Fix WriteBatchBase::DeleteRange API comment

The `DeleteRange` end key is exclusive, not inclusive. Updated API comment accordingly.

.travis.yml

@@ -90,7 +90,7 @@ script:
       OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=db_block_cache_test ROCKSDBTESTS_END=full_filter_block_test make -j4 check_some
       ;;
     2)
-      OPT=-DTRAVIS V=1 ROCKSDBTESTS_START=full_filter_block_test ROCKSDBTESTS_END=write_batch_with_index_test make -j4 check_some
+      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

CMakeLists.txt

@@ -49,6 +49,10 @@ 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
@@ -90,7 +94,13 @@ else()
   if(WITH_ZLIB)
     find_package(zlib REQUIRED)
     add_definitions(-DZLIB)
-    include_directories(${ZLIB_INCLUDE_DIR})
+    if(ZLIB_INCLUDE_DIRS)
+      # CMake 3
+      include_directories(${ZLIB_INCLUDE_DIRS})
+    else()
+      # CMake 2
+      include_directories(${ZLIB_INCLUDE_DIR})
+    endif()
     list(APPEND THIRDPARTY_LIBS ${ZLIB_LIBRARIES})
   endif()
 
@@ -336,12 +346,18 @@ 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(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}")
 
@@ -356,7 +372,7 @@ 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")
@@ -461,6 +477,7 @@ 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
@@ -493,6 +510,7 @@ set(SOURCES
         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
@@ -566,6 +584,7 @@ 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
@@ -583,6 +602,7 @@ set(SOURCES
         util/comparator.cc
         util/compression_context_cache.cc
         util/concurrent_arena.cc
+        util/concurrent_task_limiter_impl.cc
         util/crc32c.cc
         util/delete_scheduler.cc
         util/dynamic_bloom.cc
@@ -592,6 +612,7 @@ 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
@@ -622,18 +643,10 @@ set(SOURCES
         utilities/cassandra/format.cc
         utilities/cassandra/merge_operator.cc
         utilities/checkpoint/checkpoint_impl.cc
-        utilities/col_buf_decoder.cc
-        utilities/col_buf_encoder.cc
-        utilities/column_aware_encoding_util.cc
         utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc
-        utilities/date_tiered/date_tiered_db_impl.cc
         utilities/debug.cc
-        utilities/document/document_db.cc
-        utilities/document/json_document.cc
-        utilities/document/json_document_builder.cc
         utilities/env_mirror.cc
         utilities/env_timed.cc
-        utilities/geodb/geodb_impl.cc
         utilities/leveldb_options/leveldb_options.cc
         utilities/lua/rocks_lua_compaction_filter.cc
         utilities/memory/memory_util.cc
@@ -650,9 +663,7 @@ set(SOURCES
         utilities/persistent_cache/block_cache_tier_metadata.cc
         utilities/persistent_cache/persistent_cache_tier.cc
         utilities/persistent_cache/volatile_tier_impl.cc
-        utilities/redis/redis_lists.cc
         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
@@ -673,12 +684,10 @@ set(SOURCES
         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)
@@ -810,6 +819,7 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
     TARGETS ${ROCKSDB_SHARED_LIB}
     EXPORT RocksDBTargets
     COMPONENT runtime
+    ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}"
     RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}"
     LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
     INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
@@ -893,6 +903,8 @@ 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
@@ -921,6 +933,7 @@ if(WITH_TESTS)
         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
@@ -940,6 +953,7 @@ 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
@@ -951,11 +965,6 @@ if(WITH_TESTS)
         utilities/cassandra/cassandra_row_merge_test.cc
         utilities/cassandra/cassandra_serialize_test.cc
         utilities/checkpoint/checkpoint_test.cc
-        utilities/column_aware_encoding_test.cc
-        utilities/date_tiered/date_tiered_test.cc
-        utilities/document/document_db_test.cc
-        utilities/document/json_document_test.cc
-        utilities/geodb/geodb_test.cc
         utilities/lua/rocks_lua_test.cc
         utilities/memory/memory_test.cc
         utilities/merge_operators/string_append/stringappend_test.cc
@@ -964,8 +973,6 @@ if(WITH_TESTS)
         utilities/options/options_util_test.cc
         utilities/persistent_cache/hash_table_test.cc
         utilities/persistent_cache/persistent_cache_test.cc
-        utilities/redis/redis_lists_test.cc
-        utilities/spatialdb/spatial_db_test.cc
         utilities/simulator_cache/sim_cache_test.cc
         utilities/table_properties_collectors/compact_on_deletion_collector_test.cc
         utilities/transactions/optimistic_transaction_test.cc
@@ -982,9 +989,9 @@ 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
     utilities/persistent_cache/hash_table_bench.cc)
   add_library(testharness OBJECT util/testharness.cc)
   foreach(sourcefile ${BENCHMARKS})

HISTORY.md

@@ -1,17 +1,75 @@
 # Rocksdb Change Log
+## Unreleased
+### New Features
+* Enabled checkpoint on readonly db (DBImplReadOnly).
+* Make DB ignore dropped column families while committing results of atomic flush.
+* RocksDB may choose to preopen some files even if options.max_open_files != -1. This may make DB open slightly longer.
+* For users of dictionary compression with ZSTD v0.7.0+, we now reuse the same digested dictionary when compressing each of an SST file's data blocks for faster compression speeds.
+* For all users of dictionary compression who set `cache_index_and_filter_blocks == true`, we now store dictionary data used for decompression in the block cache for better control over memory usage. For users of ZSTD v1.1.4+ who compile with -DZSTD_STATIC_LINKING_ONLY, this includes a digested dictionary, which is used to increase decompression speed.
+* Add support for block checksums verification for external SST files before ingestion.
+
+### Public API Change
+* CompactionPri = kMinOverlappingRatio also uses compensated file size, which boosts file with lots of tombstones to be compacted first.
+* Transaction::GetForUpdate is extended with a do_validate parameter with default value of true. If false it skips validating the snapshot before doing the read. Similarly ::Merge, ::Put, ::Delete, and ::SingleDelete are extended with assume_tracked with default value of false. If true it indicates that call is assumed to be after a ::GetForUpdate.
+* `TableProperties::num_entries` and `TableProperties::num_deletions` now also account for number of range tombstones.
+* Remove geodb, spatial_db, document_db, json_document, date_tiered_db, and redis_lists.
+* With "ldb ----try_load_options", when wal_dir specified by the option file doesn't exist, ignore it.
+* Change time resolution in FileOperationInfo.
+* Deleting Blob files also go through SStFileManager.
+* Remove PlainTable's store_index_in_file feature. When opening an existing DB with index in SST files, the index and bloom filter will still be rebuild while SST files are opened, in the same way as there is no index in the file.
 
-# 5.16.6 (11/24/2018)
 ### Bug Fixes
-* Fix the bug that WriteBatchWithIndex's SeekForPrev() doesn't see the entries with the same key.
+* Fix a deadlock caused by compaction and file ingestion waiting for each other in the event of write stalls.
+* Fix a memory leak when files with range tombstones are read in mmap mode and block cache is enabled
+* Fix handling of corrupt range tombstone blocks such that corruptions cannot cause deleted keys to reappear
+* Lock free MultiGet
+* Fix incorrect `NotFound` point lookup result when querying the endpoint of a file that has been extended by a range tombstone.
+* Fix with pipelined write, write leaders's callback failure lead to the whole write group fail.
 
-## 5.16.5 (10/16/2018)
-### Bug Fixes
-* Fix slow flush/compaction when DB contains many snapshots. The problem became noticeable to us in DBs with 100,000+ snapshots, though it will affect others at different thresholds.
-* Properly set the stop key for a truncated manual CompactRange
+### Change Default Options
+* Change options.compaction_pri's default to kMinOverlappingRatio
+
+## 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.
+
+### 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.
 
-## 5.16.4 (10/10/2018)
 ### 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.
+
+## 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
@@ -29,11 +87,9 @@
 
 ## 5.16.0 (8/21/2018)
 ### 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.
-
 * 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.
@@ -42,7 +98,6 @@
 
 ### Bug Fixes
 * Fix a bug in misreporting the estimated partition index size in properties block.
-* Avoid creating empty SSTs and subsequently deleting them in certain cases during compaction.
 
 ## 5.15.0 (7/17/2018)
 ### Public API Change
@@ -178,7 +233,8 @@
 * `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**

Makefile

@@ -93,19 +93,38 @@ 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
 endif
 endif
 
-ifeq (,$(shell $(CXX) -fsyntax-only -faligned-new -xc++ /dev/null 2>&1))
-CXXFLAGS += -faligned-new -DHAVE_ALIGNED_NEW
-endif
-
 ifeq (,$(shell $(CXX) -fsyntax-only -maltivec -xc /dev/null 2>&1))
 CXXFLAGS += -DHAS_ALTIVEC
 CFLAGS += -DHAS_ALTIVEC
@@ -179,7 +198,7 @@ include make_config.mk
 CLEAN_FILES += make_config.mk
 
 missing_make_config_paths := $(shell				\
-	grep "\/\S*" -o $(CURDIR)/make_config.mk | 		\
+	grep "\./\S*\|/\S*" -o $(CURDIR)/make_config.mk | 	\
 	while read path;					\
 		do [ -e $$path ] || echo $$path; 		\
 	done | sort | uniq)
@@ -322,7 +341,7 @@ 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
 
@@ -384,7 +403,7 @@ BENCHTOOLOBJECTS = $(BENCH_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
 
 ANALYZETOOLOBJECTS = $(ANALYZER_LIB_SOURCES:.cc=.o)
 
-EXPOBJECTS = $(EXP_LIB_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
+EXPOBJECTS = $(LIBOBJECTS) $(TESTUTIL)
 
 TESTS = \
 	db_basic_test \
@@ -463,7 +482,6 @@ TESTS = \
 	merger_test \
 	util_merge_operators_test \
 	options_file_test \
-	redis_test \
 	reduce_levels_test \
 	plain_table_db_test \
 	comparator_db_test \
@@ -477,12 +495,8 @@ TESTS = \
 	cassandra_row_merge_test \
 	cassandra_serialize_test \
 	ttl_test \
-	date_tiered_test \
 	backupable_db_test \
-	document_db_test \
-	json_document_test \
 	sim_cache_test \
-	spatial_db_test \
 	version_edit_test \
 	version_set_test \
 	compaction_picker_test \
@@ -494,7 +508,6 @@ TESTS = \
 	deletefile_test \
 	obsolete_files_test \
 	table_test \
-	geodb_test \
 	delete_scheduler_test \
 	options_test \
 	options_settable_test \
@@ -511,7 +524,6 @@ TESTS = \
 	compaction_job_test \
 	thread_list_test \
 	sst_dump_test \
-	column_aware_encoding_test \
 	compact_files_test \
 	optimistic_transaction_test \
 	write_callback_test \
@@ -524,7 +536,6 @@ TESTS = \
 	persistent_cache_test \
 	statistics_test \
 	lua_test \
-	range_del_aggregator_test \
 	lru_cache_test \
 	object_registry_test \
 	repair_test \
@@ -533,6 +544,10 @@ TESTS = \
 	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 \
@@ -582,7 +597,7 @@ 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 persistent_cache_bench range_del_aggregator_bench
 
 # if user didn't config LIBNAME, set the default
 ifeq ($(LIBNAME),)
@@ -802,7 +817,6 @@ J ?= 100%
 # Use this regexp to select the subset of tests whose names match.
 tests-regexp = .
 
-t_run = $(wildcard t/run-*)
 .PHONY: check_0
 check_0:
 	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD); \
@@ -812,7 +826,7 @@ check_0:
 	test -t 1 && eta=--eta || eta=; \
 	{ \
 		printf './%s\n' $(filter-out $(PARALLEL_TEST),$(TESTS)); \
-		printf '%s\n' $(t_run); \
+		find t -name 'run-*' -print; \
 	} \
 	  | $(prioritize_long_running_tests)				\
 	  | grep -E '$(tests-regexp)'					\
@@ -829,7 +843,7 @@ valgrind_check_0:
 	test -t 1 && eta=--eta || eta=;					\
 	{								\
 	  printf './%s\n' $(filter-out $(PARALLEL_TEST) %skiplist_test options_settable_test, $(TESTS));		\
-	  printf '%s\n' $(t_run);					\
+	  find t -name 'run-*' -print; \
 	}								\
 	  | $(prioritize_long_running_tests)				\
 	  | grep -E '$(tests-regexp)'					\
@@ -883,6 +897,7 @@ 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),)
@@ -891,6 +906,8 @@ 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)
@@ -1002,8 +1019,10 @@ 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
 
@@ -1126,9 +1145,6 @@ cassandra_row_merge_test: utilities/cassandra/cassandra_row_merge_test.o utiliti
 cassandra_serialize_test: utilities/cassandra/cassandra_serialize_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-redis_test: utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 hash_table_test: utilities/persistent_cache/hash_table_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1267,18 +1283,9 @@ backupable_db_test: utilities/backupable/backupable_db_test.o $(LIBOBJECTS) $(TE
 checkpoint_test: utilities/checkpoint/checkpoint_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-document_db_test: utilities/document/document_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
-json_document_test: utilities/document/json_document_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 sim_cache_test: utilities/simulator_cache/sim_cache_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-spatial_db_test: utilities/spatialdb/spatial_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 env_mirror_test: utilities/env_mirror_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1296,9 +1303,6 @@ object_registry_test: utilities/object_registry_test.o $(LIBOBJECTS) $(TESTHARNE
 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-date_tiered_test: utilities/date_tiered/date_tiered_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 write_batch_with_index_test: utilities/write_batch_with_index/write_batch_with_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1425,9 +1429,6 @@ deletefile_test: db/deletefile_test.o $(LIBOBJECTS) $(TESTHARNESS)
 obsolete_files_test: db/obsolete_files_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-geodb_test: utilities/geodb/geodb_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 rocksdb_dump: tools/dump/rocksdb_dump.o $(LIBOBJECTS)
 	$(AM_LINK)
 
@@ -1476,9 +1477,6 @@ timer_queue_test: util/timer_queue_test.o $(LIBOBJECTS) $(TESTHARNESS)
 sst_dump_test: tools/sst_dump_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-column_aware_encoding_test: utilities/column_aware_encoding_test.o $(TESTHARNESS) $(EXPOBJECTS)
-	$(AM_LINK)
-
 optimistic_transaction_test: utilities/transactions/optimistic_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1518,9 +1516,6 @@ sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 blob_dump: tools/blob_dump.o $(LIBOBJECTS)
 	$(AM_LINK)
 
-column_aware_encoding_exp: utilities/column_aware_encoding_exp.o $(EXPOBJECTS)
-	$(AM_LINK)
-
 repair_test: db/repair_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1548,9 +1543,21 @@ 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
@@ -1622,14 +1629,14 @@ ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.6
 BZIP2_SHA256 ?= a2848f34fcd5d6cf47def00461fcb528a0484d8edef8208d6d2e2909dc61d9cd
 BZIP2_DOWNLOAD_BASE ?= https://web.archive.org/web/20180624184835/http://www.bzip.org
-SNAPPY_VER ?= 1.1.4
-SNAPPY_SHA256 ?= 134bfe122fd25599bb807bb8130e7ba6d9bdb851e0b16efcb83ac4f5d0b70057
-SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/releases/download
-LZ4_VER ?= 1.8.0
-LZ4_SHA256 ?= 2ca482ea7a9bb103603108b5a7510b7592b90158c151ff50a28f1ca8389fccf6
+SNAPPY_VER ?= 1.1.7
+SNAPPY_SHA256 ?= 3dfa02e873ff51a11ee02b9ca391807f0c8ea0529a4924afa645fbf97163f9d4
+SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
+LZ4_VER ?= 1.8.3
+LZ4_SHA256 ?= 33af5936ac06536805f9745e0b6d61da606a1f8b4cc5c04dd3cbaca3b9b4fc43
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.3.3
-ZSTD_SHA256 ?= a77c47153ee7de02626c5b2a097005786b71688be61e9fb81806a011f90b297b
+ZSTD_VER ?= 1.3.7
+ZSTD_SHA256 ?= 5dd1e90eb16c25425880c8a91327f63de22891ffed082fcc17e5ae84fce0d5fb
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
 CURL_SSL_OPTS ?= --tlsv1
 
@@ -1668,7 +1675,9 @@ 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 --output zlib-$(ZLIB_VER).tar.gz -L ${ZLIB_DOWNLOAD_BASE}/zlib-$(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\"; \
@@ -1680,7 +1689,9 @@ 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-$(BZIP2_VER).tar.gz))
+	curl --output bzip2-$(BZIP2_VER).tar.gz -L ${BZIP2_DOWNLOAD_BASE}/$(BZIP2_VER)/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\"; \
@@ -1692,21 +1703,24 @@ 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-$(SNAPPY_VER).tar.gz))
+	curl --output snappy-$(SNAPPY_VER).tar.gz -L ${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 .. && $(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 --output lz4-$(LZ4_VER).tar.gz -L ${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\"; \
@@ -1718,8 +1732,9 @@ 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 --output zstd-$(ZSTD_VER).tar.gz -L ${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\"; \
@@ -1888,7 +1903,8 @@ commit_prereq: build_tools/rocksdb-lego-determinator \
 ifeq ($(PLATFORM), IOS)
 # For iOS, create universal object files to be used on both the simulator and
 # a device.
-PLATFORMSROOT=/Applications/Xcode.app/Contents/Developer/Platforms
+XCODEROOT=$(shell xcode-select -print-path)
+PLATFORMSROOT=$(XCODEROOT)/Platforms
 SIMULATORROOT=$(PLATFORMSROOT)/iPhoneSimulator.platform/Developer
 DEVICEROOT=$(PLATFORMSROOT)/iPhoneOS.platform/Developer
 IOSVERSION=$(shell defaults read $(PLATFORMSROOT)/iPhoneOS.platform/version CFBundleShortVersionString)
@@ -1925,7 +1941,7 @@ 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) $(ANALYZER_LIB_SOURCES)
+all_sources = $(LIB_SOURCES) $(MAIN_SOURCES) $(MOCK_LIB_SOURCES) $(TOOL_LIB_SOURCES) $(BENCH_LIB_SOURCES) $(TEST_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

TARGETS

@@ -1,12 +1,10 @@
 load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+load("@fbcode_macros//build_defs:cpp_library.bzl", "cpp_library")
+load(":defs.bzl", "test_binary")
 
 REPO_PATH = package_name() + "/"
 
-BUCK_BINS = "buck-out/gen/" + REPO_PATH
-
-TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
-
-rocksdb_compiler_flags = [
+ROCKSDB_COMPILER_FLAGS = [
     "-fno-builtin-memcmp",
     "-DROCKSDB_PLATFORM_POSIX",
     "-DROCKSDB_LIB_IO_POSIX",
@@ -22,6 +20,7 @@ rocksdb_compiler_flags = [
     "-DBZIP2",
     "-DLZ4",
     "-DZSTD",
+    "-DZSTD_STATIC_LINKING_ONLY",
     "-DGFLAGS=gflags",
     "-DNUMA",
     "-DTBB",
@@ -33,7 +32,7 @@ rocksdb_compiler_flags = [
     "-Wnarrowing",
 ]
 
-rocksdb_external_deps = [
+ROCKSDB_EXTERNAL_DEPS = [
     ("bzip2", None, "bz2"),
     ("snappy", None, "snappy"),
     ("zlib", None, "z"),
@@ -45,13 +44,13 @@ rocksdb_external_deps = [
     ("googletest", None, "gtest"),
 ]
 
-rocksdb_preprocessor_flags = [
+ROCKSDB_PREPROCESSOR_FLAGS = [
     # Directories with files for #include
     "-I" + REPO_PATH + "include/",
     "-I" + REPO_PATH,
 ]
 
-rocksdb_arch_preprocessor_flags = {
+ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
     "x86_64": [
         "-DHAVE_SSE42",
         "-DHAVE_PCLMUL",
@@ -64,8 +63,15 @@ is_opt_mode = build_mode.startswith("opt")
 
 # -DNDEBUG is added by default in opt mode in fbcode. But adding it twice
 # doesn't harm and avoid forgetting to add it.
-if is_opt_mode:
-    rocksdb_compiler_flags.append("-DNDEBUG")
+ROCKSDB_COMPILER_FLAGS += (["-DNDEBUG"] if is_opt_mode else [])
+
+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.
+ROCKSDB_COMPILER_FLAGS += (["-DROCKSDB_JEMALLOC"] if sanitizer == "" else [])
+
+ROCKSDB_EXTERNAL_DEPS += ([("jemalloc", None, "headers")] if sanitizer == "" else [])
 
 cpp_library(
     name = "rocksdb_lib",
@@ -81,6 +87,7 @@ 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",
@@ -113,6 +120,7 @@ cpp_library(
         "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",
@@ -190,6 +198,7 @@ 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",
@@ -206,6 +215,7 @@ cpp_library(
         "util/comparator.cc",
         "util/compression_context_cache.cc",
         "util/concurrent_arena.cc",
+        "util/concurrent_task_limiter_impl.cc",
         "util/crc32c.cc",
         "util/delete_scheduler.cc",
         "util/dynamic_bloom.cc",
@@ -215,6 +225,7 @@ 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",
@@ -246,14 +257,9 @@ cpp_library(
         "utilities/checkpoint/checkpoint_impl.cc",
         "utilities/compaction_filters/remove_emptyvalue_compactionfilter.cc",
         "utilities/convenience/info_log_finder.cc",
-        "utilities/date_tiered/date_tiered_db_impl.cc",
         "utilities/debug.cc",
-        "utilities/document/document_db.cc",
-        "utilities/document/json_document.cc",
-        "utilities/document/json_document_builder.cc",
         "utilities/env_mirror.cc",
         "utilities/env_timed.cc",
-        "utilities/geodb/geodb_impl.cc",
         "utilities/leveldb_options/leveldb_options.cc",
         "utilities/lua/rocks_lua_compaction_filter.cc",
         "utilities/memory/memory_util.cc",
@@ -270,9 +276,7 @@ cpp_library(
         "utilities/persistent_cache/block_cache_tier_metadata.cc",
         "utilities/persistent_cache/persistent_cache_tier.cc",
         "utilities/persistent_cache/volatile_tier_impl.cc",
-        "utilities/redis/redis_lists.cc",
         "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",
@@ -292,12 +296,12 @@ cpp_library(
         "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,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
     deps = [],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 
 cpp_library(
@@ -310,41 +314,39 @@ cpp_library(
         "util/testharness.cc",
         "util/testutil.cc",
         "utilities/cassandra/test_utils.cc",
-        "utilities/col_buf_decoder.cc",
-        "utilities/col_buf_encoder.cc",
-        "utilities/column_aware_encoding_util.cc",
     ],
-    headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
     deps = [":rocksdb_lib"],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 
 cpp_library(
     name = "rocksdb_tools_lib",
     srcs = [
         "tools/db_bench_tool.cc",
+        "tools/trace_analyzer_tool.cc",
         "util/testutil.cc",
     ],
-    headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
     deps = [":rocksdb_lib"],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 
 cpp_library(
     name = "env_basic_test_lib",
     srcs = ["env/env_basic_test.cc"],
-    headers = AutoHeaders.RECURSIVE_GLOB,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    auto_headers = AutoHeaders.RECURSIVE_GLOB,
+    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,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 
 # [test_name, test_src, test_type]
@@ -434,11 +436,6 @@ ROCKS_TESTS = [
         "util/coding_test.cc",
         "serial",
     ],
-    [
-        "column_aware_encoding_test",
-        "utilities/column_aware_encoding_test.cc",
-        "serial",
-    ],
     [
         "column_family_test",
         "db/column_family_test.cc",
@@ -509,11 +506,6 @@ ROCKS_TESTS = [
         "table/data_block_hash_index_test.cc",
         "serial",
     ],
-    [
-        "date_tiered_test",
-        "utilities/date_tiered/date_tiered_test.cc",
-        "serial",
-    ],
     [
         "db_basic_test",
         "db/db_basic_test.cc",
@@ -674,11 +666,6 @@ ROCKS_TESTS = [
         "db/deletefile_test.cc",
         "serial",
     ],
-    [
-        "document_db_test",
-        "utilities/document/document_db_test.cc",
-        "serial",
-    ],
     [
         "dynamic_bloom_test",
         "util/dynamic_bloom_test.cc",
@@ -754,11 +741,6 @@ ROCKS_TESTS = [
         "table/full_filter_block_test.cc",
         "serial",
     ],
-    [
-        "geodb_test",
-        "utilities/geodb/geodb_test.cc",
-        "serial",
-    ],
     [
         "hash_table_test",
         "utilities/persistent_cache/hash_table_test.cc",
@@ -789,11 +771,6 @@ ROCKS_TESTS = [
         "monitoring/iostats_context_test.cc",
         "serial",
     ],
-    [
-        "json_document_test",
-        "utilities/document/json_document_test.cc",
-        "serial",
-    ],
     [
         "ldb_cmd_test",
         "tools/ldb_cmd_test.cc",
@@ -919,6 +896,11 @@ 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",
@@ -934,6 +916,11 @@ 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",
@@ -950,13 +937,13 @@ ROCKS_TESTS = [
         "serial",
     ],
     [
-        "spatial_db_test",
-        "utilities/spatialdb/spatial_db_test.cc",
+        "sst_dump_test",
+        "tools/sst_dump_test.cc",
         "serial",
     ],
     [
-        "sst_dump_test",
-        "tools/sst_dump_test.cc",
+        "sst_file_reader_test",
+        "table/sst_file_reader_test.cc",
         "serial",
     ],
     [
@@ -1074,26 +1061,16 @@ ROCKS_TESTS = [
 # Generate a test rule for each entry in ROCKS_TESTS
 # Do not build the tests in opt mode, since SyncPoint and other test code
 # will not be included.
-if not is_opt_mode:
-    for test_cfg in ROCKS_TESTS:
-        test_name = test_cfg[0]
-        test_cc = test_cfg[1]
-        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,
-        )
-
-        custom_unittest(
-          name = test_name,
-          type = ttype,
-          deps = [":" + test_bin],
-          command = [TEST_RUNNER, BUCK_BINS + test_bin]
-        )
+[
+    test_binary(
+        parallelism = parallelism,
+        rocksdb_arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+        rocksdb_compiler_flags = ROCKSDB_COMPILER_FLAGS,
+        rocksdb_external_deps = ROCKSDB_EXTERNAL_DEPS,
+        rocksdb_preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+        test_cc = test_cc,
+        test_name = test_name,
+    )
+    for test_name, test_cc, parallelism in ROCKS_TESTS
+    if not is_opt_mode
+]

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
@@ -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.

appveyor.yml

@@ -3,7 +3,7 @@ image: Visual Studio 2017
 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 ..
+- cmake -G "Visual Studio 15 Win64" -DOPTDBG=1 -DWITH_XPRESS=1 -DPORTABLE=1 -DJNI=1 ..
 - cd ..
 build:
   project: build\rocksdb.sln

buckifier/buckify_rocksdb.py

@@ -109,12 +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

@@ -2,16 +2,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
-rocksdb_target_header = """
-load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+rocksdb_target_header = """load("@fbcode_macros//build_defs:auto_headers.bzl", "AutoHeaders")
+load(":defs.bzl", "test_binary")
 
 REPO_PATH = package_name() + "/"
 
-BUCK_BINS = "buck-out/gen/" + REPO_PATH
-
-TEST_RUNNER = REPO_PATH + "buckifier/rocks_test_runner.sh"
-
-rocksdb_compiler_flags = [
+ROCKSDB_COMPILER_FLAGS = [
     "-fno-builtin-memcmp",
     "-DROCKSDB_PLATFORM_POSIX",
     "-DROCKSDB_LIB_IO_POSIX",
@@ -38,7 +34,7 @@ rocksdb_compiler_flags = [
     "-Wnarrowing",
 ]
 
-rocksdb_external_deps = [
+ROCKSDB_EXTERNAL_DEPS = [
     ("bzip2", None, "bz2"),
     ("snappy", None, "snappy"),
     ("zlib", None, "z"),
@@ -50,13 +46,13 @@ rocksdb_external_deps = [
     ("googletest", None, "gtest"),
 ]
 
-rocksdb_preprocessor_flags = [
+ROCKSDB_PREPROCESSOR_FLAGS = [
     # Directories with files for #include
     "-I" + REPO_PATH + "include/",
     "-I" + REPO_PATH,
 ]
 
-rocksdb_arch_preprocessor_flags = {
+ROCKSDB_ARCH_PREPROCESSOR_FLAGS = {
     "x86_64": [
         "-DHAVE_SSE42",
         "-DHAVE_PCLMUL",
@@ -69,21 +65,28 @@ is_opt_mode = build_mode.startswith("opt")
 
 # -DNDEBUG is added by default in opt mode in fbcode. But adding it twice
 # doesn't harm and avoid forgetting to add it.
-if is_opt_mode:
-    rocksdb_compiler_flags.append("-DNDEBUG")
+ROCKSDB_COMPILER_FLAGS += (["-DNDEBUG"] if is_opt_mode else [])
+
+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.
+ROCKSDB_COMPILER_FLAGS += (["-DROCKSDB_JEMALLOC"] if sanitizer == "" else [])
+
+ROCKSDB_EXTERNAL_DEPS += ([("jemalloc", None, "headers")] if sanitizer == "" else [])
 """
 
 
 library_template = """
 cpp_library(
-    name = "%s",
-    srcs = [%s],
-    headers = %s,
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
-    deps = [%s],
-    external_deps = rocksdb_external_deps,
+    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 = [{deps}],
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 """
 
@@ -91,11 +94,11 @@ binary_template = """
 cpp_binary(
     name = "%s",
     srcs = [%s],
-    arch_preprocessor_flags = rocksdb_arch_preprocessor_flags,
-    compiler_flags = rocksdb_compiler_flags,
-    preprocessor_flags = rocksdb_preprocessor_flags,
+    arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+    compiler_flags = ROCKSDB_COMPILER_FLAGS,
+    preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
     deps = [%s],
-    external_deps = rocksdb_external_deps,
+    external_deps = ROCKSDB_EXTERNAL_DEPS,
 )
 """
 
@@ -114,28 +117,17 @@ ROCKS_TESTS = [
 # Generate a test rule for each entry in ROCKS_TESTS
 # Do not build the tests in opt mode, since SyncPoint and other test code
 # will not be included.
-if not is_opt_mode:
-    for test_cfg in ROCKS_TESTS:
-        test_name = test_cfg[0]
-        test_cc = test_cfg[1]
-        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,
-        )
-
-        custom_unittest(
-          name = test_name,
-          type = ttype,
-          deps = [":" + test_bin],
-          command = [TEST_RUNNER, BUCK_BINS + test_bin]
-        )
-
+[
+    test_binary(
+        parallelism = parallelism,
+        rocksdb_arch_preprocessor_flags = ROCKSDB_ARCH_PREPROCESSOR_FLAGS,
+        rocksdb_compiler_flags = ROCKSDB_COMPILER_FLAGS,
+        rocksdb_external_deps = ROCKSDB_EXTERNAL_DEPS,
+        rocksdb_preprocessor_flags = ROCKSDB_PREPROCESSOR_FLAGS,
+        test_cc = test_cc,
+        test_name = test_name,
+    )
+    for test_name, test_cc, parallelism in ROCKS_TESTS
+    if not is_opt_mode
+]
 """

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 */
@@ -97,7 +97,7 @@ function getSteps($applyDiff, $diffID, $username, $test) {
   }
 
   // fbcode is a sub-repo. We cannot patch until we add it to ignore otherwise
-  // Git thinks it is an uncommited change.
+  // Git thinks it is an uncommitted change.
   $fix_git_ignore = array(
     "name" => "Fix git ignore",
     "shell" => "echo fbcode >> .git/info/exclude",

build_tools/build_detect_platform

@@ -56,10 +56,10 @@ if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
     if [ -n "$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 "$PWD/build_tools/fbcode_config.sh"
-    else
+    elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_PLATFORM007" ]; then
       source "$PWD/build_tools/fbcode_config_platform007.sh"
+    else
+      source "$PWD/build_tools/fbcode_config.sh"
     fi
 fi
 
@@ -370,6 +370,16 @@ EOF
             # JEMALLOC can be enabled either using the flag (like here) or by
             # providing direct link to the jemalloc library
             WITH_JEMALLOC_FLAG=1
+            # check for JEMALLOC installed with HomeBrew
+            if [ "$PLATFORM" == "OS_MACOSX" ]; then
+                if hash brew 2>/dev/null && brew ls --versions jemalloc > /dev/null; then
+                    JEMALLOC_VER=$(brew ls --versions jemalloc | tail -n 1 | cut -f 2 -d ' ')
+                    JEMALLOC_INCLUDE="-I/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/include"
+                    JEMALLOC_LIB="/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/lib/libjemalloc_pic.a"
+                    PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS $JEMALLOC_LIB"
+                    JAVA_STATIC_LDFLAGS="$JAVA_STATIC_LDFLAGS $JEMALLOC_LIB"
+                fi
+            fi
         fi
     fi
     if ! test $JEMALLOC && ! test $ROCKSDB_DISABLE_TCMALLOC; then
@@ -476,6 +486,17 @@ EOF
             COMMON_FLAGS="$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.

build_tools/fbcode_config.sh

@@ -51,7 +51,7 @@ if test -z $PIC_BUILD; then
 else
   ZSTD_LIBS=" $ZSTD_BASE/lib/libzstd_pic.a"
 fi
-CFLAGS+=" -DZSTD"
+CFLAGS+=" -DZSTD -DZSTD_STATIC_LINKING_ONLY"
 
 # location of gflags headers and libraries
 GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"

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/rocksdb-lego-determinator

@@ -346,7 +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
             },
@@ -664,12 +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`)

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

@@ -461,8 +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_));
@@ -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

@@ -279,7 +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;

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;

cmake/modules/Findzlib.cmake

@@ -1,21 +0,0 @@
-# - Find zlib
-# Find the zlib compression library and includes
-#
-# ZLIB_INCLUDE_DIR - where to find zlib.h, etc.
-# ZLIB_LIBRARIES - List of libraries when using zlib.
-# ZLIB_FOUND - True if zlib found.
-
-find_path(ZLIB_INCLUDE_DIR
-  NAMES zlib.h
-  HINTS ${ZLIB_ROOT_DIR}/include)
-
-find_library(ZLIB_LIBRARIES
-  NAMES z
-  HINTS ${ZLIB_ROOT_DIR}/lib)
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(zlib DEFAULT_MSG ZLIB_LIBRARIES ZLIB_INCLUDE_DIR)
-
-mark_as_advanced(
-  ZLIB_LIBRARIES
-  ZLIB_INCLUDE_DIR)

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"
@@ -65,8 +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,
@@ -86,12 +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(),
@@ -104,9 +104,9 @@ Status BuildTable(
 
   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);
@@ -121,8 +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, env,
+                                 ioptions.statistics, ioptions.listeners));
       builder = NewTableBuilder(
           ioptions, mutable_cf_options, internal_comparator,
           int_tbl_prop_collector_factories, column_family_id,
@@ -157,8 +158,10 @@ Status BuildTable(
       }
     }
 
-    for (auto it = range_del_agg->NewIterator(); it->Valid(); it->Next()) {
-      auto tombstone = it->Tombstone();
+    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(),

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"
@@ -65,8 +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;
@@ -108,8 +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" {
 
@@ -142,14 +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;
@@ -1769,11 +1784,11 @@ rocksdb_iterator_t* rocksdb_writebatch_wi_create_iterator_with_base(
 }
 
 rocksdb_iterator_t* rocksdb_writebatch_wi_create_iterator_with_base_cf(
-    rocksdb_writebatch_wi_t* wbwi,
-    rocksdb_iterator_t* base_iterator,
+    rocksdb_writebatch_wi_t* wbwi, rocksdb_iterator_t* base_iterator,
     rocksdb_column_family_handle_t* column_family) {
   rocksdb_iterator_t* result = new rocksdb_iterator_t;
-  result->rep = wbwi->rep->NewIteratorWithBase(column_family->rep, base_iterator->rep);
+  result->rep =
+      wbwi->rep->NewIteratorWithBase(column_family->rep, base_iterator->rep);
   delete base_iterator;
   return result;
 }
@@ -2052,6 +2067,15 @@ void rocksdb_set_options(
             db->rep->SetOptions(options_map));
     }
 
+void rocksdb_set_options_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* handle, int count, const char* const keys[], const char* const values[], char** errptr) {
+        std::unordered_map<std::string, std::string> options_map;
+        for (int i=0; i<count; i++)
+            options_map[keys[i]] = values[i];
+        SaveError(errptr,
+            db->rep->SetOptions(handle->rep, options_map));
+    }
+
 rocksdb_options_t* rocksdb_options_create() {
   return new rocksdb_options_t;
 }
@@ -2402,7 +2426,7 @@ 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,
@@ -2433,11 +2457,13 @@ 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) {
+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) {
+void rocksdb_options_set_max_subcompactions(rocksdb_options_t* opt,
+                                            uint32_t n) {
   opt->rep.max_subcompactions = n;
 }
 
@@ -3533,6 +3559,18 @@ 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;
@@ -4099,6 +4137,98 @@ 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);
@@ -908,7 +917,8 @@ int main(int argc, char** argv) {
     rocksdb_writebatch_wi_t* wbi = rocksdb_writebatch_wi_create(0, 1);
     rocksdb_writebatch_wi_put(wbi, "bar", 3, "b", 1);
     rocksdb_writebatch_wi_delete(wbi, "foo", 3);
-    rocksdb_iterator_t* iter = rocksdb_writebatch_wi_create_iterator_with_base(wbi, base_iter);
+    rocksdb_iterator_t* iter =
+        rocksdb_writebatch_wi_create_iterator_with_base(wbi, base_iter);
     CheckCondition(!rocksdb_iter_valid(iter));
     rocksdb_iter_seek_to_first(iter);
     CheckCondition(rocksdb_iter_valid(iter));
@@ -1334,6 +1344,47 @@ 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;
@@ -1477,7 +1528,7 @@ int main(int argc, char** argv) {
     const rocksdb_snapshot_t* snapshot;
     snapshot = rocksdb_transactiondb_create_snapshot(txn_db);
     rocksdb_readoptions_set_snapshot(roptions, snapshot);
-  
+
     rocksdb_transactiondb_put(txn_db, woptions, "foo", 3, "hey", 3,  &err);
     CheckNoError(err);
 
@@ -1675,7 +1726,7 @@ 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) {
@@ -130,14 +129,10 @@ Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
     }
   }
   if (cf_options.compression_opts.zstd_max_train_bytes > 0) {
-    if (!CompressionTypeSupported(CompressionType::kZSTD)) {
-      // Dictionary trainer is available since v0.6.1, but ZSTD was marked
-      // stable only since v0.8.0. For now we enable the feature in stable
-      // versions only.
+    if (!ZSTD_TrainDictionarySupported()) {
       return Status::InvalidArgument(
-          "zstd dictionary trainer cannot be used because " +
-          CompressionTypeToString(CompressionType::kZSTD) +
-          " is not linked with the binary.");
+          "zstd dictionary trainer cannot be used because ZSTD 1.1.3+ "
+          "is not linked with the binary.");
     }
     if (cf_options.compression_opts.max_dict_bytes == 0) {
       return Status::InvalidArgument(
@@ -945,26 +940,16 @@ 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/column_family.h

@@ -383,6 +383,8 @@ class ColumnFamilyData {
 
   Directory* GetDataDir(size_t path_id) const;
 
+  ThreadLocalPtr* TEST_GetLocalSV() { return local_sv_.get(); }
+
  private:
   friend class ColumnFamilySet;
   ColumnFamilyData(uint32_t id, const std::string& name,

db/column_family_test.cc

@@ -47,7 +47,7 @@ class EnvCounter : public EnvWrapper {
   int GetNumberOfNewWritableFileCalls() {
     return num_new_writable_file_;
   }
-  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r,
+  Status NewWritableFile(const std::string& f, std::unique_ptr<WritableFile>* r,
                          const EnvOptions& soptions) override {
     ++num_new_writable_file_;
     return EnvWrapper::NewWritableFile(f, r, soptions);
@@ -486,9 +486,9 @@ class ColumnFamilyTestBase : public testing::Test {
   void CopyFile(const std::string& source, const std::string& destination,
                 uint64_t size = 0) {
     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) {
@@ -513,6 +513,16 @@ class ColumnFamilyTestBase : public testing::Test {
     return static_cast<int>(files.size());
   }
 
+  void RecalculateWriteStallConditions(ColumnFamilyData* cfd,
+      const MutableCFOptions& mutable_cf_options)  {
+    // add lock to avoid race condition between
+    // `RecalculateWriteStallConditions` which writes to CFStats and
+    // background `DBImpl::DumpStats()` threads which read CFStats
+    dbfull()->TEST_LockMutex();
+    cfd->RecalculateWriteStallConditions(mutable_cf_options);
+    dbfull()-> TEST_UnlockMutex();
+  }
+
   std::vector<ColumnFamilyHandle*> handles_;
   std::vector<std::string> names_;
   std::vector<std::set<std::string>> keys_;
@@ -2032,7 +2042,7 @@ TEST_P(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
 }
 #endif  // !ROCKSDB_LITE
 
-#ifndef ROCKSDB_LITE  // Tailing interator not supported
+#ifndef ROCKSDB_LITE  // Tailing iterator not supported
 namespace {
 std::string IterStatus(Iterator* iter) {
   std::string result;
@@ -2500,139 +2510,139 @@ TEST_P(ColumnFamilyTest, WriteStallSingleColumnFamily) {
   mutable_cf_options.disable_auto_compactions = false;
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(50);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(201);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(400);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(500);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(450);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(205);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(202);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(201);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(198);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(399);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(599);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(2001);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(3001);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(390);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(100);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage->set_l0_delay_trigger_count(100);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(101);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->set_l0_delay_trigger_count(0);
   vstorage->TEST_set_estimated_compaction_needed_bytes(300);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage->set_l0_delay_trigger_count(101);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(200);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage->set_l0_delay_trigger_count(0);
   vstorage->TEST_set_estimated_compaction_needed_bytes(0);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   mutable_cf_options.disable_auto_compactions = true;
   dbfull()->TEST_write_controler().set_delayed_write_rate(kBaseRate);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage->set_l0_delay_trigger_count(50);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(0, GetDbDelayedWriteRate());
@@ -2640,7 +2650,7 @@ TEST_P(ColumnFamilyTest, WriteStallSingleColumnFamily) {
 
   vstorage->set_l0_delay_trigger_count(60);
   vstorage->TEST_set_estimated_compaction_needed_bytes(300);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(0, GetDbDelayedWriteRate());
@@ -2649,14 +2659,14 @@ TEST_P(ColumnFamilyTest, WriteStallSingleColumnFamily) {
   mutable_cf_options.disable_auto_compactions = false;
   vstorage->set_l0_delay_trigger_count(70);
   vstorage->TEST_set_estimated_compaction_needed_bytes(500);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->set_l0_delay_trigger_count(71);
   vstorage->TEST_set_estimated_compaction_needed_bytes(501);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
@@ -2681,31 +2691,31 @@ TEST_P(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
   mutable_cf_options.hard_pending_compaction_bytes_limit = 2000;
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(40);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(50);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(300);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(45);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(7);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(9);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(6);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   // Speed up threshold = min(4 * 2, 4 + (12 - 4)/4) = 6
@@ -2714,15 +2724,15 @@ TEST_P(ColumnFamilyTest, CompactionSpeedupSingleColumnFamily) {
   mutable_cf_options.level0_stop_writes_trigger = 30;
 
   vstorage->set_l0_delay_trigger_count(5);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(7);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(3);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }
 
@@ -2749,53 +2759,53 @@ TEST_P(ColumnFamilyTest, WriteStallTwoColumnFamilies) {
   mutable_cf_options1.soft_pending_compaction_bytes_limit = 500;
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(50);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(201);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(!dbfull()->TEST_write_controler().NeedsDelay());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(600);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(70);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate, GetDbDelayedWriteRate());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(800);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(300);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(700);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(500);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25 / 1.25, GetDbDelayedWriteRate());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(600);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_TRUE(!IsDbWriteStopped());
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_EQ(kBaseRate / 1.25, GetDbDelayedWriteRate());
@@ -2828,41 +2838,41 @@ TEST_P(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
   mutable_cf_options1.level0_slowdown_writes_trigger = 16;
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(40);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(60);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(30);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(70);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->TEST_set_estimated_compaction_needed_bytes(20);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage1->TEST_set_estimated_compaction_needed_bytes(3);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(9);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage1->set_l0_delay_trigger_count(2);
-  cfd1->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd1, mutable_cf_options);
   ASSERT_EQ(6, dbfull()->TEST_BGCompactionsAllowed());
 
   vstorage->set_l0_delay_trigger_count(0);
-  cfd->RecalculateWriteStallConditions(mutable_cf_options);
+  RecalculateWriteStallConditions(cfd, mutable_cf_options);
   ASSERT_EQ(1, dbfull()->TEST_BGCompactionsAllowed());
 }
 

db/compact_files_test.cc

@@ -309,6 +309,14 @@ TEST_F(CompactFilesTest, CompactionFilterWithGetSv) {
 }
 
 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 :
@@ -349,6 +357,51 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
   }
 }
 
+TEST_F(CompactFilesTest, GetCompactionJobInfo) {
+  Options options;
+  options.create_if_missing = true;
+  // Disable RocksDB background compaction.
+  options.compaction_style = kCompactionStyleNone;
+  options.level0_slowdown_writes_trigger = 1000;
+  options.level0_stop_writes_trigger = 1000;
+  options.write_buffer_size = 65536;
+  options.max_write_buffer_number = 2;
+  options.compression = kNoCompression;
+  options.max_compaction_bytes = 5000;
+
+  // Add listener
+  FlushedFileCollector* collector = new FlushedFileCollector();
+  options.listeners.emplace_back(collector);
+
+  DB* db = nullptr;
+  DestroyDB(db_name_, options);
+  Status s = DB::Open(options, db_name_, &db);
+  assert(s.ok());
+  assert(db);
+
+  // create couple files
+  for (int i = 0; i < 500; ++i) {
+    db->Put(WriteOptions(), ToString(i), std::string(1000, 'a' + (i % 26)));
+  }
+  reinterpret_cast<DBImpl*>(db)->TEST_WaitForFlushMemTable();
+  auto l0_files_1 = collector->GetFlushedFiles();
+  CompactionOptions co;
+  co.compression = CompressionType::kLZ4Compression;
+  CompactionJobInfo compaction_job_info;
+  ASSERT_OK(
+      db->CompactFiles(co, l0_files_1, 0, -1, nullptr, &compaction_job_info));
+  ASSERT_EQ(compaction_job_info.base_input_level, 0);
+  ASSERT_EQ(compaction_job_info.cf_id, db->DefaultColumnFamily()->GetID());
+  ASSERT_EQ(compaction_job_info.cf_name, db->DefaultColumnFamily()->GetName());
+  ASSERT_EQ(compaction_job_info.compaction_reason,
+            CompactionReason::kManualCompaction);
+  ASSERT_EQ(compaction_job_info.compression, CompressionType::kLZ4Compression);
+  ASSERT_EQ(compaction_job_info.output_level, 0);
+  ASSERT_OK(compaction_job_info.status);
+  // no assertion failure
+  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(
@@ -155,7 +235,7 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
       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_)),
@@ -293,8 +373,8 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
         if (user_cmp->Compare(user_key, f->largest.user_key()) <= 0) {
           // We've advanced far enough
           if (user_cmp->Compare(user_key, f->smallest.user_key()) >= 0) {
-            // Key falls in this file's range, so definitely
-            // exists beyond output level
+            // Key falls in this file's range, so it may
+            // exist beyond output level
             return false;
           }
           break;
@@ -331,12 +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(); }
@@ -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
@@ -262,6 +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(

db/compaction_iterator.cc

@@ -9,6 +9,23 @@
 #include "port/likely.h"
 #include "rocksdb/listener.h"
 #include "table/internal_iterator.h"
+#include "util/sync_point.h"
+
+#define DEFINITELY_IN_SNAPSHOT(seq, snapshot)                       \
+  ((seq) <= (snapshot) &&                                           \
+   (snapshot_checker_ == nullptr ||                                 \
+    LIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
+           SnapshotCheckerResult::kInSnapshot)))
+
+#define DEFINITELY_NOT_IN_SNAPSHOT(seq, snapshot)                     \
+  ((seq) > (snapshot) ||                                              \
+   (snapshot_checker_ != nullptr &&                                   \
+    UNLIKELY(snapshot_checker_->CheckInSnapshot((seq), (snapshot)) == \
+             SnapshotCheckerResult::kNotInSnapshot)))
+
+#define IN_EARLIEST_SNAPSHOT(seq) \
+  ((seq) <= earliest_snapshot_ && \
+   (snapshot_checker_ == nullptr || LIKELY(IsInEarliestSnapshot(seq))))
 
 namespace rocksdb {
 
@@ -18,8 +35,8 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* snapshot_checker, Env* env,
     bool report_detailed_time, bool expect_valid_internal_key,
-    RangeDelAggregator* range_del_agg,
-    const Compaction* compaction, const CompactionFilter* compaction_filter,
+    CompactionRangeDelAggregator* range_del_agg, const Compaction* compaction,
+    const CompactionFilter* compaction_filter,
     const std::atomic<bool>* shutting_down,
     const SequenceNumber preserve_deletes_seqnum)
     : CompactionIterator(
@@ -36,7 +53,7 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_write_conflict_snapshot,
     const SnapshotChecker* 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,
@@ -61,26 +78,28 @@ CompactionIterator::CompactionIterator(
       merge_out_iter_(merge_helper_),
       current_key_committed_(false) {
   assert(compaction_filter_ == nullptr || compaction_ != nullptr);
+  assert(snapshots_ != nullptr);
   bottommost_level_ =
       compaction_ == nullptr ? false : compaction_->bottommost_level();
   if (compaction_ != nullptr) {
     level_ptrs_ = std::vector<size_t>(compaction_->number_levels(), 0);
   }
-
   if (snapshots_->size() == 0) {
     // optimize for fast path if there are no snapshots
     visible_at_tip_ = true;
+    earliest_snapshot_iter_ = snapshots_->end();
     earliest_snapshot_ = kMaxSequenceNumber;
     latest_snapshot_ = 0;
   } else {
     visible_at_tip_ = false;
+    earliest_snapshot_iter_ = snapshots_->begin();
     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));
+    assert(snapshots_->at(i - 1) < snapshots_->at(i));
   }
 #endif
   if (compaction_filter_ != nullptr) {
@@ -91,6 +110,7 @@ CompactionIterator::CompactionIterator(
     ignore_snapshots_ = false;
   }
   input_->SetPinnedItersMgr(&pinned_iters_mgr_);
+  TEST_SYNC_POINT_CALLBACK("CompactionIterator:AfterInit", compaction_.get());
 }
 
 CompactionIterator::~CompactionIterator() {
@@ -162,10 +182,7 @@ void CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
   if (compaction_filter_ != nullptr &&
       (ikey_.type == kTypeValue || ikey_.type == kTypeBlobIndex) &&
       (visible_at_tip_ || ignore_snapshots_ ||
-       ikey_.sequence > latest_snapshot_ ||
-       (snapshot_checker_ != nullptr &&
-        UNLIKELY(!snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                  latest_snapshot_))))) {
+       DEFINITELY_NOT_IN_SNAPSHOT(ikey_.sequence, latest_snapshot_))) {
     // If the user has specified a compaction filter and the sequence
     // number is greater than any external snapshot, then invoke the
     // filter. If the return value of the compaction filter is true,
@@ -241,6 +258,7 @@ void CompactionIterator::NextFromInput() {
       valid_ = true;
       break;
     }
+    TEST_SYNC_POINT_CALLBACK("CompactionIterator:ProcessKV", &ikey_);
 
     // Update input statistics
     if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion) {
@@ -269,9 +287,7 @@ void CompactionIterator::NextFromInput() {
       has_outputted_key_ = false;
       current_user_key_sequence_ = kMaxSequenceNumber;
       current_user_key_snapshot_ = 0;
-      current_key_committed_ =
-          (snapshot_checker_ == nullptr ||
-           snapshot_checker_->IsInSnapshot(ikey_.sequence, kMaxSequenceNumber));
+      current_key_committed_ = KeyCommitted(ikey_.sequence);
 
       // Apply the compaction filter to the first committed version of the user
       // key.
@@ -293,8 +309,7 @@ void CompactionIterator::NextFromInput() {
       // to query snapshot_checker_ in that case.
       if (UNLIKELY(!current_key_committed_)) {
         assert(snapshot_checker_ != nullptr);
-        current_key_committed_ =
-            snapshot_checker_->IsInSnapshot(ikey_.sequence, kMaxSequenceNumber);
+        current_key_committed_ = KeyCommitted(ikey_.sequence);
         // Apply the compaction filter to the first committed version of the
         // user key.
         if (current_key_committed_) {
@@ -378,10 +393,8 @@ void CompactionIterator::NextFromInput() {
           cmp_->Equal(ikey_.user_key, next_ikey.user_key)) {
         // Check whether the next key belongs to the same snapshot as the
         // SingleDelete.
-        if (prev_snapshot == 0 || next_ikey.sequence > prev_snapshot ||
-            (snapshot_checker_ != nullptr &&
-             UNLIKELY(!snapshot_checker_->IsInSnapshot(next_ikey.sequence,
-                                                       prev_snapshot)))) {
+        if (prev_snapshot == 0 ||
+            DEFINITELY_NOT_IN_SNAPSHOT(next_ikey.sequence, prev_snapshot)) {
           if (next_ikey.type == kTypeSingleDeletion) {
             // We encountered two SingleDeletes in a row.  This could be due to
             // unexpected user input.
@@ -393,11 +406,8 @@ void CompactionIterator::NextFromInput() {
             ++iter_stats_.num_record_drop_obsolete;
             ++iter_stats_.num_single_del_mismatch;
           } else if (has_outputted_key_ ||
-                     (ikey_.sequence <= earliest_write_conflict_snapshot_ &&
-                      (snapshot_checker_ == nullptr ||
-                       LIKELY(snapshot_checker_->IsInSnapshot(
-                           ikey_.sequence,
-                           earliest_write_conflict_snapshot_))))) {
+                     DEFINITELY_IN_SNAPSHOT(
+                         ikey_.sequence, earliest_write_conflict_snapshot_)) {
             // Found a matching value, we can drop the single delete and the
             // value.  It is safe to drop both records since we've already
             // outputted a key in this snapshot, or there is no earlier
@@ -407,7 +417,8 @@ void CompactionIterator::NextFromInput() {
             // is an unexpected Merge or Delete.  We will compact it out
             // either way. We will maintain counts of how many mismatches
             // happened
-            if (next_ikey.type != kTypeValue) {
+            if (next_ikey.type != kTypeValue &&
+                next_ikey.type != kTypeBlobIndex) {
               ++iter_stats_.num_single_del_mismatch;
             }
 
@@ -444,10 +455,7 @@ void CompactionIterator::NextFromInput() {
         // iteration. If the next key is corrupt, we return before the
         // comparison, so the value of has_current_user_key does not matter.
         has_current_user_key_ = false;
-        if (compaction_ != nullptr && ikey_.sequence <= earliest_snapshot_ &&
-            (snapshot_checker_ == nullptr ||
-             LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                    earliest_snapshot_))) &&
+        if (compaction_ != nullptr && IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
             compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                        &level_ptrs_)) {
           // Key doesn't exist outside of this range.
@@ -466,24 +474,34 @@ void CompactionIterator::NextFromInput() {
       if (valid_) {
         at_next_ = true;
       }
-    } else if (last_snapshot == current_user_key_snapshot_) {
+    } else if (last_snapshot == current_user_key_snapshot_ ||
+               (last_snapshot > 0 &&
+                last_snapshot < current_user_key_snapshot_)) {
       // If the earliest snapshot is which this key is visible in
       // is the same as the visibility of a previous instance of the
       // same key, then this kv is not visible in any snapshot.
       // Hidden by an newer entry for same user key
-      // TODO(noetzli): why not > ?
       //
       // Note: Dropping this key will not affect TransactionDB write-conflict
       // checking since there has already been a record returned for this key
       // in this snapshot.
       assert(last_sequence >= current_user_key_sequence_);
+
+      // Note2: if last_snapshot < current_user_key_snapshot, it can only
+      // mean last_snapshot is released between we process last value and
+      // this value, and findEarliestVisibleSnapshot returns the next snapshot
+      // as current_user_key_snapshot. In this case last value and current
+      // value are both in current_user_key_snapshot currently.
+      assert(last_snapshot == current_user_key_snapshot_ ||
+             (snapshot_checker_ != nullptr &&
+              snapshot_checker_->CheckInSnapshot(current_user_key_sequence_,
+                                                 last_snapshot) ==
+                  SnapshotCheckerResult::kSnapshotReleased));
+
       ++iter_stats_.num_record_drop_hidden;  // (A)
       input_->Next();
     } else if (compaction_ != nullptr && ikey_.type == kTypeDeletion &&
-               ikey_.sequence <= earliest_snapshot_ &&
-               (snapshot_checker_ == nullptr ||
-                LIKELY(snapshot_checker_->IsInSnapshot(ikey_.sequence,
-                                                       earliest_snapshot_))) &&
+               IN_EARLIEST_SNAPSHOT(ikey_.sequence) &&
                ikeyNotNeededForIncrementalSnapshot() &&
                compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                           &level_ptrs_)) {
@@ -511,6 +529,28 @@ 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 ||
+              DEFINITELY_NOT_IN_SNAPSHOT(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(
@@ -592,13 +632,9 @@ void CompactionIterator::PrepareOutput() {
   //
   // Can we do the same for levels above bottom level as long as
   // KeyNotExistsBeyondOutputLevel() return true?
-  if ((compaction_ != nullptr &&
-      !compaction_->allow_ingest_behind()) &&
-      ikeyNotNeededForIncrementalSnapshot() &&
-      bottommost_level_ && valid_ && ikey_.sequence <= earliest_snapshot_ &&
-      (snapshot_checker_ == nullptr || LIKELY(snapshot_checker_->IsInSnapshot(
-        ikey_.sequence, earliest_snapshot_))) &&
-      ikey_.type != kTypeMerge &&
+  if ((compaction_ != nullptr && !compaction_->allow_ingest_behind()) &&
+      ikeyNotNeededForIncrementalSnapshot() && bottommost_level_ && valid_ &&
+      IN_EARLIEST_SNAPSHOT(ikey_.sequence) && ikey_.type != kTypeMerge &&
       !cmp_->Equal(compaction_->GetLargestUserKey(), ikey_.user_key)) {
     assert(ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion);
     ikey_.sequence = 0;
@@ -617,12 +653,23 @@ inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
     *prev_snapshot = *std::prev(snapshots_iter);
     assert(*prev_snapshot < in);
   }
+  if (snapshot_checker_ == nullptr) {
+    return snapshots_iter != snapshots_->end()
+      ? *snapshots_iter : kMaxSequenceNumber;
+  }
+  bool has_released_snapshot = !released_snapshots_.empty();
   for (; snapshots_iter != snapshots_->end(); ++snapshots_iter) {
     auto cur = *snapshots_iter;
     assert(in <= cur);
-    if (snapshot_checker_ == nullptr ||
-        snapshot_checker_->IsInSnapshot(in, cur)) {
+    // Skip if cur is in released_snapshots.
+    if (has_released_snapshot && released_snapshots_.count(cur) > 0) {
+      continue;
+    }
+    auto res = snapshot_checker_->CheckInSnapshot(in, cur);
+    if (res == SnapshotCheckerResult::kInSnapshot) {
       return cur;
+    } else if (res == SnapshotCheckerResult::kSnapshotReleased) {
+      released_snapshots_.insert(cur);
     }
     *prev_snapshot = cur;
   }
@@ -636,4 +683,30 @@ inline bool CompactionIterator::ikeyNotNeededForIncrementalSnapshot() {
          (ikey_.sequence < preserve_deletes_seqnum_);
 }
 
+bool CompactionIterator::IsInEarliestSnapshot(SequenceNumber sequence) {
+  assert(snapshot_checker_ != nullptr);
+  assert(earliest_snapshot_ == kMaxSequenceNumber ||
+         (earliest_snapshot_iter_ != snapshots_->end() &&
+          *earliest_snapshot_iter_ == earliest_snapshot_));
+  auto in_snapshot =
+      snapshot_checker_->CheckInSnapshot(sequence, earliest_snapshot_);
+  while (UNLIKELY(in_snapshot == SnapshotCheckerResult::kSnapshotReleased)) {
+    // Avoid the the current earliest_snapshot_ being return as
+    // earliest visible snapshot for the next value. So if a value's sequence
+    // is zero-ed out by PrepareOutput(), the next value will be compact out.
+    released_snapshots_.insert(earliest_snapshot_);
+    earliest_snapshot_iter_++;
+
+    if (earliest_snapshot_iter_ == snapshots_->end()) {
+      earliest_snapshot_ = kMaxSequenceNumber;
+    } else {
+      earliest_snapshot_ = *earliest_snapshot_iter_;
+    }
+    in_snapshot =
+        snapshot_checker_->CheckInSnapshot(sequence, earliest_snapshot_);
+  }
+  assert(in_snapshot != SnapshotCheckerResult::kSnapshotReleased);
+  return in_snapshot == SnapshotCheckerResult::kInSnapshot;
+}
+
 }  // namespace rocksdb

db/compaction_iterator.h

@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <deque>
 #include <string>
+#include <unordered_set>
 #include <vector>
 
 #include "db/compaction.h"
@@ -64,7 +65,7 @@ class CompactionIterator {
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* snapshot_checker, Env* env,
                      bool report_detailed_time, bool expect_valid_internal_key,
-                     RangeDelAggregator* range_del_agg,
+                     CompactionRangeDelAggregator* range_del_agg,
                      const Compaction* compaction = nullptr,
                      const CompactionFilter* compaction_filter = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -77,7 +78,7 @@ class CompactionIterator {
                      SequenceNumber earliest_write_conflict_snapshot,
                      const SnapshotChecker* 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 = nullptr,
                      const std::atomic<bool>* shutting_down = nullptr,
@@ -132,16 +133,31 @@ class CompactionIterator {
   // or seqnum be zero-ed out even if all other conditions for it are met.
   inline bool ikeyNotNeededForIncrementalSnapshot();
 
+  inline bool KeyCommitted(SequenceNumber sequence) {
+    return snapshot_checker_ == nullptr ||
+           snapshot_checker_->CheckInSnapshot(sequence, kMaxSequenceNumber) ==
+               SnapshotCheckerResult::kInSnapshot;
+  }
+
+  bool IsInEarliestSnapshot(SequenceNumber sequence);
+
   InternalIterator* input_;
   const Comparator* cmp_;
   MergeHelper* merge_helper_;
   const std::vector<SequenceNumber>* snapshots_;
+  // List of snapshots released during compaction.
+  // findEarliestVisibleSnapshot() find them out from return of
+  // snapshot_checker, and make sure they will not be returned as
+  // earliest visible snapshot of an older value.
+  // See WritePreparedTransactionTest::ReleaseSnapshotDuringCompaction3.
+  std::unordered_set<SequenceNumber> released_snapshots_;
+  std::vector<SequenceNumber>::const_iterator earliest_snapshot_iter_;
   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_;
@@ -151,6 +167,7 @@ class CompactionIterator {
   bool visible_at_tip_;
   SequenceNumber earliest_snapshot_;
   SequenceNumber latest_snapshot_;
+
   bool ignore_snapshots_;
 
   // State

db/compaction_iterator_test.cc

@@ -190,13 +190,17 @@ class TestSnapshotChecker : public SnapshotChecker {
       : last_committed_sequence_(last_committed_sequence),
         snapshots_(snapshots) {}
 
-  bool IsInSnapshot(SequenceNumber seq,
-                    SequenceNumber snapshot_seq) const override {
+  SnapshotCheckerResult CheckInSnapshot(
+      SequenceNumber seq, SequenceNumber snapshot_seq) const override {
     if (snapshot_seq == kMaxSequenceNumber) {
-      return seq <= last_committed_sequence_;
+      return seq <= last_committed_sequence_
+                 ? SnapshotCheckerResult::kInSnapshot
+                 : SnapshotCheckerResult::kNotInSnapshot;
     }
     assert(snapshots_.count(snapshot_seq) > 0);
-    return seq <= snapshots_.at(snapshot_seq);
+    return seq <= snapshots_.at(snapshot_seq)
+               ? SnapshotCheckerResult::kInSnapshot
+               : SnapshotCheckerResult::kNotInSnapshot;
   }
 
  private:
@@ -221,10 +225,15 @@ 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) {
@@ -247,9 +256,8 @@ 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 /* report_detailed_time */,
-        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,
@@ -293,7 +301,7 @@ 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_;
@@ -672,8 +680,12 @@ 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*/);
 }
@@ -842,13 +854,26 @@ 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

@@ -36,6 +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"
@@ -200,7 +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;
@@ -511,7 +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++;
@@ -521,7 +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;
   }
@@ -593,12 +603,15 @@ Status CompactionJob::Run() {
     thread.join();
   }
 
-  if (output_directory_) {
-    output_directory_->Fsync();
+  compaction_stats_.micros = env_->NowMicros() - start_micros;
+  compaction_stats_.cpu_micros = 0;
+  for (size_t i = 0; i < compact_->sub_compact_states.size(); i++) {
+    compaction_stats_.cpu_micros +=
+        compact_->sub_compact_states[i].compaction_job_stats.cpu_micros;
   }
 
-  compaction_stats_.micros = env_->NowMicros() - start_micros;
   MeasureTime(stats_, COMPACTION_TIME, compaction_stats_.micros);
+  MeasureTime(stats_, COMPACTION_CPU_TIME, compaction_stats_.cpu_micros);
 
   TEST_SYNC_POINT("CompactionJob::Run:BeforeVerify");
 
@@ -611,6 +624,10 @@ Status CompactionJob::Run() {
     }
   }
 
+  if (status.ok() && output_directory_) {
+    status = output_directory_->Fsync();
+  }
+
   if (status.ok()) {
     thread_pool.clear();
     std::vector<const FileMetaData*> files_meta;
@@ -757,6 +774,7 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
   stream << "job" << job_id_ << "event"
          << "compaction_finished"
          << "compaction_time_micros" << compaction_stats_.micros
+         << "compaction_time_cpu_micros" << compaction_stats_.cpu_micros
          << "output_level" << compact_->compaction->output_level()
          << "num_output_files" << compact_->NumOutputFiles()
          << "total_output_size" << compact_->total_bytes << "num_input_records"
@@ -794,11 +812,17 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options) {
 
 void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   assert(sub_compact != nullptr);
+
+  uint64_t prev_cpu_micros = env_->NowCPUNanos() / 1000;
+
   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);
@@ -810,13 +834,17 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   uint64_t prev_fsync_nanos = 0;
   uint64_t prev_range_sync_nanos = 0;
   uint64_t prev_prepare_write_nanos = 0;
+  uint64_t prev_cpu_write_nanos = 0;
+  uint64_t prev_cpu_read_nanos = 0;
   if (measure_io_stats_) {
     prev_perf_level = GetPerfLevel();
-    SetPerfLevel(PerfLevel::kEnableTime);
+    SetPerfLevel(PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);
     prev_write_nanos = IOSTATS(write_nanos);
     prev_fsync_nanos = IOSTATS(fsync_nanos);
     prev_range_sync_nanos = IOSTATS(range_sync_nanos);
     prev_prepare_write_nanos = IOSTATS(prepare_write_nanos);
+    prev_cpu_write_nanos = IOSTATS(cpu_write_nanos);
+    prev_cpu_read_nanos = IOSTATS(cpu_read_nanos);
   }
 
   const MutableCFOptions* mutable_cf_options =
@@ -887,7 +915,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
       input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
       &existing_snapshots_, earliest_write_conflict_snapshot_,
       snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_), false,
-      range_del_agg.get(), sub_compact->compaction, compaction_filter,
+      &range_del_agg, sub_compact->compaction, compaction_filter,
       shutting_down_, preserve_deletes_seqnum_));
   auto c_iter = sub_compact->c_iter.get();
   c_iter->SeekToFirst();
@@ -1025,9 +1053,9 @@ 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) {
@@ -1077,8 +1105,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   if (status.ok() && sub_compact->builder == nullptr &&
-      sub_compact->outputs.size() == 0 &&
-      !range_del_agg->IsEmpty()) {
+      sub_compact->outputs.size() == 0 && !range_del_agg.IsEmpty()) {
     // handle subcompaction containing only range deletions
     status = OpenCompactionOutputFile(sub_compact);
   }
@@ -1087,14 +1114,17 @@ 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;
     }
     RecordDroppedKeys(range_del_out_stats, &sub_compact->compaction_job_stats);
   }
 
+  sub_compact->compaction_job_stats.cpu_micros =
+      env_->NowCPUNanos() / 1000 - prev_cpu_micros;
+
   if (measure_io_stats_) {
     sub_compact->compaction_job_stats.file_write_nanos +=
         IOSTATS(write_nanos) - prev_write_nanos;
@@ -1104,7 +1134,11 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
         IOSTATS(range_sync_nanos) - prev_range_sync_nanos;
     sub_compact->compaction_job_stats.file_prepare_write_nanos +=
         IOSTATS(prepare_write_nanos) - prev_prepare_write_nanos;
-    if (prev_perf_level != PerfLevel::kEnableTime) {
+    sub_compact->compaction_job_stats.cpu_micros -=
+        (IOSTATS(cpu_write_nanos) - prev_cpu_write_nanos +
+         IOSTATS(cpu_read_nanos) - prev_cpu_read_nanos) /
+        1000;
+    if (prev_perf_level != PerfLevel::kEnableTimeAndCPUTimeExceptForMutex) {
       SetPerfLevel(prev_perf_level);
     }
   }
@@ -1153,7 +1187,7 @@ 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(
@@ -1177,10 +1211,12 @@ Status CompactionJob::FinishCompactionOutputFile(
     Slice lower_bound_guard, upper_bound_guard;
     std::string smallest_user_key;
     const Slice *lower_bound, *upper_bound;
+    bool lower_bound_from_sub_compact = false;
     if (sub_compact->outputs.size() == 1) {
       // For the first output table, include range tombstones before the min key
       // but after the subcompaction boundary.
       lower_bound = sub_compact->start;
+      lower_bound_from_sub_compact = true;
     } else if (meta->smallest.size() > 0) {
       // For subsequent output tables, only include range tombstones from min
       // key onwards since the previous file was extended to contain range
@@ -1205,18 +1241,38 @@ Status CompactionJob::FinishCompactionOutputFile(
     if (existing_snapshots_.size() > 0) {
       earliest_snapshot = existing_snapshots_[0];
     }
-    auto it = range_del_agg->NewIterator();
+    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;
+    }
+
+    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 &&
-          ucmp->Compare(*upper_bound, tombstone.start_key_) <= 0) {
-        // Tombstones starting at upper_bound or later only need to be included
-        // in the next table. Break because subsequent tombstones will start
-        // even later.
-        break;
+      if (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) {
@@ -1228,6 +1284,8 @@ Status CompactionJob::FinishCompactionOutputFile(
       }
 
       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 &&
@@ -1236,11 +1294,24 @@ Status CompactionJob::FinishCompactionOutputFile(
         // (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);
+        // When lower_bound is chosen by a subcompaction, we know that
+        // subcompactions over smaller keys cannot contain any keys at
+        // lower_bound. We also know that smaller subcompactions exist, because
+        // otherwise the subcompaction woud be unbounded on the left. As a
+        // result, we know that no other files on the output level will contain
+        // actual keys at lower_bound (an output file may have a largest key of
+        // lower_bound@kMaxSequenceNumber, but this only indicates a large range
+        // tombstone was truncated). Therefore, it is safe to use the
+        // tombstone's sequence number, to ensure that keys at lower_bound at
+        // lower levels are covered by truncated tombstones.
+        //
+        // If lower_bound was chosen by the smallest data key in the file,
+        // choose lowest seqnum so this file's smallest internal key comes after
+        // the previous file's largest. The fake seqnum is OK because the read
+        // path's file-picking code only considers user key.
+        smallest_candidate = InternalKey(
+            *lower_bound, lower_bound_from_sub_compact ? tombstone.seq_ : 0,
+            kTypeRangeDeletion);
       }
       InternalKey largest_candidate = tombstone.SerializeEndKey();
       if (upper_bound != nullptr &&
@@ -1262,9 +1333,23 @@ Status CompactionJob::FinishCompactionOutputFile(
         largest_candidate =
             InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
       }
+#ifndef NDEBUG
+      SequenceNumber smallest_ikey_seqnum = kMaxSequenceNumber;
+      if (meta->smallest.size() > 0) {
+        smallest_ikey_seqnum = GetInternalKeySeqno(meta->smallest.Encode());
+      }
+#endif
       meta->UpdateBoundariesForRange(smallest_candidate, largest_candidate,
                                      tombstone.seq_,
                                      cfd->internal_comparator());
+
+      // The smallest key in a file is used for range tombstone truncation, so
+      // it cannot have a seqnum of 0 (unless the smallest data key in a file
+      // has a seqnum of 0). Otherwise, the truncated tombstone may expose
+      // deleted keys at lower levels.
+      assert(smallest_ikey_seqnum == 0 ||
+             ExtractInternalKeyFooter(meta->smallest.Encode()) !=
+                 PackSequenceAndType(0, kTypeRangeDeletion));
     }
     meta->marked_for_compaction = sub_compact->builder->NeedCompact();
   }
@@ -1309,9 +1394,7 @@ 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;
   }
 
   if (s.ok() && (current_entries > 0 || tp.num_range_deletions > 0)) {
@@ -1344,10 +1427,7 @@ Status CompactionJob::FinishCompactionOutputFile(
   auto sfm =
       static_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
   if (sfm && meta != nullptr && meta->fd.GetPathId() == 0) {
-    auto fn =
-        TableFileName(sub_compact->compaction->immutable_cf_options()->cf_paths,
-                      meta->fd.GetNumber(), meta->fd.GetPathId());
-    sfm->OnAddFile(fn);
+    sfm->OnAddFile(fname);
     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)?
@@ -1433,7 +1513,7 @@ 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",
@@ -1465,8 +1545,11 @@ 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_,
+                             env_, 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

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"
@@ -104,7 +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);

db/compaction_job_test.cc

@@ -79,7 +79,7 @@ class CompactionJobTest : public testing::Test {
         shutting_down_(false),
         preserve_deletes_seqnum_(0),
         mock_table_factory_(new mock::MockTableFactory()),
-        error_handler_(db_options_, &mutex_) {
+        error_handler_(nullptr, db_options_, &mutex_) {
     EXPECT_OK(env_->CreateDirIfMissing(dbname_));
     db_options_.db_paths.emplace_back(dbname_,
                                       std::numeric_limits<uint64_t>::max());
@@ -200,12 +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;
@@ -456,8 +456,7 @@ TEST_F(CompactionJobTest, NonAssocMerge) {
 
   auto expected_results =
       mock::MakeMockFile({{KeyStr("a", 0U, kTypeValue), "3,4,5"},
-                          {KeyStr("b", 2U, kTypeMerge), "2"},
-                          {KeyStr("b", 1U, kTypeMerge), "1"}});
+                          {KeyStr("b", 2U, kTypeValue), "1,2"}});
 
   SetLastSequence(5U);
   auto files = cfd_->current()->storage_info()->LevelFiles(0);
@@ -484,7 +483,7 @@ TEST_F(CompactionJobTest, MergeOperandFilter) {
 
   auto expected_results =
       mock::MakeMockFile({{KeyStr("a", 0U, kTypeValue), test::EncodeInt(8U)},
-                          {KeyStr("b", 2U, kTypeMerge), test::EncodeInt(2U)}});
+                          {KeyStr("b", 2U, kTypeValue), test::EncodeInt(2U)}});
 
   SetLastSequence(5U);
   auto files = cfd_->current()->storage_info()->LevelFiles(0);

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.
@@ -943,8 +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 " +
@@ -1547,217 +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, 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;
-}
-
-#endif  // !ROCKSDB_LITE
-
 }  // namespace rocksdb

db/compaction_picker.h

@@ -235,42 +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,
@@ -308,6 +272,11 @@ 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,

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"
@@ -57,6 +59,9 @@ class CompactionPickerTest : public testing::Test {
         log_buffer_(InfoLogLevel::INFO_LEVEL, &logger_),
         file_num_(1),
         vstorage_(nullptr) {
+    // ioptions_.compaction_pri = kMinOverlappingRatio has its own set of
+    // tests to cover.
+    ioptions_.compaction_pri = kByCompensatedSize;
     fifo_options_.max_table_files_size = 1;
     mutable_cf_options_.RefreshDerivedOptions(ioptions_);
     ioptions_.cf_paths.emplace_back("dummy",
@@ -83,8 +88,8 @@ class CompactionPickerTest : public testing::Test {
 
   void Add(int level, uint32_t file_number, const char* smallest,
            const char* largest, uint64_t file_size = 1, uint32_t path_id = 0,
-           SequenceNumber smallest_seq = 100,
-           SequenceNumber largest_seq = 100) {
+           SequenceNumber smallest_seq = 100, SequenceNumber largest_seq = 100,
+           size_t compensated_file_size = 0) {
     assert(level < vstorage_->num_levels());
     FileMetaData* f = new FileMetaData;
     f->fd = FileDescriptor(file_number, path_id, file_size);
@@ -92,7 +97,8 @@ class CompactionPickerTest : public testing::Test {
     f->largest = InternalKey(largest, largest_seq, kTypeValue);
     f->fd.smallest_seqno = smallest_seq;
     f->fd.largest_seqno = largest_seq;
-    f->compensated_file_size = file_size;
+    f->compensated_file_size =
+        (compensated_file_size != 0) ? compensated_file_size : file_size;
     f->refs = 0;
     vstorage_->AddFile(level, f);
     files_.emplace_back(f);
@@ -615,6 +621,35 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping3) {
   ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
 }
 
+TEST_F(CompactionPickerTest, CompactionPriMinOverlapping4) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
+  mutable_cf_options_.max_bytes_for_level_base = 10000000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+
+  // file 7 and 8 over lap with the same file, but file 8 is smaller so
+  // it will be picked.
+  // Overlaps with file 26, 27. And the file is compensated so will be
+  // picked up.
+  Add(2, 6U, "150", "167", 60000000U, 0, 100, 100, 180000000U);
+  Add(2, 7U, "168", "169", 60000000U);  // Overlaps with file 27
+  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 28
+
+  Add(3, 26U, "160", "165", 60000000U);
+  // Boosted file size in output level is not considered.
+  Add(3, 27U, "166", "170", 60000000U, 0, 100, 100, 260000000U);
+  Add(3, 28U, "180", "400", 60000000U);
+  Add(3, 29U, "401", "500", 60000000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  // Picking file 8 because overlapping ratio is the biggest.
+  ASSERT_EQ(6U, compaction->input(0, 0)->fd.GetNumber());
+}
+
 // This test exhibits the bug where we don't properly reset parent_index in
 // PickCompaction()
 TEST_F(CompactionPickerTest, ParentIndexResetBug) {

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);
@@ -46,7 +46,7 @@ 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;

db/corruption_test.cc

@@ -24,6 +24,8 @@
 #include "rocksdb/env.h"
 #include "rocksdb/table.h"
 #include "rocksdb/write_batch.h"
+#include "table/block_based_table_builder.h"
+#include "table/meta_blocks.h"
 #include "util/filename.h"
 #include "util/string_util.h"
 #include "util/testharness.h"
@@ -37,7 +39,7 @@ 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_;
 
@@ -467,6 +469,39 @@ TEST_F(CorruptionTest, UnrelatedKeys) {
   ASSERT_EQ(Value(1000, &tmp2).ToString(), v);
 }
 
+TEST_F(CorruptionTest, RangeDeletionCorrupted) {
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  std::vector<LiveFileMetaData> metadata;
+  db_->GetLiveFilesMetaData(&metadata);
+  ASSERT_EQ(static_cast<size_t>(1), metadata.size());
+  std::string filename = dbname_ + metadata[0].name;
+
+  std::unique_ptr<RandomAccessFile> file;
+  ASSERT_OK(options_.env->NewRandomAccessFile(filename, &file, EnvOptions()));
+  std::unique_ptr<RandomAccessFileReader> file_reader(
+      new RandomAccessFileReader(std::move(file), filename));
+
+  uint64_t file_size;
+  ASSERT_OK(options_.env->GetFileSize(filename, &file_size));
+
+  BlockHandle range_del_handle;
+  ASSERT_OK(FindMetaBlock(
+      file_reader.get(), file_size, kBlockBasedTableMagicNumber,
+      ImmutableCFOptions(options_), kRangeDelBlock, &range_del_handle));
+
+  ASSERT_OK(TryReopen());
+  CorruptFile(filename, static_cast<int>(range_del_handle.offset()), 1);
+  // The test case does not fail on TryReopen because failure to preload table
+  // handlers is not considered critical.
+  ASSERT_OK(TryReopen());
+  std::string val;
+  // However, it does fail on any read involving that file since that file
+  // cannot be opened with a corrupt range deletion meta-block.
+  ASSERT_TRUE(db_->Get(ReadOptions(), "a", &val).IsCorruption());
+}
+
 TEST_F(CorruptionTest, FileSystemStateCorrupted) {
   for (int iter = 0; iter < 2; ++iter) {
     Options options;
@@ -485,7 +520,7 @@ 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/db_basic_test.cc

@@ -6,6 +6,7 @@
 // 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 <iostream>
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
@@ -794,7 +795,7 @@ 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 +810,7 @@ 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 +882,7 @@ 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();
     }
@@ -956,6 +957,184 @@ TEST_F(DBBasicTest, DBCloseFlushError) {
   Destroy(options);
 }
 
+TEST_F(DBBasicTest, MultiGetMultiCF) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  rocksdb::DBImpl* db = reinterpret_cast<DBImpl*>(db_);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (++get_sv_count == 2) {
+          // After MultiGet refs a couple of CFs, flush all CFs so MultiGet
+          // is forced to repeat the process
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                          "cf" + std::to_string(i) + "_val2"));
+          }
+        }
+        if (get_sv_count == 11) {
+          for (int i = 0; i < 8; ++i) {
+            auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                            db->GetColumnFamilyHandle(i))
+                            ->cfd();
+            ASSERT_EQ(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  values = MultiGet(cfs, keys);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j], "cf" + std::to_string(j) + "_val2");
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVObsolete);
+  }
+}
+
+TEST_F(DBBasicTest, MultiGetMultiCFMutex) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  int retries = 0;
+  bool last_try = false;
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::LastTry", [&](void* /*arg*/) {
+        last_try = true;
+        rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+      });
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (last_try) {
+          return;
+        }
+        if (++get_sv_count == 2) {
+          ++retries;
+          get_sv_count = 0;
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(
+                i, "cf" + std::to_string(i) + "_key",
+                "cf" + std::to_string(i) + "_val" + std::to_string(retries)));
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  values = MultiGet(cfs, keys);
+  ASSERT_TRUE(last_try);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j],
+              "cf" + std::to_string(j) + "_val" + std::to_string(retries));
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+  }
+}
+
+TEST_F(DBBasicTest, MultiGetMultiCFSnapshot) {
+  Options options = CurrentOptions();
+  CreateAndReopenWithCF({"pikachu", "ilya", "muromec", "dobrynia", "nikitich",
+                         "alyosha", "popovich"},
+                        options);
+
+  for (int i = 0; i < 8; ++i) {
+    ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                  "cf" + std::to_string(i) + "_val"));
+  }
+
+  int get_sv_count = 0;
+  rocksdb::DBImpl* db = reinterpret_cast<DBImpl*>(db_);
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::MultiGet::AfterRefSV", [&](void* /*arg*/) {
+        if (++get_sv_count == 2) {
+          for (int i = 0; i < 8; ++i) {
+            ASSERT_OK(Flush(i));
+            ASSERT_OK(Put(i, "cf" + std::to_string(i) + "_key",
+                          "cf" + std::to_string(i) + "_val2"));
+          }
+        }
+        if (get_sv_count == 8) {
+          for (int i = 0; i < 8; ++i) {
+            auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                            db->GetColumnFamilyHandle(i))
+                            ->cfd();
+            ASSERT_TRUE(
+                (cfd->TEST_GetLocalSV()->Get() == SuperVersion::kSVInUse) ||
+                (cfd->TEST_GetLocalSV()->Get() == SuperVersion::kSVObsolete));
+          }
+        }
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<int> cfs;
+  std::vector<std::string> keys;
+  std::vector<std::string> values;
+
+  for (int i = 0; i < 8; ++i) {
+    cfs.push_back(i);
+    keys.push_back("cf" + std::to_string(i) + "_key");
+  }
+
+  const Snapshot* snapshot = db_->GetSnapshot();
+  values = MultiGet(cfs, keys, snapshot);
+  db_->ReleaseSnapshot(snapshot);
+  ASSERT_EQ(values.size(), 8);
+  for (unsigned int j = 0; j < values.size(); ++j) {
+    ASSERT_EQ(values[j], "cf" + std::to_string(j) + "_val");
+  }
+  for (int i = 0; i < 8; ++i) {
+    auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
+                    reinterpret_cast<DBImpl*>(db_)->GetColumnFamilyHandle(i))
+                    ->cfd();
+    ASSERT_NE(cfd->TEST_GetLocalSV()->Get(), SuperVersion::kSVInUse);
+  }
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/db_block_cache_test.cc

@@ -173,7 +173,7 @@ TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
   delete iter;
   iter = nullptr;
 
-  // Release interators and access cache again.
+  // Release iterators and access cache again.
   for (size_t i = 0; i < kNumBlocks - 1; i++) {
     iterators[i].reset();
     CheckCacheCounters(options, 0, 0, 0, 0);
@@ -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);
 
@@ -631,6 +631,74 @@ TEST_F(DBBlockCacheTest, CompressedCache) {
   }
 }
 
+TEST_F(DBBlockCacheTest, CacheCompressionDict) {
+  const int kNumFiles = 4;
+  const int kNumEntriesPerFile = 32;
+  const int kNumBytesPerEntry = 1024;
+
+  // Try all the available libraries that support dictionary compression
+  std::vector<CompressionType> compression_types;
+#ifdef ZLIB
+  compression_types.push_back(kZlibCompression);
+#endif  // ZLIB
+#if LZ4_VERSION_NUMBER >= 10400
+  compression_types.push_back(kLZ4Compression);
+  compression_types.push_back(kLZ4HCCompression);
+#endif  // LZ4_VERSION_NUMBER >= 10400
+#if ZSTD_VERSION_NUMBER >= 500
+  compression_types.push_back(kZSTD);
+#endif  // ZSTD_VERSION_NUMBER >= 500
+  Random rnd(301);
+  for (auto compression_type : compression_types) {
+    Options options = CurrentOptions();
+    options.compression = compression_type;
+    options.compression_opts.max_dict_bytes = 4096;
+    options.create_if_missing = true;
+    options.num_levels = 2;
+    options.statistics = rocksdb::CreateDBStatistics();
+    options.target_file_size_base = kNumEntriesPerFile * kNumBytesPerEntry;
+    BlockBasedTableOptions table_options;
+    table_options.cache_index_and_filter_blocks = true;
+    table_options.block_cache.reset(new MockCache());
+    options.table_factory.reset(new BlockBasedTableFactory(table_options));
+    DestroyAndReopen(options);
+
+    for (int i = 0; i < kNumFiles; ++i) {
+      ASSERT_EQ(i, NumTableFilesAtLevel(0, 0));
+      for (int j = 0; j < kNumEntriesPerFile; ++j) {
+        std::string value = RandomString(&rnd, kNumBytesPerEntry);
+        ASSERT_OK(Put(Key(j * kNumFiles + i), value.c_str()));
+      }
+      ASSERT_OK(Flush());
+    }
+    dbfull()->TEST_WaitForCompact();
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(1));
+
+    // Seek to a key in a file. It should cause the SST's dictionary meta-block
+    // to be read.
+    RecordCacheCounters(options);
+    ASSERT_EQ(0,
+              TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
+    ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD));
+    ASSERT_EQ(
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
+        0);
+    ReadOptions read_options;
+    ASSERT_NE("NOT_FOUND", Get(Key(kNumFiles * kNumEntriesPerFile - 1)));
+    // Two blocks missed/added: dictionary and data block
+    // One block hit: index since it's prefetched
+    CheckCacheCounters(options, 2 /* expected_misses */, 1 /* expected_hits */,
+                       2 /* expected_inserts */, 0 /* expected_failures */);
+    ASSERT_EQ(1,
+              TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
+    ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD));
+    ASSERT_GT(
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
+        0);
+  }
+}
+
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

db/db_bloom_filter_test.cc

@@ -147,6 +147,7 @@ TEST_F(DBBloomFilterTest, GetFilterByPrefixBloomCustomPrefixExtractor) {
     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) {
@@ -171,20 +172,39 @@ TEST_F(DBBloomFilterTest, GetFilterByPrefixBloomCustomPrefixExtractor) {
 
     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();
   }
 }
 
@@ -193,6 +213,7 @@ TEST_F(DBBloomFilterTest, GetFilterByPrefixBloom) {
     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) {
@@ -231,6 +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();
   }
 }
 
@@ -239,6 +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));
@@ -386,6 +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();
   }
 }
 
@@ -465,6 +499,7 @@ TEST_P(DBBloomFilterTestWithParam, BloomFilter) {
   } while (ChangeCompactOptions());
 }
 
+#ifndef ROCKSDB_VALGRIND_RUN
 INSTANTIATE_TEST_CASE_P(
     FormatDef, DBBloomFilterTestDefFormatVersion,
     ::testing::Values(std::make_tuple(true, false, test::kDefaultFormatVersion),
@@ -485,11 +520,13 @@ INSTANTIATE_TEST_CASE_P(
                       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;
@@ -511,6 +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();
   }
 }
 
@@ -856,7 +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);
@@ -1032,6 +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;
@@ -1078,6 +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");
@@ -1187,6 +1237,7 @@ 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) {

db/db_compaction_filter_test.cc

@@ -340,9 +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()) {
@@ -429,9 +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()) {
@@ -646,9 +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()) {
@@ -848,7 +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 +8,15 @@
 // 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/concurrent_task_limiter.h"
 #include "rocksdb/experimental.h"
 #include "rocksdb/utilities/convenience.h"
+#include "util/concurrent_task_limiter_impl.h"
+#include "util/fault_injection_test_env.h"
 #include "util/sync_point.h"
+
 namespace rocksdb {
 
 // SYNC_POINT is not supported in released Windows mode.
@@ -120,13 +124,12 @@ class SstStatsCollector : public EventListener {
  public:
   SstStatsCollector() : num_ssts_creation_started_(0) {}
 
-  void OnTableFileCreationStarted(const TableFileCreationBriefInfo& /* info */) override {
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& /* info */) override {
     ++num_ssts_creation_started_;
   }
 
-  int num_ssts_creation_started() {
-    return num_ssts_creation_started_;
-  }
+  int num_ssts_creation_started() { return num_ssts_creation_started_; }
 
  private:
   std::atomic<int> num_ssts_creation_started_;
@@ -411,6 +414,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
 
   // Reopen the DB with stats-update disabled
   options.skip_stats_update_on_db_open = true;
+  options.max_open_files = 20;
   env_->random_file_open_counter_.store(0);
   Reopen(options);
 
@@ -436,7 +440,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
   Options options = CurrentOptions();
   options.env = env_;
   options.new_table_reader_for_compaction_inputs = true;
-  options.max_open_files = 100;
+  options.max_open_files = 20;
   options.level0_file_num_compaction_trigger = 3;
   DestroyAndReopen(options);
   Random rnd(301);
@@ -465,15 +469,16 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
       Flush();
       dbfull()->TEST_WaitForCompact();
       // preloading iterator issues one table cache lookup and create
-      // a new table reader.
-      ASSERT_EQ(num_table_cache_lookup, 1);
+      // a new table reader, if not preloaded.
+      int old_num_table_cache_lookup = num_table_cache_lookup;
+      ASSERT_GE(num_table_cache_lookup, 1);
       ASSERT_EQ(num_new_table_reader, 1);
 
       num_table_cache_lookup = 0;
       num_new_table_reader = 0;
       ASSERT_EQ(Key(k), Get(Key(k)));
       // lookup iterator from table cache and no need to create a new one.
-      ASSERT_EQ(num_table_cache_lookup, 1);
+      ASSERT_EQ(old_num_table_cache_lookup + num_table_cache_lookup, 2);
       ASSERT_EQ(num_new_table_reader, 0);
     }
   }
@@ -486,7 +491,10 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
   // a new table reader. One file is created for flush and one for compaction.
   // Compaction inputs make no table cache look-up for data/range deletion
   // iterators
-  ASSERT_EQ(num_table_cache_lookup, 2);
+  // May preload table cache too.
+  ASSERT_GE(num_table_cache_lookup, 2);
+  int old_num_table_cache_lookup2 = num_table_cache_lookup;
+
   // Create new iterator for:
   // (1) 1 for verifying flush results
   // (2) 3 for compaction input files
@@ -496,7 +504,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
   num_table_cache_lookup = 0;
   num_new_table_reader = 0;
   ASSERT_EQ(Key(1), Get(Key(1)));
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  ASSERT_EQ(num_table_cache_lookup + old_num_table_cache_lookup2, 3);
   ASSERT_EQ(num_new_table_reader, 0);
 
   num_table_cache_lookup = 0;
@@ -508,14 +516,16 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
   db_->CompactRange(cro, nullptr, nullptr);
   // Only verifying compaction outputs issues one table cache lookup
   // for both data block and range deletion block).
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  // May preload table cache too.
+  ASSERT_GE(num_table_cache_lookup, 1);
+  old_num_table_cache_lookup2 = num_table_cache_lookup;
   // One for compaction input, one for verifying compaction results.
   ASSERT_EQ(num_new_table_reader, 2);
 
   num_table_cache_lookup = 0;
   num_new_table_reader = 0;
   ASSERT_EQ(Key(1), Get(Key(1)));
-  ASSERT_EQ(num_table_cache_lookup, 1);
+  ASSERT_EQ(num_table_cache_lookup + old_num_table_cache_lookup2, 2);
   ASSERT_EQ(num_new_table_reader, 0);
 
   rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
@@ -1593,7 +1603,8 @@ TEST_F(DBCompactionTest, DeleteFileRange) {
   // Note that we don't delete level 0 files
   compact_options.change_level = true;
   compact_options.target_level = 1;
-  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
+  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
+  dbfull()->TEST_WaitForCompact();
 
   ASSERT_OK(
       DeleteFilesInRange(db_, db_->DefaultColumnFamily(), nullptr, nullptr));
@@ -2478,6 +2489,7 @@ 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));
@@ -2493,6 +2505,7 @@ 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));
@@ -2509,6 +2522,7 @@ 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));
@@ -2951,6 +2965,12 @@ 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;
@@ -2963,10 +2983,28 @@ 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
@@ -2976,7 +3014,7 @@ 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());
 
@@ -2993,7 +3031,7 @@ 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());
 
@@ -3011,7 +3049,7 @@ 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());
 
@@ -3029,7 +3067,7 @@ 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();
@@ -3862,6 +3900,192 @@ TEST_F(DBCompactionTest, CompactionHasEmptyOutput) {
   ASSERT_EQ(2, collector->num_ssts_creation_started());
 }
 
+TEST_F(DBCompactionTest, CompactionLimiter) {
+  const int kNumKeysPerFile = 10;
+  const int kMaxBackgroundThreads = 64;
+
+  struct CompactionLimiter {
+    std::string name;
+    int limit_tasks;
+    int max_tasks;
+    int tasks;
+    std::shared_ptr<ConcurrentTaskLimiter> limiter;
+  };
+
+  std::vector<CompactionLimiter> limiter_settings;
+  limiter_settings.push_back({"limiter_1", 1, 0, 0, nullptr});
+  limiter_settings.push_back({"limiter_2", 2, 0, 0, nullptr});
+  limiter_settings.push_back({"limiter_3", 3, 0, 0, nullptr});
+
+  for (auto& ls : limiter_settings) {
+    ls.limiter.reset(NewConcurrentTaskLimiter(ls.name, ls.limit_tasks));
+  }
+
+  std::shared_ptr<ConcurrentTaskLimiter> unique_limiter(
+    NewConcurrentTaskLimiter("unique_limiter", -1));
+
+  const char* cf_names[] = {"default", "0", "1", "2", "3", "4", "5",
+    "6", "7", "8", "9", "a", "b", "c", "d", "e", "f" };
+  const int cf_count = sizeof cf_names / sizeof cf_names[0];
+
+  std::unordered_map<std::string, CompactionLimiter*> cf_to_limiter;
+
+  Options options = CurrentOptions();
+  options.write_buffer_size = 110 * 1024;  // 110KB
+  options.arena_block_size = 4096;
+  options.num_levels = 3;
+  options.level0_file_num_compaction_trigger = 4;
+  options.level0_slowdown_writes_trigger = 64;
+  options.level0_stop_writes_trigger = 64;
+  options.max_background_jobs = kMaxBackgroundThreads; // Enough threads
+  options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
+  options.max_write_buffer_number = 10; // Enough memtables
+  DestroyAndReopen(options);
+
+  std::vector<Options> option_vector;
+  option_vector.reserve(cf_count);
+
+  for (int cf = 0; cf < cf_count; cf++) {
+    ColumnFamilyOptions cf_opt(options);
+    if (cf == 0) {
+      // "Default" CF does't use compaction limiter
+      cf_opt.compaction_thread_limiter = nullptr;
+    } else if (cf == 1) {
+      // "1" CF uses bypass compaction limiter
+      unique_limiter->SetMaxOutstandingTask(-1);
+      cf_opt.compaction_thread_limiter = unique_limiter;
+    } else {
+      // Assign limiter by mod
+      auto& ls = limiter_settings[cf % 3];
+      cf_opt.compaction_thread_limiter = ls.limiter;
+      cf_to_limiter[cf_names[cf]] = &ls;
+    }
+    option_vector.emplace_back(DBOptions(options), cf_opt);
+  }
+
+  for (int cf = 1; cf < cf_count; cf++) {
+    CreateColumnFamilies({cf_names[cf]}, option_vector[cf]);
+  }
+
+  ReopenWithColumnFamilies(std::vector<std::string>(cf_names,
+                                                    cf_names + cf_count),
+                           option_vector);
+
+  port::Mutex mutex;
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "DBImpl::BackgroundCompaction:BeforeCompaction", [&](void* arg) {
+      const auto& cf_name = static_cast<ColumnFamilyData*>(arg)->GetName();
+      auto iter = cf_to_limiter.find(cf_name);
+      if (iter != cf_to_limiter.end()) {
+        MutexLock l(&mutex);
+        ASSERT_GE(iter->second->limit_tasks, ++iter->second->tasks);
+        iter->second->max_tasks = std::max(iter->second->max_tasks,
+              iter->second->limit_tasks);
+      }
+    });
+
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+    "DBImpl::BackgroundCompaction:AfterCompaction", [&](void* arg) {
+      const auto& cf_name = static_cast<ColumnFamilyData*>(arg)->GetName();
+      auto iter = cf_to_limiter.find(cf_name);
+      if (iter != cf_to_limiter.end()) {
+        MutexLock l(&mutex);
+        ASSERT_GE(--iter->second->tasks, 0);
+      }
+    });
+
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  // Block all compact threads in thread pool.
+  const size_t kTotalFlushTasks = kMaxBackgroundThreads / 4;
+  const size_t kTotalCompactTasks = kMaxBackgroundThreads - kTotalFlushTasks;
+  env_->SetBackgroundThreads((int)kTotalFlushTasks, Env::HIGH);
+  env_->SetBackgroundThreads((int)kTotalCompactTasks, Env::LOW);
+
+  test::SleepingBackgroundTask sleeping_compact_tasks[kTotalCompactTasks];
+
+  // Block all compaction threads in thread pool.
+  for (size_t i = 0; i < kTotalCompactTasks; i++) {
+    env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
+                   &sleeping_compact_tasks[i], Env::LOW);
+    sleeping_compact_tasks[i].WaitUntilSleeping();
+  }
+
+  int keyIndex = 0;
+
+  for (int n = 0; n < options.level0_file_num_compaction_trigger; n++) {
+    for (int cf = 0; cf < cf_count; cf++) {
+      for (int i = 0; i < kNumKeysPerFile; i++) {
+        ASSERT_OK(Put(cf, Key(keyIndex++), ""));
+      }
+      // put extra key to trigger flush
+      ASSERT_OK(Put(cf, "", ""));
+    }
+
+    for (int cf = 0; cf < cf_count; cf++) {
+      dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+    }
+  }
+
+  // Enough L0 files to trigger compaction
+  for (int cf = 0; cf < cf_count; cf++) {
+    ASSERT_EQ(NumTableFilesAtLevel(0, cf),
+      options.level0_file_num_compaction_trigger);
+  }
+
+  // Create more files for one column family, which triggers speed up
+  // condition, all compactions will be scheduled.
+  for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
+    for (int i = 0; i < kNumKeysPerFile; i++) {
+      ASSERT_OK(Put(0, Key(i), ""));
+    }
+    // put extra key to trigger flush
+    ASSERT_OK(Put(0, "", ""));
+    dbfull()->TEST_WaitForFlushMemTable(handles_[0]);
+    ASSERT_EQ(options.level0_file_num_compaction_trigger + num + 1,
+              NumTableFilesAtLevel(0, 0));
+  }
+
+  // All CFs are pending compaction
+  ASSERT_EQ(cf_count, env_->GetThreadPoolQueueLen(Env::LOW));
+
+  // Unblock all compaction threads
+  for (size_t i = 0; i < kTotalCompactTasks; i++) {
+    sleeping_compact_tasks[i].WakeUp();
+    sleeping_compact_tasks[i].WaitUntilDone();
+  }
+
+  for (int cf = 0; cf < cf_count; cf++) {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
+  }
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Max outstanding compact tasks reached limit
+  for (auto& ls : limiter_settings) {
+    ASSERT_EQ(ls.limit_tasks, ls.max_tasks);
+    ASSERT_EQ(0, ls.limiter->GetOutstandingTask());
+  }
+
+  // test manual compaction under a fully throttled limiter
+  int cf_test = 1;
+  unique_limiter->SetMaxOutstandingTask(0);
+
+  // flush one more file to cf 1
+  for (int i = 0; i < kNumKeysPerFile; i++) {
+      ASSERT_OK(Put(cf_test, Key(keyIndex++), ""));
+  }
+  // put extra key to trigger flush
+  ASSERT_OK(Put(cf_test, "", ""));
+
+  dbfull()->TEST_WaitForFlushMemTable(handles_[cf_test]);
+  ASSERT_EQ(1, NumTableFilesAtLevel(0, cf_test));
+
+  Compact(cf_test, Key(0), Key(keyIndex));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+}
+
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                         ::testing::Values(std::make_tuple(1, true),
                                           std::make_tuple(1, false),
@@ -4003,6 +4227,123 @@ TEST_F(DBCompactionTest, PartialManualCompaction) {
   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) {
@@ -40,7 +46,7 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
   SyncPoint::GetInstance()->LoadDependency(
       {{"VersionSet::LogAndApply:WriteManifest",
         "DBFlushTest::FlushWhileWritingManifest:1"},
-       {"MemTableList::InstallMemtableFlushResults:InProgress",
+       {"MemTableList::TryInstallMemtableFlushResults:InProgress",
         "VersionSet::LogAndApply:WriteManifestDone"}});
   SyncPoint::GetInstance()->EnableProcessing();
 
@@ -56,7 +62,6 @@ TEST_F(DBFlushTest, FlushWhileWritingManifest) {
 #endif  // ROCKSDB_LITE
 }
 
-#ifndef TRAVIS
 // Disable this test temporarily on Travis as it fails intermittently.
 // Github issue: #4151
 TEST_F(DBFlushTest, SyncFail) {
@@ -67,11 +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())
@@ -82,6 +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");
@@ -96,7 +109,30 @@ TEST_F(DBFlushTest, SyncFail) {
   ASSERT_EQ(refs_before, cfd->current()->TEST_refs());
   Destroy(options);
 }
-#endif  // TRAVIS
+
+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
@@ -214,9 +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.cc

@@ -45,7 +45,7 @@
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
 #include "db/merge_helper.h"
-#include "db/range_del_aggregator.h"
+#include "db/range_tombstone_fragmenter.h"
 #include "db/table_cache.h"
 #include "db/table_properties_collector.h"
 #include "db/transaction_log_impl.h"
@@ -215,9 +215,12 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
       // requires a custom gc for compaction, we use that to set use_custom_gc_
       // as well.
       use_custom_gc_(seq_per_batch),
+      shutdown_initiated_(false),
+      own_sfm_(options.sst_file_manager == nullptr),
       preserve_deletes_(options.preserve_deletes),
       closed_(false),
-      error_handler_(immutable_db_options_, &mutex_) {
+      error_handler_(this, immutable_db_options_, &mutex_),
+      atomic_flush_install_cv_(&mutex_) {
   // !batch_per_trx_ implies seq_per_batch_ because it is only unset for
   // WriteUnprepared, which should use seq_per_batch_.
   assert(batch_per_txn_ || seq_per_batch_);
@@ -259,16 +262,85 @@ Status DBImpl::Resume() {
     return Status::OK();
   }
 
-  Status s = error_handler_.GetBGError();
-  if (s.severity() > Status::Severity::kHardError) {
+  if (error_handler_.IsRecoveryInProgress()) {
+    // Don't allow a mix of manual and automatic recovery
+    return Status::Busy();
+  }
+
+  mutex_.Unlock();
+  Status s = error_handler_.RecoverFromBGError(true);
+  mutex_.Lock();
+  return s;
+}
+
+// This function implements the guts of recovery from a background error. It
+// is eventually called for both manual as well as automatic recovery. It does
+// the following -
+// 1. Wait for currently scheduled background flush/compaction to exit, in
+//    order to inadvertently causing an error and thinking recovery failed
+// 2. Flush memtables if there's any data for all the CFs. This may result
+//    another error, which will be saved by error_handler_ and reported later
+//    as the recovery status
+// 3. Find and delete any obsolete files
+// 4. Schedule compactions if needed for all the CFs. This is needed as the
+//    flush in the prior step might have been a no-op for some CFs, which
+//    means a new super version wouldn't have been installed
+Status DBImpl::ResumeImpl() {
+  mutex_.AssertHeld();
+  WaitForBackgroundWork();
+
+  Status bg_error = error_handler_.GetBGError();
+  Status s;
+  if (shutdown_initiated_) {
+    // Returning shutdown status to SFM during auto recovery will cause it
+    // to abort the recovery and allow the shutdown to progress
+    s = Status::ShutdownInProgress();
+  }
+  if (s.ok() && bg_error.severity() > Status::Severity::kHardError) {
     ROCKS_LOG_INFO(immutable_db_options_.info_log,
         "DB resume requested but failed due to Fatal/Unrecoverable error");
-    return s;
+    s = bg_error;
+  }
+
+  // We cannot guarantee consistency of the WAL. So force flush Memtables of
+  // all the column families
+  if (s.ok()) {
+    FlushOptions flush_opts;
+    // We allow flush to stall write since we are trying to resume from error.
+    flush_opts.allow_write_stall = true;
+    if (immutable_db_options_.atomic_flush) {
+      autovector<ColumnFamilyData*> cfds;
+      SelectColumnFamiliesForAtomicFlush(&cfds);
+      mutex_.Unlock();
+      s = AtomicFlushMemTables(cfds, flush_opts, FlushReason::kErrorRecovery);
+      mutex_.Lock();
+    } else {
+      for (auto cfd : *versions_->GetColumnFamilySet()) {
+        if (cfd->IsDropped()) {
+          continue;
+        }
+        cfd->Ref();
+        mutex_.Unlock();
+        s = FlushMemTable(cfd, flush_opts, FlushReason::kErrorRecovery);
+        mutex_.Lock();
+        cfd->Unref();
+        if (!s.ok()) {
+          break;
+        }
+      }
+    }
+    if (!s.ok()) {
+      ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                     "DB resume requested but failed due to Flush failure [%s]",
+                     s.ToString().c_str());
+    }
   }
 
   JobContext job_context(0);
   FindObsoleteFiles(&job_context, true);
-  error_handler_.ClearBGError();
+  if (s.ok()) {
+    s = error_handler_.ClearBGError();
+  }
   mutex_.Unlock();
 
   job_context.manifest_file_number = 1;
@@ -277,32 +349,72 @@ Status DBImpl::Resume() {
   }
   job_context.Clean();
 
-  ROCKS_LOG_INFO(immutable_db_options_.info_log, "Successfully resumed DB");
+  if (s.ok()) {
+    ROCKS_LOG_INFO(immutable_db_options_.info_log, "Successfully resumed DB");
+  }
   mutex_.Lock();
-  MaybeScheduleFlushOrCompaction();
+  // Check for shutdown again before scheduling further compactions,
+  // since we released and re-acquired the lock above
+  if (shutdown_initiated_) {
+    s = Status::ShutdownInProgress();
+  }
+  if (s.ok()) {
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      SchedulePendingCompaction(cfd);
+    }
+    MaybeScheduleFlushOrCompaction();
+  }
+
+  // Wake up any waiters - in this case, it could be the shutdown thread
+  bg_cv_.SignalAll();
 
   // No need to check BGError again. If something happened, event listener would be
   // notified and the operation causing it would have failed
-  return Status::OK();
+  return s;
+}
+
+void DBImpl::WaitForBackgroundWork() {
+  // Wait for background work to finish
+  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
+         bg_flush_scheduled_) {
+    bg_cv_.Wait();
+  }
 }
 
 // Will lock the mutex_,  will wait for completion if wait is true
 void DBImpl::CancelAllBackgroundWork(bool wait) {
-  InstrumentedMutexLock l(&mutex_);
 
   ROCKS_LOG_INFO(immutable_db_options_.info_log,
                  "Shutdown: canceling all background work");
 
+  InstrumentedMutexLock l(&mutex_);
+  // To avoid deadlock, `thread_dump_stats_->cancel()` needs to be called
+  // before grabbing db mutex because the actual worker function
+  // `DBImpl::DumpStats()` also holds db mutex
+  if (thread_dump_stats_ != nullptr) {
+    mutex_.Unlock();
+    thread_dump_stats_->cancel();
+    mutex_.Lock();
+    thread_dump_stats_.reset();
+  }
   if (!shutting_down_.load(std::memory_order_acquire) &&
       has_unpersisted_data_.load(std::memory_order_relaxed) &&
       !mutable_db_options_.avoid_flush_during_shutdown) {
-    for (auto cfd : *versions_->GetColumnFamilySet()) {
-      if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
-        cfd->Ref();
-        mutex_.Unlock();
-        FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
-        mutex_.Lock();
-        cfd->Unref();
+    if (immutable_db_options_.atomic_flush) {
+      autovector<ColumnFamilyData*> cfds;
+      SelectColumnFamiliesForAtomicFlush(&cfds);
+      mutex_.Unlock();
+      AtomicFlushMemTables(cfds, FlushOptions(), FlushReason::kShutDown);
+      mutex_.Lock();
+    } else {
+      for (auto cfd : *versions_->GetColumnFamilySet()) {
+        if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
+          cfd->Ref();
+          mutex_.Unlock();
+          FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
+          mutex_.Lock();
+          cfd->Unref();
+        }
       }
     }
     versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
@@ -313,14 +425,20 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
   if (!wait) {
     return;
   }
-  // Wait for background work to finish
-  while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
-         bg_flush_scheduled_) {
-    bg_cv_.Wait();
-  }
+  WaitForBackgroundWork();
 }
 
 Status DBImpl::CloseHelper() {
+  // Guarantee that there is no background error recovery in progress before
+  // continuing with the shutdown
+  mutex_.Lock();
+  shutdown_initiated_ = true;
+  error_handler_.CancelErrorRecovery();
+  while (error_handler_.IsRecoveryInProgress()) {
+    bg_cv_.Wait();
+  }
+  mutex_.Unlock();
+
   // CancelAllBackgroundWork called with false means we just set the shutdown
   // marker. After this we do a variant of the waiting and unschedule work
   // (to consider: moving all the waiting into CancelAllBackgroundWork(true))
@@ -338,7 +456,8 @@ Status DBImpl::CloseHelper() {
   // Wait for background work to finish
   while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
          bg_flush_scheduled_ || bg_purge_scheduled_ ||
-         pending_purge_obsolete_files_) {
+         pending_purge_obsolete_files_ ||
+         error_handler_.IsRecoveryInProgress()) {
     TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
     bg_cv_.Wait();
   }
@@ -348,9 +467,12 @@ Status DBImpl::CloseHelper() {
   flush_scheduler_.Clear();
 
   while (!flush_queue_.empty()) {
-    auto cfd = PopFirstFromFlushQueue();
-    if (cfd->Unref()) {
-      delete cfd;
+    const FlushRequest& flush_req = PopFirstFromFlushQueue();
+    for (const auto& iter : flush_req) {
+      ColumnFamilyData* cfd = iter.first;
+      if (cfd->Unref()) {
+        delete cfd;
+      }
     }
   }
   while (!compaction_queue_.empty()) {
@@ -440,6 +562,17 @@ Status DBImpl::CloseHelper() {
   ROCKS_LOG_INFO(immutable_db_options_.info_log, "Shutdown complete");
   LogFlush(immutable_db_options_.info_log);
 
+#ifndef ROCKSDB_LITE
+  // If the sst_file_manager was allocated by us during DB::Open(), ccall
+  // Close() on it before closing the info_log. Otherwise, background thread
+  // in SstFileManagerImpl might try to log something
+  if (immutable_db_options_.sst_file_manager && own_sfm_) {
+    auto sfm = static_cast<SstFileManagerImpl*>(
+        immutable_db_options_.sst_file_manager.get());
+    sfm->Close();
+  }
+#endif // ROCKSDB_LITE
+
   if (immutable_db_options_.info_log && own_info_log_) {
     Status s = immutable_db_options_.info_log->Close();
     if (ret.ok()) {
@@ -485,66 +618,68 @@ void DBImpl::PrintStatistics() {
   }
 }
 
-void DBImpl::MaybeDumpStats() {
-  mutex_.Lock();
-  unsigned int stats_dump_period_sec =
-      mutable_db_options_.stats_dump_period_sec;
-  mutex_.Unlock();
-  if (stats_dump_period_sec == 0) return;
-
-  const uint64_t now_micros = env_->NowMicros();
-
-  if (last_stats_dump_time_microsec_ + stats_dump_period_sec * 1000000 <=
-      now_micros) {
-    // Multiple threads could race in here simultaneously.
-    // However, the last one will update last_stats_dump_time_microsec_
-    // atomically. We could see more than one dump during one dump
-    // period in rare cases.
-    last_stats_dump_time_microsec_ = now_micros;
+void DBImpl::StartTimedTasks() {
+  unsigned int stats_dump_period_sec = 0;
+  {
+    InstrumentedMutexLock l(&mutex_);
+    stats_dump_period_sec = mutable_db_options_.stats_dump_period_sec;
+    if (stats_dump_period_sec > 0) {
+      if (!thread_dump_stats_) {
+        thread_dump_stats_.reset(new rocksdb::RepeatableThread(
+            [this]() { DBImpl::DumpStats(); }, "dump_st", env_,
+            stats_dump_period_sec * 1000000));
+      }
+    }
+  }
+}
 
+void DBImpl::DumpStats() {
+  TEST_SYNC_POINT("DBImpl::DumpStats:1");
 #ifndef ROCKSDB_LITE
-    const DBPropertyInfo* cf_property_info =
-        GetPropertyInfo(DB::Properties::kCFStats);
-    assert(cf_property_info != nullptr);
-    const DBPropertyInfo* db_property_info =
-        GetPropertyInfo(DB::Properties::kDBStats);
-    assert(db_property_info != nullptr);
+  const DBPropertyInfo* cf_property_info =
+      GetPropertyInfo(DB::Properties::kCFStats);
+  assert(cf_property_info != nullptr);
+  const DBPropertyInfo* db_property_info =
+      GetPropertyInfo(DB::Properties::kDBStats);
+  assert(db_property_info != nullptr);
 
-    std::string stats;
-    {
-      InstrumentedMutexLock l(&mutex_);
-      default_cf_internal_stats_->GetStringProperty(
-          *db_property_info, DB::Properties::kDBStats, &stats);
-      for (auto cfd : *versions_->GetColumnFamilySet()) {
-        if (cfd->initialized()) {
-          cfd->internal_stats()->GetStringProperty(
-              *cf_property_info, DB::Properties::kCFStatsNoFileHistogram,
-              &stats);
-        }
-      }
-      for (auto cfd : *versions_->GetColumnFamilySet()) {
-        if (cfd->initialized()) {
-          cfd->internal_stats()->GetStringProperty(
-              *cf_property_info, DB::Properties::kCFFileHistogram, &stats);
-        }
+  std::string stats;
+  if (shutdown_initiated_) {
+    return;
+  }
+  {
+    InstrumentedMutexLock l(&mutex_);
+    default_cf_internal_stats_->GetStringProperty(
+        *db_property_info, DB::Properties::kDBStats, &stats);
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      if (cfd->initialized()) {
+        cfd->internal_stats()->GetStringProperty(
+            *cf_property_info, DB::Properties::kCFStatsNoFileHistogram, &stats);
       }
     }
-    ROCKS_LOG_WARN(immutable_db_options_.info_log,
-                   "------- DUMPING STATS -------");
-    ROCKS_LOG_WARN(immutable_db_options_.info_log, "%s", stats.c_str());
-    if (immutable_db_options_.dump_malloc_stats) {
-      stats.clear();
-      DumpMallocStats(&stats);
-      if (!stats.empty()) {
-        ROCKS_LOG_WARN(immutable_db_options_.info_log,
-                       "------- Malloc STATS -------");
-        ROCKS_LOG_WARN(immutable_db_options_.info_log, "%s", stats.c_str());
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      if (cfd->initialized()) {
+        cfd->internal_stats()->GetStringProperty(
+            *cf_property_info, DB::Properties::kCFFileHistogram, &stats);
       }
     }
+  }
+  TEST_SYNC_POINT("DBImpl::DumpStats:2");
+  ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                 "------- DUMPING STATS -------");
+  ROCKS_LOG_WARN(immutable_db_options_.info_log, "%s", stats.c_str());
+  if (immutable_db_options_.dump_malloc_stats) {
+    stats.clear();
+    DumpMallocStats(&stats);
+    if (!stats.empty()) {
+      ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                     "------- Malloc STATS -------");
+      ROCKS_LOG_WARN(immutable_db_options_.info_log, "%s", stats.c_str());
+    }
+  }
 #endif  // !ROCKSDB_LITE
 
-    PrintStatistics();
-  }
+  PrintStatistics();
 }
 
 void DBImpl::ScheduleBgLogWriterClose(JobContext* job_context) {
@@ -595,7 +730,6 @@ Status DBImpl::SetOptions(
   MutableCFOptions new_options;
   Status s;
   Status persist_options_status;
-  WriteThread::Writer w;
   SuperVersionContext sv_context(/* create_superversion */ true);
   {
     InstrumentedMutexLock l(&mutex_);
@@ -657,7 +791,6 @@ Status DBImpl::SetDBOptions(
   Status s;
   Status persist_options_status;
   bool wal_changed = false;
-  WriteThread::Writer w;
   WriteContext write_context;
   {
     InstrumentedMutexLock l(&mutex_);
@@ -670,7 +803,22 @@ Status DBImpl::SetDBOptions(
             new_options.max_background_compactions, Env::Priority::LOW);
         MaybeScheduleFlushOrCompaction();
       }
-
+      if (new_options.stats_dump_period_sec !=
+          mutable_db_options_.stats_dump_period_sec) {
+          if (thread_dump_stats_) {
+            mutex_.Unlock();
+            thread_dump_stats_->cancel();
+            mutex_.Lock();
+          }
+          if (new_options.stats_dump_period_sec > 0) {
+            thread_dump_stats_.reset(new rocksdb::RepeatableThread(
+                [this]() { DBImpl::DumpStats(); }, "dump_st", env_,
+                new_options.stats_dump_period_sec * 1000000));
+          }
+          else {
+            thread_dump_stats_.reset();
+          }
+      }
       write_controller_.set_max_delayed_write_rate(
           new_options.delayed_write_rate);
       table_cache_.get()->SetCapacity(new_options.max_open_files == -1
@@ -691,6 +839,7 @@ Status DBImpl::SetDBOptions(
           env_options_for_compaction_, immutable_db_options_);
       env_options_for_compaction_.compaction_readahead_size =
           mutable_db_options_.compaction_readahead_size;
+      WriteThread::Writer w;
       write_thread_.EnterUnbatched(&w, &mutex_);
       if (total_log_size_ > GetMaxTotalWalSize() || wal_changed) {
         Status purge_wal_status = SwitchWAL(&write_context);
@@ -883,7 +1032,7 @@ bool DBImpl::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) {
 }
 
 InternalIterator* DBImpl::NewInternalIterator(
-    Arena* arena, RangeDelAggregator* range_del_agg,
+    Arena* arena, RangeDelAggregator* range_del_agg, SequenceNumber sequence,
     ColumnFamilyHandle* column_family) {
   ColumnFamilyData* cfd;
   if (column_family == nullptr) {
@@ -897,8 +1046,8 @@ InternalIterator* DBImpl::NewInternalIterator(
   SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
   mutex_.Unlock();
   ReadOptions roptions;
-  return NewInternalIterator(roptions, cfd, super_version, arena,
-                             range_del_agg);
+  return NewInternalIterator(roptions, cfd, super_version, arena, range_del_agg,
+                             sequence);
 }
 
 void DBImpl::SchedulePurge() {
@@ -1000,10 +1149,12 @@ static void CleanupIteratorState(void* arg1, void* /*arg2*/) {
 }
 }  // namespace
 
-InternalIterator* DBImpl::NewInternalIterator(
-    const ReadOptions& read_options, ColumnFamilyData* cfd,
-    SuperVersion* super_version, Arena* arena,
-    RangeDelAggregator* range_del_agg) {
+InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
+                                              ColumnFamilyData* cfd,
+                                              SuperVersion* super_version,
+                                              Arena* arena,
+                                              RangeDelAggregator* range_del_agg,
+                                              SequenceNumber sequence) {
   InternalIterator* internal_iter;
   assert(arena != nullptr);
   assert(range_del_agg != nullptr);
@@ -1015,12 +1166,12 @@ InternalIterator* DBImpl::NewInternalIterator(
   // Collect iterator for mutable mem
   merge_iter_builder.AddIterator(
       super_version->mem->NewIterator(read_options, arena));
-  std::unique_ptr<InternalIterator> range_del_iter;
+  std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter;
   Status s;
   if (!read_options.ignore_range_deletions) {
     range_del_iter.reset(
-        super_version->mem->NewRangeTombstoneIterator(read_options));
-    s = range_del_agg->AddTombstones(std::move(range_del_iter));
+        super_version->mem->NewRangeTombstoneIterator(read_options, sequence));
+    range_del_agg->AddTombstones(std::move(range_del_iter));
   }
   // Collect all needed child iterators for immutable memtables
   if (s.ok()) {
@@ -1065,6 +1216,7 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
                        PinnableSlice* pinnable_val, bool* value_found,
                        ReadCallback* callback, bool* is_blob_index) {
   assert(pinnable_val != nullptr);
+  PERF_CPU_TIMER_GUARD(get_cpu_nanos, env_);
   StopWatch sw(env_, stats_, DB_GET);
   PERF_TIMER_GUARD(get_snapshot_time);
 
@@ -1121,7 +1273,7 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
 
   // Prepare to store a list of merge operations if merge occurs.
   MergeContext merge_context;
-  RangeDelAggregator range_del_agg(cfd->internal_comparator(), snapshot);
+  SequenceNumber max_covering_tombstone_seq = 0;
 
   Status s;
   // First look in the memtable, then in the immutable memtable (if any).
@@ -1135,13 +1287,14 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
   bool done = false;
   if (!skip_memtable) {
     if (sv->mem->Get(lkey, pinnable_val->GetSelf(), &s, &merge_context,
-                     &range_del_agg, read_options, callback, is_blob_index)) {
+                     &max_covering_tombstone_seq, read_options, callback,
+                     is_blob_index)) {
       done = true;
       pinnable_val->PinSelf();
       RecordTick(stats_, MEMTABLE_HIT);
     } else if ((s.ok() || s.IsMergeInProgress()) &&
                sv->imm->Get(lkey, pinnable_val->GetSelf(), &s, &merge_context,
-                            &range_del_agg, read_options, callback,
+                            &max_covering_tombstone_seq, read_options, callback,
                             is_blob_index)) {
       done = true;
       pinnable_val->PinSelf();
@@ -1155,8 +1308,8 @@ Status DBImpl::GetImpl(const ReadOptions& read_options,
   if (!done) {
     PERF_TIMER_GUARD(get_from_output_files_time);
     sv->current->Get(read_options, lkey, pinnable_val, &s, &merge_context,
-                     &range_del_agg, value_found, nullptr, nullptr, callback,
-                     is_blob_index);
+                     &max_covering_tombstone_seq, value_found, nullptr, nullptr,
+                     callback, is_blob_index);
     RecordTick(stats_, MEMTABLE_MISS);
   }
 
@@ -1178,6 +1331,7 @@ std::vector<Status> DBImpl::MultiGet(
     const ReadOptions& read_options,
     const std::vector<ColumnFamilyHandle*>& column_family,
     const std::vector<Slice>& keys, std::vector<std::string>* values) {
+  PERF_CPU_TIMER_GUARD(get_cpu_nanos, env_);
   StopWatch sw(env_, stats_, DB_MULTIGET);
   PERF_TIMER_GUARD(get_snapshot_time);
 
@@ -1186,33 +1340,96 @@ std::vector<Status> DBImpl::MultiGet(
   struct MultiGetColumnFamilyData {
     ColumnFamilyData* cfd;
     SuperVersion* super_version;
+    MultiGetColumnFamilyData(ColumnFamilyData* cf, SuperVersion* sv)
+        : cfd(cf), super_version(sv) {}
   };
-  std::unordered_map<uint32_t, MultiGetColumnFamilyData*> multiget_cf_data;
-  // fill up and allocate outside of mutex
+  std::unordered_map<uint32_t, MultiGetColumnFamilyData> multiget_cf_data(
+      column_family.size());
   for (auto cf : column_family) {
     auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(cf);
     auto cfd = cfh->cfd();
     if (multiget_cf_data.find(cfd->GetID()) == multiget_cf_data.end()) {
-      auto mgcfd = new MultiGetColumnFamilyData();
-      mgcfd->cfd = cfd;
-      multiget_cf_data.insert({cfd->GetID(), mgcfd});
+      multiget_cf_data.emplace(cfd->GetID(),
+                               MultiGetColumnFamilyData(cfd, nullptr));
     }
   }
 
-  mutex_.Lock();
-  if (read_options.snapshot != nullptr) {
-    snapshot =
-        reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
-  } else {
-    snapshot = last_seq_same_as_publish_seq_
-                   ? versions_->LastSequence()
-                   : versions_->LastPublishedSequence();
+  bool last_try = false;
+  {
+    // If we end up with the same issue of memtable geting sealed during 2
+    // consecutive retries, it means the write rate is very high. In that case
+    // its probably ok to take the mutex on the 3rd try so we can succeed for
+    // sure
+    static const int num_retries = 3;
+    for (auto i = 0; i < num_retries; ++i) {
+      last_try = (i == num_retries - 1);
+      bool retry = false;
+
+      if (i > 0) {
+        for (auto mgd_iter = multiget_cf_data.begin();
+             mgd_iter != multiget_cf_data.end(); ++mgd_iter) {
+          auto super_version = mgd_iter->second.super_version;
+          auto cfd = mgd_iter->second.cfd;
+          if (super_version != nullptr) {
+            ReturnAndCleanupSuperVersion(cfd, super_version);
+          }
+          mgd_iter->second.super_version = nullptr;
+        }
+      }
+
+      if (read_options.snapshot == nullptr) {
+        if (last_try) {
+          TEST_SYNC_POINT("DBImpl::MultiGet::LastTry");
+          // We're close to max number of retries. For the last retry,
+          // acquire the lock so we're sure to succeed
+          mutex_.Lock();
+        }
+        snapshot = last_seq_same_as_publish_seq_
+                       ? versions_->LastSequence()
+                       : versions_->LastPublishedSequence();
+      } else {
+        snapshot = reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)
+                       ->number_;
+      }
+
+      for (auto mgd_iter = multiget_cf_data.begin();
+           mgd_iter != multiget_cf_data.end(); ++mgd_iter) {
+        if (!last_try) {
+          mgd_iter->second.super_version =
+              GetAndRefSuperVersion(mgd_iter->second.cfd);
+        } else {
+          mgd_iter->second.super_version =
+              mgd_iter->second.cfd->GetSuperVersion()->Ref();
+        }
+        TEST_SYNC_POINT("DBImpl::MultiGet::AfterRefSV");
+        if (read_options.snapshot != nullptr || last_try) {
+          // If user passed a snapshot, then we don't care if a memtable is
+          // sealed or compaction happens because the snapshot would ensure
+          // that older key versions are kept around. If this is the last
+          // retry, then we have the lock so nothing bad can happen
+          continue;
+        }
+        // We could get the earliest sequence number for the whole list of
+        // memtables, which will include immutable memtables as well, but that
+        // might be tricky to maintain in case we decide, in future, to do
+        // memtable compaction.
+        if (!last_try) {
+          auto seq =
+              mgd_iter->second.super_version->mem->GetEarliestSequenceNumber();
+          if (seq > snapshot) {
+            retry = true;
+            break;
+          }
+        }
+      }
+      if (!retry) {
+        if (last_try) {
+          mutex_.Unlock();
+        }
+        break;
+      }
+    }
   }
-  for (auto mgd_iter : multiget_cf_data) {
-    mgd_iter.second->super_version =
-        mgd_iter.second->cfd->GetSuperVersion()->Ref();
-  }
-  mutex_.Unlock();
 
   // Contain a list of merge operations if merge occurs.
   MergeContext merge_context;
@@ -1238,23 +1455,23 @@ std::vector<Status> DBImpl::MultiGet(
 
     LookupKey lkey(keys[i], snapshot);
     auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family[i]);
-    RangeDelAggregator range_del_agg(cfh->cfd()->internal_comparator(),
-                                     snapshot);
+    SequenceNumber max_covering_tombstone_seq = 0;
     auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
     assert(mgd_iter != multiget_cf_data.end());
     auto mgd = mgd_iter->second;
-    auto super_version = mgd->super_version;
+    auto super_version = mgd.super_version;
     bool skip_memtable =
         (read_options.read_tier == kPersistedTier &&
          has_unpersisted_data_.load(std::memory_order_relaxed));
     bool done = false;
     if (!skip_memtable) {
       if (super_version->mem->Get(lkey, value, &s, &merge_context,
-                                  &range_del_agg, read_options)) {
+                                  &max_covering_tombstone_seq, read_options)) {
         done = true;
         RecordTick(stats_, MEMTABLE_HIT);
       } else if (super_version->imm->Get(lkey, value, &s, &merge_context,
-                                         &range_del_agg, read_options)) {
+                                         &max_covering_tombstone_seq,
+                                         read_options)) {
         done = true;
         RecordTick(stats_, MEMTABLE_HIT);
       }
@@ -1263,7 +1480,7 @@ std::vector<Status> DBImpl::MultiGet(
       PinnableSlice pinnable_val;
       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);
       value->assign(pinnable_val.data(), pinnable_val.size());
       RecordTick(stats_, MEMTABLE_MISS);
     }
@@ -1278,24 +1495,14 @@ std::vector<Status> DBImpl::MultiGet(
   PERF_TIMER_GUARD(get_post_process_time);
   autovector<SuperVersion*> superversions_to_delete;
 
-  // TODO(icanadi) do we need lock here or just around Cleanup()?
-  mutex_.Lock();
   for (auto mgd_iter : multiget_cf_data) {
     auto mgd = mgd_iter.second;
-    if (mgd->super_version->Unref()) {
-      mgd->super_version->Cleanup();
-      superversions_to_delete.push_back(mgd->super_version);
+    if (!last_try) {
+      ReturnAndCleanupSuperVersion(mgd.cfd, mgd.super_version);
+    } else {
+      mgd.cfd->GetSuperVersion()->Unref();
     }
   }
-  mutex_.Unlock();
-
-  for (auto td : superversions_to_delete) {
-    delete td;
-  }
-  for (auto mgd : multiget_cf_data) {
-    delete mgd.second;
-  }
-
   RecordTick(stats_, NUMBER_MULTIGET_CALLS);
   RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
   RecordTick(stats_, NUMBER_MULTIGET_KEYS_FOUND, num_found);
@@ -1691,7 +1898,7 @@ ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
 
   InternalIterator* internal_iter =
       NewInternalIterator(read_options, cfd, sv, db_iter->GetArena(),
-                          db_iter->GetRangeDelAggregator());
+                          db_iter->GetRangeDelAggregator(), snapshot);
   db_iter->SetIterUnderDBIter(internal_iter);
 
   return db_iter;
@@ -1714,7 +1921,7 @@ Status DBImpl::NewIterators(
   if (read_options.tailing) {
 #ifdef ROCKSDB_LITE
     return Status::InvalidArgument(
-        "Tailing interator not supported in RocksDB lite");
+        "Tailing iterator not supported in RocksDB lite");
 #else
     for (auto cfh : column_families) {
       auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
@@ -1753,21 +1960,32 @@ const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
 }
 #endif  // ROCKSDB_LITE
 
-SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
+SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
+                                      bool lock) {
   int64_t unix_time = 0;
   env_->GetCurrentTime(&unix_time);  // Ignore error
   SnapshotImpl* s = new SnapshotImpl;
 
-  InstrumentedMutexLock l(&mutex_);
+  if (lock) {
+    mutex_.Lock();
+  }
   // returns null if the underlying memtable does not support snapshot.
   if (!is_snapshot_supported_) {
+    if (lock) {
+      mutex_.Unlock();
+    }
     delete s;
     return nullptr;
   }
   auto snapshot_seq = last_seq_same_as_publish_seq_
                           ? versions_->LastSequence()
                           : versions_->LastPublishedSequence();
-  return snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
+  SnapshotImpl* snapshot =
+      snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
+  if (lock) {
+    mutex_.Unlock();
+  }
+  return snapshot;
 }
 
 void DBImpl::ReleaseSnapshot(const Snapshot* s) {
@@ -2062,17 +2280,18 @@ ColumnFamilyHandle* DBImpl::GetColumnFamilyHandle(uint32_t column_family_id) {
 }
 
 // REQUIRED: mutex is NOT held.
-ColumnFamilyHandle* DBImpl::GetColumnFamilyHandleUnlocked(
+std::unique_ptr<ColumnFamilyHandle> DBImpl::GetColumnFamilyHandleUnlocked(
     uint32_t column_family_id) {
-  ColumnFamilyMemTables* cf_memtables = column_family_memtables_.get();
-
   InstrumentedMutexLock l(&mutex_);
 
-  if (!cf_memtables->Seek(column_family_id)) {
+  auto* cfd =
+      versions_->GetColumnFamilySet()->GetColumnFamily(column_family_id);
+  if (cfd == nullptr) {
     return nullptr;
   }
 
-  return cf_memtables->GetColumnFamilyHandle();
+  return std::unique_ptr<ColumnFamilyHandleImpl>(
+      new ColumnFamilyHandleImpl(cfd, this, &mutex_));
 }
 
 void DBImpl::GetApproximateMemTableStats(ColumnFamilyHandle* column_family,
@@ -2143,9 +2362,8 @@ void DBImpl::ReleaseFileNumberFromPendingOutputs(
 
 #ifndef ROCKSDB_LITE
 Status DBImpl::GetUpdatesSince(
-    SequenceNumber seq, unique_ptr<TransactionLogIterator>* iter,
+    SequenceNumber seq, std::unique_ptr<TransactionLogIterator>* iter,
     const TransactionLogIterator::ReadOptions& read_options) {
-
   RecordTick(stats_, GET_UPDATES_SINCE_CALLS);
   if (seq > versions_->LastSequence()) {
     return Status::NotFound("Requested sequence not yet written in the db");
@@ -2235,9 +2453,9 @@ Status DBImpl::DeleteFile(std::string name) {
     status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                     &edit, &mutex_, directories_.GetDbDir());
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_contexts[0],
-          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(cfd,
+                                         &job_context.superversion_contexts[0],
+                                         *cfd->GetLatestMutableCFOptions());
     }
     FindObsoleteFiles(&job_context, false);
   }  // lock released here
@@ -2320,9 +2538,9 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
     status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
                                     &edit, &mutex_, directories_.GetDbDir());
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(
-          cfd, &job_context.superversion_contexts[0],
-          *cfd->GetLatestMutableCFOptions(), FlushReason::kDeleteFiles);
+      InstallSuperVersionAndScheduleWork(cfd,
+                                         &job_context.superversion_contexts[0],
+                                         *cfd->GetLatestMutableCFOptions());
     }
     for (auto* deleted_file : deleted_files) {
       deleted_file->being_compacted = false;
@@ -2393,10 +2611,10 @@ Status DBImpl::CheckConsistency() {
 Status DBImpl::GetDbIdentity(std::string& identity) const {
   std::string idfilename = IdentityFileName(dbname_);
   const EnvOptions soptions;
-  unique_ptr<SequentialFileReader> id_file_reader;
+  std::unique_ptr<SequentialFileReader> id_file_reader;
   Status s;
   {
-    unique_ptr<SequentialFile> idfile;
+    std::unique_ptr<SequentialFile> idfile;
     s = env_->NewSequentialFile(idfilename, &idfile, soptions);
     if (!s.ok()) {
       return s;
@@ -2410,7 +2628,7 @@ Status DBImpl::GetDbIdentity(std::string& identity) const {
   if (!s.ok()) {
     return s;
   }
-  char* buffer = reinterpret_cast<char*>(alloca(file_size));
+  char* buffer = reinterpret_cast<char*>(alloca(static_cast<size_t>(file_size)));
   Slice id;
   s = id_file_reader->Read(static_cast<size_t>(file_size), &id, buffer);
   if (!s.ok()) {
@@ -2712,11 +2930,15 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
 #ifndef ROCKSDB_LITE
   Status s;
 
-  versions_->options_file_number_ = versions_->NewFileNumber();
+  uint64_t options_file_number = versions_->NewFileNumber();
   std::string options_file_name =
-      OptionsFileName(GetName(), versions_->options_file_number_);
+      OptionsFileName(GetName(), options_file_number);
   // Retry if the file name happen to conflict with an existing one.
   s = GetEnv()->RenameFile(file_name, options_file_name);
+  if (s.ok()) {
+    InstrumentedMutexLock l(&mutex_);
+    versions_->options_file_number_ = options_file_number;
+  }
 
   if (0 == disable_delete_obsolete_files_) {
     DeleteObsoleteOptionsFiles();
@@ -2767,6 +2989,8 @@ void DumpRocksDBBuildVersion(Logger* log) {
                    ROCKSDB_MINOR, ROCKSDB_PATCH);
   ROCKS_LOG_HEADER(log, "Git sha %s", rocksdb_build_git_sha);
   ROCKS_LOG_HEADER(log, "Compile date %s", rocksdb_build_compile_date);
+#else
+    (void)log;  // ignore "-Wunused-parameter"
 #endif
 }
 
@@ -2793,8 +3017,7 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
                                        bool* is_blob_index) {
   Status s;
   MergeContext merge_context;
-  RangeDelAggregator range_del_agg(sv->mem->GetInternalKeyComparator(),
-                                   kMaxSequenceNumber);
+  SequenceNumber max_covering_tombstone_seq = 0;
 
   ReadOptions read_options;
   SequenceNumber current_seq = versions_->LastSequence();
@@ -2804,8 +3027,8 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
   *found_record_for_key = false;
 
   // Check if there is a record for this key in the latest memtable
-  sv->mem->Get(lkey, nullptr, &s, &merge_context, &range_del_agg, seq,
-               read_options, nullptr /*read_callback*/, is_blob_index);
+  sv->mem->Get(lkey, nullptr, &s, &merge_context, &max_covering_tombstone_seq,
+               seq, read_options, nullptr /*read_callback*/, is_blob_index);
 
   if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
     // unexpected error reading memtable.
@@ -2823,8 +3046,8 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
   }
 
   // Check if there is a record for this key in the immutable memtables
-  sv->imm->Get(lkey, nullptr, &s, &merge_context, &range_del_agg, seq,
-               read_options, nullptr /*read_callback*/, is_blob_index);
+  sv->imm->Get(lkey, nullptr, &s, &merge_context, &max_covering_tombstone_seq,
+               seq, read_options, nullptr /*read_callback*/, is_blob_index);
 
   if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
     // unexpected error reading memtable.
@@ -2842,8 +3065,9 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
   }
 
   // Check if there is a record for this key in the immutable memtables
-  sv->imm->GetFromHistory(lkey, nullptr, &s, &merge_context, &range_del_agg,
-                          seq, read_options, is_blob_index);
+  sv->imm->GetFromHistory(lkey, nullptr, &s, &merge_context,
+                          &max_covering_tombstone_seq, seq, read_options,
+                          is_blob_index);
 
   if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
     // unexpected error reading memtable.
@@ -2866,7 +3090,7 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
   if (!cache_only) {
     // Check tables
     sv->current->Get(read_options, lkey, nullptr, &s, &merge_context,
-                     &range_del_agg, nullptr /* value_found */,
+                     &max_covering_tombstone_seq, nullptr /* value_found */,
                      found_record_for_key, seq, nullptr /*read_callback*/,
                      is_blob_index);
 
@@ -2875,12 +3099,10 @@ Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
       ROCKS_LOG_ERROR(immutable_db_options_.info_log,
                       "Unexpected status returned from Version::Get: %s\n",
                       s.ToString().c_str());
-
-      return s;
     }
   }
 
-  return Status::OK();
+  return s;
 }
 
 Status DBImpl::IngestExternalFile(
@@ -2967,6 +3189,7 @@ Status DBImpl::IngestExternalFile(
     }
 
     num_running_ingest_file_++;
+    TEST_SYNC_POINT("DBImpl::IngestExternalFile:AfterIncIngestFileCounter");
 
     // We cannot ingest a file into a dropped CF
     if (cfd->IsDropped()) {
@@ -2981,10 +3204,21 @@ Status DBImpl::IngestExternalFile(
       TEST_SYNC_POINT_CALLBACK("DBImpl::IngestExternalFile:NeedFlush",
                                &need_flush);
       if (status.ok() && need_flush) {
-        mutex_.Unlock();
-        status = FlushMemTable(cfd, FlushOptions(),
-                               FlushReason::kExternalFileIngestion,
-                               true /* writes_stopped */);
+        FlushOptions flush_opts;
+        flush_opts.allow_write_stall = true;
+        if (immutable_db_options_.atomic_flush) {
+          autovector<ColumnFamilyData*> cfds;
+          SelectColumnFamiliesForAtomicFlush(&cfds);
+          mutex_.Unlock();
+          status = AtomicFlushMemTables(cfds, flush_opts,
+                                        FlushReason::kExternalFileIngestion,
+                                        true /* writes_stopped */);
+        } else {
+          mutex_.Unlock();
+          status = FlushMemTable(cfd, flush_opts,
+                                 FlushReason::kExternalFileIngestion,
+                                 true /* writes_stopped */);
+        }
         mutex_.Lock();
       }
     }
@@ -3002,8 +3236,7 @@ Status DBImpl::IngestExternalFile(
                                  &mutex_, directories_.GetDbDir());
     }
     if (status.ok()) {
-      InstallSuperVersionAndScheduleWork(cfd, &sv_context, *mutable_cf_options,
-                                         FlushReason::kExternalFileIngestion);
+      InstallSuperVersionAndScheduleWork(cfd, &sv_context, *mutable_cf_options);
     }
 
     // Resume writes to the DB
@@ -3118,10 +3351,10 @@ void DBImpl::WaitForIngestFile() {
   }
 }
 
-Status DBImpl::StartTrace(const TraceOptions& /* options */,
+Status DBImpl::StartTrace(const TraceOptions& trace_options,
                           std::unique_ptr<TraceWriter>&& trace_writer) {
   InstrumentedMutexLock lock(&trace_mutex_);
-  tracer_.reset(new Tracer(env_, std::move(trace_writer)));
+  tracer_.reset(new Tracer(env_, trace_options, std::move(trace_writer)));
   return Status::OK();
 }
 
@@ -3143,7 +3376,8 @@ Status DBImpl::TraceIteratorSeek(const uint32_t& cf_id, const Slice& key) {
   return s;
 }
 
-Status DBImpl::TraceIteratorSeekForPrev(const uint32_t& cf_id, const Slice& key) {
+Status DBImpl::TraceIteratorSeekForPrev(const uint32_t& cf_id,
+                                        const Slice& key) {
   Status s;
   if (tracer_) {
     InstrumentedMutexLock lock(&trace_mutex_);

db/db_impl.h

@@ -22,15 +22,16 @@
 #include "db/column_family.h"
 #include "db/compaction_job.h"
 #include "db/dbformat.h"
-#include "db/external_sst_file_ingestion_job.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"
 #include "db/internal_stats.h"
 #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"
@@ -53,6 +54,7 @@
 #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"
@@ -63,6 +65,7 @@ class Arena;
 class ArenaWrappedDBIter;
 class MemTable;
 class TableCache;
+class TaskLimiterToken;
 class Version;
 class VersionEdit;
 class VersionSet;
@@ -189,13 +192,13 @@ class DBImpl : public DB {
                               const Slice* begin, const Slice* end) override;
 
   using DB::CompactFiles;
-  virtual Status CompactFiles(const CompactionOptions& compact_options,
-                              ColumnFamilyHandle* column_family,
-                              const std::vector<std::string>& input_file_names,
-                              const int output_level,
-                              const int output_path_id = -1,
-                              std::vector<std::string>* const output_file_names
-                              = nullptr) override;
+  virtual Status CompactFiles(
+      const CompactionOptions& compact_options,
+      ColumnFamilyHandle* column_family,
+      const std::vector<std::string>& input_file_names, const int output_level,
+      const int output_path_id = -1,
+      std::vector<std::string>* const output_file_names = nullptr,
+      CompactionJobInfo* compaction_job_info = nullptr) override;
 
   virtual Status PauseBackgroundWork() override;
   virtual Status ContinueBackgroundWork() override;
@@ -227,11 +230,21 @@ 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;
 
   virtual SequenceNumber GetLatestSequenceNumber() const override;
+  virtual SequenceNumber GetLastPublishedSequence() const {
+    if (last_seq_same_as_publish_seq_) {
+      return versions_->LastSequence();
+    } else {
+      return versions_->LastPublishedSequence();
+    }
+  }
   // REQUIRES: joined the main write queue if two_write_queues is disabled, and
   // the second write queue otherwise.
   virtual void SetLastPublishedSequence(SequenceNumber seq);
@@ -255,9 +268,9 @@ 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,
@@ -369,7 +382,7 @@ 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() {
@@ -464,6 +477,8 @@ 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
 
@@ -485,6 +500,14 @@ 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
@@ -544,7 +567,8 @@ 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.
@@ -562,11 +586,9 @@ 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
@@ -678,6 +700,12 @@ class DBImpl : public DB {
 
   void SetSnapshotChecker(SnapshotChecker* snapshot_checker);
 
+  // Fill JobContext with snapshot information needed by flush and compaction.
+  void GetSnapshotContext(JobContext* job_context,
+                          std::vector<SequenceNumber>* snapshot_seqs,
+                          SequenceNumber* earliest_write_conflict_snapshot,
+                          SnapshotChecker** snapshot_checker);
+
   // Not thread-safe.
   void SetRecoverableStatePreReleaseCallback(PreReleaseCallback* callback);
 
@@ -699,7 +727,7 @@ 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_;
@@ -731,6 +759,11 @@ 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,
@@ -795,6 +828,7 @@ class DBImpl : public DB {
 
  private:
   friend class DB;
+  friend class ErrorHandler;
   friend class InternalStats;
   friend class PessimisticTransaction;
   friend class TransactionBaseImpl;
@@ -846,6 +880,8 @@ 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();
@@ -885,11 +921,46 @@ 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.
-  Status FlushMemTableToOutputFile(ColumnFamilyData* cfd,
-                                   const MutableCFOptions& mutable_cf_options,
-                                   bool* madeProgress, JobContext* job_context,
-                                   LogBuffer* log_buffer);
+  // 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,
+      std::vector<SequenceNumber>& snapshot_seqs,
+      SequenceNumber earliest_write_conflict_snapshot,
+      SnapshotChecker* snapshot_checker, 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,
@@ -903,6 +974,12 @@ 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);
@@ -912,13 +989,19 @@ 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);
@@ -927,8 +1010,22 @@ class DBImpl : public DB {
   // 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);
@@ -969,7 +1066,8 @@ class DBImpl : public DB {
                           const std::vector<std::string>& input_file_names,
                           std::vector<std::string>* const output_file_names,
                           const int output_level, int output_path_id,
-                          JobContext* job_context, LogBuffer* log_buffer);
+                          JobContext* job_context, LogBuffer* log_buffer,
+                          CompactionJobInfo* compaction_job_info);
 
   // Wait for current IngestExternalFile() calls to finish.
   // REQUIRES: mutex_ held
@@ -984,7 +1082,20 @@ 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);
@@ -1003,15 +1114,25 @@ 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);
 
+  // Request compaction tasks token from compaction thread limiter.
+  // It always succeeds if force = true or limiter is disable.
+  bool RequestCompactionToken(ColumnFamilyData* cfd, bool force,
+                              std::unique_ptr<TaskLimiterToken>* token,
+                              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.
@@ -1026,13 +1147,17 @@ 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();
+
+  // Pick the first unthrottled compaction with task token from queue.
+  ColumnFamilyData* PickCompactionFromQueue(
+      std::unique_ptr<TaskLimiterToken>* token, LogBuffer* log_buffer);
 
   // helper function to call after some of the logs_ were synced
   void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);
 
-  SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary);
+  SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary,
+                                bool lock = true);
 
   uint64_t GetMaxTotalWalSize() const;
 
@@ -1040,6 +1165,8 @@ class DBImpl : public DB {
 
   Status CloseHelper();
 
+  void WaitForBackgroundWork();
+
   // table_cache_ provides its own synchronization
   std::shared_ptr<Cache> table_cache_;
 
@@ -1090,7 +1217,7 @@ 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) {}
@@ -1118,7 +1245,7 @@ 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;
@@ -1204,7 +1331,7 @@ 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.
@@ -1253,14 +1380,14 @@ class DBImpl : public DB {
   // compacted. Consumers of these queues are flush and compaction threads. When
   // column family is put on this queue, we increase unscheduled_flushes_ and
   // unscheduled_compactions_. When these variables are bigger than zero, that
-  // means we need to schedule background threads for compaction and thread.
+  // means we need to schedule background threads for flush and compaction.
   // Once the background threads are scheduled, we decrease unscheduled_flushes_
   // and unscheduled_compactions_. That way we keep track of number of
   // compaction and flush threads we need to schedule. This scheduling is done
   // 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_;
@@ -1320,6 +1447,8 @@ class DBImpl : public DB {
     // caller retains ownership of `manual_compaction_state` as it is reused
     // across background compactions.
     ManualCompactionState* manual_compaction_state;  // nullptr if non-manual
+    // task limiter token is requested during compaction picking.
+    std::unique_ptr<TaskLimiterToken> task_token;
   };
   std::deque<ManualCompactionState*> manual_compaction_dequeue_;
 
@@ -1359,7 +1488,7 @@ class DBImpl : public DB {
   std::atomic<bool> has_unpersisted_data_;
 
   // if an attempt was made to flush all column families that
-  // the oldest log depends on but uncommited data in the oldest
+  // the oldest log depends on but uncommitted data in the oldest
   // log prevents the log from being released.
   // We must attempt to free the dependent memtables again
   // at a later time after the transaction in the oldest
@@ -1411,6 +1540,10 @@ 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&);
@@ -1424,8 +1557,7 @@ class DBImpl : public DB {
   // state needs flush or compaction.
   void InstallSuperVersionAndScheduleWork(
       ColumnFamilyData* cfd, SuperVersionContext* sv_context,
-      const MutableCFOptions& mutable_cf_options,
-      FlushReason flush_reason = FlushReason::kOthers);
+      const MutableCFOptions& mutable_cf_options);
 
 #ifndef ROCKSDB_LITE
   using DB::GetPropertiesOfAllTables;
@@ -1450,6 +1582,13 @@ class DBImpl : public DB {
   bool ShouldntRunManualCompaction(ManualCompactionState* m);
   bool HaveManualCompaction(ColumnFamilyData* cfd);
   bool MCOverlap(ManualCompactionState* m, ManualCompactionState* m1);
+#ifndef ROCKSDB_LITE
+  void BuildCompactionJobInfo(const ColumnFamilyData* cfd, Compaction* c,
+                              const Status& st,
+                              const CompactionJobStats& compaction_job_stats,
+                              const int job_id, const Version* current,
+                              CompactionJobInfo* compaction_job_info) const;
+#endif
 
   bool ShouldPurge(uint64_t file_number) const;
   void MarkAsGrabbedForPurge(uint64_t file_number);
@@ -1484,6 +1623,16 @@ 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
@@ -1496,6 +1645,17 @@ class DBImpl : public 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

@@ -123,7 +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) {
@@ -137,7 +137,7 @@ Status DBImpl::TEST_WaitForCompact(bool wait_unscheduled) {
   while ((bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
           bg_flush_scheduled_ ||
           (wait_unscheduled && unscheduled_compactions_)) &&
-         !error_handler_.IsDBStopped()) {
+         (error_handler_.GetBGError() == Status::OK())) {
     bg_cv_.Wait();
   }
   return error_handler_.GetBGError();
@@ -237,5 +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_files.cc

@@ -20,6 +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();
@@ -28,6 +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
@@ -456,7 +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

@@ -23,8 +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 =
@@ -42,6 +41,8 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
       max_max_open_files = 0x400000;
     }
     ClipToRange(&result.max_open_files, 20, max_max_open_files);
+    TEST_SYNC_POINT_CALLBACK("SanitizeOptions::AfterChangeMaxOpenFiles",
+                             &result.max_open_files);
   }
 
   if (result.info_log == nullptr) {
@@ -56,10 +57,9 @@ 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,
@@ -107,14 +107,12 @@ 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;
   }
 
@@ -134,8 +132,15 @@ 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;
 }
 
@@ -211,7 +216,7 @@ static Status ValidateOptions(
 
   return Status::OK();
 }
-} // namespace
+}  // namespace
 Status DBImpl::NewDB() {
   VersionEdit new_db;
   new_db.SetLogNumber(0);
@@ -223,7 +228,7 @@ 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()) {
@@ -231,8 +236,9 @@ 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, env_, nullptr /* stats */,
+        immutable_db_options_.listeners));
     log::Writer log(std::move(file_writer), 0, false);
     std::string record;
     new_db.EncodeTo(&record);
@@ -250,9 +256,8 @@ 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
@@ -333,8 +338,8 @@ 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
@@ -354,7 +359,7 @@ 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;
@@ -389,6 +394,16 @@ 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(
@@ -406,7 +421,10 @@ Status DBImpl::Recover(
     // produced by an older version of rocksdb.
     std::vector<std::string> filenames;
     s = env_->GetChildren(immutable_db_options_.wal_dir, &filenames);
-    if (!s.ok()) {
+    if (s.IsNotFound()) {
+      return Status::InvalidArgument("wal_dir not found",
+                                     immutable_db_options_.wal_dir);
+    } else if (!s.ok()) {
       return s;
     }
 
@@ -461,12 +479,26 @@ 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;
@@ -517,10 +549,9 @@ 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,
@@ -563,9 +594,9 @@ 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()) {
@@ -598,8 +629,8 @@ 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
@@ -870,8 +901,8 @@ 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;
@@ -879,18 +910,8 @@ 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"
@@ -899,6 +920,60 @@ 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();
@@ -940,11 +1015,17 @@ 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),
@@ -978,8 +1059,8 @@ 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;
 }
@@ -1051,6 +1132,12 @@ 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()) {
@@ -1069,16 +1156,16 @@ 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(
@@ -1086,8 +1173,10 @@ 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,
+                                   impl->env_, nullptr /* stats */, listeners));
         impl->logs_.emplace_back(
             new_log_number,
             new log::Writer(
@@ -1164,7 +1253,8 @@ 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;
@@ -1214,6 +1304,14 @@ 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
 
@@ -1229,6 +1327,9 @@ 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 {
@@ -31,6 +30,8 @@ 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);
@@ -43,16 +44,24 @@ Status DBImplReadOnly::Get(const ReadOptions& read_options,
   }
   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;
 }
 
@@ -62,18 +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;
 }
@@ -89,21 +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

@@ -77,8 +77,8 @@ class DBImplReadOnly : public DBImpl {
       ColumnFamilyHandle* /*column_family*/,
       const std::vector<std::string>& /*input_file_names*/,
       const int /*output_level*/, const int /*output_path_id*/ = -1,
-      std::vector<std::string>* const /*output_file_names*/ = nullptr
-      ) override {
+      std::vector<std::string>* const /*output_file_names*/ = nullptr,
+      CompactionJobInfo* /*compaction_job_info*/ = nullptr) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
 
@@ -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

@@ -338,8 +338,6 @@ 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;
 
@@ -477,7 +475,7 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
 
     PERF_TIMER_STOP(write_pre_and_post_process_time);
 
-    if (w.ShouldWriteToWAL()) {
+    if (w.status.ok() && !write_options.disableWAL) {
       PERF_TIMER_GUARD(write_wal_time);
       stats->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
       RecordTick(stats_, WRITE_DONE_BY_SELF, 1);
@@ -506,7 +504,7 @@ Status DBImpl::PipelinedWriteImpl(const WriteOptions& write_options,
   WriteThread::WriteGroup memtable_write_group;
   if (w.state == WriteThread::STATE_MEMTABLE_WRITER_LEADER) {
     PERF_TIMER_GUARD(write_memtable_time);
-    assert(w.status.ok());
+    assert(w.ShouldWriteToMemtable());
     write_thread_.EnterAsMemTableWriter(&w, &memtable_write_group);
     if (memtable_write_group.size > 1 &&
         immutable_db_options_.allow_concurrent_memtable_write) {
@@ -710,6 +708,10 @@ 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_ ||
@@ -728,10 +730,6 @@ Status DBImpl::PreprocessWrite(const WriteOptions& write_options,
     status = HandleWriteBufferFull(write_context);
   }
 
-  if (UNLIKELY(status.ok())) {
-    status = error_handler_.GetBGError();
-  }
-
   if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
     status = ScheduleFlushes(write_context);
   }
@@ -1016,6 +1014,28 @@ 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);
@@ -1028,22 +1048,22 @@ Status DBImpl::SwitchWAL(WriteContext* write_context) {
   auto oldest_alive_log = alive_log_files_.begin()->number;
   bool flush_wont_release_oldest_log = false;
   if (allow_2pc()) {
-    auto oldest_log_with_uncommited_prep =
+    auto oldest_log_with_uncommitted_prep =
         logs_with_prep_tracker_.FindMinLogContainingOutstandingPrep();
 
-    assert(oldest_log_with_uncommited_prep == 0 ||
-           oldest_log_with_uncommited_prep >= oldest_alive_log);
-    if (oldest_log_with_uncommited_prep > 0 &&
-        oldest_log_with_uncommited_prep == oldest_alive_log) {
+    assert(oldest_log_with_uncommitted_prep == 0 ||
+           oldest_log_with_uncommitted_prep >= oldest_alive_log);
+    if (oldest_log_with_uncommitted_prep > 0 &&
+        oldest_log_with_uncommitted_prep == oldest_alive_log) {
       if (unable_to_release_oldest_log_) {
         // we already attempted to flush all column families dependent on
-        // the oldest alive log but the log still contained uncommited
+        // the oldest alive log but the log still contained uncommitted
         // transactions so there is still nothing that we can do.
         return status;
       } else {
         ROCKS_LOG_WARN(
             immutable_db_options_.info_log,
-            "Unable to release oldest log due to uncommited transaction");
+            "Unable to release oldest log due to uncommitted transaction");
         unable_to_release_oldest_log_ = true;
         flush_wont_release_oldest_log = true;
       }
@@ -1064,20 +1084,39 @@ 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;
 }
 
@@ -1099,32 +1138,55 @@ 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;
 }
 
@@ -1173,7 +1235,11 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
       write_thread_.EndWriteStall();
     }
 
-    while (!error_handler_.IsDBStopped() && 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("Write stall");
       }
@@ -1194,7 +1260,19 @@ Status DBImpl::DelayWrite(uint64_t num_bytes,
     RecordTick(stats_, STALL_MICROS, time_delayed);
   }
 
-  return error_handler_.GetBGError();
+  // 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,
@@ -1228,17 +1306,42 @@ 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
@@ -1259,8 +1362,7 @@ 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_) {
@@ -1269,7 +1371,7 @@ 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;
 
@@ -1329,6 +1431,8 @@ 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);
@@ -1336,14 +1440,12 @@ 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
@@ -1353,8 +1455,9 @@ 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, env_, 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_);
@@ -1432,7 +1535,7 @@ 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

@@ -134,15 +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;
@@ -158,9 +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_) {
@@ -174,7 +171,7 @@ class DBIter final: public Iterator {
     iter_ = iter;
     iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
   }
-  virtual RangeDelAggregator* GetRangeDelAggregator() {
+  virtual ReadRangeDelAggregator* GetRangeDelAggregator() {
     return &range_del_agg_;
   }
 
@@ -232,7 +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;
@@ -344,7 +341,7 @@ 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_;
@@ -1115,7 +1112,7 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
 
     if (ikey.type == kTypeDeletion || ikey.type == kTypeSingleDeletion ||
         range_del_agg_.ShouldDelete(
-            ikey, RangeDelPositioningMode::kBackwardTraversal)) {
+            ikey, RangeDelPositioningMode::kForwardTraversal)) {
       break;
     } else if (ikey.type == kTypeValue) {
       const Slice val = iter_->value();
@@ -1482,7 +1479,7 @@ Iterator* NewDBIterator(Env* env, const ReadOptions& read_options,
 
 ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
 
-RangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
+ReadRangeDelAggregator* ArenaWrappedDBIter::GetRangeDelAggregator() {
   return db_iter_->GetRangeDelAggregator();
 }
 
@@ -1558,7 +1555,8 @@ Status ArenaWrappedDBIter::Refresh() {
          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);

db/db_iter.h

@@ -48,7 +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.

db/db_iterator_test.cc

@@ -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));
@@ -1951,6 +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;
@@ -1959,7 +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);
 
@@ -2373,7 +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());
@@ -2392,7 +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());
@@ -2420,7 +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).
@@ -2430,7 +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");
   }
 

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

@@ -21,7 +21,8 @@ class TestReadCallback : public ReadCallback {
       : snapshot_checker_(snapshot_checker), snapshot_seq_(snapshot_seq) {}
 
   bool IsVisible(SequenceNumber seq) override {
-    return snapshot_checker_->IsInSnapshot(seq, snapshot_seq_);
+    return snapshot_checker_->CheckInSnapshot(seq, snapshot_seq_) ==
+           SnapshotCheckerResult::kInSnapshot;
   }
 
  private:
@@ -547,8 +548,15 @@ TEST_F(DBMergeOperatorTest, SnapshotCheckerAndReadCallback) {
   DestroyAndReopen(options);
 
   class TestSnapshotChecker : public SnapshotChecker {
-    bool IsInSnapshot(SequenceNumber seq,
-                      SequenceNumber snapshot_seq) const override {
+   public:
+    SnapshotCheckerResult CheckInSnapshot(
+        SequenceNumber seq, SequenceNumber snapshot_seq) const override {
+      return IsInSnapshot(seq, snapshot_seq)
+                 ? SnapshotCheckerResult::kInSnapshot
+                 : SnapshotCheckerResult::kNotInSnapshot;
+    }
+
+    bool IsInSnapshot(SequenceNumber seq, SequenceNumber snapshot_seq) const {
       switch (snapshot_seq) {
         case 0:
           return seq == 0;

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 +170,8 @@ 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;
 }
 
@@ -179,17 +181,19 @@ void ParseTablePropertiesString(std::string tp_string, TableProperties* tp) {
   std::replace(tp_string.begin(), tp_string.end(), '=', ' ');
   ResetTableProperties(tp);
   sscanf(tp_string.c_str(),
-         "# data blocks %" SCNu64 " # entries %" SCNu64
-         " # range deletions %" 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 (user-key? %" SCNu64
          ", delta-value? %" SCNu64 ") %" SCNu64 " filter block size %" SCNu64,
-         &tp->num_data_blocks, &tp->num_entries, &tp->num_range_deletions,
-         &tp->raw_key_size, &dummy_double, &tp->raw_value_size, &dummy_double,
-         &tp->data_size, &tp->index_key_is_user_key,
-         &tp->index_value_is_delta_encoded, &tp->index_size, &tp->filter_size);
+         &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);
 }
 
 void VerifySimilar(uint64_t a, uint64_t b, double bias) {
@@ -217,28 +221,42 @@ 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 kRangeDeletionsPerTable,
+    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 kKeyCount = kTableCount * kKeysPerTable;
+  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 + kRangeDeletionCount;
   const int kAvgSuccessorSize = kKeySize / 5;
   const int kEncodingSavePerKey = kKeySize / 4;
-  expected_tp->raw_key_size = (kKeyCount + kRangeDeletionCount) * (kKeySize + 8);
-  expected_tp->raw_value_size = (kKeyCount + kRangeDeletionCount) * kValueSize;
+  expected_tp->raw_key_size = kKeyCount * (kKeySize + 8);
+  expected_tp->raw_value_size =
+      (kPutCount + kMergeCount + kRangeDeletionCount) * kValueSize;
   expected_tp->num_entries = kKeyCount;
+  expected_tp->num_deletions = kDeletionCount + kRangeDeletionCount;
+  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));
@@ -298,8 +316,10 @@ TEST_F(DBPropertiesTest, ValidateSampleNumber) {
 
 TEST_F(DBPropertiesTest, AggregatedTableProperties) {
   for (int kTableCount = 40; kTableCount <= 100; kTableCount += 30) {
+    const int kDeletionsPerTable = 5;
+    const int kMergeOperandsPerTable = 15;
     const int kRangeDeletionsPerTable = 5;
-    const int kKeysPerTable = 100;
+    const int kPutsPerTable = 100;
     const int kKeySize = 80;
     const int kValueSize = 200;
     const int kBloomBitsPerKey = 20;
@@ -308,6 +328,8 @@ 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(
@@ -323,10 +345,17 @@ TEST_F(DBPropertiesTest, AggregatedTableProperties) {
 
     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;
@@ -343,11 +372,11 @@ TEST_F(DBPropertiesTest, AggregatedTableProperties) {
     bool value_is_delta_encoded = output_tp.index_value_is_delta_encoded > 0;
 
     TableProperties expected_tp;
-    GetExpectedTableProperties(&expected_tp, kKeySize, kValueSize,
-                               kKeysPerTable, kRangeDeletionsPerTable,
-                               kTableCount, kBloomBitsPerKey,
-                               table_options.block_size, index_key_is_user_key,
-                               value_is_delta_encoded);
+    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);
   }
@@ -363,7 +392,15 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
   options.target_file_size_base = 98 << 10;
   options.max_write_buffer_number = 2;
   options.statistics = rocksdb::CreateDBStatistics();
-  options.max_open_files = 100;
+  options.max_open_files = 11;  // Make sure no proloading of table readers
+
+  // RocksDB sanitize max open files to at least 20. Modify it back.
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+        int* max_open_files = static_cast<int*>(arg);
+        *max_open_files = 11;
+      });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
   BlockBasedTableOptions table_options;
   table_options.no_block_cache = true;
@@ -411,12 +448,13 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
 
   // Reopen and issue iterating. See thee latency tracked
   ReopenWithColumnFamilies({"default", "pikachu"}, options);
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
   ASSERT_TRUE(dbfull()->GetProperty("rocksdb.cf-file-histogram", &prop));
   ASSERT_EQ(std::string::npos, prop.find("** Level 0 read latency histogram"));
   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()) {
     }
   }
@@ -469,8 +507,10 @@ TEST_F(DBPropertiesTest, ReadLatencyHistogramByLevel) {
 
 TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
   const int kTableCount = 100;
+  const int kDeletionsPerTable = 2;
+  const int kMergeOperandsPerTable = 2;
   const int kRangeDeletionsPerTable = 2;
-  const int kKeysPerTable = 10;
+  const int kPutsPerTable = 10;
   const int kKeySize = 50;
   const int kValueSize = 400;
   const int kMaxLevel = 7;
@@ -486,6 +526,8 @@ 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(
@@ -503,10 +545,17 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
   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;
@@ -528,6 +577,8 @@ 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);
@@ -541,12 +592,15 @@ TEST_F(DBPropertiesTest, AggregatedTablePropertiesAtLevel) {
     ASSERT_EQ(sum_tp.raw_value_size, tp.raw_value_size);
     ASSERT_EQ(sum_tp.num_data_blocks, tp.num_data_blocks);
     ASSERT_EQ(sum_tp.num_entries, tp.num_entries);
+    ASSERT_EQ(sum_tp.num_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, kRangeDeletionsPerTable, table,
-                                 kBloomBitsPerKey, table_options.block_size,
-                                 index_key_is_user_key, value_is_delta_encoded);
+      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(expected_tp, tp, 0.5, 0.4, 0.4, 0.25);
@@ -1483,6 +1537,64 @@ 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

@@ -1041,11 +1041,16 @@ TEST_F(DBRangeDelTest, RangeTombstoneEndKeyAsSstableUpperBound) {
     //   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)));
   }
 
   {
@@ -1103,6 +1108,395 @@ 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);
+
+  // Push dummy data to L3 so that our actual test files on L0-L2
+  // will not be considered "bottommost" level, otherwise compaction
+  // may prevent us from creating overlapping user keys
+  // as on the bottommost layer MergeHelper
+  ASSERT_OK(db_->Merge(WriteOptions(), "key", "dummy"));
+  ASSERT_OK(db_->Flush(FlushOptions()));
+  MoveFilesToLevel(3);
+
+  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());
+
+  dbfull()->TEST_CompactRange(0 /* level */, nullptr /* begin */,
+                              nullptr /* end */, nullptr /* column_family */,
+                              true /* disallow_trivial_move */);
+  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

@@ -385,6 +385,7 @@ TEST_F(DBSSTTest, RateLimitedDelete) {
 
   // Compaction will move the 4 files in L0 to trash and create 1 L1 file
   ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
   ASSERT_EQ("0,1", FilesPerLevel(0));
 
   uint64_t delete_start_time = env_->NowMicros();

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_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();
@@ -2313,6 +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;
@@ -2339,9 +2342,6 @@ static void GCThreadBody(void* arg) {
 
 }  // namespace
 
-#ifndef TRAVIS
-// Disable this test temporarily on Travis as it fails intermittently.
-// Github issue: #4151
 TEST_F(DBTest, GroupCommitTest) {
   do {
     Options options = CurrentOptions();
@@ -2351,7 +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
@@ -2604,8 +2606,8 @@ class ModelDB : public DB {
       ColumnFamilyHandle* /*column_family*/,
       const std::vector<std::string>& /*input_file_names*/,
       const int /*output_level*/, const int /*output_path_id*/ = -1,
-      std::vector<std::string>* const /*output_file_names*/ = nullptr
-      ) override {
+      std::vector<std::string>* const /*output_file_names*/ = nullptr,
+      CompactionJobInfo* /*compaction_job_info*/ = nullptr) override {
     return Status::NotSupported("Not supported operation.");
   }
 
@@ -2659,6 +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(); }
 
@@ -2682,7 +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");
@@ -3504,8 +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));
@@ -5641,41 +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) {
@@ -5696,6 +5687,41 @@ 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
@@ -5731,7 +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.
+    InstallFlushCallback();
     dbfull()->TEST_FlushMemTable(true, true);
+    WaitForFlush();
   }
   ASSERT_TRUE(dbfull()->TEST_write_controler().NeedsDelay());
   ASSERT_TRUE(listener->CheckCondition(WriteStallCondition::kDelayed));
@@ -5756,8 +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();
@@ -5766,7 +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.
+    InstallFlushCallback();
     dbfull()->TEST_FlushMemTable(true, true);
+    WaitForFlush();
   }
 
   // Wake up sleep task to enable compaction to run and waits
@@ -5787,7 +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.
+    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());
@@ -342,6 +342,7 @@ TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
     ASSERT_GE(cache->GetUsage(), 1024 * 1024);
     Close();
     options.write_buffer_manager.reset();
+    last_options_.write_buffer_manager.reset();
     ASSERT_LT(cache->GetUsage(), 1024 * 1024);
   }
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
@@ -454,6 +455,22 @@ 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 +1245,14 @@ 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 +1260,8 @@ 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 +1270,8 @@ 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 +1313,18 @@ 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();
 }
@@ -1314,11 +1346,49 @@ TEST_F(DBTest2, FirstSnapshotTest) {
   db_->ReleaseSnapshot(s1);
 }
 
-class PinL0IndexAndFilterBlocksTest : public DBTestBase,
-                                      public testing::WithParamInterface<bool> {
+#ifndef ROCKSDB_LITE
+TEST_F(DBTest2, DuplicateSnapshot) {
+  Options options;
+  options = CurrentOptions(options);
+  std::vector<const Snapshot*> snapshots;
+  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+  SequenceNumber oldest_ww_snap, first_ww_snap;
+
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+  snapshots.push_back(db_->GetSnapshot());
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+  snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
+  first_ww_snap = snapshots.back()->GetSequenceNumber();
+  Put("k", "v");  // inc seq
+  snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
+  snapshots.push_back(db_->GetSnapshot());
+  Put("k", "v");  // inc seq
+  snapshots.push_back(db_->GetSnapshot());
+
+  {
+    InstrumentedMutexLock l(dbi->mutex());
+    auto seqs = dbi->snapshots().GetAll(&oldest_ww_snap);
+    ASSERT_EQ(seqs.size(), 4);  // duplicates are not counted
+    ASSERT_EQ(oldest_ww_snap, first_ww_snap);
+  }
+
+  for (auto s : snapshots) {
+    db_->ReleaseSnapshot(s);
+  }
+}
+#endif  // ROCKSDB_LITE
+
+class PinL0IndexAndFilterBlocksTest
+    : public DBTestBase,
+      public testing::WithParamInterface<std::tuple<bool, bool>> {
  public:
   PinL0IndexAndFilterBlocksTest() : DBTestBase("/db_pin_l0_index_bloom_test") {}
-  virtual void SetUp() override { infinite_max_files_ = GetParam(); }
+  virtual void SetUp() override {
+    infinite_max_files_ = std::get<0>(GetParam());
+    disallow_preload_ = std::get<1>(GetParam());
+  }
 
   void CreateTwoLevels(Options* options, bool close_afterwards) {
     if (infinite_max_files_) {
@@ -1355,6 +1425,7 @@ class PinL0IndexAndFilterBlocksTest : public DBTestBase,
   }
 
   bool infinite_max_files_;
+  bool disallow_preload_;
 };
 
 TEST_P(PinL0IndexAndFilterBlocksTest,
@@ -1445,12 +1516,27 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
   uint64_t im = TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS);
   uint64_t ih = TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT);
 
+  if (disallow_preload_) {
+    // Now we have two files. We narrow the max open files to allow 3 entries
+    // so that preloading SST files won't happen.
+    options.max_open_files = 13;
+    // RocksDB sanitize max open files to at least 20. Modify it back.
+    rocksdb::SyncPoint::GetInstance()->SetCallBack(
+        "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+          int* max_open_files = static_cast<int*>(arg);
+          *max_open_files = 13;
+        });
+  }
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
   // Reopen database. If max_open_files is set as -1, table readers will be
   // preloaded. This will trigger a BlockBasedTable::Open() and prefetch
   // L0 index and filter. Level 1's prefetching is disabled in DB::Open()
   TryReopenWithColumnFamilies({"default", "pikachu"}, options);
 
-  if (infinite_max_files_) {
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+
+  if (!disallow_preload_) {
     // After reopen, cache miss are increased by one because we read (and only
     // read) filter and index on L0
     ASSERT_EQ(fm + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
@@ -1478,7 +1564,7 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
 
   // this should be read from L1
   value = Get(1, "a");
-  if (infinite_max_files_) {
+  if (!disallow_preload_) {
     // In inifinite max files case, there's a cache miss in executing Get()
     // because index and filter are not prefetched before.
     ASSERT_EQ(fm + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
@@ -1496,10 +1582,45 @@ TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
     ASSERT_EQ(im + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
     ASSERT_EQ(ih + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
   }
+
+  // Force a full compaction to one single file. There will be a block
+  // cache read for both of index and filter. If prefetch doesn't explicitly
+  // happen, it will happen when verifying the file.
+  Compact(1, "a", "zzzzz");
+  dbfull()->TEST_WaitForCompact();
+
+  if (!disallow_preload_) {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  } else {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  }
+
+  // Bloom and index hit will happen when a Get() happens.
+  value = Get(1, "a");
+  if (!disallow_preload_) {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  } else {
+    ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+    ASSERT_EQ(fh + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+    ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+    ASSERT_EQ(ih + 4, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
+  }
 }
 
 INSTANTIATE_TEST_CASE_P(PinL0IndexAndFilterBlocksTest,
-                        PinL0IndexAndFilterBlocksTest, ::testing::Bool());
+                        PinL0IndexAndFilterBlocksTest,
+                        ::testing::Values(std::make_tuple(true, false),
+                                          std::make_tuple(false, false),
+                                          std::make_tuple(false, true)));
 
 #ifndef ROCKSDB_LITE
 TEST_F(DBTest2, MaxCompactionBytesTest) {
@@ -1615,6 +1736,42 @@ class MockPersistentCache : public PersistentCache {
   const size_t max_size_ = 10 * 1024;  // 10KiB
 };
 
+#ifdef OS_LINUX
+// Make sure that in CPU time perf context counters, Env::NowCPUNanos()
+// is used, rather than Env::CPUNanos();
+TEST_F(DBTest2, TestPerfContextCpuTime) {
+  // force resizing table cache so table handle is not preloaded so that
+  // we can measure find_table_nanos during Get().
+  dbfull()->TEST_table_cache()->SetCapacity(0);
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+  env_->now_cpu_count_.store(0);
+
+  // CPU timing is not enabled with kEnableTimeExceptForMutex
+  SetPerfLevel(PerfLevel::kEnableTimeExceptForMutex);
+  ASSERT_EQ("bar", Get("foo"));
+  ASSERT_EQ(0, get_perf_context()->get_cpu_nanos);
+  ASSERT_EQ(0, env_->now_cpu_count_.load());
+
+  uint64_t kDummyAddonTime = uint64_t{1000000000000};
+
+  // Add time to NowNanos() reading.
+  rocksdb::SyncPoint::GetInstance()->SetCallBack(
+      "TableCache::FindTable:0",
+      [&](void* /*arg*/) { env_->addon_time_.fetch_add(kDummyAddonTime); });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  SetPerfLevel(PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);
+  ASSERT_EQ("bar", Get("foo"));
+  ASSERT_GT(env_->now_cpu_count_.load(), 2);
+  ASSERT_LT(get_perf_context()->get_cpu_nanos, kDummyAddonTime);
+  ASSERT_GT(get_perf_context()->find_table_nanos, kDummyAddonTime);
+
+  SetPerfLevel(PerfLevel::kDisable);
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+}
+#endif  // OS_LINUX
+
 #ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
 TEST_F(DBTest2, PersistentCache) {
   int num_iter = 80;
@@ -2500,6 +2657,61 @@ 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();
@@ -2602,6 +2814,74 @@ TEST_F(DBTest2, TraceAndReplay) {
   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) {
@@ -2677,7 +2957,7 @@ TEST_F(DBTest2, DISABLED_IteratorPinnedMemory) {
   // Verify that iterators don't pin more than one data block in block cache
   // at each time.
   {
-    unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
+    std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
     iter->SeekToFirst();
 
     for (int i = 0; i < 4; i++) {
@@ -2835,6 +3115,156 @@ TEST_F(DBTest2, TestBBTTailPrefetch) {
   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
+
+// TODO: figure out why this test fails in appveyor
+#ifndef OS_WIN
+TEST_F(DBTest2, MultiDBParallelOpenTest) {
+  const int kNumDbs = 2;
+  Options options = CurrentOptions();
+  std::vector<std::string> dbnames;
+  for (int i = 0; i < kNumDbs; ++i) {
+    dbnames.emplace_back(test::TmpDir(env_) + "/db" + ToString(i));
+    ASSERT_OK(DestroyDB(dbnames.back(), options));
+  }
+
+  // Verify empty DBs can be created in parallel
+  std::vector<std::thread> open_threads;
+  std::vector<DB*> dbs{static_cast<unsigned int>(kNumDbs), nullptr};
+  options.create_if_missing = true;
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads.emplace_back(
+        [&](int dbnum) {
+          ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
+        },
+        i);
+  }
+
+  // Now add some data and close, so next we can verify non-empty DBs can be
+  // recovered in parallel
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads[i].join();
+    ASSERT_OK(dbs[i]->Put(WriteOptions(), "xi", "gua"));
+    delete dbs[i];
+  }
+
+  // Verify non-empty DBs can be recovered in parallel
+  dbs.clear();
+  open_threads.clear();
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads.emplace_back(
+        [&](int dbnum) {
+          ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
+        },
+        i);
+  }
+
+  // Wait and cleanup
+  for (int i = 0; i < kNumDbs; ++i) {
+    open_threads[i].join();
+    delete dbs[i];
+    ASSERT_OK(DestroyDB(dbnames[i], options));
+  }
+}
+#endif  // OS_WIN
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/db_test_util.cc

@@ -259,6 +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 {
@@ -430,6 +471,10 @@ Options DBTestBase::GetOptions(
       table_options.checksum = kxxHash;
       break;
     }
+    case kxxHash64Checksum: {
+      table_options.checksum = kxxHash64;
+      break;
+    }
     case kFIFOCompaction: {
       options.compaction_style = kCompactionStyleFIFO;
       break;
@@ -559,6 +604,7 @@ Status DBTestBase::TryReopenWithColumnFamilies(
     column_families.push_back(ColumnFamilyDescriptor(cfs[i], options[i]));
   }
   DBOptions db_opts = DBOptions(options[0]);
+  last_options_ = options[0];
   return DB::Open(db_opts, dbname_, column_families, &handles_, &db_);
 }
 
@@ -608,30 +654,19 @@ 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 {
@@ -646,6 +681,13 @@ 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);
@@ -721,6 +763,31 @@ std::string DBTestBase::Get(int cf, const std::string& k,
   return result;
 }
 
+std::vector<std::string> DBTestBase::MultiGet(std::vector<int> cfs,
+                                              const std::vector<std::string>& k,
+                                              const Snapshot* snapshot) {
+  ReadOptions options;
+  options.verify_checksums = true;
+  options.snapshot = snapshot;
+  std::vector<ColumnFamilyHandle*> handles;
+  std::vector<Slice> keys;
+  std::vector<std::string> result;
+
+  for (unsigned int i = 0; i < cfs.size(); ++i) {
+    handles.push_back(handles_[cfs[i]]);
+    keys.push_back(k[i]);
+  }
+  std::vector<Status> s = db_->MultiGet(options, handles, keys, &result);
+  for (unsigned int i = 0; i < s.size(); ++i) {
+    if (s[i].IsNotFound()) {
+      result[i] = "NOT_FOUND";
+    } else if (!s[i].ok()) {
+      result[i] = s[i].ToString();
+    }
+  }
+  return result;
+}
+
 Status DBTestBase::Get(const std::string& k, PinnableSlice* v) {
   ReadOptions options;
   options.verify_checksums = true;
@@ -773,13 +840,15 @@ 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());
@@ -1184,15 +1253,17 @@ 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);
@@ -1212,9 +1283,9 @@ 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) {
@@ -1392,8 +1463,10 @@ 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());

db/db_test_util.h

@@ -46,6 +46,7 @@
 #include "table/scoped_arena_iterator.h"
 #include "util/compression.h"
 #include "util/filename.h"
+#include "util/mock_time_env.h"
 #include "util/mutexlock.h"
 
 #include "util/string_util.h"
@@ -168,7 +169,7 @@ class SpecialMemTableRep : public MemTableRep {
   virtual ~SpecialMemTableRep() override {}
 
  private:
-  unique_ptr<MemTableRep> memtable_;
+  std::unique_ptr<MemTableRep> memtable_;
   int num_entries_flush_;
   int num_entries_;
 };
@@ -206,15 +207,15 @@ class SpecialEnv : public EnvWrapper {
  public:
   explicit SpecialEnv(Env* base);
 
-  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r,
+  Status NewWritableFile(const std::string& f, std::unique_ptr<WritableFile>* r,
                          const EnvOptions& soptions) override {
     class SSTableFile : public WritableFile {
      private:
       SpecialEnv* env_;
-      unique_ptr<WritableFile> base_;
+      std::unique_ptr<WritableFile> base_;
 
      public:
-      SSTableFile(SpecialEnv* env, unique_ptr<WritableFile>&& base)
+      SSTableFile(SpecialEnv* env, std::unique_ptr<WritableFile>&& base)
           : env_(env), base_(std::move(base)) {}
       Status Append(const Slice& data) override {
         if (env_->table_write_callback_) {
@@ -294,7 +295,7 @@ class SpecialEnv : public EnvWrapper {
     };
     class ManifestFile : public WritableFile {
      public:
-      ManifestFile(SpecialEnv* env, unique_ptr<WritableFile>&& b)
+      ManifestFile(SpecialEnv* env, std::unique_ptr<WritableFile>&& b)
           : env_(env), base_(std::move(b)) {}
       Status Append(const Slice& data) override {
         if (env_->manifest_write_error_.load(std::memory_order_acquire)) {
@@ -315,14 +316,17 @@ class SpecialEnv : public EnvWrapper {
         }
       }
       uint64_t GetFileSize() override { return base_->GetFileSize(); }
+      Status Allocate(uint64_t offset, uint64_t len) override {
+        return base_->Allocate(offset, len);
+      }
 
      private:
       SpecialEnv* env_;
-      unique_ptr<WritableFile> base_;
+      std::unique_ptr<WritableFile> base_;
     };
     class WalFile : public WritableFile {
      public:
-      WalFile(SpecialEnv* env, unique_ptr<WritableFile>&& b)
+      WalFile(SpecialEnv* env, std::unique_ptr<WritableFile>&& b)
           : env_(env), base_(std::move(b)) {
         env_->num_open_wal_file_.fetch_add(1);
       }
@@ -368,10 +372,13 @@ class SpecialEnv : public EnvWrapper {
       bool IsSyncThreadSafe() const override {
         return env_->is_wal_sync_thread_safe_.load();
       }
+      Status Allocate(uint64_t offset, uint64_t len) override {
+        return base_->Allocate(offset, len);
+      }
 
      private:
       SpecialEnv* env_;
-      unique_ptr<WritableFile> base_;
+      std::unique_ptr<WritableFile> base_;
     };
 
     if (non_writeable_rate_.load(std::memory_order_acquire) > 0) {
@@ -413,11 +420,11 @@ class SpecialEnv : public EnvWrapper {
   }
 
   Status NewRandomAccessFile(const std::string& f,
-                             unique_ptr<RandomAccessFile>* r,
+                             std::unique_ptr<RandomAccessFile>* r,
                              const EnvOptions& soptions) override {
     class CountingFile : public RandomAccessFile {
      public:
-      CountingFile(unique_ptr<RandomAccessFile>&& target,
+      CountingFile(std::unique_ptr<RandomAccessFile>&& target,
                    anon::AtomicCounter* counter,
                    std::atomic<size_t>* bytes_read)
           : target_(std::move(target)),
@@ -432,7 +439,7 @@ class SpecialEnv : public EnvWrapper {
       }
 
      private:
-      unique_ptr<RandomAccessFile> target_;
+      std::unique_ptr<RandomAccessFile> target_;
       anon::AtomicCounter* counter_;
       std::atomic<size_t>* bytes_read_;
     };
@@ -450,11 +457,11 @@ class SpecialEnv : public EnvWrapper {
   }
 
   virtual Status NewSequentialFile(const std::string& f,
-                                   unique_ptr<SequentialFile>* r,
+                                   std::unique_ptr<SequentialFile>* r,
                                    const EnvOptions& soptions) override {
     class CountingFile : public SequentialFile {
      public:
-      CountingFile(unique_ptr<SequentialFile>&& target,
+      CountingFile(std::unique_ptr<SequentialFile>&& target,
                    anon::AtomicCounter* counter)
           : target_(std::move(target)), counter_(counter) {}
       virtual Status Read(size_t n, Slice* result, char* scratch) override {
@@ -464,7 +471,7 @@ class SpecialEnv : public EnvWrapper {
       virtual Status Skip(uint64_t n) override { return target_->Skip(n); }
 
      private:
-      unique_ptr<SequentialFile> target_;
+      std::unique_ptr<SequentialFile> target_;
       anon::AtomicCounter* counter_;
     };
 
@@ -496,6 +503,11 @@ class SpecialEnv : public EnvWrapper {
     return s;
   }
 
+  virtual uint64_t NowCPUNanos() override {
+    now_cpu_count_.fetch_add(1);
+    return target()->NowCPUNanos();
+  }
+
   virtual uint64_t NowNanos() override {
     return (time_elapse_only_sleep_ ? 0 : target()->NowNanos()) +
            addon_time_.load() * 1000;
@@ -565,46 +577,17 @@ class SpecialEnv : public EnvWrapper {
 
   std::atomic<int64_t> addon_time_;
 
+  std::atomic<int> now_cpu_count_;
+
   std::atomic<int> delete_count_;
 
-  bool time_elapse_only_sleep_;
+  std::atomic<bool> time_elapse_only_sleep_;
 
   bool no_slowdown_;
 
   std::atomic<bool> is_wal_sync_thread_safe_{true};
 
-  std::atomic<size_t> compaction_readahead_size_;
-};
-
-class MockTimeEnv : public EnvWrapper {
- public:
-  explicit MockTimeEnv(Env* base) : EnvWrapper(base) {}
-
-  virtual Status GetCurrentTime(int64_t* time) override {
-    assert(time != nullptr);
-    assert(current_time_ <=
-           static_cast<uint64_t>(std::numeric_limits<int64_t>::max()));
-    *time = static_cast<int64_t>(current_time_);
-    return Status::OK();
-  }
-
-  virtual uint64_t NowMicros() override {
-    assert(current_time_ <= std::numeric_limits<uint64_t>::max() / 1000000);
-    return current_time_ * 1000000;
-  }
-
-  virtual uint64_t NowNanos() override {
-    assert(current_time_ <= std::numeric_limits<uint64_t>::max() / 1000000000);
-    return current_time_ * 1000000000;
-  }
-
-  void set_current_time(uint64_t time) {
-    assert(time >= current_time_);
-    current_time_ = time;
-  }
-
- private:
-  std::atomic<uint64_t> current_time_{0};
+  std::atomic<size_t> compaction_readahead_size_{};
 };
 
 #ifndef ROCKSDB_LITE
@@ -699,6 +682,7 @@ class DBTestBase : public testing::Test {
     kBlockBasedTableWithPartitionedIndexFormat4,
     kPartitionedFilterWithNewTableReaderForCompactions,
     kUniversalSubcompactions,
+    kxxHash64Checksum,
     // This must be the last line
     kEnd,
   };
@@ -770,6 +754,9 @@ class DBTestBase : public testing::Test {
   // Jump from kDefault to kFilter to kFullFilter
   bool ChangeFilterOptions();
 
+  // Switch between different DB options for file ingestion tests.
+  bool ChangeOptionsForFileIngestionTest();
+
   // Return the current option configuration.
   Options CurrentOptions(const anon::OptionsOverride& options_override =
                              anon::OptionsOverride()) const;
@@ -823,6 +810,8 @@ class DBTestBase : public testing::Test {
 
   Status Flush(int cf = 0);
 
+  Status Flush(const std::vector<int>& cf_ids);
+
   Status Put(const Slice& k, const Slice& v, WriteOptions wo = WriteOptions());
 
   Status Put(int cf, const Slice& k, const Slice& v,
@@ -851,6 +840,10 @@ class DBTestBase : public testing::Test {
 
   Status Get(const std::string& k, PinnableSlice* v);
 
+  std::vector<std::string> MultiGet(std::vector<int> cfs,
+                                    const std::vector<std::string>& k,
+                                    const Snapshot* snapshot = nullptr);
+
   uint64_t GetNumSnapshots();
 
   uint64_t GetTimeOldestSnapshots();

db/db_universal_compaction_test.cc

@@ -1154,7 +1154,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio2) {
     dbfull()->TEST_WaitForFlushMemTable();
     dbfull()->TEST_WaitForCompact();
   }
-  ASSERT_LT(TotalSize(), 120000U * 12 * 0.8 + 120000 * 2);
+  ASSERT_LT(TotalSize(), 120000U * 12 * 0.82 + 120000 * 2);
 }
 
 #ifndef ROCKSDB_VALGRIND_RUN

db/db_wal_test.cc

@@ -18,13 +18,23 @@ namespace rocksdb {
 class DBWALTest : public DBTestBase {
  public:
   DBWALTest() : DBTestBase("/db_wal_test") {}
+
+#if defined(ROCKSDB_PLATFORM_POSIX)
+  uint64_t GetAllocatedFileSize(std::string file_name) {
+    struct stat sbuf;
+    int err = stat(file_name.c_str(), &sbuf);
+    assert(err == 0);
+    return sbuf.st_blocks * 512;
+  }
+#endif
 };
 
 // A SpecialEnv enriched to give more insight about deleted files
 class EnrichedSpecialEnv : public SpecialEnv {
  public:
   explicit EnrichedSpecialEnv(Env* base) : SpecialEnv(base) {}
-  Status NewSequentialFile(const std::string& f, unique_ptr<SequentialFile>* r,
+  Status NewSequentialFile(const std::string& f,
+                           std::unique_ptr<SequentialFile>* r,
                            const EnvOptions& soptions) override {
     InstrumentedMutexLock l(&env_mutex_);
     if (f == skipped_wal) {
@@ -79,6 +89,7 @@ class DBWALTestWithEnrichedEnv : public DBTestBase {
     enriched_env_ = new EnrichedSpecialEnv(env_->target());
     auto options = CurrentOptions();
     options.env = enriched_env_;
+    options.allow_2pc = true;
     Reopen(options);
     delete env_;
     // to be deleted by the parent class
@@ -272,7 +283,31 @@ TEST_F(DBWALTest, RecoverWithTableHandle) {
     ASSERT_OK(Put(1, "bar", "v4"));
     ASSERT_OK(Flush(1));
     ASSERT_OK(Put(1, "big", std::string(100, 'a')));
-    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+
+    options = CurrentOptions();
+    const int kSmallMaxOpenFiles = 13;
+    if (option_config_ == kDBLogDir) {
+      // Use this option to check not preloading files
+      // Set the max open files to be small enough so no preload will
+      // happen.
+      options.max_open_files = kSmallMaxOpenFiles;
+      // RocksDB sanitize max open files to at least 20. Modify it back.
+      rocksdb::SyncPoint::GetInstance()->SetCallBack(
+          "SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
+            int* max_open_files = static_cast<int*>(arg);
+            *max_open_files = kSmallMaxOpenFiles;
+          });
+
+    } else if (option_config_ == kWalDirAndMmapReads) {
+      // Use this option to check always loading all files.
+      options.max_open_files = 100;
+    } else {
+      options.max_open_files = -1;
+    }
+    rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+    rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 
     std::vector<std::vector<FileMetaData>> files;
     dbfull()->TEST_GetFilesMetaData(handles_[1], &files);
@@ -283,10 +318,10 @@ TEST_F(DBWALTest, RecoverWithTableHandle) {
     ASSERT_EQ(total_files, 3);
     for (const auto& level : files) {
       for (const auto& file : level) {
-        if (kInfiniteMaxOpenFiles == option_config_) {
-          ASSERT_TRUE(file.table_reader_handle != nullptr);
-        } else {
+        if (options.max_open_files == kSmallMaxOpenFiles) {
           ASSERT_TRUE(file.table_reader_handle == nullptr);
+        } else {
+          ASSERT_TRUE(file.table_reader_handle != nullptr);
         }
       }
     }
@@ -793,12 +828,12 @@ class RecoveryTestHelper {
 
     *count = 0;
 
-    shared_ptr<Cache> table_cache = NewLRUCache(50, 0);
+    std::shared_ptr<Cache> table_cache = NewLRUCache(50, 0);
     EnvOptions env_options;
     WriteBufferManager write_buffer_manager(db_options.db_write_buffer_size);
 
-    unique_ptr<VersionSet> versions;
-    unique_ptr<WalManager> wal_manager;
+    std::unique_ptr<VersionSet> versions;
+    std::unique_ptr<WalManager> wal_manager;
     WriteController write_controller;
 
     versions.reset(new VersionSet(test->dbname_, &db_options, env_options,
@@ -812,10 +847,10 @@ class RecoveryTestHelper {
     for (size_t j = kWALFileOffset; j < wal_count + kWALFileOffset; j++) {
       uint64_t current_log_number = j;
       std::string fname = LogFileName(test->dbname_, current_log_number);
-      unique_ptr<WritableFile> file;
+      std::unique_ptr<WritableFile> file;
       ASSERT_OK(db_options.env->NewWritableFile(fname, &file, env_options));
-      unique_ptr<WritableFileWriter> file_writer(
-          new WritableFileWriter(std::move(file), env_options));
+      std::unique_ptr<WritableFileWriter> file_writer(
+          new WritableFileWriter(std::move(file), fname, env_options));
       current_log_writer.reset(
           new log::Writer(std::move(file_writer), current_log_number,
                           db_options.recycle_log_file_num > 0));
@@ -1329,6 +1364,99 @@ TEST_F(DBWALTest, RecoverFromCorruptedWALWithoutFlush) {
   }
 }
 
+// Tests that total log size is recovered if we set
+// avoid_flush_during_recovery=true.
+// Flush should trigger if max_total_wal_size is reached.
+TEST_F(DBWALTest, RestoreTotalLogSizeAfterRecoverWithoutFlush) {
+  class TestFlushListener : public EventListener {
+   public:
+    std::atomic<int> count{0};
+
+    TestFlushListener() = default;
+
+    void OnFlushBegin(DB* /*db*/, const FlushJobInfo& flush_job_info) override {
+      count++;
+      assert(FlushReason::kWriteBufferManager == flush_job_info.flush_reason);
+    }
+  };
+  std::shared_ptr<TestFlushListener> test_listener =
+      std::make_shared<TestFlushListener>();
+
+  constexpr size_t kKB = 1024;
+  constexpr size_t kMB = 1024 * 1024;
+  Options options = CurrentOptions();
+  options.avoid_flush_during_recovery = true;
+  options.max_total_wal_size = 1 * kMB;
+  options.listeners.push_back(test_listener);
+  // Have to open DB in multi-CF mode to trigger flush when
+  // max_total_wal_size is reached.
+  CreateAndReopenWithCF({"one"}, options);
+  // Write some keys and we will end up with one log file which is slightly
+  // smaller than 1MB.
+  std::string value_100k(100 * kKB, 'v');
+  std::string value_300k(300 * kKB, 'v');
+  ASSERT_OK(Put(0, "foo", "v1"));
+  for (int i = 0; i < 9; i++) {
+    ASSERT_OK(Put(1, "key" + ToString(i), value_100k));
+  }
+  // Get log files before reopen.
+  VectorLogPtr log_files_before;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_before));
+  ASSERT_EQ(1, log_files_before.size());
+  uint64_t log_size_before = log_files_before[0]->SizeFileBytes();
+  ASSERT_GT(log_size_before, 900 * kKB);
+  ASSERT_LT(log_size_before, 1 * kMB);
+  ReopenWithColumnFamilies({"default", "one"}, options);
+  // Write one more value to make log larger than 1MB.
+  ASSERT_OK(Put(1, "bar", value_300k));
+  // Get log files again. A new log file will be opened.
+  VectorLogPtr log_files_after_reopen;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_after_reopen));
+  ASSERT_EQ(2, log_files_after_reopen.size());
+  ASSERT_EQ(log_files_before[0]->LogNumber(),
+            log_files_after_reopen[0]->LogNumber());
+  ASSERT_GT(log_files_after_reopen[0]->SizeFileBytes() +
+                log_files_after_reopen[1]->SizeFileBytes(),
+            1 * kMB);
+  // Write one more key to trigger flush.
+  ASSERT_OK(Put(0, "foo", "v2"));
+  dbfull()->TEST_WaitForFlushMemTable();
+  // Flushed two column families.
+  ASSERT_EQ(2, test_listener->count.load());
+}
+
+#if defined(ROCKSDB_PLATFORM_POSIX)
+#if defined(ROCKSDB_FALLOCATE_PRESENT)
+// Tests that we will truncate the preallocated space of the last log from
+// previous.
+TEST_F(DBWALTest, TruncateLastLogAfterRecoverWithoutFlush) {
+  constexpr size_t kKB = 1024;
+  Options options = CurrentOptions();
+  options.avoid_flush_during_recovery = true;
+  DestroyAndReopen(options);
+  size_t preallocated_size =
+      dbfull()->TEST_GetWalPreallocateBlockSize(options.write_buffer_size);
+  ASSERT_OK(Put("foo", "v1"));
+  VectorLogPtr log_files_before;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_before));
+  ASSERT_EQ(1, log_files_before.size());
+  auto& file_before = log_files_before[0];
+  ASSERT_LT(file_before->SizeFileBytes(), 1 * kKB);
+  // The log file has preallocated space.
+  ASSERT_GE(GetAllocatedFileSize(dbname_ + file_before->PathName()),
+            preallocated_size);
+  Reopen(options);
+  VectorLogPtr log_files_after;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(log_files_after));
+  ASSERT_EQ(1, log_files_after.size());
+  ASSERT_LT(log_files_after[0]->SizeFileBytes(), 1 * kKB);
+  // The preallocated space should be truncated.
+  ASSERT_LT(GetAllocatedFileSize(dbname_ + file_before->PathName()),
+            preallocated_size);
+}
+#endif  // ROCKSDB_FALLOCATE_PRESENT
+#endif  // ROCKSDB_PLATFORM_POSIX
+
 #endif  // ROCKSDB_LITE
 
 TEST_F(DBWALTest, WalTermTest) {

db/dbformat.h

@@ -674,5 +674,16 @@ int InternalKeyComparator::CompareKeySeq(const Slice& akey,
   return r;
 }
 
+struct ParsedInternalKeyComparator {
+  explicit ParsedInternalKeyComparator(const InternalKeyComparator* c)
+      : cmp(c) {}
+
+  bool operator()(const ParsedInternalKey& a,
+                  const ParsedInternalKey& b) const {
+    return cmp->Compare(a, b) < 0;
+  }
+
+  const InternalKeyComparator* cmp;
+};
 
 }  // namespace rocksdb

db/error_handler.cc

@@ -4,7 +4,9 @@
 //  (found in the LICENSE.Apache file in the root directory).
 //
 #include "db/error_handler.h"
+#include "db/db_impl.h"
 #include "db/event_helpers.h"
+#include "util/sst_file_manager_impl.h"
 
 namespace rocksdb {
 
@@ -33,7 +35,7 @@ std::map<std::tuple<BackgroundErrorReason, Status::Code, Status::SubCode, bool>,
         // Errors during BG flush
         {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
                          Status::SubCode::kNoSpace, true),
-         Status::Severity::kSoftError},
+         Status::Severity::kHardError},
         {std::make_tuple(BackgroundErrorReason::kFlush, Status::Code::kIOError,
                          Status::SubCode::kNoSpace, false),
          Status::Severity::kNoError},
@@ -44,11 +46,11 @@ std::map<std::tuple<BackgroundErrorReason, Status::Code, Status::SubCode, bool>,
         {std::make_tuple(BackgroundErrorReason::kWriteCallback,
                          Status::Code::kIOError, Status::SubCode::kNoSpace,
                          true),
-         Status::Severity::kFatalError},
+         Status::Severity::kHardError},
         {std::make_tuple(BackgroundErrorReason::kWriteCallback,
                          Status::Code::kIOError, Status::SubCode::kNoSpace,
                          false),
-         Status::Severity::kFatalError},
+         Status::Severity::kHardError},
 };
 
 std::map<std::tuple<BackgroundErrorReason, Status::Code, bool>, Status::Severity>
@@ -118,6 +120,45 @@ std::map<std::tuple<BackgroundErrorReason, bool>, Status::Severity>
           Status::Severity::kFatalError},
 };
 
+void ErrorHandler::CancelErrorRecovery() {
+#ifndef ROCKSDB_LITE
+  db_mutex_->AssertHeld();
+
+  // We'll release the lock before calling sfm, so make sure no new
+  // recovery gets scheduled at that point
+  auto_recovery_ = false;
+  SstFileManagerImpl* sfm = reinterpret_cast<SstFileManagerImpl*>(
+      db_options_.sst_file_manager.get());
+  if (sfm) {
+    // This may or may not cancel a pending recovery
+    db_mutex_->Unlock();
+    bool cancelled = sfm->CancelErrorRecovery(this);
+    db_mutex_->Lock();
+    if (cancelled) {
+      recovery_in_prog_ = false;
+    }
+  }
+#endif
+}
+
+// This is the main function for looking at an error during a background
+// operation and deciding the severity, and error recovery strategy. The high
+// level algorithm is as follows -
+// 1. Classify the severity of the error based on the ErrorSeverityMap,
+//    DefaultErrorSeverityMap and DefaultReasonMap defined earlier
+// 2. Call a Status code specific override function to adjust the severity
+//    if needed. The reason for this is our ability to recover may depend on
+//    the exact options enabled in DBOptions
+// 3. Determine if auto recovery is possible. A listener notification callback
+//    is called, which can disable the auto recovery even if we decide its
+//    feasible
+// 4. For Status::NoSpace() errors, rely on SstFileManagerImpl to control
+//    the actual recovery. If no sst file manager is specified in DBOptions,
+//    a default one is allocated during DB::Open(), so there will always be
+//    one.
+// This can also get called as part of a recovery operation. In that case, we
+// also track the error separately in recovery_error_ so we can tell in the
+// end whether recovery succeeded or not
 Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reason) {
   db_mutex_->AssertHeld();
 
@@ -125,6 +166,12 @@ Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reas
     return Status::OK();
   }
 
+  // Check if recovery is currently in progress. If it is, we will save this
+  // error so we can check it at the end to see if recovery succeeded or not
+  if (recovery_in_prog_ && recovery_error_.ok()) {
+    recovery_error_ = bg_err;
+  }
+
   bool paranoid = db_options_.paranoid_checks;
   Status::Severity sev = Status::Severity::kFatalError;
   Status new_bg_err;
@@ -156,15 +203,143 @@ Status ErrorHandler::SetBGError(const Status& bg_err, BackgroundErrorReason reas
   }
 
   new_bg_err = Status(bg_err, sev);
+
+  bool auto_recovery = auto_recovery_;
+  if (new_bg_err.severity() >= Status::Severity::kFatalError && auto_recovery) {
+    auto_recovery = false;
+    ;
+  }
+
+  // Allow some error specific overrides
+  if (new_bg_err == Status::NoSpace()) {
+    new_bg_err = OverrideNoSpaceError(new_bg_err, &auto_recovery);
+  }
+
   if (!new_bg_err.ok()) {
     Status s = new_bg_err;
-    EventHelpers::NotifyOnBackgroundError(db_options_.listeners, reason, &s, db_mutex_);
+    EventHelpers::NotifyOnBackgroundError(db_options_.listeners, reason, &s,
+                                          db_mutex_, &auto_recovery);
     if (!s.ok() && (s.severity() > bg_error_.severity())) {
       bg_error_ = s;
+    } else {
+      // This error is less severe than previously encountered error. Don't
+      // take any further action
+      return bg_error_;
     }
   }
 
+  if (auto_recovery) {
+    recovery_in_prog_ = true;
+
+    // Kick-off error specific recovery
+    if (bg_error_ == Status::NoSpace()) {
+      RecoverFromNoSpace();
+    }
+  }
   return bg_error_;
 }
 
+Status ErrorHandler::OverrideNoSpaceError(Status bg_error,
+                                          bool* auto_recovery) {
+#ifndef ROCKSDB_LITE
+  if (bg_error.severity() >= Status::Severity::kFatalError) {
+    return bg_error;
+  }
+
+  if (db_options_.sst_file_manager.get() == nullptr) {
+    // We rely on SFM to poll for enough disk space and recover
+    *auto_recovery = false;
+    return bg_error;
+  }
+
+  if (db_options_.allow_2pc &&
+      (bg_error.severity() <= Status::Severity::kSoftError)) {
+    // Don't know how to recover, as the contents of the current WAL file may
+    // be inconsistent, and it may be needed for 2PC. If 2PC is not enabled,
+    // we can just flush the memtable and discard the log
+    *auto_recovery = false;
+    return Status(bg_error, Status::Severity::kFatalError);
+  }
+
+  {
+    uint64_t free_space;
+    if (db_options_.env->GetFreeSpace(db_options_.db_paths[0].path,
+                                      &free_space) == Status::NotSupported()) {
+      *auto_recovery = false;
+    }
+  }
+
+  return bg_error;
+#else
+  (void)auto_recovery;
+  return Status(bg_error, Status::Severity::kFatalError);
+#endif
+}
+
+void ErrorHandler::RecoverFromNoSpace() {
+#ifndef ROCKSDB_LITE
+  SstFileManagerImpl* sfm =
+      reinterpret_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
+
+  // Inform SFM of the error, so it can kick-off the recovery
+  if (sfm) {
+    sfm->StartErrorRecovery(this, bg_error_);
+  }
+#endif
+}
+
+Status ErrorHandler::ClearBGError() {
+#ifndef ROCKSDB_LITE
+  db_mutex_->AssertHeld();
+
+  // Signal that recovery succeeded
+  if (recovery_error_.ok()) {
+    Status old_bg_error = bg_error_;
+    bg_error_ = Status::OK();
+    recovery_in_prog_ = false;
+    EventHelpers::NotifyOnErrorRecoveryCompleted(db_options_.listeners,
+                                                 old_bg_error, db_mutex_);
+  }
+  return recovery_error_;
+#else
+  return bg_error_;
+#endif
+}
+
+Status ErrorHandler::RecoverFromBGError(bool is_manual) {
+#ifndef ROCKSDB_LITE
+  InstrumentedMutexLock l(db_mutex_);
+  if (is_manual) {
+    // If its a manual recovery and there's a background recovery in progress
+    // return busy status
+    if (recovery_in_prog_) {
+      return Status::Busy();
+    }
+    recovery_in_prog_ = true;
+  }
+
+  if (bg_error_.severity() == Status::Severity::kSoftError) {
+    // Simply clear the background error and return
+    recovery_error_ = Status::OK();
+    return ClearBGError();
+  }
+
+  // Reset recovery_error_. We will use this to record any errors that happen
+  // during the recovery process. While recovering, the only operations that
+  // can generate background errors should be the flush operations
+  recovery_error_ = Status::OK();
+  Status s = db_->ResumeImpl();
+  // For manual recover, shutdown, and fatal error  cases, set
+  // recovery_in_prog_ to false. For automatic background recovery, leave it
+  // as is regardless of success or failure as it will be retried
+  if (is_manual || s.IsShutdownInProgress() ||
+      bg_error_.severity() >= Status::Severity::kFatalError) {
+    recovery_in_prog_ = false;
+  }
+  return s;
+#else
+  (void)is_manual;
+  return bg_error_;
+#endif
+}
 }

db/error_handler.h

@@ -11,42 +11,65 @@
 
 namespace rocksdb {
 
+class DBImpl;
+
 class ErrorHandler {
   public:
-    ErrorHandler(const ImmutableDBOptions& db_options,
-        InstrumentedMutex* db_mutex)
-      : db_options_(db_options),
-        bg_error_(Status::OK()),
-        db_mutex_(db_mutex)
-      {}
-    ~ErrorHandler() {}
+   ErrorHandler(DBImpl* db, const ImmutableDBOptions& db_options,
+                InstrumentedMutex* db_mutex)
+       : db_(db),
+         db_options_(db_options),
+         bg_error_(Status::OK()),
+         recovery_error_(Status::OK()),
+         db_mutex_(db_mutex),
+         auto_recovery_(false),
+         recovery_in_prog_(false) {}
+   ~ErrorHandler() {}
 
-    Status::Severity GetErrorSeverity(BackgroundErrorReason reason,
-        Status::Code code, Status::SubCode subcode);
+   void EnableAutoRecovery() { auto_recovery_ = true; }
 
-    Status SetBGError(const Status& bg_err, BackgroundErrorReason reason);
+   Status::Severity GetErrorSeverity(BackgroundErrorReason reason,
+                                     Status::Code code,
+                                     Status::SubCode subcode);
 
-    Status GetBGError()
-    {
-      return bg_error_;
-    }
+   Status SetBGError(const Status& bg_err, BackgroundErrorReason reason);
 
-    void ClearBGError() {
-      bg_error_ = Status::OK();
-    }
+   Status GetBGError() { return bg_error_; }
 
-    bool IsDBStopped() {
-      return !bg_error_.ok();
+   Status GetRecoveryError() { return recovery_error_; }
+
+   Status ClearBGError();
+
+   bool IsDBStopped() {
+     return !bg_error_.ok() &&
+            bg_error_.severity() >= Status::Severity::kHardError;
     }
 
     bool IsBGWorkStopped() {
-      return !bg_error_.ok();
+      return !bg_error_.ok() &&
+             (bg_error_.severity() >= Status::Severity::kHardError ||
+              !auto_recovery_);
     }
 
-  private:
+    bool IsRecoveryInProgress() { return recovery_in_prog_; }
+
+    Status RecoverFromBGError(bool is_manual = false);
+    void CancelErrorRecovery();
+
+   private:
+    DBImpl* db_;
     const ImmutableDBOptions& db_options_;
     Status bg_error_;
+    // A separate Status variable used to record any errors during the
+    // recovery process from hard errors
+    Status recovery_error_;
     InstrumentedMutex* db_mutex_;
+    // A flag indicating whether automatic recovery from errors is enabled
+    bool auto_recovery_;
+    bool recovery_in_prog_;
+
+    Status OverrideNoSpaceError(Status bg_error, bool* auto_recovery);
+    void RecoverFromNoSpace();
 };
 
 }

db/error_handler_test.cc

@@ -6,9 +6,12 @@
 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
+#ifndef ROCKSDB_LITE
+
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
 #include "rocksdb/perf_context.h"
+#include "rocksdb/sst_file_manager.h"
 #include "util/fault_injection_test_env.h"
 #if !defined(ROCKSDB_LITE)
 #include "util/sync_point.h"
@@ -33,36 +36,139 @@ class DBErrorHandlingEnv : public EnvWrapper {
     bool trig_io_error;
 };
 
+class ErrorHandlerListener : public EventListener {
+ public:
+  ErrorHandlerListener()
+      : mutex_(),
+        cv_(&mutex_),
+        no_auto_recovery_(false),
+        recovery_complete_(false),
+        file_creation_started_(false),
+        override_bg_error_(false),
+        file_count_(0),
+        fault_env_(nullptr) {}
+
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& /*ti*/) override {
+    InstrumentedMutexLock l(&mutex_);
+    file_creation_started_ = true;
+    if (file_count_ > 0) {
+      if (--file_count_ == 0) {
+        fault_env_->SetFilesystemActive(false, file_creation_error_);
+        file_creation_error_ = Status::OK();
+      }
+    }
+    cv_.SignalAll();
+  }
+
+  void OnErrorRecoveryBegin(BackgroundErrorReason /*reason*/,
+                            Status /*bg_error*/,
+                            bool* auto_recovery) override {
+    if (*auto_recovery && no_auto_recovery_) {
+      *auto_recovery = false;
+    }
+  }
+
+  void OnErrorRecoveryCompleted(Status /*old_bg_error*/) override {
+    InstrumentedMutexLock l(&mutex_);
+    recovery_complete_ = true;
+    cv_.SignalAll();
+  }
+
+  bool WaitForRecovery(uint64_t /*abs_time_us*/) {
+    InstrumentedMutexLock l(&mutex_);
+    while (!recovery_complete_) {
+      cv_.Wait(/*abs_time_us*/);
+    }
+    if (recovery_complete_) {
+      recovery_complete_ = false;
+      return true;
+    }
+    return false;
+  }
+
+  void WaitForTableFileCreationStarted(uint64_t /*abs_time_us*/) {
+    InstrumentedMutexLock l(&mutex_);
+    while (!file_creation_started_) {
+      cv_.Wait(/*abs_time_us*/);
+    }
+    file_creation_started_ = false;
+  }
+
+  void OnBackgroundError(BackgroundErrorReason /*reason*/,
+                         Status* bg_error) override {
+    if (override_bg_error_) {
+      *bg_error = bg_error_;
+      override_bg_error_ = false;
+    }
+  }
+
+  void EnableAutoRecovery(bool enable = true) { no_auto_recovery_ = !enable; }
+
+  void OverrideBGError(Status bg_err) {
+    bg_error_ = bg_err;
+    override_bg_error_ = true;
+  }
+
+  void InjectFileCreationError(FaultInjectionTestEnv* env, int file_count,
+                               Status s) {
+    fault_env_ = env;
+    file_count_ = file_count;
+    file_creation_error_ = s;
+  }
+
+ private:
+  InstrumentedMutex mutex_;
+  InstrumentedCondVar cv_;
+  bool no_auto_recovery_;
+  bool recovery_complete_;
+  bool file_creation_started_;
+  bool override_bg_error_;
+  int file_count_;
+  Status file_creation_error_;
+  Status bg_error_;
+  FaultInjectionTestEnv* fault_env_;
+};
+
 TEST_F(DBErrorHandlingTest, FLushWriteError) {
   std::unique_ptr<FaultInjectionTestEnv> fault_env(
       new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
   Options options = GetDefaultOptions();
   options.create_if_missing = true;
   options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
   Status s;
+
+  listener->EnableAutoRecovery(false);
   DestroyAndReopen(options);
 
-  Put(Key(0), "va;");
+  Put(Key(0), "val");
   SyncPoint::GetInstance()->SetCallBack(
       "FlushJob::Start", [&](void *) {
     fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
   });
   SyncPoint::GetInstance()->EnableProcessing();
   s = Flush();
-  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kSoftError);
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  SyncPoint::GetInstance()->DisableProcessing();
   fault_env->SetFilesystemActive(true);
   s = dbfull()->Resume();
   ASSERT_EQ(s, Status::OK());
 
+  Reopen(options);
+  ASSERT_EQ("val", Get(Key(0)));
   Destroy(options);
 }
 
 TEST_F(DBErrorHandlingTest, CompactionWriteError) {
   std::unique_ptr<FaultInjectionTestEnv> fault_env(
       new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
   Options options = GetDefaultOptions();
   options.create_if_missing = true;
   options.level0_file_num_compaction_trigger = 2;
+  options.listeners.emplace_back(listener);
   options.env = fault_env.get();
   Status s;
   DestroyAndReopen(options);
@@ -72,6 +178,10 @@ TEST_F(DBErrorHandlingTest, CompactionWriteError) {
   s = Flush();
   ASSERT_EQ(s, Status::OK());
 
+  listener->OverrideBGError(
+      Status(Status::NoSpace(), Status::Severity::kHardError)
+      );
+  listener->EnableAutoRecovery(false);
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
       {{"FlushMemTableFinished", "BackgroundCallCompaction:0"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
@@ -85,7 +195,7 @@ TEST_F(DBErrorHandlingTest, CompactionWriteError) {
   ASSERT_EQ(s, Status::OK());
 
   s = dbfull()->TEST_WaitForCompact();
-  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kSoftError);
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
 
   fault_env->SetFilesystemActive(true);
   s = dbfull()->Resume();
@@ -129,6 +239,439 @@ TEST_F(DBErrorHandlingTest, CorruptionError) {
   Destroy(options);
 }
 
+TEST_F(DBErrorHandlingTest, AutoRecoverFlushError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  Put(Key(0), "val");
+  SyncPoint::GetInstance()->SetCallBack("FlushJob::Start", [&](void*) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+
+  s = Put(Key(1), "val");
+  ASSERT_EQ(s, Status::OK());
+
+  Reopen(options);
+  ASSERT_EQ("val", Get(Key(0)));
+  ASSERT_EQ("val", Get(Key(1)));
+  Destroy(options);
+}
+
+TEST_F(DBErrorHandlingTest, FailRecoverFlushError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  Put(Key(0), "val");
+  SyncPoint::GetInstance()->SetCallBack("FlushJob::Start", [&](void*) {
+    fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  });
+  SyncPoint::GetInstance()->EnableProcessing();
+  s = Flush();
+  ASSERT_EQ(s.severity(), rocksdb::Status::Severity::kHardError);
+  // We should be able to shutdown the database while auto recovery is going
+  // on in the background
+  Close();
+  DestroyDB(dbname_, options);
+}
+
+TEST_F(DBErrorHandlingTest, WALWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.writable_file_max_buffer_size = 32768;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+  Random rnd(301);
+
+  listener->EnableAutoRecovery();
+  DestroyAndReopen(options);
+
+  {
+    WriteBatch batch;
+
+    for (auto i = 0; i<100; ++i) {
+      batch.Put(Key(i), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(dbfull()->Write(wopts, &batch), Status::OK());
+  };
+
+  {
+    WriteBatch batch;
+    int write_error = 0;
+
+    for (auto i = 100; i<199; ++i) {
+      batch.Put(Key(i), RandomString(&rnd, 1024));
+    }
+
+    SyncPoint::GetInstance()->SetCallBack("WritableFileWriter::Append:BeforePrepareWrite", [&](void*) {
+      write_error++;
+      if (write_error > 2) {
+        fault_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+      }
+    });
+    SyncPoint::GetInstance()->EnableProcessing();
+    WriteOptions wopts;
+    wopts.sync = true;
+    s = dbfull()->Write(wopts, &batch);
+    ASSERT_EQ(s, s.NoSpace());
+  }
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+  for (auto i=0; i<199; ++i) {
+    if (i < 100) {
+      ASSERT_NE(Get(Key(i)), "NOT_FOUND");
+    } else {
+      ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+    }
+  }
+  Reopen(options);
+  for (auto i=0; i<199; ++i) {
+    if (i < 100) {
+      ASSERT_NE(Get(Key(i)), "NOT_FOUND");
+    } else {
+      ASSERT_EQ(Get(Key(i)), "NOT_FOUND");
+    }
+  }
+  Close();
+}
+
+TEST_F(DBErrorHandlingTest, MultiCFWALWriteError) {
+  std::unique_ptr<FaultInjectionTestEnv> fault_env(
+      new FaultInjectionTestEnv(Env::Default()));
+  std::shared_ptr<ErrorHandlerListener> listener(new ErrorHandlerListener());
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.writable_file_max_buffer_size = 32768;
+  options.env = fault_env.get();
+  options.listeners.emplace_back(listener);
+  Status s;
+  Random rnd(301);
+
+  listener->EnableAutoRecovery();
+  CreateAndReopenWithCF({"one", "two", "three"}, options);
+
+  {
+    WriteBatch batch;
+
+    for (auto i = 1; i < 4; ++i) {
+      for (auto j = 0; j < 100; ++j) {
+        batch.Put(handles_[i], Key(j), RandomString(&rnd, 1024));
+      }
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(dbfull()->Write(wopts, &batch), Status::OK());
+  };
+
+  {
+    WriteBatch batch;
+    int write_error = 0;
+
+    // Write to one CF
+    for (auto i = 100; i < 199; ++i) {
+      batch.Put(handles_[2], Key(i), RandomString(&rnd, 1024));
+    }
+
+    SyncPoint::GetInstance()->SetCallBack(
+        "WritableFileWriter::Append:BeforePrepareWrite", [&](void*) {
+          write_error++;
+          if (write_error > 2) {
+            fault_env->SetFilesystemActive(false,
+                                           Status::NoSpace("Out of space"));
+          }
+        });
+    SyncPoint::GetInstance()->EnableProcessing();
+    WriteOptions wopts;
+    wopts.sync = true;
+    s = dbfull()->Write(wopts, &batch);
+    ASSERT_EQ(s, s.NoSpace());
+  }
+  SyncPoint::GetInstance()->DisableProcessing();
+  fault_env->SetFilesystemActive(true);
+  ASSERT_EQ(listener->WaitForRecovery(5000000), true);
+
+  for (auto i = 1; i < 4; ++i) {
+    // Every CF should have been flushed
+    ASSERT_EQ(NumTableFilesAtLevel(0, i), 1);
+  }
+
+  for (auto i = 1; i < 4; ++i) {
+    for (auto j = 0; j < 199; ++j) {
+      if (j < 100) {
+        ASSERT_NE(Get(i, Key(j)), "NOT_FOUND");
+      } else {
+        ASSERT_EQ(Get(i, Key(j)), "NOT_FOUND");
+      }
+    }
+  }
+  ReopenWithColumnFamilies({"default", "one", "two", "three"}, options);
+  for (auto i = 1; i < 4; ++i) {
+    for (auto j = 0; j < 199; ++j) {
+      if (j < 100) {
+        ASSERT_NE(Get(i, Key(j)), "NOT_FOUND");
+      } else {
+        ASSERT_EQ(Get(i, Key(j)), "NOT_FOUND");
+      }
+    }
+  }
+  Close();
+}
+
+TEST_F(DBErrorHandlingTest, MultiDBCompactionError) {
+  FaultInjectionTestEnv* def_env = new FaultInjectionTestEnv(Env::Default());
+  std::vector<std::unique_ptr<FaultInjectionTestEnv>> fault_env;
+  std::vector<Options> options;
+  std::vector<std::shared_ptr<ErrorHandlerListener>> listener;
+  std::vector<DB*> db;
+  std::shared_ptr<SstFileManager> sfm(NewSstFileManager(def_env));
+  int kNumDbInstances = 3;
+  Random rnd(301);
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    listener.emplace_back(new ErrorHandlerListener());
+    options.emplace_back(GetDefaultOptions());
+    fault_env.emplace_back(new FaultInjectionTestEnv(Env::Default()));
+    options[i].create_if_missing = true;
+    options[i].level0_file_num_compaction_trigger = 2;
+    options[i].writable_file_max_buffer_size = 32768;
+    options[i].env = fault_env[i].get();
+    options[i].listeners.emplace_back(listener[i]);
+    options[i].sst_file_manager = sfm;
+    DB* dbptr;
+    char buf[16];
+
+    listener[i]->EnableAutoRecovery();
+    // Setup for returning error for the 3rd SST, which would be level 1
+    listener[i]->InjectFileCreationError(fault_env[i].get(), 3,
+                                         Status::NoSpace("Out of space"));
+    snprintf(buf, sizeof(buf), "_%d", i);
+    DestroyDB(dbname_ + std::string(buf), options[i]);
+    ASSERT_EQ(DB::Open(options[i], dbname_ + std::string(buf), &dbptr),
+              Status::OK());
+    db.emplace_back(dbptr);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    for (auto j = 0; j <= 100; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  def_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    // Write to one CF
+    for (auto j = 100; j < 199; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    Status s = static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true);
+    ASSERT_EQ(s.severity(), Status::Severity::kSoftError);
+    fault_env[i]->SetFilesystemActive(true);
+  }
+
+  def_env->SetFilesystemActive(true);
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    std::string prop;
+    ASSERT_EQ(listener[i]->WaitForRecovery(5000000), true);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(0), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 0);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(1), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 1);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    char buf[16];
+    snprintf(buf, sizeof(buf), "_%d", i);
+    delete db[i];
+    fault_env[i]->SetFilesystemActive(true);
+    if (getenv("KEEP_DB")) {
+      printf("DB is still at %s%s\n", dbname_.c_str(), buf);
+    } else {
+      Status s = DestroyDB(dbname_ + std::string(buf), options[i]);
+    }
+  }
+  options.clear();
+  sfm.reset();
+  delete def_env;
+}
+
+TEST_F(DBErrorHandlingTest, MultiDBVariousErrors) {
+  FaultInjectionTestEnv* def_env = new FaultInjectionTestEnv(Env::Default());
+  std::vector<std::unique_ptr<FaultInjectionTestEnv>> fault_env;
+  std::vector<Options> options;
+  std::vector<std::shared_ptr<ErrorHandlerListener>> listener;
+  std::vector<DB*> db;
+  std::shared_ptr<SstFileManager> sfm(NewSstFileManager(def_env));
+  int kNumDbInstances = 3;
+  Random rnd(301);
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    listener.emplace_back(new ErrorHandlerListener());
+    options.emplace_back(GetDefaultOptions());
+    fault_env.emplace_back(new FaultInjectionTestEnv(Env::Default()));
+    options[i].create_if_missing = true;
+    options[i].level0_file_num_compaction_trigger = 2;
+    options[i].writable_file_max_buffer_size = 32768;
+    options[i].env = fault_env[i].get();
+    options[i].listeners.emplace_back(listener[i]);
+    options[i].sst_file_manager = sfm;
+    DB* dbptr;
+    char buf[16];
+
+    listener[i]->EnableAutoRecovery();
+    switch (i) {
+      case 0:
+        // Setup for returning error for the 3rd SST, which would be level 1
+        listener[i]->InjectFileCreationError(fault_env[i].get(), 3,
+                                             Status::NoSpace("Out of space"));
+        break;
+      case 1:
+        // Setup for returning error after the 1st SST, which would result
+        // in a hard error
+        listener[i]->InjectFileCreationError(fault_env[i].get(), 2,
+                                             Status::NoSpace("Out of space"));
+        break;
+      default:
+        break;
+    }
+    snprintf(buf, sizeof(buf), "_%d", i);
+    DestroyDB(dbname_ + std::string(buf), options[i]);
+    ASSERT_EQ(DB::Open(options[i], dbname_ + std::string(buf), &dbptr),
+              Status::OK());
+    db.emplace_back(dbptr);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    for (auto j = 0; j <= 100; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+  }
+
+  def_env->SetFilesystemActive(false, Status::NoSpace("Out of space"));
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    WriteBatch batch;
+
+    // Write to one CF
+    for (auto j = 100; j < 199; ++j) {
+      batch.Put(Key(j), RandomString(&rnd, 1024));
+    }
+
+    WriteOptions wopts;
+    wopts.sync = true;
+    ASSERT_EQ(db[i]->Write(wopts, &batch), Status::OK());
+    if (i != 1) {
+      ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::OK());
+    } else {
+      ASSERT_EQ(db[i]->Flush(FlushOptions()), Status::NoSpace());
+    }
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    Status s = static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true);
+    switch (i) {
+      case 0:
+        ASSERT_EQ(s.severity(), Status::Severity::kSoftError);
+        break;
+      case 1:
+        ASSERT_EQ(s.severity(), Status::Severity::kHardError);
+        break;
+      case 2:
+        ASSERT_EQ(s, Status::OK());
+        break;
+    }
+    fault_env[i]->SetFilesystemActive(true);
+  }
+
+  def_env->SetFilesystemActive(true);
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    std::string prop;
+    if (i < 2) {
+      ASSERT_EQ(listener[i]->WaitForRecovery(5000000), true);
+    }
+    if (i == 1) {
+      ASSERT_EQ(static_cast<DBImpl*>(db[i])->TEST_WaitForCompact(true),
+                Status::OK());
+    }
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(0), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 0);
+    EXPECT_TRUE(db[i]->GetProperty(
+        "rocksdb.num-files-at-level" + NumberToString(1), &prop));
+    EXPECT_EQ(atoi(prop.c_str()), 1);
+  }
+
+  for (auto i = 0; i < kNumDbInstances; ++i) {
+    char buf[16];
+    snprintf(buf, sizeof(buf), "_%d", i);
+    fault_env[i]->SetFilesystemActive(true);
+    delete db[i];
+    if (getenv("KEEP_DB")) {
+      printf("DB is still at %s%s\n", dbname_.c_str(), buf);
+    } else {
+      DestroyDB(dbname_ + std::string(buf), options[i]);
+    }
+  }
+  options.clear();
+  delete def_env;
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {
@@ -136,3 +679,13 @@ int main(int argc, char** argv) {
   ::testing::InitGoogleTest(&argc, argv);
   return RUN_ALL_TESTS();
 }
+
+#else
+#include <stdio.h>
+
+int main(int /*argc*/, char** /*argv*/) {
+  fprintf(stderr, "SKIPPED as Cuckoo table is not supported in ROCKSDB_LITE\n");
+  return 0;
+}
+
+#endif  // ROCKSDB_LITE

db/event_helpers.cc

@@ -40,8 +40,8 @@ void EventHelpers::NotifyTableFileCreationStarted(
 
 void EventHelpers::NotifyOnBackgroundError(
     const std::vector<std::shared_ptr<EventListener>>& listeners,
-    BackgroundErrorReason reason, Status* bg_error,
-    InstrumentedMutex* db_mutex) {
+    BackgroundErrorReason reason, Status* bg_error, InstrumentedMutex* db_mutex,
+    bool* auto_recovery) {
 #ifndef ROCKSDB_LITE
   if (listeners.size() == 0U) {
     return;
@@ -51,6 +51,9 @@ void EventHelpers::NotifyOnBackgroundError(
   db_mutex->Unlock();
   for (auto& listener : listeners) {
     listener->OnBackgroundError(reason, bg_error);
+    if (*auto_recovery) {
+      listener->OnErrorRecoveryBegin(reason, *bg_error, auto_recovery);
+    }
   }
   db_mutex->Lock();
 #else
@@ -58,6 +61,7 @@ void EventHelpers::NotifyOnBackgroundError(
   (void)reason;
   (void)bg_error;
   (void)db_mutex;
+  (void)auto_recovery;
 #endif  // ROCKSDB_LITE
 }
 
@@ -167,4 +171,25 @@ void EventHelpers::LogAndNotifyTableFileDeletion(
 #endif  // !ROCKSDB_LITE
 }
 
+void EventHelpers::NotifyOnErrorRecoveryCompleted(
+    const std::vector<std::shared_ptr<EventListener>>& listeners,
+    Status old_bg_error, InstrumentedMutex* db_mutex) {
+#ifndef ROCKSDB_LITE
+  if (listeners.size() == 0U) {
+    return;
+  }
+  db_mutex->AssertHeld();
+  // release lock while notifying events
+  db_mutex->Unlock();
+  for (auto& listener : listeners) {
+    listener->OnErrorRecoveryCompleted(old_bg_error);
+  }
+  db_mutex->Lock();
+#else
+  (void)listeners;
+  (void)old_bg_error;
+  (void)db_mutex;
+#endif  // ROCKSDB_LITE
+}
+
 }  // namespace rocksdb

db/event_helpers.h

@@ -28,7 +28,7 @@ class EventHelpers {
   static void NotifyOnBackgroundError(
       const std::vector<std::shared_ptr<EventListener>>& listeners,
       BackgroundErrorReason reason, Status* bg_error,
-      InstrumentedMutex* db_mutex);
+      InstrumentedMutex* db_mutex, bool* auto_recovery);
   static void LogAndNotifyTableFileCreationFinished(
       EventLogger* event_logger,
       const std::vector<std::shared_ptr<EventListener>>& listeners,
@@ -41,6 +41,9 @@ class EventHelpers {
       uint64_t file_number, const std::string& file_path,
       const Status& status, const std::string& db_name,
       const std::vector<std::shared_ptr<EventListener>>& listeners);
+  static void NotifyOnErrorRecoveryCompleted(
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      Status bg_error, InstrumentedMutex* db_mutex);
 
  private:
   static void LogAndNotifyTableFileCreation(

db/external_sst_file_basic_test.cc

@@ -14,9 +14,11 @@
 namespace rocksdb {
 
 #ifndef ROCKSDB_LITE
-class ExternalSSTFileBasicTest : public DBTestBase {
+class ExternalSSTFileBasicTest
+    : public DBTestBase,
+      public ::testing::WithParamInterface<std::tuple<bool, bool>> {
  public:
-  ExternalSSTFileBasicTest() : DBTestBase("/external_sst_file_test") {
+  ExternalSSTFileBasicTest() : DBTestBase("/external_sst_file_basic_test") {
     sst_files_dir_ = dbname_ + "/sst_files/";
     DestroyAndRecreateExternalSSTFilesDir();
   }
@@ -41,6 +43,7 @@ class ExternalSSTFileBasicTest : public DBTestBase {
       const Options options, std::vector<int> keys,
       const std::vector<ValueType>& value_types,
       std::vector<std::pair<int, int>> range_deletions, int file_id,
+      bool write_global_seqno, bool verify_checksums_before_ingest,
       std::map<std::string, std::string>* true_data) {
     assert(value_types.size() == 1 || keys.size() == value_types.size());
     std::string file_path = sst_files_dir_ + ToString(file_id);
@@ -105,6 +108,8 @@ class ExternalSSTFileBasicTest : public DBTestBase {
     if (s.ok()) {
       IngestExternalFileOptions ifo;
       ifo.allow_global_seqno = true;
+      ifo.write_global_seqno = write_global_seqno;
+      ifo.verify_checksums_before_ingest = verify_checksums_before_ingest;
       s = db_->IngestExternalFile({file_path}, ifo);
     }
     return s;
@@ -113,17 +118,20 @@ class ExternalSSTFileBasicTest : public DBTestBase {
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<int> keys,
       const std::vector<ValueType>& value_types, int file_id,
+      bool write_global_seqno, bool verify_checksums_before_ingest,
       std::map<std::string, std::string>* true_data) {
-    return GenerateAndAddExternalFile(options, keys, value_types, {}, file_id,
-                                      true_data);
+    return GenerateAndAddExternalFile(
+        options, keys, value_types, {}, file_id, write_global_seqno,
+        verify_checksums_before_ingest, true_data);
   }
 
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<int> keys, const ValueType value_type,
-      int file_id, std::map<std::string, std::string>* true_data) {
-    return GenerateAndAddExternalFile(options, keys,
-                                      std::vector<ValueType>(1, value_type),
-                                      file_id, true_data);
+      int file_id, bool write_global_seqno, bool verify_checksums_before_ingest,
+      std::map<std::string, std::string>* true_data) {
+    return GenerateAndAddExternalFile(
+        options, keys, std::vector<ValueType>(1, value_type), file_id,
+        write_global_seqno, verify_checksums_before_ingest, true_data);
   }
 
   ~ExternalSSTFileBasicTest() { test::DestroyDir(env_, sst_files_dir_); }
@@ -245,7 +253,9 @@ TEST_F(ExternalSSTFileBasicTest, NoCopy) {
   }
 }
 
-TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
+TEST_P(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   do {
     Options options = CurrentOptions();
     DestroyAndReopen(options);
@@ -253,35 +263,39 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
 
     int file_id = 1;
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
-                                         ValueType::kTypeValue, file_id++,
-                                         &true_data));
-    // File doesn't overwrite any keys, no seqno needed
-    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
-
-    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
-                                         ValueType::kTypeValue, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 2, 3, 4, 5, 6}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1, 4, 6}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {10, 11, 12, 13}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 4, 6}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {11, 15, 19}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {11, 15, 19}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {120, 130}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {120, 130}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1, 130}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {1, 130}, ValueType::kTypeValue, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
@@ -293,17 +307,20 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
     SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {60, 61, 62}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {40, 41, 42}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {40, 41, 42}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {20, 30, 40}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {20, 30, 40}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
@@ -312,34 +329,39 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithGlobalSeqnoPickedSeqno) {
     // We will need a seqno for the file regardless if the file overwrite
     // keys in the DB or not because we have a snapshot
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1000, 1002}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {1000, 1002}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {2000, 3002}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {2000, 3002}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
-                                         ValueType::kTypeValue, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 20, 40, 100, 150}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
     db_->ReleaseSnapshot(snapshot);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {5000, 5001}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {5000, 5001}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // No snapshot anymore, no need to assign a seqno
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
     size_t kcnt = 0;
     VerifyDBFromMap(true_data, &kcnt, false);
-  } while (ChangeCompactOptions());
+  } while (ChangeOptionsForFileIngestionTest());
 }
 
-TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
+TEST_P(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   do {
     Options options = CurrentOptions();
     options.merge_operator.reset(new TestPutOperator());
@@ -348,54 +370,59 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
 
     int file_id = 1;
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2, 3, 4, 5, 6},
-                                         ValueType::kTypeValue, file_id++,
-                                         &true_data));
-    // File doesn't overwrite any keys, no seqno needed
-    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
-
-    ASSERT_OK(GenerateAndAddExternalFile(options, {10, 11, 12, 13},
-                                         ValueType::kTypeValue, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 2, 3, 4, 5, 6}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1, 4, 6}, ValueType::kTypeMerge, file_id++, &true_data));
+        options, {10, 11, 12, 13}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
+    // File doesn't overwrite any keys, no seqno needed
+    ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
+
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 4, 6}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {11, 15, 19},
-                                         ValueType::kTypeDeletion, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {11, 15, 19}, ValueType::kTypeDeletion, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {120, 130}, ValueType::kTypeMerge, file_id++, &true_data));
+        options, {120, 130}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1, 130}, ValueType::kTypeDeletion, file_id++, &true_data));
+        options, {1, 130}, ValueType::kTypeDeletion, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {120},
-                                         {ValueType::kTypeValue}, {{120, 135}},
-                                         file_id++, &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {120}, {ValueType::kTypeValue}, {{120, 135}}, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{110, 120}},
-                                         file_id++, &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {}, {}, {{110, 120}}, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // The range deletion ends on a key, but it doesn't actually delete
     // this key because the largest key in the range is exclusive. Still,
     // it counts as an overlap so a new seqno will be assigned.
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{100, 109}},
-                                         file_id++, &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {}, {}, {{100, 109}}, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
 
@@ -407,18 +434,20 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
     SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {60, 61, 62}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {60, 61, 62}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {40, 41, 42}, ValueType::kTypeMerge, file_id++, &true_data));
+        options, {40, 41, 42}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {20, 30, 40},
-                                         ValueType::kTypeDeletion, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {20, 30, 40}, ValueType::kTypeDeletion, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
@@ -427,34 +456,39 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMultipleValueType) {
     // We will need a seqno for the file regardless if the file overwrite
     // keys in the DB or not because we have a snapshot
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1000, 1002}, ValueType::kTypeMerge, file_id++, &true_data));
+        options, {1000, 1002}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {2000, 3002}, ValueType::kTypeMerge, file_id++, &true_data));
+        options, {2000, 3002}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
 
-    ASSERT_OK(GenerateAndAddExternalFile(options, {1, 20, 40, 100, 150},
-                                         ValueType::kTypeMerge, file_id++,
-                                         &true_data));
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, {1, 20, 40, 100, 150}, ValueType::kTypeMerge, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
     db_->ReleaseSnapshot(snapshot);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {5000, 5001}, ValueType::kTypeValue, file_id++, &true_data));
+        options, {5000, 5001}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data));
     // No snapshot anymore, no need to assign a seqno
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
     size_t kcnt = 0;
     VerifyDBFromMap(true_data, &kcnt, false);
-  } while (ChangeCompactOptions());
+  } while (ChangeOptionsForFileIngestionTest());
 }
 
-TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
+TEST_P(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   do {
     Options options = CurrentOptions();
     options.merge_operator.reset(new TestPutOperator());
@@ -467,7 +501,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {1, 2, 3, 4, 5, 6},
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
          ValueType::kTypeMerge, ValueType::kTypeValue, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
@@ -475,33 +510,40 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {10, 11, 12, 13},
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue,
          ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 0);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {1, 4, 6}, {ValueType::kTypeDeletion, ValueType::kTypeValue,
-                             ValueType::kTypeMerge},
-        file_id++, &true_data));
+        options, {1, 4, 6},
+        {ValueType::kTypeDeletion, ValueType::kTypeValue,
+         ValueType::kTypeMerge},
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 1);
 
     ASSERT_OK(GenerateAndAddExternalFile(
-        options, {11, 15, 19}, {ValueType::kTypeDeletion, ValueType::kTypeMerge,
-                                ValueType::kTypeValue},
-        file_id++, &true_data));
+        options, {11, 15, 19},
+        {ValueType::kTypeDeletion, ValueType::kTypeMerge,
+         ValueType::kTypeValue},
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {120, 130}, {ValueType::kTypeValue, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 2);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1, 130}, {ValueType::kTypeMerge, ValueType::kTypeDeletion},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
@@ -509,14 +551,16 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {150, 151, 152},
         {ValueType::kTypeValue, ValueType::kTypeMerge,
          ValueType::kTypeDeletion},
-        {{150, 160}, {180, 190}}, file_id++, &true_data));
+        {{150, 160}, {180, 190}}, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {150, 151, 152},
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
-        {{200, 250}}, file_id++, &true_data));
+        {{200, 250}}, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4);
 
@@ -524,7 +568,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {300, 301, 302},
         {ValueType::kTypeValue, ValueType::kTypeMerge,
          ValueType::kTypeDeletion},
-        {{1, 2}, {152, 154}}, file_id++, &true_data));
+        {{1, 2}, {152, 154}}, file_id++, write_global_seqno,
+        verify_checksums_before_ingest, &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 5);
 
@@ -538,7 +583,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {60, 61, 62},
         {ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeValue},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File doesn't overwrite any keys, no seqno needed
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
 
@@ -546,7 +592,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {40, 41, 42},
         {ValueType::kTypeValue, ValueType::kTypeDeletion,
          ValueType::kTypeDeletion},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 1);
 
@@ -554,7 +601,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {20, 30, 40},
         {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
          ValueType::kTypeDeletion},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // File overwrites some keys, a seqno will be assigned
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 2);
 
@@ -564,13 +612,15 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
     // keys in the DB or not because we have a snapshot
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {1000, 1002}, {ValueType::kTypeValue, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 3);
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {2000, 3002}, {ValueType::kTypeValue, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 4);
 
@@ -578,7 +628,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
         options, {1, 20, 40, 100, 150},
         {ValueType::kTypeDeletion, ValueType::kTypeDeletion,
          ValueType::kTypeValue, ValueType::kTypeMerge, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // A global seqno will be assigned anyway because of the snapshot
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
@@ -586,13 +637,14 @@ TEST_F(ExternalSSTFileBasicTest, IngestFileWithMixedValueType) {
 
     ASSERT_OK(GenerateAndAddExternalFile(
         options, {5000, 5001}, {ValueType::kTypeValue, ValueType::kTypeMerge},
-        file_id++, &true_data));
+        file_id++, write_global_seqno, verify_checksums_before_ingest,
+        &true_data));
     // No snapshot anymore, no need to assign a seqno
     ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno + 5);
 
     size_t kcnt = 0;
     VerifyDBFromMap(true_data, &kcnt, false);
-  } while (ChangeCompactOptions());
+  } while (ChangeOptionsForFileIngestionTest());
 }
 
 TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
@@ -635,7 +687,7 @@ TEST_F(ExternalSSTFileBasicTest, FadviseTrigger) {
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
 
-TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
+TEST_P(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   int kNumLevels = 7;
   Options options = CurrentOptions();
   options.disable_auto_compactions = true;
@@ -662,12 +714,15 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 1));
 
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   // overlaps with L0 file but not memtable, so flush is skipped and file is
   // ingested into L0
   SequenceNumber last_seqno = dbfull()->GetLatestSequenceNumber();
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {60, 90}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      {{65, 70}, {70, 85}}, file_id++, &true_data));
+      {{65, 70}, {70, 85}}, file_id++, write_global_seqno,
+      verify_checksums_before_ingest, &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
   ASSERT_EQ(2, NumTableFilesAtLevel(0));
   ASSERT_EQ(0, NumTableFilesAtLevel(kNumLevels - 2));
@@ -677,7 +732,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   // file is ingested into L5
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {10, 40}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      file_id++, write_global_seqno, verify_checksums_before_ingest,
+      &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
   ASSERT_EQ(2, NumTableFilesAtLevel(0));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
@@ -685,8 +741,9 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
 
   // overlaps with L5 file but not memtable or L0 file, so flush is skipped and
   // file is ingested into L4
-  ASSERT_OK(GenerateAndAddExternalFile(options, {}, {}, {{5, 15}}, file_id++,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {}, {}, {{5, 15}}, file_id++, write_global_seqno,
+      verify_checksums_before_ingest, &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
   ASSERT_EQ(2, NumTableFilesAtLevel(0));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
@@ -698,7 +755,8 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   // count increases by two.
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {100, 140}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      file_id++, &true_data));
+      file_id++, write_global_seqno, verify_checksums_before_ingest,
+      &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), ++last_seqno);
   ASSERT_EQ(4, NumTableFilesAtLevel(0));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
@@ -711,13 +769,102 @@ TEST_F(ExternalSSTFileBasicTest, IngestionWithRangeDeletions) {
   // seqnum.
   ASSERT_OK(GenerateAndAddExternalFile(
       options, {151, 175}, {ValueType::kTypeValue, ValueType::kTypeValue},
-      {{160, 200}}, file_id++, &true_data));
+      {{160, 200}}, file_id++, write_global_seqno,
+      verify_checksums_before_ingest, &true_data));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), last_seqno);
   ASSERT_EQ(4, NumTableFilesAtLevel(0));
   ASSERT_EQ(1, NumTableFilesAtLevel(kNumLevels - 2));
   ASSERT_EQ(2, NumTableFilesAtLevel(options.num_levels - 1));
 }
 
+TEST_P(ExternalSSTFileBasicTest, IngestFileWithBadBlockChecksum) {
+  bool change_checksum_called = false;
+  const auto& change_checksum = [&](void* arg) {
+    if (!change_checksum_called) {
+      char* buf = reinterpret_cast<char*>(arg);
+      assert(nullptr != buf);
+      buf[0] ^= 0x1;
+      change_checksum_called = true;
+    }
+  };
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->SetCallBack(
+      "BlockBasedTableBuilder::WriteRawBlock:TamperWithChecksum",
+      change_checksum);
+  SyncPoint::GetInstance()->EnableProcessing();
+  int file_id = 0;
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
+  do {
+    Options options = CurrentOptions();
+    DestroyAndReopen(options);
+    std::map<std::string, std::string> true_data;
+    Status s = GenerateAndAddExternalFile(
+        options, {1, 2, 3, 4, 5, 6}, ValueType::kTypeValue, file_id++,
+        write_global_seqno, verify_checksums_before_ingest, &true_data);
+    if (verify_checksums_before_ingest) {
+      ASSERT_NOK(s);
+    } else {
+      ASSERT_OK(s);
+    }
+    change_checksum_called = false;
+  } while (ChangeOptionsForFileIngestionTest());
+}
+
+TEST_P(ExternalSSTFileBasicTest, IngestFileWithFirstByteTampered) {
+  SyncPoint::GetInstance()->DisableProcessing();
+  int file_id = 0;
+  EnvOptions env_options;
+  do {
+    Options options = CurrentOptions();
+    std::string file_path = sst_files_dir_ + ToString(file_id++);
+    SstFileWriter sst_file_writer(env_options, options);
+    Status s = sst_file_writer.Open(file_path);
+    ASSERT_OK(s);
+    for (int i = 0; i != 100; ++i) {
+      std::string key = Key(i);
+      std::string value = Key(i) + ToString(0);
+      ASSERT_OK(sst_file_writer.Put(key, value));
+    }
+    ASSERT_OK(sst_file_writer.Finish());
+    {
+      // Get file size
+      uint64_t file_size = 0;
+      ASSERT_OK(env_->GetFileSize(file_path, &file_size));
+      ASSERT_GT(file_size, 8);
+      std::unique_ptr<RandomRWFile> rwfile;
+      ASSERT_OK(env_->NewRandomRWFile(file_path, &rwfile, EnvOptions()));
+      // Manually corrupt the file
+      // We deterministically corrupt the first byte because we currently
+      // cannot choose a random offset. The reason for this limitation is that
+      // we do not checksum property block at present.
+      const uint64_t offset = 0;
+      char scratch[8] = {0};
+      Slice buf;
+      ASSERT_OK(rwfile->Read(offset, sizeof(scratch), &buf, scratch));
+      scratch[0] ^= 0xff;  // flip one bit
+      ASSERT_OK(rwfile->Write(offset, buf));
+    }
+    // Ingest file.
+    IngestExternalFileOptions ifo;
+    ifo.write_global_seqno = std::get<0>(GetParam());
+    ifo.verify_checksums_before_ingest = std::get<1>(GetParam());
+    s = db_->IngestExternalFile({file_path}, ifo);
+    if (ifo.verify_checksums_before_ingest) {
+      ASSERT_NOK(s);
+    } else {
+      ASSERT_OK(s);
+    }
+  } while (ChangeOptionsForFileIngestionTest());
+}
+
+INSTANTIATE_TEST_CASE_P(ExternalSSTFileBasicTest, ExternalSSTFileBasicTest,
+                        testing::Values(std::make_tuple(true, true),
+                                        std::make_tuple(true, false),
+                                        std::make_tuple(false, true),
+                                        std::make_tuple(false, false)));
+
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

db/external_sst_file_ingestion_job.cc

@@ -225,7 +225,7 @@ void ExternalSstFileIngestionJob::UpdateStats() {
   for (IngestedFileInfo& f : files_to_ingest_) {
     InternalStats::CompactionStats stats(CompactionReason::kExternalSstIngestion, 1);
     stats.micros = total_time;
-    // If actual copy occured for this file, then we need to count the file
+    // If actual copy occurred for this file, then we need to count the file
     // size as the actual bytes written. If the file was linked, then we ignore
     // the bytes written for file metadata.
     // TODO (yanqin) maybe account for file metadata bytes for exact accuracy?
@@ -316,6 +316,13 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
     return status;
   }
 
+  if (ingestion_options_.verify_checksums_before_ingest) {
+    status = table_reader->VerifyChecksum();
+  }
+  if (!status.ok()) {
+    return status;
+  }
+
   // Get the external file properties
   auto props = table_reader->GetTableProperties();
   const auto& uprops = props->user_collected_properties;
@@ -344,7 +351,7 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
       file_to_ingest->global_seqno_offset = 0;
       return Status::Corruption("Was not able to find file global seqno field");
     }
-    file_to_ingest->global_seqno_offset = offsets_iter->second;
+    file_to_ingest->global_seqno_offset = static_cast<size_t>(offsets_iter->second);
   } else if (file_to_ingest->version == 1) {
     // SST file V1 should not have global seqno field
     assert(seqno_iter == uprops.end());

db/external_sst_file_test.cc

@@ -15,7 +15,9 @@
 
 namespace rocksdb {
 
-class ExternalSSTFileTest : public DBTestBase {
+class ExternalSSTFileTest
+    : public DBTestBase,
+      public ::testing::WithParamInterface<std::tuple<bool, bool>> {
  public:
   ExternalSSTFileTest() : DBTestBase("/external_sst_file_test") {
     sst_files_dir_ = dbname_ + "/sst_files/";
@@ -30,7 +32,8 @@ class ExternalSSTFileTest : public DBTestBase {
   Status GenerateAndAddExternalFile(
       const Options options,
       std::vector<std::pair<std::string, std::string>> data, int file_id = -1,
-      bool allow_global_seqno = false, bool sort_data = false,
+      bool allow_global_seqno = false, bool write_global_seqno = false,
+      bool verify_checksums_before_ingest = true, bool sort_data = false,
       std::map<std::string, std::string>* true_data = nullptr,
       ColumnFamilyHandle* cfh = nullptr) {
     // Generate a file id if not provided
@@ -73,6 +76,8 @@ class ExternalSSTFileTest : public DBTestBase {
     if (s.ok()) {
       IngestExternalFileOptions ifo;
       ifo.allow_global_seqno = allow_global_seqno;
+      ifo.write_global_seqno = allow_global_seqno ? write_global_seqno : false;
+      ifo.verify_checksums_before_ingest = verify_checksums_before_ingest;
       if (cfh) {
         s = db_->IngestExternalFile(cfh, {file_path}, ifo);
       } else {
@@ -149,44 +154,47 @@ class ExternalSSTFileTest : public DBTestBase {
     return s;
   }
 
-
-
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<std::pair<int, std::string>> data,
-      int file_id = -1, bool allow_global_seqno = false, bool sort_data = false,
+      int file_id = -1, bool allow_global_seqno = false,
+      bool write_global_seqno = false,
+      bool verify_checksums_before_ingest = true, bool sort_data = false,
       std::map<std::string, std::string>* true_data = nullptr,
       ColumnFamilyHandle* cfh = nullptr) {
     std::vector<std::pair<std::string, std::string>> file_data;
     for (auto& entry : data) {
       file_data.emplace_back(Key(entry.first), entry.second);
     }
-    return GenerateAndAddExternalFile(options, file_data, file_id,
-                                      allow_global_seqno, sort_data, true_data,
-                                      cfh);
+    return GenerateAndAddExternalFile(
+        options, file_data, file_id, allow_global_seqno, write_global_seqno,
+        verify_checksums_before_ingest, sort_data, true_data, cfh);
   }
 
   Status GenerateAndAddExternalFile(
       const Options options, std::vector<int> keys, int file_id = -1,
-      bool allow_global_seqno = false, bool sort_data = false,
+      bool allow_global_seqno = false, bool write_global_seqno = false,
+      bool verify_checksums_before_ingest = true, bool sort_data = false,
       std::map<std::string, std::string>* true_data = nullptr,
       ColumnFamilyHandle* cfh = nullptr) {
     std::vector<std::pair<std::string, std::string>> file_data;
     for (auto& k : keys) {
       file_data.emplace_back(Key(k), Key(k) + ToString(file_id));
     }
-    return GenerateAndAddExternalFile(options, file_data, file_id,
-                                      allow_global_seqno, sort_data, true_data,
-                                      cfh);
+    return GenerateAndAddExternalFile(
+        options, file_data, file_id, allow_global_seqno, write_global_seqno,
+        verify_checksums_before_ingest, sort_data, true_data, cfh);
   }
 
   Status DeprecatedAddFile(const std::vector<std::string>& files,
                            bool move_files = false,
-                           bool skip_snapshot_check = false) {
+                           bool skip_snapshot_check = false,
+                           bool skip_write_global_seqno = false) {
     IngestExternalFileOptions opts;
     opts.move_files = move_files;
     opts.snapshot_consistency = !skip_snapshot_check;
     opts.allow_global_seqno = false;
     opts.allow_blocking_flush = false;
+    opts.write_global_seqno = !skip_write_global_seqno;
     return db_->IngestExternalFile(files, opts);
   }
 
@@ -453,6 +461,7 @@ TEST_F(ExternalSSTFileTest, Basic) {
   } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction |
                          kRangeDelSkipConfigs));
 }
+
 class SstFileWriterCollector : public TablePropertiesCollector {
  public:
   explicit SstFileWriterCollector(const std::string prefix) : prefix_(prefix) {
@@ -1141,7 +1150,7 @@ TEST_F(ExternalSSTFileTest, OverlappingRanges) {
   } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }
 
-TEST_F(ExternalSSTFileTest, PickedLevel) {
+TEST_P(ExternalSSTFileTest, PickedLevel) {
   Options options = CurrentOptions();
   options.disable_auto_compactions = false;
   options.level0_file_num_compaction_trigger = 4;
@@ -1151,13 +1160,13 @@ TEST_F(ExternalSSTFileTest, PickedLevel) {
   std::map<std::string, std::string> true_data;
 
   // File 0 will go to last level (L3)
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 10}, -1, false, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 10}, -1, false, false, true,
+                                       false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "0,0,0,1");
 
   // File 1 will go to level L2 (since it overlap with file 0 in L3)
-  ASSERT_OK(GenerateAndAddExternalFile(options, {2, 9}, -1, false, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {2, 9}, -1, false, false, true,
+                                       false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "0,0,1,1");
 
   rocksdb::SyncPoint::GetInstance()->LoadDependency({
@@ -1186,13 +1195,13 @@ TEST_F(ExternalSSTFileTest, PickedLevel) {
 
   // This file overlaps with file 0 (L3), file 1 (L2) and the
   // output of compaction going to L1
-  ASSERT_OK(GenerateAndAddExternalFile(options, {4, 7}, -1, false, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {4, 7}, -1, false, false, true,
+                                       false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "5,0,1,1");
 
   // This file does not overlap with any file or with the running compaction
   ASSERT_OK(GenerateAndAddExternalFile(options, {9000, 9001}, -1, false, false,
-                                       &true_data));
+                                       true, false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "5,0,1,2");
 
   // Hold compaction from finishing
@@ -1311,6 +1320,7 @@ TEST_F(ExternalSSTFileTest, IngestNonExistingFile) {
   ASSERT_OK(Flush());
 
   ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
 
   // After full compaction, there should be only 1 file.
   std::vector<std::string> files;
@@ -1422,12 +1432,12 @@ TEST_F(ExternalSSTFileTest, PickedLevelDynamic) {
   // This file overlaps with the output of the compaction (going to L3)
   // so the file will be added to L0 since L3 is the base level
   ASSERT_OK(GenerateAndAddExternalFile(options, {31, 32, 33, 34}, -1, false,
-                                       false, &true_data));
+                                       false, true, false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "5");
 
   // This file does not overlap with the current running compactiong
   ASSERT_OK(GenerateAndAddExternalFile(options, {9000, 9001}, -1, false, false,
-                                       &true_data));
+                                       true, false, &true_data));
   EXPECT_EQ(FilesPerLevel(), "5,0,0,1");
 
   // Hold compaction from finishing
@@ -1442,25 +1452,25 @@ TEST_F(ExternalSSTFileTest, PickedLevelDynamic) {
   Reopen(options);
 
   ASSERT_OK(GenerateAndAddExternalFile(options, {1, 15, 19}, -1, false, false,
-                                       &true_data));
+                                       true, false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "1,0,0,3");
 
   ASSERT_OK(GenerateAndAddExternalFile(options, {1000, 1001, 1002}, -1, false,
-                                       false, &true_data));
+                                       false, true, false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "1,0,0,4");
 
   ASSERT_OK(GenerateAndAddExternalFile(options, {500, 600, 700}, -1, false,
-                                       false, &true_data));
+                                       false, true, false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "1,0,0,5");
 
   // File 5 overlaps with file 2 (L3 / base level)
-  ASSERT_OK(GenerateAndAddExternalFile(options, {2, 10}, -1, false, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {2, 10}, -1, false, false, true,
+                                       false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "2,0,0,5");
 
   // File 6 overlaps with file 2 (L3 / base level) and file 5 (L0)
-  ASSERT_OK(GenerateAndAddExternalFile(options, {3, 9}, -1, false, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {3, 9}, -1, false, false, true,
+                                       false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "3,0,0,5");
 
   // Verify data in files
@@ -1479,7 +1489,7 @@ TEST_F(ExternalSSTFileTest, PickedLevelDynamic) {
 
   // File 7 overlaps with file 4 (L3)
   ASSERT_OK(GenerateAndAddExternalFile(options, {650, 651, 652}, -1, false,
-                                       false, &true_data));
+                                       false, true, false, &true_data));
   ASSERT_EQ(FilesPerLevel(), "5,0,0,5");
 
   VerifyDBFromMap(true_data, &kcnt, false);
@@ -1613,12 +1623,14 @@ TEST_F(ExternalSSTFileTest, SstFileWriterNonSharedKeys) {
   ASSERT_OK(DeprecatedAddFile({file_path}));
 }
 
-TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
+TEST_P(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
   Options options = CurrentOptions();
   options.IncreaseParallelism(20);
   options.level0_slowdown_writes_trigger = 256;
   options.level0_stop_writes_trigger = 256;
 
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   for (int iter = 0; iter < 2; iter++) {
     bool write_to_memtable = (iter == 0);
     DestroyAndReopen(options);
@@ -1642,8 +1654,9 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
           true_data[entry.first] = entry.second;
         }
       } else {
-        ASSERT_OK(GenerateAndAddExternalFile(options, random_data, -1, true,
-                                             true, &true_data));
+        ASSERT_OK(GenerateAndAddExternalFile(
+            options, random_data, -1, true, write_global_seqno,
+            verify_checksums_before_ingest, true, &true_data));
       }
     }
     size_t kcnt = 0;
@@ -1653,7 +1666,7 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoRandomized) {
   }
 }
 
-TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
+TEST_P(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   Options options = CurrentOptions();
   options.num_levels = 5;
   options.disable_auto_compactions = true;
@@ -1672,8 +1685,11 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   for (int i = 0; i <= 20; i++) {
     file_data.emplace_back(Key(i), "L4");
   }
-  ASSERT_OK(GenerateAndAddExternalFile(options, file_data, -1, true, false,
-                                       &true_data));
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, file_data, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
 
   // This file dont overlap with anything in the DB, will go to L4
   ASSERT_EQ("0,0,0,0,1", FilesPerLevel());
@@ -1683,8 +1699,9 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   for (int i = 80; i <= 130; i++) {
     file_data.emplace_back(Key(i), "L0");
   }
-  ASSERT_OK(GenerateAndAddExternalFile(options, file_data, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, file_data, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
 
   // This file overlap with the memtable, so it will flush it and add
   // it self to L0
@@ -1695,8 +1712,9 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   for (int i = 30; i <= 50; i++) {
     file_data.emplace_back(Key(i), "L4");
   }
-  ASSERT_OK(GenerateAndAddExternalFile(options, file_data, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, file_data, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
 
   // This file dont overlap with anything in the DB and fit in L4 as well
   ASSERT_EQ("2,0,0,0,2", FilesPerLevel());
@@ -1706,8 +1724,9 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   for (int i = 10; i <= 40; i++) {
     file_data.emplace_back(Key(i), "L3");
   }
-  ASSERT_OK(GenerateAndAddExternalFile(options, file_data, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, file_data, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
 
   // This file overlap with files in L4, we will ingest it in L3
   ASSERT_EQ("2,0,0,1,2", FilesPerLevel());
@@ -1716,7 +1735,7 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoAssignedLevel) {
   VerifyDBFromMap(true_data, &kcnt, false);
 }
 
-TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
+TEST_P(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
   Options options = CurrentOptions();
   DestroyAndReopen(options);
   uint64_t entries_in_memtable;
@@ -1730,16 +1749,20 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
                       &entries_in_memtable);
   ASSERT_GE(entries_in_memtable, 1);
 
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   // No need for flush
-  ASSERT_OK(GenerateAndAddExternalFile(options, {90, 100, 110}, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {90, 100, 110}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
   db_->GetIntProperty(DB::Properties::kNumEntriesActiveMemTable,
                       &entries_in_memtable);
   ASSERT_GE(entries_in_memtable, 1);
 
   // This file will flush the memtable
-  ASSERT_OK(GenerateAndAddExternalFile(options, {19, 20, 21}, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {19, 20, 21}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
   db_->GetIntProperty(DB::Properties::kNumEntriesActiveMemTable,
                       &entries_in_memtable);
   ASSERT_EQ(entries_in_memtable, 0);
@@ -1753,15 +1776,17 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
   ASSERT_GE(entries_in_memtable, 1);
 
   // No need for flush, this file keys fit between the memtable keys
-  ASSERT_OK(GenerateAndAddExternalFile(options, {202, 203, 204}, -1, true,
-                                       false, &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {202, 203, 204}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, false, &true_data));
   db_->GetIntProperty(DB::Properties::kNumEntriesActiveMemTable,
                       &entries_in_memtable);
   ASSERT_GE(entries_in_memtable, 1);
 
   // This file will flush the memtable
-  ASSERT_OK(GenerateAndAddExternalFile(options, {206, 207}, -1, true, false,
-                                       &true_data));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {206, 207}, -1, true, false, write_global_seqno,
+      verify_checksums_before_ingest, &true_data));
   db_->GetIntProperty(DB::Properties::kNumEntriesActiveMemTable,
                       &entries_in_memtable);
   ASSERT_EQ(entries_in_memtable, 0);
@@ -1770,7 +1795,7 @@ TEST_F(ExternalSSTFileTest, IngestFileWithGlobalSeqnoMemtableFlush) {
   VerifyDBFromMap(true_data, &kcnt, false);
 }
 
-TEST_F(ExternalSSTFileTest, L0SortingIssue) {
+TEST_P(ExternalSSTFileTest, L0SortingIssue) {
   Options options = CurrentOptions();
   options.num_levels = 2;
   DestroyAndReopen(options);
@@ -1779,10 +1804,16 @@ TEST_F(ExternalSSTFileTest, L0SortingIssue) {
   ASSERT_OK(Put(Key(1), "memtable"));
   ASSERT_OK(Put(Key(10), "memtable"));
 
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   // No Flush needed, No global seqno needed, Ingest in L1
-  ASSERT_OK(GenerateAndAddExternalFile(options, {7, 8}, -1, true, false));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {7, 8}, -1, true,
+                                       write_global_seqno,
+                                       verify_checksums_before_ingest, false));
   // No Flush needed, but need a global seqno, Ingest in L0
-  ASSERT_OK(GenerateAndAddExternalFile(options, {7, 8}, -1, true, false));
+  ASSERT_OK(GenerateAndAddExternalFile(options, {7, 8}, -1, true,
+                                       write_global_seqno,
+                                       verify_checksums_before_ingest, false));
   printf("%s\n", FilesPerLevel().c_str());
 
   // Overwrite what we added using external files
@@ -2011,16 +2042,19 @@ class TestIngestExternalFileListener : public EventListener {
   std::vector<ExternalFileIngestionInfo> ingested_files;
 };
 
-TEST_F(ExternalSSTFileTest, IngestionListener) {
+TEST_P(ExternalSSTFileTest, IngestionListener) {
   Options options = CurrentOptions();
   TestIngestExternalFileListener* listener =
       new TestIngestExternalFileListener();
   options.listeners.emplace_back(listener);
   CreateAndReopenWithCF({"koko", "toto"}, options);
 
+  bool write_global_seqno = std::get<0>(GetParam());
+  bool verify_checksums_before_ingest = std::get<1>(GetParam());
   // Ingest into default cf
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
-                                       handles_[0]));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {1, 2}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, true, nullptr, handles_[0]));
   ASSERT_EQ(listener->ingested_files.size(), 1);
   ASSERT_EQ(listener->ingested_files.back().cf_name, "default");
   ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
@@ -2030,8 +2064,9 @@ TEST_F(ExternalSSTFileTest, IngestionListener) {
             "default");
 
   // Ingest into cf1
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
-                                       handles_[1]));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {1, 2}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, true, nullptr, handles_[1]));
   ASSERT_EQ(listener->ingested_files.size(), 2);
   ASSERT_EQ(listener->ingested_files.back().cf_name, "koko");
   ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
@@ -2041,8 +2076,9 @@ TEST_F(ExternalSSTFileTest, IngestionListener) {
             "koko");
 
   // Ingest into cf2
-  ASSERT_OK(GenerateAndAddExternalFile(options, {1, 2}, -1, true, true, nullptr,
-                                       handles_[2]));
+  ASSERT_OK(GenerateAndAddExternalFile(
+      options, {1, 2}, -1, true, write_global_seqno,
+      verify_checksums_before_ingest, true, nullptr, handles_[2]));
   ASSERT_EQ(listener->ingested_files.size(), 3);
   ASSERT_EQ(listener->ingested_files.back().cf_name, "toto");
   ASSERT_EQ(listener->ingested_files.back().global_seqno, 0);
@@ -2084,7 +2120,7 @@ TEST_F(ExternalSSTFileTest, SnapshotInconsistencyBug) {
   db_->ReleaseSnapshot(snap);
 }
 
-TEST_F(ExternalSSTFileTest, IngestBehind) {
+TEST_P(ExternalSSTFileTest, IngestBehind) {
   Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.num_levels = 3;
@@ -2108,6 +2144,8 @@ TEST_F(ExternalSSTFileTest, IngestBehind) {
   IngestExternalFileOptions ifo;
   ifo.allow_global_seqno = true;
   ifo.ingest_behind = true;
+  ifo.write_global_seqno = std::get<0>(GetParam());
+  ifo.verify_checksums_before_ingest = std::get<1>(GetParam());
 
   // Can't ingest behind since allow_ingest_behind isn't set to true
   ASSERT_NOK(GenerateAndAddExternalFileIngestBehind(options, ifo,
@@ -2195,6 +2233,12 @@ TEST_F(ExternalSSTFileTest, SkipBloomFilter) {
   }
 }
 
+INSTANTIATE_TEST_CASE_P(ExternalSSTFileTest, ExternalSSTFileTest,
+                        testing::Values(std::make_tuple(false, false),
+                                        std::make_tuple(false, true),
+                                        std::make_tuple(true, false),
+                                        std::make_tuple(true, true)));
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/fault_injection_test.cc

@@ -70,7 +70,7 @@ class FaultInjectionTest
   std::unique_ptr<Env> base_env_;
   FaultInjectionTestEnv* env_;
   std::string dbname_;
-  shared_ptr<Cache> tiny_cache_;
+  std::shared_ptr<Cache> tiny_cache_;
   Options options_;
   DB* db_;
 

db/flush_job.cc

@@ -24,14 +24,15 @@
 #include "db/event_helpers.h"
 #include "db/log_reader.h"
 #include "db/log_writer.h"
+#include "db/memtable.h"
 #include "db/memtable_list.h"
 #include "db/merge_context.h"
+#include "db/range_tombstone_fragmenter.h"
 #include "db/version_set.h"
 #include "monitoring/iostats_context_imp.h"
 #include "monitoring/perf_context_imp.h"
 #include "monitoring/thread_status_util.h"
 #include "port/port.h"
-#include "db/memtable.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/statistics.h"
@@ -78,16 +79,18 @@ const char* GetFlushReasonString (FlushReason flush_reason) {
       return "Auto Compaction";
     case FlushReason::kManualFlush:
       return "Manual Flush";
+    case FlushReason::kErrorRecovery:
+      return "Error Recovery";
     default:
       return "Invalid";
   }
 }
 
-
 FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                    const ImmutableDBOptions& db_options,
                    const MutableCFOptions& mutable_cf_options,
-                   const EnvOptions env_options, VersionSet* versions,
+                   const uint64_t* max_memtable_id,
+                   const EnvOptions& env_options, VersionSet* versions,
                    InstrumentedMutex* db_mutex,
                    std::atomic<bool>* shutting_down,
                    std::vector<SequenceNumber> existing_snapshots,
@@ -96,11 +99,13 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                    LogBuffer* log_buffer, Directory* db_directory,
                    Directory* output_file_directory,
                    CompressionType output_compression, Statistics* stats,
-                   EventLogger* event_logger, bool measure_io_stats)
+                   EventLogger* event_logger, bool measure_io_stats,
+                   const bool sync_output_directory, const bool write_manifest)
     : dbname_(dbname),
       cfd_(cfd),
       db_options_(db_options),
       mutable_cf_options_(mutable_cf_options),
+      max_memtable_id_(max_memtable_id),
       env_options_(env_options),
       versions_(versions),
       db_mutex_(db_mutex),
@@ -116,6 +121,8 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
       stats_(stats),
       event_logger_(event_logger),
       measure_io_stats_(measure_io_stats),
+      sync_output_directory_(sync_output_directory),
+      write_manifest_(write_manifest),
       edit_(nullptr),
       base_(nullptr),
       pick_memtable_called(false) {
@@ -160,7 +167,7 @@ void FlushJob::PickMemTable() {
   assert(!pick_memtable_called);
   pick_memtable_called = true;
   // Save the contents of the earliest memtable as a new Table
-  cfd_->imm()->PickMemtablesToFlush(&mems_);
+  cfd_->imm()->PickMemtablesToFlush(max_memtable_id_, &mems_);
   if (mems_.empty()) {
     return;
   }
@@ -204,6 +211,8 @@ Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
   uint64_t prev_fsync_nanos = 0;
   uint64_t prev_range_sync_nanos = 0;
   uint64_t prev_prepare_write_nanos = 0;
+  uint64_t prev_cpu_write_nanos = 0;
+  uint64_t prev_cpu_read_nanos = 0;
   if (measure_io_stats_) {
     prev_perf_level = GetPerfLevel();
     SetPerfLevel(PerfLevel::kEnableTime);
@@ -211,6 +220,8 @@ Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
     prev_fsync_nanos = IOSTATS(fsync_nanos);
     prev_range_sync_nanos = IOSTATS(range_sync_nanos);
     prev_prepare_write_nanos = IOSTATS(prepare_write_nanos);
+    prev_cpu_write_nanos = IOSTATS(cpu_write_nanos);
+    prev_cpu_read_nanos = IOSTATS(cpu_read_nanos);
   }
 
   // This will release and re-acquire the mutex.
@@ -224,10 +235,10 @@ Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
 
   if (!s.ok()) {
     cfd_->imm()->RollbackMemtableFlush(mems_, meta_.fd.GetNumber());
-  } else {
+  } else if (write_manifest_) {
     TEST_SYNC_POINT("FlushJob::InstallResults");
     // Replace immutable memtable with the generated Table
-    s = cfd_->imm()->InstallMemtableFlushResults(
+    s = cfd_->imm()->TryInstallMemtableFlushResults(
         cfd_, mutable_cf_options_, mems_, prep_tracker, versions_, db_mutex_,
         meta_.fd.GetNumber(), &job_context_->memtables_to_free, db_directory_,
         log_buffer_);
@@ -262,6 +273,10 @@ Status FlushJob::Run(LogsWithPrepTracker* prep_tracker,
     stream << "file_fsync_nanos" << (IOSTATS(fsync_nanos) - prev_fsync_nanos);
     stream << "file_prepare_write_nanos"
            << (IOSTATS(prepare_write_nanos) - prev_prepare_write_nanos);
+    stream << "file_cpu_write_nanos"
+           << (IOSTATS(cpu_write_nanos) - prev_cpu_write_nanos);
+    stream << "file_cpu_read_nanos"
+           << (IOSTATS(cpu_read_nanos) - prev_cpu_read_nanos);
   }
 
   return s;
@@ -278,6 +293,7 @@ Status FlushJob::WriteLevel0Table() {
       ThreadStatus::STAGE_FLUSH_WRITE_L0);
   db_mutex_->AssertHeld();
   const uint64_t start_micros = db_options_.env->NowMicros();
+  const uint64_t start_cpu_micros = db_options_.env->NowCPUNanos() / 1000;
   Status s;
   {
     auto write_hint = cfd_->CalculateSSTWriteHint(0);
@@ -289,7 +305,8 @@ Status FlushJob::WriteLevel0Table() {
     // memtable and its associated range deletion memtable, respectively, at
     // corresponding indexes.
     std::vector<InternalIterator*> memtables;
-    std::vector<InternalIterator*> range_del_iters;
+    std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
+        range_del_iters;
     ReadOptions ro;
     ro.total_order_seek = true;
     Arena arena;
@@ -301,9 +318,10 @@ Status FlushJob::WriteLevel0Table() {
           "[%s] [JOB %d] Flushing memtable with next log file: %" PRIu64 "\n",
           cfd_->GetName().c_str(), job_context_->job_id, m->GetNextLogNumber());
       memtables.push_back(m->NewIterator(ro, &arena));
-      auto* range_del_iter = m->NewRangeTombstoneIterator(ro);
+      auto* range_del_iter =
+          m->NewRangeTombstoneIterator(ro, kMaxSequenceNumber);
       if (range_del_iter != nullptr) {
-        range_del_iters.push_back(range_del_iter);
+        range_del_iters.emplace_back(range_del_iter);
       }
       total_num_entries += m->num_entries();
       total_num_deletes += m->num_deletes();
@@ -322,10 +340,6 @@ Status FlushJob::WriteLevel0Table() {
       ScopedArenaIterator iter(
           NewMergingIterator(&cfd_->internal_comparator(), &memtables[0],
                              static_cast<int>(memtables.size()), &arena));
-      std::unique_ptr<InternalIterator> range_del_iter(NewMergingIterator(
-          &cfd_->internal_comparator(),
-          range_del_iters.empty() ? nullptr : &range_del_iters[0],
-          static_cast<int>(range_del_iters.size())));
       ROCKS_LOG_INFO(db_options_.info_log,
                      "[%s] [JOB %d] Level-0 flush table #%" PRIu64 ": started",
                      cfd_->GetName().c_str(), job_context_->job_id,
@@ -351,7 +365,7 @@ Status FlushJob::WriteLevel0Table() {
       s = BuildTable(
           dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
           env_options_, cfd_->table_cache(), iter.get(),
-          std::move(range_del_iter), &meta_, cfd_->internal_comparator(),
+          std::move(range_del_iters), &meta_, cfd_->internal_comparator(),
           cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
           cfd_->GetName(), existing_snapshots_,
           earliest_write_conflict_snapshot_, snapshot_checker_,
@@ -371,8 +385,8 @@ Status FlushJob::WriteLevel0Table() {
                    s.ToString().c_str(),
                    meta_.marked_for_compaction ? " (needs compaction)" : "");
 
-    if (output_file_directory_ != nullptr) {
-      output_file_directory_->Fsync();
+    if (s.ok() && output_file_directory_ != nullptr && sync_output_directory_) {
+      s = output_file_directory_->Fsync();
     }
     TEST_SYNC_POINT("FlushJob::WriteLevel0Table");
     db_mutex_->Lock();
@@ -396,6 +410,7 @@ Status FlushJob::WriteLevel0Table() {
   // Note that here we treat flush as level 0 compaction in internal stats
   InternalStats::CompactionStats stats(CompactionReason::kFlush, 1);
   stats.micros = db_options_.env->NowMicros() - start_micros;
+  stats.cpu_micros = db_options_.env->NowCPUNanos() / 1000 - start_cpu_micros;
   stats.bytes_written = meta_.fd.GetFileSize();
   MeasureTime(stats_, FLUSH_TIME, stats.micros);
   cfd_->internal_stats()->AddCompactionStats(0 /* level */, stats);

db/flush_job.h

@@ -59,14 +59,16 @@ class FlushJob {
   FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
            const ImmutableDBOptions& db_options,
            const MutableCFOptions& mutable_cf_options,
-           const EnvOptions env_options, VersionSet* versions,
-           InstrumentedMutex* db_mutex, std::atomic<bool>* shutting_down,
+           const uint64_t* max_memtable_id, const EnvOptions& env_options,
+           VersionSet* versions, InstrumentedMutex* db_mutex,
+           std::atomic<bool>* shutting_down,
            std::vector<SequenceNumber> existing_snapshots,
            SequenceNumber earliest_write_conflict_snapshot,
            SnapshotChecker* snapshot_checker, JobContext* job_context,
            LogBuffer* log_buffer, Directory* db_directory,
            Directory* output_file_directory, CompressionType output_compression,
-           Statistics* stats, EventLogger* event_logger, bool measure_io_stats);
+           Statistics* stats, EventLogger* event_logger, bool measure_io_stats,
+           const bool sync_output_directory, const bool write_manifest);
 
   ~FlushJob();
 
@@ -77,16 +79,24 @@ class FlushJob {
              FileMetaData* file_meta = nullptr);
   void Cancel();
   TableProperties GetTableProperties() const { return table_properties_; }
+  const autovector<MemTable*>& GetMemTables() const { return mems_; }
 
  private:
   void ReportStartedFlush();
   void ReportFlushInputSize(const autovector<MemTable*>& mems);
   void RecordFlushIOStats();
   Status WriteLevel0Table();
+
   const std::string& dbname_;
   ColumnFamilyData* cfd_;
   const ImmutableDBOptions& db_options_;
   const MutableCFOptions& mutable_cf_options_;
+  // Pointer to a variable storing the largest memtable id to flush in this
+  // flush job. RocksDB uses this variable to select the memtables to flush in
+  // this job. All memtables in this column family with an ID smaller than or
+  // equal to *max_memtable_id_ will be selected for flush. If null, then all
+  // memtables in the column family will be selected.
+  const uint64_t* max_memtable_id_;
   const EnvOptions env_options_;
   VersionSet* versions_;
   InstrumentedMutex* db_mutex_;
@@ -103,6 +113,23 @@ class FlushJob {
   EventLogger* event_logger_;
   TableProperties table_properties_;
   bool measure_io_stats_;
+  // True if this flush job should call fsync on the output directory. False
+  // otherwise.
+  // Usually sync_output_directory_ is true. A flush job needs to call sync on
+  // the output directory before committing to the MANIFEST.
+  // However, an individual flush job does not have to call sync on the output
+  // directory if it is part of an atomic flush. After all flush jobs in the
+  // atomic flush succeed, call sync once on each distinct output directory.
+  const bool sync_output_directory_;
+  // True if this flush job should write to MANIFEST after successfully
+  // flushing memtables. False otherwise.
+  // Usually write_manifest_ is true. A flush job commits to the MANIFEST after
+  // flushing the memtables.
+  // However, an individual flush job cannot rashly write to the MANIFEST
+  // immediately after it finishes the flush if it is part of an atomic flush.
+  // In this case, only after all flush jobs succeed in flush can RocksDB
+  // commit to the MANIFEST.
+  const bool write_manifest_;
 
   // Variables below are set by PickMemTable():
   FileMetaData meta_;

db/flush_job_test.cc

@@ -30,6 +30,7 @@ class FlushJobTest : public testing::Test {
         dbname_(test::PerThreadDBPath("flush_job_test")),
         options_(),
         db_options_(options_),
+        column_family_names_({kDefaultColumnFamilyName, "foo", "bar"}),
         table_cache_(NewLRUCache(50000, 16)),
         write_buffer_manager_(db_options_.db_write_buffer_size),
         versions_(new VersionSet(dbname_, &db_options_, env_options_,
@@ -45,7 +46,9 @@ class FlushJobTest : public testing::Test {
     NewDB();
     std::vector<ColumnFamilyDescriptor> column_families;
     cf_options_.table_factory = mock_table_factory_;
-    column_families.emplace_back(kDefaultColumnFamilyName, cf_options_);
+    for (const auto& cf_name : column_family_names_) {
+      column_families.emplace_back(cf_name, cf_options_);
+    }
 
     EXPECT_OK(versions_->Recover(column_families, false));
   }
@@ -56,18 +59,38 @@ class FlushJobTest : public testing::Test {
     new_db.SetNextFile(2);
     new_db.SetLastSequence(0);
 
+    autovector<VersionEdit> new_cfs;
+    SequenceNumber last_seq = 1;
+    uint32_t cf_id = 1;
+    for (size_t i = 1; i != column_family_names_.size(); ++i) {
+      VersionEdit new_cf;
+      new_cf.AddColumnFamily(column_family_names_[i]);
+      new_cf.SetColumnFamily(cf_id++);
+      new_cf.SetLogNumber(0);
+      new_cf.SetNextFile(2);
+      new_cf.SetLastSequence(last_seq++);
+      new_cfs.emplace_back(new_cf);
+    }
+
     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), EnvOptions()));
+    std::unique_ptr<WritableFileWriter> file_writer(
+        new WritableFileWriter(std::move(file), manifest, EnvOptions()));
     {
       log::Writer log(std::move(file_writer), 0, false);
       std::string record;
       new_db.EncodeTo(&record);
       s = log.AddRecord(record);
+
+      for (const auto& e : new_cfs) {
+        record.clear();
+        e.EncodeTo(&record);
+        s = log.AddRecord(record);
+        ASSERT_OK(s);
+      }
     }
     ASSERT_OK(s);
     // Make "CURRENT" file that points to the new manifest file.
@@ -79,6 +102,7 @@ class FlushJobTest : public testing::Test {
   EnvOptions env_options_;
   Options options_;
   ImmutableDBOptions db_options_;
+  const std::vector<std::string> column_family_names_;
   std::shared_ptr<Cache> table_cache_;
   WriteController write_controller_;
   WriteBufferManager write_buffer_manager_;
@@ -96,9 +120,11 @@ TEST_F(FlushJobTest, Empty) {
   SnapshotChecker* snapshot_checker = nullptr;  // not relavant
   FlushJob flush_job(
       dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
-      *cfd->GetLatestMutableCFOptions(), env_options_, versions_.get(), &mutex_,
-      &shutting_down_, {}, kMaxSequenceNumber, snapshot_checker, &job_context,
-      nullptr, nullptr, nullptr, kNoCompression, nullptr, &event_logger, false);
+      *cfd->GetLatestMutableCFOptions(), nullptr /* memtable_id */,
+      env_options_, versions_.get(), &mutex_, &shutting_down_, {},
+      kMaxSequenceNumber, snapshot_checker, &job_context, nullptr, nullptr,
+      nullptr, kNoCompression, nullptr, &event_logger, false,
+      true /* sync_output_directory */, true /* write_manifest */);
   {
     InstrumentedMutexLock l(&mutex_);
     flush_job.PickMemTable();
@@ -139,12 +165,13 @@ TEST_F(FlushJobTest, NonEmpty) {
 
   EventLogger event_logger(db_options_.info_log.get());
   SnapshotChecker* snapshot_checker = nullptr;  // not relavant
-  FlushJob flush_job(dbname_, versions_->GetColumnFamilySet()->GetDefault(),
-                     db_options_, *cfd->GetLatestMutableCFOptions(),
-                     env_options_, versions_.get(), &mutex_, &shutting_down_,
-                     {}, kMaxSequenceNumber, snapshot_checker, &job_context,
-                     nullptr, nullptr, nullptr, kNoCompression,
-                     db_options_.statistics.get(), &event_logger, true);
+  FlushJob flush_job(
+      dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
+      *cfd->GetLatestMutableCFOptions(), nullptr /* memtable_id */,
+      env_options_, versions_.get(), &mutex_, &shutting_down_, {},
+      kMaxSequenceNumber, snapshot_checker, &job_context, nullptr, nullptr,
+      nullptr, kNoCompression, db_options_.statistics.get(), &event_logger,
+      true, true /* sync_output_directory */, true /* write_manifest */);
 
   HistogramData hist;
   FileMetaData file_meta;
@@ -165,23 +192,190 @@ TEST_F(FlushJobTest, NonEmpty) {
   job_context.Clean();
 }
 
+TEST_F(FlushJobTest, FlushMemTablesSingleColumnFamily) {
+  const size_t num_mems = 2;
+  const size_t num_mems_to_flush = 1;
+  const size_t num_keys_per_table = 100;
+  JobContext job_context(0);
+  ColumnFamilyData* cfd = versions_->GetColumnFamilySet()->GetDefault();
+  std::vector<uint64_t> memtable_ids;
+  std::vector<MemTable*> new_mems;
+  for (size_t i = 0; i != num_mems; ++i) {
+    MemTable* mem = cfd->ConstructNewMemtable(*cfd->GetLatestMutableCFOptions(),
+                                              kMaxSequenceNumber);
+    mem->SetID(i);
+    mem->Ref();
+    new_mems.emplace_back(mem);
+    memtable_ids.push_back(mem->GetID());
+
+    for (size_t j = 0; j < num_keys_per_table; ++j) {
+      std::string key(ToString(j + i * num_keys_per_table));
+      std::string value("value" + key);
+      mem->Add(SequenceNumber(j + i * num_keys_per_table), kTypeValue, key,
+               value);
+    }
+  }
+
+  autovector<MemTable*> to_delete;
+  for (auto mem : new_mems) {
+    cfd->imm()->Add(mem, &to_delete);
+  }
+
+  EventLogger event_logger(db_options_.info_log.get());
+  SnapshotChecker* snapshot_checker = nullptr;  // not relavant
+
+  assert(memtable_ids.size() == num_mems);
+  uint64_t smallest_memtable_id = memtable_ids.front();
+  uint64_t flush_memtable_id = smallest_memtable_id + num_mems_to_flush - 1;
+
+  FlushJob flush_job(
+      dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
+      *cfd->GetLatestMutableCFOptions(), &flush_memtable_id, env_options_,
+      versions_.get(), &mutex_, &shutting_down_, {}, kMaxSequenceNumber,
+      snapshot_checker, &job_context, nullptr, nullptr, nullptr, kNoCompression,
+      db_options_.statistics.get(), &event_logger, true,
+      true /* sync_output_directory */, true /* write_manifest */);
+  HistogramData hist;
+  FileMetaData file_meta;
+  mutex_.Lock();
+  flush_job.PickMemTable();
+  ASSERT_OK(flush_job.Run(nullptr /* prep_tracker */, &file_meta));
+  mutex_.Unlock();
+  db_options_.statistics->histogramData(FLUSH_TIME, &hist);
+  ASSERT_GT(hist.average, 0.0);
+
+  ASSERT_EQ(ToString(0), file_meta.smallest.user_key().ToString());
+  ASSERT_EQ("99", file_meta.largest.user_key().ToString());
+  ASSERT_EQ(0, file_meta.fd.smallest_seqno);
+  ASSERT_EQ(SequenceNumber(num_mems_to_flush * num_keys_per_table - 1),
+            file_meta.fd.largest_seqno);
+
+  for (auto m : to_delete) {
+    delete m;
+  }
+  to_delete.clear();
+  job_context.Clean();
+}
+
+TEST_F(FlushJobTest, FlushMemtablesMultipleColumnFamilies) {
+  autovector<ColumnFamilyData*> all_cfds;
+  for (auto cfd : *versions_->GetColumnFamilySet()) {
+    all_cfds.push_back(cfd);
+  }
+  const std::vector<size_t> num_memtables = {2, 1, 3};
+  assert(num_memtables.size() == column_family_names_.size());
+  const size_t num_keys_per_memtable = 1000;
+  JobContext job_context(0);
+  std::vector<uint64_t> memtable_ids;
+  std::vector<SequenceNumber> smallest_seqs;
+  std::vector<SequenceNumber> largest_seqs;
+  autovector<MemTable*> to_delete;
+  SequenceNumber curr_seqno = 0;
+  size_t k = 0;
+  for (auto cfd : all_cfds) {
+    smallest_seqs.push_back(curr_seqno);
+    for (size_t i = 0; i != num_memtables[k]; ++i) {
+      MemTable* mem = cfd->ConstructNewMemtable(
+          *cfd->GetLatestMutableCFOptions(), kMaxSequenceNumber);
+      mem->SetID(i);
+      mem->Ref();
+
+      for (size_t j = 0; j != num_keys_per_memtable; ++j) {
+        std::string key(ToString(j + i * num_keys_per_memtable));
+        std::string value("value" + key);
+        mem->Add(curr_seqno++, kTypeValue, key, value);
+      }
+
+      cfd->imm()->Add(mem, &to_delete);
+    }
+    largest_seqs.push_back(curr_seqno - 1);
+    memtable_ids.push_back(num_memtables[k++] - 1);
+  }
+
+  EventLogger event_logger(db_options_.info_log.get());
+  SnapshotChecker* snapshot_checker = nullptr;  // not relevant
+  std::vector<FlushJob> flush_jobs;
+  k = 0;
+  for (auto cfd : all_cfds) {
+    std::vector<SequenceNumber> snapshot_seqs;
+    flush_jobs.emplace_back(
+        dbname_, cfd, db_options_, *cfd->GetLatestMutableCFOptions(),
+        &memtable_ids[k], env_options_, versions_.get(), &mutex_,
+        &shutting_down_, snapshot_seqs, kMaxSequenceNumber, snapshot_checker,
+        &job_context, nullptr, nullptr, nullptr, kNoCompression,
+        db_options_.statistics.get(), &event_logger, true,
+        false /* sync_output_directory */, false /* write_manifest */);
+    k++;
+  }
+  HistogramData hist;
+  autovector<FileMetaData> file_metas;
+  mutex_.Lock();
+  for (auto& job : flush_jobs) {
+    job.PickMemTable();
+  }
+  for (auto& job : flush_jobs) {
+    FileMetaData meta;
+    // Run will release and re-acquire  mutex
+    ASSERT_OK(job.Run(nullptr /**/, &meta));
+    file_metas.emplace_back(meta);
+  }
+  autovector<const autovector<MemTable*>*> mems_list;
+  for (size_t i = 0; i != all_cfds.size(); ++i) {
+    const auto& mems = flush_jobs[i].GetMemTables();
+    mems_list.push_back(&mems);
+  }
+  autovector<const MutableCFOptions*> mutable_cf_options_list;
+  for (auto cfd : all_cfds) {
+    mutable_cf_options_list.push_back(cfd->GetLatestMutableCFOptions());
+  }
+
+  Status s = InstallMemtableAtomicFlushResults(
+      nullptr /* imm_lists */, all_cfds, mutable_cf_options_list, mems_list,
+      versions_.get(), &mutex_, file_metas, &job_context.memtables_to_free,
+      nullptr /* db_directory */, nullptr /* log_buffer */);
+  ASSERT_OK(s);
+
+  mutex_.Unlock();
+  db_options_.statistics->histogramData(FLUSH_TIME, &hist);
+  ASSERT_GT(hist.average, 0.0);
+  k = 0;
+  for (const auto& file_meta : file_metas) {
+    ASSERT_EQ(ToString(0), file_meta.smallest.user_key().ToString());
+    ASSERT_EQ("999", file_meta.largest.user_key()
+                         .ToString());  // max key by bytewise comparator
+    ASSERT_EQ(smallest_seqs[k], file_meta.fd.smallest_seqno);
+    ASSERT_EQ(largest_seqs[k], file_meta.fd.largest_seqno);
+    // Verify that imm is empty
+    ASSERT_EQ(std::numeric_limits<uint64_t>::max(),
+              all_cfds[k]->imm()->GetEarliestMemTableID());
+    ASSERT_EQ(0, all_cfds[k]->imm()->GetLatestMemTableID());
+    ++k;
+  }
+
+  for (auto m : to_delete) {
+    delete m;
+  }
+  to_delete.clear();
+  job_context.Clean();
+}
+
 TEST_F(FlushJobTest, Snapshots) {
   JobContext job_context(0);
   auto cfd = versions_->GetColumnFamilySet()->GetDefault();
   auto new_mem = cfd->ConstructNewMemtable(*cfd->GetLatestMutableCFOptions(),
                                            kMaxSequenceNumber);
 
-  std::vector<SequenceNumber> snapshots;
   std::set<SequenceNumber> snapshots_set;
   int keys = 10000;
   int max_inserts_per_keys = 8;
 
   Random rnd(301);
   for (int i = 0; i < keys / 2; ++i) {
-    snapshots.push_back(rnd.Uniform(keys * (max_inserts_per_keys / 2)) + 1);
-    snapshots_set.insert(snapshots.back());
+    snapshots_set.insert(rnd.Uniform(keys * (max_inserts_per_keys / 2)) + 1);
   }
-  std::sort(snapshots.begin(), snapshots.end());
+  // set has already removed the duplicate snapshots
+  std::vector<SequenceNumber> snapshots(snapshots_set.begin(),
+                                        snapshots_set.end());
 
   new_mem->Ref();
   SequenceNumber current_seqno = 0;
@@ -213,12 +407,13 @@ TEST_F(FlushJobTest, Snapshots) {
 
   EventLogger event_logger(db_options_.info_log.get());
   SnapshotChecker* snapshot_checker = nullptr;  // not relavant
-  FlushJob flush_job(dbname_, versions_->GetColumnFamilySet()->GetDefault(),
-                     db_options_, *cfd->GetLatestMutableCFOptions(),
-                     env_options_, versions_.get(), &mutex_, &shutting_down_,
-                     snapshots, kMaxSequenceNumber, snapshot_checker,
-                     &job_context, nullptr, nullptr, nullptr, kNoCompression,
-                     db_options_.statistics.get(), &event_logger, true);
+  FlushJob flush_job(
+      dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
+      *cfd->GetLatestMutableCFOptions(), nullptr /* memtable_id */,
+      env_options_, versions_.get(), &mutex_, &shutting_down_, snapshots,
+      kMaxSequenceNumber, snapshot_checker, &job_context, nullptr, nullptr,
+      nullptr, kNoCompression, db_options_.statistics.get(), &event_logger,
+      true, true /* sync_output_directory */, true /* write_manifest */);
   mutex_.Lock();
   flush_job.PickMemTable();
   ASSERT_OK(flush_job.Run());

db/forward_iterator.cc

@@ -15,6 +15,8 @@
 #include "db/db_iter.h"
 #include "db/dbformat.h"
 #include "db/job_context.h"
+#include "db/range_del_aggregator.h"
+#include "db/range_tombstone_fragmenter.h"
 #include "rocksdb/env.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/slice_transform.h"
@@ -71,8 +73,8 @@ class ForwardLevelIterator : public InternalIterator {
       delete file_iter_;
     }
 
-    RangeDelAggregator range_del_agg(
-        cfd_->internal_comparator(), {} /* snapshots */);
+    ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
+                                         kMaxSequenceNumber /* upper_bound */);
     file_iter_ = cfd_->table_cache()->NewIterator(
         read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
         *files_[file_index_],
@@ -608,13 +610,14 @@ void ForwardIterator::RebuildIterators(bool refresh_sv) {
     // New
     sv_ = cfd_->GetReferencedSuperVersion(&(db_->mutex_));
   }
-  RangeDelAggregator range_del_agg(
-      cfd_->internal_comparator(), {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
+                                       kMaxSequenceNumber /* upper_bound */);
   mutable_iter_ = sv_->mem->NewIterator(read_options_, &arena_);
   sv_->imm->AddIterators(read_options_, &imm_iters_, &arena_);
   if (!read_options_.ignore_range_deletions) {
-    std::unique_ptr<InternalIterator> range_del_iter(
-        sv_->mem->NewRangeTombstoneIterator(read_options_));
+    std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(
+        sv_->mem->NewRangeTombstoneIterator(
+            read_options_, sv_->current->version_set()->LastSequence()));
     range_del_agg.AddTombstones(std::move(range_del_iter));
     sv_->imm->AddRangeTombstoneIterators(read_options_, &arena_,
                                          &range_del_agg);
@@ -666,11 +669,12 @@ void ForwardIterator::RenewIterators() {
 
   mutable_iter_ = svnew->mem->NewIterator(read_options_, &arena_);
   svnew->imm->AddIterators(read_options_, &imm_iters_, &arena_);
-  RangeDelAggregator range_del_agg(
-      cfd_->internal_comparator(), {} /* snapshots */);
+  ReadRangeDelAggregator range_del_agg(&cfd_->internal_comparator(),
+                                       kMaxSequenceNumber /* upper_bound */);
   if (!read_options_.ignore_range_deletions) {
-    std::unique_ptr<InternalIterator> range_del_iter(
-        svnew->mem->NewRangeTombstoneIterator(read_options_));
+    std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(
+        svnew->mem->NewRangeTombstoneIterator(
+            read_options_, sv_->current->version_set()->LastSequence()));
     range_del_agg.AddTombstones(std::move(range_del_iter));
     svnew->imm->AddRangeTombstoneIterators(read_options_, &arena_,
                                            &range_del_agg);
@@ -916,21 +920,13 @@ bool ForwardIterator::TEST_CheckDeletedIters(int* pdeleted_iters,
 uint32_t ForwardIterator::FindFileInRange(
     const std::vector<FileMetaData*>& files, const Slice& internal_key,
     uint32_t left, uint32_t right) {
-  while (left < right) {
-    uint32_t mid = (left + right) / 2;
-    const FileMetaData* f = files[mid];
-    if (cfd_->internal_comparator().InternalKeyComparator::Compare(
-          f->largest.Encode(), internal_key) < 0) {
-      // Key at "mid.largest" is < "target".  Therefore all
-      // files at or before "mid" are uninteresting.
-      left = mid + 1;
-    } else {
-      // Key at "mid.largest" is >= "target".  Therefore all files
-      // after "mid" are uninteresting.
-      right = mid;
-    }
-  }
-  return right;
+  auto cmp = [&](const FileMetaData* f, const Slice& key) -> bool {
+    return cfd_->internal_comparator().InternalKeyComparator::Compare(
+            f->largest.Encode(), key) < 0;
+  };
+  const auto &b = files.begin();
+  return static_cast<uint32_t>(std::lower_bound(b + left,
+                                 b + right, internal_key, cmp) - b);
 }
 
 void ForwardIterator::DeleteIterator(InternalIterator* iter, bool is_arena) {

db/internal_stats.cc

@@ -45,6 +45,8 @@ const std::map<LevelStatType, LevelStat> InternalStats::compaction_level_stats =
         {LevelStatType::READ_MBPS, LevelStat{"ReadMBps", "Rd(MB/s)"}},
         {LevelStatType::WRITE_MBPS, LevelStat{"WriteMBps", "Wr(MB/s)"}},
         {LevelStatType::COMP_SEC, LevelStat{"CompSec", "Comp(sec)"}},
+        {LevelStatType::COMP_CPU_SEC,
+         LevelStat{"CompMergeCPU", "CompMergeCPU(sec)"}},
         {LevelStatType::COMP_COUNT, LevelStat{"CompCount", "Comp(cnt)"}},
         {LevelStatType::AVG_SEC, LevelStat{"AvgSec", "Avg(sec)"}},
         {LevelStatType::KEY_IN, LevelStat{"KeyIn", "KeyIn"}},
@@ -64,7 +66,7 @@ void PrintLevelStatsHeader(char* buf, size_t len, const std::string& cf_name) {
   };
   int line_size = snprintf(
       buf + written_size, len - written_size,
-      "Level    %s   %s     %s %s  %s %s %s %s %s %s %s %s %s %s %s %s %s\n",
+      "Level    %s   %s     %s %s  %s %s %s %s %s %s %s %s %s %s %s %s %s %s\n",
       // Note that we skip COMPACTED_FILES and merge it with Files column
       hdr(LevelStatType::NUM_FILES), hdr(LevelStatType::SIZE_BYTES),
       hdr(LevelStatType::SCORE), hdr(LevelStatType::READ_GB),
@@ -72,9 +74,9 @@ void PrintLevelStatsHeader(char* buf, size_t len, const std::string& cf_name) {
       hdr(LevelStatType::WRITE_GB), hdr(LevelStatType::W_NEW_GB),
       hdr(LevelStatType::MOVED_GB), hdr(LevelStatType::WRITE_AMP),
       hdr(LevelStatType::READ_MBPS), hdr(LevelStatType::WRITE_MBPS),
-      hdr(LevelStatType::COMP_SEC), hdr(LevelStatType::COMP_COUNT),
-      hdr(LevelStatType::AVG_SEC), hdr(LevelStatType::KEY_IN),
-      hdr(LevelStatType::KEY_DROP));
+      hdr(LevelStatType::COMP_SEC), hdr(LevelStatType::COMP_CPU_SEC),
+      hdr(LevelStatType::COMP_COUNT), hdr(LevelStatType::AVG_SEC),
+      hdr(LevelStatType::KEY_IN), hdr(LevelStatType::KEY_DROP));
 
   written_size += line_size;
   snprintf(buf + written_size, len - written_size, "%s\n",
@@ -106,6 +108,7 @@ void PrepareLevelStats(std::map<LevelStatType, double>* level_stats,
   (*level_stats)[LevelStatType::WRITE_MBPS] =
       stats.bytes_written / kMB / elapsed;
   (*level_stats)[LevelStatType::COMP_SEC] = stats.micros / kMicrosInSec;
+  (*level_stats)[LevelStatType::COMP_CPU_SEC] = stats.cpu_micros / kMicrosInSec;
   (*level_stats)[LevelStatType::COMP_COUNT] = stats.count;
   (*level_stats)[LevelStatType::AVG_SEC] =
       stats.count == 0 ? 0 : stats.micros / kMicrosInSec / stats.count;
@@ -132,7 +135,8 @@ void PrintLevelStats(char* buf, size_t len, const std::string& name,
       "%5.1f "    /*  W-Amp */
       "%8.1f "    /*  Rd(MB/s) */
       "%8.1f "    /*  Wr(MB/s) */
-      "%9.0f "    /*  Comp(sec) */
+      "%9.2f "    /*  Comp(sec) */
+      "%17.2f "   /*  CompMergeCPU(sec) */
       "%9d "      /*  Comp(cnt) */
       "%8.3f "    /*  Avg(sec) */
       "%7s "      /*  KeyIn */
@@ -153,6 +157,7 @@ void PrintLevelStats(char* buf, size_t len, const std::string& name,
       stat_value.at(LevelStatType::READ_MBPS),
       stat_value.at(LevelStatType::WRITE_MBPS),
       stat_value.at(LevelStatType::COMP_SEC),
+      stat_value.at(LevelStatType::COMP_CPU_SEC),
       static_cast<int>(stat_value.at(LevelStatType::COMP_COUNT)),
       stat_value.at(LevelStatType::AVG_SEC),
       NumberToHumanString(
@@ -231,6 +236,8 @@ static const std::string current_version_number =
     "current-super-version-number";
 static const std::string estimate_live_data_size = "estimate-live-data-size";
 static const std::string min_log_number_to_keep_str = "min-log-number-to-keep";
+static const std::string min_obsolete_sst_number_to_keep_str =
+    "min-obsolete-sst-number-to-keep";
 static const std::string base_level_str = "base-level";
 static const std::string total_sst_files_size = "total-sst-files-size";
 static const std::string live_sst_files_size = "live-sst-files-size";
@@ -310,6 +317,8 @@ const std::string DB::Properties::kEstimateLiveDataSize =
     rocksdb_prefix + estimate_live_data_size;
 const std::string DB::Properties::kMinLogNumberToKeep =
     rocksdb_prefix + min_log_number_to_keep_str;
+const std::string DB::Properties::kMinObsoleteSstNumberToKeep =
+    rocksdb_prefix + min_obsolete_sst_number_to_keep_str;
 const std::string DB::Properties::kTotalSstFilesSize =
     rocksdb_prefix + total_sst_files_size;
 const std::string DB::Properties::kLiveSstFilesSize =
@@ -430,6 +439,9 @@ const std::unordered_map<std::string, DBPropertyInfo>
         {DB::Properties::kMinLogNumberToKeep,
          {false, nullptr, &InternalStats::HandleMinLogNumberToKeep, nullptr,
           nullptr}},
+        {DB::Properties::kMinObsoleteSstNumberToKeep,
+         {false, nullptr, &InternalStats::HandleMinObsoleteSstNumberToKeep,
+          nullptr, nullptr}},
         {DB::Properties::kBaseLevel,
          {false, nullptr, &InternalStats::HandleBaseLevel, nullptr, nullptr}},
         {DB::Properties::kTotalSstFilesSize,
@@ -826,6 +838,13 @@ bool InternalStats::HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db,
   return true;
 }
 
+bool InternalStats::HandleMinObsoleteSstNumberToKeep(uint64_t* value,
+                                                     DBImpl* db,
+                                                     Version* /*version*/) {
+  *value = db->MinObsoleteSstNumberToKeep();
+  return true;
+}
+
 bool InternalStats::HandleActualDelayedWriteRate(uint64_t* value, DBImpl* db,
                                                  Version* /*version*/) {
   const WriteController& wc = db->write_controller();

db/internal_stats.h

@@ -71,6 +71,7 @@ enum class LevelStatType {
   READ_MBPS,
   WRITE_MBPS,
   COMP_SEC,
+  COMP_CPU_SEC,
   COMP_COUNT,
   AVG_SEC,
   KEY_IN,
@@ -135,6 +136,7 @@ class InternalStats {
   // compactions that produced data for the specified "level".
   struct CompactionStats {
     uint64_t micros;
+    uint64_t cpu_micros;
 
     // The number of bytes read from all non-output levels
     uint64_t bytes_read_non_output_levels;
@@ -172,6 +174,7 @@ class InternalStats {
 
     explicit CompactionStats()
         : micros(0),
+          cpu_micros(0),
           bytes_read_non_output_levels(0),
           bytes_read_output_level(0),
           bytes_written(0),
@@ -190,6 +193,7 @@ class InternalStats {
 
     explicit CompactionStats(CompactionReason reason, int c)
         : micros(0),
+          cpu_micros(0),
           bytes_read_non_output_levels(0),
           bytes_read_output_level(0),
           bytes_written(0),
@@ -214,14 +218,14 @@ class InternalStats {
 
     explicit CompactionStats(const CompactionStats& c)
         : micros(c.micros),
+          cpu_micros(c.cpu_micros),
           bytes_read_non_output_levels(c.bytes_read_non_output_levels),
           bytes_read_output_level(c.bytes_read_output_level),
           bytes_written(c.bytes_written),
           bytes_moved(c.bytes_moved),
           num_input_files_in_non_output_levels(
               c.num_input_files_in_non_output_levels),
-          num_input_files_in_output_level(
-              c.num_input_files_in_output_level),
+          num_input_files_in_output_level(c.num_input_files_in_output_level),
           num_output_files(c.num_output_files),
           num_input_records(c.num_input_records),
           num_dropped_records(c.num_dropped_records),
@@ -234,6 +238,7 @@ class InternalStats {
 
     void Clear() {
       this->micros = 0;
+      this->cpu_micros = 0;
       this->bytes_read_non_output_levels = 0;
       this->bytes_read_output_level = 0;
       this->bytes_written = 0;
@@ -252,6 +257,7 @@ class InternalStats {
 
     void Add(const CompactionStats& c) {
       this->micros += c.micros;
+      this->cpu_micros += c.cpu_micros;
       this->bytes_read_non_output_levels += c.bytes_read_non_output_levels;
       this->bytes_read_output_level += c.bytes_read_output_level;
       this->bytes_written += c.bytes_written;
@@ -272,6 +278,7 @@ class InternalStats {
 
     void Subtract(const CompactionStats& c) {
       this->micros -= c.micros;
+      this->cpu_micros -= c.cpu_micros;
       this->bytes_read_non_output_levels -= c.bytes_read_non_output_levels;
       this->bytes_read_output_level -= c.bytes_read_output_level;
       this->bytes_written -= c.bytes_written;
@@ -533,6 +540,8 @@ class InternalStats {
   bool HandleEstimateLiveDataSize(uint64_t* value, DBImpl* db,
                                   Version* version);
   bool HandleMinLogNumberToKeep(uint64_t* value, DBImpl* db, Version* version);
+  bool HandleMinObsoleteSstNumberToKeep(uint64_t* value, DBImpl* db,
+                                        Version* version);
   bool HandleActualDelayedWriteRate(uint64_t* value, DBImpl* db,
                                     Version* version);
   bool HandleIsWriteStopped(uint64_t* value, DBImpl* db, Version* version);
@@ -593,6 +602,7 @@ class InternalStats {
 
   struct CompactionStats {
     uint64_t micros;
+    uint64_t cpu_micros;
     uint64_t bytes_read_non_output_levels;
     uint64_t bytes_read_output_level;
     uint64_t bytes_written;