_file_size, uint64_t max_compaction_bytes,
             uint32_t output_path_id, CompressionType compression,
-             CompressionOptions compression_opts, uint32_t max_subcompactions,
+             CompressionOptions compression_opts,
+             Temperature output_temperature, uint32_t max_subcompactions,
             std::vector<FileMetaData*> grandparents,
             bool manual_compaction = false, double score = -1,
             bool deletion_compaction = false,
@@ -266,6 +267,11 @@ class Compaction {
  // Should this compaction be broken up into smaller ones run in parallel?
  bool ShouldFormSubcompactions() const;

+  // Returns true iff at least one input file references a blob file.
+  //
+  // PRE: input version has been set.
+  bool DoesInputReferenceBlobFiles() const;
+
  // test function to validate the functionality of IsBottommostLevel()
  // function -- determines if compaction with inputs and storage is bottommost
  static bool TEST_IsBottommostLevel(
@@ -294,10 +300,22 @@ class Compaction {

  uint64_t max_compaction_bytes() const { return max_compaction_bytes_; }

+  Temperature output_temperature() const { return output_temperature_; }
+
  uint32_t max_subcompactions() const { return max_subcompactions_; }

  uint64_t MinInputFileOldestAncesterTime() const;

+  // Called by DBImpl::NotifyOnCompactionCompleted to make sure number of
+  // compaction begin and compaction completion callbacks match.
+  void SetNotifyOnCompactionCompleted() {
+    notify_on_compaction_completion_ = true;
+  }
+
+  bool ShouldNotifyOnCompactionCompleted() const {
+    return notify_on_compaction_completion_;
+  }
+
 private:
  // mark (or clear) all files that are being compacted
  void MarkFilesBeingCompacted(bool mark_as_compacted);
@@ -341,6 +359,7 @@ class Compaction {
  const uint32_t output_path_id_;
  CompressionType output_compression_;
  CompressionOptions output_compression_opts_;
+  Temperature output_temperature_;
  // If true, then the compaction can be done by simply deleting input files.
  const bool deletion_compaction_;

@@ -381,6 +400,10 @@ class Compaction {

  // Reason for compaction
  CompactionReason compaction_reason_;
+
+  // Notify on compaction completion only if listener was notified on compaction
+  // begin.
+  bool notify_on_compaction_completion_;
 };

 // Return sum of sizes of all files in `files`.
@@ -5,6 +5,8 @@

 #pragma once

+#include <cstdint>
+
 #include "rocksdb/rocksdb_namespace.h"

 struct CompactionIterationStats {
@@ -38,4 +40,6 @@ struct CompactionIterationStats {
  // Blob related statistics
  uint64_t num_blobs_read = 0;
  uint64_t total_blob_bytes_read = 0;
+  uint64_t num_blobs_relocated = 0;
+  uint64_t total_blob_bytes_relocated = 0;
 };
@@ -16,22 +16,6 @@
 #include "table/internal_iterator.h"
 #include "test_util/sync_point.h"

-#define DEFINITELY_IN_SNAPSHOT(seq, snapshot)                       \
-  ((seq) <= (snapshot) &&                                           \
-   (snapshot_checker_ == nullptr ||                                 \
-    LIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
-           SnapshotCheckerResult::kInSnapshot)))
-
-#define DEFINITELY_NOT_IN_SNAPSHOT(seq, snapshot)                     \
-  ((seq) > (snapshot) ||                                              \
-   (snapshot_checker_ != nullptr &&                                   \
-    UNLIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
-             SnapshotCheckerResult::kNotInSnapshot)))
-
-#define IN_EARLIEST_SNAPSHOT(seq) \
-  ((seq) <= earliest_snapshot_ && \
-   (snapshot_checker_ == nullptr || LIKELY(IsInEarliestSnapshot(seq))))
-
 namespace ROCKSDB_NAMESPACE {
 CompactionIterator::CompactionIterator(
    InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
@@ -75,10 +59,8 @@ CompactionIterator::CompactionIterator(
    const std::atomic<bool>* manual_compaction_canceled,
    const std::shared_ptr<Logger> info_log,
    const std::string* full_history_ts_low)
-    : input_(
-          input, cmp,
-          compaction ==
-              nullptr),  // Now only need to count number of entries in flush.
+    : input_(input, cmp,
+             !compaction || compaction->DoesInputReferenceBlobFiles()),
      cmp_(cmp),
      merge_helper_(merge_helper),
      snapshots_(snapshots),
@@ -513,11 +495,11 @@ void CompactionIterator::NextFromInput() {
                        "Unexpected key type %d for compaction output",
                        ikey_.type);
      }
-      assert(current_user_key_snapshot_ == last_snapshot);
-      if (current_user_key_snapshot_ != last_snapshot) {
+      assert(current_user_key_snapshot_ >= last_snapshot);
+      if (current_user_key_snapshot_ < last_snapshot) {
        ROCKS_LOG_FATAL(info_log_,
                        "current_user_key_snapshot_ (%" PRIu64
-                        ") != last_snapshot (%" PRIu64 ")",
+                        ") < last_snapshot (%" PRIu64 ")",
                        current_user_key_snapshot_, last_snapshot);
      }

@@ -573,10 +555,20 @@ void CompactionIterator::NextFromInput() {
          ParseInternalKey(input_.key(), &next_ikey, allow_data_in_errors_)
              .ok() &&
          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
+#ifndef NDEBUG
+        const Compaction* c =
+            compaction_ ? compaction_->real_compaction() : nullptr;
+#endif
+        TEST_SYNC_POINT_CALLBACK(
+            "CompactionIterator::NextFromInput:SingleDelete:1",
+            const_cast<Compaction*>(c));
+
        // Check whether the next key belongs to the same snapshot as the
        // SingleDelete.
        if (prev_snapshot == 0 ||
-            DEFINITELY_NOT_IN_SNAPSHOT(next_ikey.sequence, prev_snapshot)) {
+            DefinitelyNotInSnapshot(next_ikey.sequence, prev_snapshot)) {
+          TEST_SYNC_POINT_CALLBACK(
+              "CompactionIterator::NextFromInput:SingleDelete:2", nullptr);
          if (next_ikey.type == kTypeSingleDeletion) {
            // We encountered two SingleDeletes in a row.  This could be due to
            // unexpected user input.
@@ -588,8 +580,8 @@ void CompactionIterator::NextFromInput() {
            ++iter_stats_.num_record_drop_obsolete;
            ++iter_stats_.num_single_del_mismatch;
          } else if (has_outputted_key_ ||
-                     DEFINITELY_IN_SNAPSHOT(
-                         ikey_.sequence, earliest_write_conflict_snapshot_)) {
+                     DefinitelyInSnapshot(ikey_.sequence,
+                                          earliest_write_conflict_snapshot_)) {
            // Found a matching value, we can drop the single delete and the
            // value.  It is safe to drop both records since we've already
            // outputted a key in this snapshot, or there is no earlier
@@ -622,6 +614,9 @@ void CompactionIterator::NextFromInput() {
            // Set up the Put to be outputted in the next iteration.
            // (Optimization 3).
            clear_and_output_next_key_ = true;
+            TEST_SYNC_POINT_CALLBACK(
+                "CompactionIterator::NextFromInput:KeepSDForWW",
+                /*arg=*/nullptr);
          }
        } else {
          // We hit the next snapshot without hitting a put, so the iterator
@@ -637,7 +632,8 @@ void CompactionIterator::NextFromInput() {
        // iteration. If the next key is corrupt, we return before the
        // comparison, so the value of has_current_user_key does not matter.
        has_current_user_key_ = false;
-        if (compaction_ != nullptr && IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
+        if (compaction_ != nullptr &&
+            DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_)) {
          // Key doesn't exist outside of this range.
@@ -681,7 +677,7 @@ void CompactionIterator::NextFromInput() {
               (ikey_.type == kTypeDeletion ||
                (ikey_.type == kTypeDeletionWithTimestamp &&
                 cmp_with_history_ts_low_ < 0)) &&
-               IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
+               DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
               ikeyNotNeededForIncrementalSnapshot() &&
               compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                          &level_ptrs_)) {
@@ -724,6 +720,13 @@ void CompactionIterator::NextFromInput() {
                                 ikey_.user_key, &level_ptrs_));
      ParsedInternalKey next_ikey;
      AdvanceInputIter();
+#ifndef NDEBUG
+      const Compaction* c =
+          compaction_ ? compaction_->real_compaction() : nullptr;
+#endif
+      TEST_SYNC_POINT_CALLBACK(
+          "CompactionIterator::NextFromInput:BottommostDelete:1",
+          const_cast<Compaction*>(c));
      // Skip over all versions of this key that happen to occur in the same
      // snapshot range as the delete.
      //
@@ -736,7 +739,7 @@ void CompactionIterator::NextFromInput() {
                  .ok()) &&
             cmp_->EqualWithoutTimestamp(ikey_.user_key, next_ikey.user_key) &&
             (prev_snapshot == 0 ||
-              DEFINITELY_NOT_IN_SNAPSHOT(next_ikey.sequence, prev_snapshot))) {
+              DefinitelyNotInSnapshot(next_ikey.sequence, prev_snapshot))) {
        AdvanceInputIter();
      }
      // If you find you still need to output a row with this key, we need to output the
@@ -756,13 +759,15 @@ void CompactionIterator::NextFromInput() {
      }

      pinned_iters_mgr_.StartPinning();
+      Version* version = compaction_ ? compaction_->input_version() : nullptr;
+
      // We know the merge type entry is not hidden, otherwise we would
      // have hit (A)
      // We encapsulate the merge related state machine in a different
      // object to minimize change to the existing flow.
-      Status s =
-          merge_helper_->MergeUntil(&input_, range_del_agg_, prev_snapshot,
-                                    bottommost_level_, allow_data_in_errors_);
+      Status s = merge_helper_->MergeUntil(&input_, range_del_agg_,
+                                           prev_snapshot, bottommost_level_,
+                                           allow_data_in_errors_, version);
      merge_out_iter_.SeekToFirst();

      if (!s.ok() && !s.IsMergeInProgress()) {
@@ -914,6 +919,9 @@ void CompactionIterator::GarbageCollectBlobIfNeeded() {
    ++iter_stats_.num_blobs_read;
    iter_stats_.total_blob_bytes_read += bytes_read;

+    ++iter_stats_.num_blobs_relocated;
+    iter_stats_.total_blob_bytes_relocated += blob_index.size();
+
    value_ = blob_value_;

    if (ExtractLargeValueIfNeededImpl()) {
@@ -977,7 +985,8 @@ void CompactionIterator::PrepareOutput() {
    if (valid_ && compaction_ != nullptr &&
        !compaction_->allow_ingest_behind() &&
        ikeyNotNeededForIncrementalSnapshot() && bottommost_level_ &&
-        IN_EARLIEST_SNAPSHOT(ikey_.sequence) && ikey_.type != kTypeMerge) {
+        DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
+        ikey_.type != kTypeMerge) {
      assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
      if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion) {
        ROCKS_LOG_FATAL(info_log_,
@@ -1053,7 +1062,7 @@ inline bool CompactionIterator::ikeyNotNeededForIncrementalSnapshot() {
         (ikey_.sequence < preserve_deletes_seqnum_);
 }

-bool CompactionIterator::IsInEarliestSnapshot(SequenceNumber sequence) {
+bool CompactionIterator::IsInCurrentEarliestSnapshot(SequenceNumber sequence) {
  assert(snapshot_checker_ != nullptr);
  bool pre_condition = (earliest_snapshot_ == kMaxSequenceNumber ||
                        (earliest_snapshot_iter_ != snapshots_->end() &&
@@ -97,6 +97,10 @@ class CompactionIterator {
    virtual double blob_garbage_collection_age_cutoff() const = 0;

    virtual Version* input_version() const = 0;
+
+    virtual bool DoesInputReferenceBlobFiles() const = 0;
+
+    virtual const Compaction* real_compaction() const = 0;
  };

  class RealCompaction : public CompactionProxy {
@@ -146,6 +150,12 @@ class CompactionIterator {
      return compaction_->input_version();
    }

+    bool DoesInputReferenceBlobFiles() const override {
+      return compaction_->DoesInputReferenceBlobFiles();
+    }
+
+    const Compaction* real_compaction() const override { return compaction_; }
+
   private:
    const Compaction* compaction_;
  };
@@ -261,7 +271,13 @@ class CompactionIterator {
               SnapshotCheckerResult::kInSnapshot;
  }

-  bool IsInEarliestSnapshot(SequenceNumber sequence);
+  bool IsInCurrentEarliestSnapshot(SequenceNumber sequence);
+
+  bool DefinitelyInSnapshot(SequenceNumber seq, SequenceNumber snapshot);
+
+  bool DefinitelyNotInSnapshot(SequenceNumber seq, SequenceNumber snapshot);
+
+  bool InCurrentEarliestSnapshot(SequenceNumber seq);

  // Extract user-defined timestamp from user key if possible and compare it
  // with *full_history_ts_low_ if applicable.
@@ -406,4 +422,27 @@ class CompactionIterator {
            manual_compaction_canceled_->load(std::memory_order_relaxed));
  }
 };
+
+inline bool CompactionIterator::DefinitelyInSnapshot(SequenceNumber seq,
+                                                     SequenceNumber snapshot) {
+  return ((seq) <= (snapshot) &&
+          (snapshot_checker_ == nullptr ||
+           LIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) ==
+                  SnapshotCheckerResult::kInSnapshot)));
+}
+
+inline bool CompactionIterator::DefinitelyNotInSnapshot(
+    SequenceNumber seq, SequenceNumber snapshot) {
+  return ((seq) > (snapshot) ||
+          (snapshot_checker_ != nullptr &&
+           UNLIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) ==
+                    SnapshotCheckerResult::kNotInSnapshot)));
+}
+
+inline bool CompactionIterator::InCurrentEarliestSnapshot(SequenceNumber seq) {
+  return ((seq) <= earliest_snapshot_ &&
+          (snapshot_checker_ == nullptr ||
+           LIKELY(IsInCurrentEarliestSnapshot(seq))));
+}
+
 }  // namespace ROCKSDB_NAMESPACE
@@ -182,6 +182,10 @@ class FakeCompaction : public CompactionIterator::CompactionProxy {

  Version* input_version() const override { return nullptr; }

+  bool DoesInputReferenceBlobFiles() const override { return false; }
+
+  const Compaction* real_compaction() const override { return nullptr; }
+
  bool key_not_exists_beyond_output_level = false;

  bool is_bottommost_level = false;
@@ -20,8 +20,10 @@
 #include <utility>
 #include <vector>

+#include "db/blob/blob_counting_iterator.h"
 #include "db/blob/blob_file_addition.h"
 #include "db/blob/blob_file_builder.h"
+#include "db/blob/blob_garbage_meter.h"
 #include "db/builder.h"
 #include "db/compaction/clipping_iterator.h"
 #include "db/db_impl/db_impl.h"
@@ -105,6 +107,8 @@ const char* GetCompactionReasonString(CompactionReason compaction_reason) {
      return "ExternalSstIngestion";
    case CompactionReason::kPeriodicCompaction:
      return "PeriodicCompaction";
+    case CompactionReason::kChangeTemperature:
+      return "ChangeTemperature";
    case CompactionReason::kNumOfReasons:
      // fall through
    default:
@@ -147,6 +151,7 @@ struct CompactionJob::SubcompactionState {
  // State kept for output being generated
  std::vector<Output> outputs;
  std::vector<BlobFileAddition> blob_file_additions;
+  std::unique_ptr<BlobGarbageMeter> blob_garbage_meter;
  std::unique_ptr<WritableFileWriter> outfile;
  std::unique_ptr<TableBuilder> builder;

@@ -177,9 +182,17 @@ struct CompactionJob::SubcompactionState {
  uint64_t overlapped_bytes = 0;
  // A flag determine whether the key has been seen in ShouldStopBefore()
  bool seen_key = false;
+  // sub compaction job id, which is used to identify different sub-compaction
+  // within the same compaction job.
+  const uint32_t sub_job_id;

-  SubcompactionState(Compaction* c, Slice* _start, Slice* _end, uint64_t size)
-      : compaction(c), start(_start), end(_end), approx_size(size) {
+  SubcompactionState(Compaction* c, Slice* _start, Slice* _end, uint64_t size,
+                     uint32_t _sub_job_id)
+      : compaction(c),
+        start(_start),
+        end(_end),
+        approx_size(size),
+        sub_job_id(_sub_job_id) {
    assert(compaction != nullptr);
  }

@@ -229,6 +242,14 @@ struct CompactionJob::SubcompactionState {

    return false;
  }
+
+  Status ProcessOutFlowIfNeeded(const Slice& key, const Slice& value) {
+    if (!blob_garbage_meter) {
+      return Status::OK();
+    }
+
+    return blob_garbage_meter->ProcessOutFlow(key, value);
+  }
 };

 // Maintains state for the entire compaction
@@ -438,7 +459,8 @@ void CompactionJob::Prepare() {
    for (size_t i = 0; i <= boundaries_.size(); i++) {
      Slice* start = i == 0 ? nullptr : &boundaries_[i - 1];
      Slice* end = i == boundaries_.size() ? nullptr : &boundaries_[i];
-      compact_->sub_compact_states.emplace_back(c, start, end, sizes_[i]);
+      compact_->sub_compact_states.emplace_back(c, start, end, sizes_[i],
+                                                static_cast<uint32_t>(i));
    }
    RecordInHistogram(stats_, NUM_SUBCOMPACTIONS_SCHEDULED,
                      compact_->sub_compact_states.size());
@@ -447,7 +469,8 @@ void CompactionJob::Prepare() {
    constexpr Slice* end = nullptr;
    constexpr uint64_t size = 0;

-    compact_->sub_compact_states.emplace_back(c, start, end, size);
+    compact_->sub_compact_states.emplace_back(c, start, end, size,
+                                              /*sub_job_id*/ 0);
  }
 }

@@ -858,7 +881,7 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {

  UpdateCompactionJobStats(stats);

-  auto stream = event_logger_->LogToBuffer(log_buffer_);
+  auto stream = event_logger_->LogToBuffer(log_buffer_, 8192);
  stream << "job" << job_id_ << "event"
         << "compaction_finished"
         << "compaction_time_micros" << stats.micros
@@ -909,7 +932,8 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
 }

 #ifndef ROCKSDB_LITE
-void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
+CompactionServiceJobStatus
+CompactionJob::ProcessKeyValueCompactionWithCompactionService(
    SubcompactionState* sub_compact) {
  assert(sub_compact);
  assert(sub_compact->compaction);
@@ -946,7 +970,7 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
  Status s = compaction_input.Write(&compaction_input_binary);
  if (!s.ok()) {
    sub_compact->status = s;
-    return;
+    return CompactionServiceJobStatus::kFailure;
  }

  std::ostringstream input_files_oss;
@@ -961,38 +985,77 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
      "[%s] [JOB %d] Starting remote compaction (output level: %d): %s",
      compaction_input.column_family.name.c_str(), job_id_,
      compaction_input.output_level, input_files_oss.str().c_str());
+  CompactionServiceJobInfo info(dbname_, db_id_, db_session_id_,
+                                GetCompactionId(sub_compact), thread_pri_);
  CompactionServiceJobStatus compaction_status =
-      db_options_.compaction_service->Start(compaction_input_binary, job_id_);
-  if (compaction_status != CompactionServiceJobStatus::kSuccess) {
-    sub_compact->status =
-        Status::Incomplete("CompactionService failed to start compaction job.");
-    return;
+      db_options_.compaction_service->StartV2(info, compaction_input_binary);
+  switch (compaction_status) {
+    case CompactionServiceJobStatus::kSuccess:
+      break;
+    case CompactionServiceJobStatus::kFailure:
+      sub_compact->status = Status::Incomplete(
+          "CompactionService failed to start compaction job.");
+      ROCKS_LOG_WARN(db_options_.info_log,
+                     "[%s] [JOB %d] Remote compaction failed to start.",
+                     compaction_input.column_family.name.c_str(), job_id_);
+      return compaction_status;
+    case CompactionServiceJobStatus::kUseLocal:
+      ROCKS_LOG_INFO(
+          db_options_.info_log,
+          "[%s] [JOB %d] Remote compaction fallback to local by API Start.",
+          compaction_input.column_family.name.c_str(), job_id_);
+      return compaction_status;
+    default:
+      assert(false);  // unknown status
+      break;
  }

+  ROCKS_LOG_INFO(db_options_.info_log,
+                 "[%s] [JOB %d] Waiting for remote compaction...",
+                 compaction_input.column_family.name.c_str(), job_id_);
  std::string compaction_result_binary;
-  compaction_status = db_options_.compaction_service->WaitForComplete(
-      job_id_, &compaction_result_binary);
+  compaction_status = db_options_.compaction_service->WaitForCompleteV2(
+      info, &compaction_result_binary);
+
+  if (compaction_status == CompactionServiceJobStatus::kUseLocal) {
+    ROCKS_LOG_INFO(db_options_.info_log,
+                   "[%s] [JOB %d] Remote compaction fallback to local by API "
+                   "WaitForComplete.",
+                   compaction_input.column_family.name.c_str(), job_id_);
+    return compaction_status;
+  }

  CompactionServiceResult compaction_result;
  s = CompactionServiceResult::Read(compaction_result_binary,
                                    &compaction_result);
-  if (compaction_status != CompactionServiceJobStatus::kSuccess) {
-    sub_compact->status =
-        s.ok() ? compaction_result.status
-               : Status::Incomplete(
-                     "CompactionService failed to run compaction job.");
-    compaction_result.status.PermitUncheckedError();
+
+  if (compaction_status == CompactionServiceJobStatus::kFailure) {
+    if (s.ok()) {
+      if (compaction_result.status.ok()) {
+        sub_compact->status = Status::Incomplete(
+            "CompactionService failed to run the compaction job (even though "
+            "the internal status is okay).");
+      } else {
+        // set the current sub compaction status with the status returned from
+        // remote
+        sub_compact->status = compaction_result.status;
+      }
+    } else {
+      sub_compact->status = Status::Incomplete(
+          "CompactionService failed to run the compaction job (and no valid "
+          "result is returned).");
+      compaction_result.status.PermitUncheckedError();
+    }
    ROCKS_LOG_WARN(db_options_.info_log,
-                   "[%s] [JOB %d] Remote compaction failed, status: %s",
-                   compaction_input.column_family.name.c_str(), job_id_,
-                   s.ToString().c_str());
-    return;
+                   "[%s] [JOB %d] Remote compaction failed.",
+                   compaction_input.column_family.name.c_str(), job_id_);
+    return compaction_status;
  }

  if (!s.ok()) {
    sub_compact->status = s;
    compaction_result.status.PermitUncheckedError();
-    return;
+    return CompactionServiceJobStatus::kFailure;
  }
  sub_compact->status = compaction_result.status;

@@ -1012,7 +1075,7 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(

  if (!s.ok()) {
    sub_compact->status = s;
-    return;
+    return CompactionServiceJobStatus::kFailure;
  }

  for (const auto& file : compaction_result.output_files) {
@@ -1023,7 +1086,7 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
    s = fs_->RenameFile(src_file, tgt_file, IOOptions(), nullptr);
    if (!s.ok()) {
      sub_compact->status = s;
-      return;
+      return CompactionServiceJobStatus::kFailure;
    }

    FileMetaData meta;
@@ -1031,7 +1094,7 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
    s = fs_->GetFileSize(tgt_file, IOOptions(), &file_size, nullptr);
    if (!s.ok()) {
      sub_compact->status = s;
-      return;
+      return CompactionServiceJobStatus::kFailure;
    }
    meta.fd = FileDescriptor(file_num, compaction->output_path_id(), file_size,
                             file.smallest_seqno, file.largest_seqno);
@@ -1052,6 +1115,7 @@ void CompactionJob::ProcessKeyValueCompactionWithCompactionService(
  sub_compact->total_bytes = compaction_result.total_bytes;
  IOSTATS_ADD(bytes_written, compaction_result.bytes_written);
  IOSTATS_ADD(bytes_read, compaction_result.bytes_read);
+  return CompactionServiceJobStatus::kSuccess;
 }
 #endif  // !ROCKSDB_LITE

@@ -1061,7 +1125,14 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {

 #ifndef ROCKSDB_LITE
  if (db_options_.compaction_service) {
-    return ProcessKeyValueCompactionWithCompactionService(sub_compact);
+    CompactionServiceJobStatus comp_status =
+        ProcessKeyValueCompactionWithCompactionService(sub_compact);
+    if (comp_status == CompactionServiceJobStatus::kSuccess ||
+        comp_status == CompactionServiceJobStatus::kFailure) {
+      return;
+    }
+    // fallback to local compaction
+    assert(comp_status == CompactionServiceJobStatus::kUseLocal);
  }
 #endif  // !ROCKSDB_LITE

@@ -1136,6 +1207,15 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    input = clip.get();
  }

+  std::unique_ptr<InternalIterator> blob_counter;
+
+  if (sub_compact->compaction->DoesInputReferenceBlobFiles()) {
+    sub_compact->blob_garbage_meter.reset(new BlobGarbageMeter);
+    blob_counter.reset(
+        new BlobCountingIterator(input, sub_compact->blob_garbage_meter.get()));
+    input = blob_counter.get();
+  }
+
  input->SeekToFirst();

  AutoThreadOperationStageUpdater stage_updater(
@@ -1176,13 +1256,13 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {

  std::unique_ptr<BlobFileBuilder> blob_file_builder(
      mutable_cf_options->enable_blob_files
-          ? new BlobFileBuilder(versions_, fs_.get(),
-                                sub_compact->compaction->immutable_options(),
-                                mutable_cf_options, &file_options_, job_id_,
-                                cfd->GetID(), cfd->GetName(),
-                                Env::IOPriority::IO_LOW, write_hint_,
-                                io_tracer_, blob_callback_, &blob_file_paths,
-                                &sub_compact->blob_file_additions)
+          ? new BlobFileBuilder(
+                versions_, fs_.get(),
+                sub_compact->compaction->immutable_options(),
+                mutable_cf_options, &file_options_, job_id_, cfd->GetID(),
+                cfd->GetName(), Env::IOPriority::IO_LOW, write_hint_,
+                io_tracer_, blob_callback_, BlobFileCreationReason::kCompaction,
+                &blob_file_paths, &sub_compact->blob_file_additions)
          : nullptr);

  TEST_SYNC_POINT("CompactionJob::Run():Inprogress");
@@ -1248,6 +1328,11 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      break;
    }

+    status = sub_compact->ProcessOutFlowIfNeeded(key, value);
+    if (!status.ok()) {
+      break;
+    }
+
    sub_compact->current_output_file_size =
        sub_compact->builder->EstimatedFileSize();
    const ParsedInternalKey& ikey = c_iter->ikey();
@@ -1331,6 +1416,16 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {

  RecordTick(stats_, FILTER_OPERATION_TOTAL_TIME,
             c_iter_stats.total_filter_time);
+
+  if (c_iter_stats.num_blobs_relocated > 0) {
+    RecordTick(stats_, BLOB_DB_GC_NUM_KEYS_RELOCATED,
+               c_iter_stats.num_blobs_relocated);
+  }
+  if (c_iter_stats.total_blob_bytes_relocated > 0) {
+    RecordTick(stats_, BLOB_DB_GC_BYTES_RELOCATED,
+               c_iter_stats.total_blob_bytes_relocated);
+  }
+
  RecordDroppedKeys(c_iter_stats, &sub_compact->compaction_job_stats);
  RecordCompactionIOStats();

@@ -1378,7 +1473,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
    if (status.ok()) {
      status = blob_file_builder->Finish();
    } else {
-      blob_file_builder->Abandon();
+      blob_file_builder->Abandon(status);
    }
    blob_file_builder.reset();
  }
@@ -1415,11 +1510,16 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
 #endif  // ROCKSDB_ASSERT_STATUS_CHECKED

  sub_compact->c_iter.reset();
+  blob_counter.reset();
  clip.reset();
  raw_input.reset();
  sub_compact->status = status;
 }

+uint64_t CompactionJob::GetCompactionId(SubcompactionState* sub_compact) {
+  return (uint64_t)job_id_ << 32 | sub_compact->sub_job_id;
+}
+
 void CompactionJob::RecordDroppedKeys(
    const CompactionIterationStats& c_iter_stats,
    CompactionJobStats* compaction_job_stats) {
@@ -1784,19 +1884,6 @@ Status CompactionJob::InstallCompactionResults(
  auto* compaction = compact_->compaction;
  assert(compaction);

-  // paranoia: verify that the files that we started with
-  // still exist in the current version and in the same original level.
-  // This ensures that a concurrent compaction did not erroneously
-  // pick the same files to compact_.
-  if (!versions_->VerifyCompactionFileConsistency(compaction)) {
-    Compaction::InputLevelSummaryBuffer inputs_summary;
-
-    ROCKS_LOG_ERROR(db_options_.info_log, "[%s] [JOB %d] Compaction %s aborted",
-                    compaction->column_family_data()->GetName().c_str(),
-                    job_id_, compaction->InputLevelSummary(&inputs_summary));
-    return Status::Corruption("Compaction input files inconsistent");
-  }
-
  {
    Compaction::InputLevelSummaryBuffer inputs_summary;
    ROCKS_LOG_INFO(db_options_.info_log,
@@ -1812,6 +1899,8 @@ Status CompactionJob::InstallCompactionResults(
  // Add compaction inputs
  compaction->AddInputDeletions(edit);

+  std::unordered_map<uint64_t, BlobGarbageMeter::BlobStats> blob_total_garbage;
+
  for (const auto& sub_compact : compact_->sub_compact_states) {
    for (const auto& out : sub_compact.outputs) {
      edit->AddFile(compaction->output_level(), out.meta);
@@ -1820,6 +1909,29 @@ Status CompactionJob::InstallCompactionResults(
    for (const auto& blob : sub_compact.blob_file_additions) {
      edit->AddBlobFile(blob);
    }
+
+    if (sub_compact.blob_garbage_meter) {
+      const auto& flows = sub_compact.blob_garbage_meter->flows();
+
+      for (const auto& pair : flows) {
+        const uint64_t blob_file_number = pair.first;
+        const BlobGarbageMeter::BlobInOutFlow& flow = pair.second;
+
+        assert(flow.IsValid());
+        if (flow.HasGarbage()) {
+          blob_total_garbage[blob_file_number].Add(flow.GetGarbageCount(),
+                                                   flow.GetGarbageBytes());
+        }
+      }
+    }
+  }
+
+  for (const auto& pair : blob_total_garbage) {
+    const uint64_t blob_file_number = pair.first;
+    const BlobGarbageMeter::BlobStats& stats = pair.second;
+
+    edit->AddBlobFileGarbage(blob_file_number, stats.GetCount(),
+                             stats.GetBytes());
  }

  return versions_->LogAndApply(compaction->column_family_data(),
@@ -1874,11 +1986,12 @@ Status CompactionJob::OpenCompactionOutputFile(

  // Pass temperature of botommost files to FileSystem.
  FileOptions fo_copy = file_options_;
-  Temperature temperature = Temperature::kUnknown;
-  if (bottommost_level_) {
-    fo_copy.temperature = temperature =
+  Temperature temperature = sub_compact->compaction->output_temperature();
+  if (temperature == Temperature::kUnknown && bottommost_level_) {
+    temperature =
        sub_compact->compaction->mutable_cf_options()->bottommost_temperature;
  }
+  fo_copy.temperature = temperature;

  Status s;
  IOStatus io_s = NewWritableFile(fs_.get(), fname, &writable_file, fo_copy);
@@ -1948,7 +2061,7 @@ Status CompactionJob::OpenCompactionOutputFile(
      std::move(writable_file), fname, file_options_, db_options_.clock,
      io_tracer_, db_options_.stats, listeners,
      db_options_.file_checksum_gen_factory.get(),
-      tmp_set.Contains(FileType::kTableFile)));
+      tmp_set.Contains(FileType::kTableFile), false));

  TableBuilderOptions tboptions(
      *cfd->ioptions(), *(sub_compact->compaction->mutable_cf_options()),
@@ -2179,8 +2292,8 @@ Status CompactionServiceCompactionJob::Run() {
  Slice end = compaction_input_.end;
  compact_->sub_compact_states.emplace_back(
      c, compaction_input_.has_begin ? &begin : nullptr,
-      compaction_input_.has_end ? &end : nullptr,
-      compaction_input_.approx_size);
+      compaction_input_.has_end ? &end : nullptr, compaction_input_.approx_size,
+      /*sub_job_id*/ 0);

  log_buffer_->FlushBufferToLog();
  LogCompaction();
@@ -2219,7 +2332,7 @@ Status CompactionServiceCompactionJob::Run() {
    status = io_s;
  }
  if (status.ok()) {
-    // TODO: Add verify_table() and VerifyCompactionFileConsistency()
+    // TODO: Add verify_table()
  }

  // Finish up all book-keeping to unify the subcompaction results
@@ -146,7 +146,7 @@ class CompactionJob {
  // consecutive groups such that each group has a similar size.
  void GenSubcompactionBoundaries();

-  void ProcessKeyValueCompactionWithCompactionService(
+  CompactionServiceJobStatus ProcessKeyValueCompactionWithCompactionService(
      SubcompactionState* sub_compact);

  // update the thread status for starting a compaction.
@@ -168,7 +168,7 @@ class CompactionJob {
  void UpdateCompactionInputStatsHelper(
      int* num_files, uint64_t* bytes_read, int input_level);

-  int job_id_;
+  uint32_t job_id_;

  CompactionJobStats* compaction_job_stats_;

@@ -219,6 +219,8 @@ class CompactionJob {
  std::string full_history_ts_low_;
  BlobFileCompletionCallback* blob_callback_;

+  uint64_t GetCompactionId(SubcompactionState* sub_compact);
+
  // Get table file name in where it's outputting to, which should also be in
  // `output_directory_`.
  virtual std::string GetTableFileName(uint64_t file_number);
@@ -24,7 +24,6 @@
 #include "db/write_batch_internal.h"
 #include "env/mock_env.h"
 #include "file/filename.h"
-#include "memtable/hash_linklist_rep.h"
 #include "monitoring/statistics.h"
 #include "monitoring/thread_status_util.h"
 #include "port/stack_trace.h"
@@ -269,10 +268,10 @@ class CompactionJobStatsTest : public testing::Test,
    if (cf == 0) {
      // default cfd
      EXPECT_TRUE(db_->GetProperty(
-          "rocksdb.num-files-at-level" + NumberToString(level), &property));
+          "rocksdb.num-files-at-level" + ToString(level), &property));
    } else {
      EXPECT_TRUE(db_->GetProperty(
-          handles_[cf], "rocksdb.num-files-at-level" + NumberToString(level),
+          handles_[cf], "rocksdb.num-files-at-level" + ToString(level),
          &property));
    }
    return atoi(property.c_str());
@@ -334,8 +334,8 @@ class CompactionJobTestBase : public testing::Test {
        cfd->current()->storage_info(), *cfd->ioptions(),
        *cfd->GetLatestMutableCFOptions(), mutable_db_options_,
        compaction_input_files, output_level, 1024 * 1024, 10 * 1024 * 1024, 0,
-        kNoCompression, cfd->GetLatestMutableCFOptions()->compression_opts, 0,
-        {}, true);
+        kNoCompression, cfd->GetLatestMutableCFOptions()->compression_opts,
+        Temperature::kUnknown, 0, {}, true);
    compaction.SetInputVersion(cfd->current());

    LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
@@ -1120,7 +1120,9 @@ TEST_F(CompactionJobTest, InputSerialization) {
    input.snapshots.emplace_back(rnd64.Uniform(UINT64_MAX));
  }
  while (!rnd.OneIn(10)) {
-    input.input_files.emplace_back(rnd.RandomString(rnd.Uniform(kStrMaxLen)));
+    input.input_files.emplace_back(rnd.RandomString(
+        rnd.Uniform(kStrMaxLen - 1) +
+        1));  // input file name should have at least one character
  }
  input.output_level = 4;
  input.has_begin = rnd.OneIn(2);
@@ -358,7 +358,7 @@ Compaction* CompactionPicker::CompactFiles(
      output_level, compact_options.output_file_size_limit,
      mutable_cf_options.max_compaction_bytes, output_path_id, compression_type,
      GetCompressionOptions(mutable_cf_options, vstorage, output_level),
-      compact_options.max_subcompactions,
+      Temperature::kUnknown, compact_options.max_subcompactions,
      /* grandparents */ {}, true);
  RegisterCompaction(c);
  return c;
@@ -634,7 +634,8 @@ Compaction* CompactionPicker::CompactRange(
        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
                           output_level, 1),
        GetCompressionOptions(mutable_cf_options, vstorage, output_level),
-        compact_range_options.max_subcompactions, /* grandparents */ {},
+        Temperature::kUnknown, compact_range_options.max_subcompactions,
+        /* grandparents */ {},
        /* is manual */ true);
    RegisterCompaction(c);
    vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options);
@@ -812,7 +813,8 @@ Compaction* CompactionPicker::CompactRange(
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
                         vstorage->base_level()),
      GetCompressionOptions(mutable_cf_options, vstorage, output_level),
-      compact_range_options.max_subcompactions, std::move(grandparents),
+      Temperature::kUnknown, compact_range_options.max_subcompactions,
+      std::move(grandparents),
      /* is manual compaction */ true);

  TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
@@ -111,7 +111,7 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(
  Compaction* c = new Compaction(
      vstorage, ioptions_, mutable_cf_options, mutable_db_options,
      std::move(inputs), 0, 0, 0, 0, kNoCompression,
-      mutable_cf_options.compression_opts,
+      mutable_cf_options.compression_opts, Temperature::kUnknown,
      /* max_subcompactions */ 0, {}, /* is manual */ false,
      vstorage->CompactionScore(0),
      /* is deletion compaction */ true, CompactionReason::kFIFOTtl);
@@ -154,7 +154,8 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
            {comp_inputs}, 0, 16 * 1024 * 1024 /* output file size limit */,
            0 /* max compaction bytes, not applicable */,
            0 /* output path ID */, mutable_cf_options.compression,
-            mutable_cf_options.compression_opts, 0 /* max_subcompactions */, {},
+            mutable_cf_options.compression_opts, Temperature::kUnknown,
+            0 /* max_subcompactions */, {},
            /* is manual */ false, vstorage->CompactionScore(0),
            /* is deletion compaction */ false,
            CompactionReason::kFIFOReduceNumFiles);
@@ -203,13 +204,119 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
  Compaction* c = new Compaction(
      vstorage, ioptions_, mutable_cf_options, mutable_db_options,
      std::move(inputs), 0, 0, 0, 0, kNoCompression,
-      mutable_cf_options.compression_opts,
+      mutable_cf_options.compression_opts, Temperature::kUnknown,
      /* max_subcompactions */ 0, {}, /* is manual */ false,
      vstorage->CompactionScore(0),
      /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
  return c;
 }

+Compaction* FIFOCompactionPicker::PickCompactionToWarm(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
+    LogBuffer* log_buffer) {
+  if (mutable_cf_options.compaction_options_fifo.age_for_warm == 0) {
+    return nullptr;
+  }
+
+  const int kLevel0 = 0;
+  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
+
+  int64_t _current_time;
+  auto status = ioptions_.clock->GetCurrentTime(&_current_time);
+  if (!status.ok()) {
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "[%s] FIFO compaction: Couldn't get current time: %s. "
+                     "Not doing compactions based on warm threshold. ",
+                     cf_name.c_str(), status.ToString().c_str());
+    return nullptr;
+  }
+  const uint64_t current_time = static_cast<uint64_t>(_current_time);
+
+  if (!level0_compactions_in_progress_.empty()) {
+    ROCKS_LOG_BUFFER(
+        log_buffer,
+        "[%s] FIFO compaction: Already executing compaction. Parallel "
+        "compactions are not supported",
+        cf_name.c_str());
+    return nullptr;
+  }
+
+  std::vector<CompactionInputFiles> inputs;
+  inputs.emplace_back();
+  inputs[0].level = 0;
+
+  // avoid underflow
+  if (current_time > mutable_cf_options.compaction_options_fifo.age_for_warm) {
+    uint64_t create_time_threshold =
+        current_time - mutable_cf_options.compaction_options_fifo.age_for_warm;
+    uint64_t compaction_size = 0;
+    // We will ideally identify a file qualifying for warm tier by knowing
+    // the timestamp for the youngest entry in the file. However, right now
+    // we don't have the information. We infer it by looking at timestamp
+    // of the next file's (which is just younger) oldest entry's timestamp.
+    FileMetaData* prev_file = nullptr;
+    for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
+      FileMetaData* f = *ritr;
+      assert(f);
+      if (f->being_compacted) {
+        // Right now this probably won't happen as we never try to schedule
+        // two compactions in parallel, so here we just simply don't schedule
+        // anything.
+        return nullptr;
+      }
+      uint64_t oldest_ancester_time = f->TryGetOldestAncesterTime();
+      if (oldest_ancester_time == kUnknownOldestAncesterTime) {
+        // Older files might not have enough information. It is possible to
+        // handle these files by looking at newer files, but maintaining the
+        // logic isn't worth it.
+        break;
+      }
+      if (oldest_ancester_time > create_time_threshold) {
+        // The previous file (which has slightly older data) doesn't qualify
+        // for warm tier.
+        break;
+      }
+      if (prev_file != nullptr) {
+        compaction_size += prev_file->fd.GetFileSize();
+        if (compaction_size > mutable_cf_options.max_compaction_bytes) {
+          break;
+        }
+        inputs[0].files.push_back(prev_file);
+        ROCKS_LOG_BUFFER(log_buffer,
+                         "[%s] FIFO compaction: picking file %" PRIu64
+                         " with next file's oldest time %" PRIu64 " for warm",
+                         cf_name.c_str(), prev_file->fd.GetNumber(),
+                         oldest_ancester_time);
+      }
+      if (f->temperature == Temperature::kUnknown ||
+          f->temperature == Temperature::kHot) {
+        prev_file = f;
+      } else if (!inputs[0].files.empty()) {
+        // A warm file newer than files picked.
+        break;
+      } else {
+        assert(prev_file == nullptr);
+      }
+    }
+  }
+
+  if (inputs[0].files.empty()) {
+    return nullptr;
+  }
+
+  Compaction* c = new Compaction(
+      vstorage, ioptions_, mutable_cf_options, mutable_db_options,
+      std::move(inputs), 0, 0 /* output file size limit */,
+      0 /* max compaction bytes, not applicable */, 0 /* output path ID */,
+      mutable_cf_options.compression, mutable_cf_options.compression_opts,
+      Temperature::kWarm,
+      /* max_subcompactions */ 0, {}, /* is manual */ false,
+      vstorage->CompactionScore(0),
+      /* is deletion compaction */ false, CompactionReason::kChangeTemperature);
+  return c;
+}
+
 Compaction* FIFOCompactionPicker::PickCompaction(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
@@ -225,6 +332,10 @@ Compaction* FIFOCompactionPicker::PickCompaction(
    c = PickSizeCompaction(cf_name, mutable_cf_options, mutable_db_options,
                           vstorage, log_buffer);
  }
+  if (c == nullptr) {
+    c = PickCompactionToWarm(cf_name, mutable_cf_options, mutable_db_options,
+                             vstorage, log_buffer);
+  }
  RegisterCompaction(c);
  return c;
 }
@@ -52,6 +52,12 @@ class FIFOCompactionPicker : public CompactionPicker {
                                 const MutableDBOptions& mutable_db_options,
                                 VersionStorageInfo* version,
                                 LogBuffer* log_buffer);
+
+  Compaction* PickCompactionToWarm(const std::string& cf_name,
+                                   const MutableCFOptions& mutable_cf_options,
+                                   const MutableDBOptions& mutable_db_options,
+                                   VersionStorageInfo* version,
+                                   LogBuffer* log_buffer);
 };
 }  // namespace ROCKSDB_NAMESPACE
 #endif  // !ROCKSDB_LITE
@@ -336,6 +336,7 @@ Compaction* LevelCompactionBuilder::GetCompaction() {
      GetCompressionType(ioptions_, vstorage_, mutable_cf_options_,
                         output_level_, vstorage_->base_level()),
      GetCompressionOptions(mutable_cf_options_, vstorage_, output_level_),
+      Temperature::kUnknown,
      /* max_subcompactions */ 0, std::move(grandparents_), is_manual_,
      start_level_score_, false /* deletion_compaction */, compaction_reason_);

@@ -384,7 +385,7 @@ uint32_t LevelCompactionBuilder::GetPathId(
          if (ioptions.level_compaction_dynamic_level_bytes) {
            // Currently, level_compaction_dynamic_level_bytes is ignored when
            // multiple db paths are specified. https://github.com/facebook/
-            // rocksdb/blob/master/db/column_family.cc.
+            // rocksdb/blob/main/db/column_family.cc.
            // Still, adding this check to avoid accidentally using
            // max_bytes_for_level_multiplier_additional
            level_size = static_cast<uint64_t>(
@@ -98,7 +98,9 @@ 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, bool marked_for_compact = false) {
+           size_t compensated_file_size = 0, bool marked_for_compact = false,
+           Temperature temperature = Temperature::kUnknown,
+           uint64_t oldest_ancestor_time = kUnknownOldestAncesterTime) {
    VersionStorageInfo* vstorage;
    if (temp_vstorage_) {
      vstorage = temp_vstorage_.get();
@@ -115,6 +117,8 @@ class CompactionPickerTest : public testing::Test {
        kUnknownFileChecksum, kUnknownFileChecksumFuncName);
    f->compensated_file_size =
        (compensated_file_size != 0) ? compensated_file_size : file_size;
+    f->temperature = temperature;
+    f->oldest_ancester_time = oldest_ancestor_time;
    vstorage->AddFile(level, f);
    files_.emplace_back(f);
    file_map_.insert({file_number, {f, level}});
@@ -757,6 +761,245 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {
              vstorage_->CompactionScore(0) >= 1);
  }
 }
+
+TEST_F(CompactionPickerTest, FIFOToWarm1) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 100;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kUnknown, threshold_time - 3000);
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  ASSERT_EQ(3U, compaction->input(0, 0)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, FIFOToWarm2) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 100;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kUnknown, threshold_time - 3000);
+  Add(0, 2U, "200", "300", 4 * kFileSize, 0, 2100, 2200, 0, true,
+      Temperature::kUnknown, threshold_time - 4000);
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_files(0));
+  ASSERT_EQ(2U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(3U, compaction->input(0, 1)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, FIFOToWarmMaxSize) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 9;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kUnknown, threshold_time - 3000);
+  Add(0, 2U, "200", "300", 4 * kFileSize, 0, 2100, 2200, 0, true,
+      Temperature::kUnknown, threshold_time - 4000);
+  Add(0, 1U, "200", "300", 4 * kFileSize, 0, 2000, 2100, 0, true,
+      Temperature::kUnknown, threshold_time - 5000);
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_files(0));
+  ASSERT_EQ(1U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(2U, compaction->input(0, 1)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, FIFOToWarmWithExistingWarm) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 100;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kUnknown, threshold_time - 3000);
+  Add(0, 2U, "200", "300", 4 * kFileSize, 0, 2100, 2200, 0, true,
+      Temperature::kUnknown, threshold_time - 4000);
+  Add(0, 1U, "200", "300", 4 * kFileSize, 0, 2000, 2100, 0, true,
+      Temperature::kWarm, threshold_time - 5000);
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(2U, compaction->num_input_files(0));
+  ASSERT_EQ(2U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(3U, compaction->input(0, 1)->fd.GetNumber());
+}
+
+TEST_F(CompactionPickerTest, FIFOToWarmWithOngoing) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 100;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kUnknown, threshold_time - 3000);
+  Add(0, 2U, "200", "300", 4 * kFileSize, 0, 2100, 2200, 0, true,
+      Temperature::kUnknown, threshold_time - 4000);
+  Add(0, 1U, "200", "300", 4 * kFileSize, 0, 2000, 2100, 0, true,
+      Temperature::kWarm, threshold_time - 5000);
+  file_map_[2].first->being_compacted = true;
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  // Stop if a file is being compacted
+  ASSERT_TRUE(compaction.get() == nullptr);
+}
+
+TEST_F(CompactionPickerTest, FIFOToWarmWithHotBetweenWarms) {
+  NewVersionStorage(1, kCompactionStyleFIFO);
+  const uint64_t kFileSize = 100000;
+  const uint64_t kMaxSize = kFileSize * 100000;
+  uint64_t kWarmThreshold = 2000;
+
+  fifo_options_.max_table_files_size = kMaxSize;
+  fifo_options_.age_for_warm = kWarmThreshold;
+  mutable_cf_options_.compaction_options_fifo = fifo_options_;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 2;
+  mutable_cf_options_.max_compaction_bytes = kFileSize * 100;
+  FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
+
+  int64_t current_time = 0;
+  ASSERT_OK(Env::Default()->GetCurrentTime(&current_time));
+  uint64_t threshold_time =
+      static_cast<uint64_t>(current_time) - kWarmThreshold;
+  Add(0, 6U, "240", "290", 2 * kFileSize, 0, 2900, 3000, 0, true,
+      Temperature::kUnknown, static_cast<uint64_t>(current_time) - 100);
+  Add(0, 5U, "240", "290", 2 * kFileSize, 0, 2700, 2800, 0, true,
+      Temperature::kUnknown, threshold_time + 100);
+  Add(0, 4U, "260", "300", 1 * kFileSize, 0, 2500, 2600, 0, true,
+      Temperature::kUnknown, threshold_time - 2000);
+  Add(0, 3U, "200", "300", 4 * kFileSize, 0, 2300, 2400, 0, true,
+      Temperature::kWarm, threshold_time - 3000);
+  Add(0, 2U, "200", "300", 4 * kFileSize, 0, 2100, 2200, 0, true,
+      Temperature::kUnknown, threshold_time - 4000);
+  Add(0, 1U, "200", "300", 4 * kFileSize, 0, 2000, 2100, 0, true,
+      Temperature::kWarm, threshold_time - 5000);
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), true);
+  std::unique_ptr<Compaction> compaction(fifo_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  // Stop if a file is being compacted
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  ASSERT_EQ(2U, compaction->input(0, 0)->fd.GetNumber());
+}
+
 #endif  // ROCKSDB_LITE

 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
@@ -157,9 +157,9 @@ struct SmallestKeyHeapComparator {
  const Comparator* ucmp_;
 };

-typedef std::priority_queue<InputFileInfo, std::vector<InputFileInfo>,
-                            SmallestKeyHeapComparator>
-    SmallestKeyHeap;
+using SmallestKeyHeap =
+    std::priority_queue<InputFileInfo, std::vector<InputFileInfo>,
+                        SmallestKeyHeapComparator>;

 // This function creates the heap that is used to find if the files are
 // overlapping during universal compaction when the allow_trivial_move
@@ -728,6 +728,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSortedRuns(
                         1, enable_compression),
      GetCompressionOptions(mutable_cf_options_, vstorage_, start_level,
                            enable_compression),
+      Temperature::kUnknown,
      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
      score_, false /* deletion_compaction */, compaction_reason);
 }
@@ -955,6 +956,7 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
      GetCompressionType(ioptions_, vstorage_, mutable_cf_options_,
                         output_level, 1),
      GetCompressionOptions(mutable_cf_options_, vstorage_, output_level),
+      Temperature::kUnknown,
      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
      score_, false /* deletion_compaction */,
      CompactionReason::kFilesMarkedForCompaction);
@@ -1029,6 +1031,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionToOldest(
                         output_level, 1, true /* enable_compression */),
      GetCompressionOptions(mutable_cf_options_, vstorage_, output_level,
                            true /* enable_compression */),
+      Temperature::kUnknown,
      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
      score_, false /* deletion_compaction */, compaction_reason);
 }
@@ -10,27 +10,70 @@

 namespace ROCKSDB_NAMESPACE {

-class MyTestCompactionService : public CompactionService {
+class TestCompactionServiceBase {
 public:
-  MyTestCompactionService(const std::string& db_path,
-                          std::shared_ptr<FileSystem> fs, Options& options)
-      : db_path_(db_path), fs_(fs), options_(options) {}
+  virtual int GetCompactionNum() = 0;

-  static const char* kClassName() { return "MyTestCompactionService"; }
+  void OverrideStartStatus(CompactionServiceJobStatus s) {
+    is_override_start_status = true;
+    override_start_status = s;
+  }
+
+  void OverrideWaitStatus(CompactionServiceJobStatus s) {
+    is_override_wait_status = true;
+    override_wait_status = s;
+  }
+
+  void OverrideWaitResult(std::string str) {
+    is_override_wait_result = true;
+    override_wait_result = std::move(str);
+  }
+
+  void ResetOverride() {
+    is_override_wait_result = false;
+    is_override_start_status = false;
+    is_override_wait_status = false;
+  }
+
+  virtual ~TestCompactionServiceBase() = default;
+
+ protected:
+  bool is_override_start_status = false;
+  CompactionServiceJobStatus override_start_status =
+      CompactionServiceJobStatus::kFailure;
+  bool is_override_wait_status = false;
+  CompactionServiceJobStatus override_wait_status =
+      CompactionServiceJobStatus::kFailure;
+  bool is_override_wait_result = false;
+  std::string override_wait_result;
+};
+
+class MyTestCompactionServiceLegacy : public CompactionService,
+                                      public TestCompactionServiceBase {
+ public:
+  MyTestCompactionServiceLegacy(std::string db_path, Options& options,
+                                std::shared_ptr<Statistics>& statistics)
+      : db_path_(std::move(db_path)),
+        options_(options),
+        statistics_(statistics) {}
+
+  static const char* kClassName() { return "MyTestCompactionServiceLegacy"; }

  const char* Name() const override { return kClassName(); }

  CompactionServiceJobStatus Start(const std::string& compaction_service_input,
-                                   int job_id) override {
+                                   uint64_t job_id) override {
    InstrumentedMutexLock l(&mutex_);
    jobs_.emplace(job_id, compaction_service_input);
    CompactionServiceJobStatus s = CompactionServiceJobStatus::kSuccess;
-    TEST_SYNC_POINT_CALLBACK("MyTestCompactionService::Start::End", &s);
+    if (is_override_start_status) {
+      return override_start_status;
+    }
    return s;
  }

  CompactionServiceJobStatus WaitForComplete(
-      int job_id, std::string* compaction_service_result) override {
+      uint64_t job_id, std::string* compaction_service_result) override {
    std::string compaction_input;
    {
      InstrumentedMutexLock l(&mutex_);
@@ -42,6 +85,10 @@ class MyTestCompactionService : public CompactionService {
      jobs_.erase(i);
    }

+    if (is_override_wait_status) {
+      return override_wait_status;
+    }
+
    CompactionServiceOptionsOverride options_override;
    options_override.env = options_.env;
    options_override.file_checksum_gen_factory =
@@ -54,12 +101,14 @@ class MyTestCompactionService : public CompactionService {
    options_override.prefix_extractor = options_.prefix_extractor;
    options_override.table_factory = options_.table_factory;
    options_override.sst_partitioner_factory = options_.sst_partitioner_factory;
+    options_override.statistics = statistics_;

    Status s = DB::OpenAndCompact(
        db_path_, db_path_ + "/" + ROCKSDB_NAMESPACE::ToString(job_id),
        compaction_input, compaction_service_result, options_override);
-    TEST_SYNC_POINT_CALLBACK("MyTestCompactionService::WaitForComplete::End",
-                             compaction_service_result);
+    if (is_override_wait_result) {
+      *compaction_service_result = override_wait_result;
+    }
    compaction_num_.fetch_add(1);
    if (s.ok()) {
      return CompactionServiceJobStatus::kSuccess;
@@ -68,23 +117,155 @@ class MyTestCompactionService : public CompactionService {
    }
  }

-  int GetCompactionNum() { return compaction_num_.load(); }
+  int GetCompactionNum() override { return compaction_num_.load(); }

 private:
  InstrumentedMutex mutex_;
  std::atomic_int compaction_num_{0};
-  std::map<int, std::string> jobs_;
+  std::map<uint64_t, std::string> jobs_;
  const std::string db_path_;
-  std::shared_ptr<FileSystem> fs_;
  Options options_;
+  std::shared_ptr<Statistics> statistics_;
 };

-class CompactionServiceTest : public DBTestBase {
+class MyTestCompactionService : public CompactionService,
+                                public TestCompactionServiceBase {
+ public:
+  MyTestCompactionService(std::string db_path, Options& options,
+                          std::shared_ptr<Statistics>& statistics)
+      : db_path_(std::move(db_path)),
+        options_(options),
+        statistics_(statistics),
+        start_info_("na", "na", "na", 0, Env::TOTAL),
+        wait_info_("na", "na", "na", 0, Env::TOTAL) {}
+
+  static const char* kClassName() { return "MyTestCompactionService"; }
+
+  const char* Name() const override { return kClassName(); }
+
+  CompactionServiceJobStatus StartV2(
+      const CompactionServiceJobInfo& info,
+      const std::string& compaction_service_input) override {
+    InstrumentedMutexLock l(&mutex_);
+    start_info_ = info;
+    assert(info.db_name == db_path_);
+    jobs_.emplace(info.job_id, compaction_service_input);
+    CompactionServiceJobStatus s = CompactionServiceJobStatus::kSuccess;
+    if (is_override_start_status) {
+      return override_start_status;
+    }
+    return s;
+  }
+
+  CompactionServiceJobStatus WaitForCompleteV2(
+      const CompactionServiceJobInfo& info,
+      std::string* compaction_service_result) override {
+    std::string compaction_input;
+    assert(info.db_name == db_path_);
+    {
+      InstrumentedMutexLock l(&mutex_);
+      wait_info_ = info;
+      auto i = jobs_.find(info.job_id);
+      if (i == jobs_.end()) {
+        return CompactionServiceJobStatus::kFailure;
+      }
+      compaction_input = std::move(i->second);
+      jobs_.erase(i);
+    }
+
+    if (is_override_wait_status) {
+      return override_wait_status;
+    }
+
+    CompactionServiceOptionsOverride options_override;
+    options_override.env = options_.env;
+    options_override.file_checksum_gen_factory =
+        options_.file_checksum_gen_factory;
+    options_override.comparator = options_.comparator;
+    options_override.merge_operator = options_.merge_operator;
+    options_override.compaction_filter = options_.compaction_filter;
+    options_override.compaction_filter_factory =
+        options_.compaction_filter_factory;
+    options_override.prefix_extractor = options_.prefix_extractor;
+    options_override.table_factory = options_.table_factory;
+    options_override.sst_partitioner_factory = options_.sst_partitioner_factory;
+    options_override.statistics = statistics_;
+
+    Status s = DB::OpenAndCompact(
+        db_path_, db_path_ + "/" + ROCKSDB_NAMESPACE::ToString(info.job_id),
+        compaction_input, compaction_service_result, options_override);
+    if (is_override_wait_result) {
+      *compaction_service_result = override_wait_result;
+    }
+    compaction_num_.fetch_add(1);
+    if (s.ok()) {
+      return CompactionServiceJobStatus::kSuccess;
+    } else {
+      return CompactionServiceJobStatus::kFailure;
+    }
+  }
+
+  int GetCompactionNum() override { return compaction_num_.load(); }
+
+  CompactionServiceJobInfo GetCompactionInfoForStart() { return start_info_; }
+  CompactionServiceJobInfo GetCompactionInfoForWait() { return wait_info_; }
+
+ private:
+  InstrumentedMutex mutex_;
+  std::atomic_int compaction_num_{0};
+  std::map<uint64_t, std::string> jobs_;
+  const std::string db_path_;
+  Options options_;
+  std::shared_ptr<Statistics> statistics_;
+  CompactionServiceJobInfo start_info_;
+  CompactionServiceJobInfo wait_info_;
+};
+
+// This is only for listing test classes
+enum TestCompactionServiceType {
+  MyTestCompactionServiceType,
+  MyTestCompactionServiceLegacyType,
+};
+
+class CompactionServiceTest
+    : public DBTestBase,
+      public testing::WithParamInterface<TestCompactionServiceType> {
 public:
  explicit CompactionServiceTest()
      : DBTestBase("compaction_service_test", true) {}

 protected:
+  void ReopenWithCompactionService(Options* options) {
+    options->env = env_;
+    primary_statistics_ = CreateDBStatistics();
+    options->statistics = primary_statistics_;
+    compactor_statistics_ = CreateDBStatistics();
+    TestCompactionServiceType cs_type = GetParam();
+    switch (cs_type) {
+      case MyTestCompactionServiceType:
+        compaction_service_ = std::make_shared<MyTestCompactionService>(
+            dbname_, *options, compactor_statistics_);
+        break;
+      case MyTestCompactionServiceLegacyType:
+        compaction_service_ = std::make_shared<MyTestCompactionServiceLegacy>(
+            dbname_, *options, compactor_statistics_);
+        break;
+      default:
+        assert(false);
+    }
+    options->compaction_service = compaction_service_;
+    DestroyAndReopen(*options);
+  }
+
+  Statistics* GetCompactorStatistics() { return compactor_statistics_.get(); }
+
+  Statistics* GetPrimaryStatistics() { return primary_statistics_.get(); }
+
+  TestCompactionServiceBase* GetCompactionService() {
+    CompactionService* cs = compaction_service_.get();
+    return dynamic_cast<TestCompactionServiceBase*>(cs);
+  }
+
  void GenerateTestData() {
    // Generate 20 files @ L2
    for (int i = 0; i < 20; i++) {
@@ -118,15 +299,16 @@ class CompactionServiceTest : public DBTestBase {
      }
    }
  }
+
+ private:
+  std::shared_ptr<Statistics> compactor_statistics_;
+  std::shared_ptr<Statistics> primary_statistics_;
+  std::shared_ptr<CompactionService> compaction_service_;
 };

-TEST_F(CompactionServiceTest, BasicCompactions) {
+TEST_P(CompactionServiceTest, BasicCompactions) {
  Options options = CurrentOptions();
-  options.env = env_;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);

  for (int i = 0; i < 20; i++) {
    for (int j = 0; j < 10; j++) {
@@ -154,15 +336,22 @@ TEST_F(CompactionServiceTest, BasicCompactions) {
      ASSERT_EQ(result, "value_new" + ToString(i));
    }
  }
-  auto my_cs =
-      dynamic_cast<MyTestCompactionService*>(options.compaction_service.get());
+  auto my_cs = GetCompactionService();
+  Statistics* compactor_statistics = GetCompactorStatistics();
  ASSERT_GE(my_cs->GetCompactionNum(), 1);

+  // make sure the compaction statistics is only recorded on the remote side
+  ASSERT_GE(
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY), 1);
+  Statistics* primary_statistics = GetPrimaryStatistics();
+  ASSERT_EQ(primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+            0);
+
  // Test failed compaction
  SyncPoint::GetInstance()->SetCallBack(
      "DBImplSecondary::CompactWithoutInstallation::End", [&](void* status) {
        // override job status
-        Status* s = static_cast<Status*>(status);
+        auto s = static_cast<Status*>(status);
        *s = Status::Aborted("MyTestCompactionService failed to compact!");
      });
  SyncPoint::GetInstance()->EnableProcessing();
@@ -191,17 +380,13 @@ TEST_F(CompactionServiceTest, BasicCompactions) {
  ASSERT_TRUE(s.IsAborted());
 }

-TEST_F(CompactionServiceTest, ManualCompaction) {
+TEST_P(CompactionServiceTest, ManualCompaction) {
  Options options = CurrentOptions();
-  options.env = env_;
  options.disable_auto_compactions = true;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);
  GenerateTestData();

-  auto my_cs =
-      dynamic_cast<MyTestCompactionService*>(options.compaction_service.get());
+  auto my_cs = GetCompactionService();

  std::string start_str = Key(15);
  std::string end_str = Key(45);
@@ -232,22 +417,15 @@ TEST_F(CompactionServiceTest, ManualCompaction) {
  VerifyTestData();
 }

-TEST_F(CompactionServiceTest, FailedToStart) {
+TEST_P(CompactionServiceTest, FailedToStart) {
  Options options = CurrentOptions();
-  options.env = env_;
  options.disable_auto_compactions = true;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);
+
  GenerateTestData();

-  SyncPoint::GetInstance()->SetCallBack(
-      "MyTestCompactionService::Start::End", [&](void* status) {
-        // override job status
-        auto s = static_cast<CompactionServiceJobStatus*>(status);
-        *s = CompactionServiceJobStatus::kFailure;
-      });
-  SyncPoint::GetInstance()->EnableProcessing();
+  auto my_cs = GetCompactionService();
+  my_cs->OverrideStartStatus(CompactionServiceJobStatus::kFailure);

  std::string start_str = Key(15);
  std::string end_str = Key(45);
@@ -257,22 +435,15 @@ TEST_F(CompactionServiceTest, FailedToStart) {
  ASSERT_TRUE(s.IsIncomplete());
 }

-TEST_F(CompactionServiceTest, InvalidResult) {
+TEST_P(CompactionServiceTest, InvalidResult) {
  Options options = CurrentOptions();
-  options.env = env_;
  options.disable_auto_compactions = true;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);
+
  GenerateTestData();

-  SyncPoint::GetInstance()->SetCallBack(
-      "MyTestCompactionService::WaitForComplete::End", [&](void* result) {
-        // override job status
-        auto result_str = static_cast<std::string*>(result);
-        *result_str = "Invalid Str";
-      });
-  SyncPoint::GetInstance()->EnableProcessing();
+  auto my_cs = GetCompactionService();
+  my_cs->OverrideWaitResult("Invalid Str");

  std::string start_str = Key(15);
  std::string end_str = Key(45);
@@ -282,22 +453,27 @@ TEST_F(CompactionServiceTest, InvalidResult) {
  ASSERT_FALSE(s.ok());
 }

-// TODO: support sub-compaction
-TEST_F(CompactionServiceTest, DISABLED_SubCompaction) {
+TEST_P(CompactionServiceTest, SubCompaction) {
  Options options = CurrentOptions();
-  options.env = env_;
  options.max_subcompactions = 10;
  options.target_file_size_base = 1 << 10;  // 1KB
  options.disable_auto_compactions = true;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
+  ReopenWithCompactionService(&options);

-  DestroyAndReopen(options);
  GenerateTestData();
+  VerifyTestData();
+
+  auto my_cs = GetCompactionService();
+  int compaction_num_before = my_cs->GetCompactionNum();

  auto cro = CompactRangeOptions();
  cro.max_subcompactions = 10;
-  db_->CompactRange(cro, nullptr, nullptr);
+  Status s = db_->CompactRange(cro, nullptr, nullptr);
+  ASSERT_OK(s);
+  VerifyTestData();
+  int compaction_num = my_cs->GetCompactionNum() - compaction_num_before;
+  // make sure there's sub-compaction by checking the compaction number
+  ASSERT_GE(compaction_num, 2);
 }

 class PartialDeleteCompactionFilter : public CompactionFilter {
@@ -316,15 +492,12 @@ class PartialDeleteCompactionFilter : public CompactionFilter {
  const char* Name() const override { return "PartialDeleteCompactionFilter"; }
 };

-TEST_F(CompactionServiceTest, CompactionFilter) {
+TEST_P(CompactionServiceTest, CompactionFilter) {
  Options options = CurrentOptions();
-  options.env = env_;
-  auto delete_comp_filter = PartialDeleteCompactionFilter();
-  options.compaction_filter = &delete_comp_filter;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-
-  DestroyAndReopen(options);
+  std::unique_ptr<CompactionFilter> delete_comp_filter(
+      new PartialDeleteCompactionFilter());
+  options.compaction_filter = delete_comp_filter.get();
+  ReopenWithCompactionService(&options);

  for (int i = 0; i < 20; i++) {
    for (int j = 0; j < 10; j++) {
@@ -356,18 +529,13 @@ TEST_F(CompactionServiceTest, CompactionFilter) {
      ASSERT_EQ(result, "value_new" + ToString(i));
    }
  }
-  auto my_cs =
-      dynamic_cast<MyTestCompactionService*>(options.compaction_service.get());
+  auto my_cs = GetCompactionService();
  ASSERT_GE(my_cs->GetCompactionNum(), 1);
 }

-TEST_F(CompactionServiceTest, Snapshot) {
+TEST_P(CompactionServiceTest, Snapshot) {
  Options options = CurrentOptions();
-  options.env = env_;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
-
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);

  ASSERT_OK(Put(Key(1), "value1"));
  ASSERT_OK(Put(Key(2), "value1"));
@@ -379,23 +547,18 @@ TEST_F(CompactionServiceTest, Snapshot) {
  ASSERT_OK(Flush());

  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  auto my_cs =
-      dynamic_cast<MyTestCompactionService*>(options.compaction_service.get());
+  auto my_cs = GetCompactionService();
  ASSERT_GE(my_cs->GetCompactionNum(), 1);
  ASSERT_EQ("value1", Get(Key(1), s1));
  ASSERT_EQ("value2", Get(Key(1)));
  db_->ReleaseSnapshot(s1);
 }

-TEST_F(CompactionServiceTest, ConcurrentCompaction) {
+TEST_P(CompactionServiceTest, ConcurrentCompaction) {
  Options options = CurrentOptions();
  options.level0_file_num_compaction_trigger = 100;
-  options.env = env_;
-  options.compaction_service = std::make_shared<MyTestCompactionService>(
-      dbname_, env_->GetFileSystem(), options);
  options.max_background_jobs = 20;
-
-  DestroyAndReopen(options);
+  ReopenWithCompactionService(&options);
  GenerateTestData();

  ColumnFamilyMetaData meta;
@@ -423,12 +586,209 @@ TEST_F(CompactionServiceTest, ConcurrentCompaction) {
      ASSERT_EQ(result, "value_new" + ToString(i));
    }
  }
-  auto my_cs =
-      dynamic_cast<MyTestCompactionService*>(options.compaction_service.get());
+  auto my_cs = GetCompactionService();
  ASSERT_EQ(my_cs->GetCompactionNum(), 10);
  ASSERT_EQ(FilesPerLevel(), "0,0,10");
 }

+TEST_P(CompactionServiceTest, CompactionInfo) {
+  // only test compaction info for new compaction service interface
+  if (GetParam() != MyTestCompactionServiceType) {
+    return;
+  }
+
+  Options options = CurrentOptions();
+  ReopenWithCompactionService(&options);
+
+  for (int i = 0; i < 20; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 10 + j;
+      ASSERT_OK(Put(Key(key_id), "value" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+
+  for (int i = 0; i < 10; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 20 + j * 2;
+      ASSERT_OK(Put(Key(key_id), "value_new" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  auto my_cs =
+      static_cast_with_check<MyTestCompactionService>(GetCompactionService());
+  uint64_t comp_num = my_cs->GetCompactionNum();
+  ASSERT_GE(comp_num, 1);
+
+  CompactionServiceJobInfo info = my_cs->GetCompactionInfoForStart();
+  ASSERT_EQ(dbname_, info.db_name);
+  std::string db_id, db_session_id;
+  ASSERT_OK(db_->GetDbIdentity(db_id));
+  ASSERT_EQ(db_id, info.db_id);
+  ASSERT_OK(db_->GetDbSessionId(db_session_id));
+  ASSERT_EQ(db_session_id, info.db_session_id);
+  ASSERT_EQ(Env::LOW, info.priority);
+  info = my_cs->GetCompactionInfoForWait();
+  ASSERT_EQ(dbname_, info.db_name);
+  ASSERT_EQ(db_id, info.db_id);
+  ASSERT_EQ(db_session_id, info.db_session_id);
+  ASSERT_EQ(Env::LOW, info.priority);
+
+  // Test priority USER
+  ColumnFamilyMetaData meta;
+  db_->GetColumnFamilyMetaData(&meta);
+  SstFileMetaData file = meta.levels[1].files[0];
+  ASSERT_OK(db_->CompactFiles(CompactionOptions(),
+                              {file.db_path + "/" + file.name}, 2));
+  info = my_cs->GetCompactionInfoForStart();
+  ASSERT_EQ(Env::USER, info.priority);
+  info = my_cs->GetCompactionInfoForWait();
+  ASSERT_EQ(Env::USER, info.priority);
+
+  // Test priority BOTTOM
+  env_->SetBackgroundThreads(1, Env::BOTTOM);
+  options.num_levels = 2;
+  ReopenWithCompactionService(&options);
+  my_cs =
+      static_cast_with_check<MyTestCompactionService>(GetCompactionService());
+
+  for (int i = 0; i < 20; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 10 + j;
+      ASSERT_OK(Put(Key(key_id), "value" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+
+  for (int i = 0; i < 4; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 20 + j * 2;
+      ASSERT_OK(Put(Key(key_id), "value_new" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  info = my_cs->GetCompactionInfoForStart();
+  ASSERT_EQ(Env::BOTTOM, info.priority);
+  info = my_cs->GetCompactionInfoForWait();
+  ASSERT_EQ(Env::BOTTOM, info.priority);
+}
+
+TEST_P(CompactionServiceTest, FallbackLocalAuto) {
+  Options options = CurrentOptions();
+  ReopenWithCompactionService(&options);
+
+  auto my_cs = GetCompactionService();
+  Statistics* compactor_statistics = GetCompactorStatistics();
+  Statistics* primary_statistics = GetPrimaryStatistics();
+  uint64_t compactor_new_key =
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+  uint64_t primary_new_key =
+      primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+
+  my_cs->OverrideStartStatus(CompactionServiceJobStatus::kUseLocal);
+
+  for (int i = 0; i < 20; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 10 + j;
+      ASSERT_OK(Put(Key(key_id), "value" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+
+  for (int i = 0; i < 10; i++) {
+    for (int j = 0; j < 10; j++) {
+      int key_id = i * 20 + j * 2;
+      ASSERT_OK(Put(Key(key_id), "value_new" + ToString(key_id)));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // verify result
+  for (int i = 0; i < 200; i++) {
+    auto result = Get(Key(i));
+    if (i % 2) {
+      ASSERT_EQ(result, "value" + ToString(i));
+    } else {
+      ASSERT_EQ(result, "value_new" + ToString(i));
+    }
+  }
+
+  ASSERT_EQ(my_cs->GetCompactionNum(), 0);
+
+  // make sure the compaction statistics is only recorded on the local side
+  ASSERT_EQ(
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+      compactor_new_key);
+  ASSERT_GT(primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+            primary_new_key);
+}
+
+TEST_P(CompactionServiceTest, FallbackLocalManual) {
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  ReopenWithCompactionService(&options);
+
+  GenerateTestData();
+  VerifyTestData();
+
+  auto my_cs = GetCompactionService();
+  Statistics* compactor_statistics = GetCompactorStatistics();
+  Statistics* primary_statistics = GetPrimaryStatistics();
+  uint64_t compactor_new_key =
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+  uint64_t primary_new_key =
+      primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+
+  // re-enable remote compaction
+  my_cs->ResetOverride();
+  std::string start_str = Key(15);
+  std::string end_str = Key(45);
+  Slice start(start_str);
+  Slice end(end_str);
+  uint64_t comp_num = my_cs->GetCompactionNum();
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &start, &end));
+  ASSERT_GE(my_cs->GetCompactionNum(), comp_num + 1);
+  // make sure the compaction statistics is only recorded on the remote side
+  ASSERT_GT(
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+      compactor_new_key);
+  ASSERT_EQ(primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+            primary_new_key);
+
+  // return run local again with API WaitForComplete
+  my_cs->OverrideWaitStatus(CompactionServiceJobStatus::kUseLocal);
+  start_str = Key(120);
+  start = start_str;
+  comp_num = my_cs->GetCompactionNum();
+  compactor_new_key =
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+  primary_new_key =
+      primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY);
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &start, nullptr));
+  ASSERT_EQ(my_cs->GetCompactionNum(),
+            comp_num);  // no remote compaction is run
+  // make sure the compaction statistics is only recorded on the local side
+  ASSERT_EQ(
+      compactor_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+      compactor_new_key);
+  ASSERT_GT(primary_statistics->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
+            primary_new_key);
+
+  // verify result after 2 manual compactions
+  VerifyTestData();
+}
+
+INSTANTIATE_TEST_CASE_P(
+    CompactionServiceTest, CompactionServiceTest,
+    ::testing::Values(
+        TestCompactionServiceType::MyTestCompactionServiceType,
+        TestCompactionServiceType::MyTestCompactionServiceLegacyType));
+
 }  // namespace ROCKSDB_NAMESPACE

 #ifdef ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS
@@ -94,8 +94,8 @@ class CuckooTableDBTest : public testing::Test {

  int NumTableFilesAtLevel(int level) {
    std::string property;
-    EXPECT_TRUE(db_->GetProperty(
-        "rocksdb.num-files-at-level" + NumberToString(level), &property));
+    EXPECT_TRUE(db_->GetProperty("rocksdb.num-files-at-level" + ToString(level),
+                                 &property));
    return atoi(property.c_str());
  }

@@ -8,7 +8,6 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include <cstring>
-#include <regex>

 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
@@ -31,7 +30,7 @@ namespace ROCKSDB_NAMESPACE {

 class DBBasicTest : public DBTestBase {
 public:
-  DBBasicTest() : DBTestBase("/db_basic_test", /*env_do_fsync=*/false) {}
+  DBBasicTest() : DBTestBase("db_basic_test", /*env_do_fsync=*/false) {}
 };

 TEST_F(DBBasicTest, OpenWhenOpen) {
@@ -69,11 +68,10 @@ TEST_F(DBBasicTest, UniqueSession) {
  ASSERT_EQ(sid2, sid4);

  // Expected compact format for session ids (see notes in implementation)
-  std::regex expected("[0-9A-Z]{20}");
-  const std::string match("match");
-  EXPECT_EQ(match, std::regex_replace(sid1, expected, match));
-  EXPECT_EQ(match, std::regex_replace(sid2, expected, match));
-  EXPECT_EQ(match, std::regex_replace(sid3, expected, match));
+  TestRegex expected("[0-9A-Z]{20}");
+  EXPECT_MATCHES_REGEX(sid1, expected);
+  EXPECT_MATCHES_REGEX(sid2, expected);
+  EXPECT_MATCHES_REGEX(sid3, expected);

 #ifndef ROCKSDB_LITE
  Close();
@@ -1362,6 +1360,28 @@ TEST_P(DBMultiGetTestWithParam, MultiGetMultiCFSnapshot) {
  }
 }

+TEST_P(DBMultiGetTestWithParam, MultiGetMultiCFUnsorted) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"one", "two"}, options);
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(2, "baz", "xyz"));
+  ASSERT_OK(Put(1, "abc", "def"));
+
+  // Note: keys for the same CF do not form a consecutive range
+  std::vector<int> cfs{1, 2, 1};
+  std::vector<std::string> keys{"foo", "baz", "abc"};
+  std::vector<std::string> values;
+
+  values =
+      MultiGet(cfs, keys, /* snapshot */ nullptr, /* batched */ GetParam());
+
+  ASSERT_EQ(values.size(), 3);
+  ASSERT_EQ(values[0], "bar");
+  ASSERT_EQ(values[1], "xyz");
+  ASSERT_EQ(values[2], "def");
+}
+
 INSTANTIATE_TEST_CASE_P(DBMultiGetTestWithParam, DBMultiGetTestWithParam,
                        testing::Bool());

@@ -2576,7 +2596,7 @@ class DBBasicTestTrackWal : public DBTestBase,
                            public testing::WithParamInterface<bool> {
 public:
  DBBasicTestTrackWal()
-      : DBTestBase("/db_basic_test_track_wal", /*env_do_fsync=*/false) {}
+      : DBTestBase("db_basic_test_track_wal", /*env_do_fsync=*/false) {}

  int CountWalFiles() {
    VectorLogPtr log_files;
@@ -2734,6 +2754,11 @@ class DBBasicTestMultiGet : public DBTestBase {
        EXPECT_OK(dbfull()->Flush(FlushOptions(), handles_[cf]));
      }
    }
+    // Clear compressed cache, which is always pre-populated
+    if (compressed_cache_) {
+      compressed_cache_->SetCapacity(0);
+      compressed_cache_->SetCapacity(1048576);
+    }
  }

  bool CheckValue(int i, const std::string& value) {
@@ -3210,13 +3235,11 @@ INSTANTIATE_TEST_CASE_P(ParallelIO, DBBasicTestWithParallelIO,
 // Forward declaration
 class DeadlineFS;

-class DeadlineRandomAccessFile : public FSRandomAccessFileWrapper {
+class DeadlineRandomAccessFile : public FSRandomAccessFileOwnerWrapper {
 public:
  DeadlineRandomAccessFile(DeadlineFS& fs,
                           std::unique_ptr<FSRandomAccessFile>& file)
-      : FSRandomAccessFileWrapper(file.get()),
-        fs_(fs),
-        file_(std::move(file)) {}
+      : FSRandomAccessFileOwnerWrapper(std::move(file)), fs_(fs) {}

  IOStatus Read(uint64_t offset, size_t len, const IOOptions& opts,
                Slice* result, char* scratch,
@@ -7,6 +7,7 @@
 // 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 <cstdlib>
+#include <functional>
 #include <memory>

 #include "cache/cache_entry_roles.h"
@@ -14,9 +15,12 @@
 #include "db/column_family.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "rocksdb/statistics.h"
 #include "rocksdb/table.h"
 #include "util/compression.h"
+#include "util/defer.h"
 #include "util/random.h"
+#include "utilities/fault_injection_fs.h"

 namespace ROCKSDB_NAMESPACE {

@@ -39,7 +43,7 @@ class DBBlockCacheTest : public DBTestBase {
  const size_t kValueSize = 100;

  DBBlockCacheTest()
-      : DBTestBase("/db_block_cache_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_block_cache_test", /*env_do_fsync=*/true) {}

  BlockBasedTableOptions GetTableOptions() {
    BlockBasedTableOptions table_options;
@@ -320,8 +324,8 @@ TEST_F(DBBlockCacheTest, TestWithCompressedBlockCache) {
  // Load last key block.
  ASSERT_EQ("Result incomplete: Insert failed due to LRU cache being full.",
            Get(ToString(kNumBlocks - 1)));
-  // Failure won't record the miss counter.
-  CheckCacheCounters(options, 0, 0, 0, 1);
+  // Failure will also record the miss counter.
+  CheckCacheCounters(options, 1, 0, 0, 1);
  CheckCompressedCacheCounters(options, 1, 0, 1, 0);

  // Clear strict capacity limit flag. This time we shall hit compressed block
@@ -497,11 +501,114 @@ TEST_F(DBBlockCacheTest, WarmCacheWithDataBlocksDuringFlush) {
    ASSERT_OK(Put(ToString(i), value));
    ASSERT_OK(Flush());
    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
-
    ASSERT_EQ(value, Get(ToString(i)));
    ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_DATA_MISS));
    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_HIT));
  }
+  // Verify compaction not counted
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                              /*end=*/nullptr));
+  EXPECT_EQ(kNumBlocks,
+            options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
+}
+
+// This test cache data, index and filter blocks during flush.
+TEST_F(DBBlockCacheTest, WarmCacheWithBlocksDuringFlush) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
+
+  BlockBasedTableOptions table_options;
+  table_options.block_cache = NewLRUCache(1 << 25, 0, false);
+  table_options.cache_index_and_filter_blocks = true;
+  table_options.prepopulate_block_cache =
+      BlockBasedTableOptions::PrepopulateBlockCache::kFlushOnly;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  std::string value(kValueSize, 'a');
+  for (size_t i = 1; i <= kNumBlocks; i++) {
+    ASSERT_OK(Put(ToString(i), value));
+    ASSERT_OK(Flush());
+    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
+    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_ADD));
+    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_FILTER_ADD));
+
+    ASSERT_EQ(value, Get(ToString(i)));
+
+    ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_DATA_MISS));
+    ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_HIT));
+
+    ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(i * 3, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_HIT));
+
+    ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(i * 2,
+              options.statistics->getTickerCount(BLOCK_CACHE_FILTER_HIT));
+  }
+  // Verify compaction not counted
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                              /*end=*/nullptr));
+  EXPECT_EQ(kNumBlocks,
+            options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
+  // Index and filter blocks are automatically warmed when the new table file
+  // is automatically opened at the end of compaction. This is not easily
+  // disabled so results in the new index and filter blocks being warmed.
+  EXPECT_EQ(1 + kNumBlocks,
+            options.statistics->getTickerCount(BLOCK_CACHE_INDEX_ADD));
+  EXPECT_EQ(1 + kNumBlocks,
+            options.statistics->getTickerCount(BLOCK_CACHE_FILTER_ADD));
+}
+
+TEST_F(DBBlockCacheTest, DynamicallyWarmCacheDuringFlush) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
+
+  BlockBasedTableOptions table_options;
+  table_options.block_cache = NewLRUCache(1 << 25, 0, false);
+  table_options.cache_index_and_filter_blocks = false;
+  table_options.prepopulate_block_cache =
+      BlockBasedTableOptions::PrepopulateBlockCache::kFlushOnly;
+
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  std::string value(kValueSize, 'a');
+
+  for (size_t i = 1; i <= 5; i++) {
+    ASSERT_OK(Put(ToString(i), value));
+    ASSERT_OK(Flush());
+    ASSERT_EQ(1,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
+
+    ASSERT_EQ(value, Get(ToString(i)));
+    ASSERT_EQ(0,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
+    ASSERT_EQ(
+        0, options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_MISS));
+    ASSERT_EQ(1,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_HIT));
+  }
+
+  ASSERT_OK(dbfull()->SetOptions(
+      {{"block_based_table_factory", "{prepopulate_block_cache=kDisable;}"}}));
+
+  for (size_t i = 6; i <= kNumBlocks; i++) {
+    ASSERT_OK(Put(ToString(i), value));
+    ASSERT_OK(Flush());
+    ASSERT_EQ(0,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
+
+    ASSERT_EQ(value, Get(ToString(i)));
+    ASSERT_EQ(1,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
+    ASSERT_EQ(
+        1, options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_MISS));
+    ASSERT_EQ(0,
+              options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_HIT));
+  }
 }
 #endif

@@ -1212,13 +1319,238 @@ TEST_F(DBBlockCacheTest, CacheEntryRoleStats) {

 #endif  // ROCKSDB_LITE

+class DBBlockCacheKeyTest
+    : public DBTestBase,
+      public testing::WithParamInterface<std::tuple<bool, bool>> {
+ public:
+  DBBlockCacheKeyTest()
+      : DBTestBase("db_block_cache_test", /*env_do_fsync=*/false) {}
+
+  void SetUp() override {
+    use_compressed_cache_ = std::get<0>(GetParam());
+    exclude_file_numbers_ = std::get<1>(GetParam());
+  }
+
+  bool use_compressed_cache_;
+  bool exclude_file_numbers_;
+};
+
+// Disable LinkFile so that we can physically copy a DB using Checkpoint.
+// Disable file GetUniqueId to enable stable cache keys.
+class StableCacheKeyTestFS : public FaultInjectionTestFS {
+ public:
+  explicit StableCacheKeyTestFS(const std::shared_ptr<FileSystem>& base)
+      : FaultInjectionTestFS(base) {
+    SetFailGetUniqueId(true);
+  }
+
+  virtual ~StableCacheKeyTestFS() override {}
+
+  IOStatus LinkFile(const std::string&, const std::string&, const IOOptions&,
+                    IODebugContext*) override {
+    return IOStatus::NotSupported("Disabled");
+  }
+};
+
+TEST_P(DBBlockCacheKeyTest, StableCacheKeys) {
+  std::shared_ptr<StableCacheKeyTestFS> test_fs{
+      new StableCacheKeyTestFS(env_->GetFileSystem())};
+  std::unique_ptr<CompositeEnvWrapper> test_env{
+      new CompositeEnvWrapper(env_, test_fs)};
+
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
+  options.env = test_env.get();
+
+  BlockBasedTableOptions table_options;
+
+  int key_count = 0;
+  uint64_t expected_stat = 0;
+
+  std::function<void()> verify_stats;
+  if (use_compressed_cache_) {
+    if (!Snappy_Supported()) {
+      ROCKSDB_GTEST_SKIP("Compressed cache test requires snappy support");
+      return;
+    }
+    options.compression = CompressionType::kSnappyCompression;
+    table_options.no_block_cache = true;
+    table_options.block_cache_compressed = NewLRUCache(1 << 25, 0, false);
+    verify_stats = [&options, &expected_stat] {
+      // One for ordinary SST file and one for external SST file
+      ASSERT_EQ(expected_stat,
+                options.statistics->getTickerCount(BLOCK_CACHE_COMPRESSED_ADD));
+    };
+  } else {
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.block_cache = NewLRUCache(1 << 25, 0, false);
+    verify_stats = [&options, &expected_stat] {
+      ASSERT_EQ(expected_stat,
+                options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
+      ASSERT_EQ(expected_stat,
+                options.statistics->getTickerCount(BLOCK_CACHE_INDEX_ADD));
+      ASSERT_EQ(expected_stat,
+                options.statistics->getTickerCount(BLOCK_CACHE_FILTER_ADD));
+    };
+  }
+
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"koko"}, options);
+
+  if (exclude_file_numbers_) {
+    // Simulate something like old behavior without file numbers in properties.
+    // This is a "control" side of the test that also ensures safely degraded
+    // behavior on old files.
+    ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+        "PropertyBlockBuilder::AddTableProperty:Start", [&](void* arg) {
+          TableProperties* props = reinterpret_cast<TableProperties*>(arg);
+          props->orig_file_number = 0;
+        });
+    ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+  }
+
+  std::function<void()> perform_gets = [&key_count, &expected_stat, this]() {
+    if (exclude_file_numbers_) {
+      // No cache key reuse should happen, because we can't rely on current
+      // file number being stable
+      expected_stat += key_count;
+    } else {
+      // Cache keys should be stable
+      expected_stat = key_count;
+    }
+    for (int i = 0; i < key_count; ++i) {
+      ASSERT_EQ(Get(1, Key(i)), "abc");
+    }
+  };
+
+  // Ordinary SST files with same session id
+  const std::string something_compressible(500U, 'x');
+  for (int i = 0; i < 2; ++i) {
+    ASSERT_OK(Put(1, Key(key_count), "abc"));
+    ASSERT_OK(Put(1, Key(key_count) + "a", something_compressible));
+    ASSERT_OK(Flush(1));
+    ++key_count;
+  }
+
+#ifndef ROCKSDB_LITE
+  // Save an export of those ordinary SST files for later
+  std::string export_files_dir = dbname_ + "/exported";
+  ExportImportFilesMetaData* metadata_ptr_ = nullptr;
+  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;
+  checkpoint = nullptr;
+
+  // External SST files with same session id
+  SstFileWriter sst_file_writer(EnvOptions(), options);
+  std::vector<std::string> external;
+  for (int i = 0; i < 2; ++i) {
+    std::string f = dbname_ + "/external" + ToString(i) + ".sst";
+    external.push_back(f);
+    ASSERT_OK(sst_file_writer.Open(f));
+    ASSERT_OK(sst_file_writer.Put(Key(key_count), "abc"));
+    ASSERT_OK(
+        sst_file_writer.Put(Key(key_count) + "a", something_compressible));
+    ++key_count;
+    ExternalSstFileInfo external_info;
+    ASSERT_OK(sst_file_writer.Finish(&external_info));
+    IngestExternalFileOptions ingest_opts;
+    ASSERT_OK(db_->IngestExternalFile(handles_[1], {f}, ingest_opts));
+  }
+
+  if (exclude_file_numbers_) {
+    // FIXME(peterd): figure out where these extra two ADDs are coming from
+    options.statistics->recordTick(BLOCK_CACHE_INDEX_ADD,
+                                   uint64_t{0} - uint64_t{2});
+    options.statistics->recordTick(BLOCK_CACHE_FILTER_ADD,
+                                   uint64_t{0} - uint64_t{2});
+    options.statistics->recordTick(BLOCK_CACHE_COMPRESSED_ADD,
+                                   uint64_t{0} - uint64_t{2});
+  }
+#endif
+
+  perform_gets();
+  verify_stats();
+
+  // Make sure we can cache hit after re-open
+  ReopenWithColumnFamilies({"default", "koko"}, options);
+
+  perform_gets();
+  verify_stats();
+
+  // Make sure we can cache hit even on a full copy of the DB. Using
+  // StableCacheKeyTestFS, Checkpoint will resort to full copy not hard link.
+  // (Checkpoint  not available in LITE mode to test this.)
+#ifndef ROCKSDB_LITE
+  auto db_copy_name = dbname_ + "-copy";
+  ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
+  ASSERT_OK(checkpoint->CreateCheckpoint(db_copy_name));
+  delete checkpoint;
+
+  Close();
+  Destroy(options);
+
+  // Switch to the DB copy
+  SaveAndRestore<std::string> save_dbname(&dbname_, db_copy_name);
+  ReopenWithColumnFamilies({"default", "koko"}, options);
+
+  perform_gets();
+  verify_stats();
+
+  // And ensure that re-importing + ingesting the same files into a
+  // different DB uses same cache keys
+  DestroyAndReopen(options);
+
+  ColumnFamilyHandle* cfh = nullptr;
+  ASSERT_OK(db_->CreateColumnFamilyWithImport(ColumnFamilyOptions(), "yoyo",
+                                              ImportColumnFamilyOptions(),
+                                              *metadata_ptr_, &cfh));
+  ASSERT_NE(cfh, nullptr);
+  delete cfh;
+  cfh = nullptr;
+  delete metadata_ptr_;
+  metadata_ptr_ = nullptr;
+
+  DestroyDB(export_files_dir, options);
+
+  ReopenWithColumnFamilies({"default", "yoyo"}, options);
+
+  IngestExternalFileOptions ingest_opts;
+  ASSERT_OK(db_->IngestExternalFile(handles_[1], {external}, ingest_opts));
+
+  if (exclude_file_numbers_) {
+    // FIXME(peterd): figure out where these extra two ADDs are coming from
+    options.statistics->recordTick(BLOCK_CACHE_INDEX_ADD,
+                                   uint64_t{0} - uint64_t{2});
+    options.statistics->recordTick(BLOCK_CACHE_FILTER_ADD,
+                                   uint64_t{0} - uint64_t{2});
+  }
+
+  perform_gets();
+  verify_stats();
+#endif  // !ROCKSDB_LITE
+
+  Close();
+  Destroy(options);
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+INSTANTIATE_TEST_CASE_P(DBBlockCacheKeyTest, DBBlockCacheKeyTest,
+                        ::testing::Combine(::testing::Bool(),
+                                           ::testing::Bool()));
+
 class DBBlockCachePinningTest
    : public DBTestBase,
      public testing::WithParamInterface<
          std::tuple<bool, PinningTier, PinningTier, PinningTier>> {
 public:
  DBBlockCachePinningTest()
-      : DBTestBase("/db_block_cache_test", /*env_do_fsync=*/false) {}
+      : DBTestBase("db_block_cache_test", /*env_do_fsync=*/false) {}

  void SetUp() override {
    partition_index_and_filters_ = std::get<0>(GetParam());
@@ -15,6 +15,7 @@
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
 #include "table/block_based/filter_policy_internal.h"
+#include "test_util/testutil.h"
 #include "util/string_util.h"

 namespace ROCKSDB_NAMESPACE {
@@ -28,7 +29,7 @@ using BFP = BloomFilterPolicy;
 class DBBloomFilterTest : public DBTestBase {
 public:
  DBBloomFilterTest()
-      : DBTestBase("/db_bloom_filter_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_bloom_filter_test", /*env_do_fsync=*/true) {}
 };

 class DBBloomFilterTestWithParam : public DBTestBase,
@@ -42,7 +43,7 @@ class DBBloomFilterTestWithParam : public DBTestBase,

 public:
  DBBloomFilterTestWithParam()
-      : DBTestBase("/db_bloom_filter_tests", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_bloom_filter_tests", /*env_do_fsync=*/true) {}

  ~DBBloomFilterTestWithParam() override {}

@@ -527,7 +528,7 @@ TEST_P(DBBloomFilterTestWithParam, BloomFilter) {
  } while (ChangeCompactOptions());
 }

-#ifndef ROCKSDB_VALGRIND_RUN
+#if !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 INSTANTIATE_TEST_CASE_P(
    FormatDef, DBBloomFilterTestDefFormatVersion,
    ::testing::Values(
@@ -551,7 +552,7 @@ INSTANTIATE_TEST_CASE_P(
                        test::kLatestFormatVersion),
        std::make_tuple(BFP::kAutoBloom, true, test::kLatestFormatVersion),
        std::make_tuple(BFP::kAutoBloom, false, test::kLatestFormatVersion)));
-#endif  // ROCKSDB_VALGRIND_RUN
+#endif  // !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)

 TEST_F(DBBloomFilterTest, BloomFilterRate) {
  while (ChangeFilterOptions()) {
@@ -1094,7 +1095,7 @@ class DBBloomFilterTestVaryPrefixAndFormatVer

 public:
  DBBloomFilterTestVaryPrefixAndFormatVer()
-      : DBTestBase("/db_bloom_filter_tests", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_bloom_filter_tests", /*env_do_fsync=*/true) {}

  ~DBBloomFilterTestVaryPrefixAndFormatVer() override {}

@@ -2171,7 +2172,8 @@ TEST_F(DBBloomFilterTest, SeekForPrevWithPartitionedFilters) {
  Options options = CurrentOptions();
  constexpr size_t kNumKeys = 10000;
  static_assert(kNumKeys <= 10000, "kNumKeys have to be <= 10000");
-  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeys + 10));
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(kNumKeys + 10));
  options.create_if_missing = true;
  constexpr size_t kPrefixLength = 4;
  options.prefix_extractor.reset(NewFixedPrefixTransform(kPrefixLength));
@@ -22,7 +22,7 @@ static std::string NEW_VALUE = "NewValue";
 class DBTestCompactionFilter : public DBTestBase {
 public:
  DBTestCompactionFilter()
-      : DBTestBase("/db_compaction_filter_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_compaction_filter_test", /*env_do_fsync=*/true) {}
 };

 // Param variant of DBTestBase::ChangeCompactOptions
@@ -46,7 +46,7 @@ class DBTestCompactionFilterWithCompactParam
  }
 };

-#ifndef ROCKSDB_VALGRIND_RUN
+#if !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 INSTANTIATE_TEST_CASE_P(
    CompactionFilterWithOption, DBTestCompactionFilterWithCompactParam,
    ::testing::Values(DBTestBase::OptionConfig::kDefault,
@@ -55,11 +55,11 @@ INSTANTIATE_TEST_CASE_P(
                      DBTestBase::OptionConfig::kLevelSubcompactions,
                      DBTestBase::OptionConfig::kUniversalSubcompactions));
 #else
-// Run fewer cases in valgrind
+// Run fewer cases in non-full valgrind to save time.
 INSTANTIATE_TEST_CASE_P(CompactionFilterWithOption,
                        DBTestCompactionFilterWithCompactParam,
                        ::testing::Values(DBTestBase::OptionConfig::kDefault));
-#endif  // ROCKSDB_VALGRIND_RUN
+#endif  // !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)

 class KeepFilter : public CompactionFilter {
 public:
@@ -19,6 +19,7 @@
 #include "rocksdb/sst_file_writer.h"
 #include "rocksdb/utilities/convenience.h"
 #include "test_util/sync_point.h"
+#include "test_util/testutil.h"
 #include "util/concurrent_task_limiter_impl.h"
 #include "util/random.h"
 #include "utilities/fault_injection_env.h"
@@ -32,7 +33,7 @@ namespace ROCKSDB_NAMESPACE {
 class DBCompactionTest : public DBTestBase {
 public:
  DBCompactionTest()
-      : DBTestBase("/db_compaction_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_compaction_test", /*env_do_fsync=*/true) {}
 };

 class DBCompactionTestWithParam
@@ -40,7 +41,7 @@ class DBCompactionTestWithParam
      public testing::WithParamInterface<std::tuple<uint32_t, bool>> {
 public:
  DBCompactionTestWithParam()
-      : DBTestBase("/db_compaction_test", /*env_do_fsync=*/true) {
+      : DBTestBase("db_compaction_test", /*env_do_fsync=*/true) {
    max_subcompactions_ = std::get<0>(GetParam());
    exclusive_manual_compaction_ = std::get<1>(GetParam());
  }
@@ -58,7 +59,7 @@ class DBCompactionTestWithBottommostParam
      public testing::WithParamInterface<BottommostLevelCompaction> {
 public:
  DBCompactionTestWithBottommostParam()
-      : DBTestBase("/db_compaction_test", /*env_do_fsync=*/true) {
+      : DBTestBase("db_compaction_test", /*env_do_fsync=*/true) {
    bottommost_level_compaction_ = GetParam();
  }

@@ -301,7 +302,7 @@ const SstFileMetaData* PickFileRandomly(
 }
 }  // anonymous namespace

-#ifndef ROCKSDB_VALGRIND_RUN
+#if !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 // All the TEST_P tests run once with sub_compactions disabled (i.e.
 // options.max_subcompactions = 1) and once with it enabled
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
@@ -366,7 +367,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
    }
  }
 }
-#endif  // ROCKSDB_VALGRIND_RUN
+#endif  // !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)

 TEST_P(DBCompactionTestWithParam, CompactionsPreserveDeletes) {
  //  For each options type we test following
@@ -805,7 +806,8 @@ TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
  options.num_levels = 3;
  options.level0_file_num_compaction_trigger = 3;
  options.max_subcompactions = max_subcompactions_;
-  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(kNumKeysPerFile));
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);
@@ -852,7 +854,8 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
  options.level0_slowdown_writes_trigger = 20;
  options.soft_pending_compaction_bytes_limit = 1 << 30;  // Infinitely large
  options.max_background_compactions = 3;
-  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(kNumKeysPerFile));

  // Block all threads in thread pool.
  const size_t kTotalTasks = 4;
@@ -2094,7 +2097,7 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionThirdPath) {
  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
@@ -2203,7 +2206,7 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionPathUse) {
  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
@@ -2313,7 +2316,7 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionCFPathUse) {
  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
  options.memtable_factory.reset(
-    new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
@@ -2332,7 +2335,7 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionCFPathUse) {
  option_vector.emplace_back(DBOptions(options), cf_opt1);
  CreateColumnFamilies({"one"},option_vector[1]);

-  // Configura CF2 specific paths.
+  // Configure CF2 specific paths.
  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2", 500 * 1024);
  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_2", 4 * 1024 * 1024);
  cf_opt2.cf_paths.emplace_back(dbname_ + "cf2_3", 1024 * 1024 * 1024);
@@ -2387,13 +2390,16 @@ TEST_P(DBCompactionTestWithParam, LevelCompactionCFPathUse) {
  // Check that default column family uses db_paths.
  // And Column family "one" uses cf_paths.

-  // First three 110KB files are not going to second path.
-  // After that, (100K, 200K)
+  // The compaction in level0 outputs the sst files in level1.
+  // The first path cannot hold level1's data(400KB+400KB > 500KB),
+  // so every compaction move a sst file to second path. Please
+  // refer to LevelCompactionBuilder::GetPathId.
  for (int num = 0; num < 3; num++) {
    generate_file();
  }
+  check_sstfilecount(0, 1);
+  check_sstfilecount(1, 2);

-  // Another 110KB triggers a compaction to 400K file to fill up first path
  generate_file();
  check_sstfilecount(1, 3);

@@ -3009,7 +3015,7 @@ TEST_P(DBCompactionTestWithParam, CompressLevelCompaction) {
  }
  Options options = CurrentOptions();
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 110 << 10;  // 110KB
  options.arena_block_size = 4 << 10;
@@ -4812,7 +4818,8 @@ TEST_F(DBCompactionTest, CompactionLimiter) {
  options.level0_slowdown_writes_trigger = 64;
  options.level0_stop_writes_trigger = 64;
  options.max_background_jobs = kMaxBackgroundThreads; // Enough threads
-  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(kNumKeysPerFile));
  options.max_write_buffer_number = 10; // Enough memtables
  DestroyAndReopen(options);

@@ -5005,7 +5012,7 @@ class CompactionPriTest : public DBTestBase,
                          public testing::WithParamInterface<uint32_t> {
 public:
  CompactionPriTest()
-      : DBTestBase("/compaction_pri_test", /*env_do_fsync=*/true) {
+      : DBTestBase("compaction_pri_test", /*env_do_fsync=*/true) {
    compaction_pri_ = GetParam();
  }

@@ -5335,7 +5342,8 @@ TEST_P(DBCompactionTestWithParam, FixFileIngestionCompactionDeadlock) {
  options.level0_file_num_compaction_trigger =
      options.level0_stop_writes_trigger;
  options.max_subcompactions = max_subcompactions_;
-  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.memtable_factory.reset(
+      test::NewSpecialSkipListFactory(kNumKeysPerFile));
  DestroyAndReopen(options);
  Random rnd(301);

@@ -5861,7 +5869,7 @@ TEST_F(DBCompactionTest, ChangeLevelCompactRangeConflictsWithManual) {
  // `Status::Incomplete`.
  Options options = CurrentOptions();
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 3;
  Reopen(options);
@@ -5949,7 +5957,7 @@ TEST_F(DBCompactionTest, ChangeLevelErrorPathTest) {
  // succeeds
  Options options = CurrentOptions();
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
+      test::NewSpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 3;
  Reopen(options);
@@ -6746,6 +6754,112 @@ TEST_F(DBCompactionTest, CompactionWithChecksumHandoffManifest2) {
  Destroy(options);
 }

+TEST_F(DBCompactionTest, FIFOWarm) {
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleFIFO;
+  options.num_levels = 1;
+  options.max_open_files = -1;
+  options.level0_file_num_compaction_trigger = 2;
+  options.create_if_missing = true;
+  CompactionOptionsFIFO fifo_options;
+  fifo_options.age_for_warm = 1000;
+  fifo_options.max_table_files_size = 100000000;
+  options.compaction_options_fifo = fifo_options;
+  env_->SetMockSleep();
+  Reopen(options);
+
+  int total_warm = 0;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "NewWritableFile::FileOptions.temperature", [&](void* arg) {
+        Temperature temperature = *(static_cast<Temperature*>(arg));
+        if (temperature == Temperature::kWarm) {
+          total_warm++;
+        }
+      });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  // The file system does not support checksum handoff. The check
+  // will be ignored.
+  ASSERT_OK(Put(Key(0), "value1"));
+  env_->MockSleepForSeconds(800);
+  ASSERT_OK(Put(Key(2), "value2"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(Key(0), "value1"));
+  env_->MockSleepForSeconds(800);
+  ASSERT_OK(Put(Key(2), "value2"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(Key(0), "value1"));
+  env_->MockSleepForSeconds(800);
+  ASSERT_OK(Put(Key(2), "value2"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  ASSERT_OK(Put(Key(0), "value1"));
+  env_->MockSleepForSeconds(800);
+  ASSERT_OK(Put(Key(2), "value2"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
+
+  ColumnFamilyMetaData metadata;
+  db_->GetColumnFamilyMetaData(&metadata);
+  ASSERT_EQ(4, metadata.file_count);
+  ASSERT_EQ(Temperature::kUnknown, metadata.levels[0].files[0].temperature);
+  ASSERT_EQ(Temperature::kUnknown, metadata.levels[0].files[1].temperature);
+  ASSERT_EQ(Temperature::kWarm, metadata.levels[0].files[2].temperature);
+  ASSERT_EQ(Temperature::kWarm, metadata.levels[0].files[3].temperature);
+  ASSERT_EQ(2, total_warm);
+
+  Destroy(options);
+}
+
+TEST_F(DBCompactionTest,
+       DisableManualCompactionDoesNotWaitForDrainingAutomaticCompaction) {
+  // When `CompactRangeOptions::exclusive_manual_compaction == true`, we wait
+  // for automatic compactions to drain before starting the manual compaction.
+  // This test verifies `DisableManualCompaction()` can cancel such a compaction
+  // without waiting for the drain to complete.
+  const int kNumL0Files = 4;
+
+  // Enforces manual compaction enters wait loop due to pending automatic
+  // compaction.
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::BGWorkCompaction", "DBImpl::RunManualCompaction:NotScheduled"},
+       {"DBImpl::RunManualCompaction:WaitScheduled",
+        "BackgroundCallCompaction:0"}});
+  // The automatic compaction will cancel the waiting manual compaction.
+  // Completing this implies the cancellation did not wait on automatic
+  // compactions to finish.
+  bool callback_completed = false;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "BackgroundCallCompaction:0", [&](void* /*arg*/) {
+        db_->DisableManualCompaction();
+        callback_completed = true;
+      });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options = CurrentOptions();
+  options.level0_file_num_compaction_trigger = kNumL0Files;
+  Reopen(options);
+
+  for (int i = 0; i < kNumL0Files; ++i) {
+    ASSERT_OK(Put(Key(1), "value1"));
+    ASSERT_OK(Put(Key(2), "value2"));
+    ASSERT_OK(Flush());
+  }
+
+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = true;
+  ASSERT_TRUE(db_->CompactRange(cro, nullptr, nullptr).IsIncomplete());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  ASSERT_TRUE(callback_completed);
+}
+
 #endif  // !defined(ROCKSDB_LITE)

 }  // namespace ROCKSDB_NAMESPACE
@@ -22,7 +22,7 @@ namespace ROCKSDB_NAMESPACE {
 class DBTestDynamicLevel : public DBTestBase {
 public:
  DBTestDynamicLevel()
-      : DBTestBase("/db_dynamic_level_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_dynamic_level_test", /*env_do_fsync=*/true) {}
 };

 TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase) {
@@ -17,7 +17,7 @@ namespace ROCKSDB_NAMESPACE {
 class DBEncryptionTest : public DBTestBase {
 public:
  DBEncryptionTest()
-      : DBTestBase("/db_encryption_test", /*env_do_fsync=*/true) {}
+      : DBTestBase("db_encryption_test", /*env_do_fsync=*/true) {}
  Env* GetTargetEnv() {
    if (encrypted_env_ != nullptr) {
      return (static_cast<EnvWrapper*>(encrypted_env_))->target();
@@ -127,7 +127,22 @@ Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
      bg_cv_.Wait();
    }
  }
-  return wal_manager_.GetSortedWalFiles(files);
+
+  // Disable deletion in order to avoid the case where a file is deleted in
+  // the middle of the process so IO error is returned.
+  Status s = DisableFileDeletions();
+  bool file_deletion_supported = !s.IsNotSupported();
+  if (s.ok() || !file_deletion_supported) {
+    s = wal_manager_.GetSortedWalFiles(files);
+    if (file_deletion_supported) {
+      Status s2 = EnableFileDeletions(false);
+      if (!s2.ok() && s.ok()) {
+        s = s2;
+      }
+    }
+  }
+
+  return s;
 }

 Status DBImpl::GetCurrentWalFile(std::unique_ptr<LogFile>* current_log_file) {
@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include <atomic>
+#include <limits>

 #include "db/db_impl/db_impl.h"
 #include "db/db_test_util.h"
@@ -17,6 +18,7 @@
 #include "port/stack_trace.h"
 #include "rocksdb/utilities/transaction_db.h"
 #include "test_util/sync_point.h"
+#include "test_util/testutil.h"
 #include "util/cast_util.h"
 #include "util/mutexlock.h"
 #include "utilities/fault_injection_env.h"
@@ -24,9 +26,13 @@

 namespace ROCKSDB_NAMESPACE {

+// This is a static filter used for filtering
+// kvs during the compaction process.
+static std::string NEW_VALUE = "NewValue";
+
 class DBFlushTest : public DBTestBase {
 public:
-  DBFlushTest() : DBTestBase("/db_flush_test", /*env_do_fsync=*/true) {}
+  DBFlushTest() : DBTestBase("db_flush_test", /*env_do_fsync=*/true) {}
 };

 class DBFlushDirectIOTest : public DBFlushTest,
@@ -134,7 +140,7 @@ TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
  // scheduled in the low-pri (compaction) thread pool.
  Options options = CurrentOptions();
  options.level0_file_num_compaction_trigger = 4;
-  options.memtable_factory.reset(new SpecialSkipListFactory(1));
+  options.memtable_factory.reset(test::NewSpecialSkipListFactory(1));
  Reopen(options);
  env_->SetBackgroundThreads(0, Env::HIGH);

@@ -658,6 +664,717 @@ TEST_F(DBFlushTest, StatisticsGarbageRangeDeletes) {
  Close();
 }

+#ifndef ROCKSDB_LITE
+// This simple Listener can only handle one flush at a time.
+class TestFlushListener : public EventListener {
+ public:
+  TestFlushListener(Env* env, DBFlushTest* test)
+      : slowdown_count(0), stop_count(0), db_closed(), env_(env), test_(test) {
+    db_closed = false;
+  }
+
+  ~TestFlushListener() override {
+    prev_fc_info_.status.PermitUncheckedError();  // Ignore the status
+  }
+  void OnTableFileCreated(const TableFileCreationInfo& info) override {
+    // remember the info for later checking the FlushJobInfo.
+    prev_fc_info_ = info;
+    ASSERT_GT(info.db_name.size(), 0U);
+    ASSERT_GT(info.cf_name.size(), 0U);
+    ASSERT_GT(info.file_path.size(), 0U);
+    ASSERT_GT(info.job_id, 0);
+    ASSERT_GT(info.table_properties.data_size, 0U);
+    ASSERT_GT(info.table_properties.raw_key_size, 0U);
+    ASSERT_GT(info.table_properties.raw_value_size, 0U);
+    ASSERT_GT(info.table_properties.num_data_blocks, 0U);
+    ASSERT_GT(info.table_properties.num_entries, 0U);
+    ASSERT_EQ(info.file_checksum, kUnknownFileChecksum);
+    ASSERT_EQ(info.file_checksum_func_name, kUnknownFileChecksumFuncName);
+  }
+
+  void OnFlushCompleted(DB* db, const FlushJobInfo& info) override {
+    flushed_dbs_.push_back(db);
+    flushed_column_family_names_.push_back(info.cf_name);
+    if (info.triggered_writes_slowdown) {
+      slowdown_count++;
+    }
+    if (info.triggered_writes_stop) {
+      stop_count++;
+    }
+    // verify whether the previously created file matches the flushed file.
+    ASSERT_EQ(prev_fc_info_.db_name, db->GetName());
+    ASSERT_EQ(prev_fc_info_.cf_name, info.cf_name);
+    ASSERT_EQ(prev_fc_info_.job_id, info.job_id);
+    ASSERT_EQ(prev_fc_info_.file_path, info.file_path);
+    ASSERT_EQ(TableFileNameToNumber(info.file_path), info.file_number);
+
+    // Note: the following chunk relies on the notification pertaining to the
+    // database pointed to by DBTestBase::db_, and is thus bypassed when
+    // that assumption does not hold (see the test case MultiDBMultiListeners
+    // below).
+    ASSERT_TRUE(test_);
+    if (db == test_->db_) {
+      std::vector<std::vector<FileMetaData>> files_by_level;
+      test_->dbfull()->TEST_GetFilesMetaData(db->DefaultColumnFamily(),
+                                             &files_by_level);
+
+      ASSERT_FALSE(files_by_level.empty());
+      auto it = std::find_if(files_by_level[0].begin(), files_by_level[0].end(),
+                             [&](const FileMetaData& meta) {
+                               return meta.fd.GetNumber() == info.file_number;
+                             });
+      ASSERT_NE(it, files_by_level[0].end());
+      ASSERT_EQ(info.oldest_blob_file_number, it->oldest_blob_file_number);
+    }
+
+    ASSERT_EQ(db->GetEnv()->GetThreadID(), info.thread_id);
+    ASSERT_GT(info.thread_id, 0U);
+  }
+
+  std::vector<std::string> flushed_column_family_names_;
+  std::vector<DB*> flushed_dbs_;
+  int slowdown_count;
+  int stop_count;
+  bool db_closing;
+  std::atomic_bool db_closed;
+  TableFileCreationInfo prev_fc_info_;
+
+ protected:
+  Env* env_;
+  DBFlushTest* test_;
+};
+#endif  // !ROCKSDB_LITE
+
+TEST_F(DBFlushTest, MemPurgeBasic) {
+  Options options = CurrentOptions();
+
+  // The following options are used to enforce several values that
+  // may already exist as default values to make this test resilient
+  // to default value updates in the future.
+  options.statistics = CreateDBStatistics();
+
+  // Record all statistics.
+  options.statistics->set_stats_level(StatsLevel::kAll);
+
+  // create the DB if it's not already present
+  options.create_if_missing = true;
+
+  // Useful for now as we are trying to compare uncompressed data savings on
+  // flush().
+  options.compression = kNoCompression;
+
+  // Prevent memtable in place updates. Should already be disabled
+  // (from Wiki:
+  //  In place updates can be enabled by toggling on the bool
+  //  inplace_update_support flag. However, this flag is by default set to
+  //  false
+  //  because this thread-safe in-place update support is not compatible
+  //  with concurrent memtable writes. Note that the bool
+  //  allow_concurrent_memtable_write is set to true by default )
+  options.inplace_update_support = false;
+  options.allow_concurrent_memtable_write = true;
+
+  // Enforce size of a single MemTable to 64MB (64MB = 67108864 bytes).
+  options.write_buffer_size = 1 << 20;
+  // Activate the MemPurge prototype.
+  options.experimental_mempurge_threshold = 1.0;
+#ifndef ROCKSDB_LITE
+  TestFlushListener* listener = new TestFlushListener(options.env, this);
+  options.listeners.emplace_back(listener);
+#endif  // !ROCKSDB_LITE
+  ASSERT_OK(TryReopen(options));
+  uint32_t mempurge_count = 0;
+  uint32_t sst_count = 0;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:MemPurgeSuccessful",
+      [&](void* /*arg*/) { mempurge_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:SSTFileCreated", [&](void* /*arg*/) { sst_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::string KEY1 = "IamKey1";
+  std::string KEY2 = "IamKey2";
+  std::string KEY3 = "IamKey3";
+  std::string KEY4 = "IamKey4";
+  std::string KEY5 = "IamKey5";
+  std::string KEY6 = "IamKey6";
+  std::string KEY7 = "IamKey7";
+  std::string KEY8 = "IamKey8";
+  std::string KEY9 = "IamKey9";
+  std::string RNDKEY1, RNDKEY2, RNDKEY3;
+  const std::string NOT_FOUND = "NOT_FOUND";
+
+  // Heavy overwrite workload,
+  // more than would fit in maximum allowed memtables.
+  Random rnd(719);
+  const size_t NUM_REPEAT = 100;
+  const size_t RAND_KEYS_LENGTH = 57;
+  const size_t RAND_VALUES_LENGTH = 10240;
+  std::string p_v1, p_v2, p_v3, p_v4, p_v5, p_v6, p_v7, p_v8, p_v9, p_rv1,
+      p_rv2, p_rv3;
+
+  // Insert a very first set of keys that will be
+  // mempurged at least once.
+  p_v1 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v2 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v3 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v4 = rnd.RandomString(RAND_VALUES_LENGTH);
+  ASSERT_OK(Put(KEY1, p_v1));
+  ASSERT_OK(Put(KEY2, p_v2));
+  ASSERT_OK(Put(KEY3, p_v3));
+  ASSERT_OK(Put(KEY4, p_v4));
+  ASSERT_EQ(Get(KEY1), p_v1);
+  ASSERT_EQ(Get(KEY2), p_v2);
+  ASSERT_EQ(Get(KEY3), p_v3);
+  ASSERT_EQ(Get(KEY4), p_v4);
+
+  // Insertion of of K-V pairs, multiple times (overwrites).
+  for (size_t i = 0; i < NUM_REPEAT; i++) {
+    // Create value strings of arbitrary length RAND_VALUES_LENGTH bytes.
+    p_v5 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v6 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v7 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v8 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v9 = rnd.RandomString(RAND_VALUES_LENGTH);
+
+    ASSERT_OK(Put(KEY5, p_v5));
+    ASSERT_OK(Put(KEY6, p_v6));
+    ASSERT_OK(Put(KEY7, p_v7));
+    ASSERT_OK(Put(KEY8, p_v8));
+    ASSERT_OK(Put(KEY9, p_v9));
+
+    ASSERT_EQ(Get(KEY1), p_v1);
+    ASSERT_EQ(Get(KEY2), p_v2);
+    ASSERT_EQ(Get(KEY3), p_v3);
+    ASSERT_EQ(Get(KEY4), p_v4);
+    ASSERT_EQ(Get(KEY5), p_v5);
+    ASSERT_EQ(Get(KEY6), p_v6);
+    ASSERT_EQ(Get(KEY7), p_v7);
+    ASSERT_EQ(Get(KEY8), p_v8);
+    ASSERT_EQ(Get(KEY9), p_v9);
+  }
+
+  // Check that there was at least one mempurge
+  const uint32_t EXPECTED_MIN_MEMPURGE_COUNT = 1;
+  // Check that there was no SST files created during flush.
+  const uint32_t EXPECTED_SST_COUNT = 0;
+
+  EXPECT_GE(mempurge_count, EXPECTED_MIN_MEMPURGE_COUNT);
+  EXPECT_EQ(sst_count, EXPECTED_SST_COUNT);
+
+  const uint32_t mempurge_count_record = mempurge_count;
+
+  // Insertion of of K-V pairs, no overwrites.
+  for (size_t i = 0; i < NUM_REPEAT; i++) {
+    // Create value strings of arbitrary length RAND_VALUES_LENGTH bytes.
+    RNDKEY1 = rnd.RandomString(RAND_KEYS_LENGTH);
+    RNDKEY2 = rnd.RandomString(RAND_KEYS_LENGTH);
+    RNDKEY3 = rnd.RandomString(RAND_KEYS_LENGTH);
+    p_rv1 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_rv2 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_rv3 = rnd.RandomString(RAND_VALUES_LENGTH);
+
+    ASSERT_OK(Put(RNDKEY1, p_rv1));
+    ASSERT_OK(Put(RNDKEY2, p_rv2));
+    ASSERT_OK(Put(RNDKEY3, p_rv3));
+
+    ASSERT_EQ(Get(KEY1), p_v1);
+    ASSERT_EQ(Get(KEY2), p_v2);
+    ASSERT_EQ(Get(KEY3), p_v3);
+    ASSERT_EQ(Get(KEY4), p_v4);
+    ASSERT_EQ(Get(KEY5), p_v5);
+    ASSERT_EQ(Get(KEY6), p_v6);
+    ASSERT_EQ(Get(KEY7), p_v7);
+    ASSERT_EQ(Get(KEY8), p_v8);
+    ASSERT_EQ(Get(KEY9), p_v9);
+    ASSERT_EQ(Get(RNDKEY1), p_rv1);
+    ASSERT_EQ(Get(RNDKEY2), p_rv2);
+    ASSERT_EQ(Get(RNDKEY3), p_rv3);
+  }
+
+  // Assert that at least one flush to storage has been performed
+  ASSERT_GT(sst_count, EXPECTED_SST_COUNT);
+  // (which will consequently increase the number of mempurges recorded too).
+  ASSERT_GE(mempurge_count, mempurge_count_record);
+
+  // Assert that there is no data corruption, even with
+  // a flush to storage.
+  ASSERT_EQ(Get(KEY1), p_v1);
+  ASSERT_EQ(Get(KEY2), p_v2);
+  ASSERT_EQ(Get(KEY3), p_v3);
+  ASSERT_EQ(Get(KEY4), p_v4);
+  ASSERT_EQ(Get(KEY5), p_v5);
+  ASSERT_EQ(Get(KEY6), p_v6);
+  ASSERT_EQ(Get(KEY7), p_v7);
+  ASSERT_EQ(Get(KEY8), p_v8);
+  ASSERT_EQ(Get(KEY9), p_v9);
+  ASSERT_EQ(Get(RNDKEY1), p_rv1);
+  ASSERT_EQ(Get(RNDKEY2), p_rv2);
+  ASSERT_EQ(Get(RNDKEY3), p_rv3);
+
+  Close();
+}
+
+TEST_F(DBFlushTest, MemPurgeDeleteAndDeleteRange) {
+  Options options = CurrentOptions();
+
+  options.statistics = CreateDBStatistics();
+  options.statistics->set_stats_level(StatsLevel::kAll);
+  options.create_if_missing = true;
+  options.compression = kNoCompression;
+  options.inplace_update_support = false;
+  options.allow_concurrent_memtable_write = true;
+#ifndef ROCKSDB_LITE
+  TestFlushListener* listener = new TestFlushListener(options.env, this);
+  options.listeners.emplace_back(listener);
+#endif  // !ROCKSDB_LITE
+  // Enforce size of a single MemTable to 64MB (64MB = 67108864 bytes).
+  options.write_buffer_size = 1 << 20;
+  // Activate the MemPurge prototype.
+  options.experimental_mempurge_threshold = 1.0;
+
+  ASSERT_OK(TryReopen(options));
+
+  uint32_t mempurge_count = 0;
+  uint32_t sst_count = 0;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:MemPurgeSuccessful",
+      [&](void* /*arg*/) { mempurge_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:SSTFileCreated", [&](void* /*arg*/) { sst_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::string KEY1 = "ThisIsKey1";
+  std::string KEY2 = "ThisIsKey2";
+  std::string KEY3 = "ThisIsKey3";
+  std::string KEY4 = "ThisIsKey4";
+  std::string KEY5 = "ThisIsKey5";
+  const std::string NOT_FOUND = "NOT_FOUND";
+
+  Random rnd(117);
+  const size_t NUM_REPEAT = 100;
+  const size_t RAND_VALUES_LENGTH = 10240;
+
+  std::string key, value, p_v1, p_v2, p_v3, p_v3b, p_v4, p_v5;
+  int count = 0;
+  const int EXPECTED_COUNT_FORLOOP = 3;
+  const int EXPECTED_COUNT_END = 4;
+
+  ReadOptions ropt;
+  ropt.pin_data = true;
+  ropt.total_order_seek = true;
+  Iterator* iter = nullptr;
+
+  // Insertion of of K-V pairs, multiple times.
+  // Also insert DeleteRange
+  for (size_t i = 0; i < NUM_REPEAT; i++) {
+    // Create value strings of arbitrary length RAND_VALUES_LENGTH bytes.
+    p_v1 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v2 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v3 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v3b = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v4 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v5 = rnd.RandomString(RAND_VALUES_LENGTH);
+    ASSERT_OK(Put(KEY1, p_v1));
+    ASSERT_OK(Put(KEY2, p_v2));
+    ASSERT_OK(Put(KEY3, p_v3));
+    ASSERT_OK(Put(KEY4, p_v4));
+    ASSERT_OK(Put(KEY5, p_v5));
+    ASSERT_OK(Delete(KEY2));
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), KEY2,
+                               KEY4));
+    ASSERT_OK(Put(KEY3, p_v3b));
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), KEY1,
+                               KEY3));
+    ASSERT_OK(Delete(KEY1));
+
+    ASSERT_EQ(Get(KEY1), NOT_FOUND);
+    ASSERT_EQ(Get(KEY2), NOT_FOUND);
+    ASSERT_EQ(Get(KEY3), p_v3b);
+    ASSERT_EQ(Get(KEY4), p_v4);
+    ASSERT_EQ(Get(KEY5), p_v5);
+
+    iter = db_->NewIterator(ropt);
+    iter->SeekToFirst();
+    count = 0;
+    for (; iter->Valid(); iter->Next()) {
+      ASSERT_OK(iter->status());
+      key = (iter->key()).ToString(false);
+      value = (iter->value()).ToString(false);
+      if (key.compare(KEY3) == 0)
+        ASSERT_EQ(value, p_v3b);
+      else if (key.compare(KEY4) == 0)
+        ASSERT_EQ(value, p_v4);
+      else if (key.compare(KEY5) == 0)
+        ASSERT_EQ(value, p_v5);
+      else
+        ASSERT_EQ(value, NOT_FOUND);
+      count++;
+    }
+
+    // Expected count here is 3: KEY3, KEY4, KEY5.
+    ASSERT_EQ(count, EXPECTED_COUNT_FORLOOP);
+    if (iter) {
+      delete iter;
+    }
+  }
+
+  // Check that there was at least one mempurge
+  const uint32_t EXPECTED_MIN_MEMPURGE_COUNT = 1;
+  // Check that there was no SST files created during flush.
+  const uint32_t EXPECTED_SST_COUNT = 0;
+
+  EXPECT_GE(mempurge_count, EXPECTED_MIN_MEMPURGE_COUNT);
+  EXPECT_EQ(sst_count, EXPECTED_SST_COUNT);
+
+  // Additional test for the iterator+memPurge.
+  ASSERT_OK(Put(KEY2, p_v2));
+  iter = db_->NewIterator(ropt);
+  iter->SeekToFirst();
+  ASSERT_OK(Put(KEY4, p_v4));
+  count = 0;
+  for (; iter->Valid(); iter->Next()) {
+    ASSERT_OK(iter->status());
+    key = (iter->key()).ToString(false);
+    value = (iter->value()).ToString(false);
+    if (key.compare(KEY2) == 0)
+      ASSERT_EQ(value, p_v2);
+    else if (key.compare(KEY3) == 0)
+      ASSERT_EQ(value, p_v3b);
+    else if (key.compare(KEY4) == 0)
+      ASSERT_EQ(value, p_v4);
+    else if (key.compare(KEY5) == 0)
+      ASSERT_EQ(value, p_v5);
+    else
+      ASSERT_EQ(value, NOT_FOUND);
+    count++;
+  }
+
+  // Expected count here is 4: KEY2, KEY3, KEY4, KEY5.
+  ASSERT_EQ(count, EXPECTED_COUNT_END);
+  if (iter) delete iter;
+
+  Close();
+}
+
+// Create a Compaction Fitler that will be invoked
+// at flush time and will update the value of a KV pair
+// if the key string is "lower" than the filter_key_ string.
+class ConditionalUpdateFilter : public CompactionFilter {
+ public:
+  explicit ConditionalUpdateFilter(const std::string* filtered_key)
+      : filtered_key_(filtered_key) {}
+  bool Filter(int /*level*/, const Slice& key, const Slice& /*value*/,
+              std::string* new_value, bool* value_changed) const override {
+    // If key<filtered_key_, update the value of the KV-pair.
+    if (key.compare(*filtered_key_) < 0) {
+      assert(new_value != nullptr);
+      *new_value = NEW_VALUE;
+      *value_changed = true;
+    }
+    return false /*do not remove this KV-pair*/;
+  }
+
+  const char* Name() const override { return "ConditionalUpdateFilter"; }
+
+ private:
+  const std::string* filtered_key_;
+};
+
+class ConditionalUpdateFilterFactory : public CompactionFilterFactory {
+ public:
+  explicit ConditionalUpdateFilterFactory(const Slice& filtered_key)
+      : filtered_key_(filtered_key.ToString()) {}
+
+  std::unique_ptr<CompactionFilter> CreateCompactionFilter(
+      const CompactionFilter::Context& /*context*/) override {
+    return std::unique_ptr<CompactionFilter>(
+        new ConditionalUpdateFilter(&filtered_key_));
+  }
+
+  const char* Name() const override { return "ConditionalUpdateFilterFactory"; }
+
+  bool ShouldFilterTableFileCreation(
+      TableFileCreationReason reason) const override {
+    // This compaction filter will be invoked
+    // at flush time (and therefore at MemPurge time).
+    return (reason == TableFileCreationReason::kFlush);
+  }
+
+ private:
+  std::string filtered_key_;
+};
+
+TEST_F(DBFlushTest, MemPurgeAndCompactionFilter) {
+  Options options = CurrentOptions();
+
+  std::string KEY1 = "ThisIsKey1";
+  std::string KEY2 = "ThisIsKey2";
+  std::string KEY3 = "ThisIsKey3";
+  std::string KEY4 = "ThisIsKey4";
+  std::string KEY5 = "ThisIsKey5";
+  std::string KEY6 = "ThisIsKey6";
+  std::string KEY7 = "ThisIsKey7";
+  std::string KEY8 = "ThisIsKey8";
+  std::string KEY9 = "ThisIsKey9";
+  const std::string NOT_FOUND = "NOT_FOUND";
+
+  options.statistics = CreateDBStatistics();
+  options.statistics->set_stats_level(StatsLevel::kAll);
+  options.create_if_missing = true;
+  options.compression = kNoCompression;
+  options.inplace_update_support = false;
+  options.allow_concurrent_memtable_write = true;
+#ifndef ROCKSDB_LITE
+  TestFlushListener* listener = new TestFlushListener(options.env, this);
+  options.listeners.emplace_back(listener);
+#endif  // !ROCKSDB_LITE
+  // Create a ConditionalUpdate compaction filter
+  // that will update all the values of the KV pairs
+  // where the keys are "lower" than KEY4.
+  options.compaction_filter_factory =
+      std::make_shared<ConditionalUpdateFilterFactory>(KEY4);
+
+  // Enforce size of a single MemTable to 64MB (64MB = 67108864 bytes).
+  options.write_buffer_size = 1 << 20;
+  // Activate the MemPurge prototype.
+  options.experimental_mempurge_threshold = 1.0;
+
+  ASSERT_OK(TryReopen(options));
+
+  uint32_t mempurge_count = 0;
+  uint32_t sst_count = 0;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:MemPurgeSuccessful",
+      [&](void* /*arg*/) { mempurge_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::FlushJob:SSTFileCreated", [&](void* /*arg*/) { sst_count++; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+  Random rnd(53);
+  const size_t NUM_REPEAT = 1000;
+  const size_t RAND_VALUES_LENGTH = 10240;
+  std::string p_v1, p_v2, p_v3, p_v4, p_v5, p_v6, p_v7, p_v8, p_v9;
+
+  p_v1 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v2 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v3 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v4 = rnd.RandomString(RAND_VALUES_LENGTH);
+  p_v5 = rnd.RandomString(RAND_VALUES_LENGTH);
+  ASSERT_OK(Put(KEY1, p_v1));
+  ASSERT_OK(Put(KEY2, p_v2));
+  ASSERT_OK(Put(KEY3, p_v3));
+  ASSERT_OK(Put(KEY4, p_v4));
+  ASSERT_OK(Put(KEY5, p_v5));
+  ASSERT_OK(Delete(KEY1));
+
+  // Insertion of of K-V pairs, multiple times.
+  for (size_t i = 0; i < NUM_REPEAT; i++) {
+    // Create value strings of arbitrary
+    // length RAND_VALUES_LENGTH bytes.
+    p_v6 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v7 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v8 = rnd.RandomString(RAND_VALUES_LENGTH);
+    p_v9 = rnd.RandomString(RAND_VALUES_LENGTH);
+    ASSERT_OK(Put(KEY6, p_v6));
+    ASSERT_OK(Put(KEY7, p_v7));
+    ASSERT_OK(Put(KEY8, p_v8));
+    ASSERT_OK(Put(KEY9, p_v9));
+
+    ASSERT_OK(Delete(KEY7));
+  }
+
+  // Check that there was at least one mempurge
+  const uint32_t EXPECTED_MIN_MEMPURGE_COUNT = 1;
+  // Check that there was no SST files created during flush.
+  const uint32_t EXPECTED_SST_COUNT = 0;
+
+  EXPECT_GE(mempurge_count, EXPECTED_MIN_MEMPURGE_COUNT);
+  EXPECT_EQ(sst_count, EXPECTED_SST_COUNT);
+
+  // Verify that the ConditionalUpdateCompactionFilter
+  // updated the values of KEY2 and KEY3, and not KEY4 and KEY5.
+  ASSERT_EQ(Get(KEY1), NOT_FOUND);
+  ASSERT_EQ(Get(KEY2), NEW_VALUE);
+  ASSERT_EQ(Get(KEY3), NEW_VALUE);
+  ASSERT_EQ(Get(KEY4), p_v4);
+  ASSERT_EQ(Get(KEY5), p_v5);
+}
+
+TEST_F(DBFlushTest, MemPurgeWALSupport) {
+  Options options = CurrentOptions();
+
+  options.statistics = CreateDBStatistics();
+  options.statistics->set_stats_level(StatsLevel::kAll);
+  options.create_if_missing = true;
+  options.compression = kNoCompression;
+  options.inplace_update_support = false;
+  options.allow_concurrent_memtable_write = true;
+
+  // Enforce size of a single MemTable to 128KB.
+  options.write_buffer_size = 128 << 10;
+  // Activate the MemPurge prototype.
+  options.experimental_mempurge_threshold = 1.0;
+
+  ASSERT_OK(TryReopen(options));
+
+  const size_t KVSIZE = 10;
+
+  do {
+    CreateAndReopenWithCF({"pikachu"}, options);
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "baz", "v5"));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    ASSERT_EQ("v1", Get(1, "foo"));
+
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+    ASSERT_OK(Put(0, "bar", "v2"));
+    ASSERT_OK(Put(1, "bar", "v2"));
+    ASSERT_OK(Put(1, "foo", "v3"));
+    uint32_t mempurge_count = 0;
+    uint32_t sst_count = 0;
+    ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+        "DBImpl::FlushJob:MemPurgeSuccessful",
+        [&](void* /*arg*/) { mempurge_count++; });
+    ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+        "DBImpl::FlushJob:SSTFileCreated", [&](void* /*arg*/) { sst_count++; });
+    ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+
+    std::vector<std::string> keys;
+    for (size_t k = 0; k < KVSIZE; k++) {
+      keys.push_back("IamKey" + std::to_string(k));
+    }
+
+    std::string RNDKEY, RNDVALUE;
+    const std::string NOT_FOUND = "NOT_FOUND";
+
+    // Heavy overwrite workload,
+    // more than would fit in maximum allowed memtables.
+    Random rnd(719);
+    const size_t NUM_REPEAT = 100;
+    const size_t RAND_KEY_LENGTH = 4096;
+    const size_t RAND_VALUES_LENGTH = 1024;
+    std::vector<std::string> values_default(KVSIZE), values_pikachu(KVSIZE);
+
+    // Insert a very first set of keys that will be
+    // mempurged at least once.
+    for (size_t k = 0; k < KVSIZE / 2; k++) {
+      values_default[k] = rnd.RandomString(RAND_VALUES_LENGTH);
+      values_pikachu[k] = rnd.RandomString(RAND_VALUES_LENGTH);
+    }
+
+    // Insert keys[0:KVSIZE/2] to
+    // both 'default' and 'pikachu' CFs.
+    for (size_t k = 0; k < KVSIZE / 2; k++) {
+      ASSERT_OK(Put(0, keys[k], values_default[k]));
+      ASSERT_OK(Put(1, keys[k], values_pikachu[k]));
+    }
+
+    // Check that the insertion was seamless.
+    for (size_t k = 0; k < KVSIZE / 2; k++) {
+      ASSERT_EQ(Get(0, keys[k]), values_default[k]);
+      ASSERT_EQ(Get(1, keys[k]), values_pikachu[k]);
+    }
+
+    // Insertion of of K-V pairs, multiple times (overwrites)
+    // into 'default' CF. Will trigger mempurge.
+    for (size_t j = 0; j < NUM_REPEAT; j++) {
+      // Create value strings of arbitrary length RAND_VALUES_LENGTH bytes.
+      for (size_t k = KVSIZE / 2; k < KVSIZE; k++) {
+        values_default[k] = rnd.RandomString(RAND_VALUES_LENGTH);
+      }
+
+      // Insert K-V into default CF.
+      for (size_t k = KVSIZE / 2; k < KVSIZE; k++) {
+        ASSERT_OK(Put(0, keys[k], values_default[k]));
+      }
+
+      // Check key validity, for all keys, both in
+      // default and pikachu CFs.
+      for (size_t k = 0; k < KVSIZE; k++) {
+        ASSERT_EQ(Get(0, keys[k]), values_default[k]);
+      }
+      // Note that at this point, only keys[0:KVSIZE/2]
+      // have been inserted into Pikachu.
+      for (size_t k = 0; k < KVSIZE / 2; k++) {
+        ASSERT_EQ(Get(1, keys[k]), values_pikachu[k]);
+      }
+    }
+
+    // Insertion of of K-V pairs, multiple times (overwrites)
+    // into 'pikachu' CF. Will trigger mempurge.
+    // Check that we keep the older logs for 'default' imm().
+    for (size_t j = 0; j < NUM_REPEAT; j++) {
+      // Create value strings of arbitrary length RAND_VALUES_LENGTH bytes.
+      for (size_t k = KVSIZE / 2; k < KVSIZE; k++) {
+        values_pikachu[k] = rnd.RandomString(RAND_VALUES_LENGTH);
+      }
+
+      // Insert K-V into pikachu CF.
+      for (size_t k = KVSIZE / 2; k < KVSIZE; k++) {
+        ASSERT_OK(Put(1, keys[k], values_pikachu[k]));
+      }
+
+      // Check key validity, for all keys,
+      // both in default and pikachu.
+      for (size_t k = 0; k < KVSIZE; k++) {
+        ASSERT_EQ(Get(0, keys[k]), values_default[k]);
+        ASSERT_EQ(Get(1, keys[k]), values_pikachu[k]);
+      }
+    }
+
+    // Check that there was at least one mempurge
+    const uint32_t EXPECTED_MIN_MEMPURGE_COUNT = 1;
+    // Check that there was no SST files created during flush.
+    const uint32_t EXPECTED_SST_COUNT = 0;
+
+    EXPECT_GE(mempurge_count, EXPECTED_MIN_MEMPURGE_COUNT);
+    if (options.experimental_mempurge_threshold ==
+        std::numeric_limits<double>::max()) {
+      EXPECT_EQ(sst_count, EXPECTED_SST_COUNT);
+    }
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    // Check that there was no data corruption anywhere,
+    // not in 'default' nor in 'Pikachu' CFs.
+    ASSERT_EQ("v3", Get(1, "foo"));
+    ASSERT_OK(Put(1, "foo", "v4"));
+    ASSERT_EQ("v4", Get(1, "foo"));
+    ASSERT_EQ("v2", Get(1, "bar"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+    // Check keys in 'Default' and 'Pikachu'.
+    // keys[0:KVSIZE/2] were for sure contained
+    // in the imm() at Reopen/recovery time.
+    for (size_t k = 0; k < KVSIZE; k++) {
+      ASSERT_EQ(Get(0, keys[k]), values_default[k]);
+      ASSERT_EQ(Get(1, keys[k]), values_pikachu[k]);
+    }
+    // Insertion of random K-V pairs to trigger
+    // a flush in the Pikachu CF.
+    for (size_t j = 0; j < NUM_REPEAT; j++) {
+      RNDKEY = rnd.RandomString(RAND_KEY_LENGTH);
+      RNDVALUE = rnd.RandomString(RAND_VALUES_LENGTH);
+      ASSERT_OK(Put(1, RNDKEY, RNDVALUE));
+    }
+    // ASsert than there was at least one flush to storage.
+    EXPECT_GT(sst_count, EXPECTED_SST_COUNT);
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    ASSERT_EQ("v4", Get(1, "foo"));
+    ASSERT_EQ("v2", Get(1, "bar"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+    // Since values in default are held in mutable mem()
+    // and imm(), check if the flush in pikachu didn't
+    // affect these values.
+    for (size_t k = 0; k < KVSIZE; k++) {
+      ASSERT_EQ(Get(0, keys[k]), values_default[k]);
+      ASSERT_EQ(Get(1, keys[k]), values_pikachu[k]);
+    }
+    ASSERT_EQ(Get(1, RNDKEY), RNDVALUE);
+  } while (ChangeWalOptions());
+}
+
 TEST_P(DBFlushDirectIOTest, DirectIO) {
  Options options;
  options.create_if_missing = true;
@@ -1604,7 +2321,7 @@ TEST_P(DBAtomicFlushTest, TriggerFlushAndClose) {
  options.create_if_missing = true;
  options.atomic_flush = atomic_flush;
  options.memtable_factory.reset(
-      new SpecialSkipListFactory(kNumKeysTriggerFlush));
+      test::NewSpecialSkipListFactory(kNumKeysTriggerFlush));
  CreateAndReopenWithCF({"pikachu"}, options);

  for (int i = 0; i != kNumKeysTriggerFlush; ++i) {
@@ -53,6 +53,7 @@
 #include "db/version_set.h"
 #include "db/write_batch_internal.h"
 #include "db/write_callback.h"
+#include "env/unique_id.h"
 #include "file/file_util.h"
 #include "file/filename.h"
 #include "file/random_access_file_reader.h"
@@ -60,8 +61,6 @@
 #include "logging/auto_roll_logger.h"
 #include "logging/log_buffer.h"
 #include "logging/logging.h"
-#include "memtable/hash_linklist_rep.h"
-#include "memtable/hash_skiplist_rep.h"
 #include "monitoring/in_memory_stats_history.h"
 #include "monitoring/instrumented_mutex.h"
 #include "monitoring/iostats_context_imp.h"
@@ -94,14 +93,17 @@
 #include "table/table_builder.h"
 #include "table/two_level_iterator.h"
 #include "test_util/sync_point.h"
+#include "trace_replay/trace_replay.h"
 #include "util/autovector.h"
 #include "util/cast_util.h"
 #include "util/coding.h"
 #include "util/compression.h"
 #include "util/crc32c.h"
+#include "util/defer.h"
 #include "util/mutexlock.h"
 #include "util/stop_watch.h"
 #include "util/string_util.h"
+#include "utilities/trace/replayer_impl.h"

 namespace ROCKSDB_NAMESPACE {

@@ -235,7 +237,8 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
      closed_(false),
      atomic_flush_install_cv_(&mutex_),
      blob_callback_(immutable_db_options_.sst_file_manager.get(), &mutex_,
-                     &error_handler_) {
+                     &error_handler_, &event_logger_,
+                     immutable_db_options_.listeners, dbname_) {
  // !batch_per_trx_ implies seq_per_batch_ because it is only unset for
  // WriteUnprepared, which should use seq_per_batch_.
  assert(batch_per_txn_ || seq_per_batch_);
@@ -549,12 +552,45 @@ Status DBImpl::CloseHelper() {
  flush_scheduler_.Clear();
  trim_history_scheduler_.Clear();

+  // For now, simply trigger a manual flush at close time
+  // on all the column families.
+  // TODO(bjlemaire): Check if this is needed. Also, in the
+  // future we can contemplate doing a more fine-grained
+  // flushing by first checking if there is a need for
+  // flushing (but need to implement something
+  // else than imm()->IsFlushPending() because the output
+  // memtables added to imm() dont trigger flushes).
+  if (immutable_db_options_.experimental_mempurge_threshold > 0.0) {
+    Status flush_ret;
+    mutex_.Unlock();
+    for (ColumnFamilyData* cf : *versions_->GetColumnFamilySet()) {
+      if (immutable_db_options_.atomic_flush) {
+        flush_ret = AtomicFlushMemTables({cf}, FlushOptions(),
+                                         FlushReason::kManualFlush);
+        if (!flush_ret.ok()) {
+          ROCKS_LOG_INFO(
+              immutable_db_options_.info_log,
+              "Atomic flush memtables failed upon closing (mempurge).");
+        }
+      } else {
+        flush_ret =
+            FlushMemTable(cf, FlushOptions(), FlushReason::kManualFlush);
+        if (!flush_ret.ok()) {
+          ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                         "Flush memtables failed upon closing (mempurge).");
+        }
+      }
+    }
+    mutex_.Lock();
+  }
+
  while (!flush_queue_.empty()) {
    const FlushRequest& flush_req = PopFirstFromFlushQueue();
    for (const auto& iter : flush_req) {
      iter.first->UnrefAndTryDelete();
    }
  }
+
  while (!compaction_queue_.empty()) {
    auto cfd = PopFirstFromCompactionQueue();
    cfd->UnrefAndTryDelete();
@@ -607,7 +643,7 @@ Status DBImpl::CloseHelper() {
      ROCKS_LOG_WARN(
          immutable_db_options_.info_log,
          "Unable to Sync WAL file %s with error -- %s",
-          LogFileName(immutable_db_options_.wal_dir, log_number).c_str(),
+          LogFileName(immutable_db_options_.GetWalDir(), log_number).c_str(),
          s.ToString().c_str());
      // Retain the first error
      if (ret.ok()) {
@@ -701,7 +737,8 @@ void DBImpl::MaybeIgnoreError(Status* s) const {
 const Status DBImpl::CreateArchivalDirectory() {
  if (immutable_db_options_.WAL_ttl_seconds > 0 ||
      immutable_db_options_.WAL_size_limit_MB > 0) {
-    std::string archivalPath = ArchivalDirectory(immutable_db_options_.wal_dir);
+    std::string archivalPath =
+        ArchivalDirectory(immutable_db_options_.GetWalDir());
    return env_->CreateDirIfMissing(archivalPath);
  }
  return Status::OK();
@@ -921,8 +958,13 @@ void DBImpl::DumpStats() {
        // Release DB mutex for gathering cache entry stats. Pass over all
        // column families for this first so that other stats are dumped
        // near-atomically.
-        InstrumentedMutexUnlock u(&mutex_);
-        cfd->internal_stats()->CollectCacheEntryStats(/*foreground=*/false);
+        // Get a ref before unlocking
+        cfd->Ref();
+        {
+          InstrumentedMutexUnlock u(&mutex_);
+          cfd->internal_stats()->CollectCacheEntryStats(/*foreground=*/false);
+        }
+        cfd->UnrefAndTryDelete();
      }
    }

@@ -1103,9 +1145,19 @@ Status DBImpl::SetDBOptions(
    InstrumentedMutexLock l(&mutex_);
    s = GetMutableDBOptionsFromStrings(mutable_db_options_, options_map,
                                       &new_options);
+
    if (new_options.bytes_per_sync == 0) {
      new_options.bytes_per_sync = 1024 * 1024;
    }
+
+    if (MutableDBOptionsAreEqual(mutable_db_options_, new_options)) {
+      ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                     "SetDBOptions(), input option value is not changed, "
+                     "skipping updating.");
+      persist_options_status.PermitUncheckedError();
+      return s;
+    }
+
    DBOptions new_db_options =
        BuildDBOptions(immutable_db_options_, new_options);
    if (s.ok()) {
@@ -1751,6 +1803,8 @@ Status DBImpl::GetImpl(const ReadOptions& read_options, const Slice& key,
  }
  // If timestamp is used, we use read callback to ensure <key,t,s> is returned
  // only if t <= read_opts.timestamp and s <= snapshot.
+  // HACK: temporarily overwrite input struct field but restore
+  SaveAndRestore<ReadCallback*> restore_callback(&get_impl_options.callback);
  if (ts_sz > 0) {
    assert(!get_impl_options
                .callback);  // timestamp with callback is not supported
@@ -2108,7 +2162,7 @@ bool DBImpl::MultiCFSnapshot(
    // consecutive retries, it means the write rate is very high. In that case
    // its probably ok to take the mutex on the 3rd try so we can succeed for
    // sure
-    static const int num_retries = 3;
+    constexpr int num_retries = 3;
    for (int i = 0; i < num_retries; ++i) {
      last_try = (i == num_retries - 1);
      bool retry = false;
@@ -2138,8 +2192,9 @@ bool DBImpl::MultiCFSnapshot(
          *snapshot = versions_->LastPublishedSequence();
        }
      } else {
-        *snapshot = reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
-                        ->number_;
+        *snapshot =
+            static_cast_with_check<const SnapshotImpl>(read_options.snapshot)
+                ->number_;
      }
      for (auto cf_iter = cf_list->begin(); cf_iter != cf_list->end();
           ++cf_iter) {
@@ -2243,20 +2298,18 @@ void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
      multiget_cf_data;
  size_t cf_start = 0;
  ColumnFamilyHandle* cf = sorted_keys[0]->column_family;
+
  for (size_t i = 0; i < num_keys; ++i) {
    KeyContext* key_ctx = sorted_keys[i];
    if (key_ctx->column_family != cf) {
-      multiget_cf_data.emplace_back(
-          MultiGetColumnFamilyData(cf, cf_start, i - cf_start, nullptr));
+      multiget_cf_data.emplace_back(cf, cf_start, i - cf_start, nullptr);
      cf_start = i;
      cf = key_ctx->column_family;
    }
  }
-  {
-    // multiget_cf_data.emplace_back(
-    // MultiGetColumnFamilyData(cf, cf_start, num_keys - cf_start, nullptr));
-    multiget_cf_data.emplace_back(cf, cf_start, num_keys - cf_start, nullptr);
-  }
+
+  multiget_cf_data.emplace_back(cf, cf_start, num_keys - cf_start, nullptr);
+
  std::function<MultiGetColumnFamilyData*(
      autovector<MultiGetColumnFamilyData,
                 MultiGetContext::MAX_BATCH_SIZE>::iterator&)>
@@ -2316,7 +2369,7 @@ struct CompareKeyContext {
        static_cast<ColumnFamilyHandleImpl*>(lhs->column_family);
    uint32_t cfd_id1 = cfh->cfd()->GetID();
    const Comparator* comparator = cfh->cfd()->user_comparator();
-    cfh = static_cast<ColumnFamilyHandleImpl*>(lhs->column_family);
+    cfh = static_cast<ColumnFamilyHandleImpl*>(rhs->column_family);
    uint32_t cfd_id2 = cfh->cfd()->GetID();

    if (cfd_id1 < cfd_id2) {
@@ -2340,39 +2393,16 @@ struct CompareKeyContext {
 void DBImpl::PrepareMultiGetKeys(
    size_t num_keys, bool sorted_input,
    autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys) {
-#ifndef NDEBUG
  if (sorted_input) {
-    for (size_t index = 0; index < sorted_keys->size(); ++index) {
-      if (index > 0) {
-        KeyContext* lhs = (*sorted_keys)[index - 1];
-        KeyContext* rhs = (*sorted_keys)[index];
-        ColumnFamilyHandleImpl* cfh =
-            static_cast_with_check<ColumnFamilyHandleImpl>(lhs->column_family);
-        uint32_t cfd_id1 = cfh->cfd()->GetID();
-        const Comparator* comparator = cfh->cfd()->user_comparator();
-        cfh =
-            static_cast_with_check<ColumnFamilyHandleImpl>(lhs->column_family);
-        uint32_t cfd_id2 = cfh->cfd()->GetID();
-
-        assert(cfd_id1 <= cfd_id2);
-        if (cfd_id1 < cfd_id2) {
-          continue;
-        }
-
-        // Both keys are from the same column family
-        int cmp = comparator->CompareWithoutTimestamp(
-            *(lhs->key), /*a_has_ts=*/false, *(rhs->key), /*b_has_ts=*/false);
-        assert(cmp <= 0);
-      }
-      index++;
-    }
-  }
+#ifndef NDEBUG
+    assert(std::is_sorted(sorted_keys->begin(), sorted_keys->end(),
+                          CompareKeyContext()));
 #endif
-  if (!sorted_input) {
-    CompareKeyContext sort_comparator;
-    std::sort(sorted_keys->begin(), sorted_keys->begin() + num_keys,
-              sort_comparator);
+    return;
  }
+
+  std::sort(sorted_keys->begin(), sorted_keys->begin() + num_keys,
+            CompareKeyContext());
 }

 void DBImpl::MultiGet(const ReadOptions& read_options,
@@ -3078,7 +3108,7 @@ SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
 }

 namespace {
-typedef autovector<ColumnFamilyData*, 2> CfdList;
+using CfdList = autovector<ColumnFamilyData*, 2>;
 bool CfdListContains(const CfdList& list, ColumnFamilyData* cfd) {
  for (const ColumnFamilyData* t : list) {
    if (t == cfd) {
@@ -3799,6 +3829,17 @@ void DBImpl::GetColumnFamilyMetaData(ColumnFamilyHandle* column_family,
  ReturnAndCleanupSuperVersion(cfd, sv);
 }

+void DBImpl::GetAllColumnFamilyMetaData(
+    std::vector<ColumnFamilyMetaData>* metadata) {
+  InstrumentedMutexLock l(&mutex_);
+  for (auto cfd : *(versions_->GetColumnFamilySet())) {
+    {
+      metadata->emplace_back();
+      cfd->current()->GetColumnFamilyMetaData(&metadata->back());
+    }
+  }
+}
+
 #endif  // ROCKSDB_LITE

 Status DBImpl::CheckConsistency() {
@@ -3890,7 +3931,8 @@ Status DBImpl::GetDbIdentityFromIdentityFile(std::string* identity) const {
    return s;
  }

-  // If last character is '\n' remove it from identity
+  // If last character is '\n' remove it from identity. (Old implementations
+  // of Env::GenerateUniqueId() would include a trailing '\n'.)
  if (identity->size() > 0 && identity->back() == '\n') {
    identity->pop_back();
  }
@@ -3902,10 +3944,11 @@ Status DBImpl::GetDbSessionId(std::string& session_id) const {
  return Status::OK();
 }

-void DBImpl::SetDbSessionId() {
-  // GenerateUniqueId() generates an identifier that has a negligible
-  // probability of being duplicated, ~128 bits of entropy
-  std::string uuid = env_->GenerateUniqueId();
+std::string DBImpl::GenerateDbSessionId(Env*) {
+  // GenerateRawUniqueId() generates an identifier that has a negligible
+  // probability of being duplicated. It should have full 128 bits of entropy.
+  uint64_t a, b;
+  GenerateRawUniqueId(&a, &b);

  // Hash and reformat that down to a more compact format, 20 characters
  // in base-36 ([0-9A-Z]), which is ~103 bits of entropy, which is enough
@@ -3914,17 +3957,15 @@ void DBImpl::SetDbSessionId() {
  // * Save ~ dozen bytes per SST file
  // * Shorter shared backup file names (some platforms have low limits)
  // * Visually distinct from DB id format
-  uint64_t a = NPHash64(uuid.data(), uuid.size(), 1234U);
-  uint64_t b = NPHash64(uuid.data(), uuid.size(), 5678U);
-  db_session_id_.resize(20);
-  static const char* const base36 = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
-  size_t i = 0;
-  for (; i < 10U; ++i, a /= 36U) {
-    db_session_id_[i] = base36[a % 36];
-  }
-  for (; i < 20U; ++i, b /= 36U) {
-    db_session_id_[i] = base36[b % 36];
-  }
+  std::string db_session_id(20U, '\0');
+  char* buf = &db_session_id[0];
+  PutBaseChars<36>(&buf, 10, a, /*uppercase*/ true);
+  PutBaseChars<36>(&buf, 10, b, /*uppercase*/ true);
+  return db_session_id;
+}
+
+void DBImpl::SetDbSessionId() {
+  db_session_id_ = GenerateDbSessionId(env_);
  TEST_SYNC_POINT_CALLBACK("DBImpl::SetDbSessionId", &db_session_id_);
 }

@@ -3994,7 +4035,7 @@ Status DestroyDB(const std::string& dbname, const Options& options,
  ImmutableDBOptions soptions(SanitizeOptions(dbname, options));
  Env* env = soptions.env;
  std::vector<std::string> filenames;
-  bool wal_in_db_path = IsWalDirSameAsDBPath(&soptions);
+  bool wal_in_db_path = soptions.IsWalDirSameAsDBPath();

  // Reset the logger because it holds a handle to the
  // log file and prevents cleanup and directory removal
@@ -4060,7 +4101,7 @@ Status DestroyDB(const std::string& dbname, const Options& options,
    std::vector<std::string> walDirFiles;
    std::string archivedir = ArchivalDirectory(dbname);
    bool wal_dir_exists = false;
-    if (dbname != soptions.wal_dir) {
+    if (!soptions.IsWalDirSameAsDBPath(dbname)) {
      wal_dir_exists = env->GetChildren(soptions.wal_dir, &walDirFiles).ok();
      archivedir = ArchivalDirectory(soptions.wal_dir);
    }
@@ -4150,6 +4191,8 @@ Status DBImpl::WriteOptionsFile(bool need_mutex_lock,

  TEST_SYNC_POINT("DBImpl::WriteOptionsFile:1");
  TEST_SYNC_POINT("DBImpl::WriteOptionsFile:2");
+  TEST_SYNC_POINT_CALLBACK("DBImpl::WriteOptionsFile:PersistOptions",
+                           &db_options);

  std::string file_name =
      TempOptionsFileName(GetName(), versions_->NewFileNumber());
@@ -4316,9 +4359,7 @@ SequenceNumber DBImpl::GetEarliestMemTableSequenceNumber(SuperVersion* sv,

  return earliest_seq;
 }
-#endif  // ROCKSDB_LITE

-#ifndef ROCKSDB_LITE
 Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
                                       bool cache_only,
                                       SequenceNumber lower_bound_seq,
@@ -4638,14 +4679,11 @@ Status DBImpl::IngestExternalFiles(
    if (status.ok()) {
      int consumed_seqno_count =
          ingestion_jobs[0].ConsumedSequenceNumbersCount();
-#ifndef NDEBUG
      for (size_t i = 1; i != num_cfs; ++i) {
-        assert(!!consumed_seqno_count ==
-               !!ingestion_jobs[i].ConsumedSequenceNumbersCount());
-        consumed_seqno_count +=
-            ingestion_jobs[i].ConsumedSequenceNumbersCount();
+        consumed_seqno_count =
+            std::max(consumed_seqno_count,
+                     ingestion_jobs[i].ConsumedSequenceNumbersCount());
      }
-#endif
      if (consumed_seqno_count > 0) {
        const SequenceNumber last_seqno = versions_->LastSequence();
        versions_->SetLastAllocatedSequence(last_seqno + consumed_seqno_count);
@@ -4887,6 +4925,11 @@ Status DBImpl::VerifyChecksum(const ReadOptions& read_options) {

 Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
                                      bool use_file_checksum) {
+  // `bytes_read` stat is enabled based on compile-time support and cannot
+  // be dynamically toggled. So we do not need to worry about `PerfLevel`
+  // here, unlike many other `IOStatsContext` / `PerfContext` stats.
+  uint64_t prev_bytes_read = IOSTATS(bytes_read);
+
  Status s;

  if (use_file_checksum) {
@@ -4941,6 +4984,9 @@ Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
          s = ROCKSDB_NAMESPACE::VerifySstFileChecksum(opts, file_options_,
                                                       read_options, fname);
        }
+        RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
+                   IOSTATS(bytes_read) - prev_bytes_read);
+        prev_bytes_read = IOSTATS(bytes_read);
      }
    }

@@ -4955,6 +5001,9 @@ Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
        s = VerifyFullFileChecksum(meta->GetChecksumValue(),
                                   meta->GetChecksumMethod(), blob_file_name,
                                   read_options);
+        RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
+                   IOSTATS(bytes_read) - prev_bytes_read);
+        prev_bytes_read = IOSTATS(bytes_read);
        if (!s.ok()) {
          break;
        }
@@ -4986,6 +5035,8 @@ Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
      cfd->UnrefAndTryDelete();
    }
  }
+  RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
+             IOSTATS(bytes_read) - prev_bytes_read);
  return s;
 }

@@ -5060,11 +5111,19 @@ Status DBImpl::EndTrace() {
    s = tracer_->Close();
    tracer_.reset();
  } else {
-    return Status::IOError("No trace file to close");
+    s = Status::IOError("No trace file to close");
  }
  return s;
 }

+Status DBImpl::NewDefaultReplayer(
+    const std::vector<ColumnFamilyHandle*>& handles,
+    std::unique_ptr<TraceReader>&& reader,
+    std::unique_ptr<Replayer>* replayer) {
+  replayer->reset(new ReplayerImpl(this, handles, std::move(reader)));
+  return Status::OK();
+}
+
 Status DBImpl::StartBlockCacheTrace(
    const TraceOptions& trace_options,
    std::unique_ptr<TraceWriter>&& trace_writer) {
@@ -20,6 +20,7 @@
 #include <vector>

 #include "db/column_family.h"
+#include "db/compaction/compaction_iterator.h"
 #include "db/compaction/compaction_job.h"
 #include "db/dbformat.h"
 #include "db/error_handler.h"
@@ -50,9 +51,15 @@
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/status.h"
+#ifndef ROCKSDB_LITE
 #include "rocksdb/trace_reader_writer.h"
+#endif  // ROCKSDB_LITE
 #include "rocksdb/transaction_log.h"
+#ifndef ROCKSDB_LITE
+#include "rocksdb/utilities/replayer.h"
+#endif  // ROCKSDB_LITE
 #include "rocksdb/write_buffer_manager.h"
+#include "table/merging_iterator.h"
 #include "table/scoped_arena_iterator.h"
 #include "util/autovector.h"
 #include "util/hash.h"
@@ -400,12 +407,13 @@ class DBImpl : public DB {
      FileChecksumList* checksum_list) override;

  // Obtains the meta data of the specified column family of the DB.
-  // Status::NotFound() will be returned if the current DB does not have
-  // any column family match the specified name.
  // TODO(yhchiang): output parameter is placed in the end in this codebase.
  virtual void GetColumnFamilyMetaData(ColumnFamilyHandle* column_family,
                                       ColumnFamilyMetaData* metadata) override;

+  void GetAllColumnFamilyMetaData(
+      std::vector<ColumnFamilyMetaData>* metadata) override;
+
  Status SuggestCompactRange(ColumnFamilyHandle* column_family,
                             const Slice* begin, const Slice* end) override;

@@ -461,6 +469,12 @@ class DBImpl : public DB {
  using DB::EndTrace;
  virtual Status EndTrace() override;

+  using DB::NewDefaultReplayer;
+  virtual Status NewDefaultReplayer(
+      const std::vector<ColumnFamilyHandle*>& handles,
+      std::unique_ptr<TraceReader>&& reader,
+      std::unique_ptr<Replayer>* replayer) override;
+
  using DB::StartBlockCacheTrace;
  Status StartBlockCacheTrace(
      const TraceOptions& options,
@@ -962,7 +976,7 @@ class DBImpl : public DB {

  // Return the maximum overlapping data (in bytes) at next level for any
  // file at a level >= 1.
-  int64_t TEST_MaxNextLevelOverlappingBytes(
+  uint64_t TEST_MaxNextLevelOverlappingBytes(
      ColumnFamilyHandle* column_family = nullptr);

  // Return the current manifest file no.
@@ -974,8 +988,10 @@ class DBImpl : public DB {
  // get total level0 file size. Only for testing.
  uint64_t TEST_GetLevel0TotalSize();

-  void TEST_GetFilesMetaData(ColumnFamilyHandle* column_family,
-                             std::vector<std::vector<FileMetaData>>* metadata);
+  void TEST_GetFilesMetaData(
+      ColumnFamilyHandle* column_family,
+      std::vector<std::vector<FileMetaData>>* metadata,
+      std::vector<std::shared_ptr<BlobFileMetaData>>* blob_metadata = nullptr);

  void TEST_LockMutex();

@@ -1083,8 +1099,10 @@ class DBImpl : public DB {
    // Called from WriteBufferManager. This function changes the state_
    // to State::RUNNING indicating the stall is cleared and DB can proceed.
    void Signal() override {
-      MutexLock lock(&state_mutex_);
-      state_ = State::RUNNING;
+      {
+        MutexLock lock(&state_mutex_);
+        state_ = State::RUNNING;
+      }
      state_cv_.Signal();
    }

@@ -1098,6 +1116,8 @@ class DBImpl : public DB {
    State state_;
  };

+  static std::string GenerateDbSessionId(Env* env);
+
 protected:
  const std::string dbname_;
  std::string db_id_;
@@ -1184,11 +1204,6 @@ class DBImpl : public DB {
                              ColumnFamilyHandle* column_family,
                              const Slice* begin, const Slice* end);

-  Status GetApproximateSizesInternal(const SizeApproximationOptions& options,
-                                     ColumnFamilyHandle* column_family,
-                                     const Range* range, int n,
-                                     uint64_t* sizes);
-
  // The following two functions can only be called when:
  // 1. WriteThread::Writer::EnterUnbatched() is used.
  // 2. db_mutex is NOT held
@@ -1740,7 +1755,7 @@ class DBImpl : public DB {
  // specified value, this flush request is considered to have completed its
  // work of flushing this column family. After completing the work for all
  // column families in this request, this flush is considered complete.
-  typedef std::vector<std::pair<ColumnFamilyData*, uint64_t>> FlushRequest;
+  using FlushRequest = std::vector<std::pair<ColumnFamilyData*, uint64_t>>;

  void GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
                            FlushRequest* req);
@@ -2002,7 +2017,7 @@ class DBImpl : public DB {
  // accessed from the same write_thread_ without any locks. With
  // two_write_queues writes, where it can be updated in different threads,
  // read and writes are protected by log_write_mutex_ instead. This is to avoid
-  // expesnive mutex_ lock during WAL write, which update log_empty_.
+  // expensive mutex_ lock during WAL write, which update log_empty_.
  bool log_empty_;

  ColumnFamilyHandleImpl* persist_stats_cf_handle_;
@@ -7,6 +7,7 @@
 // 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 <cinttypes>
+#include <deque>

 #include "db/builder.h"
 #include "db/db_impl/db_impl.h"
@@ -151,7 +152,6 @@ Status DBImpl::FlushMemTableToOutputFile(
  assert(cfd);
  assert(cfd->imm()->NumNotFlushed() != 0);
  assert(cfd->imm()->IsFlushPending());
-
  FlushJob flush_job(
      dbname_, cfd, immutable_db_options_, mutable_cf_options,
      port::kMaxUint64 /* memtable_id */, file_options_for_compaction_,
@@ -199,7 +199,7 @@ Status DBImpl::FlushMemTableToOutputFile(
    need_cancel = true;
  }
  TEST_SYNC_POINT("DBImpl::FlushMemTableToOutputFile:AfterPickMemtables");
-
+  bool switched_to_mempurge = false;
  // Within flush_job.Run, rocksdb may call event listener to notify
  // file creation and deletion.
  //
@@ -207,7 +207,8 @@ Status DBImpl::FlushMemTableToOutputFile(
  // and EventListener callback will be called when the db_mutex
  // is unlocked by the current thread.
  if (s.ok()) {
-    s = flush_job.Run(&logs_with_prep_tracker_, &file_meta);
+    s = flush_job.Run(&logs_with_prep_tracker_, &file_meta,
+                      &switched_to_mempurge);
    need_cancel = false;
  }

@@ -283,7 +284,9 @@ Status DBImpl::FlushMemTableToOutputFile(
    // from never needing it or ignoring the flush job status
    io_s.PermitUncheckedError();
  }
-  if (s.ok()) {
+  // If flush ran smoothly and no mempurge happened
+  // install new SST file path.
+  if (s.ok() && (!switched_to_mempurge)) {
 #ifndef ROCKSDB_LITE
    // may temporarily unlock and lock the mutex.
    NotifyOnFlushCompleted(cfd, mutable_cf_options,
@@ -408,10 +411,13 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
        stats_, &event_logger_, mutable_cf_options.report_bg_io_stats,
        false /* sync_output_directory */, false /* write_manifest */,
        thread_pri, io_tracer_, db_id_, db_session_id_,
-        cfd->GetFullHistoryTsLow()));
+        cfd->GetFullHistoryTsLow(), &blob_callback_));
  }

  std::vector<FileMetaData> file_meta(num_cfs);
+  // Use of deque<bool> because vector<bool>
+  // is specific and doesn't allow &v[i].
+  std::deque<bool> switched_to_mempurge(num_cfs, false);
  Status s;
  IOStatus log_io_s = IOStatus::OK();
  assert(num_cfs == static_cast<int>(jobs.size()));
@@ -461,10 +467,13 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
  }

  if (s.ok()) {
+    assert(switched_to_mempurge.size() ==
+           static_cast<long unsigned int>(num_cfs));
    // TODO (yanqin): parallelize jobs with threads.
    for (int i = 1; i != num_cfs; ++i) {
      exec_status[i].second =
-          jobs[i]->Run(&logs_with_prep_tracker_, &file_meta[i]);
+          jobs[i]->Run(&logs_with_prep_tracker_, &file_meta[i],
+                       &(switched_to_mempurge.at(i)));
      exec_status[i].first = true;
      io_status[i] = jobs[i]->io_status();
    }
@@ -476,8 +485,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
    }
    assert(exec_status.size() > 0);
    assert(!file_meta.empty());
-    exec_status[0].second =
-        jobs[0]->Run(&logs_with_prep_tracker_, &file_meta[0]);
+    exec_status[0].second = jobs[0]->Run(
+        &logs_with_prep_tracker_, file_meta.data() /* &file_meta[0] */,
+        switched_to_mempurge.empty() ? nullptr : &(switched_to_mempurge.at(0)));
    exec_status[0].first = true;
    io_status[0] = jobs[0]->io_status();

@@ -591,6 +601,8 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
    autovector<const autovector<MemTable*>*> mems_list;
    autovector<const MutableCFOptions*> mutable_cf_options_list;
    autovector<FileMetaData*> tmp_file_meta;
+    autovector<std::list<std::unique_ptr<FlushJobInfo>>*>
+        committed_flush_jobs_info;
    for (int i = 0; i != num_cfs; ++i) {
      const auto& mems = jobs[i]->GetMemTables();
      if (!cfds[i]->IsDropped() && !mems.empty()) {
@@ -598,13 +610,18 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
        mems_list.emplace_back(&mems);
        mutable_cf_options_list.emplace_back(&all_mutable_cf_options[i]);
        tmp_file_meta.emplace_back(&file_meta[i]);
+#ifndef ROCKSDB_LITE
+        committed_flush_jobs_info.emplace_back(
+            jobs[i]->GetCommittedFlushJobsInfo());
+#endif  //! ROCKSDB_LITE
      }
    }

    s = InstallMemtableAtomicFlushResults(
        nullptr /* imm_lists */, tmp_cfds, mutable_cf_options_list, mems_list,
        versions_.get(), &logs_with_prep_tracker_, &mutex_, tmp_file_meta,
-        &job_context->memtables_to_free, directories_.GetDbDir(), log_buffer);
+        committed_flush_jobs_info, &job_context->memtables_to_free,
+        directories_.GetDbDir(), log_buffer);
  }

  if (s.ok()) {
@@ -650,6 +667,11 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
        immutable_db_options_.sst_file_manager.get());
    assert(all_mutable_cf_options.size() == static_cast<size_t>(num_cfs));
    for (int i = 0; s.ok() && i != num_cfs; ++i) {
+      // If mempurge happened instead of Flush,
+      // no NotifyOnFlushCompleted call (no SST file created).
+      if (switched_to_mempurge[i]) {
+        continue;
+      }
      if (cfds[i]->IsDropped()) {
        continue;
      }
@@ -1258,7 +1280,7 @@ Status DBImpl::CompactFilesImpl(
      c->mutable_cf_options()->report_bg_io_stats, dbname_,
      &compaction_job_stats, Env::Priority::USER, io_tracer_,
      &manual_compaction_paused_, nullptr, db_id_, db_session_id_,
-      c->column_family_data()->GetFullHistoryTsLow());
+      c->column_family_data()->GetFullHistoryTsLow(), &blob_callback_);

  // Creating a compaction influences the compaction score because the score
  // takes running compactions into account (by skipping files that are already
@@ -1331,10 +1353,15 @@ Status DBImpl::CompactFilesImpl(

  if (output_file_names != nullptr) {
    for (const auto& newf : c->edit()->GetNewFiles()) {
-      (*output_file_names)
-          .push_back(TableFileName(c->immutable_options()->cf_paths,
-                                   newf.second.fd.GetNumber(),
-                                   newf.second.fd.GetPathId()));
+      output_file_names->push_back(TableFileName(
+          c->immutable_options()->cf_paths, newf.second.fd.GetNumber(),
+          newf.second.fd.GetPathId()));
+    }
+
+    for (const auto& blob_file : c->edit()->GetBlobFileAdditions()) {
+      output_file_names->push_back(
+          BlobFileName(c->immutable_options()->cf_paths.front().path,
+                       blob_file.GetBlobFileNumber()));
    }
  }

@@ -1395,6 +1422,8 @@ void DBImpl::NotifyOnCompactionBegin(ColumnFamilyData* cfd, Compaction* c,
      manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
    return;
  }
+
+  c->SetNotifyOnCompactionCompleted();
  Version* current = cfd->current();
  current->Ref();
  // release lock while notifying events
@@ -1430,10 +1459,8 @@ void DBImpl::NotifyOnCompactionCompleted(
  if (shutting_down_.load(std::memory_order_acquire)) {
    return;
  }
-  // TODO: Should disabling manual compaction squash compaction completed
-  //   notifications that aren't the result of a shutdown?
-  if (c->is_manual_compaction() &&
-      manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
+
+  if (c->ShouldNotifyOnCompactionCompleted() == false) {
    return;
  }

@@ -1687,8 +1714,11 @@ Status DBImpl::RunManualCompaction(

  // When a manual compaction arrives, temporarily disable scheduling of
  // non-manual compactions and wait until the number of scheduled compaction
-  // jobs drops to zero. This is needed to ensure that this manual compaction
-  // can compact any range of keys/files.
+  // jobs drops to zero. This used to be needed to ensure that this manual
+  // compaction can compact any range of keys/files. Now it is optional
+  // (see `CompactRangeOptions::exclusive_manual_compaction`). The use case for
+  // `exclusive_manual_compaction=true` (the default) is unclear beyond not
+  // trusting the new code.
  //
  // HasPendingManualCompaction() is true when at least one thread is inside
  // RunManualCompaction(), i.e. during that time no other compaction will
@@ -1702,8 +1732,20 @@ Status DBImpl::RunManualCompaction(
  AddManualCompaction(&manual);
  TEST_SYNC_POINT_CALLBACK("DBImpl::RunManualCompaction:NotScheduled", &mutex_);
  if (exclusive) {
+    // Limitation: there's no way to wake up the below loop when user sets
+    // `*manual.canceled`. So `CompactRangeOptions::exclusive_manual_compaction`
+    // and `CompactRangeOptions::canceled` might not work well together.
    while (bg_bottom_compaction_scheduled_ > 0 ||
           bg_compaction_scheduled_ > 0) {
+      if (manual_compaction_paused_ > 0 ||
+          (manual.canceled != nullptr && *manual.canceled == true)) {
+        // Pretend the error came from compaction so the below cleanup/error
+        // handling code can process it.
+        manual.done = true;
+        manual.status =
+            Status::Incomplete(Status::SubCode::kManualCompactionPaused);
+        break;
+      }
      TEST_SYNC_POINT("DBImpl::RunManualCompaction:WaitScheduled");
      ROCKS_LOG_INFO(
          immutable_db_options_.info_log,
@@ -2196,6 +2238,10 @@ Status DBImpl::EnableAutoCompaction(
 void DBImpl::DisableManualCompaction() {
  InstrumentedMutexLock l(&mutex_);
  manual_compaction_paused_.fetch_add(1, std::memory_order_release);
+
+  // Wake up manual compactions waiting to start.
+  bg_cv_.SignalAll();
+
  // Wait for any pending manual compactions to finish (typically through
  // failing with `Status::Incomplete`) prior to returning. This way we are
  // guaranteed no pending manual compaction will commit while manual
@@ -2398,6 +2444,17 @@ void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
    assert(flush_req.size() == 1);
    ColumnFamilyData* cfd = flush_req[0].first;
    assert(cfd);
+    // Note: SchedulePendingFlush is always preceded
+    // with an imm()->FlushRequested() call. However,
+    // we want to make this code snipper more resilient to
+    // future changes. Therefore, we add the following if
+    // statement - note that calling it twice (or more)
+    // doesn't break anything.
+    if (immutable_db_options_.experimental_mempurge_threshold > 0.0) {
+      // If imm() contains silent memtables,
+      // requesting a flush will mark the imm_needed as true.
+      cfd->imm()->FlushRequested();
+    }
    if (!cfd->queued_for_flush() && cfd->imm()->IsFlushPending()) {
      cfd->Ref();
      cfd->set_queued_for_flush(true);
@@ -2539,6 +2596,11 @@ Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,

    for (const auto& iter : flush_req) {
      ColumnFamilyData* cfd = iter.first;
+      if (immutable_db_options_.experimental_mempurge_threshold > 0.0) {
+        // If imm() contains silent memtables,
+        // requesting a flush will mark the imm_needed as true.
+        cfd->imm()->FlushRequested();
+      }
      if (cfd->IsDropped() || !cfd->imm()->IsFlushPending()) {
        // can't flush this CF, try next one
        column_families_not_to_flush.push_back(cfd);
@@ -2770,9 +2832,11 @@ void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,

    if (prepicked_compaction != nullptr &&
        prepicked_compaction->task_token != nullptr) {
-      // Releasing task tokens affects the DB state, so must be done before we
-      // potentially signal the DB close process to proceed below.
-      prepicked_compaction->task_token->ReleaseOnce();
+      // Releasing task tokens affects (and asserts on) the DB state, so
+      // must be done before we potentially signal the DB close process to
+      // proceed below.
+      prepicked_compaction->task_token.reset();
+      ;
    }

    if (made_progress ||
@@ -3153,7 +3217,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
        &compaction_job_stats, thread_pri, io_tracer_,
        is_manual ? &manual_compaction_paused_ : nullptr,
        is_manual ? manual_compaction->canceled : nullptr, db_id_,
-        db_session_id_, c->column_family_data()->GetFullHistoryTsLow());
+        db_session_id_, c->column_family_data()->GetFullHistoryTsLow(),
+        &blob_callback_);
    compaction_job.Prepare();

    NotifyOnCompactionBegin(c->column_family_data(), c.get(), status,
@@ -3420,6 +3485,30 @@ void DBImpl::BuildCompactionJobInfo(
    compaction_job_info->output_file_infos.push_back(CompactionFileInfo{
        newf.first, file_number, meta.oldest_blob_file_number});
  }
+  compaction_job_info->blob_compression_type =
+      c->mutable_cf_options()->blob_compression_type;
+
+  // Update BlobFilesInfo.
+  for (const auto& blob_file : c->edit()->GetBlobFileAdditions()) {
+    BlobFileAdditionInfo blob_file_addition_info(
+        BlobFileName(c->immutable_options()->cf_paths.front().path,
+                     blob_file.GetBlobFileNumber()) /*blob_file_path*/,
+        blob_file.GetBlobFileNumber(), blob_file.GetTotalBlobCount(),
+        blob_file.GetTotalBlobBytes());
+    compaction_job_info->blob_file_addition_infos.emplace_back(
+        std::move(blob_file_addition_info));
+  }
+
+  // Update BlobFilesGarbageInfo.
+  for (const auto& blob_file : c->edit()->GetBlobFileGarbages()) {
+    BlobFileGarbageInfo blob_file_garbage_info(
+        BlobFileName(c->immutable_options()->cf_paths.front().path,
+                     blob_file.GetBlobFileNumber()) /*blob_file_path*/,
+        blob_file.GetBlobFileNumber(), blob_file.GetGarbageBlobCount(),
+        blob_file.GetGarbageBlobBytes());
+    compaction_job_info->blob_file_garbage_infos.emplace_back(
+        std::move(blob_file_garbage_info));
+  }
 }
 #endif

@@ -3452,7 +3541,7 @@ void DBImpl::InstallSuperVersionAndScheduleWork(
  if (UNLIKELY(sv_context->new_superversion == nullptr)) {
    sv_context->NewSuperVersion();
  }
-  cfd->InstallSuperVersion(sv_context, &mutex_, mutable_cf_options);
+  cfd->InstallSuperVersion(sv_context, mutable_cf_options);

  // There may be a small data race here. The snapshot tricking bottommost
  // compaction may already be released here. But assuming there will always be
@@ -43,7 +43,7 @@ bool DBImpl::TEST_WALBufferIsEmpty(bool lock) {
  return res;
 }

-int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
+uint64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
    ColumnFamilyHandle* column_family) {
  ColumnFamilyData* cfd;
  if (column_family == nullptr) {
@@ -58,7 +58,8 @@ int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(

 void DBImpl::TEST_GetFilesMetaData(
    ColumnFamilyHandle* column_family,
-    std::vector<std::vector<FileMetaData>>* metadata) {
+    std::vector<std::vector<FileMetaData>>* metadata,
+    std::vector<std::shared_ptr<BlobFileMetaData>>* blob_metadata) {
  auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
  auto cfd = cfh->cfd();
  InstrumentedMutexLock l(&mutex_);
@@ -72,6 +73,12 @@ void DBImpl::TEST_GetFilesMetaData(
      (*metadata)[level].push_back(*f);
    }
  }
+  if (blob_metadata != nullptr) {
+    blob_metadata->clear();
+    for (const auto& blob : cfd->current()->storage_info()->GetBlobFiles()) {
+      blob_metadata->push_back(blob.second);
+    }
+  }
 }

 uint64_t DBImpl::TEST_Current_Manifest_FileNo() {
@@ -38,20 +38,26 @@ uint64_t DBImpl::MinObsoleteSstNumberToKeep() {
 }

 Status DBImpl::DisableFileDeletions() {
-  InstrumentedMutexLock l(&mutex_);
-  return DisableFileDeletionsWithLock();
+  Status s;
+  int my_disable_delete_obsolete_files;
+  {
+    InstrumentedMutexLock l(&mutex_);
+    s = DisableFileDeletionsWithLock();
+    my_disable_delete_obsolete_files = disable_delete_obsolete_files_;
+  }
+  if (my_disable_delete_obsolete_files == 1) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log, "File Deletions Disabled");
+  } else {
+    ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                   "File Deletions Disabled, but already disabled. Counter: %d",
+                   my_disable_delete_obsolete_files);
+  }
+  return s;
 }

 Status DBImpl::DisableFileDeletionsWithLock() {
  mutex_.AssertHeld();
  ++disable_delete_obsolete_files_;
-  if (disable_delete_obsolete_files_ == 1) {
-    ROCKS_LOG_INFO(immutable_db_options_.info_log, "File Deletions Disabled");
-  } else {
-    ROCKS_LOG_WARN(immutable_db_options_.info_log,
-                   "File Deletions Disabled, but already disabled. Counter: %d",
-                   disable_delete_obsolete_files_);
-  }
  return Status::OK();
 }

@@ -215,7 +221,8 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
    }

    // Add log files in wal_dir
-    if (immutable_db_options_.wal_dir != dbname_) {
+
+    if (!immutable_db_options_.IsWalDirSameAsDBPath(dbname_)) {
      std::vector<std::string> log_files;
      Status s = env_->GetChildren(immutable_db_options_.wal_dir, &log_files);
      s.PermitUncheckedError();  // TODO: What should we do on error?
@@ -311,7 +318,7 @@ bool CompareCandidateFile(const JobContext::CandidateFileInfo& first,
    return (first.file_path > second.file_path);
  }
 }
-};  // namespace
+}  // namespace

 // Delete obsolete files and log status and information of file deletion
 void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
@@ -353,6 +360,11 @@ void DBImpl::DeleteObsoleteFileImpl(int job_id, const std::string& fname,
        &event_logger_, job_id, number, fname, file_deletion_status, GetName(),
        immutable_db_options_.listeners);
  }
+  if (type == kBlobFile) {
+    EventHelpers::LogAndNotifyBlobFileDeletion(
+        &event_logger_, immutable_db_options_.listeners, job_id, number, fname,
+        file_deletion_status, GetName());
+  }
 }

 // Diffs the files listed in filenames and those that do not
@@ -395,10 +407,10 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
                                 blob_file.GetPath());
  }

+  auto wal_dir = immutable_db_options_.GetWalDir();
  for (auto file_num : state.log_delete_files) {
    if (file_num > 0) {
-      candidate_files.emplace_back(LogFileName(file_num),
-                                   immutable_db_options_.wal_dir);
+      candidate_files.emplace_back(LogFileName(file_num), wal_dir);
    }
  }
  for (const auto& filename : state.manifest_delete_files) {
@@ -517,12 +529,6 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
        break;
      case kOptionsFile:
        keep = (number >= optsfile_num2);
-        TEST_SYNC_POINT_CALLBACK(
-            "DBImpl::PurgeObsoleteFiles:CheckOptionsFiles:1",
-            reinterpret_cast<void*>(&number));
-        TEST_SYNC_POINT_CALLBACK(
-            "DBImpl::PurgeObsoleteFiles:CheckOptionsFiles:2",
-            reinterpret_cast<void*>(&keep));
        break;
      case kCurrentFile:
      case kDBLockFile:
@@ -547,8 +553,7 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
      fname = BlobFileName(candidate_file.file_path, number);
      dir_to_sync = candidate_file.file_path;
    } else {
-      dir_to_sync =
-          (type == kWalFile) ? immutable_db_options_.wal_dir : dbname_;
+      dir_to_sync = (type == kWalFile) ? wal_dir : dbname_;
      fname = dir_to_sync +
              ((!dir_to_sync.empty() && dir_to_sync.back() == '/') ||
                       (!to_delete.empty() && to_delete.front() == '/')
@@ -111,16 +111,28 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src,
    result.recycle_log_file_num = 0;
  }

-  if (result.wal_dir.empty()) {
+  if (result.db_paths.size() == 0) {
+    result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
+  } else if (result.wal_dir.empty()) {
    // Use dbname as default
    result.wal_dir = dbname;
  }
-  if (result.wal_dir.back() == '/') {
-    result.wal_dir = result.wal_dir.substr(0, result.wal_dir.size() - 1);
+  if (!result.wal_dir.empty()) {
+    // If there is a wal_dir already set, check to see if the wal_dir is the
+    // same as the dbname AND the same as the db_path[0] (which must exist from
+    // a few lines ago). If the wal_dir matches both of these values, then clear
+    // the wal_dir value, which will make wal_dir == dbname.  Most likely this
+    // condition was the result of reading an old options file where we forced
+    // wal_dir to be set (to dbname).
+    auto npath = NormalizePath(dbname + "/");
+    if (npath == NormalizePath(result.wal_dir + "/") &&
+        npath == NormalizePath(result.db_paths[0].path + "/")) {
+      result.wal_dir.clear();
+    }
  }

-  if (result.db_paths.size() == 0) {
-    result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
+  if (!result.wal_dir.empty() && result.wal_dir.back() == '/') {
+    result.wal_dir = result.wal_dir.substr(0, result.wal_dir.size() - 1);
  }

  if (result.use_direct_reads && result.compaction_readahead_size == 0) {
@@ -141,7 +153,7 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src,

 #ifndef ROCKSDB_LITE
  ImmutableDBOptions immutable_db_options(result);
-  if (!IsWalDirSameAsDBPath(&immutable_db_options)) {
+  if (!immutable_db_options.IsWalDirSameAsDBPath()) {
    // Either the WAL dir and db_paths[0]/db_name are not the same, or we
    // cannot tell for sure. In either case, assume they're different and
    // explicitly cleanup the trash log files (bypass DeleteScheduler)
@@ -149,13 +161,14 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src,
    // DeleteScheduler::CleanupDirectory on the same dir later, it will be
    // safe
    std::vector<std::string> filenames;
-    Status s = result.env->GetChildren(result.wal_dir, &filenames);
+    auto wal_dir = immutable_db_options.GetWalDir();
+    Status s = result.env->GetChildren(wal_dir, &filenames);
    s.PermitUncheckedError();  //**TODO: What to do on error?
    for (std::string& filename : filenames) {
      if (filename.find(".log.trash", filename.length() -
                                          std::string(".log.trash").length()) !=
          std::string::npos) {
-        std::string trash_file = result.wal_dir + "/" + filename;
+        std::string trash_file = wal_dir + "/" + filename;
        result.env->DeleteFile(trash_file).PermitUncheckedError();
      }
    }
@@ -300,7 +313,8 @@ Status DBImpl::NewDB(std::vector<std::string>* new_filenames) {
    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
        std::move(file), manifest, file_options, immutable_db_options_.clock,
        io_tracer_, nullptr /* stats */, immutable_db_options_.listeners,
-        nullptr, tmp_set.Contains(FileType::kDescriptorFile)));
+        nullptr, tmp_set.Contains(FileType::kDescriptorFile),
+        tmp_set.Contains(FileType::kDescriptorFile)));
    log::Writer log(std::move(file_writer), 0, false);
    std::string record;
    new_db.EncodeTo(&record);
@@ -542,12 +556,12 @@ Status DBImpl::Recover(
    // Note that prev_log_number() is no longer used, but we pay
    // attention to it in case we are recovering a database
    // produced by an older version of rocksdb.
+    auto wal_dir = immutable_db_options_.GetWalDir();
    if (!immutable_db_options_.best_efforts_recovery) {
-      s = env_->GetChildren(immutable_db_options_.wal_dir, &files_in_wal_dir);
+      s = env_->GetChildren(wal_dir, &files_in_wal_dir);
    }
    if (s.IsNotFound()) {
-      return Status::InvalidArgument("wal_dir not found",
-                                     immutable_db_options_.wal_dir);
+      return Status::InvalidArgument("wal_dir not found", wal_dir);
    } else if (!s.ok()) {
      return s;
    }
@@ -563,8 +577,7 @@ Status DBImpl::Recover(
              "existing log file: ",
              file);
        } else {
-          wal_files[number] =
-              LogFileName(immutable_db_options_.wal_dir, number);
+          wal_files[number] = LogFileName(wal_dir, number);
        }
      }
    }
@@ -644,7 +657,7 @@ Status DBImpl::Recover(
    if (s.ok()) {
      const std::string normalized_dbname = NormalizePath(dbname_);
      const std::string normalized_wal_dir =
-          NormalizePath(immutable_db_options_.wal_dir);
+          NormalizePath(immutable_db_options_.GetWalDir());
      if (immutable_db_options_.best_efforts_recovery) {
        filenames = std::move(files_in_dbname);
      } else if (normalized_dbname == normalized_wal_dir) {
@@ -849,7 +862,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& wal_numbers,
    // update the file number allocation counter in VersionSet.
    versions_->MarkFileNumberUsed(wal_number);
    // Open the log file
-    std::string fname = LogFileName(immutable_db_options_.wal_dir, wal_number);
+    std::string fname =
+        LogFileName(immutable_db_options_.GetWalDir(), wal_number);

    ROCKS_LOG_INFO(immutable_db_options_.info_log,
                   "Recovering log #%" PRIu64 " mode %d", wal_number,
@@ -1274,7 +1288,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& wal_numbers,
 Status DBImpl::GetLogSizeAndMaybeTruncate(uint64_t wal_number, bool truncate,
                                          LogFileNumberSize* log_ptr) {
  LogFileNumberSize log(wal_number);
-  std::string fname = LogFileName(immutable_db_options_.wal_dir, wal_number);
+  std::string fname =
+      LogFileName(immutable_db_options_.GetWalDir(), wal_number);
  Status s;
  // This gets the appear size of the wals, not including preallocated space.
  s = env_->GetFileSize(fname, &log.size);
@@ -1409,8 +1424,9 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
          std::move(range_del_iters), &meta, &blob_file_additions,
          snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
          paranoid_file_checks, cfd->internal_stats(), &io_s, io_tracer_,
-          &event_logger_, job_id, Env::IO_HIGH, nullptr /* table_properties */,
-          write_hint, nullptr /*full_history_ts_low*/, &blob_callback_);
+          BlobFileCreationReason::kRecovery, &event_logger_, job_id,
+          Env::IO_HIGH, nullptr /* table_properties */, write_hint,
+          nullptr /*full_history_ts_low*/, &blob_callback_);
      LogFlush(immutable_db_options_.info_log);
      ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
                      "[%s] [WriteLevel0TableForRecovery]"
@@ -1419,8 +1435,10 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
                      meta.fd.GetFileSize(), s.ToString().c_str());
      mutex_.Lock();

-      io_s.PermitUncheckedError();  // TODO(AR) is this correct, or should we
-                                    // return io_s if not ok()?
+      // TODO(AR) is this ok?
+      if (!io_s.ok() && s.ok()) {
+        s = io_s;
+      }
    }
  }
  ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);
@@ -1514,15 +1532,14 @@ IOStatus DBImpl::CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
      BuildDBOptions(immutable_db_options_, mutable_db_options_);
  FileOptions opt_file_options =
      fs_->OptimizeForLogWrite(file_options_, db_options);
-  std::string log_fname =
-      LogFileName(immutable_db_options_.wal_dir, log_file_num);
+  std::string wal_dir = immutable_db_options_.GetWalDir();
+  std::string log_fname = LogFileName(wal_dir, log_file_num);

  if (recycle_log_number) {
    ROCKS_LOG_INFO(immutable_db_options_.info_log,
                   "reusing log %" PRIu64 " from recycle list\n",
                   recycle_log_number);
-    std::string old_log_fname =
-        LogFileName(immutable_db_options_.wal_dir, recycle_log_number);
+    std::string old_log_fname = LogFileName(wal_dir, recycle_log_number);
    TEST_SYNC_POINT("DBImpl::CreateWAL:BeforeReuseWritableFile1");
    TEST_SYNC_POINT("DBImpl::CreateWAL:BeforeReuseWritableFile2");
    io_s = fs_->ReuseWritableFile(log_fname, old_log_fname, opt_file_options,
@@ -1540,7 +1557,8 @@ IOStatus DBImpl::CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
        std::move(lfile), log_fname, opt_file_options,
        immutable_db_options_.clock, io_tracer_, nullptr /* stats */, listeners,
-        nullptr, tmp_set.Contains(FileType::kWalFile)));
+        nullptr, tmp_set.Contains(FileType::kWalFile),
+        tmp_set.Contains(FileType::kWalFile)));
    *new_log = new log::Writer(std::move(file_writer), log_file_num,
                               immutable_db_options_.recycle_log_file_num > 0,
                               immutable_db_options_.manual_wal_flush);
@@ -1572,7 +1590,7 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
  }

  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch, batch_per_txn);
-  s = impl->env_->CreateDirIfMissing(impl->immutable_db_options_.wal_dir);
+  s = impl->env_->CreateDirIfMissing(impl->immutable_db_options_.GetWalDir());
  if (s.ok()) {
    std::vector<std::string> paths;
    for (auto& db_path : impl->immutable_db_options_.db_paths) {
@@ -1604,7 +1622,7 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
    return s;
  }

-  impl->wal_in_db_path_ = IsWalDirSameAsDBPath(&impl->immutable_db_options_);
+  impl->wal_in_db_path_ = impl->immutable_db_options_.IsWalDirSameAsDBPath();

  impl->mutex_.Lock();
  // Handles create_if_missing, error_if_exists
@@ -1671,6 +1689,7 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,

      impl->DeleteObsoleteFiles();
      s = impl->directories_.GetDbDir()->Fsync(IOOptions(), nullptr);
+      TEST_SYNC_POINT("DBImpl::Open:AfterDeleteFilesAndSyncDir");
    }
    if (s.ok()) {
      // In WritePrepared there could be gap in sequence numbers. This breaks
@@ -41,6 +41,9 @@ Status DBImplSecondary::Recover(
          ->Recover(column_families, &manifest_reader_, &manifest_reporter_,
                    &manifest_reader_status_);
  if (!s.ok()) {
+    if (manifest_reader_status_) {
+      manifest_reader_status_->PermitUncheckedError();
+    }
    return s;
  }
  if (immutable_db_options_.paranoid_checks && s.ok()) {
@@ -97,10 +100,10 @@ Status DBImplSecondary::FindNewLogNumbers(std::vector<uint64_t>* logs) {
  assert(logs != nullptr);
  std::vector<std::string> filenames;
  Status s;
-  s = env_->GetChildren(immutable_db_options_.wal_dir, &filenames);
+  s = env_->GetChildren(immutable_db_options_.GetWalDir(), &filenames);
  if (s.IsNotFound()) {
    return Status::InvalidArgument("Failed to open wal_dir",
-                                   immutable_db_options_.wal_dir);
+                                   immutable_db_options_.GetWalDir());
  } else if (!s.ok()) {
    return s;
  }
@@ -140,7 +143,8 @@ Status DBImplSecondary::MaybeInitLogReader(
    // initialize log reader from log_number
    // TODO: min_log_number_to_keep_2pc check needed?
    // Open the log file
-    std::string fname = LogFileName(immutable_db_options_.wal_dir, log_number);
+    std::string fname =
+        LogFileName(immutable_db_options_.GetWalDir(), log_number);
    ROCKS_LOG_INFO(immutable_db_options_.info_log,
                   "Recovering log #%" PRIu64 " mode %d", log_number,
                   static_cast<int>(immutable_db_options_.wal_recovery_mode));
@@ -411,7 +415,7 @@ Iterator* DBImplSecondary::NewIterator(const ReadOptions& read_options,
    return NewErrorIterator(
        Status::NotSupported("snapshot not supported in secondary mode"));
  } else {
-    auto snapshot = versions_->LastSequence();
+    SequenceNumber snapshot(kMaxSequenceNumber);
    result = NewIteratorImpl(read_options, cfd, snapshot, read_callback);
  }
  return result;
@@ -419,14 +423,19 @@ Iterator* DBImplSecondary::NewIterator(const ReadOptions& read_options,

 ArenaWrappedDBIter* DBImplSecondary::NewIteratorImpl(
    const ReadOptions& read_options, ColumnFamilyData* cfd,
-    SequenceNumber snapshot, ReadCallback* read_callback) {
+    SequenceNumber snapshot, ReadCallback* read_callback,
+    bool expose_blob_index, bool allow_refresh) {
  assert(nullptr != cfd);
  SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
+  assert(snapshot == kMaxSequenceNumber);
+  snapshot = versions_->LastSequence();
+  assert(snapshot != kMaxSequenceNumber);
  auto db_iter = NewArenaWrappedDbIterator(
      env_, read_options, *cfd->ioptions(), super_version->mutable_cf_options,
      super_version->current, snapshot,
      super_version->mutable_cf_options.max_sequential_skip_in_iterations,
-      super_version->version_number, read_callback);
+      super_version->version_number, read_callback, this, cfd,
+      expose_blob_index, read_options.snapshot ? false : allow_refresh);
  auto internal_iter = NewInternalIterator(
      db_iter->GetReadOptions(), cfd, super_version, db_iter->GetArena(),
      db_iter->GetRangeDelAggregator(), snapshot,
@@ -627,7 +636,7 @@ Status DB::OpenAsSecondary(
      &impl->write_controller_, impl->io_tracer_));
  impl->column_family_memtables_.reset(
      new ColumnFamilyMemTablesImpl(impl->versions_->GetColumnFamilySet()));
-  impl->wal_in_db_path_ = IsWalDirSameAsDBPath(&impl->immutable_db_options_);
+  impl->wal_in_db_path_ = impl->immutable_db_options_.IsWalDirSameAsDBPath();

  impl->mutex_.Lock();
  s = impl->Recover(column_families, true, false, false);
@@ -767,6 +776,7 @@ Status DB::OpenAndCompact(
  compaction_input.db_options.env = override_options.env;
  compaction_input.db_options.file_checksum_gen_factory =
      override_options.file_checksum_gen_factory;
+  compaction_input.db_options.statistics = override_options.statistics;
  compaction_input.column_family.options.comparator =
      override_options.comparator;
  compaction_input.column_family.options.merge_operator =
@@ -97,7 +97,9 @@ class DBImplSecondary : public DBImpl {
  ArenaWrappedDBIter* NewIteratorImpl(const ReadOptions& read_options,
                                      ColumnFamilyData* cfd,
                                      SequenceNumber snapshot,
-                                      ReadCallback* read_callback);
+                                      ReadCallback* read_callback,
+                                      bool expose_blob_index = false,
+                                      bool allow_refresh = true);

  Status NewIterators(const ReadOptions& options,
                      const std::vector<ColumnFamilyHandle*>& column_families,
@@ -278,6 +280,16 @@ class DBImplSecondary : public DBImpl {
      return AddColumnFamilyId(column_family_id);
    }

+    Status MarkBeginPrepare(bool) override { return Status::OK(); }
+
+    Status MarkEndPrepare(const Slice&) override { return Status::OK(); }
+
+    Status MarkRollback(const Slice&) override { return Status::OK(); }
+
+    Status MarkCommit(const Slice&) override { return Status::OK(); }
+
+    Status MarkNoop(bool) override { return Status::OK(); }
+
    const std::unordered_set<uint32_t>& column_families() const {
      return column_family_ids_;
    }
@@ -220,7 +220,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,

  bool need_log_sync = write_options.sync;
  bool need_log_dir_sync = need_log_sync && !log_dir_synced_;
-  if (!two_write_queues_ || !disable_memtable) {
+  assert(!two_write_queues_ || !disable_memtable);
+  {
    // With concurrent writes we do preprocess only in the write thread that
    // also does write to memtable to avoid sync issue on shared data structure
    // with the other thread
@@ -1123,6 +1124,18 @@ IOStatus DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
    //    writer thread, so no one will push to logs_,
    //  - as long as other threads don't modify it, it's safe to read
    //    from std::deque from multiple threads concurrently.
+    //
+    // Sync operation should work with locked log_write_mutex_, because:
+    //   when DBOptions.manual_wal_flush_ is set,
+    //   FlushWAL function will be invoked by another thread.
+    //   if without locked log_write_mutex_, the log file may get data
+    //   corruption
+
+    const bool needs_locking = manual_wal_flush_ && !two_write_queues_;
+    if (UNLIKELY(needs_locking)) {
+      log_write_mutex_.Lock();
+    }
+
    for (auto& log : logs_) {
      io_s = log.writer->file()->Sync(immutable_db_options_.use_fsync);
      if (!io_s.ok()) {
@@ -1130,6 +1143,10 @@ IOStatus DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
      }
    }

+    if (UNLIKELY(needs_locking)) {
+      log_write_mutex_.Unlock();
+    }
+
    if (io_s.ok() && need_log_dir_sync) {
      // We only sync WAL directory the first time WAL syncing is
      // requested, so that in case users never turn on WAL sync,
@@ -1651,12 +1668,7 @@ Status DBImpl::TrimMemtableHistory(WriteContext* context) {
  for (auto& cfd : cfds) {
    autovector<MemTable*> to_delete;
    bool trimmed = cfd->imm()->TrimHistory(
-        &to_delete, cfd->mem()->ApproximateMemoryUsage());
-    if (!to_delete.empty()) {
-      for (auto m : to_delete) {
-        delete m;
-      }
-    }
+        &context->memtables_to_free_, cfd->mem()->ApproximateMemoryUsage());
    if (trimmed) {
      context->superversion_context.NewSuperVersion();
      assert(context->superversion_context.new_superversion.get() != nullptr);
@@ -1915,6 +1927,8 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
      for (auto cf : empty_cfs) {
        if (cf->IsEmpty()) {
          cf->SetLogNumber(logfile_number_);
+          // MEMPURGE: No need to change this, because new adds
+          // should still receive new sequence numbers.
          cf->mem()->SetCreationSeq(versions_->LastSequence());
        }  // cf may become non-empty.
      }
@@ -1937,11 +1951,13 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  }

  cfd->mem()->SetNextLogNumber(logfile_number_);
+  assert(new_mem != nullptr);
  cfd->imm()->Add(cfd->mem(), &context->memtables_to_free_);
  new_mem->Ref();
  cfd->SetMemtable(new_mem);
  InstallSuperVersionAndScheduleWork(cfd, &context->superversion_context,
                                     mutable_cf_options);
+
 #ifndef ROCKSDB_LITE
  mutex_.Unlock();
  // Notify client that memtable is sealed, now that we have successfully
@@ -1998,13 +2014,18 @@ Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,
  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);
-  if (!s.ok()) {
-    return s;
+  WriteBatch batch;
+  Status s;
+  if (key.data() + key.size() == ts->data()) {
+    Slice key_with_ts = Slice(key.data(), key.size() + ts_sz);
+    s = batch.Put(column_family, key_with_ts, value);
+  } else {
+    std::array<Slice, 2> key_with_ts_slices{{key, *ts}};
+    SliceParts key_with_ts(key_with_ts_slices.data(), 2);
+    std::array<Slice, 1> value_slices{{value}};
+    SliceParts values(value_slices.data(), 1);
+    s = batch.Put(column_family, key_with_ts, values);
  }
-  s = batch.AssignTimestamp(*ts);
  if (!s.ok()) {
    return s;
  }
@@ -2023,17 +2044,19 @@ Status DB::Delete(const WriteOptions& opt, ColumnFamilyHandle* column_family,
  }
  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;
+  size_t ts_sz = ts->size();
+  assert(column_family->GetComparator());
+  assert(ts_sz == column_family->GetComparator()->timestamp_size());
+  WriteBatch batch;
+  Status s;
+  if (key.data() + key.size() == ts->data()) {
+    Slice key_with_ts = Slice(key.data(), key.size() + ts_sz);
+    s = batch.Delete(column_family, key_with_ts);
+  } else {
+    std::array<Slice, 2> key_with_ts_slices{{key, *ts}};
+    SliceParts key_with_ts(key_with_ts_slices.data(), 2);
+    s = batch.Delete(column_family, key_with_ts);
  }
-  s = batch.AssignTimestamp(*ts);
  if (!s.ok()) {
    return s;
  }