URI-based Env selection for db_bench

Summary:
Added an option, --env_uri. When provided, it is used as an argument to
NewEnvFromUri(), which instantiates an Env based on it.

Test Plan:
built a simple binary that registers ChrootEnv for prefix "/", then
ran:

  $ ./tmp --env_uri /tmp/ --db /abcde

/tmp/ is the chroot directory and /abcde is the db_name. Then I verified
db_bench uses /tmp/abcde

Reviewers: sdong, kradhakrishnan, lightmark

Reviewed By: lightmark

Subscribers: andrewkr, dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D59325

.travis.yml

@@ -1,24 +1,29 @@
 sudo: false
 language: cpp
+os:
+  - linux
+  - osx
+compiler:
+  - clang
 
-matrix:
-  include:
-    - os: linux
-      compiler: clang
-      env: COMPILER=clang++-3.6
-      addons:
-         apt:
-            sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
-            packages: ['clang-3.6', 'clang-format-3.6', 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl']
-    - os: osx
-      compiler: clang
+addons:
+   apt:
+      sources: ['ubuntu-toolchain-r-test', 'llvm-toolchain-precise-3.6']
+      packages: ['clang-3.6' , 'zlib1g-dev', 'libbz2-dev', 'libsnappy-dev', 'curl']
+env:
+  # Run all tests before db_block_cache_test (db_test, db_test2)
+  - JOB_NAME=unittests ROCKSDBTESTS_END=db_block_cache_test
+  # Run all tests starting from db_block_cache_test (db_block_cache_test, db_iter_test, ...)
+  - JOB_NAME=unittests ROCKSDBTESTS_START=db_block_cache_test
+  # Run java tests
+  - JOB_NAME=java_test
+  # Build ROCKSDB_LITE
+  - JOB_NAME=lite_build
 
 install:
   # Build gflags
   # TODO(noetzli): Remove when gflags available through Travis
   - pushd /tmp/ && curl -L https://github.com/gflags/gflags/archive/v2.1.2.tar.gz -o gflags.tar.gz && tar xfz gflags.tar.gz && cd gflags-2.1.2 && cmake . && make && popd
-  # Download clang-format-diff.py to check source code formatting
-  - pushd /tmp/ && curl -L http://llvm.org/svn/llvm-project/cfe/trunk/tools/clang-format/clang-format-diff.py -o clang-format-diff.py && chmod +x clang-format-diff.py && popd
 
 before_script:
   # Add gflags to include/library paths
@@ -26,16 +31,15 @@ before_script:
   - export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/tmp/gflags-2.1.2/lib"
   - export LIBRARY_PATH="$LIBRARY_PATH:/tmp/gflags-2.1.2/lib"
   - export CPLUS_INCLUDE_PATH="$CPLUS_INCLUDE_PATH:/tmp/gflags-2.1.2/include"
-  - if [ -n "${COMPILER}" ]; then CXX=${COMPILER}; fi
+  - if [[ "${TRAVIS_OS_NAME}" == 'linux' ]]; then CXX=clang++-3.6; fi
   - if [[ "${TRAVIS_OS_NAME}" == 'osx' ]]; then brew install gflags snappy; fi
+  # Limit the maximum number of open file descriptors to 2000
   - ulimit -n 2000 || true
 
-# Lousy hack to disable use and testing of fallocate, which doesn't behave quite
-# as EnvPosixTest::AllocateTest expects within the Travis OpenVZ environment.
 script:
-  - if [[ "${TRAVIS_OS_NAME}" == 'linux' ]]; then OPT=-DTRAVIS CLANG_FORMAT_DIFF=/tmp/clang-format-diff.py make format || true; fi
-  - OPT=-DTRAVIS V=1 make -j4 check && OPT=-DTRAVIS V=1 make clean jclean rocksdbjava jtest && make clean && OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib
-
+  - if [[ "${JOB_NAME}" == 'unittests' ]]; then OPT=-DTRAVIS V=1 make -j4 check_some; fi
+  - if [[ "${JOB_NAME}" == 'java_test' ]]; then OPT=-DTRAVIS V=1 make clean jclean rocksdbjava jtest; fi
+  - if [[ "${JOB_NAME}" == 'lite_build' ]]; then OPT="-DTRAVIS -DROCKSDB_LITE" V=1 make -j4 static_lib; fi
 notifications:
     email:
       - leveldb@fb.com

CMakeLists.txt

@@ -82,7 +82,7 @@ set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Oxt /Zp8 /Gm- /Gy /MD")
 set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /DEBUG")
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /DEBUG")
 
-add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64)
+add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64 -DNOMINMAX)
 
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_SOURCE_DIR}/include)
@@ -90,7 +90,7 @@ include_directories(${PROJECT_SOURCE_DIR}/third-party/gtest-1.7.0/fused-src)
 
 set(ROCKSDB_LIBS rocksdblib${ARTIFACT_SUFFIX})
 set(THIRDPARTY_LIBS ${THIRDPARTY_LIBS} gtest)
-set(SYSTEM_LIBS Shlwapi.lib Rpcrt4.lib)
+set(SYSTEM_LIBS ${SYSTEM_LIBS} Shlwapi.lib Rpcrt4.lib)
 
 set(LIBS ${ROCKSDB_LIBS} ${THIRDPARTY_LIBS} ${SYSTEM_LIBS})
 
@@ -153,9 +153,12 @@ set(SOURCES
         memtable/skiplistrep.cc
         memtable/vectorrep.cc
         port/stack_trace.cc
+        port/win/io_win.cc
         port/win/env_win.cc
+        port/win/env_default.cc
         port/win/port_win.cc
         port/win/win_logger.cc
+        port/win/xpress_win.cc
         table/adaptive_table_factory.cc
         table/block.cc
         table/block_based_filter_block.cc
@@ -163,7 +166,6 @@ set(SOURCES
         table/block_based_table_factory.cc
         table/block_based_table_reader.cc
         table/block_builder.cc
-        table/block_hash_index.cc
         table/block_prefix_index.cc
         table/bloom_block.cc
         table/cuckoo_table_builder.cc
@@ -182,6 +184,7 @@ set(SOURCES
         table/plain_table_index.cc
         table/plain_table_key_coding.cc
         table/plain_table_reader.cc
+        table/persistent_cache_helper.cc
         table/table_properties.cc
         table/two_level_iterator.cc
         tools/sst_dump_tool.cc
@@ -199,6 +202,7 @@ set(SOURCES
         util/delete_scheduler.cc
         util/dynamic_bloom.cc
         util/env.cc
+        util/env_chroot.cc
         util/env_hdfs.cc
         util/event_logger.cc
         util/file_util.cc
@@ -207,6 +211,7 @@ set(SOURCES
         util/filter_policy.cc
         util/hash.cc
         util/histogram.cc
+        util/histogram_windowing.cc
         util/instrumented_mutex.cc
         util/iostats_context.cc
         tools/ldb_cmd.cc
@@ -217,7 +222,6 @@ set(SOURCES
         util/murmurhash.cc
         util/mutable_cf_options.cc
         util/options.cc
-        util/options_builder.cc
         util/options_helper.cc
         util/options_parser.cc
         util/options_sanity_check.cc
@@ -234,10 +238,12 @@ set(SOURCES
         util/testharness.cc
         util/testutil.cc
         util/thread_local.cc
+        util/threadpool.cc
         util/thread_status_impl.cc
         util/thread_status_updater.cc
         util/thread_status_util.cc
         util/thread_status_util_debug.cc
+        util/transaction_test_util.cc
         util/xfunc.cc
         util/xxhash.cc
         utilities/backupable/backupable_db.cc
@@ -247,6 +253,7 @@ set(SOURCES
         utilities/document/json_document.cc
         utilities/document/json_document_builder.cc
         utilities/env_mirror.cc
+        utilities/env_registry.cc
         utilities/flashcache/flashcache.cc
         utilities/geodb/geodb_impl.cc
         utilities/leveldb_options/leveldb_options.cc
@@ -254,8 +261,11 @@ set(SOURCES
         utilities/merge_operators/string_append/stringappend.cc
         utilities/merge_operators/string_append/stringappend2.cc
         utilities/merge_operators/put.cc
+        utilities/merge_operators/max.cc
         utilities/merge_operators/uint64add.cc
         utilities/options/options_util.cc
+        utilities/persistent_cache/persistent_cache_tier.cc
+        utilities/persistent_cache/volatile_tier_impl.cc
         utilities/redis/redis_lists.cc
         utilities/spatialdb/spatial_db.cc
         utilities/table_properties_collectors/compact_on_deletion_collector.cc
@@ -278,6 +288,7 @@ set(TESTUTIL_SOURCE
     db/db_test_util.cc
     table/mock_table.cc
     util/mock_env.cc
+    util/fault_injection_test_env.cc
     util/thread_status_updater_debug.cc
 )
 
@@ -308,10 +319,12 @@ set(APPS
         tools/db_stress.cc
         tools/write_stress.cc
         tools/db_repl_stress.cc
+        tools/ldb.cc
         tools/sst_dump.cc
         tools/dump/rocksdb_dump.cc
         tools/dump/rocksdb_undump.cc
         util/cache_bench.cc
+        utilities/persistent_cache/hash_table_bench.cc
 )
 
 set(C_TESTS db/c_test.c)
@@ -338,6 +351,10 @@ set(TESTS
         db/db_tailing_iter_test.cc
         db/db_test.cc
         db/db_test2.cc
+        db/db_block_cache_test.cc
+        db/db_bloom_filter_test.cc
+        db/db_iterator_test.cc
+        db/db_sst_test.cc
         db/db_universal_compaction_test.cc
         db/db_wal_test.cc
         db/dbformat_test.cc
@@ -357,6 +374,7 @@ set(TESTS
         db/perf_context_test.cc
         db/plain_table_db_test.cc
         db/prefix_test.cc
+        db/repair_test.cc
         db/skiplist_test.cc
         db/table_properties_collector_test.cc
         db/version_builder_test.cc
@@ -366,8 +384,8 @@ set(TESTS
         db/write_batch_test.cc
         db/write_callback_test.cc
         db/write_controller_test.cc
+        db/db_io_failure_test.cc
         table/block_based_filter_block_test.cc
-        table/block_hash_index_test.cc
         table/block_test.cc
         table/cuckoo_table_builder_test.cc
         table/cuckoo_table_reader_test.cc
@@ -386,6 +404,7 @@ set(TESTS
         util/crc32c_test.cc
         util/delete_scheduler_test.cc
         util/dynamic_bloom_test.cc
+        util/env_basic_test.cc
         util/env_test.cc
         util/event_logger_test.cc
         util/filelock_test.cc
@@ -393,8 +412,8 @@ set(TESTS
         util/heap_test.cc
         util/histogram_test.cc
         util/iostats_context_test.cc
-        util/memenv_test.cc
         util/mock_env_test.cc
+        util/options_settable_test.cc
         util/options_test.cc
         util/rate_limiter_test.cc
         util/slice_transform_test.cc
@@ -404,10 +423,13 @@ set(TESTS
         utilities/checkpoint/checkpoint_test.cc
         utilities/document/document_db_test.cc
         utilities/document/json_document_test.cc
+        utilities/env_registry_test.cc
         utilities/geodb/geodb_test.cc
         utilities/memory/memory_test.cc
         utilities/merge_operators/string_append/stringappend_test.cc
         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/table_properties_collectors/compact_on_deletion_collector_test.cc

DEFAULT_OPTIONS_HISTORY.md

@@ -0,0 +1,14 @@
+# RocksDB default options change log
+## 4.8.0 (5/2/2016)
+* options.max_open_files changes from 5000 to -1. It improves performance, but users need to set file descriptor limit to be large enough and watch memory usage for index and bloom filters.
+* options.base_background_compactions changes from max_background_compactions to 1. When users set higher max_background_compactions but the write throughput is not high, the writes are less spiky to disks.
+* options.wal_recovery_mode changes from kTolerateCorruptedTailRecords to kPointInTimeRecovery. Avoid some false positive when file system or hardware reorder the writes for file data and metadata.
+
+## 4.7.0 (4/8/2016)
+* options.write_buffer_size changes from 4MB to 64MB.
+* options.target_file_size_base changes from 2MB to 64MB.
+* options.max_bytes_for_level_base changes from 10MB to 256MB.
+* options.soft_pending_compaction_bytes_limit changes from 0 (disabled) to 64GB.
+* options.hard_pending_compaction_bytes_limit changes from 0 (disabled) to 256GB.
+* table_cache_numshardbits changes from 4 to 6.
+* max_file_opening_threads changes from 1 to 16.

HISTORY.md

@@ -1,7 +1,38 @@
 # Rocksdb Change Log
 ## Unreleased
+
+## 4.9.0 (6/9/2016)
+### Public API changes
+* Add bottommost_compression option, This option can be used to set a specific compression algorithm for the bottommost level (Last level containing files in the DB).
+* Introduce CompactionJobInfo::compression, This field state the compression algorithm used to generate the output files of the compaction.
+* Deprecate BlockBaseTableOptions.hash_index_allow_collision=false
+* Deprecate options builder (GetOptions()).
+
+### New Features
+* Introduce NewSimCache() in rocksdb/utilities/sim_cache.h. This function creates a block cache that is able to give simulation results (mainly hit rate) of simulating block behavior with a configurable cache size.
+
+## 4.8.0 (5/2/2016)
+### Public API Change
+* Allow preset compression dictionary for improved compression of block-based tables. This is supported for zlib, zstd, and lz4. The compression dictionary's size is configurable via CompressionOptions::max_dict_bytes.
+* Delete deprecated classes for creating backups (BackupableDB) and restoring from backups (RestoreBackupableDB). Now, BackupEngine should be used for creating backups, and BackupEngineReadOnly should be used for restorations. For more details, see https://github.com/facebook/rocksdb/wiki/How-to-backup-RocksDB%3F
+* Expose estimate of per-level compression ratio via DB property: "rocksdb.compression-ratio-at-levelN".
+* Added EventListener::OnTableFileCreationStarted. EventListener::OnTableFileCreated will be called on failure case. User can check creation status via TableFileCreationInfo::status.
+
+### New Features
+* Add ReadOptions::readahead_size. If non-zero, NewIterator will create a new table reader which performs reads of the given size.
+
+## 4.7.0 (4/8/2016)
+### Public API Change
+* rename options compaction_measure_io_stats to report_bg_io_stats and include flush too.
+* Change some default options. Now default options will optimize for server-workloads. Also enable slowdown and full stop triggers for pending compaction bytes. These changes may cause sub-optimal performance or significant increase of resource usage. To avoid these risks, users can open existing RocksDB with options extracted from RocksDB option files. See https://github.com/facebook/rocksdb/wiki/RocksDB-Options-File for how to use RocksDB option files. Or you can call Options.OldDefaults() to recover old defaults. DEFAULT_OPTIONS_HISTORY.md will track change history of default options.
+
+## 4.6.0 (3/10/2016)
 ### Public API Changes
 * Change default of BlockBasedTableOptions.format_version to 2. It means default DB created by 4.6 or up cannot be opened by RocksDB version 3.9 or earlier.
+* Added strict_capacity_limit option to NewLRUCache. If the flag is set to true, insert to cache will fail if no enough capacity can be free. Signiture of Cache::Insert() is updated accordingly.
+* Tickers [NUMBER_DB_NEXT, NUMBER_DB_PREV, NUMBER_DB_NEXT_FOUND, NUMBER_DB_PREV_FOUND, ITER_BYTES_READ] are not updated immediately. The are updated when the Iterator is deleted.
+* Add monotonically increasing counter (DB property "rocksdb.current-super-version-number") that increments upon any change to the LSM tree.
+
 ### New Features
 * Add CompactionPri::kMinOverlappingRatio, a compaction picking mode friendly to write amplification.
 * Deprecate Iterator::IsKeyPinned() and replace it with Iterator::GetProperty() with prop_name="rocksdb.iterator.is.key.pinned"
@@ -15,7 +46,7 @@
 ### New Features
 * ldb tool now supports operations to non-default column families.
 * Add kPersistedTier to ReadTier.  This option allows Get and MultiGet to read only the persited data and skip mem-tables if writes were done with disableWAL = true.
-* Add DBOptions::sst_file_manager. Use NewSstFileManager() in include/rocksdb/sst_file_manager.h to create a SstFileManager that can be used to track the total size of SST files and control the SST files deletion rate.  
+* Add DBOptions::sst_file_manager. Use NewSstFileManager() in include/rocksdb/sst_file_manager.h to create a SstFileManager that can be used to track the total size of SST files and control the SST files deletion rate.
 
 ## 4.4.0 (1/14/2016)
 ### Public API Changes

LANGUAGE-BINDINGS.md

@@ -0,0 +1,11 @@
+This is the list of all known third-party language bindings for RocksDB. If something is missing, please open a pull request to add it.
+
+* Java - https://github.com/facebook/rocksdb/tree/master/java
+* Python - http://pyrocksdb.readthedocs.org/en/latest/
+* Perl - https://metacpan.org/pod/RocksDB
+* Node.js - https://npmjs.org/package/rocksdb
+* Go - https://github.com/tecbot/gorocksdb
+* Ruby - http://rubygems.org/gems/rocksdb-ruby
+* Haskell - https://hackage.haskell.org/package/rocksdb-haskell
+* PHP - https://github.com/Photonios/rocksdb-php
+* C# - https://github.com/warrenfalk/rocksdb-sharp

Makefile

@@ -15,7 +15,9 @@ ARFLAGS = rs
 
 # Transform parallel LOG output into something more readable.
 perl_command = perl -n \
-  -e '@a=split("\t",$$_,-1); $$t=$$a[8]; $$t =~ s,^\./,,;'		\
+  -e '@a=split("\t",$$_,-1); $$t=$$a[8];'				\
+  -e '$$t =~ /.*if\s\[\[\s"(.*?\.[\w\/]+)/ and $$t=$$1;'		\
+  -e '$$t =~ s,^\./,,;'							\
   -e '$$t =~ s, >.*,,; chomp $$t;'					\
   -e '$$t =~ /.*--gtest_filter=(.*?\.[\w\/]+)/ and $$t=$$1;'		\
   -e 'printf "%7.3f %s %s\n", $$a[3], $$a[6] == 0 ? "PASS" : "FAIL", $$t'
@@ -169,12 +171,22 @@ ifdef COMPILE_WITH_TSAN
 	PLATFORM_CXXFLAGS += -fsanitize=thread -fPIC -DROCKSDB_TSAN_RUN
         # Turn off -pg when enabling TSAN testing, because that induces
         # a link failure.  TODO: find the root cause
-	pg =
-else
-	pg = -pg
+	PROFILING_FLAGS =
+endif
+
+# USAN doesn't work well with jemalloc. If we're compiling with USAN, we should use regular malloc.
+ifdef COMPILE_WITH_UBSAN
+	DISABLE_JEMALLOC=1
+	EXEC_LDFLAGS += -fsanitize=undefined
+	PLATFORM_CCFLAGS += -fsanitize=undefined
+	PLATFORM_CXXFLAGS += -fsanitize=undefined
 endif
 
 ifndef DISABLE_JEMALLOC
+	ifdef JEMALLOC
+		PLATFORM_CXXFLAGS += "-DROCKSDB_JEMALLOC"
+		PLATFORM_CCFLAGS +=  "-DROCKSDB_JEMALLOC"
+	endif
 	EXEC_LDFLAGS := $(JEMALLOC_LIB) $(EXEC_LDFLAGS)
 	PLATFORM_CXXFLAGS += $(JEMALLOC_INCLUDE)
 	PLATFORM_CCFLAGS += $(JEMALLOC_INCLUDE)
@@ -246,18 +258,22 @@ BENCHTOOLOBJECTS = $(BENCH_SOURCES:.cc=.o) $(LIBOBJECTS) $(TESTUTIL)
 TESTS = \
 	db_test \
 	db_test2 \
+	db_block_cache_test \
+	db_bloom_filter_test \
 	db_iter_test \
 	db_log_iter_test \
 	db_compaction_filter_test \
 	db_compaction_test \
 	db_dynamic_level_test \
 	db_inplace_update_test \
+	db_iterator_test \
+	db_sst_test \
 	db_tailing_iter_test \
 	db_universal_compaction_test \
 	db_wal_test \
+	db_io_failure_test \
 	db_properties_test \
 	db_table_properties_test \
-	block_hash_index_test \
 	autovector_test \
 	column_family_test \
 	table_properties_collector_test \
@@ -274,6 +290,7 @@ TESTS = \
 	crc32c_test \
 	slice_transform_test \
 	dbformat_test \
+	env_basic_test \
 	env_test \
 	fault_injection_test \
 	filelock_test \
@@ -281,17 +298,18 @@ TESTS = \
 	file_reader_writer_test \
 	block_based_filter_block_test \
 	full_filter_block_test \
+	hash_table_test \
 	histogram_test \
 	inlineskiplist_test \
 	log_test \
 	manual_compaction_test \
-	memenv_test \
 	mock_env_test \
 	memtable_list_test \
 	merge_helper_test \
 	memory_test \
 	merge_test \
 	merger_test \
+	util_merge_operators_test \
 	options_file_test \
 	redis_test \
 	reduce_levels_test \
@@ -304,6 +322,7 @@ TESTS = \
 	backupable_db_test \
 	document_db_test \
 	json_document_test \
+	sim_cache_test \
 	spatial_db_test \
 	version_edit_test \
 	version_set_test \
@@ -320,6 +339,7 @@ TESTS = \
 	rate_limiter_test \
 	delete_scheduler_test \
 	options_test \
+	options_settable_test \
 	options_util_test \
 	event_logger_test \
 	cuckoo_table_builder_test \
@@ -341,9 +361,29 @@ TESTS = \
 	compaction_job_stats_test \
 	transaction_test \
 	ldb_cmd_test \
-	iostats_context_test
+	iostats_context_test \
+	persistent_cache_test \
 
-SUBSET :=  $(shell echo $(TESTS) |sed s/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/)
+PARALLEL_TEST = \
+	backupable_db_test \
+	compact_on_deletion_collector_test \
+	db_compaction_filter_test \
+	db_compaction_test \
+	db_test \
+	db_universal_compaction_test \
+	fault_injection_test \
+	inlineskiplist_test \
+	manual_compaction_test \
+	table_test
+
+SUBSET := $(TESTS)
+ifdef ROCKSDBTESTS_START
+        SUBSET := $(shell echo $(SUBSET) | sed 's/^.*$(ROCKSDBTESTS_START)/$(ROCKSDBTESTS_START)/')
+endif
+
+ifdef ROCKSDBTESTS_END
+        SUBSET := $(shell echo $(SUBSET) | sed 's/$(ROCKSDBTESTS_END).*//')
+endif
 
 TOOLS = \
 	sst_dump \
@@ -355,6 +395,9 @@ TOOLS = \
 	rocksdb_dump \
 	rocksdb_undump
 
+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
 
@@ -368,6 +411,7 @@ else
 endif
 endif
 LIBRARY = ${LIBNAME}.a
+TOOLS_LIBRARY = ${LIBNAME}_tools.a
 
 ROCKSDB_MAJOR = $(shell egrep "ROCKSDB_MAJOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
 ROCKSDB_MINOR = $(shell egrep "ROCKSDB_MINOR.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
@@ -419,10 +463,10 @@ endif  # PLATFORM_SHARED_EXT
 .PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests package \
 	release tags valgrind_check whitebox_crash_test format static_lib shared_lib all \
 	dbg rocksdbjavastatic rocksdbjava install install-static install-shared uninstall \
-	analyze tools
+	analyze tools tools_lib
 
 
-all: $(LIBRARY) $(BENCHMARKS) tools $(TESTS)
+all: $(LIBRARY) $(BENCHMARKS) tools tools_lib test_libs $(TESTS)
 
 static_lib: $(LIBRARY)
 
@@ -430,6 +474,10 @@ shared_lib: $(SHARED)
 
 tools: $(TOOLS)
 
+tools_lib: $(TOOLS_LIBRARY)
+
+test_libs: $(TEST_LIBS)
+
 dbg: $(LIBRARY) $(BENCHMARKS) tools $(TESTS)
 
 # creates static library and programs
@@ -444,7 +492,26 @@ coverage:
         # Delete intermediate files
 	find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
 
-# Extract the names of its tests by running db_test with --gtest_list_tests.
+ifneq (,$(filter check parallel_check,$(MAKECMDGOALS)),)
+# Use /dev/shm if it has the sticky bit set (otherwise, /tmp),
+# and create a randomly-named rocksdb.XXXX directory therein.
+# We'll use that directory in the "make check" rules.
+ifeq ($(TMPD),)
+TMPD := $(shell f=/dev/shm; test -k $$f || f=/tmp;			\
+  perl -le 'use File::Temp "tempdir";'					\
+    -e 'print tempdir("'$$f'/rocksdb.XXXX", CLEANUP => 0)')
+endif
+endif
+
+# Run all tests in parallel, accumulating per-test logs in t/log-*.
+#
+# Each t/run-* file is a tiny generated bourne shell script that invokes one of
+# sub-tests. Why use a file for this?  Because that makes the invocation of
+# parallel below simpler, which in turn makes the parsing of parallel's
+# LOG simpler (the latter is for live monitoring as parallel
+# tests run).
+#
+# Test names are extracted by running tests with --gtest_list_tests.
 # This filter removes the "#"-introduced comments, and expands to
 # fully-qualified names by changing input like this:
 #
@@ -462,52 +529,33 @@ coverage:
 #   MultiThreaded/MultiThreadedDBTest.MultiThreaded/0
 #   MultiThreaded/MultiThreadedDBTest.MultiThreaded/1
 #
-test_names = \
-  ./db_test --gtest_list_tests						\
-    | perl -n								\
-      -e 's/ *\#.*//;'							\
-      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'			\
-      -e 'print qq! $$p$$2!'
 
-ifneq (,$(filter check parallel_check,$(MAKECMDGOALS)),)
-# Use /dev/shm if it has the sticky bit set (otherwise, /tmp),
-# and create a randomly-named rocksdb.XXXX directory therein.
-# We'll use that directory in the "make check" rules.
-ifeq ($(TMPD),)
-TMPD := $(shell f=/dev/shm; test -k $$f || f=/tmp;			\
-  perl -le 'use File::Temp "tempdir";'					\
-    -e 'print tempdir("'$$f'/rocksdb.XXXX", CLEANUP => 0)')
-endif
-endif
+parallel_tests = $(patsubst %,parallel_%,$(PARALLEL_TEST))
+.PHONY: gen_parallel_tests $(parallel_tests)
+$(parallel_tests): $(PARALLEL_TEST)
+	$(AM_V_at)TEST_BINARY=$(patsubst parallel_%,%,$@); \
+  TEST_NAMES=` \
+    ./$$TEST_BINARY --gtest_list_tests \
+    | perl -n \
+      -e 's/ *\#.*//;' \
+      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'	\
+      -e 'print qq! $$p$$2!'`; \
+	for TEST_NAME in $$TEST_NAMES; do \
+		TEST_SCRIPT=t/run-$$TEST_BINARY-$${TEST_NAME//\//-}; \
+		echo "  GEN     " $$TEST_SCRIPT; \
+    printf '%s\n' \
+      '#!/bin/sh' \
+      "d=\$(TMPD)$$TEST_SCRIPT" \
+      'mkdir -p $$d' \
+			"TEST_TMPDIR=\$$d ./$$TEST_BINARY --gtest_filter=$$TEST_NAME" \
+		> $$TEST_SCRIPT; \
+		chmod a=rx $$TEST_SCRIPT; \
+	done
 
-ifneq ($(T),)
-
-# Run all tests in parallel, accumulating per-test logs in t/log-*.
-
-# t_sanitized is each $(T) with "-" in place of each "/".
-t_sanitized = $(subst /,-,$(T))
-
-# t_run is each sanitized name with a leading "t/".
-t_run = $(patsubst %,t/%,$(t_sanitized))
-
-# Each t_run file is a tiny generated bourne shell script
-# that invokes one of db_tests's sub-tests. Why use a file
-# for this?  Because that makes the invocation of parallel
-# below simpler, which in turn makes the parsing of parallel's
-# LOG simpler (the latter is for live monitoring as parallel
-# tests run).
-filter = --gtest_filter=$(subst -,/,$(@F))
-$(t_run): Makefile db_test
-	$(AM_V_GEN)mkdir -p t
-	$(AM_V_at)rm -f $@ $@-t
-	$(AM_V_at)printf '%s\n'						\
-	    '#!/bin/sh'							\
-	    'd=$(TMPD)/$(@F)'						\
-	    'mkdir -p $$d'						\
-	    'TEST_TMPDIR=$$d ./db_test $(filter)'			\
-	  > $@-t
-	$(AM_V_at)chmod a=rx $@-t
-	$(AM_V_at)mv $@-t $@
+gen_parallel_tests:
+	$(AM_V_at)mkdir -p t
+	$(AM_V_at)rm -f t/run-*
+	$(MAKE) $(parallel_tests)
 
 # Reorder input lines (which are one per test) so that the
 # longest-running tests appear first in the output.
@@ -526,7 +574,7 @@ $(t_run): Makefile db_test
 # 107.816 PASS t/DBTest.EncodeDecompressedBlockSizeTest
 #
 slow_test_regexp = \
-  ^t/DBTest\.(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$
+	^t/run-table_test-HarnessTest.Randomized$$|^t/run-db_test-.*(?:FileCreationRandomFailure|EncodeDecompressedBlockSizeTest)$$
 prioritize_long_running_tests =						\
   perl -pe 's,($(slow_test_regexp)),100 $$1,'				\
     | sort -k1,1gr							\
@@ -537,26 +585,43 @@ prioritize_long_running_tests =						\
 # Run with "make J=200% check" to run two parallel jobs per core.
 # The default is to run one job per core (J=100%).
 # See "man parallel" for its "-j ..." option.
-J = 100%
+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: $(t_run)
+check_0:
+	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD); \
+	printf '%s\n' ''						\
+	  'To monitor subtest <duration,pass/fail,name>,'		\
+	  '  run "make watch-log" in a separate window' '';		\
+	test -t 1 && eta=--eta || eta=; \
+	{ \
+		printf './%s\n' $(filter-out $(PARALLEL_TEST),$(TESTS)); \
+		printf '%s\n' $(t_run); \
+	} \
+	  | $(prioritize_long_running_tests)				\
+	  | grep -E '$(tests-regexp)'					\
+	  | parallel -j$(J) --joblog=LOG $$eta --gnu '{} >& t/log-{/}'
+
+.PHONY: valgrind_check_0
+valgrind_check_0:
 	$(AM_V_GEN)export TEST_TMPDIR=$(TMPD);				\
 	printf '%s\n' ''						\
 	  'To monitor subtest <duration,pass/fail,name>,'		\
 	  '  run "make watch-log" in a separate window' '';		\
 	test -t 1 && eta=--eta || eta=;					\
 	{								\
-	  printf './%s\n' $(filter-out db_test, $(TESTS));		\
+	  printf './%s\n' $(filter-out $(PARALLEL_TEST) %skiplist_test options_settable_test, $(TESTS));		\
 	  printf '%s\n' $(t_run);					\
 	}								\
 	  | $(prioritize_long_running_tests)				\
 	  | grep -E '$(tests-regexp)'					\
-	  | parallel -j$(J) --joblog=LOG $$eta --gnu '{} >& t/log-{/}'
-endif
+	  | parallel -j$(J) --joblog=LOG $$eta --gnu \
+      'if [[ "{}" == "./"* ]] ; then $(DRIVER) {} >& t/valgrind_log-{/}; ' \
+      'else {} >& t/valgrind_log-{/}; fi'
 
 CLEAN_FILES += t LOG $(TMPD)
 
@@ -573,11 +638,11 @@ watch-log:
 # If J != 1 and GNU parallel is installed, run the tests in parallel,
 # via the check_0 rule above.  Otherwise, run them sequentially.
 check: all
+	$(MAKE) gen_parallel_tests
 	$(AM_V_GEN)if test "$(J)" != 1                                  \
 	    && (parallel --gnu --help 2>/dev/null) |                    \
 	        grep -q 'GNU Parallel';                                 \
 	then                                                            \
-	    t=$$($(test_names));                                        \
 	    $(MAKE) T="$$t" TMPD=$(TMPD) check_0;                       \
 	else                                                            \
 	    for t in $(TESTS); do                                       \
@@ -599,7 +664,7 @@ ldb_tests: ldb
 crash_test: whitebox_crash_test blackbox_crash_test
 
 blackbox_crash_test: db_stress
-	python -u tools/db_crashtest.py --simple blackbox 
+	python -u tools/db_crashtest.py --simple blackbox
 	python -u tools/db_crashtest.py blackbox
 
 whitebox_crash_test: db_stress
@@ -616,14 +681,33 @@ asan_crash_test:
 	COMPILE_WITH_ASAN=1 $(MAKE) crash_test
 	$(MAKE) clean
 
+ubsan_check:
+	$(MAKE) clean
+	COMPILE_WITH_UBSAN=1 $(MAKE) check -j32
+	$(MAKE) clean
+
+ubsan_crash_test:
+	$(MAKE) clean
+	COMPILE_WITH_UBSAN=1 $(MAKE) crash_test
+	$(MAKE) clean
+
 valgrind_check: $(TESTS)
-	for t in $(filter-out %skiplist_test,$(TESTS)); do \
-		$(VALGRIND_VER) $(VALGRIND_OPTS) ./$$t; \
-		ret_code=$$?; \
-		if [ $$ret_code -ne 0 ]; then \
-			exit $$ret_code; \
-		fi; \
-	done
+	$(MAKE) gen_parallel_tests
+	$(AM_V_GEN)if test "$(J)" != 1                                  \
+	    && (parallel --gnu --help 2>/dev/null) |                    \
+	        grep -q 'GNU Parallel';                                 \
+	then                                                            \
+      $(MAKE) TMPD=$(TMPD)                                        \
+      DRIVER="$(VALGRIND_VER) $(VALGRIND_OPTS)" valgrind_check_0; \
+	else                                                            \
+		for t in $(filter-out %skiplist_test options_settable_test,$(TESTS)); do \
+			$(VALGRIND_VER) $(VALGRIND_OPTS) ./$$t; \
+			ret_code=$$?; \
+			if [ $$ret_code -ne 0 ]; then \
+				exit $$ret_code; \
+			fi; \
+		done; \
+	fi
 
 
 ifneq ($(PAR_TEST),)
@@ -647,6 +731,13 @@ parloop:
 	exit $$ret_bad;
 endif
 
+test_names = \
+  ./db_test --gtest_list_tests						\
+    | perl -n								\
+      -e 's/ *\#.*//;'							\
+      -e '/^(\s*)(\S+)/; !$$1 and do {$$p=$$2; break};'			\
+      -e 'print qq! $$p$$2!'
+
 parallel_check: $(TESTS)
 	$(AM_V_GEN)if test "$(J)" > 1                                  \
 	    && (parallel --gnu --help 2>/dev/null) |                    \
@@ -716,6 +807,14 @@ $(LIBRARY): $(LIBOBJECTS)
 	$(AM_V_AR)rm -f $@
 	$(AM_V_at)$(AR) $(ARFLAGS) $@ $(LIBOBJECTS)
 
+$(TOOLS_LIBRARY): $(BENCH_SOURCES:.cc=.o) $(TOOL_SOURCES:.cc=.o) $(LIB_SOURCES:.cc=.o) $(TESTUTIL)
+	$(AM_V_AR)rm -f $@
+	$(AM_V_at)$(AR) $(ARFLAGS) $@ $^
+
+librocksdb_env_basic_test.a: util/env_basic_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_V_AR)rm -f $@
+	$(AM_V_at)$(AR) $(ARFLAGS) $@ $^
+
 db_bench: tools/db_bench.o $(BENCHTOOLOBJECTS)
 	$(AM_LINK)
 
@@ -725,9 +824,6 @@ cache_bench: util/cache_bench.o $(LIBOBJECTS) $(TESTUTIL)
 memtablerep_bench: db/memtablerep_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)
 
-block_hash_index_test: table/block_hash_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 db_stress: tools/db_stress.o $(LIBOBJECTS) $(TESTUTIL)
 	$(AM_LINK)
 
@@ -773,6 +869,9 @@ stringappend_test: utilities/merge_operators/string_append/stringappend_test.o $
 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)
+
 histogram_test: util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -794,6 +893,12 @@ db_test: db/db_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 db_test2: db/db_test2.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+db_block_cache_test: db/db_block_cache_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_bloom_filter_test: db/db_bloom_filter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_log_iter_test: db/db_log_iter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -809,6 +914,12 @@ db_dynamic_level_test: db/db_dynamic_level_test.o db/db_test_util.o $(LIBOBJECTS
 db_inplace_update_test: db/db_inplace_update_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+db_iterator_test: db/db_iterator_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
+db_sst_test: db/db_sst_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_tailing_iter_test: db/db_tailing_iter_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -821,6 +932,9 @@ db_universal_compaction_test: db/db_universal_compaction_test.o db/db_test_util.
 db_wal_test: db/db_wal_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+db_io_failure_test: db/db_io_failure_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 db_properties_test: db/db_properties_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -828,7 +942,7 @@ db_table_properties_test: db/db_table_properties_test.o db/db_test_util.o $(LIBO
 	$(AM_LINK)
 
 log_write_bench: util/log_write_bench.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK) $(pg)
+	$(AM_LINK) $(PROFILING_FLAGS)
 
 plain_table_db_test: db/plain_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
@@ -837,7 +951,7 @@ comparator_db_test: db/comparator_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
 table_reader_bench: table/table_reader_bench.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK) $(pg)
+	$(AM_LINK) $(PROFILING_FLAGS)
 
 perf_context_test: db/perf_context_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
@@ -857,12 +971,18 @@ document_db_test: utilities/document/document_db_test.o $(LIBOBJECTS) $(TESTHARN
 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)
 
+env_registry_test: utilities/env_registry_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -890,6 +1010,9 @@ wal_manager_test: db/wal_manager_test.o $(LIBOBJECTS) $(TESTHARNESS)
 dbformat_test: db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+env_basic_test: util/env_basic_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 env_test: util/env_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -965,6 +1088,9 @@ merge_test: db/merge_test.o $(LIBOBJECTS) $(TESTHARNESS)
 merger_test: table/merger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+util_merge_operators_test: utilities/util_merge_operators_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 options_file_test: db/options_file_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1001,18 +1127,21 @@ compact_files_test: db/compact_files_test.o $(LIBOBJECTS) $(TESTHARNESS)
 options_test: util/options_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+options_settable_test: util/options_settable_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 options_util_test: utilities/options/options_util_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
+db_bench_tool_test: tools/db_bench_tool_test.o $(BENCHTOOLOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 event_logger_test: util/event_logger_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
 sst_dump_test: tools/sst_dump_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
-memenv_test : util/memenv_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(AM_LINK)
-
 optimistic_transaction_test: utilities/transactions/optimistic_transaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1043,6 +1172,9 @@ transaction_test: utilities/transactions/transaction_test.o $(LIBOBJECTS) $(TEST
 sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 	$(AM_LINK)
 
+repair_test: db/repair_test.o db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 ldb_cmd_test: tools/ldb_cmd_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_LINK)
 
@@ -1052,6 +1184,9 @@ ldb: tools/ldb.o $(LIBOBJECTS)
 iostats_context_test: util/iostats_context_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(AM_V_CCLD)$(CXX) $^ $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
+persistent_cache_test: utilities/persistent_cache/persistent_cache_test.o  db/db_test_util.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 INSTALL_PATH ?= /usr/local
@@ -1214,7 +1349,7 @@ jdb_bench:
 
 commit_prereq: build_tools/rocksdb-lego-determinator \
                build_tools/precommit_checker.py
-	J=$(J) build_tools/precommit_checker.py unit unit_481 clang_unit tsan asan lite
+	J=$(J) build_tools/precommit_checker.py unit unit_481 clang_unit tsan asan ubsan lite
 	$(MAKE) clean && $(MAKE) jclean && $(MAKE) rocksdbjava;
 
 xfunc:

USERS.md

@@ -9,7 +9,7 @@ At Facebook, we use RocksDB as a backend for many different stateful services. W
 ## LinkedIn
 Two different use cases at Linkedin are using RocksDB as a storage engine:
 
-1. LinkedIn's follow feed for storing user's activities
+1. LinkedIn's follow feed for storing user's activities. Check out the blog post: https://engineering.linkedin.com/blog/2016/03/followfeed--linkedin-s-feed-made-faster-and-smarter
 2. Apache Samza, open source framework for stream processing
 
 Learn more about those use cases in a Tech Talk by Ankit Gupta and Naveen Somasundaram: http://www.youtube.com/watch?v=plqVp_OnSzg
@@ -52,3 +52,10 @@ Pinterest's Object Retrieval System uses RocksDB for storage: https://www.youtub
 
 ## VWO, Wingify
 [VWO's](https://vwo.com/) Smart Code checker and URL helper uses RocksDB to store all the URLs where VWO's Smart Code is installed.
+
+## quasardb
+[quasardb](https://www.quasardb.net) is a high-performance, distributed, transactional key-value database that integrates well with in-memory analytics engines such as Apache Spark. 
+quasardb uses a heavily tuned RocksDB as its persistence layer.
+
+## Netflix
+[Netflix](http://techblog.netflix.com/2016/05/application-data-caching-using-ssds.html) uses RocksDB on spinning disks to cache application data.

appveyor.yml

@@ -2,7 +2,7 @@ version: 1.0.{build}
 before_build:
 - md %APPVEYOR_BUILD_FOLDER%\build
 - cd %APPVEYOR_BUILD_FOLDER%\build
-- cmake -G "Visual Studio 12 Win64" -DOPTDBG=1 ..
+- cmake -G "Visual Studio 12 Win64" -DOPTDBG=1 -DXPRESS=1 ..
 - cd ..
 build:
   project: build\rocksdb.sln
@@ -10,6 +10,7 @@ build:
   verbosity: minimal
 test:
 test_script:
-- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test -Exclude DBTest.Randomized,DBTest.FileCreationRandomFailure -Concurrency 18
+- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test2 -Concurrency 8
+- ps: build_tools\run_ci_db_test.ps1 -EnableRerun -Run db_test -Exclude DBTest.GroupCommitTest -Concurrency 10
 - ps: build_tools\run_ci_db_test.ps1 -Run env_test -Concurrency 1
 

arcanist_util/config/FacebookArcanistConfiguration.php

@@ -92,12 +92,13 @@ class FacebookArcanistConfiguration extends ArcanistConfiguration {
     // Run the actual command
     $this->updateTest($diffID, $test);
     $cmd = $this->updateTestCommand($diffID, $test, "running") . ";"
-           . "(./build_tools/precommit_checker.py " . $test
-           . "&& "
+           . "./build_tools/precommit_checker.py " . $test
+           . "; exit_code=$?; ([[ \$exit_code -eq 0 ]] &&"
            . $this->updateTestCommand($diffID, $test, "pass") . ")"
-           . "|| " . $this->updateTestCommand($diffID, $test, "fail")
+           . "||" . $this->updateTestCommand($diffID, $test, "fail")
            . "; cat /tmp/precommit-check.log"
-           . "; for f in `ls t/log-*`; do echo \$f; cat \$f; done";
+           . "; for f in `ls t/log-*`; do echo \$f; cat \$f; done;"
+           . "[[ \$exit_code -eq 0 ]]";
 
     $run_test = array(
       "name" => "Run " . $test,
@@ -130,7 +131,7 @@ class FacebookArcanistConfiguration extends ArcanistConfiguration {
     } else {
       // manually list of tests we want to run in sandcastle
       $tests = array(
-        "unit", "unit_481", "clang_unit", "tsan", "asan", "lite", "valgrind"
+        "unit", "unit_481", "clang_unit", "tsan", "asan", "lite_test", "valgrind"
       );
     }
 

arcanist_util/cpp_linter/cpplint.py

@@ -2714,7 +2714,7 @@ def CheckSpacing(filename, clean_lines, linenum, nesting_state, error):
 
     # Look for < that is not surrounded by spaces.  This is only
     # triggered if both sides are missing spaces, even though
-    # technically should should flag if at least one side is missing a
+    # technically should flag if at least one side is missing a
     # space.  This is done to avoid some false positives with shifts.
     match = Search(r'[^\s<]<([^\s=<].*)', reduced_line)
     if (match and

build_tools/build_detect_platform

@@ -299,13 +299,14 @@ EOF
 
     # Test whether jemalloc is available
     if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null -ljemalloc \
-      2>/dev/null; then 
+      2>/dev/null; then
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ljemalloc"
         JAVA_LDFLAGS="$JAVA_LDFLAGS -ljemalloc"
+        JEMALLOC=1
     else
         # jemalloc is not available. Let's try tcmalloc
         if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null \
-          -ltcmalloc 2>/dev/null; then 
+          -ltcmalloc 2>/dev/null; then
             PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltcmalloc"
             JAVA_LDFLAGS="$JAVA_LDFLAGS -ltcmalloc"
         fi
@@ -322,6 +323,55 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
     fi
+
+    # Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <pthread.h>
+      int main() {
+        int x = PTHREAD_MUTEX_ADAPTIVE_NP;
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_PTHREAD_ADAPTIVE_MUTEX"
+    fi
+
+    # Test whether backtrace is available
+    $CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <execinfo.h>>
+      int main() {
+        void* frames[1];
+        backtrace_symbols(frames, backtrace(frames, 1));
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+    else
+        # Test whether execinfo library is installed
+        $CXX $CFLAGS -lexecinfo -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <execinfo.h>
+          int main() {
+            void* frames[1];
+            backtrace_symbols(frames, backtrace(frames, 1));
+          }
+EOF
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lexecinfo"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lexecinfo"
+        fi
+    fi
+
+    # Test if -pg is supported
+    $CXX $CFLAGS -pg -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      int main() {
+        return 0;
+      }
+EOF
+    if [ "$?" = 0 ]; then
+        PROFILING_FLAGS=-pg
+    fi
 fi
 
 # TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
@@ -402,3 +452,7 @@ echo "ROCKSDB_MINOR=$ROCKSDB_MINOR" >> "$OUTPUT"
 echo "ROCKSDB_PATCH=$ROCKSDB_PATCH" >> "$OUTPUT"
 echo "CLANG_SCAN_BUILD=$CLANG_SCAN_BUILD" >> "$OUTPUT"
 echo "CLANG_ANALYZER=$CLANG_ANALYZER" >> "$OUTPUT"
+echo "PROFILING_FLAGS=$PROFILING_FLAGS" >> "$OUTPUT"
+if test -n "$JEMALLOC"; then
+  echo "JEMALLOC=1" >> "$OUTPUT"
+fi

build_tools/dependencies.sh

@@ -1,16 +1,16 @@
-GCC_BASE=/mnt/vol/engshare/fbcode/third-party2/gcc/4.9.x/centos6-native/1317bc4/
-CLANG_BASE=/mnt/gvfs/third-party2/clang/fc904e50a9266b9d7b98cae1993afa0c5aae1440/3.7.1/centos6-native/9d9ecb9/
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/ea2fd1278810d3af2ea52218d2767e09d786dbd0/4.9.x/gcc-4.9-glibc-2.20/024dbc3
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/f5484f168c0e4d19823d41df052c5870c6e575a4/2.20/gcc-4.9-glibc-2.20/500e281
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/cbf6f1f209e5bd160bdc5d971744e039f36b1566/1.1.3/gcc-4.9-glibc-2.20/e9936bf
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/6d39cb54708049f527e713ad19f2aadb9d3667e8/1.2.8/gcc-4.9-glibc-2.20/e9936bf
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/2ddd45f0853bfc8bb1c27f0f447236a1a26c338a/1.0.6/gcc-4.9-glibc-2.20/e9936bf
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/6858fac689e0f92e584224d91bdb0e39f6c8320d/r131/gcc-4.9-glibc-2.20/e9936bf
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/cb6c4880fcb4fee471574ba6af63a3882155a16a/0.5.1/gcc-4.9-glibc-2.20/e9936bf
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/c7275a4ceae0aca0929e56964a31dafc53c1ee96/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/40791a3fef9206a77f2c4bc51f8169e5bf10d68e/master/gcc-4.9-glibc-2.20/a6c5e1e
-NUMA_BASE=/mnt/gvfs/third-party2/numa/ae54a5ed22cdabb1c6446dce4e8ffae5b4446d73/2.0.8/gcc-4.9-glibc-2.20/e9936bf
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/303048f72efc92ae079e62dfc84823401aecfd94/trunk/gcc-4.9-glibc-2.20/12266b1
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/1a48835975c66d30e47770ec419758ed3b9ba010/3.10.62-62_fbk17_03959_ge29cc63/gcc-4.9-glibc-2.20/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/a5b8152b2a15ce8a98808cf954fbccec825a97bc/2.25/centos6-native/da39a3e
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/af85c56f424cd5edfc2c97588299b44ecdec96bb/3.10.0/gcc-4.9-glibc-2.20/e9936bf
+GCC_BASE=/mnt/gvfs/third-party2/gcc/ebc96bc2fb751b5a0300b8d91a95bdf24ac1d88b/4.9.x/centos6-native/108cf83
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/b91de48a4974ec839946d824402b098d43454cef/stable/centos6-native/7aaccbe
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/53e0eac8911888a105aa98b9a35fe61cf1d8b278/4.9.x/gcc-4.9-glibc-2.20/024dbc3
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/ee36ac9a72dfac4a995f1b215bb4c0fc8a0f6f91/2.20/gcc-4.9-glibc-2.20/500e281
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-4.9-glibc-2.20/e9936bf
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2b24f1e99454f9ca7b0301720f94836dae1bf71b/1.2.8/gcc-5-glibc-2.23/9bc6787
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/af7c14c9b652cdd5ec34eadd25c3f38a9b306c5d/1.0.6/gcc-5-glibc-2.23/9bc6787
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0e790b441e2d9acd68d51e1d2e028f88c6a79ddf/r131/gcc-5-glibc-2.23/9bc6787
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/1408484d03b15492aa54b10356104e9dc22e1cc5/0.6.1/gcc-5-glibc-2.23/9bc6787
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/5a5c7a6608cb32f1e1e7f814023d5bdfbd136370/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/a4c2adecffcaa68d5585d06be2252e3efa52555b/master/gcc-5-glibc-2.23/1c32b4b
+NUMA_BASE=/mnt/gvfs/third-party2/numa/1abc0d3c01743b854676423cf2d3629912f34930/2.0.8/gcc-5-glibc-2.23/9bc6787
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/ad576de2a1ea560c4d3434304f0fc4e079bede42/trunk/gcc-5-glibc-2.23/b1847cb
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/f24be37d170e04be6e469af487644d4d62e1c6c1/4.0.9-36_fbk5_2933_gd092e3f/gcc-5-glibc-2.23/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/9d65c666b9adf8f2a989fd4b98a9a5e7d3afa233/2.26/centos6-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/9cee5a3628dc9d4b93897972c58eba865e25b270/3.10.0/gcc-4.9-glibc-2.20/e9936bf

build_tools/dependencies_4.8.1.sh

@@ -1,16 +1,16 @@
-GCC_BASE=/mnt/vol/engshare/fbcode/third-party2/gcc/4.8.1/centos6-native/cc6c9dc/
-CLANG_BASE=/mnt/gvfs/third-party2/clang/fc904e50a9266b9d7b98cae1993afa0c5aae1440/3.7.1/centos6-native/9d9ecb9/
-LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/ea2fd1278810d3af2ea52218d2767e09d786dbd0/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc
-GLIBC_BASE=/mnt/gvfs/third-party2/glibc/f5484f168c0e4d19823d41df052c5870c6e575a4/2.17/gcc-4.8.1-glibc-2.17/99df8fc
-SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/cbf6f1f209e5bd160bdc5d971744e039f36b1566/1.1.3/gcc-4.8.1-glibc-2.17/c3f970a
-ZLIB_BASE=/mnt/gvfs/third-party2/zlib/6d39cb54708049f527e713ad19f2aadb9d3667e8/1.2.8/gcc-4.8.1-glibc-2.17/c3f970a
-BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/2ddd45f0853bfc8bb1c27f0f447236a1a26c338a/1.0.6/gcc-4.8.1-glibc-2.17/c3f970a
-LZ4_BASE=/mnt/gvfs/third-party2/lz4/6858fac689e0f92e584224d91bdb0e39f6c8320d/r131/gcc-4.8.1-glibc-2.17/c3f970a
-ZSTD_BASE=/mnt/gvfs/third-party2/zstd/cb6c4880fcb4fee471574ba6af63a3882155a16a/0.5.1/gcc-4.8.1-glibc-2.17/c3f970a
-GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/c7275a4ceae0aca0929e56964a31dafc53c1ee96/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
-JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/40791a3fef9206a77f2c4bc51f8169e5bf10d68e/master/gcc-4.8.1-glibc-2.17/8d31e51
-NUMA_BASE=/mnt/gvfs/third-party2/numa/ae54a5ed22cdabb1c6446dce4e8ffae5b4446d73/2.0.8/gcc-4.8.1-glibc-2.17/c3f970a
-LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/303048f72efc92ae079e62dfc84823401aecfd94/trunk/gcc-4.8.1-glibc-2.17/675d945
-KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/1a48835975c66d30e47770ec419758ed3b9ba010/3.10.62-62_fbk17_03959_ge29cc63/gcc-4.8.1-glibc-2.17/da39a3e
-BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/a5b8152b2a15ce8a98808cf954fbccec825a97bc/2.25/centos6-native/da39a3e
-VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/af85c56f424cd5edfc2c97588299b44ecdec96bb/3.8.1/gcc-4.8.1-glibc-2.17/c3f970a
+GCC_BASE=/mnt/gvfs/third-party2/gcc/ebc96bc2fb751b5a0300b8d91a95bdf24ac1d88b/4.8.1/centos6-native/cc6c9dc
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/b91de48a4974ec839946d824402b098d43454cef/stable/centos6-native/7aaccbe
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/53e0eac8911888a105aa98b9a35fe61cf1d8b278/4.8.1/gcc-4.8.1-glibc-2.17/8aac7fc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/ee36ac9a72dfac4a995f1b215bb4c0fc8a0f6f91/2.17/gcc-4.8.1-glibc-2.17/99df8fc
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/8c38a4c1e52b4c2cc8a9cdc31b9c947ed7dbfcb4/1.1.3/gcc-4.8.1-glibc-2.17/c3f970a
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2b24f1e99454f9ca7b0301720f94836dae1bf71b/1.2.8/gcc-4.8.1-glibc-2.17/c3f970a
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/af7c14c9b652cdd5ec34eadd25c3f38a9b306c5d/1.0.6/gcc-4.8.1-glibc-2.17/c3f970a
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0e790b441e2d9acd68d51e1d2e028f88c6a79ddf/r131/gcc-4.8.1-glibc-2.17/c3f970a
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/1408484d03b15492aa54b10356104e9dc22e1cc5/0.6.1/gcc-4.8.1-glibc-2.17/c3f970a
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/5a5c7a6608cb32f1e1e7f814023d5bdfbd136370/2.1.1/gcc-4.8.1-glibc-2.17/c3f970a
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/a4c2adecffcaa68d5585d06be2252e3efa52555b/master/gcc-4.8.1-glibc-2.17/8d31e51
+NUMA_BASE=/mnt/gvfs/third-party2/numa/1abc0d3c01743b854676423cf2d3629912f34930/2.0.8/gcc-4.8.1-glibc-2.17/c3f970a
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/ad576de2a1ea560c4d3434304f0fc4e079bede42/trunk/gcc-4.8.1-glibc-2.17/675d945
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/f24be37d170e04be6e469af487644d4d62e1c6c1/4.0.9-36_fbk5_2933_gd092e3f/gcc-4.8.1-glibc-2.17/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/9d65c666b9adf8f2a989fd4b98a9a5e7d3afa233/2.26/centos6-native/da39a3e
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/9cee5a3628dc9d4b93897972c58eba865e25b270/3.8.1/gcc-4.8.1-glibc-2.17/c3f970a

build_tools/error_filter.py

@@ -0,0 +1,164 @@
+# Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree. An additional grant
+# of patent rights can be found in the PATENTS file in the same directory.
+
+'''Filter for error messages in test output:
+    - Receives merged stdout/stderr from test on stdin
+    - Finds patterns of known error messages for test name (first argument)
+    - Prints those error messages to stdout
+'''
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import re
+import sys
+
+
+class ErrorParserBase(object):
+    def parse_error(self, line):
+        '''Parses a line of test output. If it contains an error, returns a
+        formatted message describing the error; otherwise, returns None.
+        Subclasses must override this method.
+        '''
+        raise NotImplementedError
+
+
+class GTestErrorParser(ErrorParserBase):
+    '''A parser that remembers the last test that began running so it can print
+    that test's name upon detecting failure.
+    '''
+    _GTEST_NAME_PATTERN = re.compile(r'\[ RUN      \] (\S+)$')
+    # format: '<filename or "unknown file">:<line #>: Failure'
+    _GTEST_FAIL_PATTERN = re.compile(r'(unknown file|\S+:\d+): Failure$')
+
+    def __init__(self):
+        self._last_gtest_name = 'Unknown test'
+
+    def parse_error(self, line):
+        gtest_name_match = self._GTEST_NAME_PATTERN.match(line)
+        if gtest_name_match:
+            self._last_gtest_name = gtest_name_match.group(1)
+            return None
+        gtest_fail_match = self._GTEST_FAIL_PATTERN.match(line)
+        if gtest_fail_match:
+            return '%s failed: %s' % (
+                    self._last_gtest_name, gtest_fail_match.group(1))
+        return None
+
+
+class MatchErrorParser(ErrorParserBase):
+    '''A simple parser that returns the whole line if it matches the pattern.
+    '''
+    def __init__(self, pattern):
+        self._pattern = re.compile(pattern)
+
+    def parse_error(self, line):
+        if self._pattern.match(line):
+            return line
+        return None
+
+
+class CompilerErrorParser(MatchErrorParser):
+    def __init__(self):
+        # format: '<filename>:<line #>:<column #>: error: <error msg>'
+        super(CompilerErrorParser, self).__init__(r'\S+:\d+:\d+: error:')
+
+
+class ScanBuildErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(ScanBuildErrorParser, self).__init__(
+                r'scan-build: \d+ bugs found.$')
+
+
+class DbCrashErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(DbCrashErrorParser, self).__init__(r'\*\*\*.*\^$|TEST FAILED.')
+
+
+class WriteStressErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(WriteStressErrorParser, self).__init__(
+                r'ERROR: write_stress died with exitcode=\d+')
+
+
+class AsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(AsanErrorParser, self).__init__(
+                r'==\d+==ERROR: AddressSanitizer:')
+
+
+class UbsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        # format: '<filename>:<line #>:<column #>: runtime error: <error msg>'
+        super(UbsanErrorParser, self).__init__(r'\S+:\d+:\d+: runtime error:')
+
+
+class ValgrindErrorParser(MatchErrorParser):
+    def __init__(self):
+        # just grab the summary, valgrind doesn't clearly distinguish errors
+        # from other log messages.
+        super(ValgrindErrorParser, self).__init__(r'==\d+== ERROR SUMMARY:')
+
+
+class CompatErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(CompatErrorParser, self).__init__(r'==== .*[Ee]rror.* ====$')
+
+
+class TsanErrorParser(MatchErrorParser):
+    def __init__(self):
+        super(TsanErrorParser, self).__init__(r'WARNING: ThreadSanitizer:')
+
+
+_TEST_NAME_TO_PARSERS = {
+    'punit': [CompilerErrorParser, GTestErrorParser],
+    'unit': [CompilerErrorParser, GTestErrorParser],
+    'unit_481': [CompilerErrorParser, GTestErrorParser],
+    'clang_unit': [CompilerErrorParser, GTestErrorParser],
+    'clang_analyze': [CompilerErrorParser, ScanBuildErrorParser],
+    'code_cov': [CompilerErrorParser, GTestErrorParser],
+    'unity': [CompilerErrorParser, GTestErrorParser],
+    'lite': [CompilerErrorParser],
+    'lite_test': [CompilerErrorParser, GTestErrorParser],
+    'stress_crash': [CompilerErrorParser, DbCrashErrorParser],
+    'write_stress': [CompilerErrorParser, WriteStressErrorParser],
+    'asan': [CompilerErrorParser, GTestErrorParser, AsanErrorParser],
+    'asan_crash': [CompilerErrorParser, AsanErrorParser, DbCrashErrorParser],
+    'ubsan': [CompilerErrorParser, GTestErrorParser, UbsanErrorParser],
+    'ubsan_crash': [CompilerErrorParser, UbsanErrorParser, DbCrashErrorParser],
+    'valgrind': [CompilerErrorParser, GTestErrorParser, ValgrindErrorParser],
+    'tsan': [CompilerErrorParser, GTestErrorParser, TsanErrorParser],
+    'format_compatible': [CompilerErrorParser, CompatErrorParser],
+    'run_format_compatible': [CompilerErrorParser, CompatErrorParser],
+    'no_compression': [CompilerErrorParser, GTestErrorParser],
+    'run_no_compression': [CompilerErrorParser, GTestErrorParser],
+    'regression': [CompilerErrorParser],
+    'run_regression': [CompilerErrorParser],
+}
+
+
+def main():
+    if len(sys.argv) != 2:
+        return 'Usage: %s <test name>' % sys.argv[0]
+    test_name = sys.argv[1]
+    if test_name not in _TEST_NAME_TO_PARSERS:
+        return 'Unknown test name: %s' % test_name
+
+    error_parsers = []
+    for parser_cls in _TEST_NAME_TO_PARSERS[test_name]:
+        error_parsers.append(parser_cls())
+
+    for line in sys.stdin:
+        line = line.strip()
+        for error_parser in error_parsers:
+            error_msg = error_parser.parse_error(line)
+            if error_msg is not None:
+                print(error_msg)
+
+
+if __name__ == '__main__':
+    sys.exit(main())

build_tools/fbcode_config.sh

@@ -18,7 +18,7 @@ LIBGCC_LIBS=" -L $LIBGCC_BASE/libs"
 
 # glibc
 GLIBC_INCLUDE="$GLIBC_BASE/include"
-GLIBC_LIBS=" -L $GLIB_BASE/lib"
+GLIBC_LIBS=" -L $GLIBC_BASE/lib"
 
 # snappy
 SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include/"
@@ -97,9 +97,10 @@ if [ -z "$USE_CLANG" ]; then
   CFLAGS+=" -B$BINUTILS/gold"
   CFLAGS+=" -isystem $GLIBC_INCLUDE"
   CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  JEMALLOC=1
 else
   # clang
-  CLANG_INCLUDE="$CLANG_LIB/clang/*/include"
+  CLANG_INCLUDE="$CLANG_LIB/clang/stable/include"
   CC="$CLANG_BIN/clang"
   CXX="$CLANG_BIN/clang++"
 
@@ -113,6 +114,7 @@ else
   CFLAGS+=" -isystem $CLANG_INCLUDE"
   CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE/linux "
   CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE "
+  CFLAGS+=" -Wno-expansion-to-defined "
   CXXFLAGS="-nostdinc++"
 fi
 

build_tools/fbcode_config4.8.1.sh

@@ -66,6 +66,7 @@ if [ -z "$USE_CLANG" ]; then
   CFLAGS="-B$BINUTILS/gold -m64 -mtune=generic"
   CFLAGS+=" -isystem $GLIBC_INCLUDE"
   CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  JEMALLOC=1
 else
   # clang
   CLANG_BIN="$CLANG_BASE/bin"

build_tools/precommit_checker.py

@@ -184,7 +184,7 @@ parser = argparse.ArgumentParser(description='RocksDB pre-commit checker.')
 
 # <test ....>
 parser.add_argument('test', nargs='+',
-                    help='CI test(s) to run. e.g: unit punit asan tsan')
+                    help='CI test(s) to run. e.g: unit punit asan tsan ubsan')
 
 print("Please follow log %s" % Log.LOG_FILE)
 

build_tools/rocksdb-lego-determinator

@@ -65,10 +65,13 @@ SHM="TEST_TMPDIR=/dev/shm/rocksdb"
 GCC_481="ROCKSDB_FBCODE_BUILD_WITH_481=1"
 ASAN="COMPILE_WITH_ASAN=1"
 CLANG="USE_CLANG=1"
-LITE="OPT=-DROCKSDB_LITE"
+LITE="OPT=\"-DROCKSDB_LITE -g\""
 TSAN="COMPILE_WITH_TSAN=1"
+UBSAN="COMPILE_WITH_UBSAN=1"
 DISABLE_JEMALLOC="DISABLE_JEMALLOC=1"
-PARSER="'parser':'egrep \'Failure|^#|Abort|Expected|Actual|GoogleTestFailure|^==\''"
+HTTP_PROXY="https_proxy=http://fwdproxy.29.prn1:8080 http_proxy=http://fwdproxy.29.prn1:8080 ftp_proxy=http://fwdproxy.29.prn1:8080"
+SETUP_JAVA_ENV="export $HTTP_PROXY; export JAVA_HOME=/usr/local/jdk-7u10-64/; export PATH=\$PATH:\$JAVA_HOME/bin"
+PARSER="'parser':'python build_tools/error_filter.py $1'"
 
 ARTIFACTS=" 'artifacts': [
     {
@@ -254,7 +257,7 @@ LITE_BUILD_COMMANDS="[
             $CLEANUP_ENV,
             {
                 'name':'Build RocksDB debug version',
-                'shell':'$LITE $DEBUG make J=1 static_lib',
+                'shell':'$LITE make J=1 static_lib',
                 'user':'root',
                 $PARSER
             },
@@ -274,7 +277,7 @@ LITE_UNIT_TEST_COMMANDS="[
             $CLEANUP_ENV,
             {
                 'name':'Build RocksDB debug version',
-                'shell':'$SHM $LITE $DEBUG make J=1 check',
+                'shell':'$SHM $LITE make J=1 check',
                 'user':'root',
                 $PARSER
             },
@@ -376,6 +379,48 @@ ASAN_CRASH_TEST_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB test under undefined behavior sanitizer
+#
+UBSAN_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb Unit Test under UBSAN',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Test RocksDB debug under UBSAN',
+                'shell':'set -o pipefail && $SHM $UBSAN $DEBUG make J=1 ubsan_check',
+                'user':'root',
+                $PARSER
+            }
+        ],
+        $REPORT
+    }
+]"
+
+#
+# RocksDB crash testing under udnefined behavior sanitizer
+#
+UBSAN_CRASH_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb crash test under UBSAN',
+        'oncall':'$ONCALL',
+        'timeout': 86400,
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build and run RocksDB debug ubsan_crash_test',
+                'timeout': 86400,
+                'shell':'$SHM $DEBUG make J=1 ubsan_crash_test',
+                'user':'root',
+                $PARSER
+            },
+        ],
+        $REPORT
+    }
+]"
+
 #
 # RocksDB unit test under valgrind
 #
@@ -462,9 +507,9 @@ run_format_compatible()
   if [ -e "build_detect_platform" ]
   then
     sed "s/tcmalloc/nothingnothingnothing/g" build_detect_platform > $TEST_TMPDIR/temp_build_file
-    rm -rf build_detect_platform 
+    rm -rf build_detect_platform
     cp $TEST_TMPDIR/temp_build_file build_detect_platform
-    chmod +x build_detect_platform 
+    chmod +x build_detect_platform
   fi
 
   make ldb -j32
@@ -592,6 +637,27 @@ REGRESSION_COMMANDS="[
     }
 ]"
 
+#
+# RocksDB Java build
+#
+JAVA_BUILD_TEST_COMMANDS="[
+    {
+        'name':'Rocksdb Java Build',
+        'oncall':'$ONCALL',
+        'steps': [
+            $CLEANUP_ENV,
+            {
+                'name':'Build RocksDB for Java',
+                'shell':'$SETUP_JAVA_ENV; $SHM make rocksdbjava',
+                'user':'root',
+                $PARSER
+            },
+        ],
+        $REPORT
+    }
+]"
+
+
 case $1 in
   punit)
     echo $PARALLEL_UNIT_TEST_COMMANDS
@@ -632,6 +698,12 @@ case $1 in
   asan_crash)
     echo $ASAN_CRASH_TEST_COMMANDS
     ;;
+  ubsan)
+    echo $UBSAN_TEST_COMMANDS
+    ;;
+  ubsan_crash)
+    echo $UBSAN_CRASH_TEST_COMMANDS
+    ;;
   valgrind)
     echo $VALGRIND_TEST_COMMANDS
     ;;
@@ -659,6 +731,9 @@ case $1 in
   run_regression)
     run_regression
     ;;
+  java_build)
+    echo $JAVA_BUILD_TEST_COMMANDS
+    ;;
   *)
     echo "Invalid determinator command"
     ;;

build_tools/run_ci_db_test.ps1

@@ -9,7 +9,7 @@ Param(
   [string]$WorkFolder = "",  # Direct tests to use that folder
   [int]$Limit = -1, # -1 means run all otherwise limit for testing purposes
   [string]$Exclude = "", # Expect a comma separated list, no spaces
-  [string]$Run = "db_test",  # Run db_test|tests|testname1,testname2...
+  [string]$Run = "db_test",  # Run db_test|db_test2|tests|testname1,testname2...
    # Number of async tasks that would run concurrently. Recommend a number below 64.
    # However, CPU utlization really depends on the storage media. Recommend ram based disk.
    # a value of 1 will run everything serially
@@ -46,15 +46,25 @@ if($WorkFolder -eq "") {
   $Env:TEST_TMPDIR = $WorkFolder
 }
 
+Write-Output "Root: $RootFolder, WorkFolder: $WorkFolder"
+
 # Use JEMALLOC executables
-if($EnableJE) {
-    $db_test = -Join ($BinariesFolder, "db_test_je.exe")
-} else {
-    $db_test = -Join ($BinariesFolder, "db_test.exe")
+if($Run -ceq "db_test" -or
+   $Run -ceq "db_test2" ) {
+
+   $file_name = $Run
+
+   if($EnableJE) {
+     $file_name += "_je"
+   }
+
+   $file_name += ".exe"
+
+   $db_test = -Join ($BinariesFolder, $file_name)
+
+   Write-Output "Binaries: $BinariesFolder db_test: $db_test"
 }
 
-Write-Output "Root: $RootFolder, WorkFolder: $WorkFolder"
-Write-Output "Binaries: $BinariesFolder db_test: $db_test"
 
 #Exclusions that we do not want to run
 $ExcludeTests = New-Object System.Collections.Generic.HashSet[string]
@@ -153,7 +163,8 @@ function Discover-TestBinaries([string]$Pattern, $HashTable) {
 
 $TestsToRun = [ordered]@{}
 
-if($Run -ceq "db_test") {
+if($Run -ceq "db_test" -or
+   $Run -ceq "db_test2") {
     Normalize-DbTests -HashTable $TestsToRun
 } elseif($Run -ceq "tests") {
     if($EnableJE) {
@@ -225,7 +236,8 @@ function RunJobs($TestToLog, [int]$ConcurrencyVal, [bool]$AddForRerun)
             Write-Host "Starting $k"
             $log_path = ($TestToLog.$k)
 
-            if($Run -ceq "db_test") {
+            if($Run -ceq "db_test" -or
+               $Run -ceq "db_test2") {
               $job = Start-Job -Name $k -ScriptBlock $InvokeTestCase -ArgumentList @($db_test,$k,$log_path)
             } else {
               [string]$Exe =  -Join ($BinariesFolder, $k)

build_tools/update_dependencies.sh

@@ -64,16 +64,16 @@ touch "$OUTPUT"
 echo "Writing dependencies to $OUTPUT"
 
 # Compilers locations
-GCC_BASE=`ls -d1 $TP2_LATEST/gcc/4.9.x/centos6-native/*/ | head -n1`
-CLANG_BASE=`ls -d1 /mnt/gvfs/third-party2/clang/fc904e50a9266b9d7b98cae1993afa0c5aae1440/3.7.1/centos6-native/*/ | head -n1`
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/4.9.x/centos6-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos6-native/*/`
 
 log_variable GCC_BASE
 log_variable CLANG_BASE
 
 # Libraries locations
 get_lib_base libgcc     4.9.x
-get_lib_base glibc      2.20 
-get_lib_base snappy     LATEST
+get_lib_base glibc      2.20
+get_lib_base snappy     LATEST gcc-4.9-glibc-2.20
 get_lib_base zlib       LATEST
 get_lib_base bzip2      LATEST
 get_lib_base lz4        LATEST
@@ -85,7 +85,7 @@ get_lib_base libunwind  LATEST
 
 get_lib_base kernel-headers LATEST 
 get_lib_base binutils   LATEST centos6-native 
-get_lib_base valgrind   LATEST
+get_lib_base valgrind   3.10.0 gcc-4.9-glibc-2.20
 
 git diff $OUTPUT
 
@@ -101,8 +101,8 @@ touch "$OUTPUT"
 echo "Writing 4.8.1 dependencies to $OUTPUT"
 
 # Compilers locations
-GCC_BASE=`ls -d1 $TP2_LATEST/gcc/4.8.1/centos6-native/*/ | head -n1`
-CLANG_BASE=`ls -d1 /mnt/gvfs/third-party2/clang/fc904e50a9266b9d7b98cae1993afa0c5aae1440/3.7.1/centos6-native/*/ | head -n1`
+GCC_BASE=`readlink -f $TP2_LATEST/gcc/4.8.1/centos6-native/*/`
+CLANG_BASE=`readlink -f $TP2_LATEST/llvm-fb/stable/centos6-native/*/`
 
 log_variable GCC_BASE
 log_variable CLANG_BASE

db/auto_roll_logger.cc

@@ -6,8 +6,6 @@
 #include "db/auto_roll_logger.h"
 #include "util/mutexlock.h"
 
-using namespace std;
-
 namespace rocksdb {
 
 // -- AutoRollLogger
@@ -47,7 +45,8 @@ void AutoRollLogger::RollLogFile() {
   env_->RenameFile(log_fname_, old_fname);
 }
 
-string AutoRollLogger::ValistToString(const char* format, va_list args) const {
+std::string AutoRollLogger::ValistToString(const char* format,
+                                           va_list args) const {
   // Any log messages longer than 1024 will get truncated.
   // The user is responsible for chopping longer messages into multi line log
   static const int MAXBUFFERSIZE = 1024;
@@ -112,7 +111,7 @@ void AutoRollLogger::LogHeader(const char* format, va_list args) {
   // strings
   va_list tmp;
   va_copy(tmp, args);
-  string data = ValistToString(format, tmp);
+  std::string data = ValistToString(format, tmp);
   va_end(tmp);
 
   MutexLock l(&mutex_);

db/auto_roll_logger_test.cc

@@ -19,8 +19,6 @@
 #include <sys/stat.h>
 #include <errno.h>
 
-using namespace std;
-
 namespace rocksdb {
 
 class AutoRollLoggerTest : public testing::Test {
@@ -40,23 +38,22 @@ class AutoRollLoggerTest : public testing::Test {
     Env::Default()->CreateDir(kTestDir);
   }
 
-  void RollLogFileBySizeTest(AutoRollLogger* logger,
-                             size_t log_max_size,
-                             const string& log_message);
-  uint64_t RollLogFileByTimeTest(AutoRollLogger* logger,
-                                 size_t time,
-                                 const string& log_message);
+  void RollLogFileBySizeTest(AutoRollLogger* logger, size_t log_max_size,
+                             const std::string& log_message);
+  uint64_t RollLogFileByTimeTest(AutoRollLogger* logger, size_t time,
+                                 const std::string& log_message);
 
-  static const string kSampleMessage;
-  static const string kTestDir;
-  static const string kLogFile;
+  static const std::string kSampleMessage;
+  static const std::string kTestDir;
+  static const std::string kLogFile;
   static Env* env;
 };
 
-const string AutoRollLoggerTest::kSampleMessage(
+const std::string AutoRollLoggerTest::kSampleMessage(
     "this is the message to be written to the log file!!");
-const string AutoRollLoggerTest::kTestDir(test::TmpDir() + "/db_log_test");
-const string AutoRollLoggerTest::kLogFile(test::TmpDir() + "/db_log_test/LOG");
+const std::string AutoRollLoggerTest::kTestDir(test::TmpDir() + "/db_log_test");
+const std::string AutoRollLoggerTest::kLogFile(test::TmpDir() +
+                                               "/db_log_test/LOG");
 Env* AutoRollLoggerTest::env = Env::Default();
 
 // In this test we only want to Log some simple log message with
@@ -86,7 +83,7 @@ void GetFileCreateTime(const std::string& fname, uint64_t* file_ctime) {
 
 void AutoRollLoggerTest::RollLogFileBySizeTest(AutoRollLogger* logger,
                                                size_t log_max_size,
-                                               const string& log_message) {
+                                               const std::string& log_message) {
   logger->SetInfoLogLevel(InfoLogLevel::INFO_LEVEL);
   // measure the size of each message, which is supposed
   // to be equal or greater than log_message.size()
@@ -111,7 +108,7 @@ void AutoRollLoggerTest::RollLogFileBySizeTest(AutoRollLogger* logger,
 }
 
 uint64_t AutoRollLoggerTest::RollLogFileByTimeTest(
-    AutoRollLogger* logger, size_t time, const string& log_message) {
+    AutoRollLogger* logger, size_t time, const std::string& log_message) {
   uint64_t expected_create_time;
   uint64_t actual_create_time;
   uint64_t total_log_size;
@@ -361,13 +358,13 @@ TEST_F(AutoRollLoggerTest, InfoLogLevel) {
 
 // Test the logger Header function for roll over logs
 // We expect the new logs creates as roll over to carry the headers specified
-static std::vector<string> GetOldFileNames(const string& path) {
-  std::vector<string> ret;
+static std::vector<std::string> GetOldFileNames(const std::string& path) {
+  std::vector<std::string> ret;
 
-  const string dirname = path.substr(/*start=*/ 0, path.find_last_of("/"));
-  const string fname = path.substr(path.find_last_of("/") + 1);
+  const std::string dirname = path.substr(/*start=*/0, path.find_last_of("/"));
+  const std::string fname = path.substr(path.find_last_of("/") + 1);
 
-  std::vector<string> children;
+  std::vector<std::string> children;
   Env::Default()->GetChildren(dirname, &children);
 
   // We know that the old log files are named [path]<something>
@@ -382,12 +379,13 @@ static std::vector<string> GetOldFileNames(const string& path) {
 }
 
 // Return the number of lines where a given pattern was found in the file
-static size_t GetLinesCount(const string& fname, const string& pattern) {
-  stringstream ssbuf;
-  string line;
+static size_t GetLinesCount(const std::string& fname,
+                            const std::string& pattern) {
+  std::stringstream ssbuf;
+  std::string line;
   size_t count = 0;
 
-  ifstream inFile(fname.c_str());
+  std::ifstream inFile(fname.c_str());
   ssbuf << inFile.rdbuf();
 
   while (getline(ssbuf, line)) {
@@ -426,7 +424,7 @@ TEST_F(AutoRollLoggerTest, LogHeaderTest) {
       }
     }
 
-    const string newfname = logger.TEST_log_fname();
+    const std::string newfname = logger.TEST_log_fname();
 
     // Log enough data to cause a roll over
     int i = 0;
@@ -458,7 +456,16 @@ TEST_F(AutoRollLoggerTest, LogHeaderTest) {
 TEST_F(AutoRollLoggerTest, LogFileExistence) {
   rocksdb::DB* db;
   rocksdb::Options options;
-  string deleteCmd = "rm -rf " + kTestDir;
+#ifdef OS_WIN
+  // Replace all slashes in the path so windows CompSpec does not
+  // become confused
+  std::string testDir(kTestDir);
+  std::replace_if(testDir.begin(), testDir.end(),
+    [](char ch) { return ch == '/'; }, '\\');
+  std::string deleteCmd = "if exist " + testDir + " rd /s /q " + testDir;
+#else
+  std::string deleteCmd = "rm -rf " + kTestDir;
+#endif
   ASSERT_EQ(system(deleteCmd.c_str()), 0);
   options.max_log_file_size = 100 * 1024 * 1024;
   options.create_if_missing = true;

db/builder.cc

@@ -15,6 +15,7 @@
 
 #include "db/compaction_iterator.h"
 #include "db/dbformat.h"
+#include "db/event_helpers.h"
 #include "db/filename.h"
 #include "db/internal_stats.h"
 #include "db/merge_helper.h"
@@ -41,29 +42,39 @@ TableBuilder* NewTableBuilder(
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
-    uint32_t column_family_id, WritableFileWriter* file,
-    const CompressionType compression_type,
-    const CompressionOptions& compression_opts, const bool skip_filters) {
+    uint32_t column_family_id, const std::string& column_family_name,
+    WritableFileWriter* file, const CompressionType compression_type,
+    const CompressionOptions& compression_opts,
+    const std::string* compression_dict, const bool skip_filters) {
+  assert((column_family_id ==
+          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
+         column_family_name.empty());
   return ioptions.table_factory->NewTableBuilder(
       TableBuilderOptions(ioptions, internal_comparator,
                           int_tbl_prop_collector_factories, compression_type,
-                          compression_opts, skip_filters),
+                          compression_opts, compression_dict, skip_filters,
+                          column_family_name),
       column_family_id, file);
 }
 
 Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& ioptions,
-    const EnvOptions& env_options, TableCache* table_cache,
-    InternalIterator* iter, FileMetaData* meta,
+    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    TableCache* table_cache, InternalIterator* iter, FileMetaData* meta,
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
-    uint32_t column_family_id, std::vector<SequenceNumber> snapshots,
+    uint32_t column_family_id, const std::string& column_family_name,
+    std::vector<SequenceNumber> snapshots,
     SequenceNumber earliest_write_conflict_snapshot,
     const CompressionType compression,
     const CompressionOptions& compression_opts, bool paranoid_file_checks,
-    InternalStats* internal_stats, const Env::IOPriority io_priority,
-    TableProperties* table_properties) {
+    InternalStats* internal_stats, TableFileCreationReason reason,
+    EventLogger* event_logger, int job_id, const Env::IOPriority io_priority,
+    TableProperties* table_properties, int level) {
+  assert((column_family_id ==
+          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
+         column_family_name.empty());
   // Reports the IOStats for flush for every following bytes.
   const size_t kReportFlushIOStatsEvery = 1048576;
   Status s;
@@ -72,6 +83,12 @@ Status BuildTable(
 
   std::string fname = TableFileName(ioptions.db_paths, meta->fd.GetNumber(),
                                     meta->fd.GetPathId());
+#ifndef ROCKSDB_LITE
+  EventHelpers::NotifyTableFileCreationStarted(
+      ioptions.listeners, dbname, column_family_name, fname, job_id, reason);
+#endif  // !ROCKSDB_LITE
+  TableProperties tp;
+
   if (iter->Valid()) {
     TableBuilder* builder;
     unique_ptr<WritableFileWriter> file_writer;
@@ -79,6 +96,9 @@ Status BuildTable(
       unique_ptr<WritableFile> file;
       s = NewWritableFile(env, fname, &file, env_options);
       if (!s.ok()) {
+        EventHelpers::LogAndNotifyTableFileCreationFinished(
+            event_logger, ioptions.listeners, dbname, column_family_name, fname,
+            job_id, meta->fd, tp, reason, s);
         return s;
       }
       file->SetIOPriority(io_priority);
@@ -87,12 +107,13 @@ Status BuildTable(
 
       builder = NewTableBuilder(
           ioptions, internal_comparator, int_tbl_prop_collector_factories,
-          column_family_id, file_writer.get(), compression, compression_opts);
+          column_family_id, column_family_name, file_writer.get(), compression,
+          compression_opts);
     }
 
     MergeHelper merge(env, internal_comparator.user_comparator(),
                       ioptions.merge_operator, nullptr, ioptions.info_log,
-                      ioptions.min_partial_merge_operands,
+                      mutable_cf_options.min_partial_merge_operands,
                       true /* internal key corruption is not ok */,
                       snapshots.empty() ? 0 : snapshots.back());
 
@@ -125,11 +146,13 @@ Status BuildTable(
     }
 
     if (s.ok() && !empty) {
-      meta->fd.file_size = builder->FileSize();
+      uint64_t file_size = builder->FileSize();
+      meta->fd.file_size = file_size;
       meta->marked_for_compaction = builder->NeedCompact();
       assert(meta->fd.GetFileSize() > 0);
+      tp = builder->GetTableProperties();
       if (table_properties) {
-        *table_properties = builder->GetTableProperties();
+        *table_properties = tp;
       }
     }
     delete builder;
@@ -149,7 +172,8 @@ Status BuildTable(
           ReadOptions(), env_options, internal_comparator, meta->fd, nullptr,
           (internal_stats == nullptr) ? nullptr
                                       : internal_stats->GetFileReadHist(0),
-          false));
+          false /* for_compaction */, nullptr /* arena */,
+          false /* skip_filter */, level));
       s = it->status();
       if (s.ok() && paranoid_file_checks) {
         for (it->SeekToFirst(); it->Valid(); it->Next()) {
@@ -167,6 +191,12 @@ Status BuildTable(
   if (!s.ok() || meta->fd.GetFileSize() == 0) {
     env->DeleteFile(fname);
   }
+
+  // Output to event logger and fire events.
+  EventHelpers::LogAndNotifyTableFileCreationFinished(
+      event_logger, ioptions.listeners, dbname, column_family_name, fname,
+      job_id, meta->fd, tp, reason, s);
+
   return s;
 }
 

db/builder.h

@@ -12,11 +12,14 @@
 #include "db/table_properties_collector.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
-#include "rocksdb/status.h"
-#include "rocksdb/types.h"
-#include "rocksdb/options.h"
 #include "rocksdb/immutable_options.h"
+#include "rocksdb/listener.h"
+#include "rocksdb/options.h"
+#include "rocksdb/status.h"
 #include "rocksdb/table_properties.h"
+#include "rocksdb/types.h"
+#include "util/event_logger.h"
+#include "util/mutable_cf_options.h"
 
 namespace rocksdb {
 
@@ -33,14 +36,20 @@ class WritableFileWriter;
 class InternalStats;
 class InternalIterator;
 
+// @param column_family_name Name of the column family that is also identified
+//    by column_family_id, or empty string if unknown. It must outlive the
+//    TableBuilder returned by this function.
+// @param compression_dict Data for presetting the compression library's
+//    dictionary, or nullptr.
 TableBuilder* NewTableBuilder(
     const ImmutableCFOptions& options,
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
-    uint32_t column_family_id, WritableFileWriter* file,
-    const CompressionType compression_type,
+    uint32_t column_family_id, const std::string& column_family_name,
+    WritableFileWriter* file, const CompressionType compression_type,
     const CompressionOptions& compression_opts,
+    const std::string* compression_dict = nullptr,
     const bool skip_filters = false);
 
 // Build a Table file from the contents of *iter.  The generated file
@@ -48,19 +57,24 @@ TableBuilder* NewTableBuilder(
 // *meta will be filled with metadata about the generated table.
 // If no data is present in *iter, meta->file_size will be set to
 // zero, and no Table file will be produced.
+//
+// @param column_family_name Name of the column family that is also identified
+//    by column_family_id, or empty string if unknown.
 extern Status BuildTable(
     const std::string& dbname, Env* env, const ImmutableCFOptions& options,
-    const EnvOptions& env_options, TableCache* table_cache,
-    InternalIterator* iter, FileMetaData* meta,
+    const MutableCFOptions& mutable_cf_options, const EnvOptions& env_options,
+    TableCache* table_cache, InternalIterator* iter, FileMetaData* meta,
     const InternalKeyComparator& internal_comparator,
     const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
         int_tbl_prop_collector_factories,
-    uint32_t column_family_id, std::vector<SequenceNumber> snapshots,
+    uint32_t column_family_id, const std::string& column_family_name,
+    std::vector<SequenceNumber> snapshots,
     SequenceNumber earliest_write_conflict_snapshot,
     const CompressionType compression,
     const CompressionOptions& compression_opts, bool paranoid_file_checks,
-    InternalStats* internal_stats,
+    InternalStats* internal_stats, TableFileCreationReason reason,
+    EventLogger* event_logger = nullptr, int job_id = 0,
     const Env::IOPriority io_priority = Env::IO_HIGH,
-    TableProperties* table_properties = nullptr);
+    TableProperties* table_properties = nullptr, int level = -1);
 
 }  // namespace rocksdb

db/c.cc

@@ -446,6 +446,12 @@ void rocksdb_backup_engine_create_new_backup(rocksdb_backup_engine_t* be,
   SaveError(errptr, be->rep->CreateNewBackup(db->rep));
 }
 
+void rocksdb_backup_engine_purge_old_backups(rocksdb_backup_engine_t* be,
+                                             uint32_t num_backups_to_keep,
+                                             char** errptr) {
+  SaveError(errptr, be->rep->PurgeOldBackups(num_backups_to_keep));
+}
+
 rocksdb_restore_options_t* rocksdb_restore_options_create() {
   return new rocksdb_restore_options_t;
 }
@@ -825,6 +831,31 @@ rocksdb_iterator_t* rocksdb_create_iterator_cf(
   return result;
 }
 
+void rocksdb_create_iterators(
+    rocksdb_t *db,
+    rocksdb_readoptions_t* opts,
+    rocksdb_column_family_handle_t** column_families,
+    rocksdb_iterator_t** iterators,
+    size_t size,
+    char** errptr) {
+  std::vector<ColumnFamilyHandle*> column_families_vec;
+  for (size_t i = 0; i < size; i++) {
+    column_families_vec.push_back(column_families[i]->rep);
+  }
+
+  std::vector<Iterator*> res;
+  Status status = db->rep->NewIterators(opts->rep, column_families_vec, &res);
+  assert(res.size() == size);
+  if (SaveError(errptr, status)) {
+    return;
+  }
+
+  for (size_t i = 0; i < size; i++) {
+    iterators[i] = new rocksdb_iterator_t;
+    iterators[i]->rep = res[i];
+  }
+}
+
 const rocksdb_snapshot_t* rocksdb_create_snapshot(
     rocksdb_t* db) {
   rocksdb_snapshot_t* result = new rocksdb_snapshot_t;
@@ -1288,6 +1319,11 @@ void rocksdb_block_based_options_set_cache_index_and_filter_blocks(
   options->rep.cache_index_and_filter_blocks = v;
 }
 
+void rocksdb_block_based_options_set_pin_l0_filter_and_index_blocks_in_cache(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.pin_l0_filter_and_index_blocks_in_cache = v;
+}
+
 void rocksdb_block_based_options_set_skip_table_builder_flush(
     rocksdb_block_based_table_options_t* options, unsigned char v) {
   options->rep.skip_table_builder_flush = v;
@@ -1388,6 +1424,11 @@ void rocksdb_options_set_compaction_filter_factory(
       std::shared_ptr<CompactionFilterFactory>(factory);
 }
 
+void rocksdb_options_compaction_readahead_size(
+    rocksdb_options_t* opt, size_t s) {
+  opt->rep.compaction_readahead_size = s;
+}
+
 void rocksdb_options_set_comparator(
     rocksdb_options_t* opt,
     rocksdb_comparator_t* cmp) {
@@ -1531,11 +1572,13 @@ void rocksdb_options_set_compression_per_level(rocksdb_options_t* opt,
   }
 }
 
-void rocksdb_options_set_compression_options(
-    rocksdb_options_t* opt, int w_bits, int level, int strategy) {
+void rocksdb_options_set_compression_options(rocksdb_options_t* opt, int w_bits,
+                                             int level, int strategy,
+                                             int max_dict_bytes) {
   opt->rep.compression_opts.window_bits = w_bits;
   opt->rep.compression_opts.level = level;
   opt->rep.compression_opts.strategy = strategy;
+  opt->rep.compression_opts.max_dict_bytes = max_dict_bytes;
 }
 
 void rocksdb_options_set_prefix_extractor(
@@ -1832,6 +1875,11 @@ void rocksdb_options_set_inplace_update_num_locks(
   opt->rep.inplace_update_num_locks = v;
 }
 
+void rocksdb_options_set_report_bg_io_stats(
+    rocksdb_options_t* opt, int v) {
+  opt->rep.report_bg_io_stats = v;
+}
+
 void rocksdb_options_set_compaction_style(rocksdb_options_t *opt, int style) {
   opt->rep.compaction_style = static_cast<rocksdb::CompactionStyle>(style);
 }
@@ -2079,6 +2127,11 @@ void rocksdb_readoptions_set_tailing(
   opt->rep.tailing = v;
 }
 
+void rocksdb_readoptions_set_readahead_size(
+    rocksdb_readoptions_t* opt, size_t v) {
+  opt->rep.readahead_size = v;
+}
+
 rocksdb_writeoptions_t* rocksdb_writeoptions_create() {
   return new rocksdb_writeoptions_t;
 }
@@ -2120,6 +2173,10 @@ void rocksdb_cache_destroy(rocksdb_cache_t* cache) {
   delete cache;
 }
 
+void rocksdb_cache_set_capacity(rocksdb_cache_t* cache, size_t capacity) {
+  cache->rep->SetCapacity(capacity);
+}
+
 rocksdb_env_t* rocksdb_create_default_env() {
   rocksdb_env_t* result = new rocksdb_env_t;
   result->rep = Env::Default();
@@ -2127,6 +2184,13 @@ rocksdb_env_t* rocksdb_create_default_env() {
   return result;
 }
 
+rocksdb_env_t* rocksdb_create_mem_env() {
+  rocksdb_env_t* result = new rocksdb_env_t;
+  result->rep = rocksdb::NewMemEnv(Env::Default());
+  result->is_default = false;
+  return result;
+}
+
 void rocksdb_env_set_background_threads(rocksdb_env_t* env, int n) {
   env->rep->SetBackgroundThreads(n);
 }
@@ -2338,6 +2402,31 @@ void rocksdb_get_options_from_string(const rocksdb_options_t* base_options,
                                  &new_options->rep));
 }
 
+void rocksdb_delete_file_in_range(rocksdb_t* db, const char* start_key,
+                                  size_t start_key_len, const char* limit_key,
+                                  size_t limit_key_len, char** errptr) {
+  Slice a, b;
+  SaveError(
+      errptr,
+      DeleteFilesInRange(
+          db->rep, db->rep->DefaultColumnFamily(),
+          (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+          (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr)));
+}
+
+void rocksdb_delete_file_in_range_cf(
+    rocksdb_t* db, rocksdb_column_family_handle_t* column_family,
+    const char* start_key, size_t start_key_len, const char* limit_key,
+    size_t limit_key_len, char** errptr) {
+  Slice a, b;
+  SaveError(
+      errptr,
+      DeleteFilesInRange(
+          db->rep, column_family->rep,
+          (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+          (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr)));
+}
+
 void rocksdb_free(void* ptr) { free(ptr); }
 
 }  // end extern "C"

db/c_test.c

@@ -322,7 +322,7 @@ int main(int argc, char** argv) {
   rocksdb_options_set_block_based_table_factory(options, table_options);
 
   rocksdb_options_set_compression(options, rocksdb_no_compression);
-  rocksdb_options_set_compression_options(options, -14, -1, 0);
+  rocksdb_options_set_compression_options(options, -14, -1, 0, 0);
   int compression_levels[] = {rocksdb_no_compression, rocksdb_no_compression,
                               rocksdb_no_compression, rocksdb_no_compression};
   rocksdb_options_set_compression_per_level(options, compression_levels, 4);
@@ -365,6 +365,24 @@ int main(int argc, char** argv) {
     rocksdb_backup_engine_create_new_backup(be, db, &err);
     CheckNoError(err);
 
+    // need a change to trigger a new backup
+    rocksdb_delete(db, woptions, "does-not-exist", 14, &err);
+    CheckNoError(err);
+
+    rocksdb_backup_engine_create_new_backup(be, db, &err);
+    CheckNoError(err);
+
+    const rocksdb_backup_engine_info_t* bei = rocksdb_backup_engine_get_backup_info(be);
+    CheckCondition(rocksdb_backup_engine_info_count(bei) > 1);
+    rocksdb_backup_engine_info_destroy(bei);
+
+    rocksdb_backup_engine_purge_old_backups(be, 1, &err);
+    CheckNoError(err);
+
+    bei = rocksdb_backup_engine_get_backup_info(be);
+    CheckCondition(rocksdb_backup_engine_info_count(bei) == 1);
+    rocksdb_backup_engine_info_destroy(bei);
+
     rocksdb_delete(db, woptions, "foo", 3, &err);
     CheckNoError(err);
 
@@ -768,6 +786,30 @@ int main(int argc, char** argv) {
     CheckNoError(err);
     rocksdb_iter_destroy(iter);
 
+    rocksdb_column_family_handle_t* iters_cf_handles[2] = { handles[0], handles[1] };
+    rocksdb_iterator_t* iters_handles[2];
+    rocksdb_create_iterators(db, roptions, iters_cf_handles, iters_handles, 2, &err);
+    CheckNoError(err);
+
+    iter = iters_handles[0];
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_seek_to_first(iter);
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_destroy(iter);
+
+    iter = iters_handles[1];
+    CheckCondition(!rocksdb_iter_valid(iter));
+    rocksdb_iter_seek_to_first(iter);
+    CheckCondition(rocksdb_iter_valid(iter));
+
+    for (i = 0; rocksdb_iter_valid(iter) != 0; rocksdb_iter_next(iter)) {
+      i++;
+    }
+    CheckCondition(i == 1);
+    rocksdb_iter_get_error(iter, &err);
+    CheckNoError(err);
+    rocksdb_iter_destroy(iter);
+
     rocksdb_drop_column_family(db, handles[1], &err);
     CheckNoError(err);
     for (i = 0; i < 2; i++) {

db/column_family.cc

@@ -147,13 +147,10 @@ ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
                                     const ColumnFamilyOptions& src) {
   ColumnFamilyOptions result = src;
   result.comparator = icmp;
-#ifdef OS_MACOSX
-  // TODO(icanadi) make write_buffer_size uint64_t instead of size_t
-  ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, ((size_t)1) << 30);
-#else
-  ClipToRange(&result.write_buffer_size,
-              ((size_t)64) << 10, ((size_t)64) << 30);
-#endif
+  size_t clamp_max = std::conditional<
+      sizeof(size_t) == 4, std::integral_constant<size_t, 0xffffffff>,
+      std::integral_constant<uint64_t, 64ull << 30>>::type::value;
+  ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, clamp_max);
   // if user sets arena_block_size, we trust user to use this value. Otherwise,
   // calculate a proper value from writer_buffer_size;
   if (result.arena_block_size <= 0) {

db/column_family.h

@@ -209,7 +209,7 @@ class ColumnFamilyData {
   const ImmutableCFOptions* ioptions() const { return &ioptions_; }
   // REQUIRES: DB mutex held
   // This returns the MutableCFOptions used by current SuperVersion
-  // You shoul use this API to reference MutableCFOptions most of the time.
+  // You should use this API to reference MutableCFOptions most of the time.
   const MutableCFOptions* GetCurrentMutableCFOptions() const {
     return &(super_version_->mutable_cf_options);
   }
@@ -465,6 +465,8 @@ class ColumnFamilySet {
   // Don't call while iterating over ColumnFamilySet
   void FreeDeadColumnFamilies();
 
+  Cache* get_table_cache() { return table_cache_; }
+
  private:
   friend class ColumnFamilyData;
   // helper function that gets called from cfd destructor

db/column_family_test.cc

@@ -13,6 +13,7 @@
 #include <thread>
 
 #include "db/db_impl.h"
+#include "db/db_test_util.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -26,6 +27,8 @@
 
 namespace rocksdb {
 
+static const int kValueSize = 1000;
+
 namespace {
 std::string RandomString(Random* rnd, int len) {
   std::string r;
@@ -70,6 +73,77 @@ class ColumnFamilyTest : public testing::Test {
     delete env_;
   }
 
+  // Return the value to associate with the specified key
+  Slice Value(int k, std::string* storage) {
+    if (k == 0) {
+      // Ugh.  Random seed of 0 used to produce no entropy.  This code
+      // preserves the implementation that was in place when all of the
+      // magic values in this file were picked.
+      *storage = std::string(kValueSize, ' ');
+      return Slice(*storage);
+    } else {
+      Random r(k);
+      return test::RandomString(&r, kValueSize, storage);
+    }
+  }
+
+  void Build(int base, int n, int flush_every = 0) {
+    std::string key_space, value_space;
+    WriteBatch batch;
+
+    for (int i = 0; i < n; i++) {
+      if (flush_every != 0 && i != 0 && i % flush_every == 0) {
+        DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+        dbi->TEST_FlushMemTable();
+      }
+
+      int keyi = base + i;
+      Slice key(DBTestBase::Key(keyi));
+
+      batch.Clear();
+      batch.Put(handles_[0], key, Value(keyi, &value_space));
+      batch.Put(handles_[1], key, Value(keyi, &value_space));
+      batch.Put(handles_[2], key, Value(keyi, &value_space));
+      ASSERT_OK(db_->Write(WriteOptions(), &batch));
+    }
+  }
+
+  void CheckMissed() {
+    uint64_t next_expected = 0;
+    uint64_t missed = 0;
+    int bad_keys = 0;
+    int bad_values = 0;
+    int correct = 0;
+    std::string value_space;
+    for (int cf = 0; cf < 3; cf++) {
+      next_expected = 0;
+      Iterator* iter = db_->NewIterator(ReadOptions(false, true), handles_[cf]);
+      for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+        uint64_t key;
+        Slice in(iter->key());
+        in.remove_prefix(3);
+        if (!ConsumeDecimalNumber(&in, &key) || !in.empty() ||
+            key < next_expected) {
+          bad_keys++;
+          continue;
+        }
+        missed += (key - next_expected);
+        next_expected = key + 1;
+        if (iter->value() != Value(static_cast<int>(key), &value_space)) {
+          bad_values++;
+        } else {
+          correct++;
+        }
+      }
+      delete iter;
+    }
+
+    ASSERT_EQ(0, bad_keys);
+    ASSERT_EQ(0, bad_values);
+    ASSERT_EQ(0, missed);
+    (void)correct;
+  }
+
   void Close() {
     for (auto h : handles_) {
       if (h) {
@@ -447,7 +521,7 @@ TEST_F(ColumnFamilyTest, AddDrop) {
 
   std::vector<std::string> families;
   ASSERT_OK(DB::ListColumnFamilies(db_options_, dbname_, &families));
-  sort(families.begin(), families.end());
+  std::sort(families.begin(), families.end());
   ASSERT_TRUE(families ==
               std::vector<std::string>({"default", "four", "three"}));
 }
@@ -1042,6 +1116,7 @@ TEST_F(ColumnFamilyTest, AutomaticAndManualCompactions) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1134,6 +1209,7 @@ TEST_F(ColumnFamilyTest, ManualAndAutomaticCompactions) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1229,6 +1305,7 @@ TEST_F(ColumnFamilyTest, SameCFManualManualCompactions) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1327,6 +1404,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactions) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1416,6 +1494,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticCompactionsLevel) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1512,6 +1591,7 @@ TEST_F(ColumnFamilyTest, SameCFManualAutomaticConflict) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -1631,6 +1711,7 @@ TEST_F(ColumnFamilyTest, SameCFAutomaticManualCompactions) {
   db_options_.max_open_files = 20;  // only 10 files in file cache
   db_options_.disableDataSync = true;
   db_options_.max_background_compactions = 3;
+  db_options_.base_background_compactions = 3;
 
   default_cf.compaction_style = kCompactionStyleLevel;
   default_cf.num_levels = 3;
@@ -2557,6 +2638,9 @@ TEST_F(ColumnFamilyTest, CompactionSpeedupTwoColumnFamilies) {
   ASSERT_EQ(2, dbfull()->BGCompactionsAllowed());
 }
 
+// Disable on windows because SyncWAL requires env->IsSyncThreadSafe()
+// to return true which is not so in unbuffered mode.
+#ifndef OS_WIN
 TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
@@ -2586,6 +2670,81 @@ TEST_F(ColumnFamilyTest, LogSyncConflictFlush) {
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
   Close();
 }
+#endif
+
+// this test is placed here, because the infrastructure for Column Family
+// test is being used to ensure a roll of wal files.
+// Basic idea is to test that WAL truncation is being detected and not
+// ignored
+TEST_F(ColumnFamilyTest, DISABLED_LogTruncationTest) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two"});
+
+  Build(0, 100);
+
+  // Flush the 0th column family to force a roll of the wal log
+  Flush(0);
+
+  // Add some more entries
+  Build(100, 100);
+
+  std::vector<std::string> filenames;
+  ASSERT_OK(env_->GetChildren(dbname_, &filenames));
+
+  // collect wal files
+  std::vector<std::string> logfs;
+  for (size_t i = 0; i < filenames.size(); i++) {
+    uint64_t number;
+    FileType type;
+    if (!(ParseFileName(filenames[i], &number, &type))) continue;
+
+    if (type != kLogFile) continue;
+
+    logfs.push_back(filenames[i]);
+  }
+
+  std::sort(logfs.begin(), logfs.end());
+  ASSERT_GE(logfs.size(), 2);
+
+  // Take the last but one file, and truncate it
+  std::string fpath = dbname_ + "/" + logfs[logfs.size() - 2];
+  std::vector<std::string> names_save = names_;
+
+  uint64_t fsize;
+  ASSERT_OK(env_->GetFileSize(fpath, &fsize));
+  ASSERT_GT(fsize, 0);
+
+  Close();
+
+  std::string backup_logs = dbname_ + "/backup_logs";
+  std::string t_fpath = backup_logs + "/" + logfs[logfs.size() - 2];
+
+  ASSERT_OK(env_->CreateDirIfMissing(backup_logs));
+  // Not sure how easy it is to make this data driven.
+  // need to read back the WAL file and truncate last 10
+  // entries
+  CopyFile(fpath, t_fpath, fsize - 9180);
+
+  ASSERT_OK(env_->DeleteFile(fpath));
+  ASSERT_OK(env_->RenameFile(t_fpath, fpath));
+
+  db_options_.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
+
+  OpenReadOnly(names_save);
+
+  CheckMissed();
+
+  Close();
+
+  Open(names_save);
+
+  CheckMissed();
+
+  Close();
+
+  // cleanup
+  env_->DeleteDir(backup_logs);
+}
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/compacted_db_impl.cc

@@ -93,7 +93,7 @@ Status CompactedDBImpl::Init(const Options& options) {
   mutex_.Lock();
   ColumnFamilyDescriptor cf(kDefaultColumnFamilyName,
                             ColumnFamilyOptions(options));
-  Status s = Recover({ cf }, true /* read only */, false);
+  Status s = Recover({cf}, true /* read only */, false, true);
   if (s.ok()) {
     cfd_ = reinterpret_cast<ColumnFamilyHandleImpl*>(
               DefaultColumnFamily())->cfd();

db/compaction.cc

@@ -16,8 +16,8 @@
 #include <inttypes.h>
 #include <vector>
 
-#include "rocksdb/compaction_filter.h"
 #include "db/column_family.h"
+#include "rocksdb/compaction_filter.h"
 #include "util/logging.h"
 #include "util/sync_point.h"
 
@@ -162,9 +162,6 @@ Compaction::Compaction(VersionStorageInfo* vstorage,
       deletion_compaction_(_deletion_compaction),
       inputs_(std::move(_inputs)),
       grandparents_(std::move(_grandparents)),
-      grandparent_index_(0),
-      seen_key_(false),
-      overlapped_bytes_(0),
       score_(_score),
       bottommost_level_(IsBottommostLevel(output_level_, vstorage, inputs_)),
       is_full_compaction_(IsFullCompaction(vstorage, inputs_)),
@@ -206,9 +203,11 @@ Compaction::~Compaction() {
 }
 
 bool Compaction::InputCompressionMatchesOutput() const {
-  int base_level = input_version_->storage_info()->base_level();
-  bool matches = (GetCompressionType(*cfd_->ioptions(), start_level_,
-                                     base_level) == output_compression_);
+  VersionStorageInfo* vstorage = input_version_->storage_info();
+  int base_level = vstorage->base_level();
+  bool matches =
+      (GetCompressionType(*cfd_->ioptions(), vstorage, mutable_cf_options_,
+                          start_level_, base_level) == output_compression_);
   if (matches) {
     TEST_SYNC_POINT("Compaction::InputCompressionMatchesOutput:Matches");
     return true;
@@ -290,38 +289,12 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
   return true;
 }
 
-bool Compaction::ShouldStopBefore(const Slice& internal_key) {
-  // Scan to find earliest grandparent file that contains key.
-  const InternalKeyComparator* icmp = &cfd_->internal_comparator();
-  while (grandparent_index_ < grandparents_.size() &&
-      icmp->Compare(internal_key,
-                    grandparents_[grandparent_index_]->largest.Encode()) > 0) {
-    if (seen_key_) {
-      overlapped_bytes_ += grandparents_[grandparent_index_]->fd.GetFileSize();
-    }
-    assert(grandparent_index_ + 1 >= grandparents_.size() ||
-           icmp->Compare(grandparents_[grandparent_index_]->largest.Encode(),
-                         grandparents_[grandparent_index_+1]->smallest.Encode())
-                         < 0);
-    grandparent_index_++;
-  }
-  seen_key_ = true;
-
-  if (overlapped_bytes_ > max_grandparent_overlap_bytes_) {
-    // Too much overlap for current output; start new output
-    overlapped_bytes_ = 0;
-    return true;
-  } else {
-    return false;
-  }
-}
-
 // Mark (or clear) each file that is being compacted
 void Compaction::MarkFilesBeingCompacted(bool mark_as_compacted) {
   for (size_t i = 0; i < num_input_levels(); i++) {
     for (size_t j = 0; j < inputs_[i].size(); j++) {
-      assert(mark_as_compacted ? !inputs_[i][j]->being_compacted :
-                                  inputs_[i][j]->being_compacted);
+      assert(mark_as_compacted ? !inputs_[i][j]->being_compacted
+                               : inputs_[i][j]->being_compacted);
       inputs_[i][j]->being_compacted = mark_as_compacted;
     }
   }
@@ -396,10 +369,8 @@ int InputSummary(const std::vector<FileMetaData*>& files, char* output,
 
 void Compaction::Summary(char* output, int len) {
   int write =
-      snprintf(output, len, "Base version %" PRIu64
-                            " Base level %d, inputs: [",
-               input_version_->GetVersionNumber(),
-               start_level_);
+      snprintf(output, len, "Base version %" PRIu64 " Base level %d, inputs: [",
+               input_version_->GetVersionNumber(), start_level_);
   if (write < 0 || write >= len) {
     return;
   }
@@ -421,7 +392,7 @@ void Compaction::Summary(char* output, int len) {
   snprintf(output + write, len - write, "]");
 }
 
-uint64_t Compaction::OutputFilePreallocationSize() {
+uint64_t Compaction::OutputFilePreallocationSize() const {
   uint64_t preallocation_size = 0;
 
   if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel ||

db/compaction.h

@@ -104,7 +104,7 @@ class Compaction {
   }
 
   // Returns the LevelFilesBrief of the specified compaction input level.
-  LevelFilesBrief* input_levels(size_t compaction_input_level) {
+  const LevelFilesBrief* input_levels(size_t compaction_input_level) const {
     return &input_levels_[compaction_input_level];
   }
 
@@ -132,10 +132,6 @@ class Compaction {
   bool KeyNotExistsBeyondOutputLevel(const Slice& user_key,
                                      std::vector<size_t>* level_ptrs) const;
 
-  // Returns true iff we should stop building the current output
-  // before processing "internal_key".
-  bool ShouldStopBefore(const Slice& internal_key);
-
   // Clear all files to indicate that they are not being compacted
   // Delete this compaction from the list of running compactions.
   //
@@ -151,13 +147,13 @@ class Compaction {
   double score() const { return score_; }
 
   // Is this compaction creating a file in the bottom most level?
-  bool bottommost_level() { return bottommost_level_; }
+  bool bottommost_level() const { return bottommost_level_; }
 
   // Does this compaction include all sst files?
-  bool is_full_compaction() { return is_full_compaction_; }
+  bool is_full_compaction() const { return is_full_compaction_; }
 
   // Was this compaction triggered manually by the client?
-  bool is_manual_compaction() { return is_manual_compaction_; }
+  bool is_manual_compaction() const { return is_manual_compaction_; }
 
   // Used when allow_trivial_move option is set in
   // Universal compaction. If all the input files are
@@ -170,19 +166,21 @@ class Compaction {
   // Used when allow_trivial_move option is set in
   // Universal compaction. Returns true, if the input files
   // are non-overlapping and can be trivially moved.
-  bool is_trivial_move() { return is_trivial_move_; }
+  bool is_trivial_move() const { return is_trivial_move_; }
 
   // How many total levels are there?
   int number_levels() const { return number_levels_; }
 
   // Return the MutableCFOptions that should be used throughout the compaction
   // procedure
-  const MutableCFOptions* mutable_cf_options() { return &mutable_cf_options_; }
+  const MutableCFOptions* mutable_cf_options() const {
+    return &mutable_cf_options_;
+  }
 
   // Returns the size in bytes that the output file should be preallocated to.
   // In level compaction, that is max_file_size_. In universal compaction, that
   // is the sum of all input file sizes.
-  uint64_t OutputFilePreallocationSize();
+  uint64_t OutputFilePreallocationSize() const;
 
   void SetInputVersion(Version* input_version);
 
@@ -225,6 +223,14 @@ class Compaction {
 
   CompactionReason compaction_reason() { return compaction_reason_; }
 
+  const std::vector<FileMetaData*>& grandparents() const {
+    return grandparents_;
+  }
+
+  uint64_t max_grandparent_overlap_bytes() const {
+    return max_grandparent_overlap_bytes_;
+  }
+
  private:
   // mark (or clear) all files that are being compacted
   void MarkFilesBeingCompacted(bool mark_as_compacted);
@@ -265,13 +271,9 @@ class Compaction {
   // A copy of inputs_, organized more closely in memory
   autovector<LevelFilesBrief, 2> input_levels_;
 
-  // State used to check for number of of overlapping grandparent files
+  // State used to check for number of overlapping grandparent files
   // (grandparent == "output_level_ + 1")
   std::vector<FileMetaData*> grandparents_;
-  size_t grandparent_index_;   // Index in grandparent_starts_
-  bool seen_key_;              // Some output key has been seen
-  uint64_t overlapped_bytes_;  // Bytes of overlap between current output
-                               // and grandparent files
   const double score_;         // score that was used to pick this compaction.
 
   // Is this compaction creating a file in the bottom most level?

db/compaction_iterator.cc

@@ -14,7 +14,7 @@ CompactionIterator::CompactionIterator(
     InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
     SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
     SequenceNumber earliest_write_conflict_snapshot, Env* env,
-    bool expect_valid_internal_key, Compaction* compaction,
+    bool expect_valid_internal_key, const Compaction* compaction,
     const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
     : input_(input),
       cmp_(cmp),
@@ -423,15 +423,15 @@ void CompactionIterator::PrepareOutput() {
 inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
     SequenceNumber in, SequenceNumber* prev_snapshot) {
   assert(snapshots_->size());
-  SequenceNumber prev __attribute__((unused)) = 0;
+  SequenceNumber prev __attribute__((__unused__)) = kMaxSequenceNumber;
   for (const auto cur : *snapshots_) {
-    assert(prev <= cur);
+    assert(prev == kMaxSequenceNumber || prev <= cur);
     if (cur >= in) {
-      *prev_snapshot = prev;
+      *prev_snapshot = prev == kMaxSequenceNumber ? 0 : prev;
       return cur;
     }
     prev = cur;
-    assert(prev);
+    assert(prev < kMaxSequenceNumber);
   }
   *prev_snapshot = prev;
   return kMaxSequenceNumber;

db/compaction_iterator.h

@@ -42,7 +42,7 @@ class CompactionIterator {
                      std::vector<SequenceNumber>* snapshots,
                      SequenceNumber earliest_write_conflict_snapshot, Env* env,
                      bool expect_valid_internal_key,
-                     Compaction* compaction = nullptr,
+                     const Compaction* compaction = nullptr,
                      const CompactionFilter* compaction_filter = nullptr,
                      LogBuffer* log_buffer = nullptr);
 
@@ -92,7 +92,7 @@ class CompactionIterator {
   const SequenceNumber earliest_write_conflict_snapshot_;
   Env* env_;
   bool expect_valid_internal_key_;
-  Compaction* compaction_;
+  const Compaction* compaction_;
   const CompactionFilter* compaction_filter_;
   LogBuffer* log_buffer_;
   bool bottommost_level_;

db/compaction_job.cc

@@ -16,12 +16,13 @@
 #include <inttypes.h>
 #include <algorithm>
 #include <functional>
-#include <vector>
-#include <memory>
 #include <list>
+#include <memory>
+#include <random>
 #include <set>
 #include <thread>
 #include <utility>
+#include <vector>
 
 #include "db/builder.h"
 #include "db/db_iter.h"
@@ -63,7 +64,7 @@ namespace rocksdb {
 
 // Maintains state for each sub-compaction
 struct CompactionJob::SubcompactionState {
-  Compaction* compaction;
+  const Compaction* compaction;
   std::unique_ptr<CompactionIterator> c_iter;
 
   // The boundaries of the key-range this compaction is interested in. No two
@@ -104,6 +105,14 @@ struct CompactionJob::SubcompactionState {
   uint64_t num_output_records;
   CompactionJobStats compaction_job_stats;
   uint64_t approx_size;
+  // An index that used to speed up ShouldStopBefore().
+  size_t grandparent_index = 0;
+  // The number of bytes overlapping between the current output and
+  // grandparent files used in ShouldStopBefore().
+  uint64_t overlapped_bytes = 0;
+  // A flag determine whether the key has been seen in ShouldStopBefore()
+  bool seen_key = false;
+  std::string compression_dict;
 
   SubcompactionState(Compaction* c, Slice* _start, Slice* _end,
                      uint64_t size = 0)
@@ -115,7 +124,11 @@ struct CompactionJob::SubcompactionState {
         total_bytes(0),
         num_input_records(0),
         num_output_records(0),
-        approx_size(size) {
+        approx_size(size),
+        grandparent_index(0),
+        overlapped_bytes(0),
+        seen_key(false),
+        compression_dict() {
     assert(compaction != nullptr);
   }
 
@@ -134,6 +147,10 @@ struct CompactionJob::SubcompactionState {
     num_output_records = std::move(o.num_output_records);
     compaction_job_stats = std::move(o.compaction_job_stats);
     approx_size = std::move(o.approx_size);
+    grandparent_index = std::move(o.grandparent_index);
+    overlapped_bytes = std::move(o.overlapped_bytes);
+    seen_key = std::move(o.seen_key);
+    compression_dict = std::move(o.compression_dict);
     return *this;
   }
 
@@ -141,6 +158,38 @@ struct CompactionJob::SubcompactionState {
   SubcompactionState(const SubcompactionState&) = delete;
 
   SubcompactionState& operator=(const SubcompactionState&) = delete;
+
+  // Returns true iff we should stop building the current output
+  // before processing "internal_key".
+  bool ShouldStopBefore(const Slice& internal_key) {
+    const InternalKeyComparator* icmp =
+        &compaction->column_family_data()->internal_comparator();
+    const std::vector<FileMetaData*>& grandparents = compaction->grandparents();
+
+    // Scan to find earliest grandparent file that contains key.
+    while (grandparent_index < grandparents.size() &&
+           icmp->Compare(internal_key,
+                         grandparents[grandparent_index]->largest.Encode()) >
+               0) {
+      if (seen_key) {
+        overlapped_bytes += grandparents[grandparent_index]->fd.GetFileSize();
+      }
+      assert(grandparent_index + 1 >= grandparents.size() ||
+             icmp->Compare(
+                 grandparents[grandparent_index]->largest.Encode(),
+                 grandparents[grandparent_index + 1]->smallest.Encode()) <= 0);
+      grandparent_index++;
+    }
+    seen_key = true;
+
+    if (overlapped_bytes > compaction->max_grandparent_overlap_bytes()) {
+      // Too much overlap for current output; start new output
+      overlapped_bytes = 0;
+      return true;
+    }
+
+    return false;
+  }
 };
 
 // Maintains state for the entire compaction
@@ -422,9 +471,9 @@ void CompactionJob::GenSubcompactionBoundaries() {
 
   // Group the ranges into subcompactions
   const double min_file_fill_percent = 4.0 / 5;
-  uint64_t max_output_files = static_cast<uint64_t>(std::ceil(
-      sum / min_file_fill_percent /
-      cfd->GetCurrentMutableCFOptions()->MaxFileSizeForLevel(out_lvl)));
+  uint64_t max_output_files = static_cast<uint64_t>(
+      std::ceil(sum / min_file_fill_percent /
+                c->mutable_cf_options()->MaxFileSizeForLevel(out_lvl)));
   uint64_t subcompactions =
       std::min({static_cast<uint64_t>(ranges.size()),
                 static_cast<uint64_t>(db_options_.max_subcompactions),
@@ -620,6 +669,31 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   }
 
   ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  const MutableCFOptions* mutable_cf_options =
+      sub_compact->compaction->mutable_cf_options();
+
+  // To build compression dictionary, we sample the first output file, assuming
+  // it'll reach the maximum length, and then use the dictionary for compressing
+  // subsequent output files. The dictionary may be less than max_dict_bytes if
+  // the first output file's length is less than the maximum.
+  const int kSampleLenShift = 6;  // 2^6 = 64-byte samples
+  std::set<size_t> sample_begin_offsets;
+  if (bottommost_level_ &&
+      cfd->ioptions()->compression_opts.max_dict_bytes > 0) {
+    const size_t kMaxSamples =
+        cfd->ioptions()->compression_opts.max_dict_bytes >> kSampleLenShift;
+    const size_t kOutFileLen = mutable_cf_options->MaxFileSizeForLevel(
+        compact_->compaction->output_level());
+    if (kOutFileLen != port::kMaxSizet) {
+      const size_t kOutFileNumSamples = kOutFileLen >> kSampleLenShift;
+      Random64 generator{versions_->NewFileNumber()};
+      for (size_t i = 0; i < kMaxSamples; ++i) {
+        sample_begin_offsets.insert(generator.Uniform(kOutFileNumSamples)
+                                    << kSampleLenShift);
+      }
+    }
+  }
+
   auto compaction_filter = cfd->ioptions()->compaction_filter;
   std::unique_ptr<CompactionFilter> compaction_filter_from_factory = nullptr;
   if (compaction_filter == nullptr) {
@@ -630,7 +704,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   MergeHelper merge(
       env_, cfd->user_comparator(), cfd->ioptions()->merge_operator,
       compaction_filter, db_options_.info_log.get(),
-      cfd->ioptions()->min_partial_merge_operands,
+      mutable_cf_options->min_partial_merge_operands,
       false /* internal key corruption is expected */,
       existing_snapshots_.empty() ? 0 : existing_snapshots_.back(),
       compact_->compaction->level(), db_options_.statistics.get());
@@ -655,6 +729,13 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   auto c_iter = sub_compact->c_iter.get();
   c_iter->SeekToFirst();
   const auto& c_iter_stats = c_iter->iter_stats();
+  auto sample_begin_offset_iter = sample_begin_offsets.cbegin();
+  // data_begin_offset and compression_dict are only valid while generating
+  // dictionary from the first output file.
+  size_t data_begin_offset = 0;
+  std::string compression_dict;
+  compression_dict.reserve(cfd->ioptions()->compression_opts.max_dict_bytes);
+
   // TODO(noetzli): check whether we could check !shutting_down_->... only
   // only occasionally (see diff D42687)
   while (status.ok() && !shutting_down_->load(std::memory_order_acquire) &&
@@ -669,7 +750,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
     if (end != nullptr &&
         cfd->user_comparator()->Compare(c_iter->user_key(), *end) >= 0) {
       break;
-    } else if (sub_compact->compaction->ShouldStopBefore(key) &&
+    } else if (sub_compact->ShouldStopBefore(key) &&
                sub_compact->builder != nullptr) {
       status = FinishCompactionOutputFile(input->status(), sub_compact);
       if (!status.ok()) {
@@ -698,6 +779,55 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
         key, c_iter->ikey().sequence);
     sub_compact->num_output_records++;
 
+    if (sub_compact->outputs.size() == 1) {  // first output file
+      // Check if this key/value overlaps any sample intervals; if so, appends
+      // overlapping portions to the dictionary.
+      for (const auto& data_elmt : {key, value}) {
+        size_t data_end_offset = data_begin_offset + data_elmt.size();
+        while (sample_begin_offset_iter != sample_begin_offsets.cend() &&
+               *sample_begin_offset_iter < data_end_offset) {
+          size_t sample_end_offset =
+              *sample_begin_offset_iter + (1 << kSampleLenShift);
+          // Invariant: Because we advance sample iterator while processing the
+          // data_elmt containing the sample's last byte, the current sample
+          // cannot end before the current data_elmt.
+          assert(data_begin_offset < sample_end_offset);
+
+          size_t data_elmt_copy_offset, data_elmt_copy_len;
+          if (*sample_begin_offset_iter <= data_begin_offset) {
+            // The sample starts before data_elmt starts, so take bytes starting
+            // at the beginning of data_elmt.
+            data_elmt_copy_offset = 0;
+          } else {
+            // data_elmt starts before the sample starts, so take bytes starting
+            // at the below offset into data_elmt.
+            data_elmt_copy_offset =
+                *sample_begin_offset_iter - data_begin_offset;
+          }
+          if (sample_end_offset <= data_end_offset) {
+            // The sample ends before data_elmt ends, so take as many bytes as
+            // needed.
+            data_elmt_copy_len =
+                sample_end_offset - (data_begin_offset + data_elmt_copy_offset);
+          } else {
+            // data_elmt ends before the sample ends, so take all remaining
+            // bytes in data_elmt.
+            data_elmt_copy_len =
+                data_end_offset - (data_begin_offset + data_elmt_copy_offset);
+          }
+          compression_dict.append(&data_elmt.data()[data_elmt_copy_offset],
+                                  data_elmt_copy_len);
+          if (sample_end_offset > data_end_offset) {
+            // Didn't finish sample. Try to finish it with the next data_elmt.
+            break;
+          }
+          // Next sample may require bytes from same data_elmt.
+          sample_begin_offset_iter++;
+        }
+        data_begin_offset = data_end_offset;
+      }
+    }
+
     // Close output file if it is big enough
     // TODO(aekmekji): determine if file should be closed earlier than this
     // during subcompactions (i.e. if output size, estimated by input size, is
@@ -706,8 +836,12 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
     if (sub_compact->builder->FileSize() >=
         sub_compact->compaction->max_output_file_size()) {
       status = FinishCompactionOutputFile(input->status(), sub_compact);
+      if (sub_compact->outputs.size() == 1) {
+        // Use dictionary from first output file for compression of subsequent
+        // files.
+        sub_compact->compression_dict = std::move(compression_dict);
+      }
     }
-
     c_iter->Next();
   }
 
@@ -820,9 +954,10 @@ Status CompactionJob::FinishCompactionOutputFile(
   }
   sub_compact->outfile.reset();
 
+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+  TableProperties tp;
   if (s.ok() && current_entries > 0) {
     // Verify that the table is usable
-    ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
     InternalIterator* iter = cfd->table_cache()->NewIterator(
         ReadOptions(), env_options_, cfd->internal_comparator(), meta->fd,
         nullptr, cfd->internal_stats()->GetFileReadHist(
@@ -836,34 +971,30 @@ Status CompactionJob::FinishCompactionOutputFile(
     }
 
     delete iter;
+
+    // Output to event logger and fire events.
     if (s.ok()) {
-      auto tp = sub_compact->builder->GetTableProperties();
+      tp = sub_compact->builder->GetTableProperties();
       sub_compact->current_output()->table_properties =
           std::make_shared<TableProperties>(tp);
-      TableFileCreationInfo info(std::move(tp));
-      info.db_name = dbname_;
-      info.cf_name = cfd->GetName();
-      info.file_path =
-          TableFileName(cfd->ioptions()->db_paths, meta->fd.GetNumber(),
-                        meta->fd.GetPathId());
-      info.file_size = meta->fd.GetFileSize();
-      info.job_id = job_id_;
       Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
           "[%s] [JOB %d] Generated table #%" PRIu64 ": %" PRIu64
           " keys, %" PRIu64 " bytes%s",
           cfd->GetName().c_str(), job_id_, output_number, current_entries,
           current_bytes,
           meta->marked_for_compaction ? " (need compaction)" : "");
-      EventHelpers::LogAndNotifyTableFileCreation(
-          event_logger_, cfd->ioptions()->listeners, meta->fd, info);
     }
   }
+  std::string fname = TableFileName(db_options_.db_paths, meta->fd.GetNumber(),
+                                    meta->fd.GetPathId());
+  EventHelpers::LogAndNotifyTableFileCreationFinished(
+      event_logger_, cfd->ioptions()->listeners, dbname_, cfd->GetName(), fname,
+      job_id_, meta->fd, tp, TableFileCreationReason::kCompaction, s);
 
   // Report new file to SstFileManagerImpl
   auto sfm =
       static_cast<SstFileManagerImpl*>(db_options_.sst_file_manager.get());
   if (sfm && meta->fd.GetPathId() == 0) {
-    ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
     auto fn = TableFileName(cfd->ioptions()->db_paths, meta->fd.GetNumber(),
                             meta->fd.GetPathId());
     sfm->OnAddFile(fn);
@@ -939,10 +1070,17 @@ Status CompactionJob::OpenCompactionOutputFile(
   assert(sub_compact->builder == nullptr);
   // no need to lock because VersionSet::next_file_number_ is atomic
   uint64_t file_number = versions_->NewFileNumber();
-  // Make the output file
-  unique_ptr<WritableFile> writable_file;
   std::string fname = TableFileName(db_options_.db_paths, file_number,
                                     sub_compact->compaction->output_path_id());
+  // Fire events.
+  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
+#ifndef ROCKSDB_LITE
+  EventHelpers::NotifyTableFileCreationStarted(
+      cfd->ioptions()->listeners, dbname_, cfd->GetName(), fname, job_id_,
+      TableFileCreationReason::kCompaction);
+#endif  // !ROCKSDB_LITE
+  // Make the output file
+  unique_ptr<WritableFile> writable_file;
   Status s = NewWritableFile(env_, fname, &writable_file, env_options_);
   if (!s.ok()) {
     Log(InfoLogLevel::ERROR_LEVEL, db_options_.info_log,
@@ -951,8 +1089,13 @@ Status CompactionJob::OpenCompactionOutputFile(
         sub_compact->compaction->column_family_data()->GetName().c_str(),
         job_id_, file_number, s.ToString().c_str());
     LogFlush(db_options_.info_log);
+    EventHelpers::LogAndNotifyTableFileCreationFinished(
+        event_logger_, cfd->ioptions()->listeners, dbname_, cfd->GetName(),
+        fname, job_id_, FileDescriptor(), TableProperties(),
+        TableFileCreationReason::kCompaction, s);
     return s;
   }
+
   SubcompactionState::Output out;
   out.meta.fd =
       FileDescriptor(file_number, sub_compact->compaction->output_path_id(), 0);
@@ -965,7 +1108,6 @@ Status CompactionJob::OpenCompactionOutputFile(
   sub_compact->outfile.reset(
       new WritableFileWriter(std::move(writable_file), env_options_));
 
-  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
   // If the Column family flag is to only optimize filters for hits,
   // we can skip creating filters if this is the bottommost_level where
   // data is going to be found
@@ -973,9 +1115,10 @@ Status CompactionJob::OpenCompactionOutputFile(
       cfd->ioptions()->optimize_filters_for_hits && bottommost_level_;
   sub_compact->builder.reset(NewTableBuilder(
       *cfd->ioptions(), cfd->internal_comparator(),
-      cfd->int_tbl_prop_collector_factories(), cfd->GetID(),
+      cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
       sub_compact->outfile.get(), sub_compact->compaction->output_compression(),
-      cfd->ioptions()->compression_opts, skip_filters));
+      cfd->ioptions()->compression_opts, &sub_compact->compression_dict,
+      skip_filters));
   LogFlush(db_options_.info_log);
   return s;
 }

db/compaction_job.h

@@ -122,6 +122,7 @@ class CompactionJob {
   const std::string& dbname_;
   const DBOptions& db_options_;
   const EnvOptions& env_options_;
+
   Env* env_;
   VersionSet* versions_;
   std::atomic<bool>* shutting_down_;

db/compaction_job_stats_test.cc

@@ -624,7 +624,10 @@ CompressionType GetAnyCompression() {
     return kBZip2Compression;
   } else if (LZ4_Supported()) {
     return kLZ4Compression;
+  } else if (XPRESS_Supported()) {
+    return kXpressCompression;
   }
+
   return kNoCompression;
 }
 
@@ -660,7 +663,7 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
   options.max_subcompactions = max_subcompactions_;
   options.bytes_per_sync = 512 * 1024;
 
-  options.compaction_measure_io_stats = true;
+  options.report_bg_io_stats = true;
   for (int test = 0; test < 2; ++test) {
     DestroyAndReopen(options);
     CreateAndReopenWithCF({"pikachu"}, options);

db/compaction_picker.cc

@@ -70,7 +70,8 @@ struct UserKeyComparator {
 };
 
 typedef std::priority_queue<InputFileInfo, std::vector<InputFileInfo>,
-                            UserKeyComparator> SmallestKeyHeap;
+                            UserKeyComparator>
+    SmallestKeyHeap;
 
 // This function creates the heap that is used to find if the files are
 // overlapping during universal compaction when the allow_trivial_move
@@ -109,13 +110,22 @@ SmallestKeyHeap create_level_heap(Compaction* c, const Comparator* ucmp) {
 // matter what the values of the other two parameters are.
 // Otherwise, the compression type is determined based on options and level.
 CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
+                                   const VersionStorageInfo* vstorage,
+                                   const MutableCFOptions& mutable_cf_options,
                                    int level, int base_level,
                                    const bool enable_compression) {
   if (!enable_compression) {
     // disable compression
     return kNoCompression;
   }
-  // If the use has specified a different compression level for each level,
+
+  // If bottommost_compression is set and we are compacting to the
+  // bottommost level then we should use it.
+  if (ioptions.bottommost_compression != kDisableCompressionOption &&
+      level > base_level && level >= (vstorage->num_non_empty_levels() - 1)) {
+    return ioptions.bottommost_compression;
+  }
+  // If the user has specified a different compression level for each level,
   // then pick the compression for that level.
   if (!ioptions.compression_per_level.empty()) {
     assert(level == 0 || level >= base_level);
@@ -129,7 +139,7 @@ CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
     // specified compression levels, use the last value.
     return ioptions.compression_per_level[std::max(0, std::min(idx, n))];
   } else {
-    return ioptions.compression;
+    return mutable_cf_options.compression;
   }
 }
 
@@ -190,10 +200,9 @@ void CompactionPicker::GetRange(const CompactionInputFiles& inputs1,
     InternalKey smallest1, smallest2, largest1, largest2;
     GetRange(inputs1, &smallest1, &largest1);
     GetRange(inputs2, &smallest2, &largest2);
-    *smallest = icmp_->Compare(smallest1, smallest2) < 0 ?
-                smallest1 : smallest2;
-    *largest = icmp_->Compare(largest1, largest2) < 0 ?
-               largest2 : largest1;
+    *smallest =
+        icmp_->Compare(smallest1, smallest2) < 0 ? smallest1 : smallest2;
+    *largest = icmp_->Compare(largest1, largest2) < 0 ? largest2 : largest1;
   }
 }
 
@@ -258,9 +267,9 @@ Compaction* CompactionPicker::FormCompaction(
     VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options,
     uint32_t output_path_id) {
   uint64_t max_grandparent_overlap_bytes =
-      output_level + 1 < vstorage->num_levels() ?
-          mutable_cf_options.MaxGrandParentOverlapBytes(output_level + 1) :
-          std::numeric_limits<uint64_t>::max();
+      output_level + 1 < vstorage->num_levels()
+          ? mutable_cf_options.MaxGrandParentOverlapBytes(output_level + 1)
+          : std::numeric_limits<uint64_t>::max();
   assert(input_files.size());
 
   // TODO(rven ): we might be able to run concurrent level 0 compaction
@@ -284,8 +293,7 @@ Compaction* CompactionPicker::FormCompaction(
 
 Status CompactionPicker::GetCompactionInputsFromFileNumbers(
     std::vector<CompactionInputFiles>* input_files,
-    std::unordered_set<uint64_t>* input_set,
-    const VersionStorageInfo* vstorage,
+    std::unordered_set<uint64_t>* input_set, const VersionStorageInfo* vstorage,
     const CompactionOptions& compact_options) const {
   if (input_set->size() == 0U) {
     return Status::InvalidArgument(
@@ -323,8 +331,8 @@ Status CompactionPicker::GetCompactionInputsFromFileNumbers(
     return Status::InvalidArgument(message);
   }
 
-  for (int level = first_non_empty_level;
-       level <= last_non_empty_level; ++level) {
+  for (int level = first_non_empty_level; level <= last_non_empty_level;
+       ++level) {
     matched_input_files[level].level = level;
     input_files->emplace_back(std::move(matched_input_files[level]));
   }
@@ -332,8 +340,6 @@ Status CompactionPicker::GetCompactionInputsFromFileNumbers(
   return Status::OK();
 }
 
-
-
 // Returns true if any one of the parent files are being compacted
 bool CompactionPicker::RangeInCompaction(VersionStorageInfo* vstorage,
                                          const InternalKey* smallest,
@@ -505,7 +511,8 @@ Compaction* CompactionPicker::CompactRange(
         vstorage, mutable_cf_options, std::move(inputs), output_level,
         mutable_cf_options.MaxFileSizeForLevel(output_level),
         /* max_grandparent_overlap_bytes */ LLONG_MAX, output_path_id,
-        GetCompressionType(ioptions_, output_level, 1),
+        GetCompressionType(ioptions_, vstorage, mutable_cf_options,
+                           output_level, 1),
         /* grandparents */ {}, /* is manual */ true);
     if (start_level == 0) {
       level0_compactions_in_progress_.insert(c);
@@ -542,7 +549,7 @@ Compaction* CompactionPicker::CompactRange(
   // two files overlap.
   if (input_level > 0) {
     const uint64_t limit = mutable_cf_options.MaxFileSizeForLevel(input_level) *
-      mutable_cf_options.source_compaction_factor;
+                           mutable_cf_options.source_compaction_factor;
     uint64_t total = 0;
     for (size_t i = 0; i + 1 < inputs.size(); ++i) {
       uint64_t s = inputs[i]->compensated_file_size;
@@ -606,7 +613,8 @@ Compaction* CompactionPicker::CompactRange(
       mutable_cf_options.MaxFileSizeForLevel(output_level),
       mutable_cf_options.MaxGrandParentOverlapBytes(input_level),
       output_path_id,
-      GetCompressionType(ioptions_, output_level, vstorage->base_level()),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         vstorage->base_level()),
       std::move(grandparents), /* is manual compaction */ true);
 
   TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction);
@@ -618,11 +626,7 @@ Compaction* CompactionPicker::CompactRange(
   // takes running compactions into account (by skipping files that are already
   // being compacted). Since we just changed compaction score, we recalculate it
   // here
-  {  // this piece of code recomputes compaction score
-    CompactionOptionsFIFO dummy_compaction_options_fifo;
-    vstorage->ComputeCompactionScore(mutable_cf_options,
-                                     dummy_compaction_options_fifo);
-  }
+  vstorage->ComputeCompactionScore(mutable_cf_options);
 
   return compaction;
 }
@@ -630,9 +634,8 @@ Compaction* CompactionPicker::CompactRange(
 #ifndef ROCKSDB_LITE
 namespace {
 // Test whether two files have overlapping key-ranges.
-bool HaveOverlappingKeyRanges(
-    const Comparator* c,
-    const SstFileMetaData& a, const SstFileMetaData& b) {
+bool HaveOverlappingKeyRanges(const Comparator* c, const SstFileMetaData& a,
+                              const SstFileMetaData& b) {
   if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
     if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
       // b.smallestkey <= a.smallestkey <= b.largestkey
@@ -656,9 +659,8 @@ bool HaveOverlappingKeyRanges(
 }  // namespace
 
 Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
-      std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const {
+    std::unordered_set<uint64_t>* input_files,
+    const ColumnFamilyMetaData& cf_meta, const int output_level) const {
   auto& levels = cf_meta.levels;
   auto comparator = icmp_->user_comparator();
 
@@ -711,18 +713,17 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
       // has overlapping key-range with other non-compaction input
       // files in the same level.
       while (first_included > 0) {
-        if (comparator->Compare(
-                current_files[first_included - 1].largestkey,
-                current_files[first_included].smallestkey) < 0) {
+        if (comparator->Compare(current_files[first_included - 1].largestkey,
+                                current_files[first_included].smallestkey) <
+            0) {
           break;
         }
         first_included--;
       }
 
       while (last_included < static_cast<int>(current_files.size()) - 1) {
-        if (comparator->Compare(
-                current_files[last_included + 1].smallestkey,
-                current_files[last_included].largestkey) > 0) {
+        if (comparator->Compare(current_files[last_included + 1].smallestkey,
+                                current_files[last_included].largestkey) > 0) {
           break;
         }
         last_included++;
@@ -732,33 +733,31 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
     // include all files between the first and the last compaction input files.
     for (int f = first_included; f <= last_included; ++f) {
       if (current_files[f].being_compacted) {
-        return Status::Aborted(
-            "Necessary compaction input file " + current_files[f].name +
-            " is currently being compacted.");
+        return Status::Aborted("Necessary compaction input file " +
+                               current_files[f].name +
+                               " is currently being compacted.");
       }
-      input_files->insert(
-          TableFileNameToNumber(current_files[f].name));
+      input_files->insert(TableFileNameToNumber(current_files[f].name));
     }
 
     // update smallest and largest key
     if (l == 0) {
       for (int f = first_included; f <= last_included; ++f) {
-        if (comparator->Compare(
-            smallestkey, current_files[f].smallestkey) > 0) {
+        if (comparator->Compare(smallestkey, current_files[f].smallestkey) >
+            0) {
           smallestkey = current_files[f].smallestkey;
         }
-        if (comparator->Compare(
-            largestkey, current_files[f].largestkey) < 0) {
+        if (comparator->Compare(largestkey, current_files[f].largestkey) < 0) {
           largestkey = current_files[f].largestkey;
         }
       }
     } else {
-      if (comparator->Compare(
-          smallestkey, current_files[first_included].smallestkey) > 0) {
+      if (comparator->Compare(smallestkey,
+                              current_files[first_included].smallestkey) > 0) {
         smallestkey = current_files[first_included].smallestkey;
       }
-      if (comparator->Compare(
-          largestkey, current_files[last_included].largestkey) < 0) {
+      if (comparator->Compare(largestkey,
+                              current_files[last_included].largestkey) < 0) {
         largestkey = current_files[last_included].largestkey;
       }
     }
@@ -775,16 +774,15 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
     // time and not by key
     for (int m = std::max(l, 1); m <= output_level; ++m) {
       for (auto& next_lv_file : levels[m].files) {
-        if (HaveOverlappingKeyRanges(
-            comparator, aggregated_file_meta, next_lv_file)) {
+        if (HaveOverlappingKeyRanges(comparator, aggregated_file_meta,
+                                     next_lv_file)) {
           if (next_lv_file.being_compacted) {
             return Status::Aborted(
                 "File " + next_lv_file.name +
                 " that has overlapping key range with one of the compaction "
                 " input file is currently being compacted.");
           }
-          input_files->insert(
-              TableFileNameToNumber(next_lv_file.name));
+          input_files->insert(TableFileNameToNumber(next_lv_file.name));
         }
       }
     }
@@ -794,28 +792,25 @@ Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels(
 
 Status CompactionPicker::SanitizeCompactionInputFiles(
     std::unordered_set<uint64_t>* input_files,
-    const ColumnFamilyMetaData& cf_meta,
-    const int output_level) const {
+    const ColumnFamilyMetaData& cf_meta, const int output_level) const {
   assert(static_cast<int>(cf_meta.levels.size()) - 1 ==
          cf_meta.levels[cf_meta.levels.size() - 1].level);
   if (output_level >= static_cast<int>(cf_meta.levels.size())) {
     return Status::InvalidArgument(
         "Output level for column family " + cf_meta.name +
         " must between [0, " +
-        ToString(cf_meta.levels[cf_meta.levels.size() - 1].level) +
-        "].");
+        ToString(cf_meta.levels[cf_meta.levels.size() - 1].level) + "].");
   }
 
   if (output_level > MaxOutputLevel()) {
     return Status::InvalidArgument(
         "Exceed the maximum output level defined by "
         "the current compaction algorithm --- " +
-            ToString(MaxOutputLevel()));
+        ToString(MaxOutputLevel()));
   }
 
   if (output_level < 0) {
-    return Status::InvalidArgument(
-        "Output level cannot be negative.");
+    return Status::InvalidArgument("Output level cannot be negative.");
   }
 
   if (input_files->size() == 0) {
@@ -823,8 +818,8 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
         "A compaction must contain at least one file.");
   }
 
-  Status s = SanitizeCompactionInputFilesForAllLevels(
-      input_files, cf_meta, output_level);
+  Status s = SanitizeCompactionInputFilesForAllLevels(input_files, cf_meta,
+                                                      output_level);
 
   if (!s.ok()) {
     return s;
@@ -838,10 +833,9 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
       for (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 " +
-                MakeTableFileName("", file_num) +
-                " is already being compacted.");
+            return Status::Aborted("Specified compaction input file " +
+                                   MakeTableFileName("", file_num) +
+                                   " is already being compacted.");
           }
           found = true;
           break;
@@ -853,8 +847,7 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
     }
     if (!found) {
       return Status::InvalidArgument(
-          "Specified compaction input file " +
-          MakeTableFileName("", file_num) +
+          "Specified compaction input file " + MakeTableFileName("", file_num) +
           " does not exist in column family " + cf_meta.name + ".");
     }
   }
@@ -863,8 +856,8 @@ Status CompactionPicker::SanitizeCompactionInputFiles(
 }
 #endif  // !ROCKSDB_LITE
 
-bool LevelCompactionPicker::NeedsCompaction(const VersionStorageInfo* vstorage)
-    const {
+bool LevelCompactionPicker::NeedsCompaction(
+    const VersionStorageInfo* vstorage) const {
   if (!vstorage->FilesMarkedForCompaction().empty()) {
     return true;
   }
@@ -1010,7 +1003,7 @@ Compaction* LevelCompactionPicker::PickCompaction(
   CompactionInputFiles output_level_inputs;
   output_level_inputs.level = output_level;
   if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage, &inputs,
-                   &output_level_inputs, &parent_index, base_index)) {
+                        &output_level_inputs, &parent_index, base_index)) {
     return nullptr;
   }
 
@@ -1026,7 +1019,8 @@ Compaction* LevelCompactionPicker::PickCompaction(
       mutable_cf_options.MaxFileSizeForLevel(output_level),
       mutable_cf_options.MaxGrandParentOverlapBytes(level),
       GetPathId(ioptions_, mutable_cf_options, output_level),
-      GetCompressionType(ioptions_, output_level, vstorage->base_level()),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level,
+                         vstorage->base_level()),
       std::move(grandparents), is_manual, score,
       false /* deletion_compaction */, compaction_reason);
 
@@ -1040,11 +1034,7 @@ Compaction* LevelCompactionPicker::PickCompaction(
   // takes running compactions into account (by skipping files that are already
   // being compacted). Since we just changed compaction score, we recalculate it
   // here
-  {  // this piece of code recomputes compaction score
-    CompactionOptionsFIFO dummy_compaction_options_fifo;
-    vstorage->ComputeCompactionScore(mutable_cf_options,
-                                     dummy_compaction_options_fifo);
-  }
+  vstorage->ComputeCompactionScore(mutable_cf_options);
 
   TEST_SYNC_POINT_CALLBACK("LevelCompactionPicker::PickCompaction:Return", c);
 
@@ -1439,8 +1429,9 @@ uint32_t UniversalCompactionPicker::GetPathId(
   // considered in this algorithm. So the target size can be violated in
   // that case. We need to improve it.
   uint64_t accumulated_size = 0;
-  uint64_t future_size = file_size *
-    (100 - ioptions.compaction_options_universal.size_ratio) / 100;
+  uint64_t future_size =
+      file_size * (100 - ioptions.compaction_options_universal.size_ratio) /
+      100;
   uint32_t p = 0;
   assert(!ioptions.db_paths.empty());
   for (; p < ioptions.db_paths.size() - 1; p++) {
@@ -1464,17 +1455,17 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
     unsigned int max_number_of_files_to_compact,
     const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
   unsigned int min_merge_width =
-    ioptions_.compaction_options_universal.min_merge_width;
+      ioptions_.compaction_options_universal.min_merge_width;
   unsigned int max_merge_width =
-    ioptions_.compaction_options_universal.max_merge_width;
+      ioptions_.compaction_options_universal.max_merge_width;
 
   const SortedRun* sr = nullptr;
   bool done = false;
   size_t start_index = 0;
   unsigned int candidate_count = 0;
 
-  unsigned int max_files_to_compact = std::min(max_merge_width,
-                                       max_number_of_files_to_compact);
+  unsigned int max_files_to_compact =
+      std::min(max_merge_width, max_number_of_files_to_compact);
   min_merge_width = std::max(min_merge_width, 2U);
 
   // Caller checks the size before executing this function. This invariant is
@@ -1586,7 +1577,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
     uint64_t older_file_size = 0;
     for (size_t i = sorted_runs.size() - 1; i >= first_index_after; i--) {
       older_file_size += sorted_runs[i].size;
-      if (older_file_size * 100L >= total_size * (long) ratio_to_compress) {
+      if (older_file_size * 100L >= total_size * (long)ratio_to_compress) {
         enable_compression = false;
         break;
       }
@@ -1638,7 +1629,8 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
   return new Compaction(
       vstorage, mutable_cf_options, std::move(inputs), output_level,
       mutable_cf_options.MaxFileSizeForLevel(output_level), LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, start_level, 1, enable_compression),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options, start_level,
+                         1, enable_compression),
       /* grandparents */ {}, /* is manual */ false, score,
       false /* deletion_compaction */, compaction_reason);
 }
@@ -1654,8 +1646,8 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
     VersionStorageInfo* vstorage, double score,
     const std::vector<SortedRun>& sorted_runs, LogBuffer* log_buffer) {
   // percentage flexibilty while reducing size amplification
-  uint64_t ratio = ioptions_.compaction_options_universal.
-                     max_size_amplification_percent;
+  uint64_t ratio =
+      ioptions_.compaction_options_universal.max_size_amplification_percent;
 
   unsigned int candidate_count = 0;
   uint64_t candidate_size = 0;
@@ -1666,7 +1658,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
   for (size_t loop = 0; loop < sorted_runs.size() - 1; loop++) {
     sr = &sorted_runs[loop];
     if (!sr->being_compacted) {
-      start_index = loop;         // Consider this as the first candidate.
+      start_index = loop;  // Consider this as the first candidate.
       break;
     }
     char file_num_buf[kFormatFileNumberBufSize];
@@ -1678,7 +1670,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
   }
 
   if (sr == nullptr) {
-    return nullptr;             // no candidate files
+    return nullptr;  // no candidate files
   }
   {
     char file_num_buf[kFormatFileNumberBufSize];
@@ -1763,14 +1755,15 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
       vstorage->num_levels() - 1,
       mutable_cf_options.MaxFileSizeForLevel(vstorage->num_levels() - 1),
       /* max_grandparent_overlap_bytes */ LLONG_MAX, path_id,
-      GetCompressionType(ioptions_, vstorage->num_levels() - 1, 1),
+      GetCompressionType(ioptions_, vstorage, mutable_cf_options,
+                         vstorage->num_levels() - 1, 1),
       /* grandparents */ {}, /* is manual */ false, score,
       false /* deletion_compaction */,
       CompactionReason::kUniversalSizeAmplification);
 }
 
-bool FIFOCompactionPicker::NeedsCompaction(const VersionStorageInfo* vstorage)
-    const {
+bool FIFOCompactionPicker::NeedsCompaction(
+    const VersionStorageInfo* vstorage) const {
   const int kLevel0 = 0;
   return vstorage->CompactionScore(kLevel0) >= 1;
 }

db/compaction_picker.h

@@ -63,23 +63,20 @@ class CompactionPicker {
       InternalKey** compaction_end, bool* manual_conflict);
 
   // The maximum allowed output level.  Default value is NumberLevels() - 1.
-  virtual int MaxOutputLevel() const {
-    return NumberLevels() - 1;
-  }
+  virtual int MaxOutputLevel() const { return NumberLevels() - 1; }
 
   virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const = 0;
 
-  // Sanitize the input set of compaction input files.
-  // When the input parameters do not describe a valid compaction, the
-  // function will try to fix the input_files by adding necessary
-  // files.  If it's not possible to conver an invalid input_files
-  // into a valid one by adding more files, the function will return a
-  // non-ok status with specific reason.
+// Sanitize the input set of compaction input files.
+// When the input parameters do not describe a valid compaction, the
+// function will try to fix the input_files by adding necessary
+// files.  If it's not possible to conver an invalid input_files
+// into a valid one by adding more files, the function will return a
+// non-ok status with specific reason.
 #ifndef ROCKSDB_LITE
-  Status SanitizeCompactionInputFiles(
-      std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const;
+  Status SanitizeCompactionInputFiles(std::unordered_set<uint64_t>* input_files,
+                                      const ColumnFamilyMetaData& cf_meta,
+                                      const int output_level) const;
 #endif  // ROCKSDB_LITE
 
   // Free up the files that participated in a compaction
@@ -123,15 +120,15 @@ class CompactionPicker {
   // Stores the minimal range that covers all entries in inputs in
   // *smallest, *largest.
   // REQUIRES: inputs is not empty
-  void GetRange(const CompactionInputFiles& inputs,
-                InternalKey* smallest, InternalKey* largest);
+  void GetRange(const CompactionInputFiles& inputs, InternalKey* smallest,
+                InternalKey* largest);
 
   // Stores the minimal range that covers all entries in inputs1 and inputs2
   // in *smallest, *largest.
   // REQUIRES: inputs is not empty
   void GetRange(const CompactionInputFiles& inputs1,
-                const CompactionInputFiles& inputs2,
-                InternalKey* smallest, InternalKey* largest);
+                const CompactionInputFiles& inputs2, InternalKey* smallest,
+                InternalKey* largest);
 
   // Add more files to the inputs on "level" to make sure that
   // no newer version of a key is compacted to "level+1" while leaving an older
@@ -166,13 +163,12 @@ class CompactionPicker {
 
   const ImmutableCFOptions& ioptions_;
 
-  // A helper function to SanitizeCompactionInputFiles() that
-  // sanitizes "input_files" by adding necessary files.
+// A helper function to SanitizeCompactionInputFiles() that
+// sanitizes "input_files" by adding necessary files.
 #ifndef ROCKSDB_LITE
   virtual Status SanitizeCompactionInputFilesForAllLevels(
       std::unordered_set<uint64_t>* input_files,
-      const ColumnFamilyMetaData& cf_meta,
-      const int output_level) const;
+      const ColumnFamilyMetaData& cf_meta, const int output_level) const;
 #endif  // ROCKSDB_LITE
 
   // Keeps track of all compactions that are running on Level0.
@@ -192,8 +188,8 @@ class LevelCompactionPicker : public CompactionPicker {
                                      VersionStorageInfo* vstorage,
                                      LogBuffer* log_buffer) override;
 
-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;
 
   // Pick a path ID to place a newly generated file, with its level
   static uint32_t GetPathId(const ImmutableCFOptions& ioptions,
@@ -232,8 +228,8 @@ class UniversalCompactionPicker : public CompactionPicker {
 
   virtual int MaxOutputLevel() const override { return NumberLevels() - 1; }
 
-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;
 
  private:
   struct SortedRun {
@@ -307,19 +303,17 @@ class FIFOCompactionPicker : public CompactionPicker {
       InternalKey** compaction_end, bool* manual_conflict) override;
 
   // The maximum allowed output level.  Always returns 0.
-  virtual int MaxOutputLevel() const override {
-    return 0;
-  }
+  virtual int MaxOutputLevel() const override { return 0; }
 
-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override;
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override;
 };
 
 class NullCompactionPicker : public CompactionPicker {
  public:
   NullCompactionPicker(const ImmutableCFOptions& ioptions,
-                       const InternalKeyComparator* icmp) :
-      CompactionPicker(ioptions, icmp) {}
+                       const InternalKeyComparator* icmp)
+      : CompactionPicker(ioptions, icmp) {}
   virtual ~NullCompactionPicker() {}
 
   // Always return "nullptr"
@@ -342,14 +336,16 @@ class NullCompactionPicker : public CompactionPicker {
   }
 
   // Always returns false.
-  virtual bool NeedsCompaction(const VersionStorageInfo* vstorage) const
-      override {
+  virtual bool NeedsCompaction(
+      const VersionStorageInfo* vstorage) const override {
     return false;
   }
 };
 #endif  // !ROCKSDB_LITE
 
 CompressionType GetCompressionType(const ImmutableCFOptions& ioptions,
+                                   const VersionStorageInfo* vstorage,
+                                   const MutableCFOptions& mutable_cf_options,
                                    int level, int base_level,
                                    const bool enable_compression = true);
 

db/compaction_picker_test.cc

@@ -79,7 +79,7 @@ class CompactionPickerTest : public testing::Test {
   }
 
   void Add(int level, uint32_t file_number, const char* smallest,
-           const char* largest, uint64_t file_size = 0, uint32_t path_id = 0,
+           const char* largest, uint64_t file_size = 1, uint32_t path_id = 0,
            SequenceNumber smallest_seq = 100,
            SequenceNumber largest_seq = 100) {
     assert(level < vstorage_->num_levels());
@@ -119,7 +119,7 @@ class CompactionPickerTest : public testing::Test {
     vstorage_->UpdateNumNonEmptyLevels();
     vstorage_->GenerateFileIndexer();
     vstorage_->GenerateLevelFilesBrief();
-    vstorage_->ComputeCompactionScore(mutable_cf_options_, fifo_options_);
+    vstorage_->ComputeCompactionScore(mutable_cf_options_);
     vstorage_->GenerateLevel0NonOverlapping();
     vstorage_->SetFinalized();
   }
@@ -195,6 +195,7 @@ TEST_F(CompactionPickerTest, LevelMaxScore) {
   NewVersionStorage(6, kCompactionStyleLevel);
   mutable_cf_options_.target_file_size_base = 10000000;
   mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
   Add(0, 1U, "150", "200", 1000000000U);
   // Level 1 score 1.2
   Add(1, 66U, "150", "200", 6000000U);
@@ -319,6 +320,7 @@ TEST_F(CompactionPickerTest, Level0TriggerDynamic4) {
   mutable_cf_options_.level0_file_num_compaction_trigger = 2;
   mutable_cf_options_.max_bytes_for_level_base = 200;
   mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+
   NewVersionStorage(num_levels, kCompactionStyleLevel);
   Add(0, 1U, "150", "200");
   Add(0, 2U, "200", "250");
@@ -351,6 +353,7 @@ TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
   mutable_cf_options_.level0_file_num_compaction_trigger = 2;
   mutable_cf_options_.max_bytes_for_level_base = 200;
   mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
   NewVersionStorage(num_levels, kCompactionStyleLevel);
   Add(0, 1U, "150", "200");
   Add(num_levels - 1, 3U, "200", "250", 300U);
@@ -366,11 +369,9 @@ TEST_F(CompactionPickerTest, LevelTriggerDynamic4) {
       cf_name_, mutable_cf_options_, vstorage_.get(), &log_buffer_));
   ASSERT_TRUE(compaction.get() != nullptr);
   ASSERT_EQ(1U, compaction->num_input_files(0));
-  ASSERT_EQ(6U, compaction->input(0, 0)->fd.GetNumber());
-  ASSERT_EQ(2U, compaction->num_input_files(1));
-  ASSERT_EQ(3U, compaction->input(1, 0)->fd.GetNumber());
-  ASSERT_EQ(4U, compaction->input(1, 1)->fd.GetNumber());
-  ASSERT_EQ(2U, compaction->num_input_levels());
+  ASSERT_EQ(5U, compaction->input(0, 0)->fd.GetNumber());
+  ASSERT_EQ(0, compaction->num_input_files(1));
+  ASSERT_EQ(1U, compaction->num_input_levels());
   ASSERT_EQ(num_levels - 1, compaction->output_level());
 }
 
@@ -466,7 +467,6 @@ TEST_F(CompactionPickerTest, NeedsCompactionFIFO) {
   fifo_options_.max_table_files_size = kMaxSize;
   ioptions_.compaction_options_fifo = fifo_options_;
   FIFOCompactionPicker fifo_compaction_picker(ioptions_, &icmp_);
-
   UpdateVersionStorageInfo();
   // must return false when there's no files.
   ASSERT_EQ(fifo_compaction_picker.NeedsCompaction(vstorage_.get()), false);
@@ -491,6 +491,7 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping1) {
   NewVersionStorage(6, kCompactionStyleLevel);
   mutable_cf_options_.target_file_size_base = 10000000;
   mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
   mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
 
   Add(2, 6U, "150", "179", 50000000U);
@@ -517,6 +518,7 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
   NewVersionStorage(6, kCompactionStyleLevel);
   mutable_cf_options_.target_file_size_base = 10000000;
   mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10 * 1024 * 1024;
   mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
 
   Add(2, 6U, "150", "175",
@@ -544,20 +546,21 @@ TEST_F(CompactionPickerTest, CompactionPriMinOverlapping2) {
 
 TEST_F(CompactionPickerTest, CompactionPriMinOverlapping3) {
   NewVersionStorage(6, kCompactionStyleLevel);
-  mutable_cf_options_.target_file_size_base = 10000000;
-  mutable_cf_options_.target_file_size_multiplier = 10;
+  mutable_cf_options_.max_bytes_for_level_base = 10000000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
   mutable_cf_options_.compaction_pri = kMinOverlappingRatio;
 
   // file 7 and 8 over lap with the same file, but file 8 is smaller so
   // it will be picked.
-  Add(2, 6U, "150", "175", 60000000U);  // Overlaps with file 26, 27
-  Add(2, 7U, "176", "200", 60000000U);  // Overlaps with file 27
-  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 27
+  Add(2, 6U, "150", "167", 60000000U);  // Overlaps with file 26, 27
+  Add(2, 7U, "168", "169", 60000000U);  // Overlaps with file 27
+  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 28, but the file
+                                        // itself is larger. Should be picked.
 
   Add(3, 26U, "160", "165", 260000000U);
-  Add(3, 26U, "166", "170", 260000000U);
-  Add(3, 27U, "180", "400", 260000000U);
-  Add(3, 28U, "401", "500", 260000000U);
+  Add(3, 27U, "166", "170", 260000000U);
+  Add(3, 28U, "180", "400", 260000000U);
+  Add(3, 29U, "401", "500", 260000000U);
   UpdateVersionStorageInfo();
 
   std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
@@ -595,6 +598,8 @@ TEST_F(CompactionPickerTest, ParentIndexResetBug) {
 // ranges (with different sequence numbers) in the input files.
 TEST_F(CompactionPickerTest, OverlappingUserKeys) {
   NewVersionStorage(6, kCompactionStyleLevel);
+  mutable_cf_options_.compaction_pri = kByCompensatedSize;
+
   Add(1, 1U, "100", "150", 1U);
   // Overlapping user keys
   Add(1, 2U, "200", "400", 1U);
@@ -770,19 +775,22 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded1) {
   Add(1, 4U, "400", "500", 600);
   Add(1, 5U, "600", "700", 600);
   // Level 2 is less than target 10000 even added size of level 1
+  // Size ratio of L2/L1 is 9600 / 1200 = 8
   Add(2, 6U, "150", "200", 2500);
   Add(2, 7U, "201", "210", 2000);
-  Add(2, 8U, "300", "310", 2500);
+  Add(2, 8U, "300", "310", 2600);
   Add(2, 9U, "400", "500", 2500);
   // Level 3 exceeds target 100,000 of 1000
   Add(3, 10U, "400", "500", 101000);
-  // Level 4 exceeds target 1,000,000 of 500 after adding size from level 3
-  Add(4, 11U, "400", "500", 999500);
-  Add(5, 11U, "400", "500", 8000000);
+  // Level 4 exceeds target 1,000,000 by 900 after adding size from level 3
+  // Size ratio L4/L3 is 9.9
+  // After merge from L3, L4 size is 1000900
+  Add(4, 11U, "400", "500", 999900);
+  Add(5, 11U, "400", "500", 8007200);
 
   UpdateVersionStorageInfo();
 
-  ASSERT_EQ(2200u + 11000u + 5500u,
+  ASSERT_EQ(200u * 9u + 10900u + 900u * 9,
             vstorage_->estimated_compaction_needed_bytes());
 }
 
@@ -798,17 +806,42 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded2) {
   Add(0, 4U, "150", "200", 200);
   Add(0, 5U, "150", "200", 200);
   Add(0, 6U, "150", "200", 200);
-  // Level 1 is over target by
+  // Level 1 size will be 1400 after merging with L0
   Add(1, 7U, "400", "500", 200);
   Add(1, 8U, "600", "700", 200);
   // Level 2 is less than target 10000 even added size of level 1
-  Add(2, 9U, "150", "200", 9500);
+  Add(2, 9U, "150", "200", 9100);
+  // Level 3 over the target, but since level 4 is empty, we assume it will be
+  // a trivial move.
   Add(3, 10U, "400", "500", 101000);
 
   UpdateVersionStorageInfo();
 
-  ASSERT_EQ(1400u + 4400u + 11000u,
-            vstorage_->estimated_compaction_needed_bytes());
+  // estimated L1->L2 merge: 400 * (9100.0 / 1400.0 + 1.0)
+  ASSERT_EQ(1400u + 3000u, vstorage_->estimated_compaction_needed_bytes());
+}
+
+TEST_F(CompactionPickerTest, EstimateCompactionBytesNeeded3) {
+  int num_levels = ioptions_.num_levels;
+  ioptions_.level_compaction_dynamic_level_bytes = false;
+  mutable_cf_options_.level0_file_num_compaction_trigger = 3;
+  mutable_cf_options_.max_bytes_for_level_base = 1000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  NewVersionStorage(num_levels, kCompactionStyleLevel);
+  Add(0, 1U, "150", "200", 2000);
+  Add(0, 2U, "150", "200", 2000);
+  Add(0, 4U, "150", "200", 2000);
+  Add(0, 5U, "150", "200", 2000);
+  Add(0, 6U, "150", "200", 1000);
+  // Level 1 size will be 10000 after merging with L0
+  Add(1, 7U, "400", "500", 500);
+  Add(1, 8U, "600", "700", 500);
+
+  Add(2, 9U, "150", "200", 10000);
+
+  UpdateVersionStorageInfo();
+
+  ASSERT_EQ(10000u + 18000u, vstorage_->estimated_compaction_needed_bytes());
 }
 
 TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {
@@ -832,12 +865,14 @@ TEST_F(CompactionPickerTest, EstimateCompactionBytesNeededDynamicLevel) {
   // num_levels - 3 is over target by 100 + 1000
   Add(num_levels - 3, 7U, "400", "500", 300);
   Add(num_levels - 3, 8U, "600", "700", 300);
-  // Level 2 is over target by 1100 + 100
-  Add(num_levels - 2, 9U, "150", "200", 5100);
+  // num_levels - 2 is over target by 1100 + 200
+  Add(num_levels - 2, 9U, "150", "200", 5200);
 
   UpdateVersionStorageInfo();
 
-  ASSERT_EQ(1600u + 12100u + 13200u,
+  // Merging to the second last level: (5200 / 1600 + 1) * 1100
+  // Merging to the last level: (50000 / 6300 + 1) * 1300
+  ASSERT_EQ(1600u + 4675u + 11617u,
             vstorage_->estimated_compaction_needed_bytes());
 }
 

db/corruption_test.cc

@@ -42,6 +42,7 @@ class CorruptionTest : public testing::Test {
 
   CorruptionTest() {
     tiny_cache_ = NewLRUCache(100);
+    options_.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
     options_.env = &env_;
     dbname_ = test::TmpDir() + "/corruption_test";
     DestroyDB(dbname_, options_);

db/db_block_cache_test.cc

@@ -0,0 +1,495 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same 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 <cstdlib>
+#include "db/db_test_util.h"
+#include "port/stack_trace.h"
+
+namespace rocksdb {
+
+class DBBlockCacheTest : public DBTestBase {
+ private:
+  size_t miss_count_ = 0;
+  size_t hit_count_ = 0;
+  size_t insert_count_ = 0;
+  size_t failure_count_ = 0;
+  size_t compressed_miss_count_ = 0;
+  size_t compressed_hit_count_ = 0;
+  size_t compressed_insert_count_ = 0;
+  size_t compressed_failure_count_ = 0;
+
+ public:
+  const size_t kNumBlocks = 10;
+  const size_t kValueSize = 100;
+
+  DBBlockCacheTest() : DBTestBase("/db_block_cache_test") {}
+
+  BlockBasedTableOptions GetTableOptions() {
+    BlockBasedTableOptions table_options;
+    // Set a small enough block size so that each key-value get its own block.
+    table_options.block_size = 1;
+    return table_options;
+  }
+
+  Options GetOptions(const BlockBasedTableOptions& table_options) {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    // options.compression = kNoCompression;
+    options.statistics = rocksdb::CreateDBStatistics();
+    options.table_factory.reset(new BlockBasedTableFactory(table_options));
+    return options;
+  }
+
+  void InitTable(const Options& options) {
+    std::string value(kValueSize, 'a');
+    for (size_t i = 0; i < kNumBlocks; i++) {
+      ASSERT_OK(Put(ToString(i), value.c_str()));
+    }
+  }
+
+  void RecordCacheCounters(const Options& options) {
+    miss_count_ = TestGetTickerCount(options, BLOCK_CACHE_MISS);
+    hit_count_ = TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    insert_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD);
+    failure_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
+    compressed_miss_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
+    compressed_hit_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
+    compressed_insert_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
+    compressed_failure_count_ =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
+  }
+
+  void CheckCacheCounters(const Options& options, size_t expected_misses,
+                          size_t expected_hits, size_t expected_inserts,
+                          size_t expected_failures) {
+    size_t new_miss_count = TestGetTickerCount(options, BLOCK_CACHE_MISS);
+    size_t new_hit_count = TestGetTickerCount(options, BLOCK_CACHE_HIT);
+    size_t new_insert_count = TestGetTickerCount(options, BLOCK_CACHE_ADD);
+    size_t new_failure_count =
+        TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
+    ASSERT_EQ(miss_count_ + expected_misses, new_miss_count);
+    ASSERT_EQ(hit_count_ + expected_hits, new_hit_count);
+    ASSERT_EQ(insert_count_ + expected_inserts, new_insert_count);
+    ASSERT_EQ(failure_count_ + expected_failures, new_failure_count);
+    miss_count_ = new_miss_count;
+    hit_count_ = new_hit_count;
+    insert_count_ = new_insert_count;
+    failure_count_ = new_failure_count;
+  }
+
+  void CheckCompressedCacheCounters(const Options& options,
+                                    size_t expected_misses,
+                                    size_t expected_hits,
+                                    size_t expected_inserts,
+                                    size_t expected_failures) {
+    size_t new_miss_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
+    size_t new_hit_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
+    size_t new_insert_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
+    size_t new_failure_count =
+        TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
+    ASSERT_EQ(compressed_miss_count_ + expected_misses, new_miss_count);
+    ASSERT_EQ(compressed_hit_count_ + expected_hits, new_hit_count);
+    ASSERT_EQ(compressed_insert_count_ + expected_inserts, new_insert_count);
+    ASSERT_EQ(compressed_failure_count_ + expected_failures, new_failure_count);
+    compressed_miss_count_ = new_miss_count;
+    compressed_hit_count_ = new_hit_count;
+    compressed_insert_count_ = new_insert_count;
+    compressed_failure_count_ = new_failure_count;
+  }
+};
+
+TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
+  ReadOptions read_options;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(0, 0, false);
+  table_options.block_cache = cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  // Load blocks into cache.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 1, 0, 1, 0);
+    iterators[i].reset(iter);
+  }
+  size_t usage = cache->GetUsage();
+  ASSERT_LT(0, usage);
+  cache->SetCapacity(usage);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+
+  // Test with strict capacity limit.
+  cache->SetStrictCapacityLimit(true);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_TRUE(iter->status().IsIncomplete());
+  CheckCacheCounters(options, 1, 0, 0, 1);
+  delete iter;
+  iter = nullptr;
+
+  // Release interators and access cache again.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iterators[i].reset();
+    CheckCacheCounters(options, 0, 0, 0, 0);
+  }
+  ASSERT_EQ(0, cache->GetPinnedUsage());
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 0, 1, 0, 0);
+    iterators[i].reset(iter);
+  }
+}
+
+#ifdef SNAPPY
+TEST_F(DBBlockCacheTest, TestWithCompressedBlockCache) {
+  ReadOptions read_options;
+  auto table_options = GetTableOptions();
+  auto options = GetOptions(table_options);
+  options.compression = CompressionType::kSnappyCompression;
+  InitTable(options);
+
+  std::shared_ptr<Cache> cache = NewLRUCache(0, 0, false);
+  std::shared_ptr<Cache> compressed_cache = NewLRUCache(0, 0, false);
+  table_options.block_cache = cache;
+  table_options.block_cache_compressed = compressed_cache;
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  Reopen(options);
+  RecordCacheCounters(options);
+
+  std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
+  Iterator* iter = nullptr;
+
+  // Load blocks into cache.
+  for (size_t i = 0; i < kNumBlocks - 1; i++) {
+    iter = db_->NewIterator(read_options);
+    iter->Seek(ToString(i));
+    ASSERT_OK(iter->status());
+    CheckCacheCounters(options, 1, 0, 1, 0);
+    CheckCompressedCacheCounters(options, 1, 0, 1, 0);
+    iterators[i].reset(iter);
+  }
+  size_t usage = cache->GetUsage();
+  ASSERT_LT(0, usage);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+  size_t compressed_usage = compressed_cache->GetUsage();
+  ASSERT_LT(0, compressed_usage);
+  // Compressed block cache cannot be pinned.
+  ASSERT_EQ(0, compressed_cache->GetPinnedUsage());
+
+  // Set strict capacity limit flag. Now block will only load into compressed
+  // block cache.
+  cache->SetCapacity(usage);
+  cache->SetStrictCapacityLimit(true);
+  ASSERT_EQ(usage, cache->GetPinnedUsage());
+  // compressed_cache->SetCapacity(compressed_usage);
+  compressed_cache->SetCapacity(0);
+  // compressed_cache->SetStrictCapacityLimit(true);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_TRUE(iter->status().IsIncomplete());
+  CheckCacheCounters(options, 1, 0, 0, 1);
+  CheckCompressedCacheCounters(options, 1, 0, 1, 0);
+  delete iter;
+  iter = nullptr;
+
+  // Clear strict capacity limit flag. This time we shall hit compressed block
+  // cache.
+  cache->SetStrictCapacityLimit(false);
+  iter = db_->NewIterator(read_options);
+  iter->Seek(ToString(kNumBlocks - 1));
+  ASSERT_OK(iter->status());
+  CheckCacheCounters(options, 1, 0, 1, 0);
+  CheckCompressedCacheCounters(options, 0, 1, 0, 0);
+  delete iter;
+  iter = nullptr;
+}
+#endif  // SNAPPY
+
+#ifndef ROCKSDB_LITE
+
+// Make sure that when options.block_cache is set, after a new table is
+// created its index/filter blocks are added to block cache.
+TEST_F(DBBlockCacheTest, IndexAndFilterBlocksOfNewTableAddedToCache) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = true;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "key", "val"));
+  // Create a new table.
+  ASSERT_OK(Flush(1));
+
+  // index/filter blocks added to block cache right after table creation.
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(2, /* only index/filter were added */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+  ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
+  uint64_t int_num;
+  ASSERT_TRUE(
+      dbfull()->GetIntProperty("rocksdb.estimate-table-readers-mem", &int_num));
+  ASSERT_EQ(int_num, 0U);
+
+  // Make sure filter block is in cache.
+  std::string value;
+  ReadOptions ropt;
+  db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
+
+  // Miss count should remain the same.
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  // Make sure index block is in cache.
+  auto index_block_hit = TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT);
+  value = Get(1, "key");
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(index_block_hit + 1,
+            TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+
+  value = Get(1, "key");
+  ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
+  ASSERT_EQ(index_block_hit + 2,
+            TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
+}
+
+TEST_F(DBBlockCacheTest, IndexAndFilterBlocksStats) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = true;
+  // 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));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "key", "val"));
+  // Create a new table
+  ASSERT_OK(Flush(1));
+  size_t index_bytes_insert =
+      TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT);
+  size_t filter_bytes_insert =
+      TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT);
+  ASSERT_GT(index_bytes_insert, 0);
+  ASSERT_GT(filter_bytes_insert, 0);
+  ASSERT_EQ(cache->GetUsage(), index_bytes_insert + filter_bytes_insert);
+  // set the cache capacity to the current usage
+  cache->SetCapacity(index_bytes_insert + filter_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT), 0);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT), 0);
+  ASSERT_OK(Put(1, "key2", "val"));
+  // Create a new table
+  ASSERT_OK(Flush(1));
+  // cache evicted old index and block entries
+  ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT),
+            index_bytes_insert);
+  ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT),
+            filter_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT),
+            index_bytes_insert);
+  ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT),
+            filter_bytes_insert);
+}
+
+TEST_F(DBBlockCacheTest, ParanoidFileChecks) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.statistics = rocksdb::CreateDBStatistics();
+  options.level0_file_num_compaction_trigger = 2;
+  options.paranoid_file_checks = true;
+  BlockBasedTableOptions table_options;
+  table_options.cache_index_and_filter_blocks = false;
+  table_options.filter_policy.reset(NewBloomFilterPolicy(20));
+  options.table_factory.reset(new BlockBasedTableFactory(table_options));
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "1_key", "val"));
+  ASSERT_OK(Put(1, "9_key", "val"));
+  // Create a new table.
+  ASSERT_OK(Flush(1));
+  ASSERT_EQ(1, /* read and cache data block */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+  ASSERT_OK(Put(1, "1_key2", "val2"));
+  ASSERT_OK(Put(1, "9_key2", "val2"));
+  // Create a new SST file. This will further trigger a compaction
+  // and generate another file.
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(3, /* Totally 3 files created up to now */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+
+  // After disabling options.paranoid_file_checks. NO further block
+  // is added after generating a new file.
+  ASSERT_OK(
+      dbfull()->SetOptions(handles_[1], {{"paranoid_file_checks", "false"}}));
+
+  ASSERT_OK(Put(1, "1_key3", "val3"));
+  ASSERT_OK(Put(1, "9_key3", "val3"));
+  ASSERT_OK(Flush(1));
+  ASSERT_OK(Put(1, "1_key4", "val4"));
+  ASSERT_OK(Put(1, "9_key4", "val4"));
+  ASSERT_OK(Flush(1));
+  dbfull()->TEST_WaitForCompact();
+  ASSERT_EQ(3, /* Totally 3 files created up to now */
+            TestGetTickerCount(options, BLOCK_CACHE_ADD));
+}
+
+TEST_F(DBBlockCacheTest, CompressedCache) {
+  if (!Snappy_Supported()) {
+    return;
+  }
+  int num_iter = 80;
+
+  // Run this test three iterations.
+  // Iteration 1: only a uncompressed block cache
+  // Iteration 2: only a compressed block cache
+  // Iteration 3: both block cache and compressed cache
+  // Iteration 4: both block cache and compressed cache, but DB is not
+  // compressed
+  for (int iter = 0; iter < 4; iter++) {
+    Options options = CurrentOptions();
+    options.write_buffer_size = 64 * 1024;  // small write buffer
+    options.statistics = rocksdb::CreateDBStatistics();
+
+    BlockBasedTableOptions table_options;
+    switch (iter) {
+      case 0:
+        // only uncompressed block cache
+        table_options.block_cache = NewLRUCache(8 * 1024);
+        table_options.block_cache_compressed = nullptr;
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 1:
+        // no block cache, only compressed cache
+        table_options.no_block_cache = true;
+        table_options.block_cache = nullptr;
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 2:
+        // both compressed and uncompressed block cache
+        table_options.block_cache = NewLRUCache(1024);
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        break;
+      case 3:
+        // both block cache and compressed cache, but DB is not compressed
+        // also, make block cache sizes bigger, to trigger block cache hits
+        table_options.block_cache = NewLRUCache(1024 * 1024);
+        table_options.block_cache_compressed = NewLRUCache(8 * 1024 * 1024);
+        options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+        options.compression = kNoCompression;
+        break;
+      default:
+        ASSERT_TRUE(false);
+    }
+    CreateAndReopenWithCF({"pikachu"}, options);
+    // default column family doesn't have block cache
+    Options no_block_cache_opts;
+    no_block_cache_opts.statistics = options.statistics;
+    no_block_cache_opts = CurrentOptions(no_block_cache_opts);
+    BlockBasedTableOptions table_options_no_bc;
+    table_options_no_bc.no_block_cache = true;
+    no_block_cache_opts.table_factory.reset(
+        NewBlockBasedTableFactory(table_options_no_bc));
+    ReopenWithColumnFamilies(
+        {"default", "pikachu"},
+        std::vector<Options>({no_block_cache_opts, options}));
+
+    Random rnd(301);
+
+    // Write 8MB (80 values, each 100K)
+    ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
+    std::vector<std::string> values;
+    std::string str;
+    for (int i = 0; i < num_iter; i++) {
+      if (i % 4 == 0) {  // high compression ratio
+        str = RandomString(&rnd, 1000);
+      }
+      values.push_back(str);
+      ASSERT_OK(Put(1, Key(i), values[i]));
+    }
+
+    // flush all data from memtable so that reads are from block cache
+    ASSERT_OK(Flush(1));
+
+    for (int i = 0; i < num_iter; i++) {
+      ASSERT_EQ(Get(1, Key(i)), values[i]);
+    }
+
+    // check that we triggered the appropriate code paths in the cache
+    switch (iter) {
+      case 0:
+        // only uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 1:
+        // no block cache, only compressed cache
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 2:
+        // both compressed and uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        break;
+      case 3:
+        // both compressed and uncompressed block cache
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_HIT), 0);
+        ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
+        // compressed doesn't have any hits since blocks are not compressed on
+        // storage
+        ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT), 0);
+        break;
+      default:
+        ASSERT_TRUE(false);
+    }
+
+    options.create_if_missing = true;
+    DestroyAndReopen(options);
+  }
+}
+
+#endif  // ROCKSDB_LITE
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}

db/db_compaction_filter_test.cc

@@ -368,11 +368,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilter) {
 // entries are deleted. The compaction should create bunch of 'DeleteFile'
 // entries in VersionEdit, but none of the 'AddFile's.
 TEST_F(DBTestCompactionFilter, CompactionFilterDeletesAll) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_filter_factory = std::make_shared<DeleteFilterFactory>();
   options.disable_auto_compactions = true;
   options.create_if_missing = true;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   // put some data
@@ -400,11 +399,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterDeletesAll) {
 
 TEST_F(DBTestCompactionFilter, CompactionFilterWithValueChange) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.num_levels = 3;
     options.compaction_filter_factory =
       std::make_shared<ChangeFilterFactory>();
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Write 100K+1 keys, these are written to a few files
@@ -467,8 +465,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterWithMergeOperator) {
   PutFixed64(&three, 3);
   PutFixed64(&four, 4);
 
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
   options.create_if_missing = true;
   options.merge_operator = MergeOperators::CreateUInt64AddOperator();
   options.num_levels = 3;
@@ -621,11 +618,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextCfId) {
 // Compaction filters should only be applied to records that are newer than the
 // latest snapshot. This test inserts records and applies a delete filter.
 TEST_F(DBTestCompactionFilter, CompactionFilterSnapshot) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_filter_factory = std::make_shared<DeleteFilterFactory>();
   options.disable_auto_compactions = true;
   options.create_if_missing = true;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   // Put some data.
@@ -659,11 +655,10 @@ TEST_F(DBTestCompactionFilter, CompactionFilterSnapshot) {
 // records newer than the snapshot will also be processed
 TEST_F(DBTestCompactionFilter, CompactionFilterIgnoreSnapshot) {
   std::string five = ToString(5);
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_filter_factory = std::make_shared<DeleteISFilterFactory>();
   options.disable_auto_compactions = true;
   options.create_if_missing = true;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   // Put some data.

db/db_compaction_test.cc

@@ -40,34 +40,6 @@ class DBCompactionTestWithParam
 };
 
 namespace {
-class OnFileDeletionListener : public EventListener {
- public:
-  OnFileDeletionListener() :
-      matched_count_(0),
-      expected_file_name_("") {}
-
-  void SetExpectedFileName(
-      const std::string file_name) {
-    expected_file_name_ = file_name;
-  }
-
-  void VerifyMatchedCount(size_t expected_value) {
-    ASSERT_EQ(matched_count_, expected_value);
-  }
-
-  void OnTableFileDeleted(
-      const TableFileDeletionInfo& info) override {
-    if (expected_file_name_ != "") {
-      ASSERT_EQ(expected_file_name_, info.file_path);
-      expected_file_name_ = "";
-      matched_count_++;
-    }
-  }
-
- private:
-  size_t matched_count_;
-  std::string expected_file_name_;
-};
 
 class FlushedFileCollector : public EventListener {
  public:
@@ -98,8 +70,7 @@ class FlushedFileCollector : public EventListener {
 static const int kCDTValueSize = 1000;
 static const int kCDTKeysPerBuffer = 4;
 static const int kCDTNumLevels = 8;
-Options DeletionTriggerOptions() {
-  Options options;
+Options DeletionTriggerOptions(Options options) {
   options.compression = kNoCompression;
   options.write_buffer_size = kCDTKeysPerBuffer * (kCDTValueSize + 24);
   options.min_write_buffer_number_to_merge = 1;
@@ -202,7 +173,7 @@ const SstFileMetaData* PickFileRandomly(
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
   for (int tid = 0; tid < 3; ++tid) {
     uint64_t db_size[2];
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
     options.max_subcompactions = max_subcompactions_;
 
     if (tid == 1) {
@@ -245,8 +216,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
   // the compaction behavior when there are many of deletion entries.
   // The test will need to be updated if the internal behavior changes.
 
-  Options options = DeletionTriggerOptions();
-  options = CurrentOptions(options);
+  Options options = DeletionTriggerOptions(CurrentOptions());
   options.env = env_;
   DestroyAndReopen(options);
   Random rnd(301);
@@ -288,8 +258,7 @@ TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
 }
 
 TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
   options.env = env_;
   options.new_table_reader_for_compaction_inputs = true;
   options.max_open_files = 100;
@@ -378,7 +347,7 @@ TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
 TEST_P(DBCompactionTestWithParam, CompactionDeletionTriggerReopen) {
   for (int tid = 0; tid < 2; ++tid) {
     uint64_t db_size[3];
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
     options.max_subcompactions = max_subcompactions_;
 
     if (tid == 1) {
@@ -435,7 +404,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTriggerReopen) {
 TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
   uint64_t db_size[3];
   for (int test = 0; test < 2; ++test) {
-    Options options = CurrentOptions(DeletionTriggerOptions());
+    Options options = DeletionTriggerOptions(CurrentOptions());
     options.skip_stats_update_on_db_open = (test == 0);
 
     env_->random_read_counter_.Reset();
@@ -493,14 +462,13 @@ TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
 TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
   const int kNumKeysPerFile = 100;
 
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 110 << 10;  // 110KB
   options.arena_block_size = 4 << 10;
   options.num_levels = 3;
   options.level0_file_num_compaction_trigger = 3;
   options.max_subcompactions = max_subcompactions_;
   options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
-  options = CurrentOptions(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
   Random rnd(301);
@@ -538,7 +506,7 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
   // and see number of compactions scheduled to be less than allowed.
   const int kNumKeysPerFile = 100;
 
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 110 << 10;  // 110KB
   options.arena_block_size = 4 << 10;
   options.num_levels = 3;
@@ -549,7 +517,6 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
   options.base_background_compactions = 1;
   options.max_background_compactions = 3;
   options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
-  options = CurrentOptions(options);
 
   // Block all threads in thread pool.
   const size_t kTotalTasks = 4;
@@ -632,10 +599,9 @@ TEST_F(DBCompactionTest, BGCompactionsAllowed) {
 }
 
 TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 100000000;        // Large write buffer
   options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
   Random rnd(301);
@@ -662,9 +628,8 @@ TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {
 
 TEST_F(DBCompactionTest, MinorCompactionsHappen) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.write_buffer_size = 10000;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     const int N = 500;
@@ -689,7 +654,7 @@ TEST_F(DBCompactionTest, MinorCompactionsHappen) {
 }
 
 TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleLevel;
   options.level0_file_num_compaction_trigger = 3;
 
@@ -703,7 +668,6 @@ TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
   const size_t key_len =
     static_cast<size_t>(compact_opt.output_file_size_limit) / 5;
 
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   std::vector<const Snapshot*> snaps;
@@ -747,10 +711,8 @@ TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
 // if the database is shutdown during the memtable compaction.
 TEST_F(DBCompactionTest, RecoverDuringMemtableCompaction) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.env = env_;
-    options.write_buffer_size = 1000000;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Trigger a long memtable compaction and reopen the database during it
@@ -774,10 +736,9 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
       [&](void* arg) { trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 100000000;
   options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   int32_t num_keys = 80;
@@ -995,7 +956,9 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
   bool second = true;
   rocksdb::SyncPoint::GetInstance()->LoadDependency(
       {{"DBCompaction::ManualPartial:4", "DBCompaction::ManualPartial:1"},
-       {"DBCompaction::ManualPartial:2", "DBCompaction::ManualPartial:3"}});
+       {"DBCompaction::ManualPartial:2", "DBCompaction::ManualPartial:3"},
+       {"DBCompaction::ManualPartial:5",
+        "DBImpl::BackgroundCompaction:NonTrivial:AfterRun"}});
   rocksdb::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
         if (first) {
@@ -1106,6 +1069,8 @@ TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
 
   // 3 files in L0
   ASSERT_EQ("3,0,0,0,0,1,2", FilesPerLevel(0));
+  TEST_SYNC_POINT("DBCompaction::ManualPartial:5");
+
   // 1 file in L6, 1 file in L1
   dbfull()->TEST_WaitForFlushMemTable();
   dbfull()->TEST_WaitForCompact();
@@ -1369,10 +1334,9 @@ TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
       [&](void* arg) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 100000000;
   options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   int32_t value_size = 10 * 1024;  // 10 KB
@@ -1661,7 +1625,7 @@ TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
   int max_key_universal_insert = 600;
 
   // Stage 1: generate a db with level compaction
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 110 << 10;  // 110KB
   options.arena_block_size = 4 << 10;
   options.num_levels = 4;
@@ -1671,7 +1635,6 @@ TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
   options.target_file_size_base = 200 << 10;  // 200KB
   options.target_file_size_multiplier = 1;
   options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
   for (int i = 0; i <= max_key_level_insert; i++) {
@@ -2386,10 +2349,9 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
       [&](void* arg) { non_trivial_move++; });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 100000000;
   options.max_subcompactions = max_subcompactions_;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   int32_t value_size = 10 * 1024;  // 10 KB
@@ -2479,14 +2441,14 @@ class CompactionPriTest : public DBTestBase,
 };
 
 TEST_P(CompactionPriTest, Test) {
-  Options options;
+  Options options = CurrentOptions();
   options.write_buffer_size = 16 * 1024;
   options.compaction_pri = static_cast<CompactionPri>(compaction_pri_);
   options.hard_pending_compaction_bytes_limit = 256 * 1024;
   options.max_bytes_for_level_base = 64 * 1024;
   options.max_bytes_for_level_multiplier = 4;
   options.compression = kNoCompression;
-  options = CurrentOptions(options);
+
   DestroyAndReopen(options);
 
   Random rnd(301);

db/db_dynamic_level_test.cc

@@ -125,15 +125,15 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
 
   Options options = CurrentOptions();
   options.create_if_missing = true;
-  options.db_write_buffer_size = 2048;
-  options.write_buffer_size = 2048;
+  options.db_write_buffer_size = 204800;
+  options.write_buffer_size = 20480;
   options.max_write_buffer_number = 2;
   options.level0_file_num_compaction_trigger = 2;
   options.level0_slowdown_writes_trigger = 9999;
   options.level0_stop_writes_trigger = 9999;
-  options.target_file_size_base = 2048;
+  options.target_file_size_base = 9102;
   options.level_compaction_dynamic_level_bytes = true;
-  options.max_bytes_for_level_base = 10240;
+  options.max_bytes_for_level_base = 40960;
   options.max_bytes_for_level_multiplier = 4;
   options.max_background_compactions = 2;
   options.num_levels = 5;
@@ -154,10 +154,10 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(4U, int_prop);
 
-  // Put about 7K to L0
+  // Put about 28K to L0
   for (int i = 0; i < 70; i++) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
   }
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "false"},
@@ -167,14 +167,14 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(4U, int_prop);
 
-  // Insert extra about 3.5K to L0. After they are compacted to L4, base level
+  // Insert extra about 28K to L0. After they are compacted to L4, base level
   // should be changed to L3.
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
   for (int i = 0; i < 70; i++) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
   }
 
   ASSERT_OK(dbfull()->SetOptions({
@@ -199,10 +199,10 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
-  // Write about 10K more
+  // Write about 40K more
   for (int i = 0; i < 100; i++) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
   }
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "false"},
@@ -218,15 +218,15 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
 
   // 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 10K, 40K, 160K, add triggers
+  // We try to make last levels' targets to be 40K, 160K, 640K, add triggers
   // another compaction from 40K->160K.
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "true"},
   }));
-  // Write about 150K more
-  for (int i = 0; i < 1350; i++) {
+  // Write about 600K more
+  for (int i = 0; i < 1500; i++) {
     ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                  RandomString(&rnd, 80)));
+                  RandomString(&rnd, 380)));
   }
   ASSERT_OK(dbfull()->SetOptions({
       {"disable_auto_compactions", "false"},
@@ -236,29 +236,39 @@ TEST_F(DBTestDynamicLevel, DynamicLevelMaxBytesBase2) {
   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()->EnableProcessing();
-  for (int attempt = 0; attempt <= 20; attempt++) {
-    // Write about 5K more data with two flushes. It should be flush to level 2
-    // but when it is applied, base level is already 1.
-    for (int i = 0; i < 50; i++) {
-      ASSERT_OK(Put(Key(static_cast<int>(rnd.Uniform(kMaxKey))),
-                    RandomString(&rnd, 80)));
-    }
-    Flush();
-
-    ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
-    if (int_prop == 2U) {
-      env_->SleepForMicroseconds(50000);
-    } else {
-      break;
-    }
-  }
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
   rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 
-  env_->SleepForMicroseconds(200000);
+  rocksdb::SyncPoint::GetInstance()->LoadDependency({
+      {"CompactionJob::Run():Start", "DynamicLevelMaxBytesBase2:0"},
+      {"DynamicLevelMaxBytesBase2:1", "CompactionJob::Run():End"},
+      {"DynamicLevelMaxBytesBase2:compact_range_finish",
+       "FlushJob::WriteLevel0Table"},
+  });
+  rocksdb::SyncPoint::GetInstance()->EnableProcessing();
+
+  std::thread thread([this] {
+    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_start");
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:compact_range_finish");
+  });
+
+  TEST_SYNC_POINT("DynamicLevelMaxBytesBase2:0");
+  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");
+
+  Flush();
+
+  thread.join();
+
+  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
+  rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
 
   ASSERT_TRUE(db_->GetIntProperty("rocksdb.base-level", &int_prop));
   ASSERT_EQ(1U, int_prop);

db/db_impl.cc

@@ -12,12 +12,14 @@
 #ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
 #endif
-
 #include <inttypes.h>
 #include <stdint.h>
 #ifdef OS_SOLARIS
 #include <alloca.h>
 #endif
+#ifdef ROCKSDB_JEMALLOC
+#include "jemalloc/jemalloc.h"
+#endif
 
 #include <algorithm>
 #include <climits>
@@ -88,10 +90,10 @@
 #include "util/log_buffer.h"
 #include "util/logging.h"
 #include "util/mutexlock.h"
-#include "util/sst_file_manager_impl.h"
 #include "util/options_helper.h"
 #include "util/options_parser.h"
 #include "util/perf_context_imp.h"
+#include "util/sst_file_manager_impl.h"
 #include "util/stop_watch.h"
 #include "util/string_util.h"
 #include "util/sync_point.h"
@@ -167,6 +169,17 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     result.recycle_log_file_num = false;
   }
 
+  if (result.recycle_log_file_num &&
+      (result.wal_recovery_mode == WALRecoveryMode::kPointInTimeRecovery ||
+       result.wal_recovery_mode == WALRecoveryMode::kAbsoluteConsistency)) {
+    // kPointInTimeRecovery is indistinguishable from
+    // kTolerateCorruptedTailRecords in recycle mode since we define
+    // the "end" of the log as the first corrupt record we encounter.
+    // kAbsoluteConsistency doesn't make sense because even a clean
+    // shutdown leaves old junk at the end of the log file.
+    result.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
+  }
+
   if (result.wal_dir.empty()) {
     // Use dbname as default
     result.wal_dir = dbname;
@@ -201,7 +214,47 @@ Status SanitizeOptionsByTable(
   return Status::OK();
 }
 
-CompressionType GetCompressionFlush(const ImmutableCFOptions& ioptions) {
+static Status ValidateOptions(
+    const DBOptions& db_options,
+    const std::vector<ColumnFamilyDescriptor>& column_families) {
+  Status s;
+
+  for (auto& cfd : column_families) {
+    s = CheckCompressionSupported(cfd.options);
+    if (s.ok() && db_options.allow_concurrent_memtable_write) {
+      s = CheckConcurrentWritesSupported(cfd.options);
+    }
+    if (!s.ok()) {
+      return s;
+    }
+    if (db_options.db_paths.size() > 1) {
+      if ((cfd.options.compaction_style != kCompactionStyleUniversal) &&
+          (cfd.options.compaction_style != kCompactionStyleLevel)) {
+        return Status::NotSupported(
+            "More than one DB paths are only supported in "
+            "universal and level compaction styles. ");
+      }
+    }
+  }
+
+  if (db_options.db_paths.size() > 4) {
+    return Status::NotSupported(
+        "More than four DB paths are not supported yet. ");
+  }
+
+  if (db_options.allow_mmap_reads && !db_options.allow_os_buffer) {
+    // Protect against assert in PosixMMapReadableFile constructor
+    return Status::NotSupported(
+        "If memory mapped reads (allow_mmap_reads) are enabled "
+        "then os caching (allow_os_buffer) must also be enabled. ");
+  }
+
+  return Status::OK();
+}
+
+CompressionType GetCompressionFlush(
+    const ImmutableCFOptions& ioptions,
+    const MutableCFOptions& mutable_cf_options) {
   // Compressing memtable flushes might not help unless the sequential load
   // optimization is used for leveled compaction. Otherwise the CPU and
   // latency overhead is not offset by saving much space.
@@ -218,7 +271,7 @@ CompressionType GetCompressionFlush(const ImmutableCFOptions& ioptions) {
   }
 
   if (can_compress) {
-    return ioptions.compression;
+    return mutable_cf_options.compression;
   } else {
     return kNoCompression;
   }
@@ -405,6 +458,25 @@ DBImpl::~DBImpl() {
   }
   logs_.clear();
 
+  // Table cache may have table handles holding blocks from the block cache.
+  // We need to release them before the block cache is destroyed. The block
+  // cache may be destroyed inside versions_.reset(), when column family data
+  // list is destroyed, so leaving handles in table cache after
+  // versions_.reset() may cause issues.
+  // Here we clean all unreferenced handles in table cache.
+  // Now we assume all user queries have finished, so only version set itself
+  // can possibly hold the blocks from block cache. After releasing unreferenced
+  // handles here, only handles held by version set left and inside
+  // versions_.reset(), we will release them. There, we need to make sure every
+  // time a handle is released, we erase it from the cache too. By doing that,
+  // we can guarantee that after versions_.reset(), table cache is empty
+  // so the cache can be safely destroyed.
+  table_cache_->EraseUnRefEntries();
+
+  for (auto& txn_entry : recovered_transactions_) {
+    delete txn_entry.second;
+  }
+
   // versions need to be destroyed before table_cache since it can hold
   // references to table_cache.
   versions_.reset();
@@ -481,6 +553,39 @@ void DBImpl::PrintStatistics() {
   }
 }
 
+#ifndef ROCKSDB_LITE
+#ifdef ROCKSDB_JEMALLOC
+typedef struct {
+  char* cur;
+  char* end;
+} MallocStatus;
+
+static void GetJemallocStatus(void* mstat_arg, const char* status) {
+  MallocStatus* mstat = reinterpret_cast<MallocStatus*>(mstat_arg);
+  size_t status_len = status ? strlen(status) : 0;
+  size_t buf_size = (size_t)(mstat->end - mstat->cur);
+  if (!status_len || status_len > buf_size) {
+    return;
+  }
+
+  snprintf(mstat->cur, buf_size, "%s", status);
+  mstat->cur += status_len;
+}
+#endif  // ROCKSDB_JEMALLOC
+
+static void DumpMallocStats(std::string* stats) {
+#ifdef ROCKSDB_JEMALLOC
+  MallocStatus mstat;
+  const uint kMallocStatusLen = 1000000;
+  std::unique_ptr<char> buf{new char[kMallocStatusLen + 1]};
+  mstat.cur = buf.get();
+  mstat.end = buf.get() + kMallocStatusLen;
+  malloc_stats_print(GetJemallocStatus, &mstat, "");
+  stats->append(buf.get());
+#endif  // ROCKSDB_JEMALLOC
+}
+#endif  // !ROCKSDB_LITE
+
 void DBImpl::MaybeDumpStats() {
   if (db_options_.stats_dump_period_sec == 0) return;
 
@@ -513,6 +618,9 @@ void DBImpl::MaybeDumpStats() {
       default_cf_internal_stats_->GetStringProperty(
           *db_property_info, DB::Properties::kDBStats, &stats);
     }
+    if (db_options_.dump_malloc_stats) {
+      DumpMallocStats(&stats);
+    }
     Log(InfoLogLevel::WARN_LEVEL,
         db_options_.info_log, "------- DUMPING STATS -------");
     Log(InfoLogLevel::WARN_LEVEL,
@@ -523,6 +631,79 @@ void DBImpl::MaybeDumpStats() {
   }
 }
 
+uint64_t DBImpl::FindMinPrepLogReferencedByMemTable() {
+  uint64_t min_log = 0;
+
+  // we must look through the memtables for two phase transactions
+  // that have been committed but not yet flushed
+  for (auto loop_cfd : *versions_->GetColumnFamilySet()) {
+    if (loop_cfd->IsDropped()) {
+      continue;
+    }
+
+    auto log = loop_cfd->imm()->GetMinLogContainingPrepSection();
+
+    if (log > 0 && (min_log == 0 || log < min_log)) {
+      min_log = log;
+    }
+
+    log = loop_cfd->mem()->GetMinLogContainingPrepSection();
+
+    if (log > 0 && (min_log == 0 || log < min_log)) {
+      min_log = log;
+    }
+  }
+
+  return min_log;
+}
+
+void DBImpl::MarkLogAsHavingPrepSectionFlushed(uint64_t log) {
+  assert(log != 0);
+  std::lock_guard<std::mutex> lock(prep_heap_mutex_);
+  auto it = prepared_section_completed_.find(log);
+  assert(it != prepared_section_completed_.end());
+  it->second += 1;
+}
+
+void DBImpl::MarkLogAsContainingPrepSection(uint64_t log) {
+  assert(log != 0);
+  std::lock_guard<std::mutex> lock(prep_heap_mutex_);
+  min_log_with_prep_.push(log);
+  auto it = prepared_section_completed_.find(log);
+  if (it == prepared_section_completed_.end()) {
+    prepared_section_completed_[log] = 0;
+  }
+}
+
+uint64_t DBImpl::FindMinLogContainingOutstandingPrep() {
+  std::lock_guard<std::mutex> lock(prep_heap_mutex_);
+  uint64_t min_log = 0;
+
+  // first we look in the prepared heap where we keep
+  // track of transactions that have been prepared (written to WAL)
+  // but not yet committed.
+  while (!min_log_with_prep_.empty()) {
+    min_log = min_log_with_prep_.top();
+
+    auto it = prepared_section_completed_.find(min_log);
+
+    // value was marked as 'deleted' from heap
+    if (it != prepared_section_completed_.end() && it->second > 0) {
+      it->second -= 1;
+      min_log_with_prep_.pop();
+
+      // back to squere one...
+      min_log = 0;
+      continue;
+    } else {
+      // found a valid value
+      break;
+    }
+  }
+
+  return min_log;
+}
+
 // * 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
@@ -572,6 +753,7 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
   // Get obsolete files.  This function will also update the list of
   // pending files in VersionSet().
   versions_->GetObsoleteFiles(&job_context->sst_delete_files,
+                              &job_context->manifest_delete_files,
                               job_context->min_pending_output);
 
   // store the current filenum, lognum, etc
@@ -579,6 +761,32 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
   job_context->pending_manifest_file_number =
       versions_->pending_manifest_file_number();
   job_context->log_number = versions_->MinLogNumber();
+
+  if (allow_2pc()) {
+    // if are 2pc we must consider logs containing prepared
+    // sections of outstanding transactions.
+    //
+    // We must check min logs with outstanding prep before we check
+    // logs referneces by memtables because a log referenced by the
+    // first data structure could transition to the second under us.
+    //
+    // TODO(horuff): iterating over all column families under db mutex.
+    // should find more optimial solution
+    auto min_log_in_prep_heap = FindMinLogContainingOutstandingPrep();
+
+    if (min_log_in_prep_heap != 0 &&
+        min_log_in_prep_heap < job_context->log_number) {
+      job_context->log_number = min_log_in_prep_heap;
+    }
+
+    auto min_log_refed_by_mem = FindMinPrepLogReferencedByMemTable();
+
+    if (min_log_refed_by_mem != 0 &&
+        min_log_refed_by_mem < job_context->log_number) {
+      job_context->log_number = min_log_refed_by_mem;
+    }
+  }
+
   job_context->prev_log_number = versions_->prev_log_number();
 
   versions_->AddLiveFiles(&job_context->sst_live);
@@ -616,7 +824,7 @@ void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
   }
 
   if (!alive_log_files_.empty()) {
-    uint64_t min_log_number = versions_->MinLogNumber();
+    uint64_t min_log_number = job_context->log_number;
     // find newly obsoleted log files
     while (alive_log_files_.begin()->number < min_log_number) {
       auto& earliest = *alive_log_files_.begin();
@@ -689,9 +897,9 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state) {
   }
 
   auto candidate_files = state.full_scan_candidate_files;
-  candidate_files.reserve(candidate_files.size() +
-                          state.sst_delete_files.size() +
-                          state.log_delete_files.size());
+  candidate_files.reserve(
+      candidate_files.size() + state.sst_delete_files.size() +
+      state.log_delete_files.size() + state.manifest_delete_files.size());
   // We may ignore the dbname when generating the file names.
   const char* kDumbDbName = "";
   for (auto file : state.sst_delete_files) {
@@ -707,12 +915,17 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state) {
                                    0);
     }
   }
+  for (const auto& filename : state.manifest_delete_files) {
+    candidate_files.emplace_back(filename, 0);
+  }
 
   // dedup state.candidate_files so we don't try to delete the same
   // file twice
-  sort(candidate_files.begin(), candidate_files.end(), CompareCandidateFile);
-  candidate_files.erase(unique(candidate_files.begin(), candidate_files.end()),
-                        candidate_files.end());
+  std::sort(candidate_files.begin(), candidate_files.end(),
+            CompareCandidateFile);
+  candidate_files.erase(
+      std::unique(candidate_files.begin(), candidate_files.end()),
+      candidate_files.end());
 
   std::vector<std::string> old_info_log_files;
   InfoLogPrefix info_log_prefix(!db_options_.db_log_dir.empty(), dbname_);
@@ -825,7 +1038,8 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state) {
 
   // Delete old info log files.
   size_t old_info_log_file_count = old_info_log_files.size();
-  if (old_info_log_file_count >= db_options_.keep_log_file_num) {
+  if (old_info_log_file_count != 0 &&
+      old_info_log_file_count >= db_options_.keep_log_file_num) {
     std::sort(old_info_log_files.begin(), old_info_log_files.end());
     size_t end = old_info_log_file_count - db_options_.keep_log_file_num;
     for (unsigned int i = 0; i <= end; i++) {
@@ -930,7 +1144,7 @@ Directory* DBImpl::Directories::GetDataDir(size_t path_id) {
 
 Status DBImpl::Recover(
     const std::vector<ColumnFamilyDescriptor>& column_families, bool read_only,
-    bool error_if_log_file_exist) {
+    bool error_if_log_file_exist, bool error_if_data_exists_in_logs) {
   mutex_.AssertHeld();
 
   bool is_new_db = false;
@@ -987,7 +1201,7 @@ Status DBImpl::Recover(
     s = CheckConsistency();
   }
   if (s.ok()) {
-    SequenceNumber max_sequence(kMaxSequenceNumber);
+    SequenceNumber next_sequence(kMaxSequenceNumber);
     default_cf_handle_ = new ColumnFamilyHandleImpl(
         versions_->GetColumnFamilySet()->GetDefault(), this, &mutex_);
     default_cf_internal_stats_ = default_cf_handle_->cfd()->internal_stats();
@@ -1001,8 +1215,6 @@ Status DBImpl::Recover(
     // Note that prev_log_number() is no longer used, but we pay
     // attention to it in case we are recovering a database
     // produced by an older version of rocksdb.
-    const uint64_t min_log = versions_->MinLogNumber();
-    const uint64_t prev_log = versions_->prev_log_number();
     std::vector<std::string> filenames;
     s = env_->GetChildren(db_options_.wal_dir, &filenames);
     if (!s.ok()) {
@@ -1019,22 +1231,37 @@ Status DBImpl::Recover(
               "While creating a new Db, wal_dir contains "
               "existing log file: ",
               filenames[i]);
-        } else if ((number >= min_log) || (number == prev_log)) {
+        } else {
           logs.push_back(number);
         }
       }
     }
 
-    if (logs.size() > 0 && error_if_log_file_exist) {
-      return Status::Corruption(""
-          "The db was opened in readonly mode with error_if_log_file_exist"
-          "flag but a log file already exists");
+    if (logs.size() > 0) {
+      if (error_if_log_file_exist) {
+        return Status::Corruption(
+            "The db was opened in readonly mode with error_if_log_file_exist"
+            "flag but a log file already exists");
+      } else if (error_if_data_exists_in_logs) {
+        for (auto& log : logs) {
+          std::string fname = LogFileName(db_options_.wal_dir, log);
+          uint64_t bytes;
+          s = env_->GetFileSize(fname, &bytes);
+          if (s.ok()) {
+            if (bytes > 0) {
+              return Status::Corruption(
+                  "error_if_data_exists_in_logs is set but there are data "
+                  " in log files.");
+            }
+          }
+        }
+      }
     }
 
     if (!logs.empty()) {
       // Recover in the order in which the logs were generated
       std::sort(logs.begin(), logs.end());
-      s = RecoverLogFiles(logs, &max_sequence, read_only);
+      s = RecoverLogFiles(logs, &next_sequence, read_only);
       if (!s.ok()) {
         // Clear memtables if recovery failed
         for (auto cfd : *versions_->GetColumnFamilySet()) {
@@ -1059,7 +1286,7 @@ Status DBImpl::Recover(
 
 // REQUIRES: log_numbers are sorted in ascending order
 Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
-                               SequenceNumber* max_sequence, bool read_only) {
+                               SequenceNumber* next_sequence, bool read_only) {
   struct LogReporter : public log::Reader::Reporter {
     Env* env;
     Logger* info_log;
@@ -1098,6 +1325,23 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
     stream.EndArray();
   }
 
+#ifndef ROCKSDB_LITE
+  if (db_options_.wal_filter != nullptr) {
+    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_lognumber_map.insert(
+        std::make_pair(cfd->GetID(), cfd->GetLogNumber()));
+    }
+
+    db_options_.wal_filter->ColumnFamilyLogNumberMap(
+      cf_lognumber_map,
+      cf_name_id_map);
+  }
+#endif
+
   bool continue_replay_log = true;
   for (auto log_number : log_numbers) {
     // The previous incarnation may not have written any MANIFEST
@@ -1164,7 +1408,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
         continue_replay_log &&
         reader.ReadRecord(&record, &scratch, db_options_.wal_recovery_mode) &&
         status.ok()) {
-      if (record.size() < 12) {
+      if (record.size() < WriteBatchInternal::kHeader) {
         reporter.Corruption(record.size(),
                             Status::Corruption("log record too small"));
         continue;
@@ -1177,8 +1421,8 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
         bool batch_changed = false;
 
         WalFilter::WalProcessingOption wal_processing_option =
-            db_options_.wal_filter->LogRecord(batch, &new_batch,
-                                              &batch_changed);
+            db_options_.wal_filter->LogRecordFound(log_number, fname, batch,
+                                              &new_batch, &batch_changed);
 
         switch (wal_processing_option) {
           case WalFilter::WalProcessingOption::kContinueProcessing:
@@ -1245,15 +1489,19 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
       }
 #endif  // ROCKSDB_LITE
 
+      if (*next_sequence == kMaxSequenceNumber) {
+        *next_sequence = WriteBatchInternal::Sequence(&batch);
+      }
+      WriteBatchInternal::SetSequence(&batch, *next_sequence);
+
       // If column family was not found, it might mean that the WAL write
       // batch references to the column family that was dropped after the
       // insert. We don't want to fail the whole write batch in that case --
       // we just ignore the update.
       // That's why we set ignore missing column families to true
-      status =
-          WriteBatchInternal::InsertInto(&batch, column_family_memtables_.get(),
-                                         &flush_scheduler_, true, log_number);
-
+      status = WriteBatchInternal::InsertInto(
+          &batch, column_family_memtables_.get(), &flush_scheduler_, true,
+          log_number, this, true, false, next_sequence);
       MaybeIgnoreError(&status);
       if (!status.ok()) {
         // We are treating this as a failure while reading since we read valid
@@ -1262,12 +1510,6 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
         continue;
       }
 
-      const SequenceNumber last_seq = WriteBatchInternal::Sequence(&batch) +
-                                      WriteBatchInternal::Count(&batch) - 1;
-      if ((*max_sequence == kMaxSequenceNumber) || (last_seq > *max_sequence)) {
-        *max_sequence = last_seq;
-      }
-
       if (!read_only) {
         // we can do this because this is called before client has access to the
         // DB and there is only a single thread operating on DB
@@ -1289,7 +1531,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
           }
 
           cfd->CreateNewMemtable(*cfd->GetLatestMutableCFOptions(),
-                                 *max_sequence);
+                                 *next_sequence);
         }
       }
     }
@@ -1307,7 +1549,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
 
         Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
             "Point in time recovered to log #%" PRIu64 " seq #%" PRIu64,
-            log_number, *max_sequence);
+            log_number, *next_sequence);
       } else {
         assert(db_options_.wal_recovery_mode ==
                   WALRecoveryMode::kTolerateCorruptedTailRecords
@@ -1318,9 +1560,10 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
     }
 
     flush_scheduler_.Clear();
-    if ((*max_sequence != kMaxSequenceNumber) &&
-        (versions_->LastSequence() < *max_sequence)) {
-      versions_->SetLastSequence(*max_sequence);
+    auto last_sequence = *next_sequence - 1;
+    if ((*next_sequence != kMaxSequenceNumber) &&
+        (versions_->LastSequence() <= last_sequence)) {
+      versions_->SetLastSequence(last_sequence);
     }
   }
 
@@ -1352,7 +1595,7 @@ Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
         }
 
         cfd->CreateNewMemtable(*cfd->GetLatestMutableCFOptions(),
-                               *max_sequence);
+                               *next_sequence);
       }
 
       // write MANIFEST with update
@@ -1402,43 +1645,33 @@ Status DBImpl::WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
         " Level-0 table #%" PRIu64 ": started",
         cfd->GetName().c_str(), meta.fd.GetNumber());
 
+    // Get the latest mutable cf options while the mutex is still locked
+    const MutableCFOptions mutable_cf_options =
+        *cfd->GetLatestMutableCFOptions();
     bool paranoid_file_checks =
         cfd->GetLatestMutableCFOptions()->paranoid_file_checks;
     {
       mutex_.Unlock();
-      TableFileCreationInfo info;
 
       SequenceNumber earliest_write_conflict_snapshot;
       std::vector<SequenceNumber> snapshot_seqs =
           snapshots_.GetAll(&earliest_write_conflict_snapshot);
 
       s = BuildTable(
-          dbname_, env_, *cfd->ioptions(), env_options_, cfd->table_cache(),
-          iter.get(), &meta, cfd->internal_comparator(),
-          cfd->int_tbl_prop_collector_factories(), cfd->GetID(), snapshot_seqs,
-          earliest_write_conflict_snapshot,
-          GetCompressionFlush(*cfd->ioptions()),
+          dbname_, env_, *cfd->ioptions(), mutable_cf_options, env_options_,
+          cfd->table_cache(), iter.get(), &meta, cfd->internal_comparator(),
+          cfd->int_tbl_prop_collector_factories(), cfd->GetID(), cfd->GetName(),
+          snapshot_seqs, earliest_write_conflict_snapshot,
+          GetCompressionFlush(*cfd->ioptions(), mutable_cf_options),
           cfd->ioptions()->compression_opts, paranoid_file_checks,
-          cfd->internal_stats(), Env::IO_HIGH, &info.table_properties);
+          cfd->internal_stats(), TableFileCreationReason::kRecovery,
+          &event_logger_, job_id);
       LogFlush(db_options_.info_log);
       Log(InfoLogLevel::DEBUG_LEVEL, db_options_.info_log,
           "[%s] [WriteLevel0TableForRecovery]"
           " Level-0 table #%" PRIu64 ": %" PRIu64 " bytes %s",
           cfd->GetName().c_str(), meta.fd.GetNumber(), meta.fd.GetFileSize(),
           s.ToString().c_str());
-
-      // output to event logger
-      if (s.ok()) {
-        info.db_name = dbname_;
-        info.cf_name = cfd->GetName();
-        info.file_path = TableFileName(db_options_.db_paths,
-                                       meta.fd.GetNumber(),
-                                       meta.fd.GetPathId());
-        info.file_size = meta.fd.GetFileSize();
-        info.job_id = job_id;
-        EventHelpers::LogAndNotifyTableFileCreation(
-            &event_logger_, db_options_.listeners, meta.fd, info);
-      }
       mutex_.Lock();
     }
   }
@@ -1481,7 +1714,8 @@ Status DBImpl::FlushMemTableToOutputFile(
       versions_.get(), &mutex_, &shutting_down_, snapshot_seqs,
       earliest_write_conflict_snapshot, job_context, log_buffer,
       directories_.GetDbDir(), directories_.GetDataDir(0U),
-      GetCompressionFlush(*cfd->ioptions()), stats_, &event_logger_);
+      GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
+      &event_logger_, mutable_cf_options.report_bg_io_stats);
 
   FileMetaData file_meta;
 
@@ -1510,7 +1744,6 @@ Status DBImpl::FlushMemTableToOutputFile(
     // true, mark DB read-only
     bg_error_ = s;
   }
-  RecordFlushIOStats();
   if (s.ok()) {
 #ifndef ROCKSDB_LITE
     // may temporarily unlock and lock the mutex.
@@ -1830,7 +2063,7 @@ Status DBImpl::CompactFilesImpl(
       directories_.GetDataDir(c->output_path_id()), stats_, &mutex_, &bg_error_,
       snapshot_seqs, earliest_write_conflict_snapshot, table_cache_,
       &event_logger_, c->mutable_cf_options()->paranoid_file_checks,
-      c->mutable_cf_options()->compaction_measure_io_stats, dbname_,
+      c->mutable_cf_options()->report_bg_io_stats, dbname_,
       nullptr);  // Here we pass a nullptr for CompactionJobStats because
                  // CompactFiles does not trigger OnCompactionCompleted(),
                  // which is the only place where CompactionJobStats is
@@ -1844,6 +2077,13 @@ Status DBImpl::CompactFilesImpl(
                  // support for CompactFiles, we should have CompactFiles API
                  // pass a pointer of CompactionJobStats as the out-value
                  // instead of using EventListener.
+
+  // Creating a compaction influences the compaction score because the score
+  // takes running compactions into account (by skipping files that are already
+  // being compacted). Since we just changed compaction score, we recalculate it
+  // here.
+  version->storage_info()->ComputeCompactionScore(*c->mutable_cf_options());
+
   compaction_job.Prepare();
 
   mutex_.Unlock();
@@ -1929,6 +2169,7 @@ void DBImpl::NotifyOnCompactionCompleted(
   }
   // release lock while notifying events
   mutex_.Unlock();
+  TEST_SYNC_POINT("DBImpl::NotifyOnCompactionCompleted::UnlockMutex");
   {
     CompactionJobInfo info;
     info.cf_name = cfd->GetName();
@@ -1940,6 +2181,7 @@ void DBImpl::NotifyOnCompactionCompleted(
     info.stats = compaction_job_stats;
     info.table_properties = c->GetOutputTableProperties();
     info.compaction_reason = c->compaction_reason();
+    info.compression = c->output_compression();
     for (size_t i = 0; i < c->num_input_levels(); ++i) {
       for (const auto fmd : *c->inputs(i)) {
         auto fn = TableFileName(db_options_.db_paths, fmd->fd.GetNumber(),
@@ -2568,11 +2810,6 @@ void DBImpl::SchedulePendingCompaction(ColumnFamilyData* cfd) {
   }
 }
 
-void DBImpl::RecordFlushIOStats() {
-  RecordTick(stats_, FLUSH_WRITE_BYTES, IOSTATS(bytes_written));
-  IOSTATS_RESET(bytes_written);
-}
-
 void DBImpl::BGWorkFlush(void* db) {
   IOSTATS_SET_THREAD_POOL_ID(Env::Priority::HIGH);
   TEST_SYNC_POINT("DBImpl::BGWorkFlush");
@@ -2706,7 +2943,6 @@ void DBImpl::BackgroundCallFlush() {
     bg_flush_scheduled_--;
     // See if there's more work to be done
     MaybeScheduleFlushOrCompaction();
-    RecordFlushIOStats();
     bg_cv_.SignalAll();
     // IMPORTANT: there should be no code after calling SignalAll. This call may
     // signal the DB destructor that it's OK to proceed with destruction. In
@@ -3025,7 +3261,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
         &bg_error_, snapshot_seqs, earliest_write_conflict_snapshot,
         table_cache_, &event_logger_,
         c->mutable_cf_options()->paranoid_file_checks,
-        c->mutable_cf_options()->compaction_measure_io_stats, dbname_,
+        c->mutable_cf_options()->report_bg_io_stats, dbname_,
         &compaction_job_stats);
     compaction_job.Prepare();
 
@@ -3042,11 +3278,11 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
     *made_progress = true;
   }
   if (c != nullptr) {
+    c->ReleaseCompactionFiles(status);
+    *made_progress = true;
     NotifyOnCompactionCompleted(
         c->column_family_data(), c.get(), status,
         compaction_job_stats, job_context->job_id);
-    c->ReleaseCompactionFiles(status);
-    *made_progress = true;
   }
   // this will unref its input_version and column_family_data
   c.reset();
@@ -4137,19 +4373,21 @@ Status DBImpl::SingleDelete(const WriteOptions& write_options,
 }
 
 Status DBImpl::Write(const WriteOptions& write_options, WriteBatch* my_batch) {
-  return WriteImpl(write_options, my_batch, nullptr);
+  return WriteImpl(write_options, my_batch, nullptr, nullptr);
 }
 
 #ifndef ROCKSDB_LITE
 Status DBImpl::WriteWithCallback(const WriteOptions& write_options,
                                  WriteBatch* my_batch,
                                  WriteCallback* callback) {
-  return WriteImpl(write_options, my_batch, callback);
+  return WriteImpl(write_options, my_batch, callback, nullptr);
 }
 #endif  // ROCKSDB_LITE
 
 Status DBImpl::WriteImpl(const WriteOptions& write_options,
-                         WriteBatch* my_batch, WriteCallback* callback) {
+                         WriteBatch* my_batch, WriteCallback* callback,
+                         uint64_t* log_used, uint64_t log_ref,
+                         bool disable_memtable) {
   if (my_batch == nullptr) {
     return Status::Corruption("Batch is nullptr!");
   }
@@ -4174,8 +4412,10 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
   w.batch = my_batch;
   w.sync = write_options.sync;
   w.disableWAL = write_options.disableWAL;
+  w.disable_memtable = disable_memtable;
   w.in_batch_group = false;
   w.callback = callback;
+  w.log_ref = log_ref;
 
   if (!write_options.disableWAL) {
     RecordTick(stats_, WRITE_WITH_WAL);
@@ -4188,12 +4428,16 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     // we are a non-leader in a parallel group
     PERF_TIMER_GUARD(write_memtable_time);
 
-    if (!w.CallbackFailed()) {
+    if (log_used != nullptr) {
+      *log_used = w.log_used;
+    }
+
+    if (w.ShouldWriteToMemtable()) {
       ColumnFamilyMemTablesImpl column_family_memtables(
           versions_->GetColumnFamilySet());
       WriteBatchInternal::SetSequence(w.batch, w.sequence);
       w.status = WriteBatchInternal::InsertInto(
-          w.batch, &column_family_memtables, &flush_scheduler_,
+          &w, &column_family_memtables, &flush_scheduler_,
           write_options.ignore_missing_column_families, 0 /*log_number*/, this,
           true /*dont_filter_deletes*/, true /*concurrent_memtable_writes*/);
     }
@@ -4211,6 +4455,9 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     status = w.FinalStatus();
   }
   if (w.state == WriteThread::STATE_COMPLETED) {
+    if (log_used != nullptr) {
+      *log_used = w.log_used;
+    }
     // write is complete and leader has updated sequence
     RecordTick(stats_, WRITE_DONE_BY_OTHER);
     return w.FinalStatus();
@@ -4368,10 +4615,15 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
     uint64_t total_byte_size = 0;
     for (auto writer : write_group) {
       if (writer->CheckCallback(this)) {
-        total_count += WriteBatchInternal::Count(writer->batch);
-        total_byte_size = WriteBatchInternal::AppendedByteSize(
-            total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
-        parallel = parallel && !writer->batch->HasMerge();
+        if (writer->ShouldWriteToMemtable()) {
+          total_count += WriteBatchInternal::Count(writer->batch);
+          parallel = parallel && !writer->batch->HasMerge();
+        }
+
+        if (writer->ShouldWriteToWAL()) {
+          total_byte_size = WriteBatchInternal::AppendedByteSize(
+              total_byte_size, WriteBatchInternal::ByteSize(writer->batch));
+        }
       }
     }
 
@@ -4393,22 +4645,27 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
       PERF_TIMER_GUARD(write_wal_time);
 
       WriteBatch* merged_batch = nullptr;
-      if (write_group.size() == 1 && !write_group[0]->CallbackFailed()) {
+      if (write_group.size() == 1 && write_group[0]->ShouldWriteToWAL()) {
         merged_batch = write_group[0]->batch;
+        write_group[0]->log_used = logfile_number_;
       } else {
         // WAL needs all of the batches flattened into a single batch.
         // We could avoid copying here with an iov-like AddRecord
         // interface
         merged_batch = &tmp_batch_;
         for (auto writer : write_group) {
-          if (!writer->CallbackFailed()) {
+          if (writer->ShouldWriteToWAL()) {
             WriteBatchInternal::Append(merged_batch, writer->batch);
           }
+          writer->log_used = logfile_number_;
         }
       }
-      WriteBatchInternal::SetSequence(merged_batch, current_sequence);
 
-      assert(WriteBatchInternal::Count(merged_batch) == total_count);
+      if (log_used != nullptr) {
+        *log_used = logfile_number_;
+      }
+
+      WriteBatchInternal::SetSequence(merged_batch, current_sequence);
 
       Slice log_entry = WriteBatchInternal::Contents(merged_batch);
       status = logs_.back().writer->AddRecord(log_entry);
@@ -4494,14 +4751,14 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
                          std::memory_order_relaxed);
         write_thread_.LaunchParallelFollowers(&pg, current_sequence);
 
-        if (!w.CallbackFailed()) {
+        if (w.ShouldWriteToMemtable()) {
           // do leader write
           ColumnFamilyMemTablesImpl column_family_memtables(
               versions_->GetColumnFamilySet());
           assert(w.sequence == current_sequence);
           WriteBatchInternal::SetSequence(w.batch, w.sequence);
           w.status = WriteBatchInternal::InsertInto(
-              w.batch, &column_family_memtables, &flush_scheduler_,
+              &w, &column_family_memtables, &flush_scheduler_,
               write_options.ignore_missing_column_families, 0 /*log_number*/,
               this, true /*dont_filter_deletes*/,
               true /*concurrent_memtable_writes*/);
@@ -4605,6 +4862,22 @@ Status DBImpl::ScheduleFlushes(WriteContext* context) {
   return Status::OK();
 }
 
+#ifndef ROCKSDB_LITE
+void DBImpl::NotifyOnMemTableSealed(ColumnFamilyData* cfd,
+                                    const MemTableInfo& mem_table_info) {
+  if (db_options_.listeners.size() == 0U) {
+    return;
+  }
+  if (shutting_down_.load(std::memory_order_acquire)) {
+    return;
+  }
+
+  for (auto listener : db_options_.listeners) {
+    listener->OnMemTableSealed(mem_table_info);
+  }
+}
+#endif  // ROCKSDB_LITE
+
 // REQUIRES: mutex_ is held
 // REQUIRES: this thread is currently at the front of the writer queue
 Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
@@ -4627,6 +4900,19 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
       creating_new_log ? versions_->NewFileNumber() : logfile_number_;
   SuperVersion* new_superversion = nullptr;
   const MutableCFOptions mutable_cf_options = *cfd->GetLatestMutableCFOptions();
+
+  // Set current_memtble_info for memtable sealed callback
+#ifndef ROCKSDB_LITE
+  MemTableInfo memtable_info;
+  memtable_info.cf_name = cfd->GetName();
+  memtable_info.first_seqno = cfd->mem()->GetFirstSequenceNumber();
+  memtable_info.earliest_seqno = cfd->mem()->GetEarliestSequenceNumber();
+  memtable_info.num_entries = cfd->mem()->num_entries();
+  memtable_info.num_deletes = cfd->mem()->num_deletes();
+#endif  // ROCKSDB_LITE
+  // Log this later after lock release. It may be outdated, e.g., if background
+  // flush happens before logging, but that should be ok.
+  int num_imm_unflushed = cfd->imm()->NumNotFlushed();
   mutex_.Unlock();
   Status s;
   {
@@ -4663,10 +4949,18 @@ Status DBImpl::SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context) {
       new_mem = cfd->ConstructNewMemtable(mutable_cf_options, seq);
       new_superversion = new SuperVersion();
     }
+
+#ifndef ROCKSDB_LITE
+    // PLEASE NOTE: We assume that there are no failable operations
+    // after lock is acquired below since we are already notifying
+    // client about mem table becoming immutable.
+    NotifyOnMemTableSealed(cfd, memtable_info);
+#endif //ROCKSDB_LITE
   }
-  Log(InfoLogLevel::DEBUG_LEVEL, db_options_.info_log,
-      "[%s] New memtable created with log file: #%" PRIu64 "\n",
-      cfd->GetName().c_str(), new_log_number);
+  Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
+      "[%s] New memtable created with log file: #%" PRIu64
+      ". Immutable memtables: %d.\n",
+      cfd->GetName().c_str(), new_log_number, num_imm_unflushed);
   mutex_.Lock();
   if (!s.ok()) {
     // how do we fail if we're not creating new log?
@@ -5340,27 +5634,9 @@ Status DB::Open(const DBOptions& db_options, const std::string& dbname,
     return s;
   }
 
-  for (auto& cfd : column_families) {
-    s = CheckCompressionSupported(cfd.options);
-    if (s.ok() && db_options.allow_concurrent_memtable_write) {
-      s = CheckConcurrentWritesSupported(cfd.options);
-    }
-    if (!s.ok()) {
-      return s;
-    }
-    if (db_options.db_paths.size() > 1) {
-      if ((cfd.options.compaction_style != kCompactionStyleUniversal) &&
-          (cfd.options.compaction_style != kCompactionStyleLevel)) {
-        return Status::NotSupported(
-            "More than one DB paths are only supported in "
-            "universal and level compaction styles. ");
-      }
-    }
-  }
-
-  if (db_options.db_paths.size() > 4) {
-    return Status::NotSupported(
-        "More than four DB paths are not supported yet. ");
+  s = ValidateOptions(db_options, column_families);
+  if (!s.ok()) {
+    return s;
   }
 
   *dbptr = nullptr;

db/db_impl.h

@@ -10,8 +10,10 @@
 
 #include <atomic>
 #include <deque>
+#include <functional>
 #include <limits>
 #include <list>
+#include <queue>
 #include <set>
 #include <string>
 #include <utility>
@@ -55,6 +57,7 @@ class Arena;
 class WriteCallback;
 struct JobContext;
 struct ExternalSstFileInfo;
+struct MemTableInfo;
 
 class DBImpl : public DB {
  public:
@@ -232,7 +235,7 @@ class DBImpl : public DB {
   // SST files will also be checked.
   //
   // If a key is found, *found_record_for_key will be set to true and
-  // *seq will will be set to the stored sequence number for the latest
+  // *seq will be set to the stored sequence number for the latest
   // operation on this key or kMaxSequenceNumber if unknown.
   // If no key is found, *found_record_for_key will be set to false.
   //
@@ -296,7 +299,8 @@ class DBImpl : public DB {
                            bool disallow_trivial_move = false);
 
   // Force current memtable contents to be flushed.
-  Status TEST_FlushMemTable(bool wait = true);
+  Status TEST_FlushMemTable(bool wait = true,
+                            ColumnFamilyHandle* cfh = nullptr);
 
   // Wait for memtable compaction
   Status TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family = nullptr);
@@ -341,10 +345,17 @@ class DBImpl : public DB {
   Status TEST_GetAllImmutableCFOptions(
       std::unordered_map<std::string, const ImmutableCFOptions*>* iopts_map);
 
+  // Return the lastest MutableCFOptions of of a column family
+  Status TEST_GetLatestMutableCFOptions(ColumnFamilyHandle* column_family,
+                                        MutableCFOptions* mutable_cf_opitons);
+
   Cache* TEST_table_cache() { return table_cache_.get(); }
 
   WriteController& TEST_write_controler() { return write_controller_; }
 
+  uint64_t TEST_FindMinLogContainingOutstandingPrep();
+  uint64_t TEST_FindMinPrepLogReferencedByMemTable();
+
 #endif  // NDEBUG
 
   // Return maximum background compaction alowed to be scheduled based on
@@ -421,12 +432,70 @@ class DBImpl : public DB {
     return num_running_compactions_;
   }
 
+  // hollow transactions shell used for recovery.
+  // these will then be passed to TransactionDB so that
+  // locks can be reacquired before writing can resume.
+  struct RecoveredTransaction {
+    uint64_t log_number_;
+    std::string name_;
+    WriteBatch* batch_;
+    explicit RecoveredTransaction(const uint64_t log, const std::string& name,
+                                  WriteBatch* batch)
+        : log_number_(log), name_(name), batch_(batch) {}
+
+    ~RecoveredTransaction() { delete batch_; }
+  };
+
+  bool allow_2pc() const { return db_options_.allow_2pc; }
+
+  std::unordered_map<std::string, RecoveredTransaction*>
+  recovered_transactions() {
+    return recovered_transactions_;
+  }
+
+  RecoveredTransaction* GetRecoveredTransaction(const std::string& name) {
+    auto it = recovered_transactions_.find(name);
+    if (it == recovered_transactions_.end()) {
+      return nullptr;
+    } else {
+      return it->second;
+    }
+  }
+
+  void InsertRecoveredTransaction(const uint64_t log, const std::string& name,
+                                  WriteBatch* batch) {
+    recovered_transactions_[name] = new RecoveredTransaction(log, name, batch);
+    MarkLogAsContainingPrepSection(log);
+  }
+
+  void DeleteRecoveredTransaction(const std::string& name) {
+    auto it = recovered_transactions_.find(name);
+    assert(it != recovered_transactions_.end());
+    auto* trx = it->second;
+    recovered_transactions_.erase(it);
+    MarkLogAsHavingPrepSectionFlushed(trx->log_number_);
+    delete trx;
+  }
+
+  void DeleteAllRecoveredTransactions() {
+    for (auto it = recovered_transactions_.begin();
+         it != recovered_transactions_.end(); it++) {
+      delete it->second;
+    }
+    recovered_transactions_.clear();
+  }
+
+  void MarkLogAsHavingPrepSectionFlushed(uint64_t log);
+  void MarkLogAsContainingPrepSection(uint64_t log);
+
  protected:
   Env* const env_;
   const std::string dbname_;
   unique_ptr<VersionSet> versions_;
   const DBOptions db_options_;
   Statistics* stats_;
+  std::unordered_map<std::string, RecoveredTransaction*>
+      recovered_transactions_;
 
   InternalIterator* NewInternalIterator(const ReadOptions&,
                                         ColumnFamilyData* cfd,
@@ -452,6 +521,8 @@ class DBImpl : public DB {
                                    Compaction *c, const Status &st,
                                    const CompactionJobStats& job_stats,
                                    int job_id);
+  void NotifyOnMemTableSealed(ColumnFamilyData* cfd, 
+                              const MemTableInfo& mem_table_info);
 
   void NewThreadStatusCfInfo(ColumnFamilyData* cfd) const;
 
@@ -460,11 +531,17 @@ class DBImpl : public DB {
   void EraseThreadStatusDbInfo() const;
 
   Status WriteImpl(const WriteOptions& options, WriteBatch* updates,
-                   WriteCallback* callback);
+                   WriteCallback* callback = nullptr,
+                   uint64_t* log_used = nullptr, uint64_t log_ref = 0,
+                   bool disable_memtable = false);
+
+  uint64_t FindMinLogContainingOutstandingPrep();
+  uint64_t FindMinPrepLogReferencedByMemTable();
 
  private:
   friend class DB;
   friend class InternalStats;
+  friend class TransactionImpl;
 #ifndef ROCKSDB_LITE
   friend class ForwardIterator;
 #endif
@@ -483,7 +560,8 @@ class DBImpl : public DB {
   // amount of work to recover recently logged updates.  Any changes to
   // be made to the descriptor are added to *edit.
   Status Recover(const std::vector<ColumnFamilyDescriptor>& column_families,
-                 bool read_only = false, bool error_if_log_file_exist = false);
+                 bool read_only = false, bool error_if_log_file_exist = false,
+                 bool error_if_data_exists_in_logs = false);
 
   void MaybeIgnoreError(Status* s) const;
 
@@ -523,7 +601,7 @@ class DBImpl : public DB {
                          SequenceNumber* max_sequence, bool read_only);
 
   // The following two methods are used to flush a memtable to
-  // storage. The first one is used atdatabase RecoveryTime (when the
+  // storage. The first one is used at database RecoveryTime (when the
   // database is opened) and is heavyweight because it holds the mutex
   // for the entire period. The second method WriteLevel0Table supports
   // concurrent flush memtables to storage.
@@ -544,9 +622,6 @@ class DBImpl : public DB {
   // Wait for memtable flushed
   Status WaitForFlushMemTable(ColumnFamilyData* cfd);
 
-  void RecordFlushIOStats();
-  void RecordCompactionIOStats();
-
 #ifndef ROCKSDB_LITE
   Status CompactFilesImpl(
       const CompactionOptions& compact_options, ColumnFamilyData* cfd,
@@ -856,6 +931,28 @@ class DBImpl : public DB {
   // Indicate DB was opened successfully
   bool opened_successfully_;
 
+  // minmum log number still containing prepared data.
+  // this is used by FindObsoleteFiles to determine which
+  // flushed logs we must keep around because they still
+  // contain prepared data which has not been flushed or rolled back
+  std::priority_queue<uint64_t, std::vector<uint64_t>, std::greater<uint64_t>>
+      min_log_with_prep_;
+
+  // to be used in conjunction with min_log_with_prep_.
+  // once a transaction with data in log L is committed or rolled back
+  // rather than removing the value from the heap we add that value
+  // to prepared_section_completed_ which maps LOG -> instance_count
+  // since a log could contain multiple prepared sections
+  //
+  // when trying to determine the minmum log still active we first
+  // consult min_log_with_prep_. while that root value maps to
+  // a value > 0 in prepared_section_completed_ we decrement the
+  // instance_count for that log and pop the root value in
+  // min_log_with_prep_. This will work the same as a min_heap
+  // where we are deleteing arbitrary elements and the up heaping.
+  std::unordered_map<uint64_t, uint64_t> prepared_section_completed_;
+  std::mutex prep_heap_mutex_;
+
   // No copying allowed
   DBImpl(const DBImpl&);
   void operator=(const DBImpl&);

db/db_impl_debug.cc

@@ -74,10 +74,17 @@ Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
                              disallow_trivial_move);
 }
 
-Status DBImpl::TEST_FlushMemTable(bool wait) {
+Status DBImpl::TEST_FlushMemTable(bool wait, ColumnFamilyHandle* cfh) {
   FlushOptions fo;
   fo.wait = wait;
-  return FlushMemTable(default_cf_handle_->cfd(), fo);
+  ColumnFamilyData* cfd;
+  if (cfh == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfhi = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh);
+    cfd = cfhi->cfd();
+  }
+  return FlushMemTable(cfd, fo);
 }
 
 Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
@@ -154,5 +161,22 @@ Status DBImpl::TEST_GetAllImmutableCFOptions(
   return Status::OK();
 }
 
+uint64_t DBImpl::TEST_FindMinLogContainingOutstandingPrep() {
+  return FindMinLogContainingOutstandingPrep();
+}
+
+uint64_t DBImpl::TEST_FindMinPrepLogReferencedByMemTable() {
+  return FindMinPrepLogReferencedByMemTable();
+}
+
+Status DBImpl::TEST_GetLatestMutableCFOptions(
+    ColumnFamilyHandle* column_family, MutableCFOptions* mutable_cf_options) {
+  InstrumentedMutexLock l(&mutex_);
+
+  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  *mutable_cf_options = *cfh->cfd()->GetLatestMutableCFOptions();
+  return Status::OK();
+}
+
 }  // namespace rocksdb
 #endif  // NDEBUG

db/db_impl_experimental.cc

@@ -49,8 +49,7 @@ Status DBImpl::SuggestCompactRange(ColumnFamilyHandle* column_family,
     }
     // Since we have some more files to compact, we should also recompute
     // compaction score
-    vstorage->ComputeCompactionScore(*cfd->GetLatestMutableCFOptions(),
-                                     CompactionOptionsFIFO());
+    vstorage->ComputeCompactionScore(*cfd->GetLatestMutableCFOptions());
     SchedulePendingCompaction(cfd);
     MaybeScheduleFlushOrCompaction();
   }

db/db_inplace_update_test.cc

@@ -18,12 +18,11 @@ class DBTestInPlaceUpdate : public DBTestBase {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdate) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
     options.env = env_;
     options.write_buffer_size = 100000;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Update key with values of smaller size
@@ -41,12 +40,11 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdate) {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateLargeNewValue) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
     options.env = env_;
     options.write_buffer_size = 100000;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Update key with values of larger size
@@ -64,7 +62,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateLargeNewValue) {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
 
@@ -72,7 +70,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
     options.write_buffer_size = 100000;
     options.inplace_callback =
       rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerSize;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Update key with values of smaller size
@@ -92,7 +89,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerSize) {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
 
@@ -100,7 +97,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
     options.write_buffer_size = 100000;
     options.inplace_callback =
       rocksdb::DBTestInPlaceUpdate::updateInPlaceSmallerVarintSize;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Update key with values of smaller varint size
@@ -120,7 +116,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackSmallerVarintSize) {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
 
@@ -128,7 +124,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
     options.write_buffer_size = 100000;
     options.inplace_callback =
       rocksdb::DBTestInPlaceUpdate::updateInPlaceLargerSize;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Update key with values of larger size
@@ -146,7 +141,7 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackLargeNewValue) {
 
 TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackNoAction) {
   do {
-    Options options;
+    Options options = CurrentOptions();
     options.create_if_missing = true;
     options.inplace_update_support = true;
 
@@ -154,7 +149,6 @@ TEST_F(DBTestInPlaceUpdate, InPlaceUpdateCallbackNoAction) {
     options.write_buffer_size = 100000;
     options.inplace_callback =
       rocksdb::DBTestInPlaceUpdate::updateInPlaceNoAction;
-    options = CurrentOptions(options);
     CreateAndReopenWithCF({"pikachu"}, options);
 
     // Callback function requests no actions from db

db/db_io_failure_test.cc

@@ -0,0 +1,259 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same 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/db_test_util.h"
+#include "port/stack_trace.h"
+
+namespace rocksdb {
+
+class DBIOFailureTest : public DBTestBase {
+ public:
+  DBIOFailureTest() : DBTestBase("/db_io_failure_test") {}
+};
+
+#ifndef ROCKSDB_LITE
+// Check that number of files does not grow when writes are dropped
+TEST_F(DBIOFailureTest, DropWrites) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.paranoid_checks = false;
+    Reopen(options);
+
+    ASSERT_OK(Put("foo", "v1"));
+    ASSERT_EQ("v1", Get("foo"));
+    Compact("a", "z");
+    const size_t num_files = CountFiles();
+    // Force out-of-space errors
+    env_->drop_writes_.store(true, std::memory_order_release);
+    env_->sleep_counter_.Reset();
+    env_->no_sleep_ = true;
+    for (int i = 0; i < 5; i++) {
+      if (option_config_ != kUniversalCompactionMultiLevel &&
+          option_config_ != kUniversalSubcompactions) {
+        for (int level = 0; level < dbfull()->NumberLevels(); level++) {
+          if (level > 0 && level == dbfull()->NumberLevels() - 1) {
+            break;
+          }
+          dbfull()->TEST_CompactRange(level, nullptr, nullptr, nullptr,
+                                      true /* disallow trivial move */);
+        }
+      } else {
+        dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+      }
+    }
+
+    std::string property_value;
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("5", property_value);
+
+    env_->drop_writes_.store(false, std::memory_order_release);
+    ASSERT_LT(CountFiles(), num_files + 3);
+
+    // Check that compaction attempts slept after errors
+    // TODO @krad: Figure out why ASSERT_EQ 5 keeps failing in certain compiler
+    // versions
+    ASSERT_GE(env_->sleep_counter_.Read(), 4);
+  } while (ChangeCompactOptions());
+}
+
+// Check background error counter bumped on flush failures.
+TEST_F(DBIOFailureTest, DropWritesFlush) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.max_background_flushes = 1;
+    Reopen(options);
+
+    ASSERT_OK(Put("foo", "v1"));
+    // Force out-of-space errors
+    env_->drop_writes_.store(true, std::memory_order_release);
+
+    std::string property_value;
+    // Background error count is 0 now.
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("0", property_value);
+
+    dbfull()->TEST_FlushMemTable(true);
+
+    ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
+    ASSERT_EQ("1", property_value);
+
+    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
+TEST_F(DBIOFailureTest, NoSpaceCompactRange) {
+  do {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.disable_auto_compactions = true;
+    Reopen(options);
+
+    // generate 5 tables
+    for (int i = 0; i < 5; ++i) {
+      ASSERT_OK(Put(Key(i), Key(i) + "v"));
+      ASSERT_OK(Flush());
+    }
+
+    // Force out-of-space errors
+    env_->no_space_.store(true, std::memory_order_release);
+
+    Status s = dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
+                                           true /* disallow trivial move */);
+    ASSERT_TRUE(s.IsIOError());
+
+    env_->no_space_.store(false, std::memory_order_release);
+  } while (ChangeCompactOptions());
+}
+
+TEST_F(DBIOFailureTest, NonWritableFileSystem) {
+  do {
+    Options options = CurrentOptions();
+    options.write_buffer_size = 4096;
+    options.arena_block_size = 4096;
+    options.env = env_;
+    Reopen(options);
+    ASSERT_OK(Put("foo", "v1"));
+    env_->non_writeable_rate_.store(100);
+    std::string big(100000, 'x');
+    int errors = 0;
+    for (int i = 0; i < 20; i++) {
+      if (!Put("foo", big).ok()) {
+        errors++;
+        env_->SleepForMicroseconds(100000);
+      }
+    }
+    ASSERT_GT(errors, 0);
+    env_->non_writeable_rate_.store(0);
+  } while (ChangeCompactOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBIOFailureTest, ManifestWriteError) {
+  // Test for the following problem:
+  // (a) Compaction produces file F
+  // (b) Log record containing F is written to MANIFEST file, but Sync() fails
+  // (c) GC deletes F
+  // (d) After reopening DB, reads fail since deleted F is named in log record
+
+  // We iterate twice.  In the second iteration, everything is the
+  // same except the log record never makes it to the MANIFEST file.
+  for (int iter = 0; iter < 2; iter++) {
+    std::atomic<bool>* error_type = (iter == 0) ? &env_->manifest_sync_error_
+                                                : &env_->manifest_write_error_;
+
+    // Insert foo=>bar mapping
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.create_if_missing = true;
+    options.error_if_exists = false;
+    options.paranoid_checks = true;
+    DestroyAndReopen(options);
+    ASSERT_OK(Put("foo", "bar"));
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Memtable compaction (will succeed)
+    Flush();
+    ASSERT_EQ("bar", Get("foo"));
+    const int last = 2;
+    MoveFilesToLevel(2);
+    ASSERT_EQ(NumTableFilesAtLevel(last), 1);  // foo=>bar is now in last level
+
+    // Merging compaction (will fail)
+    error_type->store(true, std::memory_order_release);
+    dbfull()->TEST_CompactRange(last, nullptr, nullptr);  // Should fail
+    ASSERT_EQ("bar", Get("foo"));
+
+    error_type->store(false, std::memory_order_release);
+
+    // Since paranoid_checks=true, writes should fail
+    ASSERT_NOK(Put("foo2", "bar2"));
+
+    // Recovery: should not lose data
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Try again with paranoid_checks=false
+    Close();
+    options.paranoid_checks = false;
+    Reopen(options);
+
+    // Merging compaction (will fail)
+    error_type->store(true, std::memory_order_release);
+    dbfull()->TEST_CompactRange(last, nullptr, nullptr);  // Should fail
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Recovery: should not lose data
+    error_type->store(false, std::memory_order_release);
+    Reopen(options);
+    ASSERT_EQ("bar", Get("foo"));
+
+    // Since paranoid_checks=false, writes should succeed
+    ASSERT_OK(Put("foo2", "bar2"));
+    ASSERT_EQ("bar", Get("foo"));
+    ASSERT_EQ("bar2", Get("foo2"));
+  }
+}
+#endif  // ROCKSDB_LITE
+
+TEST_F(DBIOFailureTest, PutFailsParanoid) {
+  // Test the following:
+  // (a) A random put fails in paranoid mode (simulate by sync fail)
+  // (b) All other puts have to fail, even if writes would succeed
+  // (c) All of that should happen ONLY if paranoid_checks = true
+
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.create_if_missing = true;
+  options.error_if_exists = false;
+  options.paranoid_checks = true;
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+  Status s;
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(1, "foo1", "bar1"));
+  // simulate error
+  env_->log_write_error_.store(true, std::memory_order_release);
+  s = Put(1, "foo2", "bar2");
+  ASSERT_TRUE(!s.ok());
+  env_->log_write_error_.store(false, std::memory_order_release);
+  s = Put(1, "foo3", "bar3");
+  // the next put should fail, too
+  ASSERT_TRUE(!s.ok());
+  // but we're still able to read
+  ASSERT_EQ("bar", Get(1, "foo"));
+
+  // do the same thing with paranoid checks off
+  options.paranoid_checks = false;
+  DestroyAndReopen(options);
+  CreateAndReopenWithCF({"pikachu"}, options);
+
+  ASSERT_OK(Put(1, "foo", "bar"));
+  ASSERT_OK(Put(1, "foo1", "bar1"));
+  // simulate error
+  env_->log_write_error_.store(true, std::memory_order_release);
+  s = Put(1, "foo2", "bar2");
+  ASSERT_TRUE(!s.ok());
+  env_->log_write_error_.store(false, std::memory_order_release);
+  s = Put(1, "foo3", "bar3");
+  // the next put should NOT fail
+  ASSERT_TRUE(s.ok());
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::port::InstallStackTraceHandler();
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}

db/db_iter.cc

@@ -15,6 +15,8 @@
 
 #include "db/dbformat.h"
 #include "db/filename.h"
+#include "db/merge_context.h"
+#include "db/pinned_iterators_manager.h"
 #include "port/port.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
@@ -60,11 +62,49 @@ class DBIter: public Iterator {
     kReverse
   };
 
+  // LocalStatistics contain Statistics counters that will be aggregated per
+  // each iterator instance and then will be sent to the global statistics when
+  // the iterator is destroyed.
+  //
+  // The purpose of this approach is to avoid perf regression happening
+  // when multiple threads bump the atomic counters from a DBIter::Next().
+  struct LocalStatistics {
+    explicit LocalStatistics() { ResetCounters(); }
+
+    void ResetCounters() {
+      next_count_ = 0;
+      next_found_count_ = 0;
+      prev_count_ = 0;
+      prev_found_count_ = 0;
+      bytes_read_ = 0;
+    }
+
+    void BumpGlobalStatistics(Statistics* global_statistics) {
+      RecordTick(global_statistics, NUMBER_DB_NEXT, next_count_);
+      RecordTick(global_statistics, NUMBER_DB_NEXT_FOUND, next_found_count_);
+      RecordTick(global_statistics, NUMBER_DB_PREV, prev_count_);
+      RecordTick(global_statistics, NUMBER_DB_PREV_FOUND, prev_found_count_);
+      RecordTick(global_statistics, ITER_BYTES_READ, bytes_read_);
+      ResetCounters();
+    }
+
+    // Map to Tickers::NUMBER_DB_NEXT
+    uint64_t next_count_;
+    // Map to Tickers::NUMBER_DB_NEXT_FOUND
+    uint64_t next_found_count_;
+    // Map to Tickers::NUMBER_DB_PREV
+    uint64_t prev_count_;
+    // Map to Tickers::NUMBER_DB_PREV_FOUND
+    uint64_t prev_found_count_;
+    // Map to Tickers::ITER_BYTES_READ
+    uint64_t bytes_read_;
+  };
+
   DBIter(Env* env, const ImmutableCFOptions& ioptions, const Comparator* cmp,
          InternalIterator* iter, SequenceNumber s, bool arena_mode,
          uint64_t max_sequential_skip_in_iterations, uint64_t version_number,
          const Slice* iterate_upper_bound = nullptr,
-         bool prefix_same_as_start = false)
+         bool prefix_same_as_start = false, bool pin_data = false)
       : arena_mode_(arena_mode),
         env_(env),
         logger_(ioptions.info_log),
@@ -79,13 +119,22 @@ class DBIter: public Iterator {
         version_number_(version_number),
         iterate_upper_bound_(iterate_upper_bound),
         prefix_same_as_start_(prefix_same_as_start),
-        iter_pinned_(false) {
+        pin_thru_lifetime_(pin_data) {
     RecordTick(statistics_, NO_ITERATORS);
     prefix_extractor_ = ioptions.prefix_extractor;
     max_skip_ = max_sequential_skip_in_iterations;
+    if (pin_thru_lifetime_) {
+      pinned_iters_mgr_.StartPinning();
+    }
+    if (iter_) {
+      iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
+    }
   }
   virtual ~DBIter() {
+    // Release pinned data if any
+    pinned_iters_mgr_.ReleasePinnedIterators();
     RecordTick(statistics_, NO_ITERATORS, -1);
+    local_stats_.BumpGlobalStatistics(statistics_);
     if (!arena_mode_) {
       delete iter_;
     } else {
@@ -95,9 +144,7 @@ class DBIter: public Iterator {
   virtual void SetIter(InternalIterator* iter) {
     assert(iter_ == nullptr);
     iter_ = iter;
-    if (iter_ && iter_pinned_) {
-      iter_->PinData();
-    }
+    iter_->SetPinnedItersMgr(&pinned_iters_mgr_);
   }
   virtual bool Valid() const override { return valid_; }
   virtual Slice key() const override {
@@ -106,8 +153,13 @@ class DBIter: public Iterator {
   }
   virtual Slice value() const override {
     assert(valid_);
-    return (direction_ == kForward && !current_entry_is_merged_) ?
-      iter_->value() : saved_value_;
+    if (current_entry_is_merged_) {
+      return saved_value_;
+    } else if (direction_ == kReverse) {
+      return pinned_value_;
+    } else {
+      return iter_->value();
+    }
   }
   virtual Status status() const override {
     if (status_.ok()) {
@@ -116,28 +168,6 @@ class DBIter: public Iterator {
       return status_;
     }
   }
-  virtual Status PinData() {
-    Status s;
-    if (iter_) {
-      s = iter_->PinData();
-    }
-    if (s.ok()) {
-      // Even if iter_ is nullptr, we set iter_pinned_ to true so that when
-      // iter_ is updated using SetIter, we Pin it.
-      iter_pinned_ = true;
-    }
-    return s;
-  }
-  virtual Status ReleasePinnedData() {
-    Status s;
-    if (iter_) {
-      s = iter_->ReleasePinnedData();
-    }
-    if (s.ok()) {
-      iter_pinned_ = false;
-    }
-    return s;
-  }
 
   virtual Status GetProperty(std::string prop_name,
                              std::string* prop) override {
@@ -152,7 +182,7 @@ class DBIter: public Iterator {
       return Status::OK();
     } else if (prop_name == "rocksdb.iterator.is-key-pinned") {
       if (valid_) {
-        *prop = (iter_pinned_ && saved_key_.IsKeyPinned()) ? "1" : "0";
+        *prop = (pin_thru_lifetime_ && saved_key_.IsKeyPinned()) ? "1" : "0";
       } else {
         *prop = "Iterator is not valid.";
       }
@@ -175,11 +205,26 @@ class DBIter: public Iterator {
   bool FindValueForCurrentKeyUsingSeek();
   void FindPrevUserKey();
   void FindNextUserKey();
-  inline void FindNextUserEntry(bool skipping);
-  void FindNextUserEntryInternal(bool skipping);
+  inline void FindNextUserEntry(bool skipping, bool prefix_check);
+  void FindNextUserEntryInternal(bool skipping, bool prefix_check);
   bool ParseKey(ParsedInternalKey* key);
   void MergeValuesNewToOld();
 
+  // Temporarily pin the blocks that we encounter until ReleaseTempPinnedData()
+  // is called
+  void TempPinData() {
+    if (!pin_thru_lifetime_) {
+      pinned_iters_mgr_.StartPinning();
+    }
+  }
+
+  // Release blocks pinned by TempPinData()
+  void ReleaseTempPinnedData() {
+    if (!pin_thru_lifetime_) {
+      pinned_iters_mgr_.ReleasePinnedIterators();
+    }
+  }
+
   inline void ClearSavedValue() {
     if (saved_value_.capacity() > 1048576) {
       std::string empty;
@@ -201,6 +246,7 @@ class DBIter: public Iterator {
   Status status_;
   IterKey saved_key_;
   std::string saved_value_;
+  Slice pinned_value_;
   Direction direction_;
   bool valid_;
   bool current_entry_is_merged_;
@@ -208,11 +254,16 @@ class DBIter: public Iterator {
   uint64_t max_skip_;
   uint64_t version_number_;
   const Slice* iterate_upper_bound_;
-  IterKey prefix_start_;
-  bool prefix_same_as_start_;
-  bool iter_pinned_;
+  IterKey prefix_start_buf_;
+  Slice prefix_start_key_;
+  const bool prefix_same_as_start_;
+  // Means that we will pin all data blocks we read as long the Iterator
+  // is not deleted, will be true if ReadOptions::pin_data is true
+  const bool pin_thru_lifetime_;
   // List of operands for merge operator.
-  std::deque<std::string> merge_operands_;
+  MergeContext merge_context_;
+  LocalStatistics local_stats_;
+  PinnedIteratorsManager pinned_iters_mgr_;
 
   // No copying allowed
   DBIter(const DBIter&);
@@ -235,6 +286,8 @@ void DBIter::Next() {
   assert(valid_);
 
   if (direction_ == kReverse) {
+    // We only pin blocks when doing kReverse
+    ReleaseTempPinnedData();
     FindNextUserKey();
     direction_ = kForward;
     if (!iter_->Valid()) {
@@ -250,24 +303,19 @@ void DBIter::Next() {
     PERF_COUNTER_ADD(internal_key_skipped_count, 1);
   }
 
+  if (statistics_ != nullptr) {
+    local_stats_.next_count_++;
+  }
   // Now we point to the next internal position, for both of merge and
   // not merge cases.
   if (!iter_->Valid()) {
     valid_ = false;
     return;
   }
-  FindNextUserEntry(true /* skipping the current user key */);
-  if (statistics_ != nullptr) {
-    RecordTick(statistics_, NUMBER_DB_NEXT);
-    if (valid_) {
-      RecordTick(statistics_, NUMBER_DB_NEXT_FOUND);
-      RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
-    }
-  }
-  if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
-      prefix_extractor_->Transform(saved_key_.GetKey())
-              .compare(prefix_start_.GetKey()) != 0) {
-    valid_ = false;
+  FindNextUserEntry(true /* skipping the current user key */, prefix_same_as_start_);
+  if (statistics_ != nullptr && valid_) {
+    local_stats_.next_found_count_++;
+    local_stats_.bytes_read_ += (key().size() + value().size());
   }
 }
 
@@ -279,13 +327,18 @@ void DBIter::Next() {
 //
 // NOTE: In between, saved_key_ can point to a user key that has
 //       a delete marker
-inline void DBIter::FindNextUserEntry(bool skipping) {
+//
+// The prefix_check parameter controls whether we check the iterated
+// keys against the prefix of the seeked key. Set to false when
+// performing a seek without a key (e.g. SeekToFirst). Set to
+// prefix_same_as_start_ for other iterations.
+inline void DBIter::FindNextUserEntry(bool skipping, bool prefix_check) {
   PERF_TIMER_GUARD(find_next_user_entry_time);
-  FindNextUserEntryInternal(skipping);
+  FindNextUserEntryInternal(skipping, prefix_check);
 }
 
 // Actual implementation of DBIter::FindNextUserEntry()
-void DBIter::FindNextUserEntryInternal(bool skipping) {
+void DBIter::FindNextUserEntryInternal(bool skipping, bool prefix_check) {
   // Loop until we hit an acceptable entry to yield
   assert(iter_->Valid());
   assert(direction_ == kForward);
@@ -300,6 +353,11 @@ void DBIter::FindNextUserEntryInternal(bool skipping) {
         break;
       }
 
+      if (prefix_extractor_ && prefix_check &&
+          prefix_extractor_->Transform(ikey.user_key).compare(prefix_start_key_) != 0) {
+        break;
+      }
+
       if (ikey.sequence <= sequence_) {
         if (skipping &&
            user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
@@ -311,21 +369,24 @@ void DBIter::FindNextUserEntryInternal(bool skipping) {
             case kTypeSingleDeletion:
               // Arrange to skip all upcoming entries for this key since
               // they are hidden by this deletion.
-              saved_key_.SetKey(ikey.user_key,
-                                !iter_->IsKeyPinned() /* copy */);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
               skipping = true;
               num_skipped = 0;
               PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
               break;
             case kTypeValue:
               valid_ = true;
-              saved_key_.SetKey(ikey.user_key,
-                                !iter_->IsKeyPinned() /* copy */);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
               return;
             case kTypeMerge:
               // By now, we are sure the current ikey is going to yield a value
-              saved_key_.SetKey(ikey.user_key,
-                                !iter_->IsKeyPinned() /* copy */);
+              saved_key_.SetKey(
+                  ikey.user_key,
+                  !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
               current_entry_is_merged_ = true;
               valid_ = true;
               MergeValuesNewToOld();  // Go to a different state machine
@@ -371,9 +432,9 @@ void DBIter::MergeValuesNewToOld() {
     return;
   }
 
+  merge_context_.Clear();
   // Start the merge process by pushing the first operand
-  std::deque<std::string> operands;
-  operands.push_front(iter_->value().ToString());
+  merge_context_.PushOperand(iter_->value());
 
   ParsedInternalKey ikey;
   for (iter_->Next(); iter_->Valid(); iter_->Next()) {
@@ -398,7 +459,8 @@ void DBIter::MergeValuesNewToOld() {
       {
         StopWatchNano timer(env_, statistics_ != nullptr);
         PERF_TIMER_GUARD(merge_operator_time_nanos);
-        user_merge_operator_->FullMerge(ikey.user_key, &val, operands,
+        user_merge_operator_->FullMerge(ikey.user_key, &val,
+                                        merge_context_.GetOperands(),
                                         &saved_value_, logger_);
         RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                    timer.ElapsedNanos());
@@ -410,7 +472,7 @@ void DBIter::MergeValuesNewToOld() {
       // hit a merge, add the value as an operand and run associative merge.
       // when complete, add result to operands and continue.
       const Slice& val = iter_->value();
-      operands.push_front(val.ToString());
+      merge_context_.PushOperand(val);
     } else {
       assert(false);
     }
@@ -423,8 +485,9 @@ void DBIter::MergeValuesNewToOld() {
     // a deletion marker.
     // feed null as the existing value to the merge operator, such that
     // client can differentiate this scenario and do things accordingly.
-    user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
-                                    &saved_value_, logger_);
+    user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
+                                    merge_context_.GetOperands(), &saved_value_,
+                                    logger_);
     RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
   }
 }
@@ -434,17 +497,18 @@ void DBIter::Prev() {
   if (direction_ == kForward) {
     ReverseToBackward();
   }
+  ReleaseTempPinnedData();
   PrevInternal();
   if (statistics_ != nullptr) {
-    RecordTick(statistics_, NUMBER_DB_PREV);
+    local_stats_.prev_count_++;
     if (valid_) {
-      RecordTick(statistics_, NUMBER_DB_PREV_FOUND);
-      RecordTick(statistics_, ITER_BYTES_READ, key().size() + value().size());
+      local_stats_.prev_found_count_++;
+      local_stats_.bytes_read_ += (key().size() + value().size());
     }
   }
   if (valid_ && prefix_extractor_ && prefix_same_as_start_ &&
       prefix_extractor_->Transform(saved_key_.GetKey())
-              .compare(prefix_start_.GetKey()) != 0) {
+              .compare(prefix_start_key_) != 0) {
     valid_ = false;
   }
 }
@@ -486,7 +550,7 @@ void DBIter::PrevInternal() {
 
   while (iter_->Valid()) {
     saved_key_.SetKey(ExtractUserKey(iter_->key()),
-                      !iter_->IsKeyPinned() /* copy */);
+                      !iter_->IsKeyPinned() || !pin_thru_lifetime_ /* copy */);
     if (FindValueForCurrentKey()) {
       valid_ = true;
       if (!iter_->Valid()) {
@@ -516,7 +580,8 @@ void DBIter::PrevInternal() {
 // saved_value_
 bool DBIter::FindValueForCurrentKey() {
   assert(iter_->Valid());
-  merge_operands_.clear();
+  merge_context_.Clear();
+  current_entry_is_merged_ = false;
   // last entry before merge (could be kTypeDeletion, kTypeSingleDeletion or
   // kTypeValue)
   ValueType last_not_merge_type = kTypeDeletion;
@@ -536,19 +601,21 @@ bool DBIter::FindValueForCurrentKey() {
     last_key_entry_type = ikey.type;
     switch (last_key_entry_type) {
       case kTypeValue:
-        merge_operands_.clear();
-        saved_value_ = iter_->value().ToString();
+        merge_context_.Clear();
+        ReleaseTempPinnedData();
+        TempPinData();
+        pinned_value_ = iter_->value();
         last_not_merge_type = kTypeValue;
         break;
       case kTypeDeletion:
       case kTypeSingleDeletion:
-        merge_operands_.clear();
+        merge_context_.Clear();
         last_not_merge_type = last_key_entry_type;
         PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
         break;
       case kTypeMerge:
         assert(user_merge_operator_ != nullptr);
-        merge_operands_.push_back(iter_->value().ToString());
+        merge_context_.PushOperandBack(iter_->value());
         break;
       default:
         assert(false);
@@ -567,24 +634,23 @@ bool DBIter::FindValueForCurrentKey() {
       valid_ = false;
       return false;
     case kTypeMerge:
+      current_entry_is_merged_ = true;
       if (last_not_merge_type == kTypeDeletion) {
         StopWatchNano timer(env_, statistics_ != nullptr);
         PERF_TIMER_GUARD(merge_operator_time_nanos);
         user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
-                                        merge_operands_, &saved_value_,
-                                        logger_);
+                                        merge_context_.GetOperands(),
+                                        &saved_value_, logger_);
         RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                    timer.ElapsedNanos());
       } else {
         assert(last_not_merge_type == kTypeValue);
-        std::string last_put_value = saved_value_;
-        Slice temp_slice(last_put_value);
         {
           StopWatchNano timer(env_, statistics_ != nullptr);
           PERF_TIMER_GUARD(merge_operator_time_nanos);
-          user_merge_operator_->FullMerge(saved_key_.GetKey(), &temp_slice,
-                                          merge_operands_, &saved_value_,
-                                          logger_);
+          user_merge_operator_->FullMerge(saved_key_.GetKey(), &pinned_value_,
+                                          merge_context_.GetOperands(),
+                                          &saved_value_, logger_);
           RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME,
                      timer.ElapsedNanos());
         }
@@ -617,7 +683,9 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
   if (ikey.type == kTypeValue || ikey.type == kTypeDeletion ||
       ikey.type == kTypeSingleDeletion) {
     if (ikey.type == kTypeValue) {
-      saved_value_ = iter_->value().ToString();
+      ReleaseTempPinnedData();
+      TempPinData();
+      pinned_value_ = iter_->value();
       valid_ = true;
       return true;
     }
@@ -627,11 +695,12 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
 
   // kTypeMerge. We need to collect all kTypeMerge values and save them
   // in operands
-  std::deque<std::string> operands;
+  current_entry_is_merged_ = true;
+  merge_context_.Clear();
   while (iter_->Valid() &&
          user_comparator_->Equal(ikey.user_key, saved_key_.GetKey()) &&
          ikey.type == kTypeMerge) {
-    operands.push_front(iter_->value().ToString());
+    merge_context_.PushOperand(iter_->value());
     iter_->Next();
     FindParseableKey(&ikey, kForward);
   }
@@ -642,7 +711,8 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
     {
       StopWatchNano timer(env_, statistics_ != nullptr);
       PERF_TIMER_GUARD(merge_operator_time_nanos);
-      user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
+      user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr,
+                                      merge_context_.GetOperands(),
                                       &saved_value_, logger_);
       RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
     }
@@ -660,8 +730,9 @@ bool DBIter::FindValueForCurrentKeyUsingSeek() {
   {
     StopWatchNano timer(env_, statistics_ != nullptr);
     PERF_TIMER_GUARD(merge_operator_time_nanos);
-    user_merge_operator_->FullMerge(saved_key_.GetKey(), &val, operands,
-                                    &saved_value_, logger_);
+    user_merge_operator_->FullMerge(saved_key_.GetKey(), &val,
+                                    merge_context_.GetOperands(), &saved_value_,
+                                    logger_);
     RecordTick(statistics_, MERGE_OPERATION_TOTAL_TIME, timer.ElapsedNanos());
   }
   valid_ = true;
@@ -727,6 +798,7 @@ void DBIter::FindParseableKey(ParsedInternalKey* ikey, Direction direction) {
 
 void DBIter::Seek(const Slice& target) {
   StopWatch sw(env_, statistics_, DB_SEEK);
+  ReleaseTempPinnedData();
   saved_key_.Clear();
   // now savved_key is used to store internal key.
   saved_key_.SetInternalKey(target, sequence_);
@@ -738,9 +810,15 @@ void DBIter::Seek(const Slice& target) {
 
   RecordTick(statistics_, NUMBER_DB_SEEK);
   if (iter_->Valid()) {
+    if (prefix_extractor_ && prefix_same_as_start_) {
+      prefix_start_key_ = prefix_extractor_->Transform(target);
+    }
     direction_ = kForward;
     ClearSavedValue();
-    FindNextUserEntry(false /* not skipping */);
+    FindNextUserEntry(false /* not skipping */, prefix_same_as_start_);
+    if (!valid_) {
+      prefix_start_key_.clear();
+    }
     if (statistics_ != nullptr) {
       if (valid_) {
         RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
@@ -751,7 +829,8 @@ void DBIter::Seek(const Slice& target) {
     valid_ = false;
   }
   if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(target));
+    prefix_start_buf_.SetKey(prefix_start_key_);
+    prefix_start_key_ = prefix_start_buf_.GetKey();
   }
 }
 
@@ -762,6 +841,7 @@ void DBIter::SeekToFirst() {
     max_skip_ = std::numeric_limits<uint64_t>::max();
   }
   direction_ = kForward;
+  ReleaseTempPinnedData();
   ClearSavedValue();
 
   {
@@ -771,7 +851,7 @@ void DBIter::SeekToFirst() {
 
   RecordTick(statistics_, NUMBER_DB_SEEK);
   if (iter_->Valid()) {
-    FindNextUserEntry(false /* not skipping */);
+    FindNextUserEntry(false /* not skipping */, false /* no prefix check */);
     if (statistics_ != nullptr) {
       if (valid_) {
         RecordTick(statistics_, NUMBER_DB_SEEK_FOUND);
@@ -782,7 +862,8 @@ void DBIter::SeekToFirst() {
     valid_ = false;
   }
   if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_buf_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_key_ = prefix_start_buf_.GetKey();
   }
 }
 
@@ -793,6 +874,7 @@ void DBIter::SeekToLast() {
     max_skip_ = std::numeric_limits<uint64_t>::max();
   }
   direction_ = kReverse;
+  ReleaseTempPinnedData();
   ClearSavedValue();
 
   {
@@ -830,7 +912,8 @@ void DBIter::SeekToLast() {
     }
   }
   if (valid_ && prefix_extractor_ && prefix_same_as_start_) {
-    prefix_start_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_buf_.SetKey(prefix_extractor_->Transform(saved_key_.GetKey()));
+    prefix_start_key_ = prefix_start_buf_.GetKey();
   }
 }
 
@@ -845,10 +928,7 @@ Iterator* NewDBIterator(Env* env, const ImmutableCFOptions& ioptions,
   DBIter* db_iter =
       new DBIter(env, ioptions, user_key_comparator, internal_iter, sequence,
                  false, max_sequential_skip_in_iterations, version_number,
-                 iterate_upper_bound, prefix_same_as_start);
-  if (pin_data) {
-    db_iter->PinData();
-  }
+                 iterate_upper_bound, prefix_same_as_start, pin_data);
   return db_iter;
 }
 
@@ -871,14 +951,10 @@ inline void ArenaWrappedDBIter::Prev() { db_iter_->Prev(); }
 inline Slice ArenaWrappedDBIter::key() const { return db_iter_->key(); }
 inline Slice ArenaWrappedDBIter::value() const { return db_iter_->value(); }
 inline Status ArenaWrappedDBIter::status() const { return db_iter_->status(); }
-inline Status ArenaWrappedDBIter::PinData() { return db_iter_->PinData(); }
 inline Status ArenaWrappedDBIter::GetProperty(std::string prop_name,
                                               std::string* prop) {
   return db_iter_->GetProperty(prop_name, prop);
 }
-inline Status ArenaWrappedDBIter::ReleasePinnedData() {
-  return db_iter_->ReleasePinnedData();
-}
 void ArenaWrappedDBIter::RegisterCleanup(CleanupFunction function, void* arg1,
                                          void* arg2) {
   db_iter_->RegisterCleanup(function, arg1, arg2);
@@ -896,12 +972,9 @@ ArenaWrappedDBIter* NewArenaWrappedDbIterator(
   DBIter* db_iter =
       new (mem) DBIter(env, ioptions, user_key_comparator, nullptr, sequence,
                        true, max_sequential_skip_in_iterations, version_number,
-                       iterate_upper_bound, prefix_same_as_start);
+                       iterate_upper_bound, prefix_same_as_start, pin_data);
 
   iter->SetDBIter(db_iter);
-  if (pin_data) {
-    iter->PinData();
-  }
 
   return iter;
 }

db/db_iter.h

@@ -63,8 +63,6 @@ class ArenaWrappedDBIter : public Iterator {
   virtual Status status() const override;
 
   void RegisterCleanup(CleanupFunction function, void* arg1, void* arg2);
-  virtual Status PinData();
-  virtual Status ReleasePinnedData();
   virtual Status GetProperty(std::string prop_name, std::string* prop) override;
 
  private:

db/db_properties_test.cc

@@ -19,6 +19,7 @@
 #include "rocksdb/perf_level.h"
 #include "rocksdb/table.h"
 #include "util/random.h"
+#include "util/string_util.h"
 
 namespace rocksdb {
 
@@ -223,7 +224,7 @@ void GetExpectedTableProperties(TableProperties* expected_tp,
                                 const int kBloomBitsPerKey,
                                 const size_t kBlockSize) {
   const int kKeyCount = kTableCount * kKeysPerTable;
-  const int kAvgSuccessorSize = kKeySize / 2;
+  const int kAvgSuccessorSize = kKeySize / 5;
   const int kEncodingSavePerKey = kKeySize / 4;
   expected_tp->raw_key_size = kKeyCount * (kKeySize + 8);
   expected_tp->raw_value_size = kKeyCount * kValueSize;
@@ -253,6 +254,35 @@ TEST_F(DBPropertiesTest, ValidatePropertyInfo) {
   }
 }
 
+TEST_F(DBPropertiesTest, ValidateSampleNumber) {
+  // When "max_open_files" is -1, we read all the files for
+  // "rocksdb.estimate-num-keys" computation, which is the ground truth.
+  // Otherwise, we sample 20 newest files to make an estimation.
+  // Formula: lastest_20_files_active_key_ratio * total_files
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.level0_stop_writes_trigger = 1000;
+  DestroyAndReopen(options);
+  int key = 0;
+  for (int files = 20; files >= 10; files -= 10) {
+    for (int i = 0; i < files; i++) {
+      int rows = files / 10;
+      for (int j = 0; j < rows; j++) {
+        db_->Put(WriteOptions(), std::to_string(++key), "foo");
+      }
+      db_->Flush(FlushOptions());
+    }
+  }
+  std::string num;
+  Reopen(options);
+  ASSERT_TRUE(dbfull()->GetProperty("rocksdb.estimate-num-keys", &num));
+  ASSERT_EQ("45", num);
+  options.max_open_files = -1;
+  Reopen(options);
+  ASSERT_TRUE(dbfull()->GetProperty("rocksdb.estimate-num-keys", &num));
+  ASSERT_EQ("50", num);
+}
+
 TEST_F(DBPropertiesTest, AggregatedTableProperties) {
   for (int kTableCount = 40; kTableCount <= 100; kTableCount += 30) {
     const int kKeysPerTable = 100;
@@ -869,6 +899,48 @@ TEST_F(DBPropertiesTest, EstimatePendingCompBytes) {
       "rocksdb.estimate-pending-compaction-bytes", &int_num));
   ASSERT_EQ(int_num, 0U);
 }
+
+TEST_F(DBPropertiesTest, EstimateCompressionRatio) {
+  if (!Snappy_Supported()) {
+    return;
+  }
+  const int kNumL0Files = 3;
+  const int kNumEntriesPerFile = 1000;
+
+  Options options = CurrentOptions();
+  options.compression_per_level = {kNoCompression, kSnappyCompression};
+  options.disable_auto_compactions = true;
+  options.max_background_flushes = 0;
+  options.num_levels = 2;
+  Reopen(options);
+
+  // compression ratio is -1.0 when no open files at level
+  ASSERT_EQ(CompressionRatioAtLevel(0), -1.0);
+
+  const std::string kVal(100, 'a');
+  for (int i = 0; i < kNumL0Files; ++i) {
+    for (int j = 0; j < kNumEntriesPerFile; ++j) {
+      // Put common data ("key") at end to prevent delta encoding from
+      // compressing the key effectively
+      std::string key = ToString(i) + ToString(j) + "key";
+      ASSERT_OK(dbfull()->Put(WriteOptions(), key, kVal));
+    }
+    Flush();
+  }
+
+  // no compression at L0, so ratio is less than one
+  ASSERT_LT(CompressionRatioAtLevel(0), 1.0);
+  ASSERT_GT(CompressionRatioAtLevel(0), 0.0);
+  ASSERT_EQ(CompressionRatioAtLevel(1), -1.0);
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr);
+
+  ASSERT_EQ(CompressionRatioAtLevel(0), -1.0);
+  // Data at L1 should be highly compressed thanks to Snappy and redundant data
+  // in values (ratio is 12.846 as of 4/19/2016).
+  ASSERT_GT(CompressionRatioAtLevel(1), 10.0);
+}
+
 #endif  // ROCKSDB_LITE
 
 class CountingUserTblPropCollector : public TablePropertiesCollector {

db/db_table_properties_test.cc

@@ -215,6 +215,34 @@ TEST_F(DBTablePropertiesTest, GetPropertiesOfTablesInRange) {
     TestGetPropertiesOfTablesInRange(std::move(ranges));
   }
 }
+
+TEST_F(DBTablePropertiesTest, GetColumnFamilyNameProperty) {
+  std::string kExtraCfName = "pikachu";
+  CreateAndReopenWithCF({kExtraCfName}, CurrentOptions());
+
+  // Create one table per CF, then verify it was created with the column family
+  // name property.
+  for (int cf = 0; cf < 2; ++cf) {
+    Put(cf, "key", "val");
+    Flush(cf);
+
+    TablePropertiesCollection fname_to_props;
+    ASSERT_OK(db_->GetPropertiesOfAllTables(handles_[cf], &fname_to_props));
+    ASSERT_EQ(1U, fname_to_props.size());
+
+    std::string expected_cf_name;
+    if (cf > 0) {
+      expected_cf_name = kExtraCfName;
+    } else {
+      expected_cf_name = kDefaultColumnFamilyName;
+    }
+    ASSERT_EQ(expected_cf_name,
+              fname_to_props.begin()->second->column_family_name);
+    ASSERT_EQ(cf, static_cast<uint32_t>(
+                      fname_to_props.begin()->second->column_family_id));
+  }
+}
+
 }  // namespace rocksdb
 
 #endif  // ROCKSDB_LITE

db/db_test_util.cc

@@ -210,6 +210,13 @@ Options DBTestBase::CurrentOptions(
     const anon::OptionsOverride& options_override) {
   Options options;
   options.write_buffer_size = 4090 * 4096;
+  options.target_file_size_base = 2 * 1024 * 1024;
+  options.max_bytes_for_level_base = 10 * 1024 * 1024;
+  options.max_open_files = 5000;
+  options.base_background_compactions = -1;
+  options.wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;
+  options.compaction_pri = CompactionPri::kByCompensatedSize;
+
   return CurrentOptions(options, options_override);
 }
 
@@ -301,6 +308,7 @@ Options DBTestBase::CurrentOptions(
     case kPerfOptions:
       options.soft_rate_limit = 2.0;
       options.delayed_write_rate = 8 * 1024 * 1024;
+      options.report_bg_io_stats = true;
       // TODO(3.13) -- test more options
       break;
     case kDeletesFilterFirst:
@@ -711,6 +719,22 @@ int DBTestBase::NumTableFilesAtLevel(int level, int cf) {
   return atoi(property.c_str());
 }
 
+double DBTestBase::CompressionRatioAtLevel(int level, int cf) {
+  std::string property;
+  if (cf == 0) {
+    // default cfd
+    EXPECT_TRUE(db_->GetProperty(
+        "rocksdb.compression-ratio-at-level" + NumberToString(level),
+        &property));
+  } else {
+    EXPECT_TRUE(db_->GetProperty(
+        handles_[cf],
+        "rocksdb.compression-ratio-at-level" + NumberToString(level),
+        &property));
+  }
+  return std::stod(property);
+}
+
 int DBTestBase::TotalTableFiles(int cf, int levels) {
   if (levels == -1) {
     levels = CurrentOptions().num_levels;
@@ -1040,4 +1064,34 @@ std::unordered_map<std::string, uint64_t> DBTestBase::GetAllSSTFiles(
   return res;
 }
 
+std::vector<std::uint64_t> DBTestBase::ListTableFiles(Env* env,
+                                                      const std::string& path) {
+  std::vector<std::string> files;
+  std::vector<uint64_t> file_numbers;
+  env->GetChildren(path, &files);
+  uint64_t number;
+  FileType type;
+  for (size_t i = 0; i < files.size(); ++i) {
+    if (ParseFileName(files[i], &number, &type)) {
+      if (type == kTableFile) {
+        file_numbers.push_back(number);
+      }
+    }
+  }
+  return file_numbers;
+}
+
+#ifndef ROCKSDB_LITE
+uint64_t DBTestBase::GetNumberOfSstFilesForColumnFamily(
+    DB* db, std::string column_family_name) {
+  std::vector<LiveFileMetaData> metadata;
+  db->GetLiveFilesMetaData(&metadata);
+  uint64_t result = 0;
+  for (auto& fileMetadata : metadata) {
+    result += (fileMetadata.column_family_name == column_family_name);
+  }
+  return result;
+}
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb

db/db_test_util.h

@@ -39,6 +39,7 @@
 #include "rocksdb/filter_policy.h"
 #include "rocksdb/options.h"
 #include "rocksdb/slice.h"
+#include "rocksdb/statistics.h"
 #include "rocksdb/table.h"
 #include "rocksdb/utilities/checkpoint.h"
 #include "table/block_based_table_factory.h"
@@ -357,23 +358,30 @@ class SpecialEnv : public EnvWrapper {
     class CountingFile : public RandomAccessFile {
      public:
       CountingFile(unique_ptr<RandomAccessFile>&& target,
-                   anon::AtomicCounter* counter)
-          : target_(std::move(target)), counter_(counter) {}
+                   anon::AtomicCounter* counter,
+                   std::atomic<size_t>* bytes_read)
+          : target_(std::move(target)),
+            counter_(counter),
+            bytes_read_(bytes_read) {}
       virtual Status Read(uint64_t offset, size_t n, Slice* result,
                           char* scratch) const override {
         counter_->Increment();
-        return target_->Read(offset, n, result, scratch);
+        Status s = target_->Read(offset, n, result, scratch);
+        *bytes_read_ += result->size();
+        return s;
       }
 
      private:
       unique_ptr<RandomAccessFile> target_;
       anon::AtomicCounter* counter_;
+      std::atomic<size_t>* bytes_read_;
     };
 
     Status s = target()->NewRandomAccessFile(f, r, soptions);
     random_file_open_counter_++;
     if (s.ok() && count_random_reads_) {
-      r->reset(new CountingFile(std::move(*r), &random_read_counter_));
+      r->reset(new CountingFile(std::move(*r), &random_read_counter_,
+                                &random_read_bytes_counter_));
     }
     return s;
   }
@@ -463,6 +471,7 @@ class SpecialEnv : public EnvWrapper {
 
   bool count_random_reads_;
   anon::AtomicCounter random_read_counter_;
+  std::atomic<size_t> random_read_bytes_counter_;
   std::atomic<int> random_file_open_counter_;
 
   bool count_sequential_reads_;
@@ -491,6 +500,33 @@ class SpecialEnv : public EnvWrapper {
   std::atomic<bool> is_wal_sync_thread_safe_{true};
 };
 
+#ifndef ROCKSDB_LITE
+class OnFileDeletionListener : public EventListener {
+ public:
+  OnFileDeletionListener() : matched_count_(0), expected_file_name_("") {}
+
+  void SetExpectedFileName(const std::string file_name) {
+    expected_file_name_ = file_name;
+  }
+
+  void VerifyMatchedCount(size_t expected_value) {
+    ASSERT_EQ(matched_count_, expected_value);
+  }
+
+  void OnTableFileDeleted(const TableFileDeletionInfo& info) override {
+    if (expected_file_name_ != "") {
+      ASSERT_EQ(expected_file_name_, info.file_path);
+      expected_file_name_ = "";
+      matched_count_++;
+    }
+  }
+
+ private:
+  size_t matched_count_;
+  std::string expected_file_name_;
+};
+#endif
+
 class DBTestBase : public testing::Test {
  protected:
   // Sequence of option configurations to try
@@ -672,6 +708,8 @@ class DBTestBase : public testing::Test {
 
   int NumTableFilesAtLevel(int level, int cf = 0);
 
+  double CompressionRatioAtLevel(int level, int cf = 0);
+
   int TotalTableFiles(int cf = 0, int levels = -1);
 
   // Return spread of files per level
@@ -756,6 +794,17 @@ class DBTestBase : public testing::Test {
 
   std::unordered_map<std::string, uint64_t> GetAllSSTFiles(
       uint64_t* total_size = nullptr);
+
+  std::vector<std::uint64_t> ListTableFiles(Env* env, const std::string& path);
+
+#ifndef ROCKSDB_LITE
+  uint64_t GetNumberOfSstFilesForColumnFamily(DB* db,
+                                              std::string column_family_name);
+#endif  // ROCKSDB_LITE
+
+  uint64_t TestGetTickerCount(const Options& options, Tickers ticker_type) {
+    return options.statistics->getTickerCount(ticker_type);
+  }
 };
 
 }  // namespace rocksdb

db/db_universal_compaction_test.cc

@@ -14,11 +14,6 @@
 
 namespace rocksdb {
 
-static uint64_t TestGetTickerCount(const Options& options,
-                                   Tickers ticker_type) {
-  return options.statistics->getTickerCount(ticker_type);
-}
-
 static std::string CompressibleString(Random* rnd, int len) {
   std::string r;
   test::CompressibleString(rnd, 0.8, len, &r);
@@ -124,8 +119,7 @@ class DelayFilterFactory : public CompactionFilterFactory {
 // Make sure we don't trigger a problem if the trigger conditon is given
 // to be 0, which is invalid.
 TEST_P(DBTestUniversalCompaction, UniversalCompactionSingleSortedRun) {
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
 
   options.compaction_style = kCompactionStyleUniversal;
   options.num_levels = num_levels_;
@@ -160,8 +154,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSingleSortedRun) {
 }
 
 TEST_P(DBTestUniversalCompaction, OptimizeFiltersForHits) {
-  Options options;
-  options = CurrentOptions(options);
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.size_ratio = 5;
   options.num_levels = num_levels_;
@@ -337,13 +330,12 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrigger) {
 }
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionSizeAmplification) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.num_levels = num_levels_;
   options.write_buffer_size = 100 << 10;     // 100KB
   options.target_file_size_base = 32 << 10;  // 32KB
   options.level0_file_num_compaction_trigger = 3;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -388,12 +380,12 @@ TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
   ChangeCompactOptions();
   Options options;
   options.create_if_missing = true;
-  options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
   options.compaction_style = kCompactionStyleLevel;
   options.num_levels = 1;
   options.target_file_size_base = options.write_buffer_size;
   options.compression = kNoCompression;
   options = CurrentOptions(options);
+  options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
   CreateAndReopenWithCF({"pikachu"}, options);
   ASSERT_EQ(options.compaction_style, kCompactionStyleUniversal);
   Random rnd(301);
@@ -451,12 +443,11 @@ TEST_P(DBTestUniversalCompaction, CompactFilesOnUniversalCompaction) {
 }
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionTargetLevel) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.write_buffer_size = 100 << 10;     // 100KB
   options.num_levels = 7;
   options.disable_auto_compactions = true;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   // Generate 3 overlapping files
@@ -496,14 +487,13 @@ class DBTestUniversalCompactionMultiLevels
 };
 
 TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionMultiLevels) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.num_levels = num_levels_;
   options.write_buffer_size = 100 << 10;  // 100KB
   options.level0_file_num_compaction_trigger = 8;
   options.max_background_compactions = 3;
   options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
   // Trigger compaction if size amplification exceeds 110%
@@ -539,7 +529,7 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
       });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.allow_trivial_move = true;
   options.num_levels = 3;
@@ -547,7 +537,6 @@ TEST_P(DBTestUniversalCompactionMultiLevels, UniversalCompactionTrivialMove) {
   options.level0_file_num_compaction_trigger = 3;
   options.max_background_compactions = 2;
   options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -586,7 +575,7 @@ class DBTestUniversalCompactionParallel :
 };
 
 TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.num_levels = num_levels_;
   options.write_buffer_size = 1 << 10;  // 1KB
@@ -595,7 +584,6 @@ TEST_P(DBTestUniversalCompactionParallel, UniversalCompactionParallel) {
   options.max_background_flushes = 3;
   options.target_file_size_base = 1 * 1024;
   options.compaction_options_universal.max_size_amplification_percent = 110;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -652,7 +640,7 @@ INSTANTIATE_TEST_CASE_P(DBTestUniversalCompactionParallel,
                                            ::testing::Bool()));
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.write_buffer_size = 105 << 10;    // 105KB
   options.arena_block_size = 4 << 10;       // 4KB
@@ -660,7 +648,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionOptions) {
   options.level0_file_num_compaction_trigger = 4;
   options.num_levels = num_levels_;
   options.compaction_options_universal.compression_size_percent = -1;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -773,14 +760,13 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio1) {
     return;
   }
 
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.write_buffer_size = 100 << 10;     // 100KB
   options.target_file_size_base = 32 << 10;  // 32KB
   options.level0_file_num_compaction_trigger = 2;
   options.num_levels = num_levels_;
   options.compaction_options_universal.compression_size_percent = 70;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   Random rnd(301);
@@ -841,14 +827,13 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionCompressRatio2) {
   if (!Snappy_Supported()) {
     return;
   }
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.write_buffer_size = 100 << 10;     // 100KB
   options.target_file_size_base = 32 << 10;  // 32KB
   options.level0_file_num_compaction_trigger = 2;
   options.num_levels = num_levels_;
   options.compaction_options_universal.compression_size_percent = 95;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
 
   Random rnd(301);
@@ -884,7 +869,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
       });
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.allow_trivial_move = true;
   options.num_levels = 2;
@@ -892,7 +877,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest1) {
   options.level0_file_num_compaction_trigger = 3;
   options.max_background_compactions = 1;
   options.target_file_size_base = 32 * 1024;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -929,7 +913,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
 
   rocksdb::SyncPoint::GetInstance()->EnableProcessing();
 
-  Options options;
+  Options options = CurrentOptions();
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.allow_trivial_move = true;
   options.num_levels = 15;
@@ -937,7 +921,6 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
   options.level0_file_num_compaction_trigger = 8;
   options.max_background_compactions = 4;
   options.target_file_size_base = 64 * 1024;
-  options = CurrentOptions(options);
   DestroyAndReopen(options);
   CreateAndReopenWithCF({"pikachu"}, options);
 
@@ -963,7 +946,7 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionTrivialMoveTest2) {
 }
 
 TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
-  Options options;
+  Options options = CurrentOptions();
   options.db_paths.emplace_back(dbname_, 300 * 1024);
   options.db_paths.emplace_back(dbname_ + "_2", 300 * 1024);
   options.db_paths.emplace_back(dbname_ + "_3", 500 * 1024);
@@ -972,11 +955,10 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionFourPaths) {
       new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.size_ratio = 5;
-  options.write_buffer_size = 110 << 10;  // 105KB
+  options.write_buffer_size = 111 << 10;  // 114KB
   options.arena_block_size = 4 << 10;
   options.level0_file_num_compaction_trigger = 2;
   options.num_levels = 1;
-  options = CurrentOptions(options);
 
   std::vector<std::string> filenames;
   env_->GetChildren(options.db_paths[1].path, &filenames);
@@ -1104,6 +1086,7 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
     // each value is 10K
     ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
     dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
   }
   ASSERT_OK(Flush(1));
   dbfull()->TEST_WaitForCompact();
@@ -1121,6 +1104,7 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
     // each value is 10K
     ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
     dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
   }
   ASSERT_OK(Flush(1));
   dbfull()->TEST_WaitForCompact();
@@ -1152,6 +1136,7 @@ TEST_P(DBTestUniversalCompaction, IncreaseUniversalCompactionNumLevels) {
     // each value is 10K
     ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
     dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
+    dbfull()->TEST_WaitForCompact();
   }
   ASSERT_OK(Flush(1));
   dbfull()->TEST_WaitForCompact();
@@ -1163,19 +1148,17 @@ TEST_P(DBTestUniversalCompaction, UniversalCompactionSecondPathRatio) {
   if (!Snappy_Supported()) {
     return;
   }
-  Options options;
+  Options options = CurrentOptions();
   options.db_paths.emplace_back(dbname_, 500 * 1024);
   options.db_paths.emplace_back(dbname_ + "_2", 1024 * 1024 * 1024);
   options.compaction_style = kCompactionStyleUniversal;
   options.compaction_options_universal.size_ratio = 5;
-  options.write_buffer_size = 110 << 10;  // 105KB
-  options.arena_block_size = 4 * 1024;
+  options.write_buffer_size = 111 << 10;  // 114KB
   options.arena_block_size = 4 << 10;
   options.level0_file_num_compaction_trigger = 2;
   options.num_levels = 1;
   options.memtable_factory.reset(
       new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
-  options = CurrentOptions(options);
 
   std::vector<std::string> filenames;
   env_->GetChildren(options.db_paths[1].path, &filenames);

db/db_wal_test.cc

@@ -127,6 +127,637 @@ TEST_F(DBWALTest, SyncWALNotWaitWrite) {
   ASSERT_EQ(Get("foo2"), "bar2");
   rocksdb::SyncPoint::GetInstance()->DisableProcessing();
 }
+
+TEST_F(DBWALTest, Recover) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "baz", "v5"));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v1", Get(1, "foo"));
+
+    ASSERT_EQ("v1", Get(1, "foo"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+    ASSERT_OK(Put(1, "bar", "v2"));
+    ASSERT_OK(Put(1, "foo", "v3"));
+
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v3", Get(1, "foo"));
+    ASSERT_OK(Put(1, "foo", "v4"));
+    ASSERT_EQ("v4", Get(1, "foo"));
+    ASSERT_EQ("v2", Get(1, "bar"));
+    ASSERT_EQ("v5", Get(1, "baz"));
+  } while (ChangeOptions());
+}
+
+TEST_F(DBWALTest, RecoverWithTableHandle) {
+  do {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.disable_auto_compactions = true;
+    DestroyAndReopen(options);
+    CreateAndReopenWithCF({"pikachu"}, options);
+
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "bar", "v2"));
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(Put(1, "foo", "v3"));
+    ASSERT_OK(Put(1, "bar", "v4"));
+    ASSERT_OK(Flush(1));
+    ASSERT_OK(Put(1, "big", std::string(100, 'a')));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+
+    std::vector<std::vector<FileMetaData>> files;
+    dbfull()->TEST_GetFilesMetaData(handles_[1], &files);
+    size_t total_files = 0;
+    for (const auto& level : files) {
+      total_files += level.size();
+    }
+    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 {
+          ASSERT_TRUE(file.table_reader_handle == nullptr);
+        }
+      }
+    }
+  } while (ChangeOptions());
+}
+
+TEST_F(DBWALTest, IgnoreRecoveredLog) {
+  std::string backup_logs = dbname_ + "/backup_logs";
+
+  // delete old files in backup_logs directory
+  env_->CreateDirIfMissing(backup_logs);
+  std::vector<std::string> old_files;
+  env_->GetChildren(backup_logs, &old_files);
+  for (auto& file : old_files) {
+    if (file != "." && file != "..") {
+      env_->DeleteFile(backup_logs + "/" + file);
+    }
+  }
+
+  do {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.merge_operator = MergeOperators::CreateUInt64AddOperator();
+    options.wal_dir = dbname_ + "/logs";
+    DestroyAndReopen(options);
+
+    // fill up the DB
+    std::string one, two;
+    PutFixed64(&one, 1);
+    PutFixed64(&two, 2);
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("foo"), Slice(one)));
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("foo"), Slice(one)));
+    ASSERT_OK(db_->Merge(WriteOptions(), Slice("bar"), Slice(one)));
+
+    // copy the logs to backup
+    std::vector<std::string> logs;
+    env_->GetChildren(options.wal_dir, &logs);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(options.wal_dir + "/" + log, backup_logs + "/" + log);
+      }
+    }
+
+    // recover the DB
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+    Close();
+
+    // copy the logs from backup back to wal dir
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+      }
+    }
+    // this should ignore the log files, recovery should not happen again
+    // if the recovery happens, the same merge operator would be called twice,
+    // leading to incorrect results
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+    Close();
+    Destroy(options);
+    Reopen(options);
+    Close();
+
+    // copy the logs from backup back to wal dir
+    env_->CreateDirIfMissing(options.wal_dir);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+      }
+    }
+    // assert that we successfully recovered only from logs, even though we
+    // destroyed the DB
+    Reopen(options);
+    ASSERT_EQ(two, Get("foo"));
+    ASSERT_EQ(one, Get("bar"));
+
+    // Recovery will fail if DB directory doesn't exist.
+    Destroy(options);
+    // copy the logs from backup back to wal dir
+    env_->CreateDirIfMissing(options.wal_dir);
+    for (auto& log : logs) {
+      if (log != ".." && log != ".") {
+        CopyFile(backup_logs + "/" + log, options.wal_dir + "/" + log);
+        // we won't be needing this file no more
+        env_->DeleteFile(backup_logs + "/" + log);
+      }
+    }
+    Status s = TryReopen(options);
+    ASSERT_TRUE(!s.ok());
+  } while (ChangeOptions(kSkipHashCuckoo));
+}
+
+TEST_F(DBWALTest, RecoveryWithEmptyLog) {
+  do {
+    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v1"));
+    ASSERT_OK(Put(1, "foo", "v2"));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_OK(Put(1, "foo", "v3"));
+    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
+    ASSERT_EQ("v3", Get(1, "foo"));
+  } while (ChangeOptions());
+}
+
+#ifndef ROCKSDB_LITE
+TEST_F(DBWALTest, RecoverWithLargeLog) {
+  do {
+    {
+      Options options = CurrentOptions();
+      CreateAndReopenWithCF({"pikachu"}, options);
+      ASSERT_OK(Put(1, "big1", std::string(200000, '1')));
+      ASSERT_OK(Put(1, "big2", std::string(200000, '2')));
+      ASSERT_OK(Put(1, "small3", std::string(10, '3')));
+      ASSERT_OK(Put(1, "small4", std::string(10, '4')));
+      ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
+    }
+
+    // Make sure that if we re-open with a small write buffer size that
+    // we flush table files in the middle of a large log file.
+    Options options;
+    options.write_buffer_size = 100000;
+    options = CurrentOptions(options);
+    ReopenWithColumnFamilies({"default", "pikachu"}, options);
+    ASSERT_EQ(NumTableFilesAtLevel(0, 1), 3);
+    ASSERT_EQ(std::string(200000, '1'), Get(1, "big1"));
+    ASSERT_EQ(std::string(200000, '2'), Get(1, "big2"));
+    ASSERT_EQ(std::string(10, '3'), Get(1, "small3"));
+    ASSERT_EQ(std::string(10, '4'), Get(1, "small4"));
+    ASSERT_GT(NumTableFilesAtLevel(0, 1), 1);
+  } while (ChangeCompactOptions());
+}
+
+// In https://reviews.facebook.net/D20661 we change
+// recovery behavior: previously for each log file each column family
+// memtable was flushed, even it was empty. Now it's changed:
+// we try to create the smallest number of table files by merging
+// updates from multiple logs
+TEST_F(DBWALTest, RecoverCheckFileAmountWithSmallWriteBuffer) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 5000000;
+  CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
+
+  // Since we will reopen DB with smaller write_buffer_size,
+  // each key will go to new SST file
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+  ASSERT_OK(Put(1, Key(10), DummyString(1000000)));
+
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  // Make 'dobrynia' to be flushed and new WAL file to be created
+  ASSERT_OK(Put(2, Key(10), DummyString(7500000)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(1));
+    // Make sure 'dobrynia' was flushed: check sst files amount
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(1));
+  }
+  // New WAL file
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+  ASSERT_OK(Put(3, Key(10), DummyString(1)));
+
+  options.write_buffer_size = 4096;
+  options.arena_block_size = 4096;
+  ReopenWithColumnFamilies({"default", "pikachu", "dobrynia", "nikitich"},
+                           options);
+  {
+    // No inserts => default is empty
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
+              static_cast<uint64_t>(0));
+    // First 4 keys goes to separate SSTs + 1 more SST for 2 smaller keys
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
+              static_cast<uint64_t>(5));
+    // 1 SST for big key + 1 SST for small one
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(2));
+    // 1 SST for all keys
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(1));
+  }
+}
+
+// In https://reviews.facebook.net/D20661 we change
+// recovery behavior: previously for each log file each column family
+// memtable was flushed, even it wasn't empty. Now it's changed:
+// we try to create the smallest number of table files by merging
+// updates from multiple logs
+TEST_F(DBWALTest, RecoverCheckFileAmount) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 100000;
+  options.arena_block_size = 4 * 1024;
+  CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
+
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  // Make 'nikitich' memtable to be flushed
+  ASSERT_OK(Put(3, Key(10), DummyString(1002400)));
+  ASSERT_OK(Put(3, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  // 4 memtable are not flushed, 1 sst file
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(1));
+  }
+  // Memtable for 'nikitich' has flushed, new WAL file has opened
+  // 4 memtable still not flushed
+
+  // Write to new WAL file
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  // Fill up 'nikitich' one more time
+  ASSERT_OK(Put(3, Key(10), DummyString(1002400)));
+  // make it flush
+  ASSERT_OK(Put(3, Key(1), DummyString(1)));
+  dbfull()->TEST_WaitForFlushMemTable(handles_[3]);
+  // There are still 4 memtable not flushed, and 2 sst tables
+  ASSERT_OK(Put(0, Key(1), DummyString(1)));
+  ASSERT_OK(Put(1, Key(1), DummyString(1)));
+  ASSERT_OK(Put(2, Key(1), DummyString(1)));
+
+  {
+    auto tables = ListTableFiles(env_, dbname_);
+    ASSERT_EQ(tables.size(), static_cast<size_t>(2));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(2));
+  }
+
+  ReopenWithColumnFamilies({"default", "pikachu", "dobrynia", "nikitich"},
+                           options);
+  {
+    std::vector<uint64_t> table_files = ListTableFiles(env_, dbname_);
+    // Check, that records for 'default', 'dobrynia' and 'pikachu' from
+    // first, second and third WALs  went to the same SST.
+    // So, there is 6 SSTs: three  for 'nikitich', one for 'default', one for
+    // 'dobrynia', one for 'pikachu'
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
+              static_cast<uint64_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
+              static_cast<uint64_t>(3));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
+              static_cast<uint64_t>(1));
+    ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
+              static_cast<uint64_t>(1));
+  }
+}
+
+TEST_F(DBWALTest, SyncMultipleLogs) {
+  const uint64_t kNumBatches = 2;
+  const int kBatchSize = 1000;
+
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.write_buffer_size = 4096;
+  Reopen(options);
+
+  WriteBatch batch;
+  WriteOptions wo;
+  wo.sync = true;
+
+  for (uint64_t b = 0; b < kNumBatches; b++) {
+    batch.Clear();
+    for (int i = 0; i < kBatchSize; i++) {
+      batch.Put(Key(i), DummyString(128));
+    }
+
+    dbfull()->Write(wo, &batch);
+  }
+
+  ASSERT_OK(dbfull()->SyncWAL());
+}
+
+//
+// Test WAL recovery for the various modes available
+//
+class RecoveryTestHelper {
+ public:
+  // Number of WAL files to generate
+  static const int kWALFilesCount = 10;
+  // Starting number for the WAL file name like 00010.log
+  static const int kWALFileOffset = 10;
+  // Keys to be written per WAL file
+  static const int kKeysPerWALFile = 1024;
+  // Size of the value
+  static const int kValueSize = 10;
+
+  // Create WAL files with values filled in
+  static void FillData(DBWALTest* test, const Options& options,
+                       const size_t wal_count, size_t* count) {
+    const DBOptions& db_options = options;
+
+    *count = 0;
+
+    shared_ptr<Cache> table_cache = NewLRUCache(50000, 16);
+    EnvOptions env_options;
+    WriteBuffer write_buffer(db_options.db_write_buffer_size);
+
+    unique_ptr<VersionSet> versions;
+    unique_ptr<WalManager> wal_manager;
+    WriteController write_controller;
+
+    versions.reset(new VersionSet(test->dbname_, &db_options, env_options,
+                                  table_cache.get(), &write_buffer,
+                                  &write_controller));
+
+    wal_manager.reset(new WalManager(db_options, env_options));
+
+    std::unique_ptr<log::Writer> current_log_writer;
+
+    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;
+      ASSERT_OK(db_options.env->NewWritableFile(fname, &file, env_options));
+      unique_ptr<WritableFileWriter> file_writer(
+          new WritableFileWriter(std::move(file), env_options));
+      current_log_writer.reset(
+          new log::Writer(std::move(file_writer), current_log_number,
+                          db_options.recycle_log_file_num > 0));
+
+      for (int i = 0; i < kKeysPerWALFile; i++) {
+        std::string key = "key" + ToString((*count)++);
+        std::string value = test->DummyString(kValueSize);
+        assert(current_log_writer.get() != nullptr);
+        uint64_t seq = versions->LastSequence() + 1;
+        WriteBatch batch;
+        batch.Put(key, value);
+        WriteBatchInternal::SetSequence(&batch, seq);
+        current_log_writer->AddRecord(WriteBatchInternal::Contents(&batch));
+        versions->SetLastSequence(seq);
+      }
+    }
+  }
+
+  // Recreate and fill the store with some data
+  static size_t FillData(DBWALTest* test, Options* options) {
+    options->create_if_missing = true;
+    test->DestroyAndReopen(*options);
+    test->Close();
+
+    size_t count = 0;
+    FillData(test, *options, kWALFilesCount, &count);
+    return count;
+  }
+
+  // Read back all the keys we wrote and return the number of keys found
+  static size_t GetData(DBWALTest* test) {
+    size_t count = 0;
+    for (size_t i = 0; i < kWALFilesCount * kKeysPerWALFile; i++) {
+      if (test->Get("key" + ToString(i)) != "NOT_FOUND") {
+        ++count;
+      }
+    }
+    return count;
+  }
+
+  // Manuall corrupt the specified WAL
+  static void CorruptWAL(DBWALTest* test, const Options& options,
+                         const double off, const double len,
+                         const int wal_file_id, const bool trunc = false) {
+    Env* env = options.env;
+    std::string fname = LogFileName(test->dbname_, wal_file_id);
+    uint64_t size;
+    ASSERT_OK(env->GetFileSize(fname, &size));
+    ASSERT_GT(size, 0);
+#ifdef OS_WIN
+    // Windows disk cache behaves differently. When we truncate
+    // the original content is still in the cache due to the original
+    // handle is still open. Generally, in Windows, one prohibits
+    // shared access to files and it is not needed for WAL but we allow
+    // it to induce corruption at various tests.
+    test->Close();
+#endif
+    if (trunc) {
+      ASSERT_EQ(0, truncate(fname.c_str(), static_cast<int64_t>(size * off)));
+    } else {
+      InduceCorruption(fname, static_cast<size_t>(size * off),
+                       static_cast<size_t>(size * len));
+    }
+  }
+
+  // Overwrite data with 'a' from offset for length len
+  static void InduceCorruption(const std::string& filename, size_t offset,
+                               size_t len) {
+    ASSERT_GT(len, 0U);
+
+    int fd = open(filename.c_str(), O_RDWR);
+
+    // On windows long is 32-bit
+    ASSERT_LE(offset, std::numeric_limits<long>::max());
+
+    ASSERT_GT(fd, 0);
+    ASSERT_EQ(offset, lseek(fd, static_cast<long>(offset), SEEK_SET));
+
+    void* buf = alloca(len);
+    memset(buf, 'a', len);
+    ASSERT_EQ(len, write(fd, buf, static_cast<unsigned int>(len)));
+
+    close(fd);
+  }
+};
+
+// Test scope:
+// - We expect to open the data store when there is incomplete trailing writes
+// at the end of any of the logs
+// - We do not expect to open the data store for corruption
+TEST_F(DBWALTest, kTolerateCorruptedTailRecords) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Corruption offset position */
+      for (int j = jstart; j < jend; j++) { /* WAL file */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+        // test checksum failure or parsing
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, /*wal=*/j, trunc);
+
+        if (trunc) {
+          options.wal_recovery_mode =
+              WALRecoveryMode::kTolerateCorruptedTailRecords;
+          options.create_if_missing = false;
+          ASSERT_OK(TryReopen(options));
+          const size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+          ASSERT_TRUE(i == 0 || recovered_row_count > 0);
+          ASSERT_LT(recovered_row_count, row_count);
+        } else {
+          options.wal_recovery_mode =
+              WALRecoveryMode::kTolerateCorruptedTailRecords;
+          ASSERT_NOK(TryReopen(options));
+        }
+      }
+    }
+  }
+}
+
+// Test scope:
+// We don't expect the data store to be opened if there is any corruption
+// (leading, middle or trailing -- incomplete writes or corruption)
+TEST_F(DBWALTest, kAbsoluteConsistency) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  // Verify clean slate behavior
+  Options options = CurrentOptions();
+  const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+  options.wal_recovery_mode = WALRecoveryMode::kAbsoluteConsistency;
+  options.create_if_missing = false;
+  ASSERT_OK(TryReopen(options));
+  ASSERT_EQ(RecoveryTestHelper::GetData(this), row_count);
+
+  for (auto trunc : {true, false}) { /* Corruption style */
+    for (int i = 0; i < 4; i++) {    /* Corruption offset position */
+      if (trunc && i == 0) {
+        continue;
+      }
+
+      for (int j = jstart; j < jend; j++) { /* wal files */
+        // fill with new date
+        RecoveryTestHelper::FillData(this, &options);
+        // corrupt the wal
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+        // verify
+        options.wal_recovery_mode = WALRecoveryMode::kAbsoluteConsistency;
+        options.create_if_missing = false;
+        ASSERT_NOK(TryReopen(options));
+      }
+    }
+  }
+}
+
+// Test scope:
+// - We expect to open data store under all circumstances
+// - We expect only data upto the point where the first error was encountered
+TEST_F(DBWALTest, kPointInTimeRecovery) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+  const int maxkeys =
+      RecoveryTestHelper::kWALFilesCount * RecoveryTestHelper::kKeysPerWALFile;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Offset of corruption */
+      for (int j = jstart; j < jend; j++) { /* WAL file */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+
+        // Corrupt the wal
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+
+        // Verify
+        options.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
+        options.create_if_missing = false;
+        ASSERT_OK(TryReopen(options));
+
+        // Probe data for invariants
+        size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+        ASSERT_LT(recovered_row_count, row_count);
+
+        bool expect_data = true;
+        for (size_t k = 0; k < maxkeys; ++k) {
+          bool found = Get("key" + ToString(i)) != "NOT_FOUND";
+          if (expect_data && !found) {
+            expect_data = false;
+          }
+          ASSERT_EQ(found, expect_data);
+        }
+
+        const size_t min = RecoveryTestHelper::kKeysPerWALFile *
+                           (j - RecoveryTestHelper::kWALFileOffset);
+        ASSERT_GE(recovered_row_count, min);
+        if (!trunc && i != 0) {
+          const size_t max = RecoveryTestHelper::kKeysPerWALFile *
+                             (j - RecoveryTestHelper::kWALFileOffset + 1);
+          ASSERT_LE(recovered_row_count, max);
+        }
+      }
+    }
+  }
+}
+
+// Test scope:
+// - We expect to open the data store under all scenarios
+// - We expect to have recovered records past the corruption zone
+TEST_F(DBWALTest, kSkipAnyCorruptedRecords) {
+  const int jstart = RecoveryTestHelper::kWALFileOffset;
+  const int jend = jstart + RecoveryTestHelper::kWALFilesCount;
+
+  for (auto trunc : {true, false}) {        /* Corruption style */
+    for (int i = 0; i < 4; i++) {           /* Corruption offset */
+      for (int j = jstart; j < jend; j++) { /* wal files */
+        // Fill data for testing
+        Options options = CurrentOptions();
+        const size_t row_count = RecoveryTestHelper::FillData(this, &options);
+
+        // Corrupt the WAL
+        RecoveryTestHelper::CorruptWAL(this, options, /*off=*/i * .3,
+                                       /*len%=*/.1, j, trunc);
+
+        // Verify behavior
+        options.wal_recovery_mode = WALRecoveryMode::kSkipAnyCorruptedRecords;
+        options.create_if_missing = false;
+        ASSERT_OK(TryReopen(options));
+
+        // Probe data for invariants
+        size_t recovered_row_count = RecoveryTestHelper::GetData(this);
+        ASSERT_LT(recovered_row_count, row_count);
+
+        if (!trunc) {
+          ASSERT_TRUE(i != 0 || recovered_row_count > 0);
+        }
+      }
+    }
+  }
+}
+
+#endif  // ROCKSDB_LITE
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/dbformat.h

@@ -39,6 +39,11 @@ enum ValueType : unsigned char {
   kTypeColumnFamilyMerge = 0x6,     // WAL only.
   kTypeSingleDeletion = 0x7,
   kTypeColumnFamilySingleDeletion = 0x8,  // WAL only.
+  kTypeBeginPrepareXID = 0x9,             // WAL only.
+  kTypeEndPrepareXID = 0xA,               // WAL only.
+  kTypeCommitXID = 0xB,                   // WAL only.
+  kTypeRollbackXID = 0xC,                 // WAL only.
+  kTypeNoop = 0xD,                        // WAL only.
   kMaxValue = 0x7F                        // Not used for storing records.
 };
 
@@ -465,6 +470,12 @@ class InternalKeySliceTransform : public SliceTransform {
   const SliceTransform* const transform_;
 };
 
+// Read the key of a record from a write batch.
+// if this record represent the default column family then cf_record
+// must be passed as false, otherwise it must be passed as true.
+extern bool ReadKeyFromWriteBatchEntry(Slice* input, Slice* key,
+                                       bool cf_record);
+
 // Read record from a write batch piece from input.
 // tag, column_family, key, value and blob are return values. Callers own the
 // Slice they point to.
@@ -472,5 +483,5 @@ class InternalKeySliceTransform : public SliceTransform {
 // input will be advanced to after the record.
 extern Status ReadRecordFromWriteBatch(Slice* input, char* tag,
                                        uint32_t* column_family, Slice* key,
-                                       Slice* value, Slice* blob);
+                                       Slice* value, Slice* blob, Slice* xid);
 }  // namespace rocksdb

db/dbformat_test.cc

@@ -92,6 +92,30 @@ TEST_F(FormatTest, InternalKeyShortSeparator) {
             Shorten(IKey("foo", 100, kTypeValue),
                     IKey("hello", 200, kTypeValue)));
 
+  ASSERT_EQ(IKey("ABC2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("ABC1AAAAA", 100, kTypeValue),
+                    IKey("ABC2ABB", 200, kTypeValue)));
+
+  ASSERT_EQ(IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("AAA1AAA", 100, kTypeValue),
+                    IKey("AAA2AA", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+      Shorten(IKey("AAA1AAA", 100, kTypeValue), IKey("AAA4", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA1B", kMaxSequenceNumber, kValueTypeForSeek),
+      Shorten(IKey("AAA1AAA", 100, kTypeValue), IKey("AAA2", 200, kTypeValue)));
+
+  ASSERT_EQ(IKey("AAA2", kMaxSequenceNumber, kValueTypeForSeek),
+            Shorten(IKey("AAA1AAA", 100, kTypeValue),
+                    IKey("AAA2A", 200, kTypeValue)));
+
+  ASSERT_EQ(
+      IKey("AAA1", 100, kTypeValue),
+      Shorten(IKey("AAA1", 100, kTypeValue), IKey("AAA2", 200, kTypeValue)));
+
   // When start user key is prefix of limit user key
   ASSERT_EQ(IKey("foo", 100, kTypeValue),
             Shorten(IKey("foo", 100, kTypeValue),

db/event_helpers.cc

@@ -20,57 +20,83 @@ void EventHelpers::AppendCurrentTime(JSONWriter* jwriter) {
                   std::chrono::system_clock::now().time_since_epoch()).count();
 }
 
-void EventHelpers::LogAndNotifyTableFileCreation(
+#ifndef ROCKSDB_LITE
+void EventHelpers::NotifyTableFileCreationStarted(
+    const std::vector<std::shared_ptr<EventListener>>& listeners,
+    const std::string& db_name, const std::string& cf_name,
+    const std::string& file_path, int job_id, TableFileCreationReason reason) {
+  TableFileCreationBriefInfo info;
+  info.db_name = db_name;
+  info.cf_name = cf_name;
+  info.file_path = file_path;
+  info.job_id = job_id;
+  info.reason = reason;
+  for (auto& listener : listeners) {
+    listener->OnTableFileCreationStarted(info);
+  }
+}
+#endif  // !ROCKSDB_LITE
+
+void EventHelpers::LogAndNotifyTableFileCreationFinished(
     EventLogger* event_logger,
     const std::vector<std::shared_ptr<EventListener>>& listeners,
-    const FileDescriptor& fd, const TableFileCreationInfo& info) {
-  assert(event_logger);
-  JSONWriter jwriter;
-  AppendCurrentTime(&jwriter);
-  jwriter << "cf_name" << info.cf_name
-          << "job" << info.job_id
-          << "event" << "table_file_creation"
-          << "file_number" << fd.GetNumber()
-          << "file_size" << fd.GetFileSize();
+    const std::string& db_name, const std::string& cf_name,
+    const std::string& file_path, int job_id, const FileDescriptor& fd,
+    const TableProperties& table_properties, TableFileCreationReason reason,
+    const Status& s) {
+  if (s.ok() && event_logger) {
+    JSONWriter jwriter;
+    AppendCurrentTime(&jwriter);
+    jwriter << "cf_name" << cf_name << "job" << job_id << "event"
+            << "table_file_creation"
+            << "file_number" << fd.GetNumber() << "file_size"
+            << fd.GetFileSize();
 
-  // table_properties
-  {
-    jwriter << "table_properties";
-    jwriter.StartObject();
+    // table_properties
+    {
+      jwriter << "table_properties";
+      jwriter.StartObject();
 
-    // basic properties:
-    jwriter << "data_size" << info.table_properties.data_size
-            << "index_size" << info.table_properties.index_size
-            << "filter_size" << info.table_properties.filter_size
-            << "raw_key_size" << info.table_properties.raw_key_size
-            << "raw_average_key_size" << SafeDivide(
-                info.table_properties.raw_key_size,
-                info.table_properties.num_entries)
-            << "raw_value_size" << info.table_properties.raw_value_size
-            << "raw_average_value_size" << SafeDivide(
-               info.table_properties.raw_value_size,
-               info.table_properties.num_entries)
-            << "num_data_blocks" << info.table_properties.num_data_blocks
-            << "num_entries" << info.table_properties.num_entries
-            << "filter_policy_name" <<
-                info.table_properties.filter_policy_name;
+      // basic properties:
+      jwriter << "data_size" << table_properties.data_size << "index_size"
+              << table_properties.index_size << "filter_size"
+              << table_properties.filter_size << "raw_key_size"
+              << table_properties.raw_key_size << "raw_average_key_size"
+              << SafeDivide(table_properties.raw_key_size,
+                            table_properties.num_entries)
+              << "raw_value_size" << table_properties.raw_value_size
+              << "raw_average_value_size"
+              << SafeDivide(table_properties.raw_value_size,
+                            table_properties.num_entries)
+              << "num_data_blocks" << table_properties.num_data_blocks
+              << "num_entries" << table_properties.num_entries
+              << "filter_policy_name" << table_properties.filter_policy_name;
 
-    // user collected properties
-    for (const auto& prop : info.table_properties.readable_properties) {
-      jwriter << prop.first << prop.second;
+      // user collected properties
+      for (const auto& prop : table_properties.readable_properties) {
+        jwriter << prop.first << prop.second;
+      }
+      jwriter.EndObject();
     }
     jwriter.EndObject();
-  }
-  jwriter.EndObject();
 
-  event_logger->Log(jwriter);
+    event_logger->Log(jwriter);
+  }
 
 #ifndef ROCKSDB_LITE
   if (listeners.size() == 0) {
     return;
   }
-
-  for (auto listener : listeners) {
+  TableFileCreationInfo info;
+  info.db_name = db_name;
+  info.cf_name = cf_name;
+  info.file_path = file_path;
+  info.file_size = fd.file_size;
+  info.job_id = job_id;
+  info.table_properties = table_properties;
+  info.reason = reason;
+  info.status = s;
+  for (auto& listener : listeners) {
     listener->OnTableFileCreated(info);
   }
 #endif  // !ROCKSDB_LITE

db/event_helpers.h

@@ -19,15 +19,30 @@ namespace rocksdb {
 class EventHelpers {
  public:
   static void AppendCurrentTime(JSONWriter* json_writer);
-  static void LogAndNotifyTableFileCreation(
+#ifndef ROCKSDB_LITE
+  static void NotifyTableFileCreationStarted(
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      const std::string& db_name, const std::string& cf_name,
+      const std::string& file_path, int job_id, TableFileCreationReason reason);
+#endif  // !ROCKSDB_LITE
+  static void LogAndNotifyTableFileCreationFinished(
       EventLogger* event_logger,
       const std::vector<std::shared_ptr<EventListener>>& listeners,
-      const FileDescriptor& fd, const TableFileCreationInfo& info);
+      const std::string& db_name, const std::string& cf_name,
+      const std::string& file_path, int job_id, const FileDescriptor& fd,
+      const TableProperties& table_properties, TableFileCreationReason reason,
+      const Status& s);
   static void LogAndNotifyTableFileDeletion(
       EventLogger* event_logger, int job_id,
       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);
+
+ private:
+  static void LogAndNotifyTableFileCreation(
+      EventLogger* event_logger,
+      const std::vector<std::shared_ptr<EventListener>>& listeners,
+      const FileDescriptor& fd, const TableFileCreationInfo& info);
 };
 
 }  // namespace rocksdb

db/fault_injection_test.cc

@@ -11,8 +11,6 @@
 // the last "sync". It then checks for data loss errors by purposely dropping
 // file data (or entire files) not protected by a "sync".
 
-#include <map>
-#include <set>
 #include "db/db_impl.h"
 #include "db/filename.h"
 #include "db/log_format.h"
@@ -22,6 +20,7 @@
 #include "rocksdb/env.h"
 #include "rocksdb/table.h"
 #include "rocksdb/write_batch.h"
+#include "util/fault_injection_test_env.h"
 #include "util/logging.h"
 #include "util/mock_env.h"
 #include "util/mutexlock.h"
@@ -35,401 +34,6 @@ static const int kValueSize = 1000;
 static const int kMaxNumValues = 2000;
 static const size_t kNumIterations = 3;
 
-class TestWritableFile;
-class FaultInjectionTestEnv;
-
-namespace {
-
-// Assume a filename, and not a directory name like "/foo/bar/"
-static std::string GetDirName(const std::string filename) {
-  size_t found = filename.find_last_of("/\\");
-  if (found == std::string::npos) {
-    return "";
-  } else {
-    return filename.substr(0, found);
-  }
-}
-
-// Trim the tailing "/" in the end of `str`
-static std::string TrimDirname(const std::string& str) {
-  size_t found = str.find_last_not_of("/");
-  if (found == std::string::npos) {
-    return str;
-  }
-  return str.substr(0, found + 1);
-}
-
-// Return pair <parent directory name, file name> of a full path.
-static std::pair<std::string, std::string> GetDirAndName(
-    const std::string& name) {
-  std::string dirname = GetDirName(name);
-  std::string fname = name.substr(dirname.size() + 1);
-  return std::make_pair(dirname, fname);
-}
-
-// A basic file truncation function suitable for this test.
-Status Truncate(Env* env, const std::string& filename, uint64_t length) {
-  unique_ptr<SequentialFile> orig_file;
-  const EnvOptions options;
-  Status s = env->NewSequentialFile(filename, &orig_file, options);
-  if (!s.ok()) {
-    fprintf(stderr, "Cannot truncate file %s: %s\n", filename.c_str(),
-            s.ToString().c_str());
-    return s;
-  }
-
-  std::unique_ptr<char[]> scratch(new char[length]);
-  rocksdb::Slice result;
-  s = orig_file->Read(length, &result, scratch.get());
-#ifdef OS_WIN
-  orig_file.reset();
-#endif
-  if (s.ok()) {
-    std::string tmp_name = GetDirName(filename) + "/truncate.tmp";
-    unique_ptr<WritableFile> tmp_file;
-    s = env->NewWritableFile(tmp_name, &tmp_file, options);
-    if (s.ok()) {
-      s = tmp_file->Append(result);
-      if (s.ok()) {
-        s = env->RenameFile(tmp_name, filename);
-      } else {
-        fprintf(stderr, "Cannot rename file %s to %s: %s\n", tmp_name.c_str(),
-                filename.c_str(), s.ToString().c_str());
-        env->DeleteFile(tmp_name);
-      }
-    }
-  }
-  if (!s.ok()) {
-    fprintf(stderr, "Cannot truncate file %s: %s\n", filename.c_str(),
-            s.ToString().c_str());
-  }
-
-  return s;
-}
-
-struct FileState {
-  std::string filename_;
-  ssize_t pos_;
-  ssize_t pos_at_last_sync_;
-  ssize_t pos_at_last_flush_;
-
-  explicit FileState(const std::string& filename)
-      : filename_(filename),
-        pos_(-1),
-        pos_at_last_sync_(-1),
-        pos_at_last_flush_(-1) { }
-
-  FileState() : pos_(-1), pos_at_last_sync_(-1), pos_at_last_flush_(-1) {}
-
-  bool IsFullySynced() const { return pos_ <= 0 || pos_ == pos_at_last_sync_; }
-
-  Status DropUnsyncedData(Env* env) const;
-
-  Status DropRandomUnsyncedData(Env* env, Random* rand) const;
-};
-
-}  // anonymous namespace
-
-// A wrapper around WritableFileWriter* file
-// is written to or sync'ed.
-class TestWritableFile : public WritableFile {
- public:
-  explicit TestWritableFile(const std::string& fname,
-                            unique_ptr<WritableFile>&& f,
-                            FaultInjectionTestEnv* env);
-  virtual ~TestWritableFile();
-  virtual Status Append(const Slice& data) override;
-  virtual Status Truncate(uint64_t size) override { return target_->Truncate(size); }
-  virtual Status Close() override;
-  virtual Status Flush() override;
-  virtual Status Sync() override;
-  virtual bool IsSyncThreadSafe() const override { return true; }
-
- private:
-  FileState state_;
-  unique_ptr<WritableFile> target_;
-  bool writable_file_opened_;
-  FaultInjectionTestEnv* env_;
-};
-
-class TestDirectory : public Directory {
- public:
-  explicit TestDirectory(FaultInjectionTestEnv* env, std::string dirname,
-                         Directory* dir)
-      : env_(env), dirname_(dirname), dir_(dir) {}
-  ~TestDirectory() {}
-
-  virtual Status Fsync() override;
-
- private:
-  FaultInjectionTestEnv* env_;
-  std::string dirname_;
-  unique_ptr<Directory> dir_;
-};
-
-class FaultInjectionTestEnv : public EnvWrapper {
- public:
-  explicit FaultInjectionTestEnv(Env* base)
-      : EnvWrapper(base),
-        filesystem_active_(true) {}
-  virtual ~FaultInjectionTestEnv() { }
-
-  Status NewDirectory(const std::string& name,
-                      unique_ptr<Directory>* result) override {
-    unique_ptr<Directory> r;
-    Status s = target()->NewDirectory(name, &r);
-    EXPECT_OK(s);
-    if (!s.ok()) {
-      return s;
-    }
-    result->reset(new TestDirectory(this, TrimDirname(name), r.release()));
-    return Status::OK();
-  }
-
-  Status NewWritableFile(const std::string& fname,
-                         unique_ptr<WritableFile>* result,
-                         const EnvOptions& soptions) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    // Not allow overwriting files
-    Status s = target()->FileExists(fname);
-    if (s.ok()) {
-      return Status::Corruption("File already exists.");
-    } else if (!s.IsNotFound()) {
-      assert(s.IsIOError());
-      return s;
-    }
-    s = target()->NewWritableFile(fname, result, soptions);
-    if (s.ok()) {
-      result->reset(new TestWritableFile(fname, std::move(*result), this));
-      // WritableFileWriter* file is opened
-      // again then it will be truncated - so forget our saved state.
-      UntrackFile(fname);
-      MutexLock l(&mutex_);
-      open_files_.insert(fname);
-      auto dir_and_name = GetDirAndName(fname);
-      auto& list = dir_to_new_files_since_last_sync_[dir_and_name.first];
-      list.insert(dir_and_name.second);
-    }
-    return s;
-  }
-
-  virtual Status DeleteFile(const std::string& f) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    Status s = EnvWrapper::DeleteFile(f);
-    if (!s.ok()) {
-      fprintf(stderr, "Cannot delete file %s: %s\n", f.c_str(),
-              s.ToString().c_str());
-    }
-    EXPECT_OK(s);
-    if (s.ok()) {
-      UntrackFile(f);
-    }
-    return s;
-  }
-
-  virtual Status RenameFile(const std::string& s,
-                            const std::string& t) override {
-    if (!IsFilesystemActive()) {
-      return Status::Corruption("Not Active");
-    }
-    Status ret = EnvWrapper::RenameFile(s, t);
-
-    if (ret.ok()) {
-      MutexLock l(&mutex_);
-      if (db_file_state_.find(s) != db_file_state_.end()) {
-        db_file_state_[t] = db_file_state_[s];
-        db_file_state_.erase(s);
-      }
-
-      auto sdn = GetDirAndName(s);
-      auto tdn = GetDirAndName(t);
-      if (dir_to_new_files_since_last_sync_[sdn.first].erase(sdn.second) != 0) {
-        auto& tlist = dir_to_new_files_since_last_sync_[tdn.first];
-        assert(tlist.find(tdn.second) == tlist.end());
-        tlist.insert(tdn.second);
-      }
-    }
-
-    return ret;
-  }
-
-  void WritableFileClosed(const FileState& state) {
-    MutexLock l(&mutex_);
-    if (open_files_.find(state.filename_) != open_files_.end()) {
-      db_file_state_[state.filename_] = state;
-      open_files_.erase(state.filename_);
-    }
-  }
-
-  // For every file that is not fully synced, make a call to `func` with
-  // FileState of the file as the parameter.
-  Status DropFileData(std::function<Status(Env*, FileState)> func) {
-    Status s;
-    MutexLock l(&mutex_);
-    for (std::map<std::string, FileState>::const_iterator it =
-             db_file_state_.begin();
-         s.ok() && it != db_file_state_.end(); ++it) {
-      const FileState& state = it->second;
-      if (!state.IsFullySynced()) {
-        s = func(target(), state);
-      }
-    }
-    return s;
-  }
-
-  Status DropUnsyncedFileData() {
-    return DropFileData([&](Env* env, const FileState& state) {
-      return state.DropUnsyncedData(env);
-    });
-  }
-
-  Status DropRandomUnsyncedFileData(Random* rnd) {
-    return DropFileData([&](Env* env, const FileState& state) {
-      return state.DropRandomUnsyncedData(env, rnd);
-    });
-  }
-
-  Status DeleteFilesCreatedAfterLastDirSync() {
-    // Because DeleteFile access this container make a copy to avoid deadlock
-    std::map<std::string, std::set<std::string>> map_copy;
-    {
-      MutexLock l(&mutex_);
-      map_copy.insert(dir_to_new_files_since_last_sync_.begin(),
-                      dir_to_new_files_since_last_sync_.end());
-    }
-
-    for (auto& pair : map_copy) {
-      for (std::string name : pair.second) {
-        Status s = DeleteFile(pair.first + "/" + name);
-        if (!s.ok()) {
-          return s;
-        }
-      }
-    }
-    return Status::OK();
-  }
-  void ResetState() {
-    MutexLock l(&mutex_);
-    db_file_state_.clear();
-    dir_to_new_files_since_last_sync_.clear();
-    SetFilesystemActiveNoLock(true);
-  }
-
-  void UntrackFile(const std::string& f) {
-    MutexLock l(&mutex_);
-    auto dir_and_name = GetDirAndName(f);
-    dir_to_new_files_since_last_sync_[dir_and_name.first].erase(
-        dir_and_name.second);
-    db_file_state_.erase(f);
-    open_files_.erase(f);
-  }
-
-  void SyncDir(const std::string& dirname) {
-    MutexLock l(&mutex_);
-    dir_to_new_files_since_last_sync_.erase(dirname);
-  }
-
-  // Setting the filesystem to inactive is the test equivalent to simulating a
-  // system reset. Setting to inactive will freeze our saved filesystem state so
-  // that it will stop being recorded. It can then be reset back to the state at
-  // the time of the reset.
-  bool IsFilesystemActive() {
-    MutexLock l(&mutex_);
-    return filesystem_active_;
-  }
-  void SetFilesystemActiveNoLock(bool active) { filesystem_active_ = active; }
-  void SetFilesystemActive(bool active) {
-    MutexLock l(&mutex_);
-    SetFilesystemActiveNoLock(active);
-  }
-  void AssertNoOpenFile() { ASSERT_TRUE(open_files_.empty()); }
-
- private:
-  port::Mutex mutex_;
-  std::map<std::string, FileState> db_file_state_;
-  std::set<std::string> open_files_;
-  std::unordered_map<std::string, std::set<std::string>>
-      dir_to_new_files_since_last_sync_;
-  bool filesystem_active_;  // Record flushes, syncs, writes
-};
-
-Status FileState::DropUnsyncedData(Env* env) const {
-  ssize_t sync_pos = pos_at_last_sync_ == -1 ? 0 : pos_at_last_sync_;
-  return Truncate(env, filename_, sync_pos);
-}
-
-Status FileState::DropRandomUnsyncedData(Env* env, Random* rand) const {
-  ssize_t sync_pos = pos_at_last_sync_ == -1 ? 0 : pos_at_last_sync_;
-  assert(pos_ >= sync_pos);
-  int range = static_cast<int>(pos_ - sync_pos);
-  uint64_t truncated_size =
-      static_cast<uint64_t>(sync_pos) + rand->Uniform(range);
-  return Truncate(env, filename_, truncated_size);
-}
-
-Status TestDirectory::Fsync() {
-  env_->SyncDir(dirname_);
-  return dir_->Fsync();
-}
-
-TestWritableFile::TestWritableFile(const std::string& fname,
-                                   unique_ptr<WritableFile>&& f,
-                                   FaultInjectionTestEnv* env)
-      : state_(fname),
-        target_(std::move(f)),
-        writable_file_opened_(true),
-        env_(env) {
-  assert(target_ != nullptr);
-  state_.pos_ = 0;
-}
-
-TestWritableFile::~TestWritableFile() {
-  if (writable_file_opened_) {
-    Close();
-  }
-}
-
-Status TestWritableFile::Append(const Slice& data) {
-  if (!env_->IsFilesystemActive()) {
-    return Status::Corruption("Not Active");
-  }
-  Status s = target_->Append(data);
-  if (s.ok()) {
-    state_.pos_ += data.size();
-  }
-  return s;
-}
-
-Status TestWritableFile::Close() {
-  writable_file_opened_ = false;
-  Status s = target_->Close();
-  if (s.ok()) {
-    env_->WritableFileClosed(state_);
-  }
-  return s;
-}
-
-Status TestWritableFile::Flush() {
-  Status s = target_->Flush();
-  if (s.ok() && env_->IsFilesystemActive()) {
-    state_.pos_at_last_flush_ = state_.pos_;
-  }
-  return s;
-}
-
-Status TestWritableFile::Sync() {
-  if (!env_->IsFilesystemActive()) {
-    return Status::OK();
-  }
-  // No need to actual sync.
-  state_.pos_at_last_sync_ = state_.pos_;
-  return Status::OK();
-}
-
 class FaultInjectionTest : public testing::Test,
                            public testing::WithParamInterface<bool> {
  protected:
@@ -732,7 +336,7 @@ class FaultInjectionTest : public testing::Test,
   }
 
   void WaitCompactionFinish() {
-    static_cast<DBImpl*>(db_)->TEST_WaitForCompact();
+    static_cast<DBImpl*>(db_->GetRootDB())->TEST_WaitForCompact();
     ASSERT_OK(db_->Put(WriteOptions(), "", ""));
   }
 };

db/flush_job.cc

@@ -66,7 +66,7 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
                    JobContext* job_context, LogBuffer* log_buffer,
                    Directory* db_directory, Directory* output_file_directory,
                    CompressionType output_compression, Statistics* stats,
-                   EventLogger* event_logger)
+                   EventLogger* event_logger, bool measure_io_stats)
     : dbname_(dbname),
       cfd_(cfd),
       db_options_(db_options),
@@ -83,7 +83,8 @@ FlushJob::FlushJob(const std::string& dbname, ColumnFamilyData* cfd,
       output_file_directory_(output_file_directory),
       output_compression_(output_compression),
       stats_(stats),
-      event_logger_(event_logger) {
+      event_logger_(event_logger),
+      measure_io_stats_(measure_io_stats) {
   // Update the thread status to indicate flush.
   ReportStartedFlush();
   TEST_SYNC_POINT("FlushJob::FlushJob()");
@@ -114,13 +115,30 @@ void FlushJob::ReportFlushInputSize(const autovector<MemTable*>& mems) {
 }
 
 void FlushJob::RecordFlushIOStats() {
-  ThreadStatusUtil::SetThreadOperationProperty(
+  RecordTick(stats_, FLUSH_WRITE_BYTES, IOSTATS(bytes_written));
+  ThreadStatusUtil::IncreaseThreadOperationProperty(
       ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
+  IOSTATS_RESET(bytes_written);
 }
 
 Status FlushJob::Run(FileMetaData* file_meta) {
   AutoThreadOperationStageUpdater stage_run(
       ThreadStatus::STAGE_FLUSH_RUN);
+  // I/O measurement variables
+  PerfLevel prev_perf_level = PerfLevel::kEnableTime;
+  uint64_t prev_write_nanos = 0;
+  uint64_t prev_fsync_nanos = 0;
+  uint64_t prev_range_sync_nanos = 0;
+  uint64_t prev_prepare_write_nanos = 0;
+  if (measure_io_stats_) {
+    prev_perf_level = GetPerfLevel();
+    SetPerfLevel(PerfLevel::kEnableTime);
+    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);
+  }
+
   // Save the contents of the earliest memtable as a new Table
   FileMetaData meta;
   autovector<MemTable*> mems;
@@ -179,6 +197,19 @@ Status FlushJob::Run(FileMetaData* file_meta) {
     stream << vstorage->NumLevelFiles(level);
   }
   stream.EndArray();
+  stream << "immutable_memtables" << cfd_->imm()->NumNotFlushed();
+
+  if (measure_io_stats_) {
+    if (prev_perf_level != PerfLevel::kEnableTime) {
+      SetPerfLevel(prev_perf_level);
+    }
+    stream << "file_write_nanos" << (IOSTATS(write_nanos) - prev_write_nanos);
+    stream << "file_range_sync_nanos"
+           << (IOSTATS(range_sync_nanos) - prev_range_sync_nanos);
+    stream << "file_fsync_nanos" << (IOSTATS(fsync_nanos) - prev_fsync_nanos);
+    stream << "file_prepare_write_nanos"
+           << (IOSTATS(prepare_write_nanos) - prev_prepare_write_nanos);
+  }
 
   return s;
 }
@@ -223,7 +254,6 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
                          << total_num_deletes << "memory_usage"
                          << total_memory_usage;
 
-    TableFileCreationInfo info;
     {
       ScopedArenaIterator iter(
           NewMergingIterator(&cfd_->internal_comparator(), &memtables[0],
@@ -234,15 +264,16 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
 
       TEST_SYNC_POINT_CALLBACK("FlushJob::WriteLevel0Table:output_compression",
                                &output_compression_);
-      s = BuildTable(dbname_, db_options_.env, *cfd_->ioptions(), env_options_,
-                     cfd_->table_cache(), iter.get(), meta,
-                     cfd_->internal_comparator(),
-                     cfd_->int_tbl_prop_collector_factories(), cfd_->GetID(),
-                     existing_snapshots_, earliest_write_conflict_snapshot_,
-                     output_compression_, cfd_->ioptions()->compression_opts,
-                     mutable_cf_options_.paranoid_file_checks,
-                     cfd_->internal_stats(), Env::IO_HIGH, &table_properties_);
-      info.table_properties = table_properties_;
+      s = BuildTable(
+          dbname_, db_options_.env, *cfd_->ioptions(), mutable_cf_options_,
+          env_options_, cfd_->table_cache(), iter.get(), meta,
+          cfd_->internal_comparator(), cfd_->int_tbl_prop_collector_factories(),
+          cfd_->GetID(), cfd_->GetName(), existing_snapshots_,
+          earliest_write_conflict_snapshot_, output_compression_,
+          cfd_->ioptions()->compression_opts,
+          mutable_cf_options_.paranoid_file_checks, cfd_->internal_stats(),
+          TableFileCreationReason::kFlush, event_logger_, job_context_->job_id,
+          Env::IO_HIGH, &table_properties_, 0 /* level */);
       LogFlush(db_options_.info_log);
     }
     Log(InfoLogLevel::INFO_LEVEL, db_options_.info_log,
@@ -253,21 +284,6 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
         meta->fd.GetFileSize(), s.ToString().c_str(),
         meta->marked_for_compaction ? " (needs compaction)" : "");
 
-    // output to event logger
-    if (s.ok()) {
-      info.db_name = dbname_;
-      info.cf_name = cfd_->GetName();
-      info.file_path = TableFileName(db_options_.db_paths,
-                                     meta->fd.GetNumber(),
-                                     meta->fd.GetPathId());
-      info.file_size = meta->fd.GetFileSize();
-      info.job_id = job_context_->job_id;
-      EventHelpers::LogAndNotifyTableFileCreation(
-          event_logger_, db_options_.listeners,
-          meta->fd, info);
-      TEST_SYNC_POINT("FlushJob::LogAndNotifyTableFileCreation()");
-    }
-
     if (!db_options_.disableDataSync && output_file_directory_ != nullptr) {
       output_file_directory_->Fsync();
     }
@@ -293,13 +309,14 @@ Status FlushJob::WriteLevel0Table(const autovector<MemTable*>& mems,
                   meta->marked_for_compaction);
   }
 
+  // Note that here we treat flush as level 0 compaction in internal stats
   InternalStats::CompactionStats stats(1);
   stats.micros = db_options_.env->NowMicros() - start_micros;
   stats.bytes_written = meta->fd.GetFileSize();
   cfd_->internal_stats()->AddCompactionStats(0 /* level */, stats);
   cfd_->internal_stats()->AddCFStats(InternalStats::BYTES_FLUSHED,
                                      meta->fd.GetFileSize());
-  RecordTick(stats_, COMPACT_WRITE_BYTES, meta->fd.GetFileSize());
+  RecordFlushIOStats();
   return s;
 }
 

db/flush_job.h

@@ -62,7 +62,7 @@ class FlushJob {
            JobContext* job_context, LogBuffer* log_buffer,
            Directory* db_directory, Directory* output_file_directory,
            CompressionType output_compression, Statistics* stats,
-           EventLogger* event_logger);
+           EventLogger* event_logger, bool measure_io_stats);
 
   ~FlushJob();
 
@@ -93,6 +93,7 @@ class FlushJob {
   Statistics* stats_;
   EventLogger* event_logger_;
   TableProperties table_properties_;
+  bool measure_io_stats_;
 };
 
 }  // namespace rocksdb

db/flush_job_test.cc

@@ -93,7 +93,7 @@ TEST_F(FlushJobTest, Empty) {
                      db_options_, *cfd->GetLatestMutableCFOptions(),
                      env_options_, versions_.get(), &mutex_, &shutting_down_,
                      {}, kMaxSequenceNumber, &job_context, nullptr, nullptr,
-                     nullptr, kNoCompression, nullptr, &event_logger);
+                     nullptr, kNoCompression, nullptr, &event_logger, false);
   ASSERT_OK(flush_job.Run());
   job_context.Clean();
 }
@@ -132,7 +132,7 @@ TEST_F(FlushJobTest, NonEmpty) {
                      db_options_, *cfd->GetLatestMutableCFOptions(),
                      env_options_, versions_.get(), &mutex_, &shutting_down_,
                      {}, kMaxSequenceNumber, &job_context, nullptr, nullptr,
-                     nullptr, kNoCompression, nullptr, &event_logger);
+                     nullptr, kNoCompression, nullptr, &event_logger, true);
   FileMetaData fd;
   mutex_.Lock();
   ASSERT_OK(flush_job.Run(&fd));
@@ -192,11 +192,11 @@ TEST_F(FlushJobTest, Snapshots) {
   }
 
   EventLogger event_logger(db_options_.info_log.get());
-  FlushJob flush_job(dbname_, versions_->GetColumnFamilySet()->GetDefault(),
-                     db_options_, *cfd->GetLatestMutableCFOptions(),
-                     env_options_, versions_.get(), &mutex_, &shutting_down_,
-                     snapshots, kMaxSequenceNumber, &job_context, nullptr,
-                     nullptr, nullptr, kNoCompression, nullptr, &event_logger);
+  FlushJob flush_job(
+      dbname_, versions_->GetColumnFamilySet()->GetDefault(), db_options_,
+      *cfd->GetLatestMutableCFOptions(), env_options_, versions_.get(), &mutex_,
+      &shutting_down_, snapshots, kMaxSequenceNumber, &job_context, nullptr,
+      nullptr, nullptr, kNoCompression, nullptr, &event_logger, true);
   mutex_.Lock();
   ASSERT_OK(flush_job.Run());
   mutex_.Unlock();

db/internal_stats.cc

@@ -101,6 +101,8 @@ std::pair<Slice, Slice> GetPropertyNameAndArg(const Slice& property) {
 static const std::string rocksdb_prefix = "rocksdb.";
 
 static const std::string num_files_at_level_prefix = "num-files-at-level";
+static const std::string compression_ratio_at_level_prefix =
+    "compression-ratio-at-level";
 static const std::string allstats = "stats";
 static const std::string sstables = "sstables";
 static const std::string cfstats = "cfstats";
@@ -148,6 +150,8 @@ static const std::string num_running_flushes = "num-running-flushes";
 
 const std::string DB::Properties::kNumFilesAtLevelPrefix =
                       rocksdb_prefix + num_files_at_level_prefix;
+const std::string DB::Properties::kCompressionRatioAtLevelPrefix =
+                      rocksdb_prefix + compression_ratio_at_level_prefix;
 const std::string DB::Properties::kStats = rocksdb_prefix + allstats;
 const std::string DB::Properties::kSSTables = rocksdb_prefix + sstables;
 const std::string DB::Properties::kCFStats = rocksdb_prefix + cfstats;
@@ -211,6 +215,8 @@ const std::unordered_map<std::string,
                          DBPropertyInfo> InternalStats::ppt_name_to_info = {
     {DB::Properties::kNumFilesAtLevelPrefix,
      {false, &InternalStats::HandleNumFilesAtLevel, nullptr}},
+    {DB::Properties::kCompressionRatioAtLevelPrefix,
+     {false, &InternalStats::HandleCompressionRatioAtLevelPrefix, nullptr}},
     {DB::Properties::kLevelStats,
      {false, &InternalStats::HandleLevelStats, nullptr}},
     {DB::Properties::kStats, {false, &InternalStats::HandleStats, nullptr}},
@@ -324,6 +330,19 @@ bool InternalStats::HandleNumFilesAtLevel(std::string* value, Slice suffix) {
   }
 }
 
+bool InternalStats::HandleCompressionRatioAtLevelPrefix(std::string* value,
+                                                        Slice suffix) {
+  uint64_t level;
+  const auto* vstorage = cfd_->current()->storage_info();
+  bool ok = ConsumeDecimalNumber(&suffix, &level) && suffix.empty();
+  if (!ok || level >= static_cast<uint64_t>(number_levels_)) {
+    return false;
+  }
+  *value = ToString(
+      vstorage->GetEstimatedCompressionRatioAtLevel(static_cast<int>(level)));
+  return true;
+}
+
 bool InternalStats::HandleLevelStats(std::string* value, Slice suffix) {
   char buf[1000];
   const auto* vstorage = cfd_->current()->storage_info();
@@ -589,9 +608,6 @@ void InternalStats::DumpDBStats(std::string* value) {
   uint64_t wal_synced = GetDBStats(InternalStats::WAL_FILE_SYNCED);
   uint64_t write_with_wal = GetDBStats(InternalStats::WRITE_WITH_WAL);
   uint64_t write_stall_micros = GetDBStats(InternalStats::WRITE_STALL_MICROS);
-  uint64_t compact_bytes_read = 0;
-  uint64_t compact_bytes_write = 0;
-  uint64_t compact_micros = 0;
 
   const int kHumanMicrosLen = 32;
   char human_micros[kHumanMicrosLen];
@@ -623,20 +639,6 @@ void InternalStats::DumpDBStats(std::string* value) {
            write_with_wal / static_cast<double>(wal_synced + 1),
            wal_bytes / kGB, wal_bytes / kMB / seconds_up);
   value->append(buf);
-  // Compact
-  for (int level = 0; level < number_levels_; level++) {
-    compact_bytes_read += comp_stats_[level].bytes_read_output_level +
-                          comp_stats_[level].bytes_read_non_output_levels;
-    compact_bytes_write += comp_stats_[level].bytes_written;
-    compact_micros += comp_stats_[level].micros;
-  }
-  snprintf(buf, sizeof(buf),
-           "Cumulative compaction: %.2f GB write, %.2f MB/s write, "
-           "%.2f GB read, %.2f MB/s read, %.1f seconds\n",
-           compact_bytes_write / kGB, compact_bytes_write / kMB / seconds_up,
-           compact_bytes_read / kGB, compact_bytes_read / kMB / seconds_up,
-           compact_micros / kMicrosInSec);
-  value->append(buf);
   // Stall
   AppendHumanMicros(write_stall_micros, human_micros, kHumanMicrosLen, true);
   snprintf(buf, sizeof(buf),
@@ -681,25 +683,6 @@ void InternalStats::DumpDBStats(std::string* value) {
            interval_wal_bytes / kMB / std::max(interval_seconds_up, 0.001));
   value->append(buf);
 
-  // Compaction
-  uint64_t interval_compact_bytes_write =
-      compact_bytes_write - db_stats_snapshot_.compact_bytes_write;
-  uint64_t interval_compact_bytes_read =
-      compact_bytes_read - db_stats_snapshot_.compact_bytes_read;
-  uint64_t interval_compact_micros =
-      compact_micros - db_stats_snapshot_.compact_micros;
-
-  snprintf(
-      buf, sizeof(buf),
-      "Interval compaction: %.2f GB write, %.2f MB/s write, "
-      "%.2f GB read, %.2f MB/s read, %.1f seconds\n",
-      interval_compact_bytes_write / kGB,
-      interval_compact_bytes_write / kMB / std::max(interval_seconds_up, 0.001),
-      interval_compact_bytes_read / kGB,
-      interval_compact_bytes_read / kMB / std::max(interval_seconds_up, 0.001),
-      interval_compact_micros / kMicrosInSec);
-  value->append(buf);
-
   // Stall
   AppendHumanMicros(
       write_stall_micros - db_stats_snapshot_.write_stall_micros,
@@ -731,9 +714,6 @@ void InternalStats::DumpDBStats(std::string* value) {
   db_stats_snapshot_.wal_synced = wal_synced;
   db_stats_snapshot_.write_with_wal = write_with_wal;
   db_stats_snapshot_.write_stall_micros = write_stall_micros;
-  db_stats_snapshot_.compact_bytes_write = compact_bytes_write;
-  db_stats_snapshot_.compact_bytes_read = compact_bytes_read;
-  db_stats_snapshot_.compact_micros = compact_micros;
 }
 
 void InternalStats::DumpCFStats(std::string* value) {
@@ -789,6 +769,7 @@ void InternalStats::DumpCFStats(std::string* value) {
       value->append(buf);
     }
   }
+
   uint64_t curr_ingest = cf_stats_value_[BYTES_FLUSHED];
   // Cumulative summary
   double w_amp = stats_sum.bytes_written / static_cast<double>(curr_ingest + 1);
@@ -812,11 +793,58 @@ void InternalStats::DumpCFStats(std::string* value) {
   PrintLevelStats(buf, sizeof(buf), "Int", 0, 0, 0, 0, w_amp, interval_stats);
   value->append(buf);
 
+  double seconds_up = (env_->NowMicros() - started_at_ + 1) / kMicrosInSec;
+  double interval_seconds_up = seconds_up - cf_stats_snapshot_.seconds_up;
+  snprintf(buf, sizeof(buf), "Uptime(secs): %.1f total, %.1f interval\n",
+           seconds_up, interval_seconds_up);
+  value->append(buf);
+
   snprintf(buf, sizeof(buf),
            "Flush(GB): cumulative %.3f, interval %.3f\n",
            curr_ingest / kGB, interval_ingest / kGB);
   value->append(buf);
 
+  // Compact
+  uint64_t compact_bytes_read = 0;
+  uint64_t compact_bytes_write = 0;
+  uint64_t compact_micros = 0;
+  for (int level = 0; level < number_levels_; level++) {
+    compact_bytes_read += comp_stats_[level].bytes_read_output_level +
+                          comp_stats_[level].bytes_read_non_output_levels;
+    compact_bytes_write += comp_stats_[level].bytes_written;
+    compact_micros += comp_stats_[level].micros;
+  }
+
+  snprintf(buf, sizeof(buf),
+           "Cumulative compaction: %.2f GB write, %.2f MB/s write, "
+           "%.2f GB read, %.2f MB/s read, %.1f seconds\n",
+           compact_bytes_write / kGB, compact_bytes_write / kMB / seconds_up,
+           compact_bytes_read / kGB, compact_bytes_read / kMB / seconds_up,
+           compact_micros / kMicrosInSec);
+  value->append(buf);
+
+  // Compaction interval
+  uint64_t interval_compact_bytes_write =
+      compact_bytes_write - cf_stats_snapshot_.compact_bytes_write;
+  uint64_t interval_compact_bytes_read =
+      compact_bytes_read - cf_stats_snapshot_.compact_bytes_read;
+  uint64_t interval_compact_micros =
+      compact_micros - cf_stats_snapshot_.compact_micros;
+
+  snprintf(
+      buf, sizeof(buf),
+      "Interval compaction: %.2f GB write, %.2f MB/s write, "
+      "%.2f GB read, %.2f MB/s read, %.1f seconds\n",
+      interval_compact_bytes_write / kGB,
+      interval_compact_bytes_write / kMB / std::max(interval_seconds_up, 0.001),
+      interval_compact_bytes_read / kGB,
+      interval_compact_bytes_read / kMB / std::max(interval_seconds_up, 0.001),
+      interval_compact_micros / kMicrosInSec);
+  value->append(buf);
+  cf_stats_snapshot_.compact_bytes_write = compact_bytes_write;
+  cf_stats_snapshot_.compact_bytes_read = compact_bytes_read;
+  cf_stats_snapshot_.compact_micros = compact_micros;
+
   snprintf(buf, sizeof(buf), "Stalls(count): %" PRIu64
                              " level0_slowdown, "
                              "%" PRIu64

db/internal_stats.h

@@ -246,11 +246,20 @@ class InternalStats {
     CompactionStats comp_stats;
     uint64_t ingest_bytes;            // Bytes written to L0
     uint64_t stall_count;             // Stall count
+    // Stats from compaction jobs - bytes written, bytes read, duration.
+    uint64_t compact_bytes_write;
+    uint64_t compact_bytes_read;
+    uint64_t compact_micros;
+    double seconds_up;
 
     CFStatsSnapshot()
         : comp_stats(0),
           ingest_bytes(0),
-          stall_count(0) {}
+          stall_count(0),
+          compact_bytes_write(0),
+          compact_bytes_read(0),
+          compact_micros(0),
+          seconds_up(0) {}
   } cf_stats_snapshot_;
 
   struct DBStatsSnapshot {
@@ -263,10 +272,6 @@ class InternalStats {
     // another thread.
     uint64_t write_other;
     uint64_t write_self;
-    // Stats from compaction jobs - bytes written, bytes read, duration.
-    uint64_t compact_bytes_write;
-    uint64_t compact_bytes_read;
-    uint64_t compact_micros;
     // Total number of keys written. write_self and write_other measure number
     // of write requests written, Each of the write request can contain updates
     // to multiple keys. num_keys_written is total number of keys updated by all
@@ -283,9 +288,6 @@ class InternalStats {
           write_with_wal(0),
           write_other(0),
           write_self(0),
-          compact_bytes_write(0),
-          compact_bytes_read(0),
-          compact_micros(0),
           num_keys_written(0),
           write_stall_micros(0),
           seconds_up(0) {}
@@ -294,6 +296,7 @@ class InternalStats {
   // Handler functions for getting property values. They use "value" as a value-
   // result argument, and return true upon successfully setting "value".
   bool HandleNumFilesAtLevel(std::string* value, Slice suffix);
+  bool HandleCompressionRatioAtLevelPrefix(std::string* value, Slice suffix);
   bool HandleLevelStats(std::string* value, Slice suffix);
   bool HandleStats(std::string* value, Slice suffix);
   bool HandleCFStats(std::string* value, Slice suffix);

db/job_context.h

@@ -22,9 +22,9 @@ class MemTable;
 struct JobContext {
   inline bool HaveSomethingToDelete() const {
     return full_scan_candidate_files.size() || sst_delete_files.size() ||
-           log_delete_files.size() || new_superversion != nullptr ||
-           superversions_to_free.size() > 0 || memtables_to_free.size() > 0 ||
-           logs_to_free.size() > 0;
+           log_delete_files.size() || manifest_delete_files.size() ||
+           new_superversion != nullptr || superversions_to_free.size() > 0 ||
+           memtables_to_free.size() > 0 || logs_to_free.size() > 0;
   }
 
   // Structure to store information for candidate files to delete.
@@ -56,6 +56,9 @@ struct JobContext {
   // a list of log files that we need to delete
   std::vector<uint64_t> log_delete_files;
 
+  // a list of manifest files that we need to delete
+  std::vector<std::string> manifest_delete_files;
+
   // a list of memtables to be free
   autovector<MemTable*> memtables_to_free;
 

db/listener_test.cc

@@ -484,7 +484,10 @@ TEST_F(EventListenerTest, CompactionReasonLevel) {
   listener->compaction_reasons_.clear();
   Reopen(options);
 
-  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
+  Put("key", "value");
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
   ASSERT_GT(listener->compaction_reasons_.size(), 0);
   for (auto compaction_reason : listener->compaction_reasons_) {
     ASSERT_EQ(compaction_reason, CompactionReason::kManualCompaction);
@@ -579,7 +582,180 @@ TEST_F(EventListenerTest, CompactionReasonFIFO) {
     ASSERT_EQ(compaction_reason, CompactionReason::kFIFOMaxSize);
   }
 }
-}  // namespace rocksdb
+
+class TableFileCreationListener : public EventListener {
+ public:
+  class TestEnv : public EnvWrapper {
+   public:
+    TestEnv() : EnvWrapper(Env::Default()) {}
+
+    void SetStatus(Status s) { status_ = s; }
+
+    Status NewWritableFile(const std::string& fname,
+                           std::unique_ptr<WritableFile>* result,
+                           const EnvOptions& options) {
+      if (fname.size() > 4 && fname.substr(fname.size() - 4) == ".sst") {
+        if (!status_.ok()) {
+          return status_;
+        }
+      }
+      return Env::Default()->NewWritableFile(fname, result, options);
+    }
+
+   private:
+    Status status_;
+  };
+
+  TableFileCreationListener() {
+    for (int i = 0; i < 2; i++) {
+      started_[i] = finished_[i] = failure_[i] = 0;
+    }
+  }
+
+  int Index(TableFileCreationReason reason) {
+    int idx;
+    switch (reason) {
+      case TableFileCreationReason::kFlush:
+        idx = 0;
+        break;
+      case TableFileCreationReason::kCompaction:
+        idx = 1;
+        break;
+      default:
+        idx = -1;
+    }
+    return idx;
+  }
+
+  void CheckAndResetCounters(int flush_started, int flush_finished,
+                             int flush_failure, int compaction_started,
+                             int compaction_finished, int compaction_failure) {
+    ASSERT_EQ(started_[0], flush_started);
+    ASSERT_EQ(finished_[0], flush_finished);
+    ASSERT_EQ(failure_[0], flush_failure);
+    ASSERT_EQ(started_[1], compaction_started);
+    ASSERT_EQ(finished_[1], compaction_finished);
+    ASSERT_EQ(failure_[1], compaction_failure);
+    for (int i = 0; i < 2; i++) {
+      started_[i] = finished_[i] = failure_[i] = 0;
+    }
+  }
+
+  void OnTableFileCreationStarted(
+      const TableFileCreationBriefInfo& info) override {
+    int idx = Index(info.reason);
+    if (idx >= 0) {
+      started_[idx]++;
+    }
+    ASSERT_GT(info.db_name.size(), 0U);
+    ASSERT_GT(info.cf_name.size(), 0U);
+    ASSERT_GT(info.file_path.size(), 0U);
+    ASSERT_GT(info.job_id, 0);
+  }
+
+  void OnTableFileCreated(const TableFileCreationInfo& info) override {
+    int idx = Index(info.reason);
+    if (idx >= 0) {
+      finished_[idx]++;
+    }
+    ASSERT_GT(info.db_name.size(), 0U);
+    ASSERT_GT(info.cf_name.size(), 0U);
+    ASSERT_GT(info.file_path.size(), 0U);
+    ASSERT_GT(info.job_id, 0);
+    if (info.status.ok()) {
+      ASSERT_GT(info.table_properties.data_size, 0U);
+      ASSERT_GT(info.table_properties.raw_key_size, 0U);
+      ASSERT_GT(info.table_properties.raw_value_size, 0U);
+      ASSERT_GT(info.table_properties.num_data_blocks, 0U);
+      ASSERT_GT(info.table_properties.num_entries, 0U);
+    } else {
+      if (idx >= 0) {
+        failure_[idx]++;
+      }
+    }
+  }
+
+  TestEnv test_env;
+  int started_[2];
+  int finished_[2];
+  int failure_[2];
+};
+
+TEST_F(EventListenerTest, TableFileCreationListenersTest) {
+  auto listener = std::make_shared<TableFileCreationListener>();
+  Options options;
+  options.create_if_missing = true;
+  options.listeners.push_back(listener);
+  options.env = &listener->test_env;
+  DestroyAndReopen(options);
+
+  ASSERT_OK(Put("foo", "aaa"));
+  ASSERT_OK(Put("bar", "bbb"));
+  ASSERT_OK(Flush());
+  dbfull()->TEST_WaitForFlushMemTable();
+  listener->CheckAndResetCounters(1, 1, 0, 0, 0, 0);
+
+  ASSERT_OK(Put("foo", "aaa1"));
+  ASSERT_OK(Put("bar", "bbb1"));
+  listener->test_env.SetStatus(Status::NotSupported("not supported"));
+  ASSERT_NOK(Flush());
+  listener->CheckAndResetCounters(1, 1, 1, 0, 0, 0);
+  listener->test_env.SetStatus(Status::OK());
+
+  Reopen(options);
+  ASSERT_OK(Put("foo", "aaa2"));
+  ASSERT_OK(Put("bar", "bbb2"));
+  ASSERT_OK(Flush());
+  dbfull()->TEST_WaitForFlushMemTable();
+  listener->CheckAndResetCounters(1, 1, 0, 0, 0, 0);
+
+  const Slice kRangeStart = "a";
+  const Slice kRangeEnd = "z";
+  dbfull()->CompactRange(CompactRangeOptions(), &kRangeStart, &kRangeEnd);
+  dbfull()->TEST_WaitForCompact();
+  listener->CheckAndResetCounters(0, 0, 0, 1, 1, 0);
+
+  ASSERT_OK(Put("foo", "aaa3"));
+  ASSERT_OK(Put("bar", "bbb3"));
+  ASSERT_OK(Flush());
+  listener->test_env.SetStatus(Status::NotSupported("not supported"));
+  dbfull()->CompactRange(CompactRangeOptions(), &kRangeStart, &kRangeEnd);
+  dbfull()->TEST_WaitForCompact();
+  listener->CheckAndResetCounters(1, 1, 0, 1, 1, 1);
+}
+
+class MemTableSealedListener : public EventListener {
+private:
+  SequenceNumber latest_seq_number_;
+public:
+  MemTableSealedListener() {}
+  void OnMemTableSealed(const MemTableInfo& info) override {
+    latest_seq_number_ = info.first_seqno;
+  }
+
+  void OnFlushCompleted(DB* /*db*/,
+    const FlushJobInfo& flush_job_info) override {
+    ASSERT_LE(flush_job_info.smallest_seqno, latest_seq_number_);
+  }
+};
+
+TEST_F(EventListenerTest, MemTableSealedListenerTest) {
+  auto listener = std::make_shared<MemTableSealedListener>();
+  Options options;
+  options.create_if_missing = true;
+  options.listeners.push_back(listener);
+  DestroyAndReopen(options);
+
+  for (unsigned int i = 0; i < 10; i++) {
+    std::string tag = std::to_string(i);
+    ASSERT_OK(Put("foo"+tag, "aaa"));
+    ASSERT_OK(Put("bar"+tag, "bbb"));
+
+    ASSERT_OK(Flush());
+  }
+}
+} // namespace rocksdb
+
 
 #endif  // ROCKSDB_LITE
 

db/log_reader.cc

@@ -22,9 +22,8 @@ Reader::Reporter::~Reporter() {
 }
 
 Reader::Reader(std::shared_ptr<Logger> info_log,
-	       unique_ptr<SequentialFileReader>&& _file,
-               Reporter* reporter, bool checksum, uint64_t initial_offset,
-               uint64_t log_num)
+               unique_ptr<SequentialFileReader>&& _file, Reporter* reporter,
+               bool checksum, uint64_t initial_offset, uint64_t log_num)
     : info_log_(info_log),
       file_(std::move(_file)),
       reporter_(reporter),
@@ -37,7 +36,8 @@ Reader::Reader(std::shared_ptr<Logger> info_log,
       last_record_offset_(0),
       end_of_buffer_offset_(0),
       initial_offset_(initial_offset),
-      log_number_(log_num) {}
+      log_number_(log_num),
+      recycled_(false) {}
 
 Reader::~Reader() {
   delete[] backing_store_;
@@ -191,6 +191,25 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
         }
         break;
 
+      case kBadRecordLen:
+      case kBadRecordChecksum:
+        if (recycled_ &&
+            wal_recovery_mode ==
+                WALRecoveryMode::kTolerateCorruptedTailRecords) {
+          scratch->clear();
+          return false;
+        }
+	if (record_type == kBadRecordLen)
+	  ReportCorruption(drop_size, "bad record length");
+	else
+	  ReportCorruption(drop_size, "checksum mismatch");
+        if (in_fragmented_record) {
+          ReportCorruption(scratch->size(), "error in middle of record");
+          in_fragmented_record = false;
+          scratch->clear();
+        }
+        break;
+
       default: {
         char buf[40];
         snprintf(buf, sizeof(buf), "unknown record type %u", record_type);
@@ -337,6 +356,9 @@ unsigned int Reader::ReadPhysicalRecord(Slice* result, size_t* drop_size) {
     const uint32_t length = a | (b << 8);
     int header_size = kHeaderSize;
     if (type >= kRecyclableFullType && type <= kRecyclableLastType) {
+      if (end_of_buffer_offset_ - buffer_.size() == 0) {
+        recycled_ = true;
+      }
       header_size = kRecyclableHeaderSize;
       // We need enough for the larger header
       if (buffer_.size() < (size_t)kRecyclableHeaderSize) {
@@ -355,8 +377,7 @@ unsigned int Reader::ReadPhysicalRecord(Slice* result, size_t* drop_size) {
       *drop_size = buffer_.size();
       buffer_.clear();
       if (!eof_) {
-        ReportCorruption(*drop_size, "bad record length");
-        return kBadRecord;
+        return kBadRecordLen;
       }
       // If the end of the file has been reached without reading |length| bytes
       // of payload, assume the writer died in the middle of writing the record.
@@ -388,8 +409,7 @@ unsigned int Reader::ReadPhysicalRecord(Slice* result, size_t* drop_size) {
         // like a valid log record.
         *drop_size = buffer_.size();
         buffer_.clear();
-        ReportCorruption(*drop_size, "checksum mismatch");
-        return kBadRecord;
+        return kBadRecordChecksum;
       }
     }
 

db/log_reader.h

@@ -113,6 +113,9 @@ class Reader {
   // which log number this is
   uint64_t const log_number_;
 
+  // Whether this is a recycled log file
+  bool recycled_;
+
   // Extend record types with the following special values
   enum {
     kEof = kMaxRecordType + 1,
@@ -126,6 +129,10 @@ class Reader {
     kBadHeader = kMaxRecordType + 3,
     // Returned when we read an old record from a previous user of the log.
     kOldRecord = kMaxRecordType + 4,
+    // Returned when we get a bad record length
+    kBadRecordLen = kMaxRecordType + 5,
+    // Returned when we get a bad record checksum
+    kBadRecordChecksum = kMaxRecordType + 6,
   };
 
   // Skips all blocks that are completely before "initial_offset_".

db/log_test.cc

@@ -174,6 +174,8 @@ class LogTest : public ::testing::TestWithParam<int> {
                                              3 * header_size;
   }
 
+  Slice* get_reader_contents() { return &reader_contents_; }
+
   void Write(const std::string& msg) {
     writer_.AddRecord(Slice(msg));
   }
@@ -445,9 +447,13 @@ TEST_P(LogTest, BadLength) {
   Write("foo");
   // Least significant size byte is stored in header[4].
   IncrementByte(4, 1);
-  ASSERT_EQ("foo", Read());
-  ASSERT_EQ(kBlockSize, DroppedBytes());
-  ASSERT_EQ("OK", MatchError("bad record length"));
+  if (!GetParam()) {
+    ASSERT_EQ("foo", Read());
+    ASSERT_EQ(kBlockSize, DroppedBytes());
+    ASSERT_EQ("OK", MatchError("bad record length"));
+  } else {
+    ASSERT_EQ("EOF", Read());
+  }
 }
 
 TEST_P(LogTest, BadLengthAtEndIsIgnored) {
@@ -470,8 +476,13 @@ TEST_P(LogTest, ChecksumMismatch) {
   Write("foooooo");
   IncrementByte(0, 14);
   ASSERT_EQ("EOF", Read());
-  ASSERT_EQ(14U + 4 * !!GetParam(), DroppedBytes());
-  ASSERT_EQ("OK", MatchError("checksum mismatch"));
+  if (!GetParam()) {
+    ASSERT_EQ(14U, DroppedBytes());
+    ASSERT_EQ("OK", MatchError("checksum mismatch"));
+  } else {
+    ASSERT_EQ(0U, DroppedBytes());
+    ASSERT_EQ("", ReportMessage());
+  }
 }
 
 TEST_P(LogTest, UnexpectedMiddleType) {
@@ -568,11 +579,15 @@ TEST_P(LogTest, ErrorJoinsRecords) {
     SetByte(offset, 'x');
   }
 
-  ASSERT_EQ("correct", Read());
-  ASSERT_EQ("EOF", Read());
-  size_t dropped = DroppedBytes();
-  ASSERT_LE(dropped, 2 * kBlockSize + 100);
-  ASSERT_GE(dropped, 2 * kBlockSize);
+  if (!GetParam()) {
+    ASSERT_EQ("correct", Read());
+    ASSERT_EQ("EOF", Read());
+    size_t dropped = DroppedBytes();
+    ASSERT_LE(dropped, 2 * kBlockSize + 100);
+    ASSERT_GE(dropped, 2 * kBlockSize);
+  } else {
+    ASSERT_EQ("EOF", Read());
+  }
 }
 
 TEST_P(LogTest, ReadStart) { CheckInitialOffsetRecord(0, 0); }
@@ -690,6 +705,29 @@ TEST_P(LogTest, ClearEofError2) {
   ASSERT_EQ("OK", MatchError("read error"));
 }
 
+TEST_P(LogTest, Recycle) {
+  if (!GetParam()) {
+    return;  // test is only valid for recycled logs
+  }
+  Write("foo");
+  Write("bar");
+  Write("baz");
+  Write("bif");
+  Write("blitz");
+  while (get_reader_contents()->size() < log::kBlockSize * 2) {
+    Write("xxxxxxxxxxxxxxxx");
+  }
+  unique_ptr<WritableFileWriter> dest_holder(test::GetWritableFileWriter(
+      new test::OverwritingStringSink(get_reader_contents())));
+  Writer recycle_writer(std::move(dest_holder), 123, true);
+  recycle_writer.AddRecord(Slice("foooo"));
+  recycle_writer.AddRecord(Slice("bar"));
+  ASSERT_GE(get_reader_contents()->size(), log::kBlockSize * 2);
+  ASSERT_EQ("foooo", Read());
+  ASSERT_EQ("bar", Read());
+  ASSERT_EQ("EOF", Read());
+}
+
 INSTANTIATE_TEST_CASE_P(bool, LogTest, ::testing::Values(0, 2));
 
 }  // namespace log

db/managed_iterator.cc

@@ -79,7 +79,7 @@ ManagedIterator::ManagedIterator(DBImpl* db, const ReadOptions& read_options,
       release_supported_(true) {
   read_options_.managed = false;
   if ((!read_options_.tailing) && (read_options_.snapshot == nullptr)) {
-    assert(read_options_.snapshot = db_->GetSnapshot());
+    assert(nullptr != (read_options_.snapshot = db_->GetSnapshot()));
     snapshot_created_ = true;
   }
   cfh_.SetCFD(cfd);
@@ -210,7 +210,8 @@ void ManagedIterator::RebuildIterator() {
 void ManagedIterator::UpdateCurrent() {
   assert(mutable_iter_ != nullptr);
 
-  if (!(valid_ = mutable_iter_->Valid())) {
+  valid_ = mutable_iter_->Valid();
+  if (!valid_) {
     status_ = mutable_iter_->status();
     return;
   }

db/memtable.cc

@@ -15,6 +15,7 @@
 
 #include "db/dbformat.h"
 #include "db/merge_context.h"
+#include "db/pinned_iterators_manager.h"
 #include "db/writebuffer.h"
 #include "rocksdb/comparator.h"
 #include "rocksdb/env.h"
@@ -74,6 +75,7 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
       first_seqno_(0),
       earliest_seqno_(earliest_seq),
       mem_next_logfile_number_(0),
+      min_prep_log_referenced_(0),
       locks_(moptions_.inplace_update_support
                  ? moptions_.inplace_update_num_locks
                  : 0),
@@ -231,6 +233,12 @@ class MemTableIterator : public InternalIterator {
   }
 
   ~MemTableIterator() {
+#ifndef NDEBUG
+    // Assert that the MemTableIterator is never deleted while
+    // Pinning is Enabled.
+    assert(!pinned_iters_mgr_ ||
+           (pinned_iters_mgr_ && !pinned_iters_mgr_->PinningEnabled()));
+#endif
     if (arena_mode_) {
       iter_->~Iterator();
     } else {
@@ -238,6 +246,14 @@ class MemTableIterator : public InternalIterator {
     }
   }
 
+#ifndef NDEBUG
+  virtual void SetPinnedItersMgr(
+      PinnedIteratorsManager* pinned_iters_mgr) override {
+    pinned_iters_mgr_ = pinned_iters_mgr;
+  }
+  PinnedIteratorsManager* pinned_iters_mgr_ = nullptr;
+#endif
+
   virtual bool Valid() const override { return valid_; }
   virtual void Seek(const Slice& k) override {
     PERF_TIMER_GUARD(seek_on_memtable_time);
@@ -285,16 +301,6 @@ class MemTableIterator : public InternalIterator {
 
   virtual Status status() const override { return Status::OK(); }
 
-  virtual Status PinData() override {
-    // memtable data is always pinned
-    return Status::OK();
-  }
-
-  virtual Status ReleasePinnedData() override {
-    // memtable data is always pinned
-    return Status::OK();
-  }
-
   virtual bool IsKeyPinned() const override {
     // memtable data is always pinned
     return true;
@@ -795,4 +801,17 @@ void MemTableRep::Get(const LookupKey& k, void* callback_args,
   }
 }
 
+void MemTable::RefLogContainingPrepSection(uint64_t log) {
+  assert(log > 0);
+  auto cur = min_prep_log_referenced_.load();
+  while ((log < cur || cur == 0) &&
+         !min_prep_log_referenced_.compare_exchange_strong(cur, log)) {
+    cur = min_prep_log_referenced_.load();
+  }
+}
+
+uint64_t MemTable::GetMinLogContainingPrepSection() {
+  return min_prep_log_referenced_.load();
+}
+
 }  // namespace rocksdb

db/memtable.h

@@ -271,6 +271,13 @@ class MemTable {
   // operations on the same MemTable.
   void SetNextLogNumber(uint64_t num) { mem_next_logfile_number_ = num; }
 
+  // if this memtable contains data from a committed
+  // two phase transaction we must take note of the
+  // log which contains that data so we can know
+  // when to relese that log
+  void RefLogContainingPrepSection(uint64_t log);
+  uint64_t GetMinLogContainingPrepSection();
+
   // Notify the underlying storage that no more items will be added.
   // REQUIRES: external synchronization to prevent simultaneous
   // operations on the same MemTable.
@@ -342,6 +349,10 @@ class MemTable {
   // The log files earlier than this number can be deleted.
   uint64_t mem_next_logfile_number_;
 
+  // the earliest log containing a prepared section
+  // which has been inserted into this memtable.
+  std::atomic<uint64_t> min_prep_log_referenced_;
+
   // rw locks for inplace updates
   std::vector<port::RWMutex> locks_;
 

db/memtable_list.cc

@@ -345,8 +345,8 @@ Status MemTableList::InstallMemtableFlushResults(
         imm_flush_needed.store(true, std::memory_order_release);
       }
       ++mem_id;
-    } while (!current_->memlist_.empty() && (m = current_->memlist_.back()) &&
-             m->file_number_ == file_number);
+    } while (!current_->memlist_.empty() && (nullptr != (m = current_->memlist_.back())) &&
+             (m->file_number_ == file_number));
   }
   commit_in_progress_ = false;
   return s;
@@ -392,4 +392,24 @@ void MemTableList::InstallNewVersion() {
   }
 }
 
+uint64_t MemTableList::GetMinLogContainingPrepSection() {
+  uint64_t min_log = 0;
+
+  for (auto& m : current_->memlist_) {
+    // this mem has been flushed it no longer
+    // needs to hold on the its prep section
+    if (m->flush_completed_) {
+      continue;
+    }
+
+    auto log = m->GetMinLogContainingPrepSection();
+
+    if (log > 0 && (min_log == 0 || log < min_log)) {
+      min_log = log;
+    }
+  }
+
+  return min_log;
+}
+
 }  // namespace rocksdb

db/memtable_list.h

@@ -215,6 +215,8 @@ class MemTableList {
 
   size_t* current_memory_usage() { return &current_memory_usage_; }
 
+  uint64_t GetMinLogContainingPrepSection();
+
  private:
   // DB mutex held
   void InstallNewVersion();

db/merge_context.h

@@ -37,6 +37,11 @@ public:
     Initialize();
     operand_list->push_front(operand_slice.ToString());
   }
+  // Push back a merge operand
+  void PushOperandBack(const Slice& operand_slice) {
+    Initialize();
+    operand_list->push_back(operand_slice.ToString());
+  }
   // return total number of operands in the list
   size_t GetNumOperands() const {
     if (!operand_list) {

db/perf_context_test.cc

@@ -9,16 +9,16 @@
 #include <vector>
 
 #include "rocksdb/db.h"
+#include "rocksdb/memtablerep.h"
 #include "rocksdb/perf_context.h"
 #include "rocksdb/slice_transform.h"
-#include "rocksdb/memtablerep.h"
 #include "util/histogram.h"
 #include "util/instrumented_mutex.h"
 #include "util/stop_watch.h"
+#include "util/string_util.h"
 #include "util/testharness.h"
 #include "util/thread_status_util.h"
-#include "util/string_util.h"
-
+#include "utilities/merge_operators.h"
 
 bool FLAGS_random_key = false;
 bool FLAGS_use_set_based_memetable = false;
@@ -37,6 +37,7 @@ std::shared_ptr<DB> OpenDb(bool read_only = false) {
     DB* db;
     Options options;
     options.create_if_missing = true;
+    options.max_open_files = -1;
     options.write_buffer_size = FLAGS_write_buffer_size;
     options.max_write_buffer_number = FLAGS_max_write_buffer_number;
     options.min_write_buffer_number_to_merge =
@@ -279,14 +280,19 @@ void ProfileQueries(bool enabled_time = false) {
 #endif
 
   for (const int i : keys) {
+    if (i == kFlushFlag) {
+      continue;
+    }
     std::string key = "k" + ToString(i);
-    std::string value = "v" + ToString(i);
+    std::string expected_value = "v" + ToString(i);
+    std::string value;
 
     std::vector<Slice> multiget_keys = {Slice(key)};
     std::vector<std::string> values;
 
     perf_context.Reset();
-    db->Get(read_options, key, &value);
+    ASSERT_OK(db->Get(read_options, key, &value));
+    ASSERT_EQ(expected_value, value);
     hist_get_snapshot.Add(perf_context.get_snapshot_time);
     hist_get_memtable.Add(perf_context.get_from_memtable_time);
     hist_get_files.Add(perf_context.get_from_output_files_time);
@@ -375,14 +381,19 @@ void ProfileQueries(bool enabled_time = false) {
   hist_mget_num_memtable_checked.Clear();
 
   for (const int i : keys) {
+    if (i == kFlushFlag) {
+      continue;
+    }
     std::string key = "k" + ToString(i);
-    std::string value = "v" + ToString(i);
+    std::string expected_value = "v" + ToString(i);
+    std::string value;
 
     std::vector<Slice> multiget_keys = {Slice(key)};
     std::vector<std::string> values;
 
     perf_context.Reset();
-    db->Get(read_options, key, &value);
+    ASSERT_OK(db->Get(read_options, key, &value));
+    ASSERT_EQ(expected_value, value);
     hist_get_snapshot.Add(perf_context.get_snapshot_time);
     hist_get_memtable.Add(perf_context.get_from_memtable_time);
     hist_get_files.Add(perf_context.get_from_output_files_time);
@@ -602,6 +613,41 @@ TEST_F(PerfContextTest, ToString) {
   ASSERT_EQ(std::string::npos, zero_excluded.find("= 0"));
   ASSERT_NE(std::string::npos, zero_excluded.find("= 12345"));
 }
+
+TEST_F(PerfContextTest, MergeOperatorTime) {
+  DestroyDB(kDbName, Options());
+  DB* db;
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator = MergeOperators::CreateStringAppendOperator();
+  Status s = DB::Open(options, kDbName, &db);
+  EXPECT_OK(s);
+
+  std::string val;
+  ASSERT_OK(db->Merge(WriteOptions(), "k1", "val1"));
+  ASSERT_OK(db->Merge(WriteOptions(), "k1", "val2"));
+  ASSERT_OK(db->Merge(WriteOptions(), "k1", "val3"));
+  ASSERT_OK(db->Merge(WriteOptions(), "k1", "val4"));
+
+  SetPerfLevel(kEnableTime);
+  perf_context.Reset();
+  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+
+  ASSERT_OK(db->Flush(FlushOptions()));
+
+  perf_context.Reset();
+  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+
+  ASSERT_OK(db->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  perf_context.Reset();
+  ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
+  EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
+
+  delete db;
+}
 }
 
 int main(int argc, char** argv) {

db/pinned_iterators_manager.h

@@ -0,0 +1,66 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+//
+#pragma once
+#include <algorithm>
+#include <memory>
+#include <vector>
+
+#include "table/internal_iterator.h"
+
+namespace rocksdb {
+
+// PinnedIteratorsManager will be notified whenever we need to pin an Iterator
+// and it will be responsible for deleting pinned Iterators when they are
+// not needed anymore.
+class PinnedIteratorsManager {
+ public:
+  PinnedIteratorsManager() : pinning_enabled(false), pinned_iters_(nullptr) {}
+  ~PinnedIteratorsManager() { assert(!pinning_enabled); }
+
+  // Enable Iterators pinning
+  void StartPinning() {
+    if (!pinning_enabled) {
+      pinning_enabled = true;
+      if (!pinned_iters_) {
+        pinned_iters_.reset(new std::vector<InternalIterator*>());
+      }
+    }
+  }
+
+  // Is pinning enabled ?
+  bool PinningEnabled() { return pinning_enabled; }
+
+  // Take ownership of iter if pinning is enabled and delete it when
+  // ReleasePinnedIterators() is called
+  void PinIteratorIfNeeded(InternalIterator* iter) {
+    if (!pinning_enabled || !iter) {
+      return;
+    }
+    pinned_iters_->push_back(iter);
+  }
+
+  // Release pinned Iterators
+  void ReleasePinnedIterators() {
+    if (pinning_enabled) {
+      pinning_enabled = false;
+
+      // Remove duplicate pointers
+      std::sort(pinned_iters_->begin(), pinned_iters_->end());
+      std::unique(pinned_iters_->begin(), pinned_iters_->end());
+
+      for (auto& iter : *pinned_iters_) {
+        delete iter;
+      }
+      pinned_iters_->clear();
+    }
+  }
+
+ private:
+  bool pinning_enabled;
+  std::unique_ptr<std::vector<InternalIterator*>> pinned_iters_;
+};
+
+}  // namespace rocksdb

db/plain_table_db_test.cc

@@ -263,7 +263,9 @@ class TestPlainTableReader : public PlainTableReader {
                        const TableProperties* table_properties,
                        unique_ptr<RandomAccessFileReader>&& file,
                        const ImmutableCFOptions& ioptions,
-                       bool* expect_bloom_not_match, bool store_index_in_file)
+                       bool* expect_bloom_not_match, bool store_index_in_file,
+                       uint32_t column_family_id,
+                       const std::string& column_family_name)
       : PlainTableReader(ioptions, std::move(file), env_options, icomparator,
                          encoding_type, file_size, table_properties),
         expect_bloom_not_match_(expect_bloom_not_match) {
@@ -276,6 +278,8 @@ class TestPlainTableReader : public PlainTableReader {
     EXPECT_TRUE(s.ok());
 
     TableProperties* props = const_cast<TableProperties*>(table_properties);
+    EXPECT_EQ(column_family_id, static_cast<uint32_t>(props->column_family_id));
+    EXPECT_EQ(column_family_name, props->column_family_name);
     if (store_index_in_file) {
       auto bloom_version_ptr = props->user_collected_properties.find(
           PlainTablePropertyNames::kBloomVersion);
@@ -308,13 +312,17 @@ extern const uint64_t kPlainTableMagicNumber;
 class TestPlainTableFactory : public PlainTableFactory {
  public:
   explicit TestPlainTableFactory(bool* expect_bloom_not_match,
-                                 const PlainTableOptions& options)
+                                 const PlainTableOptions& options,
+                                 uint32_t column_family_id,
+                                 std::string column_family_name)
       : PlainTableFactory(options),
         bloom_bits_per_key_(options.bloom_bits_per_key),
         hash_table_ratio_(options.hash_table_ratio),
         index_sparseness_(options.index_sparseness),
         store_index_in_file_(options.store_index_in_file),
-        expect_bloom_not_match_(expect_bloom_not_match) {}
+        expect_bloom_not_match_(expect_bloom_not_match),
+        column_family_id_(column_family_id),
+        column_family_name_(std::move(column_family_name)) {}
 
   Status NewTableReader(const TableReaderOptions& table_reader_options,
                         unique_ptr<RandomAccessFileReader>&& file,
@@ -354,7 +362,7 @@ class TestPlainTableFactory : public PlainTableFactory {
         table_reader_options.internal_comparator, encoding_type, file_size,
         bloom_bits_per_key_, hash_table_ratio_, index_sparseness_, props,
         std::move(file), table_reader_options.ioptions, expect_bloom_not_match_,
-        store_index_in_file_));
+        store_index_in_file_, column_family_id_, column_family_name_));
 
     *table = std::move(new_reader);
     return s;
@@ -366,6 +374,8 @@ class TestPlainTableFactory : public PlainTableFactory {
   size_t index_sparseness_;
   bool store_index_in_file_;
   bool* expect_bloom_not_match_;
+  const uint32_t column_family_id_;
+  const std::string column_family_name_;
 };
 
 TEST_P(PlainTableDBTest, Flush) {
@@ -492,7 +502,8 @@ TEST_P(PlainTableDBTest, Flush2) {
         plain_table_options.encoding_type = encoding_type;
         plain_table_options.store_index_in_file = store_index_in_file;
         options.table_factory.reset(new TestPlainTableFactory(
-            &expect_bloom_not_match, plain_table_options));
+            &expect_bloom_not_match, plain_table_options,
+            0 /* column_family_id */, kDefaultColumnFamilyName));
 
         DestroyAndReopen(&options);
         ASSERT_OK(Put("0000000000000bar", "b"));
@@ -561,7 +572,8 @@ TEST_P(PlainTableDBTest, Iterator) {
           plain_table_options.encoding_type = encoding_type;
 
           options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, plain_table_options));
+              &expect_bloom_not_match, plain_table_options,
+              0 /* column_family_id */, kDefaultColumnFamilyName));
         } else {
           PlainTableOptions plain_table_options;
           plain_table_options.user_key_len = 16;
@@ -572,7 +584,8 @@ TEST_P(PlainTableDBTest, Iterator) {
           plain_table_options.encoding_type = encoding_type;
 
           options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, plain_table_options));
+              &expect_bloom_not_match, plain_table_options,
+              0 /* column_family_id */, kDefaultColumnFamilyName));
         }
         DestroyAndReopen(&options);
 

db/prefix_test.cc

@@ -446,6 +446,10 @@ TEST_F(PrefixTest, PrefixValid) {
       iter->Next();
       ASSERT_FALSE(iter->Valid());
       ASSERT_EQ(kNotFoundResult, Get(db.get(), read_options, 12346, 8));
+
+      // Verify seeking past the prefix won't return a result.
+      SeekIterator(iter.get(), 12345, 10);
+      ASSERT_TRUE(!iter->Valid());
     }
   }
 }

db/repair.cc

@@ -265,7 +265,7 @@ class Repairer {
     mem->Ref();
     int counter = 0;
     while (reader.ReadRecord(&record, &scratch)) {
-      if (record.size() < 12) {
+      if (record.size() < WriteBatchInternal::kHeader) {
         reporter.Corruption(
             record.size(), Status::Corruption("log record too small"));
         continue;
@@ -291,12 +291,15 @@ class Repairer {
       ro.total_order_seek = true;
       Arena arena;
       ScopedArenaIterator iter(mem->NewIterator(ro, &arena));
+      MutableCFOptions mutable_cf_options(options_, ioptions_);
       status = BuildTable(
-          dbname_, env_, ioptions_, env_options_, table_cache_, iter.get(),
-          &meta, icmp_, &int_tbl_prop_collector_factories_,
-          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily, {},
-          kMaxSequenceNumber, kNoCompression, CompressionOptions(), false,
-          nullptr);
+          dbname_, env_, ioptions_, mutable_cf_options, env_options_,
+          table_cache_, iter.get(), &meta, icmp_,
+          &int_tbl_prop_collector_factories_,
+          TablePropertiesCollectorFactory::Context::kUnknownColumnFamily,
+          std::string() /* column_family_name */, {}, kMaxSequenceNumber,
+          kNoCompression, CompressionOptions(), false,
+          nullptr /* internal_stats */, TableFileCreationReason::kRecovery);
     }
     delete mem->Unref();
     delete cf_mems_default;

db/repair_test.cc

@@ -0,0 +1,177 @@
+//  Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under the BSD-style license found in the
+//  LICENSE file in the root directory of this source tree. An additional grant
+//  of patent rights can be found in the PATENTS file in the same directory.
+
+#include <algorithm>
+#include <string>
+#include <vector>
+
+#include "db/db_impl.h"
+#include "db/db_test_util.h"
+#include "rocksdb/db.h"
+#include "rocksdb/transaction_log.h"
+#include "util/file_util.h"
+
+namespace rocksdb {
+
+class RepairTest : public DBTestBase {
+ public:
+  RepairTest() : DBTestBase("/repair_test") {}
+
+  std::string GetFirstSstPath() {
+    uint64_t manifest_size;
+    std::vector<std::string> files;
+    db_->GetLiveFiles(files, &manifest_size);
+    auto sst_iter =
+        std::find_if(files.begin(), files.end(), [](const std::string& file) {
+          uint64_t number;
+          FileType type;
+          bool ok = ParseFileName(file, &number, &type);
+          return ok && type == kTableFile;
+        });
+    return sst_iter == files.end() ? "" : dbname_ + *sst_iter;
+  }
+};
+
+TEST_F(RepairTest, LostManifest) {
+  // Add a couple SST files, delete the manifest, and verify RepairDB() saves
+  // the day.
+  Put("key", "val");
+  Flush();
+  Put("key2", "val2");
+  Flush();
+  // Need to get path before Close() deletes db_, but delete it after Close() to
+  // ensure Close() didn't change the manifest.
+  std::string manifest_path =
+      DescriptorFileName(dbname_, dbfull()->TEST_Current_Manifest_FileNo());
+
+  Close();
+  ASSERT_OK(env_->FileExists(manifest_path));
+  ASSERT_OK(env_->DeleteFile(manifest_path));
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  ASSERT_EQ(Get("key"), "val");
+  ASSERT_EQ(Get("key2"), "val2");
+}
+
+TEST_F(RepairTest, CorruptManifest) {
+  // Manifest is in an invalid format. Expect a full recovery.
+  Put("key", "val");
+  Flush();
+  Put("key2", "val2");
+  Flush();
+  // Need to get path before Close() deletes db_, but overwrite it after Close()
+  // to ensure Close() didn't change the manifest.
+  std::string manifest_path =
+      DescriptorFileName(dbname_, dbfull()->TEST_Current_Manifest_FileNo());
+
+  Close();
+  ASSERT_OK(env_->FileExists(manifest_path));
+  CreateFile(env_, manifest_path, "blah");
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  ASSERT_EQ(Get("key"), "val");
+  ASSERT_EQ(Get("key2"), "val2");
+}
+
+TEST_F(RepairTest, IncompleteManifest) {
+  // In this case, the manifest is valid but does not reference all of the SST
+  // files. Expect a full recovery.
+  Put("key", "val");
+  Flush();
+  std::string orig_manifest_path =
+      DescriptorFileName(dbname_, dbfull()->TEST_Current_Manifest_FileNo());
+  CopyFile(orig_manifest_path, orig_manifest_path + ".tmp");
+  Put("key2", "val2");
+  Flush();
+  // Need to get path before Close() deletes db_, but overwrite it after Close()
+  // to ensure Close() didn't change the manifest.
+  std::string new_manifest_path =
+      DescriptorFileName(dbname_, dbfull()->TEST_Current_Manifest_FileNo());
+
+  Close();
+  ASSERT_OK(env_->FileExists(new_manifest_path));
+  // Replace the manifest with one that is only aware of the first SST file.
+  CopyFile(orig_manifest_path + ".tmp", new_manifest_path);
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  ASSERT_EQ(Get("key"), "val");
+  ASSERT_EQ(Get("key2"), "val2");
+}
+
+TEST_F(RepairTest, LostSst) {
+  // Delete one of the SST files but preserve the manifest that refers to it,
+  // then verify the DB is still usable for the intact SST.
+  Put("key", "val");
+  Flush();
+  Put("key2", "val2");
+  Flush();
+  auto sst_path = GetFirstSstPath();
+  ASSERT_FALSE(sst_path.empty());
+  ASSERT_OK(env_->DeleteFile(sst_path));
+
+  Close();
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  // Exactly one of the key-value pairs should be in the DB now.
+  ASSERT_TRUE((Get("key") == "val") != (Get("key2") == "val2"));
+}
+
+TEST_F(RepairTest, CorruptSst) {
+  // Corrupt one of the SST files but preserve the manifest that refers to it,
+  // then verify the DB is still usable for the intact SST.
+  Put("key", "val");
+  Flush();
+  Put("key2", "val2");
+  Flush();
+  auto sst_path = GetFirstSstPath();
+  ASSERT_FALSE(sst_path.empty());
+  CreateFile(env_, sst_path, "blah");
+
+  Close();
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  // Exactly one of the key-value pairs should be in the DB now.
+  ASSERT_TRUE((Get("key") == "val") != (Get("key2") == "val2"));
+}
+
+TEST_F(RepairTest, UnflushedSst) {
+  // This test case invokes repair while some data is unflushed, then verifies
+  // that data is in the db.
+  Put("key", "val");
+  VectorLogPtr wal_files;
+  ASSERT_OK(dbfull()->GetSortedWalFiles(wal_files));
+  ASSERT_EQ(wal_files.size(), 1);
+  uint64_t total_ssts_size;
+  GetAllSSTFiles(&total_ssts_size);
+  ASSERT_EQ(total_ssts_size, 0);
+  // Need to get path before Close() deletes db_, but delete it after Close() to
+  // ensure Close() didn't change the manifest.
+  std::string manifest_path =
+      DescriptorFileName(dbname_, dbfull()->TEST_Current_Manifest_FileNo());
+
+  Close();
+  ASSERT_OK(env_->FileExists(manifest_path));
+  ASSERT_OK(env_->DeleteFile(manifest_path));
+  RepairDB(dbname_, CurrentOptions());
+  Reopen(CurrentOptions());
+
+  ASSERT_OK(dbfull()->GetSortedWalFiles(wal_files));
+  ASSERT_EQ(wal_files.size(), 0);
+  GetAllSSTFiles(&total_ssts_size);
+  ASSERT_GT(total_ssts_size, 0);
+  ASSERT_EQ(Get("key"), "val");
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}

db/table_cache.cc

@@ -87,15 +87,16 @@ void TableCache::ReleaseHandle(Cache::Handle* handle) {
 Status TableCache::GetTableReader(
     const EnvOptions& env_options,
     const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
-    bool sequential_mode, bool record_read_stats, HistogramImpl* file_read_hist,
-    unique_ptr<TableReader>* table_reader, bool skip_filters) {
+    bool sequential_mode, size_t readahead, bool record_read_stats,
+    HistogramImpl* file_read_hist, unique_ptr<TableReader>* table_reader,
+    bool skip_filters, int level) {
   std::string fname =
       TableFileName(ioptions_.db_paths, fd.GetNumber(), fd.GetPathId());
   unique_ptr<RandomAccessFile> file;
   Status s = ioptions_.env->NewRandomAccessFile(fname, &file, env_options);
-  if (sequential_mode && ioptions_.compaction_readahead_size > 0) {
-    file = NewReadaheadRandomAccessFile(std::move(file),
-                                        ioptions_.compaction_readahead_size);
+
+  if (readahead > 0) {
+    file = NewReadaheadRandomAccessFile(std::move(file), readahead);
   }
   RecordTick(ioptions_.statistics, NO_FILE_OPENS);
   if (s.ok()) {
@@ -109,18 +110,26 @@ Status TableCache::GetTableReader(
                                    file_read_hist));
     s = ioptions_.table_factory->NewTableReader(
         TableReaderOptions(ioptions_, env_options, internal_comparator,
-                           skip_filters),
+                           skip_filters, level),
         std::move(file_reader), fd.GetFileSize(), table_reader);
     TEST_SYNC_POINT("TableCache::GetTableReader:0");
   }
   return s;
 }
 
+void TableCache::EraseHandle(const FileDescriptor& fd, Cache::Handle* handle) {
+  ReleaseHandle(handle);
+  uint64_t number = fd.GetNumber();
+  Slice key = GetSliceForFileNumber(&number);
+  cache_->Erase(key);
+}
+
 Status TableCache::FindTable(const EnvOptions& env_options,
                              const InternalKeyComparator& internal_comparator,
                              const FileDescriptor& fd, Cache::Handle** handle,
                              const bool no_io, bool record_read_stats,
-                             HistogramImpl* file_read_hist, bool skip_filters) {
+                             HistogramImpl* file_read_hist, bool skip_filters,
+                             int level) {
   PERF_TIMER_GUARD(find_table_nanos);
   Status s;
   uint64_t number = fd.GetNumber();
@@ -135,16 +144,21 @@ Status TableCache::FindTable(const EnvOptions& env_options,
     }
     unique_ptr<TableReader> table_reader;
     s = GetTableReader(env_options, internal_comparator, fd,
-                       false /* sequential mode */, record_read_stats,
-                       file_read_hist, &table_reader, skip_filters);
+                       false /* sequential mode */, 0 /* readahead */,
+                       record_read_stats, file_read_hist, &table_reader,
+                       skip_filters, level);
     if (!s.ok()) {
       assert(table_reader == nullptr);
       RecordTick(ioptions_.statistics, NO_FILE_ERRORS);
       // We do not cache error results so that if the error is transient,
       // or somebody repairs the file, we recover automatically.
     } else {
-      *handle = cache_->Insert(key, table_reader.release(), 1,
-                               &DeleteEntry<TableReader>);
+      s = cache_->Insert(key, table_reader.get(), 1, &DeleteEntry<TableReader>,
+                         handle);
+      if (s.ok()) {
+        // Release ownership of table reader.
+        table_reader.release();
+      }
     }
   }
   return s;
@@ -154,7 +168,7 @@ InternalIterator* TableCache::NewIterator(
     const ReadOptions& options, const EnvOptions& env_options,
     const InternalKeyComparator& icomparator, const FileDescriptor& fd,
     TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
-    bool for_compaction, Arena* arena, bool skip_filters) {
+    bool for_compaction, Arena* arena, bool skip_filters, int level) {
   PERF_TIMER_GUARD(new_table_iterator_nanos);
 
   if (table_reader_ptr != nullptr) {
@@ -163,13 +177,25 @@ InternalIterator* TableCache::NewIterator(
 
   TableReader* table_reader = nullptr;
   Cache::Handle* handle = nullptr;
-  bool create_new_table_reader =
-      (for_compaction && ioptions_.new_table_reader_for_compaction_inputs);
+
+  size_t readahead = 0;
+  bool create_new_table_reader = false;
+  if (for_compaction) {
+    if (ioptions_.new_table_reader_for_compaction_inputs) {
+      readahead = ioptions_.compaction_readahead_size;
+      create_new_table_reader = true;
+    }
+  } else {
+    readahead = options.readahead_size;
+    create_new_table_reader = readahead > 0;
+  }
+
   if (create_new_table_reader) {
     unique_ptr<TableReader> table_reader_unique_ptr;
     Status s = GetTableReader(
-        env_options, icomparator, fd, /* sequential mode */ true,
-        /* record stats */ false, nullptr, &table_reader_unique_ptr);
+        env_options, icomparator, fd, true /* sequential_mode */, readahead,
+        !for_compaction /* record stats */, nullptr, &table_reader_unique_ptr,
+        false /* skip_filters */, level);
     if (!s.ok()) {
       return NewErrorInternalIterator(s, arena);
     }
@@ -180,7 +206,7 @@ InternalIterator* TableCache::NewIterator(
       Status s = FindTable(env_options, icomparator, fd, &handle,
                            options.read_tier == kBlockCacheTier /* no_io */,
                            !for_compaction /* record read_stats */,
-                           file_read_hist, skip_filters);
+                           file_read_hist, skip_filters, level);
       if (!s.ok()) {
         return NewErrorInternalIterator(s, arena);
       }
@@ -212,7 +238,7 @@ Status TableCache::Get(const ReadOptions& options,
                        const InternalKeyComparator& internal_comparator,
                        const FileDescriptor& fd, const Slice& k,
                        GetContext* get_context, HistogramImpl* file_read_hist,
-                       bool skip_filters) {
+                       bool skip_filters, int level) {
   TableReader* t = fd.table_reader;
   Status s;
   Cache::Handle* handle = nullptr;
@@ -261,7 +287,8 @@ Status TableCache::Get(const ReadOptions& options,
   if (!t) {
     s = FindTable(env_options_, internal_comparator, fd, &handle,
                   options.read_tier == kBlockCacheTier /* no_io */,
-                  true /* record_read_stats */, file_read_hist, skip_filters);
+                  true /* record_read_stats */, file_read_hist, skip_filters,
+                  level);
     if (s.ok()) {
       t = GetTableReaderFromHandle(handle);
     }
@@ -285,9 +312,8 @@ Status TableCache::Get(const ReadOptions& options,
     size_t charge =
         row_cache_key.Size() + row_cache_entry->size() + sizeof(std::string);
     void* row_ptr = new std::string(std::move(*row_cache_entry));
-    auto row_handle = ioptions_.row_cache->Insert(
-        row_cache_key.GetKey(), row_ptr, charge, &DeleteEntry<std::string>);
-    ioptions_.row_cache->Release(row_handle);
+    ioptions_.row_cache->Insert(row_cache_key.GetKey(), row_ptr, charge,
+                                &DeleteEntry<std::string>);
   }
 #endif  // ROCKSDB_LITE
 

db/table_cache.h

@@ -45,34 +45,41 @@ class TableCache {
   // the cache and should not be deleted, and is valid for as long as the
   // returned iterator is live.
   // @param skip_filters Disables loading/accessing the filter block
+  // @param level The level this table is at, -1 for "not set / don't know"
   InternalIterator* NewIterator(
       const ReadOptions& options, const EnvOptions& toptions,
       const InternalKeyComparator& internal_comparator,
       const FileDescriptor& file_fd, TableReader** table_reader_ptr = nullptr,
       HistogramImpl* file_read_hist = nullptr, bool for_compaction = false,
-      Arena* arena = nullptr, bool skip_filters = false);
+      Arena* arena = nullptr, bool skip_filters = false, int level = -1);
 
   // If a seek to internal key "k" in specified file finds an entry,
   // call (*handle_result)(arg, found_key, found_value) repeatedly until
   // it returns false.
   // @param skip_filters Disables loading/accessing the filter block
+  // @param level The level this table is at, -1 for "not set / don't know"
   Status Get(const ReadOptions& options,
              const InternalKeyComparator& internal_comparator,
              const FileDescriptor& file_fd, const Slice& k,
              GetContext* get_context, HistogramImpl* file_read_hist = nullptr,
-             bool skip_filters = false);
+             bool skip_filters = false, int level = -1);
 
   // Evict any entry for the specified file number
   static void Evict(Cache* cache, uint64_t file_number);
 
+  // Clean table handle and erase it from the table cache
+  // Used in DB close, or the file is not live anymore.
+  void EraseHandle(const FileDescriptor& fd, Cache::Handle* handle);
+
   // Find table reader
   // @param skip_filters Disables loading/accessing the filter block
+  // @param level == -1 means not specified
   Status FindTable(const EnvOptions& toptions,
                    const InternalKeyComparator& internal_comparator,
                    const FileDescriptor& file_fd, Cache::Handle**,
                    const bool no_io = false, bool record_read_stats = true,
                    HistogramImpl* file_read_hist = nullptr,
-                   bool skip_filters = false);
+                   bool skip_filters = false, int level = -1);
 
   // Get TableReader from a cache handle.
   TableReader* GetTableReaderFromHandle(Cache::Handle* handle);
@@ -104,9 +111,10 @@ class TableCache {
   Status GetTableReader(const EnvOptions& env_options,
                         const InternalKeyComparator& internal_comparator,
                         const FileDescriptor& fd, bool sequential_mode,
-                        bool record_read_stats, HistogramImpl* file_read_hist,
+                        size_t readahead, bool record_read_stats,
+                        HistogramImpl* file_read_hist,
                         unique_ptr<TableReader>* table_reader,
-                        bool skip_filters = false);
+                        bool skip_filters = false, int level = -1);
 
   const ImmutableCFOptions& ioptions_;
   const EnvOptions& env_options_;

db/table_properties_collector.cc

@@ -23,6 +23,8 @@ Status InternalKeyPropertiesCollector::InternalAdd(const Slice& key,
   if (ikey.type == ValueType::kTypeDeletion ||
       ikey.type == ValueType::kTypeSingleDeletion) {
     ++deleted_keys_;
+  } else if (ikey.type == ValueType::kTypeMerge) {
+    ++merge_operands_;
   }
 
   return Status::OK();
@@ -33,19 +35,26 @@ Status InternalKeyPropertiesCollector::Finish(
   assert(properties);
   assert(properties->find(
         InternalKeyTablePropertiesNames::kDeletedKeys) == properties->end());
-  std::string val;
+  assert(properties->find(InternalKeyTablePropertiesNames::kMergeOperands) ==
+         properties->end());
 
-  PutVarint64(&val, deleted_keys_);
-  properties->insert({ InternalKeyTablePropertiesNames::kDeletedKeys, val });
+  std::string val_deleted_keys;
+  PutVarint64(&val_deleted_keys, deleted_keys_);
+  properties->insert(
+      {InternalKeyTablePropertiesNames::kDeletedKeys, val_deleted_keys});
+
+  std::string val_merge_operands;
+  PutVarint64(&val_merge_operands, merge_operands_);
+  properties->insert(
+      {InternalKeyTablePropertiesNames::kMergeOperands, val_merge_operands});
 
   return Status::OK();
 }
 
 UserCollectedProperties
 InternalKeyPropertiesCollector::GetReadableProperties() const {
-  return {
-    { "kDeletedKeys", ToString(deleted_keys_) }
-  };
+  return {{"kDeletedKeys", ToString(deleted_keys_)},
+          {"kMergeOperands", ToString(merge_operands_)}};
 }
 
 namespace {
@@ -65,6 +74,20 @@ EntryType GetEntryType(ValueType value_type) {
   }
 }
 
+uint64_t GetUint64Property(const UserCollectedProperties& props,
+                           const std::string property_name,
+                           bool* property_present) {
+  auto pos = props.find(property_name);
+  if (pos == props.end()) {
+    *property_present = false;
+    return 0;
+  }
+  Slice raw = pos->second;
+  uint64_t val = 0;
+  *property_present = true;
+  return GetVarint64(&raw, &val) ? val : 0;
+}
+
 }  // namespace
 
 Status UserKeyTablePropertiesCollector::InternalAdd(const Slice& key,
@@ -92,16 +115,20 @@ UserKeyTablePropertiesCollector::GetReadableProperties() const {
 
 const std::string InternalKeyTablePropertiesNames::kDeletedKeys
   = "rocksdb.deleted.keys";
+const std::string InternalKeyTablePropertiesNames::kMergeOperands =
+    "rocksdb.merge.operands";
 
 uint64_t GetDeletedKeys(
     const UserCollectedProperties& props) {
-  auto pos = props.find(InternalKeyTablePropertiesNames::kDeletedKeys);
-  if (pos == props.end()) {
-    return 0;
-  }
-  Slice raw = pos->second;
-  uint64_t val = 0;
-  return GetVarint64(&raw, &val) ? val : 0;
+  bool property_present_ignored;
+  return GetUint64Property(props, InternalKeyTablePropertiesNames::kDeletedKeys,
+                           &property_present_ignored);
+}
+
+uint64_t GetMergeOperands(const UserCollectedProperties& props,
+                          bool* property_present) {
+  return GetUint64Property(
+      props, InternalKeyTablePropertiesNames::kMergeOperands, property_present);
 }
 
 }  // namespace rocksdb

db/table_properties_collector.h

@@ -16,6 +16,7 @@ namespace rocksdb {
 
 struct InternalKeyTablePropertiesNames {
   static const std::string kDeletedKeys;
+  static const std::string kMergeOperands;
 };
 
 // Base class for internal table properties collector.
@@ -65,6 +66,7 @@ class InternalKeyPropertiesCollector : public IntTblPropCollector {
 
  private:
   uint64_t deleted_keys_ = 0;
+  uint64_t merge_operands_ = 0;
 };
 
 class InternalKeyPropertiesCollectorFactory

db/table_properties_collector_test.cc

@@ -36,6 +36,7 @@ class TablePropertiesTest : public testing::Test,
 // Utilities test functions
 namespace {
 static const uint32_t kTestColumnFamilyId = 66;
+static const std::string kTestColumnFamilyName = "test_column_fam";
 
 void MakeBuilder(const Options& options, const ImmutableCFOptions& ioptions,
                  const InternalKeyComparator& internal_comparator,
@@ -48,8 +49,8 @@ void MakeBuilder(const Options& options, const ImmutableCFOptions& ioptions,
 
   builder->reset(NewTableBuilder(
       ioptions, internal_comparator, int_tbl_prop_collector_factories,
-      kTestColumnFamilyId /* column_family_id */, writable->get(),
-      options.compression, options.compression_opts));
+      kTestColumnFamilyId, kTestColumnFamilyName,
+      writable->get(), options.compression, options.compression_opts));
 }
 }  // namespace
 
@@ -367,6 +368,8 @@ void TestInternalKeyPropertiesCollector(
       InternalKey("Y       ", 5, ValueType::kTypeDeletion),
       InternalKey("Z       ", 6, ValueType::kTypeDeletion),
       InternalKey("a       ", 7, ValueType::kTypeSingleDeletion),
+      InternalKey("b       ", 8, ValueType::kTypeMerge),
+      InternalKey("c       ", 9, ValueType::kTypeMerge),
   };
 
   std::unique_ptr<TableBuilder> builder;
@@ -422,6 +425,11 @@ void TestInternalKeyPropertiesCollector(
     uint64_t deleted = GetDeletedKeys(user_collected);
     ASSERT_EQ(5u, deleted);  // deletes + single-deletes
 
+    bool property_present;
+    uint64_t merges = GetMergeOperands(user_collected, &property_present);
+    ASSERT_TRUE(property_present);
+    ASSERT_EQ(2u, merges);
+
     if (sanitized) {
       uint32_t starts_with_A = 0;
       ASSERT_NE(user_collected.find("Count"), user_collected.end());

db/transaction_log_impl.cc

@@ -8,8 +8,8 @@
 #define __STDC_FORMAT_MACROS
 #endif
 
-#include <inttypes.h>
 #include "db/transaction_log_impl.h"
+#include <inttypes.h>
 #include "db/write_batch_internal.h"
 #include "util/file_reader_writer.h"
 
@@ -107,7 +107,7 @@ void TransactionLogIteratorImpl::SeekToStartSequence(
     return;
   }
   while (RestrictedRead(&record, &scratch)) {
-    if (record.size() < 12) {
+    if (record.size() < WriteBatchInternal::kHeader) {
       reporter_.Corruption(
         record.size(), Status::Corruption("very small log record"));
       continue;
@@ -167,7 +167,7 @@ void TransactionLogIteratorImpl::NextImpl(bool internal) {
       currentLogReader_->UnmarkEOF();
     }
     while (RestrictedRead(&record, &scratch)) {
-      if (record.size() < 12) {
+      if (record.size() < WriteBatchInternal::kHeader) {
         reporter_.Corruption(
           record.size(), Status::Corruption("very small log record"));
         continue;
@@ -250,7 +250,7 @@ void TransactionLogIteratorImpl::UpdateCurrentWriteBatch(const Slice& record) {
   // currentBatchSeq_ can only change here
   assert(currentLastSeq_ <= versions_->LastSequence());
 
-  currentBatch_ = move(batch);
+  currentBatch_ = std::move(batch);
   isValid_ = true;
   currentStatus_ = Status::OK();
 }
@@ -262,10 +262,9 @@ Status TransactionLogIteratorImpl::OpenLogReader(const LogFile* logFile) {
     return s;
   }
   assert(file);
-  currentLogReader_.reset(new log::Reader(options_->info_log,
-					  std::move(file), &reporter_,
-                                          read_options_.verify_checksums_, 0,
-                                          logFile->LogNumber()));
+  currentLogReader_.reset(new log::Reader(
+      options_->info_log, std::move(file), &reporter_,
+      read_options_.verify_checksums_, 0, logFile->LogNumber()));
   return Status::OK();
 }
 }  //  namespace rocksdb

db/version_builder.cc

@@ -309,11 +309,11 @@ class VersionBuilder::Rep {
 
         auto* file_meta = files_meta[file_idx].first;
         int level = files_meta[file_idx].second;
-        table_cache_->FindTable(env_options_,
-                                *(base_vstorage_->InternalComparator()),
-                                file_meta->fd, &file_meta->table_reader_handle,
-                                false /*no_io */, true /* record_read_stats */,
-                                internal_stats->GetFileReadHist(level));
+        table_cache_->FindTable(
+            env_options_, *(base_vstorage_->InternalComparator()),
+            file_meta->fd, &file_meta->table_reader_handle, false /*no_io */,
+            true /* record_read_stats */,
+            internal_stats->GetFileReadHist(level), false, level);
         if (file_meta->table_reader_handle != nullptr) {
           // Load table_reader
           file_meta->fd.table_reader = table_cache_->GetTableReaderFromHandle(

db/version_set.cc

@@ -23,6 +23,7 @@
 #include <vector>
 #include <string>
 
+#include "db/compaction.h"
 #include "db/filename.h"
 #include "db/internal_stats.h"
 #include "db/log_reader.h"
@@ -30,22 +31,22 @@
 #include "db/memtable.h"
 #include "db/merge_context.h"
 #include "db/table_cache.h"
-#include "db/compaction.h"
 #include "db/version_builder.h"
 #include "db/writebuffer.h"
 #include "rocksdb/env.h"
 #include "rocksdb/merge_operator.h"
-#include "table/internal_iterator.h"
-#include "table/table_reader.h"
-#include "table/merger.h"
-#include "table/two_level_iterator.h"
 #include "table/format.h"
-#include "table/plain_table_factory.h"
-#include "table/meta_blocks.h"
 #include "table/get_context.h"
+#include "table/internal_iterator.h"
+#include "table/merger.h"
+#include "table/meta_blocks.h"
+#include "table/plain_table_factory.h"
+#include "table/table_reader.h"
+#include "table/two_level_iterator.h"
 #include "util/coding.h"
 #include "util/file_reader_writer.h"
 #include "util/logging.h"
+#include "util/perf_context_imp.h"
 #include "util/stop_watch.h"
 #include "util/sync_point.h"
 
@@ -90,8 +91,9 @@ class FilePicker {
              const Comparator* user_comparator,
              const InternalKeyComparator* internal_comparator)
       : num_levels_(num_levels),
-        curr_level_(-1),
-        hit_file_level_(-1),
+        curr_level_(static_cast<unsigned int>(-1)),
+        returned_file_level_(static_cast<unsigned int>(-1)),
+        hit_file_level_(static_cast<unsigned int>(-1)),
         search_left_bound_(0),
         search_right_bound_(FileIndexer::kLevelMaxIndex),
 #ifndef NDEBUG
@@ -117,6 +119,8 @@ class FilePicker {
     }
   }
 
+  int GetCurrentLevel() { return returned_file_level_; }
+
   FdWithKeyRange* GetNextFile() {
     while (!search_ended_) {  // Loops over different levels.
       while (curr_index_in_curr_level_ < curr_file_level_->num_files) {
@@ -189,6 +193,7 @@ class FilePicker {
         }
         prev_file_ = f;
 #endif
+        returned_file_level_ = curr_level_;
         if (curr_level_ > 0 && cmp_largest < 0) {
           // No more files to search in this level.
           search_ended_ = !PrepareNextLevel();
@@ -215,6 +220,7 @@ class FilePicker {
  private:
   unsigned int num_levels_;
   unsigned int curr_level_;
+  unsigned int returned_file_level_;
   unsigned int hit_file_level_;
   int32_t search_left_bound_;
   int32_t search_right_bound_;
@@ -321,7 +327,7 @@ Version::~Version() {
       f->refs--;
       if (f->refs <= 0) {
         if (f->table_reader_handle) {
-          cfd_->table_cache()->ReleaseHandle(f->table_reader_handle);
+          cfd_->table_cache()->EraseHandle(f->fd, f->table_reader_handle);
           f->table_reader_handle = nullptr;
         }
         vset_->obsolete_files_.push_back(f);
@@ -485,7 +491,7 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
                          const EnvOptions& env_options,
                          const InternalKeyComparator& icomparator,
                          HistogramImpl* file_read_hist, bool for_compaction,
-                         bool prefix_enabled, bool skip_filters)
+                         bool prefix_enabled, bool skip_filters, int level)
       : TwoLevelIteratorState(prefix_enabled),
         table_cache_(table_cache),
         read_options_(read_options),
@@ -493,7 +499,8 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
         icomparator_(icomparator),
         file_read_hist_(file_read_hist),
         for_compaction_(for_compaction),
-        skip_filters_(skip_filters) {}
+        skip_filters_(skip_filters),
+        level_(level) {}
 
   InternalIterator* NewSecondaryIterator(const Slice& meta_handle) override {
     if (meta_handle.size() != sizeof(FileDescriptor)) {
@@ -505,7 +512,7 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
       return table_cache_->NewIterator(
           read_options_, env_options_, icomparator_, *fd,
           nullptr /* don't need reference to table*/, file_read_hist_,
-          for_compaction_, nullptr /* arena */, skip_filters_);
+          for_compaction_, nullptr /* arena */, skip_filters_, level_);
     }
   }
 
@@ -521,6 +528,7 @@ class LevelFileIteratorState : public TwoLevelIteratorState {
   HistogramImpl* file_read_hist_;
   bool for_compaction_;
   bool skip_filters_;
+  int level_;
 };
 
 // A wrapper of version builder which references the current version in
@@ -771,6 +779,21 @@ uint64_t VersionStorageInfo::GetEstimatedActiveKeys() const {
   }
 }
 
+double VersionStorageInfo::GetEstimatedCompressionRatioAtLevel(
+    int level) const {
+  assert(level < num_levels_);
+  uint64_t sum_file_size_bytes = 0;
+  uint64_t sum_data_size_bytes = 0;
+  for (auto* file_meta : files_[level]) {
+    sum_file_size_bytes += file_meta->fd.GetFileSize();
+    sum_data_size_bytes += file_meta->raw_key_size + file_meta->raw_value_size;
+  }
+  if (sum_file_size_bytes == 0) {
+    return -1.0;
+  }
+  return static_cast<double>(sum_data_size_bytes) / sum_file_size_bytes;
+}
+
 void Version::AddIterators(const ReadOptions& read_options,
                            const EnvOptions& soptions,
                            MergeIteratorBuilder* merge_iter_builder) {
@@ -788,7 +811,8 @@ void Version::AddIterators(const ReadOptions& read_options,
     const auto& file = storage_info_.LevelFilesBrief(0).files[i];
     merge_iter_builder->AddIterator(cfd_->table_cache()->NewIterator(
         read_options, soptions, cfd_->internal_comparator(), file.fd, nullptr,
-        cfd_->internal_stats()->GetFileReadHist(0), false, arena));
+        cfd_->internal_stats()->GetFileReadHist(0), false, arena,
+        false /* skip_filters */, 0 /* level */));
   }
 
   // For levels > 0, we can use a concatenating iterator that sequentially
@@ -803,7 +827,7 @@ void Version::AddIterators(const ReadOptions& read_options,
                                  cfd_->internal_stats()->GetFileReadHist(level),
                                  false /* for_compaction */,
                                  cfd_->ioptions()->prefix_extractor != nullptr,
-                                 IsFilterSkipped(level));
+                                 IsFilterSkipped(level), level);
       mem = arena->AllocateAligned(sizeof(LevelFileNumIterator));
       auto* first_level_iter = new (mem) LevelFileNumIterator(
           cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level));
@@ -908,7 +932,8 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
         read_options, *internal_comparator(), f->fd, ikey, &get_context,
         cfd_->internal_stats()->GetFileReadHist(fp.GetHitFileLevel()),
         IsFilterSkipped(static_cast<int>(fp.GetHitFileLevel()),
-                        fp.IsHitFileLastInLevel()));
+                        fp.IsHitFileLastInLevel()),
+        fp.GetCurrentLevel());
     // TODO: examine the behavior for corrupted key
     if (!status->ok()) {
       return;
@@ -948,9 +973,16 @@ void Version::Get(const ReadOptions& read_options, const LookupKey& k,
     }
     // merge_operands are in saver and we hit the beginning of the key history
     // do a final merge of nullptr and operands;
-    if (merge_operator_->FullMerge(user_key, nullptr,
-                                   merge_context->GetOperands(), value,
-                                   info_log_)) {
+    bool merge_success = false;
+    {
+      StopWatchNano timer(env_, db_statistics_ != nullptr);
+      PERF_TIMER_GUARD(merge_operator_time_nanos);
+      merge_success = merge_operator_->FullMerge(
+          user_key, nullptr, merge_context->GetOperands(), value, info_log_);
+      RecordTick(db_statistics_, MERGE_OPERATION_TOTAL_TIME,
+                 timer.ElapsedNanos());
+    }
+    if (merge_success) {
       *status = Status::OK();
     } else {
       RecordTick(db_statistics_, NUMBER_MERGE_FAILURES);
@@ -1062,6 +1094,12 @@ void Version::UpdateAccumulatedStats(bool update_stats) {
         if (MaybeInitializeFileMetaData(file_meta)) {
           // each FileMeta will be initialized only once.
           storage_info_.UpdateAccumulatedStats(file_meta);
+          // when option "max_open_files" is -1, all the file metadata has
+          // already been read, so MaybeInitializeFileMetaData() won't incur
+          // any I/O cost.
+          if (vset_->db_options_->max_open_files == -1) {
+            continue;
+          }
           if (++init_count >= kMaxInitCount) {
             break;
           }
@@ -1157,10 +1195,23 @@ void VersionStorageInfo::EstimateCompactionBytesNeeded(
   }
 
   // Level 1 and up.
+  uint64_t bytes_next_level = 0;
   for (int level = base_level(); level <= MaxInputLevel(); level++) {
     uint64_t level_size = 0;
-    for (auto* f : files_[level]) {
-      level_size += f->fd.GetFileSize();
+    if (bytes_next_level > 0) {
+#ifndef NDEBUG
+      uint64_t level_size2 = 0;
+      for (auto* f : files_[level]) {
+        level_size2 += f->fd.GetFileSize();
+      }
+      assert(level_size2 == bytes_next_level);
+#endif
+      level_size = bytes_next_level;
+      bytes_next_level = 0;
+    } else {
+      for (auto* f : files_[level]) {
+        level_size += f->fd.GetFileSize();
+      }
     }
     if (level == base_level() && level0_compact_triggered) {
       // Add base level size to compaction if level0 compaction triggered.
@@ -1172,18 +1223,29 @@ void VersionStorageInfo::EstimateCompactionBytesNeeded(
     uint64_t level_target = MaxBytesForLevel(level);
     if (level_size > level_target) {
       bytes_compact_to_next_level = level_size - level_target;
-      // Simplify to assume the actual compaction fan-out ratio is always
-      // mutable_cf_options.max_bytes_for_level_multiplier.
-      estimated_compaction_needed_bytes_ +=
-          bytes_compact_to_next_level *
-          (1 + mutable_cf_options.max_bytes_for_level_multiplier);
+      // Estimate the actual compaction fan-out ratio as size ratio between
+      // the two levels.
+
+      assert(bytes_next_level == 0);
+      if (level + 1 < num_levels_) {
+        for (auto* f : files_[level + 1]) {
+          bytes_next_level += f->fd.GetFileSize();
+        }
+      }
+      if (bytes_next_level > 0) {
+        assert(level_size > 0);
+        estimated_compaction_needed_bytes_ += static_cast<uint64_t>(
+            static_cast<double>(bytes_compact_to_next_level) *
+            (static_cast<double>(bytes_next_level) /
+                 static_cast<double>(level_size) +
+             1));
+      }
     }
   }
 }
 
 void VersionStorageInfo::ComputeCompactionScore(
-    const MutableCFOptions& mutable_cf_options,
-    const CompactionOptionsFIFO& compaction_options_fifo) {
+    const MutableCFOptions& mutable_cf_options) {
   for (int level = 0; level <= MaxInputLevel(); level++) {
     double score;
     if (level == 0) {
@@ -1219,7 +1281,7 @@ void VersionStorageInfo::ComputeCompactionScore(
 
       if (compaction_style_ == kCompactionStyleFIFO) {
         score = static_cast<double>(total_size) /
-                compaction_options_fifo.max_table_files_size;
+                mutable_cf_options.compaction_options_fifo.max_table_files_size;
       } else {
         score = static_cast<double>(num_sorted_runs) /
                 mutable_cf_options.level0_file_num_compaction_trigger;
@@ -1489,11 +1551,11 @@ void VersionStorageInfo::GenerateLevel0NonOverlapping() {
   std::vector<FdWithKeyRange> level0_sorted_file(
       level_files_brief_[0].files,
       level_files_brief_[0].files + level_files_brief_[0].num_files);
-  sort(level0_sorted_file.begin(), level0_sorted_file.end(),
-       [this](const FdWithKeyRange & f1, const FdWithKeyRange & f2)->bool {
-    return (internal_comparator_->Compare(f1.smallest_key, f2.smallest_key) <
-            0);
-  });
+  std::sort(level0_sorted_file.begin(), level0_sorted_file.end(),
+            [this](const FdWithKeyRange& f1, const FdWithKeyRange& f2) -> bool {
+              return (internal_comparator_->Compare(f1.smallest_key,
+                                                    f2.smallest_key) < 0);
+            });
 
   for (size_t i = 1; i < level0_sorted_file.size(); ++i) {
     FdWithKeyRange& f = level0_sorted_file[i];
@@ -2054,9 +2116,16 @@ VersionSet::VersionSet(const std::string& dbname, const DBOptions* db_options,
       env_options_(storage_options),
       env_options_compactions_(env_options_) {}
 
+void CloseTables(void* ptr, size_t) {
+  TableReader* table_reader = reinterpret_cast<TableReader*>(ptr);
+  table_reader->Close();
+}
+
 VersionSet::~VersionSet() {
   // we need to delete column_family_set_ because its destructor depends on
   // VersionSet
+  column_family_set_->get_table_cache()->ApplyToAllCacheEntries(&CloseTables,
+                                                                false);
   column_family_set_.reset();
   for (auto file : obsolete_files_) {
     delete file;
@@ -2068,8 +2137,7 @@ void VersionSet::AppendVersion(ColumnFamilyData* column_family_data,
                                Version* v) {
   // compute new compaction score
   v->storage_info()->ComputeCompactionScore(
-      *column_family_data->GetLatestMutableCFOptions(),
-      column_family_data->ioptions()->compaction_options_fifo);
+      *column_family_data->GetLatestMutableCFOptions());
 
   // Mark v finalized
   v->storage_info_.SetFinalized();
@@ -2252,8 +2320,6 @@ Status VersionSet::LogAndApply(ColumnFamilyData* column_family_data,
     if (s.ok() && new_descriptor_log) {
       s = SetCurrentFile(env_, dbname_, pending_manifest_file_number_,
                          db_options_->disableDataSync ? nullptr : db_directory);
-      // Leave the old file behind since PurgeObsoleteFiles will take care of it
-      // later. It's unsafe to delete now since file deletion may be disabled.
     }
 
     if (s.ok()) {
@@ -2272,6 +2338,13 @@ Status VersionSet::LogAndApply(ColumnFamilyData* column_family_data,
     mu->Lock();
   }
 
+  // Append the old mainfest file to the obsolete_manifests_ list to be deleted
+  // by PurgeObsoleteFiles later.
+  if (s.ok() && new_descriptor_log) {
+    obsolete_manifests_.emplace_back(
+        DescriptorFileName("", manifest_file_number_));
+  }
+
   // Install the new version
   if (s.ok()) {
     if (edit->is_column_family_add_) {
@@ -2304,11 +2377,14 @@ Status VersionSet::LogAndApply(ColumnFamilyData* column_family_data,
     manifest_file_size_ = new_manifest_file_size;
     prev_log_number_ = edit->prev_log_number_;
   } else {
+    std::string version_edits;
+    for (auto& e : batch_edits) {
+      version_edits = version_edits + "\n" + e->DebugString(true);
+    }
     Log(InfoLogLevel::ERROR_LEVEL, db_options_->info_log,
-        "Error in committing version %lu to [%s]",
-        (unsigned long)v->GetVersionNumber(),
-        column_family_data ? column_family_data->GetName().c_str()
-                           : "<null>");
+        "[%s] Error in committing version edit to MANIFEST: %s",
+        column_family_data ? column_family_data->GetName().c_str() : "<null>",
+        version_edits.c_str());
     delete v;
     if (new_descriptor_log) {
       Log(InfoLogLevel::INFO_LEVEL, db_options_->info_log,
@@ -3238,7 +3314,7 @@ void VersionSet::AddLiveFiles(std::vector<FileDescriptor>* live_list) {
   }
 }
 
-InternalIterator* VersionSet::MakeInputIterator(Compaction* c) {
+InternalIterator* VersionSet::MakeInputIterator(const Compaction* c) {
   auto cfd = c->column_family_data();
   ReadOptions read_options;
   read_options.verify_checksums =
@@ -3265,7 +3341,8 @@ InternalIterator* VersionSet::MakeInputIterator(Compaction* c) {
               read_options, env_options_compactions_,
               cfd->internal_comparator(), flevel->files[i].fd, nullptr,
               nullptr, /* no per level latency histogram*/
-              true /* for compaction */);
+              true /* for_compaction */, nullptr /* arena */,
+              false /* skip_filters */, (int)which /* level */);
         }
       } else {
         // Create concatenating iterator for the files from this level
@@ -3275,7 +3352,7 @@ InternalIterator* VersionSet::MakeInputIterator(Compaction* c) {
                 cfd->internal_comparator(),
                 nullptr /* no per level latency histogram */,
                 true /* for_compaction */, false /* prefix enabled */,
-                false /* skip_filters */),
+                false /* skip_filters */, (int)which /* level */),
             new LevelFileNumIterator(cfd->internal_comparator(),
                                      c->input_levels(which)));
       }
@@ -3388,7 +3465,10 @@ void VersionSet::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
 }
 
 void VersionSet::GetObsoleteFiles(std::vector<FileMetaData*>* files,
+                                  std::vector<std::string>* manifest_filenames,
                                   uint64_t min_pending_output) {
+  assert(manifest_filenames->empty());
+  obsolete_manifests_.swap(*manifest_filenames);
   std::vector<FileMetaData*> pending_files;
   for (auto f : obsolete_files_) {
     if (f->fd.GetNumber() < min_pending_output) {

db/version_set.h

@@ -24,6 +24,7 @@
 #include <map>
 #include <memory>
 #include <set>
+#include <string>
 #include <utility>
 #include <vector>
 
@@ -120,9 +121,7 @@ class VersionStorageInfo {
   // We use compaction scores to figure out which compaction to do next
   // REQUIRES: db_mutex held!!
   // TODO find a better way to pass compaction_options_fifo.
-  void ComputeCompactionScore(
-      const MutableCFOptions& mutable_cf_options,
-      const CompactionOptionsFIFO& compaction_options_fifo);
+  void ComputeCompactionScore(const MutableCFOptions& mutable_cf_options);
 
   // Estimate est_comp_needed_bytes_
   void EstimateCompactionBytesNeeded(
@@ -300,6 +299,8 @@ class VersionStorageInfo {
 
   uint64_t GetEstimatedActiveKeys() const;
 
+  double GetEstimatedCompressionRatioAtLevel(int level) const;
+
   // re-initializes the index that is used to offset into
   // files_by_compaction_pri_
   // to find the next compaction candidate file.
@@ -670,7 +671,7 @@ class VersionSet {
 
   // Create an iterator that reads over the compaction inputs for "*c".
   // The caller should delete the iterator when no longer needed.
-  InternalIterator* MakeInputIterator(Compaction* c);
+  InternalIterator* MakeInputIterator(const Compaction* c);
 
   // Add all files listed in any live version to *live.
   void AddLiveFiles(std::vector<FileDescriptor>* live_list);
@@ -697,6 +698,7 @@ class VersionSet {
   void GetLiveFilesMetaData(std::vector<LiveFileMetaData> *metadata);
 
   void GetObsoleteFiles(std::vector<FileMetaData*>* files,
+                        std::vector<std::string>* manifest_filenames,
                         uint64_t min_pending_output);
 
   ColumnFamilySet* GetColumnFamilySet() { return column_family_set_.get(); }
@@ -758,6 +760,7 @@ class VersionSet {
   uint64_t manifest_file_size_;
 
   std::vector<FileMetaData*> obsolete_files_;
+  std::vector<std::string> obsolete_manifests_;
 
   // env options for all reads and writes except compactions
   const EnvOptions& env_options_;