WritePrepared: Fix bug in searching in non-cached snapshots (#4639 )

Summary: When evicting an entry form the commit_cache, it is verified against the list of old snapshots to see if it overlaps with any. The list of old snapshots is split into two lists: an efficient concurrent cache and an slow vector protected by a lock. The patch fixes a bug that would stop the search in the cache if it finds any and yet would not include the larger snapshots in the slower list. An extra info log entry is also removed. The condition to trigger that although very rare is still feasible and should not spam the LOG when that happens. Fixes #4621 Pull Request resolved: https://github.com/facebook/rocksdb/pull/4639 Differential Revision: D12934989 Pulled By: maysamyabandeh fbshipit-source-id: 4e0fe8147ba292b554ae78e94c21c2ef31e03e2d
Revert "Introduce CacheAllocator, a custom allocator for cache blocks (#4437 )"
2026-07-07 14:47:40 +08:00 · 2018-11-06 13:49:35 -08:00 · 2018-10-31 14:55:11 -07:00 · 2018-10-31 13:02:53 -07:00 · 2018-10-30 12:37:06 -07:00 · 2018-10-19 16:38:03 -07:00
505 changed files with 30909 additions and 8024 deletions
@@ -45,6 +45,8 @@ etags
 rocksdb_dump
 rocksdb_undump
 db_test2
+trace_analyzer
+trace_analyzer_test

 java/out
 java/target
@@ -0,0 +1,4 @@
+extraction:
+  cpp:
+    index:
+      build_command: make static_lib
@@ -15,8 +15,17 @@ cache:
  - apt

 addons:
-   apt:
-      packages: ['zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl', 'libgflags-dev', 'mingw-w64']
+  apt:
+    sources:
+      - ubuntu-toolchain-r-test
+    packages:
+      - curl
+      - g++-8
+      - libbz2-dev
+      - libgflags-dev
+      - libsnappy-dev
+      - mingw-w64
+      - zlib1g-dev
 env:
  - TEST_GROUP=platform_dependent # 16-18 minutes
  - TEST_GROUP=1 # 33-35 minutes
@@ -30,6 +39,7 @@ env:
  # Build examples
  - JOB_NAME=examples # 5-7 minutes
  - JOB_NAME=cmake # 3-5 minutes
+  - JOB_NAME=cmake-gcc8 # 3-5 minutes
  - JOB_NAME=cmake-mingw # 3 minutes

 matrix:
@@ -42,6 +52,8 @@ matrix:
    env: TEST_GROUP=3
  - os: osx
    env: TEST_GROUP=4
+  - os: osx
+    env: JOB_NAME=cmake-gcc8
  - os : osx
    env: JOB_NAME=cmake-mingw
  - os : linux
@@ -52,9 +64,12 @@ matrix:
 # https://docs.travis-ci.com/user/caching/#ccache-cache
 install:
  - if [ "${TRAVIS_OS_NAME}" == osx ]; then
-      brew install ccache;
+      brew install ccache zstd lz4 snappy xz;
      PATH=$PATH:/usr/local/opt/ccache/libexec;
    fi
+  - if [ "${JOB_NAME}" == cmake-gcc8 ]; then
+      CC=gcc-8 && CXX=g++-8;
+    fi
  - if [[ "${JOB_NAME}" == cmake* ]] && [ "${TRAVIS_OS_NAME}" == linux ]; then
      mkdir cmake-dist && curl -sfSL https://cmake.org/files/v3.8/cmake-3.8.1-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
    fi
@@ -67,16 +82,40 @@ before_script:
 script:
  - ${CXX} --version
  - if [ `command -v ccache` ]; then ccache -C; fi
-  - if [ "${TEST_GROUP}" == 'platform_dependent' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=db_block_cache_test make -j4 all_but_some_tests check_some; fi
-  - if [ "${TEST_GROUP}" == '1' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=full_filter_block_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '2' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=full_filter_block_test ROCKSDBTESTS_END=write_batch_with_index_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '3' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_batch_with_index_test ROCKSDBTESTS_END=write_prepared_transaction_test make -j4 check_some; fi
-  - if [ "${TEST_GROUP}" == '4' ]; then OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_prepared_transaction_test make -j4 check_some; fi
-  - if [ "${JOB_NAME}" == 'java_test' ]; then OPT=-DTRAVIS V=1 make clean jclean && make rocksdbjava jtest; fi
-  - if [ "${JOB_NAME}" == 'lite_build' ]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib tools; fi
-  - if [ "${JOB_NAME}" == 'examples' ]; then OPT=-DTRAVIS V=1 make -j4 static_lib; cd examples; make -j4; fi
-  - if [ "${JOB_NAME}" == 'cmake' ]; then mkdir build && cd build && cmake -DJNI=1 .. && make -j4 rocksdb rocksdbjni; fi
-  - if [ "${JOB_NAME}" == 'cmake-mingw' ]; then mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb rocksdbjni; fi
+  - case $TEST_GROUP in
+    platform_dependent)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_END=db_block_cache_test make -j4 all_but_some_tests check_some
+      ;;
+    1)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=full_filter_block_test make -j4 check_some
+      ;;
+    2)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=full_filter_block_test ROCKSDBTESTS_END=write_batch_with_index_test make -j4 check_some
+      ;;
+    3)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_batch_with_index_test ROCKSDBTESTS_END=write_prepared_transaction_test make -j4 check_some
+      ;;
+    4)
+      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=write_prepared_transaction_test make -j4 check_some
+      ;;
+    esac
+  - case $JOB_NAME in
+    java_test)
+      OPT=-DTRAVIS V=1 make clean jclean && make rocksdbjava jtest
+      ;;
+    lite_build)
+      OPT='-DTRAVIS -DROCKSDB_LITE' V=1 make -j4 static_lib tools
+      ;;
+    examples)
+      OPT=-DTRAVIS V=1 make -j4 static_lib && cd examples && make -j4
+      ;;
+    cmake-mingw)
+      mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_C_COMPILER=x86_64-w64-mingw32-gcc -DCMAKE_CXX_COMPILER=x86_64-w64-mingw32-g++ -DCMAKE_SYSTEM_NAME=Windows && make -j4 rocksdb rocksdbjni
+      ;;
+    cmake*)
+      mkdir build && cd build && cmake -DJNI=1 .. -DCMAKE_BUILD_TYPE=Release && make -j4 rocksdb rocksdbjni
+      ;;
+    esac
 notifications:
    email:
      - leveldb@fb.com
@@ -48,6 +48,7 @@ option(WITH_JEMALLOC "build with JeMalloc" OFF)
 option(WITH_SNAPPY "build with SNAPPY" OFF)
 option(WITH_LZ4 "build with lz4" OFF)
 option(WITH_ZLIB "build with zlib" OFF)
+option(WITH_ZSTD "build with zstd" OFF)
 if(MSVC)
  # Defaults currently different for GFLAGS.
  #  We will address find_package work a little later
@@ -108,7 +109,6 @@ else()
    list(APPEND THIRDPARTY_LIBS ${LZ4_LIBRARIES})
  endif()

-  option(WITH_ZSTD "build with zstd" OFF)
  if(WITH_ZSTD)
    find_package(zstd REQUIRED)
    add_definitions(-DZSTD)
@@ -170,7 +170,7 @@ else()
  endif()
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -fno-omit-frame-pointer")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
    include(CheckCXXCompilerFlag)
    CHECK_CXX_COMPILER_FLAG("-momit-leaf-frame-pointer" HAVE_OMIT_LEAF_FRAME_POINTER)
    if(HAVE_OMIT_LEAF_FRAME_POINTER)
@@ -307,14 +307,6 @@ if(DISABLE_STALL_NOTIF)
  add_definitions(-DROCKSDB_DISABLE_STALL_NOTIFICATION)
 endif()

-# Used to run CI build and tests so we can run faster
-set(OPTIMIZE_DEBUG_DEFAULT 0)        # Debug build is unoptimized by default use -DOPTDBG=1 to optimize
-
-if(DEFINED OPTDBG)
-   set(OPTIMIZE_DEBUG ${OPTDBG})
-else()
-   set(OPTIMIZE_DEBUG ${OPTIMIZE_DEBUG_DEFAULT})
-endif()

 if(DEFINED USE_RTTI)
  if(USE_RTTI)
@@ -322,22 +314,40 @@ if(DEFINED USE_RTTI)
    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -DROCKSDB_USE_RTTI")
  else()
-    message(STATUS "Disabling RTTI")
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-rtti")
-    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+    if(MSVC)
+      message(STATUS "Disabling RTTI in Release builds. Always on in Debug.")
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
+      set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /GR-")
+    else()
+      message(STATUS "Disabling RTTI in Release builds")
+      set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-rtti")
+      set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+    endif()
  endif()
 else()
  message(STATUS "Enabling RTTI in Debug builds only (default)")
  set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DROCKSDB_USE_RTTI")
-  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+  if(MSVC)
+     set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /GR-")
+  else()
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-rtti")
+  endif()
 endif()

+# Used to run CI build and tests so we can run faster
+option(OPTDBG "Build optimized debug build with MSVC" OFF)
+option(WITH_RUNTIME_DEBUG "build with debug version of runtime library" ON)
 if(MSVC)
-  if((${OPTIMIZE_DEBUG} EQUAL 1))
+  if(OPTDBG)
    message(STATUS "Debug optimization is enabled")
-    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt /${RUNTIME_LIBRARY}d")
+    set(CMAKE_CXX_FLAGS_DEBUG "/Oxt")
  else()
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm /${RUNTIME_LIBRARY}d")
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /RTC1 /Gm")
+  endif()
+  if(WITH_RUNTIME_DEBUG)
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /${RUNTIME_LIBRARY}d")
+  else()
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /${RUNTIME_LIBRARY}")
  endif()
  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /${RUNTIME_LIBRARY}")

@@ -471,6 +481,7 @@ set(SOURCES
        db/db_info_dumper.cc
        db/db_iter.cc
        db/dbformat.cc
+        db/error_handler.cc
        db/event_helpers.cc
        db/experimental.cc
        db/external_sst_file_ingestion_job.cc
@@ -483,7 +494,6 @@ set(SOURCES
        db/log_reader.cc
        db/log_writer.cc
        db/malloc_stats.cc
-        db/managed_iterator.cc
        db/memtable.cc
        db/memtable_list.cc
        db/merge_helper.cc
@@ -545,6 +555,8 @@ set(SOURCES
        table/cuckoo_table_builder.cc
        table/cuckoo_table_factory.cc
        table/cuckoo_table_reader.cc
+        table/data_block_hash_index.cc
+        table/data_block_footer.cc
        table/flush_block_policy.cc
        table/format.cc
        table/full_filter_block.cc
@@ -568,12 +580,14 @@ set(SOURCES
        tools/ldb_cmd.cc
        tools/ldb_tool.cc
        tools/sst_dump_tool.cc
+        tools/trace_analyzer_tool.cc
        util/arena.cc
        util/auto_roll_logger.cc
        util/bloom.cc
        util/coding.cc
        util/compaction_job_stats_impl.cc
        util/comparator.cc
+        util/compression_context_cache.cc
        util/concurrent_arena.cc
        util/crc32c.cc
        util/delete_scheduler.cc
@@ -591,25 +605,25 @@ set(SOURCES
        util/slice.cc
        util/sst_file_manager_impl.cc
        util/status.cc
-        util/status_message.cc
        util/string_util.cc
        util/sync_point.cc
        util/sync_point_impl.cc
        util/testutil.cc
        util/thread_local.cc
        util/threadpool_imp.cc
+        util/trace_replay.cc
        util/transaction_test_util.cc
        util/xxhash.cc
        utilities/backupable/backupable_db.cc
        utilities/blob_db/blob_compaction_filter.cc
        utilities/blob_db/blob_db.cc
        utilities/blob_db/blob_db_impl.cc
+        utilities/blob_db/blob_db_impl_filesnapshot.cc
        utilities/blob_db/blob_dump_tool.cc
        utilities/blob_db/blob_file.cc
        utilities/blob_db/blob_log_reader.cc
        utilities/blob_db/blob_log_writer.cc
        utilities/blob_db/blob_log_format.cc
-        utilities/blob_db/ttl_extractor.cc
        utilities/cassandra/cassandra_compaction_filter.cc
        utilities/cassandra/format.cc
        utilities/cassandra/merge_operator.cc
@@ -646,6 +660,7 @@ set(SOURCES
        utilities/simulator_cache/sim_cache.cc
        utilities/spatialdb/spatial_db.cc
        utilities/table_properties_collectors/compact_on_deletion_collector.cc
+        utilities/trace/file_trace_reader_writer.cc
        utilities/transactions/optimistic_transaction_db_impl.cc
        utilities/transactions/optimistic_transaction.cc
        utilities/transactions/pessimistic_transaction.cc
@@ -657,6 +672,8 @@ set(SOURCES
        utilities/transactions/transaction_util.cc
        utilities/transactions/write_prepared_txn.cc
        utilities/transactions/write_prepared_txn_db.cc
+        utilities/transactions/write_unprepared_txn.cc
+        utilities/transactions/write_unprepared_txn_db.cc
        utilities/ttl/db_ttl_impl.cc
        utilities/write_batch_with_index/write_batch_with_index.cc
        utilities/write_batch_with_index/write_batch_with_index_internal.cc
@@ -864,6 +881,7 @@ if(WITH_TESTS)
        db/db_write_test.cc
        db/dbformat_test.cc
        db/deletefile_test.cc
+        db/error_handler_test.cc
        db/obsolete_files_test.cc
        db/external_sst_file_basic_test.cc
        db/external_sst_file_test.cc
@@ -906,12 +924,14 @@ if(WITH_TESTS)
        table/cleanable_test.cc
        table/cuckoo_table_builder_test.cc
        table/cuckoo_table_reader_test.cc
+        table/data_block_hash_index_test.cc
        table/full_filter_block_test.cc
        table/merger_test.cc
        table/table_test.cc
        tools/ldb_cmd_test.cc
        tools/reduce_levels_test.cc
        tools/sst_dump_test.cc
+        tools/trace_analyzer_test.cc
        util/arena_test.cc
        util/auto_roll_logger_test.cc
        util/autovector_test.cc
@@ -926,6 +946,7 @@ if(WITH_TESTS)
        util/hash_test.cc
        util/heap_test.cc
        util/rate_limiter_test.cc
+        util/repeatable_thread_test.cc
        util/slice_transform_test.cc
        util/timer_queue_test.cc
        util/thread_list_test.cc
@@ -957,6 +978,7 @@ if(WITH_TESTS)
        utilities/transactions/optimistic_transaction_test.cc
        utilities/transactions/transaction_test.cc
        utilities/transactions/write_prepared_transaction_test.cc
+        utilities/transactions/write_unprepared_transaction_test.cc
        utilities/ttl/ttl_test.cc
        utilities/write_batch_with_index/write_batch_with_index_test.cc
  )
@@ -967,6 +989,7 @@ if(WITH_TESTS)
  set(BENCHMARKS
    cache/cache_bench.cc
    memtable/memtablerep_bench.cc
+    db/range_del_aggregator_bench.cc
    tools/db_bench.cc
    table/table_reader_bench.cc
    utilities/column_aware_encoding_exp.cc
@@ -1,7 +1,84 @@
 # Rocksdb Change Log
-## Unreleased
+# Unreleased
+### Bug Fixes
+* Fix the bug that WriteBatchWithIndex's SeekForPrev() doesn't see the entries with the same key.
+* Fix a bug in WritePrepared txns where if the number of old snapshots goes beyond the snapshot cache size (128 default) the rest will not be checked when evicting a commit entry from the commit cache.
+
+## 5.17.1 (10/16/2018)
+### Bug Fixes
+* Fix slow flush/compaction when DB contains many snapshots. The problem became noticeable to us in DBs with 100,000+ snapshots, though it will affect others at different thresholds.
+* Properly set the stop key for a truncated manual CompactRange
+* Fix corner case where a write group leader blocked due to write stall blocks other writers in queue with WriteOptions::no_slowdown set.
+
+### New Features
+* Introduced CacheAllocator, which lets the user specify custom allocator for memory in block cache.
+
+## 5.17.0 (10/05/2018)
 ### Public API Change
+* `OnTableFileCreated` will now be called for empty files generated during compaction. In that case, `TableFileCreationInfo::file_path` will be "(nil)" and `TableFileCreationInfo::file_size` will be zero.
+* Add `FlushOptions::allow_write_stall`, which controls whether Flush calls start working immediately, even if it causes user writes to stall, or will wait until flush can be performed without causing write stall (similar to `CompactRangeOptions::allow_write_stall`). Note that the default value is false, meaning we add delay to Flush calls until stalling can be avoided when possible. This is behavior change compared to previous RocksDB versions, where Flush calls didn't check if they might cause stall or not.
+* Application using PessimisticTransactionDB is expected to rollback/commit recovered transactions before starting new ones. This assumption is used to skip concurrency control during recovery.
+
+### New Features
+* TransactionOptions::skip_concurrency_control allows pessimistic transactions to skip the overhead of concurrency control. Could be used for optimizing certain transactions or during recovery.
+
+### Bug Fixes
+* Avoid creating empty SSTs and subsequently deleting them in certain cases during compaction.
+* Sync CURRENT file contents during checkpoint.
+
+## 5.16.3 (10/1/2018)
+### Bug Fixes
+* Fix crash caused when `CompactFiles` run with `CompactionOptions::compression == CompressionType::kDisableCompressionOption`. Now that setting causes the compression type to be chosen according to the column family-wide compression options.
+
+## 5.16.2 (9/21/2018)
+### Bug Fixes
+* Fix bug in partition filters with format_version=4.
+
+## 5.16.1 (9/17/2018)
+### Bug Fixes
+* Remove trace_analyzer_tool from rocksdb_lib target in TARGETS file.
+* Fix RocksDB Java build and tests.
+* Remove sync point in Block destructor.
+
+## 5.16.0 (8/21/2018)
+### Public API Change
+* The merge operands are passed to `MergeOperator::ShouldMerge` in the reversed order relative to how they were merged (passed to FullMerge or FullMergeV2) for performance reasons
+* GetAllKeyVersions() to take an extra argument of `max_num_ikeys`.
+* Using ZSTD dictionary trainer (i.e., setting `CompressionOptions::zstd_max_train_bytes` to a nonzero value) now requires ZSTD version 1.1.3 or later.
+
+### New Features
+* Changes the format of index blocks by delta encoding the index values, which are the block handles. This saves the encoding of BlockHandle::offset of the non-head index entries in each restart interval. The feature is backward compatible but not forward compatible. It is disabled by default unless format_version 4 or above is used.
+* Add a new tool: trace_analyzer. Trace_analyzer analyzes the trace file generated by using trace_replay API. It can convert the binary format trace file to a human readable txt file, output the statistics of the analyzed query types such as access statistics and size statistics, combining the dumped whole key space file to analyze, support query correlation analyzing, and etc. Current supported query types are: Get, Put, Delete, SingleDelete, DeleteRange, Merge, Iterator (Seek, SeekForPrev only).
+* Add hash index support to data blocks, which helps reducing the cpu utilization of point-lookup operations. This feature is backward compatible with the data block created without the hash index. It is disabled by default unless BlockBasedTableOptions::data_block_index_type is set to data_block_index_type = kDataBlockBinaryAndHash.
+
+### Bug Fixes
+* Fix a bug in misreporting the estimated partition index size in properties block.
+
+## 5.15.0 (7/17/2018)
+### Public API Change
+* Remove managed iterator. ReadOptions.managed is not effective anymore.
+* For bottommost_compression, a compatible CompressionOptions is added via `bottommost_compression_opts`. To keep backward compatible, a new boolean `enabled` is added to CompressionOptions. For compression_opts, it will be always used no matter what value of `enabled` is. For bottommost_compression_opts, it will only be used when user set `enabled=true`, otherwise, compression_opts will be used for bottommost_compression as default.
+* With LRUCache, when high_pri_pool_ratio > 0, midpoint insertion strategy will be enabled to put low-pri items to the tail of low-pri list (the midpoint) when they first inserted into the cache. This is to make cache entries never get hit age out faster, improving cache efficiency when large background scan presents.
 * For users of `Statistics` objects created via `CreateDBStatistics()`, the format of the string returned by its `ToString()` method has changed.
+* The "rocksdb.num.entries" table property no longer counts range deletion tombstones as entries.
+
+### New Features
+* Changes the format of index blocks by storing the key in their raw form rather than converting them to InternalKey. This saves 8 bytes per index key. The feature is backward compatible but not forward compatible. It is disabled by default unless format_version 3 or above is used.
+* Avoid memcpy when reading mmap files with OpenReadOnly and max_open_files==-1.
+* Support dynamically changing `ColumnFamilyOptions::ttl` via `SetOptions()`.
+* Add a new table property, "rocksdb.num.range-deletions", which counts the number of range deletion tombstones in the table.
+* Improve the performance of iterators doing long range scans by using readahead, when using direct IO.
+* pin_top_level_index_and_filter (default true) in BlockBasedTableOptions can be used in combination with cache_index_and_filter_blocks to prefetch and pin the top-level index of partitioned index and filter blocks in cache. It has no impact when cache_index_and_filter_blocks is false.
+* Write properties meta-block at the end of block-based table to save read-ahead IO.
+
+### Bug Fixes
+* Fix deadlock with enable_pipelined_write=true and max_successive_merges > 0
+* Check conflict at output level in CompactFiles.
+* Fix corruption in non-iterator reads when mmap is used for file reads
+* Fix bug with prefix search in partition filters where a shared prefix would be ignored from the later partitions. The bug could report an eixstent key as missing. The bug could be triggered if prefix_extractor is set and partition filters is enabled.
+* Change default value of `bytes_max_delete_chunk` to 0 in NewSstFileManager() as it doesn't work well with checkpoints.
+* Fix a bug caused by not copying the block trailer with compressed SST file, direct IO, prefetcher and no compressed block cache.
+* Fix write can stuck indefinitely if enable_pipelined_write=true. The issue exists since pipelined write was introduced in 5.5.0.

 ## 5.14.0 (5/16/2018)
 ### Public API Change
@@ -111,7 +188,8 @@
 * `BackupableDBOptions::max_valid_backups_to_open == 0` now means no backups will be opened during BackupEngine initialization. Previously this condition disabled limiting backups opened.
 * `DBOptions::preserve_deletes` is a new option that allows one to specify that DB should not drop tombstones for regular deletes if they have sequence number larger than what was set by the new API call `DB::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum)`. Disabled by default.
 * API call `DB::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum)` was added, users who wish to preserve deletes are expected to periodically call this function to advance the cutoff seqnum (all deletes made before this seqnum can be dropped by DB). It's user responsibility to figure out how to advance the seqnum in the way so the tombstones are kept for the desired period of time, yet are eventually processed in time and don't eat up too much space.
-* `ReadOptions::iter_start_seqnum` was added; if set to something > 0 user will see 2 changes in iterators behavior 1) only keys written with sequence larger than this parameter would be returned and 2) the `Slice` returned by iter->key() now points to the memory that keep User-oriented representation of the internal key, rather than user key. New struct `FullKey` was added to represent internal keys, along with a new helper function `ParseFullKey(const Slice& internal_key, FullKey* result);`.
+* `ReadOptions::iter_start_seqnum` was added;
+if set to something > 0 user will see 2 changes in iterators behavior 1) only keys written with sequence larger than this parameter would be returned and 2) the `Slice` returned by iter->key() now points to the memory that keep User-oriented representation of the internal key, rather than user key. New struct `FullKey` was added to represent internal keys, along with a new helper function `ParseFullKey(const Slice& internal_key, FullKey* result);`.
 * Deprecate trash_dir param in NewSstFileManager, right now we will rename deleted files to <name>.trash instead of moving them to trash directory
 * Allow setting a custom trash/DB size ratio limit in the SstFileManager, after which files that are to be scheduled for deletion are deleted immediately, regardless of any delete ratelimit.
 * Return an error on write if write_options.sync = true and write_options.disableWAL = true to warn user of inconsistent options. Previously we will not write to WAL and not respecting the sync options in this case.
@@ -10,9 +10,11 @@ This is the list of all known third-party language bindings for RocksDB. If some
 * PHP - https://github.com/Photonios/rocksdb-php
 * C# - https://github.com/warrenfalk/rocksdb-sharp
 * Rust
+    * https://github.com/pingcap/rust-rocksdb (used in production fork of https://github.com/spacejam/rust-rocksdb)
    * https://github.com/spacejam/rust-rocksdb
    * https://github.com/bh1xuw/rust-rocks
 * D programming language - https://github.com/b1naryth1ef/rocksdb
 * Erlang - https://gitlab.com/barrel-db/erlang-rocksdb
 * Elixir - https://github.com/urbint/rox
 * Nim - https://github.com/status-im/nim-rocksdb
+* Swift and Objective-C (iOS/OSX) - https://github.com/iabudiab/ObjectiveRocks 
@@ -85,6 +85,10 @@ ifeq ($(MAKECMDGOALS),rocksdbjavastaticrelease)
 	DEBUG_LEVEL=0
 endif

+ifeq ($(MAKECMDGOALS),rocksdbjavastaticreleasedocker)
+        DEBUG_LEVEL=0
+endif
+
 ifeq ($(MAKECMDGOALS),rocksdbjavastaticpublish)
 	DEBUG_LEVEL=0
 endif
@@ -98,6 +102,10 @@ OPT += -momit-leaf-frame-pointer
 endif
 endif

+ifeq (,$(shell $(CXX) -fsyntax-only -faligned-new -xc++ /dev/null 2>&1))
+CXXFLAGS += -faligned-new -DHAVE_ALIGNED_NEW
+endif
+
 ifeq (,$(shell $(CXX) -fsyntax-only -maltivec -xc /dev/null 2>&1))
 CXXFLAGS += -DHAS_ALTIVEC
 CFLAGS += -DHAS_ALTIVEC
@@ -374,6 +382,8 @@ VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full

 BENCHTOOLOBJECTS = $(BENCH_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)

+ANALYZETOOLOBJECTS = $(ANALYZER_LIB_SOURCES:.cc=.o)
+
 EXPOBJECTS = $(EXP_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)

 TESTS = \
@@ -421,6 +431,7 @@ TESTS = \
 	db_table_properties_test \
 	db_statistics_test \
 	db_write_test \
+	error_handler_test \
 	autovector_test \
 	blob_db_test \
 	cleanable_test \
@@ -428,6 +439,7 @@ TESTS = \
 	table_properties_collector_test \
 	arena_test \
 	block_test \
+	data_block_hash_index_test \
 	cache_test \
 	corruption_test \
 	slice_transform_test \
@@ -518,7 +530,10 @@ TESTS = \
 	repair_test \
 	env_timed_test \
 	write_prepared_transaction_test \
+	write_unprepared_transaction_test \
 	db_universal_compaction_test \
+	trace_analyzer_test \
+	repeatable_thread_test \

 PARALLEL_TEST = \
 	backupable_db_test \
@@ -537,6 +552,7 @@ PARALLEL_TEST = \
 	table_test \
 	transaction_test \
 	write_prepared_transaction_test \
+	write_unprepared_transaction_test \

 # options_settable_test doesn't pass with UBSAN as we use hack in the test
 ifdef COMPILE_WITH_UBSAN
@@ -561,12 +577,13 @@ TOOLS = \
 	rocksdb_dump \
 	rocksdb_undump \
 	blob_dump \
+	trace_analyzer \

 TEST_LIBS = \
 	librocksdb_env_basic_test.a

 # TODO: add back forward_iterator_bench, after making it build in all environemnts.
-BENCHMARKS = db_bench table_reader_bench cache_bench memtablerep_bench column_aware_encoding_exp persistent_cache_bench
+BENCHMARKS = db_bench table_reader_bench cache_bench memtablerep_bench column_aware_encoding_exp persistent_cache_bench range_del_aggregator_bench

 # if user didn't config LIBNAME, set the default
 ifeq ($(LIBNAME),)
@@ -654,7 +671,7 @@ $(SHARED4): $(shared_all_libobjects)
 endif  # PLATFORM_SHARED_EXT

 .PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests package \
-	release tags valgrind_check whitebox_crash_test format static_lib shared_lib all \
+	release tags tags0 valgrind_check whitebox_crash_test format static_lib shared_lib all \
 	dbg rocksdbjavastatic rocksdbjava install install-static install-shared uninstall \
 	analyze tools tools_lib

@@ -802,7 +819,7 @@ check_0:
 	  | grep -E '$(tests-regexp)'					\
 	  | build_tools/gnu_parallel -j$(J) --plain --joblog=LOG $$eta --gnu '{} >& t/log-{/}'

-valgrind-blacklist-regexp = InlineSkipTest.ConcurrentInsert|TransactionTest.DeadlockStress|DBCompactionTest.SuggestCompactRangeNoTwoLevel0Compactions|BackupableDBTest.RateLimiting|DBTest.CloseSpeedup|DBTest.ThreadStatusFlush|DBTest.RateLimitingTest|DBTest.EncodeDecompressedBlockSizeTest|FaultInjectionTest.UninstalledCompaction|HarnessTest.Randomized|ExternalSSTFileTest.CompactDuringAddFileRandom|ExternalSSTFileTest.IngestFileWithGlobalSeqnoRandomized|MySQLStyleTransactionTest.TransactionStressTest
+valgrind-blacklist-regexp = InlineSkipTest.ConcurrentInsert|TransactionStressTest.DeadlockStress|DBCompactionTest.SuggestCompactRangeNoTwoLevel0Compactions|BackupableDBTest.RateLimiting|DBTest.CloseSpeedup|DBTest.ThreadStatusFlush|DBTest.RateLimitingTest|DBTest.EncodeDecompressedBlockSizeTest|FaultInjectionTest.UninstalledCompaction|HarnessTest.Randomized|ExternalSSTFileTest.CompactDuringAddFileRandom|ExternalSSTFileTest.IngestFileWithGlobalSeqnoRandomized|MySQLStyleTransactionTest.TransactionStressTest

 .PHONY: valgrind_check_0
 valgrind_check_0:
@@ -986,8 +1003,10 @@ unity.a: unity.o
 	$(AM_V_AR)rm -f $@
 	$(AM_V_at)$(AR) $(ARFLAGS) $@ unity.o

+
+TOOLLIBOBJECTS = $(TOOL_LIB_SOURCES:.cc=.o)
 # try compiling db_test with unity
-unity_test: db/db_test.o db/db_test_util.o $(TESTHARNESS) unity.a
+unity_test: db/db_test.o db/db_test_util.o $(TESTHARNESS) $(TOOLLIBOBJECTS) unity.a
 	$(AM_LINK)
 	./unity_test

@@ -1007,6 +1026,13 @@ tags:
 	cscope -b `$(FIND) . -name '*.cc'` `$(FIND) . -name '*.h'` `$(FIND) . -name '*.c'`
 	ctags -e -R -o etags *

+tags0:
+	ctags -R .
+	cscope -b `$(FIND) . -name '*.cc' -and ! -name '*_test.cc'` \
+		  `$(FIND) . -name '*.c' -and ! -name '*_test.c'` \
+		  `$(FIND) . -name '*.h' -and ! -name '*_test.h'`
+	ctags -e -R -o etags *
+
 format:
 	build_tools/format-diff.sh

@@ -1020,7 +1046,7 @@ $(LIBRARY): $(LIBOBJECTS)
 	$(AM_V_AR)rm -f $@
 	$(AM_V_at)$(AR) $(ARFLAGS) $@ $(LIBOBJECTS)

-$(TOOLS_LIBRARY): $(BENCH_LIB_SOURCES:.cc=.o) $(TOOL_LIB_SOURCES:.cc=.o) $(LIB_SOURCES:.cc=.o) $(TESTUTIL)
+$(TOOLS_LIBRARY): $(BENCH_LIB_SOURCES:.cc=.o) $(TOOL_LIB_SOURCES:.cc=.o) $(LIB_SOURCES:.cc=.o) $(TESTUTIL) $(ANALYZER_LIB_SOURCES:.cc=.o)
 	$(AM_V_AR)rm -f $@
 	$(AM_V_at)$(AR) $(ARFLAGS) $@ $^

@@ -1031,6 +1057,9 @@ librocksdb_env_basic_test.a: env/env_basic_test.o $(LIBOBJECTS) $(TESTHARNESS)
 db_bench: tools/db_bench.o $(BENCHTOOLOBJECTS)
 	$(AM_LINK)

+trace_analyzer: tools/trace_analyzer.o $(ANALYZETOOLOBJECTS) $(LIBOBJECTS)
+	$(AM_LINK)
+
 cache_bench: cache/cache_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)

@@ -1184,6 +1213,9 @@ db_statistics_test: db/db_statistics_test.o db/db_test_util.o $(LIBOBJECTS) $(TE
 db_write_test: db/db_write_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

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

@@ -1336,6 +1368,9 @@ table_test: table/table_test.o $(LIBOBJECTS) $(TESTHARNESS)
 block_test: table/block_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+data_block_hash_index_test: table/data_block_hash_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 inlineskiplist_test: memtable/inlineskiplist_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1432,6 +1467,9 @@ options_util_test: utilities/options/options_util_test.o $(LIBOBJECTS) $(TESTHAR
 db_bench_tool_test: tools/db_bench_tool_test.o $(BENCHTOOLOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+trace_analyzer_test: tools/trace_analyzer_test.o $(LIBOBJECTS) $(ANALYZETOOLOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 event_logger_test: util/event_logger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

@@ -1474,6 +1512,9 @@ transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TEST
 write_prepared_transaction_test: utilities/transactions/write_prepared_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+write_unprepared_transaction_test: utilities/transactions/write_unprepared_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 	$(AM_LINK)

@@ -1510,9 +1551,15 @@ lua_test: utilities/lua/rocks_lua_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTH
 range_del_aggregator_test: db/range_del_aggregator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+range_del_aggregator_bench: db/range_del_aggregator_bench.o $(LIBOBJECTS) $(TESTUTIL)
+	$(AM_LINK)
+
 blob_db_test: utilities/blob_db/blob_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)

+repeatable_thread_test: util/repeatable_thread_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1583,7 +1630,7 @@ ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.6
 BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-BZIP2_DOWNLOAD_BASE ?= http://www.bzip.org
+BZIP2_DOWNLOAD_BASE ?= https://web.archive.org/web/20180624184835/http://www.bzip.org
 SNAPPY_VER ?= 1.1.4
 SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
@@ -1743,20 +1790,26 @@ rocksdbjavastaticrelease: rocksdbjavastatic
 	cd java/target;jar -uf $(ROCKSDB_JAR_ALL) librocksdbjni-*.so librocksdbjni-*.jnilib
 	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR_ALL) org/rocksdb/*.class org/rocksdb/util/*.class

-rocksdbjavastaticreleasedocker: rocksdbjavastatic
-	DOCKER_LINUX_X64_CONTAINER=`docker ps -aqf name=rocksdb_linux_x64-be`; \
-	if [ -z "$$DOCKER_LINUX_X64_CONTAINER" ]; then \
-		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x64-be evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
-	fi
-	docker start -a rocksdb_linux_x64-be
+rocksdbjavastaticreleasedocker: rocksdbjavastatic rocksdbjavastaticdockerx86 rocksdbjavastaticdockerx86_64
+	cd java;jar -cf target/$(ROCKSDB_JAR_ALL) HISTORY*.md
+	cd java/target;jar -uf $(ROCKSDB_JAR_ALL) librocksdbjni-*.so librocksdbjni-*.jnilib
+	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR_ALL) org/rocksdb/*.class org/rocksdb/util/*.class
+
+rocksdbjavastaticdockerx86:
+	mkdir -p java/target
 	DOCKER_LINUX_X86_CONTAINER=`docker ps -aqf name=rocksdb_linux_x86-be`; \
 	if [ -z "$$DOCKER_LINUX_X86_CONTAINER" ]; then \
 		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x86-be evolvedbinary/rocksjava:centos6_x86-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
 	fi
 	docker start -a rocksdb_linux_x86-be
-	cd java;jar -cf target/$(ROCKSDB_JAR_ALL) HISTORY*.md
-	cd java/target;jar -uf $(ROCKSDB_JAR_ALL) librocksdbjni-*.so librocksdbjni-*.jnilib
-	cd java/target/classes;jar -uf ../$(ROCKSDB_JAR_ALL) org/rocksdb/*.class org/rocksdb/util/*.class
+
+rocksdbjavastaticdockerx86_64:
+	mkdir -p java/target
+	DOCKER_LINUX_X64_CONTAINER=`docker ps -aqf name=rocksdb_linux_x64-be`; \
+	if [ -z "$$DOCKER_LINUX_X64_CONTAINER" ]; then \
+		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_x64-be evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
+	fi
+	docker start -a rocksdb_linux_x64-be

 rocksdbjavastaticdockerppc64le:
 	mkdir -p java/target
@@ -1812,7 +1865,7 @@ $(java_libobjects): jl/%.o: %.cc
 rocksdbjava: $(java_all_libobjects)
 	$(AM_V_GEN)cd java;$(MAKE) javalib;
 	$(AM_V_at)rm -f ./java/target/$(ROCKSDBJNILIB)
-	$(AM_V_at)$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) $(java_libobjects) $(JAVA_LDFLAGS) $(COVERAGEFLAGS)
+	$(AM_V_at)$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC -o ./java/target/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) $(java_all_libobjects) $(JAVA_LDFLAGS) $(COVERAGEFLAGS)
 	$(AM_V_at)cd java;jar -cf target/$(ROCKSDB_JAR) HISTORY*.md
 	$(AM_V_at)cd java/target;jar -uf $(ROCKSDB_JAR) $(ROCKSDBJNILIB)
 	$(AM_V_at)cd java/target/classes;jar -uf ../$(ROCKSDB_JAR) org/rocksdb/*.class org/rocksdb/util/*.class
@@ -1881,7 +1934,7 @@ endif
 #  	Source files dependencies detection
 # ---------------------------------------------------------------------------

-all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(EXP_LIB_SOURCES)
+all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_LIB_SOURCES) $(EXP_LIB_SOURCES) $(ANALYZER_LIB_SOURCES)
 DEPFILES = $(all_sources:.cc=.cc.d)

 # Add proper dependency support so changing a .h file forces a .cc file to
@@ -4,7 +4,6 @@
 [![Windows Build status](https://ci.appveyor.com/api/projects/status/fbgfu0so3afcno78/branch/master?svg=true)](https://ci.appveyor.com/project/Facebook/rocksdb/branch/master)
 [![PPC64le Build Status](http://140.211.168.68:8080/buildStatus/icon?job=Rocksdb)](http://140.211.168.68:8080/job/Rocksdb)

-
 RocksDB is developed and maintained by Facebook Database Engineering Team.
 It is built on earlier work on LevelDB by Sanjay Ghemawat (sanjay@google.com)
 and Jeff Dean (jeff@google.com)
@@ -1,3 +1,5 @@
+load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+
 REPO_PATH = package_name() + "/"

 BUCK_BINS = "buck-out/gen/" + REPO_PATH
@@ -28,7 +30,7 @@ rocksdb_compiler_flags = [
    # Added missing flags from output of build_detect_platform
    "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
    "-DROCKSDB_BACKTRACE",
-    "-Wshorten-64-to-32",
+    "-Wnarrowing",
 ]

 rocksdb_external_deps = [
@@ -93,6 +95,7 @@ cpp_library(
        "db/db_info_dumper.cc",
        "db/db_iter.cc",
        "db/dbformat.cc",
+        "db/error_handler.cc",
        "db/event_helpers.cc",
        "db/experimental.cc",
        "db/external_sst_file_ingestion_job.cc",
@@ -105,7 +108,6 @@ cpp_library(
        "db/log_writer.cc",
        "db/logs_with_prep_tracker.cc",
        "db/malloc_stats.cc",
-        "db/managed_iterator.cc",
        "db/memtable.cc",
        "db/memtable_list.cc",
        "db/merge_helper.cc",
@@ -171,6 +173,8 @@ cpp_library(
        "table/cuckoo_table_builder.cc",
        "table/cuckoo_table_factory.cc",
        "table/cuckoo_table_reader.cc",
+        "table/data_block_footer.cc",
+        "table/data_block_hash_index.cc",
        "table/flush_block_policy.cc",
        "table/format.cc",
        "table/full_filter_block.cc",
@@ -200,6 +204,7 @@ cpp_library(
        "util/coding.cc",
        "util/compaction_job_stats_impl.cc",
        "util/comparator.cc",
+        "util/compression_context_cache.cc",
        "util/concurrent_arena.cc",
        "util/crc32c.cc",
        "util/delete_scheduler.cc",
@@ -217,24 +222,24 @@ cpp_library(
        "util/slice.cc",
        "util/sst_file_manager_impl.cc",
        "util/status.cc",
-        "util/status_message.cc",
        "util/string_util.cc",
        "util/sync_point.cc",
        "util/sync_point_impl.cc",
        "util/thread_local.cc",
        "util/threadpool_imp.cc",
+        "util/trace_replay.cc",
        "util/transaction_test_util.cc",
        "util/xxhash.cc",
        "utilities/backupable/backupable_db.cc",
        "utilities/blob_db/blob_compaction_filter.cc",
        "utilities/blob_db/blob_db.cc",
        "utilities/blob_db/blob_db_impl.cc",
+        "utilities/blob_db/blob_db_impl_filesnapshot.cc",
        "utilities/blob_db/blob_dump_tool.cc",
        "utilities/blob_db/blob_file.cc",
        "utilities/blob_db/blob_log_format.cc",
        "utilities/blob_db/blob_log_reader.cc",
        "utilities/blob_db/blob_log_writer.cc",
-        "utilities/blob_db/ttl_extractor.cc",
        "utilities/cassandra/cassandra_compaction_filter.cc",
        "utilities/cassandra/format.cc",
        "utilities/cassandra/merge_operator.cc",
@@ -269,6 +274,7 @@ cpp_library(
        "utilities/simulator_cache/sim_cache.cc",
        "utilities/spatialdb/spatial_db.cc",
        "utilities/table_properties_collectors/compact_on_deletion_collector.cc",
+        "utilities/trace/file_trace_reader_writer.cc",
        "utilities/transactions/optimistic_transaction.cc",
        "utilities/transactions/optimistic_transaction_db_impl.cc",
        "utilities/transactions/pessimistic_transaction.cc",
@@ -280,6 +286,8 @@ cpp_library(
        "utilities/transactions/transaction_util.cc",
        "utilities/transactions/write_prepared_txn.cc",
        "utilities/transactions/write_prepared_txn_db.cc",
+        "utilities/transactions/write_unprepared_txn.cc",
+        "utilities/transactions/write_unprepared_txn_db.cc",
        "utilities/ttl/db_ttl_impl.cc",
        "utilities/write_batch_with_index/write_batch_with_index.cc",
        "utilities/write_batch_with_index/write_batch_with_index_internal.cc",
@@ -297,6 +305,7 @@ cpp_library(
    srcs = [
        "db/db_test_util.cc",
        "table/mock_table.cc",
+        "tools/trace_analyzer_tool.cc",
        "util/fault_injection_test_env.cc",
        "util/testharness.cc",
        "util/testutil.cc",
@@ -317,6 +326,7 @@ cpp_library(
    name = "rocksdb_tools_lib",
    srcs = [
        "tools/db_bench_tool.cc",
+        "tools/trace_analyzer_tool.cc",
        "util/testutil.cc",
    ],
    headers = AutoHeaders.RECURSIVE_GLOB,
@@ -495,6 +505,11 @@ ROCKS_TESTS = [
        "table/cuckoo_table_reader_test.cc",
        "serial",
    ],
+    [
+        "data_block_hash_index_test",
+        "table/data_block_hash_index_test.cc",
+        "serial",
+    ],
    [
        "date_tiered_test",
        "utilities/date_tiered/date_tiered_test.cc",
@@ -556,13 +571,13 @@ ROCKS_TESTS = [
        "serial",
    ],
    [
-        "db_iter_test",
-        "db/db_iter_test.cc",
+        "db_iter_stress_test",
+        "db/db_iter_stress_test.cc",
        "serial",
    ],
    [
-        "db_iter_stress_test",
-        "db/db_iter_stress_test.cc",
+        "db_iter_test",
+        "db/db_iter_test.cc",
        "serial",
    ],
    [
@@ -685,6 +700,11 @@ ROCKS_TESTS = [
        "utilities/env_timed_test.cc",
        "serial",
    ],
+    [
+        "error_handler_test",
+        "db/error_handler_test.cc",
+        "serial",
+    ],
    [
        "event_logger_test",
        "util/event_logger_test.cc",
@@ -915,6 +935,11 @@ ROCKS_TESTS = [
        "db/repair_test.cc",
        "serial",
    ],
+    [
+        "repeatable_thread_test",
+        "util/repeatable_thread_test.cc",
+        "serial",
+    ],
    [
        "sim_cache_test",
        "utilities/simulator_cache/sim_cache_test.cc",
@@ -975,6 +1000,11 @@ ROCKS_TESTS = [
        "util/timer_queue_test.cc",
        "serial",
    ],
+    [
+        "trace_analyzer_test",
+        "tools/trace_analyzer_test.cc",
+        "serial",
+    ],
    [
        "transaction_test",
        "utilities/transactions/transaction_test.cc",
@@ -1040,6 +1070,11 @@ ROCKS_TESTS = [
        "utilities/transactions/write_prepared_transaction_test.cc",
        "parallel",
    ],
+    [
+        "write_unprepared_transaction_test",
+        "utilities/transactions/write_unprepared_transaction_test.cc",
+        "parallel",
+    ],
 ]

 # Generate a test rule for each entry in ROCKS_TESTS
@@ -1052,31 +1087,19 @@ if not is_opt_mode:
        ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
        test_bin = test_name + "_bin"

-        cpp_binary (
-          name = test_bin,
-          srcs = [test_cc],
-          deps = [":rocksdb_test_lib"],
-          preprocessor_flags = rocksdb_preprocessor_flags,
-          arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-          compiler_flags = rocksdb_compiler_flags,
-          external_deps = rocksdb_external_deps,
+        cpp_binary(
+            name = test_bin,
+            srcs = [test_cc],
+            arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
+            compiler_flags = rocksdb_compiler_flags,
+            preprocessor_flags = rocksdb_preprocessor_flags,
+            deps = [":rocksdb_test_lib"],
+            external_deps = rocksdb_external_deps,
        )

        custom_unittest(
-          name = test_name,
-          type = ttype,
-          deps = [":" + test_bin],
-          command = [TEST_RUNNER, BUCK_BINS + test_bin]
+            name = test_name,
+            command = [TEST_RUNNER, BUCK_BINS + test_bin],
+            type = ttype,
+            deps = [":" + test_bin],
        )
-
-custom_unittest(
-    name = "make_rocksdbjavastatic",
-    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
-    type = "simple",
-)
-
-custom_unittest(
-    name = "make_rocksdb_lite_release",
-    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
-    type = "simple",
-)
@@ -12,6 +12,7 @@ At Facebook, we use RocksDB as storage engines in multiple data management servi
 6. LogDevice -- a distributed data store for logs [2]

 [1] https://research.facebook.com/publications/realtime-data-processing-at-facebook/
+
 [2] https://code.facebook.com/posts/357056558062811/logdevice-a-distributed-data-store-for-logs/

 ## LinkedIn
@@ -26,7 +27,7 @@ Learn more about those use cases in a Tech Talk by Ankit Gupta and Naveen Somasu
 Yahoo is using RocksDB as a storage engine for their biggest distributed data store Sherpa. Learn more about it here: http://yahooeng.tumblr.com/post/120730204806/sherpa-scales-new-heights

 ## CockroachDB
-CockroachDB is an open-source geo-replicated transactional database (still in development). They are using RocksDB as their storage engine. Check out their github: https://github.com/cockroachdb/cockroach
+CockroachDB is an open-source geo-replicated transactional database. They are using RocksDB as their storage engine. Check out their github: https://github.com/cockroachdb/cockroach

 ## DNANexus
 DNANexus is using RocksDB to speed up processing of genomics data.
@@ -88,3 +89,6 @@ LzLabs is using RocksDB as a storage engine in their multi-database distributed

 ## ProfaneDB
 [ProfaneDB](https://profanedb.gitlab.io/) is a database for Protocol Buffers, and uses RocksDB for storage. It is accessible via gRPC, and the schema is defined using directly `.proto` files.
+
+## IOTA Foundation
+ [IOTA Foundation](https://www.iota.org/) is using RocksDB in the [IOTA Reference Implementation (IRI)](https://github.com/iotaledger/iri) to store the local state of the Tangle. The Tangle is the first open-source distributed ledger powering the future of the Internet of Things.
@@ -45,7 +45,7 @@ We plan to use this port for our business purposes here at Bing and this provide
 * All posix specific headers were replaced to port/port.h which worked well
 * Replaced `dirent.h` for `port/dirent.h` (very few places) with the implementation of the relevant interfaces within `rocksdb::port` namespace
 * Replaced `sys/time.h` to `port/sys_time.h` (few places) implemented equivalents within `rocksdb::port`
-* `printf %z` specification is not supported on Windows. To imitate existing standards we came up with a string macro `ROCKSDB_PRIszt` which expands to `%z` on posix systems and to Iu on windows.
+* `printf %z` specification is not supported on Windows. To imitate existing standards we came up with a string macro `ROCKSDB_PRIszt` which expands to `zu` on posix systems and to `Iu` on windows.
 * in class member initialization were moved to a __ctors in some cases
 * `constexpr` is not supported. We had to replace `std::numeric_limits<>::max/min()` to its C macros for constants. Sometimes we had to make class members `static const` and place a definition within a .cc file.
 * `constexpr` for functions was replaced to a template specialization (1 place)
@@ -3,14 +3,11 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
 from targets_builder import TARGETSBuilder
-from optparse import OptionParser
 import os
 import fnmatch
 import sys
-import tempfile

 from util import ColorString
-import util

 # tests to export as libraries for inclusion in other projects
 _EXPORTED_TEST_LIBS = ["env_basic_test"]
@@ -112,12 +109,14 @@ def generate_targets(repo_path):
        "rocksdb_test_lib",
        src_mk.get("MOCK_LIB_SOURCES", []) +
        src_mk.get("TEST_LIB_SOURCES", []) +
-        src_mk.get("EXP_LIB_SOURCES", []),
+        src_mk.get("EXP_LIB_SOURCES", []) +
+        src_mk.get("ANALYZER_LIB_SOURCES", []),
        [":rocksdb_lib"])
    # rocksdb_tools_lib
    TARGETS.add_library(
        "rocksdb_tools_lib",
        src_mk.get("BENCH_LIB_SOURCES", []) +
+        src_mk.get("ANALYZER_LIB_SOURCES", []) +
        ["util/testutil.cc"],
        [":rocksdb_lib"])

@@ -2,7 +2,9 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
-rocksdb_target_header = """REPO_PATH = package_name() + "/"
+rocksdb_target_header = """load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+
+REPO_PATH = package_name() + "/"

 BUCK_BINS = "buck-out/gen/" + REPO_PATH

@@ -32,7 +34,7 @@ rocksdb_compiler_flags = [
    # Added missing flags from output of build_detect_platform
    "-DROCKSDB_PTHREAD_ADAPTIVE_MUTEX",
    "-DROCKSDB_BACKTRACE",
-    "-Wshorten-64-to-32",
+    "-Wnarrowing",
 ]

 rocksdb_external_deps = [
@@ -118,32 +120,20 @@ if not is_opt_mode:
        ttype = "gtest" if test_cfg[2] == "parallel" else "simple"
        test_bin = test_name + "_bin"

-        cpp_binary (
-          name = test_bin,
-          srcs = [test_cc],
-          deps = [":rocksdb_test_lib"],
-          preprocessor_flags = rocksdb_preprocessor_flags,
-          arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-          compiler_flags = rocksdb_compiler_flags,
-          external_deps = rocksdb_external_deps,
+        cpp_binary(
+            name = test_bin,
+            srcs = [test_cc],
+            arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
+            compiler_flags = rocksdb_compiler_flags,
+            preprocessor_flags = rocksdb_preprocessor_flags,
+            deps = [":rocksdb_test_lib"],
+            external_deps = rocksdb_external_deps,
        )

        custom_unittest(
-          name = test_name,
-          type = ttype,
-          deps = [":" + test_bin],
-          command = [TEST_RUNNER, BUCK_BINS + test_bin]
+            name = test_name,
+            command = [TEST_RUNNER, BUCK_BINS + test_bin],
+            type = ttype,
+            deps = [":" + test_bin],
        )
-
-custom_unittest(
-    name = "make_rocksdbjavastatic",
-    command = ["internal_repo_rocksdb/make_rocksdbjavastatic.sh"],
-    type = "simple",
-)
-
-custom_unittest(
-    name = "make_rocksdb_lite_release",
-    command = ["internal_repo_rocksdb/make_rocksdb_lite_release.sh"],
-    type = "simple",
-)
 """
@@ -1,19 +1,18 @@
-# shellcheck disable=SC2148
-GCC_BASE=/mnt/gvfs/third-party2/gcc/8219ec1bcedf8ad9da05e121e193364de2cc4f61/5.x/centos6-native/c447969
-CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/64d8d58e3d84f8bde7a029763d4f5baf39d0d5b9/stable/centos6-native/6aaf4de
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/ba9be983c81de7299b59fe71950c664a84dcb5f8/5.x/gcc-5-glibc-2.23/339d858
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/f20197cf3d4bd50339c9777aaa0b2ccadad9e2cb/2.23/gcc-5-glibc-2.23/ca1d1c0
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/6427ce8c7496e4ab06c2da81543b94c0de8be3d0/1.1.3/gcc-5-glibc-2.23/9bc6787
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/8f1e8b867d26efef93eac2fabbdb2e1d512665d7/1.2.8/gcc-5-glibc-2.23/9bc6787
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/70471c0571559fe0af7db6d7e8860b93a7eadfe1/1.0.6/gcc-5-glibc-2.23/9bc6787
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/453c89d6f0e68cdf1c151c769197fabedad9cac8/r131/gcc-5-glibc-2.23/9bc6787
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/00a40fa5f8bd2cd0622f2e868552793aef37ccf4/1.3.0/gcc-5-glibc-2.23/03859b5
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/47eef08f9acb77de982fbda6047c26d330739538/2.2.0/gcc-5-glibc-2.23/9bc6787
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/4414ddc78df8008b35cc4adac23590ad29148584/master/gcc-5-glibc-2.23/d506c82
-NUMA_BASE=/mnt/gvfs/third-party2/numa/9d7ae2693d05d62f9a579cb21e6b717cf257a75d/2.0.11/gcc-5-glibc-2.23/9bc6787
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/2b2dd58e3a52ccf2c1d827def59e5f740de0ad15/1.2/gcc-5-glibc-2.23/b443de1
-TBB_BASE=/mnt/gvfs/third-party2/tbb/379addf7ab2468a2b4293b47456cfcd1c9cb318d/4.3/gcc-5-glibc-2.23/9bc6787
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/3f68f5fe65a85b7c2d3e66852268fbd1efdb3151/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/b9fab0aec99d9c36408e810b2677e91c12807afd/2.28/centos6-native/da39a3e
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/423431d61786b20bcc3bde8972901130cb29e6b3/3.11.0/gcc-5-glibc-2.23/9bc6787
-LUA_BASE=/mnt/gvfs/third-party2/lua/3b0bb3bd9a0f690a069c479fcc0f7424fc7456d2/5.2.3/gcc-5-glibc-2.23/65372bd
+GCC_BASE=/mnt/gvfs/third-party2/gcc/112ec378fec7002ad3e09afde022e656049f7191/5.x/centos7-native/c447969
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/04999bdb3ce81a11073535dcb00b5e13dc1cbaf5/stable/centos7-native/c9f9104
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/92b0c8e5c8eecc71eb042594ce1ab3413799b385/5.x/gcc-5-glibc-2.23/339d858
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/3d8698d5973ba94f41620a80a67e4457fdf01e90/2.23/gcc-5-glibc-2.23/ca1d1c0
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/gcc-5-glibc-2.23/9bc6787
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/22c2d65676fb7c23cfa797c4f6937f38b026f3cf/1.2.8/gcc-5-glibc-2.23/9bc6787
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/gcc-5-glibc-2.23/9bc6787
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/907b498203d297947f3bb70b9466f47e100f1873/r131/gcc-5-glibc-2.23/9bc6787
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/af6628a46758f1a15484a1760cd7294164bc5ba1/1.3.5/gcc-5-glibc-2.23/03859b5
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/gcc-5-glibc-2.23/9bc6787
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/b1a0e56c1e3e6929813a4331ade3a58ff083afbb/master/gcc-5-glibc-2.23/aa64d6b
+NUMA_BASE=/mnt/gvfs/third-party2/numa/9cbf2460284c669ed19c3ccb200a71f7dd7e53c7/2.0.11/gcc-5-glibc-2.23/9bc6787
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/bf3d7497fe4e6d007354f0adffa16ce3003f8338/1.3/gcc-5-glibc-2.23/b443de1
+TBB_BASE=/mnt/gvfs/third-party2/tbb/ff4e0b093534704d8abab678a4fd7f5ea7b094c7/2018_U5/gcc-5-glibc-2.23/9bc6787
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/b5c4a61a5c483ba24722005ae07895971a2ac707/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/55031de95a2b46c82948743419a603b3d6aefe28/2.29.1/centos7-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/f3f697a28122e6bcd513273dd9c1ff23852fc59f/3.13.0/gcc-5-glibc-2.23/9bc6787
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.2.3/gcc-5-glibc-2.23/65372bd
@@ -64,8 +64,12 @@ class MatchErrorParser(ErrorParserBase):

 class CompilerErrorParser(MatchErrorParser):
    def __init__(self):
-        # format: '<filename>:<line #>:<column #>: error: <error msg>'
-        super(CompilerErrorParser, self).__init__(r'\S+:\d+:\d+: error:')
+        # format (compile error):
+        #   '<filename>:<line #>:<column #>: error: <error msg>'
+        # format (link error):
+        #   '<filename>:<line #>: error: <error msg>'
+        # The below regex catches both
+        super(CompilerErrorParser, self).__init__(r'\S+:\d+: error:')


 class ScanBuildErrorParser(MatchErrorParser):
@@ -43,12 +43,16 @@ if test -z $PIC_BUILD; then
  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
  CFLAGS+=" -DLZ4"
-
-  ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
-  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
-  CFLAGS+=" -DZSTD"
 fi

+ZSTD_INCLUDE=" -I $ZSTD_BASE/include/"
+if test -z $PIC_BUILD; then
+  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd.a"
+else
+  ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
+fi
+CFLAGS+=" -DZSTD"
+
 # location of gflags headers and libraries
 GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
 if test -z $PIC_BUILD; then
@@ -88,6 +88,8 @@ CLANG="USE_CLANG=1"
 LITE="OPT=\"-DROCKSDB_LITE -g\""
 TSAN="COMPILE_WITH_TSAN=1"
 UBSAN="COMPILE_WITH_UBSAN=1"
+TSAN_CRASH='CRASH_TEST_EXT_ARGS="--compression_type=zstd --log2_keys_per_lock=22"'
+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"
@@ -380,14 +382,14 @@ STRESS_CRASH_TEST_COMMANDS="[
            $CLEANUP_ENV,
            {
                'name':'Build and run RocksDB debug stress tests',
-                'shell':'$SHM $DEBUG make J=1 db_stress || $CONTRUN_NAME=db_stress $TASK_CREATION_TOOL',
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 db_stress || $CONTRUN_NAME=db_stress $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            },
            {
                'name':'Build and run RocksDB debug crash tests',
                'timeout': 86400,
-                'shell':'$SHM $DEBUG make J=1 crash_test || $CONTRUN_NAME=crash_test $TASK_CREATION_TOOL',
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 crash_test || $CONTRUN_NAME=crash_test $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            }
@@ -452,7 +454,7 @@ ASAN_CRASH_TEST_COMMANDS="[
            {
                'name':'Build and run RocksDB debug asan_crash_test',
                'timeout': 86400,
-                'shell':'$SHM $DEBUG make J=1 asan_crash_test || $CONTRUN_NAME=asan_crash_test $TASK_CREATION_TOOL',
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 asan_crash_test || $CONTRUN_NAME=asan_crash_test $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            },
@@ -494,7 +496,7 @@ UBSAN_CRASH_TEST_COMMANDS="[
            {
                'name':'Build and run RocksDB debug ubsan_crash_test',
                'timeout': 86400,
-                'shell':'$SHM $DEBUG make J=1 ubsan_crash_test || $CONTRUN_NAME=ubsan_crash_test $TASK_CREATION_TOOL',
+                'shell':'$SHM $DEBUG $NON_TSAN_CRASH make J=1 ubsan_crash_test || $CONTRUN_NAME=ubsan_crash_test $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            },
@@ -560,7 +562,7 @@ TSAN_CRASH_TEST_COMMANDS="[
            {
                'name':'Compile and run',
                'timeout': 86400,
-                'shell':'set -o pipefail && $SHM $DEBUG $TSAN CRASH_TEST_KILL_ODD=1887 CRASH_TEST_EXT_ARGS=--log2_keys_per_lock=22  make J=1 crash_test || $CONTRUN_NAME=tsan_crash_test $TASK_CREATION_TOOL',
+                'shell':'set -o pipefail && $SHM $DEBUG $TSAN $TSAN_CRASH CRASH_TEST_KILL_ODD=1887 make J=1 crash_test || $CONTRUN_NAME=tsan_crash_test $TASK_CREATION_TOOL',
                'user':'root',
                $PARSER
            },
@@ -65,8 +65,8 @@ touch "$OUTPUT"
 echo "Writing dependencies to $OUTPUT"

 # Compilers locations
-GCC_BASE=`readlink -f $TP2_LATEST/gcc/5.x/centos6-native/*/`
-CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos6-native/*/`
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/5.x/centos7-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos7-native/*/`

 log_variable GCC_BASE
 log_variable CLANG_BASE
@@ -86,7 +86,7 @@ get_lib_base libunwind  LATEST  gcc-5-glibc-2.23
 get_lib_base tbb        LATEST  gcc-5-glibc-2.23

 get_lib_base kernel-headers 4.0.9-36_fbk5_2933_gd092e3f gcc-5-glibc-2.23
-get_lib_base binutils   LATEST centos6-native 
+get_lib_base binutils   LATEST centos7-native
 get_lib_base valgrind   LATEST gcc-5-glibc-2.23
 get_lib_base lua        5.2.3 gcc-5-glibc-2.23

@@ -199,7 +199,7 @@ void LRUCacheShard::LRU_Remove(LRUHandle* e) {
 void LRUCacheShard::LRU_Insert(LRUHandle* e) {
  assert(e->next == nullptr);
  assert(e->prev == nullptr);
-  if (high_pri_pool_ratio_ > 0 && e->IsHighPri()) {
+  if (high_pri_pool_ratio_ > 0 && (e->IsHighPri() || e->HasHit())) {
    // Inset "e" to head of LRU list.
    e->next = &lru_;
    e->prev = lru_.prev;
@@ -246,18 +246,6 @@ void LRUCacheShard::EvictFromLRU(size_t charge,
  }
 }

-void* LRUCacheShard::operator new(size_t size) {
-  return port::cacheline_aligned_alloc(size);
-}
-
-void* LRUCacheShard::operator new(size_t /*size*/, void* ptr) { return ptr; }
-
-void LRUCacheShard::operator delete(void *memblock) {
-  port::cacheline_aligned_free(memblock);
-}
-
-void LRUCacheShard::operator delete(void* /*memblock*/, void* /*ptr*/) {}
-
 void LRUCacheShard::SetCapacity(size_t capacity) {
  autovector<LRUHandle*> last_reference_list;
  {
@@ -287,6 +275,7 @@ Cache::Handle* LRUCacheShard::Lookup(const Slice& key, uint32_t hash) {
      LRU_Remove(e);
    }
    e->refs++;
+    e->SetHit();
  }
  return reinterpret_cast<Cache::Handle*>(e);
 }
@@ -485,10 +474,13 @@ LRUCache::LRUCache(size_t capacity, int num_shard_bits,
 }

 LRUCache::~LRUCache() {
-  for (int i = 0; i < num_shards_; i++) {
-    shards_[i].~LRUCacheShard();
+  if (shards_ != nullptr) {
+    assert(num_shards_ > 0);
+    for (int i = 0; i < num_shards_; i++) {
+      shards_[i].~LRUCacheShard();
+    }
+    port::cacheline_aligned_free(shards_);
  }
-  port::cacheline_aligned_free(shards_);
 }

 CacheShard* LRUCache::GetShard(int shard) {
@@ -513,9 +505,17 @@ uint32_t LRUCache::GetHash(Handle* handle) const {

 void LRUCache::DisownData() {
 // Do not drop data if compile with ASAN to suppress leak warning.
+#if defined(__clang__)
+#if !defined(__has_feature) || !__has_feature(address_sanitizer)
+  shards_ = nullptr;
+  num_shards_ = 0;
+#endif
+#else  // __clang__
 #ifndef __SANITIZE_ADDRESS__
  shards_ = nullptr;
+  num_shards_ = 0;
 #endif  // !__SANITIZE_ADDRESS__
+#endif  // __clang__
 }

 size_t LRUCache::TEST_GetLRUSize() {
@@ -77,6 +77,7 @@ struct LRUHandle {
  bool InCache() { return flags & 1; }
  bool IsHighPri() { return flags & 2; }
  bool InHighPriPool() { return flags & 4; }
+  bool HasHit() { return flags & 8; }

  void SetInCache(bool in_cache) {
    if (in_cache) {
@@ -102,6 +103,8 @@ struct LRUHandle {
    }
  }

+  void SetHit() { flags |= 8; }
+
  void Free() {
    assert((refs == 1 && InCache()) || (refs == 0 && !InCache()));
    if (deleter) {
@@ -206,18 +209,6 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard : public CacheShard {
  //  Retrives high pri pool ratio
  double GetHighPriPoolRatio();

-  // Overloading to aligned it to cache line size
-  // They are used by tests.
-  void* operator new(size_t);
-
-  // placement new
-  void* operator new(size_t, void*);
-
-  void operator delete(void *);
-
-  // placement delete, does nothing.
-  void operator delete(void*, void*);
-
 private:
  void LRU_Remove(LRUHandle* e);
  void LRU_Insert(LRUHandle* e);
@@ -304,7 +295,7 @@ class LRUCache : public ShardedCache {
  double GetHighPriPoolRatio();

 private:
-  LRUCacheShard* shards_;
+  LRUCacheShard* shards_ = nullptr;
  int num_shards_ = 0;
 };

@@ -15,11 +15,22 @@ namespace rocksdb {
 class LRUCacheTest : public testing::Test {
 public:
  LRUCacheTest() {}
-  ~LRUCacheTest() {}
+  ~LRUCacheTest() { DeleteCache(); }
+
+  void DeleteCache() {
+    if (cache_ != nullptr) {
+      cache_->~LRUCacheShard();
+      port::cacheline_aligned_free(cache_);
+      cache_ = nullptr;
+    }
+  }

  void NewCache(size_t capacity, double high_pri_pool_ratio = 0.0) {
-    cache_.reset(new LRUCacheShard(capacity, false /*strict_capcity_limit*/,
-                                   high_pri_pool_ratio));
+    DeleteCache();
+    cache_ = reinterpret_cast<LRUCacheShard*>(
+        port::cacheline_aligned_alloc(sizeof(LRUCacheShard)));
+    new (cache_) LRUCacheShard(capacity, false /*strict_capcity_limit*/,
+                               high_pri_pool_ratio);
  }

  void Insert(const std::string& key,
@@ -75,7 +86,7 @@ class LRUCacheTest : public testing::Test {
  }

 private:
-  std::unique_ptr<LRUCacheShard> cache_;
+  LRUCacheShard* cache_ = nullptr;
 };

 TEST_F(LRUCacheTest, BasicLRU) {
@@ -104,6 +115,29 @@ TEST_F(LRUCacheTest, BasicLRU) {
  ValidateLRUList({"e", "z", "d", "u", "v"});
 }

+TEST_F(LRUCacheTest, MidpointInsertion) {
+  // Allocate 2 cache entries to high-pri pool.
+  NewCache(5, 0.45);
+
+  Insert("a", Cache::Priority::LOW);
+  Insert("b", Cache::Priority::LOW);
+  Insert("c", Cache::Priority::LOW);
+  Insert("x", Cache::Priority::HIGH);
+  Insert("y", Cache::Priority::HIGH);
+  ValidateLRUList({"a", "b", "c", "x", "y"}, 2);
+
+  // Low-pri entries inserted to the tail of low-pri list (the midpoint).
+  // After lookup, it will move to the tail of the full list.
+  Insert("d", Cache::Priority::LOW);
+  ValidateLRUList({"b", "c", "d", "x", "y"}, 2);
+  ASSERT_TRUE(Lookup("d"));
+  ValidateLRUList({"b", "c", "x", "y", "d"}, 2);
+
+  // High-pri entries will be inserted to the tail of full list.
+  Insert("z", Cache::Priority::HIGH);
+  ValidateLRUList({"c", "x", "y", "d", "z"}, 2);
+}
+
 TEST_F(LRUCacheTest, EntriesWithPriority) {
  // Allocate 2 cache entries to high-pri pool.
  NewCache(5, 0.45);
@@ -130,15 +164,15 @@ TEST_F(LRUCacheTest, EntriesWithPriority) {
  Insert("a", Cache::Priority::LOW);
  ValidateLRUList({"v", "X", "a", "Y", "Z"}, 2);

-  // Low-pri entries will be inserted to head of low-pri pool after lookup.
+  // Low-pri entries will be inserted to head of high-pri pool after lookup.
  ASSERT_TRUE(Lookup("v"));
-  ValidateLRUList({"X", "a", "v", "Y", "Z"}, 2);
+  ValidateLRUList({"X", "a", "Y", "Z", "v"}, 2);

  // High-pri entries will be inserted to the head of the list after lookup.
  ASSERT_TRUE(Lookup("X"));
-  ValidateLRUList({"a", "v", "Y", "Z", "X"}, 2);
+  ValidateLRUList({"a", "Y", "Z", "v", "X"}, 2);
  ASSERT_TRUE(Lookup("Z"));
-  ValidateLRUList({"a", "v", "Y", "X", "Z"}, 2);
+  ValidateLRUList({"a", "Y", "v", "X", "Z"}, 2);

  Erase("Y");
  ValidateLRUList({"a", "v", "X", "Z"}, 2);
@@ -151,7 +185,7 @@ TEST_F(LRUCacheTest, EntriesWithPriority) {
  Insert("g", Cache::Priority::LOW);
  ValidateLRUList({"d", "e", "f", "g", "Z"}, 1);
  ASSERT_TRUE(Lookup("d"));
-  ValidateLRUList({"e", "f", "g", "d", "Z"}, 1);
+  ValidateLRUList({"e", "f", "g", "Z", "d"}, 2);
 }

 }  // namespace rocksdb
@@ -1,4 +1,3 @@
-import optparse
 import re
 import sys

@@ -28,6 +28,7 @@
 #include "rocksdb/options.h"
 #include "rocksdb/table.h"
 #include "table/block_based_table_builder.h"
+#include "table/format.h"
 #include "table/internal_iterator.h"
 #include "util/file_reader_writer.h"
 #include "util/filename.h"
@@ -101,7 +102,7 @@ Status BuildTable(
 #endif  // !ROCKSDB_LITE
  TableProperties tp;

-  if (iter->Valid() || range_del_agg->ShouldAddTombstones()) {
+  if (iter->Valid() || !range_del_agg->IsEmpty()) {
    TableBuilder* builder;
    unique_ptr<WritableFileWriter> file_writer;
    {
@@ -120,8 +121,8 @@ Status BuildTable(
      file->SetIOPriority(io_priority);
      file->SetWriteLifeTimeHint(write_hint);

-      file_writer.reset(new WritableFileWriter(std::move(file), env_options,
-                                               ioptions.statistics));
+      file_writer.reset(new WritableFileWriter(
+          std::move(file), fname, env_options, ioptions.statistics));
      builder = NewTableBuilder(
          ioptions, mutable_cf_options, internal_comparator,
          int_tbl_prop_collector_factories, column_family_id,
@@ -139,6 +140,7 @@ Status BuildTable(
    CompactionIterator c_iter(
        iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
        &snapshots, earliest_write_conflict_snapshot, snapshot_checker, env,
+        ShouldReportDetailedTime(env, ioptions.statistics),
        true /* internal key corruption is not ok */, range_del_agg.get());
    c_iter.SeekToFirst();
    for (; c_iter.Valid(); c_iter.Next()) {
@@ -154,12 +156,18 @@ Status BuildTable(
            ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
      }
    }
-    // nullptr for table_{min,max} so all range tombstones will be flushed
-    range_del_agg->AddToBuilder(builder, nullptr /* lower_bound */,
-                                nullptr /* upper_bound */, meta);
+
+    for (auto it = range_del_agg->NewIterator(); it->Valid(); it->Next()) {
+      auto tombstone = it->Tombstone();
+      auto kv = tombstone.Serialize();
+      builder->Add(kv.first.Encode(), kv.second);
+      meta->UpdateBoundariesForRange(kv.first, tombstone.SerializeEndKey(),
+                                     tombstone.seq_, internal_comparator);
+    }

    // Finish and check for builder errors
-    bool empty = builder->NumEntries() == 0;
+    tp = builder->GetTableProperties();
+    bool empty = builder->NumEntries() == 0 && tp.num_range_deletions == 0;
    s = c_iter.status();
    if (!s.ok() || empty) {
      builder->Abandon();
@@ -172,7 +180,7 @@ Status BuildTable(
      meta->fd.file_size = file_size;
      meta->marked_for_compaction = builder->NeedCompact();
      assert(meta->fd.GetFileSize() > 0);
-      tp = builder->GetTableProperties();
+      tp = builder->GetTableProperties(); // refresh now that builder is finished
      if (table_properties) {
        *table_properties = tp;
      }
@@ -196,7 +204,7 @@ Status BuildTable(
      // we will regrad this verification as user reads since the goal is
      // to cache it here for further user reads
      std::unique_ptr<InternalIterator> it(table_cache->NewIterator(
-          ReadOptions(), env_options, internal_comparator, meta->fd,
+          ReadOptions(), env_options, internal_comparator, *meta,
          nullptr /* range_del_agg */,
          mutable_cf_options.prefix_extractor.get(), nullptr,
          (internal_stats == nullptr) ? nullptr
@@ -35,7 +35,6 @@ class VersionEdit;
 class TableBuilder;
 class WritableFileWriter;
 class InternalStats;
-class InternalIterator;

 // @param column_family_name Name of the column family that is also identified
 //    by column_family_id, or empty string if unknown. It must outlive the
@@ -33,6 +33,7 @@
 #include "rocksdb/utilities/backupable_db.h"
 #include "rocksdb/utilities/checkpoint.h"
 #include "rocksdb/utilities/db_ttl.h"
+#include "rocksdb/utilities/memory_util.h"
 #include "rocksdb/utilities/optimistic_transaction_db.h"
 #include "rocksdb/utilities/transaction.h"
 #include "rocksdb/utilities/transaction_db.h"
@@ -41,6 +42,10 @@
 #include "rocksdb/perf_context.h"
 #include "utilities/merge_operators.h"

+#include <vector>
+#include <unordered_set>
+#include <map>
+
 using rocksdb::BytewiseComparator;
 using rocksdb::Cache;
 using rocksdb::ColumnFamilyDescriptor;
@@ -90,8 +95,10 @@ using rocksdb::LiveFileMetaData;
 using rocksdb::BackupEngine;
 using rocksdb::BackupableDBOptions;
 using rocksdb::BackupInfo;
+using rocksdb::BackupID;
 using rocksdb::RestoreOptions;
 using rocksdb::CompactRangeOptions;
+using rocksdb::BottommostLevelCompaction;
 using rocksdb::RateLimiter;
 using rocksdb::NewGenericRateLimiter;
 using rocksdb::PinnableSlice;
@@ -106,8 +113,12 @@ using rocksdb::TransactionLogIterator;
 using rocksdb::BatchResult;
 using rocksdb::PerfLevel;
 using rocksdb::PerfContext;
+using rocksdb::MemoryUtil;

 using std::shared_ptr;
+using std::vector;
+using std::unordered_set;
+using std::map;

 extern "C" {

@@ -531,10 +542,18 @@ rocksdb_backup_engine_t* rocksdb_backup_engine_open(
 }

 void rocksdb_backup_engine_create_new_backup(rocksdb_backup_engine_t* be,
-                                             rocksdb_t* db, char** errptr) {
+                                             rocksdb_t* db,
+                                             char** errptr) {
  SaveError(errptr, be->rep->CreateNewBackup(db->rep));
 }

+void rocksdb_backup_engine_create_new_backup_flush(rocksdb_backup_engine_t* be,
+                                                   rocksdb_t* db,
+                                                   unsigned char flush_before_backup,
+                                                   char** errptr) {
+  SaveError(errptr, be->rep->CreateNewBackup(db->rep, flush_before_backup));
+}
+
 void rocksdb_backup_engine_purge_old_backups(rocksdb_backup_engine_t* be,
                                             uint32_t num_backups_to_keep,
                                             char** errptr) {
@@ -554,6 +573,12 @@ void rocksdb_restore_options_set_keep_log_files(rocksdb_restore_options_t* opt,
  opt->rep.keep_log_files = v;
 }

+
+void rocksdb_backup_engine_verify_backup(rocksdb_backup_engine_t* be,
+    uint32_t backup_id, char** errptr) {
+  SaveError(errptr, be->rep->VerifyBackup(static_cast<BackupID>(backup_id)));
+}
+
 void rocksdb_backup_engine_restore_db_from_latest_backup(
    rocksdb_backup_engine_t* be, const char* db_dir, const char* wal_dir,
    const rocksdb_restore_options_t* restore_options, char** errptr) {
@@ -1969,6 +1994,11 @@ void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
  options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
 }

+void rocksdb_block_based_options_set_pin_top_level_index_and_filter(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.pin_top_level_index_and_filter = v;
+}
+
 void rocksdb_options_set_block_based_table_factory(
    rocksdb_options_t *opt,
    rocksdb_block_based_table_options_t* table_options) {
@@ -2247,6 +2277,18 @@ void rocksdb_options_set_compression_per_level(rocksdb_options_t* opt,
  }
 }

+void rocksdb_options_set_bottommost_compression_options(rocksdb_options_t* opt,
+                                                        int w_bits, int level,
+                                                        int strategy,
+                                                        int max_dict_bytes,
+                                                        bool enabled) {
+  opt->rep.bottommost_compression_opts.window_bits = w_bits;
+  opt->rep.bottommost_compression_opts.level = level;
+  opt->rep.bottommost_compression_opts.strategy = strategy;
+  opt->rep.bottommost_compression_opts.max_dict_bytes = max_dict_bytes;
+  opt->rep.bottommost_compression_opts.enabled = enabled;
+}
+
 void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
                                             int level, int strategy,
                                             int max_dict_bytes) {
@@ -2369,7 +2411,7 @@ void rocksdb_options_set_bytes_per_sync(

 void rocksdb_options_set_writable_file_max_buffer_size(rocksdb_options_t* opt,
                                                       uint64_t v) {
-  opt->rep.writable_file_max_buffer_size = v;
+  opt->rep.writable_file_max_buffer_size = static_cast<size_t>(v);
 }

 void rocksdb_options_set_allow_concurrent_memtable_write(rocksdb_options_t* opt,
@@ -2400,6 +2442,20 @@ void rocksdb_options_set_max_write_buffer_number_to_maintain(
  opt->rep.max_write_buffer_number_to_maintain = n;
 }

+void rocksdb_options_set_enable_pipelined_write(rocksdb_options_t* opt,
+                                                unsigned char v) {
+  opt->rep.enable_pipelined_write = v;
+}
+
+void rocksdb_options_set_max_subcompactions(rocksdb_options_t* opt,
+                                            uint32_t n) {
+  opt->rep.max_subcompactions = n;
+}
+
+void rocksdb_options_set_max_background_jobs(rocksdb_options_t* opt, int n) {
+  opt->rep.max_background_jobs = n;
+}
+
 void rocksdb_options_set_max_background_compactions(rocksdb_options_t* opt, int n) {
  opt->rep.max_background_compactions = n;
 }
@@ -3094,6 +3150,11 @@ void rocksdb_compactoptions_destroy(rocksdb_compactoptions_t* opt) {
  delete opt;
 }

+void rocksdb_compactoptions_set_bottommost_level_compaction(
+    rocksdb_compactoptions_t* opt, unsigned char v) {
+  opt->rep.bottommost_level_compaction = static_cast<BottommostLevelCompaction>(v);
+}
+
 void rocksdb_compactoptions_set_exclusive_manual_compaction(
    rocksdb_compactoptions_t* opt, unsigned char v) {
  opt->rep.exclusive_manual_compaction = v;
@@ -3243,6 +3304,11 @@ void rocksdb_sstfilewriter_finish(rocksdb_sstfilewriter_t* writer,
  SaveError(errptr, writer->rep->Finish(nullptr));
 }

+void rocksdb_sstfilewriter_file_size(rocksdb_sstfilewriter_t* writer,
+                                  uint64_t* file_size) {
+  *file_size = writer->rep->FileSize();
+}
+
 void rocksdb_sstfilewriter_destroy(rocksdb_sstfilewriter_t* writer) {
  delete writer->rep;
  delete writer;
@@ -4044,6 +4110,98 @@ const char* rocksdb_pinnableslice_value(const rocksdb_pinnableslice_t* v,
  *vlen = v->rep.size();
  return v->rep.data();
 }
+
+// container to keep databases and caches in order to use rocksdb::MemoryUtil
+struct rocksdb_memory_consumers_t {
+  std::vector<rocksdb_t*> dbs;
+  std::unordered_set<rocksdb_cache_t*> caches;
+};
+
+// initializes new container of memory consumers
+rocksdb_memory_consumers_t* rocksdb_memory_consumers_create() {
+  return new rocksdb_memory_consumers_t;
+}
+
+// adds datatabase to the container of memory consumers
+void rocksdb_memory_consumers_add_db(rocksdb_memory_consumers_t* consumers,
+                                     rocksdb_t* db) {
+  consumers->dbs.push_back(db);
+}
+
+// adds cache to the container of memory consumers
+void rocksdb_memory_consumers_add_cache(rocksdb_memory_consumers_t* consumers,
+                                        rocksdb_cache_t* cache) {
+  consumers->caches.insert(cache);
+}
+
+// deletes container with memory consumers
+void rocksdb_memory_consumers_destroy(rocksdb_memory_consumers_t* consumers) {
+  delete consumers;
+}
+
+// contains memory usage statistics provided by rocksdb::MemoryUtil
+struct rocksdb_memory_usage_t {
+  uint64_t mem_table_total;
+  uint64_t mem_table_unflushed;
+  uint64_t mem_table_readers_total;
+  uint64_t cache_total;
+};
+
+// estimates amount of memory occupied by consumers (dbs and caches)
+rocksdb_memory_usage_t* rocksdb_approximate_memory_usage_create(
+    rocksdb_memory_consumers_t* consumers, char** errptr) {
+
+  vector<DB*> dbs;
+  for (auto db : consumers->dbs) {
+    dbs.push_back(db->rep);
+  }
+
+  unordered_set<const Cache*> cache_set;
+  for (auto cache : consumers->caches) {
+    cache_set.insert(const_cast<const Cache*>(cache->rep.get()));
+  }
+
+  std::map<rocksdb::MemoryUtil::UsageType, uint64_t> usage_by_type;
+
+  auto status = MemoryUtil::GetApproximateMemoryUsageByType(dbs, cache_set,
+                                                            &usage_by_type);
+  if (SaveError(errptr, status)) {
+    return nullptr;
+  }
+
+  auto result = new rocksdb_memory_usage_t;
+  result->mem_table_total = usage_by_type[MemoryUtil::kMemTableTotal];
+  result->mem_table_unflushed = usage_by_type[MemoryUtil::kMemTableUnFlushed];
+  result->mem_table_readers_total = usage_by_type[MemoryUtil::kTableReadersTotal];
+  result->cache_total = usage_by_type[MemoryUtil::kCacheTotal];
+  return result;
+}
+
+uint64_t rocksdb_approximate_memory_usage_get_mem_table_total(
+    rocksdb_memory_usage_t* memory_usage) {
+  return memory_usage->mem_table_total;
+}
+
+uint64_t rocksdb_approximate_memory_usage_get_mem_table_unflushed(
+    rocksdb_memory_usage_t* memory_usage) {
+  return memory_usage->mem_table_unflushed;
+}
+
+uint64_t rocksdb_approximate_memory_usage_get_mem_table_readers_total(
+    rocksdb_memory_usage_t* memory_usage) {
+  return memory_usage->mem_table_readers_total;
+}
+
+uint64_t rocksdb_approximate_memory_usage_get_cache_total(
+    rocksdb_memory_usage_t* memory_usage) {
+  return memory_usage->cache_total;
+}
+
+// deletes container with memory usage estimates
+void rocksdb_approximate_memory_usage_destroy(rocksdb_memory_usage_t* usage) {
+  delete usage;
+}
+
 }  // end extern "C"

 #endif  // !ROCKSDB_LITE
@@ -19,11 +19,8 @@

 // Can not use port/port.h macros as this is a c file
 #ifdef OS_WIN
-
 #include <windows.h>

-#define snprintf _snprintf
-
 // Ok for uniqueness
 int geteuid() {
  int result = 0;
@@ -34,6 +31,11 @@ int geteuid() {
  return result;
 }

+// VS < 2015
+#if defined(_MSC_VER) && (_MSC_VER < 1900)
+#define snprintf _snprintf
+#endif
+
 #endif

 const char* phase = "";
@@ -47,12 +49,19 @@ static void StartPhase(const char* name) {
  fprintf(stderr, "=== Test %s\n", name);
  phase = name;
 }
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning (disable: 4996) // getenv security warning
+#endif
 static const char* GetTempDir(void) {
    const char* ret = getenv("TEST_TMPDIR");
    if (ret == NULL || ret[0] == '\0')
        ret = "/tmp";
    return ret;
 }
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif

 #define CheckNoError(err)                                               \
  if ((err) != NULL) {                                                  \
@@ -643,7 +652,7 @@ int main(int argc, char** argv) {
    rocksdb_sstfilewriter_t* writer =
        rocksdb_sstfilewriter_create(env_opt, io_options);

-    unlink(sstfilename);
+    remove(sstfilename);
    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
    CheckNoError(err);
    rocksdb_sstfilewriter_put(writer, "sstk1", 5, "v1", 2, &err);
@@ -664,7 +673,7 @@ int main(int argc, char** argv) {
    CheckGet(db, roptions, "sstk2", "v2");
    CheckGet(db, roptions, "sstk3", "v3");

-    unlink(sstfilename);
+    remove(sstfilename);
    rocksdb_sstfilewriter_open(writer, sstfilename, &err);
    CheckNoError(err);
    rocksdb_sstfilewriter_put(writer, "sstk2", 5, "v4", 2, &err);
@@ -1334,6 +1343,47 @@ int main(int argc, char** argv) {
    rocksdb_destroy_db(options, dbname, &err);
  }

+  // Check memory usage stats
+  StartPhase("approximate_memory_usage");
+  {
+    // Create database
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+
+    rocksdb_memory_consumers_t* consumers;
+    consumers = rocksdb_memory_consumers_create();
+    rocksdb_memory_consumers_add_db(consumers, db);
+    rocksdb_memory_consumers_add_cache(consumers, cache);
+
+    // take memory usage report before write-read operation
+    rocksdb_memory_usage_t* mu1;
+    mu1 = rocksdb_approximate_memory_usage_create(consumers, &err);
+    CheckNoError(err);
+
+    // Put data (this should affect memtables)
+    rocksdb_put(db, woptions, "memory", 6, "test", 4, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "memory", "test");
+
+    // take memory usage report after write-read operation
+    rocksdb_memory_usage_t* mu2;
+    mu2 = rocksdb_approximate_memory_usage_create(consumers, &err);
+    CheckNoError(err);
+
+    // amount of memory used within memtables should grow
+    CheckCondition(rocksdb_approximate_memory_usage_get_mem_table_total(mu2) >=
+                   rocksdb_approximate_memory_usage_get_mem_table_total(mu1));
+    CheckCondition(rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu2) >=
+                   rocksdb_approximate_memory_usage_get_mem_table_unflushed(mu1));
+
+    rocksdb_memory_consumers_destroy(consumers);
+    rocksdb_approximate_memory_usage_destroy(mu1);
+    rocksdb_approximate_memory_usage_destroy(mu2);
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+    CheckNoError(err);
+  }
+
  StartPhase("cuckoo_options");
  {
    rocksdb_cuckoo_table_options_t* cuckoo_options;
@@ -1675,7 +1725,7 @@ int main(int argc, char** argv) {
    db = rocksdb_open(options, dbname, &err);
    CheckNoError(err);
  }
-  
+
  StartPhase("cleanup");
  rocksdb_close(db);
  rocksdb_options_destroy(options);
@@ -57,18 +57,18 @@ class EnvCounter : public EnvWrapper {
  std::atomic<int> num_new_writable_file_;
 };

-class ColumnFamilyTest : public testing::Test {
+class ColumnFamilyTestBase : public testing::Test {
 public:
-  ColumnFamilyTest() : rnd_(139) {
+  ColumnFamilyTestBase(uint32_t format) : rnd_(139), format_(format) {
    env_ = new EnvCounter(Env::Default());
-    dbname_ = test::TmpDir() + "/column_family_test";
+    dbname_ = test::PerThreadDBPath("column_family_test");
    db_options_.create_if_missing = true;
    db_options_.fail_if_options_file_error = true;
    db_options_.env = env_;
    DestroyDB(dbname_, Options(db_options_, column_family_options_));
  }

-  ~ColumnFamilyTest() {
+  virtual ~ColumnFamilyTestBase() {
    std::vector<ColumnFamilyDescriptor> column_families;
    for (auto h : handles_) {
      ColumnFamilyDescriptor cfdescriptor;
@@ -81,6 +81,12 @@ class ColumnFamilyTest : public testing::Test {
    delete env_;
  }

+  BlockBasedTableOptions GetBlockBasedTableOptions() {
+    BlockBasedTableOptions options;
+    options.format_version = format_;
+    return options;
+  }
+
  // Return the value to associate with the specified key
  Slice Value(int k, std::string* storage) {
    if (k == 0) {
@@ -516,9 +522,22 @@ class ColumnFamilyTest : public testing::Test {
  DB* db_ = nullptr;
  EnvCounter* env_;
  Random rnd_;
+  uint32_t format_;
 };

-TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
+class ColumnFamilyTest
+    : public ColumnFamilyTestBase,
+      virtual public ::testing::WithParamInterface<uint32_t> {
+ public:
+  ColumnFamilyTest() : ColumnFamilyTestBase(GetParam()) {}
+};
+
+INSTANTIATE_TEST_CASE_P(FormatDef, ColumnFamilyTest,
+                        testing::Values(test::kDefaultFormatVersion));
+INSTANTIATE_TEST_CASE_P(FormatLatest, ColumnFamilyTest,
+                        testing::Values(test::kLatestFormatVersion));
+
+TEST_P(ColumnFamilyTest, DontReuseColumnFamilyID) {
  for (int iter = 0; iter < 3; ++iter) {
    Open();
    CreateColumnFamilies({"one", "two", "three"});
@@ -537,7 +556,8 @@ TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
      Reopen();
    }
    CreateColumnFamilies({"three2"});
-    // ID 3 that was used for dropped column family "three" should not be reused
+    // ID 3 that was used for dropped column family "three" should not be
+    // reused
    auto cfh3 = reinterpret_cast<ColumnFamilyHandleImpl*>(handles_[3]);
    ASSERT_EQ(4U, cfh3->GetID());
    Close();
@@ -546,7 +566,7 @@ TEST_F(ColumnFamilyTest, DontReuseColumnFamilyID) {
 }

 #ifndef ROCKSDB_LITE
-TEST_F(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
+TEST_P(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
  Open();

  rocksdb::SyncPoint::GetInstance()->LoadDependency(
@@ -569,10 +589,13 @@ TEST_F(ColumnFamilyTest, CreateCFRaceWithGetAggProperty) {
 }
 #endif  // !ROCKSDB_LITE

-class FlushEmptyCFTestWithParam : public ColumnFamilyTest,
-                                  public testing::WithParamInterface<bool> {
+class FlushEmptyCFTestWithParam
+    : public ColumnFamilyTestBase,
+      virtual public testing::WithParamInterface<std::tuple<uint32_t, bool>> {
 public:
-  FlushEmptyCFTestWithParam() { allow_2pc_ = GetParam(); }
+  FlushEmptyCFTestWithParam()
+      : ColumnFamilyTestBase(std::get<0>(GetParam())),
+        allow_2pc_(std::get<1>(GetParam())) {}

  // Required if inheriting from testing::WithParamInterface<>
  static void SetUpTestCase() {}
@@ -697,10 +720,16 @@ TEST_P(FlushEmptyCFTestWithParam, FlushEmptyCFTest2) {
  db_options_.env = env_;
 }

-INSTANTIATE_TEST_CASE_P(FlushEmptyCFTestWithParam, FlushEmptyCFTestWithParam,
-                        ::testing::Bool());
+INSTANTIATE_TEST_CASE_P(
+    FormatDef, FlushEmptyCFTestWithParam,
+    testing::Values(std::make_tuple(test::kDefaultFormatVersion, true),
+                    std::make_tuple(test::kDefaultFormatVersion, false)));
+INSTANTIATE_TEST_CASE_P(
+    FormatLatest, FlushEmptyCFTestWithParam,
+    testing::Values(std::make_tuple(test::kLatestFormatVersion, true),
+                    std::make_tuple(test::kLatestFormatVersion, false)));

-TEST_F(ColumnFamilyTest, AddDrop) {
+TEST_P(ColumnFamilyTest, AddDrop) {
  Open();
  CreateColumnFamilies({"one", "two", "three"});
  ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
@@ -726,7 +755,7 @@ TEST_F(ColumnFamilyTest, AddDrop) {
              std::vector<std::string>({"default", "four", "three"}));
 }

-TEST_F(ColumnFamilyTest, BulkAddDrop) {
+TEST_P(ColumnFamilyTest, BulkAddDrop) {
  constexpr int kNumCF = 1000;
  ColumnFamilyOptions cf_options;
  WriteOptions write_options;
@@ -764,7 +793,7 @@ TEST_F(ColumnFamilyTest, BulkAddDrop) {
  ASSERT_TRUE(families == std::vector<std::string>({"default"}));
 }

-TEST_F(ColumnFamilyTest, DropTest) {
+TEST_P(ColumnFamilyTest, DropTest) {
  // first iteration - dont reopen DB before dropping
  // second iteration - reopen DB before dropping
  for (int iter = 0; iter < 2; ++iter) {
@@ -788,7 +817,7 @@ TEST_F(ColumnFamilyTest, DropTest) {
  }
 }

-TEST_F(ColumnFamilyTest, WriteBatchFailure) {
+TEST_P(ColumnFamilyTest, WriteBatchFailure) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});
  WriteBatch batch;
@@ -806,7 +835,7 @@ TEST_F(ColumnFamilyTest, WriteBatchFailure) {
  Close();
 }

-TEST_F(ColumnFamilyTest, ReadWrite) {
+TEST_P(ColumnFamilyTest, ReadWrite) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});
  ASSERT_OK(Put(0, "foo", "v1"));
@@ -830,7 +859,7 @@ TEST_F(ColumnFamilyTest, ReadWrite) {
  Close();
 }

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

  // delete old files in backup_logs directory
@@ -906,7 +935,7 @@ TEST_F(ColumnFamilyTest, IgnoreRecoveredLog) {
 }

 #ifndef ROCKSDB_LITE  // TEST functions used are not supported
-TEST_F(ColumnFamilyTest, FlushTest) {
+TEST_P(ColumnFamilyTest, FlushTest) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});
  ASSERT_OK(Put(0, "foo", "v1"));
@@ -954,7 +983,7 @@ TEST_F(ColumnFamilyTest, FlushTest) {
 }

 // Makes sure that obsolete log files get deleted
-TEST_F(ColumnFamilyTest, LogDeletionTest) {
+TEST_P(ColumnFamilyTest, LogDeletionTest) {
  db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
  column_family_options_.arena_block_size = 4 * 1024;
  column_family_options_.write_buffer_size = 128000;  // 128KB
@@ -1022,7 +1051,7 @@ TEST_F(ColumnFamilyTest, LogDeletionTest) {
 }
 #endif  // !ROCKSDB_LITE

-TEST_F(ColumnFamilyTest, CrashAfterFlush) {
+TEST_P(ColumnFamilyTest, CrashAfterFlush) {
  std::unique_ptr<FaultInjectionTestEnv> fault_env(
      new FaultInjectionTestEnv(env_));
  db_options_.env = fault_env.get();
@@ -1054,7 +1083,7 @@ TEST_F(ColumnFamilyTest, CrashAfterFlush) {
  db_options_.env = env_;
 }

-TEST_F(ColumnFamilyTest, OpenNonexistentColumnFamily) {
+TEST_P(ColumnFamilyTest, OpenNonexistentColumnFamily) {
  ASSERT_OK(TryOpen({"default"}));
  Close();
  ASSERT_TRUE(TryOpen({"default", "dne"}).IsInvalidArgument());
@@ -1062,7 +1091,7 @@ TEST_F(ColumnFamilyTest, OpenNonexistentColumnFamily) {

 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
 // Makes sure that obsolete log files get deleted
-TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
+TEST_P(ColumnFamilyTest, DifferentWriteBufferSizes) {
  // disable flushing stale column families
  db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
  Open();
@@ -1168,7 +1197,7 @@ TEST_F(ColumnFamilyTest, DifferentWriteBufferSizes) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // Cuckoo is not supported in lite
-TEST_F(ColumnFamilyTest, MemtableNotSupportSnapshot) {
+TEST_P(ColumnFamilyTest, MemtableNotSupportSnapshot) {
  db_options_.allow_concurrent_memtable_write = false;
  Open();
  auto* s1 = dbfull()->GetSnapshot();
@@ -1206,7 +1235,7 @@ static TestComparator third_comparator;
 static TestComparator fourth_comparator;

 // Test that we can retrieve the comparator from a created CF
-TEST_F(ColumnFamilyTest, GetComparator) {
+TEST_P(ColumnFamilyTest, GetComparator) {
  Open();
  // Add a column family with no comparator specified
  CreateColumnFamilies({"first"});
@@ -1225,7 +1254,7 @@ TEST_F(ColumnFamilyTest, GetComparator) {
  Close();
 }

-TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
+TEST_P(ColumnFamilyTest, DifferentMergeOperators) {
  Open();
  CreateColumnFamilies({"first", "second"});
  ColumnFamilyOptions default_cf, first, second;
@@ -1256,7 +1285,7 @@ TEST_F(ColumnFamilyTest, DifferentMergeOperators) {
 }

 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
-TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
+TEST_P(ColumnFamilyTest, DifferentCompactionStyles) {
  Open();
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
@@ -1268,7 +1297,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = static_cast<uint64_t>(1) << 60;

-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1328,7 +1357,7 @@ TEST_F(ColumnFamilyTest, DifferentCompactionStyles) {
 #ifndef ROCKSDB_LITE
 // Sync points not supported in RocksDB Lite

-TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
+TEST_P(ColumnFamilyTest, MultipleManualCompactions) {
  Open();
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
@@ -1340,7 +1369,7 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1422,10 +1451,12 @@ TEST_F(ColumnFamilyTest, MultipleManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
  Close();
 }

-TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
+TEST_P(ColumnFamilyTest, AutomaticAndManualCompactions) {
  Open();
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
@@ -1437,7 +1468,8 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1516,9 +1548,11 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

-TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
+TEST_P(ColumnFamilyTest, ManualAndAutomaticCompactions) {
  Open();
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
@@ -1530,7 +1564,8 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1612,9 +1647,11 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

-TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
+TEST_P(ColumnFamilyTest, SameCFManualManualCompactions) {
  Open();
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
@@ -1626,7 +1663,8 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1711,9 +1749,11 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

-TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
+TEST_P(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
  Open();
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
@@ -1725,7 +1765,8 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1801,9 +1842,11 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

-TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
+TEST_P(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  Open();
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
@@ -1815,7 +1858,8 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1891,6 +1935,8 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

 // In this test, we generate enough files to trigger automatic compactions.
@@ -1899,7 +1945,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
 // This will wait because the automatic compaction has files it needs.
 // Once the conflict is hit, the automatic compaction starts and ends
 // Then the manual will run and end.
-TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
+TEST_P(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
  Open();
  CreateColumnFamilies({"one"});
  ColumnFamilyOptions default_cf, one;
@@ -1911,7 +1957,8 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
  default_cf.write_buffer_size = 64 << 10;  // 64KB
  default_cf.target_file_size_base = 30 << 10;
  default_cf.max_compaction_bytes = default_cf.target_file_size_base * 1100;
-  BlockBasedTableOptions table_options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  ;
  table_options.no_block_cache = true;
  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -1980,6 +2027,8 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
    ASSERT_NE("NOT_FOUND", Get(1, *key_iter));
    key_iter++;
  }
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 #endif  // !ROCKSDB_LITE

@@ -1996,7 +2045,7 @@ std::string IterStatus(Iterator* iter) {
 }
 }  // anonymous namespace

-TEST_F(ColumnFamilyTest, NewIteratorsTest) {
+TEST_P(ColumnFamilyTest, NewIteratorsTest) {
  // iter == 0 -- no tailing
  // iter == 2 -- tailing
  for (int iter = 0; iter < 2; ++iter) {
@@ -2043,7 +2092,7 @@ TEST_F(ColumnFamilyTest, NewIteratorsTest) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // ReadOnlyDB is not supported
-TEST_F(ColumnFamilyTest, ReadOnlyDBTest) {
+TEST_P(ColumnFamilyTest, ReadOnlyDBTest) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
  ASSERT_OK(Put(0, "a", "b"));
@@ -2095,7 +2144,7 @@ TEST_F(ColumnFamilyTest, ReadOnlyDBTest) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  //  WaitForFlush() is not supported in lite
-TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
+TEST_P(ColumnFamilyTest, DontRollEmptyLogs) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});

@@ -2119,7 +2168,7 @@ TEST_F(ColumnFamilyTest, DontRollEmptyLogs) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  //  WaitForCompaction() is not supported in lite
-TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
+TEST_P(ColumnFamilyTest, FlushStaleColumnFamilies) {
  Open();
  CreateColumnFamilies({"one", "two"});
  ColumnFamilyOptions default_cf, one, two;
@@ -2154,7 +2203,7 @@ TEST_F(ColumnFamilyTest, FlushStaleColumnFamilies) {
 }
 #endif  // !ROCKSDB_LITE

-TEST_F(ColumnFamilyTest, CreateMissingColumnFamilies) {
+TEST_P(ColumnFamilyTest, CreateMissingColumnFamilies) {
  Status s = TryOpen({"one", "two"});
  ASSERT_TRUE(!s.ok());
  db_options_.create_missing_column_families = true;
@@ -2163,7 +2212,7 @@ TEST_F(ColumnFamilyTest, CreateMissingColumnFamilies) {
  Close();
 }

-TEST_F(ColumnFamilyTest, SanitizeOptions) {
+TEST_P(ColumnFamilyTest, SanitizeOptions) {
  DBOptions db_options;
  for (int s = kCompactionStyleLevel; s <= kCompactionStyleUniversal; ++s) {
    for (int l = 0; l <= 2; l++) {
@@ -2212,7 +2261,7 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
  }
 }

-TEST_F(ColumnFamilyTest, ReadDroppedColumnFamily) {
+TEST_P(ColumnFamilyTest, ReadDroppedColumnFamily) {
  // iter 0 -- drop CF, don't reopen
  // iter 1 -- delete CF, reopen
  for (int iter = 0; iter < 2; ++iter) {
@@ -2284,7 +2333,7 @@ TEST_F(ColumnFamilyTest, ReadDroppedColumnFamily) {
  }
 }

-TEST_F(ColumnFamilyTest, FlushAndDropRaceCondition) {
+TEST_P(ColumnFamilyTest, FlushAndDropRaceCondition) {
  db_options_.create_missing_column_families = true;
  Open({"default", "one"});
  ColumnFamilyOptions options;
@@ -2350,12 +2399,13 @@ TEST_F(ColumnFamilyTest, FlushAndDropRaceCondition) {
 // skipped as persisting options is not supported in ROCKSDB_LITE
 namespace {
 std::atomic<int> test_stage(0);
+std::atomic<bool> ordered_by_writethread(false);
 const int kMainThreadStartPersistingOptionsFile = 1;
 const int kChildThreadFinishDroppingColumnFamily = 2;
-const int kChildThreadWaitingMainThreadPersistOptions = 3;
 void DropSingleColumnFamily(ColumnFamilyTest* cf_test, int cf_id,
                            std::vector<Comparator*>* comparators) {
-  while (test_stage < kMainThreadStartPersistingOptionsFile) {
+  while (test_stage < kMainThreadStartPersistingOptionsFile &&
+         !ordered_by_writethread) {
    Env::Default()->SleepForMicroseconds(100);
  }
  cf_test->DropColumnFamilies({cf_id});
@@ -2367,7 +2417,7 @@ void DropSingleColumnFamily(ColumnFamilyTest* cf_test, int cf_id,
 }
 }  // namespace

-TEST_F(ColumnFamilyTest, CreateAndDropRace) {
+TEST_P(ColumnFamilyTest, CreateAndDropRace) {
  const int kCfCount = 5;
  std::vector<ColumnFamilyOptions> cf_opts;
  std::vector<Comparator*> comparators;
@@ -2388,7 +2438,8 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
    // will be blocked until some other thread update the test_stage.
    if (main_thread_id == current_thread_id) {
      test_stage = kMainThreadStartPersistingOptionsFile;
-      while (test_stage < kChildThreadFinishDroppingColumnFamily) {
+      while (test_stage < kChildThreadFinishDroppingColumnFamily &&
+             !ordered_by_writethread) {
        Env::Default()->SleepForMicroseconds(100);
      }
    }
@@ -2399,11 +2450,7 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
        // This means a thread doing DropColumnFamily() is waiting for
        // other thread to finish persisting options.
        // In such case, we update the test_stage to unblock the main thread.
-        test_stage = kChildThreadWaitingMainThreadPersistOptions;
-
-        // Note that based on the test setting, this must not be the
-        // main thread.
-        ASSERT_NE(main_thread_id, std::this_thread::get_id());
+        ordered_by_writethread = true;
      });

  // Create a database with four column families
@@ -2414,7 +2461,8 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {

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

  DropColumnFamilies({2});

@@ -2426,10 +2474,13 @@ TEST_F(ColumnFamilyTest, CreateAndDropRace) {
      delete comparator;
    }
  }
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 }
 #endif  // !ROCKSDB_LITE

-TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
+TEST_P(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  const uint64_t kBaseRate = 800000u;
  db_options_.delayed_write_rate = kBaseRate;
  db_options_.max_background_compactions = 6;
@@ -2611,7 +2662,7 @@ TEST_F(ColumnFamilyTest, WriteStallSingleColumnFamily) {
  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 }

-TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
+TEST_P(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
  db_options_.max_background_compactions = 6;
  Open({"default"});
  ColumnFamilyData* cfd =
@@ -2675,7 +2726,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }

-TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
+TEST_P(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
  const uint64_t kBaseRate = 810000u;
  db_options_.delayed_write_rate = kBaseRate;
  Open();
@@ -2750,7 +2801,7 @@ TEST_F(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
  ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 }

-TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
+TEST_P(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
  db_options_.max_background_compactions = 6;
  column_family_options_.soft_pending_compaction_bytes_limit = 200;
  column_family_options_.hard_pending_compaction_bytes_limit = 2000;
@@ -2815,7 +2866,7 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
  ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }

-TEST_F(ColumnFamilyTest, CreateAndDestoryOptions) {
+TEST_P(ColumnFamilyTest, CreateAndDestoryOptions) {
  std::unique_ptr<ColumnFamilyOptions> cfo(new ColumnFamilyOptions());
  ColumnFamilyHandle* cfh;
  Open();
@@ -2827,8 +2878,29 @@ TEST_F(ColumnFamilyTest, CreateAndDestoryOptions) {
  ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
 }

+TEST_P(ColumnFamilyTest, CreateDropAndDestroy) {
+  ColumnFamilyHandle* cfh;
+  Open();
+  ASSERT_OK(db_->CreateColumnFamily(ColumnFamilyOptions(), "yoyo", &cfh));
+  ASSERT_OK(db_->Put(WriteOptions(), cfh, "foo", "bar"));
+  ASSERT_OK(db_->Flush(FlushOptions(), cfh));
+  ASSERT_OK(db_->DropColumnFamily(cfh));
+  ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
+}
+
 #ifndef ROCKSDB_LITE
-TEST_F(ColumnFamilyTest, FlushCloseWALFiles) {
+TEST_P(ColumnFamilyTest, CreateDropAndDestroyWithoutFileDeletion) {
+  ColumnFamilyHandle* cfh;
+  Open();
+  ASSERT_OK(db_->CreateColumnFamily(ColumnFamilyOptions(), "yoyo", &cfh));
+  ASSERT_OK(db_->Put(WriteOptions(), cfh, "foo", "bar"));
+  ASSERT_OK(db_->Flush(FlushOptions(), cfh));
+  ASSERT_OK(db_->DisableFileDeletions());
+  ASSERT_OK(db_->DropColumnFamily(cfh));
+  ASSERT_OK(db_->DestroyColumnFamilyHandle(cfh));
+}
+
+TEST_P(ColumnFamilyTest, FlushCloseWALFiles) {
  SpecialEnv env(Env::Default());
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
@@ -2870,7 +2942,7 @@ TEST_F(ColumnFamilyTest, FlushCloseWALFiles) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // WaitForFlush() is not supported
-TEST_F(ColumnFamilyTest, IteratorCloseWALFile1) {
+TEST_P(ColumnFamilyTest, IteratorCloseWALFile1) {
  SpecialEnv env(Env::Default());
  db_options_.env = &env;
  db_options_.max_background_flushes = 1;
@@ -2915,7 +2987,7 @@ TEST_F(ColumnFamilyTest, IteratorCloseWALFile1) {
  Close();
 }

-TEST_F(ColumnFamilyTest, IteratorCloseWALFile2) {
+TEST_P(ColumnFamilyTest, IteratorCloseWALFile2) {
  SpecialEnv env(Env::Default());
  // Allow both of flush and purge job to schedule.
  env.SetBackgroundThreads(2, Env::HIGH);
@@ -2972,7 +3044,7 @@ TEST_F(ColumnFamilyTest, IteratorCloseWALFile2) {
 #endif  // !ROCKSDB_LITE

 #ifndef ROCKSDB_LITE  // TEST functions are not supported in lite
-TEST_F(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
+TEST_P(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
  SpecialEnv env(Env::Default());
  // Allow both of flush and purge job to schedule.
  env.SetBackgroundThreads(2, Env::HIGH);
@@ -3049,7 +3121,7 @@ TEST_F(ColumnFamilyTest, ForwardIteratorCloseWALFile) {
 // Disable on windows because SyncWAL requires env->IsSyncThreadSafe()
 // to return true which is not so in unbuffered mode.
 #ifndef OS_WIN
-TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
+TEST_P(ColumnFamilyTest, LogSyncConflictFlush) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});

@@ -3084,7 +3156,7 @@ TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
 // test is being used to ensure a roll of wal files.
 // Basic idea is to test that WAL truncation is being detected and not
 // ignored
-TEST_F(ColumnFamilyTest, DISABLED_LogTruncationTest) {
+TEST_P(ColumnFamilyTest, DISABLED_LogTruncationTest) {
  Open();
  CreateColumnFamiliesAndReopen({"one", "two"});

@@ -3154,7 +3226,7 @@ TEST_F(ColumnFamilyTest, DISABLED_LogTruncationTest) {
  env_->DeleteDir(backup_logs);
 }

-TEST_F(ColumnFamilyTest, DefaultCfPathsTest) {
+TEST_P(ColumnFamilyTest, DefaultCfPathsTest) {
  Open();
  // Leave cf_paths for one column families to be empty.
  // Files should be generated according to db_paths for that
@@ -3181,7 +3253,7 @@ TEST_F(ColumnFamilyTest, DefaultCfPathsTest) {
  ASSERT_EQ(1, GetSstFileCount(dbname_));
 }

-TEST_F(ColumnFamilyTest, MultipleCFPathsTest) {
+TEST_P(ColumnFamilyTest, MultipleCFPathsTest) {
  Open();
  // Configure Column family specific paths.
  ColumnFamilyOptions cf_opt1, cf_opt2;
@@ -24,7 +24,7 @@ class CompactFilesTest : public testing::Test {
 public:
  CompactFilesTest() {
    env_ = Env::Default();
-    db_name_ = test::TmpDir(env_) + "/compact_files_test";
+    db_name_ = test::PerThreadDBPath("compact_files_test");
  }

  std::string db_name_;
@@ -308,6 +308,55 @@ TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
  delete db;
 }

+TEST_F(CompactFilesTest, SentinelCompressionType) {
+  if (!Zlib_Supported()) {
+    fprintf(stderr, "zlib compression not supported, skip this test\n");
+    return;
+  }
+  if (!Snappy_Supported()) {
+    fprintf(stderr, "snappy compression not supported, skip this test\n");
+    return;
+  }
+  // Check that passing `CompressionType::kDisableCompressionOption` to
+  // `CompactFiles` causes it to use the column family compression options.
+  for (auto compaction_style :
+       {CompactionStyle::kCompactionStyleLevel,
+        CompactionStyle::kCompactionStyleUniversal,
+        CompactionStyle::kCompactionStyleNone}) {
+    DestroyDB(db_name_, Options());
+    Options options;
+    options.compaction_style = compaction_style;
+    // L0: Snappy, L1: ZSTD, L2: Snappy
+    options.compression_per_level = {CompressionType::kSnappyCompression,
+                                     CompressionType::kZlibCompression,
+                                     CompressionType::kSnappyCompression};
+    options.create_if_missing = true;
+    FlushedFileCollector* collector = new FlushedFileCollector();
+    options.listeners.emplace_back(collector);
+    DB* db = nullptr;
+    ASSERT_OK(DB::Open(options, db_name_, &db));
+
+    db->Put(WriteOptions(), "key", "val");
+    db->Flush(FlushOptions());
+
+    auto l0_files = collector->GetFlushedFiles();
+    ASSERT_EQ(1, l0_files.size());
+
+    // L0->L1 compaction, so output should be ZSTD-compressed
+    CompactionOptions compaction_opts;
+    compaction_opts.compression = CompressionType::kDisableCompressionOption;
+    ASSERT_OK(db->CompactFiles(compaction_opts, l0_files, 1));
+
+    rocksdb::TablePropertiesCollection all_tables_props;
+    ASSERT_OK(db->GetPropertiesOfAllTables(&all_tables_props));
+    for (const auto& name_and_table_props : all_tables_props) {
+      ASSERT_EQ(CompressionTypeToString(CompressionType::kZlibCompression),
+                name_and_table_props.second->compression_name);
+    }
+    delete db;
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -25,22 +25,12 @@ CompactedDBImpl::~CompactedDBImpl() {
 }

 size_t CompactedDBImpl::FindFile(const Slice& key) {
-  size_t left = 0;
  size_t right = files_.num_files - 1;
-  while (left < right) {
-    size_t mid = (left + right) >> 1;
-    const FdWithKeyRange& f = files_.files[mid];
-    if (user_comparator_->Compare(ExtractUserKey(f.largest_key), key) < 0) {
-      // Key at "mid.largest" is < "target".  Therefore all
-      // files at or before "mid" are uninteresting.
-      left = mid + 1;
-    } else {
-      // Key at "mid.largest" is >= "target".  Therefore all files
-      // after "mid" are uninteresting.
-      right = mid;
-    }
-  }
-  return right;
+  auto cmp = [&](const FdWithKeyRange& f, const Slice& k) -> bool {
+    return user_comparator_->Compare(ExtractUserKey(f.largest_key), k) < 0;
+  };
+  return static_cast<size_t>(std::lower_bound(files_.files,
+                            files_.files + right, key, cmp) - files_.files);
 }

 Status CompactedDBImpl::Get(const ReadOptions& options, ColumnFamilyHandle*,
@@ -134,6 +134,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
                       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,
@@ -152,6 +153,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
      cfd_(nullptr),
      output_path_id_(_output_path_id),
      output_compression_(_compression),
+      output_compression_opts_(_compression_opts),
      deletion_compaction_(_deletion_compaction),
      inputs_(std::move(_inputs)),
      grandparents_(std::move(_grandparents)),
@@ -40,7 +40,7 @@ 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,
-             uint32_t max_subcompactions,
+             CompressionOptions compression_opts, uint32_t max_subcompactions,
             std::vector<FileMetaData*> grandparents,
             bool manual_compaction = false, double score = -1,
             bool deletion_compaction = false,
@@ -119,6 +119,11 @@ class Compaction {
  // What compression for output
  CompressionType output_compression() const { return output_compression_; }

+  // What compression options for output
+  CompressionOptions output_compression_opts() const {
+    return output_compression_opts_;
+  }
+
  // Whether need to write output file to second DB path.
  uint32_t output_path_id() const { return output_path_id_; }

@@ -283,6 +288,7 @@ class Compaction {

  const uint32_t output_path_id_;
  CompressionType output_compression_;
+  CompressionOptions output_compression_opts_;
  // If true, then the comaction can be done by simply deleting input files.
  const bool deletion_compaction_;

@@ -17,14 +17,15 @@ CompactionIterator::CompactionIterator(
    SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
    SequenceNumber earliest_write_conflict_snapshot,
    const SnapshotChecker* snapshot_checker, Env* env,
-    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
-    const Compaction* compaction, const CompactionFilter* compaction_filter,
+    bool report_detailed_time, bool expect_valid_internal_key,
+    RangeDelAggregator* range_del_agg, const Compaction* compaction,
+    const CompactionFilter* compaction_filter,
    const std::atomic<bool>* shutting_down,
    const SequenceNumber preserve_deletes_seqnum)
    : CompactionIterator(
          input, cmp, merge_helper, last_sequence, snapshots,
          earliest_write_conflict_snapshot, snapshot_checker, env,
-          expect_valid_internal_key, range_del_agg,
+          report_detailed_time, expect_valid_internal_key, range_del_agg,
          std::unique_ptr<CompactionProxy>(
              compaction ? new CompactionProxy(compaction) : nullptr),
          compaction_filter, shutting_down, preserve_deletes_seqnum) {}
@@ -34,7 +35,8 @@ CompactionIterator::CompactionIterator(
    SequenceNumber /*last_sequence*/, std::vector<SequenceNumber>* snapshots,
    SequenceNumber earliest_write_conflict_snapshot,
    const SnapshotChecker* snapshot_checker, Env* env,
-    bool expect_valid_internal_key, RangeDelAggregator* range_del_agg,
+    bool report_detailed_time, bool expect_valid_internal_key,
+    RangeDelAggregator* range_del_agg,
    std::unique_ptr<CompactionProxy> compaction,
    const CompactionFilter* compaction_filter,
    const std::atomic<bool>* shutting_down,
@@ -46,6 +48,7 @@ CompactionIterator::CompactionIterator(
      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      snapshot_checker_(snapshot_checker),
      env_(env),
+      report_detailed_time_(report_detailed_time),
      expect_valid_internal_key_(expect_valid_internal_key),
      range_del_agg_(range_del_agg),
      compaction_(std::move(compaction)),
@@ -74,6 +77,12 @@ CompactionIterator::CompactionIterator(
    earliest_snapshot_ = snapshots_->at(0);
    latest_snapshot_ = snapshots_->back();
  }
+#ifndef NDEBUG
+  // findEarliestVisibleSnapshot assumes this ordering.
+  for (size_t i = 1; i < snapshots_->size(); ++i) {
+    assert(snapshots_->at(i - 1) <= snapshots_->at(i));
+  }
+#endif
  if (compaction_filter_ != nullptr) {
    if (compaction_filter_->IgnoreSnapshots()) {
      ignore_snapshots_ = true;
@@ -171,12 +180,12 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
    // to get sequence number.
    Slice& filter_key = ikey_.type == kTypeValue ? ikey_.user_key : key_;
    {
-      StopWatchNano timer(env_, true);
+      StopWatchNano timer(env_, report_detailed_time_);
      filter = compaction_filter_->FilterV2(
          compaction_->level(), filter_key, value_type, value_,
          &compaction_filter_value_, compaction_filter_skip_until_.rep());
      iter_stats_.total_filter_time +=
-          env_ != nullptr ? timer.ElapsedNanos() : 0;
+          env_ != nullptr && report_detailed_time_ ? timer.ElapsedNanos() : 0;
    }

    if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil &&
@@ -502,6 +511,31 @@ void CompactionIterator::NextFromInput() {
        ++iter_stats_.num_optimized_del_drop_obsolete;
      }
      input_->Next();
+    } else if ((ikey_.type == kTypeDeletion) && bottommost_level_ &&
+               ikeyNotNeededForIncrementalSnapshot()) {
+      // Handle the case where we have a delete key at the bottom most level
+      // We can skip outputting the key iff there are no subsequent puts for this
+      // key
+      ParsedInternalKey next_ikey;
+      input_->Next();
+      // Skip over all versions of this key that happen to occur in the same snapshot
+      // range as the delete
+      while (input_->Valid() &&
+             ParseInternalKey(input_->key(), &next_ikey) &&
+             cmp_->Equal(ikey_.user_key, next_ikey.user_key) &&
+             (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
+              (snapshot_checker_ != nullptr &&
+               UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
+                                                         prev_snapshot))))) {
+        input_->Next();
+      }
+      // If you find you still need to output a row with this key, we need to output the
+      // delete too
+      if (input_->Valid() && ParseInternalKey(input_->key(), &next_ikey) &&
+          cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
+        valid_ = true;
+        at_next_ = true;
+      }
    } else if (ikey_.type == kTypeMerge) {
      if (!merge_helper_->HasOperator()) {
        status_ = Status::InvalidArgument(
@@ -551,7 +585,7 @@ void CompactionIterator::NextFromInput() {
      // 1. new user key -OR-
      // 2. different snapshot stripe
      bool should_delete = range_del_agg_->ShouldDelete(
-          key_, RangeDelAggregator::RangePositioningMode::kForwardTraversal);
+          key_, RangeDelPositioningMode::kForwardTraversal);
      if (should_delete) {
        ++iter_stats_.num_record_drop_hidden;
        ++iter_stats_.num_record_drop_range_del;
@@ -600,18 +634,23 @@ void CompactionIterator::PrepareOutput() {
 inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
    SequenceNumber in, SequenceNumber* prev_snapshot) {
  assert(snapshots_->size());
-  SequenceNumber prev = kMaxSequenceNumber;
-  for (const auto cur : *snapshots_) {
-    assert(prev == kMaxSequenceNumber || prev <= cur);
-    if (cur >= in && (snapshot_checker_ == nullptr ||
-                      snapshot_checker_->IsInSnapshot(in, cur))) {
-      *prev_snapshot = prev == kMaxSequenceNumber ? 0 : prev;
+  auto snapshots_iter = std::lower_bound(
+      snapshots_->begin(), snapshots_->end(), in);
+  if (snapshots_iter == snapshots_->begin()) {
+    *prev_snapshot = 0;
+  } else {
+    *prev_snapshot = *std::prev(snapshots_iter);
+    assert(*prev_snapshot < in);
+  }
+  for (; snapshots_iter != snapshots_->end(); ++snapshots_iter) {
+    auto cur = *snapshots_iter;
+    assert(in <= cur);
+    if (snapshot_checker_ == nullptr ||
+        snapshot_checker_->IsInSnapshot(in, cur)) {
      return cur;
    }
-    prev = cur;
-    assert(prev < kMaxSequenceNumber);
+    *prev_snapshot = cur;
  }
-  *prev_snapshot = prev;
  return kMaxSequenceNumber;
 }

@@ -63,7 +63,7 @@ class CompactionIterator {
                     std::vector<SequenceNumber>* snapshots,
                     SequenceNumber earliest_write_conflict_snapshot,
                     const SnapshotChecker* snapshot_checker, Env* env,
-                     bool expect_valid_internal_key,
+                     bool report_detailed_time, bool expect_valid_internal_key,
                     RangeDelAggregator* range_del_agg,
                     const Compaction* compaction = nullptr,
                     const CompactionFilter* compaction_filter = nullptr,
@@ -76,7 +76,7 @@ class CompactionIterator {
                     std::vector<SequenceNumber>* snapshots,
                     SequenceNumber earliest_write_conflict_snapshot,
                     const SnapshotChecker* snapshot_checker, Env* env,
-                     bool expect_valid_internal_key,
+                     bool report_detailed_time, bool expect_valid_internal_key,
                     RangeDelAggregator* range_del_agg,
                     std::unique_ptr<CompactionProxy> compaction,
                     const CompactionFilter* compaction_filter = nullptr,
@@ -139,6 +139,7 @@ class CompactionIterator {
  const SequenceNumber earliest_write_conflict_snapshot_;
  const SnapshotChecker* const snapshot_checker_;
  Env* env_;
+  bool report_detailed_time_;
  bool expect_valid_internal_key_;
  RangeDelAggregator* range_del_agg_;
  std::unique_ptr<CompactionProxy> compaction_;
@@ -247,8 +247,8 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
    c_iter_.reset(new CompactionIterator(
        iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
        earliest_write_conflict_snapshot, snapshot_checker_.get(),
-        Env::Default(), false, range_del_agg_.get(), std::move(compaction),
-        filter, &shutting_down_));
+        Env::Default(), false /* report_detailed_time */, false,
+        range_del_agg_.get(), std::move(compaction), filter, &shutting_down_));
  }

  void AddSnapshot(SequenceNumber snapshot,
@@ -671,8 +671,12 @@ TEST_P(CompactionIteratorTest, ZeroOutSequenceAtBottomLevel) {
 TEST_P(CompactionIteratorTest, RemoveDeletionAtBottomLevel) {
  AddSnapshot(1);
  RunTest({test::KeyStr("a", 1, kTypeDeletion),
-           test::KeyStr("b", 2, kTypeDeletion)},
-          {"", ""}, {test::KeyStr("b", 2, kTypeDeletion)}, {""},
+           test::KeyStr("b", 3, kTypeDeletion),
+           test::KeyStr("b", 1, kTypeValue)},
+          {"", "", ""},
+          {test::KeyStr("b", 3, kTypeDeletion),
+           test::KeyStr("b", 0, kTypeValue)},
+          {"", ""},
          kMaxSequenceNumber /*last_commited_seq*/, nullptr /*merge_operator*/,
          nullptr /*compaction_filter*/, true /*bottommost_level*/);
 }
@@ -841,13 +845,26 @@ TEST_F(CompactionIteratorWithSnapshotCheckerTest,
      {test::KeyStr("a", 1, kTypeDeletion), test::KeyStr("b", 2, kTypeDeletion),
       test::KeyStr("c", 3, kTypeDeletion)},
      {"", "", ""},
-      {test::KeyStr("b", 2, kTypeDeletion),
-       test::KeyStr("c", 3, kTypeDeletion)},
+      {},
      {"", ""}, kMaxSequenceNumber /*last_commited_seq*/,
      nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
      true /*bottommost_level*/);
 }

+TEST_F(CompactionIteratorWithSnapshotCheckerTest,
+       NotRemoveDeletionIfValuePresentToEarlierSnapshot) {
+  AddSnapshot(2,1);
+  RunTest(
+      {test::KeyStr("a", 4, kTypeDeletion), test::KeyStr("a", 1, kTypeValue),
+          test::KeyStr("b", 3, kTypeValue)},
+      {"", "", ""},
+      {test::KeyStr("a", 4, kTypeDeletion), test::KeyStr("a", 0, kTypeValue),
+            test::KeyStr("b", 3, kTypeValue)},
+      {"", "", ""}, kMaxSequenceNumber /*last_commited_seq*/,
+      nullptr /*merge_operator*/, nullptr /*compaction_filter*/,
+      true /*bottommost_level*/);
+}
+
 TEST_F(CompactionIteratorWithSnapshotCheckerTest,
       NotRemoveSingleDeletionIfNotVisibleToEarliestSnapshot) {
  AddSnapshot(2, 1);
@@ -25,8 +25,10 @@
 #include <vector>

 #include "db/builder.h"
+#include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/dbformat.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "db/log_reader.h"
 #include "db/log_writer.h"
@@ -307,7 +309,7 @@ CompactionJob::CompactionJob(
    const std::atomic<bool>* shutting_down,
    const SequenceNumber preserve_deletes_seqnum, LogBuffer* log_buffer,
    Directory* db_directory, Directory* output_directory, Statistics* stats,
-    InstrumentedMutex* db_mutex, Status* db_bg_error,
+    InstrumentedMutex* db_mutex, ErrorHandler* db_error_handler,
    std::vector<SequenceNumber> existing_snapshots,
    SequenceNumber earliest_write_conflict_snapshot,
    const SnapshotChecker* snapshot_checker, std::shared_ptr<Cache> table_cache,
@@ -331,7 +333,7 @@ CompactionJob::CompactionJob(
      output_directory_(output_directory),
      stats_(stats),
      db_mutex_(db_mutex),
-      db_bg_error_(db_bg_error),
+      db_error_handler_(db_error_handler),
      existing_snapshots_(std::move(existing_snapshots)),
      earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
      snapshot_checker_(snapshot_checker),
@@ -591,13 +593,11 @@ Status CompactionJob::Run() {
    thread.join();
  }

-  if (output_directory_) {
-    output_directory_->Fsync();
-  }
-
  compaction_stats_.micros = env_->NowMicros() - start_micros;
  MeasureTime(stats_, COMPACTION_TIME, compaction_stats_.micros);

+  TEST_SYNC_POINT("CompactionJob::Run:BeforeVerify");
+
  // Check if any thread encountered an error during execution
  Status status;
  for (const auto& state : compact_->sub_compact_states) {
@@ -607,6 +607,73 @@ Status CompactionJob::Run() {
    }
  }

+  if (status.ok() && output_directory_) {
+    status = output_directory_->Fsync();
+  }
+
+  if (status.ok()) {
+    thread_pool.clear();
+    std::vector<const FileMetaData*> files_meta;
+    for (const auto& state : compact_->sub_compact_states) {
+      for (const auto& output : state.outputs) {
+        files_meta.emplace_back(&output.meta);
+      }
+    }
+    ColumnFamilyData* cfd = compact_->compaction->column_family_data();
+    auto prefix_extractor =
+        compact_->compaction->mutable_cf_options()->prefix_extractor.get();
+    std::atomic<size_t> next_file_meta_idx(0);
+    auto verify_table = [&](Status& output_status) {
+      while (true) {
+        size_t file_idx = next_file_meta_idx.fetch_add(1);
+        if (file_idx >= files_meta.size()) {
+          break;
+        }
+        // Verify that the table is usable
+        // We set for_compaction to false and don't OptimizeForCompactionTableRead
+        // here because this is a special case after we finish the table building
+        // No matter whether use_direct_io_for_flush_and_compaction is true,
+        // we will regard this verification as user reads since the goal is
+        // to cache it here for further user reads
+        InternalIterator* iter = cfd->table_cache()->NewIterator(
+            ReadOptions(), env_options_, cfd->internal_comparator(),
+            *files_meta[file_idx], nullptr /* range_del_agg */,
+            prefix_extractor, nullptr,
+            cfd->internal_stats()->GetFileReadHist(
+                compact_->compaction->output_level()),
+            false, nullptr /* arena */, false /* skip_filters */,
+            compact_->compaction->output_level());
+        auto s = iter->status();
+
+        if (s.ok() && paranoid_file_checks_) {
+          for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {}
+          s = iter->status();
+        }
+
+        delete iter;
+
+        if (!s.ok()) {
+          output_status = s;
+          break;
+        }
+      }
+    };
+    for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
+      thread_pool.emplace_back(verify_table,
+                               std::ref(compact_->sub_compact_states[i].status));
+    }
+    verify_table(compact_->sub_compact_states[0].status);
+    for (auto& thread : thread_pool) {
+      thread.join();
+    }
+    for (const auto& state : compact_->sub_compact_states) {
+      if (!state.status.ok()) {
+        status = state.status;
+        break;
+      }
+    }
+  }
+
  TablePropertiesCollection tp;
  for (const auto& state : compact_->sub_compact_states) {
    for (const auto& output : state.outputs) {
@@ -760,10 +827,13 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  // zstd's dictionary trainer, or just use them directly. Then, the dictionary
  // is used for compressing subsequent output files in the same subcompaction.
  const bool kUseZstdTrainer =
-      cfd->ioptions()->compression_opts.zstd_max_train_bytes > 0;
+      sub_compact->compaction->output_compression_opts().zstd_max_train_bytes >
+      0;
  const size_t kSampleBytes =
-      kUseZstdTrainer ? cfd->ioptions()->compression_opts.zstd_max_train_bytes
-                      : cfd->ioptions()->compression_opts.max_dict_bytes;
+      kUseZstdTrainer
+          ? sub_compact->compaction->output_compression_opts()
+                .zstd_max_train_bytes
+          : sub_compact->compaction->output_compression_opts().max_dict_bytes;
  const int kSampleLenShift = 6;  // 2^6 = 64-byte samples
  std::set<size_t> sample_begin_offsets;
  if (bottommost_level_ && kSampleBytes > 0) {
@@ -816,9 +886,9 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
  sub_compact->c_iter.reset(new CompactionIterator(
      input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
      &existing_snapshots_, earliest_write_conflict_snapshot_,
-      snapshot_checker_, env_, false, range_del_agg.get(),
-      sub_compact->compaction, compaction_filter, shutting_down_,
-      preserve_deletes_seqnum_));
+      snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_), false,
+      range_del_agg.get(), sub_compact->compaction, compaction_filter,
+      shutting_down_, preserve_deletes_seqnum_));
  auto c_iter = sub_compact->c_iter.get();
  c_iter->SeekToFirst();
  if (c_iter->Valid() && sub_compact->compaction->output_level() != 0) {
@@ -966,7 +1036,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
        if (kUseZstdTrainer) {
          sub_compact->compression_dict = ZSTD_TrainDictionary(
              dict_sample_data, kSampleLenShift,
-              cfd->ioptions()->compression_opts.max_dict_bytes);
+              sub_compact->compaction->output_compression_opts()
+                  .max_dict_bytes);
        } else {
          sub_compact->compression_dict = std::move(dict_sample_data);
        }
@@ -1007,7 +1078,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {

  if (status.ok() && sub_compact->builder == nullptr &&
      sub_compact->outputs.size() == 0 &&
-      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
+      !range_del_agg->IsEmpty()) {
    // handle subcompaction containing only range deletions
    status = OpenCompactionOutputFile(sub_compact);
  }
@@ -1095,9 +1166,13 @@ Status CompactionJob::FinishCompactionOutputFile(
  uint64_t output_number = sub_compact->current_output()->meta.fd.GetNumber();
  assert(output_number != 0);

+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  const Comparator* ucmp = cfd->user_comparator();
+
  // Check for iterator errors
  Status s = input_status;
  auto meta = &sub_compact->current_output()->meta;
+  assert(meta != nullptr);
  if (s.ok()) {
    Slice lower_bound_guard, upper_bound_guard;
    std::string smallest_user_key;
@@ -1126,9 +1201,71 @@ Status CompactionJob::FinishCompactionOutputFile(
      // subcompaction ends.
      upper_bound = sub_compact->end;
    }
-    range_del_agg->AddToBuilder(sub_compact->builder.get(), lower_bound,
-                                upper_bound, meta, range_del_out_stats,
-                                bottommost_level_);
+    auto earliest_snapshot = kMaxSequenceNumber;
+    if (existing_snapshots_.size() > 0) {
+      earliest_snapshot = existing_snapshots_[0];
+    }
+    auto it = range_del_agg->NewIterator();
+    if (lower_bound != nullptr) {
+      it->Seek(*lower_bound);
+    }
+    for (; it->Valid(); it->Next()) {
+      auto tombstone = it->Tombstone();
+      if (upper_bound != nullptr &&
+          ucmp->Compare(*upper_bound, tombstone.start_key_) <= 0) {
+        // Tombstones starting at upper_bound or later only need to be included
+        // in the next table. Break because subsequent tombstones will start
+        // even later.
+        break;
+      }
+
+      if (bottommost_level_ && tombstone.seq_ <= earliest_snapshot) {
+        // TODO(andrewkr): tombstones that span multiple output files are
+        // counted for each compaction output file, so lots of double counting.
+        range_del_out_stats->num_range_del_drop_obsolete++;
+        range_del_out_stats->num_record_drop_obsolete++;
+        continue;
+      }
+
+      auto kv = tombstone.Serialize();
+      sub_compact->builder->Add(kv.first.Encode(), kv.second);
+      InternalKey smallest_candidate = std::move(kv.first);
+      if (lower_bound != nullptr &&
+          ucmp->Compare(smallest_candidate.user_key(), *lower_bound) <= 0) {
+        // Pretend the smallest key has the same user key as lower_bound
+        // (the max key in the previous table or subcompaction) in order for
+        // files to appear key-space partitioned.
+        //
+        // Choose lowest seqnum so this file's smallest internal key comes
+        // after the previous file's/subcompaction's largest. The fake seqnum
+        // is OK because the read path's file-picking code only considers user
+        // key.
+        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+      }
+      InternalKey largest_candidate = tombstone.SerializeEndKey();
+      if (upper_bound != nullptr &&
+          ucmp->Compare(*upper_bound, largest_candidate.user_key()) <= 0) {
+        // Pretend the largest key has the same user key as upper_bound (the
+        // min key in the following table or subcompaction) in order for files
+        // to appear key-space partitioned.
+        //
+        // Choose highest seqnum so this file's largest internal key comes
+        // before the next file's/subcompaction's smallest. The fake seqnum is
+        // OK because the read path's file-picking code only considers the user
+        // key portion.
+        //
+        // Note Seek() also creates InternalKey with (user_key,
+        // kMaxSequenceNumber), but with kTypeDeletion (0x7) instead of
+        // kTypeRangeDeletion (0xF), so the range tombstone comes before the
+        // Seek() key in InternalKey's ordering. So Seek() will look in the
+        // next file for the user key.
+        largest_candidate =
+            InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      }
+      meta->UpdateBoundariesForRange(smallest_candidate, largest_candidate,
+                                     tombstone.seq_,
+                                     cfd->internal_comparator());
+    }
    meta->marked_for_compaction = sub_compact->builder->NeedCompact();
  }
  const uint64_t current_entries = sub_compact->builder->NumEntries();
@@ -1154,7 +1291,12 @@ Status CompactionJob::FinishCompactionOutputFile(
  }
  sub_compact->outfile.reset();

-  if (s.ok() && current_entries == 0) {
+  TableProperties tp;
+  if (s.ok()) {
+    tp = sub_compact->builder->GetTableProperties();
+  }
+
+  if (s.ok() && current_entries == 0 && tp.num_range_deletions == 0) {
    // If there is nothing to output, no necessary to generate a sst file.
    // This happens when the output level is bottom level, at the same time
    // the sub_compact output nothing.
@@ -1167,51 +1309,19 @@ Status CompactionJob::FinishCompactionOutputFile(
    // VersionEdit.
    assert(!sub_compact->outputs.empty());
    sub_compact->outputs.pop_back();
-    sub_compact->builder.reset();
-    sub_compact->current_output_file_size = 0;
-    return s;
+    meta = nullptr;
  }

-  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
-  TableProperties tp;
-  if (s.ok() && current_entries > 0) {
-    // Verify that the table is usable
-    // We set for_compaction to false and don't OptimizeForCompactionTableRead
-    // here because this is a special case after we finish the table building
-    // No matter whether use_direct_io_for_flush_and_compaction is true,
-    // we will regard this verification as user reads since the goal is
-    // to cache it here for further user reads
-    InternalIterator* iter = cfd->table_cache()->NewIterator(
-        ReadOptions(), env_options_, cfd->internal_comparator(), meta->fd,
-        nullptr /* range_del_agg */,
-        sub_compact->compaction->mutable_cf_options()->prefix_extractor.get(),
-        nullptr,
-        cfd->internal_stats()->GetFileReadHist(
-            compact_->compaction->output_level()),
-        false, nullptr /* arena */, false /* skip_filters */,
-        compact_->compaction->output_level());
-    s = iter->status();
-
-    if (s.ok() && paranoid_file_checks_) {
-      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-      }
-      s = iter->status();
-    }
-
-    delete iter;
-
+  if (s.ok() && (current_entries > 0 || tp.num_range_deletions > 0)) {
    // Output to event logger and fire events.
-    if (s.ok()) {
-      tp = sub_compact->builder->GetTableProperties();
-      sub_compact->current_output()->table_properties =
-          std::make_shared<TableProperties>(tp);
-      ROCKS_LOG_INFO(db_options_.info_log,
-                     "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
-                     " keys, %" PRIu64 " bytes%s",
-                     cfd->GetName().c_str(), job_id_, output_number,
-                     current_entries, current_bytes,
-                     meta->marked_for_compaction ? " (need compaction)" : "");
-    }
+    sub_compact->current_output()->table_properties =
+        std::make_shared<TableProperties>(tp);
+    ROCKS_LOG_INFO(db_options_.info_log,
+                   "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
+                   " keys, %" PRIu64 " bytes%s",
+                   cfd->GetName().c_str(), job_id_, output_number,
+                   current_entries, current_bytes,
+                   meta->marked_for_compaction ? " (need compaction)" : "");
  }
  std::string fname;
  FileDescriptor output_fd;
@@ -1239,21 +1349,12 @@ Status CompactionJob::FinishCompactionOutputFile(
    if (sfm->IsMaxAllowedSpaceReached()) {
      // TODO(ajkr): should we return OK() if max space was reached by the final
      // compaction output file (similarly to how flush works when full)?
-      s = Status::NoSpace("Max allowed space was reached");
+      s = Status::SpaceLimit("Max allowed space was reached");
      TEST_SYNC_POINT(
          "CompactionJob::FinishCompactionOutputFile:"
          "MaxAllowedSpaceReached");
      InstrumentedMutexLock l(db_mutex_);
-      if (db_bg_error_->ok()) {
-        Status new_bg_error = s;
-        // may temporarily unlock and lock the mutex.
-        EventHelpers::NotifyOnBackgroundError(
-            cfd->ioptions()->listeners, BackgroundErrorReason::kCompaction,
-            &new_bg_error, db_mutex_);
-        if (!new_bg_error.ok()) {
-          *db_bg_error_ = new_bg_error;
-        }
-      }
+      db_error_handler_->SetBGError(s, BackgroundErrorReason::kCompaction);
    }
  }
 #endif
@@ -1362,8 +1463,9 @@ Status CompactionJob::OpenCompactionOutputFile(
  writable_file->SetWriteLifeTimeHint(write_hint_);
  writable_file->SetPreallocationBlockSize(static_cast<size_t>(
      sub_compact->compaction->OutputFilePreallocationSize()));
-  sub_compact->outfile.reset(new WritableFileWriter(
-      std::move(writable_file), env_options_, db_options_.statistics.get()));
+  sub_compact->outfile.reset(
+      new WritableFileWriter(std::move(writable_file), fname, env_options_,
+                             db_options_.statistics.get()));

  // If the Column family flag is to only optimize filters for hits,
  // we can skip creating filters if this is the bottommost_level where
@@ -1392,7 +1494,7 @@ Status CompactionJob::OpenCompactionOutputFile(
      cfd->internal_comparator(), cfd->int_tbl_prop_collector_factories(),
      cfd->GetID(), cfd->GetName(), sub_compact->outfile.get(),
      sub_compact->compaction->output_compression(),
-      cfd->ioptions()->compression_opts,
+      sub_compact->compaction->output_compression_opts(),
      sub_compact->compaction->output_level(), &sub_compact->compression_dict,
      skip_filters, output_file_creation_time));
  LogFlush(db_options_.info_log);
@@ -47,6 +47,7 @@
 namespace rocksdb {

 class Arena;
+class ErrorHandler;
 class MemTable;
 class SnapshotChecker;
 class TableCache;
@@ -64,7 +65,7 @@ class CompactionJob {
                LogBuffer* log_buffer,
                Directory* db_directory, Directory* output_directory,
                Statistics* stats, InstrumentedMutex* db_mutex,
-                Status* db_bg_error,
+                ErrorHandler* db_error_handler,
                std::vector<SequenceNumber> existing_snapshots,
                SequenceNumber earliest_write_conflict_snapshot,
                const SnapshotChecker* snapshot_checker,
@@ -145,7 +146,7 @@ class CompactionJob {
  Directory* output_directory_;
  Statistics* stats_;
  InstrumentedMutex* db_mutex_;
-  Status* db_bg_error_;
+  ErrorHandler* db_error_handler_;
  // If there were two snapshots with seq numbers s1 and
  // s2 and s1 < s2, and if we find two instances of a key k1 then lies
  // entirely within s1 and s2, then the earlier version of k1 can be safely
@@ -98,7 +98,7 @@ class CompactionJobStatsTest : public testing::Test,
  CompactionJobStatsTest() : env_(Env::Default()) {
    env_->SetBackgroundThreads(1, Env::LOW);
    env_->SetBackgroundThreads(1, Env::HIGH);
-    dbname_ = test::TmpDir(env_) + "/compaction_job_stats_test";
+    dbname_ = test::PerThreadDBPath("compaction_job_stats_test");
    alternative_wal_dir_ = dbname_ + "/wal";
    Options options;
    options.create_if_missing = true;
@@ -12,6 +12,7 @@

 #include "db/column_family.h"
 #include "db/compaction_job.h"
+#include "db/error_handler.h"
 #include "db/version_set.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/db.h"
@@ -67,7 +68,7 @@ class CompactionJobTest : public testing::Test {
 public:
  CompactionJobTest()
      : env_(Env::Default()),
-        dbname_(test::TmpDir() + "/compaction_job_test"),
+        dbname_(test::PerThreadDBPath("compaction_job_test")),
        db_options_(),
        mutable_cf_options_(cf_options_),
        table_cache_(NewLRUCache(50000, 16)),
@@ -77,7 +78,8 @@ class CompactionJobTest : public testing::Test {
                                 &write_controller_)),
        shutting_down_(false),
        preserve_deletes_seqnum_(0),
-        mock_table_factory_(new mock::MockTableFactory()) {
+        mock_table_factory_(new mock::MockTableFactory()),
+        error_handler_(nullptr, db_options_, &mutex_) {
    EXPECT_OK(env_->CreateDirIfMissing(dbname_));
    db_options_.db_paths.emplace_back(dbname_,
                                      std::numeric_limits<uint64_t>::max());
@@ -203,7 +205,7 @@ class CompactionJobTest : public testing::Test {
        manifest, &file, env_->OptimizeForManifestWrite(env_options_));
    ASSERT_OK(s);
    unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), env_options_));
+        new WritableFileWriter(std::move(file), manifest, env_options_));
    {
      log::Writer log(std::move(file_writer), 0, false);
      std::string record;
@@ -246,7 +248,8 @@ class CompactionJobTest : public testing::Test {
    Compaction compaction(cfd->current()->storage_info(), *cfd->ioptions(),
                          *cfd->GetLatestMutableCFOptions(),
                          compaction_input_files, 1, 1024 * 1024,
-                          10 * 1024 * 1024, 0, kNoCompression, 0, {}, true);
+                          10 * 1024 * 1024, 0, kNoCompression,
+                          cfd->ioptions()->compression_opts, 0, {}, true);
    compaction.SetInputVersion(cfd->current());

    LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
@@ -254,13 +257,12 @@ class CompactionJobTest : public testing::Test {
    EventLogger event_logger(db_options_.info_log.get());
    // TODO(yiwu) add a mock snapshot checker and add test for it.
    SnapshotChecker* snapshot_checker = nullptr;
-    CompactionJob compaction_job(0, &compaction, db_options_, env_options_,
-                                 versions_.get(), &shutting_down_,
-                                 preserve_deletes_seqnum_, &log_buffer,
-                                 nullptr, nullptr, nullptr, &mutex_, &bg_error_,
-                                 snapshots, earliest_write_conflict_snapshot,
-                                 snapshot_checker, table_cache_, &event_logger,
-                                 false, false, dbname_, &compaction_job_stats_);
+    CompactionJob compaction_job(
+        0, &compaction, db_options_, env_options_, versions_.get(),
+        &shutting_down_, preserve_deletes_seqnum_, &log_buffer, nullptr,
+        nullptr, nullptr, &mutex_, &error_handler_, snapshots,
+        earliest_write_conflict_snapshot, snapshot_checker, table_cache_,
+        &event_logger, false, false, dbname_, &compaction_job_stats_);
    VerifyInitializationOfCompactionJobStats(compaction_job_stats_);

    compaction_job.Prepare();
@@ -302,7 +304,7 @@ class CompactionJobTest : public testing::Test {
  ColumnFamilyData* cfd_;
  std::unique_ptr<CompactionFilter> compaction_filter_;
  std::shared_ptr<MergeOperator> merge_op_;
-  Status bg_error_;
+  ErrorHandler error_handler_;
 };

 TEST_F(CompactionJobTest, Simple) {
@@ -49,7 +49,7 @@ bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
  // increasing.
  size_t new_compact_bytes_per_del_file = 0;
  for (span_len = 1; span_len < level_files.size(); ++span_len) {
-    compact_bytes += level_files[span_len]->fd.file_size;
+    compact_bytes += static_cast<size_t>(level_files[span_len]->fd.file_size);
    new_compact_bytes_per_del_file = compact_bytes / span_len;
    if (level_files[span_len]->being_compacted ||
        new_compact_bytes_per_del_file > compact_bytes_per_del_file) {
@@ -110,6 +110,24 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
  }
 }

+CompressionOptions GetCompressionOptions(const ImmutableCFOptions& ioptions,
+                                         const VersionStorageInfo* vstorage,
+                                         int level,
+                                         const bool enable_compression) {
+  if (!enable_compression) {
+    return ioptions.compression_opts;
+  }
+  // If bottommost_compression is set and we are compacting to the
+  // bottommost level then we should use the specified compression options
+  // for the bottmomost_compression.
+  if (ioptions.bottommost_compression != kDisableCompressionOption &&
+      level >= (vstorage->num_non_empty_levels() - 1) &&
+      ioptions.bottommost_compression_opts.enabled) {
+    return ioptions.bottommost_compression_opts;
+  }
+  return ioptions.compression_opts;
+}
+
 CompactionPicker::CompactionPicker(const ImmutableCFOptions& ioptions,
                                   const InternalKeyComparator* icmp)
    : ioptions_(ioptions), icmp_(icmp) {}
@@ -201,7 +219,8 @@ void CompactionPicker::GetRange(const std::vector<CompactionInputFiles>& inputs,

 bool CompactionPicker::ExpandInputsToCleanCut(const std::string& /*cf_name*/,
                                              VersionStorageInfo* vstorage,
-                                              CompactionInputFiles* inputs) {
+                                              CompactionInputFiles* inputs,
+                                              InternalKey** next_smallest) {
  // This isn't good compaction
  assert(!inputs->empty());

@@ -224,7 +243,8 @@ bool CompactionPicker::ExpandInputsToCleanCut(const std::string& /*cf_name*/,
    GetRange(*inputs, &smallest, &largest);
    inputs->clear();
    vstorage->GetOverlappingInputs(level, &smallest, &largest, &inputs->files,
-                                   hint_index, &hint_index);
+                                   hint_index, &hint_index, true,
+                                   next_smallest);
  } while (inputs->size() > old_size);

  // we started off with inputs non-empty and the previous loop only grew
@@ -297,13 +317,29 @@ Compaction* CompactionPicker::CompactFiles(
  // shouldn't have been released since.
  assert(!FilesRangeOverlapWithCompaction(input_files, output_level));

-  auto c =
-      new Compaction(vstorage, ioptions_, mutable_cf_options, input_files,
-                     output_level, compact_options.output_file_size_limit,
-                     mutable_cf_options.max_compaction_bytes, output_path_id,
-                     compact_options.compression,
-                     compact_options.max_subcompactions,
-                     /* grandparents */ {}, true);
+  CompressionType compression_type;
+  if (compact_options.compression == kDisableCompressionOption) {
+    int base_level;
+    if (ioptions_.compaction_style == kCompactionStyleLevel) {
+      base_level = vstorage->base_level();
+    } else {
+      base_level = 1;
+    }
+    compression_type =
+        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
+                           output_level, base_level);
+  } else {
+    // TODO(ajkr): `CompactionOptions` offers configurable `CompressionType`
+    // without configurable `CompressionOptions`, which is inconsistent.
+    compression_type = compact_options.compression;
+  }
+  auto c = new Compaction(
+      vstorage, ioptions_, mutable_cf_options, input_files, output_level,
+      compact_options.output_file_size_limit,
+      mutable_cf_options.max_compaction_bytes, output_path_id, compression_type,
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      compact_options.max_subcompactions,
+      /* grandparents */ {}, true);
  RegisterCompaction(c);
  return c;
 }
@@ -366,7 +402,7 @@ bool CompactionPicker::IsRangeInCompaction(VersionStorageInfo* vstorage,
  assert(level < NumberLevels());

  vstorage->GetOverlappingInputs(level, smallest, largest, &inputs,
-                                 *level_index, level_index);
+                                 level_index ? *level_index : 0, level_index);
  return AreFilesInCompaction(inputs);
 }

@@ -569,11 +605,13 @@ Compaction* CompactionPicker::CompactRange(

    Compaction* c = new Compaction(
        vstorage, ioptions_, mutable_cf_options, std::move(inputs),
-        output_level, MaxFileSizeForLevel(mutable_cf_options, output_level,
-            ioptions_.compaction_style),
+        output_level,
+        MaxFileSizeForLevel(mutable_cf_options, output_level,
+                            ioptions_.compaction_style),
        /* max_compaction_bytes */ LLONG_MAX, output_path_id,
        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
                           output_level, 1),
+        GetCompressionOptions(ioptions_, vstorage, output_level),
        max_subcompactions, /* grandparents */ {}, /* is manual */ true);
    RegisterCompaction(c);
    return c;
@@ -613,7 +651,6 @@ Compaction* CompactionPicker::CompactRange(
      uint64_t s = inputs[i]->compensated_file_size;
      total += s;
      if (total >= limit) {
-        **compaction_end = inputs[i + 1]->smallest;
        covering_the_whole_range = false;
        inputs.files.resize(i + 1);
        break;
@@ -622,7 +659,10 @@ Compaction* CompactionPicker::CompactRange(
  }
  assert(output_path_id < static_cast<uint32_t>(ioptions_.cf_paths.size()));

-  if (ExpandInputsToCleanCut(cf_name, vstorage, &inputs) == false) {
+  InternalKey key_storage;
+  InternalKey* next_smallest = &key_storage;
+  if (ExpandInputsToCleanCut(cf_name, vstorage, &inputs, &next_smallest) ==
+      false) {
    // manual compaction is now multi-threaded, so it can
    // happen that ExpandWhileOverlapping fails
    // we handle it higher in RunManualCompaction
@@ -630,8 +670,10 @@ Compaction* CompactionPicker::CompactRange(
    return nullptr;
  }

-  if (covering_the_whole_range) {
+  if (covering_the_whole_range || !next_smallest) {
    *compaction_end = nullptr;
+  } else {
+    **compaction_end = *next_smallest;
  }

  CompactionInputFiles output_level_inputs;
@@ -679,12 +721,13 @@ Compaction* CompactionPicker::CompactRange(
      vstorage, ioptions_, mutable_cf_options, std::move(compaction_inputs),
      output_level,
      MaxFileSizeForLevel(mutable_cf_options, output_level,
-          ioptions_.compaction_style, vstorage->base_level(),
-          ioptions_.level_compaction_dynamic_level_bytes),
+                          ioptions_.compaction_style, vstorage->base_level(),
+                          ioptions_.level_compaction_dynamic_level_bytes),
      mutable_cf_options.max_compaction_bytes, output_path_id,
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
                         vstorage->base_level()),
-      /* max_subcompactions */ 0, std::move(grandparents),
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      max_subcompactions, std::move(grandparents),
      /* is manual compaction */ true);

  TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
@@ -853,6 +896,11 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
      }
    }
  }
+  if (RangeOverlapWithCompaction(smallestkey, largestkey, output_level)) {
+    return Status::Aborted(
+        "A running compaction is writing to the same output level in an "
+        "overlapping key range");
+  }
  return Status::OK();
 }

@@ -947,6 +995,78 @@ void CompactionPicker::UnregisterCompaction(Compaction* c) {
  compactions_in_progress_.erase(c);
 }

+void CompactionPicker::PickFilesMarkedForCompaction(
+    const std::string& cf_name, VersionStorageInfo* vstorage, int* start_level,
+    int* output_level, CompactionInputFiles* start_level_inputs) {
+  if (vstorage->FilesMarkedForCompaction().empty()) {
+    return;
+  }
+
+  auto continuation = [&, cf_name](std::pair<int, FileMetaData*> level_file) {
+    // If it's being compacted it has nothing to do here.
+    // If this assert() fails that means that some function marked some
+    // files as being_compacted, but didn't call ComputeCompactionScore()
+    assert(!level_file.second->being_compacted);
+    *start_level = level_file.first;
+    *output_level =
+        (*start_level == 0) ? vstorage->base_level() : *start_level + 1;
+
+    if (*start_level == 0 && !level0_compactions_in_progress()->empty()) {
+      return false;
+    }
+
+    start_level_inputs->files = {level_file.second};
+    start_level_inputs->level = *start_level;
+    return ExpandInputsToCleanCut(cf_name, vstorage, start_level_inputs);
+  };
+
+  // take a chance on a random file first
+  Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage));
+  size_t random_file_index = static_cast<size_t>(rnd.Uniform(
+      static_cast<uint64_t>(vstorage->FilesMarkedForCompaction().size())));
+
+  if (continuation(vstorage->FilesMarkedForCompaction()[random_file_index])) {
+    // found the compaction!
+    return;
+  }
+
+  for (auto& level_file : vstorage->FilesMarkedForCompaction()) {
+    if (continuation(level_file)) {
+      // found the compaction!
+      return;
+    }
+  }
+  start_level_inputs->files.clear();
+}
+
+bool CompactionPicker::GetOverlappingL0Files(
+    VersionStorageInfo* vstorage, CompactionInputFiles* start_level_inputs,
+    int output_level, int* parent_index) {
+  // Two level 0 compaction won't run at the same time, so don't need to worry
+  // about files on level 0 being compacted.
+  assert(level0_compactions_in_progress()->empty());
+  InternalKey smallest, largest;
+  GetRange(*start_level_inputs, &smallest, &largest);
+  // Note that the next call will discard the file we placed in
+  // c->inputs_[0] earlier and replace it with an overlapping set
+  // which will include the picked file.
+  start_level_inputs->files.clear();
+  vstorage->GetOverlappingInputs(0, &smallest, &largest,
+                                 &(start_level_inputs->files));
+
+  // If we include more L0 files in the same compaction run it can
+  // cause the 'smallest' and 'largest' key to get extended to a
+  // larger range. So, re-invoke GetRange to get the new key range
+  GetRange(*start_level_inputs, &smallest, &largest);
+  if (IsRangeInCompaction(vstorage, &smallest, &largest, output_level,
+                          parent_index)) {
+    return false;
+  }
+  assert(!start_level_inputs->files.empty());
+
+  return true;
+}
+
 bool LevelCompactionPicker::NeedsCompaction(
    const VersionStorageInfo* vstorage) const {
  if (!vstorage->ExpiredTtlFiles().empty()) {
@@ -1018,9 +1138,6 @@ class LevelCompactionBuilder {
  // otherwise, returns false.
  bool PickIntraL0Compaction();

-  // If there is any file marked for compaction, put put it into inputs.
-  void PickFilesMarkedForCompaction();
-
  void PickExpiredTtlFiles();

  const std::string& cf_name_;
@@ -1049,50 +1166,6 @@ class LevelCompactionBuilder {
  static const int kMinFilesForIntraL0Compaction = 4;
 };

-void LevelCompactionBuilder::PickFilesMarkedForCompaction() {
-  if (vstorage_->FilesMarkedForCompaction().empty()) {
-    return;
-  }
-
-  auto continuation = [&](std::pair<int, FileMetaData*> level_file) {
-    // If it's being compacted it has nothing to do here.
-    // If this assert() fails that means that some function marked some
-    // files as being_compacted, but didn't call ComputeCompactionScore()
-    assert(!level_file.second->being_compacted);
-    start_level_ = level_file.first;
-    output_level_ =
-        (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1;
-
-    if (start_level_ == 0 &&
-        !compaction_picker_->level0_compactions_in_progress()->empty()) {
-      return false;
-    }
-
-    start_level_inputs_.files = {level_file.second};
-    start_level_inputs_.level = start_level_;
-    return compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_,
-                                                      &start_level_inputs_);
-  };
-
-  // take a chance on a random file first
-  Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage_));
-  size_t random_file_index = static_cast<size_t>(rnd.Uniform(
-      static_cast<uint64_t>(vstorage_->FilesMarkedForCompaction().size())));
-
-  if (continuation(vstorage_->FilesMarkedForCompaction()[random_file_index])) {
-    // found the compaction!
-    return;
-  }
-
-  for (auto& level_file : vstorage_->FilesMarkedForCompaction()) {
-    if (continuation(level_file)) {
-      // found the compaction!
-      return;
-    }
-  }
-  start_level_inputs_.files.clear();
-}
-
 void LevelCompactionBuilder::PickExpiredTtlFiles() {
  if (vstorage_->ExpiredTtlFiles().empty()) {
    return;
@@ -1182,7 +1255,9 @@ void LevelCompactionBuilder::SetupInitialFiles() {
  if (start_level_inputs_.empty()) {
    parent_index_ = base_index_ = -1;

-    PickFilesMarkedForCompaction();
+    // PickFilesMarkedForCompaction();
+    compaction_picker_->PickFilesMarkedForCompaction(
+        cf_name_, vstorage_, &start_level_, &output_level_, &start_level_inputs_);
    if (!start_level_inputs_.empty()) {
      is_manual_ = true;
      compaction_reason_ = CompactionReason::kFilesMarkedForCompaction;
@@ -1220,29 +1295,9 @@ void LevelCompactionBuilder::SetupInitialFiles() {

 bool LevelCompactionBuilder::SetupOtherL0FilesIfNeeded() {
  if (start_level_ == 0 && output_level_ != 0) {
-    // Two level 0 compaction won't run at the same time, so don't need to worry
-    // about files on level 0 being compacted.
-    assert(compaction_picker_->level0_compactions_in_progress()->empty());
-    InternalKey smallest, largest;
-    compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest);
-    // Note that the next call will discard the file we placed in
-    // c->inputs_[0] earlier and replace it with an overlapping set
-    // which will include the picked file.
-    start_level_inputs_.files.clear();
-    vstorage_->GetOverlappingInputs(0, &smallest, &largest,
-                                    &start_level_inputs_.files);
-
-    // If we include more L0 files in the same compaction run it can
-    // cause the 'smallest' and 'largest' key to get extended to a
-    // larger range. So, re-invoke GetRange to get the new key range
-    compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest);
-    if (compaction_picker_->IsRangeInCompaction(
-            vstorage_, &smallest, &largest, output_level_, &parent_index_)) {
-      return false;
-    }
+    return compaction_picker_->GetOverlappingL0Files(
+        vstorage_, &start_level_inputs_, output_level_, &parent_index_);
  }
-  assert(!start_level_inputs_.files.empty());
-
  return true;
 }

@@ -1317,12 +1372,13 @@ Compaction* LevelCompactionBuilder::GetCompaction() {
      vstorage_, ioptions_, mutable_cf_options_, std::move(compaction_inputs_),
      output_level_,
      MaxFileSizeForLevel(mutable_cf_options_, output_level_,
-          ioptions_.compaction_style, vstorage_->base_level(),
-          ioptions_.level_compaction_dynamic_level_bytes),
+                          ioptions_.compaction_style, vstorage_->base_level(),
+                          ioptions_.level_compaction_dynamic_level_bytes),
      mutable_cf_options_.max_compaction_bytes,
      GetPathId(ioptions_, mutable_cf_options_, output_level_),
      GetCompressionType(ioptions_, vstorage_, mutable_cf_options_,
                         output_level_, vstorage_->base_level()),
+      GetCompressionOptions(ioptions_, vstorage_, output_level_),
      /* max_subcompactions */ 0, std::move(grandparents_), is_manual_,
      start_level_score_, false /* deletion_compaction */, compaction_reason_);

@@ -1572,9 +1628,9 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(

  Compaction* c = new Compaction(
      vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
-      kNoCompression, /* max_subcompactions */ 0, {}, /* is manual */ false,
-      vstorage->CompactionScore(0), /* is deletion compaction */ true,
-      CompactionReason::kFIFOTtl);
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
+      /* is deletion compaction */ true, CompactionReason::kFIFOTtl);
  return c;
 }

@@ -1612,8 +1668,9 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
            16 * 1024 * 1024 /* output file size limit */,
            0 /* max compaction bytes, not applicable */,
            0 /* output path ID */, mutable_cf_options.compression,
-            0 /* max_subcompactions */, {}, /* is manual */ false,
-            vstorage->CompactionScore(0), /* is deletion compaction */ false,
+            ioptions_.compression_opts, 0 /* max_subcompactions */, {},
+            /* is manual */ false, vstorage->CompactionScore(0),
+            /* is deletion compaction */ false,
            CompactionReason::kFIFOReduceNumFiles);
        return c;
      }
@@ -1659,9 +1716,9 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(

  Compaction* c = new Compaction(
      vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
-      kNoCompression, /* max_subcompactions */ 0, {}, /* is manual */ false,
-      vstorage->CompactionScore(0), /* is deletion compaction */ true,
-      CompactionReason::kFIFOMaxSize);
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
+      /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
  return c;
 }

@@ -151,7 +151,8 @@ class CompactionPicker {
  // Will return false if it is impossible to apply this compaction.
  bool ExpandInputsToCleanCut(const std::string& cf_name,
                              VersionStorageInfo* vstorage,
-                              CompactionInputFiles* inputs);
+                              CompactionInputFiles* inputs,
+                              InternalKey** next_smallest = nullptr);

  // Returns true if any one of the parent files are being compacted
  bool IsRangeInCompaction(VersionStorageInfo* vstorage,
@@ -175,6 +176,15 @@ class CompactionPicker {
                       const CompactionInputFiles& output_level_inputs,
                       std::vector<FileMetaData*>* grandparents);

+  void PickFilesMarkedForCompaction(const std::string& cf_name,
+                                    VersionStorageInfo* vstorage,
+                                    int* start_level, int* output_level,
+                                    CompactionInputFiles* start_level_inputs);
+
+  bool GetOverlappingL0Files(VersionStorageInfo* vstorage,
+                             CompactionInputFiles* start_level_inputs,
+                             int output_level, int* parent_index);
+
  // Register this compaction in the set of running compactions
  void RegisterCompaction(Compaction* c);

@@ -304,4 +314,9 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
                                   int level, int base_level,
                                   const bool enable_compression = true);

+CompressionOptions GetCompressionOptions(const ImmutableCFOptions& ioptions,
+                                         const VersionStorageInfo* vstorage,
+                                         int level,
+                                         const bool enable_compression = true);
+
 }  // namespace rocksdb
@@ -90,8 +90,8 @@ class CompactionPickerTest : public testing::Test {
    f->fd = FileDescriptor(file_number, path_id, file_size);
    f->smallest = InternalKey(smallest, smallest_seq, kTypeValue);
    f->largest = InternalKey(largest, largest_seq, kTypeValue);
-    f->smallest_seqno = smallest_seq;
-    f->largest_seqno = largest_seq;
+    f->fd.smallest_seqno = smallest_seq;
+    f->fd.largest_seqno = largest_seq;
    f->compensated_file_size = file_size;
    f->refs = 0;
    vstorage_->AddFile(level, f);
@@ -391,10 +391,10 @@ TEST_F(CompactionPickerTest, NeedsCompactionUniversal) {
  NewVersionStorage(1, kCompactionStyleUniversal);
  UniversalCompactionPicker universal_compaction_picker(
      ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
  // must return false when there's no files.
  ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
            false);
-  UpdateVersionStorageInfo();

  // verify the trigger given different number of L0 files.
  for (int i = 1;
@@ -415,6 +415,7 @@ TEST_F(CompactionPickerTest, CompactionUniversalIngestBehindReservedLevel) {
  ioptions_.allow_ingest_behind = true;
  ioptions_.num_levels = 3;
  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
  // must return false when there's no files.
  ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
            false);
@@ -448,6 +449,7 @@ TEST_F(CompactionPickerTest, CannotTrivialMoveUniversal) {
  mutable_cf_options_.compaction_options_universal.allow_trivial_move = true;
  NewVersionStorage(1, kCompactionStyleUniversal);
  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+  UpdateVersionStorageInfo();
  // must return false when there's no files.
  ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
            false);
@@ -97,17 +97,17 @@ void GetSmallestLargestSeqno(const std::vector<FileMetaData*>& files,
                             SequenceNumber* largest_seqno) {
  bool is_first = true;
  for (FileMetaData* f : files) {
-    assert(f->smallest_seqno <= f->largest_seqno);
+    assert(f->fd.smallest_seqno <= f->fd.largest_seqno);
    if (is_first) {
      is_first = false;
-      *smallest_seqno = f->smallest_seqno;
-      *largest_seqno = f->largest_seqno;
+      *smallest_seqno = f->fd.smallest_seqno;
+      *largest_seqno = f->fd.largest_seqno;
    } else {
-      if (f->smallest_seqno < *smallest_seqno) {
-        *smallest_seqno = f->smallest_seqno;
+      if (f->fd.smallest_seqno < *smallest_seqno) {
+        *smallest_seqno = f->fd.smallest_seqno;
      }
-      if (f->largest_seqno > *largest_seqno) {
-        *largest_seqno = f->largest_seqno;
+      if (f->fd.largest_seqno > *largest_seqno) {
+        *largest_seqno = f->fd.largest_seqno;
      }
    }
  }
@@ -162,7 +162,13 @@ bool UniversalCompactionPicker::IsInputFilesNonOverlapping(Compaction* c) {
 bool UniversalCompactionPicker::NeedsCompaction(
    const VersionStorageInfo* vstorage) const {
  const int kLevel0 = 0;
-  return vstorage->CompactionScore(kLevel0) >= 1;
+  if (vstorage->CompactionScore(kLevel0) >= 1) {
+    return true;
+  }
+  if (!vstorage->FilesMarkedForCompaction().empty()) {
+    return true;
+  }
+  return false;
 }

 void UniversalCompactionPicker::SortedRun::Dump(char* out_buf,
@@ -257,8 +263,9 @@ Compaction* UniversalCompactionPicker::PickCompaction(
      CalculateSortedRuns(*vstorage, ioptions_, mutable_cf_options);

  if (sorted_runs.size() == 0 ||
-      sorted_runs.size() <
-          (unsigned int)mutable_cf_options.level0_file_num_compaction_trigger) {
+      (vstorage->FilesMarkedForCompaction().empty() &&
+       sorted_runs.size() < (unsigned int)mutable_cf_options
+                                .level0_file_num_compaction_trigger)) {
    ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: nothing to do\n",
                     cf_name.c_str());
    TEST_SYNC_POINT_CALLBACK("UniversalCompactionPicker::PickCompaction:Return",
@@ -272,58 +279,73 @@ Compaction* UniversalCompactionPicker::PickCompaction(
      cf_name.c_str(), sorted_runs.size(), vstorage->LevelSummary(&tmp));

  // Check for size amplification first.
-  Compaction* c;
-  if ((c = PickCompactionToReduceSizeAmp(cf_name, mutable_cf_options, vstorage,
-                                         score, sorted_runs, log_buffer)) !=
-      nullptr) {
-    ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: compacting for size amp\n",
-                     cf_name.c_str());
-  } else {
-    // Size amplification is within limits. Try reducing read
-    // amplification while maintaining file size ratios.
-    unsigned int ratio =
-        mutable_cf_options.compaction_options_universal.size_ratio;
-
-    if ((c = PickCompactionToReduceSortedRuns(
-             cf_name, mutable_cf_options, vstorage, score, ratio, UINT_MAX,
-             sorted_runs, log_buffer)) != nullptr) {
-      ROCKS_LOG_BUFFER(log_buffer,
-                       "[%s] Universal: compacting for size ratio\n",
+  Compaction* c = nullptr;
+  if (sorted_runs.size() >=
+      static_cast<size_t>(
+          mutable_cf_options.level0_file_num_compaction_trigger)) {
+    if ((c = PickCompactionToReduceSizeAmp(cf_name, mutable_cf_options,
+                                           vstorage, score, sorted_runs,
+                                           log_buffer)) != nullptr) {
+      ROCKS_LOG_BUFFER(log_buffer, "[%s] Universal: compacting for size amp\n",
                       cf_name.c_str());
    } else {
-      // Size amplification and file size ratios are within configured limits.
-      // If max read amplification is exceeding configured limits, then force
-      // compaction without looking at filesize ratios and try to reduce
-      // the number of files to fewer than level0_file_num_compaction_trigger.
-      // This is guaranteed by NeedsCompaction()
-      assert(sorted_runs.size() >=
-             static_cast<size_t>(
-                 mutable_cf_options.level0_file_num_compaction_trigger));
-      // Get the total number of sorted runs that are not being compacted
-      int num_sr_not_compacted = 0;
-      for (size_t i = 0; i < sorted_runs.size(); i++) {
-        if (sorted_runs[i].being_compacted == false) {
-          num_sr_not_compacted++;
-        }
-      }
+      // Size amplification is within limits. Try reducing read
+      // amplification while maintaining file size ratios.
+      unsigned int ratio =
+          mutable_cf_options.compaction_options_universal.size_ratio;

-      // The number of sorted runs that are not being compacted is greater than
-      // the maximum allowed number of sorted runs
-      if (num_sr_not_compacted >
-          mutable_cf_options.level0_file_num_compaction_trigger) {
-        unsigned int num_files =
-            num_sr_not_compacted -
-            mutable_cf_options.level0_file_num_compaction_trigger + 1;
-        if ((c = PickCompactionToReduceSortedRuns(
-                 cf_name, mutable_cf_options, vstorage, score, UINT_MAX,
-                 num_files, sorted_runs, log_buffer)) != nullptr) {
-          ROCKS_LOG_BUFFER(log_buffer,
-                           "[%s] Universal: compacting for file num -- %u\n",
-                           cf_name.c_str(), num_files);
+      if ((c = PickCompactionToReduceSortedRuns(
+               cf_name, mutable_cf_options, vstorage, score, ratio, UINT_MAX,
+               sorted_runs, log_buffer)) != nullptr) {
+        ROCKS_LOG_BUFFER(log_buffer,
+                         "[%s] Universal: compacting for size ratio\n",
+                         cf_name.c_str());
+      } else {
+        // Size amplification and file size ratios are within configured limits.
+        // If max read amplification is exceeding configured limits, then force
+        // compaction without looking at filesize ratios and try to reduce
+        // the number of files to fewer than level0_file_num_compaction_trigger.
+        // This is guaranteed by NeedsCompaction()
+        assert(sorted_runs.size() >=
+               static_cast<size_t>(
+                   mutable_cf_options.level0_file_num_compaction_trigger));
+        // Get the total number of sorted runs that are not being compacted
+        int num_sr_not_compacted = 0;
+        for (size_t i = 0; i < sorted_runs.size(); i++) {
+          if (sorted_runs[i].being_compacted == false) {
+            num_sr_not_compacted++;
+          }
+        }
+
+        // The number of sorted runs that are not being compacted is greater
+        // than the maximum allowed number of sorted runs
+        if (num_sr_not_compacted >
+            mutable_cf_options.level0_file_num_compaction_trigger) {
+          unsigned int num_files =
+              num_sr_not_compacted -
+              mutable_cf_options.level0_file_num_compaction_trigger + 1;
+          if ((c = PickCompactionToReduceSortedRuns(
+                   cf_name, mutable_cf_options, vstorage, score, UINT_MAX,
+                   num_files, sorted_runs, log_buffer)) != nullptr) {
+            ROCKS_LOG_BUFFER(log_buffer,
+                             "[%s] Universal: compacting for file num -- %u\n",
+                             cf_name.c_str(), num_files);
+          }
        }
      }
    }
  }
+
+  if (c == nullptr) {
+    if ((c = PickDeleteTriggeredCompaction(cf_name, mutable_cf_options,
+                                           vstorage, score, sorted_runs,
+                                           log_buffer)) != nullptr) {
+      ROCKS_LOG_BUFFER(log_buffer,
+                       "[%s] Universal: delete triggered compaction\n",
+                       cf_name.c_str());
+    }
+  }
+
  if (c == nullptr) {
    TEST_SYNC_POINT_CALLBACK("UniversalCompactionPicker::PickCompaction:Return",
                             nullptr);
@@ -343,11 +365,11 @@ Compaction* UniversalCompactionPicker::PickCompaction(
  size_t level_index = 0U;
  if (c->start_level() == 0) {
    for (auto f : *c->inputs(0)) {
-      assert(f->smallest_seqno <= f->largest_seqno);
+      assert(f->fd.smallest_seqno <= f->fd.largest_seqno);
      if (is_first) {
        is_first = false;
      }
-      prev_smallest_seqno = f->smallest_seqno;
+      prev_smallest_seqno = f->fd.smallest_seqno;
    }
    level_index = 1U;
  }
@@ -614,6 +636,8 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSortedRuns(
      LLONG_MAX, path_id,
      GetCompressionType(ioptions_, vstorage, mutable_cf_options, start_level,
                         1, enable_compression),
+      GetCompressionOptions(ioptions_, vstorage, start_level,
+                            enable_compression),
      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
      score, false /* deletion_compaction */, compaction_reason);
 }
@@ -637,6 +661,10 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
  size_t start_index = 0;
  const SortedRun* sr = nullptr;

+  if (sorted_runs.back().being_compacted) {
+    return nullptr;
+  }
+
  // Skip files that are already being compacted
  for (size_t loop = 0; loop < sorted_runs.size() - 1; loop++) {
    sr = &sorted_runs[loop];
@@ -743,16 +771,137 @@ Compaction* UniversalCompactionPicker::PickCompactionToReduceSizeAmp(
  }

  return new Compaction(
-      vstorage, ioptions_, mutable_cf_options, std::move(inputs),
-      output_level, MaxFileSizeForLevel(mutable_cf_options, output_level,
-          kCompactionStyleUniversal),
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          kCompactionStyleUniversal),
      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, vstorage, mutable_cf_options,
-                         output_level, 1),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         1),
+      GetCompressionOptions(ioptions_, vstorage, output_level),
      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ false,
      score, false /* deletion_compaction */,
      CompactionReason::kUniversalSizeAmplification);
 }
+
+// Pick files marked for compaction. Typically, files are marked by
+// CompactOnDeleteCollector due to the presence of tombstones.
+Compaction* UniversalCompactionPicker::PickDeleteTriggeredCompaction(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, double score,
+    const std::vector<SortedRun>& /*sorted_runs*/, LogBuffer* /*log_buffer*/) {
+  CompactionInputFiles start_level_inputs;
+  int output_level;
+  std::vector<CompactionInputFiles> inputs;
+
+  if (vstorage->num_levels() == 1) {
+    // This is single level universal. Since we're basically trying to reclaim
+    // space by processing files marked for compaction due to high tombstone
+    // density, let's do the same thing as compaction to reduce size amp which
+    // has the same goals.
+    bool compact = false;
+
+    start_level_inputs.level = 0;
+    start_level_inputs.files.clear();
+    output_level = 0;
+    for (FileMetaData* f : vstorage->LevelFiles(0)) {
+      if (f->marked_for_compaction) {
+        compact = true;
+      }
+      if (compact) {
+        start_level_inputs.files.push_back(f);
+      }
+    }
+    if (start_level_inputs.size() <= 1) {
+      // If only the last file in L0 is marked for compaction, ignore it
+      return nullptr;
+    }
+    inputs.push_back(start_level_inputs);
+  } else {
+    int start_level;
+
+    // For multi-level universal, the strategy is to make this look more like
+    // leveled. We pick one of the files marked for compaction and compact with
+    // overlapping files in the adjacent level.
+    PickFilesMarkedForCompaction(cf_name, vstorage, &start_level, &output_level,
+                                 &start_level_inputs);
+    if (start_level_inputs.empty()) {
+      return nullptr;
+    }
+
+    // Pick the first non-empty level after the start_level
+    for (output_level = start_level + 1; output_level < vstorage->num_levels();
+         output_level++) {
+      if (vstorage->NumLevelFiles(output_level) != 0) {
+        break;
+      }
+    }
+
+    // If all higher levels are empty, pick the highest level as output level
+    if (output_level == vstorage->num_levels()) {
+      if (start_level == 0) {
+        output_level = vstorage->num_levels() - 1;
+      } else {
+        // If start level is non-zero and all higher levels are empty, this
+        // compaction will translate into a trivial move. Since the idea is
+        // to reclaim space and trivial move doesn't help with that, we
+        // skip compaction in this case and return nullptr
+        return nullptr;
+      }
+    }
+    if (ioptions_.allow_ingest_behind &&
+        output_level == vstorage->num_levels() - 1) {
+      assert(output_level > 1);
+      output_level--;
+    }
+
+    if (output_level != 0) {
+      if (start_level == 0) {
+        if (!GetOverlappingL0Files(vstorage, &start_level_inputs, output_level,
+                                   nullptr)) {
+          return nullptr;
+        }
+      }
+
+      CompactionInputFiles output_level_inputs;
+      int parent_index = -1;
+
+      output_level_inputs.level = output_level;
+      if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage,
+                            &start_level_inputs, &output_level_inputs,
+                            &parent_index, -1)) {
+        return nullptr;
+      }
+      inputs.push_back(start_level_inputs);
+      if (!output_level_inputs.empty()) {
+        inputs.push_back(output_level_inputs);
+      }
+      if (FilesRangeOverlapWithCompaction(inputs, output_level)) {
+        return nullptr;
+      }
+    } else {
+      inputs.push_back(start_level_inputs);
+    }
+  }
+
+  uint64_t estimated_total_size = 0;
+  // Use size of the output level as estimated file size
+  for (FileMetaData* f : vstorage->LevelFiles(output_level)) {
+    estimated_total_size += f->fd.GetFileSize();
+  }
+  uint32_t path_id =
+      GetPathId(ioptions_, mutable_cf_options, estimated_total_size);
+  return new Compaction(
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), output_level,
+      MaxFileSizeForLevel(mutable_cf_options, output_level,
+                          kCompactionStyleUniversal),
+      /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         1),
+      GetCompressionOptions(ioptions_, vstorage, output_level),
+      /* max_subcompactions */ 0, /* grandparents */ {}, /* is manual */ true,
+      score, false /* deletion_compaction */,
+      CompactionReason::kFilesMarkedForCompaction);
+}
 }  // namespace rocksdb

 #endif  // !ROCKSDB_LITE
@@ -73,6 +73,11 @@ class UniversalCompactionPicker : public CompactionPicker {
      VersionStorageInfo* vstorage, double score,
      const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer);

+  Compaction* PickDeleteTriggeredCompaction(
+      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+      VersionStorageInfo* vstorage, double score,
+      const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer);
+
  // Used in universal compaction when the enabled_trivial_move
  // option is set. Checks whether there are any overlapping files
  // in the input. Returns true if the input files are non
@@ -251,7 +251,9 @@ class TwoStrComparator : public Comparator {
 };
 }  // namespace

-class ComparatorDBTest : public testing::Test {
+class ComparatorDBTest
+    : public testing::Test,
+      virtual public ::testing::WithParamInterface<uint32_t> {
 private:
  std::string dbname_;
  Env* env_;
@@ -262,7 +264,11 @@ class ComparatorDBTest : public testing::Test {
 public:
  ComparatorDBTest() : env_(Env::Default()), db_(nullptr) {
    comparator = BytewiseComparator();
-    dbname_ = test::TmpDir() + "/comparator_db_test";
+    dbname_ = test::PerThreadDBPath("comparator_db_test");
+    BlockBasedTableOptions toptions;
+    toptions.format_version = GetParam();
+    last_options_.table_factory.reset(
+        rocksdb::NewBlockBasedTableFactory(toptions));
    EXPECT_OK(DestroyDB(dbname_, last_options_));
  }

@@ -274,8 +280,12 @@ class ComparatorDBTest : public testing::Test {

  DB* GetDB() { return db_; }

-  void SetOwnedComparator(const Comparator* cmp) {
-    comparator_guard.reset(cmp);
+  void SetOwnedComparator(const Comparator* cmp, bool owner = true) {
+    if (owner) {
+      comparator_guard.reset(cmp);
+    } else {
+      comparator_guard.reset();
+    }
    comparator = cmp;
    last_options_.comparator = cmp;
  }
@@ -304,7 +314,12 @@ class ComparatorDBTest : public testing::Test {
  }
 };

-TEST_F(ComparatorDBTest, Bytewise) {
+INSTANTIATE_TEST_CASE_P(FormatDef, ComparatorDBTest,
+                        testing::Values(test::kDefaultFormatVersion));
+INSTANTIATE_TEST_CASE_P(FormatLatest, ComparatorDBTest,
+                        testing::Values(test::kLatestFormatVersion));
+
+TEST_P(ComparatorDBTest, Bytewise) {
  for (int rand_seed = 301; rand_seed < 306; rand_seed++) {
    DestroyAndReopen();
    Random rnd(rand_seed);
@@ -314,7 +329,7 @@ TEST_F(ComparatorDBTest, Bytewise) {
  }
 }

-TEST_F(ComparatorDBTest, SimpleSuffixReverseComparator) {
+TEST_P(ComparatorDBTest, SimpleSuffixReverseComparator) {
  SetOwnedComparator(new test::SimpleSuffixReverseComparator());

  for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -340,8 +355,8 @@ TEST_F(ComparatorDBTest, SimpleSuffixReverseComparator) {
  }
 }

-TEST_F(ComparatorDBTest, Uint64Comparator) {
-  SetOwnedComparator(test::Uint64Comparator());
+TEST_P(ComparatorDBTest, Uint64Comparator) {
+  SetOwnedComparator(test::Uint64Comparator(), false /* owner */);

  for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
    Options* opt = GetOptions();
@@ -364,7 +379,7 @@ TEST_F(ComparatorDBTest, Uint64Comparator) {
  }
 }

-TEST_F(ComparatorDBTest, DoubleComparator) {
+TEST_P(ComparatorDBTest, DoubleComparator) {
  SetOwnedComparator(new DoubleComparator());

  for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -389,7 +404,7 @@ TEST_F(ComparatorDBTest, DoubleComparator) {
  }
 }

-TEST_F(ComparatorDBTest, HashComparator) {
+TEST_P(ComparatorDBTest, HashComparator) {
  SetOwnedComparator(new HashComparator());

  for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -408,7 +423,7 @@ TEST_F(ComparatorDBTest, HashComparator) {
  }
 }

-TEST_F(ComparatorDBTest, TwoStrComparator) {
+TEST_P(ComparatorDBTest, TwoStrComparator) {
  SetOwnedComparator(new TwoStrComparator());

  for (int rnd_seed = 301; rnd_seed < 316; rnd_seed++) {
@@ -434,7 +449,71 @@ TEST_F(ComparatorDBTest, TwoStrComparator) {
  }
 }

-TEST_F(ComparatorDBTest, FindShortestSeparator) {
+TEST_P(ComparatorDBTest, IsSameLengthImmediateSuccessor) {
+  {
+    // different length
+    Slice s("abcxy");
+    Slice t("abcxyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxyz");
+    Slice t("abcxy");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    // not last byte different
+    Slice s("abc1xyz");
+    Slice t("abc2xyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    // same string
+    Slice s("abcxyz");
+    Slice t("abcxyz");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxy");
+    Slice t("abcxz");
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    Slice s("abcxz");
+    Slice t("abcxy");
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xac";
+    const char t_array[] = "\x50\x8a\xad";
+    Slice s(s_array);
+    Slice t(t_array);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff";
+    const char t_array[] = "\x50\x8b\x00";
+    Slice s(s_array, 3);
+    Slice t(t_array, 3);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff\xff";
+    const char t_array[] = "\x50\x8b\x00\x00";
+    Slice s(s_array, 4);
+    Slice t(t_array, 4);
+    ASSERT_TRUE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+  {
+    const char s_array[] = "\x50\x8a\xff\xff";
+    const char t_array[] = "\x50\x8b\x00\x01";
+    Slice s(s_array, 4);
+    Slice t(t_array, 4);
+    ASSERT_FALSE(BytewiseComparator()->IsSameLengthImmediateSuccessor(s, t));
+  }
+}
+
+TEST_P(ComparatorDBTest, FindShortestSeparator) {
  std::string s1 = "abc1xyz";
  std::string s2 = "abc3xy";

@@ -468,7 +547,7 @@ TEST_F(ComparatorDBTest, FindShortestSeparator) {
  ASSERT_TRUE(s1 > s2);
 }

-TEST_F(ComparatorDBTest, SeparatorSuccessorRandomizeTest) {
+TEST_P(ComparatorDBTest, SeparatorSuccessorRandomizeTest) {
  // Char list for boundary cases.
  std::array<unsigned char, 6> char_list{{0, 1, 2, 253, 254, 255}};
  Random rnd(301);
@@ -49,10 +49,11 @@ Status VerifySstFileChecksum(const Options& options,
  unique_ptr<TableReader> table_reader;
  std::unique_ptr<RandomAccessFileReader> file_reader(
      new RandomAccessFileReader(std::move(file), file_path));
+  const bool kImmortal = true;
  s = ioptions.table_factory->NewTableReader(
      TableReaderOptions(ioptions, options.prefix_extractor.get(), env_options,
                         internal_comparator, false /* skip_filters */,
-                         -1 /* level */),
+                         !kImmortal, -1 /* level */),
      std::move(file_reader), file_size, &table_reader,
      false /* prefetch_index_and_filter_in_cache */);
  if (!s.ok()) {
@@ -48,7 +48,7 @@ class CorruptionTest : public testing::Test {
    tiny_cache_ = NewLRUCache(100, 4);
    options_.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
    options_.env = &env_;
-    dbname_ = test::TmpDir() + "/corruption_test";
+    dbname_ = test::PerThreadDBPath("corruption_test");
    DestroyDB(dbname_, options_);

    db_ = nullptr;
@@ -25,7 +25,7 @@ class CuckooTableDBTest : public testing::Test {

 public:
  CuckooTableDBTest() : env_(Env::Default()) {
-    dbname_ = test::TmpDir() + "/cuckoo_table_db_test";
+    dbname_ = test::PerThreadDBPath("cuckoo_table_db_test");
    EXPECT_OK(DestroyDB(dbname_, Options()));
    db_ = nullptr;
    Reopen();
@@ -892,7 +892,7 @@ class TestEnv : public EnvWrapper {

 TEST_F(DBBasicTest, DBClose) {
  Options options = GetDefaultOptions();
-  std::string dbname = test::TmpDir(env_) + "/db_close_test";
+  std::string dbname = test::PerThreadDBPath("db_close_test");
  ASSERT_OK(DestroyDB(dbname, options));

  DB* db = nullptr;
@@ -390,7 +390,10 @@ class MockCache : public LRUCache {
  static uint32_t high_pri_insert_count;
  static uint32_t low_pri_insert_count;

-  MockCache() : LRUCache(1 << 25, 0, false, 0.0) {}
+  MockCache()
+      : LRUCache((size_t)1 << 25 /*capacity*/, 0 /*num_shard_bits*/,
+                 false /*strict_capacity_limit*/, 0.0 /*high_pri_pool_ratio*/) {
+  }

  virtual Status Insert(const Slice& key, void* value, size_t charge,
                        void (*deleter)(const Slice& key, void* value),
@@ -22,11 +22,12 @@ class DBBloomFilterTest : public DBTestBase {

 class DBBloomFilterTestWithParam
    : public DBTestBase,
-      public testing::WithParamInterface<std::tuple<bool, bool>> {
+      public testing::WithParamInterface<std::tuple<bool, bool, uint32_t>> {
  //                             public testing::WithParamInterface<bool> {
 protected:
  bool use_block_based_filter_;
  bool partition_filters_;
+  uint32_t format_version_;

 public:
  DBBloomFilterTestWithParam() : DBTestBase("/db_bloom_filter_tests") {}
@@ -36,13 +37,36 @@ class DBBloomFilterTestWithParam
  void SetUp() override {
    use_block_based_filter_ = std::get<0>(GetParam());
    partition_filters_ = std::get<1>(GetParam());
+    format_version_ = std::get<2>(GetParam());
+  }
+};
+
+class DBBloomFilterTestDefFormatVersion : public DBBloomFilterTestWithParam {};
+
+class SliceTransformLimitedDomainGeneric : public SliceTransform {
+  const char* Name() const override {
+    return "SliceTransformLimitedDomainGeneric";
+  }
+
+  Slice Transform(const Slice& src) const override {
+    return Slice(src.data(), 5);
+  }
+
+  bool InDomain(const Slice& src) const override {
+    // prefix will be x????
+    return src.size() >= 5;
+  }
+
+  bool InRange(const Slice& dst) const override {
+    // prefix will be x????
+    return dst.size() == 5;
  }
 };

 // KeyMayExist can lead to a few false positives, but not false negatives.
 // To make test deterministic, use a much larger number of bits per key-20 than
 // bits in the key, so that false positives are eliminated
-TEST_P(DBBloomFilterTestWithParam, KeyMayExist) {
+TEST_P(DBBloomFilterTestDefFormatVersion, KeyMayExist) {
  do {
    ReadOptions ropts;
    std::string value;
@@ -117,6 +141,53 @@ TEST_P(DBBloomFilterTestWithParam, KeyMayExist) {
      ChangeOptions(kSkipPlainTable | kSkipHashIndex | kSkipFIFOCompaction));
 }

+TEST_F(DBBloomFilterTest, GetFilterByPrefixBloomCustomPrefixExtractor) {
+  for (bool partition_filters : {true, false}) {
+    Options options = last_options_;
+    options.prefix_extractor =
+        std::make_shared<SliceTransformLimitedDomainGeneric>();
+    options.statistics = rocksdb::CreateDBStatistics();
+    BlockBasedTableOptions bbto;
+    bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
+    if (partition_filters) {
+      bbto.partition_filters = true;
+      bbto.index_type = BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
+    }
+    bbto.whole_key_filtering = false;
+    options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+    DestroyAndReopen(options);
+
+    WriteOptions wo;
+    ReadOptions ro;
+    FlushOptions fo;
+    fo.wait = true;
+    std::string value;
+
+    ASSERT_OK(dbfull()->Put(wo, "barbarbar", "foo"));
+    ASSERT_OK(dbfull()->Put(wo, "barbarbar2", "foo2"));
+    ASSERT_OK(dbfull()->Put(wo, "foofoofoo", "bar"));
+
+    dbfull()->Flush(fo);
+
+    ASSERT_EQ("foo", Get("barbarbar"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ("foo2", Get("barbarbar2"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ("NOT_FOUND", Get("barbarbar3"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+
+    ASSERT_EQ("NOT_FOUND", Get("barfoofoo"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 1);
+
+    ASSERT_EQ("NOT_FOUND", Get("foobarbar"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+
+    ro.total_order_seek = true;
+    ASSERT_TRUE(db_->Get(ro, "foobarbar", &value).IsNotFound());
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+  }
+}
+
 TEST_F(DBBloomFilterTest, GetFilterByPrefixBloom) {
  for (bool partition_filters : {true, false}) {
    Options options = last_options_;
@@ -334,6 +405,11 @@ TEST_P(DBBloomFilterTestWithParam, BloomFilter) {
      table_options.index_type =
          BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
    }
+    table_options.format_version = format_version_;
+    if (format_version_ >= 4) {
+      // value delta encoding challenged more with index interval > 1
+      table_options.index_block_restart_interval = 8;
+    }
    table_options.metadata_block_size = 32;
    options.table_factory.reset(NewBlockBasedTableFactory(table_options));

@@ -389,10 +465,26 @@ TEST_P(DBBloomFilterTestWithParam, BloomFilter) {
  } while (ChangeCompactOptions());
 }

-INSTANTIATE_TEST_CASE_P(DBBloomFilterTestWithParam, DBBloomFilterTestWithParam,
-                        ::testing::Values(std::make_tuple(true, false),
-                                          std::make_tuple(false, true),
-                                          std::make_tuple(false, false)));
+INSTANTIATE_TEST_CASE_P(
+    FormatDef, DBBloomFilterTestDefFormatVersion,
+    ::testing::Values(std::make_tuple(true, false, test::kDefaultFormatVersion),
+                      std::make_tuple(false, true, test::kDefaultFormatVersion),
+                      std::make_tuple(false, false,
+                                      test::kDefaultFormatVersion)));
+
+INSTANTIATE_TEST_CASE_P(
+    FormatDef, DBBloomFilterTestWithParam,
+    ::testing::Values(std::make_tuple(true, false, test::kDefaultFormatVersion),
+                      std::make_tuple(false, true, test::kDefaultFormatVersion),
+                      std::make_tuple(false, false,
+                                      test::kDefaultFormatVersion)));
+
+INSTANTIATE_TEST_CASE_P(
+    FormatLatest, DBBloomFilterTestWithParam,
+    ::testing::Values(std::make_tuple(true, false, test::kLatestFormatVersion),
+                      std::make_tuple(false, true, test::kLatestFormatVersion),
+                      std::make_tuple(false, false,
+                                      test::kLatestFormatVersion)));

 TEST_F(DBBloomFilterTest, BloomFilterRate) {
  while (ChangeFilterOptions()) {
@@ -1097,6 +1189,413 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
            TestGetTickerCount(options, BLOCK_CACHE_ADD));
 }

+int CountIter(std::unique_ptr<Iterator>& iter, const Slice& key) {
+  int count = 0;
+  for (iter->Seek(key); iter->Valid() && iter->status() == Status::OK();
+       iter->Next()) {
+    count++;
+  }
+  return count;
+}
+
+// use iterate_upper_bound to hint compatiability of existing bloom filters.
+// The BF is considered compatible if 1) upper bound and seek key transform
+// into the same string, or 2) the transformed seek key is of the same length
+// as the upper bound and two keys are adjacent according to the comparator.
+TEST_F(DBBloomFilterTest, DynamicBloomFilterUpperBound) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewCappedPrefixTransform(4));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    ASSERT_OK(Put("abcdxxx0", "val1"));
+    ASSERT_OK(Put("abcdxxx1", "val2"));
+    ASSERT_OK(Put("abcdxxx2", "val3"));
+    ASSERT_OK(Put("abcdxxx3", "val4"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // prefix_extractor has not changed, BF will always be read
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcd0000"), 4);
+    }
+    {
+      Slice upper_bound("abcdzzzz");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcd0000"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:5"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.FixedPrefix.5"));
+    {
+      // BF changed, [abcdxx00, abce) is a valid bound, will trigger BF read
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx00"), 4);
+      // should check bloom filter since upper bound meets requirement
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [abcdxx01, abcey) is not valid bound since upper bound is too long for
+      // the BF in SST (capped:4)
+      Slice upper_bound("abcey");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx01"), 4);
+      // should skip bloom filter since upper bound is too long
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [abcdxx02, abcdy) is a valid bound since the prefix is the same
+      Slice upper_bound("abcdy");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abcdxx02"), 4);
+      // should check bloom filter since upper bound matches transformed seek
+      // key
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    {
+      // [aaaaaaaa, abce) is not a valid bound since 1) they don't share the
+      // same prefix, 2) the prefixes are not consecutive
+      Slice upper_bound("abce");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "aaaaaaaa"), 0);
+      // should skip bloom filter since mismatch is found
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:3"}}));
+    {
+      // [abc, abd) is not a valid bound since the upper bound is too short
+      // for BF (capped:4)
+      Slice upper_bound("abd");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abc"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:4"}}));
+    {
+      // set back to capped:4 and verify BF is always read
+      Slice upper_bound("abd");
+      ReadOptions read_options;
+      read_options.prefix_same_as_start = true;
+      read_options.iterate_upper_bound = &upper_bound;
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "abc"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                3 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+    }
+    iteration++;
+  }
+}
+
+// Create multiple SST files each with a different prefix_extractor config,
+// verify iterators can read all SST files using the latest config.
+TEST_F(DBBloomFilterTest, DynamicBloomFilterMultipleSST) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    table_options.cache_index_and_filter_blocks = true;
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    Slice upper_bound("foz90000");
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+
+    // first SST with fixed:1 BF
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_OK(Put("foq1", "bar1"));
+    ASSERT_OK(Put("fpa", "0"));
+    dbfull()->Flush(FlushOptions());
+    std::unique_ptr<Iterator> iter_old(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter_old, "foo"), 4);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 1);
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    read_options.iterate_upper_bound = &upper_bound;
+    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter, "foo"), 2);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              1 + iteration);
+    ASSERT_EQ(CountIter(iter, "gpk"), 0);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              1 + iteration);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+
+    // second SST with capped:3 BF
+    ASSERT_OK(Put("foo3", "bar3"));
+    ASSERT_OK(Put("foo4", "bar4"));
+    ASSERT_OK(Put("foq5", "bar5"));
+    ASSERT_OK(Put("fpb", "1"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // BF is cappped:3 now
+      std::unique_ptr<Iterator> iter_tmp(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_tmp, "foo"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                2 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+      ASSERT_EQ(CountIter(iter_tmp, "gpk"), 0);
+      // both counters are incremented because BF is "not changed" for 1 of the
+      // 2 SST files, so filter is checked once and found no match.
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                3 + iteration * 2);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+    }
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:2"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.FixedPrefix.2"));
+    // third SST with fixed:2 BF
+    ASSERT_OK(Put("foo6", "bar6"));
+    ASSERT_OK(Put("foo7", "bar7"));
+    ASSERT_OK(Put("foq8", "bar8"));
+    ASSERT_OK(Put("fpc", "2"));
+    dbfull()->Flush(FlushOptions());
+    {
+      // BF is fixed:2 now
+      std::unique_ptr<Iterator> iter_tmp(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_tmp, "foo"), 9);
+      // the first and last BF are checked
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                4 + iteration * 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 1);
+      ASSERT_EQ(CountIter(iter_tmp, "gpk"), 0);
+      // only last BF is checked and not found
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                5 + iteration * 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 2);
+    }
+
+    // iter_old can only see the first SST, so checked plus 1
+    ASSERT_EQ(CountIter(iter_old, "foo"), 4);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              6 + iteration * 3);
+    // iter was created after the first setoptions call so only full filter
+    // will check the filter
+    ASSERT_EQ(CountIter(iter, "foo"), 2);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+              6 + iteration * 4);
+
+    {
+      // keys in all three SSTs are visible to iterator
+      // The range of [foo, foz90000] is compatible with (fixed:1) and (fixed:2)
+      // so +2 for checked counter
+      std::unique_ptr<Iterator> iter_all(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_all, "foo"), 9);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                7 + iteration * 5);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 2);
+      ASSERT_EQ(CountIter(iter_all, "gpk"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                8 + iteration * 5);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+    }
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    {
+      std::unique_ptr<Iterator> iter_all(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter_all, "foo"), 6);
+      // all three SST are checked because the current options has the same as
+      // the remaining SST (capped:3)
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                9 + iteration * 7);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+      ASSERT_EQ(CountIter(iter_all, "gpk"), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED),
+                10 + iteration * 7);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 4);
+    }
+    // TODO(Zhongyi): Maybe also need to add Get calls to test point look up?
+    iteration++;
+  }
+}
+
+// Create a new column family in a running DB, change prefix_extractor
+// dynamically, verify the iterator created on the new column family behaves
+// as expected
+TEST_F(DBBloomFilterTest, DynamicBloomFilterNewColumnFamily) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    CreateAndReopenWithCF({"pikachu" + std::to_string(iteration)}, options);
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+    // create a new CF and set prefix_extractor dynamically
+    options.prefix_extractor.reset(NewCappedPrefixTransform(3));
+    CreateColumnFamilies({"ramen_dojo_" + std::to_string(iteration)}, options);
+    ASSERT_EQ(0,
+              strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
+                     "rocksdb.CappedPrefix.3"));
+    ASSERT_OK(Put(2, "foo3", "bar3"));
+    ASSERT_OK(Put(2, "foo4", "bar4"));
+    ASSERT_OK(Put(2, "foo5", "bar5"));
+    ASSERT_OK(Put(2, "foq6", "bar6"));
+    ASSERT_OK(Put(2, "fpq7", "bar7"));
+    dbfull()->Flush(FlushOptions());
+    {
+      std::unique_ptr<Iterator> iter(
+          db_->NewIterator(read_options, handles_[2]));
+      ASSERT_EQ(CountIter(iter, "foo"), 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(
+        dbfull()->SetOptions(handles_[2], {{"prefix_extractor", "fixed:2"}}));
+    ASSERT_EQ(0,
+              strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
+                     "rocksdb.FixedPrefix.2"));
+    {
+      std::unique_ptr<Iterator> iter(
+          db_->NewIterator(read_options, handles_[2]));
+      ASSERT_EQ(CountIter(iter, "foo"), 4);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 0);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[2]));
+    dbfull()->DestroyColumnFamilyHandle(handles_[2]);
+    handles_[2] = nullptr;
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+    dbfull()->DestroyColumnFamilyHandle(handles_[1]);
+    handles_[1] = nullptr;
+    iteration++;
+  }
+}
+
+// Verify it's possible to change prefix_extractor at runtime and iterators
+// behaves as expected
+TEST_F(DBBloomFilterTest, DynamicBloomFilterOptions) {
+  int iteration = 0;
+  for (bool use_block_based_builder : {true, false}) {
+    Options options;
+    options.create_if_missing = true;
+    options.prefix_extractor.reset(NewFixedPrefixTransform(1));
+    options.disable_auto_compactions = true;
+    options.statistics = CreateDBStatistics();
+    // Enable prefix bloom for SST files
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.filter_policy.reset(
+        NewBloomFilterPolicy(10, use_block_based_builder));
+    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_OK(Put("foo1", "bar1"));
+    ASSERT_OK(Put("fpa", "0"));
+    dbfull()->Flush(FlushOptions());
+    ASSERT_OK(Put("foo3", "bar3"));
+    ASSERT_OK(Put("foo4", "bar4"));
+    ASSERT_OK(Put("foo5", "bar5"));
+    ASSERT_OK(Put("fpb", "1"));
+    dbfull()->Flush(FlushOptions());
+    ASSERT_OK(Put("foo6", "bar6"));
+    ASSERT_OK(Put("foo7", "bar7"));
+    ASSERT_OK(Put("foo8", "bar8"));
+    ASSERT_OK(Put("fpc", "2"));
+    dbfull()->Flush(FlushOptions());
+
+    ReadOptions read_options;
+    read_options.prefix_same_as_start = true;
+    {
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      ASSERT_EQ(CountIter(iter, "foo"), 12);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 3);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+    std::unique_ptr<Iterator> iter_old(db_->NewIterator(read_options));
+    ASSERT_EQ(CountIter(iter_old, "foo"), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 6);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+
+    ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
+    ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
+                        "rocksdb.CappedPrefix.3"));
+    {
+      std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
+      // "fp*" should be skipped
+      ASSERT_EQ(CountIter(iter, "foo"), 9);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 6);
+      ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    }
+
+    // iterator created before should not be affected and see all keys
+    ASSERT_EQ(CountIter(iter_old, "foo"), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 9);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 0);
+    ASSERT_EQ(CountIter(iter_old, "abc"), 0);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED), 12);
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_PREFIX_USEFUL), 3);
+    iteration++;
+  }
+}
+
 #endif  // ROCKSDB_LITE

 }  // namespace rocksdb
@@ -45,6 +45,7 @@ class DBTestCompactionFilterWithCompactParam
  }
 };

+#ifndef ROCKSDB_VALGRIND_RUN
 INSTANTIATE_TEST_CASE_P(
    DBTestCompactionFilterWithCompactOption,
    DBTestCompactionFilterWithCompactParam,
@@ -53,6 +54,12 @@ INSTANTIATE_TEST_CASE_P(
                      DBTestBase::OptionConfig::kUniversalCompactionMultiLevel,
                      DBTestBase::OptionConfig::kLevelSubcompactions,
                      DBTestBase::OptionConfig::kUniversalSubcompactions));
+#else
+// Run fewer cases in valgrind
+INSTANTIATE_TEST_CASE_P(DBTestCompactionFilterWithCompactOption,
+                        DBTestCompactionFilterWithCompactParam,
+                        ::testing::Values(DBTestBase::OptionConfig::kDefault));
+#endif  // ROCKSDB_VALGRIND_RUN

 class KeepFilter : public CompactionFilter {
 public:
@@ -116,6 +116,21 @@ private:
  std::vector<std::atomic<int>> compaction_completed_;
 };

+class SstStatsCollector : public EventListener {
+ public:
+  SstStatsCollector() : num_ssts_creation_started_(0) {}
+
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& /* info */) override {
+    ++num_ssts_creation_started_;
+  }
+
+  int num_ssts_creation_started() { return num_ssts_creation_started_; }
+
+ private:
+  std::atomic<int> num_ssts_creation_started_;
+};
+
 static const int kCDTValueSize = 1000;
 static const int kCDTKeysPerBuffer = 4;
 static const int kCDTNumLevels = 8;
@@ -251,6 +266,7 @@ const SstFileMetaData* PickFileRandomly(
 }
 }  // anonymous namespace

+#ifndef ROCKSDB_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) {
@@ -293,6 +309,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
    ASSERT_GT(db_size[0] / 3, db_size[1]);
  }
 }
+#endif  // ROCKSDB_VALGRIND_RUN

 TEST_P(DBCompactionTestWithParam, CompactionsPreserveDeletes) {
  //  For each options type we test following
@@ -2460,6 +2477,7 @@ TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {

    // Compaction range overlaps files
    Compact(1, "p1", "p9", 1);
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
    ASSERT_EQ("0,1", FilesPerLevel(1));
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
@@ -2475,6 +2493,7 @@ TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {

    // Compact just the new range
    Compact(1, "b", "f", 1);
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
    ASSERT_EQ("0,2", FilesPerLevel(1));
    ASSERT_EQ(2, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
@@ -2491,6 +2510,7 @@ TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {
    compact_options.target_path_id = 1;
    compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
    db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());

    ASSERT_EQ("0,1", FilesPerLevel(1));
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
@@ -3374,14 +3394,13 @@ TEST_F(DBCompactionTest, LevelCompactExpiredTtlFiles) {
    }
    Flush();
  }
-  Flush();
  dbfull()->TEST_WaitForCompact();
  MoveFilesToLevel(3);
  ASSERT_EQ("0,0,0,2", FilesPerLevel());

+  // Delete previously written keys.
  for (int i = 0; i < kNumLevelFiles; ++i) {
    for (int j = 0; j < kNumKeysPerFile; ++j) {
-      // Overwrite previous keys with smaller, but predictable, values.
      ASSERT_OK(Delete(Key(i * kNumKeysPerFile + j)));
    }
    Flush();
@@ -3394,7 +3413,7 @@ TEST_F(DBCompactionTest, LevelCompactExpiredTtlFiles) {
  env_->addon_time_.fetch_add(36 * 60 * 60);  // 36 hours
  ASSERT_EQ("0,2,0,2", FilesPerLevel());

-  // Just do a siimple write + flush so that the Ttl expired files get
+  // Just do a simple write + flush so that the Ttl expired files get
  // compacted.
  ASSERT_OK(Put("a", "1"));
  Flush();
@@ -3408,6 +3427,57 @@ TEST_F(DBCompactionTest, LevelCompactExpiredTtlFiles) {
  // All non-L0 files are deleted, as they contained only deleted data.
  ASSERT_EQ("1", FilesPerLevel());
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  // Test dynamically changing ttl.
+
+  env_->addon_time_.store(0);
+  DestroyAndReopen(options);
+
+  for (int i = 0; i < kNumLevelFiles; ++i) {
+    for (int j = 0; j < kNumKeysPerFile; ++j) {
+      ASSERT_OK(
+          Put(Key(i * kNumKeysPerFile + j), RandomString(&rnd, kValueSize)));
+    }
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+  MoveFilesToLevel(3);
+  ASSERT_EQ("0,0,0,2", FilesPerLevel());
+
+  // Delete previously written keys.
+  for (int i = 0; i < kNumLevelFiles; ++i) {
+    for (int j = 0; j < kNumKeysPerFile; ++j) {
+      ASSERT_OK(Delete(Key(i * kNumKeysPerFile + j)));
+    }
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ("2,0,0,2", FilesPerLevel());
+  MoveFilesToLevel(1);
+  ASSERT_EQ("0,2,0,2", FilesPerLevel());
+
+  // Move time forward by 12 hours, and make sure that compaction still doesn't
+  // trigger as ttl is set to 24 hours.
+  env_->addon_time_.fetch_add(12 * 60 * 60);
+  ASSERT_OK(Put("a", "1"));
+  Flush();
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ("1,2,0,2", FilesPerLevel());
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
+        Compaction* compaction = reinterpret_cast<Compaction*>(arg);
+        ASSERT_TRUE(compaction->compaction_reason() == CompactionReason::kTtl);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Dynamically change ttl to 10 hours.
+  // This should trigger a ttl compaction, as 12 hours have already passed.
+  ASSERT_OK(dbfull()->SetOptions({{"ttl", "36000"}}));
+  dbfull()->TEST_WaitForCompact();
+  // All non-L0 files are deleted, as they contained only deleted data.
+  ASSERT_EQ("1", FilesPerLevel());
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

 TEST_F(DBCompactionTest, CompactRangeDelayedByL0FileCount) {
@@ -3433,12 +3503,13 @@ TEST_F(DBCompactionTest, CompactRangeDelayedByL0FileCount) {
      // ensure the auto compaction doesn't finish until manual compaction has
      // had a chance to be delayed.
      rocksdb::SyncPoint::GetInstance()->LoadDependency(
-          {{"DBImpl::CompactRange:StallWait", "CompactionJob::Run():End"}});
+          {{"DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait",
+            "CompactionJob::Run():End"}});
    } else {
      // ensure the auto-compaction doesn't finish until manual compaction has
      // continued without delay.
      rocksdb::SyncPoint::GetInstance()->LoadDependency(
-          {{"DBImpl::CompactRange:StallWaitDone", "CompactionJob::Run():End"}});
+          {{"DBImpl::FlushMemTable:StallWaitDone", "CompactionJob::Run():End"}});
    }
    rocksdb::SyncPoint::GetInstance()->EnableProcessing();

@@ -3486,12 +3557,13 @@ TEST_F(DBCompactionTest, CompactRangeDelayedByImmMemTableCount) {
      // ensure the flush doesn't finish until manual compaction has had a
      // chance to be delayed.
      rocksdb::SyncPoint::GetInstance()->LoadDependency(
-          {{"DBImpl::CompactRange:StallWait", "FlushJob::WriteLevel0Table"}});
+          {{"DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait",
+            "FlushJob::WriteLevel0Table"}});
    } else {
      // ensure the flush doesn't finish until manual compaction has continued
      // without delay.
      rocksdb::SyncPoint::GetInstance()->LoadDependency(
-          {{"DBImpl::CompactRange:StallWaitDone",
+          {{"DBImpl::FlushMemTable:StallWaitDone",
            "FlushJob::WriteLevel0Table"}});
    }
    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
@@ -3501,6 +3573,7 @@ TEST_F(DBCompactionTest, CompactRangeDelayedByImmMemTableCount) {
      ASSERT_OK(Put(Key(0), RandomString(&rnd, 1024)));
      FlushOptions flush_opts;
      flush_opts.wait = false;
+      flush_opts.allow_write_stall = true;
      dbfull()->Flush(flush_opts);
    }

@@ -3536,7 +3609,7 @@ TEST_F(DBCompactionTest, CompactRangeShutdownWhileDelayed) {
    // The auto-compaction waits until the manual compaction finishes to ensure
    // the signal comes from closing CF/DB, not from compaction making progress.
    rocksdb::SyncPoint::GetInstance()->LoadDependency(
-        {{"DBImpl::CompactRange:StallWait",
+        {{"DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait",
          "DBCompactionTest::CompactRangeShutdownWhileDelayed:PreShutdown"},
         {"DBCompactionTest::CompactRangeShutdownWhileDelayed:PostManual",
          "CompactionJob::Run():End"}});
@@ -3587,18 +3660,21 @@ TEST_F(DBCompactionTest, CompactRangeSkipFlushAfterDelay) {
  // began. So it unblocks CompactRange and precludes its flush. Throughout the
  // test, stall conditions are upheld via high L0 file count.
  rocksdb::SyncPoint::GetInstance()->LoadDependency(
-      {{"DBImpl::CompactRange:StallWait",
+      {{"DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait",
        "DBCompactionTest::CompactRangeSkipFlushAfterDelay:PreFlush"},
       {"DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush",
-        "DBImpl::CompactRange:StallWaitDone"},
-       {"DBImpl::CompactRange:StallWaitDone", "CompactionJob::Run():End"}});
+        "DBImpl::FlushMemTable:StallWaitDone"},
+       {"DBImpl::FlushMemTable:StallWaitDone", "CompactionJob::Run():End"}});
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

+  //used for the delayable flushes
+  FlushOptions flush_opts;
+  flush_opts.allow_write_stall = true;
  for (int i = 0; i < kNumL0FilesLimit - 1; ++i) {
    for (int j = 0; j < 2; ++j) {
      ASSERT_OK(Put(Key(j), RandomString(&rnd, 1024)));
    }
-    Flush();
+    dbfull()->Flush(flush_opts);
  }
  auto manual_compaction_thread = port::Thread([this]() {
    CompactRangeOptions cro;
@@ -3608,7 +3684,7 @@ TEST_F(DBCompactionTest, CompactRangeSkipFlushAfterDelay) {

  TEST_SYNC_POINT("DBCompactionTest::CompactRangeSkipFlushAfterDelay:PreFlush");
  Put(ToString(0), RandomString(&rnd, 1024));
-  Flush();
+  dbfull()->Flush(flush_opts);
  Put(ToString(0), RandomString(&rnd, 1024));
  TEST_SYNC_POINT("DBCompactionTest::CompactRangeSkipFlushAfterDelay:PostFlush");
  manual_compaction_thread.join();
@@ -3702,6 +3778,92 @@ TEST_F(DBCompactionTest, CompactionStatsTest) {
  VerifyCompactionStats(*cfd, *collector);
 }

+TEST_F(DBCompactionTest, CompactFilesOutputRangeConflict) {
+  // LSM setup:
+  // L1:      [ba bz]
+  // L2: [a b]       [c d]
+  // L3: [a b]       [c d]
+  //
+  // Thread 1:                        Thread 2:
+  // Begin compacting all L2->L3
+  //                                  Compact [ba bz] L1->L3
+  // End compacting all L2->L3
+  //
+  // The compaction operation in thread 2 should be disallowed because the range
+  // overlaps with the compaction in thread 1, which also covers that range in
+  // L3.
+  Options options = CurrentOptions();
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+  Reopen(options);
+
+  for (int level = 3; level >= 2; --level) {
+    ASSERT_OK(Put("a", "val"));
+    ASSERT_OK(Put("b", "val"));
+    ASSERT_OK(Flush());
+    ASSERT_OK(Put("c", "val"));
+    ASSERT_OK(Put("d", "val"));
+    ASSERT_OK(Flush());
+    MoveFilesToLevel(level);
+  }
+  ASSERT_OK(Put("ba", "val"));
+  ASSERT_OK(Put("bz", "val"));
+  ASSERT_OK(Flush());
+  MoveFilesToLevel(1);
+
+  SyncPoint::GetInstance()->LoadDependency({
+      {"CompactFilesImpl:0",
+       "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2Begin"},
+      {"DBCompactionTest::CompactFilesOutputRangeConflict:Thread2End",
+       "CompactFilesImpl:1"},
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  auto bg_thread = port::Thread([&]() {
+    // Thread 1
+    std::vector<std::string> filenames = collector->GetFlushedFiles();
+    filenames.pop_back();
+    ASSERT_OK(db_->CompactFiles(CompactionOptions(), filenames,
+                                3 /* output_level */));
+  });
+
+  // Thread 2
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2Begin");
+  std::string filename = collector->GetFlushedFiles().back();
+  ASSERT_FALSE(
+      db_->CompactFiles(CompactionOptions(), {filename}, 3 /* output_level */)
+          .ok());
+  TEST_SYNC_POINT(
+      "DBCompactionTest::CompactFilesOutputRangeConflict:Thread2End");
+
+  bg_thread.join();
+}
+
+TEST_F(DBCompactionTest, CompactionHasEmptyOutput) {
+  Options options = CurrentOptions();
+  SstStatsCollector* collector = new SstStatsCollector();
+  options.level0_file_num_compaction_trigger = 2;
+  options.listeners.emplace_back(collector);
+  Reopen(options);
+
+  // Make sure the L0 files overlap to prevent trivial move.
+  ASSERT_OK(Put("a", "val"));
+  ASSERT_OK(Put("b", "val"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Delete("a"));
+  ASSERT_OK(Delete("b"));
+  ASSERT_OK(Flush());
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(NumTableFilesAtLevel(0), 0);
+  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
+
+  // Expect one file creation to start for each flush, and zero for compaction
+  // since no keys are written.
+  ASSERT_EQ(2, collector->num_ssts_creation_started());
+}
+
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                        ::testing::Values(std::make_tuple(1, true),
                                          std::make_tuple(1, false),
@@ -3799,6 +3961,50 @@ INSTANTIATE_TEST_CASE_P(
                      CompactionPri::kOldestSmallestSeqFirst,
                      CompactionPri::kMinOverlappingRatio));

+class NoopMergeOperator : public MergeOperator {
+ public:
+  NoopMergeOperator() {}
+
+  virtual bool FullMergeV2(const MergeOperationInput& /*merge_in*/,
+                           MergeOperationOutput* merge_out) const override {
+    std::string val("bar");
+    merge_out->new_value = val;
+    return true;
+  }
+
+  virtual const char* Name() const override { return "Noop"; }
+};
+
+TEST_F(DBCompactionTest, PartialManualCompaction) {
+  Options opts = CurrentOptions();
+  opts.num_levels = 3;
+  opts.level0_file_num_compaction_trigger = 10;
+  opts.compression = kNoCompression;
+  opts.merge_operator.reset(new NoopMergeOperator());
+  opts.target_file_size_base = 10240;
+  DestroyAndReopen(opts);
+
+  Random rnd(301);
+  for (auto i = 0; i < 8; ++i) {
+    for (auto j = 0; j < 10; ++j) {
+      Merge("foo", RandomString(&rnd, 1024));
+    }
+    Flush();
+  }
+
+  MoveFilesToLevel(2);
+
+  std::string prop;
+  EXPECT_TRUE(dbfull()->GetProperty(DB::Properties::kLiveSstFilesSize, &prop));
+  uint64_t max_compaction_bytes = atoi(prop.c_str()) / 2;
+  ASSERT_OK(dbfull()->SetOptions(
+      {{"max_compaction_bytes", std::to_string(max_compaction_bytes)}}));
+
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  dbfull()->CompactRange(cro, nullptr, nullptr);
+}
+
 #endif // !defined(ROCKSDB_LITE)
 }  // namespace rocksdb

@@ -35,6 +35,7 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
  Reopen(options);
  FlushOptions no_wait;
  no_wait.wait = false;
+  no_wait.allow_write_stall=true;

  SyncPoint::GetInstance()->LoadDependency(
      {{"VersionSet::LogAndApply:WriteManifest",
@@ -55,6 +56,9 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
 #endif  // ROCKSDB_LITE
 }

+#ifndef TRAVIS
+// Disable this test temporarily on Travis as it fails intermittently.
+// Github issue: #4151
 TEST_F(DBFlushTest, SyncFail) {
  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
      new FaultInjectionTestEnv(env_));
@@ -92,6 +96,7 @@ TEST_F(DBFlushTest, SyncFail) {
  ASSERT_EQ(refs_before, cfd->current()->TEST_refs());
  Destroy(options);
 }
+#endif  // TRAVIS

 TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
  // Verify setting an empty high-pri (flush) thread pool causes flushes to be
@@ -32,6 +32,7 @@
 #include "db/db_info_dumper.h"
 #include "db/db_iter.h"
 #include "db/dbformat.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "db/external_sst_file_ingestion_job.h"
 #include "db/flush_job.h"
@@ -40,7 +41,6 @@
 #include "db/log_reader.h"
 #include "db/log_writer.h"
 #include "db/malloc_stats.h"
-#include "db/managed_iterator.h"
 #include "db/memtable.h"
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
@@ -138,7 +138,7 @@ int64_t kDefaultLowPriThrottledRate = 2 * 1024 * 1024;
 }  // namespace

 DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
-               const bool seq_per_batch)
+               const bool seq_per_batch, const bool batch_per_txn)
    : env_(options.env),
      dbname_(dbname),
      own_info_log_(options.info_log == nullptr),
@@ -199,6 +199,7 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
      two_write_queues_(options.two_write_queues),
      manual_wal_flush_(options.manual_wal_flush),
      seq_per_batch_(seq_per_batch),
+      batch_per_txn_(batch_per_txn),
      // last_sequencee_ is always maintained by the main queue that also writes
      // to the memtable. When two_write_queues_ is disabled last seq in
      // memtable is the same as last seq published to the readers. When it is
@@ -214,8 +215,14 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
      // requires a custom gc for compaction, we use that to set use_custom_gc_
      // as well.
      use_custom_gc_(seq_per_batch),
+      shutdown_initiated_(false),
+      own_sfm_(options.sst_file_manager == nullptr),
      preserve_deletes_(options.preserve_deletes),
-      closed_(false) {
+      closed_(false),
+      error_handler_(this, immutable_db_options_, &mutex_) {
+  // !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_);
  env_->GetAbsolutePath(dbname, &db_absolute_path_);

  // Reserve ten files or so for other uses and give the rest to TableCache.
@@ -244,6 +251,112 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
  preserve_deletes_seqnum_.store(0);
 }

+Status DBImpl::Resume() {
+  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Resuming DB");
+
+  InstrumentedMutexLock db_mutex(&mutex_);
+
+  if (!error_handler_.IsDBStopped() && !error_handler_.IsBGWorkStopped()) {
+    // Nothing to do
+    return Status::OK();
+  }
+
+  if (error_handler_.IsRecoveryInProgress()) {
+    // Don't allow a mix of manual and automatic recovery
+    return Status::Busy();
+  }
+
+  mutex_.Unlock();
+  Status s = error_handler_.RecoverFromBGError(true);
+  mutex_.Lock();
+  return s;
+}
+
+// This function implements the guts of recovery from a background error. It
+// is eventually called for both manual as well as automatic recovery. It does
+// the following -
+// 1. Wait for currently scheduled background flush/compaction to exit, in
+//    order to inadvertently causing an error and thinking recovery failed
+// 2. Flush memtables if there's any data for all the CFs. This may result
+//    another error, which will be saved by error_handler_ and reported later
+//    as the recovery status
+// 3. Find and delete any obsolete files
+// 4. Schedule compactions if needed for all the CFs. This is needed as the
+//    flush in the prior step might have been a no-op for some CFs, which
+//    means a new super version wouldn't have been installed
+Status DBImpl::ResumeImpl() {
+  mutex_.AssertHeld();
+  WaitForBackgroundWork();
+
+  Status bg_error = error_handler_.GetBGError();
+  Status s;
+  if (shutdown_initiated_) {
+    // Returning shutdown status to SFM during auto recovery will cause it
+    // to abort the recovery and allow the shutdown to progress
+    s = Status::ShutdownInProgress();
+  }
+  if (s.ok() && bg_error.severity() > Status::Severity::kHardError) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log,
+        "DB resume requested but failed due to Fatal/Unrecoverable error");
+    s = bg_error;
+  }
+
+  // We cannot guarantee consistency of the WAL. So force flush Memtables of
+  // all the column families
+  if (s.ok()) {
+    s = FlushAllCFs(FlushReason::kErrorRecovery);
+    if (!s.ok()) {
+      ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                     "DB resume requested but failed due to Flush failure [%s]",
+                     s.ToString().c_str());
+    }
+  }
+
+  JobContext job_context(0);
+  FindObsoleteFiles(&job_context, true);
+  if (s.ok()) {
+    s = error_handler_.ClearBGError();
+  }
+  mutex_.Unlock();
+
+  job_context.manifest_file_number = 1;
+  if (job_context.HaveSomethingToDelete()) {
+    PurgeObsoleteFiles(job_context);
+  }
+  job_context.Clean();
+
+  if (s.ok()) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log, "Successfully resumed DB");
+  }
+  mutex_.Lock();
+  // Check for shutdown again before scheduling further compactions,
+  // since we released and re-acquired the lock above
+  if (shutdown_initiated_) {
+    s = Status::ShutdownInProgress();
+  }
+  if (s.ok()) {
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      SchedulePendingCompaction(cfd);
+    }
+    MaybeScheduleFlushOrCompaction();
+  }
+
+  // Wake up any waiters - in this case, it could be the shutdown thread
+  bg_cv_.SignalAll();
+
+  // No need to check BGError again. If something happened, event listener would be
+  // notified and the operation causing it would have failed
+  return s;
+}
+
+void DBImpl::WaitForBackgroundWork() {
+  // Wait for background work to finish
+  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
+         bg_flush_scheduled_) {
+    bg_cv_.Wait();
+  }
+}
+
 // Will lock the mutex_,  will wait for completion if wait is true
 void DBImpl::CancelAllBackgroundWork(bool wait) {
  InstrumentedMutexLock l(&mutex_);
@@ -271,14 +384,20 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
  if (!wait) {
    return;
  }
-  // Wait for background work to finish
-  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
-         bg_flush_scheduled_) {
-    bg_cv_.Wait();
-  }
+  WaitForBackgroundWork();
 }

 Status DBImpl::CloseHelper() {
+  // Guarantee that there is no background error recovery in progress before
+  // continuing with the shutdown
+  mutex_.Lock();
+  shutdown_initiated_ = true;
+  error_handler_.CancelErrorRecovery();
+  while (error_handler_.IsRecoveryInProgress()) {
+    bg_cv_.Wait();
+  }
+  mutex_.Unlock();
+
  // CancelAllBackgroundWork called with false means we just set the shutdown
  // marker. After this we do a variant of the waiting and unschedule work
  // (to consider: moving all the waiting into CancelAllBackgroundWork(true))
@@ -296,7 +415,8 @@ Status DBImpl::CloseHelper() {
  // Wait for background work to finish
  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
         bg_flush_scheduled_ || bg_purge_scheduled_ ||
-         pending_purge_obsolete_files_) {
+         pending_purge_obsolete_files_ ||
+         error_handler_.IsRecoveryInProgress()) {
    TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
    bg_cv_.Wait();
  }
@@ -306,9 +426,12 @@ Status DBImpl::CloseHelper() {
  flush_scheduler_.Clear();

  while (!flush_queue_.empty()) {
-    auto cfd = PopFirstFromFlushQueue();
-    if (cfd->Unref()) {
-      delete cfd;
+    const FlushRequest& flush_req = PopFirstFromFlushQueue();
+    for (const auto& iter : flush_req) {
+      ColumnFamilyData* cfd = iter.first;
+      if (cfd->Unref()) {
+        delete cfd;
+      }
    }
  }
  while (!compaction_queue_.empty()) {
@@ -398,6 +521,17 @@ Status DBImpl::CloseHelper() {
  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Shutdown complete");
  LogFlush(immutable_db_options_.info_log);

+#ifndef ROCKSDB_LITE
+  // If the sst_file_manager was allocated by us during DB::Open(), ccall
+  // Close() on it before closing the info_log. Otherwise, background thread
+  // in SstFileManagerImpl might try to log something
+  if (immutable_db_options_.sst_file_manager && own_sfm_) {
+    auto sfm = static_cast<SstFileManagerImpl*>(
+        immutable_db_options_.sst_file_manager.get());
+    sfm->Close();
+  }
+#endif // ROCKSDB_LITE
+
  if (immutable_db_options_.info_log && own_info_log_) {
    Status s = immutable_db_options_.info_log->Close();
    if (ret.ok()) {
@@ -1005,7 +1139,7 @@ InternalIterator* DBImpl::NewInternalIterator(
  } else {
    CleanupSuperVersion(super_version);
  }
-  return NewErrorInternalIterator(s, arena);
+  return NewErrorInternalIterator<Slice>(s, arena);
 }

 ColumnFamilyHandle* DBImpl::DefaultColumnFamily() const {
@@ -1029,6 +1163,15 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();

+  if (tracer_) {
+    // TODO: This mutex should be removed later, to improve performance when
+    // tracing is enabled.
+    InstrumentedMutexLock lock(&trace_mutex_);
+    if (tracer_) {
+      tracer_->Get(column_family, key);
+    }
+  }
+
  // Acquire SuperVersion
  SuperVersion* sv = GetAndRefSuperVersion(cfd);

@@ -1037,12 +1180,19 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,

  SequenceNumber snapshot;
  if (read_options.snapshot != nullptr) {
-    // Note: In WritePrepared txns this is not necessary but not harmful either.
-    // Because prep_seq > snapshot => commit_seq > snapshot so if a snapshot is
-    // specified we should be fine with skipping seq numbers that are greater
-    // than that.
+    // Note: In WritePrepared txns this is not necessary but not harmful
+    // either.  Because prep_seq > snapshot => commit_seq > snapshot so if
+    // a snapshot is specified we should be fine with skipping seq numbers
+    // that are greater than that.
+    //
+    // In WriteUnprepared, we cannot set snapshot in the lookup key because we
+    // may skip uncommitted data that should be visible to the transaction for
+    // reading own writes.
    snapshot =
        reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
+    if (callback) {
+      snapshot = std::max(snapshot, callback->MaxUnpreparedSequenceNumber());
+    }
  } else {
    // Since we get and reference the super version before getting
    // the snapshot number, without a mutex protection, it is possible
@@ -1050,10 +1200,10 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
    // data for this snapshot is available. But it will contain all
    // the data available in the super version we have, which is also
    // a valid snapshot to read from.
-    // We shouldn't get snapshot before finding and referencing the
-    // super versipon because a flush happening in between may compact
-    // away data for the snapshot, but the snapshot is earlier than the
-    // data overwriting it, so users may see wrong results.
+    // We shouldn't get snapshot before finding and referencing the super
+    // version because a flush happening in between may compact away data for
+    // the snapshot, but the snapshot is earlier than the data overwriting it,
+    // so users may see wrong results.
    snapshot = last_seq_same_as_publish_seq_
                   ? versions_->LastSequence()
                   : versions_->LastPublishedSequence();
@@ -1528,6 +1678,10 @@ bool DBImpl::KeyMayExist(const ReadOptions& read_options,

 Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
                              ColumnFamilyHandle* column_family) {
+  if (read_options.managed) {
+    return NewErrorIterator(
+        Status::NotSupported("Managed iterator is not supported anymore."));
+  }
  Iterator* result = nullptr;
  if (read_options.read_tier == kPersistedTier) {
    return NewErrorIterator(Status::NotSupported(
@@ -1545,22 +1699,7 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();
  ReadCallback* read_callback = nullptr;  // No read callback provided.
-  if (read_options.managed) {
-#ifdef ROCKSDB_LITE
-    // not supported in lite version
-    result = NewErrorIterator(Status::InvalidArgument(
-        "Managed Iterators not supported in RocksDBLite."));
-#else
-    if ((read_options.tailing) || (read_options.snapshot != nullptr) ||
-        (is_snapshot_supported_)) {
-      result = new ManagedIterator(this, read_options, cfd);
-    } else {
-      // Managed iter not supported
-      result = NewErrorIterator(Status::InvalidArgument(
-          "Managed Iterators not supported without snapshots."));
-    }
-#endif
-  } else if (read_options.tailing) {
+if (read_options.tailing) {
 #ifdef ROCKSDB_LITE
    // not supported in lite version
    result = nullptr;
@@ -1571,8 +1710,8 @@ Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
    result = NewDBIterator(
        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
        cfd->user_comparator(), iter, kMaxSequenceNumber,
-        sv->mutable_cf_options.max_sequential_skip_in_iterations,
-        read_callback);
+        sv->mutable_cf_options.max_sequential_skip_in_iterations, read_callback,
+        this, cfd);
 #endif
  } else {
    // Note: no need to consider the special case of
@@ -1639,9 +1778,8 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
  ArenaWrappedDBIter* db_iter = NewArenaWrappedDbIterator(
      env_, read_options, *cfd->ioptions(), sv->mutable_cf_options, snapshot,
      sv->mutable_cf_options.max_sequential_skip_in_iterations,
-      sv->version_number, read_callback,
-      ((read_options.snapshot != nullptr) ? nullptr : this), cfd, allow_blob,
-      allow_refresh);
+      sv->version_number, read_callback, this, cfd, allow_blob,
+      ((read_options.snapshot != nullptr) ? false : allow_refresh));

  InternalIterator* internal_iter =
      NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
@@ -1655,6 +1793,9 @@ Status DBImpl::NewIterators(
    const ReadOptions& read_options,
    const std::vector<ColumnFamilyHandle*>& column_families,
    std::vector<Iterator*>* iterators) {
+  if (read_options.managed) {
+    return Status::NotSupported("Managed iterator is not supported anymore.");
+  }
  if (read_options.read_tier == kPersistedTier) {
    return Status::NotSupported(
        "ReadTier::kPersistedData is not yet supported in iterators.");
@@ -1662,23 +1803,7 @@ Status DBImpl::NewIterators(
  ReadCallback* read_callback = nullptr;  // No read callback provided.
  iterators->clear();
  iterators->reserve(column_families.size());
-  if (read_options.managed) {
-#ifdef ROCKSDB_LITE
-    return Status::InvalidArgument(
-        "Managed interator not supported in RocksDB lite");
-#else
-    if ((!read_options.tailing) && (read_options.snapshot == nullptr) &&
-        (!is_snapshot_supported_)) {
-      return Status::InvalidArgument(
-          "Managed interator not supported without snapshots");
-    }
-    for (auto cfh : column_families) {
-      auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
-      auto iter = new ManagedIterator(this, read_options, cfd);
-      iterators->push_back(iter);
-    }
-#endif
-  } else if (read_options.tailing) {
+  if (read_options.tailing) {
 #ifdef ROCKSDB_LITE
    return Status::InvalidArgument(
        "Tailing interator not supported in RocksDB lite");
@@ -1691,7 +1816,7 @@ Status DBImpl::NewIterators(
          env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
          cfd->user_comparator(), iter, kMaxSequenceNumber,
          sv->mutable_cf_options.max_sequential_skip_in_iterations,
-          read_callback));
+          read_callback, this, cfd));
    }
 #endif
  } else {
@@ -1845,6 +1970,13 @@ bool DBImpl::GetProperty(ColumnFamilyHandle* column_family,
    InstrumentedMutexLock l(&mutex_);
    return cfd->internal_stats()->GetStringProperty(*property_info, property,
                                                    value);
+  } else if (property_info->handle_string_dbimpl) {
+    std::string tmp_value;
+    bool ret_value = (this->*(property_info->handle_string_dbimpl))(&tmp_value);
+    if (ret_value) {
+      *value = tmp_value;
+    }
+    return ret_value;
  }
  // Shouldn't reach here since exactly one of handle_string and handle_int
  // should be non-nullptr.
@@ -1911,6 +2043,16 @@ bool DBImpl::GetIntPropertyInternal(ColumnFamilyData* cfd,
  }
 }

+bool DBImpl::GetPropertyHandleOptionsStatistics(std::string* value) {
+  assert(value != nullptr);
+  Statistics* statistics = immutable_db_options_.statistics.get();
+  if (!statistics) {
+    return false;
+  }
+  *value = statistics->ToString();
+  return true;
+}
+
 #ifndef ROCKSDB_LITE
 Status DBImpl::ResetStats() {
  InstrumentedMutexLock l(&mutex_);
@@ -2185,9 +2327,9 @@ Status DBImpl::DeleteFile(std::string name) {
    status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                    &edit, &mutex_, directories_.GetDbDir());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(cfd, &job_context.superversion_context,
-                                         *cfd->GetLatestMutableCFOptions(),
-                                         FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(
+          cfd, &job_context.superversion_contexts[0],
+          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
    }
    FindObsoleteFiles(&job_context, false);
  }  // lock released here
@@ -2270,9 +2412,9 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
    status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                    &edit, &mutex_, directories_.GetDbDir());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(cfd, &job_context.superversion_context,
-                                         *cfd->GetLatestMutableCFOptions(),
-                                         FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(
+          cfd, &job_context.superversion_contexts[0],
+          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
    }
    for (auto* deleted_file : deleted_files) {
      deleted_file->being_compacted = false;
@@ -2351,7 +2493,8 @@ Status DBImpl::GetDbIdentity(std::string& identity) const {
    if (!s.ok()) {
      return s;
    }
-    id_file_reader.reset(new SequentialFileReader(std::move(idfile)));
+    id_file_reader.reset(
+        new SequentialFileReader(std::move(idfile), idfilename));
  }

  uint64_t file_size;
@@ -2359,7 +2502,7 @@ Status DBImpl::GetDbIdentity(std::string& identity) const {
  if (!s.ok()) {
    return s;
  }
-  char* buffer = reinterpret_cast<char*>(alloca(file_size));
+  char* buffer = reinterpret_cast<char*>(alloca(static_cast<size_t>(file_size)));
  Slice id;
  s = id_file_reader->Read(static_cast<size_t>(file_size), &id, buffer);
  if (!s.ok()) {
@@ -2667,7 +2810,9 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
  // Retry if the file name happen to conflict with an existing one.
  s = GetEnv()->RenameFile(file_name, options_file_name);

-  DeleteObsoleteOptionsFiles();
+  if (0 == disable_delete_obsolete_files_) {
+    DeleteObsoleteOptionsFiles();
+  }
  return s;
 #else
  (void)file_name;
@@ -2834,6 +2979,10 @@ Status DBImpl::IngestExternalFile(
    ColumnFamilyHandle* column_family,
    const std::vector<std::string>& external_files,
    const IngestExternalFileOptions& ingestion_options) {
+  if (external_files.empty()) {
+    return Status::InvalidArgument("external_files is empty");
+  }
+
  Status status;
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();
@@ -2851,22 +3000,46 @@ Status DBImpl::IngestExternalFile(
                                            immutable_db_options_, env_options_,
                                            &snapshots_, ingestion_options);

+  SuperVersionContext dummy_sv_ctx(/* create_superversion */ true);
+  VersionEdit dummy_edit;
+  uint64_t next_file_number = 0;
  std::list<uint64_t>::iterator pending_output_elem;
  {
    InstrumentedMutexLock l(&mutex_);
-    if (!bg_error_.ok()) {
+    if (error_handler_.IsDBStopped()) {
      // Don't ingest files when there is a bg_error
-      return bg_error_;
+      return error_handler_.GetBGError();
    }

    // Make sure that bg cleanup wont delete the files that we are ingesting
    pending_output_elem = CaptureCurrentFileNumberInPendingOutputs();
+
+    // If crash happen after a hard link established, Recover function may
+    // reuse the file number that has already assigned to the internal file,
+    // and this will overwrite the external file. To protect the external
+    // file, we have to make sure the file number will never being reused.
+    next_file_number = versions_->FetchAddFileNumber(external_files.size());
+    auto cf_options = cfd->GetLatestMutableCFOptions();
+    status = versions_->LogAndApply(cfd, *cf_options, &dummy_edit, &mutex_,
+                                    directories_.GetDbDir());
+    if (status.ok()) {
+      InstallSuperVersionAndScheduleWork(cfd, &dummy_sv_ctx, *cf_options);
+    }
+  }
+  dummy_sv_ctx.Clean();
+  if (!status.ok()) {
+    InstrumentedMutexLock l(&mutex_);
+    ReleaseFileNumberFromPendingOutputs(pending_output_elem);
+    return status;
  }

  SuperVersion* super_version = cfd->GetReferencedSuperVersion(&mutex_);
-  status = ingestion_job.Prepare(external_files, super_version);
+  status =
+      ingestion_job.Prepare(external_files, next_file_number, super_version);
  CleanupSuperVersion(super_version);
  if (!status.ok()) {
+    InstrumentedMutexLock l(&mutex_);
+    ReleaseFileNumberFromPendingOutputs(pending_output_elem);
    return status;
  }

@@ -2960,8 +3133,6 @@ Status DBImpl::IngestExternalFile(

 Status DBImpl::VerifyChecksum() {
  Status s;
-  Options options;
-  EnvOptions env_options;
  std::vector<ColumnFamilyData*> cfd_list;
  {
    InstrumentedMutexLock l(&mutex_);
@@ -2979,13 +3150,19 @@ Status DBImpl::VerifyChecksum() {
  for (auto& sv : sv_list) {
    VersionStorageInfo* vstorage = sv->current->storage_info();
    ColumnFamilyData* cfd = sv->current->cfd();
+    Options opts;
+    {
+      InstrumentedMutexLock l(&mutex_);
+      opts = Options(BuildDBOptions(immutable_db_options_,
+         mutable_db_options_), cfd->GetLatestCFOptions());
+    }
    for (int i = 0; i < vstorage->num_non_empty_levels() && s.ok(); i++) {
      for (size_t j = 0; j < vstorage->LevelFilesBrief(i).num_files && s.ok();
           j++) {
        const auto& fd = vstorage->LevelFilesBrief(i).files[j].fd;
        std::string fname = TableFileName(cfd->ioptions()->cf_paths,
                                          fd.GetNumber(), fd.GetPathId());
-        s = rocksdb::VerifySstFileChecksum(options, env_options, fname);
+        s = rocksdb::VerifySstFileChecksum(opts, env_options_, fname);
      }
    }
    if (!s.ok()) {
@@ -3009,7 +3186,6 @@ Status DBImpl::VerifyChecksum() {

 void DBImpl::NotifyOnExternalFileIngested(
    ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job) {
-#ifndef ROCKSDB_LITE
  if (immutable_db_options_.listeners.empty()) {
    return;
  }
@@ -3025,8 +3201,6 @@ void DBImpl::NotifyOnExternalFileIngested(
      listener->OnExternalFileIngested(this, info);
    }
  }
-
-#endif
 }

 void DBImpl::WaitForIngestFile() {
@@ -3036,5 +3210,43 @@ void DBImpl::WaitForIngestFile() {
  }
 }

+Status DBImpl::StartTrace(const TraceOptions& /* options */,
+                          std::unique_ptr<TraceWriter>&& trace_writer) {
+  InstrumentedMutexLock lock(&trace_mutex_);
+  tracer_.reset(new Tracer(env_, std::move(trace_writer)));
+  return Status::OK();
+}
+
+Status DBImpl::EndTrace() {
+  InstrumentedMutexLock lock(&trace_mutex_);
+  Status s = tracer_->Close();
+  tracer_.reset();
+  return s;
+}
+
+Status DBImpl::TraceIteratorSeek(const uint32_t& cf_id, const Slice& key) {
+  Status s;
+  if (tracer_) {
+    InstrumentedMutexLock lock(&trace_mutex_);
+    if (tracer_) {
+      s = tracer_->IteratorSeek(cf_id, key);
+    }
+  }
+  return s;
+}
+
+Status DBImpl::TraceIteratorSeekForPrev(const uint32_t& cf_id,
+                                        const Slice& key) {
+  Status s;
+  if (tracer_) {
+    InstrumentedMutexLock lock(&trace_mutex_);
+    if (tracer_) {
+      s = tracer_->IteratorSeekForPrev(cf_id, key);
+    }
+  }
+  return s;
+}
+
 #endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb
@@ -22,6 +22,8 @@
 #include "db/column_family.h"
 #include "db/compaction_job.h"
 #include "db/dbformat.h"
+#include "db/error_handler.h"
+#include "db/event_helpers.h"
 #include "db/external_sst_file_ingestion_job.h"
 #include "db/flush_job.h"
 #include "db/flush_scheduler.h"
@@ -44,6 +46,7 @@
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/status.h"
+#include "rocksdb/trace_reader_writer.h"
 #include "rocksdb/transaction_log.h"
 #include "rocksdb/write_buffer_manager.h"
 #include "table/scoped_arena_iterator.h"
@@ -52,6 +55,7 @@
 #include "util/hash.h"
 #include "util/stop_watch.h"
 #include "util/thread_local.h"
+#include "util/trace_replay.h"

 namespace rocksdb {

@@ -70,9 +74,12 @@ struct MemTableInfo;
 class DBImpl : public DB {
 public:
  DBImpl(const DBOptions& options, const std::string& dbname,
-         const bool seq_per_batch = false);
+         const bool seq_per_batch = false, const bool batch_per_txn = true);
  virtual ~DBImpl();

+  using DB::Resume;
+  virtual Status Resume() override;
+
  // Implementations of the DB interface
  using DB::Put;
  virtual Status Put(const WriteOptions& options,
@@ -328,6 +335,15 @@ class DBImpl : public DB {

  virtual Status VerifyChecksum() override;

+  using DB::StartTrace;
+  virtual Status StartTrace(
+      const TraceOptions& options,
+      std::unique_ptr<TraceWriter>&& trace_writer) override;
+
+  using DB::EndTrace;
+  virtual Status EndTrace() override;
+  Status TraceIteratorSeek(const uint32_t& cf_id, const Slice& key);
+  Status TraceIteratorSeekForPrev(const uint32_t& cf_id, const Slice& key);
 #endif  // ROCKSDB_LITE

  // Similar to GetSnapshot(), but also lets the db know that this snapshot
@@ -380,7 +396,7 @@ class DBImpl : public DB {
  Status TEST_SwitchMemtable(ColumnFamilyData* cfd = nullptr);

  // Force current memtable contents to be flushed.
-  Status TEST_FlushMemTable(bool wait = true,
+  Status TEST_FlushMemTable(bool wait = true, bool allow_write_stall = false,
                            ColumnFamilyHandle* cfh = nullptr);

  // Wait for memtable compaction
@@ -448,6 +464,7 @@ class DBImpl : public DB {

  int TEST_BGCompactionsAllowed() const;
  int TEST_BGFlushesAllowed() const;
+  size_t TEST_GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;

 #endif  // NDEBUG

@@ -556,25 +573,50 @@ class DBImpl : public DB {
  // these will then be passed to TransactionDB so that
  // locks can be reacquired before writing can resume.
  struct RecoveredTransaction {
-    uint64_t log_number_;
    std::string name_;
-    WriteBatch* batch_;
-    // The seq number of the first key in the batch
-    SequenceNumber seq_;
-    // Number of sub-batched. A new sub-batch is created if we txn attempts to
-    // inserts a duplicate key,seq to memtable. This is currently used in
-    // WritePrparedTxn
-    size_t batch_cnt_;
+    bool unprepared_;
+
+    struct BatchInfo {
+      uint64_t log_number_;
+      // TODO(lth): For unprepared, the memory usage here can be big for
+      // unprepared transactions. This is only useful for rollbacks, and we
+      // can in theory just keep keyset for that.
+      WriteBatch* batch_;
+      // Number of sub-batches. A new sub-batch is created if txn attempts to
+      // insert a duplicate key,seq to memtable. This is currently used in
+      // WritePreparedTxn/WriteUnpreparedTxn.
+      size_t batch_cnt_;
+    };
+
+    // This maps the seq of the first key in the batch to BatchInfo, which
+    // contains WriteBatch and other information relevant to the batch.
+    //
+    // For WriteUnprepared, batches_ can have size greater than 1, but for
+    // other write policies, it must be of size 1.
+    std::map<SequenceNumber, BatchInfo> batches_;
+
    explicit RecoveredTransaction(const uint64_t log, const std::string& name,
                                  WriteBatch* batch, SequenceNumber seq,
-                                  size_t batch_cnt)
-        : log_number_(log),
-          name_(name),
-          batch_(batch),
-          seq_(seq),
-          batch_cnt_(batch_cnt) {}
+                                  size_t batch_cnt, bool unprepared)
+        : name_(name), unprepared_(unprepared) {
+      batches_[seq] = {log, batch, batch_cnt};
+    }

-    ~RecoveredTransaction() { delete batch_; }
+    ~RecoveredTransaction() {
+      for (auto& it : batches_) {
+        delete it.second.batch_;
+      }
+    }
+
+    void AddBatch(SequenceNumber seq, uint64_t log_number, WriteBatch* batch,
+                  size_t batch_cnt, bool unprepared) {
+      assert(batches_.count(seq) == 0);
+      batches_[seq] = {log_number, batch, batch_cnt};
+      // Prior state must be unprepared, since the prepare batch must be the
+      // last batch.
+      assert(unprepared_);
+      unprepared_ = unprepared;
+    }
  };

  bool allow_2pc() const { return immutable_db_options_.allow_2pc; }
@@ -595,9 +637,19 @@ class DBImpl : public DB {

  void InsertRecoveredTransaction(const uint64_t log, const std::string& name,
                                  WriteBatch* batch, SequenceNumber seq,
-                                  size_t batch_cnt) {
-    recovered_transactions_[name] =
-        new RecoveredTransaction(log, name, batch, seq, batch_cnt);
+                                  size_t batch_cnt, bool unprepared_batch) {
+    // For WriteUnpreparedTxn, InsertRecoveredTransaction is called multiple
+    // times for every unprepared batch encountered during recovery.
+    //
+    // If the transaction is prepared, then the last call to
+    // InsertRecoveredTransaction will have unprepared_batch = false.
+    auto rtxn = recovered_transactions_.find(name);
+    if (rtxn == recovered_transactions_.end()) {
+      recovered_transactions_[name] = new RecoveredTransaction(
+          log, name, batch, seq, batch_cnt, unprepared_batch);
+    } else {
+      rtxn->second->AddBatch(seq, log, batch, batch_cnt, unprepared_batch);
+    }
    logs_with_prep_tracker_.MarkLogAsContainingPrepSection(log);
  }

@@ -606,7 +658,10 @@ class DBImpl : public DB {
    assert(it != recovered_transactions_.end());
    auto* trx = it->second;
    recovered_transactions_.erase(it);
-    logs_with_prep_tracker_.MarkLogAsHavingPrepSectionFlushed(trx->log_number_);
+    for (const auto& info : trx->batches_) {
+      logs_with_prep_tracker_.MarkLogAsHavingPrepSectionFlushed(
+          info.second.log_number_);
+    }
    delete trx;
  }

@@ -635,7 +690,7 @@ class DBImpl : public DB {
  static Status Open(const DBOptions& db_options, const std::string& name,
                     const std::vector<ColumnFamilyDescriptor>& column_families,
                     std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
-                     const bool seq_per_batch);
+                     const bool seq_per_batch, const bool batch_per_txn);

  virtual Status Close() override;

@@ -654,6 +709,8 @@ class DBImpl : public DB {
  Statistics* stats_;
  std::unordered_map<std::string, RecoveredTransaction*>
      recovered_transactions_;
+  std::unique_ptr<Tracer> tracer_;
+  InstrumentedMutex trace_mutex_;

  // Except in DB::Open(), WriteOptionsFile can only be called when:
  // Persist options to options file.
@@ -739,6 +796,7 @@ class DBImpl : public DB {

 private:
  friend class DB;
+  friend class ErrorHandler;
  friend class InternalStats;
  friend class PessimisticTransaction;
  friend class TransactionBaseImpl;
@@ -746,17 +804,22 @@ class DBImpl : public DB {
  friend class WritePreparedTxn;
  friend class WritePreparedTxnDB;
  friend class WriteBatchWithIndex;
+  friend class WriteUnpreparedTxnDB;
+  friend class WriteUnpreparedTxn;
+
 #ifndef ROCKSDB_LITE
  friend class ForwardIterator;
 #endif
  friend struct SuperVersion;
  friend class CompactedDBImpl;
  friend class DBTest_ConcurrentFlushWAL_Test;
+  friend class DBTest_MixedSlowdownOptionsStop_Test;
 #ifndef NDEBUG
  friend class DBTest2_ReadCallbackTest_Test;
  friend class WriteCallbackTest_WriteWithCallbackTest_Test;
  friend class XFTransactionWriteHandler;
  friend class DBBlobIndexTest;
+  friend class WriteUnpreparedTransactionTest_RecoveryTest_Test;
 #endif
  struct CompactionState;

@@ -785,6 +848,8 @@ class DBImpl : public DB {
                 bool read_only = false, bool error_if_log_file_exist = false,
                 bool error_if_data_exists_in_logs = false);

+  Status ResumeImpl();
+
  void MaybeIgnoreError(Status* s) const;

  const Status CreateArchivalDirectory();
@@ -824,12 +889,41 @@ class DBImpl : public DB {
  Status SyncClosedLogs(JobContext* job_context);

  // Flush the in-memory write buffer to storage.  Switches to a new
-  // log-file/memtable and writes a new descriptor iff successful.
+  // log-file/memtable and writes a new descriptor iff successful. Then
+  // installs a new super version for the column family.
  Status FlushMemTableToOutputFile(ColumnFamilyData* cfd,
                                   const MutableCFOptions& mutable_cf_options,
                                   bool* madeProgress, JobContext* job_context,
+                                   SuperVersionContext* superversion_context,
                                   LogBuffer* log_buffer);

+  // Argument required by background flush thread.
+  struct BGFlushArg {
+    BGFlushArg()
+        : cfd_(nullptr), memtable_id_(0), superversion_context_(nullptr) {}
+    BGFlushArg(ColumnFamilyData* cfd, uint64_t memtable_id,
+               SuperVersionContext* superversion_context)
+        : cfd_(cfd),
+          memtable_id_(memtable_id),
+          superversion_context_(superversion_context) {}
+
+    // Column family to flush.
+    ColumnFamilyData* cfd_;
+    // Maximum ID of memtable to flush. In this column family, memtables with
+    // IDs smaller than this value must be flushed before this flush completes.
+    uint64_t memtable_id_;
+    // Pointer to a SuperVersionContext object. After flush completes, RocksDB
+    // installs a new superversion for the column family. This operation
+    // requires a SuperVersionContext object (currently embedded in JobContext).
+    SuperVersionContext* superversion_context_;
+  };
+
+  // Flush the memtables of (multiple) column families to multiple files on
+  // persistent storage.
+  Status FlushMemTablesToOutputFiles(
+      const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
+      JobContext* job_context, LogBuffer* log_buffer);
+
  // REQUIRES: log_numbers are sorted in ascending order
  Status RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
                         SequenceNumber* next_sequence, bool read_only);
@@ -842,6 +936,12 @@ class DBImpl : public DB {
  Status WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
                                     MemTable* mem, VersionEdit* edit);

+  // Restore alive_log_files_ and total_log_size_ after recovery.
+  // It needs to run only when there's no flush during recovery
+  // (e.g. avoid_flush_during_recovery=true). May also trigger flush
+  // in case total_log_size > max_total_wal_size.
+  Status RestoreAliveLogFiles(const std::vector<uint64_t>& log_numbers);
+
  // num_bytes: for slowdown case, delay time is calculated based on
  //            `num_bytes` going through.
  Status DelayWrite(uint64_t num_bytes, const WriteOptions& write_options);
@@ -851,19 +951,28 @@ class DBImpl : public DB {

  Status ScheduleFlushes(WriteContext* context);

-  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
-                        FlushReason flush_reason = FlushReason::kOthers);
+  Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context);

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

+  // Wait until flushing this column family won't stall writes
+  Status WaitUntilFlushWouldNotStallWrites(ColumnFamilyData* cfd,
+                                           bool* flush_needed);
+
  // Wait for memtable flushed.
  // If flush_memtable_id is non-null, wait until the memtable with the ID
  // gets flush. Otherwise, wait until the column family don't have any
  // memtable pending flush.
  Status WaitForFlushMemTable(ColumnFamilyData* cfd,
-                              const uint64_t* flush_memtable_id = nullptr);
+                              const uint64_t* flush_memtable_id = nullptr) {
+    return WaitForFlushMemTables({cfd}, {flush_memtable_id});
+  }
+  // Wait for memtables to be flushed for multiple column families.
+  Status WaitForFlushMemTables(
+      const autovector<ColumnFamilyData*>& cfds,
+      const autovector<const uint64_t*>& flush_memtable_ids);

  // REQUIRES: mutex locked
  Status SwitchWAL(WriteContext* write_context);
@@ -919,7 +1028,17 @@ class DBImpl : public DB {
  ColumnFamilyData* GetColumnFamilyDataByName(const std::string& cf_name);

  void MaybeScheduleFlushOrCompaction();
-  void SchedulePendingFlush(ColumnFamilyData* cfd, FlushReason flush_reason);
+
+  // A flush request specifies the column families to flush as well as the
+  // largest memtable id to persist for each column family. Once all the
+  // memtables whose IDs are smaller than or equal to this per-column-family
+  // 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;
+
+  void SchedulePendingFlush(const FlushRequest& req, FlushReason flush_reason);
+
  void SchedulePendingCompaction(ColumnFamilyData* cfd);
  void SchedulePendingPurge(std::string fname, std::string dir_to_sync,
                            FileType type, uint64_t number, int job_id);
@@ -938,9 +1057,10 @@ class DBImpl : public DB {
                              LogBuffer* log_buffer,
                              PrepickedCompaction* prepicked_compaction);
  Status BackgroundFlush(bool* madeProgress, JobContext* job_context,
-                         LogBuffer* log_buffer);
+                         LogBuffer* log_buffer, FlushReason* reason);

-  bool EnoughRoomForCompaction(const std::vector<CompactionInputFiles>& inputs,
+  bool EnoughRoomForCompaction(ColumnFamilyData* cfd,
+                               const std::vector<CompactionInputFiles>& inputs,
                               bool* sfm_bookkeeping, LogBuffer* log_buffer);

  void PrintStatistics();
@@ -961,8 +1081,7 @@ class DBImpl : public DB {
  // helper functions for adding and removing from flush & compaction queues
  void AddToCompactionQueue(ColumnFamilyData* cfd);
  ColumnFamilyData* PopFirstFromCompactionQueue();
-  void AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason);
-  ColumnFamilyData* PopFirstFromFlushQueue();
+  FlushRequest PopFirstFromFlushQueue();

  // helper function to call after some of the logs_ were synced
  void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);
@@ -975,6 +1094,10 @@ class DBImpl : public DB {

  Status CloseHelper();

+  Status FlushAllCFs(FlushReason flush_reason);
+
+  void WaitForBackgroundWork();
+
  // table_cache_ provides its own synchronization
  std::shared_ptr<Cache> table_cache_;

@@ -1195,7 +1318,7 @@ class DBImpl : public DB {
  // in MaybeScheduleFlushOrCompaction()
  // invariant(column family present in flush_queue_ <==>
  // ColumnFamilyData::pending_flush_ == true)
-  std::deque<ColumnFamilyData*> flush_queue_;
+  std::deque<FlushRequest> flush_queue_;
  // invariant(column family present in compaction_queue_ <==>
  // ColumnFamilyData::pending_compaction_ == true)
  std::deque<ColumnFamilyData*> compaction_queue_;
@@ -1265,9 +1388,6 @@ class DBImpl : public DB {
    PrepickedCompaction* prepicked_compaction;
  };

-  // Have we encountered a background error in paranoid mode?
-  Status bg_error_;
-
  // shall we disable deletion of obsolete files
  // if 0 the deletion is enabled.
  // if non-zero, files will not be getting deleted
@@ -1379,6 +1499,7 @@ class DBImpl : public DB {
  bool GetIntPropertyInternal(ColumnFamilyData* cfd,
                              const DBPropertyInfo& property_info,
                              bool is_locked, uint64_t* value);
+  bool GetPropertyHandleOptionsStatistics(std::string* value);

  bool HasPendingManualCompaction();
  bool HasExclusiveManualCompaction();
@@ -1407,6 +1528,13 @@ class DBImpl : public DB {
  //
  // Default: false
  const bool seq_per_batch_;
+  // This determines during recovery whether we expect one writebatch per
+  // recovered transaction, or potentially multiple writebatches per
+  // transaction. For WriteUnprepared, this is set to false, since multiple
+  // batches can exist per transaction.
+  //
+  // Default: true
+  const bool batch_per_txn_;
  // LastSequence also indicates last published sequence visibile to the
  // readers. Otherwise LastPublishedSequence should be used.
  const bool last_seq_same_as_publish_seq_;
@@ -1414,6 +1542,16 @@ class DBImpl : public DB {
  // flush/compaction and if it is not provided vis SnapshotChecker, we should
  // disable gc to be safe.
  const bool use_custom_gc_;
+  // Flag to indicate that the DB instance shutdown has been initiated. This
+  // different from shutting_down_ atomic in that it is set at the beginning
+  // of shutdown sequence, specifically in order to prevent any background
+  // error recovery from going on in parallel. The latter, shutting_down_,
+  // is set a little later during the shutdown after scheduling memtable
+  // flushes
+  bool shutdown_initiated_;
+  // Flag to indicate whether sst_file_manager object was allocated in
+  // DB::Open() or passed to us
+  bool own_sfm_;

  // Clients must periodically call SetPreserveDeletesSequenceNumber()
  // to advance this seqnum. Default value is 0 which means ALL deletes are
@@ -1424,6 +1562,8 @@ class DBImpl : public DB {

  // Flag to check whether Close() has been called on this DB
  bool closed_;
+
+  ErrorHandler error_handler_;
 };

 extern Options SanitizeOptions(const std::string& db,
@@ -14,6 +14,7 @@
 #include <inttypes.h>

 #include "db/builder.h"
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
@@ -25,7 +26,7 @@
 namespace rocksdb {

 bool DBImpl::EnoughRoomForCompaction(
-    const std::vector<CompactionInputFiles>& inputs,
+    ColumnFamilyData* cfd, const std::vector<CompactionInputFiles>& inputs,
    bool* sfm_reserved_compact_space, LogBuffer* log_buffer) {
  // Check if we have enough room to do the compaction
  bool enough_room = true;
@@ -33,12 +34,17 @@ bool DBImpl::EnoughRoomForCompaction(
  auto sfm = static_cast<SstFileManagerImpl*>(
      immutable_db_options_.sst_file_manager.get());
  if (sfm) {
-    enough_room = sfm->EnoughRoomForCompaction(inputs);
+    // Pass the current bg_error_ to SFM so it can decide what checks to
+    // perform. If this DB instance hasn't seen any error yet, the SFM can be
+    // optimistic and not do disk space checks
+    enough_room =
+        sfm->EnoughRoomForCompaction(cfd, inputs, error_handler_.GetBGError());
    if (enough_room) {
      *sfm_reserved_compact_space = true;
    }
  }
 #else
+  (void)cfd;
  (void)inputs;
  (void)sfm_reserved_compact_space;
 #endif  // ROCKSDB_LITE
@@ -93,14 +99,7 @@ Status DBImpl::SyncClosedLogs(JobContext* job_context) {
    // "number < current_log_number".
    MarkLogsSynced(current_log_number - 1, true, s);
    if (!s.ok()) {
-      Status new_bg_error = s;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kFlush,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
-      }
+      error_handler_.SetBGError(s, BackgroundErrorReason::kFlush);
      TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Failed");
      return s;
    }
@@ -110,7 +109,8 @@ Status DBImpl::SyncClosedLogs(JobContext* job_context) {

 Status DBImpl::FlushMemTableToOutputFile(
    ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
-    bool* made_progress, JobContext* job_context, LogBuffer* log_buffer) {
+    bool* made_progress, JobContext* job_context,
+    SuperVersionContext* superversion_context, LogBuffer* log_buffer) {
  mutex_.AssertHeld();
  assert(cfd->imm()->NumNotFlushed() != 0);
  assert(cfd->imm()->IsFlushPending());
@@ -166,7 +166,7 @@ Status DBImpl::FlushMemTableToOutputFile(
  }

  if (s.ok()) {
-    InstallSuperVersionAndScheduleWork(cfd, &job_context->superversion_context,
+    InstallSuperVersionAndScheduleWork(cfd, superversion_context,
                                       mutable_cf_options);
    if (made_progress) {
      *made_progress = 1;
@@ -177,18 +177,9 @@ Status DBImpl::FlushMemTableToOutputFile(
                     cfd->current()->storage_info()->LevelSummary(&tmp));
  }

-  if (!s.ok() && !s.IsShutdownInProgress() &&
-      immutable_db_options_.paranoid_checks && bg_error_.ok()) {
+  if (!s.ok() && !s.IsShutdownInProgress()) {
    Status new_bg_error = s;
-    // may temporarily unlock and lock the mutex.
-    EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                          BackgroundErrorReason::kFlush,
-                                          &new_bg_error, &mutex_);
-    if (!new_bg_error.ok()) {
-      // if a bad error happened (not ShutdownInProgress), paranoid_checks is
-      // true, and the error isn't handled by callback, mark DB read-only
-      bg_error_ = new_bg_error;
-    }
+    error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
  }
  if (s.ok()) {
 #ifndef ROCKSDB_LITE
@@ -202,18 +193,12 @@ Status DBImpl::FlushMemTableToOutputFile(
      std::string file_path = MakeTableFileName(
          cfd->ioptions()->cf_paths[0].path, file_meta.fd.GetNumber());
      sfm->OnAddFile(file_path);
-      if (sfm->IsMaxAllowedSpaceReached() && bg_error_.ok()) {
-        Status new_bg_error = Status::NoSpace("Max allowed space was reached");
+      if (sfm->IsMaxAllowedSpaceReached()) {
+        Status new_bg_error = Status::SpaceLimit("Max allowed space was reached");
        TEST_SYNC_POINT_CALLBACK(
            "DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
            &new_bg_error);
-        // may temporarily unlock and lock the mutex.
-        EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                              BackgroundErrorReason::kFlush,
-                                              &new_bg_error, &mutex_);
-        if (!new_bg_error.ok()) {
-          bg_error_ = new_bg_error;
-        }
+        error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
      }
    }
 #endif  // ROCKSDB_LITE
@@ -221,6 +206,25 @@ Status DBImpl::FlushMemTableToOutputFile(
  return s;
 }

+Status DBImpl::FlushMemTablesToOutputFiles(
+    const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
+    JobContext* job_context, LogBuffer* log_buffer) {
+  Status s;
+  for (auto& arg : bg_flush_args) {
+    ColumnFamilyData* cfd = arg.cfd_;
+    const MutableCFOptions& mutable_cf_options =
+        *cfd->GetLatestMutableCFOptions();
+    SuperVersionContext* superversion_context = arg.superversion_context_;
+    s = FlushMemTableToOutputFile(cfd, mutable_cf_options, made_progress,
+                                  job_context, superversion_context,
+                                  log_buffer);
+    if (!s.ok()) {
+      break;
+    }
+  }
+  return s;
+}
+
 void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
                                const MutableCFOptions& mutable_cf_options,
                                int job_id, TableProperties prop) {
@@ -251,8 +255,8 @@ void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
    info.job_id = job_id;
    info.triggered_writes_slowdown = triggered_writes_slowdown;
    info.triggered_writes_stop = triggered_writes_stop;
-    info.smallest_seqno = file_meta->smallest_seqno;
-    info.largest_seqno = file_meta->largest_seqno;
+    info.smallest_seqno = file_meta->fd.smallest_seqno;
+    info.largest_seqno = file_meta->fd.largest_seqno;
    info.table_properties = prop;
    info.flush_reason = cfd->GetFlushReason();
    for (auto listener : immutable_db_options_.listeners) {
@@ -302,8 +306,8 @@ void DBImpl::NotifyOnFlushCompleted(ColumnFamilyData* cfd,
    info.job_id = job_id;
    info.triggered_writes_slowdown = triggered_writes_slowdown;
    info.triggered_writes_stop = triggered_writes_stop;
-    info.smallest_seqno = file_meta->smallest_seqno;
-    info.largest_seqno = file_meta->largest_seqno;
+    info.smallest_seqno = file_meta->fd.smallest_seqno;
+    info.largest_seqno = file_meta->fd.largest_seqno;
    info.table_properties = prop;
    info.flush_reason = cfd->GetFlushReason();
    for (auto listener : immutable_db_options_.listeners) {
@@ -345,60 +349,12 @@ Status DBImpl::CompactRange(const CompactRangeOptions& options,
    CleanupSuperVersion(super_version);
  }

-  if (!options.allow_write_stall && flush_needed) {
-    InstrumentedMutexLock l(&mutex_);
-    uint64_t orig_active_memtable_id = cfd->mem()->GetID();
-    WriteStallCondition write_stall_condition = WriteStallCondition::kNormal;
-    do {
-      if (write_stall_condition != WriteStallCondition::kNormal) {
-        TEST_SYNC_POINT("DBImpl::CompactRange:StallWait");
-        ROCKS_LOG_INFO(immutable_db_options_.info_log,
-                       "[%s] CompactRange waiting on stall conditions to clear",
-                       cfd->GetName().c_str());
-        bg_cv_.Wait();
-      }
-      if (cfd->IsDropped() || shutting_down_.load(std::memory_order_acquire)) {
-        return Status::ShutdownInProgress();
-      }
-
-      uint64_t earliest_memtable_id =
-          std::min(cfd->mem()->GetID(), cfd->imm()->GetEarliestMemTableID());
-      if (earliest_memtable_id > orig_active_memtable_id) {
-        // We waited so long that the memtable we were originally waiting on was
-        // flushed.
-        flush_needed = false;
-        break;
-      }
-
-      const auto& mutable_cf_options = *cfd->GetLatestMutableCFOptions();
-      const auto* vstorage = cfd->current()->storage_info();
-
-      // Skip stalling check if we're below auto-flush and auto-compaction
-      // triggers. If it stalled in these conditions, that'd mean the stall
-      // triggers are so low that stalling is needed for any background work. In
-      // that case we shouldn't wait since background work won't be scheduled.
-      if (cfd->imm()->NumNotFlushed() <
-              cfd->ioptions()->min_write_buffer_number_to_merge &&
-          vstorage->l0_delay_trigger_count() <
-              mutable_cf_options.level0_file_num_compaction_trigger) {
-        break;
-      }
-
-      // check whether one extra immutable memtable or an extra L0 file would
-      // cause write stalling mode to be entered. It could still enter stall
-      // mode due to pending compaction bytes, but that's less common
-      write_stall_condition =
-          ColumnFamilyData::GetWriteStallConditionAndCause(
-              cfd->imm()->NumNotFlushed() + 1,
-              vstorage->l0_delay_trigger_count() + 1,
-              vstorage->estimated_compaction_needed_bytes(), mutable_cf_options)
-              .first;
-    } while (write_stall_condition != WriteStallCondition::kNormal);
-  }
-  TEST_SYNC_POINT("DBImpl::CompactRange:StallWaitDone");
  Status s;
  if (flush_needed) {
-    s = FlushMemTable(cfd, FlushOptions(), FlushReason::kManualCompaction);
+    FlushOptions fo;
+    fo.allow_write_stall = options.allow_write_stall;
+    s = FlushMemTable(cfd, fo, FlushReason::kManualCompaction,
+      false /* writes_stopped*/);
    if (!s.ok()) {
      LogFlush(immutable_db_options_.info_log);
      return s;
@@ -633,7 +589,7 @@ Status DBImpl::CompactFilesImpl(
  bool sfm_reserved_compact_space = false;
  // First check if we have enough room to do the compaction
  bool enough_room = EnoughRoomForCompaction(
-      input_files, &sfm_reserved_compact_space, log_buffer);
+      cfd, input_files, &sfm_reserved_compact_space, log_buffer);

  if (!enough_room) {
    // m's vars will get set properly at the end of this function,
@@ -674,7 +630,7 @@ Status DBImpl::CompactFilesImpl(
      env_options_for_compaction_, versions_.get(), &shutting_down_,
      preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
      GetDataDir(c->column_family_data(), c->output_path_id()), stats_, &mutex_,
-      &bg_error_, snapshot_seqs, earliest_write_conflict_snapshot,
+      &error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
      snapshot_checker, table_cache_, &event_logger_,
      c->mutable_cf_options()->paranoid_file_checks,
      c->mutable_cf_options()->report_bg_io_stats, dbname_,
@@ -712,7 +668,7 @@ Status DBImpl::CompactFilesImpl(
  Status status = compaction_job.Install(*c->mutable_cf_options());
  if (status.ok()) {
    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_context,
+        c->column_family_data(), &job_context->superversion_contexts[0],
        *c->mutable_cf_options(), FlushReason::kManualCompaction);
  }
  c->ReleaseCompactionFiles(s);
@@ -736,16 +692,7 @@ Status DBImpl::CompactFilesImpl(
                   "[%s] [JOB %d] Compaction error: %s",
                   c->column_family_data()->GetName().c_str(),
                   job_context->job_id, status.ToString().c_str());
-    if (immutable_db_options_.paranoid_checks && bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kCompaction,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
-      }
-    }
+    error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
  }

  if (output_file_names != nullptr) {
@@ -915,7 +862,7 @@ Status DBImpl::ReFitLevel(ColumnFamilyData* cfd, int level, int target_level) {
      edit.DeleteFile(level, f->fd.GetNumber());
      edit.AddFile(to_level, f->fd.GetNumber(), f->fd.GetPathId(),
                   f->fd.GetFileSize(), f->smallest, f->largest,
-                   f->smallest_seqno, f->largest_seqno,
+                   f->fd.smallest_seqno, f->fd.largest_seqno,
                   f->marked_for_compaction);
    }
    ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
@@ -972,6 +919,68 @@ Status DBImpl::Flush(const FlushOptions& flush_options,
  return s;
 }

+
+Status DBImpl::FlushAllCFs(FlushReason flush_reason) {
+  Status s;
+  WriteContext context;
+  WriteThread::Writer w;
+
+  mutex_.AssertHeld();
+  write_thread_.EnterUnbatched(&w, &mutex_);
+
+  FlushRequest flush_req;
+  for (auto cfd : *versions_->GetColumnFamilySet()) {
+    if (cfd->imm()->NumNotFlushed() == 0 && cfd->mem()->IsEmpty() &&
+        cached_recoverable_state_empty_.load()) {
+      // Nothing to flush
+      continue;
+    }
+
+    // SwitchMemtable() will release and reacquire mutex during execution
+    s = SwitchMemtable(cfd, &context);
+    if (!s.ok()) {
+      break;
+    }
+
+    cfd->imm()->FlushRequested();
+
+    flush_req.emplace_back(cfd, cfd->imm()->GetLatestMemTableID());
+  }
+
+  // schedule flush
+  if (s.ok() && !flush_req.empty()) {
+    SchedulePendingFlush(flush_req, flush_reason);
+    MaybeScheduleFlushOrCompaction();
+  }
+
+  write_thread_.ExitUnbatched(&w);
+
+  if (s.ok()) {
+    for (auto& flush : flush_req) {
+      auto cfd = flush.first;
+      auto flush_memtable_id = flush.second;
+      while (cfd->imm()->NumNotFlushed() > 0 &&
+             cfd->imm()->GetEarliestMemTableID() <= flush_memtable_id) {
+        if (!error_handler_.GetRecoveryError().ok()) {
+          break;
+        }
+        if (cfd->IsDropped()) {
+          // FlushJob cannot flush a dropped CF, if we did not break here
+          // we will loop forever since cfd->imm()->NumNotFlushed() will never
+          // drop to zero
+          continue;
+        }
+        cfd->Ref();
+        bg_cv_.Wait();
+        cfd->Unref();
+      }
+    }
+  }
+
+  flush_req.clear();
+  return s;
+}
+
 Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
                                   int output_level, uint32_t output_path_id,
                                   uint32_t max_subcompactions,
@@ -1107,64 +1116,166 @@ Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
                             FlushReason flush_reason, bool writes_stopped) {
  Status s;
  uint64_t flush_memtable_id = 0;
+  if (!flush_options.allow_write_stall) {
+    bool flush_needed = true;
+    s = WaitUntilFlushWouldNotStallWrites(cfd, &flush_needed);
+    TEST_SYNC_POINT("DBImpl::FlushMemTable:StallWaitDone");
+    if (!s.ok() || !flush_needed) {
+      return s;
+    }
+  }
+  FlushRequest flush_req;
  {
    WriteContext context;
    InstrumentedMutexLock guard_lock(&mutex_);

-    if (cfd->imm()->NumNotFlushed() == 0 && cfd->mem()->IsEmpty() &&
-        cached_recoverable_state_empty_.load()) {
-      // Nothing to flush
-      return Status::OK();
-    }
-
    WriteThread::Writer w;
    if (!writes_stopped) {
      write_thread_.EnterUnbatched(&w, &mutex_);
    }

-    // SwitchMemtable() will release and reacquire mutex during execution
-    s = SwitchMemtable(cfd, &context);
-    flush_memtable_id = cfd->imm()->GetLatestMemTableID();
+    if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
+        !cached_recoverable_state_empty_.load()) {
+      s = SwitchMemtable(cfd, &context);
+      flush_memtable_id = cfd->imm()->GetLatestMemTableID();
+      flush_req.emplace_back(cfd, flush_memtable_id);
+    }
+
+    if (s.ok() && !flush_req.empty()) {
+      for (auto& elem : flush_req) {
+        ColumnFamilyData* loop_cfd = elem.first;
+        loop_cfd->imm()->FlushRequested();
+      }
+      SchedulePendingFlush(flush_req, flush_reason);
+      MaybeScheduleFlushOrCompaction();
+    }

    if (!writes_stopped) {
      write_thread_.ExitUnbatched(&w);
    }
-
-    cfd->imm()->FlushRequested();
-
-    // schedule flush
-    SchedulePendingFlush(cfd, flush_reason);
-    MaybeScheduleFlushOrCompaction();
  }

  if (s.ok() && flush_options.wait) {
-    // Wait until the compaction completes
-    s = WaitForFlushMemTable(cfd, &flush_memtable_id);
+    autovector<ColumnFamilyData*> cfds;
+    autovector<const uint64_t*> flush_memtable_ids;
+    for (auto& iter : flush_req) {
+      cfds.push_back(iter.first);
+      flush_memtable_ids.push_back(&(iter.second));
+    }
+    s = WaitForFlushMemTables(cfds, flush_memtable_ids);
  }
+  TEST_SYNC_POINT("FlushMemTableFinished");
  return s;
 }

-Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd,
-                                    const uint64_t* flush_memtable_id) {
-  Status s;
+// Calling FlushMemTable(), whether from DB::Flush() or from Backup Engine, can
+// cause write stall, for example if one memtable is being flushed already.
+// This method tries to avoid write stall (similar to CompactRange() behavior)
+// it emulates how the SuperVersion / LSM would change if flush happens, checks
+// it against various constrains and delays flush if it'd cause write stall.
+// Called should check status and flush_needed to see if flush already happened.
+Status DBImpl::WaitUntilFlushWouldNotStallWrites(ColumnFamilyData* cfd,
+    bool* flush_needed) {
+  {
+    *flush_needed = true;
+    InstrumentedMutexLock l(&mutex_);
+    uint64_t orig_active_memtable_id = cfd->mem()->GetID();
+    WriteStallCondition write_stall_condition = WriteStallCondition::kNormal;
+    do {
+      if (write_stall_condition != WriteStallCondition::kNormal) {
+        TEST_SYNC_POINT("DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait");
+        ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                       "[%s] WaitUntilFlushWouldNotStallWrites"
+                       " waiting on stall conditions to clear",
+                       cfd->GetName().c_str());
+        bg_cv_.Wait();
+      }
+      if (cfd->IsDropped() || shutting_down_.load(std::memory_order_acquire)) {
+        return Status::ShutdownInProgress();
+      }
+
+      uint64_t earliest_memtable_id =
+          std::min(cfd->mem()->GetID(), cfd->imm()->GetEarliestMemTableID());
+      if (earliest_memtable_id > orig_active_memtable_id) {
+        // We waited so long that the memtable we were originally waiting on was
+        // flushed.
+        *flush_needed = false;
+        return Status::OK();
+      }
+
+      const auto& mutable_cf_options = *cfd->GetLatestMutableCFOptions();
+      const auto* vstorage = cfd->current()->storage_info();
+
+      // Skip stalling check if we're below auto-flush and auto-compaction
+      // triggers. If it stalled in these conditions, that'd mean the stall
+      // triggers are so low that stalling is needed for any background work. In
+      // that case we shouldn't wait since background work won't be scheduled.
+      if (cfd->imm()->NumNotFlushed() <
+              cfd->ioptions()->min_write_buffer_number_to_merge &&
+          vstorage->l0_delay_trigger_count() <
+              mutable_cf_options.level0_file_num_compaction_trigger) {
+        break;
+      }
+
+      // check whether one extra immutable memtable or an extra L0 file would
+      // cause write stalling mode to be entered. It could still enter stall
+      // mode due to pending compaction bytes, but that's less common
+      write_stall_condition =
+          ColumnFamilyData::GetWriteStallConditionAndCause(
+              cfd->imm()->NumNotFlushed() + 1,
+              vstorage->l0_delay_trigger_count() + 1,
+              vstorage->estimated_compaction_needed_bytes(), mutable_cf_options)
+              .first;
+    } while (write_stall_condition != WriteStallCondition::kNormal);
+  }
+  return Status::OK();
+}
+
+// Wait for memtables to be flushed for multiple column families.
+// let N = cfds.size()
+// for i in [0, N),
+//  1) if flush_memtable_ids[i] is not null, then the memtables with lower IDs
+//     have to be flushed for THIS column family;
+//  2) if flush_memtable_ids[i] is null, then all memtables in THIS column
+//     family have to be flushed.
+// Finish waiting when ALL column families finish flushing memtables.
+Status DBImpl::WaitForFlushMemTables(
+    const autovector<ColumnFamilyData*>& cfds,
+    const autovector<const uint64_t*>& flush_memtable_ids) {
+  int num = static_cast<int>(cfds.size());
  // Wait until the compaction completes
  InstrumentedMutexLock l(&mutex_);
-  while (cfd->imm()->NumNotFlushed() > 0 && bg_error_.ok() &&
-         (flush_memtable_id == nullptr ||
-          cfd->imm()->GetEarliestMemTableID() <= *flush_memtable_id)) {
+  while (!error_handler_.IsDBStopped()) {
    if (shutting_down_.load(std::memory_order_acquire)) {
      return Status::ShutdownInProgress();
    }
-    if (cfd->IsDropped()) {
-      // FlushJob cannot flush a dropped CF, if we did not break here
-      // we will loop forever since cfd->imm()->NumNotFlushed() will never
-      // drop to zero
+    // Number of column families that have been dropped.
+    int num_dropped = 0;
+    // Number of column families that have finished flush.
+    int num_finished = 0;
+    for (int i = 0; i < num; ++i) {
+      if (cfds[i]->IsDropped()) {
+        ++num_dropped;
+      } else if (cfds[i]->imm()->NumNotFlushed() == 0 ||
+                 (flush_memtable_ids[i] != nullptr &&
+                  cfds[i]->imm()->GetEarliestMemTableID() >
+                      *flush_memtable_ids[i])) {
+        ++num_finished;
+      }
+    }
+    if (1 == num_dropped && 1 == num) {
      return Status::InvalidArgument("Cannot flush a dropped CF");
    }
+    // Column families involved in this flush request have either been dropped
+    // or finished flush. Then it's time to finish waiting.
+    if (num_dropped + num_finished == num) {
+      break;
+    }
    bg_cv_.Wait();
  }
-  if (!bg_error_.ok()) {
-    s = bg_error_;
+  Status s;
+  if (error_handler_.IsDBStopped()) {
+    s = error_handler_.GetBGError();
  }
  return s;
 }
@@ -1192,6 +1303,12 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
  if (bg_work_paused_ > 0) {
    // we paused the background work
    return;
+  } else if (error_handler_.IsBGWorkStopped() &&
+      !error_handler_.IsRecoveryInProgress()) {
+    // There has been a hard error and this call is not part of the recovery
+    // sequence. Bail out here so we don't get into an endless loop of
+    // scheduling BG work which will again call this function
+    return;
  } else if (shutting_down_.load(std::memory_order_acquire)) {
    // DB is being deleted; no more background compactions
    return;
@@ -1201,7 +1318,6 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
      env_->GetBackgroundThreads(Env::Priority::HIGH) == 0;
  while (!is_flush_pool_empty && unscheduled_flushes_ > 0 &&
         bg_flush_scheduled_ < bg_job_limits.max_flushes) {
-    unscheduled_flushes_--;
    bg_flush_scheduled_++;
    env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::HIGH, this);
  }
@@ -1212,7 +1328,6 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
    while (unscheduled_flushes_ > 0 &&
           bg_flush_scheduled_ + bg_compaction_scheduled_ <
               bg_job_limits.max_flushes) {
-      unscheduled_flushes_--;
      bg_flush_scheduled_++;
      env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::LOW, this);
    }
@@ -1221,6 +1336,12 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
  if (bg_compaction_paused_ > 0) {
    // we paused the background compaction
    return;
+  } else if (error_handler_.IsBGWorkStopped()) {
+    // Compaction is not part of the recovery sequence from a hard error. We
+    // might get here because recovery might do a flush and install a new
+    // super version, which will try to schedule pending compactions. Bail
+    // out here and let the higher level recovery handle compactions
+    return;
  }

  if (HasExclusiveManualCompaction()) {
@@ -1289,30 +1410,28 @@ ColumnFamilyData* DBImpl::PopFirstFromCompactionQueue() {
  return cfd;
 }

-void DBImpl::AddToFlushQueue(ColumnFamilyData* cfd, FlushReason flush_reason) {
-  assert(!cfd->queued_for_flush());
-  cfd->Ref();
-  flush_queue_.push_back(cfd);
-  cfd->set_queued_for_flush(true);
-  cfd->SetFlushReason(flush_reason);
-}
-
-ColumnFamilyData* DBImpl::PopFirstFromFlushQueue() {
+DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
  assert(!flush_queue_.empty());
-  auto cfd = *flush_queue_.begin();
+  FlushRequest flush_req = flush_queue_.front();
+  assert(unscheduled_flushes_ >= static_cast<int>(flush_req.size()));
+  unscheduled_flushes_ -= static_cast<int>(flush_req.size());
  flush_queue_.pop_front();
-  assert(cfd->queued_for_flush());
-  cfd->set_queued_for_flush(false);
  // TODO: need to unset flush reason?
-  return cfd;
+  return flush_req;
 }

-void DBImpl::SchedulePendingFlush(ColumnFamilyData* cfd,
+void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
                                  FlushReason flush_reason) {
-  if (!cfd->queued_for_flush() && cfd->imm()->IsFlushPending()) {
-    AddToFlushQueue(cfd, flush_reason);
-    ++unscheduled_flushes_;
+  if (flush_req.empty()) {
+    return;
  }
+  for (auto& iter : flush_req) {
+    ColumnFamilyData* cfd = iter.first;
+    cfd->Ref();
+    cfd->SetFlushReason(flush_reason);
+  }
+  unscheduled_flushes_ += static_cast<int>(flush_req.size());
+  flush_queue_.push_back(flush_req);
 }

 void DBImpl::SchedulePendingCompaction(ColumnFamilyData* cfd) {
@@ -1380,52 +1499,77 @@ void DBImpl::UnscheduleCallback(void* arg) {
 }

 Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
-                               LogBuffer* log_buffer) {
+                               LogBuffer* log_buffer, FlushReason* reason) {
  mutex_.AssertHeld();

-  Status status = bg_error_;
-  if (status.ok() && shutting_down_.load(std::memory_order_acquire)) {
-    status = Status::ShutdownInProgress();
+  Status status;
+  *reason = FlushReason::kOthers;
+  // If BG work is stopped due to an error, but a recovery is in progress,
+  // that means this flush is part of the recovery. So allow it to go through
+  if (!error_handler_.IsBGWorkStopped()) {
+    if (shutting_down_.load(std::memory_order_acquire)) {
+      status = Status::ShutdownInProgress();
+    }
+  } else if (!error_handler_.IsRecoveryInProgress()) {
+    status = error_handler_.GetBGError();
  }

  if (!status.ok()) {
    return status;
  }

-  ColumnFamilyData* cfd = nullptr;
+  autovector<BGFlushArg> bg_flush_args;
+  std::vector<SuperVersionContext>& superversion_contexts =
+      job_context->superversion_contexts;
  while (!flush_queue_.empty()) {
    // This cfd is already referenced
-    auto first_cfd = PopFirstFromFlushQueue();
+    const FlushRequest& flush_req = PopFirstFromFlushQueue();
+    superversion_contexts.clear();
+    superversion_contexts.reserve(flush_req.size());

-    if (first_cfd->IsDropped() || !first_cfd->imm()->IsFlushPending()) {
-      // can't flush this CF, try next one
-      if (first_cfd->Unref()) {
-        delete first_cfd;
+    for (const auto& iter : flush_req) {
+      ColumnFamilyData* cfd = iter.first;
+      if (cfd->IsDropped() || !cfd->imm()->IsFlushPending()) {
+        // can't flush this CF, try next one
+        if (cfd->Unref()) {
+          delete cfd;
+        }
+        continue;
      }
-      continue;
+      superversion_contexts.emplace_back(SuperVersionContext(true));
+      bg_flush_args.emplace_back(cfd, iter.second,
+                                 &(superversion_contexts.back()));
+    }
+    if (!bg_flush_args.empty()) {
+      break;
    }
-
-    // found a flush!
-    cfd = first_cfd;
-    break;
  }

-  if (cfd != nullptr) {
-    const MutableCFOptions mutable_cf_options =
-        *cfd->GetLatestMutableCFOptions();
+  if (!bg_flush_args.empty()) {
    auto bg_job_limits = GetBGJobLimits();
-    ROCKS_LOG_BUFFER(
-        log_buffer,
-        "Calling FlushMemTableToOutputFile with column "
-        "family [%s], flush slots available %d, compaction slots available %d, "
-        "flush slots scheduled %d, compaction slots scheduled %d",
-        cfd->GetName().c_str(), bg_job_limits.max_flushes,
-        bg_job_limits.max_compactions, bg_flush_scheduled_,
-        bg_compaction_scheduled_);
-    status = FlushMemTableToOutputFile(cfd, mutable_cf_options, made_progress,
-                                       job_context, log_buffer);
-    if (cfd->Unref()) {
-      delete cfd;
+    for (const auto& arg : bg_flush_args) {
+      ColumnFamilyData* cfd = arg.cfd_;
+      ROCKS_LOG_BUFFER(
+          log_buffer,
+          "Calling FlushMemTableToOutputFile with column "
+          "family [%s], flush slots available %d, compaction slots available "
+          "%d, "
+          "flush slots scheduled %d, compaction slots scheduled %d",
+          cfd->GetName().c_str(), bg_job_limits.max_flushes,
+          bg_job_limits.max_compactions, bg_flush_scheduled_,
+          bg_compaction_scheduled_);
+    }
+    status = FlushMemTablesToOutputFiles(bg_flush_args, made_progress,
+                                         job_context, log_buffer);
+    // All the CFDs in the FlushReq must have the same flush reason, so just
+    // grab the first one
+    *reason = bg_flush_args[0].cfd_->GetFlushReason();
+    for (auto& arg : bg_flush_args) {
+      ColumnFamilyData* cfd = arg.cfd_;
+      if (cfd->Unref()) {
+        delete cfd;
+        arg.cfd_ = nullptr;
+      }
    }
  }
  return status;
@@ -1446,9 +1590,12 @@ void DBImpl::BackgroundCallFlush() {

    auto pending_outputs_inserted_elem =
        CaptureCurrentFileNumberInPendingOutputs();
+    FlushReason reason;

-    Status s = BackgroundFlush(&made_progress, &job_context, &log_buffer);
-    if (!s.ok() && !s.IsShutdownInProgress()) {
+    Status s =
+        BackgroundFlush(&made_progress, &job_context, &log_buffer, &reason);
+    if (!s.ok() && !s.IsShutdownInProgress() &&
+        reason != FlushReason::kErrorRecovery) {
      // Wait a little bit before retrying background flush in
      // case this is an environmental problem and we do not want to
      // chew up resources for failed flushes for the duration of
@@ -1631,9 +1778,18 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
      is_manual && manual_compaction->disallow_trivial_move;

  CompactionJobStats compaction_job_stats;
-  Status status = bg_error_;
-  if (status.ok() && shutting_down_.load(std::memory_order_acquire)) {
-    status = Status::ShutdownInProgress();
+  Status status;
+  if (!error_handler_.IsBGWorkStopped()) {
+    if (shutting_down_.load(std::memory_order_acquire)) {
+      status = Status::ShutdownInProgress();
+    }
+  } else {
+    status = error_handler_.GetBGError();
+    // If we get here, it means a hard error happened after this compaction
+    // was scheduled by MaybeScheduleFlushOrCompaction(), but before it got
+    // a chance to execute. Since we didn't pop a cfd from the compaction
+    // queue, increment unscheduled_compactions_
+    unscheduled_compactions_++;
  }

  if (!status.ok()) {
@@ -1669,7 +1825,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    } else {
      // First check if we have enough room to do the compaction
      bool enough_room = EnoughRoomForCompaction(
-          *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
+          m->cfd, *(c->inputs()), &sfm_reserved_compact_space, log_buffer);

      if (!enough_room) {
        // Then don't do the compaction
@@ -1732,7 +1888,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,

      if (c != nullptr) {
        bool enough_room = EnoughRoomForCompaction(
-            *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
+            cfd, *(c->inputs()), &sfm_reserved_compact_space, log_buffer);

        if (!enough_room) {
          // Then don't do the compaction
@@ -1796,7 +1952,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                                    *c->mutable_cf_options(), c->edit(),
                                    &mutex_, directories_.GetDbDir());
    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_context,
+        c->column_family_data(), &job_context->superversion_contexts[0],
        *c->mutable_cf_options(), FlushReason::kAutoCompaction);
    ROCKS_LOG_BUFFER(log_buffer, "[%s] Deleted %d files\n",
                     c->column_family_data()->GetName().c_str(),
@@ -1825,8 +1981,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
        c->edit()->DeleteFile(c->level(l), f->fd.GetNumber());
        c->edit()->AddFile(c->output_level(), f->fd.GetNumber(),
                           f->fd.GetPathId(), f->fd.GetFileSize(), f->smallest,
-                           f->largest, f->smallest_seqno, f->largest_seqno,
-                           f->marked_for_compaction);
+                           f->largest, f->fd.smallest_seqno,
+                           f->fd.largest_seqno, f->marked_for_compaction);

        ROCKS_LOG_BUFFER(
            log_buffer,
@@ -1843,7 +1999,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                                    &mutex_, directories_.GetDbDir());
    // Use latest MutableCFOptions
    InstallSuperVersionAndScheduleWork(
-        c->column_family_data(), &job_context->superversion_context,
+        c->column_family_data(), &job_context->superversion_contexts[0],
        *c->mutable_cf_options(), FlushReason::kAutoCompaction);

    VersionStorageInfo::LevelSummaryStorage tmp;
@@ -1905,7 +2061,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
        env_options_for_compaction_, versions_.get(), &shutting_down_,
        preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
        GetDataDir(c->column_family_data(), c->output_path_id()), stats_,
-        &mutex_, &bg_error_, snapshot_seqs, earliest_write_conflict_snapshot,
+        &mutex_, &error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
        snapshot_checker, table_cache_, &event_logger_,
        c->mutable_cf_options()->paranoid_file_checks,
        c->mutable_cf_options()->report_bg_io_stats, dbname_,
@@ -1920,7 +2076,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    status = compaction_job.Install(*c->mutable_cf_options());
    if (status.ok()) {
      InstallSuperVersionAndScheduleWork(
-          c->column_family_data(), &job_context->superversion_context,
+          c->column_family_data(), &job_context->superversion_contexts[0],
          *c->mutable_cf_options(), FlushReason::kAutoCompaction);
    }
    *made_progress = true;
@@ -1941,8 +2097,6 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
    NotifyOnCompactionCompleted(c->column_family_data(), c.get(), status,
                                compaction_job_stats, job_context->job_id);
  }
-  // this will unref its input_version and column_family_data
-  c.reset();

  if (status.ok() || status.IsCompactionTooLarge()) {
    // Done
@@ -1951,17 +2105,27 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
  } else {
    ROCKS_LOG_WARN(immutable_db_options_.info_log, "Compaction error: %s",
                   status.ToString().c_str());
-    if (immutable_db_options_.paranoid_checks && bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kCompaction,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;
+    error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
+    if (c != nullptr && !is_manual && !error_handler_.IsBGWorkStopped()) {
+      // Put this cfd back in the compaction queue so we can retry after some
+      // time
+      auto cfd = c->column_family_data();
+      assert(cfd != nullptr);
+      // Since this compaction failed, we need to recompute the score so it
+      // takes the original input files into account
+      c->column_family_data()
+          ->current()
+          ->storage_info()
+          ->ComputeCompactionScore(*(c->immutable_cf_options()),
+                                   *(c->mutable_cf_options()));
+      if (!cfd->queued_for_compaction()) {
+        AddToCompactionQueue(cfd);
+        ++unscheduled_compactions_;
      }
    }
  }
+  // this will unref its input_version and column_family_data
+  c.reset();

  if (is_manual) {
    ManualCompactionState* m = manual_compaction;
@@ -2110,7 +2274,10 @@ bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {

 void DBImpl::InstallSuperVersionAndScheduleWork(
    ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-    const MutableCFOptions& mutable_cf_options, FlushReason flush_reason) {
+    const MutableCFOptions& mutable_cf_options,
+    FlushReason /* flush_reason */) {
+  // TODO(yanqin) investigate if 'flush_reason' can be removed since it's not
+  // used.
  mutex_.AssertHeld();

  // Update max_total_in_memory_state_
@@ -2121,14 +2288,14 @@ void DBImpl::InstallSuperVersionAndScheduleWork(
                        old_sv->mutable_cf_options.max_write_buffer_number;
  }

-  if (sv_context->new_superversion == nullptr) {
+  // this branch is unlikely to step in
+  if (UNLIKELY(sv_context->new_superversion == nullptr)) {
    sv_context->NewSuperVersion();
  }
  cfd->InstallSuperVersion(sv_context, &mutex_, mutable_cf_options);

  // Whenever we install new SuperVersion, we might need to issue new flushes or
  // compactions.
-  SchedulePendingFlush(cfd, flush_reason);
  SchedulePendingCompaction(cfd);
  MaybeScheduleFlushOrCompaction();

@@ -10,6 +10,7 @@
 #ifndef NDEBUG

 #include "db/db_impl.h"
+#include "db/error_handler.h"
 #include "monitoring/thread_status_updater.h"

 namespace rocksdb {
@@ -99,9 +100,11 @@ Status DBImpl::TEST_SwitchMemtable(ColumnFamilyData* cfd) {
  return SwitchMemtable(cfd, &write_context);
 }

-Status DBImpl::TEST_FlushMemTable(bool wait, ColumnFamilyHandle* cfh) {
+Status DBImpl::TEST_FlushMemTable(bool wait, bool allow_write_stall,
+    ColumnFamilyHandle* cfh) {
  FlushOptions fo;
  fo.wait = wait;
+  fo.allow_write_stall = allow_write_stall;
  ColumnFamilyData* cfd;
  if (cfh == nullptr) {
    cfd = default_cf_handle_->cfd();
@@ -134,10 +137,10 @@ Status DBImpl::TEST_WaitForCompact(bool wait_unscheduled) {
  while ((bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
          bg_flush_scheduled_ ||
          (wait_unscheduled && unscheduled_compactions_)) &&
-         bg_error_.ok()) {
+         (error_handler_.GetBGError() == Status::OK())) {
    bg_cv_.Wait();
  }
-  return bg_error_;
+  return error_handler_.GetBGError();
 }

 void DBImpl::TEST_LockMutex() {
@@ -234,5 +237,11 @@ SequenceNumber DBImpl::TEST_GetLastVisibleSequence() const {
  }
 }

+size_t DBImpl::TEST_GetWalPreallocateBlockSize(
+    uint64_t write_buffer_size) const {
+  InstrumentedMutexLock l(&mutex_);
+  return GetWalPreallocateBlockSize(write_buffer_size);
+}
+
 }  // namespace rocksdb
 #endif  // NDEBUG
@@ -131,16 +131,16 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
      edit.DeleteFile(0, f->fd.GetNumber());
      edit.AddFile(target_level, f->fd.GetNumber(), f->fd.GetPathId(),
                   f->fd.GetFileSize(), f->smallest, f->largest,
-                   f->smallest_seqno, f->largest_seqno,
+                   f->fd.smallest_seqno, f->fd.largest_seqno,
                   f->marked_for_compaction);
    }

    status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                    &edit, &mutex_, directories_.GetDbDir());
    if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_context,
-          *cfd->GetLatestMutableCFOptions());
+      InstallSuperVersionAndScheduleWork(cfd,
+                                         &job_context.superversion_contexts[0],
+                                         *cfd->GetLatestMutableCFOptions());
    }
  }  // lock released here
  LogFlush(immutable_db_options_.info_log);
@@ -12,6 +12,7 @@
 #define __STDC_FORMAT_MACROS
 #endif
 #include <inttypes.h>
+#include <set>
 #include <unordered_set>
 #include "db/event_helpers.h"
 #include "db/memtable_list.h"
@@ -99,10 +100,10 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
  if (doing_the_full_scan) {
    InfoLogPrefix info_log_prefix(!immutable_db_options_.db_log_dir.empty(),
        dbname_);
-    std::vector<std::string> paths;
+    std::set<std::string> paths;
    for (size_t path_id = 0; path_id < immutable_db_options_.db_paths.size();
         path_id++) {
-      paths.emplace_back(immutable_db_options_.db_paths[path_id].path);
+      paths.insert(immutable_db_options_.db_paths[path_id].path);
    }

    // Note that if cf_paths is not specified in the ColumnFamilyOptions
@@ -113,7 +114,11 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
    for (auto cfd : *versions_->GetColumnFamilySet()) {
      for (size_t path_id = 0; path_id < cfd->ioptions()->cf_paths.size();
           path_id++) {
-        paths.emplace_back(cfd->ioptions()->cf_paths[path_id].path);
+        auto& path = cfd->ioptions()->cf_paths[path_id].path;
+
+        if (paths.find(path) == paths.end()) {
+          paths.insert(path);
+        }
      }
    }

@@ -345,9 +350,32 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
  std::vector<std::string> old_info_log_files;
  InfoLogPrefix info_log_prefix(!immutable_db_options_.db_log_dir.empty(),
                                dbname_);
+
+  // File numbers of most recent two OPTIONS file in candidate_files (found in
+  // previos FindObsoleteFiles(full_scan=true))
+  // At this point, there must not be any duplicate file numbers in
+  // candidate_files.
+  uint64_t optsfile_num1 = std::numeric_limits<uint64_t>::min();
+  uint64_t optsfile_num2 = std::numeric_limits<uint64_t>::min();
+  for (const auto& candidate_file : candidate_files) {
+    const std::string& fname = candidate_file.file_name;
+    uint64_t number;
+    FileType type;
+    if (!ParseFileName(fname, &number, info_log_prefix.prefix, &type) ||
+        type != kOptionsFile) {
+      continue;
+    }
+    if (number > optsfile_num1) {
+      optsfile_num2 = optsfile_num1;
+      optsfile_num1 = number;
+    } else if (number > optsfile_num2) {
+      optsfile_num2 = number;
+    }
+  }
+
  std::unordered_set<uint64_t> files_to_del;
  for (const auto& candidate_file : candidate_files) {
-    std::string to_delete = candidate_file.file_name;
+    const std::string& to_delete = candidate_file.file_name;
    uint64_t number;
    FileType type;
    // Ignore file if we cannot recognize it.
@@ -396,11 +424,19 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
          old_info_log_files.push_back(to_delete);
        }
        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:
      case kIdentityFile:
      case kMetaDatabase:
-      case kOptionsFile:
      case kBlobFile:
        keep = true;
        break;
@@ -14,6 +14,7 @@
 #include <inttypes.h>

 #include "db/builder.h"
+#include "db/error_handler.h"
 #include "options/options_helper.h"
 #include "rocksdb/wal_filter.h"
 #include "table/block_based_table_factory.h"
@@ -133,8 +134,15 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
  for (size_t i = 0; i < result.db_paths.size(); i++) {
    DeleteScheduler::CleanupDirectory(result.env, sfm, result.db_paths[i].path);
  }
-#endif

+  // Create a default SstFileManager for purposes of tracking compaction size
+  // and facilitating recovery from out of space errors.
+  if (result.sst_file_manager.get() == nullptr) {
+    std::shared_ptr<SstFileManager> sst_file_manager(
+        NewSstFileManager(result.env, result.info_log));
+    result.sst_file_manager = sst_file_manager;
+  }
+#endif
  return result;
 }

@@ -231,7 +239,7 @@ Status DBImpl::NewDB() {
    file->SetPreallocationBlockSize(
        immutable_db_options_.manifest_preallocation_size);
    unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), env_options));
+        new WritableFileWriter(std::move(file), manifest, env_options));
    log::Writer log(std::move(file_writer), 0, false);
    std::string record;
    new_db.EncodeTo(&record);
@@ -360,7 +368,7 @@ Status DBImpl::Recover(
      s = env_->NewRandomAccessFile(IdentityFileName(dbname_), &idfile,
                                    customized_env);
      if (!s.ok()) {
-        const char* error_msg = s.ToString().c_str();
+        std::string error_str = s.ToString();
        // Check if unsupported Direct I/O is the root cause
        customized_env.use_direct_reads = false;
        s = env_->NewRandomAccessFile(IdentityFileName(dbname_), &idfile,
@@ -370,7 +378,7 @@ Status DBImpl::Recover(
              "Direct I/O is not supported by the specified DB.");
        } else {
          return Status::InvalidArgument(
-              "Found options incompatible with filesystem", error_msg);
+              "Found options incompatible with filesystem", error_str.c_str());
        }
      }
    }
@@ -388,6 +396,16 @@ Status DBImpl::Recover(
      }
    }
  }
+
+  // Initial max_total_in_memory_state_ before recovery logs. Log recovery
+  // may check this value to decide whether to flush.
+  max_total_in_memory_state_ = 0;
+  for (auto cfd : *versions_->GetColumnFamilySet()) {
+    auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
+    max_total_in_memory_state_ += mutable_cf_options->write_buffer_size *
+                                  mutable_cf_options->max_write_buffer_number;
+  }
+
  if (s.ok()) {
    SequenceNumber next_sequence(kMaxSequenceNumber);
    default_cf_handle_ = new ColumnFamilyHandleImpl(
@@ -460,14 +478,6 @@ Status DBImpl::Recover(
    }
  }

-  // Initial value
-  max_total_in_memory_state_ = 0;
-  for (auto cfd : *versions_->GetColumnFamilySet()) {
-    auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
-    max_total_in_memory_state_ += mutable_cf_options->write_buffer_size *
-                                  mutable_cf_options->max_write_buffer_number;
-  }
-
  return s;
 }

@@ -577,7 +587,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
          continue;
        }
      }
-      file_reader.reset(new SequentialFileReader(std::move(file)));
+      file_reader.reset(new SequentialFileReader(std::move(file), fname));
    }

    // Create the log reader.
@@ -597,8 +607,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
    // to be skipped instead of propagating bad information (like overly
    // large sequence numbers).
    log::Reader reader(immutable_db_options_.info_log, std::move(file_reader),
-                       &reporter, true /*checksum*/, 0 /*initial_offset*/,
-                       log_number);
+                       &reporter, true /*checksum*/, log_number);

    // Determine if we should tolerate incomplete records at the tail end of the
    // Read all the records and add to a memtable
@@ -719,7 +728,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
      status = WriteBatchInternal::InsertInto(
          &batch, column_family_memtables_.get(), &flush_scheduler_, true,
          log_number, this, false /* concurrent_memtable_writes */,
-          next_sequence, &has_valid_writes, seq_per_batch_);
+          next_sequence, &has_valid_writes, seq_per_batch_, batch_per_txn_);
      MaybeIgnoreError(&status);
      if (!status.ok()) {
        // We are treating this as a failure while reading since we read valid
@@ -878,18 +887,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
    }
  }

-  if (data_seen && !flushed) {
-    // Mark these as alive so they'll be considered for deletion later by
-    // FindObsoleteFiles()
-    if (two_write_queues_) {
-      log_write_mutex_.Lock();
-    }
-    for (auto log_number : log_numbers) {
-      alive_log_files_.push_back(LogFileNumberSize(log_number));
-    }
-    if (two_write_queues_) {
-      log_write_mutex_.Unlock();
-    }
+  if (status.ok() && data_seen && !flushed) {
+    status = RestoreAliveLogFiles(log_numbers);
  }

  event_logger_.Log() << "job" << job_id << "event"
@@ -898,6 +897,60 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
  return status;
 }

+Status DBImpl::RestoreAliveLogFiles(const std::vector<uint64_t>& log_numbers) {
+  if (log_numbers.empty()) {
+    return Status::OK();
+  }
+  Status s;
+  mutex_.AssertHeld();
+  assert(immutable_db_options_.avoid_flush_during_recovery);
+  if (two_write_queues_) {
+    log_write_mutex_.Lock();
+  }
+  // Mark these as alive so they'll be considered for deletion later by
+  // FindObsoleteFiles()
+  total_log_size_ = 0;
+  log_empty_ = false;
+  for (auto log_number : log_numbers) {
+    LogFileNumberSize log(log_number);
+    std::string fname = LogFileName(immutable_db_options_.wal_dir, log_number);
+    // This gets the appear size of the logs, not including preallocated space.
+    s = env_->GetFileSize(fname, &log.size);
+    if (!s.ok()) {
+      break;
+    }
+    total_log_size_ += log.size;
+    alive_log_files_.push_back(log);
+    // We preallocate space for logs, but then after a crash and restart, those
+    // preallocated space are not needed anymore. It is likely only the last
+    // log has such preallocated space, so we only truncate for the last log.
+    if (log_number == log_numbers.back()) {
+      std::unique_ptr<WritableFile> last_log;
+      Status truncate_status = env_->ReopenWritableFile(
+          fname, &last_log,
+          env_->OptimizeForLogWrite(
+              env_options_,
+              BuildDBOptions(immutable_db_options_, mutable_db_options_)));
+      if (truncate_status.ok()) {
+        truncate_status = last_log->Truncate(log.size);
+      }
+      if (truncate_status.ok()) {
+        truncate_status = last_log->Close();
+      }
+      // Not a critical error if fail to truncate.
+      if (!truncate_status.ok()) {
+        ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                       "Failed to truncate log #%" PRIu64 ": %s", log_number,
+                       truncate_status.ToString().c_str());
+      }
+    }
+  }
+  if (two_write_queues_) {
+    log_write_mutex_.Unlock();
+  }
+  return s;
+}
+
 Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
                                           MemTable* mem, VersionEdit* edit) {
  mutex_.AssertHeld();
@@ -968,7 +1021,7 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
  if (s.ok() && meta.fd.GetFileSize() > 0) {
    edit->AddFile(level, meta.fd.GetNumber(), meta.fd.GetPathId(),
                  meta.fd.GetFileSize(), meta.smallest, meta.largest,
-                  meta.smallest_seqno, meta.largest_seqno,
+                  meta.fd.smallest_seqno, meta.fd.largest_seqno,
                  meta.marked_for_compaction);
  }

@@ -1003,15 +1056,16 @@ Status DB::Open(const Options& options, const std::string& dbname, DB** dbptr) {
 Status DB::Open(const DBOptions& db_options, const std::string& dbname,
                const std::vector<ColumnFamilyDescriptor>& column_families,
                std::vector<ColumnFamilyHandle*>* handles, DB** dbptr) {
-  const bool seq_per_batch = true;
+  const bool kSeqPerBatch = true;
+  const bool kBatchPerTxn = true;
  return DBImpl::Open(db_options, dbname, column_families, handles, dbptr,
-                      !seq_per_batch);
+                      !kSeqPerBatch, kBatchPerTxn);
 }

 Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
                    const std::vector<ColumnFamilyDescriptor>& column_families,
                    std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
-                    const bool seq_per_batch) {
+                    const bool seq_per_batch, const bool batch_per_txn) {
  Status s = SanitizeOptionsByTable(db_options, column_families);
  if (!s.ok()) {
    return s;
@@ -1031,7 +1085,7 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
        std::max(max_write_buffer_size, cf.options.write_buffer_size);
  }

-  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch);
+  DBImpl* impl = new DBImpl(db_options, dbname, seq_per_batch, batch_per_txn);
  s = impl->env_->CreateDirIfMissing(impl->immutable_db_options_.wal_dir);
  if (s.ok()) {
    std::vector<std::string> paths;
@@ -1049,6 +1103,12 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
        break;
      }
    }
+
+    // For recovery from NoSpace() error, we can only handle
+    // the case where the database is stored in a single path
+    if (paths.size() <= 1) {
+      impl->error_handler_.EnableAutoRecovery();
+    }
  }

  if (!s.ok()) {
@@ -1073,10 +1133,10 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
        impl->immutable_db_options_.env->OptimizeForLogWrite(
            soptions, BuildDBOptions(impl->immutable_db_options_,
                                     impl->mutable_db_options_));
-    s = NewWritableFile(
-        impl->immutable_db_options_.env,
-        LogFileName(impl->immutable_db_options_.wal_dir, new_log_number),
-        &lfile, opt_env_options);
+    std::string log_fname =
+        LogFileName(impl->immutable_db_options_.wal_dir, new_log_number);
+    s = NewWritableFile(impl->immutable_db_options_.env, log_fname, &lfile,
+                        opt_env_options);
    if (s.ok()) {
      lfile->SetWriteLifeTimeHint(write_hint);
      lfile->SetPreallocationBlockSize(
@@ -1084,8 +1144,8 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
      {
        InstrumentedMutexLock wl(&impl->log_write_mutex_);
        impl->logfile_number_ = new_log_number;
-        unique_ptr<WritableFileWriter> file_writer(
-            new WritableFileWriter(std::move(lfile), opt_env_options));
+        unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+            std::move(lfile), log_fname, opt_env_options));
        impl->logs_.emplace_back(
            new_log_number,
            new log::Writer(
@@ -1212,6 +1272,14 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
        }
      }
    }
+
+    // Reserve some disk buffer space. This is a heuristic - when we run out
+    // of disk space, this ensures that there is atleast write_buffer_size
+    // amount of free space before we resume DB writes. In low disk space
+    // conditions, we want to avoid a lot of small L0 files due to frequent
+    // WAL write failures and resultant forced flushes
+    sfm->ReserveDiskBuffer(max_write_buffer_size,
+                           impl->immutable_db_options_.db_paths[0].path);
  }
 #endif  // !ROCKSDB_LITE

@@ -31,22 +31,38 @@ Status DBImplReadOnly::Get(const ReadOptions& read_options,
                           ColumnFamilyHandle* column_family, const Slice& key,
                           PinnableSlice* pinnable_val) {
  assert(pinnable_val != nullptr);
+  // TODO: stopwatch DB_GET needed?, perf timer needed?
+  PERF_TIMER_GUARD(get_snapshot_time);
  Status s;
  SequenceNumber snapshot = versions_->LastSequence();
  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  auto cfd = cfh->cfd();
+  if (tracer_) {
+    InstrumentedMutexLock lock(&trace_mutex_);
+    if (tracer_) {
+      tracer_->Get(column_family, key);
+    }
+  }
  SuperVersion* super_version = cfd->GetSuperVersion();
  MergeContext merge_context;
  RangeDelAggregator range_del_agg(cfd->internal_comparator(), snapshot);
  LookupKey lkey(key, snapshot);
+  PERF_TIMER_STOP(get_snapshot_time);
  if (super_version->mem->Get(lkey, pinnable_val->GetSelf(), &s, &merge_context,
                              &range_del_agg, read_options)) {
    pinnable_val->PinSelf();
+    RecordTick(stats_, MEMTABLE_HIT);
  } else {
    PERF_TIMER_GUARD(get_from_output_files_time);
    super_version->current->Get(read_options, lkey, pinnable_val, &s,
                                &merge_context, &range_del_agg);
+    RecordTick(stats_, MEMTABLE_MISS);
  }
+  RecordTick(stats_, NUMBER_KEYS_READ);
+  size_t size = pinnable_val->size();
+  RecordTick(stats_, BYTES_READ, size);
+  MeasureTime(stats_, BYTES_PER_READ, size);
+  PERF_COUNTER_ADD(get_read_bytes, size);
  return s;
 }

@@ -12,6 +12,7 @@
 #define __STDC_FORMAT_MACROS
 #endif
 #include <inttypes.h>
+#include "db/error_handler.h"
 #include "db/event_helpers.h"
 #include "monitoring/perf_context_imp.h"
 #include "options/options_helper.h"
@@ -75,6 +76,12 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
  if (my_batch == nullptr) {
    return Status::Corruption("Batch is nullptr!");
  }
+  if (tracer_) {
+    InstrumentedMutexLock lock(&trace_mutex_);
+    if (tracer_) {
+      tracer_->Write(my_batch);
+    }
+  }
  if (write_options.sync && write_options.disableWAL) {
    return Status::InvalidArgument("Sync writes has to enable WAL.");
  }
@@ -310,7 +317,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
        w.status = WriteBatchInternal::InsertInto(
            write_group, current_sequence, column_family_memtables_.get(),
            &flush_scheduler_, write_options.ignore_missing_column_families,
-            0 /*recovery_log_number*/, this, parallel, seq_per_batch_);
+            0 /*recovery_log_number*/, this, parallel, seq_per_batch_,
+            batch_per_txn_);
      } else {
        SequenceNumber next_sequence = current_sequence;
        // Note: the logic for advancing seq here must be consistent with the
@@ -345,7 +353,7 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
              &w, w.sequence, &column_family_memtables, &flush_scheduler_,
              write_options.ignore_missing_column_families, 0 /*log_number*/,
              this, true /*concurrent_memtable_writes*/, seq_per_batch_,
-              w.batch_cnt);
+              w.batch_cnt, batch_per_txn_);
        }
      }
      if (seq_used != nullptr) {
@@ -507,7 +515,8 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
      memtable_write_group.status = WriteBatchInternal::InsertInto(
          memtable_write_group, w.sequence, column_family_memtables_.get(),
          &flush_scheduler_, write_options.ignore_missing_column_families,
-          0 /*log_number*/, this, seq_per_batch_);
+          0 /*log_number*/, this, false /*concurrent_memtable_writes*/,
+          seq_per_batch_, batch_per_txn_);
      versions_->SetLastSequence(memtable_write_group.last_sequence);
      write_thread_.ExitAsMemTableWriter(&w, memtable_write_group);
    }
@@ -564,7 +573,6 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
  }
  // else we are the leader of the write batch group
  assert(w.state == WriteThread::STATE_GROUP_LEADER);
-  WriteContext write_context;
  WriteThread::WriteGroup write_group;
  uint64_t last_sequence;
  nonmem_write_thread_.EnterAsBatchGroupLeader(&w, &write_group);
@@ -676,16 +684,7 @@ void DBImpl::WriteStatusCheck(const Status& status) {
  if (immutable_db_options_.paranoid_checks && !status.ok() &&
      !status.IsBusy() && !status.IsIncomplete()) {
    mutex_.Lock();
-    if (bg_error_.ok()) {
-      Status new_bg_error = status;
-      // may temporarily unlock and lock the mutex.
-      EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                            BackgroundErrorReason::kWriteCallback,
-                                            &new_bg_error, &mutex_);
-      if (!new_bg_error.ok()) {
-        bg_error_ = new_bg_error;  // stop compaction & fail any further writes
-      }
-    }
+    error_handler_.SetBGError(status, BackgroundErrorReason::kWriteCallback);
    mutex_.Unlock();
  }
 }
@@ -698,15 +697,8 @@ void DBImpl::MemTableInsertStatusCheck(const Status& status) {
  // ignore_missing_column_families.
  if (!status.ok()) {
    mutex_.Lock();
-    assert(bg_error_.ok());
-    Status new_bg_error = status;
-    // may temporarily unlock and lock the mutex.
-    EventHelpers::NotifyOnBackgroundError(immutable_db_options_.listeners,
-                                          BackgroundErrorReason::kMemTable,
-                                          &new_bg_error, &mutex_);
-    if (!new_bg_error.ok()) {
-      bg_error_ = new_bg_error;  // stop compaction & fail any further writes
-    }
+    assert(!error_handler_.IsBGWorkStopped());
+    error_handler_.SetBGError(status, BackgroundErrorReason::kMemTable);
    mutex_.Unlock();
  }
 }
@@ -718,6 +710,10 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
  assert(write_context != nullptr && need_log_sync != nullptr);
  Status status;

+  if (error_handler_.IsDBStopped()) {
+    status = error_handler_.GetBGError();
+  }
+
  PERF_TIMER_GUARD(write_scheduling_flushes_compactions_time);

  assert(!single_column_family_mode_ ||
@@ -736,10 +732,6 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
    status = HandleWriteBufferFull(write_context);
  }

-  if (UNLIKELY(status.ok() && !bg_error_.ok())) {
-    status = bg_error_;
-  }
-
  if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
    status = ScheduleFlushes(write_context);
  }
@@ -1072,6 +1064,7 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
                 oldest_alive_log, total_log_size_.load(), GetMaxTotalWalSize());
  // no need to refcount because drop is happening in write thread, so can't
  // happen while we're in the write thread
+  FlushRequest flush_req;
  for (auto cfd : *versions_->GetColumnFamilySet()) {
    if (cfd->IsDropped()) {
      continue;
@@ -1081,11 +1074,14 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
      if (!status.ok()) {
        break;
      }
+      flush_req.emplace_back(cfd, cfd->imm()->GetLatestMemTableID());
      cfd->imm()->FlushRequested();
-      SchedulePendingFlush(cfd, FlushReason::kWriteBufferManager);
    }
  }
-  MaybeScheduleFlushOrCompaction();
+  if (status.ok()) {
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferManager);
+    MaybeScheduleFlushOrCompaction();
+  }
  return status;
 }

@@ -1124,14 +1120,26 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
      }
    }
  }
+
+  autovector<ColumnFamilyData*> cfds;
  if (cfd_picked != nullptr) {
-    status = SwitchMemtable(cfd_picked, write_context,
-                            FlushReason::kWriteBufferFull);
-    if (status.ok()) {
-      cfd_picked->imm()->FlushRequested();
-      SchedulePendingFlush(cfd_picked, FlushReason::kWriteBufferFull);
-      MaybeScheduleFlushOrCompaction();
+    cfds.push_back(cfd_picked);
+  }
+  FlushRequest flush_req;
+  for (const auto cfd : cfds) {
+    cfd->Ref();
+    status = SwitchMemtable(cfd, write_context);
+    cfd->Unref();
+    if (!status.ok()) {
+      break;
    }
+    uint64_t flush_memtable_id = cfd->imm()->GetLatestMemTableID();
+    cfd->imm()->FlushRequested();
+    flush_req.emplace_back(cfd, flush_memtable_id);
+  }
+  if (status.ok()) {
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+    MaybeScheduleFlushOrCompaction();
  }
  return status;
 }
@@ -1154,10 +1162,14 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
    uint64_t delay = write_controller_.GetDelay(env_, num_bytes);
    if (delay > 0) {
      if (write_options.no_slowdown) {
-        return Status::Incomplete();
+        return Status::Incomplete("Write stall");
      }
      TEST_SYNC_POINT("DBImpl::DelayWrite:Sleep");

+      // Notify write_thread_ about the stall so it can setup a barrier and
+      // fail any pending writers with no_slowdown
+      write_thread_.BeginWriteStall();
+      TEST_SYNC_POINT("DBImpl::DelayWrite:BeginWriteStallDone");
      mutex_.Unlock();
      // We will delay the write until we have slept for delay ms or
      // we don't need a delay anymore
@@ -1174,15 +1186,25 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
        env_->SleepForMicroseconds(kDelayInterval);
      }
      mutex_.Lock();
+      write_thread_.EndWriteStall();
    }

-    while (bg_error_.ok() && write_controller_.IsStopped()) {
+    // Don't wait if there's a background error, even if its a soft error. We
+    // might wait here indefinitely as the background compaction may never
+    // finish successfully, resulting in the stall condition lasting
+    // indefinitely
+    while (error_handler_.GetBGError().ok() && write_controller_.IsStopped()) {
      if (write_options.no_slowdown) {
-        return Status::Incomplete();
+        return Status::Incomplete("Write stall");
      }
      delayed = true;
+
+      // Notify write_thread_ about the stall so it can setup a barrier and
+      // fail any pending writers with no_slowdown
+      write_thread_.BeginWriteStall();
      TEST_SYNC_POINT("DBImpl::DelayWrite:Wait");
      bg_cv_.Wait();
+      write_thread_.EndWriteStall();
    }
  }
  assert(!delayed || !write_options.no_slowdown);
@@ -1192,7 +1214,19 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
    RecordTick(stats_, STALL_MICROS, time_delayed);
  }

-  return bg_error_;
+  // If DB is not in read-only mode and write_controller is not stopping
+  // writes, we can ignore any background errors and allow the write to
+  // proceed
+  Status s;
+  if (write_controller_.IsStopped()) {
+    // If writes are still stopped, it means we bailed due to a background
+    // error
+    s = Status::Incomplete(error_handler_.GetBGError().ToString());
+  }
+  if (error_handler_.IsDBStopped()) {
+    s = error_handler_.GetBGError();
+  }
+  return s;
 }

 Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
@@ -1227,16 +1261,28 @@ Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,

 Status DBImpl::ScheduleFlushes(WriteContext* context) {
  ColumnFamilyData* cfd;
+  FlushRequest flush_req;
+  Status status;
  while ((cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
-    auto status = SwitchMemtable(cfd, context, FlushReason::kWriteBufferFull);
+    status = SwitchMemtable(cfd, context);
+    bool should_schedule = true;
    if (cfd->Unref()) {
      delete cfd;
+      should_schedule = false;
    }
    if (!status.ok()) {
-      return status;
+      break;
+    }
+    if (should_schedule) {
+      uint64_t flush_memtable_id = cfd->imm()->GetLatestMemTableID();
+      flush_req.emplace_back(cfd, flush_memtable_id);
    }
  }
-  return Status::OK();
+  if (status.ok()) {
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+    MaybeScheduleFlushOrCompaction();
+  }
+  return status;
 }

 #ifndef ROCKSDB_LITE
@@ -1257,8 +1303,7 @@ void DBImpl::NotifyOnMemTableSealed(ColumnFamilyData* /*cfd*/,

 // REQUIRES: mutex_ is held
 // REQUIRES: this thread is currently at the front of the writer queue
-Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
-                              FlushReason flush_reason) {
+Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
  mutex_.AssertHeld();
  WriteThread::Writer nonmem_w;
  if (two_write_queues_) {
@@ -1281,7 +1326,11 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
  // In case of pipelined write is enabled, wait for all pending memtable
  // writers.
  if (immutable_db_options_.enable_pipelined_write) {
+    // Memtable writers may call DB::Get in case max_successive_merges > 0,
+    // which may lock mutex. Unlocking mutex here to avoid deadlock.
+    mutex_.Unlock();
    write_thread_.WaitForMemTableWriters();
+    mutex_.Lock();
  }

  // Attempt to switch to a new memtable and trigger flush of old.
@@ -1323,6 +1372,8 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
  auto write_hint = CalculateWALWriteHint();
  mutex_.Unlock();
  {
+    std::string log_fname =
+        LogFileName(immutable_db_options_.wal_dir, new_log_number);
    if (creating_new_log) {
      EnvOptions opt_env_opt =
          env_->OptimizeForLogWrite(env_options_, db_options);
@@ -1330,14 +1381,12 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
        ROCKS_LOG_INFO(immutable_db_options_.info_log,
                       "reusing log %" PRIu64 " from recycle list\n",
                       recycle_log_number);
-        s = env_->ReuseWritableFile(
-            LogFileName(immutable_db_options_.wal_dir, new_log_number),
-            LogFileName(immutable_db_options_.wal_dir, recycle_log_number),
-            &lfile, opt_env_opt);
+        std::string old_log_fname =
+            LogFileName(immutable_db_options_.wal_dir, recycle_log_number);
+        s = env_->ReuseWritableFile(log_fname, old_log_fname, &lfile,
+                                    opt_env_opt);
      } else {
-        s = NewWritableFile(
-            env_, LogFileName(immutable_db_options_.wal_dir, new_log_number),
-            &lfile, opt_env_opt);
+        s = NewWritableFile(env_, log_fname, &lfile, opt_env_opt);
      }
      if (s.ok()) {
        // Our final size should be less than write_buffer_size
@@ -1348,7 +1397,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
        lfile->SetPreallocationBlockSize(preallocate_block_size);
        lfile->SetWriteLifeTimeHint(write_hint);
        unique_ptr<WritableFileWriter> file_writer(
-            new WritableFileWriter(std::move(lfile), opt_env_opt));
+            new WritableFileWriter(std::move(lfile), log_fname, opt_env_opt));
        new_log = new log::Writer(
            std::move(file_writer), new_log_number,
            immutable_db_options_.recycle_log_file_num > 0, manual_wal_flush_);
@@ -1426,7 +1475,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
  new_mem->Ref();
  cfd->SetMemtable(new_mem);
  InstallSuperVersionAndScheduleWork(cfd, &context->superversion_context,
-                                     mutable_cf_options, flush_reason);
+                                     mutable_cf_options);
  if (two_write_queues_) {
    nonmem_write_thread_.ExitUnbatched(&nonmem_w);
  }
@@ -27,6 +27,7 @@
 #include "util/logging.h"
 #include "util/mutexlock.h"
 #include "util/string_util.h"
+#include "util/trace_replay.h"

 namespace rocksdb {

@@ -114,7 +115,8 @@ class DBIter final: public Iterator {
         const MutableCFOptions& mutable_cf_options, const Comparator* cmp,
         InternalIterator* iter, SequenceNumber s, bool arena_mode,
         uint64_t max_sequential_skip_in_iterations,
-         ReadCallback* read_callback, bool allow_blob)
+         ReadCallback* read_callback, DBImpl* db_impl, ColumnFamilyData* cfd,
+         bool allow_blob)
      : arena_mode_(arena_mode),
        env_(_env),
        logger_(cf_options.info_log),
@@ -135,6 +137,8 @@ class DBIter final: public Iterator {
        range_del_agg_(cf_options.internal_comparator, s,
                       true /* collapse_deletions */),
        read_callback_(read_callback),
+        db_impl_(db_impl),
+        cfd_(cfd),
        allow_blob_(allow_blob),
        is_blob_(false),
        start_seqnum_(read_options.iter_start_seqnum) {
@@ -260,6 +264,18 @@ class DBIter final: public Iterator {
  bool TooManyInternalKeysSkipped(bool increment = true);
  bool IsVisible(SequenceNumber sequence);

+  // CanReseekToSkip() returns whether the iterator can use the optimization
+  // where it reseek by sequence number to get the next key when there are too
+  // many versions. This is disabled for write unprepared because seeking to
+  // sequence number does not guarantee that it is visible.
+  inline bool CanReseekToSkip();
+
+  // MaxVisibleSequenceNumber() returns the maximum visible sequence number
+  // for this snapshot. This sequence number may be greater than snapshot
+  // seqno because uncommitted data written to DB for write unprepared will
+  // have a higher sequence number.
+  inline SequenceNumber MaxVisibleSequenceNumber();
+
  // Temporarily pin the blocks that we encounter until ReleaseTempPinnedData()
  // is called
  void TempPinData() {
@@ -332,6 +348,8 @@ class DBIter final: public Iterator {
  LocalStatistics local_stats_;
  PinnedIteratorsManager pinned_iters_mgr_;
  ReadCallback* read_callback_;
+  DBImpl* db_impl_;
+  ColumnFamilyData* cfd_;
  bool allow_blob_;
  bool is_blob_;
  // for diff snapshots we want the lower bound on the seqnum;
@@ -506,8 +524,7 @@ bool DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
                  ikey_.user_key,
                  !pin_thru_lifetime_ || !iter_->IsKeyPinned() /* copy */);
              if (range_del_agg_.ShouldDelete(
-                      ikey_, RangeDelAggregator::RangePositioningMode::
-                                 kForwardTraversal)) {
+                      ikey_, RangeDelPositioningMode::kForwardTraversal)) {
                // Arrange to skip all upcoming entries for this key since
                // they are hidden by this deletion.
                skipping = true;
@@ -537,8 +554,7 @@ bool DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
                ikey_.user_key,
                !pin_thru_lifetime_ || !iter_->IsKeyPinned() /* copy */);
            if (range_del_agg_.ShouldDelete(
-                    ikey_, RangeDelAggregator::RangePositioningMode::
-                               kForwardTraversal)) {
+                    ikey_, RangeDelPositioningMode::kForwardTraversal)) {
              // Arrange to skip all upcoming entries for this key since
              // they are hidden by this deletion.
              skipping = true;
@@ -578,7 +594,7 @@ bool DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {

    // If we have sequentially iterated via numerous equal keys, then it's
    // better to seek so that we can avoid too many key comparisons.
-    if (num_skipped > max_skip_) {
+    if (num_skipped > max_skip_ && CanReseekToSkip()) {
      num_skipped = 0;
      std::string last_key;
      if (skipping) {
@@ -645,8 +661,7 @@ bool DBIter::MergeValuesNewToOld() {
      break;
    } else if (kTypeDeletion == ikey.type || kTypeSingleDeletion == ikey.type ||
               range_del_agg_.ShouldDelete(
-                   ikey, RangeDelAggregator::RangePositioningMode::
-                             kForwardTraversal)) {
+                   ikey, RangeDelPositioningMode::kForwardTraversal)) {
      // hit a delete with the same user key, stop right here
      // iter_ is positioned after delete
      iter_->Next();
@@ -895,7 +910,7 @@ bool DBIter::FindValueForCurrentKey() {
    // This user key has lots of entries.
    // We're going from old to new, and it's taking too long. Let's do a Seek()
    // and go from new to old. This helps when a key was overwritten many times.
-    if (num_skipped >= max_skip_) {
+    if (num_skipped >= max_skip_ && CanReseekToSkip()) {
      return FindValueForCurrentKeyUsingSeek();
    }

@@ -904,8 +919,7 @@ bool DBIter::FindValueForCurrentKey() {
      case kTypeValue:
      case kTypeBlobIndex:
        if (range_del_agg_.ShouldDelete(
-                ikey,
-                RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+                ikey, RangeDelPositioningMode::kBackwardTraversal)) {
          last_key_entry_type = kTypeRangeDeletion;
          PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
        } else {
@@ -923,8 +937,7 @@ bool DBIter::FindValueForCurrentKey() {
        break;
      case kTypeMerge:
        if (range_del_agg_.ShouldDelete(
-                ikey,
-                RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+                ikey, RangeDelPositioningMode::kBackwardTraversal)) {
          merge_context_.Clear();
          last_key_entry_type = kTypeRangeDeletion;
          last_not_merge_type = last_key_entry_type;
@@ -1057,7 +1070,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {

  if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
      range_del_agg_.ShouldDelete(
-          ikey, RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+          ikey, RangeDelPositioningMode::kBackwardTraversal)) {
    valid_ = false;
    return true;
  }
@@ -1102,8 +1115,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {

    if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
        range_del_agg_.ShouldDelete(
-            ikey,
-            RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+            ikey, RangeDelPositioningMode::kBackwardTraversal)) {
      break;
    } else if (ikey.type == kTypeValue) {
      const Slice val = iter_->value();
@@ -1194,7 +1206,7 @@ bool DBIter::FindUserKeyBeforeSavedKey() {
      PERF_COUNTER_ADD(internal_key_skipped_count, 1);
    }

-    if (num_skipped >= max_skip_) {
+    if (num_skipped >= max_skip_ && CanReseekToSkip()) {
      num_skipped = 0;
      IterKey last_key;
      last_key.SetInternalKey(ParsedInternalKey(
@@ -1234,8 +1246,21 @@ bool DBIter::TooManyInternalKeysSkipped(bool increment) {
 }

 bool DBIter::IsVisible(SequenceNumber sequence) {
-  return sequence <= sequence_ &&
-         (read_callback_ == nullptr || read_callback_->IsCommitted(sequence));
+  return sequence <= MaxVisibleSequenceNumber() &&
+         (read_callback_ == nullptr || read_callback_->IsVisible(sequence));
+}
+
+bool DBIter::CanReseekToSkip() {
+  return read_callback_ == nullptr ||
+         read_callback_->MaxUnpreparedSequenceNumber() == 0;
+}
+
+SequenceNumber DBIter::MaxVisibleSequenceNumber() {
+  if (read_callback_ == nullptr) {
+    return sequence_;
+  }
+
+  return std::max(sequence_, read_callback_->MaxUnpreparedSequenceNumber());
 }

 void DBIter::Seek(const Slice& target) {
@@ -1243,20 +1268,28 @@ void DBIter::Seek(const Slice& target) {
  status_ = Status::OK();
  ReleaseTempPinnedData();
  ResetInternalKeysSkippedCounter();
+
+  SequenceNumber seq = MaxVisibleSequenceNumber();
  saved_key_.Clear();
-  saved_key_.SetInternalKey(target, sequence_);
+  saved_key_.SetInternalKey(target, seq);
+
+#ifndef ROCKSDB_LITE
+  if (db_impl_ != nullptr && cfd_ != nullptr) {
+    db_impl_->TraceIteratorSeek(cfd_->GetID(), target);
+  }
+#endif  // ROCKSDB_LITE

  if (iterate_lower_bound_ != nullptr &&
      user_comparator_->Compare(saved_key_.GetUserKey(),
                                *iterate_lower_bound_) < 0) {
    saved_key_.Clear();
-    saved_key_.SetInternalKey(*iterate_lower_bound_, sequence_);
+    saved_key_.SetInternalKey(*iterate_lower_bound_, seq);
  }

  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->Seek(saved_key_.GetInternalKey());
-    range_del_agg_.InvalidateTombstoneMapPositions();
+    range_del_agg_.InvalidateRangeDelMapPositions();
  }
  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
@@ -1307,9 +1340,15 @@ void DBIter::SeekForPrev(const Slice& target) {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekForPrev(saved_key_.GetInternalKey());
-    range_del_agg_.InvalidateTombstoneMapPositions();
+    range_del_agg_.InvalidateRangeDelMapPositions();
  }

+#ifndef ROCKSDB_LITE
+  if (db_impl_ != nullptr && cfd_ != nullptr) {
+    db_impl_->TraceIteratorSeekForPrev(cfd_->GetID(), target);
+  }
+#endif  // ROCKSDB_LITE
+
  RecordTick(statistics_, NUMBER_DB_SEEK);
  if (iter_->Valid()) {
    if (prefix_extractor_ && prefix_same_as_start_) {
@@ -1356,7 +1395,7 @@ void DBIter::SeekToFirst() {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekToFirst();
-    range_del_agg_.InvalidateTombstoneMapPositions();
+    range_del_agg_.InvalidateRangeDelMapPositions();
  }

  RecordTick(statistics_, NUMBER_DB_SEEK);
@@ -1407,7 +1446,7 @@ void DBIter::SeekToLast() {
  {
    PERF_TIMER_GUARD(seek_internal_seek_time);
    iter_->SeekToLast();
-    range_del_agg_.InvalidateTombstoneMapPositions();
+    range_del_agg_.InvalidateRangeDelMapPositions();
  }
  PrevInternal();
  if (statistics_ != nullptr) {
@@ -1432,11 +1471,12 @@ Iterator* NewDBIterator(Env* env, const ReadOptions& read_options,
                        InternalIterator* internal_iter,
                        const SequenceNumber& sequence,
                        uint64_t max_sequential_skip_in_iterations,
-                        ReadCallback* read_callback, bool allow_blob) {
-  DBIter* db_iter =
-      new DBIter(env, read_options, cf_options, mutable_cf_options,
-                 user_key_comparator, internal_iter, sequence, false,
-                 max_sequential_skip_in_iterations, read_callback, allow_blob);
+                        ReadCallback* read_callback, DBImpl* db_impl,
+                        ColumnFamilyData* cfd, bool allow_blob) {
+  DBIter* db_iter = new DBIter(
+      env, read_options, cf_options, mutable_cf_options, user_key_comparator,
+      internal_iter, sequence, false, max_sequential_skip_in_iterations,
+      read_callback, db_impl, cfd, allow_blob);
  return db_iter;
 }

@@ -1483,13 +1523,14 @@ void ArenaWrappedDBIter::Init(Env* env, const ReadOptions& read_options,
                              const SequenceNumber& sequence,
                              uint64_t max_sequential_skip_in_iteration,
                              uint64_t version_number,
-                              ReadCallback* read_callback, bool allow_blob,
+                              ReadCallback* read_callback, DBImpl* db_impl,
+                              ColumnFamilyData* cfd, bool allow_blob,
                              bool allow_refresh) {
  auto mem = arena_.AllocateAligned(sizeof(DBIter));
-  db_iter_ = new (mem)
-      DBIter(env, read_options, cf_options, mutable_cf_options,
-             cf_options.user_comparator, nullptr, sequence, true,
-             max_sequential_skip_in_iteration, read_callback, allow_blob);
+  db_iter_ = new (mem) DBIter(env, read_options, cf_options, mutable_cf_options,
+                              cf_options.user_comparator, nullptr, sequence,
+                              true, max_sequential_skip_in_iteration,
+                              read_callback, db_impl, cfd, allow_blob);
  sv_number_ = version_number;
  allow_refresh_ = allow_refresh;
 }
@@ -1513,7 +1554,8 @@ Status ArenaWrappedDBIter::Refresh() {
    SuperVersion* sv = cfd_->GetReferencedSuperVersion(db_impl_->mutex());
    Init(env, read_options_, *(cfd_->ioptions()), sv->mutable_cf_options,
         latest_seq, sv->mutable_cf_options.max_sequential_skip_in_iterations,
-         cur_sv_number, read_callback_, allow_blob_, allow_refresh_);
+         cur_sv_number, read_callback_, db_impl_, cfd_, allow_blob_,
+         allow_refresh_);

    InternalIterator* internal_iter = db_impl_->NewInternalIterator(
        read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator());
@@ -1535,7 +1577,7 @@ ArenaWrappedDBIter* NewArenaWrappedDbIterator(
  ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
  iter->Init(env, read_options, cf_options, mutable_cf_options, sequence,
             max_sequential_skip_in_iterations, version_number, read_callback,
-             allow_blob, allow_refresh);
+             db_impl, cfd, allow_blob, allow_refresh);
  if (db_impl != nullptr && cfd != nullptr && allow_refresh) {
    iter->StoreRefreshInfo(read_options, db_impl, cfd, read_callback,
                           allow_blob);
@@ -23,20 +23,18 @@ namespace rocksdb {

 class Arena;
 class DBIter;
-class InternalIterator;

 // Return a new iterator that converts internal keys (yielded by
 // "*internal_iter") that were live at the specified "sequence" number
 // into appropriate user keys.
-extern Iterator* NewDBIterator(Env* env, const ReadOptions& read_options,
-                               const ImmutableCFOptions& cf_options,
-                               const MutableCFOptions& mutable_cf_options,
-                               const Comparator* user_key_comparator,
-                               InternalIterator* internal_iter,
-                               const SequenceNumber& sequence,
-                               uint64_t max_sequential_skip_in_iterations,
-                               ReadCallback* read_callback,
-                               bool allow_blob = false);
+extern Iterator* NewDBIterator(
+    Env* env, const ReadOptions& read_options,
+    const ImmutableCFOptions& cf_options,
+    const MutableCFOptions& mutable_cf_options,
+    const Comparator* user_key_comparator, InternalIterator* internal_iter,
+    const SequenceNumber& sequence, uint64_t max_sequential_skip_in_iterations,
+    ReadCallback* read_callback, DBImpl* db_impl = nullptr,
+    ColumnFamilyData* cfd = nullptr, bool allow_blob = false);

 // A wrapper iterator which wraps DB Iterator and the arena, with which the DB
 // iterator is supposed be allocated. This class is used as an entry point of
@@ -75,7 +73,8 @@ class ArenaWrappedDBIter : public Iterator {
            const MutableCFOptions& mutable_cf_options,
            const SequenceNumber& sequence,
            uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
-            ReadCallback* read_callback, bool allow_blob, bool allow_refresh);
+            ReadCallback* read_callback, DBImpl* db_impl, ColumnFamilyData* cfd,
+            bool allow_blob, bool allow_refresh);

  void StoreRefreshInfo(const ReadOptions& read_options, DBImpl* db_impl,
                        ColumnFamilyData* cfd, ReadCallback* read_callback,
@@ -36,7 +36,9 @@ class TestIterator : public InternalIterator {
        valid_(false),
        sequence_number_(0),
        iter_(0),
-        cmp(comparator) {}
+        cmp(comparator) {
+    data_.reserve(16);
+  }

  void AddPut(std::string argkey, std::string argvalue) {
    Add(argkey, kTypeValue, argvalue);
@@ -2660,7 +2662,7 @@ TEST_F(DBIterWithMergeIterTest, InnerMergeIteratorDataRace1) {
  // MergeIterator::Prev() realized the mem table iterator is at its end
  // and before an SeekToLast() is called.
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "MergeIterator::Prev:BeforeSeekToLast",
+      "MergeIterator::Prev:BeforePrev",
      [&](void* /*arg*/) { internal_iter2_->Add("z", kTypeValue, "7", 12u); });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

@@ -2696,7 +2698,7 @@ TEST_F(DBIterWithMergeIterTest, InnerMergeIteratorDataRace2) {
  // mem table after MergeIterator::Prev() realized the mem tableiterator is at
  // its end and before an SeekToLast() is called.
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "MergeIterator::Prev:BeforeSeekToLast", [&](void* /*arg*/) {
+      "MergeIterator::Prev:BeforePrev", [&](void* /*arg*/) {
        internal_iter2_->Add("z", kTypeValue, "7", 12u);
        internal_iter2_->Add("z", kTypeValue, "7", 11u);
      });
@@ -2734,7 +2736,7 @@ TEST_F(DBIterWithMergeIterTest, InnerMergeIteratorDataRace3) {
  // mem table after MergeIterator::Prev() realized the mem table iterator is at
  // its end and before an SeekToLast() is called.
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "MergeIterator::Prev:BeforeSeekToLast", [&](void* /*arg*/) {
+      "MergeIterator::Prev:BeforePrev", [&](void* /*arg*/) {
        internal_iter2_->Add("z", kTypeValue, "7", 16u, true);
        internal_iter2_->Add("z", kTypeValue, "7", 15u, true);
        internal_iter2_->Add("z", kTypeValue, "7", 14u, true);
@@ -20,7 +20,7 @@ namespace rocksdb {

 // A dumb ReadCallback which saying every key is committed.
 class DummyReadCallback : public ReadCallback {
-  bool IsCommitted(SequenceNumber /*seq*/) { return true; }
+  bool IsVisible(SequenceNumber /*seq*/) override { return true; }
 };

 // Test param:
@@ -184,73 +184,6 @@ TEST_P(DBIteratorTest, NonBlockingIteration) {
                         kSkipMmapReads));
 }

-#ifndef ROCKSDB_LITE
-TEST_P(DBIteratorTest, ManagedNonBlockingIteration) {
-  do {
-    ReadOptions non_blocking_opts, regular_opts;
-    Options options = CurrentOptions();
-    options.statistics = rocksdb::CreateDBStatistics();
-    non_blocking_opts.read_tier = kBlockCacheTier;
-    non_blocking_opts.managed = true;
-    CreateAndReopenWithCF({"pikachu"}, options);
-    // write one kv to the database.
-    ASSERT_OK(Put(1, "a", "b"));
-
-    // scan using non-blocking iterator. We should find it because
-    // it is in memtable.
-    Iterator* iter = NewIterator(non_blocking_opts, handles_[1]);
-    int count = 0;
-    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-      ASSERT_OK(iter->status());
-      count++;
-    }
-    ASSERT_EQ(count, 1);
-    delete iter;
-
-    // flush memtable to storage. Now, the key should not be in the
-    // memtable neither in the block cache.
-    ASSERT_OK(Flush(1));
-
-    // verify that a non-blocking iterator does not find any
-    // kvs. Neither does it do any IOs to storage.
-    int64_t numopen = TestGetTickerCount(options, NO_FILE_OPENS);
-    int64_t cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = NewIterator(non_blocking_opts, handles_[1]);
-    count = 0;
-    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-      count++;
-    }
-    ASSERT_EQ(count, 0);
-    ASSERT_TRUE(iter->status().IsIncomplete());
-    ASSERT_EQ(numopen, TestGetTickerCount(options, NO_FILE_OPENS));
-    ASSERT_EQ(cache_added, TestGetTickerCount(options, BLOCK_CACHE_ADD));
-    delete iter;
-
-    // read in the specified block via a regular get
-    ASSERT_EQ(Get(1, "a"), "b");
-
-    // verify that we can find it via a non-blocking scan
-    numopen = TestGetTickerCount(options, NO_FILE_OPENS);
-    cache_added = TestGetTickerCount(options, BLOCK_CACHE_ADD);
-    iter = NewIterator(non_blocking_opts, handles_[1]);
-    count = 0;
-    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-      ASSERT_OK(iter->status());
-      count++;
-    }
-    ASSERT_EQ(count, 1);
-    ASSERT_EQ(numopen, TestGetTickerCount(options, NO_FILE_OPENS));
-    ASSERT_EQ(cache_added, TestGetTickerCount(options, BLOCK_CACHE_ADD));
-    delete iter;
-
-    // This test verifies block cache behaviors, which is not used by plain
-    // table format.
-    // Exclude kHashCuckoo as it does not support iteration currently
-  } while (ChangeOptions(kSkipPlainTable | kSkipNoSeekToLast | kSkipHashCuckoo |
-                         kSkipMmapReads));
-}
-#endif  // ROCKSDB_LITE
-
 TEST_P(DBIteratorTest, IterSeekBeforePrev) {
  ASSERT_OK(Put("a", "b"));
  ASSERT_OK(Put("c", "d"));
@@ -1035,6 +968,60 @@ TEST_P(DBIteratorTest, DBIteratorBoundTest) {
    ASSERT_EQ(static_cast<int>(get_perf_context()->internal_delete_skipped_count), 0);
  }
 }
+
+TEST_P(DBIteratorTest, DBIteratorBoundMultiSeek) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.prefix_extractor = nullptr;
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("a", "0"));
+  ASSERT_OK(Put("z", "0"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("foo1", "bar1"));
+  ASSERT_OK(Put("foo2", "bar2"));
+  ASSERT_OK(Put("foo3", "bar3"));
+  ASSERT_OK(Put("foo4", "bar4"));
+
+  {
+    std::string up_str = "foo5";
+    Slice up(up_str);
+    ReadOptions ro;
+    ro.iterate_upper_bound = &up;
+    std::unique_ptr<Iterator> iter(NewIterator(ro));
+
+    iter->Seek("foo1");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo1")), 0);
+
+    uint64_t prev_block_cache_hit =
+        TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    uint64_t prev_block_cache_miss =
+        TestGetTickerCount(options, BLOCK_CACHE_MISS);
+
+    ASSERT_GT(prev_block_cache_hit + prev_block_cache_miss, 0);
+
+    iter->Seek("foo4");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo4")), 0);
+    ASSERT_EQ(prev_block_cache_hit,
+              TestGetTickerCount(options, BLOCK_CACHE_HIT));
+    ASSERT_EQ(prev_block_cache_miss,
+              TestGetTickerCount(options, BLOCK_CACHE_MISS));
+
+    iter->Seek("foo2");
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo2")), 0);
+    iter->Next();
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(iter->key().compare(Slice("foo3")), 0);
+    ASSERT_EQ(prev_block_cache_hit,
+              TestGetTickerCount(options, BLOCK_CACHE_HIT));
+    ASSERT_EQ(prev_block_cache_miss,
+              TestGetTickerCount(options, BLOCK_CACHE_MISS));
+  }
+}
 #endif

 TEST_P(DBIteratorTest, DBIteratorBoundOptimizationTest) {
@@ -1483,6 +1470,26 @@ TEST_P(DBIteratorTest, PinnedDataIteratorReadAfterUpdate) {
  delete iter;
 }

+class SliceTransformLimitedDomainGeneric : public SliceTransform {
+  const char* Name() const override {
+    return "SliceTransformLimitedDomainGeneric";
+  }
+
+  Slice Transform(const Slice& src) const override {
+    return Slice(src.data(), 1);
+  }
+
+  bool InDomain(const Slice& src) const override {
+    // prefix will be x????
+    return src.size() >= 1;
+  }
+
+  bool InRange(const Slice& dst) const override {
+    // prefix will be x????
+    return dst.size() == 1;
+  }
+};
+
 TEST_P(DBIteratorTest, IterSeekForPrevCrossingFiles) {
  Options options = CurrentOptions();
  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
@@ -1537,6 +1544,61 @@ TEST_P(DBIteratorTest, IterSeekForPrevCrossingFiles) {
  }
 }

+TEST_P(DBIteratorTest, IterSeekForPrevCrossingFilesCustomPrefixExtractor) {
+  Options options = CurrentOptions();
+  options.prefix_extractor =
+      std::make_shared<SliceTransformLimitedDomainGeneric>();
+  options.disable_auto_compactions = true;
+  // Enable prefix bloom for SST files
+  BlockBasedTableOptions table_options;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put("a1", "va1"));
+  ASSERT_OK(Put("a2", "va2"));
+  ASSERT_OK(Put("a3", "va3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b1", "vb1"));
+  ASSERT_OK(Put("b2", "vb2"));
+  ASSERT_OK(Put("b3", "vb3"));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put("b4", "vb4"));
+  ASSERT_OK(Put("d1", "vd1"));
+  ASSERT_OK(Put("d2", "vd2"));
+  ASSERT_OK(Put("d4", "vd4"));
+  ASSERT_OK(Flush());
+
+  MoveFilesToLevel(1);
+  {
+    ReadOptions ro;
+    Iterator* iter = NewIterator(ro);
+
+    iter->SeekForPrev("a4");
+    ASSERT_EQ(iter->key().ToString(), "a3");
+    ASSERT_EQ(iter->value().ToString(), "va3");
+
+    iter->SeekForPrev("c2");
+    ASSERT_EQ(iter->key().ToString(), "b3");
+    iter->SeekForPrev("d3");
+    ASSERT_EQ(iter->key().ToString(), "d2");
+    iter->SeekForPrev("b5");
+    ASSERT_EQ(iter->key().ToString(), "b4");
+    delete iter;
+  }
+
+  {
+    ReadOptions ro;
+    ro.prefix_same_as_start = true;
+    Iterator* iter = NewIterator(ro);
+    iter->SeekForPrev("c2");
+    ASSERT_TRUE(!iter->Valid());
+    delete iter;
+  }
+}
+
 TEST_P(DBIteratorTest, IterPrevKeyCrossingBlocks) {
  Options options = CurrentOptions();
  BlockBasedTableOptions table_options;
@@ -2031,6 +2093,34 @@ TEST_P(DBIteratorTest, Refresh) {
  iter.reset();
 }

+TEST_P(DBIteratorTest, RefreshWithSnapshot) {
+  ASSERT_OK(Put("x", "y"));
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ReadOptions options;
+  options.snapshot = snapshot;
+  Iterator* iter = NewIterator(options);
+
+  iter->Seek(Slice("a"));
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().compare(Slice("x")), 0);
+  iter->Next();
+  ASSERT_FALSE(iter->Valid());
+
+  ASSERT_OK(Put("c", "d"));
+
+  iter->Seek(Slice("a"));
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(iter->key().compare(Slice("x")), 0);
+  iter->Next();
+  ASSERT_FALSE(iter->Valid());
+
+  Status s;
+  s = iter->Refresh();
+  ASSERT_TRUE(s.IsNotSupported());
+  db_->ReleaseSnapshot(snapshot);
+  delete iter;
+}
+
 TEST_P(DBIteratorTest, CreationFailure) {
  SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::NewInternalIterator:StatusCallback", [](void* arg) {
@@ -2044,6 +2134,43 @@ TEST_P(DBIteratorTest, CreationFailure) {
  delete iter;
 }

+TEST_P(DBIteratorTest, UpperBoundWithChangeDirection) {
+  Options options = CurrentOptions();
+  options.max_sequential_skip_in_iterations = 3;
+  DestroyAndReopen(options);
+
+  // write a bunch of kvs to the database.
+  ASSERT_OK(Put("a", "1"));
+  ASSERT_OK(Put("y", "1"));
+  ASSERT_OK(Put("y1", "1"));
+  ASSERT_OK(Put("y2", "1"));
+  ASSERT_OK(Put("y3", "1"));
+  ASSERT_OK(Put("z", "1"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("a", "1"));
+  ASSERT_OK(Put("z", "1"));
+  ASSERT_OK(Put("bar", "1"));
+  ASSERT_OK(Put("foo", "1"));
+
+  std::string upper_bound = "x";
+  Slice ub_slice(upper_bound);
+  ReadOptions ro;
+  ro.iterate_upper_bound = &ub_slice;
+  ro.max_skippable_internal_keys = 1000;
+
+  Iterator* iter = NewIterator(ro);
+  iter->Seek("foo");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("foo", iter->key().ToString());
+
+  iter->Prev();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_OK(iter->status());
+  ASSERT_EQ("bar", iter->key().ToString());
+
+  delete iter;
+}
+
 TEST_P(DBIteratorTest, TableFilter) {
  ASSERT_OK(Put("a", "1"));
  dbfull()->Flush(FlushOptions());
@@ -2110,6 +2237,47 @@ TEST_P(DBIteratorTest, TableFilter) {
  }
 }

+TEST_P(DBIteratorTest, UpperBoundWithPrevReseek) {
+  Options options = CurrentOptions();
+  options.max_sequential_skip_in_iterations = 3;
+  DestroyAndReopen(options);
+
+  // write a bunch of kvs to the database.
+  ASSERT_OK(Put("a", "1"));
+  ASSERT_OK(Put("y", "1"));
+  ASSERT_OK(Put("z", "1"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("a", "1"));
+  ASSERT_OK(Put("z", "1"));
+  ASSERT_OK(Put("bar", "1"));
+  ASSERT_OK(Put("foo", "1"));
+  ASSERT_OK(Put("foo", "2"));
+
+  ASSERT_OK(Put("foo", "3"));
+  ASSERT_OK(Put("foo", "4"));
+  ASSERT_OK(Put("foo", "5"));
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(Put("foo", "6"));
+
+  std::string upper_bound = "x";
+  Slice ub_slice(upper_bound);
+  ReadOptions ro;
+  ro.snapshot = snapshot;
+  ro.iterate_upper_bound = &ub_slice;
+
+  Iterator* iter = NewIterator(ro);
+  iter->SeekForPrev("goo");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("foo", iter->key().ToString());
+  iter->Prev();
+
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("bar", iter->key().ToString());
+
+  delete iter;
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_P(DBIteratorTest, SkipStatistics) {
  Options options = CurrentOptions();
  options.statistics = rocksdb::CreateDBStatistics();
@@ -2273,6 +2441,30 @@ TEST_P(DBIteratorTest, NonBlockingIterationBugRepro) {
  EXPECT_TRUE(iter->status().IsIncomplete());
 }

+TEST_P(DBIteratorTest, SeekBackwardAfterOutOfUpperBound) {
+  Put("a", "");
+  Put("b", "");
+  Flush();
+
+  ReadOptions ropt;
+  Slice ub = "b";
+  ropt.iterate_upper_bound = &ub;
+
+  std::unique_ptr<Iterator> it(dbfull()->NewIterator(ropt));
+  it->SeekForPrev("a");
+  ASSERT_TRUE(it->Valid());
+  ASSERT_OK(it->status());
+  ASSERT_EQ("a", it->key().ToString());
+  it->Next();
+  ASSERT_FALSE(it->Valid());
+  ASSERT_OK(it->status());
+  it->SeekForPrev("a");
+  ASSERT_OK(it->status());
+
+  ASSERT_TRUE(it->Valid());
+  ASSERT_EQ("a", it->key().ToString());
+}
+
 INSTANTIATE_TEST_CASE_P(DBIteratorTestInstance, DBIteratorTest,
                        testing::Values(true, false));

@@ -2285,7 +2477,7 @@ TEST_F(DBIteratorWithReadCallbackTest, ReadCallback) {
    explicit TestReadCallback(SequenceNumber last_visible_seq)
        : last_visible_seq_(last_visible_seq) {}

-    bool IsCommitted(SequenceNumber seq) override {
+    bool IsVisible(SequenceNumber seq) override {
      return seq <= last_visible_seq_;
    }

@@ -20,7 +20,7 @@ class TestReadCallback : public ReadCallback {
                   SequenceNumber snapshot_seq)
      : snapshot_checker_(snapshot_checker), snapshot_seq_(snapshot_seq) {}

-  bool IsCommitted(SequenceNumber seq) override {
+  bool IsVisible(SequenceNumber seq) override {
    return snapshot_checker_->IsInSnapshot(seq, snapshot_seq_);
  }

@@ -170,6 +170,7 @@ void ResetTableProperties(TableProperties* tp) {
  tp->raw_value_size = 0;
  tp->num_data_blocks = 0;
  tp->num_entries = 0;
+  tp->num_range_deletions = 0;
 }

 void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
@@ -177,17 +178,18 @@ void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
  std::replace(tp_string.begin(), tp_string.end(), ';', ' ');
  std::replace(tp_string.begin(), tp_string.end(), '=', ' ');
  ResetTableProperties(tp);
-
  sscanf(tp_string.c_str(),
-         "# data blocks %" SCNu64 " # entries %" SCNu64 " raw key size %" SCNu64
+         "# data blocks %" SCNu64 " # entries %" SCNu64
+         " # range deletions %" SCNu64 " raw key size %" SCNu64
         " raw average key size %lf "
         " raw value size %" SCNu64
         " raw average value size %lf "
-         " data block size %" SCNu64 " index block size %" SCNu64
-         " filter block size %" SCNu64,
-         &tp->num_data_blocks, &tp->num_entries, &tp->raw_key_size,
-         &dummy_double, &tp->raw_value_size, &dummy_double, &tp->data_size,
-         &tp->index_size, &tp->filter_size);
+         " data block size %" SCNu64 " index block size (user-key? %" SCNu64
+         ", delta-value? %" SCNu64 ") %" SCNu64 " filter block size %" SCNu64,
+         &tp->num_data_blocks, &tp->num_entries, &tp->num_range_deletions,
+         &tp->raw_key_size, &dummy_double, &tp->raw_value_size, &dummy_double,
+         &tp->data_size, &tp->index_key_is_user_key,
+         &tp->index_value_is_delta_encoded, &tp->index_size, &tp->filter_size);
 }

 void VerifySimilar(uint64_t a, uint64_t b, double bias) {
@@ -218,19 +220,22 @@ void VerifyTableProperties(const TableProperties& base_tp,
  ASSERT_EQ(base_tp.raw_key_size, new_tp.raw_key_size);
  ASSERT_EQ(base_tp.raw_value_size, new_tp.raw_value_size);
  ASSERT_EQ(base_tp.num_entries, new_tp.num_entries);
+  ASSERT_EQ(base_tp.num_range_deletions, new_tp.num_range_deletions);
 }

-void GetExpectedTableProperties(TableProperties* expected_tp,
-                                const int kKeySize, const int kValueSize,
-                                const int kKeysPerTable, const int kTableCount,
-                                const int kBloomBitsPerKey,
-                                const size_t kBlockSize) {
+void GetExpectedTableProperties(
+    TableProperties* expected_tp, const int kKeySize, const int kValueSize,
+    const int kKeysPerTable, const int kRangeDeletionsPerTable,
+    const int kTableCount, const int kBloomBitsPerKey, const size_t kBlockSize,
+    const bool index_key_is_user_key, const bool value_delta_encoding) {
  const int kKeyCount = kTableCount * kKeysPerTable;
+  const int kRangeDeletionCount = kTableCount * kRangeDeletionsPerTable;
  const int kAvgSuccessorSize = kKeySize / 5;
  const int kEncodingSavePerKey = kKeySize / 4;
-  expected_tp->raw_key_size = kKeyCount * (kKeySize + 8);
-  expected_tp->raw_value_size = kKeyCount * kValueSize;
+  expected_tp->raw_key_size = (kKeyCount + kRangeDeletionCount) * (kKeySize + 8);
+  expected_tp->raw_value_size = (kKeyCount + kRangeDeletionCount) * kValueSize;
  expected_tp->num_entries = kKeyCount;
+  expected_tp->num_range_deletions = kRangeDeletionCount;
  expected_tp->num_data_blocks =
      kTableCount *
      (kKeysPerTable * (kKeySize - kEncodingSavePerKey + kValueSize)) /
@@ -238,7 +243,10 @@ void GetExpectedTableProperties(TableProperties* expected_tp,
  expected_tp->data_size =
      kTableCount * (kKeysPerTable * (kKeySize + 8 + kValueSize));
  expected_tp->index_size =
-      expected_tp->num_data_blocks * (kAvgSuccessorSize + 8);
+      expected_tp->num_data_blocks *
+      (kAvgSuccessorSize + (index_key_is_user_key ? 0 : 8) -
+       // discount 1 byte as value size is not encoded in value delta encoding
+       (value_delta_encoding ? 1 : 0));
  expected_tp->filter_size =
      kTableCount * (kKeysPerTable * kBloomBitsPerKey / 8);
 }
@@ -251,8 +259,11 @@ TEST_F(DBPropertiesTest, ValidatePropertyInfo) {
    ASSERT_TRUE(ppt_name_and_info.first.empty() ||
                !isdigit(ppt_name_and_info.first.back()));

-    ASSERT_TRUE((ppt_name_and_info.second.handle_string == nullptr) !=
-                (ppt_name_and_info.second.handle_int == nullptr));
+    int count = 0;
+    count += (ppt_name_and_info.second.handle_string == nullptr) ? 0 : 1;
+    count += (ppt_name_and_info.second.handle_int == nullptr) ? 0 : 1;
+    count += (ppt_name_and_info.second.handle_string_dbimpl == nullptr) ? 0 : 1;
+    ASSERT_TRUE(count == 1);
  }
 }

@@ -287,6 +298,7 @@ TEST_F(DBPropertiesTest, ValidateSampleNumber) {

 TEST_F(DBPropertiesTest, AggregatedTableProperties) {
  for (int kTableCount = 40; kTableCount <= 100; kTableCount += 30) {
+    const int kRangeDeletionsPerTable = 5;
    const int kKeysPerTable = 100;
    const int kKeySize = 80;
    const int kValueSize = 200;
@@ -305,24 +317,37 @@ TEST_F(DBPropertiesTest, AggregatedTableProperties) {

    DestroyAndReopen(options);

+    // Hold open a snapshot to prevent range tombstones from being compacted
+    // away.
+    ManagedSnapshot snapshot(db_);
+
    Random rnd(5632);
    for (int table = 1; table <= kTableCount; ++table) {
      for (int i = 0; i < kKeysPerTable; ++i) {
        db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
                 RandomString(&rnd, kValueSize));
      }
+      for (int i = 0; i < kRangeDeletionsPerTable; i++) {
+        std::string start = RandomString(&rnd, kKeySize);
+        std::string end = start;
+        end.resize(kValueSize);
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), start, end);
+      }
      db_->Flush(FlushOptions());
    }
    std::string property;
    db_->GetProperty(DB::Properties::kAggregatedTableProperties, &property);
+    TableProperties output_tp;
+    ParseTablePropertiesString(property, &output_tp);
+    bool index_key_is_user_key = output_tp.index_key_is_user_key > 0;
+    bool value_is_delta_encoded = output_tp.index_value_is_delta_encoded > 0;

    TableProperties expected_tp;
    GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                               kKeysPerTable, kTableCount, kBloomBitsPerKey,
-                               table_options.block_size);
-
-    TableProperties output_tp;
-    ParseTablePropertiesString(property, &output_tp);
+                               kKeysPerTable, kRangeDeletionsPerTable,
+                               kTableCount, kBloomBitsPerKey,
+                               table_options.block_size, index_key_is_user_key,
+                               value_is_delta_encoded);

    VerifyTableProperties(expected_tp, output_tp);
  }
@@ -371,6 +396,13 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
  for (int key = 0; key < key_index; key++) {
    Get(Key(key));
  }
+
+  // Test for getting immutable_db_options_.statistics
+  ASSERT_TRUE(dbfull()->GetProperty(dbfull()->DefaultColumnFamily(),
+                                    "rocksdb.options-statistics", &prop));
+  ASSERT_NE(std::string::npos, prop.find("rocksdb.block.cache.miss"));
+  ASSERT_EQ(std::string::npos, prop.find("rocksdb.db.f.micros"));
+
  ASSERT_TRUE(dbfull()->GetProperty(dbfull()->DefaultColumnFamily(),
                                    "rocksdb.cf-file-histogram", &prop));
  ASSERT_NE(std::string::npos, prop.find("** Level 0 read latency histogram"));
@@ -437,6 +469,7 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {

 TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
  const int kTableCount = 100;
+  const int kRangeDeletionsPerTable = 2;
  const int kKeysPerTable = 10;
  const int kKeySize = 50;
  const int kValueSize = 400;
@@ -462,6 +495,9 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {

  DestroyAndReopen(options);

+  // Hold open a snapshot to prevent range tombstones from being compacted away.
+  ManagedSnapshot snapshot(db_);
+
  std::string level_tp_strings[kMaxLevel];
  std::string tp_string;
  TableProperties level_tps[kMaxLevel];
@@ -471,6 +507,12 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
      db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
               RandomString(&rnd, kValueSize));
    }
+    for (int i = 0; i < kRangeDeletionsPerTable; i++) {
+      std::string start = RandomString(&rnd, kKeySize);
+      std::string end = start;
+      end.resize(kValueSize);
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), start, end);
+    }
    db_->Flush(FlushOptions());
    db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
    ResetTableProperties(&sum_tp);
@@ -486,9 +528,12 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
      sum_tp.raw_value_size += level_tps[level].raw_value_size;
      sum_tp.num_data_blocks += level_tps[level].num_data_blocks;
      sum_tp.num_entries += level_tps[level].num_entries;
+      sum_tp.num_range_deletions += level_tps[level].num_range_deletions;
    }
    db_->GetProperty(DB::Properties::kAggregatedTableProperties, &tp_string);
    ParseTablePropertiesString(tp_string, &tp);
+    bool index_key_is_user_key = tp.index_key_is_user_key > 0;
+    bool value_is_delta_encoded = tp.index_value_is_delta_encoded > 0;
    ASSERT_EQ(sum_tp.data_size, tp.data_size);
    ASSERT_EQ(sum_tp.index_size, tp.index_size);
    ASSERT_EQ(sum_tp.filter_size, tp.filter_size);
@@ -496,13 +541,15 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
    ASSERT_EQ(sum_tp.raw_value_size, tp.raw_value_size);
    ASSERT_EQ(sum_tp.num_data_blocks, tp.num_data_blocks);
    ASSERT_EQ(sum_tp.num_entries, tp.num_entries);
+    ASSERT_EQ(sum_tp.num_range_deletions, tp.num_range_deletions);
    if (table > 3) {
      GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                                 kKeysPerTable, table, kBloomBitsPerKey,
-                                 table_options.block_size);
+                                 kKeysPerTable, kRangeDeletionsPerTable, table,
+                                 kBloomBitsPerKey, table_options.block_size,
+                                 index_key_is_user_key, value_is_delta_encoded);
      // Gives larger bias here as index block size, filter block size,
      // and data block size become much harder to estimate in this test.
-      VerifyTableProperties(tp, expected_tp, 0.5, 0.4, 0.4, 0.25);
+      VerifyTableProperties(expected_tp, tp, 0.5, 0.4, 0.4, 0.25);
    }
  }
 }
@@ -27,47 +27,53 @@ class DBRangeDelTest : public DBTestBase {
 // ROCKSDB_LITE
 #ifndef ROCKSDB_LITE
 TEST_F(DBRangeDelTest, NonBlockBasedTableNotSupported) {
-  if (!IsMemoryMappedAccessSupported()) {
-    return;
+  // TODO: figure out why MmapReads trips the iterator pinning assertion in
+  // RangeDelAggregator. Ideally it would be supported; otherwise it should at
+  // least be explicitly unsupported.
+  for (auto config : {kPlainTableAllBytesPrefix, /* kWalDirAndMmapReads */}) {
+    option_config_ = config;
+    DestroyAndReopen(CurrentOptions());
+    ASSERT_TRUE(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                                 "dr1", "dr1")
+                    .IsNotSupported());
  }
-  Options opts = CurrentOptions();
-  opts.table_factory.reset(new PlainTableFactory());
-  opts.prefix_extractor.reset(NewNoopTransform());
-  opts.allow_mmap_reads = true;
-  opts.max_sequential_skip_in_iterations = 999999;
-  Reopen(opts);
-
-  ASSERT_TRUE(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1", "dr1")
-          .IsNotSupported());
 }

 TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
-  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
-                             "dr2"));
-  ASSERT_OK(db_->Flush(FlushOptions()));
-  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                               "dr1", "dr2"));
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    ASSERT_EQ(1, NumTableFilesAtLevel(0));
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }

 TEST_F(DBRangeDelTest, CompactionOutputHasOnlyRangeTombstone) {
-  Options opts = CurrentOptions();
-  opts.disable_auto_compactions = true;
-  opts.statistics = CreateDBStatistics();
-  Reopen(opts);
+  do {
+    Options opts = CurrentOptions();
+    opts.disable_auto_compactions = true;
+    opts.statistics = CreateDBStatistics();
+    DestroyAndReopen(opts);

-  // snapshot protects range tombstone from dropping due to becoming obsolete.
-  const Snapshot* snapshot = db_->GetSnapshot();
-  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
-  db_->Flush(FlushOptions());
+    // snapshot protects range tombstone from dropping due to becoming obsolete.
+    const Snapshot* snapshot = db_->GetSnapshot();
+    db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
+    db_->Flush(FlushOptions());

-  ASSERT_EQ(1, NumTableFilesAtLevel(0));
-  ASSERT_EQ(0, NumTableFilesAtLevel(1));
-  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
-                              true /* disallow_trivial_move */);
-  ASSERT_EQ(0, NumTableFilesAtLevel(0));
-  ASSERT_EQ(1, NumTableFilesAtLevel(1));
-  ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
-  db_->ReleaseSnapshot(snapshot);
+    ASSERT_EQ(1, NumTableFilesAtLevel(0));
+    ASSERT_EQ(0, NumTableFilesAtLevel(1));
+    dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                                true /* disallow_trivial_move */);
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_EQ(1, NumTableFilesAtLevel(1));
+    ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
+    db_->ReleaseSnapshot(snapshot);
+    // Skip cuckoo memtables, which do not support snapshots. Skip non-leveled
+    // compactions as the above assertions about the number of files in a level
+    // do not hold true.
+  } while (ChangeOptions(kRangeDelSkipConfigs | kSkipHashCuckoo |
+                         kSkipUniversalCompaction | kSkipFIFOCompaction));
 }

 TEST_F(DBRangeDelTest, CompactionOutputFilesExactlyFilled) {
@@ -185,7 +191,7 @@ TEST_F(DBRangeDelTest, SentinelsOmittedFromOutputFile) {

  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
-  ASSERT_GT(files[0][0].smallest_seqno, 0);
+  ASSERT_GT(files[0][0].fd.smallest_seqno, 0);

  db_->ReleaseSnapshot(snapshot);
 }
@@ -496,12 +502,12 @@ TEST_F(DBRangeDelTest, ObsoleteTombstoneCleanup) {
  Reopen(opts);

  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
-                   "dr1");  // obsolete after compaction
+                   "dr10");  // obsolete after compaction
  db_->Put(WriteOptions(), "key", "val");
  db_->Flush(FlushOptions());
  const Snapshot* snapshot = db_->GetSnapshot();
  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr2",
-                   "dr2");  // protected by snapshot
+                   "dr20");  // protected by snapshot
  db_->Put(WriteOptions(), "key", "val");
  db_->Flush(FlushOptions());

@@ -590,48 +596,57 @@ TEST_F(DBRangeDelTest, TableEvictedDuringScan) {
 }

 TEST_F(DBRangeDelTest, GetCoveredKeyFromMutableMemtable) {
-  db_->Put(WriteOptions(), "key", "val");
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    db_->Put(WriteOptions(), "key", "val");
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));

-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }

 TEST_F(DBRangeDelTest, GetCoveredKeyFromImmutableMemtable) {
-  Options opts = CurrentOptions();
-  opts.max_write_buffer_number = 3;
-  opts.min_write_buffer_number_to_merge = 2;
-  // SpecialSkipListFactory lets us specify maximum number of elements the
-  // memtable can hold. It switches the active memtable to immutable (flush is
-  // prevented by the above options) upon inserting an element that would
-  // overflow the memtable.
-  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
-  Reopen(opts);
+  do {
+    Options opts = CurrentOptions();
+    opts.max_write_buffer_number = 3;
+    opts.min_write_buffer_number_to_merge = 2;
+    // SpecialSkipListFactory lets us specify maximum number of elements the
+    // memtable can hold. It switches the active memtable to immutable (flush is
+    // prevented by the above options) upon inserting an element that would
+    // overflow the memtable.
+    opts.memtable_factory.reset(new SpecialSkipListFactory(1));
+    DestroyAndReopen(opts);

-  db_->Put(WriteOptions(), "key", "val");
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
-  db_->Put(WriteOptions(), "blah", "val");
+    db_->Put(WriteOptions(), "key", "val");
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+    db_->Put(WriteOptions(), "blah", "val");

-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+  } while (ChangeOptions(kRangeDelSkipConfigs));
 }

 TEST_F(DBRangeDelTest, GetCoveredKeyFromSst) {
-  db_->Put(WriteOptions(), "key", "val");
-  // snapshot prevents key from being deleted during flush
-  const Snapshot* snapshot = db_->GetSnapshot();
-  ASSERT_OK(
-      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
-  ASSERT_OK(db_->Flush(FlushOptions()));
+  do {
+    DestroyAndReopen(CurrentOptions());
+    db_->Put(WriteOptions(), "key", "val");
+    // snapshot prevents key from being deleted during flush
+    const Snapshot* snapshot = db_->GetSnapshot();
+    ASSERT_OK(
+        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
+    ASSERT_OK(db_->Flush(FlushOptions()));

-  ReadOptions read_opts;
-  std::string value;
-  ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
-  db_->ReleaseSnapshot(snapshot);
+    ReadOptions read_opts;
+    std::string value;
+    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
+    db_->ReleaseSnapshot(snapshot);
+    // Cuckoo memtables do not support snapshots.
+  } while (ChangeOptions(kRangeDelSkipConfigs | kSkipHashCuckoo));
 }

 TEST_F(DBRangeDelTest, GetCoveredMergeOperandFromMemtable) {
@@ -895,11 +910,14 @@ TEST_F(DBRangeDelTest, MemtableBloomFilter) {
 }

 TEST_F(DBRangeDelTest, CompactionTreatsSplitInputLevelDeletionAtomically) {
-  // make sure compaction treats files containing a split range deletion in the
-  // input level as an atomic unit. I.e., compacting any input-level file(s)
-  // containing a portion of the range deletion causes all other input-level
-  // files containing portions of that same range deletion to be included in the
-  // compaction.
+  // This test originally verified that compaction treated files containing a
+  // split range deletion in the input level as an atomic unit. I.e.,
+  // compacting any input-level file(s) containing a portion of the range
+  // deletion causes all other input-level files containing portions of that
+  // same range deletion to be included in the compaction. Range deletion
+  // tombstones are now truncated to sstable boundaries which removed the need
+  // for that behavior (which could lead to excessively large
+  // compactions).
  const int kNumFilesPerLevel = 4, kValueBytes = 4 << 10;
  Options options = CurrentOptions();
  options.compression = kNoCompression;
@@ -946,22 +964,111 @@ TEST_F(DBRangeDelTest, CompactionTreatsSplitInputLevelDeletionAtomically) {
    if (i == 0) {
      ASSERT_OK(db_->CompactFiles(
          CompactionOptions(), {meta.levels[1].files[0].name}, 2 /* level */));
+      ASSERT_EQ(0, NumTableFilesAtLevel(1));
    } else if (i == 1) {
      auto begin_str = Key(0), end_str = Key(1);
      Slice begin = begin_str, end = end_str;
      ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &begin, &end));
+      ASSERT_EQ(3, NumTableFilesAtLevel(1));
    } else if (i == 2) {
      ASSERT_OK(db_->SetOptions(db_->DefaultColumnFamily(),
                                {{"max_bytes_for_level_base", "10000"}}));
      dbfull()->TEST_WaitForCompact();
+      ASSERT_EQ(1, NumTableFilesAtLevel(1));
    }
-    ASSERT_EQ(0, NumTableFilesAtLevel(1));
    ASSERT_GT(NumTableFilesAtLevel(2), 0);

    db_->ReleaseSnapshot(snapshot);
  }
 }

+TEST_F(DBRangeDelTest, RangeTombstoneEndKeyAsSstableUpperBound) {
+  // Test the handling of the range-tombstone end-key as the
+  // upper-bound for an sstable.
+
+  const int kNumFilesPerLevel = 2, kValueBytes = 4 << 10;
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.level0_file_num_compaction_trigger = kNumFilesPerLevel;
+  options.memtable_factory.reset(
+      new SpecialSkipListFactory(2 /* num_entries_flush */));
+  options.target_file_size_base = kValueBytes;
+  options.disable_auto_compactions = true;
+
+  DestroyAndReopen(options);
+
+  // Create an initial sstable at L2:
+  //   [key000000#1,1, key000000#1,1]
+  ASSERT_OK(Put(Key(0), ""));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  MoveFilesToLevel(2);
+  ASSERT_EQ(1, NumTableFilesAtLevel(2));
+
+  // A snapshot protects the range tombstone from dropping due to
+  // becoming obsolete.
+  const Snapshot* snapshot = db_->GetSnapshot();
+  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                   Key(0), Key(2 * kNumFilesPerLevel));
+
+  // Create 2 additional sstables in L0. Note that the first sstable
+  // contains the range tombstone.
+  //   [key000000#3,1, key000004#72057594037927935,15]
+  //   [key000001#5,1, key000002#6,1]
+  Random rnd(301);
+  std::string value = RandomString(&rnd, kValueBytes);
+  for (int j = 0; j < kNumFilesPerLevel; ++j) {
+    // Give files overlapping key-ranges to prevent a trivial move when we
+    // compact from L0 to L1.
+    ASSERT_OK(Put(Key(j), value));
+    ASSERT_OK(Put(Key(2 * kNumFilesPerLevel - 1 - j), value));
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    ASSERT_EQ(j + 1, NumTableFilesAtLevel(0));
+  }
+  // Compact the 2 L0 sstables to L1, resulting in the following LSM. There
+  // are 2 sstables generated in L1 due to the target_file_size_base setting.
+  //   L1:
+  //     [key000000#3,1, key000002#72057594037927935,15]
+  //     [key000002#6,1, key000004#72057594037927935,15]
+  //   L2:
+  //     [key000000#1,1, key000000#1,1]
+  MoveFilesToLevel(1);
+  ASSERT_EQ(2, NumTableFilesAtLevel(1));
+
+  {
+    // Compact the second sstable in L1:
+    //   L1:
+    //     [key000000#3,1, key000002#72057594037927935,15]
+    //   L2:
+    //     [key000000#1,1, key000000#1,1]
+    //     [key000002#6,1, key000004#72057594037927935,15]
+    auto begin_str = Key(3);
+    const rocksdb::Slice begin = begin_str;
+    dbfull()->TEST_CompactRange(1, &begin, nullptr);
+    ASSERT_EQ(1, NumTableFilesAtLevel(1));
+    ASSERT_EQ(2, NumTableFilesAtLevel(2));
+  }
+
+  {
+    // Compact the first sstable in L1. This should be copacetic, but
+    // was previously resulting in overlapping sstables in L2 due to
+    // mishandling of the range tombstone end-key when used as the
+    // largest key for an sstable. The resulting LSM structure should
+    // be:
+    //
+    //   L2:
+    //     [key000000#1,1, key000001#72057594037927935,15]
+    //     [key000001#5,1, key000002#72057594037927935,15]
+    //     [key000002#6,1, key000004#72057594037927935,15]
+    auto begin_str = Key(0);
+    const rocksdb::Slice begin = begin_str;
+    dbfull()->TEST_CompactRange(1, &begin, &begin);
+    ASSERT_EQ(0, NumTableFilesAtLevel(1));
+    ASSERT_EQ(3, NumTableFilesAtLevel(2));
+  }
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_F(DBRangeDelTest, UnorderedTombstones) {
  // Regression test for #2752. Range delete tombstones between
  // different snapshot stripes are not stored in order, so the first
@@ -436,7 +436,7 @@ TEST_F(DBSSTTest, OpenDBWithExistingTrash) {
 // deleted from first db_path were deleted using DeleteScheduler and
 // files in the second path were not.
 TEST_F(DBSSTTest, DeleteSchedulerMultipleDBPaths) {
-  int bg_delete_file = 0;
+  std::atomic<int> bg_delete_file(0);
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DeleteScheduler::DeleteTrashFile:DeleteFile",
      [&](void* /*arg*/) { bg_delete_file++; });
@@ -703,26 +703,19 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
  // When bg_error_ is set we will verify that the DB size is greater
  // than the limit.

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

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

  rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -735,15 +728,11 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {
      "CompactionJob::FinishCompactionOutputFile:MaxAllowedSpaceReached",
      [&](void* /*arg*/) {
        bg_error_set = true;
-        GetAllSSTFiles(&total_sst_files_size);
        reached_max_space_on_compaction++;
      });

  for (auto limit_mb : max_space_limits_mbs) {
    bg_error_set = false;
-    total_sst_files_size = 0;
-    estimate_multiplier = 1;
-    limit_mb_cb = limit_mb;
    rocksdb::SyncPoint::GetInstance()->ClearTrace();
    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
    std::shared_ptr<SstFileManager> sst_file_manager(NewSstFileManager(env_));
@@ -757,21 +746,17 @@ TEST_F(DBSSTTest, DBWithMaxSpaceAllowedRandomized) {

    sfm->SetMaxAllowedSpaceUsage(limit_mb * 1024 * 1024);

-    int keys_written = 0;
-    uint64_t estimated_db_size = 0;
+    // It is easy to detect if the test is stuck in a loop. No need for
+    // complex termination logic.
    while (true) {
      auto s = Put(RandomString(&rnd, 10), RandomString(&rnd, 50));
      if (!s.ok()) {
        break;
      }
-      keys_written++;
-      // Check the estimated db size vs the db limit just to make sure we
-      // dont run into an infinite loop
-      estimated_db_size = keys_written * 60;  // ~60 bytes per key
-      ASSERT_LT(estimated_db_size,
-                estimate_multiplier * limit_mb * 1024 * 1024 * 2);
    }
    ASSERT_TRUE(bg_error_set);
+    uint64_t total_sst_files_size = 0;
+    GetAllSSTFiles(&total_sst_files_size);
    ASSERT_GE(total_sst_files_size, limit_mb * 1024 * 1024);
    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
  }
@@ -252,13 +252,14 @@ TEST_F(DBTablePropertiesTest, GetColumnFamilyNameProperty) {
 }

 TEST_F(DBTablePropertiesTest, DeletionTriggeredCompactionMarking) {
-  const int kNumKeys = 1000;
-  const int kWindowSize = 100;
-  const int kNumDelsTrigger = 90;
+  int kNumKeys = 1000;
+  int kWindowSize = 100;
+  int kNumDelsTrigger = 90;
+  std::shared_ptr<TablePropertiesCollectorFactory> compact_on_del =
+    NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger);

  Options opts = CurrentOptions();
-  opts.table_properties_collector_factories.emplace_back(
-      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.table_properties_collector_factories.emplace_back(compact_on_del);
  Reopen(opts);

  // add an L1 file to prevent tombstones from dropping due to obsolescence
@@ -280,6 +281,48 @@ TEST_F(DBTablePropertiesTest, DeletionTriggeredCompactionMarking) {
  dbfull()->TEST_WaitForCompact();
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_GT(NumTableFilesAtLevel(1), 0);
+
+  // Change the window size and deletion trigger and ensure new values take
+  // effect
+  kWindowSize = 50;
+  kNumDelsTrigger = 40;
+  static_cast<CompactOnDeletionCollectorFactory*>
+    (compact_on_del.get())->SetWindowSize(kWindowSize);
+  static_cast<CompactOnDeletionCollectorFactory*>
+    (compact_on_del.get())->SetDeletionTrigger(kNumDelsTrigger);
+  for (int i = 0; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(1), 0);
+
+  // Change the window size to disable delete triggered compaction
+  kWindowSize = 0;
+  static_cast<CompactOnDeletionCollectorFactory*>
+    (compact_on_del.get())->SetWindowSize(kWindowSize);
+  static_cast<CompactOnDeletionCollectorFactory*>
+    (compact_on_del.get())->SetDeletionTrigger(kNumDelsTrigger);
+  for (int i = 0; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+
 }

 }  // namespace rocksdb
@@ -479,275 +479,6 @@ TEST_F(DBTestTailingIterator, TailingIteratorGap) {
  ASSERT_EQ("40", it->key().ToString());
 }

-TEST_F(DBTestTailingIterator, ManagedTailingIteratorSingle) {
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
-  iter->SeekToFirst();
-  ASSERT_TRUE(!iter->Valid());
-
-  // add a record and check that iter can see it
-  ASSERT_OK(db_->Put(WriteOptions(), "mirko", "fodor"));
-  iter->SeekToFirst();
-  ASSERT_TRUE(iter->Valid());
-  ASSERT_EQ(iter->key().ToString(), "mirko");
-
-  iter->Next();
-  ASSERT_TRUE(!iter->Valid());
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorKeepAdding) {
-  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options, handles_[1]));
-  std::string value(1024, 'a');
-
-  const int num_records = 10000;
-  for (int i = 0; i < num_records; ++i) {
-    char buf[32];
-    snprintf(buf, sizeof(buf), "%016d", i);
-
-    Slice key(buf, 16);
-    ASSERT_OK(Put(1, key, value));
-
-    iter->Seek(key);
-    ASSERT_TRUE(iter->Valid());
-    ASSERT_EQ(iter->key().compare(key), 0);
-  }
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorSeekToNext) {
-  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options, handles_[1]));
-  std::string value(1024, 'a');
-
-  const int num_records = 1000;
-  for (int i = 1; i < num_records; ++i) {
-    char buf1[32];
-    char buf2[32];
-    snprintf(buf1, sizeof(buf1), "00a0%016d", i * 5);
-
-    Slice key(buf1, 20);
-    ASSERT_OK(Put(1, key, value));
-
-    if (i % 100 == 99) {
-      ASSERT_OK(Flush(1));
-    }
-
-    snprintf(buf2, sizeof(buf2), "00a0%016d", i * 5 - 2);
-    Slice target(buf2, 20);
-    iter->Seek(target);
-    ASSERT_TRUE(iter->Valid());
-    ASSERT_EQ(iter->key().compare(key), 0);
-  }
-  for (int i = 2 * num_records; i > 0; --i) {
-    char buf1[32];
-    char buf2[32];
-    snprintf(buf1, sizeof(buf1), "00a0%016d", i * 5);
-
-    Slice key(buf1, 20);
-    ASSERT_OK(Put(1, key, value));
-
-    if (i % 100 == 99) {
-      ASSERT_OK(Flush(1));
-    }
-
-    snprintf(buf2, sizeof(buf2), "00a0%016d", i * 5 - 2);
-    Slice target(buf2, 20);
-    iter->Seek(target);
-    ASSERT_TRUE(iter->Valid());
-    ASSERT_EQ(iter->key().compare(key), 0);
-  }
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorDeletes) {
-  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options, handles_[1]));
-
-  // write a single record, read it using the iterator, then delete it
-  ASSERT_OK(Put(1, "0test", "test"));
-  iter->SeekToFirst();
-  ASSERT_TRUE(iter->Valid());
-  ASSERT_EQ(iter->key().ToString(), "0test");
-  ASSERT_OK(Delete(1, "0test"));
-
-  // write many more records
-  const int num_records = 10000;
-  std::string value(1024, 'A');
-
-  for (int i = 0; i < num_records; ++i) {
-    char buf[32];
-    snprintf(buf, sizeof(buf), "1%015d", i);
-
-    Slice key(buf, 16);
-    ASSERT_OK(Put(1, key, value));
-  }
-
-  // force a flush to make sure that no records are read from memtable
-  ASSERT_OK(Flush(1));
-
-  // skip "0test"
-  iter->Next();
-
-  // make sure we can read all new records using the existing iterator
-  int count = 0;
-  for (; iter->Valid(); iter->Next(), ++count) {
-  }
-
-  ASSERT_EQ(count, num_records);
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorPrefixSeek) {
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  Options options = CurrentOptions();
-  options.create_if_missing = true;
-  options.disable_auto_compactions = true;
-  options.prefix_extractor.reset(NewFixedPrefixTransform(2));
-  options.memtable_factory.reset(NewHashSkipListRepFactory(16));
-  options.allow_concurrent_memtable_write = false;
-  DestroyAndReopen(options);
-  CreateAndReopenWithCF({"pikachu"}, options);
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options, handles_[1]));
-  ASSERT_OK(Put(1, "0101", "test"));
-
-  ASSERT_OK(Flush(1));
-
-  ASSERT_OK(Put(1, "0202", "test"));
-
-  // Seek(0102) shouldn't find any records since 0202 has a different prefix
-  iter->Seek("0102");
-  ASSERT_TRUE(!iter->Valid());
-
-  iter->Seek("0202");
-  ASSERT_TRUE(iter->Valid());
-  ASSERT_EQ(iter->key().ToString(), "0202");
-
-  iter->Next();
-  ASSERT_TRUE(!iter->Valid());
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorIncomplete) {
-  CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-  read_options.read_tier = kBlockCacheTier;
-
-  std::string key = "key";
-  std::string value = "value";
-
-  ASSERT_OK(db_->Put(WriteOptions(), key, value));
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
-  iter->SeekToFirst();
-  // we either see the entry or it's not in cache
-  ASSERT_TRUE(iter->Valid() || iter->status().IsIncomplete());
-
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  iter->SeekToFirst();
-  // should still be true after compaction
-  ASSERT_TRUE(iter->Valid() || iter->status().IsIncomplete());
-}
-
-TEST_F(DBTestTailingIterator, ManagedTailingIteratorSeekToSame) {
-  Options options = CurrentOptions();
-  options.compaction_style = kCompactionStyleUniversal;
-  options.write_buffer_size = 1000;
-  CreateAndReopenWithCF({"pikachu"}, options);
-
-  ReadOptions read_options;
-  read_options.tailing = true;
-  read_options.managed = true;
-
-  const int NROWS = 10000;
-  // Write rows with keys 00000, 00002, 00004 etc.
-  for (int i = 0; i < NROWS; ++i) {
-    char buf[100];
-    snprintf(buf, sizeof(buf), "%05d", 2 * i);
-    std::string key(buf);
-    std::string value("value");
-    ASSERT_OK(db_->Put(WriteOptions(), key, value));
-  }
-
-  std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
-  // Seek to 00001.  We expect to find 00002.
-  std::string start_key = "00001";
-  iter->Seek(start_key);
-  ASSERT_TRUE(iter->Valid());
-
-  std::string found = iter->key().ToString();
-  ASSERT_EQ("00002", found);
-
-  // Now seek to the same key.  The iterator should remain in the same
-  // position.
-  iter->Seek(found);
-  ASSERT_TRUE(iter->Valid());
-  ASSERT_EQ(found, iter->key().ToString());
-}
-
-TEST_F(DBTestTailingIterator, ForwardIteratorVersionProperty) {
-  Options options = CurrentOptions();
-  options.write_buffer_size = 1000;
-
-  ReadOptions read_options;
-  read_options.tailing = true;
-
-  Put("foo", "bar");
-
-  uint64_t v1, v2, v3, v4;
-  {
-    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
-    iter->Seek("foo");
-    std::string prop_value;
-    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
-                                &prop_value));
-    v1 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
-
-    Put("foo1", "bar1");
-    Flush();
-
-    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
-                                &prop_value));
-    v2 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
-
-    iter->Seek("f");
-
-    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
-                                &prop_value));
-    v3 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
-
-    ASSERT_EQ(v1, v2);
-    ASSERT_GT(v3, v2);
-  }
-
-  {
-    std::unique_ptr<Iterator> iter(db_->NewIterator(read_options));
-    iter->Seek("foo");
-    std::string prop_value;
-    ASSERT_OK(iter->GetProperty("rocksdb.iterator.super-version-number",
-                                &prop_value));
-    v4 = static_cast<uint64_t>(std::atoi(prop_value.c_str()));
-  }
-  ASSERT_EQ(v3, v4);
-}
-
 TEST_F(DBTestTailingIterator, SeekWithUpperBoundBug) {
  ReadOptions read_options;
  read_options.tailing = true;
@@ -262,6 +262,196 @@ TEST_F(DBTest, SkipDelay) {
  }
 }

+TEST_F(DBTest, MixedSlowdownOptions) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.write_buffer_size = 100000;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  std::vector<port::Thread> threads;
+  std::atomic<int> thread_num(0);
+
+  std::function<void()> write_slowdown_func = [&]() {
+    int a = thread_num.fetch_add(1);
+    std::string key = "foo" + std::to_string(a);
+    WriteOptions wo;
+    wo.no_slowdown = false;
+    ASSERT_OK(dbfull()->Put(wo, key, "bar"));
+  };
+  std::function<void()> write_no_slowdown_func = [&]() {
+    int a = thread_num.fetch_add(1);
+    std::string key = "foo" + std::to_string(a);
+    WriteOptions wo;
+    wo.no_slowdown = true;
+    ASSERT_NOK(dbfull()->Put(wo, key, "bar"));
+  };
+  // Use a small number to ensure a large delay that is still effective
+  // when we do Put
+  // TODO(myabandeh): this is time dependent and could potentially make
+  // the test flaky
+  auto token = dbfull()->TEST_write_controler().GetDelayToken(1);
+  std::atomic<int> sleep_count(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DelayWrite:BeginWriteStallDone",
+      [&](void* /*arg*/) {
+        sleep_count.fetch_add(1);
+        if (threads.empty()) {
+          for (int i = 0; i < 2; ++i) {
+            threads.emplace_back(write_slowdown_func);
+          }
+          for (int i = 0; i < 2; ++i) {
+            threads.emplace_back(write_no_slowdown_func);
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  WriteOptions wo;
+  wo.sync = false;
+  wo.disableWAL = false;
+  wo.no_slowdown = false;
+  dbfull()->Put(wo, "foo", "bar");
+  // We need the 2nd write to trigger delay. This is because delay is
+  // estimated based on the last write size which is 0 for the first write.
+  ASSERT_OK(dbfull()->Put(wo, "foo2", "bar2"));
+          token.reset();
+
+  for (auto& t : threads) {
+    t.join();
+  }
+  ASSERT_GE(sleep_count.load(), 1);
+
+  wo.no_slowdown = true;
+  ASSERT_OK(dbfull()->Put(wo, "foo3", "bar"));
+}
+
+TEST_F(DBTest, MixedSlowdownOptionsInQueue) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.write_buffer_size = 100000;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  std::vector<port::Thread> threads;
+  std::atomic<int> thread_num(0);
+
+  std::function<void()> write_no_slowdown_func = [&]() {
+    int a = thread_num.fetch_add(1);
+    std::string key = "foo" + std::to_string(a);
+    WriteOptions wo;
+    wo.no_slowdown = true;
+    ASSERT_NOK(dbfull()->Put(wo, key, "bar"));
+  };
+  // Use a small number to ensure a large delay that is still effective
+  // when we do Put
+  // TODO(myabandeh): this is time dependent and could potentially make
+  // the test flaky
+  auto token = dbfull()->TEST_write_controler().GetDelayToken(1);
+  std::atomic<int> sleep_count(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DelayWrite:Sleep",
+      [&](void* /*arg*/) {
+        sleep_count.fetch_add(1);
+        if (threads.empty()) {
+          for (int i = 0; i < 2; ++i) {
+            threads.emplace_back(write_no_slowdown_func);
+          }
+          // Sleep for 2s to allow the threads to insert themselves into the
+          // write queue
+          env_->SleepForMicroseconds(3000000ULL);
+        }
+      });
+  std::atomic<int> wait_count(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DelayWrite:Wait",
+      [&](void* /*arg*/) { wait_count.fetch_add(1); });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  WriteOptions wo;
+  wo.sync = false;
+  wo.disableWAL = false;
+  wo.no_slowdown = false;
+  dbfull()->Put(wo, "foo", "bar");
+  // We need the 2nd write to trigger delay. This is because delay is
+  // estimated based on the last write size which is 0 for the first write.
+  ASSERT_OK(dbfull()->Put(wo, "foo2", "bar2"));
+          token.reset();
+
+  for (auto& t : threads) {
+    t.join();
+  }
+  ASSERT_EQ(sleep_count.load(), 1);
+  ASSERT_GE(wait_count.load(), 0);
+}
+
+TEST_F(DBTest, MixedSlowdownOptionsStop) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.write_buffer_size = 100000;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  std::vector<port::Thread> threads;
+  std::atomic<int> thread_num(0);
+
+  std::function<void()> write_slowdown_func = [&]() {
+    int a = thread_num.fetch_add(1);
+    std::string key = "foo" + std::to_string(a);
+    WriteOptions wo;
+    wo.no_slowdown = false;
+    ASSERT_OK(dbfull()->Put(wo, key, "bar"));
+  };
+  std::function<void()> write_no_slowdown_func = [&]() {
+    int a = thread_num.fetch_add(1);
+    std::string key = "foo" + std::to_string(a);
+    WriteOptions wo;
+    wo.no_slowdown = true;
+    ASSERT_NOK(dbfull()->Put(wo, key, "bar"));
+  };
+  std::function<void()> wakeup_writer = [&]() {
+    dbfull()->mutex_.Lock();
+    dbfull()->bg_cv_.SignalAll();
+    dbfull()->mutex_.Unlock();
+  };
+  // Use a small number to ensure a large delay that is still effective
+  // when we do Put
+  // TODO(myabandeh): this is time dependent and could potentially make
+  // the test flaky
+  auto token = dbfull()->TEST_write_controler().GetStopToken();
+  std::atomic<int> wait_count(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DelayWrite:Wait",
+      [&](void* /*arg*/) {
+        wait_count.fetch_add(1);
+        if (threads.empty()) {
+          for (int i = 0; i < 2; ++i) {
+            threads.emplace_back(write_slowdown_func);
+          }
+          for (int i = 0; i < 2; ++i) {
+            threads.emplace_back(write_no_slowdown_func);
+          }
+          // Sleep for 2s to allow the threads to insert themselves into the
+          // write queue
+          env_->SleepForMicroseconds(3000000ULL);
+        }
+        token.reset();
+        threads.emplace_back(wakeup_writer);
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  WriteOptions wo;
+  wo.sync = false;
+  wo.disableWAL = false;
+  wo.no_slowdown = false;
+  dbfull()->Put(wo, "foo", "bar");
+  // We need the 2nd write to trigger delay. This is because delay is
+  // estimated based on the last write size which is 0 for the first write.
+  ASSERT_OK(dbfull()->Put(wo, "foo2", "bar2"));
+          token.reset();
+
+  for (auto& t : threads) {
+    t.join();
+  }
+  ASSERT_GE(wait_count.load(), 1);
+
+  wo.no_slowdown = true;
+  ASSERT_OK(dbfull()->Put(wo, "foo3", "bar"));
+}
 #ifndef ROCKSDB_LITE

 TEST_F(DBTest, LevelLimitReopen) {
@@ -1682,7 +1872,7 @@ TEST_F(DBTest, CustomComparator) {

 TEST_F(DBTest, DBOpen_Options) {
  Options options = CurrentOptions();
-  std::string dbname = test::TmpDir(env_) + "/db_options_test";
+  std::string dbname = test::PerThreadDBPath("db_options_test");
  ASSERT_OK(DestroyDB(dbname, options));

  // Does not exist, and create_if_missing == false: error
@@ -1740,7 +1930,7 @@ TEST_F(DBTest, DBOpen_Change_NumLevels) {
 }

 TEST_F(DBTest, DestroyDBMetaDatabase) {
-  std::string dbname = test::TmpDir(env_) + "/db_meta";
+  std::string dbname = test::PerThreadDBPath("db_meta");
  ASSERT_OK(env_->CreateDirIfMissing(dbname));
  std::string metadbname = MetaDatabaseName(dbname, 0);
  ASSERT_OK(env_->CreateDirIfMissing(metadbname));
@@ -2149,6 +2339,9 @@ static void GCThreadBody(void* arg) {

 }  // namespace

+#ifndef TRAVIS
+// Disable this test temporarily on Travis as it fails intermittently.
+// Github issue: #4151
 TEST_F(DBTest, GroupCommitTest) {
  do {
    Options options = CurrentOptions();
@@ -2195,6 +2388,7 @@ TEST_F(DBTest, GroupCommitTest) {
    ASSERT_GT(hist_data.average, 0.0);
  } while (ChangeOptions(kSkipNoSeekToLast));
 }
+#endif  // TRAVIS

 namespace {
 typedef std::map<std::string, std::string> KVMap;
@@ -2561,6 +2755,7 @@ class ModelDB : public DB {
  std::string name_ = "";
 };

+#ifndef ROCKSDB_VALGRIND_RUN
 static std::string RandomKey(Random* rnd, int minimum = 0) {
  int len;
  do {
@@ -2717,6 +2912,7 @@ TEST_P(DBTestRandomized, Randomized) {
  if (model_snap != nullptr) model.ReleaseSnapshot(model_snap);
  if (db_snap != nullptr) db_->ReleaseSnapshot(db_snap);
 }
+#endif  // ROCKSDB_VALGRIND_RUN

 TEST_F(DBTest, BlockBasedTablePrefixIndexTest) {
  // create a DB with block prefix index
@@ -4325,7 +4521,7 @@ TEST_F(DBTest, DynamicCompactionOptions) {
  // Clean up memtable and L0. Block compaction threads. If continue to write
  // and flush memtables. We should see put stop after 8 memtable flushes
  // since level0_stop_writes_trigger = 8
-  dbfull()->TEST_FlushMemTable(true);
+  dbfull()->TEST_FlushMemTable(true, true);
  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
  // Block compaction
  test::SleepingBackgroundTask sleeping_task_low;
@@ -4338,7 +4534,7 @@ TEST_F(DBTest, DynamicCompactionOptions) {
  WriteOptions wo;
  while (count < 64) {
    ASSERT_OK(Put(Key(count), RandomString(&rnd, 1024), wo));
-    dbfull()->TEST_FlushMemTable(true);
+    dbfull()->TEST_FlushMemTable(true, true);
    count++;
    if (dbfull()->TEST_write_controler().IsStopped()) {
      sleeping_task_low.WakeUp();
@@ -4366,7 +4562,7 @@ TEST_F(DBTest, DynamicCompactionOptions) {
  count = 0;
  while (count < 64) {
    ASSERT_OK(Put(Key(count), RandomString(&rnd, 1024), wo));
-    dbfull()->TEST_FlushMemTable(true);
+    dbfull()->TEST_FlushMemTable(true, true);
    count++;
    if (dbfull()->TEST_write_controler().IsStopped()) {
      sleeping_task_low.WakeUp();
@@ -4605,180 +4801,6 @@ TEST_F(DBTest, FileCreationRandomFailure) {
 }

 #ifndef ROCKSDB_LITE
-int CountIter(Iterator* iter, const Slice& key) {
-  int count = 0;
-  for (iter->Seek(key); iter->Valid() && iter->status() == Status::OK();
-       iter->Next()) {
-    count++;
-  }
-  return count;
-}
-
-// Create multiple SST files each with a different prefix_extractor config,
-// verify iterators can read all SST files using the latest config.
-TEST_F(DBTest, DynamicBloomFilterMultipleSST) {
-  Options options;
-  options.create_if_missing = true;
-  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
-  options.disable_auto_compactions = true;
-  // Enable prefix bloom for SST files
-  BlockBasedTableOptions table_options;
-  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
-  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
-  DestroyAndReopen(options);
-
-  ReadOptions read_options;
-  read_options.prefix_same_as_start = true;
-
-  // first SST with fixed:1 BF
-  ASSERT_OK(Put("foo2", "bar2"));
-  ASSERT_OK(Put("foo", "bar"));
-  ASSERT_OK(Put("foq1", "bar1"));
-  ASSERT_OK(Put("fpa", "0"));
-  dbfull()->Flush(FlushOptions());
-  Iterator* iter_old = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_old, "foo"), 4);
-
-  ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
-  ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
-                      "rocksdb.CappedPrefix.3"));
-  Iterator* iter = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter, "foo"), 2);
-
-  // second SST with capped:3 BF
-  ASSERT_OK(Put("foo3", "bar3"));
-  ASSERT_OK(Put("foo4", "bar4"));
-  ASSERT_OK(Put("foq5", "bar5"));
-  ASSERT_OK(Put("fpb", "1"));
-  dbfull()->Flush(FlushOptions());
-  // BF is cappped:3 now
-  Iterator* iter_tmp = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_tmp, "foo"), 4);
-  delete iter_tmp;
-
-  ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "fixed:2"}}));
-  ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
-                      "rocksdb.FixedPrefix.2"));
-  // third SST with fixed:2 BF
-  ASSERT_OK(Put("foo6", "bar6"));
-  ASSERT_OK(Put("foo7", "bar7"));
-  ASSERT_OK(Put("foq8", "bar8"));
-  ASSERT_OK(Put("fpc", "2"));
-  dbfull()->Flush(FlushOptions());
-  // BF is fixed:2 now
-  iter_tmp = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_tmp, "foo"), 9);
-  delete iter_tmp;
-
-  // TODO(Zhongyi): verify existing iterator cannot see newly inserted keys
-  ASSERT_EQ(CountIter(iter_old, "foo"), 4);
-  ASSERT_EQ(CountIter(iter, "foo"), 2);
-  delete iter;
-  delete iter_old;
-
-  // keys in all three SSTs are visible to iterator
-  Iterator* iter_all = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_all, "foo"), 9);
-  delete iter_all;
-  ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
-  ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
-                      "rocksdb.CappedPrefix.3"));
-  iter_all = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_all, "foo"), 6);
-  delete iter_all;
-  // TODO(Zhongyi): add test for cases where certain SST are skipped
-  // Also verify BF related counters like BLOOM_FILTER_USEFUL
-}
-
-// Create a new column family in a running DB, change prefix_extractor
-// dynamically, verify the iterator created on the new column family behaves
-// as expected
-TEST_F(DBTest, DynamicBloomFilterNewColumnFamily) {
-  Options options = CurrentOptions();
-  options.create_if_missing = true;
-  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
-  options.disable_auto_compactions = true;
-  // Enable prefix bloom for SST files
-  BlockBasedTableOptions table_options;
-  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
-  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
-  CreateAndReopenWithCF({"pikachu"}, options);
-  ReadOptions read_options;
-  read_options.prefix_same_as_start = true;
-  // create a new CF and set prefix_extractor dynamically
-  options.prefix_extractor.reset(NewCappedPrefixTransform(3));
-  CreateColumnFamilies({"ramen_dojo"}, options);
-  ASSERT_EQ(0,
-            strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
-                   "rocksdb.CappedPrefix.3"));
-  ASSERT_OK(Put(2, "foo3", "bar3"));
-  ASSERT_OK(Put(2, "foo4", "bar4"));
-  ASSERT_OK(Put(2, "foo5", "bar5"));
-  ASSERT_OK(Put(2, "foq6", "bar6"));
-  ASSERT_OK(Put(2, "fpq7", "bar7"));
-  dbfull()->Flush(FlushOptions());
-  Iterator* iter = db_->NewIterator(read_options, handles_[2]);
-  ASSERT_EQ(CountIter(iter, "foo"), 3);
-  delete iter;
-  ASSERT_OK(
-      dbfull()->SetOptions(handles_[2], {{"prefix_extractor", "fixed:2"}}));
-  ASSERT_EQ(0,
-            strcmp(dbfull()->GetOptions(handles_[2]).prefix_extractor->Name(),
-                   "rocksdb.FixedPrefix.2"));
-  iter = db_->NewIterator(read_options, handles_[2]);
-  ASSERT_EQ(CountIter(iter, "foo"), 4);
-  delete iter;
-}
-
-// Verify it's possible to change prefix_extractor at runtime and iterators
-// behaves as expected
-TEST_F(DBTest, DynamicBloomFilterOptions) {
-  Options options;
-  options.create_if_missing = true;
-  options.prefix_extractor.reset(NewFixedPrefixTransform(1));
-  options.disable_auto_compactions = true;
-  // Enable prefix bloom for SST files
-  BlockBasedTableOptions table_options;
-  table_options.filter_policy.reset(NewBloomFilterPolicy(10, true));
-  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
-  DestroyAndReopen(options);
-
-  ASSERT_OK(Put("foo2", "bar2"));
-  ASSERT_OK(Put("foo", "bar"));
-  ASSERT_OK(Put("foo1", "bar1"));
-  ASSERT_OK(Put("fpa", "0"));
-  dbfull()->Flush(FlushOptions());
-  ASSERT_OK(Put("foo3", "bar3"));
-  ASSERT_OK(Put("foo4", "bar4"));
-  ASSERT_OK(Put("foo5", "bar5"));
-  ASSERT_OK(Put("fpb", "1"));
-  dbfull()->Flush(FlushOptions());
-  ASSERT_OK(Put("foo6", "bar6"));
-  ASSERT_OK(Put("foo7", "bar7"));
-  ASSERT_OK(Put("foo8", "bar8"));
-  ASSERT_OK(Put("fpc", "2"));
-  dbfull()->Flush(FlushOptions());
-
-  ReadOptions read_options;
-  read_options.prefix_same_as_start = true;
-  Iterator* iter = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter, "foo"), 12);
-  delete iter;
-  Iterator* iter_old = db_->NewIterator(read_options);
-  ASSERT_EQ(CountIter(iter_old, "foo"), 12);
-
-  ASSERT_OK(dbfull()->SetOptions({{"prefix_extractor", "capped:3"}}));
-  ASSERT_EQ(0, strcmp(dbfull()->GetOptions().prefix_extractor->Name(),
-                      "rocksdb.CappedPrefix.3"));
-  iter = db_->NewIterator(read_options);
-  // "fp*" should be skipped
-  ASSERT_EQ(CountIter(iter, "foo"), 9);
-  delete iter;
-
-  // iterator created before should not be affected and see all keys
-  ASSERT_EQ(CountIter(iter_old, "foo"), 12);
-  delete iter_old;
-}

 TEST_F(DBTest, DynamicMiscOptions) {
  // Test max_sequential_skip_in_iterations
@@ -5097,6 +5119,7 @@ TEST_P(DBTestWithParam, FilterCompactionTimeTest) {
  options.disable_auto_compactions = true;
  options.create_if_missing = true;
  options.statistics = rocksdb::CreateDBStatistics();
+  options.statistics->stats_level_ = kExceptTimeForMutex;
  options.max_subcompactions = max_subcompactions_;
  DestroyAndReopen(options);

@@ -5679,7 +5702,7 @@ TEST_F(DBTest, SoftLimit) {
  for (int i = 0; i < 72; i++) {
    Put(Key(i), std::string(5000, 'x'));
    if (i % 10 == 0) {
-      Flush();
+      dbfull()->TEST_FlushMemTable(true, true);
    }
  }
  dbfull()->TEST_WaitForCompact();
@@ -5689,7 +5712,7 @@ TEST_F(DBTest, SoftLimit) {
  for (int i = 0; i < 72; i++) {
    Put(Key(i), std::string(5000, 'x'));
    if (i % 10 == 0) {
-      Flush();
+      dbfull()->TEST_FlushMemTable(true, true);
    }
  }
  dbfull()->TEST_WaitForCompact();
@@ -5708,7 +5731,7 @@ TEST_F(DBTest, SoftLimit) {
    Put(Key(i), std::string(5000, 'x'));
    Put(Key(100 - i), std::string(5000, 'x'));
    // Flush the file. File size is around 30KB.
-    Flush();
+    dbfull()->TEST_FlushMemTable(true, true);
  }
  ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
  ASSERT_TRUE(listener->CheckCondition(WriteStallCondition::kDelayed));
@@ -5743,7 +5766,7 @@ TEST_F(DBTest, SoftLimit) {
    Put(Key(10 + i), std::string(5000, 'x'));
    Put(Key(90 - i), std::string(5000, 'x'));
    // Flush the file. File size is around 30KB.
-    Flush();
+    dbfull()->TEST_FlushMemTable(true, true);
  }

  // Wake up sleep task to enable compaction to run and waits
@@ -5764,7 +5787,7 @@ TEST_F(DBTest, SoftLimit) {
    Put(Key(20 + i), std::string(5000, 'x'));
    Put(Key(80 - i), std::string(5000, 'x'));
    // Flush the file. File size is around 30KB.
-    Flush();
+    dbfull()->TEST_FlushMemTable(true, true);
  }
  // Wake up sleep task to enable compaction to run and waits
  // for it to go to sleep state again to make sure one compaction
@@ -354,7 +354,7 @@ INSTANTIATE_TEST_CASE_P(DBTestSharedWriteBufferAcrossCFs,
                                          std::make_tuple(false, true)));

 TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
-  std::string dbname2 = test::TmpDir(env_) + "/db_shared_wb_db2";
+  std::string dbname2 = test::PerThreadDBPath("db_shared_wb_db2");
  Options options = CurrentOptions();
  options.arena_block_size = 4096;
  // Avoid undeterministic value by malloc_usable_size();
@@ -2321,9 +2321,9 @@ TEST_F(DBTest2, RateLimitedCompactionReads) {
        options.rate_limiter->GetTotalBytesThrough(Env::IO_LOW);
    // Include the explicit prefetch of the footer in direct I/O case.
    size_t direct_io_extra = use_direct_io ? 512 * 1024 : 0;
-    ASSERT_GE(rate_limited_bytes,
-              static_cast<size_t>(kNumKeysPerFile * kBytesPerKey * kNumL0Files +
-                                  direct_io_extra));
+    ASSERT_GE(
+        rate_limited_bytes,
+        static_cast<size_t>(kNumKeysPerFile * kBytesPerKey * kNumL0Files));
    ASSERT_LT(
        rate_limited_bytes,
        static_cast<size_t>(2 * kNumKeysPerFile * kBytesPerKey * kNumL0Files +
@@ -2422,7 +2422,7 @@ TEST_F(DBTest2, ReadCallbackTest) {
  class TestReadCallback : public ReadCallback {
   public:
    explicit TestReadCallback(SequenceNumber snapshot) : snapshot_(snapshot) {}
-    virtual bool IsCommitted(SequenceNumber seq) override {
+    virtual bool IsVisible(SequenceNumber seq) override {
      return seq <= snapshot_;
    }

@@ -2500,8 +2500,341 @@ TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

+TEST_F(DBTest2, TraceAndReplay) {
+  Options options = CurrentOptions();
+  options.merge_operator = MergeOperators::CreatePutOperator();
+  ReadOptions ro;
+  WriteOptions wo;
+  TraceOptions trace_opts;
+  EnvOptions env_opts;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  Random rnd(301);
+  Iterator* single_iter = nullptr;
+
+  std::string trace_filename = dbname_ + "/rocksdb.trace";
+  std::unique_ptr<TraceWriter> trace_writer;
+  ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
+  ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
+
+  ASSERT_OK(Put(0, "a", "1"));
+  ASSERT_OK(Merge(0, "b", "2"));
+  ASSERT_OK(Delete(0, "c"));
+  ASSERT_OK(SingleDelete(0, "d"));
+  ASSERT_OK(db_->DeleteRange(wo, dbfull()->DefaultColumnFamily(), "e", "f"));
+
+  WriteBatch batch;
+  ASSERT_OK(batch.Put("f", "11"));
+  ASSERT_OK(batch.Merge("g", "12"));
+  ASSERT_OK(batch.Delete("h"));
+  ASSERT_OK(batch.SingleDelete("i"));
+  ASSERT_OK(batch.DeleteRange("j", "k"));
+  ASSERT_OK(db_->Write(wo, &batch));
+
+  single_iter = db_->NewIterator(ro);
+  single_iter->Seek("f");
+  single_iter->SeekForPrev("g");
+  delete single_iter;
+
+  ASSERT_EQ("1", Get(0, "a"));
+  ASSERT_EQ("12", Get(0, "g"));
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(1, "rocksdb", "rocks"));
+  ASSERT_EQ("NOT_FOUND", Get(1, "leveldb"));
+
+  ASSERT_OK(db_->EndTrace());
+  // These should not get into the trace file as it is after EndTrace.
+  Put("hello", "world");
+  Merge("foo", "bar");
+
+  // Open another db, replay, and verify the data
+  std::string value;
+  std::string dbname2 = test::TmpDir(env_) + "/db_replay";
+  ASSERT_OK(DestroyDB(dbname2, options));
+
+  // Using a different name than db2, to pacify infer's use-after-lifetime
+  // warnings (http://fbinfer.com).
+  DB* db2_init = nullptr;
+  options.create_if_missing = true;
+  ASSERT_OK(DB::Open(options, dbname2, &db2_init));
+  ColumnFamilyHandle* cf;
+  ASSERT_OK(
+      db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
+  delete cf;
+  delete db2_init;
+
+  DB* db2 = nullptr;
+  std::vector<ColumnFamilyDescriptor> column_families;
+  ColumnFamilyOptions cf_options;
+  cf_options.merge_operator = MergeOperators::CreatePutOperator();
+  column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
+  column_families.push_back(
+      ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
+  std::vector<ColumnFamilyHandle*> handles;
+  ASSERT_OK(DB::Open(DBOptions(), dbname2, column_families, &handles, &db2));
+
+  env_->SleepForMicroseconds(100);
+  // Verify that the keys don't already exist
+  ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
+  ASSERT_TRUE(db2->Get(ro, handles[0], "g", &value).IsNotFound());
+
+  std::unique_ptr<TraceReader> trace_reader;
+  ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
+  Replayer replayer(db2, handles_, std::move(trace_reader));
+  ASSERT_OK(replayer.Replay());
+
+  ASSERT_OK(db2->Get(ro, handles[0], "a", &value));
+  ASSERT_EQ("1", value);
+  ASSERT_OK(db2->Get(ro, handles[0], "g", &value));
+  ASSERT_EQ("12", value);
+  ASSERT_TRUE(db2->Get(ro, handles[0], "hello", &value).IsNotFound());
+  ASSERT_TRUE(db2->Get(ro, handles[0], "world", &value).IsNotFound());
+
+  ASSERT_OK(db2->Get(ro, handles[1], "foo", &value));
+  ASSERT_EQ("bar", value);
+  ASSERT_OK(db2->Get(ro, handles[1], "rocksdb", &value));
+  ASSERT_EQ("rocks", value);
+
+  for (auto handle : handles) {
+    delete handle;
+  }
+  delete db2;
+  ASSERT_OK(DestroyDB(dbname2, options));
+}
+
 #endif  // ROCKSDB_LITE

+TEST_F(DBTest2, PinnableSliceAndMmapReads) {
+  Options options = CurrentOptions();
+  options.allow_mmap_reads = true;
+  options.max_open_files = 100;
+  options.compression = kNoCompression;
+  Reopen(options);
+
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+
+  PinnableSlice pinned_value;
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  // It is not safe to pin mmap files as they might disappear by compaction
+  ASSERT_FALSE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+  dbfull()->TEST_CompactRange(0 /* level */, nullptr /* begin */,
+                              nullptr /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
+
+  // Ensure pinned_value doesn't rely on memory munmap'd by the above
+  // compaction. It crashes if it does.
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+#ifndef ROCKSDB_LITE
+  pinned_value.Reset();
+  // Unsafe to pin mmap files when they could be kicked out of table cache
+  Close();
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  ASSERT_FALSE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+
+  pinned_value.Reset();
+  // In read-only mode with infinite capacity on table cache it should pin the
+  // value and avoid the memcpy
+  Close();
+  options.max_open_files = -1;
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
+  ASSERT_TRUE(pinned_value.IsPinned());
+  ASSERT_EQ(pinned_value.ToString(), "bar");
+#endif
+}
+
+TEST_F(DBTest2, DISABLED_IteratorPinnedMemory) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  BlockBasedTableOptions bbto;
+  bbto.no_block_cache = false;
+  bbto.cache_index_and_filter_blocks = false;
+  bbto.block_cache = NewLRUCache(100000);
+  bbto.block_size = 400;  // small block size
+  options.table_factory.reset(new BlockBasedTableFactory(bbto));
+  Reopen(options);
+
+  Random rnd(301);
+  std::string v = RandomString(&rnd, 400);
+
+  // Since v is the size of a block, each key should take a block
+  // of 400+ bytes.
+  Put("1", v);
+  Put("3", v);
+  Put("5", v);
+  Put("7", v);
+  ASSERT_OK(Flush());
+
+  ASSERT_EQ(0, bbto.block_cache->GetPinnedUsage());
+
+  // Verify that iterators don't pin more than one data block in block cache
+  // at each time.
+  {
+    unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+    iter->SeekToFirst();
+
+    for (int i = 0; i < 4; i++) {
+      ASSERT_TRUE(iter->Valid());
+      // Block cache should contain exactly one block.
+      ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
+      ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
+      iter->Next();
+    }
+    ASSERT_FALSE(iter->Valid());
+
+    iter->Seek("4");
+    ASSERT_TRUE(iter->Valid());
+
+    ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
+    ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
+
+    iter->Seek("3");
+    ASSERT_TRUE(iter->Valid());
+
+    ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
+    ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
+  }
+  ASSERT_EQ(0, bbto.block_cache->GetPinnedUsage());
+
+  // Test compaction case
+  Put("2", v);
+  Put("5", v);
+  Put("6", v);
+  Put("8", v);
+  ASSERT_OK(Flush());
+
+  // Clear existing data in block cache
+  bbto.block_cache->SetCapacity(0);
+  bbto.block_cache->SetCapacity(100000);
+
+  // Verify compaction input iterators don't hold more than one data blocks at
+  // one time.
+  std::atomic<bool> finished(false);
+  std::atomic<int> block_newed(0);
+  std::atomic<int> block_destroyed(0);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Block::Block:0", [&](void* /*arg*/) {
+        if (finished) {
+          return;
+        }
+        // Two iterators. At most 2 outstanding blocks.
+        EXPECT_GE(block_newed.load(), block_destroyed.load());
+        EXPECT_LE(block_newed.load(), block_destroyed.load() + 1);
+        block_newed.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "Block::~Block", [&](void* /*arg*/) {
+        if (finished) {
+          return;
+        }
+        // Two iterators. At most 2 outstanding blocks.
+        EXPECT_GE(block_newed.load(), block_destroyed.load() + 1);
+        EXPECT_LE(block_newed.load(), block_destroyed.load() + 2);
+        block_destroyed.fetch_add(1);
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::Run:BeforeVerify",
+      [&](void* /*arg*/) { finished = true; });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  // Two input files. Each of them has 4 data blocks.
+  ASSERT_EQ(8, block_newed.load());
+  ASSERT_EQ(8, block_destroyed.load());
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_F(DBTest2, TestBBTTailPrefetch) {
+  std::atomic<bool> called(false);
+  size_t expected_lower_bound = 512 * 1024;
+  size_t expected_higher_bound = 512 * 1024;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BlockBasedTable::Open::TailPrefetchLen", [&](void* arg) {
+        size_t* prefetch_size = static_cast<size_t*>(arg);
+        EXPECT_LE(expected_lower_bound, *prefetch_size);
+        EXPECT_GE(expected_higher_bound, *prefetch_size);
+        called = true;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put("1", "1");
+  Put("9", "1");
+  Flush();
+
+  expected_lower_bound = 0;
+  expected_higher_bound = 8 * 1024;
+
+  Put("1", "1");
+  Put("9", "1");
+  Flush();
+
+  Put("1", "1");
+  Put("9", "1");
+  Flush();
+
+  // Full compaction to make sure there is no L0 file after the open.
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  ASSERT_TRUE(called.load());
+  called = false;
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  std::atomic<bool> first_call(true);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BlockBasedTable::Open::TailPrefetchLen", [&](void* arg) {
+        size_t* prefetch_size = static_cast<size_t*>(arg);
+        if (first_call) {
+          EXPECT_EQ(4 * 1024, *prefetch_size);
+          first_call = false;
+        } else {
+          EXPECT_GE(4 * 1024, *prefetch_size);
+        }
+        called = true;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options = CurrentOptions();
+  options.max_file_opening_threads = 1;  // one thread
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  options.max_open_files = -1;
+  Reopen(options);
+
+  Put("1", "1");
+  Put("9", "1");
+  Flush();
+
+  Put("1", "1");
+  Put("9", "1");
+  Flush();
+
+  ASSERT_TRUE(called.load());
+  called = false;
+
+  // Parallel loading SST files
+  options.max_file_opening_threads = 16;
+  Reopen(options);
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  ASSERT_TRUE(called.load());
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -63,7 +63,7 @@ DBTestBase::DBTestBase(const std::string path)
      option_config_(kDefault) {
  env_->SetBackgroundThreads(1, Env::LOW);
  env_->SetBackgroundThreads(1, Env::HIGH);
-  dbname_ = test::TmpDir(env_) + path;
+  dbname_ = test::PerThreadDBPath(env_, path);
  alternative_wal_dir_ = dbname_ + "/wal";
  alternative_db_log_dir_ = dbname_ + "/db_log_dir";
  auto options = CurrentOptions();
@@ -118,7 +118,8 @@ bool DBTestBase::ShouldSkipOptions(int option_config, int skip_mask) {

    if ((skip_mask & kSkipUniversalCompaction) &&
        (option_config == kUniversalCompaction ||
-         option_config == kUniversalCompactionMultiLevel)) {
+         option_config == kUniversalCompactionMultiLevel ||
+         option_config == kUniversalSubcompactions)) {
      return true;
    }
    if ((skip_mask & kSkipMergePut) && option_config == kMergePut) {
@@ -346,6 +347,26 @@ Options DBTestBase::GetOptions(
          NewHashCuckooRepFactory(options.write_buffer_size));
      options.allow_concurrent_memtable_write = false;
      break;
+      case kDirectIO: {
+        options.use_direct_reads = true;
+        options.use_direct_io_for_flush_and_compaction = true;
+        options.compaction_readahead_size = 2 * 1024 * 1024;
+  #if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
+      !defined(OS_AIX) && !defined(OS_OPENBSD)
+        rocksdb::SyncPoint::GetInstance()->SetCallBack(
+            "NewWritableFile:O_DIRECT", [&](void* arg) {
+              int* val = static_cast<int*>(arg);
+              *val &= ~O_DIRECT;
+            });
+        rocksdb::SyncPoint::GetInstance()->SetCallBack(
+            "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
+              int* val = static_cast<int*>(arg);
+              *val &= ~O_DIRECT;
+            });
+        rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  #endif
+        break;
+      }
 #endif  // ROCKSDB_LITE
    case kMergePut:
      options.merge_operator = MergeOperators::CreatePutOperator();
@@ -428,6 +449,18 @@ Options DBTestBase::GetOptions(
      options.prefix_extractor.reset(NewNoopTransform());
      break;
    }
+    case kBlockBasedTableWithPartitionedIndexFormat4: {
+      table_options.format_version = 4;
+      // Format 4 changes the binary index format. Since partitioned index is a
+      // super-set of simple indexes, we are also using kTwoLevelIndexSearch to
+      // test this format.
+      table_options.index_type = BlockBasedTableOptions::kTwoLevelIndexSearch;
+      // The top-level index in partition filters are also affected by format 4.
+      table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
+      table_options.partition_filters = true;
+      table_options.index_block_restart_interval = 8;
+      break;
+    }
    case kBlockBasedTableWithIndexRestartInterval: {
      table_options.index_block_restart_interval = 8;
      break;
@@ -460,26 +493,6 @@ Options DBTestBase::GetOptions(
      options.enable_write_thread_adaptive_yield = true;
      break;
    }
-    case kDirectIO: {
-      options.use_direct_reads = true;
-      options.use_direct_io_for_flush_and_compaction = true;
-      options.compaction_readahead_size = 2 * 1024 * 1024;
-#if !defined(OS_MACOSX) && !defined(OS_WIN) && !defined(OS_SOLARIS) && \
-    !defined(OS_AIX) && !defined(OS_OPENBSD)
-      rocksdb::SyncPoint::GetInstance()->SetCallBack(
-          "NewWritableFile:O_DIRECT", [&](void* arg) {
-            int* val = static_cast<int*>(arg);
-            *val &= ~O_DIRECT;
-          });
-      rocksdb::SyncPoint::GetInstance()->SetCallBack(
-          "NewRandomAccessFile:O_DIRECT", [&](void* arg) {
-            int* val = static_cast<int*>(arg);
-            *val &= ~O_DIRECT;
-          });
-      rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-#endif
-      break;
-    }
    case kPipelinedWrite: {
      options.enable_pipelined_write = true;
      break;
@@ -109,8 +109,6 @@ struct OptionsOverride {
  // These will be used only if filter_policy is set
  bool partition_filters = false;
  uint64_t metadata_block_size = 1024;
-  BlockBasedTableOptions::IndexType index_type =
-      BlockBasedTableOptions::IndexType::kBinarySearch;

  // Used as a bit mask of individual enums in which to skip an XF test point
  int skip_policy = 0;
@@ -317,6 +315,9 @@ class SpecialEnv : public EnvWrapper {
        }
      }
      uint64_t GetFileSize() override { return base_->GetFileSize(); }
+      Status Allocate(uint64_t offset, uint64_t len) override {
+        return base_->Allocate(offset, len);
+      }

     private:
      SpecialEnv* env_;
@@ -370,6 +371,9 @@ class SpecialEnv : public EnvWrapper {
      bool IsSyncThreadSafe() const override {
        return env_->is_wal_sync_thread_safe_.load();
      }
+      Status Allocate(uint64_t offset, uint64_t len) override {
+        return base_->Allocate(offset, len);
+      }

     private:
      SpecialEnv* env_;
@@ -575,7 +579,7 @@ class SpecialEnv : public EnvWrapper {

  std::atomic<bool> is_wal_sync_thread_safe_{true};

-  std::atomic<size_t> compaction_readahead_size_;
+  std::atomic<size_t> compaction_readahead_size_{};
 };

 class MockTimeEnv : public EnvWrapper {
@@ -694,13 +698,15 @@ class DBTestBase : public testing::Test {
    kConcurrentSkipList = 29,
    kPipelinedWrite = 30,
    kConcurrentWALWrites = 31,
-    kEnd = 32,
-    kDirectIO = 33,
-    kLevelSubcompactions = 34,
-    kUniversalSubcompactions = 35,
-    kBlockBasedTableWithIndexRestartInterval = 36,
-    kBlockBasedTableWithPartitionedIndex = 37,
-    kPartitionedFilterWithNewTableReaderForCompactions = 38,
+    kDirectIO,
+    kLevelSubcompactions,
+    kBlockBasedTableWithIndexRestartInterval,
+    kBlockBasedTableWithPartitionedIndex,
+    kBlockBasedTableWithPartitionedIndexFormat4,
+    kPartitionedFilterWithNewTableReaderForCompactions,
+    kUniversalSubcompactions,
+    // This must be the last line
+    kEnd,
  };

 public:
@@ -731,6 +737,13 @@ class DBTestBase : public testing::Test {
    kSkipMmapReads = 256,
  };

+  const int kRangeDelSkipConfigs =
+      // Plain tables do not support range deletions.
+      kSkipPlainTable |
+      // MmapReads disables the iterator pinning that RangeDelAggregator
+      // requires.
+      kSkipMmapReads;
+
  explicit DBTestBase(const std::string path);

  ~DBTestBase();
@@ -10,6 +10,7 @@
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #if !defined(ROCKSDB_LITE)
+#include "rocksdb/utilities/table_properties_collectors.h"
 #include "util/sync_point.h"

 namespace rocksdb {
@@ -40,6 +41,12 @@ class DBTestUniversalCompaction : public DBTestUniversalCompactionBase {
      DBTestUniversalCompactionBase("/db_universal_compaction_test") {}
 };

+class DBTestUniversalDeleteTrigCompaction : public DBTestBase {
+ public:
+  DBTestUniversalDeleteTrigCompaction()
+      : DBTestBase("/db_universal_compaction_test") {}
+};
+
 namespace {
 void VerifyCompactionResult(
    const ColumnFamilyMetaData& cf_meta,
@@ -655,7 +662,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTargetLevel) {
  ASSERT_EQ("0,0,0,0,1", FilesPerLevel(0));
 }

-
+#ifndef ROCKSDB_VALGRIND_RUN
 class DBTestUniversalCompactionMultiLevels
    : public DBTestUniversalCompactionBase {
 public:
@@ -691,6 +698,7 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionMultiLevels) {
    ASSERT_EQ(Get(1, Key(i % num_keys)), Key(i));
  }
 }
+
 // Tests universal compaction with trivial move enabled
 TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
  int32_t trivial_move = 0;
@@ -933,6 +941,7 @@ INSTANTIATE_TEST_CASE_P(DBTestUniversalCompactionParallel,
                        DBTestUniversalCompactionParallel,
                        ::testing::Combine(::testing::Values(1, 10),
                                           ::testing::Values(false)));
+#endif  // ROCKSDB_VALGRIND_RUN

 TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
  Options options = CurrentOptions();
@@ -1148,6 +1157,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio2) {
  ASSERT_LT(TotalSize(), 120000U * 12 * 0.8 + 120000 * 2);
 }

+#ifndef ROCKSDB_VALGRIND_RUN
 // Test that checks trivial move in universal compaction
 TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
  int32_t trivial_move = 0;
@@ -1240,6 +1250,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
+#endif  // ROCKSDB_VALGRIND_RUN

 TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
  Options options = CurrentOptions();
@@ -1352,7 +1363,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCFPathUse) {
  options.memtable_factory.reset(
      new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
  options.compaction_style = kCompactionStyleUniversal;
-  options.compaction_options_universal.size_ratio = 5;
+  options.compaction_options_universal.size_ratio = 10;
  options.write_buffer_size = 111 << 10;  // 114KB
  options.arena_block_size = 4 << 10;
  options.level0_file_num_compaction_trigger = 2;
@@ -1770,6 +1781,63 @@ TEST_P(DBTestUniversalCompaction, RecalculateScoreAfterPicking) {
  ASSERT_EQ(NumSortedRuns(), 5);
 }

+TEST_P(DBTestUniversalCompaction, FinalSortedRunCompactFilesConflict) {
+  // Regression test for conflict between:
+  // (1) Running CompactFiles including file in the final sorted run; and
+  // (2) Picking universal size-amp-triggered compaction, which always includes
+  //     the final sorted run.
+  if (exclusive_manual_compaction_) {
+    return;
+  }
+
+  Options opts = CurrentOptions();
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.compaction_options_universal.max_size_amplification_percent = 50;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compression = kNoCompression;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.max_background_compactions = 2;
+  opts.num_levels = num_levels_;
+  Reopen(opts);
+
+  // make sure compaction jobs can be parallelized
+  auto stop_token =
+      dbfull()->TEST_write_controler().GetCompactionPressureToken();
+
+  Put("key", "val");
+  Flush();
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  ASSERT_EQ(NumTableFilesAtLevel(num_levels_ - 1), 1);
+  ColumnFamilyMetaData cf_meta;
+  ColumnFamilyHandle* default_cfh = db_->DefaultColumnFamily();
+  dbfull()->GetColumnFamilyMetaData(default_cfh, &cf_meta);
+  ASSERT_EQ(1, cf_meta.levels[num_levels_ - 1].files.size());
+  std::string first_sst_filename =
+      cf_meta.levels[num_levels_ - 1].files[0].name;
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"CompactFilesImpl:0",
+        "DBTestUniversalCompaction:FinalSortedRunCompactFilesConflict:0"},
+       {"DBImpl::BackgroundCompaction():AfterPickCompaction",
+        "CompactFilesImpl:1"}});
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  port::Thread compact_files_thread([&]() {
+    ASSERT_OK(dbfull()->CompactFiles(CompactionOptions(), default_cfh,
+                                     {first_sst_filename}, num_levels_ - 1));
+  });
+
+  TEST_SYNC_POINT(
+      "DBTestUniversalCompaction:FinalSortedRunCompactFilesConflict:0");
+  for (int i = 0; i < 2; ++i) {
+    Put("key", "val");
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+
+  compact_files_thread.join();
+}
+
 INSTANTIATE_TEST_CASE_P(UniversalCompactionNumLevels, DBTestUniversalCompaction,
                        ::testing::Combine(::testing::Values(1, 3, 5),
                                           ::testing::Bool()));
@@ -1845,6 +1913,241 @@ INSTANTIATE_TEST_CASE_P(DBTestUniversalManualCompactionOutputPathId,
                        ::testing::Combine(::testing::Values(1, 8),
                                           ::testing::Bool()));

+TEST_F(DBTestUniversalDeleteTrigCompaction, BasicL0toL1) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  //  MoveFilesToLevel(6);
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, SingleLevel) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.num_levels = 1;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, MultipleLevels) {
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 4;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 500; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 500; i < 1000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 1000; i < 1500; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 1500; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+
+  for (i = 1999; i < 2333; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2333; i < 2666; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2666; i < 2999; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+  ASSERT_GT(NumTableFilesAtLevel(5), 0);
+
+  for (i = 1900; i < 2100; ++i) {
+    Delete(Key(i));
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(1));
+  ASSERT_EQ(0, NumTableFilesAtLevel(2));
+  ASSERT_EQ(0, NumTableFilesAtLevel(3));
+  ASSERT_EQ(0, NumTableFilesAtLevel(4));
+  ASSERT_EQ(0, NumTableFilesAtLevel(5));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, OverlappingL0) {
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 5;
+  opts.compression = kNoCompression;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2000; i < 3000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 3500; i < 4000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  for (i = 2900; i < 3100; ++i) {
+    Delete(Key(i));
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(6), 0);
+}
+
+TEST_F(DBTestUniversalDeleteTrigCompaction, IngestBehind) {
+  const int kNumKeys = 3000;
+  const int kWindowSize = 100;
+  const int kNumDelsTrigger = 90;
+
+  Options opts = CurrentOptions();
+  opts.table_properties_collector_factories.emplace_back(
+      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+  opts.compaction_style = kCompactionStyleUniversal;
+  opts.level0_file_num_compaction_trigger = 2;
+  opts.compression = kNoCompression;
+  opts.allow_ingest_behind = true;
+  opts.compaction_options_universal.size_ratio = 10;
+  opts.compaction_options_universal.min_merge_width = 2;
+  opts.compaction_options_universal.max_size_amplification_percent = 200;
+  Reopen(opts);
+
+  // add an L1 file to prevent tombstones from dropping due to obsolescence
+  // during flush
+  int i;
+  for (i = 0; i < 2000; ++i) {
+    Put(Key(i), "val");
+  }
+  Flush();
+  //  MoveFilesToLevel(6);
+  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+
+  for (i = 1999; i < kNumKeys; ++i) {
+    if (i >= kNumKeys - kWindowSize &&
+        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+      Delete(Key(i));
+    } else {
+      Put(Key(i), "val");
+    }
+  }
+  Flush();
+
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(6));
+  ASSERT_GT(NumTableFilesAtLevel(5), 0);
+}
+
 }  // namespace rocksdb

 #endif  // !defined(ROCKSDB_LITE)
@@ -18,6 +18,15 @@ namespace rocksdb {
 class DBWALTest : public DBTestBase {
 public:
  DBWALTest() : DBTestBase("/db_wal_test") {}
+
+#if defined(ROCKSDB_PLATFORM_POSIX)
+  uint64_t GetAllocatedFileSize(std::string file_name) {
+    struct stat sbuf;
+    int err = stat(file_name.c_str(), &sbuf);
+    assert(err == 0);
+    return sbuf.st_blocks * 512;
+  }
+#endif
 };

 // A SpecialEnv enriched to give more insight about deleted files
@@ -815,7 +824,7 @@ class RecoveryTestHelper {
      unique_ptr<WritableFile> file;
      ASSERT_OK(db_options.env->NewWritableFile(fname, &file, env_options));
      unique_ptr<WritableFileWriter> file_writer(
-          new WritableFileWriter(std::move(file), env_options));
+          new WritableFileWriter(std::move(file), fname, env_options));
      current_log_writer.reset(
          new log::Writer(std::move(file_writer), current_log_number,
                          db_options.recycle_log_file_num > 0));
@@ -1329,6 +1338,99 @@ TEST_F(DBWALTest, RecoverFromCorruptedWALWithoutFlush) {
  }
 }

+// Tests that total log size is recovered if we set
+// avoid_flush_during_recovery=true.
+// Flush should trigger if max_total_wal_size is reached.
+TEST_F(DBWALTest, RestoreTotalLogSizeAfterRecoverWithoutFlush) {
+  class TestFlushListener : public EventListener {
+   public:
+    std::atomic<int> count{0};
+
+    TestFlushListener() = default;
+
+    void OnFlushBegin(DB* /*db*/, const FlushJobInfo& flush_job_info) override {
+      count++;
+      assert(FlushReason::kWriteBufferManager == flush_job_info.flush_reason);
+    }
+  };
+  std::shared_ptr<TestFlushListener> test_listener =
+      std::make_shared<TestFlushListener>();
+
+  constexpr size_t kKB = 1024;
+  constexpr size_t kMB = 1024 * 1024;
+  Options options = CurrentOptions();
+  options.avoid_flush_during_recovery = true;
+  options.max_total_wal_size = 1 * kMB;
+  options.listeners.push_back(test_listener);
+  // Have to open DB in multi-CF mode to trigger flush when
+  // max_total_wal_size is reached.
+  CreateAndReopenWithCF({"one"}, options);
+  // Write some keys and we will end up with one log file which is slightly
+  // smaller than 1MB.
+  std::string value_100k(100 * kKB, 'v');
+  std::string value_300k(300 * kKB, 'v');
+  ASSERT_OK(Put(0, "foo", "v1"));
+  for (int i = 0; i < 9; i++) {
+    ASSERT_OK(Put(1, "key" + ToString(i), value_100k));
+  }
+  // Get log files before reopen.
+  VectorLogPtr log_files_before;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_before));
+  ASSERT_EQ(1, log_files_before.size());
+  uint64_t log_size_before = log_files_before[0]->SizeFileBytes();
+  ASSERT_GT(log_size_before, 900 * kKB);
+  ASSERT_LT(log_size_before, 1 * kMB);
+  ReopenWithColumnFamilies({"default", "one"}, options);
+  // Write one more value to make log larger than 1MB.
+  ASSERT_OK(Put(1, "bar", value_300k));
+  // Get log files again. A new log file will be opened.
+  VectorLogPtr log_files_after_reopen;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_after_reopen));
+  ASSERT_EQ(2, log_files_after_reopen.size());
+  ASSERT_EQ(log_files_before[0]->LogNumber(),
+            log_files_after_reopen[0]->LogNumber());
+  ASSERT_GT(log_files_after_reopen[0]->SizeFileBytes() +
+                log_files_after_reopen[1]->SizeFileBytes(),
+            1 * kMB);
+  // Write one more key to trigger flush.
+  ASSERT_OK(Put(0, "foo", "v2"));
+  dbfull()->TEST_WaitForFlushMemTable();
+  // Flushed two column families.
+  ASSERT_EQ(2, test_listener->count.load());
+}
+
+#if defined(ROCKSDB_PLATFORM_POSIX)
+#if defined(ROCKSDB_FALLOCATE_PRESENT)
+// Tests that we will truncate the preallocated space of the last log from
+// previous.
+TEST_F(DBWALTest, TruncateLastLogAfterRecoverWithoutFlush) {
+  constexpr size_t kKB = 1024;
+  Options options = CurrentOptions();
+  options.avoid_flush_during_recovery = true;
+  DestroyAndReopen(options);
+  size_t preallocated_size =
+      dbfull()->TEST_GetWalPreallocateBlockSize(options.write_buffer_size);
+  ASSERT_OK(Put("foo", "v1"));
+  VectorLogPtr log_files_before;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_before));
+  ASSERT_EQ(1, log_files_before.size());
+  auto& file_before = log_files_before[0];
+  ASSERT_LT(file_before->SizeFileBytes(), 1 * kKB);
+  // The log file has preallocated space.
+  ASSERT_GE(GetAllocatedFileSize(dbname_ + file_before->PathName()),
+            preallocated_size);
+  Reopen(options);
+  VectorLogPtr log_files_after;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_after));
+  ASSERT_EQ(1, log_files_after.size());
+  ASSERT_LT(log_files_after[0]->SizeFileBytes(), 1 * kKB);
+  // The preallocated space should be truncated.
+  ASSERT_LT(GetAllocatedFileSize(dbname_ + file_before->PathName()),
+            preallocated_size);
+}
+#endif  // ROCKSDB_FALLOCATE_PRESENT
+#endif  // ROCKSDB_PLATFORM_POSIX
+
 #endif  // ROCKSDB_LITE

 TEST_F(DBWALTest, WalTermTest) {
@@ -48,6 +48,8 @@ EntryType GetEntryType(ValueType value_type) {
      return kEntryMerge;
    case kTypeRangeDeletion:
      return kEntryRangeDeletion;
+    case kTypeBlobIndex:
+      return kEntryBlobIndex;
    default:
      return kEntryOther;
  }
@@ -55,7 +55,11 @@ enum ValueType : unsigned char {
  // is not mistakenly read by another, which would result into data
  // inconsistency.
  kTypeBeginPersistedPrepareXID = 0x12,  // WAL only.
-  kMaxValue = 0x7F                       // Not used for storing records.
+  // Similar to kTypeBeginPersistedPrepareXID, this is to ensure that WAL
+  // generated by WriteUnprepared write policy is not mistakenly read by
+  // another.
+  kTypeBeginUnprepareXID = 0x13,  // WAL only.
+  kMaxValue = 0x7F                // Not used for storing records.
 };

 // Defined in dbformat.cc
@@ -136,10 +140,14 @@ inline Slice ExtractUserKey(const Slice& internal_key) {
  return Slice(internal_key.data(), internal_key.size() - 8);
 }

-inline ValueType ExtractValueType(const Slice& internal_key) {
+inline uint64_t ExtractInternalKeyFooter(const Slice& internal_key) {
  assert(internal_key.size() >= 8);
  const size_t n = internal_key.size();
-  uint64_t num = DecodeFixed64(internal_key.data() + n - 8);
+  return DecodeFixed64(internal_key.data() + n - 8);
+}
+
+inline ValueType ExtractValueType(const Slice& internal_key) {
+  uint64_t num = ExtractInternalKeyFooter(internal_key);
  unsigned char c = num & 0xff;
  return static_cast<ValueType>(c);
 }
@@ -338,6 +346,12 @@ class IterKey {

  ~IterKey() { ResetBuffer(); }

+  // The bool will be picked up by the next calls to SetKey
+  void SetIsUserKey(bool is_user_key) { is_user_key_ = is_user_key; }
+
+  // Returns the key in whichever format that was provided to KeyIter
+  Slice GetKey() const { return Slice(key_, key_size_); }
+
  Slice GetInternalKey() const {
    assert(!IsUserKey());
    return Slice(key_, key_size_);
@@ -387,6 +401,11 @@ class IterKey {
    key_size_ = total_size;
  }

+  Slice SetKey(const Slice& key, bool copy = true) {
+    // is_user_key_ expected to be set already via SetIsUserKey
+    return SetKeyImpl(key, copy);
+  }
+
  Slice SetUserKey(const Slice& key, bool copy = true) {
    is_user_key_ = true;
    return SetKeyImpl(key, copy);
@@ -602,9 +621,15 @@ struct RangeTombstone {
    return InternalKey(start_key_, seq_, kTypeRangeDeletion);
  }

+  // The tombstone end-key is exclusive, so we generate an internal-key here
+  // which has a similar property. Using kMaxSequenceNumber guarantees that
+  // the returned internal-key will compare less than any other internal-key
+  // with the same user-key. This in turn guarantees that the serialized
+  // end-key for a tombstone such as [a-b] will compare less than the key "b".
+  //
  // be careful to use SerializeEndKey(), allocates new memory
  InternalKey SerializeEndKey() const {
-    return InternalKey(end_key_, seq_, kTypeRangeDeletion);
+    return InternalKey(end_key_, kMaxSequenceNumber, kTypeRangeDeletion);
  }
 };

@@ -192,6 +192,13 @@ TEST_F(FormatTest, UpdateInternalKey) {
  ASSERT_EQ(new_val_type, decoded.type);
 }

+TEST_F(FormatTest, RangeTombstoneSerializeEndKey) {
+  RangeTombstone t("a", "b", 2);
+  InternalKey k("b", 3, kTypeValue);
+  const InternalKeyComparator cmp(BytewiseComparator());
+  ASSERT_LT(cmp.Compare(t.SerializeEndKey(), k), 0);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -45,7 +45,7 @@ class DeleteFileTest : public testing::Test {
    options_.max_bytes_for_level_base = 1024*1024*1000;
    options_.WAL_ttl_seconds = 300; // Used to test log files
    options_.WAL_size_limit_MB = 1024; // Used to test log files
-    dbname_ = test::TmpDir() + "/deletefile_test";
+    dbname_ = test::PerThreadDBPath("deletefile_test");
    options_.wal_dir = dbname_ + "/wal_files";

    // clean up all the files that might have been there before
@@ -0,0 +1,345 @@
+//  Copyright (c) 2018-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/error_handler.h"
+#include "db/db_impl.h"
+#include "db/event_helpers.h"
+#include "util/sst_file_manager_impl.h"
+
+namespace rocksdb {
+
+// Maps to help decide the severity of an error based on the
+// BackgroundErrorReason, Code, SubCode and whether db_options.paranoid_checks
+// is set or not. There are 3 maps, going from most specific to least specific
+// (i.e from all 4 fields in a tuple to only the BackgroundErrorReason and
+// paranoid_checks). The less specific map serves as a catch all in case we miss
+// a specific error code or subcode.
+std::map<std::tuple<BackgroundErrorReason, Status::Code, Status::SubCode, bool>,
+         Status::Severity>
+    ErrorSeverityMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         true),
+         Status::Severity::kSoftError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, Status::SubCode::kSpaceLimit,
+                         true),
+         Status::Severity::kHardError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kNoSpace, true),
+         Status::Severity::kHardError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kNoSpace, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
+                         Status::SubCode::kSpaceLimit, true),
+         Status::Severity::kHardError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         true),
+         Status::Severity::kHardError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, Status::SubCode::kNoSpace,
+                         false),
+         Status::Severity::kHardError},
+};
+
+std::map<std::tuple<BackgroundErrorReason, Status::Code, bool>, Status::Severity>
+    DefaultErrorSeverityMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kFlush,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kCorruption, true),
+         Status::Severity::kUnrecoverableError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kCorruption, false),
+         Status::Severity::kNoError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, true),
+         Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback,
+                         Status::Code::kIOError, false),
+         Status::Severity::kNoError},
+};
+
+std::map<std::tuple<BackgroundErrorReason, bool>, Status::Severity>
+    DefaultReasonMap = {
+        // Errors during BG compaction
+        {std::make_tuple(BackgroundErrorReason::kCompaction, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kCompaction, false),
+          Status::Severity::kNoError},
+        // Errors during BG flush
+        {std::make_tuple(BackgroundErrorReason::kFlush, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kFlush, false),
+          Status::Severity::kNoError},
+        // Errors during Write
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kWriteCallback, false),
+          Status::Severity::kFatalError},
+        // Errors during Memtable update
+        {std::make_tuple(BackgroundErrorReason::kMemTable, true),
+          Status::Severity::kFatalError},
+        {std::make_tuple(BackgroundErrorReason::kMemTable, false),
+          Status::Severity::kFatalError},
+};
+
+void ErrorHandler::CancelErrorRecovery() {
+#ifndef ROCKSDB_LITE
+  db_mutex_->AssertHeld();
+
+  // We'll release the lock before calling sfm, so make sure no new
+  // recovery gets scheduled at that point
+  auto_recovery_ = false;
+  SstFileManagerImpl* sfm = reinterpret_cast<SstFileManagerImpl*>(
+      db_options_.sst_file_manager.get());
+  if (sfm) {
+    // This may or may not cancel a pending recovery
+    db_mutex_->Unlock();
+    bool cancelled = sfm->CancelErrorRecovery(this);
+    db_mutex_->Lock();
+    if (cancelled) {
+      recovery_in_prog_ = false;
+    }
+  }
+#endif
+}
+
+// This is the main function for looking at an error during a background
+// operation and deciding the severity, and error recovery strategy. The high
+// level algorithm is as follows -
+// 1. Classify the severity of the error based on the ErrorSeverityMap,
+//    DefaultErrorSeverityMap and DefaultReasonMap defined earlier
+// 2. Call a Status code specific override function to adjust the severity
+//    if needed. The reason for this is our ability to recover may depend on
+//    the exact options enabled in DBOptions
+// 3. Determine if auto recovery is possible. A listener notification callback
+//    is called, which can disable the auto recovery even if we decide its
+//    feasible
+// 4. For Status::NoSpace() errors, rely on SstFileManagerImpl to control
+//    the actual recovery. If no sst file manager is specified in DBOptions,
+//    a default one is allocated during DB::Open(), so there will always be
+//    one.
+// This can also get called as part of a recovery operation. In that case, we
+// also track the error seperately in recovery_error_ so we can tell in the
+// end whether recovery succeeded or not
+Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reason) {
+  db_mutex_->AssertHeld();
+
+  if (bg_err.ok()) {
+    return Status::OK();
+  }
+
+  // Check if recovery is currently in progress. If it is, we will save this
+  // error so we can check it at the end to see if recovery succeeded or not
+  if (recovery_in_prog_ && recovery_error_.ok()) {
+    recovery_error_ = bg_err;
+  }
+
+  bool paranoid = db_options_.paranoid_checks;
+  Status::Severity sev = Status::Severity::kFatalError;
+  Status new_bg_err;
+  bool found = false;
+
+  {
+    auto entry = ErrorSeverityMap.find(std::make_tuple(reason, bg_err.code(),
+          bg_err.subcode(), paranoid));
+    if (entry != ErrorSeverityMap.end()) {
+      sev = entry->second;
+      found = true;
+    }
+  }
+
+  if (!found) {
+    auto entry = DefaultErrorSeverityMap.find(std::make_tuple(reason,
+          bg_err.code(), paranoid));
+    if (entry != DefaultErrorSeverityMap.end()) {
+      sev = entry->second;
+      found = true;
+    }
+  }
+
+  if (!found) {
+    auto entry = DefaultReasonMap.find(std::make_tuple(reason, paranoid));
+    if (entry != DefaultReasonMap.end()) {
+      sev = entry->second;
+    }
+  }
+
+  new_bg_err = Status(bg_err, sev);
+
+  bool auto_recovery = auto_recovery_;
+  if (new_bg_err.severity() >= Status::Severity::kFatalError && auto_recovery) {
+    auto_recovery = false;
+    ;
+  }
+
+  // Allow some error specific overrides
+  if (new_bg_err == Status::NoSpace()) {
+    new_bg_err = OverrideNoSpaceError(new_bg_err, &auto_recovery);
+  }
+
+  if (!new_bg_err.ok()) {
+    Status s = new_bg_err;
+    EventHelpers::NotifyOnBackgroundError(db_options_.listeners, reason, &s,
+                                          db_mutex_, &auto_recovery);
+    if (!s.ok() && (s.severity() > bg_error_.severity())) {
+      bg_error_ = s;
+    } else {
+      // This error is less severe than previously encountered error. Don't
+      // take any further action
+      return bg_error_;
+    }
+  }
+
+  if (auto_recovery) {
+    recovery_in_prog_ = true;
+
+    // Kick-off error specific recovery
+    if (bg_error_ == Status::NoSpace()) {
+      RecoverFromNoSpace();
+    }
+  }
+  return bg_error_;
+}
+
+Status ErrorHandler::OverrideNoSpaceError(Status bg_error,
+                                          bool* auto_recovery) {
+#ifndef ROCKSDB_LITE
+  if (bg_error.severity() >= Status::Severity::kFatalError) {
+    return bg_error;
+  }
+
+  if (db_options_.sst_file_manager.get() == nullptr) {
+    // We rely on SFM to poll for enough disk space and recover
+    *auto_recovery = false;
+    return bg_error;
+  }
+
+  if (db_options_.allow_2pc &&
+      (bg_error.severity() <= Status::Severity::kSoftError)) {
+    // Don't know how to recover, as the contents of the current WAL file may
+    // be inconsistent, and it may be needed for 2PC. If 2PC is not enabled,
+    // we can just flush the memtable and discard the log
+    *auto_recovery = false;
+    return Status(bg_error, Status::Severity::kFatalError);
+  }
+
+  {
+    uint64_t free_space;
+    if (db_options_.env->GetFreeSpace(db_options_.db_paths[0].path,
+                                      &free_space) == Status::NotSupported()) {
+      *auto_recovery = false;
+    }
+  }
+
+  return bg_error;
+#else
+  (void)auto_recovery;
+  return Status(bg_error, Status::Severity::kFatalError);
+#endif
+}
+
+void ErrorHandler::RecoverFromNoSpace() {
+#ifndef ROCKSDB_LITE
+  SstFileManagerImpl* sfm =
+      reinterpret_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
+
+  // Inform SFM of the error, so it can kick-off the recovery
+  if (sfm) {
+    sfm->StartErrorRecovery(this, bg_error_);
+  }
+#endif
+}
+
+Status ErrorHandler::ClearBGError() {
+#ifndef ROCKSDB_LITE
+  db_mutex_->AssertHeld();
+
+  // Signal that recovery succeeded
+  if (recovery_error_.ok()) {
+    Status old_bg_error = bg_error_;
+    bg_error_ = Status::OK();
+    recovery_in_prog_ = false;
+    EventHelpers::NotifyOnErrorRecoveryCompleted(db_options_.listeners,
+                                                 old_bg_error, db_mutex_);
+  }
+  return recovery_error_;
+#else
+  return bg_error_;
+#endif
+}
+
+Status ErrorHandler::RecoverFromBGError(bool is_manual) {
+#ifndef ROCKSDB_LITE
+  InstrumentedMutexLock l(db_mutex_);
+  if (is_manual) {
+    // If its a manual recovery and there's a background recovery in progress
+    // return busy status
+    if (recovery_in_prog_) {
+      return Status::Busy();
+    }
+    recovery_in_prog_ = true;
+  }
+
+  if (bg_error_.severity() == Status::Severity::kSoftError) {
+    // Simply clear the background error and return
+    recovery_error_ = Status::OK();
+    return ClearBGError();
+  }
+
+  // Reset recovery_error_. We will use this to record any errors that happen
+  // during the recovery process. While recovering, the only operations that
+  // can generate background errors should be the flush operations
+  recovery_error_ = Status::OK();
+  Status s = db_->ResumeImpl();
+  // For manual recover, shutdown, and fatal error  cases, set
+  // recovery_in_prog_ to false. For automatic background recovery, leave it
+  // as is regardless of success or failure as it will be retried
+  if (is_manual || s.IsShutdownInProgress() ||
+      bg_error_.severity() >= Status::Severity::kFatalError) {
+    recovery_in_prog_ = false;
+  }
+  return s;
+#else
+  (void)is_manual;
+  return bg_error_;
+#endif
+}
+}
@@ -0,0 +1,75 @@
+//  Copyright (c) 2018-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 "monitoring/instrumented_mutex.h"
+#include "options/db_options.h"
+#include "rocksdb/listener.h"
+#include "rocksdb/status.h"
+
+namespace rocksdb {
+
+class DBImpl;
+
+class ErrorHandler {
+  public:
+   ErrorHandler(DBImpl* db, const ImmutableDBOptions& db_options,
+                InstrumentedMutex* db_mutex)
+       : db_(db),
+         db_options_(db_options),
+         bg_error_(Status::OK()),
+         recovery_error_(Status::OK()),
+         db_mutex_(db_mutex),
+         auto_recovery_(false),
+         recovery_in_prog_(false) {}
+   ~ErrorHandler() {}
+
+   void EnableAutoRecovery() { auto_recovery_ = true; }
+
+   Status::Severity GetErrorSeverity(BackgroundErrorReason reason,
+                                     Status::Code code,
+                                     Status::SubCode subcode);
+
+   Status SetBGError(const Status& bg_err, BackgroundErrorReason reason);
+
+   Status GetBGError() { return bg_error_; }
+
+   Status GetRecoveryError() { return recovery_error_; }
+
+   Status ClearBGError();
+
+   bool IsDBStopped() {
+     return !bg_error_.ok() &&
+            bg_error_.severity() >= Status::Severity::kHardError;
+    }
+
+    bool IsBGWorkStopped() {
+      return !bg_error_.ok() &&
+             (bg_error_.severity() >= Status::Severity::kHardError ||
+              !auto_recovery_);
+    }
+
+    bool IsRecoveryInProgress() { return recovery_in_prog_; }
+
+    Status RecoverFromBGError(bool is_manual = false);
+    void CancelErrorRecovery();
+
+   private:
+    DBImpl* db_;
+    const ImmutableDBOptions& db_options_;
+    Status bg_error_;
+    // A seperate Status variable used to record any errors during the
+    // recovery process from hard errors
+    Status recovery_error_;
+    InstrumentedMutex* db_mutex_;
+    // A flag indicating whether automatic recovery from errors is enabled
+    bool auto_recovery_;
+    bool recovery_in_prog_;
+
+    Status OverrideNoSpaceError(Status bg_error, bool* auto_recovery);
+    void RecoverFromNoSpace();
+};
+
+}
@@ -0,0 +1,691 @@
+//  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.
+#ifndef ROCKSDB_LITE
+
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+#include "rocksdb/perf_context.h"
+#include "rocksdb/sst_file_manager.h"
+#include "util/fault_injection_test_env.h"
+#if !defined(ROCKSDB_LITE)
+#include "util/sync_point.h"
+#endif
+
+namespace rocksdb {
+
+class DBErrorHandlingTest : public DBTestBase {
+ public:
+  DBErrorHandlingTest() : DBTestBase("/db_error_handling_test") {}
+};
+
+class DBErrorHandlingEnv : public EnvWrapper {
+  public:
+    DBErrorHandlingEnv() : EnvWrapper(Env::Default()),
+      trig_no_space(false), trig_io_error(false) {}
+
+    void SetTrigNoSpace() {trig_no_space = true;}
+    void SetTrigIoError() {trig_io_error = true;}
+  private:
+    bool trig_no_space;
+    bool trig_io_error;
+};
+
+class ErrorHandlerListener : public EventListener {
+ public:
+  ErrorHandlerListener()
+      : mutex_(),
+        cv_(&mutex_),
+        no_auto_recovery_(false),
+        recovery_complete_(false),
+        file_creation_started_(false),
+        override_bg_error_(false),
+        file_count_(0),
+        fault_env_(nullptr) {}
+
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& /*ti*/) override {
+    InstrumentedMutexLock l(&mutex_);
+    file_creation_started_ = true;
+    if (file_count_ > 0) {
+      if (--file_count_ == 0) {
+        fault_env_->SetFilesystemActive(false, file_creation_error_);
+        file_creation_error_ = Status::OK();
+      }
+    }
+    cv_.SignalAll();
+  }
+
+  void OnErrorRecoveryBegin(BackgroundErrorReason /*reason*/,
+                            Status /*bg_error*/,
+                            bool* auto_recovery) override {
+    if (*auto_recovery && no_auto_recovery_) {
+      *auto_recovery = false;
+    }
+  }
+
+  void OnErrorRecoveryCompleted(Status /*old_bg_error*/) override {
+    InstrumentedMutexLock l(&mutex_);
+    recovery_complete_ = true;
+    cv_.SignalAll();
+  }
+
+  bool WaitForRecovery(uint64_t /*abs_time_us*/) {
+    InstrumentedMutexLock l(&mutex_);
+    while (!recovery_complete_) {
+      cv_.Wait(/*abs_time_us*/);
+    }
+    if (recovery_complete_) {
+      recovery_complete_ = false;
+      return true;
+    }
+    return false;
+  }
+
+  void WaitForTableFileCreationStarted(uint64_t /*abs_time_us*/) {
+    InstrumentedMutexLock l(&mutex_);
+    while (!file_creation_started_) {
+      cv_.Wait(/*abs_time_us*/);
+    }
+    file_creation_started_ = false;
+  }
+
+  void OnBackgroundError(BackgroundErrorReason /*reason*/,
+                         Status* bg_error) override {
+    if (override_bg_error_) {
+      *bg_error = bg_error_;
+      override_bg_error_ = false;
+    }
+  }
+
+  void EnableAutoRecovery(bool enable = true) { no_auto_recovery_ = !enable; }
+
+  void OverrideBGError(Status bg_err) {
+    bg_error_ = bg_err;
+    override_bg_error_ = true;
+  }
+
+  void InjectFileCreationError(FaultInjectionTestEnv* env, int file_count,
+                               Status s) {
+    fault_env_ = env;
+    file_count_ = file_count;
+    file_creation_error_ = s;
+  }
+
+ private:
+  InstrumentedMutex mutex_;
+  InstrumentedCondVar cv_;
+  bool no_auto_recovery_;
+  bool recovery_complete_;
+  bool file_creation_started_;
+  bool override_bg_error_;
+  int file_count_;
+  Status file_creation_error_;
+  Status bg_error_;
+  FaultInjectionTestEnv* fault_env_;
+};
+
+TEST_F(DBErrorHandlingTest, FLushWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+
+  listener->EnableAutoRecovery(false);
+  DestroyAndReopen(options);
+
+  Put(Key(0), "val");
+  SyncPoint::GetInstance()->SetCallBack(
+      "FlushJob::Start", [&](void *) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_EQ(s, Status::OK());
+
+  Reopen(options);
+  ASSERT_EQ("val", Get(Key(0)));
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, CompactionWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.level0_file_num_compaction_trigger = 2;
+  options.listeners.emplace_back(listener);
+  options.env = fault_env.get();
+  Status s;
+  DestroyAndReopen(options);
+
+  Put(Key(0), "va;");
+  Put(Key(2), "va;");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  listener->OverrideBGError(
+      Status(Status::NoSpace(), Status::Severity::kHardError)
+      );
+  listener->EnableAutoRecovery(false);
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BackgroundCallCompaction:0", [&](void *) {
+      fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put(Key(1), "val");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  s = dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_EQ(s, Status::OK());
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, CorruptionError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.level0_file_num_compaction_trigger = 2;
+  options.env = fault_env.get();
+  Status s;
+  DestroyAndReopen(options);
+
+  Put(Key(0), "va;");
+  Put(Key(2), "va;");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "BackgroundCallCompaction:0", [&](void *) {
+      fault_env->SetFilesystemActive(false, Status::Corruption("Corruption"));
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  Put(Key(1), "val");
+  s = Flush();
+  ASSERT_EQ(s, Status::OK());
+
+  s = dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kUnrecoverableError);
+
+  fault_env->SetFilesystemActive(true);
+  s = dbfull()->Resume();
+  ASSERT_NE(s, Status::OK());
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, AutoRecoverFlushError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  Put(Key(0), "val");
+  SyncPoint::GetInstance()->SetCallBack("FlushJob::Start", [&](void*) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+
+  s = Put(Key(1), "val");
+  ASSERT_EQ(s, Status::OK());
+
+  Reopen(options);
+  ASSERT_EQ("val", Get(Key(0)));
+  ASSERT_EQ("val", Get(Key(1)));
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, FailRecoverFlushError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  Put(Key(0), "val");
+  SyncPoint::GetInstance()->SetCallBack("FlushJob::Start", [&](void*) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  // We should be able to shutdown the database while auto recovery is going
+  // on in the background
+  Close();
+  DestroyDB(dbname_, options);
+}
+
+TEST_F(DBErrorHandlingTest, WALWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.writable_file_max_buffer_size = 32768;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+  Random rnd(301);
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  {
+    WriteBatch batch;
+
+    for (auto i = 0; i<100; ++i) {
+      batch.Put(Key(i), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(dbfull()->Write(wopts, &batch), Status::OK());
+  };
+
+  {
+    WriteBatch batch;
+    int write_error = 0;
+
+    for (auto i = 100; i<199; ++i) {
+      batch.Put(Key(i), RandomString(&rnd, 1024));
+    }
+
+    SyncPoint::GetInstance()->SetCallBack("WritableFileWriter::Append:BeforePrepareWrite", [&](void*) {
+      write_error++;
+      if (write_error > 2) {
+        fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+      }
+    });
+    SyncPoint::GetInstance()->EnableProcessing();
+    WriteOptions wopts;
+    wopts.sync = true;
+    s = dbfull()->Write(wopts, &batch);
+    ASSERT_EQ(s, s.NoSpace());
+  }
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+  for (auto i=0; i<199; ++i) {
+    if (i < 100) {
+      ASSERT_NE(Get(Key(i)), "NOT_FOUND");
+    } else {
+      ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+    }
+  }
+  Reopen(options);
+  for (auto i=0; i<199; ++i) {
+    if (i < 100) {
+      ASSERT_NE(Get(Key(i)), "NOT_FOUND");
+    } else {
+      ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+    }
+  }
+  Close();
+}
+
+TEST_F(DBErrorHandlingTest, MultiCFWALWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.writable_file_max_buffer_size = 32768;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+  Random rnd(301);
+
+  listener->EnableAutoRecovery();
+  CreateAndReopenWithCF({"one", "two", "three"}, options);
+
+  {
+    WriteBatch batch;
+
+    for (auto i = 1; i < 4; ++i) {
+      for (auto j = 0; j < 100; ++j) {
+        batch.Put(handles_[i], Key(j), RandomString(&rnd, 1024));
+      }
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(dbfull()->Write(wopts, &batch), Status::OK());
+  };
+
+  {
+    WriteBatch batch;
+    int write_error = 0;
+
+    // Write to one CF
+    for (auto i = 100; i < 199; ++i) {
+      batch.Put(handles_[2], Key(i), RandomString(&rnd, 1024));
+    }
+
+    SyncPoint::GetInstance()->SetCallBack(
+        "WritableFileWriter::Append:BeforePrepareWrite", [&](void*) {
+          write_error++;
+          if (write_error > 2) {
+            fault_env->SetFilesystemActive(false,
+                                           Status::NoSpace("Out of space"));
+          }
+        });
+    SyncPoint::GetInstance()->EnableProcessing();
+    WriteOptions wopts;
+    wopts.sync = true;
+    s = dbfull()->Write(wopts, &batch);
+    ASSERT_EQ(s, s.NoSpace());
+  }
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+
+  for (auto i = 1; i < 4; ++i) {
+    // Every CF should have been flushed
+    ASSERT_EQ(NumTableFilesAtLevel(0, i), 1);
+  }
+
+  for (auto i = 1; i < 4; ++i) {
+    for (auto j = 0; j < 199; ++j) {
+      if (j < 100) {
+        ASSERT_NE(Get(i, Key(j)), "NOT_FOUND");
+      } else {
+        ASSERT_EQ(Get(i, Key(j)), "NOT_FOUND");
+      }
+    }
+  }
+  ReopenWithColumnFamilies({"default", "one", "two", "three"}, options);
+  for (auto i = 1; i < 4; ++i) {
+    for (auto j = 0; j < 199; ++j) {
+      if (j < 100) {
+        ASSERT_NE(Get(i, Key(j)), "NOT_FOUND");
+      } else {
+        ASSERT_EQ(Get(i, Key(j)), "NOT_FOUND");
+      }
+    }
+  }
+  Close();
+}
+
+TEST_F(DBErrorHandlingTest, MultiDBCompactionError) {
+  FaultInjectionTestEnv* def_env = new FaultInjectionTestEnv(Env::Default());
+  std::vector<std::unique_ptr<FaultInjectionTestEnv>> fault_env;
+  std::vector<Options> options;
+  std::vector<std::shared_ptr<ErrorHandlerListener>> listener;
+  std::vector<DB*> db;
+  std::shared_ptr<SstFileManager> sfm(NewSstFileManager(def_env));
+  int kNumDbInstances = 3;
+  Random rnd(301);
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    listener.emplace_back(new ErrorHandlerListener());
+    options.emplace_back(GetDefaultOptions());
+    fault_env.emplace_back(new FaultInjectionTestEnv(Env::Default()));
+    options[i].create_if_missing = true;
+    options[i].level0_file_num_compaction_trigger = 2;
+    options[i].writable_file_max_buffer_size = 32768;
+    options[i].env = fault_env[i].get();
+    options[i].listeners.emplace_back(listener[i]);
+    options[i].sst_file_manager = sfm;
+    DB* dbptr;
+    char buf[16];
+
+    listener[i]->EnableAutoRecovery();
+    // Setup for returning error for the 3rd SST, which would be level 1
+    listener[i]->InjectFileCreationError(fault_env[i].get(), 3,
+                                         Status::NoSpace("Out of space"));
+    snprintf(buf, sizeof(buf), "_%d", i);
+    DestroyDB(dbname_ + std::string(buf), options[i]);
+    ASSERT_EQ(DB::Open(options[i], dbname_ + std::string(buf), &dbptr),
+              Status::OK());
+    db.emplace_back(dbptr);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    for (auto j = 0; j <= 100; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  def_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    // Write to one CF
+    for (auto j = 100; j < 199; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    Status s = static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true);
+    ASSERT_EQ(s.severity(), Status::Severity::kSoftError);
+    fault_env[i]->SetFilesystemActive(true);
+  }
+
+  def_env->SetFilesystemActive(true);
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    std::string prop;
+    ASSERT_EQ(listener[i]->WaitForRecovery(5000000), true);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(0), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 0);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(1), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 1);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    char buf[16];
+    snprintf(buf, sizeof(buf), "_%d", i);
+    delete db[i];
+    fault_env[i]->SetFilesystemActive(true);
+    if (getenv("KEEP_DB")) {
+      printf("DB is still at %s%s\n", dbname_.c_str(), buf);
+    } else {
+      Status s = DestroyDB(dbname_ + std::string(buf), options[i]);
+    }
+  }
+  options.clear();
+  sfm.reset();
+  delete def_env;
+}
+
+TEST_F(DBErrorHandlingTest, MultiDBVariousErrors) {
+  FaultInjectionTestEnv* def_env = new FaultInjectionTestEnv(Env::Default());
+  std::vector<std::unique_ptr<FaultInjectionTestEnv>> fault_env;
+  std::vector<Options> options;
+  std::vector<std::shared_ptr<ErrorHandlerListener>> listener;
+  std::vector<DB*> db;
+  std::shared_ptr<SstFileManager> sfm(NewSstFileManager(def_env));
+  int kNumDbInstances = 3;
+  Random rnd(301);
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    listener.emplace_back(new ErrorHandlerListener());
+    options.emplace_back(GetDefaultOptions());
+    fault_env.emplace_back(new FaultInjectionTestEnv(Env::Default()));
+    options[i].create_if_missing = true;
+    options[i].level0_file_num_compaction_trigger = 2;
+    options[i].writable_file_max_buffer_size = 32768;
+    options[i].env = fault_env[i].get();
+    options[i].listeners.emplace_back(listener[i]);
+    options[i].sst_file_manager = sfm;
+    DB* dbptr;
+    char buf[16];
+
+    listener[i]->EnableAutoRecovery();
+    switch (i) {
+      case 0:
+        // Setup for returning error for the 3rd SST, which would be level 1
+        listener[i]->InjectFileCreationError(fault_env[i].get(), 3,
+                                             Status::NoSpace("Out of space"));
+        break;
+      case 1:
+        // Setup for returning error after the 1st SST, which would result
+        // in a hard error
+        listener[i]->InjectFileCreationError(fault_env[i].get(), 2,
+                                             Status::NoSpace("Out of space"));
+        break;
+      default:
+        break;
+    }
+    snprintf(buf, sizeof(buf), "_%d", i);
+    DestroyDB(dbname_ + std::string(buf), options[i]);
+    ASSERT_EQ(DB::Open(options[i], dbname_ + std::string(buf), &dbptr),
+              Status::OK());
+    db.emplace_back(dbptr);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    for (auto j = 0; j <= 100; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  def_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    // Write to one CF
+    for (auto j = 100; j < 199; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    if (i != 1) {
+      ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+    } else {
+      ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::NoSpace());
+    }
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    Status s = static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true);
+    switch (i) {
+      case 0:
+        ASSERT_EQ(s.severity(), Status::Severity::kSoftError);
+        break;
+      case 1:
+        ASSERT_EQ(s.severity(), Status::Severity::kHardError);
+        break;
+      case 2:
+        ASSERT_EQ(s, Status::OK());
+        break;
+    }
+    fault_env[i]->SetFilesystemActive(true);
+  }
+
+  def_env->SetFilesystemActive(true);
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    std::string prop;
+    if (i < 2) {
+      ASSERT_EQ(listener[i]->WaitForRecovery(5000000), true);
+    }
+    if (i == 1) {
+      ASSERT_EQ(static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true),
+                Status::OK());
+    }
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(0), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 0);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(1), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 1);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    char buf[16];
+    snprintf(buf, sizeof(buf), "_%d", i);
+    fault_env[i]->SetFilesystemActive(true);
+    delete db[i];
+    if (getenv("KEEP_DB")) {
+      printf("DB is still at %s%s\n", dbname_.c_str(), buf);
+    } else {
+      DestroyDB(dbname_ + std::string(buf), options[i]);
+    }
+  }
+  options.clear();
+  delete def_env;
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
+
+#else
+#include <stdio.h>
+
+int main(int /*argc*/, char** /*argv*/) {
+  fprintf(stderr, "SKIPPED as Cuckoo table is not supported in ROCKSDB_LITE\n");
+  return 0;
+}
+
+#endif  // ROCKSDB_LITE
@@ -40,8 +40,8 @@ void EventHelpers::NotifyTableFileCreationStarted(

 void EventHelpers::NotifyOnBackgroundError(
    const std::vector<std::shared_ptr<EventListener>>& listeners,
-    BackgroundErrorReason reason, Status* bg_error,
-    InstrumentedMutex* db_mutex) {
+    BackgroundErrorReason reason, Status* bg_error, InstrumentedMutex* db_mutex,
+    bool* auto_recovery) {
 #ifndef ROCKSDB_LITE
  if (listeners.size() == 0U) {
    return;
@@ -51,6 +51,9 @@ void EventHelpers::NotifyOnBackgroundError(
  db_mutex->Unlock();
  for (auto& listener : listeners) {
    listener->OnBackgroundError(reason, bg_error);
+    if (*auto_recovery) {
+      listener->OnErrorRecoveryBegin(reason, *bg_error, auto_recovery);
+    }
  }
  db_mutex->Lock();
 #else
@@ -58,6 +61,7 @@ void EventHelpers::NotifyOnBackgroundError(
  (void)reason;
  (void)bg_error;
  (void)db_mutex;
+  (void)auto_recovery;
 #endif  // ROCKSDB_LITE
 }

@@ -167,4 +171,25 @@ void EventHelpers::LogAndNotifyTableFileDeletion(
 #endif  // !ROCKSDB_LITE
 }

+void EventHelpers::NotifyOnErrorRecoveryCompleted(
+    const std::vector<std::shared_ptr<EventListener>>& listeners,
+    Status old_bg_error, InstrumentedMutex* db_mutex) {
+#ifndef ROCKSDB_LITE
+  if (listeners.size() == 0U) {
+    return;
+  }
+  db_mutex->AssertHeld();
+  // release lock while notifying events
+  db_mutex->Unlock();
+  for (auto& listener : listeners) {
+    listener->OnErrorRecoveryCompleted(old_bg_error);
+  }
+  db_mutex->Lock();
+#else
+  (void)listeners;
+  (void)old_bg_error;
+  (void)db_mutex;
+#endif  // ROCKSDB_LITE
+}
+
 }  // namespace rocksdb
@@ -28,7 +28,7 @@ class EventHelpers {
  static void NotifyOnBackgroundError(
      const std::vector<std::shared_ptr<EventListener>>& listeners,
      BackgroundErrorReason reason, Status* bg_error,
-      InstrumentedMutex* db_mutex);
+      InstrumentedMutex* db_mutex, bool* auto_recovery);
  static void LogAndNotifyTableFileCreationFinished(
      EventLogger* event_logger,
      const std::vector<std::shared_ptr<EventListener>>& listeners,
@@ -41,6 +41,9 @@ class EventHelpers {
      uint64_t file_number, const std::string& file_path,
      const Status& status, const std::string& db_name,
      const std::vector<std::shared_ptr<EventListener>>& listeners);
+  static void NotifyOnErrorRecoveryCompleted(
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      Status bg_error, InstrumentedMutex* db_mutex);

 private:
  static void LogAndNotifyTableFileCreation(
@@ -39,7 +39,8 @@ class ExternalSSTFileBasicTest : public DBTestBase {

  Status GenerateAndAddExternalFile(
      const Options options, std::vector<int> keys,
-      const std::vector<ValueType>& value_types, int file_id,
+      const std::vector<ValueType>& value_types,
+      std::vector<std::pair<int, int>> range_deletions, int file_id,
      std::map<std::string, std::string>* true_data) {
    assert(value_types.size() == 1 || keys.size() == value_types.size());
    std::string file_path = sst_files_dir_ + ToString(file_id);
@@ -49,6 +50,29 @@ class ExternalSSTFileBasicTest : public DBTestBase {
    if (!s.ok()) {
      return s;
    }
+    for (size_t i = 0; i < range_deletions.size(); i++) {
+      // Account for the effect of range deletions on true_data before
+      // all point operators, even though sst_file_writer.DeleteRange
+      // must be called before other sst_file_writer methods. This is
+      // because point writes take precedence over range deletions
+      // in the same ingested sst.
+      std::string start_key = Key(range_deletions[i].first);
+      std::string end_key = Key(range_deletions[i].second);
+      s = sst_file_writer.DeleteRange(start_key, end_key);
+      if (!s.ok()) {
+        sst_file_writer.Finish();
+        return s;
+      }
+      auto start_key_it = true_data->find(start_key);
+      if (start_key_it == true_data->end()) {
+        start_key_it = true_data->upper_bound(start_key);
+      }
+      auto end_key_it = true_data->find(end_key);
+      if (end_key_it == true_data->end()) {
+        end_key_it = true_data->upper_bound(end_key);
+      }
+      true_data->erase(start_key_it, end_key_it);
+    }
    for (size_t i = 0; i < keys.size(); i++) {
      std::string key = Key(keys[i]);
      std::string value = Key(keys[i]) + ToString(file_id);
@@ -86,6 +110,14 @@ class ExternalSSTFileBasicTest : public DBTestBase {
    return s;
  }

+  Status GenerateAndAddExternalFile(
+      const Options options, std::vector<int> keys,
+      const std::vector<ValueType>& value_types, int file_id,
+      std::map<std::string, std::string>* true_data) {
+    return GenerateAndAddExternalFile(options, keys, value_types, {}, file_id,
+                                      true_data);
+  }
+
  Status GenerateAndAddExternalFile(
      const Options options, std::vector<int> keys, const ValueType value_type,
      int file_id, std::map<std::string, std::string>* true_data) {
@@ -126,9 +158,14 @@ TEST_F(ExternalSSTFileBasicTest, Basic) {
  ASSERT_EQ(file1_info.num_entries, 100);
  ASSERT_EQ(file1_info.smallest_key, Key(0));
  ASSERT_EQ(file1_info.largest_key, Key(99));
+  ASSERT_EQ(file1_info.num_range_del_entries, 0);
+  ASSERT_EQ(file1_info.smallest_range_del_key, "");
+  ASSERT_EQ(file1_info.largest_range_del_key, "");
  // sst_file_writer already finished, cannot add this value
  s = sst_file_writer.Put(Key(100), "bad_val");
  ASSERT_FALSE(s.ok()) << s.ToString();
+  s = sst_file_writer.DeleteRange(Key(100), Key(200));
+  ASSERT_FALSE(s.ok()) << s.ToString();

  DestroyAndReopen(options);
  // Add file using file path
@@ -189,6 +226,7 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
  ASSERT_EQ(file3_info.num_entries, 15);
  ASSERT_EQ(file3_info.smallest_key, Key(110));
  ASSERT_EQ(file3_info.largest_key, Key(124));
+
  s = DeprecatedAddFile({file1}, true /* move file */);
  ASSERT_TRUE(s.ok()) << s.ToString();
  ASSERT_EQ(Status::NotFound(), env_->FileExists(file1));
@@ -197,8 +235,8 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
  ASSERT_TRUE(s.ok()) << s.ToString();
  ASSERT_OK(env_->FileExists(file2));

-  // This file have overlapping values with the existing data
-  s = DeprecatedAddFile({file2}, true /* move file */);
+  // This file has overlapping values with the existing data
+  s = DeprecatedAddFile({file3}, true /* move file */);
  ASSERT_FALSE(s.ok()) << s.ToString();
  ASSERT_OK(env_->FileExists(file3));

@@ -218,34 +256,33 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
                                         ValueType::kTypeValue, file_id++,
                                         &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
                                         ValueType::kTypeValue, file_id++,
-
                                         &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 4, 6}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {11, 15, 19}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {120, 130}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 130}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);

    // Write some keys through normal write path
@@ -257,17 +294,17 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {40, 41, 42}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {20, 30, 40}, ValueType::kTypeValue, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);

    const Snapshot* snapshot = db_->GetSnapshot();
@@ -314,37 +351,54 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
                                         ValueType::kTypeValue, file_id++,
                                         &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
                                         ValueType::kTypeValue, file_id++,
-
                                         &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 4, 6}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);

    ASSERT_OK(GenerateAndAddExternalFile(options, {11, 15, 19},
                                         ValueType::kTypeDeletion, file_id++,
                                         &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {120, 130}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 130}, ValueType::kTypeDeletion, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);

+    ASSERT_OK(GenerateAndAddExternalFile(options, {120},
+                                         {ValueType::kTypeValue}, {{120, 135}},
+                                         file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{110, 120}},
+                                         file_id++, &true_data));
+    // The range deletion ends on a key, but it doesn't actually delete
+    // this key because the largest key in the range is exclusive. Still,
+    // it counts as an overlap so a new seqno will be assigned.
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
+    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{100, 109}},
+                                         file_id++, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
    // Write some keys through normal write path
    for (int i = 0; i < 50; i++) {
      ASSERT_OK(Put(Key(i), "memtable"));
@@ -354,18 +408,18 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {40, 41, 42}, ValueType::kTypeMerge, file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);

    ASSERT_OK(GenerateAndAddExternalFile(options, {20, 30, 40},
                                         ValueType::kTypeDeletion, file_id++,
                                         &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);

    const Snapshot* snapshot = db_->GetSnapshot();
@@ -414,7 +468,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
         ValueType::kTypeMerge, ValueType::kTypeValue, ValueType::kTypeMerge},
        file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(
@@ -422,35 +476,58 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
         ValueType::kTypeMerge},
        file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 4, 6}, {ValueType::kTypeDeletion, ValueType::kTypeValue,
                             ValueType::kTypeMerge},
        file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {11, 15, 19}, {ValueType::kTypeDeletion, ValueType::kTypeMerge,
                                ValueType::kTypeValue},
        file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {120, 130}, {ValueType::kTypeValue, ValueType::kTypeMerge},
        file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);

    ASSERT_OK(GenerateAndAddExternalFile(
        options, {1, 130}, {ValueType::kTypeMerge, ValueType::kTypeDeletion},
        file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);

+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {150, 151, 152},
+        {ValueType::kTypeValue, ValueType::kTypeMerge,
+         ValueType::kTypeDeletion},
+        {{150, 160}, {180, 190}}, file_id++, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {150, 151, 152},
+        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
+        {{200, 250}}, file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {300, 301, 302},
+        {ValueType::kTypeValue, ValueType::kTypeMerge,
+         ValueType::kTypeDeletion},
+        {{1, 2}, {152, 154}}, file_id++, &true_data));
+    // File overwrites some keys, a seqno will be assigned
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
+
    // Write some keys through normal write path
    for (int i = 0; i < 50; i++) {
      ASSERT_OK(Put(Key(i), "memtable"));
@@ -462,14 +539,15 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
        options, {60, 61, 62},
        {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
        file_id++, &true_data));
-    // File dont overwrite any keys, No seqno needed
+    // File doesn't overwrite any keys, no seqno needed
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);

    ASSERT_OK(GenerateAndAddExternalFile(
-        options, {40, 41, 42}, {ValueType::kTypeValue, ValueType::kTypeDeletion,
-                                ValueType::kTypeDeletion},
+        options, {40, 41, 42},
+        {ValueType::kTypeValue, ValueType::kTypeDeletion,
+         ValueType::kTypeDeletion},
        file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);

    ASSERT_OK(GenerateAndAddExternalFile(
@@ -477,7 +555,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
        {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
         ValueType::kTypeDeletion},
        file_id++, &true_data));
-    // File overwrite some keys, a seqno will be assigned
+    // File overwrites some keys, a seqno will be assigned
    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);

    const Snapshot* snapshot = db_->GetSnapshot();
@@ -589,7 +667,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
  SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
  ASSERT_OK(GenerateAndAddExternalFile(
      options, {60, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      {{65, 70}, {70, 85}}, file_id++, &true_data));
  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
  ASSERT_EQ(2, NumTableFilesAtLevel(0));
  ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
@@ -605,6 +683,16 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));

+  // overlaps with L5 file but not memtable or L0 file, so flush is skipped and
+  // file is ingested into L4
+  ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{5, 15}}, file_id++,
+                                       &true_data));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 2));
+  ASSERT_EQ(1, NumTableFilesAtLevel(options.num_levels - 1));
+
  // ingested file overlaps with memtable, so flush is triggered before the file
  // is ingested such that the ingested data is considered newest. So L0 file
  // count increases by two.
@@ -623,7 +711,7 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
  // seqnum.
  ASSERT_OK(GenerateAndAddExternalFile(
      options, {151, 175}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      {{160, 200}}, file_id++, &true_data));
  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
  ASSERT_EQ(4, NumTableFilesAtLevel(0));
  ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
@@ -29,7 +29,8 @@
 namespace rocksdb {

 Status ExternalSstFileIngestionJob::Prepare(
-    const std::vector<std::string>& external_files_paths, SuperVersion* sv) {
+    const std::vector<std::string>& external_files_paths,
+    uint64_t next_file_number, SuperVersion* sv) {
  Status status;

  // Read the information of files we are ingesting
@@ -78,7 +79,7 @@ Status ExternalSstFileIngestionJob::Prepare(
  }

  for (IngestedFileInfo& f : files_to_ingest_) {
-    if (f.num_entries == 0) {
+    if (f.num_entries == 0 && f.num_range_deletions == 0) {
      return Status::InvalidArgument("File contain no entries");
    }

@@ -90,7 +91,7 @@ Status ExternalSstFileIngestionJob::Prepare(

  // Copy/Move external files into DB
  for (IngestedFileInfo& f : files_to_ingest_) {
-    f.fd = FileDescriptor(versions_->NewFileNumber(), 0, f.file_size);
+    f.fd = FileDescriptor(next_file_number++, 0, f.file_size);

    const std::string path_outside_db = f.external_file_path;
    const std::string path_inside_db =
@@ -336,12 +337,14 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(

    // Set the global sequence number
    file_to_ingest->original_seqno = DecodeFixed64(seqno_iter->second.c_str());
-    file_to_ingest->global_seqno_offset = props->properties_offsets.at(
+    auto offsets_iter = props->properties_offsets.find(
        ExternalSstFilePropertyNames::kGlobalSeqno);
-
-    if (file_to_ingest->global_seqno_offset == 0) {
+    if (offsets_iter == props->properties_offsets.end() ||
+        offsets_iter->second == 0) {
+      file_to_ingest->global_seqno_offset = 0;
      return Status::Corruption("Was not able to find file global seqno field");
    }
+    file_to_ingest->global_seqno_offset = static_cast<size_t>(offsets_iter->second);
  } else if (file_to_ingest->version == 1) {
    // SST file V1 should not have global seqno field
    assert(seqno_iter == uprops.end());
@@ -356,6 +359,7 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
  }
  // Get number of entries in table
  file_to_ingest->num_entries = props->num_entries;
+  file_to_ingest->num_range_deletions = props->num_range_deletions;

  ParsedInternalKey key;
  ReadOptions ro;
@@ -367,26 +371,55 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
  ro.fill_cache = false;
  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(
      ro, sv->mutable_cf_options.prefix_extractor.get()));
+  std::unique_ptr<InternalIterator> range_del_iter(
+      table_reader->NewRangeTombstoneIterator(ro));

-  // Get first (smallest) key from file
+  // Get first (smallest) and last (largest) key from file.
+  bool bounds_set = false;
  iter->SeekToFirst();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("external file have corrupted keys");
-  }
-  if (key.sequence != 0) {
-    return Status::Corruption("external file have non zero sequence number");
-  }
-  file_to_ingest->smallest_user_key = key.user_key.ToString();
+  if (iter->Valid()) {
+    if (!ParseInternalKey(iter->key(), &key)) {
+      return Status::Corruption("external file have corrupted keys");
+    }
+    if (key.sequence != 0) {
+      return Status::Corruption("external file have non zero sequence number");
+    }
+    file_to_ingest->smallest_user_key = key.user_key.ToString();

-  // Get last (largest) key from file
-  iter->SeekToLast();
-  if (!ParseInternalKey(iter->key(), &key)) {
-    return Status::Corruption("external file have corrupted keys");
+    iter->SeekToLast();
+    if (!ParseInternalKey(iter->key(), &key)) {
+      return Status::Corruption("external file have corrupted keys");
+    }
+    if (key.sequence != 0) {
+      return Status::Corruption("external file have non zero sequence number");
+    }
+    file_to_ingest->largest_user_key = key.user_key.ToString();
+
+    bounds_set = true;
  }
-  if (key.sequence != 0) {
-    return Status::Corruption("external file have non zero sequence number");
+
+  // We may need to adjust these key bounds, depending on whether any range
+  // deletion tombstones extend past them.
+  const Comparator* ucmp = cfd_->internal_comparator().user_comparator();
+  if (range_del_iter != nullptr) {
+    for (range_del_iter->SeekToFirst(); range_del_iter->Valid();
+         range_del_iter->Next()) {
+      if (!ParseInternalKey(range_del_iter->key(), &key)) {
+        return Status::Corruption("external file have corrupted keys");
+      }
+      RangeTombstone tombstone(key, range_del_iter->value());
+
+      if (!bounds_set || ucmp->Compare(tombstone.start_key_,
+                                       file_to_ingest->smallest_user_key) < 0) {
+        file_to_ingest->smallest_user_key = tombstone.start_key_.ToString();
+      }
+      if (!bounds_set || ucmp->Compare(tombstone.end_key_,
+                                       file_to_ingest->largest_user_key) > 0) {
+        file_to_ingest->largest_user_key = tombstone.end_key_.ToString();
+      }
+      bounds_set = true;
+    }
  }
-  file_to_ingest->largest_user_key = key.user_key.ToString();

  file_to_ingest->cf_id = static_cast<uint32_t>(props->column_family_id);

@@ -443,9 +476,9 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
        const SequenceNumber level_largest_seqno =
            (*max_element(level_files.begin(), level_files.end(),
                          [](FileMetaData* f1, FileMetaData* f2) {
-                            return f1->largest_seqno < f2->largest_seqno;
+                            return f1->fd.largest_seqno < f2->fd.largest_seqno;
                          }))
-                ->largest_seqno;
+                ->fd.largest_seqno;
        // should only assign seqno to current level's largest seqno when
        // the file fits
        if (level_largest_seqno != 0 &&
@@ -490,7 +523,7 @@ Status ExternalSstFileIngestionJob::CheckLevelForIngestedBehindFile(
  // at some upper level
  for (int lvl = 0; lvl < cfd_->NumberLevels() - 1; lvl++) {
    for (auto file : vstorage->LevelFiles(lvl)) {
-      if (file->smallest_seqno == 0) {
+      if (file->fd.smallest_seqno == 0) {
        return Status::InvalidArgument(
          "Can't ingest_behind file as despite allow_ingest_behind=true "
          "there are files with 0 seqno in database at upper levels!");
@@ -515,24 +548,27 @@ Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
        "field");
  }

-  std::unique_ptr<RandomRWFile> rwfile;
-  Status status = env_->NewRandomRWFile(file_to_ingest->internal_file_path,
-                                        &rwfile, env_options_);
-  if (!status.ok()) {
-    return status;
+  if (ingestion_options_.write_global_seqno) {
+    // Determine if we can write global_seqno to a given offset of file.
+    // If the file system does not support random write, then we should not.
+    // Otherwise we should.
+    std::unique_ptr<RandomRWFile> rwfile;
+    Status status = env_->NewRandomRWFile(file_to_ingest->internal_file_path,
+                                          &rwfile, env_options_);
+    if (status.ok()) {
+      std::string seqno_val;
+      PutFixed64(&seqno_val, seqno);
+      status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
+      if (!status.ok()) {
+        return status;
+      }
+    } else if (!status.IsNotSupported()) {
+      return status;
+    }
  }

-  // Write the new seqno in the global sequence number field in the file
-  std::string seqno_val;
-  PutFixed64(&seqno_val, seqno);
-  status = rwfile->Write(file_to_ingest->global_seqno_offset, seqno_val);
-  if (status.ok()) {
-    status = rwfile->Fsync();
-  }
-  if (status.ok()) {
-    file_to_ingest->assigned_seqno = seqno;
-  }
-  return status;
+  file_to_ingest->assigned_seqno = seqno;
+  return Status::OK();
 }

 bool ExternalSstFileIngestionJob::IngestedFileFitInLevel(
@@ -36,6 +36,8 @@ struct IngestedFileInfo {
  uint64_t file_size;
  // total number of keys in external file
  uint64_t num_entries;
+  // total number of range deletions in external file
+  uint64_t num_range_deletions;
  // Id of column family this file shoule be ingested into
  uint32_t cf_id;
  // TableProperties read from external file
@@ -87,7 +89,7 @@ class ExternalSstFileIngestionJob {

  // Prepare the job by copying external files into the DB.
  Status Prepare(const std::vector<std::string>& external_files_paths,
-                 SuperVersion* sv);
+                 uint64_t next_file_number, SuperVersion* sv);

  // Check if we need to flush the memtable before running the ingestion job
  // This will be true if the files we are ingesting are overlapping with any
@@ -10,6 +10,7 @@
 #include "port/port.h"
 #include "port/stack_trace.h"
 #include "rocksdb/sst_file_writer.h"
+#include "util/filename.h"
 #include "util/testutil.h"

 namespace rocksdb {
@@ -222,6 +223,9 @@ TEST_F(ExternalSSTFileTest, Basic) {
    ASSERT_EQ(file1_info.num_entries, 100);
    ASSERT_EQ(file1_info.smallest_key, Key(0));
    ASSERT_EQ(file1_info.largest_key, Key(99));
+    ASSERT_EQ(file1_info.num_range_del_entries, 0);
+    ASSERT_EQ(file1_info.smallest_range_del_key, "");
+    ASSERT_EQ(file1_info.largest_range_del_key, "");
    // sst_file_writer already finished, cannot add this value
    s = sst_file_writer.Put(Key(100), "bad_val");
    ASSERT_FALSE(s.ok()) << s.ToString();
@@ -290,6 +294,58 @@ TEST_F(ExternalSSTFileTest, Basic) {
    ASSERT_EQ(file5_info.smallest_key, Key(400));
    ASSERT_EQ(file5_info.largest_key, Key(499));

+    // file6.sst (delete 400 => 500)
+    std::string file6 = sst_files_dir_ + "file6.sst";
+    ASSERT_OK(sst_file_writer.Open(file6));
+    sst_file_writer.DeleteRange(Key(400), Key(500));
+    ExternalSstFileInfo file6_info;
+    s = sst_file_writer.Finish(&file6_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file6_info.file_path, file6);
+    ASSERT_EQ(file6_info.num_entries, 0);
+    ASSERT_EQ(file6_info.smallest_key, "");
+    ASSERT_EQ(file6_info.largest_key, "");
+    ASSERT_EQ(file6_info.num_range_del_entries, 1);
+    ASSERT_EQ(file6_info.smallest_range_del_key, Key(400));
+    ASSERT_EQ(file6_info.largest_range_del_key, Key(500));
+
+    // file7.sst (delete 500 => 570, put 520 => 599 divisible by 2)
+    std::string file7 = sst_files_dir_ + "file7.sst";
+    ASSERT_OK(sst_file_writer.Open(file7));
+    sst_file_writer.DeleteRange(Key(500), Key(550));
+    for (int k = 520; k < 560; k += 2) {
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
+    }
+    sst_file_writer.DeleteRange(Key(525), Key(575));
+    for (int k = 560; k < 600; k += 2) {
+      ASSERT_OK(sst_file_writer.Put(Key(k), Key(k) + "_val"));
+    }
+    ExternalSstFileInfo file7_info;
+    s = sst_file_writer.Finish(&file7_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file7_info.file_path, file7);
+    ASSERT_EQ(file7_info.num_entries, 40);
+    ASSERT_EQ(file7_info.smallest_key, Key(520));
+    ASSERT_EQ(file7_info.largest_key, Key(598));
+    ASSERT_EQ(file7_info.num_range_del_entries, 2);
+    ASSERT_EQ(file7_info.smallest_range_del_key, Key(500));
+    ASSERT_EQ(file7_info.largest_range_del_key, Key(575));
+
+    // file8.sst (delete 600 => 700)
+    std::string file8 = sst_files_dir_ + "file8.sst";
+    ASSERT_OK(sst_file_writer.Open(file8));
+    sst_file_writer.DeleteRange(Key(600), Key(700));
+    ExternalSstFileInfo file8_info;
+    s = sst_file_writer.Finish(&file8_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file8_info.file_path, file8);
+    ASSERT_EQ(file8_info.num_entries, 0);
+    ASSERT_EQ(file8_info.smallest_key, "");
+    ASSERT_EQ(file8_info.largest_key, "");
+    ASSERT_EQ(file8_info.num_range_del_entries, 1);
+    ASSERT_EQ(file8_info.smallest_range_del_key, Key(600));
+    ASSERT_EQ(file8_info.largest_range_del_key, Key(700));
+
    // Cannot create an empty sst file
    std::string file_empty = sst_files_dir_ + "file_empty.sst";
    ExternalSstFileInfo file_empty_info;
@@ -336,6 +392,16 @@ TEST_F(ExternalSSTFileTest, Basic) {
    // Key range of file5 (400 => 499) dont overlap with any keys in DB
    ASSERT_OK(DeprecatedAddFile({file5}));

+    // This file has overlapping values with the existing data
+    s = DeprecatedAddFile({file6});
+    ASSERT_FALSE(s.ok()) << s.ToString();
+
+    // Key range of file7 (500 => 598) dont overlap with any keys in DB
+    ASSERT_OK(DeprecatedAddFile({file7}));
+
+    // Key range of file7 (600 => 700) dont overlap with any keys in DB
+    ASSERT_OK(DeprecatedAddFile({file8}));
+
    // Make sure values are correct before and after flush/compaction
    for (int i = 0; i < 2; i++) {
      for (int k = 0; k < 200; k++) {
@@ -349,6 +415,13 @@ TEST_F(ExternalSSTFileTest, Basic) {
        std::string value = Key(k) + "_val";
        ASSERT_EQ(Get(Key(k)), value);
      }
+      for (int k = 500; k < 600; k++) {
+        std::string value = Key(k) + "_val";
+        if (k < 520 || k % 2 == 1) {
+          value = "NOT_FOUND";
+        }
+        ASSERT_EQ(Get(Key(k)), value);
+      }
      ASSERT_OK(Flush());
      ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
    }
@@ -377,7 +450,8 @@ TEST_F(ExternalSSTFileTest, Basic) {
      ASSERT_EQ(Get(Key(k)), value);
    }
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction |
+                         kRangeDelSkipConfigs));
 }
 class SstFileWriterCollector : public TablePropertiesCollector {
 public:
@@ -518,17 +592,57 @@ TEST_F(ExternalSSTFileTest, AddList) {
    ASSERT_EQ(file5_info.smallest_key, Key(200));
    ASSERT_EQ(file5_info.largest_key, Key(299));

+    // file6.sst (delete 0 => 100)
+    std::string file6 = sst_files_dir_ + "file6.sst";
+    ASSERT_OK(sst_file_writer.Open(file6));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(0), Key(75)));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(25), Key(100)));
+    ExternalSstFileInfo file6_info;
+    s = sst_file_writer.Finish(&file6_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file6_info.file_path, file6);
+    ASSERT_EQ(file6_info.num_entries, 0);
+    ASSERT_EQ(file6_info.smallest_key, "");
+    ASSERT_EQ(file6_info.largest_key, "");
+    ASSERT_EQ(file6_info.num_range_del_entries, 2);
+    ASSERT_EQ(file6_info.smallest_range_del_key, Key(0));
+    ASSERT_EQ(file6_info.largest_range_del_key, Key(100));
+
+    // file7.sst (delete 100 => 200)
+    std::string file7 = sst_files_dir_ + "file7.sst";
+    ASSERT_OK(sst_file_writer.Open(file7));
+    ASSERT_OK(sst_file_writer.DeleteRange(Key(100), Key(200)));
+    ExternalSstFileInfo file7_info;
+    s = sst_file_writer.Finish(&file7_info);
+    ASSERT_TRUE(s.ok()) << s.ToString();
+    ASSERT_EQ(file7_info.file_path, file7);
+    ASSERT_EQ(file7_info.num_entries, 0);
+    ASSERT_EQ(file7_info.smallest_key, "");
+    ASSERT_EQ(file7_info.largest_key, "");
+    ASSERT_EQ(file7_info.num_range_del_entries, 1);
+    ASSERT_EQ(file7_info.smallest_range_del_key, Key(100));
+    ASSERT_EQ(file7_info.largest_range_del_key, Key(200));
+
    // list 1 has internal key range conflict
    std::vector<std::string> file_list0({file1, file2});
    std::vector<std::string> file_list1({file3, file2, file1});
    std::vector<std::string> file_list2({file5});
    std::vector<std::string> file_list3({file3, file4});
+    std::vector<std::string> file_list4({file5, file7});
+    std::vector<std::string> file_list5({file6, file7});

    DestroyAndReopen(options);

-    // This list of files have key ranges are overlapping with each other
+    // These lists of files have key ranges that overlap with each other
    s = DeprecatedAddFile(file_list1);
    ASSERT_FALSE(s.ok()) << s.ToString();
+    // Both of the following overlap on the end key of a range deletion
+    // tombstone. This is a limitation because these tombstones have exclusive
+    // end keys that should not count as overlapping with other keys.
+    s = DeprecatedAddFile(file_list4);
+    ASSERT_FALSE(s.ok()) << s.ToString();
+    s = DeprecatedAddFile(file_list5);
+    ASSERT_FALSE(s.ok()) << s.ToString();

    // Add files using file path list
    s = DeprecatedAddFile(file_list0);
@@ -619,7 +733,8 @@ TEST_F(ExternalSSTFileTest, AddList) {
      ASSERT_EQ(Get(Key(k)), value);
    }
    DestroyAndRecreateExternalSSTFilesDir();
-  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction |
+                         kRangeDelSkipConfigs));
 }

 TEST_F(ExternalSSTFileTest, AddListAtomicity) {
@@ -897,11 +1012,11 @@ TEST_F(ExternalSSTFileTest, MultiThreaded) {

 TEST_F(ExternalSSTFileTest, OverlappingRanges) {
  Random rnd(301);
-  int picked_level = 0;
+  SequenceNumber assigned_seqno = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-    "ExternalSstFileIngestionJob::Run", [&picked_level](void* arg) {
+    "ExternalSstFileIngestionJob::Run", [&assigned_seqno](void* arg) {
      ASSERT_TRUE(arg != nullptr);
-      picked_level = *(static_cast<int*>(arg));
+      assigned_seqno = *(static_cast<SequenceNumber*>(arg));
    });
  bool need_flush = false;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -925,7 +1040,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {

    printf("Option config = %d\n", option_config_);
    std::vector<std::pair<int, int>> key_ranges;
-    for (int i = 0; i < 500; i++) {
+    for (int i = 0; i < 100; i++) {
      int range_start = rnd.Uniform(20000);
      int keys_per_range = 10 + rnd.Uniform(41);

@@ -971,7 +1086,8 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
        s = DeprecatedAddFile({file_name});
        auto it = true_data.lower_bound(Key(range_start));
        if (option_config_ != kUniversalCompaction &&
-            option_config_ != kUniversalCompactionMultiLevel) {
+            option_config_ != kUniversalCompactionMultiLevel &&
+            option_config_ != kUniversalSubcompactions) {
          if (it != true_data.end() && it->first <= Key(range_end)) {
            // This range overlap with data already exist in DB
            ASSERT_NOK(s);
@@ -982,7 +1098,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
          }
        } else {
          if ((it != true_data.end() && it->first <= Key(range_end)) ||
-              need_flush || picked_level > 0 || overlap_with_db) {
+              need_flush || assigned_seqno > 0 || overlap_with_db) {
            // This range overlap with data already exist in DB
            ASSERT_NOK(s);
            failed_add_file++;
@@ -1177,6 +1293,39 @@ TEST_F(ExternalSSTFileTest, PickedLevelBug) {
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }

+TEST_F(ExternalSSTFileTest, IngestNonExistingFile) {
+  Options options = CurrentOptions();
+  DestroyAndReopen(options);
+
+  Status s = db_->IngestExternalFile({"non_existing_file"},
+                                     IngestExternalFileOptions());
+  ASSERT_NOK(s);
+
+  // Verify file deletion is not impacted (verify a bug fix)
+  ASSERT_OK(Put(Key(1), Key(1)));
+  ASSERT_OK(Put(Key(9), Key(9)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(Put(Key(1), Key(1)));
+  ASSERT_OK(Put(Key(9), Key(9)));
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  // After full compaction, there should be only 1 file.
+  std::vector<std::string> files;
+  env_->GetChildren(dbname_, &files);
+  int num_sst_files = 0;
+  for (auto& f : files) {
+    uint64_t number;
+    FileType type;
+    if (ParseFileName(f, &number, &type) && type == kTableFile) {
+      num_sst_files++;
+    }
+  }
+  ASSERT_EQ(1, num_sst_files);
+}
+
 TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
  Options options = CurrentOptions();
  options.disable_auto_compactions = false;
@@ -1194,8 +1343,9 @@ TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
    ASSERT_OK(GenerateAndAddExternalFile(options, file_keys, range_id));
  };

+  const int num_of_ranges = 1000;
  std::vector<port::Thread> threads;
-  while (range_id < 5000) {
+  while (range_id < num_of_ranges) {
    int range_start = range_id * 10;
    int range_end = range_start + 10;

@@ -1220,7 +1370,7 @@ TEST_F(ExternalSSTFileTest, CompactDuringAddFileRandom) {
    range_id++;
  }

-  for (int rid = 0; rid < 5000; rid++) {
+  for (int rid = 0; rid < num_of_ranges; rid++) {
    int range_start = rid * 10;
    int range_end = range_start + 10;

@@ -1475,7 +1625,7 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {

    Random rnd(301);
    std::map<std::string, std::string> true_data;
-    for (int i = 0; i < 2000; i++) {
+    for (int i = 0; i < 500; i++) {
      std::vector<std::pair<std::string, std::string>> random_data;
      for (int j = 0; j < 100; j++) {
        std::string k;
@@ -107,18 +107,18 @@ class FaultInjectionTest
    Options options;
    switch (option_config_) {
      case kWalDir:
-        options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
+        options.wal_dir = test::PerThreadDBPath(env_, "fault_test_wal");
        break;
      case kDifferentDataDir:
-        options.db_paths.emplace_back(test::TmpDir(env_) + "/fault_test_data",
-                                      1000000U);
+        options.db_paths.emplace_back(
+            test::PerThreadDBPath(env_, "fault_test_data"), 1000000U);
        break;
      case kSyncWal:
        sync_use_wal_ = true;
        sync_use_compact_ = false;
        break;
      case kWalDirSyncWal:
-        options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
+        options.wal_dir = test::PerThreadDBPath(env_, "/fault_test_wal");
        sync_use_wal_ = true;
        sync_use_compact_ = false;
        break;
@@ -158,7 +158,7 @@ class FaultInjectionTest
    table_options.block_cache = tiny_cache_;
    options_.table_factory.reset(NewBlockBasedTableFactory(table_options));

-    dbname_ = test::TmpDir() + "/fault_test";
+    dbname_ = test::PerThreadDBPath("fault_test");

    EXPECT_OK(DestroyDB(dbname_, options_));

@@ -78,6 +78,8 @@ const char* GetFlushReasonString (FlushReason flush_reason) {
      return "Auto Compaction";
    case FlushReason::kManualFlush:
      return "Manual Flush";
+    case FlushReason::kErrorRecovery:
+      return "Error Recovery";
    default:
      return "Invalid";
  }
@@ -187,6 +189,7 @@ void FlushJob::PickMemTable() {

 Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
                     FileMetaData* file_meta) {
+  TEST_SYNC_POINT("FlushJob::Start");
  db_mutex_->AssertHeld();
  assert(pick_memtable_called);
  AutoThreadOperationStageUpdater stage_run(
@@ -370,8 +373,8 @@ Status FlushJob::WriteLevel0Table() {
                   s.ToString().c_str(),
                   meta_.marked_for_compaction ? " (needs compaction)" : "");

-    if (output_file_directory_ != nullptr) {
-      output_file_directory_->Fsync();
+    if (s.ok() && output_file_directory_ != nullptr) {
+      s = output_file_directory_->Fsync();
    }
    TEST_SYNC_POINT("FlushJob::WriteLevel0Table");
    db_mutex_->Lock();
@@ -388,7 +391,7 @@ Status FlushJob::WriteLevel0Table() {
    // Add file to L0
    edit_->AddFile(0 /* level */, meta_.fd.GetNumber(), meta_.fd.GetPathId(),
                   meta_.fd.GetFileSize(), meta_.smallest, meta_.largest,
-                   meta_.smallest_seqno, meta_.largest_seqno,
+                   meta_.fd.smallest_seqno, meta_.fd.largest_seqno,
                   meta_.marked_for_compaction);
  }

@@ -27,7 +27,7 @@ class FlushJobTest : public testing::Test {
 public:
  FlushJobTest()
      : env_(Env::Default()),
-        dbname_(test::TmpDir() + "/flush_job_test"),
+        dbname_(test::PerThreadDBPath("flush_job_test")),
        options_(),
        db_options_(options_),
        table_cache_(NewLRUCache(50000, 16)),
@@ -62,7 +62,7 @@ class FlushJobTest : public testing::Test {
        manifest, &file, env_->OptimizeForManifestWrite(env_options_));
    ASSERT_OK(s);
    unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), EnvOptions()));
+        new WritableFileWriter(std::move(file), manifest, EnvOptions()));
    {
      log::Writer log(std::move(file_writer), 0, false);
      std::string record;
@@ -147,19 +147,20 @@ TEST_F(FlushJobTest, NonEmpty) {
                     db_options_.statistics.get(), &event_logger, true);

  HistogramData hist;
-  FileMetaData fd;
+  FileMetaData file_meta;
  mutex_.Lock();
  flush_job.PickMemTable();
-  ASSERT_OK(flush_job.Run(nullptr, &fd));
+  ASSERT_OK(flush_job.Run(nullptr, &file_meta));
  mutex_.Unlock();
  db_options_.statistics->histogramData(FLUSH_TIME, &hist);
  ASSERT_GT(hist.average, 0.0);

-  ASSERT_EQ(ToString(0), fd.smallest.user_key().ToString());
-  ASSERT_EQ("9999a",
-            fd.largest.user_key().ToString());  // range tombstone end key
-  ASSERT_EQ(1, fd.smallest_seqno);
-  ASSERT_EQ(10000, fd.largest_seqno);  // range tombstone seqnum 10000
+  ASSERT_EQ(ToString(0), file_meta.smallest.user_key().ToString());
+  ASSERT_EQ(
+      "9999a",
+      file_meta.largest.user_key().ToString());  // range tombstone end key
+  ASSERT_EQ(1, file_meta.fd.smallest_seqno);
+  ASSERT_EQ(10000, file_meta.fd.largest_seqno);  // range tombstone seqnum 10000
  mock_table_factory_->AssertSingleFile(inserted_keys);
  job_context.Clean();
 }
@@ -75,7 +75,7 @@ class ForwardLevelIterator : public InternalIterator {
        cfd_->internal_comparator(), {} /* snapshots */);
    file_iter_ = cfd_->table_cache()->NewIterator(
        read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
-        files_[file_index_]->fd,
+        *files_[file_index_],
        read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
        prefix_extractor_, nullptr /* table_reader_ptr */, nullptr, false);
    file_iter_->SetPinnedItersMgr(pinned_iters_mgr_);
@@ -635,7 +635,7 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
      continue;
    }
    l0_iters_.push_back(cfd_->table_cache()->NewIterator(
-        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), l0->fd,
+        read_options_, *cfd_->soptions(), cfd_->internal_comparator(), *l0,
        read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
        sv_->mutable_cf_options.prefix_extractor.get()));
  }
@@ -706,7 +706,7 @@ void ForwardIterator::RenewIterators() {
    }
    l0_iters_new.push_back(cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files_new[inew]->fd,
+        *l0_files_new[inew],
        read_options_.ignore_range_deletions ? nullptr : &range_del_agg,
        svnew->mutable_cf_options.prefix_extractor.get()));
  }
@@ -766,7 +766,7 @@ void ForwardIterator::ResetIncompleteIterators() {
    DeleteIterator(l0_iters_[i]);
    l0_iters_[i] = cfd_->table_cache()->NewIterator(
        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
-        l0_files[i]->fd, nullptr /* range_del_agg */,
+        *l0_files[i], nullptr /* range_del_agg */,
        sv_->mutable_cf_options.prefix_extractor.get());
    l0_iters_[i]->SetPinnedItersMgr(pinned_iters_mgr_);
  }
@@ -916,21 +916,13 @@ bool ForwardIterator::TEST_CheckDeletedIters(int* pdeleted_iters,
 uint32_t ForwardIterator::FindFileInRange(
    const std::vector<FileMetaData*>& files, const Slice& internal_key,
    uint32_t left, uint32_t right) {
-  while (left < right) {
-    uint32_t mid = (left + right) / 2;
-    const FileMetaData* f = files[mid];
-    if (cfd_->internal_comparator().InternalKeyComparator::Compare(
-          f->largest.Encode(), internal_key) < 0) {
-      // Key at "mid.largest" is < "target".  Therefore all
-      // files at or before "mid" are uninteresting.
-      left = mid + 1;
-    } else {
-      // Key at "mid.largest" is >= "target".  Therefore all files
-      // after "mid" are uninteresting.
-      right = mid;
-    }
-  }
-  return right;
+  auto cmp = [&](const FileMetaData* f, const Slice& key) -> bool {
+    return cfd_->internal_comparator().InternalKeyComparator::Compare(
+            f->largest.Encode(), key) < 0;
+  };
+  const auto &b = files.begin();
+  return static_cast<uint32_t>(std::lower_bound(b + left,
+                                 b + right, internal_key, cmp) - b);
 }

 void ForwardIterator::DeleteIterator(InternalIterator* iter, bool is_arena) {
@@ -323,7 +323,7 @@ int main(int argc, char** argv) {

  std::mt19937 rng{std::random_device()()};
  rocksdb::Status status;
-  std::string path = rocksdb::test::TmpDir() + "/forward_iterator_test";
+  std::string path = rocksdb::test::PerThreadDBPath("forward_iterator_test");
  fprintf(stderr, "db path is %s\n", path.c_str());
  rocksdb::Options options;
  options.create_if_missing = true;
@@ -249,6 +249,7 @@ static const std::string estimate_oldest_key_time = "estimate-oldest-key-time";
 static const std::string block_cache_capacity = "block-cache-capacity";
 static const std::string block_cache_usage = "block-cache-usage";
 static const std::string block_cache_pinned_usage = "block-cache-pinned-usage";
+static const std::string options_statistics = "options-statistics";

 const std::string DB::Properties::kNumFilesAtLevelPrefix =
    rocksdb_prefix + num_files_at_level_prefix;
@@ -332,116 +333,141 @@ const std::string DB::Properties::kBlockCacheUsage =
    rocksdb_prefix + block_cache_usage;
 const std::string DB::Properties::kBlockCachePinnedUsage =
    rocksdb_prefix + block_cache_pinned_usage;
+const std::string DB::Properties::kOptionsStatistics =
+    rocksdb_prefix + options_statistics;

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

 const DBPropertyInfo* GetPropertyInfo(const Slice& property) {
@@ -788,8 +814,8 @@ bool InternalStats::HandleEstimateTableReadersMem(uint64_t* value,
 }

 bool InternalStats::HandleEstimateLiveDataSize(uint64_t* value, DBImpl* /*db*/,
-                                               Version* /*version*/) {
-  const auto* vstorage = cfd_->current()->storage_info();
+                                               Version* version) {
+  const auto* vstorage = version->storage_info();
  *value = vstorage->EstimateLiveDataSize();
  return true;
 }
@@ -45,6 +45,10 @@ struct DBPropertyInfo {

  // @param props Map of general properties to populate
  bool (InternalStats::*handle_map)(std::map<std::string, std::string>* props);
+
+  // handle the string type properties rely on DBImpl methods
+  // @param value Value-result argument for storing the property's string value
+  bool (DBImpl::*handle_string_dbimpl)(std::string* value);
 };

 extern const DBPropertyInfo* GetPropertyInfo(const Slice& property);
@@ -538,7 +542,6 @@ class InternalStats {
  bool HandleBlockCacheUsage(uint64_t* value, DBImpl* db, Version* version);
  bool HandleBlockCachePinnedUsage(uint64_t* value, DBImpl* db,
                                   Version* version);
-
  // Total number of background errors encountered. Every time a flush task
  // or compaction task fails, this counter is incremented. The failure can
  // be caused by any possible reason, including file system errors, out of
@@ -35,6 +35,14 @@ struct SuperVersionContext {
  explicit SuperVersionContext(bool create_superversion = false)
    : new_superversion(create_superversion ? new SuperVersion() : nullptr) {}

+  explicit SuperVersionContext(SuperVersionContext&& other)
+      : superversions_to_free(std::move(other.superversions_to_free)),
+#ifndef ROCKSDB_DISABLE_STALL_NOTIFICATION
+        write_stall_notifications(std::move(other.write_stall_notifications)),
+#endif
+        new_superversion(std::move(other.new_superversion)) {
+  }
+
  void NewSuperVersion() {
    new_superversion = unique_ptr<SuperVersion>(new SuperVersion());
  }
@@ -98,8 +106,15 @@ struct JobContext {
  }

  inline bool HaveSomethingToClean() const {
+    bool sv_have_sth = false;
+    for (const auto& sv_ctx : superversion_contexts) {
+      if (sv_ctx.HaveSomethingToDelete()) {
+        sv_have_sth = true;
+        break;
+      }
+    }
    return memtables_to_free.size() > 0 || logs_to_free.size() > 0 ||
-           superversion_context.HaveSomethingToDelete();
+           sv_have_sth;
  }

  // Structure to store information for candidate files to delete.
@@ -142,7 +157,8 @@ struct JobContext {
  // a list of memtables to be free
  autovector<MemTable*> memtables_to_free;

-  SuperVersionContext superversion_context;
+  // contexts for installing superversions for multiple column families
+  std::vector<SuperVersionContext> superversion_contexts;

  autovector<log::Writer*> logs_to_free;

@@ -158,13 +174,14 @@ struct JobContext {
  size_t num_alive_log_files = 0;
  uint64_t size_log_to_delete = 0;

-  explicit JobContext(int _job_id, bool create_superversion = false)
-    : superversion_context(create_superversion) {
+  explicit JobContext(int _job_id, bool create_superversion = false) {
    job_id = _job_id;
    manifest_file_number = 0;
    pending_manifest_file_number = 0;
    log_number = 0;
    prev_log_number = 0;
+    superversion_contexts.emplace_back(
+        SuperVersionContext(create_superversion));
  }

  // For non-empty JobContext Clean() has to be called at least once before
@@ -173,7 +190,9 @@ struct JobContext {
  // doing potentially slow Clean() with locked DB mutex.
  void Clean() {
    // free superversions
-    superversion_context.Clean();
+    for (auto& sv_context : superversion_contexts) {
+      sv_context.Clean();
+    }
    // free pending memtables
    for (auto m : memtables_to_free) {
      delete m;
@@ -417,7 +417,9 @@ TEST_F(EventListenerTest, DisableBGCompaction) {
  for (int i = 0; static_cast<int>(cf_meta.file_count) < kSlowdownTrigger * 10;
       ++i) {
    Put(1, ToString(i), std::string(10000, 'x'), WriteOptions());
-    db_->Flush(FlushOptions(), handles_[1]);
+    FlushOptions fo;
+    fo.allow_write_stall = true;
+    db_->Flush(fo, handles_[1]);
    db_->GetColumnFamilyMetaData(handles_[1], &cf_meta);
  }
  ASSERT_GE(listener->slowdown_count, kSlowdownTrigger * 9);
@@ -880,10 +882,13 @@ TEST_F(EventListenerTest, BackgroundErrorListenerFailedCompactionTest) {
  ASSERT_EQ(1, listener->counter());

  // trigger flush so compaction is triggered again; this time it succeeds
+  // The previous failed compaction may get retried automatically, so we may
+  // be left with 0 or 1 files in level 1, depending on when the retry gets
+  // scheduled
  ASSERT_OK(Put("key0", "val"));
  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_LE(1, NumTableFilesAtLevel(0));
 }

 }  // namespace rocksdb
--- a/Show More
+++ b/Show More