Fix build on MSVC

Summary:
Fixes some RocksJava regressions recently introduced, whereby RocksJava would not build on JDK 7.
These should have been visible on Travis-CI!
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4768

Differential Revision: D13418173

Pulled By: sagar0

fbshipit-source-id: 57bf223188887f84d9e072031af2e0d2c8a69c30

.gitignore

@@ -45,6 +45,8 @@ etags
 rocksdb_dump
 rocksdb_undump
 db_test2
+trace_analyzer
+trace_analyzer_test
 
 java/out
 java/target

.lgtm.yml

@@ -0,0 +1,4 @@
+extraction:
+  cpp:
+    index:
+      build_command: make static_lib

.travis.yml

@@ -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
@@ -52,11 +64,17 @@ 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:$PATH;
+      mkdir cmake-dist && curl --silent --fail --show-error --location https://github.com/Kitware/CMake/releases/download/v3.14.5/cmake-3.14.5-Linux-x86_64.tar.gz | tar --strip-components=1 -C cmake-dist -xz && export PATH=$PWD/cmake-dist/bin:$PATH;
+    fi
+  - if [[ "${JOB_NAME}" == java_test ]]; then
+      java -version && echo "JAVA_HOME=${JAVA_HOME}";
     fi
 
 before_script:
@@ -67,16 +85,40 @@ 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 make -j4 tools && 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

CMakeLists.txt

@@ -48,6 +48,11 @@ 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)
+option(WITH_WINDOWS_UTF8_FILENAMES "use UTF8 as characterset for opening files, regardles of the system code page" OFF)
+if (WITH_WINDOWS_UTF8_FILENAMES)
+  add_definitions(-DROCKSDB_WINDOWS_UTF8_FILENAMES)
+endif()
 if(MSVC)
   # Defaults currently different for GFLAGS.
   #  We will address find_package work a little later
@@ -108,7 +113,6 @@ 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 +174,7 @@ 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 +311,6 @@ 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 +318,40 @@ 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}")
 
@@ -352,7 +366,7 @@ endif()
 option(ROCKSDB_LITE "Build RocksDBLite version" OFF)
 if(ROCKSDB_LITE)
   add_definitions(-DROCKSDB_LITE)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-exceptions")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-exceptions -Os")
 endif()
 
 if(CMAKE_SYSTEM_NAME MATCHES "Cygwin")
@@ -457,6 +471,7 @@ set(SOURCES
         db/compaction_iterator.cc
         db/compaction_job.cc
         db/compaction_picker.cc
+        db/compaction_picker_fifo.cc
         db/compaction_picker_universal.cc
         db/convenience.cc
         db/db_filesnapshot.cc
@@ -471,6 +486,7 @@ 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
@@ -483,12 +499,12 @@ 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
         db/merge_operator.cc
         db/range_del_aggregator.cc
+        db/range_tombstone_fragmenter.cc
         db/repair.cc
         db/snapshot_impl.cc
         db/table_cache.cc
@@ -545,6 +561,8 @@ 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
@@ -560,6 +578,7 @@ set(SOURCES
         table/plain_table_index.cc
         table/plain_table_key_coding.cc
         table/plain_table_reader.cc
+        table/sst_file_reader.cc
         table/sst_file_writer.cc
         table/table_properties.cc
         table/two_level_iterator.cc
@@ -568,12 +587,14 @@ 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
@@ -584,6 +605,7 @@ set(SOURCES
         util/filename.cc
         util/filter_policy.cc
         util/hash.cc
+        util/jemalloc_nodump_allocator.cc
         util/log_buffer.cc
         util/murmurhash.cc
         util/random.cc
@@ -591,25 +613,25 @@ 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 +668,7 @@ 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
@@ -657,17 +680,17 @@ 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
         $<TARGET_OBJECTS:build_version>)
 
-if(HAVE_SSE42 AND NOT FORCE_SSE42)
-  if(NOT MSVC)
-set_source_files_properties(
-  util/crc32c.cc
-      PROPERTIES COMPILE_FLAGS "-msse4.2 -mpclmul")
-  endif()
+if(HAVE_SSE42 AND NOT MSVC)
+  set_source_files_properties(
+    util/crc32c.cc
+    PROPERTIES COMPILE_FLAGS "-msse4.2 -mpclmul")
 endif()
 
 if(HAVE_POWER8)
@@ -864,6 +887,7 @@ 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
@@ -881,6 +905,8 @@ if(WITH_TESTS)
         db/perf_context_test.cc
         db/plain_table_db_test.cc
         db/prefix_test.cc
+        db/range_del_aggregator_test.cc
+        db/range_tombstone_fragmenter_test.cc
         db/repair_test.cc
         db/table_properties_collector_test.cc
         db/version_builder_test.cc
@@ -906,12 +932,15 @@ 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/sst_file_reader_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 +955,7 @@ 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 +987,7 @@ 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 +998,7 @@ 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

HISTORY.md

@@ -1,7 +1,116 @@
 # Rocksdb Change Log
-## Unreleased
+## 5.18.4 (3/3/2020)
+* Various fixes for ARM64 support (#6250)
+* Fix JEMALLOC_CXX_THROW macro missing from older Jemalloc versions, causing build failures on some platforms.
+
+## 5.18.3 (2/11/2019)
+### Bug Fixes
+* Fix possible LSM corruption when both range deletions and subcompactions are used. The symptom of this corruption is L1+ files overlapping in the user key space.
+
+## 5.18.2 (01/31/2019)
+
 ### Public API Change
+* Change time resolution in FileOperationInfo.
+* Deleting Blob files also go through SStFileManager.
+
+## 5.18.0 (11/30/2018)
+### New Features
+* Introduced `JemallocNodumpAllocator` memory allocator. When being use, block cache will be excluded from core dump.
+* Introduced `PerfContextByLevel` as part of `PerfContext` which allows storing perf context at each level. Also replaced `__thread` with `thread_local` keyword for perf_context. Added per-level perf context for bloom filter and `Get` query.
+* With level_compaction_dynamic_level_bytes = true, level multiplier may be adjusted automatically when Level 0 to 1 compaction is lagged behind.
+* Introduced DB option `atomic_flush`. If true, RocksDB supports flushing multiple column families and atomically committing the result to MANIFEST. Useful when WAL is disabled.
+* Added `num_deletions` and `num_merge_operands` members to `TableProperties`.
+* Added "rocksdb.min-obsolete-sst-number-to-keep" DB property that reports the lower bound on SST file numbers that are being kept from deletion, even if the SSTs are obsolete.
+* Add xxhash64 checksum support
+* Introduced `MemoryAllocator`, which lets the user specify custom memory allocator for block based table.
+* Improved `DeleteRange` to prevent read performance degradation. The feature is no longer marked as experimental.
+* Enabled checkpoint on readonly db (DBImplReadOnly).
+
+### Public API Change
+* `DBOptions::use_direct_reads` now affects reads issued by `BackupEngine` on the database's SSTs.
+* `NO_ITERATORS` is divided into two counters `NO_ITERATOR_CREATED` and `NO_ITERATOR_DELETE`. Both of them are only increasing now, just as other counters.
+
+### Bug Fixes
+* Fix corner case where a write group leader blocked due to write stall blocks other writers in queue with WriteOptions::no_slowdown set.
+* Fix in-memory range tombstone truncation to avoid erroneously covering newer keys at a lower level, and include range tombstones in compacted files whose largest key is the range tombstone's start key.
+* Properly set the stop key for a truncated manual CompactRange
+* 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.
+* Fix the bug that WriteBatchWithIndex's SeekForPrev() doesn't see the entries with the same key.
+* Fix the bug where user comparator was sometimes fed with InternalKey instead of the user key. The bug manifests when during GenerateBottommostFiles.
+* 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.
+* Fixed Get correctness bug in the presence of range tombstones where merge operands covered by a range tombstone always result in NotFound.
+* Start populating `NO_FILE_CLOSES` ticker statistic, which was always zero previously.
+* The default value of NewBloomFilterPolicy()'s argument use_block_based_builder is changed to false. Note that this new default may cause large temp memory usage when building very large SST files.
+* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
+* Make DB ignore dropped column families while committing results of atomic flush.
+
+## 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.
+* Expose column family id to `OnCompactionCompleted`.
+
+### 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 +220,8 @@
 * `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.

INSTALL.md

@@ -43,6 +43,8 @@ to build a portable binary, add `PORTABLE=1` before your make commands, like thi
       command line flags processing. You can compile rocksdb library even
       if you don't have gflags installed.
 
+* If you wish to build the RocksJava static target, then cmake is required for building Snappy.
+
 ## Supported platforms
 
 * **Linux - Ubuntu**

LANGUAGE-BINDINGS.md

@@ -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 

Makefile

@@ -85,13 +85,40 @@ ifeq ($(MAKECMDGOALS),rocksdbjavastaticrelease)
 	DEBUG_LEVEL=0
 endif
 
+ifeq ($(MAKECMDGOALS),rocksdbjavastaticreleasedocker)
+        DEBUG_LEVEL=0
+endif
+
 ifeq ($(MAKECMDGOALS),rocksdbjavastaticpublish)
 	DEBUG_LEVEL=0
 endif
 
+# Lite build flag.
+LITE ?= 0
+ifeq ($(LITE), 0)
+ifneq ($(filter -DROCKSDB_LITE,$(OPT)),)
+  # Be backward compatible and support older format where OPT=-DROCKSDB_LITE is
+  # specified instead of LITE=1 on the command line.
+  LITE=1
+endif
+else ifeq ($(LITE), 1)
+ifeq ($(filter -DROCKSDB_LITE,$(OPT)),)
+	OPT += -DROCKSDB_LITE
+endif
+endif
+
+# Figure out optimize level.
+ifneq ($(DEBUG_LEVEL), 2)
+ifeq ($(LITE), 0)
+	OPT += -O2
+else
+	OPT += -Os
+endif
+endif
+
 # compile with -O2 if debug level is not 2
 ifneq ($(DEBUG_LEVEL), 2)
-OPT += -O2 -fno-omit-frame-pointer
+OPT += -fno-omit-frame-pointer
 # Skip for archs that don't support -momit-leaf-frame-pointer
 ifeq (,$(shell $(CXX) -fsyntax-only -momit-leaf-frame-pointer -xc /dev/null 2>&1))
 OPT += -momit-leaf-frame-pointer
@@ -110,6 +137,12 @@ CFLAGS +=  -DHAVE_POWER8
 HAVE_POWER8=1
 endif
 
+ifeq (,$(shell $(CXX) -fsyntax-only -march=armv8-a+crc -xc /dev/null 2>&1))
+CXXFLAGS += -march=armv8-a+crc
+CFLAGS += -march=armv8-a+crc
+ARMCRC_SOURCE=1
+endif
+
 # if we're compiling for release, compile without debug code (-DNDEBUG)
 ifeq ($(DEBUG_LEVEL),0)
 OPT += -DNDEBUG
@@ -314,7 +347,7 @@ endif
 ifeq ("$(wildcard $(LUA_LIB))", "") # LUA_LIB does not exist
 $(error $(LUA_LIB) does not exist.  Try to specify both LUA_PATH and LUA_LIB manually)
 endif
-LDFLAGS += $(LUA_LIB)
+EXEC_LDFLAGS += $(LUA_LIB)
 
 endif
 
@@ -374,6 +407,8 @@ 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 +456,7 @@ TESTS = \
 	db_table_properties_test \
 	db_statistics_test \
 	db_write_test \
+	error_handler_test \
 	autovector_test \
 	blob_db_test \
 	cleanable_test \
@@ -428,6 +464,7 @@ TESTS = \
 	table_properties_collector_test \
 	arena_test \
 	block_test \
+	data_block_hash_index_test \
 	cache_test \
 	corruption_test \
 	slice_transform_test \
@@ -512,13 +549,18 @@ TESTS = \
 	persistent_cache_test \
 	statistics_test \
 	lua_test \
-	range_del_aggregator_test \
 	lru_cache_test \
 	object_registry_test \
 	repair_test \
 	env_timed_test \
 	write_prepared_transaction_test \
+	write_unprepared_transaction_test \
 	db_universal_compaction_test \
+	trace_analyzer_test \
+	repeatable_thread_test \
+	range_tombstone_fragmenter_test \
+	range_del_aggregator_test \
+	sst_file_reader_test \
 
 PARALLEL_TEST = \
 	backupable_db_test \
@@ -537,6 +579,7 @@ 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 +604,13 @@ 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 +698,7 @@ $(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 +846,7 @@ 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:
@@ -867,6 +911,7 @@ crash_test: whitebox_crash_test blackbox_crash_test
 
 blackbox_crash_test: db_stress
 	python -u tools/db_crashtest.py --simple blackbox $(CRASH_TEST_EXT_ARGS)
+	python -u tools/db_crashtest.py --enable_atomic_flush blackbox $(CRASH_TEST_EXT_ARGS)
 	python -u tools/db_crashtest.py blackbox $(CRASH_TEST_EXT_ARGS)
 
 ifeq ($(CRASH_TEST_KILL_ODD),)
@@ -875,6 +920,8 @@ endif
 
 whitebox_crash_test: db_stress
 	python -u tools/db_crashtest.py --simple whitebox --random_kill_odd \
+      $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
+	python -u tools/db_crashtest.py --enable_atomic_flush whitebox  --random_kill_odd \
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
 	python -u tools/db_crashtest.py whitebox  --random_kill_odd \
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
@@ -986,8 +1033,10 @@ 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 +1056,13 @@ 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 +1076,7 @@ $(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 +1087,9 @@ 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 +1243,9 @@ 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 +1398,9 @@ 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 +1497,9 @@ 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 +1542,9 @@ 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 +1581,21 @@ 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)
+
+range_tombstone_fragmenter_test: db/range_tombstone_fragmenter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+sst_file_reader_test: table/sst_file_reader_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1558,7 +1641,7 @@ JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
 ifeq ($(PLATFORM), OS_SOLARIS)
 	ARCH := $(shell isainfo -b)
 else ifeq ($(PLATFORM), OS_OPENBSD)
-	ifneq (,$(filter $(MACHINE), amd64 arm64 sparc64))
+	ifneq (,$(filter amd64 ppc64 ppc64le arm64 aarch64 sparc64, $(MACHINE)))
 		ARCH := 64
 	else
 		ARCH := 32
@@ -1567,10 +1650,10 @@ else
 	ARCH := $(shell getconf LONG_BIT)
 endif
 
-ifeq (,$(findstring ppc,$(MACHINE)))
-        ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
+ifneq (,$(filter ppc% arm64 aarch64 sparc64, $(MACHINE)))
+	ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE).so
 else
-        ROCKSDBJNILIB = librocksdbjni-linux-$(MACHINE).so
+	ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
 endif
 ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux$(ARCH).jar
 ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
@@ -1583,15 +1666,15 @@ ZLIB_SHA256 ?= c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1
 ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.6
 BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
-BZIP2_DOWNLOAD_BASE ?= http://www.bzip.org
-SNAPPY_VER ?= 1.1.4
-SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
-SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
-LZ4_VER ?= 1.8.0
-LZ4_SHA256 ?= 2ca482ea7a9bb103603108b5a7510b7592b90158c151ff50a28f1ca8389fccf6
+BZIP2_DOWNLOAD_BASE ?= https://downloads.sourceforge.net/project/bzip2
+SNAPPY_VER ?= 1.1.8
+SNAPPY_SHA256 ?= 16b677f07832a612b0836178db7f374e414f94657c138e6993cbfc5dcc58651f
+SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
+LZ4_VER ?= 1.9.2
+LZ4_SHA256 ?= 658ba6191fa44c92280d4aa2c271b0f4fbc0e34d249578dd05e50e76d0e5efcc
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.3.3
-ZSTD_SHA256 ?= a77c47153ee7de02626c5b2a097005786b71688be61e9fb81806a011f90b297b
+ZSTD_VER ?= 1.4.4
+ZSTD_SHA256 ?= a364f5162c7d1a455cc915e8e3cf5f4bd8b75d09bc0f53965b0c9ca1383c52c8
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
 CURL_SSL_OPTS ?= --tlsv1
 
@@ -1630,7 +1713,9 @@ endif
 
 libz.a:
 	-rm -rf zlib-$(ZLIB_VER)
-	curl -O -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+ifeq (,$(wildcard ./zlib-$(ZLIB_VER).tar.gz))
+	curl --fail --output zlib-$(ZLIB_VER).tar.gz --location ${ZLIB_DOWNLOAD_BASE}/zlib-$(ZLIB_VER).tar.gz
+endif
 	ZLIB_SHA256_ACTUAL=`$(SHA256_CMD) zlib-$(ZLIB_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZLIB_SHA256)" != "$$ZLIB_SHA256_ACTUAL" ]; then \
 		echo zlib-$(ZLIB_VER).tar.gz checksum mismatch, expected=\"$(ZLIB_SHA256)\" actual=\"$$ZLIB_SHA256_ACTUAL\"; \
@@ -1642,7 +1727,9 @@ libz.a:
 
 libbz2.a:
 	-rm -rf bzip2-$(BZIP2_VER)
-	curl -O -L ${BZIP2_DOWNLOAD_BASE}/$(BZIP2_VER)/bzip2-$(BZIP2_VER).tar.gz
+ifeq (,$(wildcard ./bzip2-$(BZIP2_VER).tar.gz))
+	curl --fail --output bzip2-$(BZIP2_VER).tar.gz --location ${CURL_SSL_OPTS} ${BZIP2_DOWNLOAD_BASE}/bzip2-$(BZIP2_VER).tar.gz
+endif
 	BZIP2_SHA256_ACTUAL=`$(SHA256_CMD) bzip2-$(BZIP2_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(BZIP2_SHA256)" != "$$BZIP2_SHA256_ACTUAL" ]; then \
 		echo bzip2-$(BZIP2_VER).tar.gz checksum mismatch, expected=\"$(BZIP2_SHA256)\" actual=\"$$BZIP2_SHA256_ACTUAL\"; \
@@ -1654,21 +1741,24 @@ libbz2.a:
 
 libsnappy.a:
 	-rm -rf snappy-$(SNAPPY_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER)/snappy-$(SNAPPY_VER).tar.gz
+ifeq (,$(wildcard ./snappy-$(SNAPPY_VER).tar.gz))
+	curl --fail --output snappy-$(SNAPPY_VER).tar.gz --location ${CURL_SSL_OPTS} ${SNAPPY_DOWNLOAD_BASE}/$(SNAPPY_VER).tar.gz
+endif
 	SNAPPY_SHA256_ACTUAL=`$(SHA256_CMD) snappy-$(SNAPPY_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(SNAPPY_SHA256)" != "$$SNAPPY_SHA256_ACTUAL" ]; then \
 		echo snappy-$(SNAPPY_VER).tar.gz checksum mismatch, expected=\"$(SNAPPY_SHA256)\" actual=\"$$SNAPPY_SHA256_ACTUAL\"; \
 		exit 1; \
 	fi
 	tar xvzf snappy-$(SNAPPY_VER).tar.gz
-	cd snappy-$(SNAPPY_VER) && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' ./configure --with-pic --enable-static --disable-shared
-	cd snappy-$(SNAPPY_VER) && $(MAKE) ${SNAPPY_MAKE_TARGET}
-	cp snappy-$(SNAPPY_VER)/.libs/libsnappy.a .
+	mkdir snappy-$(SNAPPY_VER)/build
+	cd snappy-$(SNAPPY_VER)/build && CFLAGS='${EXTRA_CFLAGS}' CXXFLAGS='${EXTRA_CXXFLAGS}' LDFLAGS='${EXTRA_LDFLAGS}' cmake -DCMAKE_POSITION_INDEPENDENT_CODE=ON .. && $(MAKE) ${SNAPPY_MAKE_TARGET}
+	cp snappy-$(SNAPPY_VER)/build/libsnappy.a .
 
 liblz4.a:
 	-rm -rf lz4-$(LZ4_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
-	mv v$(LZ4_VER).tar.gz lz4-$(LZ4_VER).tar.gz
+ifeq (,$(wildcard ./lz4-$(LZ4_VER).tar.gz))
+	curl --fail --output lz4-$(LZ4_VER).tar.gz --location ${CURL_SSL_OPTS} ${LZ4_DOWNLOAD_BASE}/v$(LZ4_VER).tar.gz
+endif
 	LZ4_SHA256_ACTUAL=`$(SHA256_CMD) lz4-$(LZ4_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(LZ4_SHA256)" != "$$LZ4_SHA256_ACTUAL" ]; then \
 		echo lz4-$(LZ4_VER).tar.gz checksum mismatch, expected=\"$(LZ4_SHA256)\" actual=\"$$LZ4_SHA256_ACTUAL\"; \
@@ -1680,8 +1770,9 @@ liblz4.a:
 
 libzstd.a:
 	-rm -rf zstd-$(ZSTD_VER)
-	curl -O -L ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
-	mv v$(ZSTD_VER).tar.gz zstd-$(ZSTD_VER).tar.gz
+ifeq (,$(wildcard ./zstd-$(ZSTD_VER).tar.gz))
+	curl --fail --output zstd-$(ZSTD_VER).tar.gz --location ${CURL_SSL_OPTS} ${ZSTD_DOWNLOAD_BASE}/v$(ZSTD_VER).tar.gz
+endif
 	ZSTD_SHA256_ACTUAL=`$(SHA256_CMD) zstd-$(ZSTD_VER).tar.gz | cut -d ' ' -f 1`; \
 	if [ "$(ZSTD_SHA256)" != "$$ZSTD_SHA256_ACTUAL" ]; then \
 		echo zstd-$(ZSTD_VER).tar.gz checksum mismatch, expected=\"$(ZSTD_SHA256)\" actual=\"$$ZSTD_SHA256_ACTUAL\"; \
@@ -1743,20 +1834,26 @@ 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
@@ -1766,6 +1863,14 @@ rocksdbjavastaticdockerppc64le:
 	fi
 	docker start -a rocksdb_linux_ppc64le-be
 
+rocksdbjavastaticdockerarm64v8:
+	mkdir -p java/target
+	DOCKER_LINUX_ARM64V8_CONTAINER=`docker ps -aqf name=rocksdb_linux_arm64v8-be`; \
+	if [ -z "$$DOCKER_LINUX_ARM64V8_CONTAINER" ]; then \
+		docker container create --attach stdin --attach stdout --attach stderr --volume `pwd`:/rocksdb-host --name rocksdb_linux_arm64v8-be evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux-centos.sh; \
+	fi
+	docker start -a rocksdb_linux_arm64v8-be
+
 rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral
 
 rocksdbjavastaticpublishdocker: rocksdbjavastaticreleasedocker rocksdbjavastaticpublishcentral
@@ -1812,7 +1917,7 @@ $(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 +1986,7 @@ 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

README.md

@@ -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)

TARGETS

@@ -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 = [
@@ -65,6 +67,14 @@ is_opt_mode = build_mode.startswith("opt")
 if is_opt_mode:
     rocksdb_compiler_flags.append("-DNDEBUG")
 
+sanitizer = read_config("fbcode", "sanitizer")
+
+# Do not enable jemalloc if sanitizer presents. RocksDB will further detect
+# whether the binary is linked with jemalloc at runtime.
+if sanitizer == "":
+    rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
+    rocksdb_external_deps.append(("jemalloc", None, "headers"))
+
 cpp_library(
     name = "rocksdb_lib",
     srcs = [
@@ -79,6 +89,7 @@ cpp_library(
         "db/compaction_iterator.cc",
         "db/compaction_job.cc",
         "db/compaction_picker.cc",
+        "db/compaction_picker_fifo.cc",
         "db/compaction_picker_universal.cc",
         "db/convenience.cc",
         "db/db_filesnapshot.cc",
@@ -93,6 +104,7 @@ 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",
@@ -105,12 +117,12 @@ 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",
         "db/merge_operator.cc",
         "db/range_del_aggregator.cc",
+        "db/range_tombstone_fragmenter.cc",
         "db/repair.cc",
         "db/snapshot_impl.cc",
         "db/table_cache.cc",
@@ -171,6 +183,8 @@ 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",
@@ -186,6 +200,7 @@ cpp_library(
         "table/plain_table_index.cc",
         "table/plain_table_key_coding.cc",
         "table/plain_table_reader.cc",
+        "table/sst_file_reader.cc",
         "table/sst_file_writer.cc",
         "table/table_properties.cc",
         "table/two_level_iterator.cc",
@@ -200,6 +215,7 @@ 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",
@@ -210,6 +226,7 @@ cpp_library(
         "util/filename.cc",
         "util/filter_policy.cc",
         "util/hash.cc",
+        "util/jemalloc_nodump_allocator.cc",
         "util/log_buffer.cc",
         "util/murmurhash.cc",
         "util/random.cc",
@@ -217,24 +234,24 @@ 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 +286,7 @@ 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",
@@ -280,11 +298,13 @@ 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",
     ],
-    headers = AutoHeaders.RECURSIVE_GLOB,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
     arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
     compiler_flags = rocksdb_compiler_flags,
     preprocessor_flags = rocksdb_preprocessor_flags,
@@ -297,6 +317,7 @@ 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",
@@ -305,7 +326,7 @@ cpp_library(
         "utilities/col_buf_encoder.cc",
         "utilities/column_aware_encoding_util.cc",
     ],
-    headers = AutoHeaders.RECURSIVE_GLOB,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
     arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
     compiler_flags = rocksdb_compiler_flags,
     preprocessor_flags = rocksdb_preprocessor_flags,
@@ -317,9 +338,10 @@ cpp_library(
     name = "rocksdb_tools_lib",
     srcs = [
         "tools/db_bench_tool.cc",
+        "tools/trace_analyzer_tool.cc",
         "util/testutil.cc",
     ],
-    headers = AutoHeaders.RECURSIVE_GLOB,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
     arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
     compiler_flags = rocksdb_compiler_flags,
     preprocessor_flags = rocksdb_preprocessor_flags,
@@ -330,7 +352,7 @@ cpp_library(
 cpp_library(
     name = "env_basic_test_lib",
     srcs = ["env/env_basic_test.cc"],
-    headers = AutoHeaders.RECURSIVE_GLOB,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
     arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
     compiler_flags = rocksdb_compiler_flags,
     preprocessor_flags = rocksdb_preprocessor_flags,
@@ -495,6 +517,11 @@ 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 +583,13 @@ 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 +712,11 @@ 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",
@@ -900,6 +932,11 @@ ROCKS_TESTS = [
         "db/range_del_aggregator_test.cc",
         "serial",
     ],
+    [
+        "range_tombstone_fragmenter_test",
+        "db/range_tombstone_fragmenter_test.cc",
+        "serial",
+    ],
     [
         "rate_limiter_test",
         "util/rate_limiter_test.cc",
@@ -915,6 +952,11 @@ 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",
@@ -940,6 +982,11 @@ ROCKS_TESTS = [
         "tools/sst_dump_test.cc",
         "serial",
     ],
+    [
+        "sst_file_reader_test",
+        "table/sst_file_reader_test.cc",
+        "serial",
+    ],
     [
         "statistics_test",
         "monitoring/statistics_test.cc",
@@ -975,6 +1022,11 @@ 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 +1092,11 @@ 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 +1109,19 @@ 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",
-)

USERS.md

@@ -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.

WINDOWS_PORT.md

@@ -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)

appveyor.yml

@@ -1,15 +1,75 @@
 version: 1.0.{build}
+
 image: Visual Studio 2017
+
+environment:
+  JAVA_HOME: C:\Program Files\Java\jdk1.8.0
+  THIRDPARTY_HOME: $(APPVEYOR_BUILD_FOLDER)\thirdparty
+  SNAPPY_HOME: $(THIRDPARTY_HOME)\snappy-1.1.7
+  SNAPPY_INCLUDE: $(SNAPPY_HOME);$(SNAPPY_HOME)\build
+  SNAPPY_LIB_DEBUG: $(SNAPPY_HOME)\build\Debug\snappy.lib
+  SNAPPY_LIB_RELEASE: $(SNAPPY_HOME)\build\Release\snappy.lib
+  LZ4_HOME: $(THIRDPARTY_HOME)\lz4-1.8.3
+  LZ4_INCLUDE: $(LZ4_HOME)\lib
+  LZ4_LIB_DEBUG: $(LZ4_HOME)\visual\VS2010\bin\x64_Debug\liblz4_static.lib
+  LZ4_LIB_RELEASE: $(LZ4_HOME)\visual\VS2010\bin\x64_Release\liblz4_static.lib
+  ZSTD_HOME: $(THIRDPARTY_HOME)\zstd-1.4.0
+  ZSTD_INCLUDE: $(ZSTD_HOME)\lib;$(ZSTD_HOME)\lib\dictBuilder
+  ZSTD_LIB_DEBUG: $(ZSTD_HOME)\build\VS2010\bin\x64_Debug\libzstd_static.lib
+  ZSTD_LIB_RELEASE: $(ZSTD_HOME)\build\VS2010\bin\x64_Release\libzstd_static.lib
+  matrix:
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
+      CMAKE_GENERATOR: Visual Studio 14 Win64
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio 14.0\Common7\IDE\devenv.com
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
+      CMAKE_GENERATOR: Visual Studio 15 Win64
+      DEV_ENV: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\devenv.com
+
+install:
+  - md %THIRDPARTY_HOME%
+  - echo "Building Snappy dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output snappy-1.1.7.zip --location https://github.com/google/snappy/archive/1.1.7.zip
+  - unzip snappy-1.1.7.zip
+  - cd snappy-1.1.7
+  - mkdir build
+  - cd build
+  - cmake -G "%CMAKE_GENERATOR%" ..
+  - msbuild Snappy.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild Snappy.sln /p:Configuration=Release /p:Platform=x64
+  - echo "Building LZ4 dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output lz4-1.8.3.zip --location https://github.com/lz4/lz4/archive/v1.8.3.zip
+  - unzip lz4-1.8.3.zip
+  - cd lz4-1.8.3\visual\VS2010
+  - ps: $CMD="$Env:DEV_ENV"; & $CMD lz4.sln /upgrade
+  - msbuild lz4.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild lz4.sln /p:Configuration=Release /p:Platform=x64
+  - echo "Building ZStd dependency..."
+  - cd %THIRDPARTY_HOME%
+  - curl --fail --silent --show-error --output zstd-1.4.0.zip --location https://github.com/facebook/zstd/archive/v1.4.0.zip
+  - unzip zstd-1.4.0.zip
+  - cd zstd-1.4.0\build\VS2010
+  - ps: $CMD="$Env:DEV_ENV"; & $CMD zstd.sln /upgrade
+  - msbuild zstd.sln /p:Configuration=Debug /p:Platform=x64
+  - msbuild zstd.sln /p:Configuration=Release /p:Platform=x64
+
 before_build:
-- md %APPVEYOR_BUILD_FOLDER%\build
-- cd %APPVEYOR_BUILD_FOLDER%\build
-- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 ..
-- cd ..
+  - md %APPVEYOR_BUILD_FOLDER%\build
+  - cd %APPVEYOR_BUILD_FOLDER%\build
+  - cmake -G "%CMAKE_GENERATOR%" -DCMAKE_BUILD_TYPE=Debug -DOPTDBG=1 -DPORTABLE=1 -DSNAPPY=1 -DLZ4=1 -DZSTD=1 -DXPRESS=1 -DJNI=1 ..
+  - cd ..
+
 build:
   project: build\rocksdb.sln
   parallel: true
   verbosity: normal
-test:
-test_script:
-- ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test -Concurrency 8
+
+test:
+
+test_script:
+  - ps: build_tools\run_ci_db_test.ps1 -SuiteRun db_basic_test,db_test2,db_test,env_basic_test,env_test,db_merge_operand_test -Concurrency 8
+
+on_failure:
+  - cmd: 7z a build-failed.zip %APPVEYOR_BUILD_FOLDER%\build\ && appveyor PushArtifact build-failed.zip
 

buckifier/buckify_rocksdb.py

@@ -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"])
 

buckifier/targets_builder.py

@@ -10,7 +10,7 @@ def pretty_list(lst, indent=8):
 
     if len(lst) == 1:
         return "\"%s\"" % lst[0]
-    
+
     separator = "\",\n%s\"" % (" " * indent)
     res = separator.join(sorted(lst))
     res = "\n" + (" " * indent) + "\"" + res + "\",\n" + (" " * (indent - 4))
@@ -31,13 +31,16 @@ class TARGETSBuilder:
         self.targets_file.close()
 
     def add_library(self, name, srcs, deps=None, headers=None):
+        headers_attr_prefix = ""
         if headers is None:
+            headers_attr_prefix = "auto_"
             headers = "AutoHeaders.RECURSIVE_GLOB"
-        self.targets_file.write(targets_cfg.library_template % (
-            name,
-            pretty_list(srcs),
-            headers,
-            pretty_list(deps)))
+        self.targets_file.write(targets_cfg.library_template.format(
+            name=name,
+            srcs=pretty_list(srcs),
+            headers_attr_prefix=headers_attr_prefix,
+            headers=headers,
+            deps=pretty_list(deps)))
         self.total_lib = self.total_lib + 1
 
     def add_binary(self, name, srcs, deps=None):

buckifier/targets_cfg.py

@@ -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 = [
@@ -68,18 +70,26 @@ is_opt_mode = build_mode.startswith("opt")
 # doesn't harm and avoid forgetting to add it.
 if is_opt_mode:
     rocksdb_compiler_flags.append("-DNDEBUG")
+
+sanitizer = read_config("fbcode", "sanitizer")
+
+# Do not enable jemalloc if sanitizer presents. RocksDB will further detect
+# whether the binary is linked with jemalloc at runtime.
+if sanitizer == "":
+    rocksdb_compiler_flags.append("-DROCKSDB_JEMALLOC")
+    rocksdb_external_deps.append(("jemalloc", None, "headers"))
 """
 
 
 library_template = """
 cpp_library(
-    name = "%s",
-    srcs = [%s],
-    headers = %s,
+    name = "{name}",
+    srcs = [{srcs}],
+    {headers_attr_prefix}headers = {headers},
     arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
     compiler_flags = rocksdb_compiler_flags,
     preprocessor_flags = rocksdb_preprocessor_flags,
-    deps = [%s],
+    deps = [{deps}],
     external_deps = rocksdb_external_deps,
 )
 """
@@ -118,32 +128,20 @@ 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",
-)
 """

build_tools/RocksDBCommonHelper.php

@@ -7,12 +7,12 @@
 // Name of the environment variables which need to be set by the entity which
 // triggers continuous runs so that code at the end of the file gets executed
 // and Sandcastle run starts.
-define("ENV_POST_RECEIVE_HOOK", "POST_RECEIVE_HOOK");
-define("ENV_HTTPS_APP_VALUE", "HTTPS_APP_VALUE");
-define("ENV_HTTPS_TOKEN_VALUE", "HTTPS_TOKEN_VALUE");
+const ENV_POST_RECEIVE_HOOK = "POST_RECEIVE_HOOK";
+const ENV_HTTPS_APP_VALUE = "HTTPS_APP_VALUE";
+const ENV_HTTPS_TOKEN_VALUE = "HTTPS_TOKEN_VALUE";
 
-define("PRIMARY_TOKEN_FILE", '/home/krad/.sandcastle');
-define("CONT_RUN_ALIAS", "leveldb");
+const PRIMARY_TOKEN_FILE = '/home/krad/.sandcastle';
+const CONT_RUN_ALIAS = "leveldb";
 
 //////////////////////////////////////////////////////////////////////
 /*  Run tests in sandcastle */

build_tools/build_detect_platform

@@ -53,11 +53,13 @@ if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
     FBCODE_BUILD="true"
     # If we're compiling with TSAN we need pic build
     PIC_BUILD=$COMPILE_WITH_TSAN
-    if [ -z "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
-      source "$PWD/build_tools/fbcode_config.sh"
-    else
+    if [ -n "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
       # we need this to build with MySQL. Don't use for other purposes.
       source "$PWD/build_tools/fbcode_config4.8.1.sh"
+    elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_5xx" ]; then
+      source "$PWD/build_tools/fbcode_config.sh"
+    else
+      source "$PWD/build_tools/fbcode_config_platform007.sh"
     fi
 fi
 
@@ -474,6 +476,17 @@ EOF
             COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SCHED_GETCPU_PRESENT"
         fi
     fi
+
+    if ! test $ROCKSDB_DISABLE_ALIGNED_NEW; then
+        # Test whether c++17 aligned-new is supported
+        $CXX $PLATFORM_CXXFLAGS -faligned-new -x c++ - -o /dev/null 2>/dev/null <<EOF
+            struct alignas(1024) t {int a;};
+            int main() {}
+EOF
+        if [ "$?" = 0 ]; then
+            PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS -faligned-new -DHAVE_ALIGNED_NEW"
+        fi
+    fi
 fi
 
 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -514,6 +527,8 @@ if test -z "$PORTABLE"; then
   elif test -n "`echo $TARGET_ARCHITECTURE | grep ^arm`"; then
     # TODO: Handle this with approprite options.
     COMMON_FLAGS="$COMMON_FLAGS"
+  elif test -n "`echo $TARGET_ARCHITECTURE | grep ^aarch64`"; then
+    COMMON_FLAGS="$COMMON_FLAGS"
   elif [ "$TARGET_OS" == "IOS" ]; then
     COMMON_FLAGS="$COMMON_FLAGS"
   elif [ "$TARGET_OS" != "AIX" ] && [ "$TARGET_OS" != "SunOS" ]; then

build_tools/dependencies.sh

@@ -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

build_tools/dependencies_platform007.sh

@@ -0,0 +1,18 @@
+GCC_BASE=/mnt/gvfs/third-party2/gcc/6e8e715624fd15256a7970073387793dfcf79b46/7.x/centos7-native/b2ef2b6
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/ef37e1faa1c29782abfac1ae65a291b9b7966f6d/stable/centos7-native/c9f9104
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/c67031f0f739ac61575a061518d6ef5038f99f90/7.x/platform007/5620abc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/60d6f124a78798b73944f5ba87c2306ae3460153/2.26/platform007/f259413
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/22c2d65676fb7c23cfa797c4f6937f38b026f3cf/1.2.8/platform007/ca4da3d
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/907b498203d297947f3bb70b9466f47e100f1873/r131/platform007/ca4da3d
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/3ee276cbacfad3074e3f07bf826ac47f06970f4e/1.3.5/platform007/15a3614
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/9c910d36d6235cc40e8ff559358f1833452300ca/master/platform007/5b0f53e
+NUMA_BASE=/mnt/gvfs/third-party2/numa/9cbf2460284c669ed19c3ccb200a71f7dd7e53c7/2.0.11/platform007/ca4da3d
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/bf3d7497fe4e6d007354f0adffa16ce3003f8338/1.3/platform007/6f3e0a9
+TBB_BASE=/mnt/gvfs/third-party2/tbb/ff4e0b093534704d8abab678a4fd7f5ea7b094c7/2018_U5/platform007/ca4da3d
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/b5c4a61a5c483ba24722005ae07895971a2ac707/fb/platform007/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/92ff90349e2f43ea0a8246d8b1cf17b6869013e3/2.29.1/centos7-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/f3f697a28122e6bcd513273dd9c1ff23852fc59f/3.13.0/platform007/ca4da3d
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832

build_tools/error_filter.py

@@ -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):

build_tools/fbcode_config.sh

@@ -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

build_tools/fbcode_config_platform007.sh

@@ -0,0 +1,157 @@
+#!/bin/sh
+#
+# Set environment variables so that we can compile rocksdb using
+# fbcode settings.  It uses the latest g++ and clang compilers and also
+# uses jemalloc
+# Environment variables that change the behavior of this script:
+# PIC_BUILD -- if true, it will only take pic versions of libraries from fbcode. libraries that don't have pic variant will not be included
+
+
+BASEDIR=`dirname $BASH_SOURCE`
+source "$BASEDIR/dependencies_platform007.sh"
+
+CFLAGS=""
+
+# libgcc
+LIBGCC_INCLUDE="$LIBGCC_BASE/include/c++/7.3.0"
+LIBGCC_LIBS=" -L $LIBGCC_BASE/lib"
+
+# glibc
+GLIBC_INCLUDE="$GLIBC_BASE/include"
+GLIBC_LIBS=" -L $GLIBC_BASE/lib"
+
+# snappy
+SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include/"
+if test -z $PIC_BUILD; then
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy.a"
+else
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy_pic.a"
+fi
+CFLAGS+=" -DSNAPPY"
+
+if test -z $PIC_BUILD; then
+  # location of zlib headers and libraries
+  ZLIB_INCLUDE=" -I $ZLIB_BASE/include/"
+  ZLIB_LIBS=" $ZLIB_BASE/lib/libz.a"
+  CFLAGS+=" -DZLIB"
+
+  # location of bzip headers and libraries
+  BZIP_INCLUDE=" -I $BZIP2_BASE/include/"
+  BZIP_LIBS=" $BZIP2_BASE/lib/libbz2.a"
+  CFLAGS+=" -DBZIP2"
+
+  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
+  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
+  CFLAGS+=" -DLZ4"
+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
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags.a"
+else
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
+fi
+CFLAGS+=" -DGFLAGS=gflags"
+
+# location of jemalloc
+JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
+JEMALLOC_LIB=" $JEMALLOC_BASE/lib/libjemalloc.a"
+
+if test -z $PIC_BUILD; then
+  # location of numa
+  NUMA_INCLUDE=" -I $NUMA_BASE/include/"
+  NUMA_LIB=" $NUMA_BASE/lib/libnuma.a"
+  CFLAGS+=" -DNUMA"
+
+  # location of libunwind
+  LIBUNWIND="$LIBUNWIND_BASE/lib/libunwind.a"
+fi
+
+# location of TBB
+TBB_INCLUDE=" -isystem $TBB_BASE/include/"
+if test -z $PIC_BUILD; then
+  TBB_LIBS="$TBB_BASE/lib/libtbb.a"
+else
+  TBB_LIBS="$TBB_BASE/lib/libtbb_pic.a"
+fi
+CFLAGS+=" -DTBB"
+
+# use Intel SSE support for checksum calculations
+export USE_SSE=1
+export PORTABLE=1
+
+BINUTILS="$BINUTILS_BASE/bin"
+AR="$BINUTILS/ar"
+
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $ZSTD_INCLUDE $GFLAGS_INCLUDE $NUMA_INCLUDE $TBB_INCLUDE"
+
+STDLIBS="-L $GCC_BASE/lib64"
+
+CLANG_BIN="$CLANG_BASE/bin"
+CLANG_LIB="$CLANG_BASE/lib"
+CLANG_SRC="$CLANG_BASE/../../src"
+
+CLANG_ANALYZER="$CLANG_BIN/clang++"
+CLANG_SCAN_BUILD="$CLANG_SRC/llvm/tools/clang/tools/scan-build/bin/scan-build"
+
+if [ -z "$USE_CLANG" ]; then
+  # gcc
+  CC="$GCC_BASE/bin/gcc"
+  CXX="$GCC_BASE/bin/g++"
+
+  CFLAGS+=" -B$BINUTILS/gold"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  JEMALLOC=1
+else
+  # clang
+  CLANG_INCLUDE="$CLANG_LIB/clang/stable/include"
+  CC="$CLANG_BIN/clang"
+  CXX="$CLANG_BIN/clang++"
+
+  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include"
+
+  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x/x86_64-facebook-linux "
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $CLANG_INCLUDE"
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE/linux "
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE "
+  CFLAGS+=" -Wno-expansion-to-defined "
+  CXXFLAGS="-nostdinc++"
+fi
+
+CFLAGS+=" $DEPS_INCLUDE"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
+CXXFLAGS+=" $CFLAGS"
+
+EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
+EXEC_LDFLAGS+=" -B$BINUTILS/gold"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
+EXEC_LDFLAGS+=" $LIBUNWIND"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
+# required by libtbb
+EXEC_LDFLAGS+=" -ldl"
+
+PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
+
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
+
+VALGRIND_VER="$VALGRIND_BASE/bin/"
+
+# lua not supported because it's on track for deprecation, I think
+LUA_PATH=
+LUA_LIB=
+
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB

build_tools/gnu_parallel

@@ -5082,8 +5082,8 @@ sub openoutputfiles {
 	# Set reading FD if using --group (--ungroup does not need)
 	for my $fdno (1,2) {
 	    # Re-open the file for reading
-	    # so fdw can be closed seperately
-	    # and fdr can be seeked seperately (for --line-buffer)
+	    # so fdw can be closed separately
+	    # and fdr can be seeked separately (for --line-buffer)
 	    open(my $fdr,"<", $self->fh($fdno,'name')) ||
 		::die_bug("fdr: Cannot open ".$self->fh($fdno,'name'));
 	    $self->set_fh($fdno,'r',$fdr);

build_tools/regression_build_test.sh

@@ -377,7 +377,7 @@ function send_to_ods {
     echo >&2 "ERROR: Key $key doesn't have a value."
     return
   fi
-  curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
+  curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build$git_br&key=$key&value=$value" \
     --connect-timeout 60
 }
 

build_tools/rocksdb-lego-determinator

@@ -85,9 +85,12 @@ NON_SHM="TMPD=/tmp/rocksdb_test_tmp"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-LITE="OPT=\"-DROCKSDB_LITE -g\""
-TSAN="COMPILE_WITH_TSAN=1"
+# in gcc-5 there are known problems with TSAN like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71090.
+# using platform007 gives us gcc-8 or higher which has that bug fixed.
+TSAN="ROCKSDB_FBCODE_BUILD_WITH_PLATFORM007=1 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"
@@ -343,7 +346,7 @@ LITE_BUILD_COMMANDS="[
             $CLEANUP_ENV,
             {
                 'name':'Build RocksDB debug version',
-                'shell':'$LITE make J=1 all check || $CONTRUN_NAME=lite $TASK_CREATION_TOOL',
+                'shell':'make J=1 LITE=1 all check || $CONTRUN_NAME=lite $TASK_CREATION_TOOL',
                 'user':'root',
                 $PARSER
             },
@@ -380,14 +383,14 @@ 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',
                 $PARSER
             }
@@ -452,7 +455,7 @@ ASAN_CRASH_TEST_COMMANDS="[
             {
                 'name':'Build and run RocksDB debug asan_crash_test',
                 'timeout': 86400,
-                'shell':'$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',
                 $PARSER
             },
@@ -494,7 +497,7 @@ UBSAN_CRASH_TEST_COMMANDS="[
             {
                 'name':'Build and run RocksDB debug ubsan_crash_test',
                 'timeout': 86400,
-                'shell':'$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 +563,7 @@ 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
             },
@@ -643,7 +646,7 @@ run_regression()
 
   # parameters: $1 -- key, $2 -- value
   function send_size_to_ods {
-    curl -s "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
+    curl --silent "https://www.intern.facebook.com/intern/agent/ods_set.php?entity=rocksdb_build&key=rocksdb.build_size.$1&value=$2" \
       --connect-timeout 60
   }
 
@@ -661,12 +664,12 @@ run_regression()
 
   # === lite build ===
   make clean
-  OPT=-DROCKSDB_LITE make -j$(nproc) static_lib
+  make LITE=1 -j$(nproc) static_lib
   send_size_to_ods static_lib_lite $(stat --printf="%s" librocksdb.a)
   strip librocksdb.a
   send_size_to_ods static_lib_lite_stripped $(stat --printf="%s" librocksdb.a)
 
-  OPT=-DROCKSDB_LITE make -j$(nproc) shared_lib
+  make LITE=1 -j$(nproc) shared_lib
   send_size_to_ods shared_lib_lite $(stat --printf="%s" `readlink -f librocksdb.so`)
   strip `readlink -f librocksdb.so`
   send_size_to_ods shared_lib_lite_stripped $(stat --printf="%s" `readlink -f librocksdb.so`)

build_tools/update_dependencies.sh

@@ -53,6 +53,45 @@ function get_lib_base()
   log_variable $__res_var
 }
 
+###########################################################
+#                platform007 dependencies                 #
+###########################################################
+
+OUTPUT="$BASEDIR/dependencies_platform007.sh"
+
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+echo "Writing dependencies to $OUTPUT"
+
+# Compilers locations
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/7.x/centos7-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos7-native/*/`
+
+log_variable GCC_BASE
+log_variable CLANG_BASE
+
+# Libraries locations
+get_lib_base libgcc     7.x     platform007
+get_lib_base glibc      2.26    platform007
+get_lib_base snappy     LATEST  platform007
+get_lib_base zlib       LATEST  platform007
+get_lib_base bzip2      LATEST  platform007
+get_lib_base lz4        LATEST  platform007
+get_lib_base zstd       LATEST  platform007
+get_lib_base gflags     LATEST  platform007
+get_lib_base jemalloc   LATEST  platform007
+get_lib_base numa       LATEST  platform007
+get_lib_base libunwind  LATEST  platform007
+get_lib_base tbb        LATEST  platform007
+
+get_lib_base kernel-headers fb platform007
+get_lib_base binutils   LATEST centos7-native
+get_lib_base valgrind   LATEST platform007
+get_lib_base lua        5.3.4  platform007
+
+git diff $OUTPUT
+
 ###########################################################
 #                   5.x dependencies                      #
 ###########################################################
@@ -65,8 +104,8 @@ touch "$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 +125,7 @@ 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
 

cache/cache_test.cc

@@ -64,8 +64,8 @@ class CacheTest : public testing::TestWithParam<std::string> {
 
   std::vector<int> deleted_keys_;
   std::vector<int> deleted_values_;
-  shared_ptr<Cache> cache_;
-  shared_ptr<Cache> cache2_;
+  std::shared_ptr<Cache> cache_;
+  std::shared_ptr<Cache> cache2_;
 
   CacheTest()
       : cache_(NewCache(kCacheSize, kNumShardBits, false)),
@@ -145,7 +145,7 @@ class CacheTest : public testing::TestWithParam<std::string> {
 CacheTest* CacheTest::current_;
 
 TEST_P(CacheTest, UsageTest) {
-  // cache is shared_ptr and will be automatically cleaned up.
+  // cache is std::shared_ptr and will be automatically cleaned up.
   const uint64_t kCapacity = 100000;
   auto cache = NewCache(kCapacity, 8, false);
 
@@ -173,7 +173,7 @@ TEST_P(CacheTest, UsageTest) {
 }
 
 TEST_P(CacheTest, PinnedUsageTest) {
-  // cache is shared_ptr and will be automatically cleaned up.
+  // cache is std::shared_ptr and will be automatically cleaned up.
   const uint64_t kCapacity = 100000;
   auto cache = NewCache(kCapacity, 8, false);
 

cache/lru_cache.cc

@@ -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);
 }
@@ -472,8 +461,10 @@ std::string LRUCacheShard::GetPrintableOptions() const {
 }
 
 LRUCache::LRUCache(size_t capacity, int num_shard_bits,
-                   bool strict_capacity_limit, double high_pri_pool_ratio)
-    : ShardedCache(capacity, num_shard_bits, strict_capacity_limit) {
+                   bool strict_capacity_limit, double high_pri_pool_ratio,
+                   std::shared_ptr<MemoryAllocator> allocator)
+    : ShardedCache(capacity, num_shard_bits, strict_capacity_limit,
+                   std::move(allocator)) {
   num_shards_ = 1 << num_shard_bits;
   shards_ = reinterpret_cast<LRUCacheShard*>(
       port::cacheline_aligned_alloc(sizeof(LRUCacheShard) * num_shards_));
@@ -485,10 +476,13 @@ 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 +507,17 @@ 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() {
@@ -537,12 +539,14 @@ double LRUCache::GetHighPriPoolRatio() {
 std::shared_ptr<Cache> NewLRUCache(const LRUCacheOptions& cache_opts) {
   return NewLRUCache(cache_opts.capacity, cache_opts.num_shard_bits,
                      cache_opts.strict_capacity_limit,
-                     cache_opts.high_pri_pool_ratio);
+                     cache_opts.high_pri_pool_ratio,
+                     cache_opts.memory_allocator);
 }
 
-std::shared_ptr<Cache> NewLRUCache(size_t capacity, int num_shard_bits,
-                                   bool strict_capacity_limit,
-                                   double high_pri_pool_ratio) {
+std::shared_ptr<Cache> NewLRUCache(
+    size_t capacity, int num_shard_bits, bool strict_capacity_limit,
+    double high_pri_pool_ratio,
+    std::shared_ptr<MemoryAllocator> memory_allocator) {
   if (num_shard_bits >= 20) {
     return nullptr;  // the cache cannot be sharded into too many fine pieces
   }
@@ -554,7 +558,8 @@ std::shared_ptr<Cache> NewLRUCache(size_t capacity, int num_shard_bits,
     num_shard_bits = GetDefaultCacheShardBits(capacity);
   }
   return std::make_shared<LRUCache>(capacity, num_shard_bits,
-                                    strict_capacity_limit, high_pri_pool_ratio);
+                                    strict_capacity_limit, high_pri_pool_ratio,
+                                    std::move(memory_allocator));
 }
 
 }  // namespace rocksdb

cache/lru_cache.h

@@ -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);
@@ -288,7 +279,8 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard : public CacheShard {
 class LRUCache : public ShardedCache {
  public:
   LRUCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit,
-           double high_pri_pool_ratio);
+           double high_pri_pool_ratio,
+           std::shared_ptr<MemoryAllocator> memory_allocator = nullptr);
   virtual ~LRUCache();
   virtual const char* Name() const override { return "LRUCache"; }
   virtual CacheShard* GetShard(int shard) override;
@@ -304,7 +296,7 @@ class LRUCache : public ShardedCache {
   double GetHighPriPoolRatio();
 
  private:
-  LRUCacheShard* shards_;
+  LRUCacheShard* shards_ = nullptr;
   int num_shards_ = 0;
 };
 

cache/lru_cache_test.cc

@@ -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

cache/sharded_cache.cc

@@ -20,8 +20,10 @@
 namespace rocksdb {
 
 ShardedCache::ShardedCache(size_t capacity, int num_shard_bits,
-                           bool strict_capacity_limit)
-    : num_shard_bits_(num_shard_bits),
+                           bool strict_capacity_limit,
+                           std::shared_ptr<MemoryAllocator> allocator)
+    : Cache(std::move(allocator)),
+      num_shard_bits_(num_shard_bits),
       capacity_(capacity),
       strict_capacity_limit_(strict_capacity_limit),
       last_id_(1) {}
@@ -142,6 +144,9 @@ std::string ShardedCache::GetPrintableOptions() const {
              strict_capacity_limit_);
     ret.append(buffer);
   }
+  snprintf(buffer, kBufferSize, "    memory_allocator : %s\n",
+           memory_allocator() ? memory_allocator()->Name() : "None");
+  ret.append(buffer);
   ret.append(GetShard(0)->GetPrintableOptions());
   return ret;
 }

cache/sharded_cache.h

@@ -47,7 +47,8 @@ class CacheShard {
 // Keys are sharded by the highest num_shard_bits bits of hash value.
 class ShardedCache : public Cache {
  public:
-  ShardedCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit);
+  ShardedCache(size_t capacity, int num_shard_bits, bool strict_capacity_limit,
+               std::shared_ptr<MemoryAllocator> memory_allocator = nullptr);
   virtual ~ShardedCache() = default;
   virtual const char* Name() const override = 0;
   virtual CacheShard* GetShard(int shard) = 0;

coverage/parse_gcov_output.py

@@ -1,4 +1,3 @@
-import optparse
 import re
 import sys
 

db/builder.cc

@@ -18,6 +18,7 @@
 #include "db/event_helpers.h"
 #include "db/internal_stats.h"
 #include "db/merge_helper.h"
+#include "db/range_del_aggregator.h"
 #include "db/table_cache.h"
 #include "db/version_edit.h"
 #include "monitoring/iostats_context_imp.h"
@@ -28,6 +29,7 @@
 #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"
@@ -64,8 +66,9 @@ Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
     const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
     TableCache* table_cache, InternalIterator* iter,
-    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
-    const InternalKeyComparator& internal_comparator,
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+        range_del_iters,
+    FileMetaData* meta, const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
     uint32_t column_family_id, const std::string& column_family_name,
@@ -85,12 +88,10 @@ Status BuildTable(
   Status s;
   meta->fd.file_size = 0;
   iter->SeekToFirst();
-  std::unique_ptr<RangeDelAggregator> range_del_agg(
-      new RangeDelAggregator(internal_comparator, snapshots));
-  s = range_del_agg->AddTombstones(std::move(range_del_iter));
-  if (!s.ok()) {
-    // may be non-ok if a range tombstone key is unparsable
-    return s;
+  std::unique_ptr<CompactionRangeDelAggregator> range_del_agg(
+      new CompactionRangeDelAggregator(&internal_comparator, snapshots));
+  for (auto& range_del_iter : range_del_iters) {
+    range_del_agg->AddTombstones(std::move(range_del_iter));
   }
 
   std::string fname = TableFileName(ioptions.cf_paths, meta->fd.GetNumber(),
@@ -101,11 +102,11 @@ 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;
+    std::unique_ptr<WritableFileWriter> file_writer;
     {
-      unique_ptr<WritableFile> file;
+      std::unique_ptr<WritableFile> file;
 #ifndef NDEBUG
       bool use_direct_writes = env_options.use_direct_writes;
       TEST_SYNC_POINT_CALLBACK("BuildTable:create_file", &use_direct_writes);
@@ -120,8 +121,9 @@ 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,
+                                               ioptions.listeners));
       builder = NewTableBuilder(
           ioptions, mutable_cf_options, internal_comparator,
           int_tbl_prop_collector_factories, column_family_id,
@@ -139,6 +141,7 @@ 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 +157,20 @@ 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);
+
+    auto range_del_it = range_del_agg->NewIterator();
+    for (range_del_it->SeekToFirst(); range_del_it->Valid();
+         range_del_it->Next()) {
+      auto tombstone = range_del_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 +183,7 @@ 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 +207,7 @@ 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

db/builder.h

@@ -9,6 +9,7 @@
 #include <string>
 #include <utility>
 #include <vector>
+#include "db/range_tombstone_fragmenter.h"
 #include "db/table_properties_collector.h"
 #include "options/cf_options.h"
 #include "rocksdb/comparator.h"
@@ -35,7 +36,6 @@ 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
@@ -66,8 +66,9 @@ extern Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& options,
     const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
     TableCache* table_cache, InternalIterator* iter,
-    std::unique_ptr<InternalIterator> range_del_iter, FileMetaData* meta,
-    const InternalKeyComparator& internal_comparator,
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+        range_del_iters,
+    FileMetaData* meta, const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
     uint32_t column_family_id, const std::string& column_family_name,

db/c.cc

@@ -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" {
 
@@ -140,14 +151,20 @@ struct rocksdb_writablefile_t    { WritableFile*     rep; };
 struct rocksdb_wal_iterator_t { TransactionLogIterator* rep; };
 struct rocksdb_wal_readoptions_t { TransactionLogIterator::ReadOptions rep; };
 struct rocksdb_filelock_t        { FileLock*         rep; };
-struct rocksdb_logger_t          { shared_ptr<Logger>  rep; };
-struct rocksdb_cache_t           { shared_ptr<Cache>   rep; };
+struct rocksdb_logger_t {
+  std::shared_ptr<Logger> rep;
+};
+struct rocksdb_cache_t {
+  std::shared_ptr<Cache> rep;
+};
 struct rocksdb_livefiles_t       { std::vector<LiveFileMetaData> rep; };
 struct rocksdb_column_family_handle_t  { ColumnFamilyHandle* rep; };
 struct rocksdb_envoptions_t      { EnvOptions        rep; };
 struct rocksdb_ingestexternalfileoptions_t  { IngestExternalFileOptions rep; };
 struct rocksdb_sstfilewriter_t   { SstFileWriter*    rep; };
-struct rocksdb_ratelimiter_t     { shared_ptr<RateLimiter>      rep; };
+struct rocksdb_ratelimiter_t {
+  std::shared_ptr<RateLimiter> rep;
+};
 struct rocksdb_perfcontext_t     { PerfContext*      rep; };
 struct rocksdb_pinnableslice_t {
   PinnableSlice rep;
@@ -531,10 +548,18 @@ 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 +579,12 @@ 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 +2000,11 @@ 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 +2283,18 @@ 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 +2417,7 @@ 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 +2448,20 @@ 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 +3156,11 @@ 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 +3310,11 @@ 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;
@@ -3478,6 +3550,18 @@ const char* rocksdb_livefiles_largestkey(
   return lf->rep[index].largestkey.data();
 }
 
+uint64_t rocksdb_livefiles_entries(
+    const rocksdb_livefiles_t* lf,
+    int index) {
+  return lf->rep[index].num_entries;
+}
+
+uint64_t rocksdb_livefiles_deletions(
+    const rocksdb_livefiles_t* lf,
+    int index) {
+  return lf->rep[index].num_deletions;
+}
+
 extern void rocksdb_livefiles_destroy(
   const rocksdb_livefiles_t* lf) {
   delete lf;
@@ -4044,6 +4128,98 @@ 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

db/c_test.c

@@ -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);

db/column_family.cc

@@ -20,10 +20,12 @@
 #include <limits>
 
 #include "db/compaction_picker.h"
+#include "db/compaction_picker_fifo.h"
 #include "db/compaction_picker_universal.h"
 #include "db/db_impl.h"
 #include "db/internal_stats.h"
 #include "db/job_context.h"
+#include "db/range_del_aggregator.h"
 #include "db/table_properties_collector.h"
 #include "db/version_set.h"
 #include "db/write_controller.h"
@@ -105,9 +107,6 @@ void GetIntTblPropCollectorFactory(
     int_tbl_prop_collector_factories->emplace_back(
         new UserKeyTablePropertiesCollectorFactory(collector_factories[i]));
   }
-  // Add collector to collect internal key statistics
-  int_tbl_prop_collector_factories->emplace_back(
-      new InternalKeyPropertiesCollectorFactory);
 }
 
 Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
@@ -945,26 +944,16 @@ Status ColumnFamilyData::RangesOverlapWithMemtables(
   super_version->imm->AddIterators(read_opts, &merge_iter_builder);
   ScopedArenaIterator memtable_iter(merge_iter_builder.Finish());
 
-  std::vector<InternalIterator*> memtable_range_del_iters;
+  auto read_seq = super_version->current->version_set()->LastSequence();
+  ReadRangeDelAggregator range_del_agg(&internal_comparator_, read_seq);
   auto* active_range_del_iter =
-      super_version->mem->NewRangeTombstoneIterator(read_opts);
-  if (active_range_del_iter != nullptr) {
-    memtable_range_del_iters.push_back(active_range_del_iter);
-  }
-  super_version->imm->AddRangeTombstoneIterators(read_opts,
-                                                 &memtable_range_del_iters);
-  RangeDelAggregator range_del_agg(internal_comparator_, {} /* snapshots */,
-                                   false /* collapse_deletions */);
+      super_version->mem->NewRangeTombstoneIterator(read_opts, read_seq);
+  range_del_agg.AddTombstones(
+      std::unique_ptr<FragmentedRangeTombstoneIterator>(active_range_del_iter));
+  super_version->imm->AddRangeTombstoneIterators(read_opts, nullptr /* arena */,
+                                                 &range_del_agg);
+
   Status status;
-  {
-    std::unique_ptr<InternalIterator> memtable_range_del_iter(
-        NewMergingIterator(&internal_comparator_,
-                           memtable_range_del_iters.empty()
-                               ? nullptr
-                               : &memtable_range_del_iters[0],
-                           static_cast<int>(memtable_range_del_iters.size())));
-    status = range_del_agg.AddTombstones(std::move(memtable_range_del_iter));
-  }
   for (size_t i = 0; i < ranges.size() && status.ok() && !*overlap; ++i) {
     auto* vstorage = super_version->current->storage_info();
     auto* ucmp = vstorage->InternalComparator()->user_comparator();

db/compact_files_test.cc

@@ -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) {

db/compacted_db_impl.cc

@@ -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*,
@@ -157,6 +147,7 @@ Status CompactedDBImpl::Open(const Options& options,
   std::unique_ptr<CompactedDBImpl> db(new CompactedDBImpl(db_options, dbname));
   Status s = db->Init(options);
   if (s.ok()) {
+    db->StartTimedTasks();
     ROCKS_LOG_INFO(db->immutable_db_options_.info_log,
                    "Opened the db as fully compacted mode");
     LogFlush(db->immutable_db_options_.info_log);

db/compacted_db_impl.h

@@ -67,10 +67,11 @@ class CompactedDBImpl : public DBImpl {
   virtual Status EnableFileDeletions(bool /*force*/) override {
     return Status::NotSupported("Not supported in compacted db mode.");
   }
-  virtual Status GetLiveFiles(std::vector<std::string>&,
-                              uint64_t* /*manifest_file_size*/,
-                              bool /*flush_memtable*/ = true) override {
-    return Status::NotSupported("Not supported in compacted db mode.");
+  virtual Status GetLiveFiles(std::vector<std::string>& ret,
+                              uint64_t* manifest_file_size,
+                              bool /*flush_memtable*/) override {
+    return DBImpl::GetLiveFiles(ret, manifest_file_size,
+                                false /* flush_memtable */);
   }
   using DBImpl::Flush;
   virtual Status Flush(const FlushOptions& /*options*/,

db/compaction.cc

@@ -23,6 +23,43 @@
 
 namespace rocksdb {
 
+const uint64_t kRangeTombstoneSentinel =
+    PackSequenceAndType(kMaxSequenceNumber, kTypeRangeDeletion);
+
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey& a,
+                      const InternalKey& b) {
+  auto c = user_cmp->Compare(a.user_key(), b.user_key());
+  if (c != 0) {
+    return c;
+  }
+  auto a_footer = ExtractInternalKeyFooter(a.Encode());
+  auto b_footer = ExtractInternalKeyFooter(b.Encode());
+  if (a_footer == kRangeTombstoneSentinel) {
+    if (b_footer != kRangeTombstoneSentinel) {
+      return -1;
+    }
+  } else if (b_footer == kRangeTombstoneSentinel) {
+    return 1;
+  }
+  return 0;
+}
+
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey* a,
+                      const InternalKey& b) {
+  if (a == nullptr) {
+    return -1;
+  }
+  return sstableKeyCompare(user_cmp, *a, b);
+}
+
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey& a,
+                      const InternalKey* b) {
+  if (b == nullptr) {
+    return -1;
+  }
+  return sstableKeyCompare(user_cmp, a, *b);
+}
+
 uint64_t TotalFileSize(const std::vector<FileMetaData*>& files) {
   uint64_t sum = 0;
   for (size_t i = 0; i < files.size() && files[i]; i++) {
@@ -81,6 +118,49 @@ void Compaction::GetBoundaryKeys(
   }
 }
 
+std::vector<CompactionInputFiles> Compaction::PopulateWithAtomicBoundaries(
+    VersionStorageInfo* vstorage, std::vector<CompactionInputFiles> inputs) {
+  const Comparator* ucmp = vstorage->InternalComparator()->user_comparator();
+  for (size_t i = 0; i < inputs.size(); i++) {
+    if (inputs[i].level == 0 || inputs[i].files.empty()) {
+      continue;
+    }
+    inputs[i].atomic_compaction_unit_boundaries.reserve(inputs[i].files.size());
+    AtomicCompactionUnitBoundary cur_boundary;
+    size_t first_atomic_idx = 0;
+    auto add_unit_boundary = [&](size_t to) {
+      if (first_atomic_idx == to) return;
+      for (size_t k = first_atomic_idx; k < to; k++) {
+        inputs[i].atomic_compaction_unit_boundaries.push_back(cur_boundary);
+      }
+      first_atomic_idx = to;
+    };
+    for (size_t j = 0; j < inputs[i].files.size(); j++) {
+      const auto* f = inputs[i].files[j];
+      if (j == 0) {
+        // First file in a level.
+        cur_boundary.smallest = &f->smallest;
+        cur_boundary.largest = &f->largest;
+      } else if (sstableKeyCompare(ucmp, *cur_boundary.largest, f->smallest) ==
+                 0) {
+        // SSTs overlap but the end key of the previous file was not
+        // artificially extended by a range tombstone. Extend the current
+        // boundary.
+        cur_boundary.largest = &f->largest;
+      } else {
+        // Atomic compaction unit has ended.
+        add_unit_boundary(j);
+        cur_boundary.smallest = &f->smallest;
+        cur_boundary.largest = &f->largest;
+      }
+    }
+    add_unit_boundary(inputs[i].files.size());
+    assert(inputs[i].files.size() ==
+           inputs[i].atomic_compaction_unit_boundaries.size());
+  }
+  return inputs;
+}
+
 // helper function to determine if compaction is creating files at the
 // bottommost level
 bool Compaction::IsBottommostLevel(
@@ -134,6 +214,7 @@ 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,
@@ -152,8 +233,9 @@ 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)),
+      inputs_(PopulateWithAtomicBoundaries(vstorage, std::move(_inputs))),
       grandparents_(std::move(_grandparents)),
       score_(_score),
       bottommost_level_(IsBottommostLevel(output_level_, vstorage, inputs_)),
@@ -329,12 +411,14 @@ const char* Compaction::InputLevelSummary(
     if (!is_first) {
       len +=
           snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len, " + ");
+      len = std::min(len, static_cast<int>(sizeof(scratch->buffer)));
     } else {
       is_first = false;
     }
     len += snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len,
                     "%" ROCKSDB_PRIszt "@%d", input_level.size(),
                     input_level.level);
+    len = std::min(len, static_cast<int>(sizeof(scratch->buffer)));
   }
   snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len,
            " files to L%d", output_level());

db/compaction.h

@@ -15,11 +15,43 @@
 
 namespace rocksdb {
 
+// Utility for comparing sstable boundary keys. Returns -1 if either a or b is
+// null which provides the property that a==null indicates a key that is less
+// than any key and b==null indicates a key that is greater than any key. Note
+// that the comparison is performed primarily on the user-key portion of the
+// key. If the user-keys compare equal, an additional test is made to sort
+// range tombstone sentinel keys before other keys with the same user-key. The
+// result is that 2 user-keys will compare equal if they differ purely on
+// their sequence number and value, but the range tombstone sentinel for that
+// user-key will compare not equal. This is necessary because the range
+// tombstone sentinel key is set as the largest key for an sstable even though
+// that key never appears in the database. We don't want adjacent sstables to
+// be considered overlapping if they are separated by the range tombstone
+// sentinel.
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey& a,
+                      const InternalKey& b);
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey* a,
+                      const InternalKey& b);
+int sstableKeyCompare(const Comparator* user_cmp, const InternalKey& a,
+                      const InternalKey* b);
+
+// An AtomicCompactionUnitBoundary represents a range of keys [smallest,
+// largest] that exactly spans one ore more neighbouring SSTs on the same
+// level. Every pair of  SSTs in this range "overlap" (i.e., the largest
+// user key of one file is the smallest user key of the next file). These
+// boundaries are propagated down to RangeDelAggregator during compaction
+// to provide safe truncation boundaries for range tombstones.
+struct AtomicCompactionUnitBoundary {
+  const InternalKey* smallest = nullptr;
+  const InternalKey* largest = nullptr;
+};
+
 // The structure that manages compaction input files associated
 // with the same physical level.
 struct CompactionInputFiles {
   int level;
   std::vector<FileMetaData*> files;
+  std::vector<AtomicCompactionUnitBoundary> atomic_compaction_unit_boundaries;
   inline bool empty() const { return files.empty(); }
   inline size_t size() const { return files.size(); }
   inline void clear() { files.clear(); }
@@ -40,7 +72,7 @@ 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,
@@ -96,6 +128,12 @@ class Compaction {
     return inputs_[compaction_input_level][i];
   }
 
+  const std::vector<AtomicCompactionUnitBoundary>* boundaries(
+      size_t compaction_input_level) const {
+    assert(compaction_input_level < inputs_.size());
+    return &inputs_[compaction_input_level].atomic_compaction_unit_boundaries;
+  }
+
   // Returns the list of file meta data of the specified compaction
   // input level.
   // REQUIREMENT: "compaction_input_level" must be >= 0 and
@@ -119,6 +157,11 @@ 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_; }
 
@@ -257,6 +300,13 @@ class Compaction {
                               const std::vector<CompactionInputFiles>& inputs,
                               Slice* smallest_key, Slice* largest_key);
 
+  // Get the atomic file boundaries for all files in the compaction. Necessary
+  // in order to avoid the scenario described in
+  // https://github.com/facebook/rocksdb/pull/4432#discussion_r221072219 and plumb
+  // down appropriate key boundaries to RangeDelAggregator during compaction.
+  static std::vector<CompactionInputFiles> PopulateWithAtomicBoundaries(
+      VersionStorageInfo* vstorage, std::vector<CompactionInputFiles> inputs);
+
   // helper function to determine if compaction with inputs and storage is
   // bottommost
   static bool IsBottommostLevel(
@@ -283,6 +333,7 @@ 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_;
 

db/compaction_iterator.cc

@@ -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,
+    CompactionRangeDelAggregator* 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,
+    CompactionRangeDelAggregator* 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;
 }
 

db/compaction_iterator.h

@@ -63,8 +63,8 @@ class CompactionIterator {
                      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,
+                     CompactionRangeDelAggregator* range_del_agg,
                      const Compaction* compaction = nullptr,
                      const CompactionFilter* compaction_filter = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -76,8 +76,8 @@ class CompactionIterator {
                      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,
+                     CompactionRangeDelAggregator* range_del_agg,
                      std::unique_ptr<CompactionProxy> compaction,
                      const CompactionFilter* compaction_filter = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -139,8 +139,9 @@ 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_;
+  CompactionRangeDelAggregator* range_del_agg_;
   std::unique_ptr<CompactionProxy> compaction_;
   const CompactionFilter* compaction_filter_;
   const std::atomic<bool>* shutting_down_;

db/compaction_iterator_test.cc

@@ -221,10 +221,15 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
       MergeOperator* merge_op = nullptr, CompactionFilter* filter = nullptr,
       bool bottommost_level = false,
       SequenceNumber earliest_write_conflict_snapshot = kMaxSequenceNumber) {
-    std::unique_ptr<InternalIterator> range_del_iter(
+    std::unique_ptr<InternalIterator> unfragmented_range_del_iter(
         new test::VectorIterator(range_del_ks, range_del_vs));
-    range_del_agg_.reset(new RangeDelAggregator(icmp_, snapshots_));
-    ASSERT_OK(range_del_agg_->AddTombstones(std::move(range_del_iter)));
+    auto tombstone_list = std::make_shared<FragmentedRangeTombstoneList>(
+        std::move(unfragmented_range_del_iter), icmp_);
+    std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(
+        new FragmentedRangeTombstoneIterator(tombstone_list, icmp_,
+                                             kMaxSequenceNumber));
+    range_del_agg_.reset(new CompactionRangeDelAggregator(&icmp_, snapshots_));
+    range_del_agg_->AddTombstones(std::move(range_del_iter));
 
     std::unique_ptr<CompactionIterator::CompactionProxy> compaction;
     if (filter || bottommost_level) {
@@ -247,8 +252,8 @@ 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,
@@ -292,7 +297,7 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
   std::unique_ptr<MergeHelper> merge_helper_;
   std::unique_ptr<LoggingForwardVectorIterator> iter_;
   std::unique_ptr<CompactionIterator> c_iter_;
-  std::unique_ptr<RangeDelAggregator> range_del_agg_;
+  std::unique_ptr<CompactionRangeDelAggregator> range_del_agg_;
   std::unique_ptr<SnapshotChecker> snapshot_checker_;
   std::atomic<bool> shutting_down_{false};
   FakeCompaction* compaction_proxy_;
@@ -671,8 +676,12 @@ 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 +850,26 @@ 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);

db/compaction_job.cc

@@ -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"
@@ -34,6 +36,7 @@
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
 #include "db/merge_helper.h"
+#include "db/range_del_aggregator.h"
 #include "db/version_set.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
@@ -198,7 +201,7 @@ struct CompactionJob::SubcompactionState {
     return *this;
   }
 
-  // Because member unique_ptrs do not have these.
+  // Because member std::unique_ptrs do not have these.
   SubcompactionState(const SubcompactionState&) = delete;
 
   SubcompactionState& operator=(const SubcompactionState&) = delete;
@@ -307,7 +310,7 @@ 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 +334,7 @@ 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),
@@ -509,7 +512,10 @@ void CompactionJob::GenSubcompactionBoundaries() {
   // size of data covered by keys in that range
   uint64_t sum = 0;
   std::vector<RangeWithSize> ranges;
-  auto* v = cfd->current();
+  // Get input version from CompactionState since it's already referenced
+  // earlier in SetInputVersioCompaction::SetInputVersion and will not change
+  // when db_mutex_ is released below
+  auto* v = compact_->compaction->input_version();
   for (auto it = bounds.begin();;) {
     const Slice a = *it;
     it++;
@@ -519,7 +525,13 @@ void CompactionJob::GenSubcompactionBoundaries() {
     }
 
     const Slice b = *it;
+
+    // ApproximateSize could potentially create table reader iterator to seek
+    // to the index block and may incur I/O cost in the process. Unlock db
+    // mutex to reduce contention
+    db_mutex_->Unlock();
     uint64_t size = versions_->ApproximateSize(v, a, b, start_lvl, out_lvl + 1);
+    db_mutex_->Lock();
     ranges.emplace_back(a, b, size);
     sum += size;
   }
@@ -591,13 +603,11 @@ 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 +617,73 @@ 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) {
@@ -728,10 +805,13 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
 void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   assert(sub_compact != nullptr);
   ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
-  std::unique_ptr<RangeDelAggregator> range_del_agg(
-      new RangeDelAggregator(cfd->internal_comparator(), existing_snapshots_));
+  CompactionRangeDelAggregator range_del_agg(&cfd->internal_comparator(),
+                                             existing_snapshots_);
+
+  // Although the v2 aggregator is what the level iterator(s) know about,
+  // the AddTombstones calls will be propagated down to the v1 aggregator.
   std::unique_ptr<InternalIterator> input(versions_->MakeInputIterator(
-      sub_compact->compaction, range_del_agg.get(), env_optiosn_for_read_));
+      sub_compact->compaction, &range_del_agg, env_optiosn_for_read_));
 
   AutoThreadOperationStageUpdater stage_updater(
       ThreadStatus::STAGE_COMPACTION_PROCESS_KV);
@@ -760,10 +840,13 @@ 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 +899,9 @@ 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, 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) {
@@ -955,9 +1038,9 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
         next_key = &c_iter->key();
       }
       CompactionIterationStats range_del_out_stats;
-      status = FinishCompactionOutputFile(input_status, sub_compact,
-                                          range_del_agg.get(),
-                                          &range_del_out_stats, next_key);
+      status =
+          FinishCompactionOutputFile(input_status, sub_compact, &range_del_agg,
+                                     &range_del_out_stats, next_key);
       RecordDroppedKeys(range_del_out_stats,
                         &sub_compact->compaction_job_stats);
       if (sub_compact->outputs.size() == 1) {
@@ -966,7 +1049,8 @@ 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);
         }
@@ -1006,8 +1090,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   if (status.ok() && sub_compact->builder == nullptr &&
-      sub_compact->outputs.size() == 0 &&
-      range_del_agg->ShouldAddTombstones(bottommost_level_)) {
+      sub_compact->outputs.size() == 0 && !range_del_agg.IsEmpty()) {
     // handle subcompaction containing only range deletions
     status = OpenCompactionOutputFile(sub_compact);
   }
@@ -1016,8 +1099,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   // close the output file.
   if (sub_compact->builder != nullptr) {
     CompactionIterationStats range_del_out_stats;
-    Status s = FinishCompactionOutputFile(
-        status, sub_compact, range_del_agg.get(), &range_del_out_stats);
+    Status s = FinishCompactionOutputFile(status, sub_compact, &range_del_agg,
+                                          &range_del_out_stats);
     if (status.ok()) {
       status = s;
     }
@@ -1082,7 +1165,7 @@ void CompactionJob::RecordDroppedKeys(
 
 Status CompactionJob::FinishCompactionOutputFile(
     const Status& input_status, SubcompactionState* sub_compact,
-    RangeDelAggregator* range_del_agg,
+    CompactionRangeDelAggregator* range_del_agg,
     CompactionIterationStats* range_del_out_stats,
     const Slice* next_table_min_key /* = nullptr */) {
   AutoThreadOperationStageUpdater stage_updater(
@@ -1095,9 +1178,13 @@ 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;
@@ -1117,18 +1204,118 @@ Status CompactionJob::FinishCompactionOutputFile(
       lower_bound = nullptr;
     }
     if (next_table_min_key != nullptr) {
-      // This isn't the last file in the subcompaction, so extend until the next
-      // file starts.
+      // This may be the last file in the subcompaction in some cases, so we
+      // need to compare the end key of subcompaction with the next file start
+      // key. When the end key is chosen by the subcompaction, we know that
+      // it must be the biggest key in output file. Therefore, it is safe to
+      // use the smaller key as the upper bound of the output file, to ensure
+      // that there is no overlapping between different output files.
       upper_bound_guard = ExtractUserKey(*next_table_min_key);
-      upper_bound = &upper_bound_guard;
+      if (sub_compact->end != nullptr &&
+          ucmp->Compare(upper_bound_guard, *sub_compact->end) >= 0) {
+        upper_bound = sub_compact->end;
+      } else {
+        upper_bound = &upper_bound_guard;
+      }
     } else {
       // This is the last file in the subcompaction, so extend until the
       // 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];
+    }
+    bool has_overlapping_endpoints;
+    if (upper_bound != nullptr && meta->largest.size() > 0) {
+      has_overlapping_endpoints =
+          ucmp->Compare(meta->largest.user_key(), *upper_bound) == 0;
+    } else {
+      has_overlapping_endpoints = false;
+    }
+
+    // The end key of the subcompaction must be bigger or equal to the upper
+    // bound. If the end of subcompaction is null or the upper bound is null,
+    // it means that this file is the last file in the compaction. So there
+    // will be no overlapping between this file and others.
+    assert(sub_compact->end == nullptr ||
+           upper_bound == nullptr ||
+           ucmp->Compare(*upper_bound , *sub_compact->end) <= 0);
+    auto it = range_del_agg->NewIterator(lower_bound, upper_bound,
+                                         has_overlapping_endpoints);
+    // Position the range tombstone output iterator. There may be tombstone
+    // fragments that are entirely out of range, so make sure that we do not
+    // include those.
+    if (lower_bound != nullptr) {
+      it->Seek(*lower_bound);
+    } else {
+      it->SeekToFirst();
+    }
+    for (; it->Valid(); it->Next()) {
+      auto tombstone = it->Tombstone();
+      if (upper_bound != nullptr) {
+        int cmp = ucmp->Compare(*upper_bound, tombstone.start_key_);
+        if ((has_overlapping_endpoints && cmp < 0) ||
+            (!has_overlapping_endpoints && cmp <= 0)) {
+          // Tombstones starting after upper_bound only need to be included in
+          // the next table. If the current SST ends before upper_bound, i.e.,
+          // `has_overlapping_endpoints == false`, we can also skip over range
+          // tombstones that start exactly at upper_bound. Such range tombstones
+          // will be included in the next file and are not relevant to the point
+          // keys or endpoints of the current file.
+          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();
+      assert(lower_bound == nullptr ||
+             ucmp->Compare(*lower_bound, kv.second) < 0);
+      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 +1341,12 @@ 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 +1359,19 @@ 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 +1399,12 @@ 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
@@ -1330,7 +1481,7 @@ Status CompactionJob::OpenCompactionOutputFile(
       TableFileCreationReason::kCompaction);
 #endif  // !ROCKSDB_LITE
   // Make the output file
-  unique_ptr<WritableFile> writable_file;
+  std::unique_ptr<WritableFile> writable_file;
 #ifndef NDEBUG
   bool syncpoint_arg = env_options_.use_direct_writes;
   TEST_SYNC_POINT_CALLBACK("CompactionJob::OpenCompactionOutputFile",
@@ -1362,8 +1513,11 @@ 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()));
+  const auto& listeners =
+      sub_compact->compaction->immutable_cf_options()->listeners;
+  sub_compact->outfile.reset(
+      new WritableFileWriter(std::move(writable_file), fname, env_options_,
+                             db_options_.statistics.get(), listeners));
 
   // 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 +1546,7 @@ 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);

db/compaction_job.h

@@ -29,8 +29,8 @@
 #include "db/version_edit.h"
 #include "db/write_controller.h"
 #include "db/write_thread.h"
-#include "options/db_options.h"
 #include "options/cf_options.h"
+#include "options/db_options.h"
 #include "port/port.h"
 #include "rocksdb/compaction_filter.h"
 #include "rocksdb/compaction_job_stats.h"
@@ -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,
@@ -103,7 +104,7 @@ class CompactionJob {
 
   Status FinishCompactionOutputFile(
       const Status& input_status, SubcompactionState* sub_compact,
-      RangeDelAggregator* range_del_agg,
+      CompactionRangeDelAggregator* range_del_agg,
       CompactionIterationStats* range_del_out_stats,
       const Slice* next_table_min_key = nullptr);
   Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options);
@@ -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

db/compaction_job_stats_test.cc

@@ -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;

db/compaction_job_test.cc

@@ -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());
@@ -198,12 +200,12 @@ class CompactionJobTest : public testing::Test {
     new_db.SetLastSequence(0);
 
     const std::string manifest = DescriptorFileName(dbname_, 1);
-    unique_ptr<WritableFile> file;
+    std::unique_ptr<WritableFile> file;
     Status s = env_->NewWritableFile(
         manifest, &file, env_->OptimizeForManifestWrite(env_options_));
     ASSERT_OK(s);
-    unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), env_options_));
+    std::unique_ptr<WritableFileWriter> file_writer(
+        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) {

db/compaction_picker.cc

@@ -18,6 +18,7 @@
 #include <queue>
 #include <string>
 #include <utility>
+#include <vector>
 #include "db/column_family.h"
 #include "monitoring/statistics.h"
 #include "util/filename.h"
@@ -36,6 +37,7 @@ uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) {
   }
   return sum;
 }
+}  // anonymous namespace
 
 bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
                            size_t min_files_to_compact,
@@ -49,7 +51,7 @@ 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) {
@@ -69,7 +71,6 @@ bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
   }
   return false;
 }
-}  // anonymous namespace
 
 // Determine compression type, based on user options, level of the output
 // file and whether compression is disabled.
@@ -110,6 +111,24 @@ 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 +220,8 @@ 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 +244,8 @@ 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 +318,29 @@ 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 +403,7 @@ 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 +606,13 @@ 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 +652,6 @@ 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 +660,10 @@ 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 +671,10 @@ 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 +722,13 @@ 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 +897,11 @@ 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();
 }
 
@@ -895,8 +944,8 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
   // any currently-existing files.
   for (auto file_num : *input_files) {
     bool found = false;
-    for (auto level_meta : cf_meta.levels) {
-      for (auto file_meta : level_meta.files) {
+    for (const auto& level_meta : cf_meta.levels) {
+      for (const auto& file_meta : level_meta.files) {
         if (file_num == TableFileNameToNumber(file_meta.name)) {
           if (file_meta.being_compacted) {
             return Status::Aborted("Specified compaction input file " +
@@ -947,6 +996,78 @@ 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 +1139,6 @@ 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 +1167,6 @@ 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 +1256,9 @@ 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 +1296,9 @@ 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 +1373,13 @@ 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_);
 
@@ -1491,216 +1548,4 @@ Compaction* LevelCompactionPicker::PickCompaction(
   return builder.PickCompaction();
 }
 
-#ifndef ROCKSDB_LITE
-bool FIFOCompactionPicker::NeedsCompaction(
-    const VersionStorageInfo* vstorage) const {
-  const int kLevel0 = 0;
-  return vstorage->CompactionScore(kLevel0) >= 1;
-}
-
-namespace {
-uint64_t GetTotalFilesSize(
-    const std::vector<FileMetaData*>& files) {
-  uint64_t total_size = 0;
-  for (const auto& f : files) {
-    total_size += f->fd.file_size;
-  }
-  return total_size;
-}
-}  // anonymous namespace
-
-Compaction* FIFOCompactionPicker::PickTTLCompaction(
-    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
-  assert(mutable_cf_options.compaction_options_fifo.ttl > 0);
-
-  const int kLevel0 = 0;
-  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
-  uint64_t total_size = GetTotalFilesSize(level_files);
-
-  int64_t _current_time;
-  auto status = ioptions_.env->GetCurrentTime(&_current_time);
-  if (!status.ok()) {
-    ROCKS_LOG_BUFFER(log_buffer,
-                     "[%s] FIFO compaction: Couldn't get current time: %s. "
-                     "Not doing compactions based on TTL. ",
-                     cf_name.c_str(), status.ToString().c_str());
-    return nullptr;
-  }
-  const uint64_t current_time = static_cast<uint64_t>(_current_time);
-
-  std::vector<CompactionInputFiles> inputs;
-  inputs.emplace_back();
-  inputs[0].level = 0;
-
-  // avoid underflow
-  if (current_time > mutable_cf_options.compaction_options_fifo.ttl) {
-    for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
-      auto f = *ritr;
-      if (f->fd.table_reader != nullptr &&
-          f->fd.table_reader->GetTableProperties() != nullptr) {
-        auto creation_time =
-            f->fd.table_reader->GetTableProperties()->creation_time;
-        if (creation_time == 0 ||
-            creation_time >= (current_time -
-                              mutable_cf_options.compaction_options_fifo.ttl)) {
-          break;
-        }
-        total_size -= f->compensated_file_size;
-        inputs[0].files.push_back(f);
-      }
-    }
-  }
-
-  // Return a nullptr and proceed to size-based FIFO compaction if:
-  // 1. there are no files older than ttl OR
-  // 2. there are a few files older than ttl, but deleting them will not bring
-  //    the total size to be less than max_table_files_size threshold.
-  if (inputs[0].files.empty() ||
-      total_size >
-          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
-    return nullptr;
-  }
-
-  for (const auto& f : inputs[0].files) {
-    ROCKS_LOG_BUFFER(log_buffer,
-                     "[%s] FIFO compaction: picking file %" PRIu64
-                     " with creation time %" PRIu64 " for deletion",
-                     cf_name.c_str(), f->fd.GetNumber(),
-                     f->fd.table_reader->GetTableProperties()->creation_time);
-  }
-
-  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);
-  return c;
-}
-
-Compaction* FIFOCompactionPicker::PickSizeCompaction(
-    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
-  const int kLevel0 = 0;
-  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
-  uint64_t total_size = GetTotalFilesSize(level_files);
-
-  if (total_size <=
-          mutable_cf_options.compaction_options_fifo.max_table_files_size ||
-      level_files.size() == 0) {
-    // total size not exceeded
-    if (mutable_cf_options.compaction_options_fifo.allow_compaction &&
-        level_files.size() > 0) {
-      CompactionInputFiles comp_inputs;
-      // try to prevent same files from being compacted multiple times, which
-      // could produce large files that may never TTL-expire. Achieve this by
-      // disallowing compactions with files larger than memtable (inflate its
-      // size by 10% to account for uncompressed L0 files that may have size
-      // slightly greater than memtable size limit).
-      size_t max_compact_bytes_per_del_file =
-          static_cast<size_t>(MultiplyCheckOverflow(
-              static_cast<uint64_t>(mutable_cf_options.write_buffer_size),
-              1.1));
-      if (FindIntraL0Compaction(
-              level_files,
-              mutable_cf_options
-                  .level0_file_num_compaction_trigger /* min_files_to_compact */
-              ,
-              max_compact_bytes_per_del_file, &comp_inputs)) {
-        Compaction* c = new Compaction(
-            vstorage, ioptions_, mutable_cf_options, {comp_inputs}, 0,
-            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,
-            CompactionReason::kFIFOReduceNumFiles);
-        return c;
-      }
-    }
-
-    ROCKS_LOG_BUFFER(
-        log_buffer,
-        "[%s] FIFO compaction: nothing to do. Total size %" PRIu64
-        ", max size %" PRIu64 "\n",
-        cf_name.c_str(), total_size,
-        mutable_cf_options.compaction_options_fifo.max_table_files_size);
-    return nullptr;
-  }
-
-  if (!level0_compactions_in_progress_.empty()) {
-    ROCKS_LOG_BUFFER(
-        log_buffer,
-        "[%s] FIFO compaction: Already executing compaction. No need "
-        "to run parallel compactions since compactions are very fast",
-        cf_name.c_str());
-    return nullptr;
-  }
-
-  std::vector<CompactionInputFiles> inputs;
-  inputs.emplace_back();
-  inputs[0].level = 0;
-
-  for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
-    auto f = *ritr;
-    total_size -= f->compensated_file_size;
-    inputs[0].files.push_back(f);
-    char tmp_fsize[16];
-    AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
-    ROCKS_LOG_BUFFER(log_buffer,
-                     "[%s] FIFO compaction: picking file %" PRIu64
-                     " with size %s for deletion",
-                     cf_name.c_str(), f->fd.GetNumber(), tmp_fsize);
-    if (total_size <=
-        mutable_cf_options.compaction_options_fifo.max_table_files_size) {
-      break;
-    }
-  }
-
-  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);
-  return c;
-}
-
-Compaction* FIFOCompactionPicker::PickCompaction(
-    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
-  assert(vstorage->num_levels() == 1);
-
-  Compaction* c = nullptr;
-  if (mutable_cf_options.compaction_options_fifo.ttl > 0) {
-    c = PickTTLCompaction(cf_name, mutable_cf_options, vstorage, log_buffer);
-  }
-  if (c == nullptr) {
-    c = PickSizeCompaction(cf_name, mutable_cf_options, vstorage, log_buffer);
-  }
-  RegisterCompaction(c);
-  return c;
-}
-
-Compaction* FIFOCompactionPicker::CompactRange(
-    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-    VersionStorageInfo* vstorage, int input_level, int output_level,
-    uint32_t /*output_path_id*/, uint32_t /*max_subcompactions*/,
-    const InternalKey* /*begin*/, const InternalKey* /*end*/,
-    InternalKey** compaction_end, bool* /*manual_conflict*/) {
-#ifdef NDEBUG
-  (void)input_level;
-  (void)output_level;
-#endif
-  assert(input_level == 0);
-  assert(output_level == 0);
-  *compaction_end = nullptr;
-  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, ioptions_.info_log);
-  Compaction* c =
-      PickCompaction(cf_name, mutable_cf_options, vstorage, &log_buffer);
-  log_buffer.FlushBufferToLog();
-  return c;
-}
-
-#endif  // !ROCKSDB_LITE
-
 }  // namespace rocksdb

db/compaction_picker.h

@@ -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);
 
@@ -225,42 +235,6 @@ class LevelCompactionPicker : public CompactionPicker {
 };
 
 #ifndef ROCKSDB_LITE
-class FIFOCompactionPicker : public CompactionPicker {
- public:
-  FIFOCompactionPicker(const ImmutableCFOptions& ioptions,
-                       const InternalKeyComparator* icmp)
-      : CompactionPicker(ioptions, icmp) {}
-
-  virtual Compaction* PickCompaction(const std::string& cf_name,
-                                     const MutableCFOptions& mutable_cf_options,
-                                     VersionStorageInfo* version,
-                                     LogBuffer* log_buffer) override;
-
-  virtual Compaction* CompactRange(
-      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
-      VersionStorageInfo* vstorage, int input_level, int output_level,
-      uint32_t output_path_id, uint32_t max_subcompactions,
-      const InternalKey* begin, const InternalKey* end,
-      InternalKey** compaction_end, bool* manual_conflict) override;
-
-  // The maximum allowed output level.  Always returns 0.
-  virtual int MaxOutputLevel() const override { return 0; }
-
-  virtual bool NeedsCompaction(
-      const VersionStorageInfo* vstorage) const override;
-
- private:
-  Compaction* PickTTLCompaction(const std::string& cf_name,
-                                const MutableCFOptions& mutable_cf_options,
-                                VersionStorageInfo* version,
-                                LogBuffer* log_buffer);
-
-  Compaction* PickSizeCompaction(const std::string& cf_name,
-                                 const MutableCFOptions& mutable_cf_options,
-                                 VersionStorageInfo* version,
-                                 LogBuffer* log_buffer);
-};
-
 class NullCompactionPicker : public CompactionPicker {
  public:
   NullCompactionPicker(const ImmutableCFOptions& ioptions,
@@ -298,10 +272,20 @@ class NullCompactionPicker : public CompactionPicker {
 };
 #endif  // !ROCKSDB_LITE
 
+bool FindIntraL0Compaction(const std::vector<FileMetaData*>& level_files,
+                           size_t min_files_to_compact,
+                           uint64_t max_compact_bytes_per_del_file,
+                           CompactionInputFiles* comp_inputs);
+
 CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
                                    const VersionStorageInfo* vstorage,
                                    const MutableCFOptions& mutable_cf_options,
                                    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

db/compaction_picker_fifo.cc

@@ -0,0 +1,235 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#include "db/compaction_picker_fifo.h"
+#ifndef ROCKSDB_LITE
+
+#ifndef __STDC_FORMAT_MACROS
+#define __STDC_FORMAT_MACROS
+#endif
+
+#include <inttypes.h>
+#include <string>
+#include <vector>
+#include "db/column_family.h"
+#include "util/log_buffer.h"
+#include "util/string_util.h"
+
+namespace rocksdb {
+namespace {
+uint64_t GetTotalFilesSize(const std::vector<FileMetaData*>& files) {
+  uint64_t total_size = 0;
+  for (const auto& f : files) {
+    total_size += f->fd.file_size;
+  }
+  return total_size;
+}
+}  // anonymous namespace
+
+bool FIFOCompactionPicker::NeedsCompaction(
+    const VersionStorageInfo* vstorage) const {
+  const int kLevel0 = 0;
+  return vstorage->CompactionScore(kLevel0) >= 1;
+}
+
+Compaction* FIFOCompactionPicker::PickTTLCompaction(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
+  assert(mutable_cf_options.compaction_options_fifo.ttl > 0);
+
+  const int kLevel0 = 0;
+  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
+  uint64_t total_size = GetTotalFilesSize(level_files);
+
+  int64_t _current_time;
+  auto status = ioptions_.env->GetCurrentTime(&_current_time);
+  if (!status.ok()) {
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "[%s] FIFO compaction: Couldn't get current time: %s. "
+                     "Not doing compactions based on TTL. ",
+                     cf_name.c_str(), status.ToString().c_str());
+    return nullptr;
+  }
+  const uint64_t current_time = static_cast<uint64_t>(_current_time);
+
+  std::vector<CompactionInputFiles> inputs;
+  inputs.emplace_back();
+  inputs[0].level = 0;
+
+  // avoid underflow
+  if (current_time > mutable_cf_options.compaction_options_fifo.ttl) {
+    for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
+      auto f = *ritr;
+      if (f->fd.table_reader != nullptr &&
+          f->fd.table_reader->GetTableProperties() != nullptr) {
+        auto creation_time =
+            f->fd.table_reader->GetTableProperties()->creation_time;
+        if (creation_time == 0 ||
+            creation_time >= (current_time -
+                              mutable_cf_options.compaction_options_fifo.ttl)) {
+          break;
+        }
+        total_size -= f->compensated_file_size;
+        inputs[0].files.push_back(f);
+      }
+    }
+  }
+
+  // Return a nullptr and proceed to size-based FIFO compaction if:
+  // 1. there are no files older than ttl OR
+  // 2. there are a few files older than ttl, but deleting them will not bring
+  //    the total size to be less than max_table_files_size threshold.
+  if (inputs[0].files.empty() ||
+      total_size >
+          mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+    return nullptr;
+  }
+
+  for (const auto& f : inputs[0].files) {
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "[%s] FIFO compaction: picking file %" PRIu64
+                     " with creation time %" PRIu64 " for deletion",
+                     cf_name.c_str(), f->fd.GetNumber(),
+                     f->fd.table_reader->GetTableProperties()->creation_time);
+  }
+
+  Compaction* c = new Compaction(
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
+      /* is deletion compaction */ true, CompactionReason::kFIFOTtl);
+  return c;
+}
+
+Compaction* FIFOCompactionPicker::PickSizeCompaction(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
+  const int kLevel0 = 0;
+  const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0);
+  uint64_t total_size = GetTotalFilesSize(level_files);
+
+  if (total_size <=
+          mutable_cf_options.compaction_options_fifo.max_table_files_size ||
+      level_files.size() == 0) {
+    // total size not exceeded
+    if (mutable_cf_options.compaction_options_fifo.allow_compaction &&
+        level_files.size() > 0) {
+      CompactionInputFiles comp_inputs;
+      // try to prevent same files from being compacted multiple times, which
+      // could produce large files that may never TTL-expire. Achieve this by
+      // disallowing compactions with files larger than memtable (inflate its
+      // size by 10% to account for uncompressed L0 files that may have size
+      // slightly greater than memtable size limit).
+      size_t max_compact_bytes_per_del_file =
+          static_cast<size_t>(MultiplyCheckOverflow(
+              static_cast<uint64_t>(mutable_cf_options.write_buffer_size),
+              1.1));
+      if (FindIntraL0Compaction(
+              level_files,
+              mutable_cf_options
+                  .level0_file_num_compaction_trigger /* min_files_to_compact */
+              ,
+              max_compact_bytes_per_del_file, &comp_inputs)) {
+        Compaction* c = new Compaction(
+            vstorage, ioptions_, mutable_cf_options, {comp_inputs}, 0,
+            16 * 1024 * 1024 /* output file size limit */,
+            0 /* max compaction bytes, not applicable */,
+            0 /* output path ID */, mutable_cf_options.compression,
+            ioptions_.compression_opts, 0 /* max_subcompactions */, {},
+            /* is manual */ false, vstorage->CompactionScore(0),
+            /* is deletion compaction */ false,
+            CompactionReason::kFIFOReduceNumFiles);
+        return c;
+      }
+    }
+
+    ROCKS_LOG_BUFFER(
+        log_buffer,
+        "[%s] FIFO compaction: nothing to do. Total size %" PRIu64
+        ", max size %" PRIu64 "\n",
+        cf_name.c_str(), total_size,
+        mutable_cf_options.compaction_options_fifo.max_table_files_size);
+    return nullptr;
+  }
+
+  if (!level0_compactions_in_progress_.empty()) {
+    ROCKS_LOG_BUFFER(
+        log_buffer,
+        "[%s] FIFO compaction: Already executing compaction. No need "
+        "to run parallel compactions since compactions are very fast",
+        cf_name.c_str());
+    return nullptr;
+  }
+
+  std::vector<CompactionInputFiles> inputs;
+  inputs.emplace_back();
+  inputs[0].level = 0;
+
+  for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) {
+    auto f = *ritr;
+    total_size -= f->compensated_file_size;
+    inputs[0].files.push_back(f);
+    char tmp_fsize[16];
+    AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
+    ROCKS_LOG_BUFFER(log_buffer,
+                     "[%s] FIFO compaction: picking file %" PRIu64
+                     " with size %s for deletion",
+                     cf_name.c_str(), f->fd.GetNumber(), tmp_fsize);
+    if (total_size <=
+        mutable_cf_options.compaction_options_fifo.max_table_files_size) {
+      break;
+    }
+  }
+
+  Compaction* c = new Compaction(
+      vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0,
+      kNoCompression, ioptions_.compression_opts, /* max_subcompactions */ 0,
+      {}, /* is manual */ false, vstorage->CompactionScore(0),
+      /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize);
+  return c;
+}
+
+Compaction* FIFOCompactionPicker::PickCompaction(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, LogBuffer* log_buffer) {
+  assert(vstorage->num_levels() == 1);
+
+  Compaction* c = nullptr;
+  if (mutable_cf_options.compaction_options_fifo.ttl > 0) {
+    c = PickTTLCompaction(cf_name, mutable_cf_options, vstorage, log_buffer);
+  }
+  if (c == nullptr) {
+    c = PickSizeCompaction(cf_name, mutable_cf_options, vstorage, log_buffer);
+  }
+  RegisterCompaction(c);
+  return c;
+}
+
+Compaction* FIFOCompactionPicker::CompactRange(
+    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+    VersionStorageInfo* vstorage, int input_level, int output_level,
+    uint32_t /*output_path_id*/, uint32_t /*max_subcompactions*/,
+    const InternalKey* /*begin*/, const InternalKey* /*end*/,
+    InternalKey** compaction_end, bool* /*manual_conflict*/) {
+#ifdef NDEBUG
+  (void)input_level;
+  (void)output_level;
+#endif
+  assert(input_level == 0);
+  assert(output_level == 0);
+  *compaction_end = nullptr;
+  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, ioptions_.info_log);
+  Compaction* c =
+      PickCompaction(cf_name, mutable_cf_options, vstorage, &log_buffer);
+  log_buffer.FlushBufferToLog();
+  return c;
+}
+
+}  // namespace rocksdb
+#endif  // !ROCKSDB_LITE

db/compaction_picker_fifo.h

@@ -0,0 +1,52 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#pragma once
+#ifndef ROCKSDB_LITE
+
+#include "db/compaction_picker.h"
+
+namespace rocksdb {
+class FIFOCompactionPicker : public CompactionPicker {
+ public:
+  FIFOCompactionPicker(const ImmutableCFOptions& ioptions,
+                       const InternalKeyComparator* icmp)
+      : CompactionPicker(ioptions, icmp) {}
+
+  virtual Compaction* PickCompaction(const std::string& cf_name,
+                                     const MutableCFOptions& mutable_cf_options,
+                                     VersionStorageInfo* version,
+                                     LogBuffer* log_buffer) override;
+
+  virtual Compaction* CompactRange(
+      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+      VersionStorageInfo* vstorage, int input_level, int output_level,
+      uint32_t output_path_id, uint32_t max_subcompactions,
+      const InternalKey* begin, const InternalKey* end,
+      InternalKey** compaction_end, bool* manual_conflict) override;
+
+  // The maximum allowed output level.  Always returns 0.
+  virtual int MaxOutputLevel() const override { return 0; }
+
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;
+
+ private:
+  Compaction* PickTTLCompaction(const std::string& cf_name,
+                                const MutableCFOptions& mutable_cf_options,
+                                VersionStorageInfo* version,
+                                LogBuffer* log_buffer);
+
+  Compaction* PickSizeCompaction(const std::string& cf_name,
+                                 const MutableCFOptions& mutable_cf_options,
+                                 VersionStorageInfo* version,
+                                 LogBuffer* log_buffer);
+};
+}  // namespace rocksdb
+#endif  // !ROCKSDB_LITE

db/compaction_picker_test.cc

@@ -4,10 +4,12 @@
 //  (found in the LICENSE.Apache file in the root directory).
 
 #include "db/compaction_picker.h"
+
 #include <limits>
 #include <string>
 #include <utility>
 #include "db/compaction.h"
+#include "db/compaction_picker_fifo.h"
 #include "db/compaction_picker_universal.h"
 
 #include "util/logging.h"
@@ -90,8 +92,8 @@ 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 +393,10 @@ 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 +417,7 @@ 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 +451,7 @@ 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);

db/compaction_picker_universal.cc

@@ -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

db/compaction_picker_universal.h

@@ -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

db/comparator_db_test.cc

@@ -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);

db/convenience.cc

@@ -35,7 +35,7 @@ Status DeleteFilesInRanges(DB* db, ColumnFamilyHandle* column_family,
 Status VerifySstFileChecksum(const Options& options,
                              const EnvOptions& env_options,
                              const std::string& file_path) {
-  unique_ptr<RandomAccessFile> file;
+  std::unique_ptr<RandomAccessFile> file;
   uint64_t file_size;
   InternalKeyComparator internal_comparator(options.comparator);
   ImmutableCFOptions ioptions(options);
@@ -46,13 +46,14 @@ Status VerifySstFileChecksum(const Options& options,
   } else {
     return s;
   }
-  unique_ptr<TableReader> table_reader;
+  std::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()) {

db/corruption_test.cc

@@ -37,7 +37,7 @@ class CorruptionTest : public testing::Test {
  public:
   test::ErrorEnv env_;
   std::string dbname_;
-  shared_ptr<Cache> tiny_cache_;
+  std::shared_ptr<Cache> tiny_cache_;
   Options options_;
   DB* db_;
 
@@ -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;
@@ -485,7 +485,7 @@ TEST_F(CorruptionTest, FileSystemStateCorrupted) {
     db_ = nullptr;
 
     if (iter == 0) {  // corrupt file size
-      unique_ptr<WritableFile> file;
+      std::unique_ptr<WritableFile> file;
       env_.NewWritableFile(filename, &file, EnvOptions());
       file->Append(Slice("corrupted sst"));
       file.reset();

db/cuckoo_table_db_test.cc

@@ -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();

db/db_basic_test.cc

@@ -794,7 +794,7 @@ TEST_F(DBBasicTest, ChecksumTest) {
   BlockBasedTableOptions table_options;
   Options options = CurrentOptions();
   // change when new checksum type added
-  int max_checksum = static_cast<int>(kxxHash);
+  int max_checksum = static_cast<int>(kxxHash64);
   const int kNumPerFile = 2;
 
   // generate one table with each type of checksum
@@ -809,7 +809,7 @@ TEST_F(DBBasicTest, ChecksumTest) {
   }
 
   // verify data with each type of checksum
-  for (int i = 0; i <= kxxHash; ++i) {
+  for (int i = 0; i <= kxxHash64; ++i) {
     table_options.checksum = static_cast<ChecksumType>(i);
     options.table_factory.reset(NewBlockBasedTableFactory(table_options));
     Reopen(options);
@@ -881,7 +881,7 @@ class TestEnv : public EnvWrapper {
     int GetCloseCount() { return close_count; }
 
     virtual Status NewLogger(const std::string& /*fname*/,
-                             shared_ptr<Logger>* result) {
+                             std::shared_ptr<Logger>* result) {
       result->reset(new TestLogger(this));
       return Status::OK();
     }
@@ -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;

db/db_block_cache_test.cc

@@ -349,7 +349,7 @@ TEST_F(DBBlockCacheTest, IndexAndFilterBlocksStats) {
   // 200 bytes are enough to hold the first two blocks
   std::shared_ptr<Cache> cache = NewLRUCache(200, 0, false);
   table_options.block_cache = cache;
-  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20, true));
   options.table_factory.reset(new BlockBasedTableFactory(table_options));
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -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),

db/db_bloom_filter_test.cc

@@ -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;
@@ -117,11 +141,79 @@ 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();
+    get_perf_context()->EnablePerLevelPerfContext();
+    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(
+        0,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+    ASSERT_EQ("foo2", Get("barbarbar2"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ(
+        0,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+    ASSERT_EQ("NOT_FOUND", Get("barbarbar3"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 0);
+    ASSERT_EQ(
+        0,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+
+    ASSERT_EQ("NOT_FOUND", Get("barfoofoo"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 1);
+    ASSERT_EQ(
+        1,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+
+    ASSERT_EQ("NOT_FOUND", Get("foobarbar"));
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+    ASSERT_EQ(
+        2,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+
+    ro.total_order_seek = true;
+    ASSERT_TRUE(db_->Get(ro, "foobarbar", &value).IsNotFound());
+    ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+    ASSERT_EQ(
+        2,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+    get_perf_context()->Reset();
+  }
+}
+
 TEST_F(DBBloomFilterTest, GetFilterByPrefixBloom) {
   for (bool partition_filters : {true, false}) {
     Options options = last_options_;
     options.prefix_extractor.reset(NewFixedPrefixTransform(8));
     options.statistics = rocksdb::CreateDBStatistics();
+    get_perf_context()->EnablePerLevelPerfContext();
     BlockBasedTableOptions bbto;
     bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
     if (partition_filters) {
@@ -160,6 +252,10 @@ TEST_F(DBBloomFilterTest, GetFilterByPrefixBloom) {
     ro.total_order_seek = true;
     ASSERT_TRUE(db_->Get(ro, "foobarbar", &value).IsNotFound());
     ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 2);
+    ASSERT_EQ(
+        2,
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful);
+    get_perf_context()->Reset();
   }
 }
 
@@ -168,6 +264,7 @@ TEST_F(DBBloomFilterTest, WholeKeyFilterProp) {
     Options options = last_options_;
     options.prefix_extractor.reset(NewFixedPrefixTransform(3));
     options.statistics = rocksdb::CreateDBStatistics();
+    get_perf_context()->EnablePerLevelPerfContext();
 
     BlockBasedTableOptions bbto;
     bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
@@ -315,6 +412,14 @@ TEST_F(DBBloomFilterTest, WholeKeyFilterProp) {
     ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 12);
     ASSERT_EQ("bar", Get("barfoo"));
     ASSERT_EQ(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 12);
+    uint64_t bloom_filter_useful_all_levels = 0;
+    for (auto& kv : (*(get_perf_context()->level_to_perf_context))) {
+      if (kv.second.bloom_filter_useful > 0) {
+        bloom_filter_useful_all_levels += kv.second.bloom_filter_useful;
+      }
+    }
+    ASSERT_EQ(12, bloom_filter_useful_all_levels);
+    get_perf_context()->Reset();
   }
 }
 
@@ -334,6 +439,11 @@ 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));
 
@@ -389,15 +499,34 @@ 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)));
+#ifndef ROCKSDB_VALGRIND_RUN
+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)));
+#endif  // ROCKSDB_VALGRIND_RUN
 
 TEST_F(DBBloomFilterTest, BloomFilterRate) {
   while (ChangeFilterOptions()) {
     Options options = CurrentOptions();
     options.statistics = rocksdb::CreateDBStatistics();
+    get_perf_context()->EnablePerLevelPerfContext();
     CreateAndReopenWithCF({"pikachu"}, options);
 
     const int maxKey = 10000;
@@ -419,6 +548,10 @@ TEST_F(DBBloomFilterTest, BloomFilterRate) {
       ASSERT_EQ("NOT_FOUND", Get(1, Key(i + 33333)));
     }
     ASSERT_GE(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), maxKey * 0.98);
+    ASSERT_GE(
+        (*(get_perf_context()->level_to_perf_context))[0].bloom_filter_useful,
+        maxKey * 0.98);
+    get_perf_context()->Reset();
   }
 }
 
@@ -764,7 +897,7 @@ TEST_P(BloomStatsTestWithParam, BloomStatsTestWithIter) {
   ASSERT_OK(Put(key1, value1, WriteOptions()));
   ASSERT_OK(Put(key3, value3, WriteOptions()));
 
-  unique_ptr<Iterator> iter(dbfull()->NewIterator(ReadOptions()));
+  std::unique_ptr<Iterator> iter(dbfull()->NewIterator(ReadOptions()));
 
   // check memtable bloom stats
   iter->Seek(key1);
@@ -940,6 +1073,7 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
   options.table_factory.reset(NewBlockBasedTableFactory(bbto));
   options.optimize_filters_for_hits = true;
   options.statistics = rocksdb::CreateDBStatistics();
+  get_perf_context()->EnablePerLevelPerfContext();
   CreateAndReopenWithCF({"mypikachu"}, options);
 
   int numkeys = 200000;
@@ -986,6 +1120,14 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
   // no bloom filter. Most keys be checked bloom filters twice.
   ASSERT_GT(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 65000 * 2);
   ASSERT_LT(TestGetTickerCount(options, BLOOM_FILTER_USEFUL), 120000 * 2);
+  uint64_t bloom_filter_useful_all_levels = 0;
+  for (auto& kv : (*(get_perf_context()->level_to_perf_context))) {
+    if (kv.second.bloom_filter_useful > 0) {
+      bloom_filter_useful_all_levels += kv.second.bloom_filter_useful;
+    }
+  }
+  ASSERT_GT(bloom_filter_useful_all_levels, 65000 * 2);
+  ASSERT_LT(bloom_filter_useful_all_levels, 120000 * 2);
 
   for (int i = 0; i < numkeys; i += 2) {
     ASSERT_EQ(Get(1, Key(i)), "val");
@@ -1095,6 +1237,414 @@ TEST_F(DBBloomFilterTest, OptimizeFiltersForHits) {
   ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
   ASSERT_EQ(2 /* index and data block */,
             TestGetTickerCount(options, BLOCK_CACHE_ADD));
+  get_perf_context()->Reset();
+}
+
+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

db/db_compaction_filter_test.cc

@@ -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:
@@ -333,9 +340,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
   Arena arena;
   {
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
-    ScopedArenaIterator iter(
-        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[1]));
+    ReadRangeDelAggregator range_del_agg(&icmp,
+                                         kMaxSequenceNumber /* upper_bound */);
+    ScopedArenaIterator iter(dbfull()->NewInternalIterator(
+        &arena, &range_del_agg, kMaxSequenceNumber, handles_[1]));
     iter->SeekToFirst();
     ASSERT_OK(iter->status());
     while (iter->Valid()) {
@@ -422,9 +430,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
   count = 0;
   {
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
-    ScopedArenaIterator iter(
-        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[1]));
+    ReadRangeDelAggregator range_del_agg(&icmp,
+                                         kMaxSequenceNumber /* upper_bound */);
+    ScopedArenaIterator iter(dbfull()->NewInternalIterator(
+        &arena, &range_del_agg, kMaxSequenceNumber, handles_[1]));
     iter->SeekToFirst();
     ASSERT_OK(iter->status());
     while (iter->Valid()) {
@@ -639,9 +648,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextManual) {
     int total = 0;
     Arena arena;
     InternalKeyComparator icmp(options.comparator);
-    RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
-    ScopedArenaIterator iter(
-        dbfull()->NewInternalIterator(&arena, &range_del_agg));
+    ReadRangeDelAggregator range_del_agg(&icmp,
+                                         kMaxSequenceNumber /* snapshots */);
+    ScopedArenaIterator iter(dbfull()->NewInternalIterator(
+        &arena, &range_del_agg, kMaxSequenceNumber));
     iter->SeekToFirst();
     ASSERT_OK(iter->status());
     while (iter->Valid()) {
@@ -841,7 +851,7 @@ TEST_F(DBTestCompactionFilter, SkipUntilWithBloomFilter) {
   DestroyAndReopen(options);
 
   Put("0000000010", "v10");
-  Put("0000000020", "v20"); // skipped
+  Put("0000000020", "v20");  // skipped
   Put("0000000050", "v50");
   Flush();
 

db/db_compaction_test.cc

@@ -8,11 +8,13 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "db/db_test_util.h"
-#include "port/stack_trace.h"
 #include "port/port.h"
+#include "port/stack_trace.h"
 #include "rocksdb/experimental.h"
 #include "rocksdb/utilities/convenience.h"
+#include "util/fault_injection_test_env.h"
 #include "util/sync_point.h"
+
 namespace rocksdb {
 
 // SYNC_POINT is not supported in released Windows mode.
@@ -116,6 +118,21 @@ 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 +268,7 @@ 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 +311,7 @@ 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 +2479,7 @@ 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 +2495,7 @@ 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 +2512,7 @@ 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));
@@ -2933,6 +2955,12 @@ TEST_F(DBCompactionTest, SuggestCompactRangeNoTwoLevel0Compactions) {
   dbfull()->TEST_WaitForCompact();
 }
 
+static std::string ShortKey(int i) {
+  assert(i < 10000);
+  char buf[100];
+  snprintf(buf, sizeof(buf), "key%04d", i);
+  return std::string(buf);
+}
 
 TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   int32_t trivial_move = 0;
@@ -2945,10 +2973,28 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
       [&](void* /*arg*/) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
+  // The key size is guaranteed to be <= 8
+  class ShortKeyComparator : public Comparator {
+    int Compare(const rocksdb::Slice& a,
+                const rocksdb::Slice& b) const override {
+      assert(a.size() <= 8);
+      assert(b.size() <= 8);
+      return BytewiseComparator()->Compare(a, b);
+    }
+    const char* Name() const override { return "ShortKeyComparator"; }
+    void FindShortestSeparator(std::string* start,
+                               const rocksdb::Slice& limit) const override {
+      return BytewiseComparator()->FindShortestSeparator(start, limit);
+    }
+    void FindShortSuccessor(std::string* key) const override {
+      return BytewiseComparator()->FindShortSuccessor(key);
+    }
+  } short_key_cmp;
   Options options = CurrentOptions();
   options.target_file_size_base = 100000000;
   options.write_buffer_size = 100000000;
   options.max_subcompactions = max_subcompactions_;
+  options.comparator = &short_key_cmp;
   DestroyAndReopen(options);
 
   int32_t value_size = 10 * 1024;  // 10 KB
@@ -2958,7 +3004,7 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   // File with keys [ 0 => 99 ]
   for (int i = 0; i < 100; i++) {
     values.push_back(RandomString(&rnd, value_size));
-    ASSERT_OK(Put(Key(i), values[i]));
+    ASSERT_OK(Put(ShortKey(i), values[i]));
   }
   ASSERT_OK(Flush());
 
@@ -2975,7 +3021,7 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   // File with keys [ 100 => 199 ]
   for (int i = 100; i < 200; i++) {
     values.push_back(RandomString(&rnd, value_size));
-    ASSERT_OK(Put(Key(i), values[i]));
+    ASSERT_OK(Put(ShortKey(i), values[i]));
   }
   ASSERT_OK(Flush());
 
@@ -2993,7 +3039,7 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   // File with keys [ 200 => 299 ]
   for (int i = 200; i < 300; i++) {
     values.push_back(RandomString(&rnd, value_size));
-    ASSERT_OK(Put(Key(i), values[i]));
+    ASSERT_OK(Put(ShortKey(i), values[i]));
   }
   ASSERT_OK(Flush());
 
@@ -3011,7 +3057,7 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
   ASSERT_EQ(non_trivial_move, 0);
 
   for (int i = 0; i < 300; i++) {
-    ASSERT_EQ(Get(Key(i)), values[i]);
+    ASSERT_EQ(Get(ShortKey(i)), values[i]);
   }
 
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
@@ -3374,14 +3420,13 @@ 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 +3439,7 @@ 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 +3453,57 @@ 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 +3529,13 @@ 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 +3583,13 @@ 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 +3599,7 @@ 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 +3635,7 @@ 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 +3686,21 @@ 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 +3710,7 @@ 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 +3804,92 @@ 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 +3987,167 @@ 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);
+}
+
+TEST_F(DBCompactionTest, ManualCompactionFailsInReadOnlyMode) {
+  // Regression test for bug where manual compaction hangs forever when the DB
+  // is in read-only mode. Verify it now at least returns, despite failing.
+  const int kNumL0Files = 4;
+  std::unique_ptr<FaultInjectionTestEnv> mock_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  Options opts = CurrentOptions();
+  opts.disable_auto_compactions = true;
+  opts.env = mock_env.get();
+  DestroyAndReopen(opts);
+
+  Random rnd(301);
+  for (int i = 0; i < kNumL0Files; ++i) {
+    // Make sure files are overlapping in key-range to prevent trivial move.
+    Put("key1", RandomString(&rnd, 1024));
+    Put("key2", RandomString(&rnd, 1024));
+    Flush();
+  }
+  ASSERT_EQ(kNumL0Files, NumTableFilesAtLevel(0));
+
+  // Enter read-only mode by failing a write.
+  mock_env->SetFilesystemActive(false);
+  // Make sure this is outside `CompactRange`'s range so that it doesn't fail
+  // early trying to flush memtable.
+  ASSERT_NOK(Put("key3", RandomString(&rnd, 1024)));
+
+  // In the bug scenario, the first manual compaction would fail and forget to
+  // unregister itself, causing the second one to hang forever due to conflict
+  // with a non-running compaction.
+  CompactRangeOptions cro;
+  cro.exclusive_manual_compaction = false;
+  Slice begin_key("key1");
+  Slice end_key("key2");
+  ASSERT_NOK(dbfull()->CompactRange(cro, &begin_key, &end_key));
+  ASSERT_NOK(dbfull()->CompactRange(cro, &begin_key, &end_key));
+
+  // Close before mock_env destruct.
+  Close();
+}
+
+// FixFileIngestionCompactionDeadlock tests and verifies that compaction and
+// file ingestion do not cause deadlock in the event of write stall triggered
+// by number of L0 files reaching level0_stop_writes_trigger.
+TEST_P(DBCompactionTestWithParam, FixFileIngestionCompactionDeadlock) {
+  const int kNumKeysPerFile = 100;
+  // Generate SST files.
+  Options options = CurrentOptions();
+
+  // Generate an external SST file containing a single key, i.e. 99
+  std::string sst_files_dir = dbname_ + "/sst_files/";
+  test::DestroyDir(env_, sst_files_dir);
+  ASSERT_OK(env_->CreateDir(sst_files_dir));
+  SstFileWriter sst_writer(EnvOptions(), options);
+  const std::string sst_file_path = sst_files_dir + "test.sst";
+  ASSERT_OK(sst_writer.Open(sst_file_path));
+  ASSERT_OK(sst_writer.Put(Key(kNumKeysPerFile - 1), "value"));
+  ASSERT_OK(sst_writer.Finish());
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->LoadDependency({
+      {"DBImpl::IngestExternalFile:AfterIncIngestFileCounter",
+       "BackgroundCallCompaction:0"},
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  options.write_buffer_size = 110 << 10;  // 110KB
+  options.level0_file_num_compaction_trigger =
+      options.level0_stop_writes_trigger;
+  options.max_subcompactions = max_subcompactions_;
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  DestroyAndReopen(options);
+  Random rnd(301);
+
+  // Generate level0_stop_writes_trigger L0 files to trigger write stop
+  for (int i = 0; i != options.level0_file_num_compaction_trigger; ++i) {
+    for (int j = 0; j != kNumKeysPerFile; ++j) {
+      ASSERT_OK(Put(Key(j), RandomString(&rnd, 990)));
+    }
+    if (0 == i) {
+      // When we reach here, the memtables have kNumKeysPerFile keys. Note that
+      // flush is not yet triggered. We need to write an extra key so that the
+      // write path will call PreprocessWrite and flush the previous key-value
+      // pairs to e flushed. After that, there will be the newest key in the
+      // memtable, and a bunch of L0 files. Since there is already one key in
+      // the memtable, then for i = 1, 2, ..., we do not have to write this
+      // extra key to trigger flush.
+      ASSERT_OK(Put("", ""));
+    }
+    dbfull()->TEST_WaitForFlushMemTable();
+    ASSERT_EQ(NumTableFilesAtLevel(0 /*level*/, 0 /*cf*/), i + 1);
+  }
+  // When we reach this point, there will be level0_stop_writes_trigger L0
+  // files and one extra key (99) in memory, which overlaps with the external
+  // SST file. Write stall triggers, and can be cleared only after compaction
+  // reduces the number of L0 files.
+
+  // Compaction will also be triggered since we have reached the threshold for
+  // auto compaction. Note that compaction may begin after the following file
+  // ingestion thread and waits for ingestion to finish.
+
+  // Thread to ingest file with overlapping key range with the current
+  // memtable. Consequently ingestion will trigger a flush. The flush MUST
+  // proceed without waiting for the write stall condition to clear, otherwise
+  // deadlock can happen.
+  port::Thread ingestion_thr([&]() {
+    IngestExternalFileOptions ifo;
+    Status s = db_->IngestExternalFile({sst_file_path}, ifo);
+    ASSERT_OK(s);
+  });
+
+  // More write to trigger write stop
+  ingestion_thr.join();
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+  Close();
+}
+
 #endif // !defined(ROCKSDB_LITE)
 }  // namespace rocksdb
 

db/db_dynamic_level_test.cc

@@ -27,7 +27,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase) {
     return;
   }
   // Use InMemoryEnv, or it would be too slow.
-  unique_ptr<Env> env(new MockEnv(env_));
+  std::unique_ptr<Env> env(new MockEnv(env_));
 
   const int kNKeys = 1000;
   int keys[kNKeys];
@@ -125,6 +125,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   int kMaxKey = 1000000;
 
   Options options = CurrentOptions();
+  options.compression = kNoCompression;
   options.create_if_missing = true;
   options.write_buffer_size = 20480;
   options.max_write_buffer_number = 2;
@@ -167,8 +168,8 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(4U, int_prop);
 
-  // Insert extra about 28K to L0. After they are compacted to L4, base level
-  // should be changed to L3.
+  // Insert extra about 28K to L0. After they are compacted to L4, the base
+  // level should be changed to L3.
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
@@ -189,13 +190,7 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   ASSERT_TRUE(db_->GetProperty("rocksdb.num-files-at-level2", &str_prop));
   ASSERT_EQ("0", str_prop);
 
-  // Trigger parallel compaction, and the first one would change the base
-  // level.
-  // Hold compaction jobs to make sure
-  rocksdb::SyncPoint::GetInstance()->SetCallBack(
-      "CompactionJob::Run():Start",
-      [&](void* /*arg*/) { env_->SleepForMicroseconds(100000); });
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  // Write even more data while leaving the base level at L3.
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
@@ -208,18 +203,12 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
       {"disable_auto_compactions", "false"},
   }));
   Flush();
-  // Wait for 200 milliseconds before proceeding compactions to make sure two
-  // parallel ones are executed.
-  env_->SleepForMicroseconds(200000);
   dbfull()->TEST_WaitForCompact();
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(3U, int_prop);
-  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 
-  // Trigger a condition that the compaction changes base level and L0->Lbase
-  // happens at the same time.
-  // We try to make last levels' targets to be 40K, 160K, 640K, add triggers
-  // another compaction from 40K->160K.
+  // Fill up L0, and then run an (auto) L0->Lmax compaction to raise the base
+  // level to 2.
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
@@ -229,23 +218,31 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
                   RandomString(&rnd, 380)));
   }
+
+  // Make sure that the compaction starts before the last bit of data is
+  // flushed, so that the base level isn't raised to L1.
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"CompactionJob::Run():Start", "DynamicLevelMaxBytesBase2:0"},
+  });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "false"},
   }));
+
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:0");
   Flush();
   dbfull()->TEST_WaitForCompact();
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(2U, int_prop);
-
-  // A manual compaction will trigger the base level to become L2
-  // Keep Writing data until base level changed 2->1. There will be L0->L2
-  // compaction going on at the same time.
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
   rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 
+  // Write more data until the base level changes to L1. There will be
+  // a manual compaction going on at the same time.
   rocksdb::SyncPoint::GetInstance()->LoadDependency({
-      {"CompactionJob::Run():Start", "DynamicLevelMaxBytesBase2:0"},
-      {"DynamicLevelMaxBytesBase2:1", "CompactionJob::Run():End"},
+      {"CompactionJob::Run():Start", "DynamicLevelMaxBytesBase2:1"},
+      {"DynamicLevelMaxBytesBase2:2", "CompactionJob::Run():End"},
       {"DynamicLevelMaxBytesBase2:compact_range_finish",
        "FlushJob::WriteLevel0Table"},
   });
@@ -257,12 +254,12 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
     TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_finish");
   });
 
-  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:0");
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:1");
   for (int i = 0; i < 2; i++) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
                   RandomString(&rnd, 380)));
   }
-  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:1");
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:2");
 
   Flush();
 

db/db_encryption_test.cc

@@ -40,7 +40,7 @@ TEST_F(DBEncryptionTest, CheckEncrypted) {
       continue;
     }
     auto filePath = dbname_ + "/" + *it;
-    unique_ptr<SequentialFile> seqFile;
+    std::unique_ptr<SequentialFile> seqFile;
     auto envOptions = EnvOptions(CurrentOptions());
     status = defaultEnv->NewSequentialFile(filePath, &seqFile, envOptions);
     ASSERT_OK(status);

db/db_filesnapshot.cc

@@ -44,7 +44,7 @@ Status DBImpl::EnableFileDeletions(bool force) {
   // Job id == 0 means that this is not our background process, but rather
   // user thread
   JobContext job_context(0);
-  bool should_purge_files = false;
+  bool file_deletion_enabled = false;
   {
     InstrumentedMutexLock l(&mutex_);
     if (force) {
@@ -54,19 +54,18 @@ Status DBImpl::EnableFileDeletions(bool force) {
       --disable_delete_obsolete_files_;
     }
     if (disable_delete_obsolete_files_ == 0)  {
-      ROCKS_LOG_INFO(immutable_db_options_.info_log, "File Deletions Enabled");
-      should_purge_files = true;
+      file_deletion_enabled = true;
       FindObsoleteFiles(&job_context, true);
       bg_cv_.SignalAll();
-    } else {
-      ROCKS_LOG_WARN(
-          immutable_db_options_.info_log,
-          "File Deletions Enable, but not really enabled. Counter: %d",
-          disable_delete_obsolete_files_);
     }
   }
-  if (should_purge_files)  {
+  if (file_deletion_enabled) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log, "File Deletions Enabled");
     PurgeObsoleteFiles(job_context);
+  } else {
+    ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                   "File Deletions Enable, but not really enabled. Counter: %d",
+                   disable_delete_obsolete_files_);
   }
   job_context.Clean();
   LogFlush(immutable_db_options_.info_log);
@@ -87,19 +86,28 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
   if (flush_memtable) {
     // flush all dirty data to disk.
     Status status;
-    for (auto cfd : *versions_->GetColumnFamilySet()) {
-      if (cfd->IsDropped()) {
-        continue;
-      }
-      cfd->Ref();
+    if (immutable_db_options_.atomic_flush) {
+      autovector<ColumnFamilyData*> cfds;
+      SelectColumnFamiliesForAtomicFlush(&cfds);
       mutex_.Unlock();
-      status = FlushMemTable(cfd, FlushOptions(), FlushReason::kGetLiveFiles);
-      TEST_SYNC_POINT("DBImpl::GetLiveFiles:1");
-      TEST_SYNC_POINT("DBImpl::GetLiveFiles:2");
+      status = AtomicFlushMemTables(cfds, FlushOptions(),
+                                    FlushReason::kGetLiveFiles);
       mutex_.Lock();
-      cfd->Unref();
-      if (!status.ok()) {
-        break;
+    } else {
+      for (auto cfd : *versions_->GetColumnFamilySet()) {
+        if (cfd->IsDropped()) {
+          continue;
+        }
+        cfd->Ref();
+        mutex_.Unlock();
+        status = FlushMemTable(cfd, FlushOptions(), FlushReason::kGetLiveFiles);
+        TEST_SYNC_POINT("DBImpl::GetLiveFiles:1");
+        TEST_SYNC_POINT("DBImpl::GetLiveFiles:2");
+        mutex_.Lock();
+        cfd->Unref();
+        if (!status.ok()) {
+          break;
+        }
       }
     }
     versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
@@ -126,7 +134,7 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
 
   // create names of the live files. The names are not absolute
   // paths, instead they are relative to dbname_;
-  for (auto live_file : live) {
+  for (const auto& live_file : live) {
     ret.push_back(MakeTableFileName("", live_file.GetNumber()));
   }
 

db/db_flush_test.cc

@@ -25,6 +25,12 @@ class DBFlushDirectIOTest : public DBFlushTest,
   DBFlushDirectIOTest() : DBFlushTest() {}
 };
 
+class DBAtomicFlushTest : public DBFlushTest,
+                          public ::testing::WithParamInterface<bool> {
+ public:
+  DBAtomicFlushTest() : DBFlushTest() {}
+};
+
 // We had issue when two background threads trying to flush at the same time,
 // only one of them get committed. The test verifies the issue is fixed.
 TEST_F(DBFlushTest, FlushWhileWritingManifest) {
@@ -35,11 +41,12 @@ 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",
         "DBFlushTest::FlushWhileWritingManifest:1"},
-       {"MemTableList::InstallMemtableFlushResults:InProgress",
+       {"MemTableList::TryInstallMemtableFlushResults:InProgress",
         "VersionSet::LogAndApply:WriteManifestDone"}});
   SyncPoint::GetInstance()->EnableProcessing();
 
@@ -55,6 +62,8 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
 #endif  // ROCKSDB_LITE
 }
 
+// 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_));
@@ -63,11 +72,15 @@ TEST_F(DBFlushTest, SyncFail) {
   options.env = fault_injection_env.get();
 
   SyncPoint::GetInstance()->LoadDependency(
-      {{"DBFlushTest::SyncFail:1", "DBImpl::SyncClosedLogs:Start"},
+      {{"DBFlushTest::SyncFail:GetVersionRefCount:1",
+        "DBImpl::FlushMemTableToOutputFile:BeforePickMemtables"},
+       {"DBImpl::FlushMemTableToOutputFile:AfterPickMemtables",
+        "DBFlushTest::SyncFail:GetVersionRefCount:2"},
+       {"DBFlushTest::SyncFail:1", "DBImpl::SyncClosedLogs:Start"},
        {"DBImpl::SyncClosedLogs:Failed", "DBFlushTest::SyncFail:2"}});
   SyncPoint::GetInstance()->EnableProcessing();
 
-  Reopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
   Put("key", "value");
   auto* cfd =
       reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
@@ -78,6 +91,10 @@ TEST_F(DBFlushTest, SyncFail) {
   // Flush installs a new super-version. Get the ref count after that.
   auto current_before = cfd->current();
   int refs_before = cfd->current()->TEST_refs();
+  TEST_SYNC_POINT("DBFlushTest::SyncFail:GetVersionRefCount:1");
+  TEST_SYNC_POINT("DBFlushTest::SyncFail:GetVersionRefCount:2");
+  int refs_after_picking_memtables = cfd->current()->TEST_refs();
+  ASSERT_EQ(refs_before + 1, refs_after_picking_memtables);
   fault_injection_env->SetFilesystemActive(false);
   TEST_SYNC_POINT("DBFlushTest::SyncFail:1");
   TEST_SYNC_POINT("DBFlushTest::SyncFail:2");
@@ -93,6 +110,30 @@ TEST_F(DBFlushTest, SyncFail) {
   Destroy(options);
 }
 
+TEST_F(DBFlushTest, SyncSkip) {
+  Options options = CurrentOptions();
+
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBFlushTest::SyncSkip:1", "DBImpl::SyncClosedLogs:Skip"},
+       {"DBImpl::SyncClosedLogs:Skip", "DBFlushTest::SyncSkip:2"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  Reopen(options);
+  Put("key", "value");
+
+  FlushOptions flush_options;
+  flush_options.wait = false;
+  ASSERT_OK(dbfull()->Flush(flush_options));
+
+  TEST_SYNC_POINT("DBFlushTest::SyncSkip:1");
+  TEST_SYNC_POINT("DBFlushTest::SyncSkip:2");
+
+  // Now the background job will do the flush; wait for it.
+  dbfull()->TEST_WaitForFlushMemTable();
+
+  Destroy(options);
+}
+
 TEST_F(DBFlushTest, FlushInLowPriThreadPool) {
   // Verify setting an empty high-pri (flush) thread pool causes flushes to be
   // scheduled in the low-pri (compaction) thread pool.
@@ -209,9 +250,249 @@ TEST_F(DBFlushTest, FlushError) {
   ASSERT_NE(s, Status::OK());
 }
 
+TEST_F(DBFlushTest, ManualFlushFailsInReadOnlyMode) {
+  // Regression test for bug where manual flush hangs forever when the DB
+  // is in read-only mode. Verify it now at least returns, despite failing.
+  Options options;
+  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
+      new FaultInjectionTestEnv(env_));
+  options.env = fault_injection_env.get();
+  options.max_write_buffer_number = 2;
+  Reopen(options);
+
+  // Trigger a first flush but don't let it run
+  ASSERT_OK(db_->PauseBackgroundWork());
+  ASSERT_OK(Put("key1", "value1"));
+  FlushOptions flush_opts;
+  flush_opts.wait = false;
+  ASSERT_OK(db_->Flush(flush_opts));
+
+  // Write a key to the second memtable so we have something to flush later
+  // after the DB is in read-only mode.
+  ASSERT_OK(Put("key2", "value2"));
+
+  // Let the first flush continue, hit an error, and put the DB in read-only
+  // mode.
+  fault_injection_env->SetFilesystemActive(false);
+  ASSERT_OK(db_->ContinueBackgroundWork());
+  dbfull()->TEST_WaitForFlushMemTable();
+#ifndef ROCKSDB_LITE
+  uint64_t num_bg_errors;
+  ASSERT_TRUE(db_->GetIntProperty(DB::Properties::kBackgroundErrors,
+                                  &num_bg_errors));
+  ASSERT_GT(num_bg_errors, 0);
+#endif  // ROCKSDB_LITE
+
+  // In the bug scenario, triggering another flush would cause the second flush
+  // to hang forever. After the fix we expect it to return an error.
+  ASSERT_NOK(db_->Flush(FlushOptions()));
+
+  Close();
+}
+
+TEST_P(DBAtomicFlushTest, ManualAtomicFlush) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = GetParam();
+  options.write_buffer_size = (static_cast<size_t>(64) << 20);
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    ASSERT_OK(Put(static_cast<int>(i) /*cf*/, "key", "value", wopts));
+  }
+  std::vector<int> cf_ids;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    cf_ids.emplace_back(static_cast<int>(i));
+  }
+  ASSERT_OK(Flush(cf_ids));
+  for (size_t i = 0; i != num_cfs; ++i) {
+    auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
+    ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
+    ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
+  }
+}
+
+TEST_P(DBAtomicFlushTest, AtomicFlushTriggeredByMemTableFull) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = GetParam();
+  // 4KB so that we can easily trigger auto flush.
+  options.write_buffer_size = 4096;
+
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::BackgroundCallFlush:FlushFinish:0",
+        "DBAtomicFlushTest::AtomicFlushTriggeredByMemTableFull:BeforeCheck"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    ASSERT_OK(Put(static_cast<int>(i) /*cf*/, "key", "value", wopts));
+  }
+  // Keep writing to one of them column families to trigger auto flush.
+  for (int i = 0; i != 4000; ++i) {
+    ASSERT_OK(Put(static_cast<int>(num_cfs) - 1 /*cf*/,
+                  "key" + std::to_string(i), "value" + std::to_string(i),
+                  wopts));
+  }
+
+  TEST_SYNC_POINT(
+      "DBAtomicFlushTest::AtomicFlushTriggeredByMemTableFull:BeforeCheck");
+  if (options.atomic_flush) {
+    for (size_t i = 0; i != num_cfs - 1; ++i) {
+      auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
+      ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
+      ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
+    }
+  } else {
+    for (size_t i = 0; i != num_cfs - 1; ++i) {
+      auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
+      ASSERT_EQ(0, cfh->cfd()->imm()->NumNotFlushed());
+      ASSERT_FALSE(cfh->cfd()->mem()->IsEmpty());
+    }
+  }
+  SyncPoint::GetInstance()->DisableProcessing();
+}
+
+TEST_P(DBAtomicFlushTest, AtomicFlushRollbackSomeJobs) {
+  bool atomic_flush = GetParam();
+  if (!atomic_flush) {
+    return;
+  }
+  std::unique_ptr<FaultInjectionTestEnv> fault_injection_env(
+      new FaultInjectionTestEnv(env_));
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = atomic_flush;
+  options.env = fault_injection_env.get();
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:1",
+        "DBAtomicFlushTest::AtomicFlushRollbackSomeJobs:1"},
+       {"DBAtomicFlushTest::AtomicFlushRollbackSomeJobs:2",
+        "DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:2"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_OK(Put(cf_id, "key", "value", wopts));
+  }
+  FlushOptions flush_opts;
+  flush_opts.wait = false;
+  ASSERT_OK(dbfull()->Flush(flush_opts, handles_));
+  TEST_SYNC_POINT("DBAtomicFlushTest::AtomicFlushRollbackSomeJobs:1");
+  fault_injection_env->SetFilesystemActive(false);
+  TEST_SYNC_POINT("DBAtomicFlushTest::AtomicFlushRollbackSomeJobs:2");
+  for (auto* cfh : handles_) {
+    dbfull()->TEST_WaitForFlushMemTable(cfh);
+  }
+  for (size_t i = 0; i != num_cfs; ++i) {
+    auto cfh = static_cast<ColumnFamilyHandleImpl*>(handles_[i]);
+    ASSERT_EQ(1, cfh->cfd()->imm()->NumNotFlushed());
+    ASSERT_TRUE(cfh->cfd()->mem()->IsEmpty());
+  }
+  fault_injection_env->SetFilesystemActive(true);
+  Destroy(options);
+}
+
+TEST_P(DBAtomicFlushTest, FlushMultipleCFs_DropSomeBeforeRequestFlush) {
+  bool atomic_flush = GetParam();
+  if (!atomic_flush) {
+    return;
+  }
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = atomic_flush;
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  std::vector<int> cf_ids;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_OK(Put(cf_id, "key", "value", wopts));
+    cf_ids.push_back(cf_id);
+  }
+  ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+  ASSERT_TRUE(Flush(cf_ids).IsShutdownInProgress());
+  Destroy(options);
+}
+
+TEST_P(DBAtomicFlushTest,
+       FlushMultipleCFs_DropSomeAfterScheduleFlushBeforeFlushJobRun) {
+  bool atomic_flush = GetParam();
+  if (!atomic_flush) {
+    return;
+  }
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.atomic_flush = atomic_flush;
+
+  CreateAndReopenWithCF({"pikachu", "eevee"}, options);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DBImpl::AtomicFlushMemTables:AfterScheduleFlush",
+        "DBAtomicFlushTest::BeforeDropCF"},
+       {"DBAtomicFlushTest::AfterDropCF",
+        "DBImpl::BackgroundCallFlush:start"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  size_t num_cfs = handles_.size();
+  ASSERT_EQ(3, num_cfs);
+  WriteOptions wopts;
+  wopts.disableWAL = true;
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_OK(Put(cf_id, "key", "value", wopts));
+  }
+  port::Thread user_thread([&]() {
+    TEST_SYNC_POINT("DBAtomicFlushTest::BeforeDropCF");
+    ASSERT_OK(dbfull()->DropColumnFamily(handles_[1]));
+    TEST_SYNC_POINT("DBAtomicFlushTest::AfterDropCF");
+  });
+  FlushOptions flush_opts;
+  flush_opts.wait = true;
+  ASSERT_OK(dbfull()->Flush(flush_opts, handles_));
+  user_thread.join();
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_EQ("value", Get(cf_id, "key"));
+  }
+
+  ReopenWithColumnFamilies({kDefaultColumnFamilyName, "eevee"}, options);
+  num_cfs = handles_.size();
+  ASSERT_EQ(2, num_cfs);
+  for (size_t i = 0; i != num_cfs; ++i) {
+    int cf_id = static_cast<int>(i);
+    ASSERT_EQ("value", Get(cf_id, "key"));
+  }
+  Destroy(options);
+}
+
 INSTANTIATE_TEST_CASE_P(DBFlushDirectIOTest, DBFlushDirectIOTest,
                         testing::Bool());
 
+INSTANTIATE_TEST_CASE_P(DBAtomicFlushTest, DBAtomicFlushTest, testing::Bool());
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/db_impl.h

@@ -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"
@@ -29,6 +31,7 @@
 #include "db/log_writer.h"
 #include "db/logs_with_prep_tracker.h"
 #include "db/pre_release_callback.h"
+#include "db/range_del_aggregator.h"
 #include "db/read_callback.h"
 #include "db/snapshot_checker.h"
 #include "db/snapshot_impl.h"
@@ -44,14 +47,17 @@
 #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"
 #include "util/autovector.h"
 #include "util/event_logger.h"
 #include "util/hash.h"
+#include "util/repeatable_thread.h"
 #include "util/stop_watch.h"
 #include "util/thread_local.h"
+#include "util/trace_replay.h"
 
 namespace rocksdb {
 
@@ -70,9 +76,12 @@ 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,
@@ -220,6 +229,9 @@ class DBImpl : public DB {
   using DB::Flush;
   virtual Status Flush(const FlushOptions& options,
                        ColumnFamilyHandle* column_family) override;
+  virtual Status Flush(
+      const FlushOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_families) override;
   virtual Status FlushWAL(bool sync) override;
   bool TEST_WALBufferIsEmpty();
   virtual Status SyncWAL() override;
@@ -248,9 +260,9 @@ class DBImpl : public DB {
   virtual Status GetSortedWalFiles(VectorLogPtr& files) override;
 
   virtual Status GetUpdatesSince(
-      SequenceNumber seq_number, unique_ptr<TransactionLogIterator>* iter,
-      const TransactionLogIterator::ReadOptions&
-          read_options = TransactionLogIterator::ReadOptions()) override;
+      SequenceNumber seq_number, std::unique_ptr<TransactionLogIterator>* iter,
+      const TransactionLogIterator::ReadOptions& read_options =
+          TransactionLogIterator::ReadOptions()) override;
   virtual Status DeleteFile(std::string name) override;
   Status DeleteFilesInRanges(ColumnFamilyHandle* column_family,
                              const RangePtr* ranges, size_t n,
@@ -328,6 +340,15 @@ 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
@@ -353,7 +374,7 @@ class DBImpl : public DB {
   // The keys of this iterator are internal keys (see format.h).
   // The returned iterator should be deleted when no longer needed.
   InternalIterator* NewInternalIterator(
-      Arena* arena, RangeDelAggregator* range_del_agg,
+      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
       ColumnFamilyHandle* column_family = nullptr);
 
   LogsWithPrepTracker* logs_with_prep_tracker() {
@@ -380,7 +401,7 @@ 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 +469,8 @@ class DBImpl : public DB {
 
   int TEST_BGCompactionsAllowed() const;
   int TEST_BGFlushesAllowed() const;
+  size_t TEST_GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
+  void TEST_WaitForTimedTaskRun(std::function<void()> callback) const;
 
 #endif  // NDEBUG
 
@@ -469,6 +492,14 @@ class DBImpl : public DB {
 
   uint64_t MinLogNumberToKeep();
 
+  // Returns the lower bound file number for SSTs that won't be deleted, even if
+  // they're obsolete. This lower bound is used internally to prevent newly
+  // created flush/compaction output files from being deleted before they're
+  // installed. This technique avoids the need for tracking the exact numbers of
+  // files pending creation, although it prevents more files than necessary from
+  // being deleted.
+  uint64_t MinObsoleteSstNumberToKeep();
+
   // Returns the list of live files in 'live' and the list
   // of all files in the filesystem in 'candidate_files'.
   // If force == false and the last call was less than
@@ -528,7 +559,8 @@ class DBImpl : public DB {
   ColumnFamilyHandle* GetColumnFamilyHandle(uint32_t column_family_id);
 
   // Same as above, should called without mutex held and not on write thread.
-  ColumnFamilyHandle* GetColumnFamilyHandleUnlocked(uint32_t column_family_id);
+  std::unique_ptr<ColumnFamilyHandle> GetColumnFamilyHandleUnlocked(
+      uint32_t column_family_id);
 
   // Returns the number of currently running flushes.
   // REQUIREMENT: mutex_ must be held when calling this function.
@@ -546,35 +578,58 @@ class DBImpl : public DB {
 
   const WriteController& write_controller() { return write_controller_; }
 
-  InternalIterator* NewInternalIterator(const ReadOptions&,
-                                        ColumnFamilyData* cfd,
-                                        SuperVersion* super_version,
-                                        Arena* arena,
-                                        RangeDelAggregator* range_del_agg);
+  InternalIterator* NewInternalIterator(
+      const ReadOptions&, ColumnFamilyData* cfd, SuperVersion* super_version,
+      Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence);
 
   // hollow transactions shell used for recovery.
   // 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 +650,19 @@ 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 +671,10 @@ 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 +703,7 @@ 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;
 
@@ -645,7 +713,7 @@ class DBImpl : public DB {
  protected:
   Env* const env_;
   const std::string dbname_;
-  unique_ptr<VersionSet> versions_;
+  std::unique_ptr<VersionSet> versions_;
   // Flag to check whether we allocated and own the info log file
   bool own_info_log_;
   const DBOptions initial_db_options_;
@@ -654,6 +722,8 @@ 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.
@@ -675,6 +745,11 @@ class DBImpl : public DB {
                               const MutableCFOptions& mutable_cf_options,
                               int job_id, TableProperties prop);
 
+  void NotifyOnCompactionBegin(ColumnFamilyData* cfd,
+                               Compaction *c, const Status &st,
+                               const CompactionJobStats& job_stats,
+                               int job_id);
+
   void NotifyOnCompactionCompleted(ColumnFamilyData* cfd,
                                    Compaction *c, const Status &st,
                                    const CompactionJobStats& job_stats,
@@ -739,6 +814,7 @@ class DBImpl : public DB {
 
  private:
   friend class DB;
+  friend class ErrorHandler;
   friend class InternalStats;
   friend class PessimisticTransaction;
   friend class TransactionBaseImpl;
@@ -746,17 +822,22 @@ 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 +866,8 @@ 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 +907,45 @@ 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), max_memtable_id_(0), superversion_context_(nullptr) {}
+    BGFlushArg(ColumnFamilyData* cfd, uint64_t max_memtable_id,
+               SuperVersionContext* superversion_context)
+        : cfd_(cfd),
+          max_memtable_id_(max_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 max_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);
+
+  Status AtomicFlushMemTablesToOutputFiles(
+      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 +958,12 @@ 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 +973,43 @@ 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);
+
+  void SelectColumnFamiliesForAtomicFlush(autovector<ColumnFamilyData*>* cfds);
 
   // Force current memtable contents to be flushed.
   Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options,
                        FlushReason flush_reason, bool writes_stopped = false);
 
+  Status AtomicFlushMemTables(
+      const autovector<ColumnFamilyData*>& column_family_datas,
+      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.
+  // resuming_from_bg_err indicates whether the caller is attempting to resume
+  // from background error.
   Status WaitForFlushMemTable(ColumnFamilyData* cfd,
-                              const uint64_t* flush_memtable_id = nullptr);
+                              const uint64_t* flush_memtable_id = nullptr,
+                              bool resuming_from_bg_err = false) {
+    return WaitForFlushMemTables({cfd}, {flush_memtable_id},
+                                 resuming_from_bg_err);
+  }
+  // Wait for memtables to be flushed for multiple column families.
+  Status WaitForFlushMemTables(
+      const autovector<ColumnFamilyData*>& cfds,
+      const autovector<const uint64_t*>& flush_memtable_ids,
+      bool resuming_from_bg_err);
+
+  // REQUIRES: mutex locked and in write thread.
+  void AssignAtomicFlushSeq(const autovector<ColumnFamilyData*>& cfds);
 
   // REQUIRES: mutex locked
   Status SwitchWAL(WriteContext* write_context);
@@ -919,7 +1065,20 @@ 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 GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
+                            FlushRequest* req);
+
+  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);
@@ -938,15 +1097,19 @@ 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);
 
+  // Schedule background tasks
+  void StartTimedTasks();
+
   void PrintStatistics();
 
   // dump rocksdb.stats to LOG
-  void MaybeDumpStats();
+  void DumpStats();
 
   // Return the minimum empty level that could hold the total data in the
   // input level. Return the input level, if such level could not be found.
@@ -961,8 +1124,7 @@ 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 +1137,8 @@ class DBImpl : public DB {
 
   Status CloseHelper();
 
+  void WaitForBackgroundWork();
+
   // table_cache_ provides its own synchronization
   std::shared_ptr<Cache> table_cache_;
 
@@ -1025,7 +1189,7 @@ class DBImpl : public DB {
   bool log_empty_;
   ColumnFamilyHandleImpl* default_cf_handle_;
   InternalStats* default_cf_internal_stats_;
-  unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
+  std::unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
   struct LogFileNumberSize {
     explicit LogFileNumberSize(uint64_t _number)
         : number(_number) {}
@@ -1053,7 +1217,7 @@ class DBImpl : public DB {
 
     uint64_t number;
     // Visual Studio doesn't support deque's member to be noncopyable because
-    // of a unique_ptr as a member.
+    // of a std::unique_ptr as a member.
     log::Writer* writer;  // own
     // true for some prefix of logs_
     bool getting_synced = false;
@@ -1139,7 +1303,7 @@ class DBImpl : public DB {
 
   WriteController write_controller_;
 
-  unique_ptr<RateLimiter> low_pri_write_rate_limiter_;
+  std::unique_ptr<RateLimiter> low_pri_write_rate_limiter_;
 
   // Size of the last batch group. In slowdown mode, next write needs to
   // sleep if it uses up the quota.
@@ -1195,7 +1359,7 @@ 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 +1429,6 @@ 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
@@ -1349,6 +1510,10 @@ class DBImpl : public DB {
   // Only to be set during initialization
   std::unique_ptr<PreReleaseCallback> recoverable_state_pre_release_callback_;
 
+  // handle for scheduling jobs at fixed intervals
+  // REQUIRES: mutex locked
+  std::unique_ptr<rocksdb::RepeatableThread> thread_dump_stats_;
+
   // No copying allowed
   DBImpl(const DBImpl&);
   void operator=(const DBImpl&);
@@ -1379,6 +1544,7 @@ 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 +1573,13 @@ 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 +1587,16 @@ 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
+  std::atomic<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 +1607,19 @@ class DBImpl : public DB {
 
   // Flag to check whether Close() has been called on this DB
   bool closed_;
+
+  ErrorHandler error_handler_;
+
+  // Conditional variable to coordinate installation of atomic flush results.
+  // With atomic flush, each bg thread installs the result of flushing multiple
+  // column families, and different threads can flush different column
+  // families. It's difficult to rely on one thread to perform batch
+  // installation for all threads. This is different from the non-atomic flush
+  // case.
+  // atomic_flush_install_cv_ makes sure that threads install atomic flush
+  // results sequentially. Flush results of memtables with lower IDs get
+  // installed to MANIFEST first.
+  InstrumentedCondVar atomic_flush_install_cv_;
 };
 
 extern Options SanitizeOptions(const std::string& db,

db/db_impl_debug.cc

@@ -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();
@@ -120,7 +123,7 @@ Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
     auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
     cfd = cfh->cfd();
   }
-  return WaitForFlushMemTable(cfd);
+  return WaitForFlushMemTable(cfd, nullptr, false);
 }
 
 Status DBImpl::TEST_WaitForCompact(bool wait_unscheduled) {
@@ -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 +237,16 @@ SequenceNumber DBImpl::TEST_GetLastVisibleSequence() const {
   }
 }
 
+size_t DBImpl::TEST_GetWalPreallocateBlockSize(
+    uint64_t write_buffer_size) const {
+  InstrumentedMutexLock l(&mutex_);
+  return GetWalPreallocateBlockSize(write_buffer_size);
+}
+
+void DBImpl::TEST_WaitForTimedTaskRun(std::function<void()> callback) const {
+  if (thread_dump_stats_ != nullptr) {
+    thread_dump_stats_->TEST_WaitForRun(callback);
+  }
+}
 }  // namespace rocksdb
 #endif  // NDEBUG

db/db_impl_experimental.cc

@@ -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);

db/db_impl_files.cc

@@ -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"
@@ -19,6 +20,7 @@
 #include "util/sst_file_manager_impl.h"
 
 namespace rocksdb {
+
 uint64_t DBImpl::MinLogNumberToKeep() {
   if (allow_2pc()) {
     return versions_->min_log_number_to_keep_2pc();
@@ -27,6 +29,14 @@ uint64_t DBImpl::MinLogNumberToKeep() {
   }
 }
 
+uint64_t DBImpl::MinObsoleteSstNumberToKeep() {
+  mutex_.AssertHeld();
+  if (!pending_outputs_.empty()) {
+    return *pending_outputs_.begin();
+  }
+  return std::numeric_limits<uint64_t>::max();
+}
+
 // * Returns the list of live files in 'sst_live'
 // If it's doing full scan:
 // * Returns the list of all files in the filesystem in
@@ -99,10 +109,10 @@ 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 +123,11 @@ 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 +359,32 @@ 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 +433,19 @@ 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;
@@ -420,7 +465,13 @@ void DBImpl::PurgeObsoleteFiles(JobContext& state, bool schedule_only) {
     } else {
       dir_to_sync =
           (type == kLogFile) ? immutable_db_options_.wal_dir : dbname_;
-      fname = dir_to_sync + "/" + to_delete;
+      fname = dir_to_sync
+            + (
+                (!dir_to_sync.empty() && dir_to_sync.back() == '/') ||
+                (!to_delete.empty() && to_delete.front() == '/')
+                ? "" : "/"
+              )
+            + to_delete;
     }
 
 #ifndef ROCKSDB_LITE

db/db_impl_open.cc

@@ -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"
@@ -22,8 +23,7 @@
 #include "util/sync_point.h"
 
 namespace rocksdb {
-Options SanitizeOptions(const std::string& dbname,
-                        const Options& src) {
+Options SanitizeOptions(const std::string& dbname, const Options& src) {
   auto db_options = SanitizeOptions(dbname, DBOptions(src));
   ImmutableDBOptions immutable_db_options(db_options);
   auto cf_options =
@@ -55,10 +55,9 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.write_buffer_manager.reset(
         new WriteBufferManager(result.db_write_buffer_size));
   }
-  auto bg_job_limits = DBImpl::GetBGJobLimits(result.max_background_flushes,
-                                              result.max_background_compactions,
-                                              result.max_background_jobs,
-                                              true /* parallelize_compactions */);
+  auto bg_job_limits = DBImpl::GetBGJobLimits(
+      result.max_background_flushes, result.max_background_compactions,
+      result.max_background_jobs, true /* parallelize_compactions */);
   result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_compactions,
                                            Env::Priority::LOW);
   result.env->IncBackgroundThreadsIfNeeded(bg_job_limits.max_flushes,
@@ -106,14 +105,12 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
   }
 
-  if (result.use_direct_reads &&
-      result.compaction_readahead_size == 0) {
+  if (result.use_direct_reads && result.compaction_readahead_size == 0) {
     TEST_SYNC_POINT_CALLBACK("SanitizeOptions:direct_io", nullptr);
     result.compaction_readahead_size = 1024 * 1024 * 2;
   }
 
-  if (result.compaction_readahead_size > 0 ||
-      result.use_direct_reads) {
+  if (result.compaction_readahead_size > 0 || result.use_direct_reads) {
     result.new_table_reader_for_compaction_inputs = true;
   }
 
@@ -133,8 +130,15 @@ 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;
 }
 
@@ -210,7 +214,7 @@ static Status ValidateOptions(
 
   return Status::OK();
 }
-} // namespace
+}  // namespace
 Status DBImpl::NewDB() {
   VersionEdit new_db;
   new_db.SetLogNumber(0);
@@ -222,7 +226,7 @@ Status DBImpl::NewDB() {
   ROCKS_LOG_INFO(immutable_db_options_.info_log, "Creating manifest 1 \n");
   const std::string manifest = DescriptorFileName(dbname_, 1);
   {
-    unique_ptr<WritableFile> file;
+    std::unique_ptr<WritableFile> file;
     EnvOptions env_options = env_->OptimizeForManifestWrite(env_options_);
     s = NewWritableFile(env_, manifest, &file, env_options);
     if (!s.ok()) {
@@ -230,8 +234,9 @@ Status DBImpl::NewDB() {
     }
     file->SetPreallocationBlockSize(
         immutable_db_options_.manifest_preallocation_size);
-    unique_ptr<WritableFileWriter> file_writer(
-        new WritableFileWriter(std::move(file), env_options));
+    std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+        std::move(file), manifest, env_options, nullptr /* stats */,
+        immutable_db_options_.listeners));
     log::Writer log(std::move(file_writer), 0, false);
     std::string record;
     new_db.EncodeTo(&record);
@@ -249,9 +254,8 @@ Status DBImpl::NewDB() {
   return s;
 }
 
-Status DBImpl::CreateAndNewDirectory(
-    Env* env, const std::string& dirname,
-    std::unique_ptr<Directory>* directory) {
+Status DBImpl::CreateAndNewDirectory(Env* env, const std::string& dirname,
+                                     std::unique_ptr<Directory>* directory) {
   // We call CreateDirIfMissing() as the directory may already exist (if we
   // are reopening a DB), when this happens we don't want creating the
   // directory to cause an error. However, we need to check if creating the
@@ -332,8 +336,8 @@ Status DBImpl::Recover(
       }
     } else if (s.ok()) {
       if (immutable_db_options_.error_if_exists) {
-        return Status::InvalidArgument(
-            dbname_, "exists (error_if_exists is true)");
+        return Status::InvalidArgument(dbname_,
+                                       "exists (error_if_exists is true)");
       }
     } else {
       // Unexpected error reading file
@@ -353,14 +357,14 @@ Status DBImpl::Recover(
     }
     // Verify compatibility of env_options_ and filesystem
     {
-      unique_ptr<RandomAccessFile> idfile;
+      std::unique_ptr<RandomAccessFile> idfile;
       EnvOptions customized_env(env_options_);
       customized_env.use_direct_reads |=
           immutable_db_options_.use_direct_io_for_flush_and_compaction;
       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 +374,7 @@ 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 +392,16 @@ 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(
@@ -460,12 +474,26 @@ 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;
+  if (read_only) {
+    // If we are opening as read-only, we need to update options_file_number_
+    // to reflect the most recent OPTIONS file. It does not matter for regular
+    // read-write db instance because options_file_number_ will later be
+    // updated to versions_->NewFileNumber() in RenameTempFileToOptionsFile.
+    std::vector<std::string> file_names;
+    if (s.ok()) {
+      s = env_->GetChildren(GetName(), &file_names);
+    }
+    if (s.ok()) {
+      uint64_t number = 0;
+      uint64_t options_file_number = 0;
+      FileType type;
+      for (const auto& fname : file_names) {
+        if (ParseFileName(fname, &number, &type) && type == kOptionsFile) {
+          options_file_number = std::max(number, options_file_number);
+        }
+      }
+      versions_->options_file_number_ = options_file_number;
+    }
   }
 
   return s;
@@ -516,10 +544,9 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
     std::map<std::string, uint32_t> cf_name_id_map;
     std::map<uint32_t, uint64_t> cf_lognumber_map;
     for (auto cfd : *versions_->GetColumnFamilySet()) {
-      cf_name_id_map.insert(
-        std::make_pair(cfd->GetName(), cfd->GetID()));
+      cf_name_id_map.insert(std::make_pair(cfd->GetName(), cfd->GetID()));
       cf_lognumber_map.insert(
-        std::make_pair(cfd->GetID(), cfd->GetLogNumber()));
+          std::make_pair(cfd->GetID(), cfd->GetLogNumber()));
     }
 
     immutable_db_options_.wal_filter->ColumnFamilyLogNumberMap(cf_lognumber_map,
@@ -562,9 +589,9 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
       continue;
     }
 
-    unique_ptr<SequentialFileReader> file_reader;
+    std::unique_ptr<SequentialFileReader> file_reader;
     {
-      unique_ptr<SequentialFile> file;
+      std::unique_ptr<SequentialFile> file;
       status = env_->NewSequentialFile(fname, &file,
                                        env_->OptimizeForLogRead(env_options_));
       if (!status.ok()) {
@@ -577,7 +604,7 @@ 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 +624,8 @@ 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,
+                       false /* retry_after_eof */);
 
     // Determine if we should tolerate incomplete records at the tail end of the
     // Read all the records and add to a memtable
@@ -719,7 +746,7 @@ 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
@@ -869,8 +896,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
       // VersionSet::next_file_number_ always to be strictly greater than any
       // log number
       versions_->MarkFileNumberUsed(max_log_number + 1);
-      status = versions_->LogAndApply(
-          cfd, *cfd->GetLatestMutableCFOptions(), edit, &mutex_);
+      status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
+                                      edit, &mutex_);
       if (!status.ok()) {
         // Recovery failed
         break;
@@ -878,18 +905,8 @@ 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 +915,60 @@ 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();
@@ -939,11 +1010,17 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
       if (use_custom_gc_ && snapshot_checker == nullptr) {
         snapshot_checker = DisableGCSnapshotChecker::Instance();
       }
+      std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+          range_del_iters;
+      auto range_del_iter =
+          mem->NewRangeTombstoneIterator(ro, kMaxSequenceNumber);
+      if (range_del_iter != nullptr) {
+        range_del_iters.emplace_back(range_del_iter);
+      }
       s = BuildTable(
           dbname_, env_, *cfd->ioptions(), mutable_cf_options,
           env_options_for_compaction_, cfd->table_cache(), iter.get(),
-          std::unique_ptr<InternalIterator>(mem->NewRangeTombstoneIterator(ro)),
-          &meta, cfd->internal_comparator(),
+          std::move(range_del_iters), &meta, cfd->internal_comparator(),
           cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
           snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
           GetCompressionFlush(*cfd->ioptions(), mutable_cf_options),
@@ -968,7 +1045,7 @@ 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);
   }
 
@@ -977,8 +1054,8 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
   stats.bytes_written = meta.fd.GetFileSize();
   stats.num_output_files = 1;
   cfd->internal_stats()->AddCompactionStats(level, stats);
-  cfd->internal_stats()->AddCFStats(
-      InternalStats::BYTES_FLUSHED, meta.fd.GetFileSize());
+  cfd->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
+                                    meta.fd.GetFileSize());
   RecordTick(stats_, COMPACT_WRITE_BYTES, meta.fd.GetFileSize());
   return s;
 }
@@ -1003,15 +1080,16 @@ 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 +1109,7 @@ 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 +1127,12 @@ 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()) {
@@ -1067,16 +1151,16 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
   s = impl->Recover(column_families);
   if (s.ok()) {
     uint64_t new_log_number = impl->versions_->NewFileNumber();
-    unique_ptr<WritableFile> lfile;
+    std::unique_ptr<WritableFile> lfile;
     EnvOptions soptions(db_options);
     EnvOptions opt_env_options =
         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 +1168,10 @@ 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));
+        const auto& listeners = impl->immutable_db_options_.listeners;
+        std::unique_ptr<WritableFileWriter> file_writer(
+            new WritableFileWriter(std::move(lfile), log_fname, opt_env_options,
+                                   nullptr /* stats */, listeners));
         impl->logs_.emplace_back(
             new_log_number,
             new log::Writer(
@@ -1162,7 +1248,8 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
           !cfd->mem()->IsMergeOperatorSupported()) {
         s = Status::InvalidArgument(
             "The memtable of column family %s does not support merge operator "
-            "its options.merge_operator is non-null", cfd->GetName().c_str());
+            "its options.merge_operator is non-null",
+            cfd->GetName().c_str());
       }
       if (!s.ok()) {
         break;
@@ -1212,6 +1299,14 @@ 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
 
@@ -1227,6 +1322,9 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
           persist_options_status.ToString());
     }
   }
+  if (s.ok()) {
+    impl->StartTimedTasks();
+  }
   if (!s.ok()) {
     for (auto* h : *handles) {
       delete h;

db/db_impl_readonly.cc

@@ -9,7 +9,6 @@
 #include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/merge_context.h"
-#include "db/range_del_aggregator.h"
 #include "monitoring/perf_context_imp.h"
 
 namespace rocksdb {
@@ -31,22 +30,38 @@ 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);
+  SequenceNumber max_covering_tombstone_seq = 0;
   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)) {
+                              &max_covering_tombstone_seq, 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);
+                                &merge_context, &max_covering_tombstone_seq);
+    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;
 }
 
@@ -56,18 +71,20 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
   auto cfd = cfh->cfd();
   SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
   SequenceNumber latest_snapshot = versions_->LastSequence();
+  SequenceNumber read_seq =
+      read_options.snapshot != nullptr
+          ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                ->number_
+          : latest_snapshot;
   ReadCallback* read_callback = nullptr;  // No read callback provided.
   auto db_iter = NewArenaWrappedDbIterator(
       env_, read_options, *cfd->ioptions(), super_version->mutable_cf_options,
-      (read_options.snapshot != nullptr
-           ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
-                 ->number_
-           : latest_snapshot),
+      read_seq,
       super_version->mutable_cf_options.max_sequential_skip_in_iterations,
       super_version->version_number, read_callback);
   auto internal_iter =
       NewInternalIterator(read_options, cfd, super_version, db_iter->GetArena(),
-                          db_iter->GetRangeDelAggregator());
+                          db_iter->GetRangeDelAggregator(), read_seq);
   db_iter->SetIterUnderDBIter(internal_iter);
   return db_iter;
 }
@@ -83,21 +100,22 @@ Status DBImplReadOnly::NewIterators(
   iterators->clear();
   iterators->reserve(column_families.size());
   SequenceNumber latest_snapshot = versions_->LastSequence();
+  SequenceNumber read_seq =
+      read_options.snapshot != nullptr
+          ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                ->number_
+          : latest_snapshot;
 
   for (auto cfh : column_families) {
     auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
     auto* sv = cfd->GetSuperVersion()->Ref();
     auto* db_iter = NewArenaWrappedDbIterator(
-        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
-        (read_options.snapshot != nullptr
-             ? reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
-                   ->number_
-             : latest_snapshot),
+        env_, read_options, *cfd->ioptions(), sv->mutable_cf_options, read_seq,
         sv->mutable_cf_options.max_sequential_skip_in_iterations,
         sv->version_number, read_callback);
     auto* internal_iter =
         NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
-                            db_iter->GetRangeDelAggregator());
+                            db_iter->GetRangeDelAggregator(), read_seq);
     db_iter->SetIterUnderDBIter(internal_iter);
     iterators->push_back(db_iter);
   }

db/db_impl_readonly.h

@@ -89,10 +89,11 @@ class DBImplReadOnly : public DBImpl {
   virtual Status EnableFileDeletions(bool /*force*/) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
-  virtual Status GetLiveFiles(std::vector<std::string>&,
-                              uint64_t* /*manifest_file_size*/,
-                              bool /*flush_memtable*/ = true) override {
-    return Status::NotSupported("Not supported operation in read only mode.");
+  virtual Status GetLiveFiles(std::vector<std::string>& ret,
+                              uint64_t* manifest_file_size,
+                              bool /*flush_memtable*/) override {
+    return DBImpl::GetLiveFiles(ret, manifest_file_size,
+                                false /* flush_memtable */);
   }
 
   using DBImpl::Flush;

db/db_impl_write.cc

@@ -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.");
   }
@@ -258,18 +265,19 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     // We're optimistic, updating the stats before we successfully
     // commit.  That lets us release our leader status early.
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count,
+    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count,
                       concurrent_update);
     RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
+    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
                       concurrent_update);
     RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-    stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
+    stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
+                      concurrent_update);
     RecordTick(stats_, WRITE_DONE_BY_SELF);
     auto write_done_by_other = write_group.size - 1;
     if (write_done_by_other > 0) {
-      stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
-                        concurrent_update);
+      stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
+                        write_done_by_other, concurrent_update);
       RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
     }
     MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -310,7 +318,8 @@ 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
@@ -330,8 +339,6 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
           }
         }
         write_group.last_sequence = last_sequence;
-        write_group.running.store(static_cast<uint32_t>(write_group.size),
-                                  std::memory_order_relaxed);
         write_thread_.LaunchParallelMemTableWriters(&write_group);
         in_parallel_group = true;
 
@@ -345,7 +352,7 @@ 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) {
@@ -461,9 +468,9 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
     }
 
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN, total_count);
+    stats->AddDBStats(InternalStats::kIntStatsNumKeysWritten, total_count);
     RecordTick(stats_, NUMBER_KEYS_WRITTEN, total_count);
-    stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size);
+    stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size);
     RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
     MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
 
@@ -471,10 +478,10 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
 
     if (w.ShouldWriteToWAL()) {
       PERF_TIMER_GUARD(write_wal_time);
-      stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
+      stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1);
       RecordTick(stats_, WRITE_DONE_BY_SELF, 1);
       if (wal_write_group.size > 1) {
-        stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
+        stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
                           wal_write_group.size - 1);
         RecordTick(stats_, WRITE_DONE_BY_OTHER, wal_write_group.size - 1);
       }
@@ -507,7 +514,8 @@ 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 +572,6 @@ 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);
@@ -585,15 +592,16 @@ Status DBImpl::WriteImplWALOnly(const WriteOptions& write_options,
   // We're optimistic, updating the stats before we successfully
   // commit.  That lets us release our leader status early.
   auto stats = default_cf_internal_stats_;
-  stats->AddDBStats(InternalStats::BYTES_WRITTEN, total_byte_size,
+  stats->AddDBStats(InternalStats::kIntStatsBytesWritten, total_byte_size,
                     concurrent_update);
   RecordTick(stats_, BYTES_WRITTEN, total_byte_size);
-  stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1, concurrent_update);
+  stats->AddDBStats(InternalStats::kIntStatsWriteDoneBySelf, 1,
+                    concurrent_update);
   RecordTick(stats_, WRITE_DONE_BY_SELF);
   auto write_done_by_other = write_group.size - 1;
   if (write_done_by_other > 0) {
-    stats->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER, write_done_by_other,
-                      concurrent_update);
+    stats->AddDBStats(InternalStats::kIntStatsWriteDoneByOther,
+                      write_done_by_other, concurrent_update);
     RecordTick(stats_, WRITE_DONE_BY_OTHER, write_done_by_other);
   }
   MeasureTime(stats_, BYTES_PER_WRITE, total_byte_size);
@@ -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);
   }
@@ -918,12 +910,12 @@ Status DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
   if (status.ok()) {
     auto stats = default_cf_internal_stats_;
     if (need_log_sync) {
-      stats->AddDBStats(InternalStats::WAL_FILE_SYNCED, 1);
+      stats->AddDBStats(InternalStats::kIntStatsWalFileSynced, 1);
       RecordTick(stats_, WAL_FILE_SYNCED);
     }
-    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size);
+    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size);
     RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal);
+    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal);
     RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
   }
   return status;
@@ -969,9 +961,10 @@ Status DBImpl::ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
   if (status.ok()) {
     const bool concurrent = true;
     auto stats = default_cf_internal_stats_;
-    stats->AddDBStats(InternalStats::WAL_FILE_BYTES, log_size, concurrent);
+    stats->AddDBStats(InternalStats::kIntStatsWalFileBytes, log_size,
+                      concurrent);
     RecordTick(stats_, WAL_FILE_BYTES, log_size);
-    stats->AddDBStats(InternalStats::WRITE_WITH_WAL, write_with_wal,
+    stats->AddDBStats(InternalStats::kIntStatsWriteWithWal, write_with_wal,
                       concurrent);
     RecordTick(stats_, WRITE_WITH_WAL, write_with_wal);
   }
@@ -1024,6 +1017,28 @@ Status DBImpl::WriteRecoverableState() {
   return Status::OK();
 }
 
+void DBImpl::SelectColumnFamiliesForAtomicFlush(
+    autovector<ColumnFamilyData*>* cfds) {
+  for (ColumnFamilyData* cfd : *versions_->GetColumnFamilySet()) {
+    if (cfd->IsDropped()) {
+      continue;
+    }
+    if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
+        !cached_recoverable_state_empty_.load()) {
+      cfds->push_back(cfd);
+    }
+  }
+}
+
+// Assign sequence number for atomic flush.
+void DBImpl::AssignAtomicFlushSeq(const autovector<ColumnFamilyData*>& cfds) {
+  assert(immutable_db_options_.atomic_flush);
+  auto seq = versions_->LastSequence();
+  for (auto cfd : cfds) {
+    cfd->imm()->AssignAtomicFlushSeq(seq);
+  }
+}
+
 Status DBImpl::SwitchWAL(WriteContext* write_context) {
   mutex_.AssertHeld();
   assert(write_context != nullptr);
@@ -1072,20 +1087,39 @@ 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
-  for (auto cfd : *versions_->GetColumnFamilySet()) {
-    if (cfd->IsDropped()) {
-      continue;
-    }
-    if (cfd->OldestLogToKeep() <= oldest_alive_log) {
-      status = SwitchMemtable(cfd, write_context);
-      if (!status.ok()) {
-        break;
+  autovector<ColumnFamilyData*> cfds;
+  if (immutable_db_options_.atomic_flush) {
+    SelectColumnFamiliesForAtomicFlush(&cfds);
+  } else {
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      if (cfd->IsDropped()) {
+        continue;
+      }
+      if (cfd->OldestLogToKeep() <= oldest_alive_log) {
+        cfds.push_back(cfd);
       }
-      cfd->imm()->FlushRequested();
-      SchedulePendingFlush(cfd, FlushReason::kWriteBufferManager);
     }
   }
-  MaybeScheduleFlushOrCompaction();
+  for (const auto cfd : cfds) {
+    cfd->Ref();
+    status = SwitchMemtable(cfd, write_context);
+    cfd->Unref();
+    if (!status.ok()) {
+      break;
+    }
+  }
+  if (status.ok()) {
+    if (immutable_db_options_.atomic_flush) {
+      AssignAtomicFlushSeq(cfds);
+    }
+    for (auto cfd : cfds) {
+      cfd->imm()->FlushRequested();
+    }
+    FlushRequest flush_req;
+    GenerateFlushRequest(cfds, &flush_req);
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferManager);
+    MaybeScheduleFlushOrCompaction();
+  }
   return status;
 }
 
@@ -1107,32 +1141,55 @@ Status DBImpl::HandleWriteBufferFull(WriteContext* write_context) {
       write_buffer_manager_->buffer_size());
   // no need to refcount because drop is happening in write thread, so can't
   // happen while we're in the write thread
-  ColumnFamilyData* cfd_picked = nullptr;
-  SequenceNumber seq_num_for_cf_picked = kMaxSequenceNumber;
+  autovector<ColumnFamilyData*> cfds;
+  if (immutable_db_options_.atomic_flush) {
+    SelectColumnFamiliesForAtomicFlush(&cfds);
+  } else {
+    ColumnFamilyData* cfd_picked = nullptr;
+    SequenceNumber seq_num_for_cf_picked = kMaxSequenceNumber;
 
-  for (auto cfd : *versions_->GetColumnFamilySet()) {
-    if (cfd->IsDropped()) {
-      continue;
-    }
-    if (!cfd->mem()->IsEmpty()) {
-      // We only consider active mem table, hoping immutable memtable is
-      // already in the process of flushing.
-      uint64_t seq = cfd->mem()->GetCreationSeq();
-      if (cfd_picked == nullptr || seq < seq_num_for_cf_picked) {
-        cfd_picked = cfd;
-        seq_num_for_cf_picked = seq;
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      if (cfd->IsDropped()) {
+        continue;
+      }
+      if (!cfd->mem()->IsEmpty()) {
+        // We only consider active mem table, hoping immutable memtable is
+        // already in the process of flushing.
+        uint64_t seq = cfd->mem()->GetCreationSeq();
+        if (cfd_picked == nullptr || seq < seq_num_for_cf_picked) {
+          cfd_picked = cfd;
+          seq_num_for_cf_picked = seq;
+        }
       }
     }
-  }
-  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();
+    if (cfd_picked != nullptr) {
+      cfds.push_back(cfd_picked);
     }
   }
+
+  for (const auto cfd : cfds) {
+    if (cfd->mem()->IsEmpty()) {
+      continue;
+    }
+    cfd->Ref();
+    status = SwitchMemtable(cfd, write_context);
+    cfd->Unref();
+    if (!status.ok()) {
+      break;
+    }
+  }
+  if (status.ok()) {
+    if (immutable_db_options_.atomic_flush) {
+      AssignAtomicFlushSeq(cfds);
+    }
+    for (const auto cfd : cfds) {
+      cfd->imm()->FlushRequested();
+    }
+    FlushRequest flush_req;
+    GenerateFlushRequest(cfds, &flush_req);
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+    MaybeScheduleFlushOrCompaction();
+  }
   return status;
 }
 
@@ -1154,10 +1211,14 @@ 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
@@ -1174,25 +1235,47 @@ 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);
   if (delayed) {
-    default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_STALL_MICROS,
-                                           time_delayed);
+    default_cf_internal_stats_->AddDBStats(
+        InternalStats::kIntStatsWriteStallMicros, time_delayed);
     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,
@@ -1226,17 +1309,42 @@ Status DBImpl::ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
 }
 
 Status DBImpl::ScheduleFlushes(WriteContext* context) {
-  ColumnFamilyData* cfd;
-  while ((cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
-    auto status = SwitchMemtable(cfd, context, FlushReason::kWriteBufferFull);
-    if (cfd->Unref()) {
-      delete cfd;
+  autovector<ColumnFamilyData*> cfds;
+  if (immutable_db_options_.atomic_flush) {
+    SelectColumnFamiliesForAtomicFlush(&cfds);
+    for (auto cfd : cfds) {
+      cfd->Ref();
     }
-    if (!status.ok()) {
-      return status;
+    flush_scheduler_.Clear();
+  } else {
+    ColumnFamilyData* tmp_cfd;
+    while ((tmp_cfd = flush_scheduler_.TakeNextColumnFamily()) != nullptr) {
+      cfds.push_back(tmp_cfd);
     }
   }
-  return Status::OK();
+  Status status;
+  for (auto& cfd : cfds) {
+    if (!cfd->mem()->IsEmpty()) {
+      status = SwitchMemtable(cfd, context);
+    }
+    if (cfd->Unref()) {
+      delete cfd;
+      cfd = nullptr;
+    }
+    if (!status.ok()) {
+      break;
+    }
+  }
+  if (status.ok()) {
+    if (immutable_db_options_.atomic_flush) {
+      AssignAtomicFlushSeq(cfds);
+    }
+    FlushRequest flush_req;
+    GenerateFlushRequest(cfds, &flush_req);
+    SchedulePendingFlush(flush_req, FlushReason::kWriteBufferFull);
+    MaybeScheduleFlushOrCompaction();
+  }
+  return status;
 }
 
 #ifndef ROCKSDB_LITE
@@ -1257,8 +1365,7 @@ 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_) {
@@ -1267,7 +1374,7 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
     nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
   }
 
-  unique_ptr<WritableFile> lfile;
+  std::unique_ptr<WritableFile> lfile;
   log::Writer* new_log = nullptr;
   MemTable* new_mem = nullptr;
 
@@ -1281,7 +1388,11 @@ 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 +1434,8 @@ 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 +1443,12 @@ 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
@@ -1347,8 +1458,9 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context,
         // of calling GetWalPreallocateBlockSize()
         lfile->SetPreallocationBlockSize(preallocate_block_size);
         lfile->SetWriteLifeTimeHint(write_hint);
-        unique_ptr<WritableFileWriter> file_writer(
-            new WritableFileWriter(std::move(lfile), opt_env_opt));
+        std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
+            std::move(lfile), log_fname, opt_env_opt, nullptr /* stats */,
+            immutable_db_options_.listeners));
         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 +1538,7 @@ 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);
   }

db/db_info_dumper.cc

@@ -42,7 +42,7 @@ void DumpDBFileSummary(const ImmutableDBOptions& options,
           "Error when reading %s dir\n", dbname.c_str());
   }
   std::sort(files.begin(), files.end());
-  for (std::string file : files) {
+  for (const std::string& file : files) {
     if (!ParseFileName(file, &number, &type)) {
       continue;
     }
@@ -85,7 +85,7 @@ void DumpDBFileSummary(const ImmutableDBOptions& options,
         continue;
       }
       std::sort(files.begin(), files.end());
-      for (std::string file : files) {
+      for (const std::string& file : files) {
         if (ParseFileName(file, &number, &type)) {
           if (type == kTableFile && ++file_num < 10) {
             file_info.append(file).append(" ");
@@ -109,7 +109,7 @@ void DumpDBFileSummary(const ImmutableDBOptions& options,
       return;
     }
     wal_info.clear();
-    for (std::string file : files) {
+    for (const std::string& file : files) {
       if (ParseFileName(file, &number, &type)) {
         if (type == kLogFile) {
           env->GetFileSize(options.wal_dir + "/" + file, &file_size);

db/db_io_failure_test.cc

@@ -88,7 +88,6 @@ TEST_F(DBIOFailureTest, DropWritesFlush) {
     env_->drop_writes_.store(false, std::memory_order_release);
   } while (ChangeCompactOptions());
 }
-#endif  // ROCKSDB_LITE
 
 // Check that CompactRange() returns failure if there is not enough space left
 // on device
@@ -116,6 +115,7 @@ TEST_F(DBIOFailureTest, NoSpaceCompactRange) {
     env_->no_space_.store(false, std::memory_order_release);
   } while (ChangeCompactOptions());
 }
+#endif  // ROCKSDB_LITE
 
 TEST_F(DBIOFailureTest, NonWritableFileSystem) {
   do {

db/db_iter.cc

@@ -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),
@@ -132,13 +134,14 @@ class DBIter final: public Iterator {
         prefix_same_as_start_(read_options.prefix_same_as_start),
         pin_thru_lifetime_(read_options.pin_data),
         total_order_seek_(read_options.total_order_seek),
-        range_del_agg_(cf_options.internal_comparator, s,
-                       true /* collapse_deletions */),
+        range_del_agg_(&cf_options.internal_comparator, s),
         read_callback_(read_callback),
+        db_impl_(db_impl),
+        cfd_(cfd),
         allow_blob_(allow_blob),
         is_blob_(false),
         start_seqnum_(read_options.iter_start_seqnum) {
-    RecordTick(statistics_, NO_ITERATORS);
+    RecordTick(statistics_, NO_ITERATOR_CREATED);
     prefix_extractor_ = mutable_cf_options.prefix_extractor.get();
     max_skip_ = max_sequential_skip_in_iterations;
     max_skippable_internal_keys_ = read_options.max_skippable_internal_keys;
@@ -154,9 +157,7 @@ class DBIter final: public Iterator {
     if (pinned_iters_mgr_.PinningEnabled()) {
       pinned_iters_mgr_.ReleasePinnedData();
     }
-    // Compiler warning issue filed:
-    // https://github.com/facebook/rocksdb/issues/3013
-    RecordTick(statistics_, NO_ITERATORS, uint64_t(-1));
+    RecordTick(statistics_, NO_ITERATOR_DELETED);
     ResetInternalKeysSkippedCounter();
     local_stats_.BumpGlobalStatistics(statistics_);
     if (!arena_mode_) {
@@ -170,7 +171,7 @@ class DBIter final: public Iterator {
     iter_ = iter;
     iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
   }
-  virtual RangeDelAggregator* GetRangeDelAggregator() {
+  virtual ReadRangeDelAggregator* GetRangeDelAggregator() {
     return &range_del_agg_;
   }
 
@@ -228,7 +229,7 @@ class DBIter final: public Iterator {
       *prop = saved_key_.GetUserKey().ToString();
       return Status::OK();
     }
-    return Status::InvalidArgument("Undentified property.");
+    return Status::InvalidArgument("Unidentified property.");
   }
 
   virtual void Next() override;
@@ -260,6 +261,18 @@ 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() {
@@ -328,10 +341,12 @@ class DBIter final: public Iterator {
   const bool total_order_seek_;
   // List of operands for merge operator.
   MergeContext merge_context_;
-  RangeDelAggregator range_del_agg_;
+  ReadRangeDelAggregator range_del_agg_;
   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 +521,7 @@ 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 +551,7 @@ 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 +591,7 @@ 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 +658,7 @@ 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 +907,7 @@ 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 +916,7 @@ 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 +934,7 @@ 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 +1067,7 @@ 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 +1112,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
 
     if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
         range_del_agg_.ShouldDelete(
-            ikey,
-            RangeDelAggregator::RangePositioningMode::kBackwardTraversal)) {
+            ikey, RangeDelPositioningMode::kForwardTraversal)) {
       break;
     } else if (ikey.type == kTypeValue) {
       const Slice val = iter_->value();
@@ -1194,7 +1203,7 @@ 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 +1243,21 @@ 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 +1265,28 @@ 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 +1337,15 @@ 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 +1392,7 @@ 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 +1443,7 @@ void DBIter::SeekToLast() {
   {
     PERF_TIMER_GUARD(seek_internal_seek_time);
     iter_->SeekToLast();
-    range_del_agg_.InvalidateTombstoneMapPositions();
+    range_del_agg_.InvalidateRangeDelMapPositions();
   }
   PrevInternal();
   if (statistics_ != nullptr) {
@@ -1432,17 +1468,18 @@ 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;
 }
 
 ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
 
-RangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
+ReadRangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
   return db_iter_->GetRangeDelAggregator();
 }
 
@@ -1483,13 +1520,14 @@ 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;
 }
@@ -1513,10 +1551,12 @@ 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());
+        read_options_, cfd_, sv, &arena_, db_iter_->GetRangeDelAggregator(),
+        latest_seq);
     SetIterUnderDBIter(internal_iter);
   } else {
     db_iter_->set_sequence(latest_seq);
@@ -1535,7 +1575,7 @@ 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);

db/db_iter.h

@@ -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
@@ -50,7 +48,7 @@ class ArenaWrappedDBIter : public Iterator {
   // Get the arena to be used to allocate memory for DBIter to be wrapped,
   // as well as child iterators in it.
   virtual Arena* GetArena() { return &arena_; }
-  virtual RangeDelAggregator* GetRangeDelAggregator();
+  virtual ReadRangeDelAggregator* GetRangeDelAggregator();
 
   // Set the internal iterator wrapped inside the DB Iterator. Usually it is
   // a merging iterator.
@@ -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,

db/db_iter_test.cc

@@ -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);

db/db_iterator_test.cc

@@ -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:
@@ -85,7 +85,7 @@ TEST_P(DBIteratorTest, IteratorProperty) {
   ReadOptions ropt;
   ropt.pin_data = false;
   {
-    unique_ptr<Iterator> iter(NewIterator(ropt, handles_[1]));
+    std::unique_ptr<Iterator> iter(NewIterator(ropt, handles_[1]));
     iter->SeekToFirst();
     std::string prop_value;
     ASSERT_NOK(iter->GetProperty("non_existing.value", &prop_value));
@@ -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;
@@ -1889,6 +1951,7 @@ TEST_P(DBIteratorTest, ReadAhead) {
   size_t bytes_read = env_->random_read_bytes_counter_;
   delete iter;
 
+  int64_t num_file_closes = TestGetTickerCount(options, NO_FILE_CLOSES);
   env_->random_read_bytes_counter_ = 0;
   options.statistics->setTickerCount(NO_FILE_OPENS, 0);
   read_options.readahead_size = 1024 * 10;
@@ -1897,7 +1960,10 @@ TEST_P(DBIteratorTest, ReadAhead) {
   int64_t num_file_opens_readahead = TestGetTickerCount(options, NO_FILE_OPENS);
   size_t bytes_read_readahead = env_->random_read_bytes_counter_;
   delete iter;
+  int64_t num_file_closes_readahead =
+      TestGetTickerCount(options, NO_FILE_CLOSES);
   ASSERT_EQ(num_file_opens + 3, num_file_opens_readahead);
+  ASSERT_EQ(num_file_closes + 3, num_file_closes_readahead);
   ASSERT_GT(bytes_read_readahead, bytes_read);
   ASSERT_GT(bytes_read_readahead, read_options.readahead_size * 3);
 
@@ -2031,6 +2097,34 @@ 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 +2138,43 @@ 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 +2241,47 @@ 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();
@@ -2205,7 +2377,7 @@ TEST_P(DBIteratorTest, SeekAfterHittingManyInternalKeys) {
   Delete("5");
   Put("6", "val_6");
 
-  unique_ptr<Iterator> iter(NewIterator(ropts));
+  std::unique_ptr<Iterator> iter(NewIterator(ropts));
   iter->SeekToFirst();
 
   ASSERT_TRUE(iter->Valid());
@@ -2224,7 +2396,7 @@ TEST_P(DBIteratorTest, SeekAfterHittingManyInternalKeys) {
   ASSERT_EQ("4", prop_value);
 
   // Create a new iterator to seek to the internal key.
-  unique_ptr<Iterator> iter2(NewIterator(ropts));
+  std::unique_ptr<Iterator> iter2(NewIterator(ropts));
   iter2->Seek(prop_value);
   ASSERT_TRUE(iter2->Valid());
   ASSERT_OK(iter2->status());
@@ -2252,7 +2424,7 @@ TEST_P(DBIteratorTest, NonBlockingIterationBugRepro) {
   // Create a nonblocking iterator before writing to memtable.
   ReadOptions ropt;
   ropt.read_tier = kBlockCacheTier;
-  unique_ptr<Iterator> iter(NewIterator(ropt));
+  std::unique_ptr<Iterator> iter(NewIterator(ropt));
 
   // Overwrite a key in memtable many times to hit
   // max_sequential_skip_in_iterations (which is 8 by default).
@@ -2262,7 +2434,7 @@ TEST_P(DBIteratorTest, NonBlockingIterationBugRepro) {
 
   // Load the second block in sst file into the block cache.
   {
-    unique_ptr<Iterator> iter2(NewIterator(ReadOptions()));
+    std::unique_ptr<Iterator> iter2(NewIterator(ReadOptions()));
     iter2->Seek("d");
   }
 
@@ -2273,6 +2445,30 @@ 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 +2481,7 @@ 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_;
     }
 

db/db_log_iter_test.cc

@@ -23,7 +23,7 @@ class DBTestXactLogIterator : public DBTestBase {
 
   std::unique_ptr<TransactionLogIterator> OpenTransactionLogIter(
       const SequenceNumber seq) {
-    unique_ptr<TransactionLogIterator> iter;
+    std::unique_ptr<TransactionLogIterator> iter;
     Status status = dbfull()->GetUpdatesSince(seq, &iter);
     EXPECT_OK(status);
     EXPECT_TRUE(iter->Valid());

db/db_memtable_test.cc

@@ -8,6 +8,7 @@
 
 #include "db/db_test_util.h"
 #include "db/memtable.h"
+#include "db/range_del_aggregator.h"
 #include "port/stack_trace.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/slice_transform.h"
@@ -135,7 +136,8 @@ TEST_F(DBMemTableTest, DuplicateSeq) {
   MergeContext merge_context;
   Options options;
   InternalKeyComparator ikey_cmp(options.comparator);
-  RangeDelAggregator range_del_agg(ikey_cmp, {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&ikey_cmp,
+                                       kMaxSequenceNumber /* upper_bound */);
 
   // Create a MemTable
   InternalKeyComparator cmp(BytewiseComparator());

db/db_merge_operator_test.cc

@@ -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_);
   }
 

db/db_options_test.cc

@@ -514,6 +514,33 @@ TEST_F(DBOptionsTest, SetStatsDumpPeriodSec) {
   }
 }
 
+TEST_F(DBOptionsTest, RunStatsDumpPeriodSec) {
+  Options options;
+  options.create_if_missing = true;
+  options.stats_dump_period_sec = 5;
+  std::unique_ptr<rocksdb::MockTimeEnv> mock_env;
+  mock_env.reset(new rocksdb::MockTimeEnv(env_));
+  mock_env->set_current_time(0); // in seconds
+  options.env = mock_env.get();
+  int counter = 0;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::DumpStats:1", [&](void* /*arg*/) {
+        counter++;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+  Reopen(options);
+  ASSERT_EQ(5, dbfull()->GetDBOptions().stats_dump_period_sec);
+  dbfull()->TEST_WaitForTimedTaskRun([&] { mock_env->set_current_time(5); });
+  ASSERT_GE(counter, 1);
+
+  // Test cacel job through SetOptions
+  ASSERT_OK(dbfull()->SetDBOptions({{"stats_dump_period_sec", "0"}}));
+  int old_val = counter;
+  env_->SleepForMicroseconds(10000000);
+  ASSERT_EQ(counter, old_val);
+  Close();
+}
+
 static void assert_candidate_files_empty(DBImpl* dbfull, const bool empty) {
   dbfull->TEST_LockMutex();
   JobContext job_context(0);

db/db_properties_test.cc

@@ -170,6 +170,9 @@ void ResetTableProperties(TableProperties* tp) {
   tp->raw_value_size = 0;
   tp->num_data_blocks = 0;
   tp->num_entries = 0;
+  tp->num_deletions = 0;
+  tp->num_merge_operands = 0;
+  tp->num_range_deletions = 0;
 }
 
 void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
@@ -177,16 +180,19 @@ 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 " # deletions %" SCNu64
+         " # merge operands %" 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,
+         " 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_deletions,
+         &tp->num_merge_operands, &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);
 }
 
@@ -215,30 +221,51 @@ void VerifyTableProperties(const TableProperties& base_tp,
   VerifySimilar(base_tp.filter_size, new_tp.filter_size, filter_size_bias);
   VerifySimilar(base_tp.num_data_blocks, new_tp.num_data_blocks,
                 num_data_blocks_bias);
+
   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_deletions, new_tp.num_deletions);
+  ASSERT_EQ(base_tp.num_range_deletions, new_tp.num_range_deletions);
+
+  // Merge operands may become Puts, so we only have an upper bound the exact
+  // number of merge operands.
+  ASSERT_GE(base_tp.num_merge_operands, new_tp.num_merge_operands);
 }
 
-void GetExpectedTableProperties(TableProperties* expected_tp,
-                                const int kKeySize, const int kValueSize,
-                                const int kKeysPerTable, const int kTableCount,
-                                const int kBloomBitsPerKey,
-                                const size_t kBlockSize) {
-  const int kKeyCount = kTableCount * kKeysPerTable;
+void GetExpectedTableProperties(
+    TableProperties* expected_tp, const int kKeySize, const int kValueSize,
+    const int kPutsPerTable, const int kDeletionsPerTable,
+    const int kMergeOperandsPerTable, 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 kKeysPerTable =
+      kPutsPerTable + kDeletionsPerTable + kMergeOperandsPerTable;
+  const int kPutCount = kTableCount * kPutsPerTable;
+  const int kDeletionCount = kTableCount * kDeletionsPerTable;
+  const int kMergeCount = kTableCount * kMergeOperandsPerTable;
+  const int kRangeDeletionCount = kTableCount * kRangeDeletionsPerTable;
+  const int kKeyCount = kPutCount + kDeletionCount + kMergeCount;
   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 =
+      (kPutCount + kMergeCount + kRangeDeletionCount) * kValueSize;
   expected_tp->num_entries = kKeyCount;
+  expected_tp->num_deletions = kDeletionCount;
+  expected_tp->num_merge_operands = kMergeCount;
+  expected_tp->num_range_deletions = kRangeDeletionCount;
   expected_tp->num_data_blocks =
-      kTableCount *
-      (kKeysPerTable * (kKeySize - kEncodingSavePerKey + kValueSize)) /
+      kTableCount * (kKeysPerTable * (kKeySize - kEncodingSavePerKey + kValueSize)) /
       kBlockSize;
   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 +278,11 @@ 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);
   }
 }
 
@@ -287,7 +317,10 @@ TEST_F(DBPropertiesTest, ValidateSampleNumber) {
 
 TEST_F(DBPropertiesTest, AggregatedTableProperties) {
   for (int kTableCount = 40; kTableCount <= 100; kTableCount += 30) {
-    const int kKeysPerTable = 100;
+    const int kDeletionsPerTable = 5;
+    const int kMergeOperandsPerTable = 15;
+    const int kRangeDeletionsPerTable = 5;
+    const int kPutsPerTable = 100;
     const int kKeySize = 80;
     const int kValueSize = 200;
     const int kBloomBitsPerKey = 20;
@@ -296,6 +329,8 @@ TEST_F(DBPropertiesTest, AggregatedTableProperties) {
     options.level0_file_num_compaction_trigger = 8;
     options.compression = kNoCompression;
     options.create_if_missing = true;
+    options.preserve_deletes = true;
+    options.merge_operator.reset(new TestPutOperator());
 
     BlockBasedTableOptions table_options;
     table_options.filter_policy.reset(
@@ -305,24 +340,44 @@ 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) {
+      for (int i = 0; i < kPutsPerTable; ++i) {
         db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
                  RandomString(&rnd, kValueSize));
       }
+      for (int i = 0; i < kDeletionsPerTable; i++) {
+        db_->Delete(WriteOptions(), RandomString(&rnd, kKeySize));
+      }
+      for (int i = 0; i < kMergeOperandsPerTable; i++) {
+        db_->Merge(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 expected_tp;
-    GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                               kKeysPerTable, kTableCount, kBloomBitsPerKey,
-                               table_options.block_size);
-
     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, kPutsPerTable, kDeletionsPerTable,
+        kMergeOperandsPerTable, kRangeDeletionsPerTable, kTableCount,
+        kBloomBitsPerKey, table_options.block_size, index_key_is_user_key,
+        value_is_delta_encoded);
 
     VerifyTableProperties(expected_tp, output_tp);
   }
@@ -371,6 +426,13 @@ 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"));
@@ -384,7 +446,7 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
   ASSERT_EQ(std::string::npos, prop.find("** Level 1 read latency histogram"));
   ASSERT_EQ(std::string::npos, prop.find("** Level 2 read latency histogram"));
   {
-    unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+    std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
     for (iter->Seek(Key(0)); iter->Valid(); iter->Next()) {
     }
   }
@@ -437,7 +499,10 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
 
 TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
   const int kTableCount = 100;
-  const int kKeysPerTable = 10;
+  const int kDeletionsPerTable = 2;
+  const int kMergeOperandsPerTable = 2;
+  const int kRangeDeletionsPerTable = 2;
+  const int kPutsPerTable = 10;
   const int kKeySize = 50;
   const int kValueSize = 400;
   const int kMaxLevel = 7;
@@ -453,6 +518,8 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
   options.max_bytes_for_level_multiplier = 2;
   // This ensures there no compaction happening when we call GetProperty().
   options.disable_auto_compactions = true;
+  options.preserve_deletes = true;
+  options.merge_operator.reset(new TestPutOperator());
 
   BlockBasedTableOptions table_options;
   table_options.filter_policy.reset(
@@ -462,15 +529,31 @@ 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];
   TableProperties tp, sum_tp, expected_tp;
   for (int table = 1; table <= kTableCount; ++table) {
-    for (int i = 0; i < kKeysPerTable; ++i) {
+    for (int i = 0; i < kPutsPerTable; ++i) {
       db_->Put(WriteOptions(), RandomString(&rnd, kKeySize),
                RandomString(&rnd, kValueSize));
     }
+    for (int i = 0; i < kDeletionsPerTable; i++) {
+      db_->Delete(WriteOptions(), RandomString(&rnd, kKeySize));
+    }
+    for (int i = 0; i < kMergeOperandsPerTable; i++) {
+      db_->Merge(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 +569,14 @@ 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_deletions += level_tps[level].num_deletions;
+      sum_tp.num_merge_operands += level_tps[level].num_merge_operands;
+      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 +584,18 @@ 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_deletions, tp.num_deletions);
+    ASSERT_EQ(sum_tp.num_merge_operands, tp.num_merge_operands);
+    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);
+      GetExpectedTableProperties(
+          &expected_tp, kKeySize, kValueSize, kPutsPerTable, kDeletionsPerTable,
+          kMergeOperandsPerTable, 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);
     }
   }
 }
@@ -1436,6 +1529,64 @@ TEST_F(DBPropertiesTest, SstFilesSize) {
   ASSERT_TRUE(listener->callback_triggered);
 }
 
+TEST_F(DBPropertiesTest, MinObsoleteSstNumberToKeep) {
+  class TestListener : public EventListener {
+   public:
+    void OnTableFileCreated(const TableFileCreationInfo& info) override {
+      if (info.reason == TableFileCreationReason::kCompaction) {
+        // Verify the property indicates that SSTs created by a running
+        // compaction cannot be deleted.
+        uint64_t created_file_num;
+        FileType created_file_type;
+        std::string filename =
+            info.file_path.substr(info.file_path.rfind('/') + 1);
+        ASSERT_TRUE(
+            ParseFileName(filename, &created_file_num, &created_file_type));
+        ASSERT_EQ(kTableFile, created_file_type);
+
+        uint64_t keep_sst_lower_bound;
+        ASSERT_TRUE(
+            db_->GetIntProperty(DB::Properties::kMinObsoleteSstNumberToKeep,
+                                &keep_sst_lower_bound));
+
+        ASSERT_LE(keep_sst_lower_bound, created_file_num);
+        validated_ = true;
+      }
+    }
+
+    void SetDB(DB* db) { db_ = db; }
+
+    int GetNumCompactions() { return num_compactions_; }
+
+    // True if we've verified the property for at least one output file
+    bool Validated() { return validated_; }
+
+   private:
+    int num_compactions_ = 0;
+    bool validated_ = false;
+    DB* db_ = nullptr;
+  };
+
+  const int kNumL0Files = 4;
+
+  std::shared_ptr<TestListener> listener = std::make_shared<TestListener>();
+
+  Options options = CurrentOptions();
+  options.listeners.push_back(listener);
+  options.level0_file_num_compaction_trigger = kNumL0Files;
+  DestroyAndReopen(options);
+  listener->SetDB(db_);
+
+  for (int i = 0; i < kNumL0Files; ++i) {
+    // Make sure they overlap in keyspace to prevent trivial move
+    Put("key1", "val");
+    Put("key2", "val");
+    Flush();
+  }
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_TRUE(listener->Validated());
+}
+
 TEST_F(DBPropertiesTest, BlockCacheProperties) {
   Options options;
   uint64_t value;

db/db_range_del_test.cc

@@ -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 +964,116 @@ 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]
+    //
+    // At the same time, verify the compaction does not cause the key at the
+    // endpoint (key000002#6,1) to disappear.
+    ASSERT_EQ(value, Get(Key(2)));
+    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));
+    ASSERT_EQ(value, Get(Key(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
@@ -996,6 +1108,386 @@ TEST_F(DBRangeDelTest, UnorderedTombstones) {
   ASSERT_TRUE(s.IsNotFound());
 }
 
+class MockMergeOperator : public MergeOperator {
+  // Mock non-associative operator. Non-associativity is expressed by lack of
+  // implementation for any `PartialMerge*` functions.
+ public:
+  virtual bool FullMergeV2(const MergeOperationInput& merge_in,
+                           MergeOperationOutput* merge_out) const override {
+    assert(merge_out != nullptr);
+    merge_out->new_value = merge_in.operand_list.back().ToString();
+    return true;
+  }
+
+  virtual const char* Name() const override { return "MockMergeOperator"; }
+};
+
+TEST_F(DBRangeDelTest, KeyAtOverlappingEndpointReappears) {
+  // This test uses a non-associative merge operator since that is a convenient
+  // way to get compaction to write out files with overlapping user-keys at the
+  // endpoints. Note, however, overlapping endpoints can also occur with other
+  // value types (Put, etc.), assuming the right snapshots are present.
+  const int kFileBytes = 1 << 20;
+  const int kValueBytes = 1 << 10;
+  const int kNumFiles = 4;
+
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.disable_auto_compactions = true;
+  options.merge_operator.reset(new MockMergeOperator());
+  options.target_file_size_base = kFileBytes;
+  Reopen(options);
+
+  Random rnd(301);
+  const Snapshot* snapshot = nullptr;
+  for (int i = 0; i < kNumFiles; ++i) {
+    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
+      auto value = RandomString(&rnd, kValueBytes);
+      ASSERT_OK(db_->Merge(WriteOptions(), "key", value));
+    }
+    if (i == kNumFiles - 1) {
+      // Take snapshot to prevent covered merge operands from being dropped by
+      // compaction.
+      snapshot = db_->GetSnapshot();
+      // The DeleteRange is the last write so all merge operands are covered.
+      ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                                 "key", "key_"));
+    }
+    ASSERT_OK(db_->Flush(FlushOptions()));
+  }
+  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
+  std::string value;
+  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
+                              nullptr /* end_key */));
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  // Now we have multiple files at L1 all containing a single user key, thus
+  // guaranteeing overlap in the file endpoints.
+  ASSERT_GT(NumTableFilesAtLevel(1), 1);
+
+  // Verify no merge operands reappeared after the compaction.
+  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());
+
+  // Compact and verify again. It's worthwhile because now the files have
+  // tighter endpoints, so we can verify that doesn't mess anything up.
+  dbfull()->TEST_CompactRange(1 /* level */, nullptr /* begin */,
+                              nullptr /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
+  ASSERT_GT(NumTableFilesAtLevel(2), 1);
+  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, UntruncatedTombstoneDoesNotDeleteNewerKey) {
+  // Verify a key newer than a range tombstone cannot be deleted by being
+  // compacted to the bottom level (and thus having its seqnum zeroed) before
+  // the range tombstone. This used to happen when range tombstones were
+  // untruncated on reads such that they extended past their file boundaries.
+  //
+  // Test summary:
+  //
+  // - L1 is bottommost.
+  // - A couple snapshots are strategically taken to prevent seqnums from being
+  //   zeroed, range tombstone from being dropped, merge operands from being
+  //   dropped, and merge operands from being combined.
+  // - Left half of files in L1 all have same user key, ensuring their file
+  //   boundaries overlap. In the past this would cause range tombstones to be
+  //   untruncated.
+  // - Right half of L1 files all have different keys, ensuring no overlap.
+  // - A range tombstone spans all L1 keys, so it is stored in every L1 file.
+  // - Keys in the right side of the key-range are overwritten. These are
+  //   compacted down to L1 after releasing snapshots such that their seqnums
+  //   will be zeroed.
+  // - A full range scan is performed. If the tombstone in the left L1 files
+  //   were untruncated, it would now cover keys newer than it (but with zeroed
+  //   seqnums) in the right L1 files.
+  const int kFileBytes = 1 << 20;
+  const int kValueBytes = 1 << 10;
+  const int kNumFiles = 4;
+  const int kMaxKey = kNumFiles* kFileBytes / kValueBytes;
+  const int kKeysOverwritten = 10;
+
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.disable_auto_compactions = true;
+  options.merge_operator.reset(new MockMergeOperator());
+  options.num_levels = 2;
+  options.target_file_size_base = kFileBytes;
+  Reopen(options);
+
+  Random rnd(301);
+  // - snapshots[0] prevents merge operands from being combined during
+  //   compaction.
+  // - snapshots[1] prevents merge operands from being dropped due to the
+  //   covering range tombstone.
+  const Snapshot* snapshots[] = {nullptr, nullptr};
+  for (int i = 0; i < kNumFiles; ++i) {
+    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
+      auto value = RandomString(&rnd, kValueBytes);
+      std::string key;
+      if (i < kNumFiles / 2) {
+        key = Key(0);
+      } else {
+        key = Key(1 + i * kFileBytes / kValueBytes + j);
+      }
+      ASSERT_OK(db_->Merge(WriteOptions(), key, value));
+    }
+    if (i == 0) {
+      snapshots[0] = db_->GetSnapshot();
+    }
+    if (i == kNumFiles - 1) {
+      snapshots[1] = db_->GetSnapshot();
+      // The DeleteRange is the last write so all merge operands are covered.
+      ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                                 Key(0), Key(kMaxKey + 1)));
+    }
+    ASSERT_OK(db_->Flush(FlushOptions()));
+  }
+  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
+
+  auto get_key_count = [this]() -> int {
+    auto* iter = db_->NewIterator(ReadOptions());
+    iter->SeekToFirst();
+    int keys_found = 0;
+    for (; iter->Valid(); iter->Next()) {
+      ++keys_found;
+    }
+    delete iter;
+    return keys_found;
+  };
+
+  // All keys should be covered
+  ASSERT_EQ(0, get_key_count());
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
+                              nullptr /* end_key */));
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  // Roughly the left half of L1 files should have overlapping boundary keys,
+  // while the right half should not.
+  ASSERT_GE(NumTableFilesAtLevel(1), kNumFiles);
+
+  // Now overwrite a few keys that are in L1 files that definitely don't have
+  // overlapping boundary keys.
+  for (int i = kMaxKey; i > kMaxKey - kKeysOverwritten; --i) {
+    auto value = RandomString(&rnd, kValueBytes);
+    ASSERT_OK(db_->Merge(WriteOptions(), Key(i), value));
+  }
+  ASSERT_OK(db_->Flush(FlushOptions()));
+
+  // The overwritten keys are in L0 now, so clearly aren't covered by the range
+  // tombstone in L1.
+  ASSERT_EQ(kKeysOverwritten, get_key_count());
+
+  // Release snapshots so seqnums can be zeroed when L0->L1 happens.
+  db_->ReleaseSnapshot(snapshots[0]);
+  db_->ReleaseSnapshot(snapshots[1]);
+
+  auto begin_key_storage = Key(kMaxKey - kKeysOverwritten + 1);
+  auto end_key_storage = Key(kMaxKey);
+  Slice begin_key(begin_key_storage);
+  Slice end_key(end_key_storage);
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &begin_key, &end_key));
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GE(NumTableFilesAtLevel(1), kNumFiles);
+
+  ASSERT_EQ(kKeysOverwritten, get_key_count());
+}
+
+TEST_F(DBRangeDelTest, DeletedMergeOperandReappearsIterPrev) {
+  // Exposes a bug where we were using
+  // `RangeDelPositioningMode::kBackwardTraversal` while scanning merge operands
+  // in the forward direction. Confusingly, this case happened during
+  // `DBIter::Prev`. It could cause assertion failure, or reappearing keys.
+  const int kFileBytes = 1 << 20;
+  const int kValueBytes = 1 << 10;
+  // Need multiple keys so we can get results when calling `Prev()` after
+  // `SeekToLast()`.
+  const int kNumKeys = 3;
+  const int kNumFiles = 4;
+
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.disable_auto_compactions = true;
+  options.merge_operator.reset(new MockMergeOperator());
+  options.target_file_size_base = kFileBytes;
+  Reopen(options);
+
+  Random rnd(301);
+  const Snapshot* snapshot = nullptr;
+  for (int i = 0; i < kNumFiles; ++i) {
+    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
+      auto value = RandomString(&rnd, kValueBytes);
+      ASSERT_OK(db_->Merge(WriteOptions(), Key(j % kNumKeys), value));
+      if (i == 0 && j == kNumKeys) {
+        // Take snapshot to prevent covered merge operands from being dropped or
+        // merged by compaction.
+        snapshot = db_->GetSnapshot();
+        // Do a DeleteRange near the beginning so only the oldest merge operand
+        // for each key is covered. This ensures the sequence of events:
+        //
+        // - `DBIter::Prev()` is called
+        // - After several same versions of the same user key are encountered,
+        //   it decides to seek using `DBIter::FindValueForCurrentKeyUsingSeek`.
+        // - Binary searches to the newest version of the key, which is in the
+        //   leftmost file containing the user key.
+        // - Scans forwards to collect all merge operands. Eventually reaches
+        //   the rightmost file containing the oldest merge operand, which
+        //   should be covered by the `DeleteRange`. If `RangeDelAggregator`
+        //   were not properly using `kForwardTraversal` here, that operand
+        //   would reappear.
+        ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                                   Key(0), Key(kNumKeys + 1)));
+      }
+    }
+    ASSERT_OK(db_->Flush(FlushOptions()));
+  }
+  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
+                              nullptr /* end_key */));
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_GT(NumTableFilesAtLevel(1), 1);
+
+  auto* iter = db_->NewIterator(ReadOptions());
+  iter->SeekToLast();
+  int keys_found = 0;
+  for (; iter->Valid(); iter->Prev()) {
+    ++keys_found;
+  }
+  delete iter;
+  ASSERT_EQ(kNumKeys, keys_found);
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, SnapshotPreventsDroppedKeys) {
+  const int kFileBytes = 1 << 20;
+
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.disable_auto_compactions = true;
+  options.target_file_size_base = kFileBytes;
+  Reopen(options);
+
+  ASSERT_OK(Put(Key(0), "a"));
+  const Snapshot* snapshot = db_->GetSnapshot();
+
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
+                             Key(10)));
+
+  db_->Flush(FlushOptions());
+
+  ReadOptions read_opts;
+  read_opts.snapshot = snapshot;
+  auto* iter = db_->NewIterator(read_opts);
+
+  iter->SeekToFirst();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(Key(0), iter->key());
+
+  iter->Next();
+  ASSERT_FALSE(iter->Valid());
+
+  delete iter;
+  db_->ReleaseSnapshot(snapshot);
+}
+
+TEST_F(DBRangeDelTest, RangeTombstoneWrittenToMinimalSsts) {
+  // Adapted from
+  // https://github.com/cockroachdb/cockroach/blob/de8b3ea603dd1592d9dc26443c2cc92c356fbc2f/pkg/storage/engine/rocksdb_test.go#L1267-L1398.
+  // Regression test for issue where range tombstone was written to more files
+  // than necessary when it began exactly at the begin key in the next
+  // compaction output file.
+  const int kFileBytes = 1 << 20;
+  const int kValueBytes = 4 << 10;
+  Options options = CurrentOptions();
+  options.compression = kNoCompression;
+  options.disable_auto_compactions = true;
+  // Have a bit of slack in the size limits but we enforce them more strictly
+  // when manually flushing/compacting.
+  options.max_compaction_bytes = 2 * kFileBytes;
+  options.target_file_size_base = 2 * kFileBytes;
+  options.write_buffer_size = 2 * kFileBytes;
+  Reopen(options);
+
+  Random rnd(301);
+  for (char first_char : {'a', 'b', 'c'}) {
+    for (int i = 0; i < kFileBytes / kValueBytes; ++i) {
+      std::string key(1, first_char);
+      key.append(Key(i));
+      std::string value = RandomString(&rnd, kValueBytes);
+      ASSERT_OK(Put(key, value));
+    }
+    db_->Flush(FlushOptions());
+    MoveFilesToLevel(2);
+  }
+  ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  ASSERT_EQ(3, NumTableFilesAtLevel(2));
+
+  // Populate the memtable lightly while spanning the whole key-space. The
+  // setting of `max_compaction_bytes` will cause the L0->L1 to output multiple
+  // files to prevent a large L1->L2 compaction later.
+  ASSERT_OK(Put("a", "val"));
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                             "c" + Key(1), "d"));
+  // Our compaction output file cutting logic currently only considers point
+  // keys. So, in order for the range tombstone to have a chance at landing at
+  // the start of a new file, we need a point key at the range tombstone's
+  // start.
+  // TODO(ajkr): remove this `Put` after file cutting accounts for range
+  // tombstones (#3977).
+  ASSERT_OK(Put("c" + Key(1), "value"));
+  db_->Flush(FlushOptions());
+
+  // Ensure manual L0->L1 compaction cuts the outputs before the range tombstone
+  // and the range tombstone is only placed in the second SST.
+  std::string begin_key_storage("c" + Key(1));
+  Slice begin_key(begin_key_storage);
+  std::string end_key_storage("d");
+  Slice end_key(end_key_storage);
+  dbfull()->TEST_CompactRange(0 /* level */, &begin_key /* begin */,
+                              &end_key /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
+  ASSERT_EQ(2, NumTableFilesAtLevel(1));
+
+  std::vector<LiveFileMetaData> all_metadata;
+  std::vector<LiveFileMetaData> l1_metadata;
+  db_->GetLiveFilesMetaData(&all_metadata);
+  for (const auto& metadata : all_metadata) {
+    if (metadata.level == 1) {
+      l1_metadata.push_back(metadata);
+    }
+  }
+  std::sort(l1_metadata.begin(), l1_metadata.end(),
+            [&](const LiveFileMetaData& a, const LiveFileMetaData& b) {
+              return options.comparator->Compare(a.smallestkey, b.smallestkey) <
+                     0;
+            });
+  ASSERT_EQ("a", l1_metadata[0].smallestkey);
+  ASSERT_EQ("a", l1_metadata[0].largestkey);
+  ASSERT_EQ("c" + Key(1), l1_metadata[1].smallestkey);
+  ASSERT_EQ("d", l1_metadata[1].largestkey);
+
+  TablePropertiesCollection all_table_props;
+  ASSERT_OK(db_->GetPropertiesOfAllTables(&all_table_props));
+  int64_t num_range_deletions = 0;
+  for (const auto& name_and_table_props : all_table_props) {
+    const auto& name = name_and_table_props.first;
+    const auto& table_props = name_and_table_props.second;
+    // The range tombstone should only be output to the second L1 SST.
+    if (name.size() >= l1_metadata[1].name.size() &&
+        name.substr(name.size() - l1_metadata[1].name.size()).compare(l1_metadata[1].name) == 0) {
+      ASSERT_EQ(1, table_props->num_range_deletions);
+      ++num_range_deletions;
+    } else {
+      ASSERT_EQ(0, table_props->num_range_deletions);
+    }
+  }
+  ASSERT_EQ(1, num_range_deletions);
+}
+
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

db/db_sst_test.cc

@@ -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();
   }

db/db_table_properties_test.cc

@@ -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

db/db_tailing_iter_test.cc

@@ -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;

db/db_test.cc

@@ -93,7 +93,7 @@ class DBTestWithParam
 };
 
 TEST_F(DBTest, MockEnvTest) {
-  unique_ptr<MockEnv> env{new MockEnv(Env::Default())};
+  std::unique_ptr<MockEnv> env{new MockEnv(Env::Default())};
   Options options;
   options.create_if_missing = true;
   options.env = env.get();
@@ -143,7 +143,7 @@ TEST_F(DBTest, MockEnvTest) {
 // defined.
 #ifndef ROCKSDB_LITE
 TEST_F(DBTest, MemEnvTest) {
-  unique_ptr<Env> env{NewMemEnv(Env::Default())};
+  std::unique_ptr<Env> env{NewMemEnv(Env::Default())};
   Options options;
   options.create_if_missing = true;
   options.env = env.get();
@@ -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));
@@ -2123,6 +2313,9 @@ INSTANTIATE_TEST_CASE_P(
 #endif  // ROCKSDB_LITE
 
 // Group commit test:
+#if !defined(TRAVIS) && !defined(OS_WIN)
+// Disable this test temporarily on Travis and appveyor as it fails
+// intermittently. Github issue: #4151
 namespace {
 
 static const int kGCNumThreads = 4;
@@ -2158,7 +2351,9 @@ TEST_F(DBTest, GroupCommitTest) {
 
     rocksdb::SyncPoint::GetInstance()->LoadDependency(
         {{"WriteThread::JoinBatchGroup:BeganWaiting",
-          "DBImpl::WriteImpl:BeforeLeaderEnters"}});
+          "DBImpl::WriteImpl:BeforeLeaderEnters"},
+          {"WriteThread::AwaitState:BlockingWaiting",
+          "WriteThread::EnterAsBatchGroupLeader:End"}});
     rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
     // Start threads
@@ -2195,6 +2390,7 @@ 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;
@@ -2465,6 +2661,11 @@ class ModelDB : public DB {
     Status ret;
     return ret;
   }
+  virtual Status Flush(
+      const rocksdb::FlushOptions& /*options*/,
+      const std::vector<ColumnFamilyHandle*>& /*column_families*/) override {
+    return Status::OK();
+  }
 
   virtual Status SyncWAL() override { return Status::OK(); }
 
@@ -2488,7 +2689,8 @@ class ModelDB : public DB {
   }
 
   virtual Status GetUpdatesSince(
-      rocksdb::SequenceNumber, unique_ptr<rocksdb::TransactionLogIterator>*,
+      rocksdb::SequenceNumber,
+      std::unique_ptr<rocksdb::TransactionLogIterator>*,
       const TransactionLogIterator::ReadOptions& /*read_options*/ =
           TransactionLogIterator::ReadOptions()) override {
     return Status::NotSupported("Not supported in Model DB");
@@ -2561,6 +2763,7 @@ 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 +2920,7 @@ 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
@@ -3308,8 +3512,14 @@ TEST_F(DBTest, SanitizeNumThreads) {
                      (i < 4) ? Env::Priority::LOW : Env::Priority::HIGH);
     }
 
-    // Wait 100 milliseconds for they are scheduled.
-    env_->SleepForMicroseconds(100000);
+    // Wait until 10s for they are scheduled.
+    for (int i = 0; i < 10000; i++) {
+      if (options.env->GetThreadPoolQueueLen(Env::Priority::LOW) <= 1 &&
+          options.env->GetThreadPoolQueueLen(Env::Priority::HIGH) <= 2) {
+        break;
+      }
+      env_->SleepForMicroseconds(1000);
+    }
 
     // pool size 3, total task 4. Queue size should be 1.
     ASSERT_EQ(1U, options.env->GetThreadPoolQueueLen(Env::Priority::LOW));
@@ -4325,7 +4535,7 @@ 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 +4548,7 @@ 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 +4576,7 @@ 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 +4815,6 @@ 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 +5133,7 @@ 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);
 
@@ -5618,41 +5655,18 @@ TEST_F(DBTest, HardLimit) {
 #if !defined(ROCKSDB_LITE) && !defined(ROCKSDB_DISABLE_STALL_NOTIFICATION)
 class WriteStallListener : public EventListener {
  public:
-  WriteStallListener()
-      : cond_(&mutex_),
-        condition_(WriteStallCondition::kNormal),
-        expected_(WriteStallCondition::kNormal),
-        expected_set_(false) {}
+  WriteStallListener() : condition_(WriteStallCondition::kNormal) {}
   void OnStallConditionsChanged(const WriteStallInfo& info) override {
     MutexLock l(&mutex_);
     condition_ = info.condition.cur;
-    if (expected_set_ && condition_ == expected_) {
-      cond_.Signal();
-      expected_set_ = false;
-    }
   }
   bool CheckCondition(WriteStallCondition expected) {
     MutexLock l(&mutex_);
-    if (expected != condition_) {
-      expected_ = expected;
-      expected_set_ = true;
-      while (expected != condition_) {
-        // We bail out on timeout 500 milliseconds
-        const uint64_t timeout_us = 500000;
-        if (cond_.TimedWait(timeout_us)) {
-          expected_set_ = false;
-          return false;
-        }
-      }
-    }
-    return true;
+    return expected == condition_;
   }
  private:
   port::Mutex   mutex_;
-  port::CondVar cond_;
   WriteStallCondition condition_;
-  WriteStallCondition expected_;
-  bool                expected_set_;
 };
 
 TEST_F(DBTest, SoftLimit) {
@@ -5673,13 +5687,48 @@ TEST_F(DBTest, SoftLimit) {
   WriteStallListener* listener = new WriteStallListener();
   options.listeners.emplace_back(listener);
 
+  // FlushMemtable with opt.wait=true does not wait for
+  // `OnStallConditionsChanged` being called. The event listener is triggered
+  // on `JobContext::Clean`, which happens after flush result is installed.
+  // We use sync point to create a custom WaitForFlush that waits for
+  // context cleanup.
+  port::Mutex flush_mutex;
+  port::CondVar flush_cv(&flush_mutex);
+  bool flush_finished = false;
+  auto InstallFlushCallback = [&]() {
+    {
+      MutexLock l(&flush_mutex);
+      flush_finished = false;
+    }
+    SyncPoint::GetInstance()->SetCallBack(
+        "DBImpl::BackgroundCallFlush:ContextCleanedUp", [&](void*) {
+          {
+            MutexLock l(&flush_mutex);
+            flush_finished = true;
+          }
+          flush_cv.SignalAll();
+        });
+  };
+  auto WaitForFlush = [&]() {
+    {
+      MutexLock l(&flush_mutex);
+      while (!flush_finished) {
+        flush_cv.Wait();
+      }
+    }
+    SyncPoint::GetInstance()->ClearCallBack(
+        "DBImpl::BackgroundCallFlush:ContextCleanedUp");
+  };
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
   Reopen(options);
 
   // Generating 360KB in Level 3
   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 +5738,7 @@ 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 +5757,9 @@ 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();
+    InstallFlushCallback();
+    dbfull()->TEST_FlushMemTable(true, true);
+    WaitForFlush();
   }
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_TRUE(listener->CheckCondition(WriteStallCondition::kDelayed));
@@ -5733,8 +5784,6 @@ TEST_F(DBTest, SoftLimit) {
                        &sleeping_task_low, Env::Priority::LOW);
       });
 
-  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
-
   env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
                  Env::Priority::LOW);
   sleeping_task_low.WaitUntilSleeping();
@@ -5743,7 +5792,9 @@ 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();
+    InstallFlushCallback();
+    dbfull()->TEST_FlushMemTable(true, true);
+    WaitForFlush();
   }
 
   // Wake up sleep task to enable compaction to run and waits
@@ -5764,7 +5815,9 @@ 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();
+    InstallFlushCallback();
+    dbfull()->TEST_FlushMemTable(true, true);
+    WaitForFlush();
   }
   // Wake up sleep task to enable compaction to run and waits
   // for it to go to sleep state again to make sure one compaction

db/db_test2.cc

@@ -61,7 +61,7 @@ TEST_P(PrefixFullBloomWithReverseComparator,
     bbto.block_cache->EraseUnRefEntries();
   }
 
-  unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+  std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
   iter->Seek("bar345");
   ASSERT_OK(iter->status());
   ASSERT_TRUE(iter->Valid());
@@ -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();
@@ -454,6 +454,22 @@ TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
 
+TEST_F(DBTest2, TestWriteBufferNoLimitWithCache) {
+  Options options = CurrentOptions();
+  options.arena_block_size = 4096;
+  std::shared_ptr<Cache> cache =
+      NewLRUCache(LRUCacheOptions(10000000, 1, false, 0.0));
+  options.write_buffer_size = 50000;  // this is never hit
+  // Use a write buffer total size so that the soft limit is about
+  // 105000.
+  options.write_buffer_manager.reset(new WriteBufferManager(0, cache));
+  Reopen(options);
+
+  ASSERT_OK(Put("foo", "bar"));
+  // One dummy entry is 1MB.
+  ASSERT_GT(cache->GetUsage(), 500000);
+}
+
 namespace {
   void ValidateKeyExistence(DB* db, const std::vector<Slice>& keys_must_exist,
     const std::vector<Slice>& keys_must_not_exist) {
@@ -1228,7 +1244,14 @@ TEST_F(DBTest2, CompressionOptions) {
 
 class CompactionStallTestListener : public EventListener {
  public:
-  CompactionStallTestListener() : compacted_files_cnt_(0) {}
+  CompactionStallTestListener() : compacting_files_cnt_(0), compacted_files_cnt_(0) {}
+
+  void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& ci) override {
+    ASSERT_EQ(ci.cf_name, "default");
+    ASSERT_EQ(ci.base_input_level, 0);
+    ASSERT_EQ(ci.compaction_reason, CompactionReason::kLevelL0FilesNum);
+    compacting_files_cnt_ += ci.input_files.size();
+  }
 
   void OnCompactionCompleted(DB* /*db*/, const CompactionJobInfo& ci) override {
     ASSERT_EQ(ci.cf_name, "default");
@@ -1236,6 +1259,8 @@ class CompactionStallTestListener : public EventListener {
     ASSERT_EQ(ci.compaction_reason, CompactionReason::kLevelL0FilesNum);
     compacted_files_cnt_ += ci.input_files.size();
   }
+
+  std::atomic<size_t> compacting_files_cnt_;
   std::atomic<size_t> compacted_files_cnt_;
 };
 
@@ -1244,6 +1269,8 @@ TEST_F(DBTest2, CompactionStall) {
       {{"DBImpl::BGWorkCompaction", "DBTest2::CompactionStall:0"},
        {"DBImpl::BGWorkCompaction", "DBTest2::CompactionStall:1"},
        {"DBTest2::CompactionStall:2",
+        "DBImpl::NotifyOnCompactionBegin::UnlockMutex"},
+       {"DBTest2::CompactionStall:3",
         "DBImpl::NotifyOnCompactionCompleted::UnlockMutex"}});
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
@@ -1285,14 +1312,18 @@ TEST_F(DBTest2, CompactionStall) {
   // Wait for another compaction to be triggered
   TEST_SYNC_POINT("DBTest2::CompactionStall:1");
 
-  // Hold NotifyOnCompactionCompleted in the unlock mutex section
+  // Hold NotifyOnCompactionBegin in the unlock mutex section
   TEST_SYNC_POINT("DBTest2::CompactionStall:2");
 
+  // Hold NotifyOnCompactionCompleted in the unlock mutex section
+  TEST_SYNC_POINT("DBTest2::CompactionStall:3");
+
   dbfull()->TEST_WaitForCompact();
   ASSERT_LT(NumTableFilesAtLevel(0),
             options.level0_file_num_compaction_trigger);
   ASSERT_GT(listener->compacted_files_cnt_.load(),
             10 - options.level0_file_num_compaction_trigger);
+  ASSERT_EQ(listener->compacting_files_cnt_.load(), listener->compacted_files_cnt_.load());
 
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
@@ -2321,9 +2352,9 @@ 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 +2453,7 @@ 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_;
     }
 
@@ -2500,6 +2531,561 @@ TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
 
+TEST_F(DBTest2, TestNumPread) {
+  Options options = CurrentOptions();
+  // disable block cache
+  BlockBasedTableOptions table_options;
+  table_options.no_block_cache = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  Reopen(options);
+  env_->count_random_reads_ = true;
+
+  env_->random_file_open_counter_.store(0);
+  ASSERT_OK(Put("bar", "foo"));
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+  // After flush, we'll open the file and read footer, meta block,
+  // property block and index block.
+  ASSERT_EQ(4, env_->random_read_counter_.Read());
+  ASSERT_EQ(1, env_->random_file_open_counter_.load());
+
+  // One pread per a normal data block read
+  env_->random_file_open_counter_.store(0);
+  env_->random_read_counter_.Reset();
+  ASSERT_EQ("bar", Get("foo"));
+  ASSERT_EQ(1, env_->random_read_counter_.Read());
+  // All files are already opened.
+  ASSERT_EQ(0, env_->random_file_open_counter_.load());
+
+  env_->random_file_open_counter_.store(0);
+  env_->random_read_counter_.Reset();
+  ASSERT_OK(Put("bar2", "foo2"));
+  ASSERT_OK(Put("foo2", "bar2"));
+  ASSERT_OK(Flush());
+  // After flush, we'll open the file and read footer, meta block,
+  // property block and index block.
+  ASSERT_EQ(4, env_->random_read_counter_.Read());
+  ASSERT_EQ(1, env_->random_file_open_counter_.load());
+
+  // Compaction needs two input blocks, which requires 2 preads, and
+  // generate a new SST file which needs 4 preads (footer, meta block,
+  // property block and index block). In total 6.
+  env_->random_file_open_counter_.store(0);
+  env_->random_read_counter_.Reset();
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_EQ(6, env_->random_read_counter_.Read());
+  // All compactin input files should have already been opened.
+  ASSERT_EQ(1, env_->random_file_open_counter_.load());
+
+  // One pread per a normal data block read
+  env_->random_file_open_counter_.store(0);
+  env_->random_read_counter_.Reset();
+  ASSERT_EQ("foo2", Get("bar2"));
+  ASSERT_EQ(1, env_->random_read_counter_.Read());
+  // SST files are already opened.
+  ASSERT_EQ(0, env_->random_file_open_counter_.load());
+}
+
+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));
+}
+
+TEST_F(DBTest2, TraceWithLimit) {
+  Options options = CurrentOptions();
+  options.merge_operator = MergeOperators::CreatePutOperator();
+  ReadOptions ro;
+  WriteOptions wo;
+  TraceOptions trace_opts;
+  EnvOptions env_opts;
+  CreateAndReopenWithCF({"pikachu"}, options);
+  Random rnd(301);
+
+  // test the max trace file size options
+  trace_opts.max_trace_file_size = 5;
+  std::string trace_filename = dbname_ + "/rocksdb.trace1";
+  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(Put(0, "b", "1"));
+  ASSERT_OK(Put(0, "c", "1"));
+  ASSERT_OK(db_->EndTrace());
+
+  std::string dbname2 = test::TmpDir(env_) + "/db_replay2";
+  std::string value;
+  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], "b", &value).IsNotFound());
+  ASSERT_TRUE(db2->Get(ro, handles[0], "c", &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_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
+  ASSERT_TRUE(db2->Get(ro, handles[0], "b", &value).IsNotFound());
+  ASSERT_TRUE(db2->Get(ro, handles[0], "c", &value).IsNotFound());
+
+  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.
+  {
+    std::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();
+}
+
+TEST_F(DBTest2, TestGetColumnFamilyHandleUnlocked) {
+  // Setup sync point dependency to reproduce the race condition of
+  // DBImpl::GetColumnFamilyHandleUnlocked
+  rocksdb::SyncPoint::GetInstance()->LoadDependency(
+      { {"TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked1",
+         "TestGetColumnFamilyHandleUnlocked::PreGetColumnFamilyHandleUnlocked2"},
+        {"TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked2",
+         "TestGetColumnFamilyHandleUnlocked::ReadColumnFamilyHandle1"},
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  CreateColumnFamilies({"test1", "test2"}, Options());
+  ASSERT_EQ(handles_.size(), 2);
+
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  port::Thread user_thread1([&]() {
+    auto cfh = dbi->GetColumnFamilyHandleUnlocked(handles_[0]->GetID());
+    ASSERT_EQ(cfh->GetID(), handles_[0]->GetID());
+    TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked1");
+    TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::ReadColumnFamilyHandle1");
+    ASSERT_EQ(cfh->GetID(), handles_[0]->GetID());
+  });
+
+  port::Thread user_thread2([&]() {
+    TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::PreGetColumnFamilyHandleUnlocked2");
+    auto cfh = dbi->GetColumnFamilyHandleUnlocked(handles_[1]->GetID());
+    ASSERT_EQ(cfh->GetID(), handles_[1]->GetID());
+    TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked2");
+    ASSERT_EQ(cfh->GetID(), handles_[1]->GetID());
+  });
+
+  user_thread1.join();
+  user_thread2.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBTest2, TestCompactFiles) {
+  // Setup sync point dependency to reproduce the race condition of
+  // DBImpl::GetColumnFamilyHandleUnlocked
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"TestCompactFiles::IngestExternalFile1",
+       "TestCompactFiles::IngestExternalFile2"},
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  Options options;
+  options.num_levels = 2;
+  options.disable_auto_compactions = true;
+  Reopen(options);
+  auto* handle = db_->DefaultColumnFamily();
+  ASSERT_EQ(db_->NumberLevels(handle), 2);
+
+  rocksdb::SstFileWriter sst_file_writer{rocksdb::EnvOptions(), options};
+  std::string external_file1 = dbname_ + "/test_compact_files1.sst_t";
+  std::string external_file2 = dbname_ + "/test_compact_files2.sst_t";
+  std::string external_file3 = dbname_ + "/test_compact_files3.sst_t";
+
+  ASSERT_OK(sst_file_writer.Open(external_file1));
+  ASSERT_OK(sst_file_writer.Put("1", "1"));
+  ASSERT_OK(sst_file_writer.Put("2", "2"));
+  ASSERT_OK(sst_file_writer.Finish());
+
+  ASSERT_OK(sst_file_writer.Open(external_file2));
+  ASSERT_OK(sst_file_writer.Put("3", "3"));
+  ASSERT_OK(sst_file_writer.Put("4", "4"));
+  ASSERT_OK(sst_file_writer.Finish());
+
+  ASSERT_OK(sst_file_writer.Open(external_file3));
+  ASSERT_OK(sst_file_writer.Put("5", "5"));
+  ASSERT_OK(sst_file_writer.Put("6", "6"));
+  ASSERT_OK(sst_file_writer.Finish());
+
+  ASSERT_OK(db_->IngestExternalFile(handle, {external_file1, external_file3},
+                                    IngestExternalFileOptions()));
+  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 2);
+  std::vector<std::string> files;
+  GetSstFiles(env_, dbname_, &files);
+  ASSERT_EQ(files.size(), 2);
+
+  port::Thread user_thread1(
+      [&]() { db_->CompactFiles(CompactionOptions(), handle, files, 1); });
+
+  port::Thread user_thread2([&]() {
+    ASSERT_OK(db_->IngestExternalFile(handle, {external_file2},
+                                      IngestExternalFileOptions()));
+    TEST_SYNC_POINT("TestCompactFiles::IngestExternalFile1");
+  });
+
+  user_thread1.join();
+  user_thread2.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
+}
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

db/db_test_util.cc

@@ -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) {
@@ -258,6 +259,47 @@ bool DBTestBase::ChangeFilterOptions() {
   return true;
 }
 
+// Switch between different DB options for file ingestion tests.
+bool DBTestBase::ChangeOptionsForFileIngestionTest() {
+  if (option_config_ == kDefault) {
+    option_config_ = kUniversalCompaction;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    options.create_if_missing = true;
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kUniversalCompaction) {
+    option_config_ = kUniversalCompactionMultiLevel;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    options.create_if_missing = true;
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kUniversalCompactionMultiLevel) {
+    option_config_ = kLevelSubcompactions;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    assert(options.max_subcompactions > 1);
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kLevelSubcompactions) {
+    option_config_ = kUniversalSubcompactions;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    assert(options.max_subcompactions > 1);
+    TryReopen(options);
+    return true;
+  } else if (option_config_ == kUniversalSubcompactions) {
+    option_config_ = kDirectIO;
+    Destroy(last_options_);
+    auto options = CurrentOptions();
+    TryReopen(options);
+    return true;
+  } else {
+    return false;
+  }
+}
+
 // Return the current option configuration.
 Options DBTestBase::CurrentOptions(
     const anon::OptionsOverride& options_override) const {
@@ -346,6 +388,26 @@ 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();
@@ -409,6 +471,10 @@ Options DBTestBase::GetOptions(
       table_options.checksum = kxxHash;
       break;
     }
+    case kxxHash64Checksum: {
+      table_options.checksum = kxxHash64;
+      break;
+    }
     case kFIFOCompaction: {
       options.compaction_style = kCompactionStyleFIFO;
       break;
@@ -428,6 +494,18 @@ 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 +538,6 @@ 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;
@@ -595,30 +653,19 @@ Status DBTestBase::ReadOnlyReopen(const Options& options) {
 Status DBTestBase::TryReopen(const Options& options) {
   Close();
   last_options_.table_factory.reset();
-  // Note: operator= is an unsafe approach here since it destructs shared_ptr in
-  // the same order of their creation, in contrast to destructors which
-  // destructs them in the opposite order of creation. One particular problme is
-  // that the cache destructor might invoke callback functions that use Option
-  // members such as statistics. To work around this problem, we manually call
-  // destructor of table_facotry which eventually clears the block cache.
+  // Note: operator= is an unsafe approach here since it destructs
+  // std::shared_ptr in the same order of their creation, in contrast to
+  // destructors which destructs them in the opposite order of creation. One
+  // particular problme is that the cache destructor might invoke callback
+  // functions that use Option members such as statistics. To work around this
+  // problem, we manually call destructor of table_facotry which eventually
+  // clears the block cache.
   last_options_ = options;
   return DB::Open(options, dbname_, &db_);
 }
 
 bool DBTestBase::IsDirectIOSupported() {
-  EnvOptions env_options;
-  env_options.use_mmap_writes = false;
-  env_options.use_direct_writes = true;
-  std::string tmp = TempFileName(dbname_, 999);
-  Status s;
-  {
-    unique_ptr<WritableFile> file;
-    s = env_->NewWritableFile(tmp, &file, env_options);
-  }
-  if (s.ok()) {
-    s = env_->DeleteFile(tmp);
-  }
-  return s.ok();
+  return test::IsDirectIOSupported(env_, dbname_);
 }
 
 bool DBTestBase::IsMemoryMappedAccessSupported() const {
@@ -633,6 +680,13 @@ Status DBTestBase::Flush(int cf) {
   }
 }
 
+Status DBTestBase::Flush(const std::vector<int>& cf_ids) {
+  std::vector<ColumnFamilyHandle*> cfhs;
+  std::for_each(cf_ids.begin(), cf_ids.end(),
+                [&cfhs, this](int id) { cfhs.emplace_back(handles_[id]); });
+  return db_->Flush(FlushOptions(), cfhs);
+}
+
 Status DBTestBase::Put(const Slice& k, const Slice& v, WriteOptions wo) {
   if (kMergePut == option_config_) {
     return db_->Merge(wo, k, v);
@@ -760,13 +814,15 @@ std::string DBTestBase::AllEntriesFor(const Slice& user_key, int cf) {
   Arena arena;
   auto options = CurrentOptions();
   InternalKeyComparator icmp(options.comparator);
-  RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&icmp,
+                                       kMaxSequenceNumber /* upper_bound */);
   ScopedArenaIterator iter;
   if (cf == 0) {
-    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg,
+                                           kMaxSequenceNumber));
   } else {
-    iter.set(
-        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[cf]));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg,
+                                           kMaxSequenceNumber, handles_[cf]));
   }
   InternalKey target(user_key, kMaxSequenceNumber, kTypeValue);
   iter->Seek(target.Encode());
@@ -1171,15 +1227,17 @@ void DBTestBase::validateNumberOfEntries(int numValues, int cf) {
   Arena arena;
   auto options = CurrentOptions();
   InternalKeyComparator icmp(options.comparator);
-  RangeDelAggregator range_del_agg(icmp, {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&icmp,
+                                       kMaxSequenceNumber /* upper_bound */);
   // This should be defined after range_del_agg so that it destructs the
   // assigned iterator before it range_del_agg is already destructed.
   ScopedArenaIterator iter;
   if (cf != 0) {
-    iter.set(
-        dbfull()->NewInternalIterator(&arena, &range_del_agg, handles_[cf]));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg,
+                                           kMaxSequenceNumber, handles_[cf]));
   } else {
-    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg));
+    iter.set(dbfull()->NewInternalIterator(&arena, &range_del_agg,
+                                           kMaxSequenceNumber));
   }
   iter->SeekToFirst();
   ASSERT_EQ(iter->status().ok(), true);
@@ -1199,9 +1257,9 @@ void DBTestBase::validateNumberOfEntries(int numValues, int cf) {
 void DBTestBase::CopyFile(const std::string& source,
                           const std::string& destination, uint64_t size) {
   const EnvOptions soptions;
-  unique_ptr<SequentialFile> srcfile;
+  std::unique_ptr<SequentialFile> srcfile;
   ASSERT_OK(env_->NewSequentialFile(source, &srcfile, soptions));
-  unique_ptr<WritableFile> destfile;
+  std::unique_ptr<WritableFile> destfile;
   ASSERT_OK(env_->NewWritableFile(destination, &destfile, soptions));
 
   if (size == 0) {
@@ -1379,8 +1437,10 @@ void DBTestBase::VerifyDBInternal(
     std::vector<std::pair<std::string, std::string>> true_data) {
   Arena arena;
   InternalKeyComparator icmp(last_options_.comparator);
-  RangeDelAggregator range_del_agg(icmp, {});
-  auto iter = dbfull()->NewInternalIterator(&arena, &range_del_agg);
+  ReadRangeDelAggregator range_del_agg(&icmp,
+                                       kMaxSequenceNumber /* upper_bound */);
+  auto iter =
+      dbfull()->NewInternalIterator(&arena, &range_del_agg, kMaxSequenceNumber);
   iter->SeekToFirst();
   for (auto p : true_data) {
     ASSERT_TRUE(iter->Valid());