An odd fix for GCC 7

Summary:
There are two statuses in DoCompactionWork could be hidden.
This path fix this bug.

This seems to be a bug since CompactionFilterV2 is introduced:
https://github.com/facebook/rocksdb/blame/3.6.fb/db/db_impl.cc#L3294

Test Plan: make

Reviewers: nkg-, igor, sdong

Reviewed By: igor, sdong

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D28989

.gitignore

@@ -28,7 +28,13 @@ util/build_version.cc
 build_tools/VALGRIND_LOGS/
 coverage/COVERAGE_REPORT
 .gdbhistory
+package/
 .phutil_module_cache
 tags
 java/*.log
 java/include/org_rocksdb_*.h
+unity.cc
+java/crossbuild/.vagrant
+.vagrant/
+java/**.asc
+java/javadoc

.travis.yml

@@ -14,7 +14,6 @@ before_install:
  - sudo dpkg -i libgflags-dev_2.0-1_amd64.deb
 # Lousy hack to disable use and testing of fallocate, which doesn't behave quite
 # as EnvPosixTest::AllocateTest expects within the Travis OpenVZ environment.
- - sed -i "s/fallocate(/HACK_NO_fallocate(/" build_tools/build_detect_platform
-script: make check -j8
+script: OPT=-DTRAVIS make unity  && OPT=-DTRAVIS make check -j8
 notifications:
     email: false

AUTHORS

@@ -0,0 +1,11 @@
+Facebook Inc.
+Facebook Engineering Team
+
+Google Inc.
+# Initial version authors:
+Jeffrey Dean <jeff@google.com>
+Sanjay Ghemawat <sanjay@google.com>
+
+# Partial list of contributors:
+Kevin Regan <kevin.d.regan@gmail.com>
+Johan Bilien <jobi@litl.com>

HISTORY.md

@@ -1,5 +1,45 @@
 # Rocksdb Change Log
 
+## 3.6.0 (10/7/2014)
+### Disk format changes
+* If you're using RocksDB on ARM platforms and you're using default bloom filter, there is a disk format change you need to be aware of. There are three steps you need to do when you convert to new release: 1. turn off filter policy, 2. compact the whole database, 3. turn on filter policy
+
+### Behavior changes
+* We have refactored our system of stalling writes.  Any stall-related statistics' meanings are changed. Instead of per-write stall counts, we now count stalls per-epoch, where epochs are periods between flushes and compactions. You'll find more information in our Tuning Perf Guide once we release RocksDB 3.6.
+* When disableDataSync=true, we no longer sync the MANIFEST file.
+* Add identity_as_first_hash property to CuckooTable. SST file needs to be rebuilt to be opened by reader properly.
+
+### Public API changes
+* Change target_file_size_base type to uint64_t from int.
+* Remove allow_thread_local. This feature was proved to be stable, so we are turning it always-on.
+
+## 3.5.0 (9/3/2014)
+### New Features
+* Add include/utilities/write_batch_with_index.h, providing a utilitiy class to query data out of WriteBatch when building it.
+* Move BlockBasedTable related options to BlockBasedTableOptions from Options. Change corresponding JNI interface. Options affected include:
+  no_block_cache, block_cache, block_cache_compressed, block_size, block_size_deviation, block_restart_interval, filter_policy, whole_key_filtering. filter_policy is changed to shared_ptr from a raw pointer.
+* Remove deprecated options: disable_seek_compaction and db_stats_log_interval
+* OptimizeForPointLookup() takes one parameter for block cache size. It now builds hash index, bloom filter, and block cache.
+
+### Public API changes
+* The Prefix Extractor used with V2 compaction filters is now passed user key to SliceTransform::Transform instead of unparsed RocksDB key.
+
+## 3.4.0 (8/18/2014)
+### New Features
+* Support Multiple DB paths in universal style compactions
+* Add feature of storing plain table index and bloom filter in SST file.
+* CompactRange() will never output compacted files to level 0. This used to be the case when all the compaction input files were at level 0.
+* Added iterate_upper_bound to define the extent upto which the forward iterator will return entries. This will prevent iterating over delete markers and overwritten entries for edge cases where you want to break out the iterator anyways. This may improve perfomance in case there are a large number of delete markers or overwritten entries.
+
+### Public API changes
+* DBOptions.db_paths now is a vector of a DBPath structure which indicates both of path and target size
+* NewPlainTableFactory instead of bunch of parameters now accepts PlainTableOptions, which is defined in include/rocksdb/table.h
+* Moved include/utilities/*.h to include/rocksdb/utilities/*.h
+* Statistics APIs now take uint32_t as type instead of Tickers. Also make two access functions getTickerCount and histogramData const
+* Add DB property rocksdb.estimate-num-keys, estimated number of live keys in DB.
+* Add DB::GetIntProperty(), which returns DB properties that are integer as uint64_t.
+* The Prefix Extractor used with V2 compaction filters is now passed user key to SliceTransform::Transform instead of unparsed RocksDB key.
+
 ## 3.3.0 (7/10/2014)
 ### New Features
 * Added JSON API prototype.

INSTALL.md

@@ -15,6 +15,10 @@ There are few options when compiling RocksDB:
 * `make all` will compile our static library, and all our tools and unit tests. Our tools
 depend on gflags. You will need to have gflags installed to run `make all`.
 
+* if Intel SSE instruction set is supported, set USE_SSE=" -msse -msse4.2 " to make sure
+SSE4.2 is used to speed up CRC32 when calculating data checksum.
+
+
 ## Dependencies
 
 * You can link RocksDB with following compression libraries:
@@ -81,4 +85,4 @@ depend on gflags. You will need to have gflags installed to run `make all`.
     We did not run any production workloads on it.
 
 * **iOS**:
-  * Run: `TARGET_OS=IOS make static_lib`
+  * Run: `TARGET_OS=IOS make static_lib`. When building the project which uses rocksdb iOS library, make sure to define two important pre-processing macros: `ROCKSDB_LITE` and `IOS_CROSS_COMPILE`.

Makefile

@@ -3,7 +3,6 @@
 # found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 # Inherit some settings from environment variables, if available
-INSTALL_PATH ?= $(CURDIR)
 
 #-----------------------------------------------
 
@@ -49,7 +48,27 @@ else
 	PLATFORM_CCFLAGS += $(JEMALLOC_INCLUDE) -DHAVE_JEMALLOC
 endif
 
-WARNING_FLAGS = -Wall -Werror
+#-------------------------------------------------
+# make install related stuff
+INSTALL_PATH ?= /usr/local
+
+uninstall:
+	@rm -rf $(INSTALL_PATH)/include/rocksdb
+	@rm -rf $(INSTALL_PATH)/lib/$(LIBRARY)
+	@rm -rf $(INSTALL_PATH)/lib/$(SHARED)
+
+install:
+	@install -d $(INSTALL_PATH)/lib
+	@for header_dir in `find "include/rocksdb" -type d`; do \
+		install -d $(INSTALL_PATH)/$$header_dir; \
+	done
+	@for header in `find "include/rocksdb" -type f -name *.h`; do \
+		install -C -m 644 $$header $(INSTALL_PATH)/$$header; \
+	done
+	@[ ! -e $(LIBRARY) ] || install -C -m 644 $(LIBRARY) $(INSTALL_PATH)/lib
+	@[ ! -e $(SHARED) ] || install -C -m 644 $(SHARED) $(INSTALL_PATH)/lib
+#-------------------------------------------------
+
 CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual
 
@@ -69,6 +88,7 @@ VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full
 
 TESTS = \
 	db_test \
+	db_iter_test \
 	block_hash_index_test \
 	autovector_test \
 	column_family_test \
@@ -89,32 +109,40 @@ TESTS = \
 	blob_store_test \
 	filelock_test \
 	filename_test \
-	filter_block_test \
+	block_based_filter_block_test \
+	full_filter_block_test \
 	histogram_test \
 	log_test \
 	manual_compaction_test \
 	memenv_test \
 	merge_test \
+	merger_test \
 	redis_test \
 	reduce_levels_test \
 	plain_table_db_test \
 	prefix_test \
-	simple_table_db_test \
 	skiplist_test \
 	stringappend_test \
 	ttl_test \
 	backupable_db_test \
 	document_db_test \
 	json_document_test \
+	spatial_db_test \
 	version_edit_test \
 	version_set_test \
 	file_indexer_test \
-	write_batch_test\
+	write_batch_test \
+	write_controller_test\
 	deletefile_test \
 	table_test \
 	thread_local_test \
-        geodb_test \
-        rate_limiter_test
+	geodb_test \
+	rate_limiter_test \
+	options_test \
+	cuckoo_table_builder_test \
+	cuckoo_table_reader_test \
+	cuckoo_table_db_test \
+	write_batch_with_index_test
 
 TOOLS = \
         sst_dump \
@@ -122,9 +150,10 @@ TOOLS = \
         db_stress \
         ldb \
 	db_repl_stress \
+  options_test \
 	blob_store_bench
 
-PROGRAMS = db_bench signal_test table_reader_bench log_and_apply_bench $(TOOLS)
+PROGRAMS = db_bench signal_test table_reader_bench log_and_apply_bench cache_bench perf_context_test $(TOOLS)
 
 # The library name is configurable since we are maintaining libraries of both
 # debug/release mode.
@@ -134,6 +163,10 @@ endif
 LIBRARY = ${LIBNAME}.a
 MEMENVLIBRARY = libmemenv.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)
+ROCKSDB_PATCH = $(shell egrep "ROCKSDB_PATCH.[0-9]" include/rocksdb/version.h | cut -d ' ' -f 3)
+
 default: all
 
 #-----------------------------------------------
@@ -145,29 +178,33 @@ ifneq ($(PLATFORM_SHARED_VERSIONED),true)
 SHARED1 = ${LIBNAME}.$(PLATFORM_SHARED_EXT)
 SHARED2 = $(SHARED1)
 SHARED3 = $(SHARED1)
+SHARED4 = $(SHARED1)
 SHARED = $(SHARED1)
 else
-# Update db.h if you change these.
-SHARED_MAJOR = 3
-SHARED_MINOR = 3
+SHARED_MAJOR = $(ROCKSDB_MAJOR)
+SHARED_MINOR = $(ROCKSDB_MINOR)
+SHARED_PATCH = $(ROCKSDB_PATCH)
 SHARED1 = ${LIBNAME}.$(PLATFORM_SHARED_EXT)
 SHARED2 = $(SHARED1).$(SHARED_MAJOR)
 SHARED3 = $(SHARED1).$(SHARED_MAJOR).$(SHARED_MINOR)
-SHARED = $(SHARED1) $(SHARED2) $(SHARED3)
-$(SHARED1): $(SHARED3)
-	ln -fs $(SHARED3) $(SHARED1)
-$(SHARED2): $(SHARED3)
-	ln -fs $(SHARED3) $(SHARED2)
+SHARED4 = $(SHARED1).$(SHARED_MAJOR).$(SHARED_MINOR).$(SHARED_PATCH)
+SHARED = $(SHARED1) $(SHARED2) $(SHARED3) $(SHARED4)
+$(SHARED1): $(SHARED4)
+	ln -fs $(SHARED4) $(SHARED1)
+$(SHARED2): $(SHARED4)
+	ln -fs $(SHARED4) $(SHARED2)
+$(SHARED3): $(SHARED4)
+	ln -fs $(SHARED4) $(SHARED3)
 endif
 
-$(SHARED3):
+$(SHARED4):
 	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED2) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(SOURCES) $(LDFLAGS) -o $@
 
 endif  # PLATFORM_SHARED_EXT
 
-.PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests \
+.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
+	dbg rocksdbjavastatic rocksdbjava install uninstall
 
 all: $(LIBRARY) $(PROGRAMS) $(TESTS)
 
@@ -228,11 +265,18 @@ valgrind_check: all $(PROGRAMS) $(TESTS)
 		echo $$t $$((etime - stime)) >> $(VALGRIND_DIR)/valgrind_tests_times; \
 	done
 
+unity.cc:
+	$(shell (export ROCKSDB_ROOT="$(CURDIR)"; "$(CURDIR)/build_tools/unity" "$(CURDIR)/unity.cc"))
+
+unity: unity.cc unity.o
+	$(CXX) unity.o $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 clean:
-	-rm -f $(PROGRAMS) $(TESTS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) build_config.mk
+	-rm -f $(PROGRAMS) $(TESTS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) build_config.mk unity.cc
 	-rm -rf ios-x86/* ios-arm/*
-	-find . -name "*.[od]" -exec rm {} \;
+	-find . -name "*.[oda]" -exec rm {} \;
 	-find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
+	-rm -rf bzip2* snappy* zlib*
 tags:
 	ctags * -R
 	cscope -b `find . -name '*.cc'` `find . -name '*.h'`
@@ -240,6 +284,9 @@ tags:
 format:
 	build_tools/format-diff.sh
 
+package:
+	bash build_tools/make_package.sh $(SHARED_MAJOR).$(SHARED_MINOR)
+
 # ---------------------------------------------------------------------------
 # 	Unit tests and tools
 # ---------------------------------------------------------------------------
@@ -250,6 +297,9 @@ $(LIBRARY): $(LIBOBJECTS)
 db_bench: db/db_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(CXX) db/db_bench.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
 
+cache_bench: util/cache_bench.o $(LIBOBJECTS) $(TESTUTIL)
+	$(CXX) util/cache_bench.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
+
 block_hash_index_test: table/block_hash_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	 $(CXX) table/block_hash_index_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
@@ -322,15 +372,15 @@ crc32c_test: util/crc32c_test.o $(LIBOBJECTS) $(TESTHARNESS)
 db_test: db/db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/db_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
+db_iter_test: db/db_iter_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) db/db_iter_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 log_write_bench: util/log_write_bench.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/log_write_bench.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) -pg
 
 plain_table_db_test: db/plain_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/plain_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
-simple_table_db_test: db/simple_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(CXX) db/simple_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
-
 table_reader_bench: table/table_reader_bench.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) table/table_reader_bench.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) -pg
 
@@ -352,9 +402,15 @@ document_db_test: utilities/document/document_db_test.o $(LIBOBJECTS) $(TESTHARN
 json_document_test: utilities/document/json_document_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) utilities/document/json_document_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
 
+spatial_db_test: utilities/spatialdb/spatial_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) utilities/spatialdb/spatial_db_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
+
 ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
 
+write_batch_with_index_test: utilities/write_batch_with_index/write_batch_with_index_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) utilities/write_batch_with_index/write_batch_with_index_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
+
 dbformat_test: db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
@@ -367,8 +423,11 @@ rate_limiter_test: util/rate_limiter_test.o $(LIBOBJECTS) $(TESTHARNESS)
 filename_test: db/filename_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/filename_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
-filter_block_test: table/filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS)
-	$(CXX) table/filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+block_based_filter_block_test: table/block_based_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) table/block_based_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
+full_filter_block_test: table/full_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) table/full_filter_block_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
 log_test: db/log_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/log_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
@@ -397,15 +456,33 @@ reduce_levels_test: tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS)
 write_batch_test: db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
+write_controller_test: db/write_controller_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) db/write_controller_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 merge_test: db/merge_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/merge_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
+merger_test: table/merger_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) table/merger_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 deletefile_test: db/deletefile_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/deletefile_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
 geodb_test: utilities/geodb/geodb_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) utilities/geodb/geodb_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
 
+cuckoo_table_builder_test: table/cuckoo_table_builder_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) table/cuckoo_table_builder_test.o  $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
+cuckoo_table_reader_test: table/cuckoo_table_reader_test.o $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS)
+	$(CXX) table/cuckoo_table_reader_test.o  $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
+cuckoo_table_db_test: db/cuckoo_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) db/cuckoo_table_db_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
+options_test: util/options_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) util/options_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 $(MEMENVLIBRARY) : $(MEMENVOBJECTS)
 	rm -f $@
 	$(AR) -rs $@ $(MEMENVOBJECTS)
@@ -440,18 +517,69 @@ ldb: tools/ldb.o $(LIBOBJECTS)
 
 JNI_NATIVE_SOURCES = ./java/rocksjni/*.cc
 JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
-ROCKSDBJNILIB = ./java/librocksdbjni.so
+ARCH := $(shell getconf LONG_BIT)
+ROCKSDBJNILIB = librocksdbjni-linux$(ARCH).so
+ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux$(ARCH).jar
+ROCKSDB_JAR_ALL = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
+ROCKSDB_JAVADOCS_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-javadoc.jar
+ROCKSDB_SOURCES_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar
 
 ifeq ($(PLATFORM), OS_MACOSX)
-ROCKSDBJNILIB = ./java/librocksdbjni.jnilib
+ROCKSDBJNILIB = librocksdbjni-osx.jnilib
+ROCKSDB_JAR = rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar
 JAVA_INCLUDE = -I/System/Library/Frameworks/JavaVM.framework/Headers/
 endif
 
+libz.a:
+	-rm -rf zlib-1.2.8
+	curl -O http://zlib.net/zlib-1.2.8.tar.gz
+	tar xvzf zlib-1.2.8.tar.gz
+	cd zlib-1.2.8 && CFLAGS='-fPIC' ./configure --static && make
+	cp zlib-1.2.8/libz.a .
+
+libbz2.a:
+	-rm -rf bzip2-1.0.6
+	curl -O  http://www.bzip.org/1.0.6/bzip2-1.0.6.tar.gz
+	tar xvzf bzip2-1.0.6.tar.gz
+	cd bzip2-1.0.6 && make CFLAGS='-fPIC -Wall -Winline -O2 -g -D_FILE_OFFSET_BITS=64'
+	cp bzip2-1.0.6/libbz2.a .
+
+libsnappy.a:
+	-rm -rf snappy-1.1.1
+	curl -O https://snappy.googlecode.com/files/snappy-1.1.1.tar.gz
+	tar xvzf snappy-1.1.1.tar.gz
+	cd snappy-1.1.1 && ./configure --with-pic --enable-static
+	cd snappy-1.1.1 && make
+	cp snappy-1.1.1/.libs/libsnappy.a .
+
+
+rocksdbjavastatic: libz.a libbz2.a libsnappy.a
+	OPT="-fPIC -DNDEBUG -O2" $(MAKE) $(LIBRARY) -j
+	cd java;$(MAKE) java;
+	rm -f ./java/$(ROCKSDBJNILIB)
+	$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC -o ./java/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) $(LIBOBJECTS) $(COVERAGEFLAGS) libz.a libbz2.a libsnappy.a
+	cd java;jar -cf $(ROCKSDB_JAR) org/rocksdb/*.class org/rocksdb/util/*.class HISTORY*.md $(ROCKSDBJNILIB)
+	cd java/javadoc;jar -cf ../$(ROCKSDB_JAVADOCS_JAR) *
+	cd java;jar -cf $(ROCKSDB_SOURCES_JAR) org
+
+rocksdbjavastaticrelease: rocksdbjavastatic
+	cd java/crossbuild && vagrant destroy -f && vagrant up linux32 && vagrant halt linux32 && vagrant up linux64 && vagrant halt linux64
+	cd java;jar -cf $(ROCKSDB_JAR_ALL) org/rocksdb/*.class org/rocksdb/util/*.class HISTORY*.md librocksdbjni-*.so librocksdbjni-*.jnilib
+
+rocksdbjavastaticpublish: rocksdbjavastaticrelease
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-javadoc.jar -Dclassifier=javadoc
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-sources.jar -Dclassifier=sources
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux64.jar -Dclassifier=linux64
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-linux32.jar -Dclassifier=linux32
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH)-osx.jar -Dclassifier=osx
+	mvn gpg:sign-and-deploy-file -Durl=https://oss.sonatype.org/service/local/staging/deploy/maven2/ -DrepositoryId=sonatype-nexus-staging -DpomFile=java/rocksjni.pom -Dfile=java/rocksdbjni-$(ROCKSDB_MAJOR).$(ROCKSDB_MINOR).$(ROCKSDB_PATCH).jar
+
 rocksdbjava:
 	OPT="-fPIC -DNDEBUG -O2" $(MAKE) $(LIBRARY) -j32
 	cd java;$(MAKE) java;
-	rm -f $(ROCKSDBJNILIB)
-	$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC -o $(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) $(LIBOBJECTS) $(LDFLAGS) $(COVERAGEFLAGS)
+	rm -f ./java/$(ROCKSDBJNILIB)
+	$(CXX) $(CXXFLAGS) -I./java/. $(JAVA_INCLUDE) -shared -fPIC -o ./java/$(ROCKSDBJNILIB) $(JNI_NATIVE_SOURCES) $(LIBOBJECTS) $(JAVA_LDFLAGS) $(COVERAGEFLAGS)
+	cd java;jar -cf $(ROCKSDB_JAR) org/rocksdb/*.class org/rocksdb/util/*.class HISTORY*.md $(ROCKSDBJNILIB)
 
 jclean:
 	cd java;$(MAKE) clean;
@@ -528,8 +656,10 @@ ifneq ($(MAKECMDGOALS),clean)
 ifneq ($(MAKECMDGOALS),format)
 ifneq ($(MAKECMDGOALS),jclean)
 ifneq ($(MAKECMDGOALS),jtest)
+ifneq ($(MAKECMDGOALS),package)
 -include $(DEPFILES)
 endif
 endif
 endif
 endif
+endif

README.md

@@ -3,7 +3,7 @@
 [![Build Status](https://travis-ci.org/facebook/rocksdb.svg?branch=master)](https://travis-ci.org/facebook/rocksdb)
 
 RocksDB is developed and maintained by Facebook Database Engineering Team.
-It is built on on earlier work on LevelDB by Sanjay Ghemawat (sanjay@google.com)
+It is built on earlier work on LevelDB by Sanjay Ghemawat (sanjay@google.com)
 and Jeff Dean (jeff@google.com)
 
 This code is a library that forms the core building block for a fast

Vagrantfile

@@ -0,0 +1,16 @@
+Vagrant.configure("2") do |config|
+
+  config.vm.provider "virtualbox" do |v|
+    v.memory = 4096
+    v.cpus = 2
+  end
+
+  config.vm.define "ubuntu14" do |box|
+    box.vm.box = "ubuntu/trusty64"
+  end
+
+  config.vm.define "centos65" do |box|
+    box.vm.box = "chef/centos-6.5"
+  end
+
+end

build_tools/build_detect_platform

@@ -7,6 +7,7 @@
 #   CC                          C Compiler path
 #   CXX                         C++ Compiler path
 #   PLATFORM_LDFLAGS            Linker flags
+#   JAVA_LDFLAGS                Linker flags for RocksDBJava
 #   PLATFORM_SHARED_EXT         Extension for shared libraries
 #   PLATFORM_SHARED_LDFLAGS     Flags for building shared library
 #   PLATFORM_SHARED_CFLAGS      Flags for compiling objects for shared library
@@ -45,16 +46,10 @@ PLATFORM_CXXFLAGS="-std=c++11"
 COMMON_FLAGS="-DROCKSDB_PLATFORM_POSIX"
 
 # Default to fbcode gcc on internal fb machines
-if [ -d /mnt/gvfs/third-party -a -z "$CXX" ]; then
+if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
     FBCODE_BUILD="true"
     if [ -z "$USE_CLANG" ]; then
-        CENTOS_VERSION=`rpm -q --qf "%{VERSION}" \
-          $(rpm -q --whatprovides redhat-release)`
-        if [ "$CENTOS_VERSION" = "6" ]; then
-          source "$PWD/build_tools/fbcode.gcc481.sh"
-        else
-          source "$PWD/build_tools/fbcode.gcc471.sh"
-        fi
+        source $PWD/build_tools/fbcode_config.sh
     else
         source "$PWD/build_tools/fbcode.clang31.sh"
     fi
@@ -157,6 +152,8 @@ case "$TARGET_OS" in
         exit 1
 esac
 
+JAVA_LDFLAGS="$PLATFORM_LDFLAGS"
+
 if test -z "$DO_NOT_RUN_BUILD_DETECT_VERSION"; then
   "$PWD/build_tools/build_detect_version"
 fi
@@ -215,6 +212,7 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DSNAPPY"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lsnappy"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -lsnappy"
     fi
 
 
@@ -238,7 +236,7 @@ EOF
       int main() {}
 EOF
     if [ "$?" = 0 ]; then
-        COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=google"
+        COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=gflags"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
     fi
 
@@ -250,6 +248,7 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DZLIB"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lz"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -lz"
     fi
 
     # Test whether bzip library is installed
@@ -260,6 +259,7 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DBZIP2"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbz2"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -lbz2"
     fi
 
     # Test whether lz4 library is installed
@@ -271,6 +271,7 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DLZ4"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -llz4"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -llz4"
     fi
 
     # Test whether numa is available
@@ -282,6 +283,7 @@ EOF
     if [ "$?" = 0 ]; then
         COMMON_FLAGS="$COMMON_FLAGS -DNUMA"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lnuma"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -lnuma"
     fi
 
     # Test whether tcmalloc is available
@@ -290,6 +292,7 @@ EOF
 EOF
     if [ "$?" = 0 ]; then
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltcmalloc"
+        JAVA_LDFLAGS="$JAVA_LDFLAGS -ltcmalloc"
     fi
 fi
 
@@ -306,6 +309,7 @@ if test "$USE_HDFS"; then
   HDFS_LDFLAGS="$HDFS_LDFLAGS -ldl -lverify -ljava -ljvm"
   COMMON_FLAGS="$COMMON_FLAGS $HDFS_CCFLAGS"
   PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS $HDFS_LDFLAGS"
+  JAVA_LDFLAGS="$JAVA_LDFLAGS $HDFS_LDFLAGS"
 fi
 
 # if Intel SSE instruction set is supported, set USE_SSE=" -msse -msse4.2 "
@@ -316,10 +320,15 @@ PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS $COMMON_FLAGS"
 
 VALGRIND_VER="$VALGRIND_VER"
 
+ROCKSDB_MAJOR=`build_tools/version.sh major`
+ROCKSDB_MINOR=`build_tools/version.sh minor`
+ROCKSDB_PATCH=`build_tools/version.sh patch`
+
 echo "CC=$CC" >> "$OUTPUT"
 echo "CXX=$CXX" >> "$OUTPUT"
 echo "PLATFORM=$PLATFORM" >> "$OUTPUT"
 echo "PLATFORM_LDFLAGS=$PLATFORM_LDFLAGS" >> "$OUTPUT"
+echo "JAVA_LDFLAGS=$JAVA_LDFLAGS" >> "$OUTPUT"
 echo "VALGRIND_VER=$VALGRIND_VER" >> "$OUTPUT"
 echo "PLATFORM_CCFLAGS=$PLATFORM_CCFLAGS" >> "$OUTPUT"
 echo "PLATFORM_CXXFLAGS=$PLATFORM_CXXFLAGS" >> "$OUTPUT"
@@ -330,3 +339,6 @@ echo "PLATFORM_SHARED_VERSIONED=$PLATFORM_SHARED_VERSIONED" >> "$OUTPUT"
 echo "EXEC_LDFLAGS=$EXEC_LDFLAGS" >> "$OUTPUT"
 echo "JEMALLOC_INCLUDE=$JEMALLOC_INCLUDE" >> "$OUTPUT"
 echo "JEMALLOC_LIB=$JEMALLOC_LIB" >> "$OUTPUT"
+echo "ROCKSDB_MAJOR=$ROCKSDB_MAJOR" >> "$OUTPUT"
+echo "ROCKSDB_MINOR=$ROCKSDB_MINOR" >> "$OUTPUT"
+echo "ROCKSDB_PATCH=$ROCKSDB_PATCH" >> "$OUTPUT"

build_tools/dependencies.sh

@@ -0,0 +1,19 @@
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+GCC_BASE=/mnt/gvfs/third-party2/gcc/7331085db891a2ef4a88a48a751d834e8d68f4cb/7.x/centos7-native/b2ef2b6
+CLANG_BASE=/mnt/gvfs/third-party2/llvm-fb/963d9aeda70cc4779885b1277484fe7544a04e3e/9.0.0/platform007/9e92d53/
+LIBGCC_BASE=/mnt/gvfs/third-party2/libgcc/6ace84e956873d53638c738b6f65f3f469cca74c/7.x/platform007/5620abc
+GLIBC_BASE=/mnt/gvfs/third-party2/glibc/192b0f42d63dcf6210d6ceae387b49af049e6e0c/2.26/platform007/f259413
+SNAPPY_BASE=/mnt/gvfs/third-party2/snappy/7f9bdaada18f59bc27ec2b0871eb8a6144343aef/1.1.3/platform007/ca4da3d
+ZLIB_BASE=/mnt/gvfs/third-party2/zlib/2d9f0b9a4274cc21f61272a9e89bdb859bce8f1f/1.2.8/platform007/ca4da3d
+BZIP2_BASE=/mnt/gvfs/third-party2/bzip2/dc49a21c5fceec6456a7a28a94dcd16690af1337/1.0.6/platform007/ca4da3d
+LZ4_BASE=/mnt/gvfs/third-party2/lz4/0f607f8fc442ea7d6b876931b1898bb573d5e5da/1.9.1/platform007/ca4da3d
+ZSTD_BASE=/mnt/gvfs/third-party2/zstd/ca22bc441a4eb709e9e0b1f9fec9750fed7b31c5/1.4.x/platform007/15a3614
+GFLAGS_BASE=/mnt/gvfs/third-party2/gflags/0b9929d2588991c65a57168bf88aff2db87c5d48/2.2.0/platform007/ca4da3d
+JEMALLOC_BASE=/mnt/gvfs/third-party2/jemalloc/c26f08f47ac35fc31da2633b7da92d6b863246eb/master/platform007/c26c002
+NUMA_BASE=/mnt/gvfs/third-party2/numa/3f3fb57a5ccc5fd21c66416c0b83e0aa76a05376/2.0.11/platform007/ca4da3d
+LIBUNWIND_BASE=/mnt/gvfs/third-party2/libunwind/40c73d874898b386a71847f1b99115d93822d11f/1.4/platform007/6f3e0a9
+TBB_BASE=/mnt/gvfs/third-party2/tbb/4ce8e8dba77cdbd81b75d6f0c32fd7a1b76a11ec/2018_U5/platform007/ca4da3d
+KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/fb251ecd2f5ae16f8671f7014c246e52a748fe0b/fb/platform007/da39a3e
+BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/ab9f09bba370e7066cafd4eb59752db93f2e8312/2.29.1/platform007/15a3614
+VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/d42d152a15636529b0861ec493927200ebebca8e/3.15.0/platform007/ca4da3d
+LUA_BASE=/mnt/gvfs/third-party2/lua/f0cd714433206d5139df61659eb7b28b1dea6683/5.3.4/platform007/5007832

build_tools/fbcode_config.sh

@@ -0,0 +1,134 @@
+#!/bin/sh
+#
+# Set environment variables so that we can compile rocksdb using
+# fbcode settings.  It uses the latest g++ and clang compilers and also
+# uses jemalloc
+# Environment variables that change the behavior of this script:
+# PIC_BUILD -- if true, it will only take pic versions of libraries from fbcode. libraries that don't have pic variant will not be included
+
+
+BASEDIR=`dirname $BASH_SOURCE`
+source "$BASEDIR/dependencies.sh"
+
+CFLAGS=""
+
+# libgcc
+LIBGCC_INCLUDE="$LIBGCC_BASE/include/c++/7.3.0"
+LIBGCC_LIBS=" -L $LIBGCC_BASE/lib"
+
+# glibc
+GLIBC_INCLUDE="$GLIBC_BASE/include"
+GLIBC_LIBS=" -L $GLIBC_BASE/lib"
+
+# snappy
+SNAPPY_INCLUDE=" -I $SNAPPY_BASE/include/"
+if test -z $PIC_BUILD; then
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy.a"
+else
+  SNAPPY_LIBS=" $SNAPPY_BASE/lib/libsnappy_pic.a"
+fi
+CFLAGS+=" -DSNAPPY"
+
+if test -z $PIC_BUILD; then
+  # location of zlib headers and libraries
+  ZLIB_INCLUDE=" -I $ZLIB_BASE/include/"
+  ZLIB_LIBS=" $ZLIB_BASE/lib/libz.a"
+  CFLAGS+=" -DZLIB"
+
+  LZ4_INCLUDE=" -I $LZ4_BASE/include/"
+  LZ4_LIBS=" $LZ4_BASE/lib/liblz4.a"
+  CFLAGS+=" -DLZ4"
+fi
+
+# location of gflags headers and libraries
+GFLAGS_INCLUDE=" -I $GFLAGS_BASE/include/"
+if test -z $PIC_BUILD; then
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags.a"
+else
+  GFLAGS_LIBS=" $GFLAGS_BASE/lib/libgflags_pic.a"
+fi
+CFLAGS+=" -DGFLAGS=gflags"
+
+# location of jemalloc
+JEMALLOC_INCLUDE=" -I $JEMALLOC_BASE/include/"
+JEMALLOC_LIB=" $JEMALLOC_BASE/lib/libjemalloc.a"
+
+if test -z $PIC_BUILD; then
+  # location of libunwind
+  LIBUNWIND="$LIBUNWIND_BASE/lib/libunwind.a"
+fi
+
+# location of TBB
+TBB_INCLUDE=" -isystem $TBB_BASE/include/"
+if test -z $PIC_BUILD; then
+  TBB_LIBS="$TBB_BASE/lib/libtbb.a"
+else
+  TBB_LIBS="$TBB_BASE/lib/libtbb_pic.a"
+fi
+CFLAGS+=" -DTBB"
+
+# use Intel SSE support for checksum calculations
+export USE_SSE=" -msse -msse4.2 "
+
+BINUTILS="$BINUTILS_BASE/bin"
+AR="$BINUTILS/ar"
+
+DEPS_INCLUDE="$SNAPPY_INCLUDE $ZLIB_INCLUDE $LZ4_INCLUDE $GFLAGS_INCLUDE"
+
+STDLIBS="-L $GCC_BASE/lib64"
+
+CLANG_BIN="$CLANG_BASE/bin"
+CLANG_LIB="$CLANG_BASE/lib"
+CLANG_SRC="$CLANG_BASE/../../src"
+
+CLANG_ANALYZER="$CLANG_BIN/clang++"
+CLANG_SCAN_BUILD="$CLANG_SRC/llvm/tools/clang/tools/scan-build/bin/scan-build"
+
+if [ -z "$USE_CLANG" ]; then
+  # gcc
+  CC="$GCC_BASE/bin/gcc"
+  CXX="$GCC_BASE/bin/g++"
+
+  CFLAGS+=" -B$BINUTILS/gold"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  JEMALLOC=1
+else
+  # clang
+  CLANG_INCLUDE="$CLANG_LIB/clang/stable/include"
+  CC="$CLANG_BIN/clang"
+  CXX="$CLANG_BIN/clang++"
+
+  KERNEL_HEADERS_INCLUDE="$KERNEL_HEADERS_BASE/include"
+
+  CFLAGS+=" -B$BINUTILS/gold -nostdinc -nostdlib"
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x "
+  CFLAGS+=" -isystem $LIBGCC_BASE/include/c++/7.x/x86_64-facebook-linux "
+  CFLAGS+=" -isystem $GLIBC_INCLUDE"
+  CFLAGS+=" -isystem $LIBGCC_INCLUDE"
+  CFLAGS+=" -isystem $CLANG_INCLUDE"
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE/linux "
+  CFLAGS+=" -isystem $KERNEL_HEADERS_INCLUDE "
+  CFLAGS+=" -Wno-expansion-to-defined "
+  CXXFLAGS="-nostdinc++"
+fi
+
+CFLAGS+=" $DEPS_INCLUDE"
+CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX -DROCKSDB_FALLOCATE_PRESENT -DROCKSDB_MALLOC_USABLE_SIZE -DROCKSDB_RANGESYNC_PRESENT -DROCKSDB_SCHED_GETCPU_PRESENT -DROCKSDB_SUPPORT_THREAD_LOCAL -DHAVE_SSE42"
+CXXFLAGS+=" $CFLAGS"
+
+EXEC_LDFLAGS=" $SNAPPY_LIBS $ZLIB_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $NUMA_LIB $TBB_LIBS"
+EXEC_LDFLAGS+=" -B$BINUTILS/gold"
+EXEC_LDFLAGS+=" -Wl,--dynamic-linker,/usr/local/fbcode/platform007/lib/ld.so"
+EXEC_LDFLAGS+=" $LIBUNWIND"
+EXEC_LDFLAGS+=" -Wl,-rpath=/usr/local/fbcode/platform007/lib"
+# required by libtbb
+EXEC_LDFLAGS+=" -ldl"
+
+PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS $STDLIBS -lgcc -lstdc++"
+
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $LZ4_LIBS $ZSTD_LIBS $GFLAGS_LIBS $TBB_LIBS"
+
+VALGRIND_VER="$VALGRIND_BASE/bin/"
+
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD 

build_tools/make_package.sh

@@ -0,0 +1,116 @@
+#/usr/bin/env bash
+
+set -e
+
+function log() {
+  echo "[+] $1"
+}
+
+function fatal() {
+  echo "[!] $1"
+  exit 1
+}
+
+function platform() {
+  local  __resultvar=$1
+  if [[ -f "/etc/yum.conf" ]]; then
+    eval $__resultvar="centos"
+  elif [[ -f "/etc/dpkg/dpkg.cfg" ]]; then
+    eval $__resultvar="ubuntu"
+  else
+    fatal "Unknwon operating system"
+  fi
+}
+platform OS
+
+function package() {
+  if [[ $OS = "ubuntu" ]]; then
+    if dpkg --get-selections | grep --quiet $1; then
+      log "$1 is already installed. skipping."
+    else
+      apt-get install $@ -y
+    fi
+  elif [[ $OS = "centos" ]]; then
+    if rpm -qa | grep --quiet $1; then
+      log "$1 is already installed. skipping."
+    else
+      yum install $@ -y
+    fi
+  fi
+}
+
+function detect_fpm_output() {
+  if [[ $OS = "ubuntu" ]]; then
+    export FPM_OUTPUT=deb
+  elif [[ $OS = "centos" ]]; then
+    export FPM_OUTPUT=rpm
+  fi
+}
+detect_fpm_output
+
+function gem_install() {
+  if gem list | grep --quiet $1; then
+    log "$1 is already installed. skipping."
+  else
+    gem install $@
+  fi
+}
+
+function main() {
+  if [[ $# -ne 1 ]]; then
+    fatal "Usage: $0 <rocksdb_version>"
+  else
+    log "using rocksdb version: $1"
+  fi
+
+  if [[ -d /vagrant ]]; then
+    if [[ $OS = "ubuntu" ]]; then
+      package g++-4.7
+      export CXX=g++-4.7
+
+      # the deb would depend on libgflags2, but the static lib is the only thing
+      # installed by make install
+      package libgflags-dev
+
+      package ruby-all-dev
+    elif [[ $OS = "centos" ]]; then
+      pushd /etc/yum.repos.d
+      if [[ ! -f /etc/yum.repos.d/devtools-1.1.repo ]]; then
+        wget http://people.centos.org/tru/devtools-1.1/devtools-1.1.repo
+      fi
+      package devtoolset-1.1-gcc --enablerepo=testing-1.1-devtools-6
+      package devtoolset-1.1-gcc-c++ --enablerepo=testing-1.1-devtools-6
+      export CC=/opt/centos/devtoolset-1.1/root/usr/bin/gcc
+      export CPP=/opt/centos/devtoolset-1.1/root/usr/bin/cpp
+      export CXX=/opt/centos/devtoolset-1.1/root/usr/bin/c++
+      export PATH=$PATH:/opt/centos/devtoolset-1.1/root/usr/bin
+      popd
+      if ! rpm -qa | grep --quiet gflags; then
+        rpm -i https://github.com/schuhschuh/gflags/releases/download/v2.1.0/gflags-devel-2.1.0-1.amd64.rpm
+      fi
+
+      package ruby
+      package ruby-devel
+      package rubygems
+      package rpm-build
+    fi
+  fi
+  gem_install fpm
+
+  make static_lib
+  make install INSTALL_PATH=package
+  fpm \
+    -s dir \
+    -t $FPM_OUTPUT \
+    -n rocksdb \
+    -v $1 \
+    --prefix /usr \
+    --url http://rocksdb.org/ \
+    -m rocksdb@fb.com \
+    --license BSD \
+    --vendor Facebook \
+    --description "RocksDB is an embeddable persistent key-value store for fast storage." \
+    package
+}
+
+main $@

build_tools/regression_build_test.sh

@@ -344,6 +344,38 @@ common_in_mem_args="--db=/dev/shm/rocksdb \
     --threads=32 \
     --writes_per_second=81920 > ${STAT_FILE}.seekwhilewriting_in_ram
 
+# measure fillseq with bunch of column families
+./db_bench \
+    --benchmarks=fillseq \
+    --num_column_families=500 \
+    --write_buffer_size=1048576 \
+    --db=$DATA_DIR \
+    --use_existing_db=0 \
+    --num=$NUM \
+    --writes=$NUM \
+    --open_files=55000 \
+    --statistics=1 \
+    --histogram=1 \
+    --disable_data_sync=1 \
+    --disable_wal=1 \
+    --sync=0  > ${STAT_FILE}.fillseq_lots_column_families
+
+# measure overwrite performance with bunch of column families
+./db_bench \
+    --benchmarks=overwrite \
+    --num_column_families=500 \
+    --write_buffer_size=1048576 \
+    --db=$DATA_DIR \
+    --use_existing_db=1 \
+    --num=$NUM \
+    --writes=$((NUM / 10)) \
+    --open_files=55000 \
+    --statistics=1 \
+    --histogram=1 \
+    --disable_data_sync=1 \
+    --disable_wal=1 \
+    --sync=0 \
+    --threads=8 > ${STAT_FILE}.overwrite_lots_column_families
 
 # send data to ods
 function send_to_ods {
@@ -392,3 +424,5 @@ send_benchmark_to_ods readrandom memtablereadrandom $STAT_FILE.memtablefillreadr
 send_benchmark_to_ods readwhilewriting readwhilewriting $STAT_FILE.readwhilewriting
 send_benchmark_to_ods readwhilewriting readwhilewriting_in_ram ${STAT_FILE}.readwhilewriting_in_ram
 send_benchmark_to_ods seekrandomwhilewriting seekwhilewriting_in_ram ${STAT_FILE}.seekwhilewriting_in_ram
+send_benchmark_to_ods fillseq fillseq_lots_column_families ${STAT_FILE}.fillseq_lots_column_families
+send_benchmark_to_ods overwrite overwrite_lots_column_families ${STAT_FILE}.overwrite_lots_column_families

build_tools/unity

@@ -0,0 +1,78 @@
+#!/bin/sh
+#
+# Create the unity file
+#
+
+OUTPUT=$1
+if test -z "$OUTPUT"; then
+  echo "usage: $0 <output-filename>" >&2
+  exit 1
+fi
+
+# Delete existing file, if it exists
+rm -f "$OUTPUT"
+touch "$OUTPUT"
+
+# Detect OS
+if test -z "$TARGET_OS"; then
+    TARGET_OS=`uname -s`
+fi
+
+# generic port files (working on all platform by #ifdef) go directly in /port
+GENERIC_PORT_FILES=`cd "$ROCKSDB_ROOT"; find port -name '*.cc' | tr "\n" " "`
+
+# On GCC, we pick libc's memcmp over GCC's memcmp via -fno-builtin-memcmp
+case "$TARGET_OS" in
+    Darwin)
+        # PORT_FILES=port/darwin/darwin_specific.cc
+        ;;
+    IOS)
+        ;;
+    Linux)
+        # PORT_FILES=port/linux/linux_specific.cc
+        ;;
+    SunOS)
+        # PORT_FILES=port/sunos/sunos_specific.cc
+        ;;
+    FreeBSD)
+        # PORT_FILES=port/freebsd/freebsd_specific.cc
+        ;;
+    NetBSD)
+        # PORT_FILES=port/netbsd/netbsd_specific.cc
+        ;;
+    OpenBSD)
+        # PORT_FILES=port/openbsd/openbsd_specific.cc
+        ;;
+    DragonFly)
+        # PORT_FILES=port/dragonfly/dragonfly_specific.cc
+        ;;
+    OS_ANDROID_CROSSCOMPILE)
+        # PORT_FILES=port/android/android.cc
+        ;;
+    *)
+        echo "Unknown platform!" >&2
+        exit 1
+esac
+
+# We want to make a list of all cc files within util, db, table, and helpers
+# except for the test and benchmark files. By default, find will output a list
+# of all files matching either rule, so we need to append -print to make the
+# prune take effect.
+DIRS="util db table utilities"
+
+set -f # temporarily disable globbing so that our patterns arent expanded
+PRUNE_TEST="-name *test*.cc -prune"
+PRUNE_BENCH="-name *bench*.cc -prune"
+PORTABLE_FILES=`cd "$ROCKSDB_ROOT"; find $DIRS $PRUNE_TEST -o $PRUNE_BENCH -o -name '*.cc' -print | sort`
+PORTABLE_CPP=`cd "$ROCKSDB_ROOT"; find $DIRS $PRUNE_TEST -o $PRUNE_BENCH -o -name '*.cpp' -print | sort`
+set +f # re-enable globbing
+
+# The sources consist of the portable files, plus the platform-specific port
+# file.
+for SOURCE_FILE in $PORTABLE_FILES $GENERIC_PORT_FILES $PORT_FILES $PORTABLE_CPP
+do
+	echo "#include <$SOURCE_FILE>" >> "$OUTPUT"
+done
+
+echo "int main(int argc, char** argv){ return 0; }" >> "$OUTPUT"
+

build_tools/version.sh

@@ -0,0 +1,14 @@
+#!/bin/sh
+if [ "$#" = "0" ]; then
+  echo "Usage: $0 major|minor|patch"
+  exit 1
+fi
+if [ "$1" = "major" ]; then
+  cat include/rocksdb/version.h  | grep MAJOR | head -n1 | awk '{print $3}'
+fi
+if [ "$1" = "minor" ]; then
+  cat include/rocksdb/version.h  | grep MINOR | head -n1 | awk '{print $3}'
+fi
+if [ "$1" = "patch" ]; then
+  cat include/rocksdb/version.h  | grep PATCH | head -n1 | awk '{print $3}'
+fi

db/builder.cc

@@ -26,21 +26,24 @@ namespace rocksdb {
 
 class TableFactory;
 
-TableBuilder* NewTableBuilder(const Options& options,
+TableBuilder* NewTableBuilder(const ImmutableCFOptions& ioptions,
                               const InternalKeyComparator& internal_comparator,
                               WritableFile* file,
-                              CompressionType compression_type) {
-  return options.table_factory->NewTableBuilder(options, internal_comparator,
-                                                file, compression_type);
+                              const CompressionType compression_type,
+                              const CompressionOptions& compression_opts) {
+  return ioptions.table_factory->NewTableBuilder(
+      ioptions, internal_comparator, file, compression_type, compression_opts);
 }
 
-Status BuildTable(const std::string& dbname, Env* env, const Options& options,
-                  const EnvOptions& soptions, TableCache* table_cache,
+Status BuildTable(const std::string& dbname, Env* env,
+                  const ImmutableCFOptions& ioptions,
+                  const EnvOptions& env_options, TableCache* table_cache,
                   Iterator* iter, FileMetaData* meta,
                   const InternalKeyComparator& internal_comparator,
                   const SequenceNumber newest_snapshot,
                   const SequenceNumber earliest_seqno_in_memtable,
                   const CompressionType compression,
+                  const CompressionOptions& compression_opts,
                   const Env::IOPriority io_priority) {
   Status s;
   meta->fd.file_size = 0;
@@ -50,23 +53,24 @@ Status BuildTable(const std::string& dbname, Env* env, const Options& options,
   // If the sequence number of the smallest entry in the memtable is
   // smaller than the most recent snapshot, then we do not trigger
   // removal of duplicate/deleted keys as part of this builder.
-  bool purge = options.purge_redundant_kvs_while_flush;
+  bool purge = ioptions.purge_redundant_kvs_while_flush;
   if (earliest_seqno_in_memtable <= newest_snapshot) {
     purge = false;
   }
 
-  std::string fname = TableFileName(options.db_paths, meta->fd.GetNumber(),
+  std::string fname = TableFileName(ioptions.db_paths, meta->fd.GetNumber(),
                                     meta->fd.GetPathId());
   if (iter->Valid()) {
     unique_ptr<WritableFile> file;
-    s = env->NewWritableFile(fname, &file, soptions);
+    s = env->NewWritableFile(fname, &file, env_options);
     if (!s.ok()) {
       return s;
     }
     file->SetIOPriority(io_priority);
 
-    TableBuilder* builder =
-        NewTableBuilder(options, internal_comparator, file.get(), compression);
+    TableBuilder* builder = NewTableBuilder(
+        ioptions, internal_comparator, file.get(),
+        compression, compression_opts);
 
     // the first key is the smallest key
     Slice key = iter->key();
@@ -75,8 +79,8 @@ Status BuildTable(const std::string& dbname, Env* env, const Options& options,
     meta->largest_seqno = meta->smallest_seqno;
 
     MergeHelper merge(internal_comparator.user_comparator(),
-                      options.merge_operator.get(), options.info_log.get(),
-                      options.min_partial_merge_operands,
+                      ioptions.merge_operator, ioptions.info_log,
+                      ioptions.min_partial_merge_operands,
                       true /* internal key corruption is not ok */);
 
     if (purge) {
@@ -113,6 +117,13 @@ Status BuildTable(const std::string& dbname, Env* env, const Options& options,
           is_first_key = false;
 
           if (this_ikey.type == kTypeMerge) {
+            // TODO(tbd): Add a check here to prevent RocksDB from crash when
+            // reopening a DB w/o properly specifying the merge operator.  But
+            // currently we observed a memory leak on failing in RocksDB
+            // recovery, so we decide to let it crash instead of causing
+            // memory leak for now before we have identified the real cause
+            // of the memory leak.
+
             // Handle merge-type keys using the MergeHelper
             // TODO: pass statistics to MergeUntil
             merge.MergeUntil(iter, 0 /* don't worry about snapshot */);
@@ -189,12 +200,12 @@ Status BuildTable(const std::string& dbname, Env* env, const Options& options,
     delete builder;
 
     // Finish and check for file errors
-    if (s.ok() && !options.disableDataSync) {
-      if (options.use_fsync) {
-        StopWatch sw(env, options.statistics.get(), TABLE_SYNC_MICROS);
+    if (s.ok() && !ioptions.disable_data_sync) {
+      if (ioptions.use_fsync) {
+        StopWatch sw(env, ioptions.statistics, TABLE_SYNC_MICROS);
         s = file->Fsync();
       } else {
-        StopWatch sw(env, options.statistics.get(), TABLE_SYNC_MICROS);
+        StopWatch sw(env, ioptions.statistics, TABLE_SYNC_MICROS);
         s = file->Sync();
       }
     }
@@ -204,7 +215,7 @@ Status BuildTable(const std::string& dbname, Env* env, const Options& options,
 
     if (s.ok()) {
       // Verify that the table is usable
-      Iterator* it = table_cache->NewIterator(ReadOptions(), soptions,
+      Iterator* it = table_cache->NewIterator(ReadOptions(), env_options,
                                               internal_comparator, meta->fd);
       s = it->status();
       delete it;

db/builder.h

@@ -11,6 +11,7 @@
 #include "rocksdb/status.h"
 #include "rocksdb/types.h"
 #include "rocksdb/options.h"
+#include "rocksdb/immutable_options.h"
 
 namespace rocksdb {
 
@@ -26,8 +27,10 @@ class TableBuilder;
 class WritableFile;
 
 extern TableBuilder* NewTableBuilder(
-    const Options& options, const InternalKeyComparator& internal_comparator,
-    WritableFile* file, CompressionType compression_type);
+    const ImmutableCFOptions& options,
+    const InternalKeyComparator& internal_comparator,
+    WritableFile* file, const CompressionType compression_type,
+    const CompressionOptions& compression_opts);
 
 // Build a Table file from the contents of *iter.  The generated file
 // will be named according to number specified in meta. On success, the rest of
@@ -35,13 +38,15 @@ extern TableBuilder* NewTableBuilder(
 // If no data is present in *iter, meta->file_size will be set to
 // zero, and no Table file will be produced.
 extern Status BuildTable(const std::string& dbname, Env* env,
-                         const Options& options, const EnvOptions& soptions,
+                         const ImmutableCFOptions& options,
+                         const EnvOptions& env_options,
                          TableCache* table_cache, Iterator* iter,
                          FileMetaData* meta,
                          const InternalKeyComparator& internal_comparator,
                          const SequenceNumber newest_snapshot,
                          const SequenceNumber earliest_seqno_in_memtable,
                          const CompressionType compression,
+                         const CompressionOptions& compression_opts,
                          const Env::IOPriority io_priority = Env::IO_HIGH);
 
 }  // namespace rocksdb

db/c.cc

@@ -36,6 +36,9 @@ using rocksdb::ColumnFamilyHandle;
 using rocksdb::ColumnFamilyOptions;
 using rocksdb::CompactionFilter;
 using rocksdb::CompactionFilterFactory;
+using rocksdb::CompactionFilterV2;
+using rocksdb::CompactionFilterFactoryV2;
+using rocksdb::CompactionFilterContext;
 using rocksdb::CompactionOptionsFIFO;
 using rocksdb::Comparator;
 using rocksdb::CompressionType;
@@ -52,6 +55,7 @@ using rocksdb::MergeOperator;
 using rocksdb::NewBloomFilterPolicy;
 using rocksdb::NewLRUCache;
 using rocksdb::Options;
+using rocksdb::BlockBasedTableOptions;
 using rocksdb::RandomAccessFile;
 using rocksdb::Range;
 using rocksdb::ReadOptions;
@@ -78,6 +82,7 @@ struct rocksdb_fifo_compaction_options_t { CompactionOptionsFIFO rep; };
 struct rocksdb_readoptions_t     { ReadOptions       rep; };
 struct rocksdb_writeoptions_t    { WriteOptions      rep; };
 struct rocksdb_options_t         { Options           rep; };
+struct rocksdb_block_based_table_options_t  { BlockBasedTableOptions rep; };
 struct rocksdb_seqfile_t         { SequentialFile*   rep; };
 struct rocksdb_randomfile_t      { RandomAccessFile* rep; };
 struct rocksdb_writablefile_t    { WritableFile*     rep; };
@@ -113,7 +118,7 @@ struct rocksdb_compactionfilter_t : public CompactionFilter {
       const Slice& existing_value,
       std::string* new_value,
       bool* value_changed) const {
-    char* c_new_value = NULL;
+    char* c_new_value = nullptr;
     size_t new_value_length = 0;
     unsigned char c_value_changed = 0;
     unsigned char result = (*filter_)(
@@ -154,6 +159,104 @@ struct rocksdb_compactionfilterfactory_t : public CompactionFilterFactory {
   virtual const char* Name() const { return (*name_)(state_); }
 };
 
+struct rocksdb_compactionfilterv2_t : public CompactionFilterV2 {
+  void* state_;
+  void (*destructor_)(void*);
+  const char* (*name_)(void*);
+  void (*filter_)(void*, int level, size_t num_keys,
+                  const char* const* keys_list, const size_t* keys_list_sizes,
+                  const char* const* existing_values_list, const size_t* existing_values_list_sizes,
+                  char** new_values_list, size_t* new_values_list_sizes,
+                  unsigned char* to_delete_list);
+
+  virtual ~rocksdb_compactionfilterv2_t() {
+    (*destructor_)(state_);
+  }
+
+  virtual const char* Name() const {
+    return (*name_)(state_);
+  }
+
+  virtual std::vector<bool> Filter(int level,
+                                   const SliceVector& keys,
+                                   const SliceVector& existing_values,
+                                   std::vector<std::string>* new_values,
+                                   std::vector<bool>* values_changed) const {
+    // Make a vector pointing to the underlying key data.
+    size_t num_keys = keys.size();
+    std::vector<const char*> keys_list(num_keys);
+    std::vector<size_t> keys_list_sizes(num_keys);
+    for (size_t i = 0; i < num_keys; ++i) {
+      keys_list[i] = keys[i].data();
+      keys_list_sizes[i] = keys[i].size();
+    }
+    // Make a vector pointing to the underlying value data.
+    std::vector<const char*> existing_values_list(num_keys);
+    std::vector<size_t> existing_values_list_sizes(num_keys);
+    for (size_t i = 0; i < num_keys; ++i) {
+      existing_values_list[i] = existing_values[i].data();
+      existing_values_list_sizes[i] = existing_values[i].size();
+    }
+    // Make a vector which will accept newly-allocated char* arrays
+    // which we will take ownership of and assign to strings in new_values.
+    new_values->clear();
+    std::vector<char*> new_values_list(num_keys);
+    std::vector<size_t> new_values_list_sizes(num_keys);
+    // Resize values_changed to hold all keys.
+    values_changed->resize(num_keys);
+    // Make a vector for bools indicating a value should be deleted
+    // on compaction (true) or maintained (false).
+    std::vector<unsigned char> to_delete_list(num_keys);
+
+    (*filter_)(
+        state_, level, num_keys, &keys_list[0], &keys_list_sizes[0],
+        &existing_values_list[0], &existing_values_list_sizes[0],
+        &new_values_list[0], &new_values_list_sizes[0], &to_delete_list[0]);
+
+    // Now, we transfer any changed values, setting values_changed and
+    // initializing new_values in the event a value changed.
+    std::vector<bool> to_delete(num_keys);
+    for (size_t i = 0; i < num_keys; ++i) {
+      to_delete[i] = to_delete_list[i];
+      (*values_changed)[i] = new_values_list[i] != nullptr;
+      if ((*values_changed)[i]) {
+        new_values->push_back(std::string(new_values_list[i], new_values_list_sizes[i]));
+        free(new_values_list[i]);
+      }
+    }
+    return to_delete;
+  }
+};
+
+struct rocksdb_compactionfilterfactoryv2_t : public CompactionFilterFactoryV2 {
+  void* state_;
+  void (*destructor_)(void*);
+  const char* (*name_)(void*);
+  rocksdb_compactionfilterv2_t* (*create_compaction_filter_v2_)(
+      void* state, const rocksdb_compactionfiltercontext_t* context);
+
+  rocksdb_compactionfilterfactoryv2_t(const SliceTransform* prefix_extractor)
+      : CompactionFilterFactoryV2(prefix_extractor) {
+  }
+
+  virtual ~rocksdb_compactionfilterfactoryv2_t() {
+    (*destructor_)(state_);
+  }
+
+  virtual const char* Name() const {
+    return (*name_)(state_);
+  }
+
+  virtual std::unique_ptr<CompactionFilterV2> CreateCompactionFilterV2(
+      const CompactionFilterContext& context) {
+    struct rocksdb_compactionfiltercontext_t c_context;
+    c_context.rep.is_full_compaction = context.is_full_compaction;
+    c_context.rep.is_manual_compaction = context.is_manual_compaction;
+    return std::unique_ptr<CompactionFilterV2>(
+        (*create_compaction_filter_v2_)(state_, &c_context));
+  }
+};
+
 struct rocksdb_comparator_t : public Comparator {
   void* state_;
   void (*destructor_)(void*);
@@ -952,6 +1055,74 @@ const char* rocksdb_writebatch_data(rocksdb_writebatch_t* b, size_t* size) {
   return b->rep.Data().c_str();
 }
 
+rocksdb_block_based_table_options_t*
+rocksdb_block_based_options_create() {
+  return new rocksdb_block_based_table_options_t;
+}
+
+void rocksdb_block_based_options_destroy(
+    rocksdb_block_based_table_options_t* options) {
+  delete options;
+}
+
+void rocksdb_block_based_options_set_block_size(
+    rocksdb_block_based_table_options_t* options, size_t block_size) {
+  options->rep.block_size = block_size;
+}
+
+void rocksdb_block_based_options_set_block_size_deviation(
+    rocksdb_block_based_table_options_t* options, int block_size_deviation) {
+  options->rep.block_size_deviation = block_size_deviation;
+}
+
+void rocksdb_block_based_options_set_block_restart_interval(
+    rocksdb_block_based_table_options_t* options, int block_restart_interval) {
+  options->rep.block_restart_interval = block_restart_interval;
+}
+
+void rocksdb_block_based_options_set_filter_policy(
+    rocksdb_block_based_table_options_t* options,
+    rocksdb_filterpolicy_t* filter_policy) {
+  options->rep.filter_policy.reset(filter_policy);
+}
+
+void rocksdb_block_based_options_set_no_block_cache(
+    rocksdb_block_based_table_options_t* options,
+    unsigned char no_block_cache) {
+  options->rep.no_block_cache = no_block_cache;
+}
+
+void rocksdb_block_based_options_set_block_cache(
+    rocksdb_block_based_table_options_t* options,
+    rocksdb_cache_t* block_cache) {
+  if (block_cache) {
+    options->rep.block_cache = block_cache->rep;
+  }
+}
+
+void rocksdb_block_based_options_set_block_cache_compressed(
+    rocksdb_block_based_table_options_t* options,
+    rocksdb_cache_t* block_cache_compressed) {
+  if (block_cache_compressed) {
+    options->rep.block_cache_compressed = block_cache_compressed->rep;
+  }
+}
+
+void rocksdb_block_based_options_set_whole_key_filtering(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.whole_key_filtering = v;
+}
+
+void rocksdb_options_set_block_based_table_factory(
+    rocksdb_options_t *opt,
+    rocksdb_block_based_table_options_t* table_options) {
+  if (table_options) {
+    opt->rep.table_factory.reset(
+        rocksdb::NewBlockBasedTableFactory(table_options->rep));
+  }
+}
+
+
 rocksdb_options_t* rocksdb_options_create() {
   return new rocksdb_options_t;
 }
@@ -966,8 +1137,8 @@ void rocksdb_options_increase_parallelism(
 }
 
 void rocksdb_options_optimize_for_point_lookup(
-    rocksdb_options_t* opt) {
-  opt->rep.OptimizeForPointLookup();
+    rocksdb_options_t* opt, uint64_t block_cache_size_mb) {
+  opt->rep.OptimizeForPointLookup(block_cache_size_mb);
 }
 
 void rocksdb_options_optimize_level_style_compaction(
@@ -1004,10 +1175,10 @@ void rocksdb_options_set_merge_operator(
   opt->rep.merge_operator = std::shared_ptr<MergeOperator>(merge_operator);
 }
 
-void rocksdb_options_set_filter_policy(
+void rocksdb_options_set_compaction_filter_factory_v2(
     rocksdb_options_t* opt,
-    rocksdb_filterpolicy_t* policy) {
-  opt->rep.filter_policy = policy;
+    rocksdb_compactionfilterfactoryv2_t* compaction_filter_factory_v2) {
+  opt->rep.compaction_filter_factory_v2 = std::shared_ptr<CompactionFilterFactoryV2>(compaction_filter_factory_v2);
 }
 
 void rocksdb_options_set_create_if_missing(
@@ -1053,26 +1224,6 @@ void rocksdb_options_set_max_open_files(rocksdb_options_t* opt, int n) {
   opt->rep.max_open_files = n;
 }
 
-void rocksdb_options_set_cache(rocksdb_options_t* opt, rocksdb_cache_t* c) {
-  if (c) {
-    opt->rep.block_cache = c->rep;
-  }
-}
-
-void rocksdb_options_set_cache_compressed(rocksdb_options_t* opt, rocksdb_cache_t* c) {
-  if (c) {
-    opt->rep.block_cache_compressed = c->rep;
-  }
-}
-
-void rocksdb_options_set_block_size(rocksdb_options_t* opt, size_t s) {
-  opt->rep.block_size = s;
-}
-
-void rocksdb_options_set_block_restart_interval(rocksdb_options_t* opt, int n) {
-  opt->rep.block_restart_interval = n;
-}
-
 void rocksdb_options_set_target_file_size_base(
     rocksdb_options_t* opt, uint64_t n) {
   opt->rep.target_file_size_base = n;
@@ -1165,11 +1316,6 @@ void rocksdb_options_set_prefix_extractor(
   opt->rep.prefix_extractor.reset(prefix_extractor);
 }
 
-void rocksdb_options_set_whole_key_filtering(
-    rocksdb_options_t* opt, unsigned char v) {
-  opt->rep.whole_key_filtering = v;
-}
-
 void rocksdb_options_set_disable_data_sync(
     rocksdb_options_t* opt, int disable_data_sync) {
   opt->rep.disableDataSync = disable_data_sync;
@@ -1180,11 +1326,6 @@ void rocksdb_options_set_use_fsync(
   opt->rep.use_fsync = use_fsync;
 }
 
-void rocksdb_options_set_db_stats_log_interval(
-    rocksdb_options_t* opt, int db_stats_log_interval) {
-  opt->rep.db_stats_log_interval = db_stats_log_interval;
-}
-
 void rocksdb_options_set_db_log_dir(
     rocksdb_options_t* opt, const char* db_log_dir) {
   opt->rep.db_log_dir = db_log_dir;
@@ -1214,8 +1355,8 @@ void rocksdb_options_set_purge_redundant_kvs_while_flush(
   opt->rep.purge_redundant_kvs_while_flush = v;
 }
 
-void rocksdb_options_set_allow_os_buffer(
-    rocksdb_options_t* opt, unsigned char v) {
+void rocksdb_options_set_allow_os_buffer(rocksdb_options_t* opt,
+                                         unsigned char v) {
   opt->rep.allow_os_buffer = v;
 }
 
@@ -1244,11 +1385,6 @@ void rocksdb_options_set_stats_dump_period_sec(
   opt->rep.stats_dump_period_sec = v;
 }
 
-void rocksdb_options_set_block_size_deviation(
-    rocksdb_options_t* opt, int v) {
-  opt->rep.block_size_deviation = v;
-}
-
 void rocksdb_options_set_advise_random_on_open(
     rocksdb_options_t* opt, unsigned char v) {
   opt->rep.advise_random_on_open = v;
@@ -1343,11 +1479,6 @@ void rocksdb_options_set_max_manifest_file_size(
   opt->rep.max_manifest_file_size = v;
 }
 
-void rocksdb_options_set_no_block_cache(
-    rocksdb_options_t* opt, unsigned char v) {
-  opt->rep.no_block_cache = v;
-}
-
 void rocksdb_options_set_table_cache_numshardbits(
     rocksdb_options_t* opt, int v) {
   opt->rep.table_cache_numshardbits = v;
@@ -1367,10 +1498,6 @@ void rocksdb_options_set_disable_auto_compactions(rocksdb_options_t* opt, int di
   opt->rep.disable_auto_compactions = disable;
 }
 
-void rocksdb_options_set_disable_seek_compaction(rocksdb_options_t* opt, int disable) {
-  opt->rep.disable_seek_compaction = disable;
-}
-
 void rocksdb_options_set_delete_obsolete_files_period_micros(
     rocksdb_options_t* opt, uint64_t v) {
   opt->rep.delete_obsolete_files_period_micros = v;
@@ -1428,9 +1555,13 @@ void rocksdb_options_set_plain_table_factory(
     double hash_table_ratio, size_t index_sparseness) {
   static rocksdb::TableFactory* factory = 0;
   if (!factory) {
-    factory = rocksdb::NewPlainTableFactory(
-        user_key_len, bloom_bits_per_key,
-        hash_table_ratio, index_sparseness);
+    rocksdb::PlainTableOptions options;
+    options.user_key_len = user_key_len;
+    options.bloom_bits_per_key = bloom_bits_per_key;
+    options.hash_table_ratio = hash_table_ratio;
+    options.index_sparseness = index_sparseness;
+
+    factory = rocksdb::NewPlainTableFactory(options);
   }
   opt->rep.table_factory.reset(factory);
 }
@@ -1450,11 +1581,6 @@ void rocksdb_options_set_bloom_locality(
   opt->rep.bloom_locality = v;
 }
 
-void rocksdb_options_set_allow_thread_local(
-    rocksdb_options_t* opt, unsigned char v) {
-  opt->rep.allow_thread_local = v;
-}
-
 void rocksdb_options_set_inplace_update_support(
     rocksdb_options_t* opt, unsigned char v) {
   opt->rep.inplace_update_support = v;
@@ -1546,6 +1672,46 @@ void rocksdb_compactionfilterfactory_destroy(
   delete factory;
 }
 
+rocksdb_compactionfilterv2_t* rocksdb_compactionfilterv2_create(
+    void* state,
+    void (*destructor)(void*),
+    void (*filter)(void*, int level, size_t num_keys,
+                   const char* const* keys_list, const size_t* keys_list_sizes,
+                   const char* const* existing_values_list, const size_t* existing_values_list_sizes,
+                   char** new_values_list, size_t* new_values_list_sizes,
+                   unsigned char* to_delete_list),
+    const char* (*name)(void*)) {
+  rocksdb_compactionfilterv2_t* result = new rocksdb_compactionfilterv2_t;
+  result->state_ = state;
+  result->destructor_ = destructor;
+  result->filter_ = filter;
+  result->name_ = name;
+  return result;
+}
+
+void rocksdb_compactionfilterv2_destroy(rocksdb_compactionfilterv2_t* filter) {
+  delete filter;
+}
+
+rocksdb_compactionfilterfactoryv2_t* rocksdb_compactionfilterfactoryv2_create(
+    void* state,
+    rocksdb_slicetransform_t* prefix_extractor,
+    void (*destructor)(void*),
+    rocksdb_compactionfilterv2_t* (*create_compaction_filter_v2)(
+        void* state, const rocksdb_compactionfiltercontext_t* context),
+    const char* (*name)(void*)) {
+  rocksdb_compactionfilterfactoryv2_t* result = new rocksdb_compactionfilterfactoryv2_t(prefix_extractor);
+  result->state_ = state;
+  result->destructor_ = destructor;
+  result->create_compaction_filter_v2_ = create_compaction_filter_v2;
+  result->name_ = name;
+  return result;
+}
+
+void rocksdb_compactionfilterfactoryv2_destroy(rocksdb_compactionfilterfactoryv2_t* factory) {
+  delete factory;
+}
+
 rocksdb_comparator_t* rocksdb_comparator_create(
     void* state,
     void (*destructor)(void*),
@@ -1673,6 +1839,13 @@ void rocksdb_readoptions_set_snapshot(
   opt->rep.snapshot = (snap ? snap->rep : nullptr);
 }
 
+void rocksdb_readoptions_set_iterate_upper_bound(
+    rocksdb_readoptions_t* opt,
+    const char* key, size_t keylen) {
+  Slice prefix = Slice(key, keylen);
+  opt->rep.iterate_upper_bound = &prefix;
+}
+
 void rocksdb_readoptions_set_read_tier(
     rocksdb_readoptions_t* opt, int v) {
   opt->rep.read_tier = static_cast<rocksdb::ReadTier>(v);

db/c_test.c

@@ -225,6 +225,79 @@ static rocksdb_t* CheckCompaction(rocksdb_t* db, rocksdb_options_t* options,
   return db;
 }
 
+// Custom compaction filter V2.
+static void CompactionFilterV2Destroy(void* arg) { }
+static const char* CompactionFilterV2Name(void* arg) {
+  return "TestCompactionFilterV2";
+}
+static void CompactionFilterV2Filter(
+    void* arg, int level, size_t num_keys,
+    const char* const* keys_list, const size_t* keys_list_sizes,
+    const char* const* existing_values_list, const size_t* existing_values_list_sizes,
+    char** new_values_list, size_t* new_values_list_sizes,
+    unsigned char* to_delete_list) {
+  size_t i;
+  for (i = 0; i < num_keys; i++) {
+    // If any value is "gc", it's removed.
+    if (existing_values_list_sizes[i] == 2 && memcmp(existing_values_list[i], "gc", 2) == 0) {
+      to_delete_list[i] = 1;
+    } else if (existing_values_list_sizes[i] == 6 && memcmp(existing_values_list[i], "gc all", 6) == 0) {
+      // If any value is "gc all", all keys are removed.
+      size_t j;
+      for (j = 0; j < num_keys; j++) {
+        to_delete_list[j] = 1;
+      }
+      return;
+    } else if (existing_values_list_sizes[i] == 6 && memcmp(existing_values_list[i], "change", 6) == 0) {
+      // If value is "change", set changed value to "changed".
+      size_t len;
+      len = strlen("changed");
+      new_values_list[i] = malloc(len);
+      memcpy(new_values_list[i], "changed", len);
+      new_values_list_sizes[i] = len;
+    } else {
+      // Otherwise, no keys are removed.
+    }
+  }
+}
+
+// Custom prefix extractor for compaction filter V2 which extracts first 3 characters.
+static void CFV2PrefixExtractorDestroy(void* arg) { }
+static char* CFV2PrefixExtractorTransform(void* arg, const char* key, size_t length, size_t* dst_length) {
+  // Verify keys are maximum length 4; this verifies fix for a
+  // prior bug which was passing the RocksDB-encoded key with
+  // logical timestamp suffix instead of parsed user key.
+  if (length > 4) {
+    fprintf(stderr, "%s:%d: %s: key %s is not user key\n", __FILE__, __LINE__, phase, key);
+    abort();
+  }
+  *dst_length = length < 3 ? length : 3;
+  return (char*)key;
+}
+static unsigned char CFV2PrefixExtractorInDomain(void* state, const char* key, size_t length) {
+  return 1;
+}
+static unsigned char CFV2PrefixExtractorInRange(void* state, const char* key, size_t length) {
+  return 1;
+}
+static const char* CFV2PrefixExtractorName(void* state) {
+  return "TestCFV2PrefixExtractor";
+}
+
+// Custom compaction filter factory V2.
+static void CompactionFilterFactoryV2Destroy(void* arg) {
+  rocksdb_slicetransform_destroy((rocksdb_slicetransform_t*)arg);
+}
+static const char* CompactionFilterFactoryV2Name(void* arg) {
+  return "TestCompactionFilterFactoryV2";
+}
+static rocksdb_compactionfilterv2_t* CompactionFilterFactoryV2Create(
+    void* state, const rocksdb_compactionfiltercontext_t* context) {
+  return rocksdb_compactionfilterv2_create(state, CompactionFilterV2Destroy,
+                                           CompactionFilterV2Filter,
+                                           CompactionFilterV2Name);
+}
+
 // Custom merge operator
 static void MergeOperatorDestroy(void* arg) { }
 static const char* MergeOperatorName(void* arg) {
@@ -262,6 +335,7 @@ int main(int argc, char** argv) {
   rocksdb_cache_t* cache;
   rocksdb_env_t* env;
   rocksdb_options_t* options;
+  rocksdb_block_based_table_options_t* table_options;
   rocksdb_readoptions_t* roptions;
   rocksdb_writeoptions_t* woptions;
   char* err = NULL;
@@ -280,14 +354,15 @@ int main(int argc, char** argv) {
   options = rocksdb_options_create();
   rocksdb_options_set_comparator(options, cmp);
   rocksdb_options_set_error_if_exists(options, 1);
-  rocksdb_options_set_cache(options, cache);
   rocksdb_options_set_env(options, env);
   rocksdb_options_set_info_log(options, NULL);
   rocksdb_options_set_write_buffer_size(options, 100000);
   rocksdb_options_set_paranoid_checks(options, 1);
   rocksdb_options_set_max_open_files(options, 10);
-  rocksdb_options_set_block_size(options, 1024);
-  rocksdb_options_set_block_restart_interval(options, 8);
+  table_options = rocksdb_block_based_options_create();
+  rocksdb_block_based_options_set_block_cache(table_options, cache);
+  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);
   int compression_levels[] = {rocksdb_no_compression, rocksdb_no_compression,
@@ -467,10 +542,12 @@ int main(int argc, char** argv) {
       policy = rocksdb_filterpolicy_create_bloom(10);
     }
 
+    rocksdb_block_based_options_set_filter_policy(table_options, policy);
+
     // Create new database
     rocksdb_close(db);
     rocksdb_destroy_db(options, dbname, &err);
-    rocksdb_options_set_filter_policy(options, policy);
+    rocksdb_options_set_block_based_table_factory(options, table_options);
     db = rocksdb_open(options, dbname, &err);
     CheckNoError(err);
     rocksdb_put(db, woptions, "foo", 3, "foovalue", 8, &err);
@@ -492,8 +569,9 @@ int main(int argc, char** argv) {
       CheckGet(db, roptions, "foo", "foovalue");
       CheckGet(db, roptions, "bar", "barvalue");
     }
-    rocksdb_options_set_filter_policy(options, NULL);
-    rocksdb_filterpolicy_destroy(policy);
+    // Reset the policy
+    rocksdb_block_based_options_set_filter_policy(table_options, NULL);
+    rocksdb_options_set_block_based_table_factory(options, table_options);
   }
 
   StartPhase("compaction_filter");
@@ -531,6 +609,50 @@ int main(int argc, char** argv) {
     rocksdb_options_destroy(options);
   }
 
+  StartPhase("compaction_filter_v2");
+  {
+    rocksdb_compactionfilterfactoryv2_t* factory;
+    rocksdb_slicetransform_t* prefix_extractor;
+    prefix_extractor = rocksdb_slicetransform_create(
+        NULL, CFV2PrefixExtractorDestroy, CFV2PrefixExtractorTransform,
+        CFV2PrefixExtractorInDomain, CFV2PrefixExtractorInRange,
+        CFV2PrefixExtractorName);
+    factory = rocksdb_compactionfilterfactoryv2_create(
+        prefix_extractor, prefix_extractor, CompactionFilterFactoryV2Destroy,
+        CompactionFilterFactoryV2Create, CompactionFilterFactoryV2Name);
+    // Create new database
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+    rocksdb_options_set_compaction_filter_factory_v2(options, factory);
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+    // Only foo2 is GC'd, foo3 is changed.
+    rocksdb_put(db, woptions, "foo1", 4, "no gc", 5, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo2", 4, "gc", 2, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo3", 4, "change", 6, &err);
+    CheckNoError(err);
+    // All bars are GC'd.
+    rocksdb_put(db, woptions, "bar1", 4, "no gc", 5, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "bar2", 4, "gc all", 6, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "bar3", 4, "no gc", 5, &err);
+    CheckNoError(err);
+    // Compact the DB to garbage collect.
+    rocksdb_compact_range(db, NULL, 0, NULL, 0);
+
+    // Verify foo entries.
+    CheckGet(db, roptions, "foo1", "no gc");
+    CheckGet(db, roptions, "foo2", NULL);
+    CheckGet(db, roptions, "foo3", "changed");
+    // Verify bar entries were all deleted.
+    CheckGet(db, roptions, "bar1", NULL);
+    CheckGet(db, roptions, "bar2", NULL);
+    CheckGet(db, roptions, "bar3", NULL);
+  }
+
   StartPhase("merge_operator");
   {
     rocksdb_mergeoperator_t* merge_operator;
@@ -640,8 +762,7 @@ int main(int argc, char** argv) {
   StartPhase("prefix");
   {
     // Create new database
-    rocksdb_filterpolicy_t* policy = rocksdb_filterpolicy_create_bloom(10);
-    rocksdb_options_set_filter_policy(options, policy);
+    rocksdb_options_set_allow_mmap_reads(options, 1);
     rocksdb_options_set_prefix_extractor(options, rocksdb_slicetransform_create_fixed_prefix(3));
     rocksdb_options_set_hash_skip_list_rep(options, 5000, 4, 4);
     rocksdb_options_set_plain_table_factory(options, 4, 10, 0.75, 16);
@@ -678,13 +799,13 @@ int main(int argc, char** argv) {
     rocksdb_iter_get_error(iter, &err);
     CheckNoError(err);
     rocksdb_iter_destroy(iter);
-    rocksdb_filterpolicy_destroy(policy);
   }
 
 
   StartPhase("cleanup");
   rocksdb_close(db);
   rocksdb_options_destroy(options);
+  rocksdb_block_based_options_destroy(table_options);
   rocksdb_readoptions_destroy(roptions);
   rocksdb_writeoptions_destroy(woptions);
   rocksdb_cache_destroy(cache);

db/column_family.cc

@@ -9,6 +9,11 @@
 
 #include "db/column_family.h"
 
+#ifndef __STDC_FORMAT_MACROS
+#define __STDC_FORMAT_MACROS
+#endif
+
+#include <inttypes.h>
 #include <vector>
 #include <string>
 #include <algorithm>
@@ -19,11 +24,43 @@
 #include "db/internal_stats.h"
 #include "db/compaction_picker.h"
 #include "db/table_properties_collector.h"
+#include "db/write_controller.h"
 #include "util/autovector.h"
 #include "util/hash_skiplist_rep.h"
+#include "util/options_helper.h"
 
 namespace rocksdb {
 
+namespace {
+// This function computes the amount of time in microseconds by which a write
+// should be delayed based on the number of level-0 files according to the
+// following formula:
+// if n < bottom, return 0;
+// if n >= top, return 1000;
+// otherwise, let r = (n - bottom) /
+//                    (top - bottom)
+//  and return r^2 * 1000.
+// The goal of this formula is to gradually increase the rate at which writes
+// are slowed. We also tried linear delay (r * 1000), but it seemed to do
+// slightly worse. There is no other particular reason for choosing quadratic.
+uint64_t SlowdownAmount(int n, double bottom, double top) {
+  uint64_t delay;
+  if (n >= top) {
+    delay = 1000;
+  } else if (n < bottom) {
+    delay = 0;
+  } else {
+    // If we are here, we know that:
+    //   level0_start_slowdown <= n < level0_slowdown
+    // since the previous two conditions are false.
+    double how_much = static_cast<double>(n - bottom) / (top - bottom);
+    delay = std::max(how_much * how_much * 1000, 100.0);
+  }
+  assert(delay <= 1000);
+  return delay;
+}
+}  // namespace
+
 ColumnFamilyHandleImpl::ColumnFamilyHandleImpl(ColumnFamilyData* cfd,
                                                DBImpl* db, port::Mutex* mutex)
     : cfd_(cfd), db_(db), mutex_(mutex) {
@@ -49,21 +86,14 @@ ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
 
 uint32_t ColumnFamilyHandleImpl::GetID() const { return cfd()->GetID(); }
 
-namespace {
-// Fix user-supplied options to be reasonable
-template <class T, class V>
-static void ClipToRange(T* ptr, V minvalue, V maxvalue) {
-  if (static_cast<V>(*ptr) > maxvalue) *ptr = maxvalue;
-  if (static_cast<V>(*ptr) < minvalue) *ptr = minvalue;
+const Comparator* ColumnFamilyHandleImpl::user_comparator() const {
+  return cfd()->user_comparator();
 }
-}  // anonymous namespace
 
 ColumnFamilyOptions SanitizeOptions(const InternalKeyComparator* icmp,
-                                    const InternalFilterPolicy* ipolicy,
                                     const ColumnFamilyOptions& src) {
   ColumnFamilyOptions result = src;
   result.comparator = icmp;
-  result.filter_policy = (src.filter_policy != nullptr) ? ipolicy : nullptr;
 #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);
@@ -79,13 +109,7 @@ ColumnFamilyOptions SanitizeOptions(const InternalKeyComparator* icmp,
   result.min_write_buffer_number_to_merge =
       std::min(result.min_write_buffer_number_to_merge,
                result.max_write_buffer_number - 1);
-  if (result.block_cache == nullptr && !result.no_block_cache) {
-    result.block_cache = NewLRUCache(8 << 20);
-  }
   result.compression_per_level = src.compression_per_level;
-  if (result.block_size_deviation < 0 || result.block_size_deviation > 100) {
-    result.block_size_deviation = 0;
-  }
   if (result.max_mem_compaction_level >= result.num_levels) {
     result.max_mem_compaction_level = result.num_levels - 1;
   }
@@ -191,12 +215,11 @@ void SuperVersionUnrefHandle(void* ptr) {
 }
 }  // anonymous namespace
 
-ColumnFamilyData::ColumnFamilyData(const std::string& dbname, uint32_t id,
-                                   const std::string& name,
+ColumnFamilyData::ColumnFamilyData(uint32_t id, const std::string& name,
                                    Version* dummy_versions, Cache* table_cache,
-                                   const ColumnFamilyOptions& options,
+                                   const ColumnFamilyOptions& cf_options,
                                    const DBOptions* db_options,
-                                   const EnvOptions& storage_options,
+                                   const EnvOptions& env_options,
                                    ColumnFamilySet* column_family_set)
     : id_(id),
       name_(name),
@@ -204,10 +227,10 @@ ColumnFamilyData::ColumnFamilyData(const std::string& dbname, uint32_t id,
       current_(nullptr),
       refs_(0),
       dropped_(false),
-      internal_comparator_(options.comparator),
-      internal_filter_policy_(options.filter_policy),
-      options_(*db_options, SanitizeOptions(&internal_comparator_,
-                                            &internal_filter_policy_, options)),
+      internal_comparator_(cf_options.comparator),
+      options_(*db_options, SanitizeOptions(&internal_comparator_, cf_options)),
+      ioptions_(options_),
+      mutable_cf_options_(options_, ioptions_),
       mem_(nullptr),
       imm_(options_.min_write_buffer_number_to_merge),
       super_version_(nullptr),
@@ -216,34 +239,33 @@ ColumnFamilyData::ColumnFamilyData(const std::string& dbname, uint32_t id,
       next_(nullptr),
       prev_(nullptr),
       log_number_(0),
-      need_slowdown_for_num_level0_files_(false),
       column_family_set_(column_family_set) {
   Ref();
 
   // if dummy_versions is nullptr, then this is a dummy column family.
   if (dummy_versions != nullptr) {
-    internal_stats_.reset(new InternalStats(
-        options_.num_levels, db_options->env, db_options->statistics.get()));
-    table_cache_.reset(new TableCache(&options_, storage_options, table_cache));
-    if (options_.compaction_style == kCompactionStyleUniversal) {
+    internal_stats_.reset(
+        new InternalStats(ioptions_.num_levels, db_options->env, this));
+    table_cache_.reset(new TableCache(ioptions_, env_options, table_cache));
+    if (ioptions_.compaction_style == kCompactionStyleUniversal) {
       compaction_picker_.reset(
-          new UniversalCompactionPicker(&options_, &internal_comparator_));
-    } else if (options_.compaction_style == kCompactionStyleLevel) {
+          new UniversalCompactionPicker(ioptions_, &internal_comparator_));
+    } else if (ioptions_.compaction_style == kCompactionStyleLevel) {
       compaction_picker_.reset(
-          new LevelCompactionPicker(&options_, &internal_comparator_));
+          new LevelCompactionPicker(ioptions_, &internal_comparator_));
     } else {
-      assert(options_.compaction_style == kCompactionStyleFIFO);
+      assert(ioptions_.compaction_style == kCompactionStyleFIFO);
       compaction_picker_.reset(
-          new FIFOCompactionPicker(&options_, &internal_comparator_));
+          new FIFOCompactionPicker(ioptions_, &internal_comparator_));
     }
 
-    Log(options_.info_log, "Options for column family \"%s\":\n",
+    Log(ioptions_.info_log, "Options for column family \"%s\":\n",
         name.c_str());
     const ColumnFamilyOptions* cf_options = &options_;
-    cf_options->Dump(options_.info_log.get());
+    cf_options->Dump(ioptions_.info_log);
   }
 
-  RecalculateWriteStallConditions();
+  RecalculateWriteStallConditions(mutable_cf_options_);
 }
 
 // DB mutex held
@@ -296,82 +318,107 @@ ColumnFamilyData::~ColumnFamilyData() {
   }
 }
 
-void ColumnFamilyData::RecalculateWriteStallConditions() {
-  need_wait_for_num_memtables_ =
-    (imm()->size() == options()->max_write_buffer_number - 1);
-
+void ColumnFamilyData::RecalculateWriteStallConditions(
+      const MutableCFOptions& mutable_cf_options) {
   if (current_ != nullptr) {
-    need_wait_for_num_level0_files_ =
-      (current_->NumLevelFiles(0) >= options()->level0_stop_writes_trigger);
-  } else {
-    need_wait_for_num_level0_files_ = false;
-  }
+    const double score = current_->MaxCompactionScore();
+    const int max_level = current_->MaxCompactionScoreLevel();
 
-  RecalculateWriteStallRateLimitsConditions();
-}
+    auto write_controller = column_family_set_->write_controller_;
 
-void ColumnFamilyData::RecalculateWriteStallRateLimitsConditions() {
-  if (current_ != nullptr) {
-    exceeds_hard_rate_limit_ =
-        (options()->hard_rate_limit > 1.0 &&
-         current_->MaxCompactionScore() > options()->hard_rate_limit);
-
-    exceeds_soft_rate_limit_ =
-        (options()->soft_rate_limit > 0.0 &&
-         current_->MaxCompactionScore() > options()->soft_rate_limit);
-  } else {
-    exceeds_hard_rate_limit_ = false;
-    exceeds_soft_rate_limit_ = false;
+    if (imm()->size() == options_.max_write_buffer_number) {
+      write_controller_token_ = write_controller->GetStopToken();
+      internal_stats_->AddCFStats(InternalStats::MEMTABLE_COMPACTION, 1);
+      Log(ioptions_.info_log,
+          "[%s] Stopping writes because we have %d immutable memtables "
+          "(waiting for flush)",
+          name_.c_str(), imm()->size());
+    } else if (current_->NumLevelFiles(0) >=
+               mutable_cf_options.level0_stop_writes_trigger) {
+      write_controller_token_ = write_controller->GetStopToken();
+      internal_stats_->AddCFStats(InternalStats::LEVEL0_NUM_FILES, 1);
+      Log(ioptions_.info_log,
+          "[%s] Stopping writes because we have %d level-0 files",
+          name_.c_str(), current_->NumLevelFiles(0));
+    } else if (mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
+               current_->NumLevelFiles(0) >=
+                   mutable_cf_options.level0_slowdown_writes_trigger) {
+      uint64_t slowdown = SlowdownAmount(
+          current_->NumLevelFiles(0),
+          mutable_cf_options.level0_slowdown_writes_trigger,
+          mutable_cf_options.level0_stop_writes_trigger);
+      write_controller_token_ = write_controller->GetDelayToken(slowdown);
+      internal_stats_->AddCFStats(InternalStats::LEVEL0_SLOWDOWN, slowdown);
+      Log(ioptions_.info_log,
+          "[%s] Stalling writes because we have %d level-0 files (%" PRIu64
+          "us)",
+          name_.c_str(), current_->NumLevelFiles(0), slowdown);
+    } else if (options_.hard_rate_limit > 1.0 &&
+               score > options_.hard_rate_limit) {
+      uint64_t kHardLimitSlowdown = 1000;
+      write_controller_token_ =
+          write_controller->GetDelayToken(kHardLimitSlowdown);
+      internal_stats_->RecordLevelNSlowdown(max_level, kHardLimitSlowdown,
+                                            false);
+      Log(ioptions_.info_log,
+          "[%s] Stalling writes because we hit hard limit on level %d. "
+          "(%" PRIu64 "us)",
+          name_.c_str(), max_level, kHardLimitSlowdown);
+    } else if (options_.soft_rate_limit > 0.0 &&
+               score > options_.soft_rate_limit) {
+      uint64_t slowdown = SlowdownAmount(score, options_.soft_rate_limit,
+                                         options_.hard_rate_limit);
+      write_controller_token_ = write_controller->GetDelayToken(slowdown);
+      internal_stats_->RecordLevelNSlowdown(max_level, slowdown, true);
+      Log(ioptions_.info_log,
+          "[%s] Stalling writes because we hit soft limit on level %d (%" PRIu64
+          "us)",
+          name_.c_str(), max_level, slowdown);
+    } else {
+      write_controller_token_.reset();
+    }
   }
 }
 
 const EnvOptions* ColumnFamilyData::soptions() const {
-  return &(column_family_set_->storage_options_);
+  return &(column_family_set_->env_options_);
 }
 
-void ColumnFamilyData::SetCurrent(Version* current) {
-  current_ = current;
-  need_slowdown_for_num_level0_files_ =
-      (options_.level0_slowdown_writes_trigger >= 0 &&
-       current_->NumLevelFiles(0) >= options_.level0_slowdown_writes_trigger);
-}
+void ColumnFamilyData::SetCurrent(Version* current) { current_ = current; }
 
-void ColumnFamilyData::CreateNewMemtable() {
+void ColumnFamilyData::CreateNewMemtable(const MemTableOptions& moptions) {
   assert(current_ != nullptr);
   if (mem_ != nullptr) {
     delete mem_->Unref();
   }
-  mem_ = new MemTable(internal_comparator_, options_);
+  mem_ = new MemTable(internal_comparator_, ioptions_, moptions);
   mem_->Ref();
 }
 
-Compaction* ColumnFamilyData::PickCompaction(LogBuffer* log_buffer) {
-  auto result = compaction_picker_->PickCompaction(current_, log_buffer);
-  RecalculateWriteStallRateLimitsConditions();
+Compaction* ColumnFamilyData::PickCompaction(
+    const MutableCFOptions& mutable_options, LogBuffer* log_buffer) {
+  auto result = compaction_picker_->PickCompaction(
+      mutable_options, current_, log_buffer);
   return result;
 }
 
-Compaction* ColumnFamilyData::CompactRange(int input_level, int output_level,
-                                           const InternalKey* begin,
-                                           const InternalKey* end,
-                                           InternalKey** compaction_end) {
-  return compaction_picker_->CompactRange(current_, input_level, output_level,
-                                          begin, end, compaction_end);
+Compaction* ColumnFamilyData::CompactRange(
+    const MutableCFOptions& mutable_cf_options,
+    int input_level, int output_level, uint32_t output_path_id,
+    const InternalKey* begin, const InternalKey* end,
+    InternalKey** compaction_end) {
+  return compaction_picker_->CompactRange(
+      mutable_cf_options, current_, input_level, output_level,
+      output_path_id, begin, end, compaction_end);
 }
 
 SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(
     port::Mutex* db_mutex) {
   SuperVersion* sv = nullptr;
-  if (LIKELY(column_family_set_->db_options_->allow_thread_local)) {
-    sv = GetThreadLocalSuperVersion(db_mutex);
-    sv->Ref();
-    if (!ReturnThreadLocalSuperVersion(sv)) {
-      sv->Unref();
-    }
-  } else {
-    db_mutex->Lock();
-    sv = super_version_->Ref();
-    db_mutex->Unlock();
+  sv = GetThreadLocalSuperVersion(db_mutex);
+  sv->Ref();
+  if (!ReturnThreadLocalSuperVersion(sv)) {
+    sv->Unref();
   }
   return sv;
 }
@@ -400,11 +447,11 @@ SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
   sv = static_cast<SuperVersion*>(ptr);
   if (sv == SuperVersion::kSVObsolete ||
       sv->version_number != super_version_number_.load()) {
-    RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_ACQUIRES);
+    RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_ACQUIRES);
     SuperVersion* sv_to_delete = nullptr;
 
     if (sv && sv->Unref()) {
-      RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_CLEANUPS);
+      RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_CLEANUPS);
       db_mutex->Lock();
       // NOTE: underlying resources held by superversion (sst files) might
       // not be released until the next background job.
@@ -442,18 +489,24 @@ bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
 
 SuperVersion* ColumnFamilyData::InstallSuperVersion(
     SuperVersion* new_superversion, port::Mutex* db_mutex) {
+  db_mutex->AssertHeld();
+  return InstallSuperVersion(new_superversion, db_mutex, mutable_cf_options_);
+}
+
+SuperVersion* ColumnFamilyData::InstallSuperVersion(
+    SuperVersion* new_superversion, port::Mutex* db_mutex,
+    const MutableCFOptions& mutable_cf_options) {
   new_superversion->db_mutex = db_mutex;
+  new_superversion->mutable_cf_options = mutable_cf_options;
   new_superversion->Init(mem_, imm_.current(), current_);
   SuperVersion* old_superversion = super_version_;
   super_version_ = new_superversion;
   ++super_version_number_;
   super_version_->version_number = super_version_number_;
   // Reset SuperVersions cached in thread local storage
-  if (column_family_set_->db_options_->allow_thread_local) {
-    ResetThreadLocalSuperVersions();
-  }
+  ResetThreadLocalSuperVersions();
 
-  RecalculateWriteStallConditions();
+  RecalculateWriteStallConditions(mutable_cf_options);
 
   if (old_superversion != nullptr && old_superversion->Unref()) {
     old_superversion->Cleanup();
@@ -478,19 +531,33 @@ void ColumnFamilyData::ResetThreadLocalSuperVersions() {
   }
 }
 
+bool ColumnFamilyData::SetOptions(
+      const std::unordered_map<std::string, std::string>& options_map) {
+  MutableCFOptions new_mutable_cf_options;
+  if (GetMutableOptionsFromStrings(mutable_cf_options_, options_map,
+                                   &new_mutable_cf_options)) {
+    mutable_cf_options_ = new_mutable_cf_options;
+    mutable_cf_options_.RefreshDerivedOptions(ioptions_);
+    return true;
+  }
+  return false;
+}
+
 ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
                                  const DBOptions* db_options,
-                                 const EnvOptions& storage_options,
-                                 Cache* table_cache)
+                                 const EnvOptions& env_options,
+                                 Cache* table_cache,
+                                 WriteController* write_controller)
     : max_column_family_(0),
-      dummy_cfd_(new ColumnFamilyData(dbname, 0, "", nullptr, nullptr,
+      dummy_cfd_(new ColumnFamilyData(0, "", nullptr, nullptr,
                                       ColumnFamilyOptions(), db_options,
-                                      storage_options_, nullptr)),
+                                      env_options, nullptr)),
       default_cfd_cache_(nullptr),
       db_name_(dbname),
       db_options_(db_options),
-      storage_options_(storage_options),
+      env_options_(env_options),
       table_cache_(table_cache),
+      write_controller_(write_controller),
       spin_lock_(ATOMIC_FLAG_INIT) {
   // initialize linked list
   dummy_cfd_->prev_ = dummy_cfd_;
@@ -554,8 +621,8 @@ ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
     const ColumnFamilyOptions& options) {
   assert(column_families_.find(name) == column_families_.end());
   ColumnFamilyData* new_cfd =
-      new ColumnFamilyData(db_name_, id, name, dummy_versions, table_cache_,
-                           options, db_options_, storage_options_, this);
+      new ColumnFamilyData(id, name, dummy_versions, table_cache_, options,
+                           db_options_, env_options_, this);
   Lock();
   column_families_.insert({name, id});
   column_family_data_.insert({id, new_cfd});
@@ -614,6 +681,11 @@ bool ColumnFamilyMemTablesImpl::Seek(uint32_t column_family_id) {
     column_family_set_->Lock();
     current_ = column_family_set_->GetColumnFamily(column_family_id);
     column_family_set_->Unlock();
+    // TODO(icanadi) Maybe remove column family from the hash table when it's
+    // dropped?
+    if (current_ != nullptr && current_->IsDropped()) {
+      current_ = nullptr;
+    }
   }
   handle_.SetCFD(current_);
   return current_ != nullptr;
@@ -639,4 +711,29 @@ ColumnFamilyHandle* ColumnFamilyMemTablesImpl::GetColumnFamilyHandle() {
   return &handle_;
 }
 
+void ColumnFamilyMemTablesImpl::CheckMemtableFull() {
+  if (current_ != nullptr && current_->mem()->ShouldScheduleFlush()) {
+    flush_scheduler_->ScheduleFlush(current_);
+    current_->mem()->MarkFlushScheduled();
+  }
+}
+
+uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family) {
+  uint32_t column_family_id = 0;
+  if (column_family != nullptr) {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    column_family_id = cfh->GetID();
+  }
+  return column_family_id;
+}
+
+const Comparator* GetColumnFamilyUserComparator(
+    ColumnFamilyHandle* column_family) {
+  if (column_family != nullptr) {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    return cfh->user_comparator();
+  }
+  return nullptr;
+}
+
 }  // namespace rocksdb

db/column_family.h

@@ -19,8 +19,11 @@
 #include "rocksdb/env.h"
 #include "db/memtable_list.h"
 #include "db/write_batch_internal.h"
+#include "db/write_controller.h"
 #include "db/table_cache.h"
 #include "util/thread_local.h"
+#include "db/flush_scheduler.h"
+#include "util/mutable_cf_options.h"
 
 namespace rocksdb {
 
@@ -46,6 +49,7 @@ class ColumnFamilyHandleImpl : public ColumnFamilyHandle {
   // destroy without mutex
   virtual ~ColumnFamilyHandleImpl();
   virtual ColumnFamilyData* cfd() const { return cfd_; }
+  virtual const Comparator* user_comparator() const;
 
   virtual uint32_t GetID() const;
 
@@ -78,6 +82,7 @@ struct SuperVersion {
   MemTable* mem;
   MemTableListVersion* imm;
   Version* current;
+  MutableCFOptions mutable_cf_options;
   std::atomic<uint32_t> refs;
   // We need to_delete because during Cleanup(), imm->Unref() returns
   // all memtables that we need to free through this vector. We then
@@ -113,7 +118,6 @@ struct SuperVersion {
 };
 
 extern ColumnFamilyOptions SanitizeOptions(const InternalKeyComparator* icmp,
-                                           const InternalFilterPolicy* ipolicy,
                                            const ColumnFamilyOptions& src);
 
 class ColumnFamilySet;
@@ -133,7 +137,7 @@ class ColumnFamilyData {
   void Ref() { ++refs_; }
   // will just decrease reference count to 0, but will not delete it. returns
   // true if the ref count was decreased to zero. in that case, it can be
-  // deleted by the caller immediatelly, or later, by calling
+  // deleted by the caller immediately, or later, by calling
   // FreeDeadColumnFamilies()
   bool Unref() {
     assert(refs_ > 0);
@@ -157,6 +161,7 @@ class ColumnFamilyData {
     // can't drop default CF
     assert(id_ != 0);
     dropped_ = true;
+    write_controller_token_.reset();
   }
   bool IsDropped() const { return dropped_; }
 
@@ -169,6 +174,21 @@ class ColumnFamilyData {
   // thread-safe
   const Options* options() const { return &options_; }
   const EnvOptions* soptions() const;
+  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.
+  const MutableCFOptions* mutable_cf_options() const {
+    return &(super_version_->mutable_cf_options);
+  }
+  // REQUIRES: DB mutex held
+  // This returns the latest MutableCFOptions, which may be not in effect yet.
+  const MutableCFOptions* GetLatestMutableCFOptions() const {
+    return &mutable_cf_options_;
+  }
+  // REQUIRES: DB mutex held
+  bool SetOptions(
+      const std::unordered_map<std::string, std::string>& options_map);
 
   InternalStats* internal_stats() { return internal_stats_.get(); }
 
@@ -178,15 +198,19 @@ class ColumnFamilyData {
   Version* dummy_versions() { return dummy_versions_; }
   void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }
   void SetCurrent(Version* current);
-  void CreateNewMemtable();
+  void CreateNewMemtable(const MemTableOptions& moptions);
 
   TableCache* table_cache() const { return table_cache_.get(); }
 
   // See documentation in compaction_picker.h
-  Compaction* PickCompaction(LogBuffer* log_buffer);
-  Compaction* CompactRange(int input_level, int output_level,
-                           const InternalKey* begin, const InternalKey* end,
-                           InternalKey** compaction_end);
+  // REQUIRES: DB mutex held
+  Compaction* PickCompaction(const MutableCFOptions& mutable_options,
+                             LogBuffer* log_buffer);
+  Compaction* CompactRange(
+      const MutableCFOptions& mutable_cf_options,
+      int input_level, int output_level, uint32_t output_path_id,
+      const InternalKey* begin, const InternalKey* end,
+      InternalKey** compaction_end);
 
   CompactionPicker* compaction_picker() { return compaction_picker_.get(); }
   // thread-safe
@@ -218,40 +242,20 @@ class ColumnFamilyData {
   // if its reference count is zero and needs deletion or nullptr if not
   // As argument takes a pointer to allocated SuperVersion to enable
   // the clients to allocate SuperVersion outside of mutex.
+  SuperVersion* InstallSuperVersion(SuperVersion* new_superversion,
+                                    port::Mutex* db_mutex,
+                                    const MutableCFOptions& mutable_cf_options);
   SuperVersion* InstallSuperVersion(SuperVersion* new_superversion,
                                     port::Mutex* db_mutex);
 
   void ResetThreadLocalSuperVersions();
 
-  // A Flag indicating whether write needs to slowdown because of there are
-  // too many number of level0 files.
-  bool NeedSlowdownForNumLevel0Files() const {
-    return need_slowdown_for_num_level0_files_;
-  }
-
-  bool NeedWaitForNumLevel0Files() const {
-    return need_wait_for_num_level0_files_;
-  }
-
-  bool NeedWaitForNumMemtables() const {
-    return need_wait_for_num_memtables_;
-  }
-
-  bool ExceedsSoftRateLimit() const {
-    return exceeds_soft_rate_limit_;
-  }
-
-  bool ExceedsHardRateLimit() const {
-    return exceeds_hard_rate_limit_;
-  }
-
  private:
   friend class ColumnFamilySet;
-  ColumnFamilyData(const std::string& dbname, uint32_t id,
-                   const std::string& name, Version* dummy_versions,
-                   Cache* table_cache, const ColumnFamilyOptions& options,
-                   const DBOptions* db_options,
-                   const EnvOptions& storage_options,
+  ColumnFamilyData(uint32_t id, const std::string& name,
+                   Version* dummy_versions, Cache* table_cache,
+                   const ColumnFamilyOptions& options,
+                   const DBOptions* db_options, const EnvOptions& env_options,
                    ColumnFamilySet* column_family_set);
 
   // Recalculate some small conditions, which are changed only during
@@ -259,8 +263,8 @@ class ColumnFamilyData {
   // recalculation of compaction score. These values are used in
   // DBImpl::MakeRoomForWrite function to decide, if it need to make
   // a write stall
-  void RecalculateWriteStallConditions();
-  void RecalculateWriteStallRateLimitsConditions();
+  void RecalculateWriteStallConditions(
+      const MutableCFOptions& mutable_cf_options);
 
   uint32_t id_;
   const std::string name_;
@@ -271,9 +275,10 @@ class ColumnFamilyData {
   bool dropped_;               // true if client dropped it
 
   const InternalKeyComparator internal_comparator_;
-  const InternalFilterPolicy internal_filter_policy_;
 
-  Options const options_;
+  const Options options_;
+  const ImmutableCFOptions ioptions_;
+  MutableCFOptions mutable_cf_options_;
 
   std::unique_ptr<TableCache> table_cache_;
 
@@ -302,31 +307,13 @@ class ColumnFamilyData {
   // recovered from
   uint64_t log_number_;
 
-  // A flag indicating whether we should delay writes because
-  // we have too many level 0 files
-  bool need_slowdown_for_num_level0_files_;
-
-  // These 4 variables are updated only after compaction,
-  // adding new memtable, flushing memtables to files
-  // and/or add recalculation of compaction score.
-  // That's why theirs values are cached in ColumnFamilyData.
-  // Recalculation is made by RecalculateWriteStallConditions and
-  // RecalculateWriteStallRateLimitsConditions function. They are used
-  // in DBImpl::MakeRoomForWrite function to decide, if it need
-  // to sleep during write operation
-  bool need_wait_for_num_memtables_;
-
-  bool need_wait_for_num_level0_files_;
-
-  bool exceeds_hard_rate_limit_;
-
-  bool exceeds_soft_rate_limit_;
-
   // An object that keeps all the compaction stats
   // and picks the next compaction
   std::unique_ptr<CompactionPicker> compaction_picker_;
 
   ColumnFamilySet* column_family_set_;
+
+  std::unique_ptr<WriteControllerToken> write_controller_token_;
 };
 
 // ColumnFamilySet has interesting thread-safety requirements
@@ -368,7 +355,8 @@ class ColumnFamilySet {
   };
 
   ColumnFamilySet(const std::string& dbname, const DBOptions* db_options,
-                  const EnvOptions& storage_options, Cache* table_cache);
+                  const EnvOptions& env_options, Cache* table_cache,
+                  WriteController* write_controller);
   ~ColumnFamilySet();
 
   ColumnFamilyData* GetDefault() const;
@@ -421,8 +409,9 @@ class ColumnFamilySet {
 
   const std::string db_name_;
   const DBOptions* const db_options_;
-  const EnvOptions storage_options_;
+  const EnvOptions env_options_;
   Cache* table_cache_;
+  WriteController* write_controller_;
   std::atomic_flag spin_lock_;
 };
 
@@ -430,8 +419,11 @@ class ColumnFamilySet {
 // memtables of different column families (specified by ID in the write batch)
 class ColumnFamilyMemTablesImpl : public ColumnFamilyMemTables {
  public:
-  explicit ColumnFamilyMemTablesImpl(ColumnFamilySet* column_family_set)
-      : column_family_set_(column_family_set), current_(nullptr) {}
+  explicit ColumnFamilyMemTablesImpl(ColumnFamilySet* column_family_set,
+                                     FlushScheduler* flush_scheduler)
+      : column_family_set_(column_family_set),
+        current_(nullptr),
+        flush_scheduler_(flush_scheduler) {}
 
   // sets current_ to ColumnFamilyData with column_family_id
   // returns false if column family doesn't exist
@@ -450,10 +442,18 @@ class ColumnFamilyMemTablesImpl : public ColumnFamilyMemTables {
   // Returns column family handle for the selected column family
   virtual ColumnFamilyHandle* GetColumnFamilyHandle() override;
 
+  virtual void CheckMemtableFull() override;
+
  private:
   ColumnFamilySet* column_family_set_;
   ColumnFamilyData* current_;
+  FlushScheduler* flush_scheduler_;
   ColumnFamilyHandleInternal handle_;
 };
 
+extern uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family);
+
+extern const Comparator* GetColumnFamilyUserComparator(
+    ColumnFamilyHandle* column_family);
+
 }  // namespace rocksdb

db/column_family_test.cc

@@ -408,9 +408,15 @@ TEST(ColumnFamilyTest, WriteBatchFailure) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two"});
   WriteBatch batch;
+  batch.Put(handles_[0], Slice("existing"), Slice("column-family"));
   batch.Put(handles_[1], Slice("non-existing"), Slice("column-family"));
   ASSERT_OK(db_->Write(WriteOptions(), &batch));
   DropColumnFamilies({1});
+  WriteOptions woptions_ignore_missing_cf;
+  woptions_ignore_missing_cf.ignore_missing_column_families = true;
+  batch.Put(handles_[0], Slice("still here"), Slice("column-family"));
+  ASSERT_OK(db_->Write(woptions_ignore_missing_cf, &batch));
+  ASSERT_EQ("column-family", Get(0, "still here"));
   Status s = db_->Write(WriteOptions(), &batch);
   ASSERT_TRUE(s.IsInvalidArgument());
   Close();
@@ -746,9 +752,10 @@ TEST(ColumnFamilyTest, DifferentCompactionStyles) {
   default_cf.num_levels = 3;
   default_cf.write_buffer_size = 64 << 10;  // 64KB
   default_cf.target_file_size_base = 30 << 10;
-  default_cf.filter_policy = nullptr;
-  default_cf.no_block_cache = true;
   default_cf.source_compaction_factor = 100;
+  BlockBasedTableOptions table_options;
+  table_options.no_block_cache = true;
+  default_cf.table_factory.reset(NewBlockBasedTableFactory(table_options));
 
   one.compaction_style = kCompactionStyleUniversal;
   // trigger compaction if there are >= 4 files
@@ -857,6 +864,7 @@ TEST(ColumnFamilyTest, NewIteratorsTest) {
 TEST(ColumnFamilyTest, ReadOnlyDBTest) {
   Open();
   CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
+  ASSERT_OK(Put(0, "a", "b"));
   ASSERT_OK(Put(1, "foo", "bla"));
   ASSERT_OK(Put(2, "foo", "blabla"));
   ASSERT_OK(Put(3, "foo", "blablabla"));
@@ -870,6 +878,29 @@ TEST(ColumnFamilyTest, ReadOnlyDBTest) {
   ASSERT_EQ("bla", Get(1, "foo"));
   ASSERT_EQ("blablablabla", Get(2, "foo"));
 
+
+  // test newiterators
+  {
+    std::vector<Iterator*> iterators;
+    ASSERT_OK(db_->NewIterators(ReadOptions(), handles_, &iterators));
+    for (auto it : iterators) {
+      it->SeekToFirst();
+    }
+    ASSERT_EQ(IterStatus(iterators[0]), "a->b");
+    ASSERT_EQ(IterStatus(iterators[1]), "foo->bla");
+    ASSERT_EQ(IterStatus(iterators[2]), "foo->blablablabla");
+    for (auto it : iterators) {
+      it->Next();
+    }
+    ASSERT_EQ(IterStatus(iterators[0]), "(invalid)");
+    ASSERT_EQ(IterStatus(iterators[1]), "(invalid)");
+    ASSERT_EQ(IterStatus(iterators[2]), "(invalid)");
+
+    for (auto it : iterators) {
+      delete it;
+    }
+  }
+
   Close();
   // can't open dropped column family
   Status s = OpenReadOnly({"default", "one", "two"});
@@ -889,7 +920,9 @@ TEST(ColumnFamilyTest, DontRollEmptyLogs) {
   }
   int num_writable_file_start = env_->GetNumberOfNewWritableFileCalls();
   // this will trigger the flushes
-  ASSERT_OK(db_->Write(WriteOptions(), nullptr));
+  for (size_t i = 0; i <= 4; ++i) {
+    ASSERT_OK(Flush(i));
+  }
 
   for (int i = 0; i < 4; ++i) {
     dbfull()->TEST_WaitForFlushMemTable(handles_[i]);

db/compaction.cc

@@ -9,7 +9,10 @@
 
 #include "db/compaction.h"
 
+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif
+
 #include <inttypes.h>
 #include <vector>
 
@@ -18,7 +21,7 @@
 
 namespace rocksdb {
 
-static uint64_t TotalFileSize(const std::vector<FileMetaData*>& files) {
+uint64_t TotalFileSize(const std::vector<FileMetaData*>& files) {
   uint64_t sum = 0;
   for (size_t i = 0; i < files.size() && files[i]; i++) {
     sum += files[i]->fd.GetFileSize();
@@ -26,14 +29,14 @@ static uint64_t TotalFileSize(const std::vector<FileMetaData*>& files) {
   return sum;
 }
 
-Compaction::Compaction(Version* input_version, int level, int out_level,
+Compaction::Compaction(Version* input_version, int start_level, int out_level,
                        uint64_t target_file_size,
                        uint64_t max_grandparent_overlap_bytes,
                        uint32_t output_path_id,
                        CompressionType output_compression, bool seek_compaction,
                        bool deletion_compaction)
-    : level_(level),
-      out_level_(out_level),
+    : start_level_(start_level),
+      output_level_(out_level),
       max_output_file_size_(target_file_size),
       max_grandparent_overlap_bytes_(max_grandparent_overlap_bytes),
       input_version_(input_version),
@@ -53,7 +56,6 @@ Compaction::Compaction(Version* input_version, int level, int out_level,
       is_full_compaction_(false),
       is_manual_compaction_(false),
       level_ptrs_(std::vector<size_t>(number_levels_)) {
-
   cfd_->Ref();
   input_version_->Ref();
   edit_ = new VersionEdit();
@@ -61,6 +63,12 @@ Compaction::Compaction(Version* input_version, int level, int out_level,
   for (int i = 0; i < number_levels_; i++) {
     level_ptrs_[i] = 0;
   }
+  int num_levels = output_level_ - start_level_ + 1;
+  input_levels_.resize(num_levels);
+  inputs_.resize(num_levels);
+  for (int i = 0; i < num_levels; ++i) {
+    inputs_[i].level = start_level_ + i;
+  }
 }
 
 Compaction::~Compaction() {
@@ -75,43 +83,50 @@ Compaction::~Compaction() {
   }
 }
 
+void Compaction::GenerateFileLevels() {
+  input_levels_.resize(num_input_levels());
+  for (int which = 0; which < num_input_levels(); which++) {
+    DoGenerateFileLevel(&input_levels_[which], inputs_[which].files, &arena_);
+  }
+}
+
 bool Compaction::IsTrivialMove() const {
   // Avoid a move if there is lots of overlapping grandparent data.
   // Otherwise, the move could create a parent file that will require
   // a very expensive merge later on.
-  // If level_== out_level_, the purpose is to force compaction filter to be
-  // applied to that level, and thus cannot be a trivia move.
-  return (level_ != out_level_ &&
+  // If start_level_== output_level_, the purpose is to force compaction
+  // filter to be applied to that level, and thus cannot be a trivia move.
+  return (start_level_ != output_level_ &&
+          num_input_levels() == 2 &&
           num_input_files(0) == 1 &&
           num_input_files(1) == 0 &&
           TotalFileSize(grandparents_) <= max_grandparent_overlap_bytes_);
 }
 
-bool Compaction::IsDeletionCompaction() const { return deletion_compaction_; }
-
 void Compaction::AddInputDeletions(VersionEdit* edit) {
-  for (int which = 0; which < 2; which++) {
+  for (int which = 0; which < num_input_levels(); which++) {
     for (size_t i = 0; i < inputs_[which].size(); i++) {
-      edit->DeleteFile(level_ + which, inputs_[which][i]->fd.GetNumber());
+      edit->DeleteFile(level(which), inputs_[which][i]->fd.GetNumber());
     }
   }
 }
 
-bool Compaction::IsBaseLevelForKey(const Slice& user_key) {
-  assert(cfd_->options()->compaction_style != kCompactionStyleFIFO);
-  if (cfd_->options()->compaction_style == kCompactionStyleUniversal) {
+bool Compaction::KeyNotExistsBeyondOutputLevel(const Slice& user_key) {
+  assert(cfd_->ioptions()->compaction_style != kCompactionStyleFIFO);
+  if (cfd_->ioptions()->compaction_style == kCompactionStyleUniversal) {
     return bottommost_level_;
   }
   // Maybe use binary search to find right entry instead of linear search?
   const Comparator* user_cmp = cfd_->user_comparator();
-  for (int lvl = level_ + 2; lvl < number_levels_; lvl++) {
+  for (int lvl = output_level_ + 1; lvl < number_levels_; lvl++) {
     const std::vector<FileMetaData*>& files = input_version_->files_[lvl];
     for (; level_ptrs_[lvl] < files.size(); ) {
       FileMetaData* f = files[level_ptrs_[lvl]];
       if (user_cmp->Compare(user_key, f->largest.user_key()) <= 0) {
         // We've advanced far enough
         if (user_cmp->Compare(user_key, f->smallest.user_key()) >= 0) {
-          // Key falls in this file's range, so definitely not base level
+          // Key falls in this file's range, so definitely
+          // exists beyond output level
           return false;
         }
         break;
@@ -149,34 +164,34 @@ bool Compaction::ShouldStopBefore(const Slice& internal_key) {
 }
 
 // Mark (or clear) each file that is being compacted
-void Compaction::MarkFilesBeingCompacted(bool value) {
-  for (int i = 0; i < 2; i++) {
-    std::vector<FileMetaData*> v = inputs_[i];
+void Compaction::MarkFilesBeingCompacted(bool mark_as_compacted) {
+  for (int i = 0; i < num_input_levels(); i++) {
     for (unsigned int j = 0; j < inputs_[i].size(); j++) {
-      assert(value ? !inputs_[i][j]->being_compacted :
-                      inputs_[i][j]->being_compacted);
-      inputs_[i][j]->being_compacted = value;
+      assert(mark_as_compacted ? !inputs_[i][j]->being_compacted :
+                                  inputs_[i][j]->being_compacted);
+      inputs_[i][j]->being_compacted = mark_as_compacted;
     }
   }
 }
 
 // Is this compaction producing files at the bottommost level?
-void Compaction::SetupBottomMostLevel(bool isManual) {
-  assert(cfd_->options()->compaction_style != kCompactionStyleFIFO);
-  if (cfd_->options()->compaction_style == kCompactionStyleUniversal) {
+void Compaction::SetupBottomMostLevel(bool is_manual) {
+  assert(cfd_->ioptions()->compaction_style != kCompactionStyleFIFO);
+  if (cfd_->ioptions()->compaction_style == kCompactionStyleUniversal) {
     // If universal compaction style is used and manual
     // compaction is occuring, then we are guaranteed that
     // all files will be picked in a single compaction
     // run. We can safely set bottommost_level_ = true.
     // If it is not manual compaction, then bottommost_level_
     // is already set when the Compaction was created.
-    if (isManual) {
+    if (is_manual) {
       bottommost_level_ = true;
     }
     return;
   }
   bottommost_level_ = true;
-  for (int i = output_level() + 1; i < number_levels_; i++) {
+  // checks whether there are files living beyond the output_level.
+  for (int i = output_level_ + 1; i < number_levels_; i++) {
     if (input_version_->NumLevelFiles(i) > 0) {
       bottommost_level_ = false;
       break;
@@ -202,7 +217,7 @@ void Compaction::ReleaseCompactionFiles(Status status) {
 }
 
 void Compaction::ResetNextCompactionIndex() {
-  input_version_->ResetNextCompactionIndex(level_);
+  input_version_->ResetNextCompactionIndex(start_level_);
 }
 
 namespace {
@@ -229,38 +244,39 @@ void Compaction::Summary(char* output, int len) {
   int write =
       snprintf(output, len, "Base version %" PRIu64
                             " Base level %d, seek compaction:%d, inputs: [",
-               input_version_->GetVersionNumber(), level_, seek_compaction_);
+               input_version_->GetVersionNumber(),
+               start_level_, seek_compaction_);
   if (write < 0 || write >= len) {
     return;
   }
 
-  write += InputSummary(inputs_[0], output + write, len - write);
-  if (write < 0 || write >= len) {
-    return;
-  }
-
-  write += snprintf(output + write, len - write, "], [");
-  if (write < 0 || write >= len) {
-    return;
-  }
-
-  write += InputSummary(inputs_[1], output + write, len - write);
-  if (write < 0 || write >= len) {
-    return;
+  for (int level = 0; level < num_input_levels(); ++level) {
+    if (level > 0) {
+      write += snprintf(output + write, len - write, "], [");
+      if (write < 0 || write >= len) {
+        return;
+      }
+    }
+    write += InputSummary(inputs_[level].files, output + write, len - write);
+    if (write < 0 || write >= len) {
+      return;
+    }
   }
 
   snprintf(output + write, len - write, "]");
 }
 
-uint64_t Compaction::OutputFilePreallocationSize() {
+uint64_t Compaction::OutputFilePreallocationSize(
+    const MutableCFOptions& mutable_options) {
   uint64_t preallocation_size = 0;
 
-  if (cfd_->options()->compaction_style == kCompactionStyleLevel) {
-    preallocation_size =
-        cfd_->compaction_picker()->MaxFileSizeForLevel(output_level());
+  if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel) {
+    preallocation_size = mutable_options.MaxFileSizeForLevel(output_level());
   } else {
-    for (const auto& f : inputs_[0]) {
-      preallocation_size += f->fd.GetFileSize();
+    for (int level = 0; level < num_input_levels(); ++level) {
+      for (const auto& f : inputs_[level].files) {
+        preallocation_size += f->fd.GetFileSize();
+      }
     }
   }
   // Over-estimate slightly so we don't end up just barely crossing

db/compaction.h

@@ -8,41 +8,91 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #pragma once
+#include "util/arena.h"
+#include "util/autovector.h"
+#include "util/mutable_cf_options.h"
 #include "db/version_set.h"
 
 namespace rocksdb {
 
+// The structure that manages compaction input files associated
+// with the same physical level.
+struct CompactionInputFiles {
+  int level;
+  std::vector<FileMetaData*> files;
+  inline bool empty() const { return files.empty(); }
+  inline size_t size() const { return files.size(); }
+  inline void clear() { files.clear(); }
+  inline FileMetaData* operator[](int i) const { return files[i]; }
+};
+
 class Version;
 class ColumnFamilyData;
 
 // A Compaction encapsulates information about a compaction.
 class Compaction {
  public:
+  // No copying allowed
+  Compaction(const Compaction&) = delete;
+  void operator=(const Compaction&) = delete;
+
   ~Compaction();
 
-  // Return the level that is being compacted.  Inputs from "level"
-  // will be merged.
-  int level() const { return level_; }
+  // Returns the level associated to the specified compaction input level.
+  // If compaction_input_level is not specified, then input_level is set to 0.
+  int level(int compaction_input_level = 0) const {
+    return inputs_[compaction_input_level].level;
+  }
 
   // Outputs will go to this level
-  int output_level() const { return out_level_; }
+  int output_level() const { return output_level_; }
+
+  // Returns the number of input levels in this compaction.
+  int num_input_levels() const { return inputs_.size(); }
 
   // Return the object that holds the edits to the descriptor done
   // by this compaction.
-  VersionEdit* edit() { return edit_; }
+  VersionEdit* edit() const { return edit_; }
 
-  // "which" must be either 0 or 1
-  int num_input_files(int which) const { return inputs_[which].size(); }
+  // Returns the number of input files associated to the specified
+  // compaction input level.
+  // The function will return 0 if when "compaction_input_level" < 0
+  // or "compaction_input_level" >= "num_input_levels()".
+  int num_input_files(size_t compaction_input_level) const {
+    if (compaction_input_level < inputs_.size()) {
+      return inputs_[compaction_input_level].size();
+    }
+    return 0;
+  }
 
   // Returns input version of the compaction
   Version* input_version() const { return input_version_; }
 
+  // Returns the ColumnFamilyData associated with the compaction.
   ColumnFamilyData* column_family_data() const { return cfd_; }
 
-  // Return the ith input file at "level()+which" ("which" must be 0 or 1).
-  FileMetaData* input(int which, int i) const { return inputs_[which][i]; }
+  // Returns the file meta data of the 'i'th input file at the
+  // specified compaction input level.
+  // REQUIREMENT: "compaction_input_level" must be >= 0 and
+  //              < "input_levels()"
+  FileMetaData* input(size_t compaction_input_level, int i) const {
+    assert(compaction_input_level < inputs_.size());
+    return inputs_[compaction_input_level][i];
+  }
 
-  std::vector<FileMetaData*>* inputs(int which) { return &inputs_[which]; }
+  // Returns the list of file meta data of the specified compaction
+  // input level.
+  // REQUIREMENT: "compaction_input_level" must be >= 0 and
+  //              < "input_levels()"
+  std::vector<FileMetaData*>* const inputs(size_t compaction_input_level) {
+    assert(compaction_input_level < inputs_.size());
+    return &inputs_[compaction_input_level].files;
+  }
+
+  // Returns the FileLevel of the specified compaction input level.
+  FileLevel* input_levels(int compaction_input_level) {
+    return &input_levels_[compaction_input_level];
+  }
 
   // Maximum size of files to build during this compaction.
   uint64_t MaxOutputFileSize() const { return max_output_file_size_; }
@@ -53,20 +103,25 @@ class Compaction {
   // Whether need to write output file to second DB path.
   uint32_t GetOutputPathId() const { return output_path_id_; }
 
-    // Is this a trivial compaction that can be implemented by just
+  // Generate input_levels_ from inputs_
+  // Should be called when inputs_ is stable
+  void GenerateFileLevels();
+
+  // Is this a trivial compaction that can be implemented by just
   // moving a single input file to the next level (no merging or splitting)
   bool IsTrivialMove() const;
 
-  // If true, just delete all files in inputs_[0]
-  bool IsDeletionCompaction() const;
+  // If true, then the comaction can be done by simply deleting input files.
+  bool IsDeletionCompaction() const {
+    return deletion_compaction_;
+  }
 
   // Add all inputs to this compaction as delete operations to *edit.
   void AddInputDeletions(VersionEdit* edit);
 
-  // Returns true if the information we have available guarantees that
-  // the compaction is producing data in "level+1" for which no data exists
-  // in levels greater than "level+1".
-  bool IsBaseLevelForKey(const Slice& user_key);
+  // Returns true if the available information we have guarantees that
+  // the input "user_key" does not exist in any level beyond "output_level()".
+  bool KeyNotExistsBeyondOutputLevel(const Slice& user_key);
 
   // Returns true iff we should stop building the current output
   // before processing "internal_key".
@@ -80,6 +135,9 @@ class Compaction {
   // Delete this compaction from the list of running compactions.
   void ReleaseCompactionFiles(Status status);
 
+  // Returns the summary of the compaction in "output" with maximum "len"
+  // in bytes.  The caller is responsible for the memory management of
+  // "output".
   void Summary(char* output, int len);
 
   // Return the score that was used to pick this compaction run.
@@ -94,10 +152,14 @@ class Compaction {
   // Was this compaction triggered manually by the client?
   bool IsManualCompaction() { return is_manual_compaction_; }
 
-  // Returns a number of byte that the output file should be preallocated to
+  // Return the MutableCFOptions that should be used throughout the compaction
+  // procedure
+  const MutableCFOptions* mutable_cf_options() { 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();
+  // is the sum of all input file sizes.
+  uint64_t OutputFilePreallocationSize(const MutableCFOptions& mutable_options);
 
  private:
   friend class CompactionPicker;
@@ -105,39 +167,46 @@ class Compaction {
   friend class FIFOCompactionPicker;
   friend class LevelCompactionPicker;
 
-  Compaction(Version* input_version, int level, int out_level,
+  Compaction(Version* input_version, int start_level, int out_level,
              uint64_t target_file_size, uint64_t max_grandparent_overlap_bytes,
              uint32_t output_path_id, CompressionType output_compression,
              bool seek_compaction = false, bool deletion_compaction = false);
 
-  int level_;
-  int out_level_; // levels to which output files are stored
+  const int start_level_;    // the lowest level to be compacted
+  const int output_level_;  // levels to which output files are stored
   uint64_t max_output_file_size_;
   uint64_t max_grandparent_overlap_bytes_;
+  MutableCFOptions mutable_cf_options_;
   Version* input_version_;
   VersionEdit* edit_;
   int number_levels_;
   ColumnFamilyData* cfd_;
+  Arena arena_;          // Arena used to allocate space for file_levels_
 
   uint32_t output_path_id_;
   CompressionType output_compression_;
   bool seek_compaction_;
-  // if true, just delete files in inputs_[0]
+  // If true, then the comaction can be done by simply deleting input files.
   bool deletion_compaction_;
 
-  // Each compaction reads inputs from "level_" and "level_+1"
-  std::vector<FileMetaData*> inputs_[2];      // The two sets of inputs
+  // Compaction input files organized by level.
+  autovector<CompactionInputFiles> inputs_;
+
+  // A copy of inputs_, organized more closely in memory
+  autovector<FileLevel, 2> input_levels_;
 
   // State used to check for number of of overlapping grandparent files
-  // (parent == level_ + 1, grandparent == level_ + 2)
+  // (grandparent == "output_level_ + 1")
+  // This vector is updated by Version::GetOverlappingInputs().
   std::vector<FileMetaData*> grandparents_;
-  size_t grandparent_index_;  // Index in grandparent_starts_
-  bool seen_key_;             // Some output key has been seen
+  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
-  int base_index_;   // index of the file in files_[level_]
-  int parent_index_; // index of some file with same range in files_[level_+1]
-  double score_;     // score that was used to pick this compaction.
+                               // and grandparent files
+  int base_index_;    // index of the file in files_[start_level_]
+  int parent_index_;  // index of some file with same range in
+                      // files_[start_level_+1]
+  double score_;      // score that was used to pick this compaction.
 
   // Is this compaction creating a file in the bottom most level?
   bool bottommost_level_;
@@ -147,21 +216,28 @@ class Compaction {
   // Is this compaction requested by the client?
   bool is_manual_compaction_;
 
-  // level_ptrs_ holds indices into input_version_->levels_: our state
-  // is that we are positioned at one of the file ranges for each
-  // higher level than the ones involved in this compaction (i.e. for
-  // all L >= level_ + 2).
+  // "level_ptrs_" holds indices into "input_version_->levels_", where each
+  // index remembers which file of an associated level we are currently used
+  // to check KeyNotExistsBeyondOutputLevel() for deletion operation.
+  // As it is for checking KeyNotExistsBeyondOutputLevel(), it only
+  // records indices for all levels beyond "output_level_".
   std::vector<size_t> level_ptrs_;
 
   // mark (or clear) all files that are being compacted
-  void MarkFilesBeingCompacted(bool);
+  void MarkFilesBeingCompacted(bool mark_as_compacted);
 
-  // Initialize whether compaction producing files at the bottommost level
-  void SetupBottomMostLevel(bool isManual);
+  // Initialize whether the compaction is producing files at the
+  // bottommost level.
+  //
+  // @see BottomMostLevel()
+  void SetupBottomMostLevel(bool is_manual);
 
   // In case of compaction error, reset the nextIndex that is used
   // to pick up the next file to be compacted from files_by_size_
   void ResetNextCompactionIndex();
 };
 
+// Utility function
+extern uint64_t TotalFileSize(const std::vector<FileMetaData*>& files);
+
 }  // namespace rocksdb

db/compaction_picker.cc

@@ -9,7 +9,10 @@
 
 #include "db/compaction_picker.h"
 
+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif
+
 #include <inttypes.h>
 #include <limits>
 #include "db/filename.h"
@@ -18,33 +21,6 @@
 
 namespace rocksdb {
 
-namespace {
-// Determine compression type, based on user options, level of the output
-// file and whether compression is disabled.
-// If enable_compression is false, then compression is always disabled no
-// matter what the values of the other two parameters are.
-// Otherwise, the compression type is determined based on options and level.
-CompressionType GetCompressionType(const Options& options, int level,
-                                   const bool enable_compression = true) {
-  if (!enable_compression) {
-    // disable compression
-    return kNoCompression;
-  }
-  // If the use has specified a different compression level for each level,
-  // then pick the compresison for that level.
-  if (!options.compression_per_level.empty()) {
-    const int n = options.compression_per_level.size() - 1;
-    // It is possible for level_ to be -1; in that case, we use level
-    // 0's compression.  This occurs mostly in backwards compatibility
-    // situations when the builder doesn't know what level the file
-    // belongs to.  Likewise, if level_ is beyond the end of the
-    // specified compression levels, use the last value.
-    return options.compression_per_level[std::max(0, std::min(level, n))];
-  } else {
-    return options.compression;
-  }
-}
-
 uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) {
   uint64_t sum = 0;
   for (size_t i = 0; i < files.size() && files[i]; i++) {
@@ -53,49 +29,42 @@ uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) {
   return sum;
 }
 
-// Multiple two operands. If they overflow, return op1.
-uint64_t MultiplyCheckOverflow(uint64_t op1, int op2) {
-  if (op1 == 0) {
-    return 0;
+namespace {
+// Determine compression type, based on user options, level of the output
+// file and whether compression is disabled.
+// If enable_compression is false, then compression is always disabled no
+// 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, int level,
+    const bool enable_compression = true) {
+  if (!enable_compression) {
+    // disable compression
+    return kNoCompression;
   }
-  if (op2 <= 0) {
-    return op1;
+  // If the use has specified a different compression level for each level,
+  // then pick the compression for that level.
+  if (!ioptions.compression_per_level.empty()) {
+    const int n = ioptions.compression_per_level.size() - 1;
+    // It is possible for level_ to be -1; in that case, we use level
+    // 0's compression.  This occurs mostly in backwards compatibility
+    // situations when the builder doesn't know what level the file
+    // belongs to.  Likewise, if level is beyond the end of the
+    // specified compression levels, use the last value.
+    return ioptions.compression_per_level[std::max(0, std::min(level, n))];
+  } else {
+    return ioptions.compression;
   }
-  uint64_t casted_op2 = (uint64_t) op2;
-  if (std::numeric_limits<uint64_t>::max() / op1 < casted_op2) {
-    return op1;
-  }
-  return op1 * casted_op2;
 }
 
+
 }  // anonymous namespace
 
-CompactionPicker::CompactionPicker(const Options* options,
+CompactionPicker::CompactionPicker(const ImmutableCFOptions& ioptions,
                                    const InternalKeyComparator* icmp)
-    : compactions_in_progress_(options->num_levels),
-      options_(options),
-      num_levels_(options->num_levels),
+    : ioptions_(ioptions),
+      compactions_in_progress_(ioptions_.num_levels),
       icmp_(icmp) {
-
-  max_file_size_.reset(new uint64_t[NumberLevels()]);
-  level_max_bytes_.reset(new uint64_t[NumberLevels()]);
-  int target_file_size_multiplier = options_->target_file_size_multiplier;
-  int max_bytes_multiplier = options_->max_bytes_for_level_multiplier;
-  for (int i = 0; i < NumberLevels(); i++) {
-    if (i == 0 && options_->compaction_style == kCompactionStyleUniversal) {
-      max_file_size_[i] = ULLONG_MAX;
-      level_max_bytes_[i] = options_->max_bytes_for_level_base;
-    } else if (i > 1) {
-      max_file_size_[i] = MultiplyCheckOverflow(max_file_size_[i - 1],
-                                                target_file_size_multiplier);
-      level_max_bytes_[i] = MultiplyCheckOverflow(
-          MultiplyCheckOverflow(level_max_bytes_[i - 1], max_bytes_multiplier),
-          options_->max_bytes_for_level_multiplier_additional[i - 1]);
-    } else {
-      max_file_size_[i] = options_->target_file_size_base;
-      level_max_bytes_[i] = options_->max_bytes_for_level_base;
-    }
-  }
 }
 
 CompactionPicker::~CompactionPicker() {}
@@ -123,26 +92,6 @@ void CompactionPicker::ReleaseCompactionFiles(Compaction* c, Status status) {
   }
 }
 
-uint64_t CompactionPicker::MaxFileSizeForLevel(int level) const {
-  assert(level >= 0);
-  assert(level < NumberLevels());
-  return max_file_size_[level];
-}
-
-uint64_t CompactionPicker::MaxGrandParentOverlapBytes(int level) {
-  uint64_t result = MaxFileSizeForLevel(level);
-  result *= options_->max_grandparent_overlap_factor;
-  return result;
-}
-
-double CompactionPicker::MaxBytesForLevel(int level) {
-  // Note: the result for level zero is not really used since we set
-  // the level-0 compaction threshold based on number of files.
-  assert(level >= 0);
-  assert(level < NumberLevels());
-  return level_max_bytes_[level];
-}
-
 void CompactionPicker::GetRange(const std::vector<FileMetaData*>& inputs,
                                 InternalKey* smallest, InternalKey* largest) {
   assert(!inputs.empty());
@@ -173,9 +122,12 @@ void CompactionPicker::GetRange(const std::vector<FileMetaData*>& inputs1,
 }
 
 bool CompactionPicker::ExpandWhileOverlapping(Compaction* c) {
+  assert(c != nullptr);
   // If inputs are empty then there is nothing to expand.
-  if (!c || c->inputs_[0].empty()) {
-    return true;
+  if (c->inputs_[0].empty()) {
+    assert(c->inputs_[1].empty());
+    // This isn't good compaction
+    return false;
   }
 
   // GetOverlappingInputs will always do the right thing for level-0.
@@ -194,24 +146,25 @@ bool CompactionPicker::ExpandWhileOverlapping(Compaction* c) {
   size_t old_size;
   do {
     old_size = c->inputs_[0].size();
-    GetRange(c->inputs_[0], &smallest, &largest);
+    GetRange(c->inputs_[0].files, &smallest, &largest);
     c->inputs_[0].clear();
     c->input_version_->GetOverlappingInputs(
-        level, &smallest, &largest, &c->inputs_[0], hint_index, &hint_index);
+        level, &smallest, &largest, &c->inputs_[0].files,
+        hint_index, &hint_index);
   } while(c->inputs_[0].size() > old_size);
 
   // Get the new range
-  GetRange(c->inputs_[0], &smallest, &largest);
+  GetRange(c->inputs_[0].files, &smallest, &largest);
 
   // If, after the expansion, there are files that are already under
   // compaction, then we must drop/cancel this compaction.
   int parent_index = -1;
   if (c->inputs_[0].empty()) {
-    Log(options_->info_log,
+    Log(ioptions_.info_log,
         "[%s] ExpandWhileOverlapping() failure because zero input files",
         c->column_family_data()->GetName().c_str());
   }
-  if (c->inputs_[0].empty() || FilesInCompaction(c->inputs_[0]) ||
+  if (c->inputs_[0].empty() || FilesInCompaction(c->inputs_[0].files) ||
       (c->level() != c->output_level() &&
        ParentRangeInCompaction(c->input_version_, &smallest, &largest, level,
                                &parent_index))) {
@@ -222,12 +175,6 @@ bool CompactionPicker::ExpandWhileOverlapping(Compaction* c) {
   return true;
 }
 
-uint64_t CompactionPicker::ExpandedCompactionByteSizeLimit(int level) {
-  uint64_t result = MaxFileSizeForLevel(level);
-  result *= options_->expanded_compaction_factor;
-  return result;
-}
-
 // Returns true if any one of specified files are being compacted
 bool CompactionPicker::FilesInCompaction(std::vector<FileMetaData*>& files) {
   for (unsigned int i = 0; i < files.size(); i++) {
@@ -255,7 +202,8 @@ bool CompactionPicker::ParentRangeInCompaction(Version* version,
 // Will also attempt to expand "level" if that doesn't expand "level+1"
 // or cause "level" to include a file for compaction that has an overlapping
 // user-key with another file.
-void CompactionPicker::SetupOtherInputs(Compaction* c) {
+void CompactionPicker::SetupOtherInputs(
+    const MutableCFOptions& mutable_cf_options, Compaction* c) {
   // If inputs are empty, then there is nothing to expand.
   // If both input and output levels are the same, no need to consider
   // files at level "level+1"
@@ -267,16 +215,17 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
   InternalKey smallest, largest;
 
   // Get the range one last time.
-  GetRange(c->inputs_[0], &smallest, &largest);
+  GetRange(c->inputs_[0].files, &smallest, &largest);
 
   // Populate the set of next-level files (inputs_[1]) to include in compaction
-  c->input_version_->GetOverlappingInputs(level + 1, &smallest, &largest,
-                                          &c->inputs_[1], c->parent_index_,
-                                          &c->parent_index_);
+  c->input_version_->GetOverlappingInputs(
+      level + 1, &smallest, &largest,
+      &c->inputs_[1].files, c->parent_index_,
+      &c->parent_index_);
 
   // Get entire range covered by compaction
   InternalKey all_start, all_limit;
-  GetRange(c->inputs_[0], c->inputs_[1], &all_start, &all_limit);
+  GetRange(c->inputs_[0].files, c->inputs_[1].files, &all_start, &all_limit);
 
   // See if we can further grow the number of inputs in "level" without
   // changing the number of "level+1" files we pick up. We also choose NOT
@@ -287,10 +236,10 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
     std::vector<FileMetaData*> expanded0;
     c->input_version_->GetOverlappingInputs(
         level, &all_start, &all_limit, &expanded0, c->base_index_, nullptr);
-    const uint64_t inputs0_size = TotalCompensatedFileSize(c->inputs_[0]);
-    const uint64_t inputs1_size = TotalCompensatedFileSize(c->inputs_[1]);
+    const uint64_t inputs0_size = TotalCompensatedFileSize(c->inputs_[0].files);
+    const uint64_t inputs1_size = TotalCompensatedFileSize(c->inputs_[1].files);
     const uint64_t expanded0_size = TotalCompensatedFileSize(expanded0);
-    uint64_t limit = ExpandedCompactionByteSizeLimit(level);
+    uint64_t limit = mutable_cf_options.ExpandedCompactionByteSizeLimit(level);
     if (expanded0.size() > c->inputs_[0].size() &&
         inputs1_size + expanded0_size < limit &&
         !FilesInCompaction(expanded0) &&
@@ -303,7 +252,7 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
                                               &c->parent_index_);
       if (expanded1.size() == c->inputs_[1].size() &&
           !FilesInCompaction(expanded1)) {
-        Log(options_->info_log,
+        Log(ioptions_.info_log,
             "[%s] Expanding@%d %zu+%zu (%" PRIu64 "+%" PRIu64
             " bytes) to %zu+%zu (%" PRIu64 "+%" PRIu64 "bytes)\n",
             c->column_family_data()->GetName().c_str(), level,
@@ -312,9 +261,10 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
             inputs1_size);
         smallest = new_start;
         largest = new_limit;
-        c->inputs_[0] = expanded0;
-        c->inputs_[1] = expanded1;
-        GetRange(c->inputs_[0], c->inputs_[1], &all_start, &all_limit);
+        c->inputs_[0].files = expanded0;
+        c->inputs_[1].files = expanded1;
+        GetRange(c->inputs_[0].files, c->inputs_[1].files,
+                 &all_start, &all_limit);
       }
     }
   }
@@ -327,21 +277,20 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
   }
 }
 
-
-Compaction* CompactionPicker::CompactRange(Version* version, int input_level,
-                                           int output_level,
-                                           const InternalKey* begin,
-                                           const InternalKey* end,
-                                           InternalKey** compaction_end) {
+Compaction* CompactionPicker::CompactRange(
+    const MutableCFOptions& mutable_cf_options, Version* version,
+    int input_level, int output_level, uint32_t output_path_id,
+    const InternalKey* begin, const InternalKey* end,
+    InternalKey** compaction_end) {
   // CompactionPickerFIFO has its own implementation of compact range
-  assert(options_->compaction_style != kCompactionStyleFIFO);
+  assert(ioptions_.compaction_style != kCompactionStyleFIFO);
 
   std::vector<FileMetaData*> inputs;
   bool covering_the_whole_range = true;
 
   // All files are 'overlapping' in universal style compaction.
   // We have to compact the entire range in one shot.
-  if (options_->compaction_style == kCompactionStyleUniversal) {
+  if (ioptions_.compaction_style == kCompactionStyleUniversal) {
     begin = nullptr;
     end = nullptr;
   }
@@ -355,8 +304,8 @@ Compaction* CompactionPicker::CompactRange(Version* version, int input_level,
   // and we must not pick one file and drop another older file if the
   // two files overlap.
   if (input_level > 0) {
-    const uint64_t limit =
-        MaxFileSizeForLevel(input_level) * options_->source_compaction_factor;
+    const uint64_t limit = mutable_cf_options.MaxFileSizeForLevel(input_level) *
+      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;
@@ -369,21 +318,24 @@ Compaction* CompactionPicker::CompactRange(Version* version, int input_level,
       }
     }
   }
-  Compaction* c = new Compaction(version, input_level, output_level,
-                                 MaxFileSizeForLevel(output_level),
-                                 MaxGrandParentOverlapBytes(input_level), 0,
-                                 GetCompressionType(*options_, output_level));
+  assert(output_path_id < static_cast<uint32_t>(ioptions_.db_paths.size()));
+  Compaction* c = new Compaction(
+      version, input_level, output_level,
+      mutable_cf_options.MaxFileSizeForLevel(output_level),
+      mutable_cf_options.MaxGrandParentOverlapBytes(input_level),
+      output_path_id,
+      GetCompressionType(ioptions_, output_level));
 
-  c->inputs_[0] = inputs;
+  c->inputs_[0].files = inputs;
   if (ExpandWhileOverlapping(c) == false) {
     delete c;
-    Log(options_->info_log,
+    Log(ioptions_.info_log,
         "[%s] Could not compact due to expansion failure.\n",
         version->cfd_->GetName().c_str());
     return nullptr;
   }
 
-  SetupOtherInputs(c);
+  SetupOtherInputs(mutable_cf_options, c);
 
   if (covering_the_whole_range) {
     *compaction_end = nullptr;
@@ -398,12 +350,14 @@ Compaction* CompactionPicker::CompactRange(Version* version, int input_level,
   c->SetupBottomMostLevel(true);
 
   c->is_manual_compaction_ = true;
+  c->mutable_cf_options_ = mutable_cf_options;
 
   return c;
 }
 
-Compaction* LevelCompactionPicker::PickCompaction(Version* version,
-                                                  LogBuffer* log_buffer) {
+Compaction* LevelCompactionPicker::PickCompaction(
+    const MutableCFOptions& mutable_cf_options,
+    Version* version, LogBuffer* log_buffer) {
   Compaction* c = nullptr;
   int level = -1;
 
@@ -411,7 +365,7 @@ Compaction* LevelCompactionPicker::PickCompaction(Version* version,
   // and also in LogAndApply(), otherwise the values could be stale.
   std::vector<uint64_t> size_being_compacted(NumberLevels() - 1);
   SizeBeingCompacted(size_being_compacted);
-  version->ComputeCompactionScore(size_being_compacted);
+  version->ComputeCompactionScore(mutable_cf_options, size_being_compacted);
 
   // We prefer compactions triggered by too much data in a level over
   // the compactions triggered by seeks.
@@ -422,8 +376,9 @@ Compaction* LevelCompactionPicker::PickCompaction(Version* version,
            version->compaction_score_[i] <= version->compaction_score_[i - 1]);
     level = version->compaction_level_[i];
     if ((version->compaction_score_[i] >= 1)) {
-      c = PickCompactionBySize(version, level, version->compaction_score_[i]);
-      if (ExpandWhileOverlapping(c) == false) {
+      c = PickCompactionBySize(mutable_cf_options, version, level,
+          version->compaction_score_[i]);
+      if (c == nullptr || ExpandWhileOverlapping(c) == false) {
         delete c;
         c = nullptr;
       } else {
@@ -441,18 +396,18 @@ Compaction* LevelCompactionPicker::PickCompaction(Version* version,
   if (level == 0) {
     assert(compactions_in_progress_[0].empty());
     InternalKey smallest, largest;
-    GetRange(c->inputs_[0], &smallest, &largest);
+    GetRange(c->inputs_[0].files, &smallest, &largest);
     // Note that the next call will discard the file we placed in
     // c->inputs_[0] earlier and replace it with an overlapping set
     // which will include the picked file.
     c->inputs_[0].clear();
     c->input_version_->GetOverlappingInputs(0, &smallest, &largest,
-                                            &c->inputs_[0]);
+                                            &c->inputs_[0].files);
 
     // If we include more L0 files in the same compaction run it can
     // cause the 'smallest' and 'largest' key to get extended to a
     // larger range. So, re-invoke GetRange to get the new key range
-    GetRange(c->inputs_[0], &smallest, &largest);
+    GetRange(c->inputs_[0].files, &smallest, &largest);
     if (ParentRangeInCompaction(c->input_version_, &smallest, &largest, level,
                                 &c->parent_index_)) {
       delete c;
@@ -462,7 +417,7 @@ Compaction* LevelCompactionPicker::PickCompaction(Version* version,
   }
 
   // Setup "level+1" files (inputs_[1])
-  SetupOtherInputs(c);
+  SetupOtherInputs(mutable_cf_options, c);
 
   // mark all the files that are being compacted
   c->MarkFilesBeingCompacted(true);
@@ -473,12 +428,13 @@ Compaction* LevelCompactionPicker::PickCompaction(Version* version,
   // remember this currently undergoing compaction
   compactions_in_progress_[level].insert(c);
 
+  c->mutable_cf_options_ = mutable_cf_options;
   return c;
 }
 
-Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
-                                                        int level,
-                                                        double score) {
+Compaction* LevelCompactionPicker::PickCompactionBySize(
+    const MutableCFOptions& mutable_cf_options,
+    Version* version, int level, double score) {
   Compaction* c = nullptr;
 
   // level 0 files are overlapping. So we cannot pick more
@@ -491,9 +447,10 @@ Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
 
   assert(level >= 0);
   assert(level + 1 < NumberLevels());
-  c = new Compaction(version, level, level + 1, MaxFileSizeForLevel(level + 1),
-                     MaxGrandParentOverlapBytes(level), 0,
-                     GetCompressionType(*options_, level + 1));
+  c = new Compaction(version, level, level + 1,
+                     mutable_cf_options.MaxFileSizeForLevel(level + 1),
+                     mutable_cf_options.MaxGrandParentOverlapBytes(level), 0,
+                     GetCompressionType(ioptions_, level + 1));
   c->score_ = score;
 
   // Pick the largest file in this level that is not already
@@ -533,7 +490,7 @@ Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
                                 level, &parent_index)) {
       continue;
     }
-    c->inputs_[0].push_back(f);
+    c->inputs_[0].files.push_back(f);
     c->base_index_ = index;
     c->parent_index_ = parent_index;
     break;
@@ -553,35 +510,37 @@ Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
 // Universal style of compaction. Pick files that are contiguous in
 // time-range to compact.
 //
-Compaction* UniversalCompactionPicker::PickCompaction(Version* version,
-                                                      LogBuffer* log_buffer) {
+Compaction* UniversalCompactionPicker::PickCompaction(
+    const MutableCFOptions& mutable_cf_options,
+    Version* version, LogBuffer* log_buffer) {
   int level = 0;
   double score = version->compaction_score_[0];
 
   if ((version->files_[level].size() <
-       (unsigned int)options_->level0_file_num_compaction_trigger)) {
+       (unsigned int)mutable_cf_options.level0_file_num_compaction_trigger)) {
     LogToBuffer(log_buffer, "[%s] Universal: nothing to do\n",
                 version->cfd_->GetName().c_str());
     return nullptr;
   }
   Version::FileSummaryStorage tmp;
-  LogToBuffer(log_buffer, "[%s] Universal: candidate files(%zu): %s\n",
+  LogToBuffer(log_buffer, 3072, "[%s] Universal: candidate files(%zu): %s\n",
               version->cfd_->GetName().c_str(), version->files_[level].size(),
               version->LevelFileSummary(&tmp, 0));
 
   // Check for size amplification first.
   Compaction* c;
-  if ((c = PickCompactionUniversalSizeAmp(version, score, log_buffer)) !=
-      nullptr) {
+  if ((c = PickCompactionUniversalSizeAmp(
+          mutable_cf_options, version, score, log_buffer)) != nullptr) {
     LogToBuffer(log_buffer, "[%s] Universal: compacting for size amp\n",
                 version->cfd_->GetName().c_str());
   } else {
     // Size amplification is within limits. Try reducing read
     // amplification while maintaining file size ratios.
-    unsigned int ratio = options_->compaction_options_universal.size_ratio;
+    unsigned int ratio = ioptions_.compaction_options_universal.size_ratio;
 
-    if ((c = PickCompactionUniversalReadAmp(version, score, ratio, UINT_MAX,
-                                            log_buffer)) != nullptr) {
+    if ((c = PickCompactionUniversalReadAmp(
+            mutable_cf_options, version, score, ratio,
+            UINT_MAX, log_buffer)) != nullptr) {
       LogToBuffer(log_buffer, "[%s] Universal: compacting for size ratio\n",
                   version->cfd_->GetName().c_str());
     } else {
@@ -590,9 +549,10 @@ Compaction* UniversalCompactionPicker::PickCompaction(Version* version,
       // compaction without looking at filesize ratios and try to reduce
       // the number of files to fewer than level0_file_num_compaction_trigger.
       unsigned int num_files = version->files_[level].size() -
-                               options_->level0_file_num_compaction_trigger;
+          mutable_cf_options.level0_file_num_compaction_trigger;
       if ((c = PickCompactionUniversalReadAmp(
-               version, score, UINT_MAX, num_files, log_buffer)) != nullptr) {
+               mutable_cf_options, version, score, UINT_MAX,
+               num_files, log_buffer)) != nullptr) {
         LogToBuffer(log_buffer, "[%s] Universal: compacting for file num\n",
                     version->cfd_->GetName().c_str());
       }
@@ -615,11 +575,11 @@ Compaction* UniversalCompactionPicker::PickCompaction(Version* version,
 
   // Is the earliest file part of this compaction?
   FileMetaData* last_file = c->input_version_->files_[level].back();
-  c->bottommost_level_ = c->inputs_[0].back() == last_file;
+  c->bottommost_level_ = c->inputs_[0].files.back() == last_file;
 
   // update statistics
-  MeasureTime(options_->statistics.get(), NUM_FILES_IN_SINGLE_COMPACTION,
-              c->inputs_[0].size());
+  MeasureTime(ioptions_.statistics,
+              NUM_FILES_IN_SINGLE_COMPACTION, c->inputs_[0].size());
 
   // mark all the files that are being compacted
   c->MarkFilesBeingCompacted(true);
@@ -632,22 +592,55 @@ Compaction* UniversalCompactionPicker::PickCompaction(Version* version,
   c->is_full_compaction_ =
       (c->inputs_[0].size() == c->input_version_->files_[0].size());
 
+  c->mutable_cf_options_ = mutable_cf_options;
   return c;
 }
 
+uint32_t UniversalCompactionPicker::GetPathId(
+    const ImmutableCFOptions& ioptions, uint64_t file_size) {
+  // Two conditions need to be satisfied:
+  // (1) the target path needs to be able to hold the file's size
+  // (2) Total size left in this and previous paths need to be not
+  //     smaller than expected future file size before this new file is
+  //     compacted, which is estimated based on size_ratio.
+  // For example, if now we are compacting files of size (1, 1, 2, 4, 8),
+  // we will make sure the target file, probably with size of 16, will be
+  // placed in a path so that eventually when new files are generated and
+  // compacted to (1, 1, 2, 4, 8, 16), all those files can be stored in or
+  // before the path we chose.
+  //
+  // TODO(sdong): now the case of multiple column families is not
+  // 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;
+  uint32_t p = 0;
+  for (; p < ioptions.db_paths.size() - 1; p++) {
+    uint64_t target_size = ioptions.db_paths[p].target_size;
+    if (target_size > file_size &&
+        accumulated_size + (target_size - file_size) > future_size) {
+      return p;
+    }
+    accumulated_size += target_size;
+  }
+  return p;
+}
+
 //
 // Consider compaction files based on their size differences with
 // the next file in time order.
 //
 Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
-    Version* version, double score, unsigned int ratio,
+    const MutableCFOptions& mutable_cf_options, Version* version,
+    double score, unsigned int ratio,
     unsigned int max_number_of_files_to_compact, LogBuffer* log_buffer) {
   int level = 0;
 
   unsigned int min_merge_width =
-    options_->compaction_options_universal.min_merge_width;
+    ioptions_.compaction_options_universal.min_merge_width;
   unsigned int max_merge_width =
-    options_->compaction_options_universal.max_merge_width;
+    ioptions_.compaction_options_universal.max_merge_width;
 
   // The files are sorted from newest first to oldest last.
   const auto& files = version->files_[level];
@@ -685,10 +678,11 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
     // first candidate to be compacted.
     uint64_t candidate_size =  f != nullptr? f->compensated_file_size : 0;
     if (f != nullptr) {
-      LogToBuffer(
-          log_buffer, "[%s] Universal: Possible candidate file %s[%d].",
-          version->cfd_->GetName().c_str(),
-          FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId()).c_str(), loop);
+      char file_num_buf[kFormatFileNumberBufSize];
+      FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId(), file_num_buf,
+                       sizeof(file_num_buf));
+      LogToBuffer(log_buffer, "[%s] Universal: Possible candidate file %s[%d].",
+                  version->cfd_->GetName().c_str(), file_num_buf, loop);
     }
 
     // Check if the suceeding files need compaction.
@@ -704,15 +698,16 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
       // default kCompactionStopStyleTotalSize; with
       // kCompactionStopStyleSimilarSize, it's simply the size of the last
       // picked file.
-      uint64_t sz = (candidate_size * (100L + ratio)) /100;
-      if (sz < f->fd.GetFileSize()) {
+      double sz = candidate_size * (100.0 + ratio) / 100.0;
+      if (sz < static_cast<double>(f->fd.GetFileSize())) {
         break;
       }
-      if (options_->compaction_options_universal.stop_style == kCompactionStopStyleSimilarSize) {
+      if (ioptions_.compaction_options_universal.stop_style ==
+          kCompactionStopStyleSimilarSize) {
         // Similar-size stopping rule: also check the last picked file isn't
         // far larger than the next candidate file.
-        sz = (f->fd.GetFileSize() * (100L + ratio)) / 100;
-        if (sz < candidate_size) {
+        sz = (f->fd.GetFileSize() * (100.0 + ratio)) / 100.0;
+        if (sz < static_cast<double>(candidate_size)) {
           // If the small file we've encountered begins a run of similar-size
           // files, we'll pick them up on a future iteration of the outer
           // loop. If it's some lonely straggler, it'll eventually get picked
@@ -752,7 +747,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
   // size ratio of compression.
   bool enable_compression = true;
   int ratio_to_compress =
-      options_->compaction_options_universal.compression_size_percent;
+      ioptions_.compaction_options_universal.compression_size_percent;
   if (ratio_to_compress >= 0) {
     uint64_t total_size = version->NumLevelBytes(level);
     uint64_t older_file_size = 0;
@@ -765,20 +760,30 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
       }
     }
   }
+
+  uint64_t estimated_total_size = 0;
+  for (unsigned int i = 0; i < first_index_after; i++) {
+    estimated_total_size += files[i]->fd.GetFileSize();
+  }
+  uint32_t path_id = GetPathId(ioptions_, estimated_total_size);
+
   Compaction* c = new Compaction(
-      version, level, level, MaxFileSizeForLevel(level), LLONG_MAX, 0,
-      GetCompressionType(*options_, level, enable_compression));
+      version, level, level, mutable_cf_options.MaxFileSizeForLevel(level),
+      LLONG_MAX, path_id, GetCompressionType(ioptions_, level,
+      enable_compression));
   c->score_ = score;
 
   for (unsigned int i = start_index; i < first_index_after; i++) {
     FileMetaData* f = c->input_version_->files_[level][i];
-    c->inputs_[0].push_back(f);
+    c->inputs_[0].files.push_back(f);
+    char file_num_buf[kFormatFileNumberBufSize];
+    FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId(), file_num_buf,
+                     sizeof(file_num_buf));
     LogToBuffer(log_buffer,
                 "[%s] Universal: Picking file %s[%d] "
                 "with size %" PRIu64 " (compensated size %" PRIu64 ")\n",
-                version->cfd_->GetName().c_str(),
-                FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId()).c_str(),
-                i, f->fd.GetFileSize(), f->compensated_file_size);
+                version->cfd_->GetName().c_str(), file_num_buf, i,
+                f->fd.GetFileSize(), f->compensated_file_size);
   }
   return c;
 }
@@ -790,11 +795,12 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
 // min_merge_width and max_merge_width).
 //
 Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
-    Version* version, double score, LogBuffer* log_buffer) {
+    const MutableCFOptions& mutable_cf_options, Version* version,
+    double score, LogBuffer* log_buffer) {
   int level = 0;
 
   // percentage flexibilty while reducing size amplification
-  uint64_t ratio = options_->compaction_options_universal.
+  uint64_t ratio = ioptions_.compaction_options_universal.
                      max_size_amplification_percent;
 
   // The files are sorted from newest first to oldest last.
@@ -812,29 +818,35 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
       start_index = loop;         // Consider this as the first candidate.
       break;
     }
+    char file_num_buf[kFormatFileNumberBufSize];
+    FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId(), file_num_buf,
+                     sizeof(file_num_buf));
     LogToBuffer(log_buffer, "[%s] Universal: skipping file %s[%d] compacted %s",
-                version->cfd_->GetName().c_str(),
-                FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId()).c_str(),
-                loop, " cannot be a candidate to reduce size amp.\n");
+                version->cfd_->GetName().c_str(), file_num_buf, loop,
+                " cannot be a candidate to reduce size amp.\n");
     f = nullptr;
   }
   if (f == nullptr) {
     return nullptr;             // no candidate files
   }
 
+  char file_num_buf[kFormatFileNumberBufSize];
+  FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId(), file_num_buf,
+                   sizeof(file_num_buf));
   LogToBuffer(log_buffer, "[%s] Universal: First candidate file %s[%d] %s",
-              version->cfd_->GetName().c_str(),
-              FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId()).c_str(),
-              start_index, " to reduce size amp.\n");
+              version->cfd_->GetName().c_str(), file_num_buf, start_index,
+              " to reduce size amp.\n");
 
   // keep adding up all the remaining files
   for (unsigned int loop = start_index; loop < files.size() - 1; loop++) {
     f = files[loop];
     if (f->being_compacted) {
+      char file_num_buf[kFormatFileNumberBufSize];
+      FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId(), file_num_buf,
+                       sizeof(file_num_buf));
       LogToBuffer(
           log_buffer, "[%s] Universal: Possible candidate file %s[%d] %s.",
-          version->cfd_->GetName().c_str(),
-          FormatFileNumber(f->fd.GetNumber(), f->fd.GetPathId()).c_str(), loop,
+          version->cfd_->GetName().c_str(), file_num_buf, loop,
           " is already being compacted. No size amp reduction possible.\n");
       return nullptr;
     }
@@ -863,17 +875,25 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
         "earliest-file-size %" PRIu64,
         version->cfd_->GetName().c_str(), candidate_size, earliest_file_size);
   }
-  assert(start_index >= 0 && start_index < files.size() - 1);
+  assert(start_index < files.size() - 1);
+
+  // Estimate total file size
+  uint64_t estimated_total_size = 0;
+  for (unsigned int loop = start_index; loop < files.size(); loop++) {
+    estimated_total_size += files[loop]->fd.GetFileSize();
+  }
+  uint32_t path_id = GetPathId(ioptions_, estimated_total_size);
 
   // create a compaction request
   // We always compact all the files, so always compress.
   Compaction* c =
-      new Compaction(version, level, level, MaxFileSizeForLevel(level),
-                     LLONG_MAX, 0, GetCompressionType(*options_, level));
+      new Compaction(version, level, level,
+                     mutable_cf_options.MaxFileSizeForLevel(level),
+                     LLONG_MAX, path_id, GetCompressionType(ioptions_, level));
   c->score_ = score;
   for (unsigned int loop = start_index; loop < files.size(); loop++) {
     f = c->input_version_->files_[level][loop];
-    c->inputs_[0].push_back(f);
+    c->inputs_[0].files.push_back(f);
     LogToBuffer(log_buffer,
         "[%s] Universal: size amp picking file %" PRIu64 "[%d] "
         "with size %" PRIu64 " (compensated size %" PRIu64 ")",
@@ -884,22 +904,23 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
   return c;
 }
 
-Compaction* FIFOCompactionPicker::PickCompaction(Version* version,
-                                                 LogBuffer* log_buffer) {
+Compaction* FIFOCompactionPicker::PickCompaction(
+    const MutableCFOptions& mutable_cf_options,
+    Version* version, LogBuffer* log_buffer) {
   assert(version->NumberLevels() == 1);
   uint64_t total_size = 0;
   for (const auto& file : version->files_[0]) {
     total_size += file->compensated_file_size;
   }
 
-  if (total_size <= options_->compaction_options_fifo.max_table_files_size ||
+  if (total_size <= ioptions_.compaction_options_fifo.max_table_files_size ||
       version->files_[0].size() == 0) {
     // total size not exceeded
     LogToBuffer(log_buffer,
                 "[%s] FIFO compaction: nothing to do. Total size %" PRIu64
                 ", max size %" PRIu64 "\n",
                 version->cfd_->GetName().c_str(), total_size,
-                options_->compaction_options_fifo.max_table_files_size);
+                ioptions_.compaction_options_fifo.max_table_files_size);
     return nullptr;
   }
 
@@ -918,34 +939,37 @@ Compaction* FIFOCompactionPicker::PickCompaction(Version* version,
        ritr != version->files_[0].rend(); ++ritr) {
     auto f = *ritr;
     total_size -= f->compensated_file_size;
-    c->inputs_[0].push_back(f);
+    c->inputs_[0].files.push_back(f);
     char tmp_fsize[16];
     AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize));
     LogToBuffer(log_buffer, "[%s] FIFO compaction: picking file %" PRIu64
                             " with size %s for deletion",
                 version->cfd_->GetName().c_str(), f->fd.GetNumber(), tmp_fsize);
-    if (total_size <= options_->compaction_options_fifo.max_table_files_size) {
+    if (total_size <= ioptions_.compaction_options_fifo.max_table_files_size) {
       break;
     }
   }
 
   c->MarkFilesBeingCompacted(true);
   compactions_in_progress_[0].insert(c);
-
+  c->mutable_cf_options_ = mutable_cf_options;
   return c;
 }
 
-Compaction* FIFOCompactionPicker::CompactRange(Version* version,
-                                               int input_level,
-                                               int output_level,
-                                               const InternalKey* begin,
-                                               const InternalKey* end,
-                                               InternalKey** compaction_end) {
+Compaction* FIFOCompactionPicker::CompactRange(
+    const MutableCFOptions& mutable_cf_options,
+    Version* version, int input_level, int output_level,
+    uint32_t output_path_id, const InternalKey* begin, const InternalKey* end,
+    InternalKey** compaction_end) {
   assert(input_level == 0);
   assert(output_level == 0);
   *compaction_end = nullptr;
-  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, options_->info_log.get());
-  auto c = PickCompaction(version, &log_buffer);
+  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, ioptions_.info_log);
+  Compaction* c = PickCompaction(mutable_cf_options, version, &log_buffer);
+  if (c != nullptr) {
+    assert(output_path_id < static_cast<uint32_t>(ioptions_.db_paths.size()));
+    c->output_path_id_ = output_path_id;
+  }
   log_buffer.FlushBufferToLog();
   return c;
 }

db/compaction_picker.h

@@ -13,6 +13,7 @@
 #include "rocksdb/status.h"
 #include "rocksdb/options.h"
 #include "rocksdb/env.h"
+#include "util/mutable_cf_options.h"
 
 #include <vector>
 #include <memory>
@@ -26,15 +27,17 @@ class Version;
 
 class CompactionPicker {
  public:
-  CompactionPicker(const Options* options, const InternalKeyComparator* icmp);
+  CompactionPicker(const ImmutableCFOptions& ioptions,
+                   const InternalKeyComparator* icmp);
   virtual ~CompactionPicker();
 
   // Pick level and inputs for a new compaction.
   // Returns nullptr if there is no compaction to be done.
   // Otherwise returns a pointer to a heap-allocated object that
   // describes the compaction.  Caller should delete the result.
-  virtual Compaction* PickCompaction(Version* version,
-                                     LogBuffer* log_buffer) = 0;
+  virtual Compaction* PickCompaction(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, LogBuffer* log_buffer) = 0;
 
   // Return a compaction object for compacting the range [begin,end] in
   // the specified level.  Returns nullptr if there is nothing in that
@@ -47,10 +50,15 @@ class CompactionPicker {
   // compaction_end will be set to nullptr.
   // Client is responsible for compaction_end storage -- when called,
   // *compaction_end should point to valid InternalKey!
-  virtual Compaction* CompactRange(Version* version, int input_level,
-                                   int output_level, const InternalKey* begin,
-                                   const InternalKey* end,
-                                   InternalKey** compaction_end);
+  virtual Compaction* CompactRange(
+      const MutableCFOptions& mutable_cf_options, Version* version,
+      int input_level, int output_level, uint32_t output_path_id,
+      const InternalKey* begin, const InternalKey* end,
+      InternalKey** compaction_end);
+
+  // Given the current number of levels, returns the lowest allowed level
+  // for compaction input.
+  virtual int MaxInputLevel(int current_num_levels) const = 0;
 
   // Free up the files that participated in a compaction
   void ReleaseCompactionFiles(Compaction* c, Status status);
@@ -59,19 +67,8 @@ class CompactionPicker {
   // compactions per level
   void SizeBeingCompacted(std::vector<uint64_t>& sizes);
 
-  // Returns maximum total overlap bytes with grandparent
-  // level (i.e., level+2) before we stop building a single
-  // file in level->level+1 compaction.
-  uint64_t MaxGrandParentOverlapBytes(int level);
-
-  // Returns maximum total bytes of data on a given level.
-  double MaxBytesForLevel(int level);
-
-  // Get the max file size in a given level.
-  uint64_t MaxFileSizeForLevel(int level) const;
-
  protected:
-  int NumberLevels() const { return num_levels_; }
+  int NumberLevels() const { return ioptions_.num_levels; }
 
   // Stores the minimal range that covers all entries in inputs in
   // *smallest, *largest.
@@ -98,8 +95,6 @@ class CompactionPicker {
   // Will return false if it is impossible to apply this compaction.
   bool ExpandWhileOverlapping(Compaction* c);
 
-  uint64_t ExpandedCompactionByteSizeLimit(int level);
-
   // Returns true if any one of the specified files are being compacted
   bool FilesInCompaction(std::vector<FileMetaData*>& files);
 
@@ -108,74 +103,98 @@ class CompactionPicker {
                                const InternalKey* largest, int level,
                                int* index);
 
-  void SetupOtherInputs(Compaction* c);
+  void SetupOtherInputs(const MutableCFOptions& mutable_cf_options,
+                        Compaction* c);
+
+  const ImmutableCFOptions& ioptions_;
 
   // record all the ongoing compactions for all levels
   std::vector<std::set<Compaction*>> compactions_in_progress_;
 
-  // Per-level target file size.
-  std::unique_ptr<uint64_t[]> max_file_size_;
-
-  // Per-level max bytes
-  std::unique_ptr<uint64_t[]> level_max_bytes_;
-
-  const Options* const options_;
 
  private:
-  int num_levels_;
-
   const InternalKeyComparator* const icmp_;
 };
 
 class UniversalCompactionPicker : public CompactionPicker {
  public:
-  UniversalCompactionPicker(const Options* options,
+  UniversalCompactionPicker(const ImmutableCFOptions& ioptions,
                             const InternalKeyComparator* icmp)
-      : CompactionPicker(options, icmp) {}
-  virtual Compaction* PickCompaction(Version* version,
-                                     LogBuffer* log_buffer) override;
+      : CompactionPicker(ioptions, icmp) {}
+  virtual Compaction* PickCompaction(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, LogBuffer* log_buffer) override;
+
+  // The maxinum allowed input level.  Always return 0.
+  virtual int MaxInputLevel(int current_num_levels) const override {
+    return 0;
+  }
 
  private:
   // Pick Universal compaction to limit read amplification
-  Compaction* PickCompactionUniversalReadAmp(Version* version, double score,
-                                             unsigned int ratio,
-                                             unsigned int num_files,
-                                             LogBuffer* log_buffer);
+  Compaction* PickCompactionUniversalReadAmp(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, double score, unsigned int ratio,
+      unsigned int num_files, LogBuffer* log_buffer);
 
   // Pick Universal compaction to limit space amplification.
-  Compaction* PickCompactionUniversalSizeAmp(Version* version, double score,
-                                             LogBuffer* log_buffer);
+  Compaction* PickCompactionUniversalSizeAmp(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, double score, LogBuffer* log_buffer);
+
+  // Pick a path ID to place a newly generated file, with its estimated file
+  // size.
+  static uint32_t GetPathId(const ImmutableCFOptions& ioptions,
+                            uint64_t file_size);
 };
 
 class LevelCompactionPicker : public CompactionPicker {
  public:
-  LevelCompactionPicker(const Options* options,
+  LevelCompactionPicker(const ImmutableCFOptions& ioptions,
                         const InternalKeyComparator* icmp)
-      : CompactionPicker(options, icmp) {}
-  virtual Compaction* PickCompaction(Version* version,
-                                     LogBuffer* log_buffer) override;
+      : CompactionPicker(ioptions, icmp) {}
+  virtual Compaction* PickCompaction(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, LogBuffer* log_buffer) override;
+
+  // Returns current_num_levels - 2, meaning the last level cannot be
+  // compaction input level.
+  virtual int MaxInputLevel(int current_num_levels) const override {
+    return current_num_levels - 2;
+  }
 
  private:
   // For the specfied level, pick a compaction.
   // Returns nullptr if there is no compaction to be done.
   // If level is 0 and there is already a compaction on that level, this
   // function will return nullptr.
-  Compaction* PickCompactionBySize(Version* version, int level, double score);
+  Compaction* PickCompactionBySize(const MutableCFOptions& mutable_cf_options,
+      Version* version, int level, double score);
 };
 
 class FIFOCompactionPicker : public CompactionPicker {
  public:
-  FIFOCompactionPicker(const Options* options,
+  FIFOCompactionPicker(const ImmutableCFOptions& ioptions,
                        const InternalKeyComparator* icmp)
-      : CompactionPicker(options, icmp) {}
+      : CompactionPicker(ioptions, icmp) {}
 
-  virtual Compaction* PickCompaction(Version* version,
-                                     LogBuffer* log_buffer) override;
+  virtual Compaction* PickCompaction(
+      const MutableCFOptions& mutable_cf_options,
+      Version* version, LogBuffer* log_buffer) override;
 
-  virtual Compaction* CompactRange(Version* version, int input_level,
-                                   int output_level, const InternalKey* begin,
-                                   const InternalKey* end,
-                                   InternalKey** compaction_end) override;
+  virtual Compaction* CompactRange(
+      const MutableCFOptions& mutable_cf_options, Version* version,
+      int input_level, int output_level, uint32_t output_path_id,
+      const InternalKey* begin, const InternalKey* end,
+      InternalKey** compaction_end) override;
+
+  // The maxinum allowed input level.  Always return 0.
+  virtual int MaxInputLevel(int current_num_levels) const override {
+    return 0;
+  }
 };
 
+// Utility function
+extern uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files);
+
 }  // namespace rocksdb

db/corruption_test.cc

@@ -45,7 +45,9 @@ class CorruptionTest {
 
     db_ = nullptr;
     options_.create_if_missing = true;
-    options_.block_size_deviation = 0; // make unit test pass for now
+    BlockBasedTableOptions table_options;
+    table_options.block_size_deviation = 0;  // make unit test pass for now
+    options_.table_factory.reset(NewBlockBasedTableFactory(table_options));
     Reopen();
     options_.create_if_missing = false;
   }
@@ -60,9 +62,11 @@ class CorruptionTest {
     db_ = nullptr;
     Options opt = (options ? *options : options_);
     opt.env = &env_;
-    opt.block_cache = tiny_cache_;
-    opt.block_size_deviation = 0;
     opt.arena_block_size = 4096;
+    BlockBasedTableOptions table_options;
+    table_options.block_cache = tiny_cache_;
+    table_options.block_size_deviation = 0;
+    opt.table_factory.reset(NewBlockBasedTableFactory(table_options));
     return DB::Open(opt, dbname_, &db_);
   }
 
@@ -127,7 +131,7 @@ class CorruptionTest {
     ASSERT_GE(max_expected, correct);
   }
 
-  void CorruptFile(const std::string fname, int offset, int bytes_to_corrupt) {
+  void CorruptFile(const std::string& fname, int offset, int bytes_to_corrupt) {
     struct stat sbuf;
     if (stat(fname.c_str(), &sbuf) != 0) {
       const char* msg = strerror(errno);
@@ -328,6 +332,9 @@ TEST(CorruptionTest, CorruptedDescriptor) {
 }
 
 TEST(CorruptionTest, CompactionInputError) {
+  Options options;
+  options.max_background_flushes = 0;
+  Reopen(&options);
   Build(10);
   DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
   dbi->TEST_FlushMemTable();
@@ -347,6 +354,7 @@ TEST(CorruptionTest, CompactionInputErrorParanoid) {
   options.paranoid_checks = true;
   options.write_buffer_size = 131072;
   options.max_write_buffer_number = 2;
+  options.max_background_flushes = 0;
   Reopen(&options);
   DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
 

db/cuckoo_table_db_test.cc

@@ -0,0 +1,346 @@
+//  Copyright (c) 2014, 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 "db/db_impl.h"
+#include "rocksdb/db.h"
+#include "rocksdb/env.h"
+#include "table/meta_blocks.h"
+#include "table/cuckoo_table_factory.h"
+#include "table/cuckoo_table_reader.h"
+#include "util/testharness.h"
+#include "util/testutil.h"
+
+namespace rocksdb {
+
+class CuckooTableDBTest {
+ private:
+  std::string dbname_;
+  Env* env_;
+  DB* db_;
+
+ public:
+  CuckooTableDBTest() : env_(Env::Default()) {
+    dbname_ = test::TmpDir() + "/cuckoo_table_db_test";
+    ASSERT_OK(DestroyDB(dbname_, Options()));
+    db_ = nullptr;
+    Reopen();
+  }
+
+  ~CuckooTableDBTest() {
+    delete db_;
+    ASSERT_OK(DestroyDB(dbname_, Options()));
+  }
+
+  Options CurrentOptions() {
+    Options options;
+    options.table_factory.reset(NewCuckooTableFactory());
+    options.memtable_factory.reset(NewHashLinkListRepFactory(4, 0, 3, true));
+    options.allow_mmap_reads = true;
+    options.create_if_missing = true;
+    options.max_mem_compaction_level = 0;
+    return options;
+  }
+
+  DBImpl* dbfull() {
+    return reinterpret_cast<DBImpl*>(db_);
+  }
+
+  // The following util methods are copied from plain_table_db_test.
+  void Reopen(Options* options = nullptr) {
+    delete db_;
+    db_ = nullptr;
+    Options opts;
+    if (options != nullptr) {
+      opts = *options;
+    } else {
+      opts = CurrentOptions();
+      opts.create_if_missing = true;
+    }
+    ASSERT_OK(DB::Open(opts, dbname_, &db_));
+  }
+
+  Status Put(const Slice& k, const Slice& v) {
+    return db_->Put(WriteOptions(), k, v);
+  }
+
+  Status Delete(const std::string& k) {
+    return db_->Delete(WriteOptions(), k);
+  }
+
+  std::string Get(const std::string& k) {
+    ReadOptions options;
+    std::string result;
+    Status s = db_->Get(options, k, &result);
+    if (s.IsNotFound()) {
+      result = "NOT_FOUND";
+    } else if (!s.ok()) {
+      result = s.ToString();
+    }
+    return result;
+  }
+
+  int NumTableFilesAtLevel(int level) {
+    std::string property;
+    ASSERT_TRUE(
+        db_->GetProperty("rocksdb.num-files-at-level" + NumberToString(level),
+                         &property));
+    return atoi(property.c_str());
+  }
+
+  // Return spread of files per level
+  std::string FilesPerLevel() {
+    std::string result;
+    int last_non_zero_offset = 0;
+    for (int level = 0; level < db_->NumberLevels(); level++) {
+      int f = NumTableFilesAtLevel(level);
+      char buf[100];
+      snprintf(buf, sizeof(buf), "%s%d", (level ? "," : ""), f);
+      result += buf;
+      if (f > 0) {
+        last_non_zero_offset = result.size();
+      }
+    }
+    result.resize(last_non_zero_offset);
+    return result;
+  }
+};
+
+TEST(CuckooTableDBTest, Flush) {
+  // Try with empty DB first.
+  ASSERT_TRUE(dbfull() != nullptr);
+  ASSERT_EQ("NOT_FOUND", Get("key2"));
+
+  // Add some values to db.
+  Options options = CurrentOptions();
+  Reopen(&options);
+
+  ASSERT_OK(Put("key1", "v1"));
+  ASSERT_OK(Put("key2", "v2"));
+  ASSERT_OK(Put("key3", "v3"));
+  dbfull()->TEST_FlushMemTable();
+
+  TablePropertiesCollection ptc;
+  reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
+  ASSERT_EQ(1U, ptc.size());
+  ASSERT_EQ(3U, ptc.begin()->second->num_entries);
+  ASSERT_EQ("1", FilesPerLevel());
+
+  ASSERT_EQ("v1", Get("key1"));
+  ASSERT_EQ("v2", Get("key2"));
+  ASSERT_EQ("v3", Get("key3"));
+  ASSERT_EQ("NOT_FOUND", Get("key4"));
+
+  // Now add more keys and flush.
+  ASSERT_OK(Put("key4", "v4"));
+  ASSERT_OK(Put("key5", "v5"));
+  ASSERT_OK(Put("key6", "v6"));
+  dbfull()->TEST_FlushMemTable();
+
+  reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
+  ASSERT_EQ(2U, ptc.size());
+  auto row = ptc.begin();
+  ASSERT_EQ(3U, row->second->num_entries);
+  ASSERT_EQ(3U, (++row)->second->num_entries);
+  ASSERT_EQ("2", FilesPerLevel());
+  ASSERT_EQ("v1", Get("key1"));
+  ASSERT_EQ("v2", Get("key2"));
+  ASSERT_EQ("v3", Get("key3"));
+  ASSERT_EQ("v4", Get("key4"));
+  ASSERT_EQ("v5", Get("key5"));
+  ASSERT_EQ("v6", Get("key6"));
+
+  ASSERT_OK(Delete("key6"));
+  ASSERT_OK(Delete("key5"));
+  ASSERT_OK(Delete("key4"));
+  dbfull()->TEST_FlushMemTable();
+  reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
+  ASSERT_EQ(3U, ptc.size());
+  row = ptc.begin();
+  ASSERT_EQ(3U, row->second->num_entries);
+  ASSERT_EQ(3U, (++row)->second->num_entries);
+  ASSERT_EQ(3U, (++row)->second->num_entries);
+  ASSERT_EQ("3", FilesPerLevel());
+  ASSERT_EQ("v1", Get("key1"));
+  ASSERT_EQ("v2", Get("key2"));
+  ASSERT_EQ("v3", Get("key3"));
+  ASSERT_EQ("NOT_FOUND", Get("key4"));
+  ASSERT_EQ("NOT_FOUND", Get("key5"));
+  ASSERT_EQ("NOT_FOUND", Get("key6"));
+}
+
+TEST(CuckooTableDBTest, FlushWithDuplicateKeys) {
+  Options options = CurrentOptions();
+  Reopen(&options);
+  ASSERT_OK(Put("key1", "v1"));
+  ASSERT_OK(Put("key2", "v2"));
+  ASSERT_OK(Put("key1", "v3"));  // Duplicate
+  dbfull()->TEST_FlushMemTable();
+
+  TablePropertiesCollection ptc;
+  reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
+  ASSERT_EQ(1U, ptc.size());
+  ASSERT_EQ(2U, ptc.begin()->second->num_entries);
+  ASSERT_EQ("1", FilesPerLevel());
+  ASSERT_EQ("v3", Get("key1"));
+  ASSERT_EQ("v2", Get("key2"));
+}
+
+namespace {
+static std::string Key(int i) {
+  char buf[100];
+  snprintf(buf, sizeof(buf), "key_______%06d", i);
+  return std::string(buf);
+}
+static std::string Uint64Key(uint64_t i) {
+  std::string str;
+  str.resize(8);
+  memcpy(&str[0], static_cast<void*>(&i), 8);
+  return str;
+}
+}  // namespace.
+
+TEST(CuckooTableDBTest, Uint64Comparator) {
+  Options options = CurrentOptions();
+  options.comparator = test::Uint64Comparator();
+  Reopen(&options);
+
+  ASSERT_OK(Put(Uint64Key(1), "v1"));
+  ASSERT_OK(Put(Uint64Key(2), "v2"));
+  ASSERT_OK(Put(Uint64Key(3), "v3"));
+  dbfull()->TEST_FlushMemTable();
+
+  ASSERT_EQ("v1", Get(Uint64Key(1)));
+  ASSERT_EQ("v2", Get(Uint64Key(2)));
+  ASSERT_EQ("v3", Get(Uint64Key(3)));
+  ASSERT_EQ("NOT_FOUND", Get(Uint64Key(4)));
+
+  // Add more keys.
+  ASSERT_OK(Delete(Uint64Key(2)));  // Delete.
+  dbfull()->TEST_FlushMemTable();
+  ASSERT_OK(Put(Uint64Key(3), "v0"));  // Update.
+  ASSERT_OK(Put(Uint64Key(4), "v4"));
+  dbfull()->TEST_FlushMemTable();
+  ASSERT_EQ("v1", Get(Uint64Key(1)));
+  ASSERT_EQ("NOT_FOUND", Get(Uint64Key(2)));
+  ASSERT_EQ("v0", Get(Uint64Key(3)));
+  ASSERT_EQ("v4", Get(Uint64Key(4)));
+}
+
+TEST(CuckooTableDBTest, CompactionTrigger) {
+  Options options = CurrentOptions();
+  options.write_buffer_size = 100 << 10;  // 100KB
+  options.level0_file_num_compaction_trigger = 2;
+  Reopen(&options);
+
+  // Write 11 values, each 10016 B
+  for (int idx = 0; idx < 11; ++idx) {
+    ASSERT_OK(Put(Key(idx), std::string(10000, 'a' + idx)));
+  }
+  dbfull()->TEST_WaitForFlushMemTable();
+  ASSERT_EQ("1", FilesPerLevel());
+
+  // Generate one more file in level-0, and should trigger level-0 compaction
+  for (int idx = 11; idx < 22; ++idx) {
+    ASSERT_OK(Put(Key(idx), std::string(10000, 'a' + idx)));
+  }
+  dbfull()->TEST_WaitForFlushMemTable();
+  ASSERT_EQ("2", FilesPerLevel());
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr);
+  ASSERT_EQ("0,2", FilesPerLevel());
+  for (int idx = 0; idx < 22; ++idx) {
+    ASSERT_EQ(std::string(10000, 'a' + idx), Get(Key(idx)));
+  }
+}
+
+TEST(CuckooTableDBTest, CompactionIntoMultipleFiles) {
+  // Create a big L0 file and check it compacts into multiple files in L1.
+  Options options = CurrentOptions();
+  options.write_buffer_size = 270 << 10;
+  // Two SST files should be created, each containing 14 keys.
+  // Number of buckets will be 16. Total size ~156 KB.
+  options.target_file_size_base = 160 << 10;
+  Reopen(&options);
+
+  // Write 28 values, each 10016 B ~ 10KB
+  for (int idx = 0; idx < 28; ++idx) {
+    ASSERT_OK(Put(Key(idx), std::string(10000, 'a' + idx)));
+  }
+  dbfull()->TEST_WaitForFlushMemTable();
+  ASSERT_EQ("1", FilesPerLevel());
+
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr);
+  ASSERT_EQ("0,2", FilesPerLevel());
+  for (int idx = 0; idx < 28; ++idx) {
+    ASSERT_EQ(std::string(10000, 'a' + idx), Get(Key(idx)));
+  }
+}
+
+TEST(CuckooTableDBTest, SameKeyInsertedInTwoDifferentFilesAndCompacted) {
+  // Insert same key twice so that they go to different SST files. Then wait for
+  // compaction and check if the latest value is stored and old value removed.
+  Options options = CurrentOptions();
+  options.write_buffer_size = 100 << 10;  // 100KB
+  options.level0_file_num_compaction_trigger = 2;
+  Reopen(&options);
+
+  // Write 11 values, each 10016 B
+  for (int idx = 0; idx < 11; ++idx) {
+    ASSERT_OK(Put(Key(idx), std::string(10000, 'a')));
+  }
+  dbfull()->TEST_WaitForFlushMemTable();
+  ASSERT_EQ("1", FilesPerLevel());
+
+  // Generate one more file in level-0, and should trigger level-0 compaction
+  for (int idx = 0; idx < 11; ++idx) {
+    ASSERT_OK(Put(Key(idx), std::string(10000, 'a' + idx)));
+  }
+  dbfull()->TEST_WaitForFlushMemTable();
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr);
+
+  ASSERT_EQ("0,1", FilesPerLevel());
+  for (int idx = 0; idx < 11; ++idx) {
+    ASSERT_EQ(std::string(10000, 'a' + idx), Get(Key(idx)));
+  }
+}
+
+TEST(CuckooTableDBTest, AdaptiveTable) {
+  Options options = CurrentOptions();
+
+  // Write some keys using cuckoo table.
+  options.table_factory.reset(NewCuckooTableFactory());
+  Reopen(&options);
+
+  ASSERT_OK(Put("key1", "v1"));
+  ASSERT_OK(Put("key2", "v2"));
+  ASSERT_OK(Put("key3", "v3"));
+  dbfull()->TEST_FlushMemTable();
+
+  // Write some keys using plain table.
+  options.create_if_missing = false;
+  options.table_factory.reset(NewPlainTableFactory());
+  Reopen(&options);
+  ASSERT_OK(Put("key4", "v4"));
+  ASSERT_OK(Put("key1", "v5"));
+  dbfull()->TEST_FlushMemTable();
+
+  // Write some keys using block based table.
+  std::shared_ptr<TableFactory> block_based_factory(
+      NewBlockBasedTableFactory());
+  options.table_factory.reset(NewAdaptiveTableFactory(block_based_factory));
+  Reopen(&options);
+  ASSERT_OK(Put("key5", "v6"));
+  ASSERT_OK(Put("key2", "v7"));
+  dbfull()->TEST_FlushMemTable();
+
+  ASSERT_EQ("v5", Get("key1"));
+  ASSERT_EQ("v7", Get("key2"));
+  ASSERT_EQ("v3", Get("key3"));
+  ASSERT_EQ("v4", Get("key4"));
+  ASSERT_EQ("v6", Get("key5"));
+}
+}  // namespace rocksdb
+
+int main(int argc, char** argv) { return rocksdb::test::RunAllTests(); }

db/db_bench.cc

@@ -7,7 +7,9 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif
 
 #ifndef GFLAGS
 #include <cstdio>
@@ -30,7 +32,6 @@ int main() {
 #include <gflags/gflags.h>
 #include "db/db_impl.h"
 #include "db/version_set.h"
-#include "rocksdb/statistics.h"
 #include "rocksdb/options.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/db.h"
@@ -38,8 +39,8 @@ int main() {
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/write_batch.h"
 #include "rocksdb/slice.h"
+#include "rocksdb/filter_policy.h"
 #include "rocksdb/slice_transform.h"
-#include "rocksdb/statistics.h"
 #include "rocksdb/perf_context.h"
 #include "port/port.h"
 #include "port/stack_trace.h"
@@ -85,7 +86,8 @@ DEFINE_string(benchmarks,
               "xxhash,"
               "compress,"
               "uncompress,"
-              "acquireload,",
+              "acquireload,"
+              "fillseekseq,",
 
               "Comma-separated list of operations to run in the specified order"
               "Actual benchmarks:\n"
@@ -128,6 +130,8 @@ DEFINE_string(benchmarks,
               "\tcrc32c        -- repeated crc32c of 4K of data\n"
               "\txxhash        -- repeated xxHash of 4K of data\n"
               "\tacquireload   -- load N*1000 times\n"
+              "\tfillseekseq   -- write N values in sequential key, then read "
+              "them by seeking to each key\n"
               "Meta operations:\n"
               "\tcompact     -- Compact the entire DB\n"
               "\tstats       -- Print DB stats\n"
@@ -147,6 +151,7 @@ DEFINE_int64(merge_keys, -1,
              "Number of distinct keys to use for MergeRandom and "
              "ReadRandomMergeRandom. "
              "If negative, there will be FLAGS_num keys.");
+DEFINE_int32(num_column_families, 1, "Number of Column Families to use.");
 
 DEFINE_int64(reads, -1, "Number of read operations to do.  "
              "If negative, do FLAGS_num reads.");
@@ -163,6 +168,12 @@ DEFINE_int32(duration, 0, "Time in seconds for the random-ops tests to run."
 
 DEFINE_int32(value_size, 100, "Size of each value");
 
+DEFINE_int32(seekseq_next, 0, "How many times to call Next() after Seek() in "
+             "fillseekseq");
+
+DEFINE_bool(use_uint64_comparator, false, "use Uint64 user comparator");
+
+DEFINE_int64(batch_size, 1, "Batch size");
 
 static bool ValidateKeySize(const char* flagname, int32_t value) {
   return true;
@@ -239,9 +250,14 @@ DEFINE_int32(universal_compression_size_percent, -1,
 DEFINE_int64(cache_size, -1, "Number of bytes to use as a cache of uncompressed"
              "data. Negative means use default settings.");
 
-DEFINE_int32(block_size, rocksdb::Options().block_size,
+DEFINE_int32(block_size, rocksdb::BlockBasedTableOptions().block_size,
              "Number of bytes in a block.");
 
+DEFINE_int32(block_restart_interval,
+             rocksdb::BlockBasedTableOptions().block_restart_interval,
+             "Number of keys between restart points "
+             "for delta encoding of keys.");
+
 DEFINE_int64(compressed_cache_size, -1,
              "Number of bytes to use as a cache of compressed data.");
 
@@ -299,7 +315,7 @@ DEFINE_string(wal_dir, "", "If not empty, use the given dir for WAL");
 
 DEFINE_int32(num_levels, 7, "The total number of levels");
 
-DEFINE_int32(target_file_size_base, 2 * 1048576, "Target file size at level-1");
+DEFINE_int64(target_file_size_base, 2 * 1048576, "Target file size at level-1");
 
 DEFINE_int32(target_file_size_multiplier, 1,
              "A multiplier to compute target level-N file size (N >= 2)");
@@ -504,6 +520,11 @@ DEFINE_int32(prefix_size, 0, "control the prefix size for HashSkipList and "
 DEFINE_int64(keys_per_prefix, 0, "control average number of keys generated "
              "per prefix, 0 means no special handling of the prefix, "
              "i.e. use the prefix comes with the generated random number.");
+DEFINE_bool(enable_io_prio, false, "Lower the background flush/compaction "
+            "threads' IO priority");
+DEFINE_bool(identity_as_first_hash, false, "the first hash function of cuckoo "
+            "table becomes an identity function. This is only valid when key "
+            "is 8 bytes");
 
 enum RepFactory {
   kSkipList,
@@ -538,11 +559,21 @@ DEFINE_string(memtablerep, "skip_list", "");
 DEFINE_int64(hash_bucket_count, 1024 * 1024, "hash bucket count");
 DEFINE_bool(use_plain_table, false, "if use plain table "
             "instead of block-based table format");
-
+DEFINE_bool(use_cuckoo_table, false, "if use cuckoo table format");
+DEFINE_double(cuckoo_hash_ratio, 0.9, "Hash ratio for Cuckoo SST table.");
+DEFINE_bool(use_hash_search, false, "if use kHashSearch "
+            "instead of kBinarySearch. "
+            "This is valid if only we use BlockTable");
+DEFINE_bool(use_block_based_filter, false, "if use kBlockBasedFilter "
+            "instead of kFullFilter for filter block. "
+            "This is valid if only we use BlockTable");
 DEFINE_string(merge_operator, "", "The merge operator to use with the database."
               "If a new merge operator is specified, be sure to use fresh"
               " database The possible merge operators are defined in"
               " utilities/merge_operators.h");
+DEFINE_int32(skip_list_lookahead, 0, "Used with skip_list memtablerep; try "
+             "linear search first for this many steps from the previous "
+             "position");
 
 static const bool FLAGS_soft_rate_limit_dummy __attribute__((unused)) =
     RegisterFlagValidator(&FLAGS_soft_rate_limit, &ValidateRateLimit);
@@ -614,6 +645,14 @@ static void AppendWithSpace(std::string* str, Slice msg) {
   str->append(msg.data(), msg.size());
 }
 
+struct DBWithColumnFamilies {
+  std::vector<ColumnFamilyHandle*> cfh;
+  DB* db;
+  DBWithColumnFamilies() : db(nullptr) {
+    cfh.clear();
+  }
+};
+
 class Stats {
  private:
   int id_;
@@ -677,7 +716,7 @@ class Stats {
   void SetId(int id) { id_ = id; }
   void SetExcludeFromMerge() { exclude_from_merge_ = true; }
 
-  void FinishedOps(DB* db, int64_t num_ops) {
+  void FinishedOps(DBWithColumnFamilies* db_with_cfh, DB* db, int64_t num_ops) {
     if (FLAGS_histogram) {
       double now = FLAGS_env->NowMicros();
       double micros = now - last_op_finish_;
@@ -717,8 +756,17 @@ class Stats {
 
         if (FLAGS_stats_per_interval) {
           std::string stats;
-          if (db && db->GetProperty("rocksdb.stats", &stats))
+
+          if (db_with_cfh && db_with_cfh->cfh.size()) {
+            for (size_t i = 0; i < db_with_cfh->cfh.size(); ++i) {
+              if (db->GetProperty(db_with_cfh->cfh[i], "rocksdb.cfstats",
+                                  &stats))
+                fprintf(stderr, "%s\n", stats.c_str());
+            }
+
+          } else if (db && db->GetProperty("rocksdb.stats", &stats)) {
             fprintf(stderr, "%s\n", stats.c_str());
+          }
         }
 
         fflush(stderr);
@@ -833,12 +881,12 @@ class Duration {
 
 class Benchmark {
  private:
-  shared_ptr<Cache> cache_;
-  shared_ptr<Cache> compressed_cache_;
-  const FilterPolicy* filter_policy_;
+  std::shared_ptr<Cache> cache_;
+  std::shared_ptr<Cache> compressed_cache_;
+  std::shared_ptr<const FilterPolicy> filter_policy_;
   const SliceTransform* prefix_extractor_;
-  DB* db_;
-  std::vector<DB*> multi_dbs_;
+  DBWithColumnFamilies db_;
+  std::vector<DBWithColumnFamilies> multi_dbs_;
   int64_t num_;
   int value_size_;
   int key_size_;
@@ -1054,11 +1102,10 @@ class Benchmark {
            (FLAGS_cache_numshardbits >= 1 ?
             NewLRUCache(FLAGS_compressed_cache_size, FLAGS_cache_numshardbits) :
             NewLRUCache(FLAGS_compressed_cache_size)) : nullptr),
-    filter_policy_(FLAGS_bloom_bits >= 0
-                   ? NewBloomFilterPolicy(FLAGS_bloom_bits)
-                   : nullptr),
+    filter_policy_(FLAGS_bloom_bits >= 0 ?
+        NewBloomFilterPolicy(FLAGS_bloom_bits, FLAGS_use_block_based_filter)
+        : nullptr),
     prefix_extractor_(NewFixedPrefixTransform(FLAGS_prefix_size)),
-    db_(nullptr),
     num_(FLAGS_num),
     value_size_(FLAGS_value_size),
     key_size_(FLAGS_key_size),
@@ -1089,8 +1136,9 @@ class Benchmark {
   }
 
   ~Benchmark() {
-    delete db_;
-    delete filter_policy_;
+    std::for_each(db_.cfh.begin(), db_.cfh.end(),
+                  [](ColumnFamilyHandle* cfh) { delete cfh; });
+    delete db_.db;
     delete prefix_extractor_;
   }
 
@@ -1149,6 +1197,16 @@ class Benchmark {
     return base_name + std::to_string(id);
   }
 
+  std::string ColumnFamilyName(int i) {
+    if (i == 0) {
+      return kDefaultColumnFamilyName;
+    } else {
+      char name[100];
+      snprintf(name, sizeof(name), "column_family_name_%06d", i);
+      return std::string(name);
+    }
+  }
+
   void Run() {
     if (!SanityCheck()) {
       exit(1);
@@ -1225,7 +1283,12 @@ class Benchmark {
         method = &Benchmark::ReadReverse;
       } else if (name == Slice("readrandom")) {
         method = &Benchmark::ReadRandom;
+      } else if (name == Slice("readrandomfast")) {
+        method = &Benchmark::ReadRandomFast;
       } else if (name == Slice("multireadrandom")) {
+        entries_per_batch_ = FLAGS_batch_size;
+        fprintf(stderr, "entries_per_batch = %" PRIi64 "\n",
+                entries_per_batch_);
         method = &Benchmark::MultiReadRandom;
       } else if (name == Slice("readmissing")) {
         ++key_size_;
@@ -1272,6 +1335,8 @@ class Benchmark {
         method = &Benchmark::MergeRandom;
       } else if (name == Slice("randomwithverify")) {
         method = &Benchmark::RandomWithVerify;
+      } else if (name == Slice("fillseekseq")) {
+        method = &Benchmark::WriteSeqSeekSeq;
       } else if (name == Slice("compact")) {
         method = &Benchmark::Compact;
       } else if (name == Slice("crc32c")) {
@@ -1303,13 +1368,16 @@ class Benchmark {
                   name.ToString().c_str());
           method = nullptr;
         } else {
-          if (db_ != nullptr) {
-            delete db_;
-            db_ = nullptr;
+          if (db_.db != nullptr) {
+            std::for_each(db_.cfh.begin(), db_.cfh.end(),
+                          [](ColumnFamilyHandle* cfh) { delete cfh; });
+            delete db_.db;
+            db_.db = nullptr;
+            db_.cfh.clear();
             DestroyDB(FLAGS_db, Options());
           }
           for (size_t i = 0; i < multi_dbs_.size(); i++) {
-            delete multi_dbs_[i];
+            delete multi_dbs_[i].db;
             DestroyDB(GetDbNameForMultiple(FLAGS_db, i), Options());
           }
           multi_dbs_.clear();
@@ -1434,7 +1502,7 @@ class Benchmark {
     uint32_t crc = 0;
     while (bytes < 500 * 1048576) {
       crc = crc32c::Value(data.data(), size);
-      thread->stats.FinishedOps(nullptr, 1);
+      thread->stats.FinishedOps(nullptr, nullptr, 1);
       bytes += size;
     }
     // Print so result is not dead
@@ -1453,7 +1521,7 @@ class Benchmark {
     unsigned int xxh32 = 0;
     while (bytes < 500 * 1048576) {
       xxh32 = XXH32(data.data(), size, 0);
-      thread->stats.FinishedOps(nullptr, 1);
+      thread->stats.FinishedOps(nullptr, nullptr, 1);
       bytes += size;
     }
     // Print so result is not dead
@@ -1474,14 +1542,14 @@ class Benchmark {
         ptr = ap.Acquire_Load();
       }
       count++;
-      thread->stats.FinishedOps(nullptr, 1);
+      thread->stats.FinishedOps(nullptr, nullptr, 1);
     }
     if (ptr == nullptr) exit(1); // Disable unused variable warning.
   }
 
   void Compress(ThreadState *thread) {
     RandomGenerator gen;
-    Slice input = gen.Generate(Options().block_size);
+    Slice input = gen.Generate(FLAGS_block_size);
     int64_t bytes = 0;
     int64_t produced = 0;
     bool ok = true;
@@ -1515,7 +1583,7 @@ class Benchmark {
       }
       produced += compressed.size();
       bytes += input.size();
-      thread->stats.FinishedOps(nullptr, 1);
+      thread->stats.FinishedOps(nullptr, nullptr, 1);
     }
 
     if (!ok) {
@@ -1531,7 +1599,7 @@ class Benchmark {
 
   void Uncompress(ThreadState *thread) {
     RandomGenerator gen;
-    Slice input = gen.Generate(Options().block_size);
+    Slice input = gen.Generate(FLAGS_block_size);
     std::string compressed;
 
     bool ok;
@@ -1596,7 +1664,7 @@ class Benchmark {
       }
       delete[] uncompressed;
       bytes += input.size();
-      thread->stats.FinishedOps(nullptr, 1);
+      thread->stats.FinishedOps(nullptr, nullptr, 1);
     }
 
     if (!ok) {
@@ -1607,14 +1675,10 @@ class Benchmark {
   }
 
   void Open() {
-    assert(db_ == nullptr);
+    assert(db_.db == nullptr);
     Options options;
     options.create_if_missing = !FLAGS_use_existing_db;
-    options.block_cache = cache_;
-    options.block_cache_compressed = compressed_cache_;
-    if (cache_ == nullptr) {
-      options.no_block_cache = true;
-    }
+    options.create_missing_column_families = FLAGS_num_column_families > 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 =
@@ -1622,16 +1686,25 @@ class Benchmark {
     options.max_background_compactions = FLAGS_max_background_compactions;
     options.max_background_flushes = FLAGS_max_background_flushes;
     options.compaction_style = FLAGS_compaction_style_e;
-    options.block_size = FLAGS_block_size;
-    options.filter_policy = filter_policy_;
-    if (FLAGS_use_plain_table) {
+    if (FLAGS_prefix_size != 0) {
       options.prefix_extractor.reset(
           NewFixedPrefixTransform(FLAGS_prefix_size));
     }
+    if (FLAGS_use_uint64_comparator) {
+      options.comparator = test::Uint64Comparator();
+      if (FLAGS_key_size != 8) {
+        fprintf(stderr, "Using Uint64 comparator but key size is not 8.\n");
+        exit(1);
+      }
+    }
     options.memtable_prefix_bloom_bits = FLAGS_memtable_bloom_bits;
     options.bloom_locality = FLAGS_bloom_locality;
     options.max_open_files = FLAGS_open_files;
     options.statistics = dbstats;
+    if (FLAGS_enable_io_prio) {
+      FLAGS_env->LowerThreadPoolIOPriority(Env::LOW);
+      FLAGS_env->LowerThreadPoolIOPriority(Env::HIGH);
+    }
     options.env = FLAGS_env;
     options.disableDataSync = FLAGS_disable_data_sync;
     options.use_fsync = FLAGS_use_fsync;
@@ -1655,7 +1728,8 @@ class Benchmark {
             FLAGS_hash_bucket_count));
         break;
       case kSkipList:
-        // no need to do anything
+        options.memtable_factory.reset(new SkipListFactory(
+            FLAGS_skip_list_lookahead));
         break;
       case kHashLinkedList:
         options.memtable_factory.reset(NewHashLinkListRepFactory(
@@ -1685,8 +1759,45 @@ class Benchmark {
       if (bloom_bits_per_key < 0) {
         bloom_bits_per_key = 0;
       }
+
+      PlainTableOptions plain_table_options;
+      plain_table_options.user_key_len = FLAGS_key_size;
+      plain_table_options.bloom_bits_per_key = bloom_bits_per_key;
+      plain_table_options.hash_table_ratio = 0.75;
       options.table_factory = std::shared_ptr<TableFactory>(
-          NewPlainTableFactory(FLAGS_key_size, bloom_bits_per_key, 0.75));
+          NewPlainTableFactory(plain_table_options));
+    } else if (FLAGS_use_cuckoo_table) {
+      if (FLAGS_cuckoo_hash_ratio > 1 || FLAGS_cuckoo_hash_ratio < 0) {
+        fprintf(stderr, "Invalid cuckoo_hash_ratio\n");
+        exit(1);
+      }
+      rocksdb::CuckooTableOptions table_options;
+      table_options.hash_table_ratio = FLAGS_cuckoo_hash_ratio;
+      table_options.identity_as_first_hash = FLAGS_identity_as_first_hash;
+      options.table_factory = std::shared_ptr<TableFactory>(
+          NewCuckooTableFactory(table_options));
+    } else {
+      BlockBasedTableOptions block_based_options;
+      if (FLAGS_use_hash_search) {
+        if (FLAGS_prefix_size == 0) {
+          fprintf(stderr,
+              "prefix_size not assigned when enable use_hash_search \n");
+          exit(1);
+        }
+        block_based_options.index_type = BlockBasedTableOptions::kHashSearch;
+      } else {
+        block_based_options.index_type = BlockBasedTableOptions::kBinarySearch;
+      }
+      if (cache_ == nullptr) {
+        block_based_options.no_block_cache = true;
+      }
+      block_based_options.block_cache = cache_;
+      block_based_options.block_cache_compressed = compressed_cache_;
+      block_based_options.block_size = FLAGS_block_size;
+      block_based_options.block_restart_interval = FLAGS_block_restart_interval;
+      block_based_options.filter_policy = filter_policy_;
+      options.table_factory.reset(
+          NewBlockBasedTableFactory(block_based_options));
     }
     if (FLAGS_max_bytes_for_level_multiplier_additional_v.size() > 0) {
       if (FLAGS_max_bytes_for_level_multiplier_additional_v.size() !=
@@ -1775,10 +1886,9 @@ class Benchmark {
       OpenDb(options, FLAGS_db, &db_);
     } else {
       multi_dbs_.clear();
+      multi_dbs_.resize(FLAGS_num_multi_db);
       for (int i = 0; i < FLAGS_num_multi_db; i++) {
-        DB* db;
-        OpenDb(options, GetDbNameForMultiple(FLAGS_db, i), &db);
-        multi_dbs_.push_back(db);
+        OpenDb(options, GetDbNameForMultiple(FLAGS_db, i), &multi_dbs_[i]);
       }
     }
     if (FLAGS_min_level_to_compress >= 0) {
@@ -1786,12 +1896,27 @@ class Benchmark {
     }
   }
 
-  void OpenDb(Options options, std::string db_name, DB** db) {
+  void OpenDb(const Options& options, const std::string& db_name,
+      DBWithColumnFamilies* db) {
     Status s;
-    if(FLAGS_readonly) {
-      s = DB::OpenForReadOnly(options, db_name, db);
+    // Open with column families if necessary.
+    if (FLAGS_num_column_families > 1) {
+      db->cfh.resize(FLAGS_num_column_families);
+      std::vector<ColumnFamilyDescriptor> column_families;
+      for (int i = 0; i < FLAGS_num_column_families; i++) {
+        column_families.push_back(ColumnFamilyDescriptor(
+              ColumnFamilyName(i), ColumnFamilyOptions(options)));
+      }
+      if (FLAGS_readonly) {
+        s = DB::OpenForReadOnly(options, db_name, column_families,
+            &db->cfh, &db->db);
+      } else {
+        s = DB::Open(options, db_name, column_families, &db->cfh, &db->db);
+      }
+    } else if (FLAGS_readonly) {
+      s = DB::OpenForReadOnly(options, db_name, &db->db);
     } else {
-      s = DB::Open(options, db_name, db);
+      s = DB::Open(options, db_name, &db->db);
     }
     if (!s.ok()) {
       fprintf(stderr, "open error: %s\n", s.ToString().c_str());
@@ -1859,10 +1984,18 @@ class Benchmark {
   };
 
   DB* SelectDB(ThreadState* thread) {
-    if (db_ != nullptr) {
-      return db_;
-    } else {
-      return multi_dbs_[thread->rand.Next() % multi_dbs_.size()];
+    return SelectDBWithCfh(thread)->db;
+  }
+
+  DBWithColumnFamilies* SelectDBWithCfh(ThreadState* thread) {
+    return SelectDBWithCfh(thread->rand.Next());
+  }
+
+  DBWithColumnFamilies* SelectDBWithCfh(uint64_t rand_int) {
+    if (db_.db != nullptr) {
+      return &db_;
+    } else  {
+      return &multi_dbs_[rand_int % multi_dbs_.size()];
     }
   }
 
@@ -1871,7 +2004,7 @@ class Benchmark {
     const int64_t num_ops = writes_ == 0 ? num_ : writes_;
 
     size_t num_key_gens = 1;
-    if (db_ == nullptr) {
+    if (db_.db == nullptr) {
       num_key_gens = multi_dbs_.size();
     }
     std::vector<std::unique_ptr<KeyGenerator>> key_gens(num_key_gens);
@@ -1894,20 +2027,26 @@ class Benchmark {
     Slice key = AllocateKey();
     std::unique_ptr<const char[]> key_guard(key.data());
     while (!duration.Done(entries_per_batch_)) {
-      size_t id = 0;
-      DB* db_to_write = db_;
-      if (db_to_write == nullptr) {
-        id = thread->rand.Next() % num_key_gens;
-        db_to_write = multi_dbs_[id];
-      }
+      size_t id = thread->rand.Next() % num_key_gens;
+      DBWithColumnFamilies* db_with_cfh = SelectDBWithCfh(id);
       batch.Clear();
       for (int64_t j = 0; j < entries_per_batch_; j++) {
-        GenerateKeyFromInt(key_gens[id]->Next(), FLAGS_num, &key);
-        batch.Put(key, gen.Generate(value_size_));
+        int64_t rand_num = key_gens[id]->Next();
+        GenerateKeyFromInt(rand_num, FLAGS_num, &key);
+        if (FLAGS_num_column_families <= 1) {
+          batch.Put(key, gen.Generate(value_size_));
+        } else {
+          // We use same rand_num as seed for key and column family so that we
+          // can deterministically find the cfh corresponding to a particular
+          // key while reading the key.
+          batch.Put(db_with_cfh->cfh[rand_num % db_with_cfh->cfh.size()],
+              key, gen.Generate(value_size_));
+        }
         bytes += value_size_ + key_size_;
       }
-      s = db_to_write->Write(write_options_, &batch);
-      thread->stats.FinishedOps(db_to_write, entries_per_batch_);
+      s = db_with_cfh->db->Write(write_options_, &batch);
+      thread->stats.FinishedOps(db_with_cfh, db_with_cfh->db,
+                                entries_per_batch_);
       if (!s.ok()) {
         fprintf(stderr, "put error: %s\n", s.ToString().c_str());
         exit(1);
@@ -1917,11 +2056,11 @@ class Benchmark {
   }
 
   void ReadSequential(ThreadState* thread) {
-    if (db_ != nullptr) {
-      ReadSequential(thread, db_);
+    if (db_.db != nullptr) {
+      ReadSequential(thread, db_.db);
     } else {
-      for (DB* db : multi_dbs_) {
-        ReadSequential(thread, db);
+      for (const auto& db_with_cfh : multi_dbs_) {
+        ReadSequential(thread, db_with_cfh.db);
       }
     }
   }
@@ -1932,7 +2071,7 @@ class Benchmark {
     int64_t bytes = 0;
     for (iter->SeekToFirst(); i < reads_ && iter->Valid(); iter->Next()) {
       bytes += iter->key().size() + iter->value().size();
-      thread->stats.FinishedOps(db, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
       ++i;
     }
     delete iter;
@@ -1940,11 +2079,11 @@ class Benchmark {
   }
 
   void ReadReverse(ThreadState* thread) {
-    if (db_ != nullptr) {
-      ReadReverse(thread, db_);
+    if (db_.db != nullptr) {
+      ReadReverse(thread, db_.db);
     } else {
-      for (DB* db : multi_dbs_) {
-        ReadReverse(thread, db);
+      for (const auto& db_with_cfh : multi_dbs_) {
+        ReadReverse(thread, db_with_cfh.db);
       }
     }
   }
@@ -1955,13 +2094,56 @@ class Benchmark {
     int64_t bytes = 0;
     for (iter->SeekToLast(); i < reads_ && iter->Valid(); iter->Prev()) {
       bytes += iter->key().size() + iter->value().size();
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
       ++i;
     }
     delete iter;
     thread->stats.AddBytes(bytes);
   }
 
+  void ReadRandomFast(ThreadState* thread) {
+    int64_t read = 0;
+    int64_t found = 0;
+    int64_t nonexist = 0;
+    ReadOptions options(FLAGS_verify_checksum, true);
+    Slice key = AllocateKey();
+    std::unique_ptr<const char[]> key_guard(key.data());
+    std::string value;
+    DB* db = SelectDBWithCfh(thread)->db;
+
+    int64_t pot = 1;
+    while (pot < FLAGS_num) {
+      pot <<= 1;
+    }
+
+    Duration duration(FLAGS_duration, reads_);
+    do {
+      for (int i = 0; i < 100; ++i) {
+        int64_t key_rand = thread->rand.Next() & (pot - 1);
+        GenerateKeyFromInt(key_rand, FLAGS_num, &key);
+        ++read;
+        if (db->Get(options, key, &value).ok()) {
+          ++found;
+        }
+        if (key_rand >= FLAGS_num) {
+          ++nonexist;
+        }
+      }
+      thread->stats.FinishedOps(nullptr, db, 100);
+    } while (!duration.Done(100));
+
+    char msg[100];
+    snprintf(msg, sizeof(msg), "(%" PRIu64 " of %" PRIu64 " found, "
+             "issued %" PRIu64 " non-exist keys)\n",
+             found, read, nonexist);
+
+    thread->stats.AddMessage(msg);
+
+    if (FLAGS_perf_level > 0) {
+      thread->stats.AddMessage(perf_context.ToString());
+    }
+  }
+
   void ReadRandom(ThreadState* thread) {
     int64_t read = 0;
     int64_t found = 0;
@@ -1972,13 +2154,24 @@ class Benchmark {
 
     Duration duration(FLAGS_duration, reads_);
     while (!duration.Done(1)) {
-      DB* db = SelectDB(thread);
-      GenerateKeyFromInt(thread->rand.Next() % FLAGS_num, FLAGS_num, &key);
+      DBWithColumnFamilies* db_with_cfh = SelectDBWithCfh(thread);
+      // We use same key_rand as seed for key and column family so that we can
+      // deterministically find the cfh corresponding to a particular key, as it
+      // is done in DoWrite method.
+      int64_t key_rand = thread->rand.Next() % FLAGS_num;
+      GenerateKeyFromInt(key_rand, FLAGS_num, &key);
       read++;
-      if (db->Get(options, key, &value).ok()) {
+      Status s;
+      if (FLAGS_num_column_families > 1) {
+        s = db_with_cfh->db->Get(options,
+            db_with_cfh->cfh[key_rand % db_with_cfh->cfh.size()], key, &value);
+      } else {
+        s = db_with_cfh->db->Get(options, key, &value);
+      }
+      if (s.ok()) {
         found++;
       }
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(db_with_cfh, db_with_cfh->db, 1);
     }
 
     char msg[100];
@@ -2020,6 +2213,7 @@ class Benchmark {
           ++found;
         }
       }
+      thread->stats.FinishedOps(nullptr, db, entries_per_batch_);
     }
     for (auto& k : keys) {
       delete k.data();
@@ -2038,7 +2232,7 @@ class Benchmark {
       DB* db = SelectDB(thread);
       Iterator* iter = db->NewIterator(options);
       delete iter;
-      thread->stats.FinishedOps(db, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
   }
 
@@ -2058,11 +2252,11 @@ class Benchmark {
 
     Iterator* single_iter = nullptr;
     std::vector<Iterator*> multi_iters;
-    if (db_ != nullptr) {
-      single_iter = db_->NewIterator(options);
+    if (db_.db != nullptr) {
+      single_iter = db_.db->NewIterator(options);
     } else {
-      for (DB* db : multi_dbs_) {
-        multi_iters.push_back(db->NewIterator(options));
+      for (const auto& db_with_cfh : multi_dbs_) {
+        multi_iters.push_back(db_with_cfh.db->NewIterator(options));
       }
     }
     uint64_t last_refresh = FLAGS_env->NowMicros();
@@ -2075,16 +2269,16 @@ class Benchmark {
       if (!FLAGS_use_tailing_iterator && FLAGS_iter_refresh_interval_us >= 0) {
         uint64_t now = FLAGS_env->NowMicros();
         if (now - last_refresh > (uint64_t)FLAGS_iter_refresh_interval_us) {
-          if (db_ != nullptr) {
+          if (db_.db != nullptr) {
             delete single_iter;
-            single_iter = db_->NewIterator(options);
+            single_iter = db_.db->NewIterator(options);
           } else {
             for (auto iter : multi_iters) {
               delete iter;
             }
             multi_iters.clear();
-            for (DB* db : multi_dbs_) {
-              multi_iters.push_back(db->NewIterator(options));
+            for (const auto& db_with_cfh : multi_dbs_) {
+              multi_iters.push_back(db_with_cfh.db->NewIterator(options));
             }
           }
         }
@@ -2102,7 +2296,7 @@ class Benchmark {
       if (iter_to_use->Valid() && iter_to_use->key().compare(key) == 0) {
         found++;
       }
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(&db_, db_.db, 1);
     }
     delete single_iter;
     for (auto iter : multi_iters) {
@@ -2142,7 +2336,7 @@ class Benchmark {
         batch.Delete(key);
       }
       auto s = db->Write(write_options_, &batch);
-      thread->stats.FinishedOps(db, entries_per_batch_);
+      thread->stats.FinishedOps(nullptr, db, entries_per_batch_);
       if (!s.ok()) {
         fprintf(stderr, "del error: %s\n", s.ToString().c_str());
         exit(1);
@@ -2202,7 +2396,7 @@ class Benchmark {
         fprintf(stderr, "put error: %s\n", s.ToString().c_str());
         exit(1);
       }
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(&db_, db_.db, 1);
 
       ++num_writes;
       if (writes_per_second_by_10 && num_writes >= writes_per_second_by_10) {
@@ -2362,7 +2556,7 @@ class Benchmark {
         deletes_done++;
       }
 
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(&db_, db_.db, 1);
     }
     char msg[100];
     snprintf(msg, sizeof(msg),
@@ -2420,7 +2614,7 @@ class Benchmark {
         put_weight--;
         writes_done++;
       }
-      thread->stats.FinishedOps(db, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
     char msg[100];
     snprintf(msg, sizeof(msg), "( reads:%" PRIu64 " writes:%" PRIu64 \
@@ -2454,7 +2648,7 @@ class Benchmark {
         fprintf(stderr, "put error: %s\n", s.ToString().c_str());
         exit(1);
       }
-      thread->stats.FinishedOps(db, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
     char msg[100];
     snprintf(msg, sizeof(msg),
@@ -2501,7 +2695,7 @@ class Benchmark {
         fprintf(stderr, "put error: %s\n", s.ToString().c_str());
         exit(1);
       }
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
 
     char msg[100];
@@ -2537,7 +2731,7 @@ class Benchmark {
         fprintf(stderr, "merge error: %s\n", s.ToString().c_str());
         exit(1);
       }
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
 
     // Print some statistics
@@ -2598,7 +2792,7 @@ class Benchmark {
 
       }
 
-      thread->stats.FinishedOps(db_, 1);
+      thread->stats.FinishedOps(nullptr, db, 1);
     }
 
     char msg[100];
@@ -2609,17 +2803,47 @@ class Benchmark {
     thread->stats.AddMessage(msg);
   }
 
+  void WriteSeqSeekSeq(ThreadState* thread) {
+    writes_ = FLAGS_num;
+    DoWrite(thread, SEQUENTIAL);
+    // exclude writes from the ops/sec calculation
+    thread->stats.Start(thread->tid);
+
+    DB* db = SelectDB(thread);
+    std::unique_ptr<Iterator> iter(
+      db->NewIterator(ReadOptions(FLAGS_verify_checksum, true)));
+
+    Slice key = AllocateKey();
+    for (int64_t i = 0; i < FLAGS_num; ++i) {
+      GenerateKeyFromInt(i, FLAGS_num, &key);
+      iter->Seek(key);
+      assert(iter->Valid() && iter->key() == key);
+      thread->stats.FinishedOps(nullptr, db, 1);
+
+      for (int j = 0; j < FLAGS_seekseq_next && i+1 < FLAGS_num; ++j) {
+        iter->Next();
+        GenerateKeyFromInt(++i, FLAGS_num, &key);
+        assert(iter->Valid() && iter->key() == key);
+        thread->stats.FinishedOps(nullptr, db, 1);
+      }
+
+      iter->Seek(key);
+      assert(iter->Valid() && iter->key() == key);
+      thread->stats.FinishedOps(nullptr, db, 1);
+    }
+  }
+
   void Compact(ThreadState* thread) {
     DB* db = SelectDB(thread);
     db->CompactRange(nullptr, nullptr);
   }
 
   void PrintStats(const char* key) {
-    if (db_ != nullptr) {
-      PrintStats(db_, key, false);
+    if (db_.db != nullptr) {
+      PrintStats(db_.db, key, false);
     }
-    for (DB* db : multi_dbs_) {
-      PrintStats(db, key, true);
+    for (const auto& db_with_cfh : multi_dbs_) {
+      PrintStats(db_with_cfh.db, key, true);
     }
   }
 

db/db_filesnapshot.cc

@@ -9,7 +9,10 @@
 
 #ifndef ROCKSDB_LITE
 
+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif
+
 #include <inttypes.h>
 #include <algorithm>
 #include <string>
@@ -29,9 +32,9 @@ Status DBImpl::DisableFileDeletions() {
   MutexLock l(&mutex_);
   ++disable_delete_obsolete_files_;
   if (disable_delete_obsolete_files_ == 1) {
-    Log(options_.info_log, "File Deletions Disabled");
+    Log(db_options_.info_log, "File Deletions Disabled");
   } else {
-    Log(options_.info_log,
+    Log(db_options_.info_log,
         "File Deletions Disabled, but already disabled. Counter: %d",
         disable_delete_obsolete_files_);
   }
@@ -50,11 +53,11 @@ Status DBImpl::EnableFileDeletions(bool force) {
       --disable_delete_obsolete_files_;
     }
     if (disable_delete_obsolete_files_ == 0)  {
-      Log(options_.info_log, "File Deletions Enabled");
+      Log(db_options_.info_log, "File Deletions Enabled");
       should_purge_files = true;
       FindObsoleteFiles(deletion_state, true);
     } else {
-      Log(options_.info_log,
+      Log(db_options_.info_log,
           "File Deletions Enable, but not really enabled. Counter: %d",
           disable_delete_obsolete_files_);
     }
@@ -62,10 +65,14 @@ Status DBImpl::EnableFileDeletions(bool force) {
   if (should_purge_files)  {
     PurgeObsoleteFiles(deletion_state);
   }
-  LogFlush(options_.info_log);
+  LogFlush(db_options_.info_log);
   return Status::OK();
 }
 
+int DBImpl::IsFileDeletionsEnabled() const {
+  return disable_delete_obsolete_files_;
+}
+
 Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
                             uint64_t* manifest_file_size,
                             bool flush_memtable) {
@@ -91,7 +98,7 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
 
     if (!status.ok()) {
       mutex_.Unlock();
-      Log(options_.info_log, "Cannot Flush data %s\n",
+      Log(db_options_.info_log, "Cannot Flush data %s\n",
           status.ToString().c_str());
       return status;
     }
@@ -129,7 +136,7 @@ Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
   Status s;
   // list wal files in main db dir.
   VectorLogPtr logs;
-  s = GetSortedWalsOfType(options_.wal_dir, logs, kAliveLogFile);
+  s = GetSortedWalsOfType(db_options_.wal_dir, logs, kAliveLogFile);
   if (!s.ok()) {
     return s;
   }
@@ -142,7 +149,7 @@ Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
 
   files.clear();
   // list wal files in archive dir.
-  std::string archivedir = ArchivalDirectory(options_.wal_dir);
+  std::string archivedir = ArchivalDirectory(db_options_.wal_dir);
   if (env_->FileExists(archivedir)) {
     s = GetSortedWalsOfType(archivedir, files, kArchivedLogFile);
     if (!s.ok()) {
@@ -153,7 +160,7 @@ Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
   uint64_t latest_archived_log_number = 0;
   if (!files.empty()) {
     latest_archived_log_number = files.back()->LogNumber();
-    Log(options_.info_log, "Latest Archived log: %" PRIu64,
+    Log(db_options_.info_log, "Latest Archived log: %" PRIu64,
         latest_archived_log_number);
   }
 
@@ -166,7 +173,7 @@ Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
       // same log in both db dir and archived dir. Simply
       // ignore the one in db dir. Note that, if we read
       // archived dir first, we would have missed the log file.
-      Log(options_.info_log, "%s already moved to archive",
+      Log(db_options_.info_log, "%s already moved to archive",
           log->PathName().c_str());
     }
   }

db/db_impl.h

@@ -28,10 +28,13 @@
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/transaction_log.h"
 #include "util/autovector.h"
-#include "util/stats_logger.h"
 #include "util/stop_watch.h"
 #include "util/thread_local.h"
+#include "util/scoped_arena_iterator.h"
 #include "db/internal_stats.h"
+#include "db/write_controller.h"
+#include "db/flush_scheduler.h"
+#include "db/write_thread.h"
 
 namespace rocksdb {
 
@@ -97,13 +100,21 @@ class DBImpl : public DB {
   using DB::GetProperty;
   virtual bool GetProperty(ColumnFamilyHandle* column_family,
                            const Slice& property, std::string* value);
+  using DB::GetIntProperty;
+  virtual bool GetIntProperty(ColumnFamilyHandle* column_family,
+                              const Slice& property, uint64_t* value) override;
   using DB::GetApproximateSizes;
   virtual void GetApproximateSizes(ColumnFamilyHandle* column_family,
                                    const Range* range, int n, uint64_t* sizes);
   using DB::CompactRange;
   virtual Status CompactRange(ColumnFamilyHandle* column_family,
                               const Slice* begin, const Slice* end,
-                              bool reduce_level = false, int target_level = -1);
+                              bool reduce_level = false, int target_level = -1,
+                              uint32_t target_path_id = 0);
+
+  using DB::SetOptions;
+  bool SetOptions(ColumnFamilyHandle* column_family,
+      const std::unordered_map<std::string, std::string>& options_map);
 
   using DB::NumberLevels;
   virtual int NumberLevels(ColumnFamilyHandle* column_family);
@@ -124,6 +135,7 @@ class DBImpl : public DB {
 #ifndef ROCKSDB_LITE
   virtual Status DisableFileDeletions();
   virtual Status EnableFileDeletions(bool force);
+  virtual int IsFileDeletionsEnabled() const;
   // All the returned filenames start with "/"
   virtual Status GetLiveFiles(std::vector<std::string>&,
                               uint64_t* manifest_file_size,
@@ -146,8 +158,8 @@ class DBImpl : public DB {
   virtual Status GetDbIdentity(std::string& identity);
 
   Status RunManualCompaction(ColumnFamilyData* cfd, int input_level,
-                             int output_level, const Slice* begin,
-                             const Slice* end);
+                             int output_level, uint32_t output_path_id,
+                             const Slice* begin, const Slice* end);
 
 #ifndef ROCKSDB_LITE
   // Extra methods (for testing) that are not in the public DB interface
@@ -169,8 +181,8 @@ class DBImpl : public DB {
   // Return an internal iterator over the current state of the database.
   // The keys of this iterator are internal keys (see format.h).
   // The returned iterator should be deleted when no longer needed.
-  Iterator* TEST_NewInternalIterator(ColumnFamilyHandle* column_family =
-                                         nullptr);
+  Iterator* TEST_NewInternalIterator(
+      Arena* arena, ColumnFamilyHandle* column_family = nullptr);
 
   // Return the maximum overlapping data (in bytes) at next level for any
   // file at a level >= 1.
@@ -198,6 +210,17 @@ class DBImpl : public DB {
                               SequenceNumber* sequence);
 
   Status TEST_ReadFirstLine(const std::string& fname, SequenceNumber* sequence);
+
+  void TEST_LockMutex();
+
+  void TEST_UnlockMutex();
+
+  // REQUIRES: mutex locked
+  void* TEST_BeginWrite();
+
+  // REQUIRES: mutex locked
+  // pass the pointer that you got from TEST_BeginWrite()
+  void TEST_EndWrite(void* w);
 #endif  // NDEBUG
 
   // Structure to store information for candidate files to delete.
@@ -271,7 +294,7 @@ class DBImpl : public DB {
   // Returns the list of live files in 'live' and the list
   // of all files in the filesystem in 'candidate_files'.
   // If force == false and the last call was less than
-  // options_.delete_obsolete_files_period_micros microseconds ago,
+  // db_options_.delete_obsolete_files_period_micros microseconds ago,
   // it will not fill up the deletion_state
   void FindObsoleteFiles(DeletionState& deletion_state,
                          bool force,
@@ -289,22 +312,23 @@ class DBImpl : public DB {
   Env* const env_;
   const std::string dbname_;
   unique_ptr<VersionSet> versions_;
-  const DBOptions options_;
+  const DBOptions db_options_;
+  Statistics* stats_;
 
   Iterator* NewInternalIterator(const ReadOptions&, ColumnFamilyData* cfd,
-                                SuperVersion* super_version,
-                                Arena* arena = nullptr);
+                                SuperVersion* super_version, Arena* arena);
 
  private:
   friend class DB;
   friend class InternalStats;
 #ifndef ROCKSDB_LITE
-  friend class TailingIterator;
   friend class ForwardIterator;
 #endif
   friend struct SuperVersion;
+  friend class CompactedDBImpl;
   struct CompactionState;
-  struct Writer;
+
+  struct WriteContext;
 
   Status NewDB();
 
@@ -323,12 +347,13 @@ class DBImpl : public DB {
 
   // Flush the in-memory write buffer to storage.  Switches to a new
   // log-file/memtable and writes a new descriptor iff successful.
-  Status FlushMemTableToOutputFile(ColumnFamilyData* cfd, bool* madeProgress,
-                                   DeletionState& deletion_state,
-                                   LogBuffer* log_buffer);
+  Status FlushMemTableToOutputFile(
+      ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
+      bool* madeProgress, DeletionState& deletion_state, LogBuffer* log_buffer);
 
-  Status RecoverLogFile(uint64_t log_number, SequenceNumber* max_sequence,
-                        bool read_only);
+  // REQUIRES: log_numbers are sorted in ascending order
+  Status RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
+                         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
@@ -337,21 +362,17 @@ class DBImpl : public DB {
   // concurrent flush memtables to storage.
   Status WriteLevel0TableForRecovery(ColumnFamilyData* cfd, MemTable* mem,
                                      VersionEdit* edit);
-  Status WriteLevel0Table(ColumnFamilyData* cfd, autovector<MemTable*>& mems,
-                          VersionEdit* edit, uint64_t* filenumber,
-                          LogBuffer* log_buffer);
+  Status WriteLevel0Table(ColumnFamilyData* cfd,
+      const MutableCFOptions& mutable_cf_options,
+      const autovector<MemTable*>& mems,
+      VersionEdit* edit, uint64_t* filenumber, LogBuffer* log_buffer);
 
-  uint64_t SlowdownAmount(int n, double bottom, double top);
+  void DelayWrite(uint64_t expiration_time);
 
-  // TODO(icanadi) free superversion_to_free and old_log outside of mutex
-  Status MakeRoomForWrite(ColumnFamilyData* cfd,
-                          bool force /* flush even if there is room? */,
-                          autovector<SuperVersion*>* superversions_to_free,
-                          autovector<log::Writer*>* logs_to_free,
-                          uint64_t expiration_time);
+  Status ScheduleFlushes(WriteContext* context);
 
-  void BuildBatchGroup(Writer** last_writer,
-                       autovector<WriteBatch*>* write_batch_group);
+  Status SetNewMemtableAndNewLogFile(ColumnFamilyData* cfd,
+                                     WriteContext* context);
 
   // Force current memtable contents to be flushed.
   Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options);
@@ -362,13 +383,6 @@ class DBImpl : public DB {
   void RecordFlushIOStats();
   void RecordCompactionIOStats();
 
-  void MaybeScheduleLogDBDeployStats();
-
-#ifndef ROCKSDB_LITE
-  static void BGLogDBDeployStats(void* db);
-  void LogDBDeployStats();
-#endif  // ROCKSDB_LITE
-
   void MaybeScheduleFlushOrCompaction();
   static void BGWorkCompaction(void* db);
   static void BGWorkFlush(void* db);
@@ -380,6 +394,7 @@ class DBImpl : public DB {
                          LogBuffer* log_buffer);
   void CleanupCompaction(CompactionState* compact, Status status);
   Status DoCompactionWork(CompactionState* compact,
+                          const MutableCFOptions& mutable_cf_options,
                           DeletionState& deletion_state,
                           LogBuffer* log_buffer);
 
@@ -387,12 +402,13 @@ class DBImpl : public DB {
   // preempt compaction, since it's higher prioirty
   // Returns: micros spent executing
   uint64_t CallFlushDuringCompaction(ColumnFamilyData* cfd,
-                                     DeletionState& deletion_state,
-                                     LogBuffer* log_buffer);
+      const MutableCFOptions& mutable_cf_options, DeletionState& deletion_state,
+      LogBuffer* log_buffer);
 
   // Call compaction filter if is_compaction_v2 is not true. Then iterate
   // through input and compact the kv-pairs
   Status ProcessKeyValueCompaction(
+    const MutableCFOptions& mutable_cf_options,
     bool is_snapshot_supported,
     SequenceNumber visible_at_tip,
     SequenceNumber earliest_snapshot,
@@ -403,16 +419,18 @@ class DBImpl : public DB {
     Iterator* input,
     CompactionState* compact,
     bool is_compaction_v2,
+    int* num_output_records,
     LogBuffer* log_buffer);
 
   // Call compaction_filter_v2->Filter() on kv-pairs in compact
   void CallCompactionFilterV2(CompactionState* compact,
     CompactionFilterV2* compaction_filter_v2);
 
-  Status OpenCompactionOutputFile(CompactionState* compact);
+  Status OpenCompactionOutputFile(CompactionState* compact,
+      const MutableCFOptions& mutable_cf_options);
   Status FinishCompactionOutputFile(CompactionState* compact, Iterator* input);
   Status InstallCompactionResults(CompactionState* compact,
-                                  LogBuffer* log_buffer);
+      const MutableCFOptions& mutable_cf_options, LogBuffer* log_buffer);
   void AllocateCompactionOutputFileNumbers(CompactionState* compact);
   void ReleaseCompactionUnusedFileNumbers(CompactionState* compact);
 
@@ -454,7 +472,8 @@ class DBImpl : public DB {
 
   // Return the minimum empty level that could hold the total data in the
   // input level. Return the input level, if such level could not be found.
-  int FindMinimumEmptyLevelFitting(ColumnFamilyData* cfd, int level);
+  int FindMinimumEmptyLevelFitting(ColumnFamilyData* cfd,
+      const MutableCFOptions& mutable_cf_options, int level);
 
   // Move the files in the input level to the target level.
   // If target_level < 0, automatically calculate the minimum level that could
@@ -478,12 +497,12 @@ class DBImpl : public DB {
   // * whenever bg_flush_scheduled_ value decreases (i.e. whenever a flush is
   // done, even if it didn't make any progress)
   // * whenever there is an error in background flush or compaction
-  // * whenever bg_logstats_scheduled_ turns to false
   port::CondVar bg_cv_;
   uint64_t logfile_number_;
   unique_ptr<log::Writer> log_;
   bool log_empty_;
   ColumnFamilyHandleImpl* default_cf_handle_;
+  InternalStats* default_cf_internal_stats_;
   unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
   struct LogFileNumberSize {
     explicit LogFileNumberSize(uint64_t _number)
@@ -502,14 +521,15 @@ class DBImpl : public DB {
   // some code-paths
   bool single_column_family_mode_;
 
-  std::string host_name_;
-
   std::unique_ptr<Directory> db_directory_;
 
-  // Queue of writers.
-  std::deque<Writer*> writers_;
+  WriteThread write_thread_;
+
   WriteBatch tmp_batch_;
 
+  WriteController write_controller_;
+  FlushScheduler flush_scheduler_;
+
   SnapshotList snapshots_;
 
   // cache for ReadFirstRecord() calls
@@ -536,14 +556,12 @@ class DBImpl : public DB {
   // number of background memtable flush jobs, submitted to the HIGH pool
   int bg_flush_scheduled_;
 
-  // Has a background stats log thread scheduled?
-  bool bg_logstats_scheduled_;
-
   // Information for a manual compaction
   struct ManualCompaction {
     ColumnFamilyData* cfd;
     int input_level;
     int output_level;
+    uint32_t output_path_id;
     bool done;
     Status status;
     bool in_progress;           // compaction request being processed?
@@ -556,10 +574,6 @@ class DBImpl : public DB {
   // Have we encountered a background error in paranoid mode?
   Status bg_error_;
 
-  std::unique_ptr<StatsLogger> logger_;
-
-  int64_t volatile last_log_ts;
-
   // shall we disable deletion of obsolete files
   // if 0 the deletion is enabled.
   // if non-zero, files will not be getting deleted
@@ -584,14 +598,10 @@ class DBImpl : public DB {
   bool flush_on_destroy_; // Used when disableWAL is true.
 
   static const int KEEP_LOG_FILE_NUM = 1000;
-  static const uint64_t kNoTimeOut = std::numeric_limits<uint64_t>::max();
   std::string db_absolute_path_;
 
-  // count of the number of contiguous delaying writes
-  int delayed_writes_;
-
   // The options to access storage files
-  const EnvOptions storage_options_;
+  const EnvOptions env_options_;
 
   // A value of true temporarily disables scheduling of background work
   bool bg_work_gate_closed_;
@@ -606,9 +616,6 @@ class DBImpl : public DB {
   DBImpl(const DBImpl&);
   void operator=(const DBImpl&);
 
-  // dump the delayed_writes_ to the log file and reset counter.
-  void DelayLoggingAndReset();
-
   // Return the earliest snapshot where seqno is visible.
   // Store the snapshot right before that, if any, in prev_snapshot
   inline SequenceNumber findEarliestVisibleSnapshot(
@@ -620,7 +627,18 @@ class DBImpl : public DB {
   // the cfd->InstallSuperVersion() function. Background threads carry
   // deletion_state which can have new_superversion already allocated.
   void InstallSuperVersion(ColumnFamilyData* cfd,
-                           DeletionState& deletion_state);
+                           DeletionState& deletion_state,
+                           const MutableCFOptions& mutable_cf_options);
+
+  // Find Super version and reference it. Based on options, it might return
+  // the thread local cached one.
+  inline SuperVersion* GetAndRefSuperVersion(ColumnFamilyData* cfd);
+
+  // Un-reference the super version and return it to thread local cache if
+  // needed. If it is the last reference of the super version. Clean it up
+  // after un-referencing it.
+  inline void ReturnAndCleanupSuperVersion(ColumnFamilyData* cfd,
+                                           SuperVersion* sv);
 
 #ifndef ROCKSDB_LITE
   using DB::GetPropertiesOfAllTables;
@@ -634,14 +652,28 @@ class DBImpl : public DB {
   Status GetImpl(const ReadOptions& options, ColumnFamilyHandle* column_family,
                  const Slice& key, std::string* value,
                  bool* value_found = nullptr);
+
+  bool GetIntPropertyInternal(ColumnFamilyHandle* column_family,
+                              DBPropertyType property_type,
+                              bool need_out_of_mutex, uint64_t* value);
 };
 
 // Sanitize db options.  The caller should delete result.info_log if
 // it is not equal to src.info_log.
 extern Options SanitizeOptions(const std::string& db,
                                const InternalKeyComparator* icmp,
-                               const InternalFilterPolicy* ipolicy,
                                const Options& src);
 extern DBOptions SanitizeOptions(const std::string& db, const DBOptions& src);
 
+// Fix user-supplied options to be reasonable
+template <class T, class V>
+static void ClipToRange(T* ptr, V minvalue, V maxvalue) {
+  if (static_cast<V>(*ptr) > maxvalue) *ptr = maxvalue;
+  if (static_cast<V>(*ptr) < minvalue) *ptr = minvalue;
+}
+
+// Dump db file summary, implemented in util/
+extern void DumpDBFileSummary(const DBOptions& options,
+                              const std::string& dbname);
+
 }  // namespace rocksdb

db/db_impl_debug.cc

@@ -20,7 +20,8 @@ uint64_t DBImpl::TEST_GetLevel0TotalSize() {
   return default_cf_handle_->cfd()->current()->NumLevelBytes(0);
 }
 
-Iterator* DBImpl::TEST_NewInternalIterator(ColumnFamilyHandle* column_family) {
+Iterator* DBImpl::TEST_NewInternalIterator(Arena* arena,
+                                           ColumnFamilyHandle* column_family) {
   ColumnFamilyData* cfd;
   if (column_family == nullptr) {
     cfd = default_cf_handle_->cfd();
@@ -33,7 +34,7 @@ Iterator* DBImpl::TEST_NewInternalIterator(ColumnFamilyHandle* column_family) {
   SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
   mutex_.Unlock();
   ReadOptions roptions;
-  return NewInternalIterator(roptions, cfd, super_version);
+  return NewInternalIterator(roptions, cfd, super_version, arena);
 }
 
 int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
@@ -85,7 +86,7 @@ Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
        cfd->options()->compaction_style == kCompactionStyleFIFO)
           ? level
           : level + 1;
-  return RunManualCompaction(cfd, level, output_level, begin, end);
+  return RunManualCompaction(cfd, level, output_level, 0, begin, end);
 }
 
 Status DBImpl::TEST_FlushMemTable(bool wait) {
@@ -129,5 +130,27 @@ Status DBImpl::TEST_ReadFirstLine(const std::string& fname,
                                   SequenceNumber* sequence) {
   return ReadFirstLine(fname, sequence);
 }
+
+void DBImpl::TEST_LockMutex() {
+  mutex_.Lock();
+}
+
+void DBImpl::TEST_UnlockMutex() {
+  mutex_.Unlock();
+}
+
+void* DBImpl::TEST_BeginWrite() {
+  auto w = new WriteThread::Writer(&mutex_);
+  Status s = write_thread_.EnterWriteThread(w, 0);
+  assert(s.ok() && !w->done);  // No timeout and nobody should do our job
+  return reinterpret_cast<void*>(w);
+}
+
+void DBImpl::TEST_EndWrite(void* w) {
+  auto writer = reinterpret_cast<WriteThread::Writer*>(w);
+  write_thread_.ExitWriteThread(writer, writer, Status::OK());
+  delete writer;
+}
+
 }  // namespace rocksdb
 #endif  // ROCKSDB_LITE

db/db_impl_readonly.cc

@@ -2,57 +2,28 @@
 //  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) 2012 Facebook. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
 
 #include "db/db_impl_readonly.h"
+#include "utilities/compacted_db/compacted_db_impl.h"
 #include "db/db_impl.h"
-
-#include <algorithm>
-#include <set>
-#include <string>
-#include <stdint.h>
-#include <stdio.h>
-#include <vector>
-#include <algorithm>
-#include "db/db_iter.h"
-#include "db/dbformat.h"
-#include "db/filename.h"
-#include "db/log_reader.h"
-#include "db/log_writer.h"
-#include "db/memtable.h"
 #include "db/merge_context.h"
-#include "db/table_cache.h"
-#include "db/version_set.h"
-#include "db/write_batch_internal.h"
-#include "rocksdb/db.h"
-#include "rocksdb/env.h"
-#include "rocksdb/status.h"
-#include "rocksdb/table.h"
-#include "rocksdb/merge_operator.h"
-#include "port/port.h"
-#include "table/block.h"
-#include "table/merger.h"
-#include "table/two_level_iterator.h"
-#include "util/coding.h"
-#include "util/logging.h"
-#include "util/build_version.h"
+#include "db/db_iter.h"
+#include "util/perf_context_imp.h"
 
 namespace rocksdb {
 
-DBImplReadOnly::DBImplReadOnly(const DBOptions& options,
+DBImplReadOnly::DBImplReadOnly(const DBOptions& db_options,
                                const std::string& dbname)
-    : DBImpl(options, dbname) {
-  Log(options_.info_log, "Opening the db in read only mode");
+    : DBImpl(db_options, dbname) {
+  Log(INFO_LEVEL, db_options_.info_log, "Opening the db in read only mode");
+  LogFlush(db_options_.info_log);
 }
 
 DBImplReadOnly::~DBImplReadOnly() {
 }
 
 // Implementations of the DB interface
-Status DBImplReadOnly::Get(const ReadOptions& options,
+Status DBImplReadOnly::Get(const ReadOptions& read_options,
                            ColumnFamilyHandle* column_family, const Slice& key,
                            std::string* value) {
   Status s;
@@ -62,32 +33,75 @@ Status DBImplReadOnly::Get(const ReadOptions& options,
   SuperVersion* super_version = cfd->GetSuperVersion();
   MergeContext merge_context;
   LookupKey lkey(key, snapshot);
-  if (super_version->mem->Get(lkey, value, &s, merge_context,
-                              *cfd->options())) {
+  if (super_version->mem->Get(lkey, value, &s, &merge_context)) {
   } else {
-    super_version->current->Get(options, lkey, value, &s, &merge_context);
+    PERF_TIMER_GUARD(get_from_output_files_time);
+    super_version->current->Get(read_options, lkey, value, &s, &merge_context);
   }
   return s;
 }
 
-Iterator* DBImplReadOnly::NewIterator(const ReadOptions& options,
+Iterator* DBImplReadOnly::NewIterator(const ReadOptions& read_options,
                                       ColumnFamilyHandle* column_family) {
   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
   auto cfd = cfh->cfd();
   SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
   SequenceNumber latest_snapshot = versions_->LastSequence();
-  Iterator* internal_iter = NewInternalIterator(options, cfd, super_version);
-  return NewDBIterator(
-      env_, *cfd->options(), cfd->user_comparator(), internal_iter,
-      (options.snapshot != nullptr
-           ? reinterpret_cast<const SnapshotImpl*>(options.snapshot)->number_
-           : latest_snapshot));
+  auto db_iter = NewArenaWrappedDbIterator(
+      env_, *cfd->ioptions(), cfd->user_comparator(),
+      (read_options.snapshot != nullptr
+           ? reinterpret_cast<const SnapshotImpl*>(
+                read_options.snapshot)->number_
+           : latest_snapshot),
+      cfd->options()->max_sequential_skip_in_iterations);
+  auto internal_iter = NewInternalIterator(
+      read_options, cfd, super_version, db_iter->GetArena());
+  db_iter->SetIterUnderDBIter(internal_iter);
+  return db_iter;
+}
+
+Status DBImplReadOnly::NewIterators(
+    const ReadOptions& read_options,
+    const std::vector<ColumnFamilyHandle*>& column_families,
+    std::vector<Iterator*>* iterators) {
+  if (iterators == nullptr) {
+    return Status::InvalidArgument("iterators not allowed to be nullptr");
+  }
+  iterators->clear();
+  iterators->reserve(column_families.size());
+  SequenceNumber latest_snapshot = versions_->LastSequence();
+
+  for (auto cfh : column_families) {
+    auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
+    auto db_iter = NewArenaWrappedDbIterator(
+        env_, *cfd->ioptions(), cfd->user_comparator(),
+        (read_options.snapshot != nullptr
+            ? reinterpret_cast<const SnapshotImpl*>(
+                  read_options.snapshot)->number_
+            : latest_snapshot),
+        cfd->options()->max_sequential_skip_in_iterations);
+    auto internal_iter = NewInternalIterator(
+        read_options, cfd, cfd->GetSuperVersion()->Ref(), db_iter->GetArena());
+    db_iter->SetIterUnderDBIter(internal_iter);
+    iterators->push_back(db_iter);
+  }
+
+  return Status::OK();
 }
 
 Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
                            DB** dbptr, bool error_if_log_file_exist) {
   *dbptr = nullptr;
 
+  // Try to first open DB as fully compacted DB
+  Status s;
+#ifndef ROCKSDB_LITE
+  s = CompactedDBImpl::Open(options, dbname, dbptr);
+  if (s.ok()) {
+    return s;
+  }
+#endif
+
   DBOptions db_options(options);
   ColumnFamilyOptions cf_options(options);
   std::vector<ColumnFamilyDescriptor> column_families;
@@ -95,8 +109,7 @@ Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
       ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
   std::vector<ColumnFamilyHandle*> handles;
 
-  Status s =
-      DB::OpenForReadOnly(db_options, dbname, column_families, &handles, dbptr);
+  s = DB::OpenForReadOnly(db_options, dbname, column_families, &handles, dbptr);
   if (s.ok()) {
     assert(handles.size() == 1);
     // i can delete the handle since DBImpl is always holding a

db/db_impl_readonly.h

@@ -2,25 +2,11 @@
 //  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) 2012 Facebook. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
 
 #pragma once
 #include "db/db_impl.h"
-
-#include <deque>
-#include <set>
 #include <vector>
 #include <string>
-#include "db/dbformat.h"
-#include "db/log_writer.h"
-#include "db/snapshot.h"
-#include "rocksdb/db.h"
-#include "rocksdb/env.h"
-#include "port/port.h"
-#include "util/stats_logger.h"
 
 namespace rocksdb {
 
@@ -33,63 +19,66 @@ class DBImplReadOnly : public DBImpl {
   using DB::Get;
   virtual Status Get(const ReadOptions& options,
                      ColumnFamilyHandle* column_family, const Slice& key,
-                     std::string* value);
+                     std::string* value) override;
 
   // TODO: Implement ReadOnly MultiGet?
 
   using DBImpl::NewIterator;
   virtual Iterator* NewIterator(const ReadOptions&,
-                                ColumnFamilyHandle* column_family);
+                                ColumnFamilyHandle* column_family) override;
 
   virtual Status NewIterators(
       const ReadOptions& options,
-      const std::vector<ColumnFamilyHandle*>& column_family,
-      std::vector<Iterator*>* iterators) {
-   // TODO
-    return Status::NotSupported("Not supported yet.");
-  }
+      const std::vector<ColumnFamilyHandle*>& column_families,
+      std::vector<Iterator*>* iterators) override;
 
   using DBImpl::Put;
   virtual Status Put(const WriteOptions& options,
                      ColumnFamilyHandle* column_family, const Slice& key,
-                     const Slice& value) {
+                     const Slice& value) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
   using DBImpl::Merge;
   virtual Status Merge(const WriteOptions& options,
                        ColumnFamilyHandle* column_family, const Slice& key,
-                       const Slice& value) {
+                       const Slice& value) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
   using DBImpl::Delete;
   virtual Status Delete(const WriteOptions& options,
-                        ColumnFamilyHandle* column_family, const Slice& key) {
+                        ColumnFamilyHandle* column_family,
+                        const Slice& key) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
-  virtual Status Write(const WriteOptions& options, WriteBatch* updates) {
+  virtual Status Write(const WriteOptions& options,
+                       WriteBatch* updates) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
   using DBImpl::CompactRange;
   virtual Status CompactRange(ColumnFamilyHandle* column_family,
                               const Slice* begin, const Slice* end,
-                              bool reduce_level = false,
-                              int target_level = -1) {
+                              bool reduce_level = false, int target_level = -1,
+                              uint32_t target_path_id = 0) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
-  virtual Status DisableFileDeletions() {
+
+#ifndef ROCKSDB_LITE
+  virtual Status DisableFileDeletions() override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
-  virtual Status EnableFileDeletions(bool force) {
+  virtual Status EnableFileDeletions(bool force) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
   virtual Status GetLiveFiles(std::vector<std::string>&,
                               uint64_t* manifest_file_size,
-                              bool flush_memtable = true) {
+                              bool flush_memtable = true) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
+#endif  // ROCKSDB_LITE
+
   using DBImpl::Flush;
   virtual Status Flush(const FlushOptions& options,
-                       ColumnFamilyHandle* column_family) {
+                       ColumnFamilyHandle* column_family) override {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
 

db/db_iter.cc

@@ -10,6 +10,8 @@
 #include "db/db_iter.h"
 #include <stdexcept>
 #include <deque>
+#include <string>
+#include <limits>
 
 #include "db/filename.h"
 #include "db/dbformat.h"
@@ -56,21 +58,25 @@ class DBIter: public Iterator {
     kReverse
   };
 
-  DBIter(Env* env, const Options& options, const Comparator* cmp,
-         Iterator* iter, SequenceNumber s, bool arena_mode)
+  DBIter(Env* env, const ImmutableCFOptions& ioptions,
+         const Comparator* cmp, Iterator* iter, SequenceNumber s,
+         bool arena_mode, uint64_t max_sequential_skip_in_iterations,
+         const Slice* iterate_upper_bound = nullptr)
       : arena_mode_(arena_mode),
         env_(env),
-        logger_(options.info_log.get()),
+        logger_(ioptions.info_log),
         user_comparator_(cmp),
-        user_merge_operator_(options.merge_operator.get()),
+        user_merge_operator_(ioptions.merge_operator),
         iter_(iter),
         sequence_(s),
         direction_(kForward),
         valid_(false),
         current_entry_is_merged_(false),
-        statistics_(options.statistics.get()) {
-    RecordTick(statistics_, NO_ITERATORS, 1);
-    max_skip_ = options.max_sequential_skip_in_iterations;
+        statistics_(ioptions.statistics),
+        iterate_upper_bound_(iterate_upper_bound) {
+    RecordTick(statistics_, NO_ITERATORS);
+    prefix_extractor_ = ioptions.prefix_extractor;
+    max_skip_ = max_sequential_skip_in_iterations;
   }
   virtual ~DBIter() {
     RecordTick(statistics_, NO_ITERATORS, -1);
@@ -109,9 +115,14 @@ class DBIter: public Iterator {
   virtual void SeekToLast();
 
  private:
+  void PrevInternal();
+  void FindParseableKey(ParsedInternalKey* ikey, Direction direction);
+  bool FindValueForCurrentKey();
+  bool FindValueForCurrentKeyUsingSeek();
+  void FindPrevUserKey();
+  void FindNextUserKey();
   inline void FindNextUserEntry(bool skipping);
   void FindNextUserEntryInternal(bool skipping);
-  void FindPrevUserEntry();
   bool ParseKey(ParsedInternalKey* key);
   void MergeValuesNewToOld();
 
@@ -124,6 +135,7 @@ class DBIter: public Iterator {
     }
   }
 
+  const SliceTransform* prefix_extractor_;
   bool arena_mode_;
   Env* const env_;
   Logger* logger_;
@@ -133,13 +145,14 @@ class DBIter: public Iterator {
   SequenceNumber const sequence_;
 
   Status status_;
-  IterKey saved_key_;   // == current key when direction_==kReverse
-  std::string saved_value_;   // == current raw value when direction_==kReverse
+  IterKey saved_key_;
+  std::string saved_value_;
   Direction direction_;
   bool valid_;
   bool current_entry_is_merged_;
   Statistics* statistics_;
   uint64_t max_skip_;
+  const Slice* iterate_upper_bound_;
 
   // No copying allowed
   DBIter(const DBIter&);
@@ -160,20 +173,11 @@ inline bool DBIter::ParseKey(ParsedInternalKey* ikey) {
 void DBIter::Next() {
   assert(valid_);
 
-  if (direction_ == kReverse) {  // Switch directions?
+  if (direction_ == kReverse) {
+    FindNextUserKey();
     direction_ = kForward;
-    // iter_ is pointing just before the entries for this->key(),
-    // so advance into the range of entries for this->key() and then
-    // use the normal skipping code below.
     if (!iter_->Valid()) {
       iter_->SeekToFirst();
-    } else {
-      iter_->Next();
-    }
-    if (!iter_->Valid()) {
-      valid_ = false;
-      saved_key_.Clear();
-      return;
     }
   }
 
@@ -185,7 +189,6 @@ void DBIter::Next() {
   FindNextUserEntry(true /* skipping the current user key */);
 }
 
-
 // PRE: saved_key_ has the current user key if skipping
 // POST: saved_key_ should have the next user key if valid_,
 //       if the current entry is a result of merge
@@ -195,9 +198,8 @@ 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) {
-  PERF_TIMER_AUTO(find_next_user_entry_time);
+  PERF_TIMER_GUARD(find_next_user_entry_time);
   FindNextUserEntryInternal(skipping);
-  PERF_TIMER_STOP(find_next_user_entry_time);
 }
 
 // Actual implementation of DBIter::FindNextUserEntry()
@@ -209,36 +211,44 @@ void DBIter::FindNextUserEntryInternal(bool skipping) {
   uint64_t num_skipped = 0;
   do {
     ParsedInternalKey ikey;
-    if (ParseKey(&ikey) && ikey.sequence <= sequence_) {
-      if (skipping &&
-          user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
-        num_skipped++; // skip this entry
-        PERF_COUNTER_ADD(internal_key_skipped_count, 1);
-      } else {
-        skipping = false;
-        switch (ikey.type) {
-          case kTypeDeletion:
-            // Arrange to skip all upcoming entries for this key since
-            // they are hidden by this deletion.
-            saved_key_.SetKey(ikey.user_key);
-            skipping = true;
-            num_skipped = 0;
-            PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
-            break;
-          case kTypeValue:
-            valid_ = true;
-            saved_key_.SetKey(ikey.user_key);
-            return;
-          case kTypeMerge:
-            // By now, we are sure the current ikey is going to yield a value
-            saved_key_.SetKey(ikey.user_key);
-            current_entry_is_merged_ = true;
-            valid_ = true;
-            MergeValuesNewToOld();  // Go to a different state machine
-            return;
-          default:
-            assert(false);
-            break;
+
+    if (ParseKey(&ikey)) {
+      if (iterate_upper_bound_ != nullptr &&
+          ikey.user_key.compare(*iterate_upper_bound_) >= 0) {
+        break;
+      }
+
+      if (ikey.sequence <= sequence_) {
+        if (skipping &&
+           user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
+          num_skipped++;  // skip this entry
+          PERF_COUNTER_ADD(internal_key_skipped_count, 1);
+        } else {
+          skipping = false;
+          switch (ikey.type) {
+            case kTypeDeletion:
+              // Arrange to skip all upcoming entries for this key since
+              // they are hidden by this deletion.
+              saved_key_.SetKey(ikey.user_key);
+              skipping = true;
+              num_skipped = 0;
+              PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+              break;
+            case kTypeValue:
+              valid_ = true;
+              saved_key_.SetKey(ikey.user_key);
+              return;
+            case kTypeMerge:
+              // By now, we are sure the current ikey is going to yield a value
+              saved_key_.SetKey(ikey.user_key);
+              current_entry_is_merged_ = true;
+              valid_ = true;
+              MergeValuesNewToOld();  // Go to a different state machine
+              return;
+            default:
+              assert(false);
+              break;
+          }
         }
       }
     }
@@ -277,7 +287,6 @@ void DBIter::MergeValuesNewToOld() {
   std::deque<std::string> operands;
   operands.push_front(iter_->value().ToString());
 
-  std::string merge_result;   // Temporary string to hold merge result later
   ParsedInternalKey ikey;
   for (iter_->Next(); iter_->Valid(); iter_->Next()) {
     if (!ParseKey(&ikey)) {
@@ -327,110 +336,259 @@ void DBIter::MergeValuesNewToOld() {
 
 void DBIter::Prev() {
   assert(valid_);
-
-  // Throw an exception now if merge_operator is provided
-  // TODO: support backward iteration
-  if (user_merge_operator_) {
-    Log(logger_, "Prev not supported yet if merge_operator is provided");
-    throw std::logic_error("DBIter::Prev backward iteration not supported"
-                           " if merge_operator is provided");
-  }
-
-  if (direction_ == kForward) {  // Switch directions?
-    // iter_ is pointing at the current entry.  Scan backwards until
-    // the key changes so we can use the normal reverse scanning code.
-    assert(iter_->Valid());  // Otherwise valid_ would have been false
-    saved_key_.SetKey(ExtractUserKey(iter_->key()));
-    while (true) {
-      iter_->Prev();
-      if (!iter_->Valid()) {
-        valid_ = false;
-        saved_key_.Clear();
-        ClearSavedValue();
-        return;
-      }
-      if (user_comparator_->Compare(ExtractUserKey(iter_->key()),
-                                    saved_key_.GetKey()) < 0) {
-        break;
-      }
-    }
+  if (direction_ == kForward) {
+    FindPrevUserKey();
     direction_ = kReverse;
   }
-
-  FindPrevUserEntry();
+  PrevInternal();
 }
 
-void DBIter::FindPrevUserEntry() {
-  assert(direction_ == kReverse);
-  uint64_t num_skipped = 0;
-
-  ValueType value_type = kTypeDeletion;
-  bool saved_key_valid = true;
-  if (iter_->Valid()) {
-    do {
-      ParsedInternalKey ikey;
-      if (ParseKey(&ikey) && ikey.sequence <= sequence_) {
-        if ((value_type != kTypeDeletion) &&
-            user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) < 0) {
-          // We encountered a non-deleted value in entries for previous keys,
-          break;
-        }
-        value_type = ikey.type;
-        if (value_type == kTypeDeletion) {
-          saved_key_.Clear();
-          ClearSavedValue();
-          saved_key_valid = false;
-        } else {
-          Slice raw_value = iter_->value();
-          if (saved_value_.capacity() > raw_value.size() + 1048576) {
-            std::string empty;
-            swap(empty, saved_value_);
-          }
-          saved_key_.SetKey(ExtractUserKey(iter_->key()));
-          saved_value_.assign(raw_value.data(), raw_value.size());
-        }
-      } else {
-        // In the case of ikey.sequence > sequence_, we might have already
-        // iterated to a different user key.
-        saved_key_valid = false;
-      }
-      num_skipped++;
-      // If we have sequentially iterated via numerous keys and still not
-      // found the prev user-key, then it is better to seek so that we can
-      // avoid too many key comparisons. We seek to the first occurence of
-      // our current key by looking for max sequence number.
-      if (saved_key_valid && num_skipped > max_skip_) {
-        num_skipped = 0;
-        std::string last_key;
-        AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(),
-                                                       kMaxSequenceNumber,
-                                                       kValueTypeForSeek));
-        iter_->Seek(last_key);
-        RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
-      } else {
-        iter_->Prev();
-      }
-    } while (iter_->Valid());
+void DBIter::PrevInternal() {
+  if (!iter_->Valid()) {
+    valid_ = false;
+    return;
   }
 
-  if (value_type == kTypeDeletion) {
-    // End
+  ParsedInternalKey ikey;
+
+  while (iter_->Valid()) {
+    saved_key_.SetKey(ExtractUserKey(iter_->key()));
+    if (FindValueForCurrentKey()) {
+      valid_ = true;
+      if (!iter_->Valid()) {
+        return;
+      }
+      FindParseableKey(&ikey, kReverse);
+      if (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0) {
+        FindPrevUserKey();
+      }
+      return;
+    }
+    if (!iter_->Valid()) {
+      break;
+    }
+    FindParseableKey(&ikey, kReverse);
+    if (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0) {
+
+      FindPrevUserKey();
+    }
+  }
+  // We haven't found any key - iterator is not valid
+  assert(!iter_->Valid());
+  valid_ = false;
+}
+
+// This function checks, if the entry with biggest sequence_number <= sequence_
+// is non kTypeDeletion. If it's not, we save value in saved_value_
+bool DBIter::FindValueForCurrentKey() {
+  assert(iter_->Valid());
+  // Contains operands for merge operator.
+  std::deque<std::string> operands;
+  // last entry before merge (could be kTypeDeletion or kTypeValue)
+  ValueType last_not_merge_type = kTypeDeletion;
+  ValueType last_key_entry_type = kTypeDeletion;
+
+  ParsedInternalKey ikey;
+  FindParseableKey(&ikey, kReverse);
+
+  size_t num_skipped = 0;
+  while (iter_->Valid() && ikey.sequence <= sequence_ &&
+         (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0)) {
+    // We iterate too much: let's use Seek() to avoid too much key comparisons
+    if (num_skipped >= max_skip_) {
+      return FindValueForCurrentKeyUsingSeek();
+    }
+
+    last_key_entry_type = ikey.type;
+    switch (last_key_entry_type) {
+      case kTypeValue:
+        operands.clear();
+        saved_value_ = iter_->value().ToString();
+        last_not_merge_type = kTypeValue;
+        break;
+      case kTypeDeletion:
+        operands.clear();
+        last_not_merge_type = kTypeDeletion;
+        PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
+        break;
+      case kTypeMerge:
+        assert(user_merge_operator_ != nullptr);
+        operands.push_back(iter_->value().ToString());
+        break;
+      default:
+        assert(false);
+    }
+
+    PERF_COUNTER_ADD(internal_key_skipped_count, 1);
+    assert(user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0);
+    iter_->Prev();
+    ++num_skipped;
+    FindParseableKey(&ikey, kReverse);
+  }
+
+  switch (last_key_entry_type) {
+    case kTypeDeletion:
+      valid_ = false;
+      return false;
+    case kTypeMerge:
+      if (last_not_merge_type == kTypeDeletion) {
+        user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
+                                        &saved_value_, logger_);
+      } else {
+        assert(last_not_merge_type == kTypeValue);
+        std::string last_put_value = saved_value_;
+        Slice temp_slice(last_put_value);
+        user_merge_operator_->FullMerge(saved_key_.GetKey(), &temp_slice,
+                                        operands, &saved_value_, logger_);
+      }
+      break;
+    case kTypeValue:
+      // do nothing - we've already has value in saved_value_
+      break;
+    default:
+      assert(false);
+      break;
+  }
+  valid_ = true;
+  return true;
+}
+
+// This function is used in FindValueForCurrentKey.
+// We use Seek() function instead of Prev() to find necessary value
+bool DBIter::FindValueForCurrentKeyUsingSeek() {
+  std::string last_key;
+  AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(), sequence_,
+                                                 kValueTypeForSeek));
+  iter_->Seek(last_key);
+  RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
+
+  // assume there is at least one parseable key for this user key
+  ParsedInternalKey ikey;
+  FindParseableKey(&ikey, kForward);
+
+  if (ikey.type == kTypeValue || ikey.type == kTypeDeletion) {
+    if (ikey.type == kTypeValue) {
+      saved_value_ = iter_->value().ToString();
+      valid_ = true;
+      return true;
+    }
     valid_ = false;
-    saved_key_.Clear();
-    ClearSavedValue();
-    direction_ = kForward;
-  } else {
+    return false;
+  }
+
+  // kTypeMerge. We need to collect all kTypeMerge values and save them
+  // in operands
+  std::deque<std::string> operands;
+  while (iter_->Valid() &&
+         (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0) &&
+         ikey.type == kTypeMerge) {
+    operands.push_front(iter_->value().ToString());
+    iter_->Next();
+    FindParseableKey(&ikey, kForward);
+  }
+
+  if (!iter_->Valid() ||
+      (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) != 0) ||
+      ikey.type == kTypeDeletion) {
+    user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
+                                    &saved_value_, logger_);
+
+    // Make iter_ valid and point to saved_key_
+    if (!iter_->Valid() ||
+        (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) != 0)) {
+      iter_->Seek(last_key);
+      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
+    }
     valid_ = true;
+    return true;
+  }
+
+  const Slice& value = iter_->value();
+  user_merge_operator_->FullMerge(saved_key_.GetKey(), &value, operands,
+                                  &saved_value_, logger_);
+  valid_ = true;
+  return true;
+}
+
+// Used in Next to change directions
+// Go to next user key
+// Don't use Seek(),
+// because next user key will be very close
+void DBIter::FindNextUserKey() {
+  if (!iter_->Valid()) {
+    return;
+  }
+  ParsedInternalKey ikey;
+  FindParseableKey(&ikey, kForward);
+  while (iter_->Valid() &&
+         user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) != 0) {
+    iter_->Next();
+    FindParseableKey(&ikey, kForward);
+  }
+}
+
+// Go to previous user_key
+void DBIter::FindPrevUserKey() {
+  if (!iter_->Valid()) {
+    return;
+  }
+  size_t num_skipped = 0;
+  ParsedInternalKey ikey;
+  FindParseableKey(&ikey, kReverse);
+  while (iter_->Valid() &&
+         user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) == 0) {
+    if (num_skipped >= max_skip_) {
+      num_skipped = 0;
+      IterKey last_key;
+      last_key.SetInternalKey(ParsedInternalKey(
+          saved_key_.GetKey(), kMaxSequenceNumber, kValueTypeForSeek));
+      iter_->Seek(last_key.GetKey());
+      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
+    }
+
+    iter_->Prev();
+    ++num_skipped;
+    FindParseableKey(&ikey, kReverse);
+  }
+}
+
+// Skip all unparseable keys
+void DBIter::FindParseableKey(ParsedInternalKey* ikey, Direction direction) {
+  while (iter_->Valid() && !ParseKey(ikey)) {
+    if (direction == kReverse) {
+      iter_->Prev();
+    } else {
+      iter_->Next();
+    }
   }
 }
 
 void DBIter::Seek(const Slice& target) {
+  StopWatch sw(env_, statistics_, DB_SEEK);
+
+  // total ordering is not guaranteed if prefix_extractor is set
+  // hence prefix based seeks will not give correct results
+  if (iterate_upper_bound_ != nullptr && prefix_extractor_ != nullptr) {
+    if (!prefix_extractor_->InDomain(*iterate_upper_bound_) ||
+        !prefix_extractor_->InDomain(target) ||
+        prefix_extractor_->Transform(*iterate_upper_bound_).compare(
+          prefix_extractor_->Transform(target)) != 0) {
+      status_ = Status::InvalidArgument("read_options.iterate_*_bound "
+                  " and seek target need to have the same prefix.");
+      valid_ = false;
+      return;
+    }
+  }
+
   saved_key_.Clear();
   // now savved_key is used to store internal key.
   saved_key_.SetInternalKey(target, sequence_);
-  PERF_TIMER_AUTO(seek_internal_seek_time);
-  iter_->Seek(saved_key_.GetKey());
-  PERF_TIMER_STOP(seek_internal_seek_time);
+
+  {
+    PERF_TIMER_GUARD(seek_internal_seek_time);
+    iter_->Seek(saved_key_.GetKey());
+  }
+
   if (iter_->Valid()) {
     direction_ = kForward;
     ClearSavedValue();
@@ -441,11 +599,19 @@ void DBIter::Seek(const Slice& target) {
 }
 
 void DBIter::SeekToFirst() {
+  // Don't use iter_::Seek() if we set a prefix extractor
+  // because prefix seek wiil be used.
+  if (prefix_extractor_ != nullptr) {
+    max_skip_ = std::numeric_limits<uint64_t>::max();
+  }
   direction_ = kForward;
   ClearSavedValue();
-  PERF_TIMER_AUTO(seek_internal_seek_time);
-  iter_->SeekToFirst();
-  PERF_TIMER_STOP(seek_internal_seek_time);
+
+  {
+    PERF_TIMER_GUARD(seek_internal_seek_time);
+    iter_->SeekToFirst();
+  }
+
   if (iter_->Valid()) {
     FindNextUserEntry(false /* not skipping */);
   } else {
@@ -454,28 +620,31 @@ void DBIter::SeekToFirst() {
 }
 
 void DBIter::SeekToLast() {
-  // Throw an exception for now if merge_operator is provided
-  // TODO: support backward iteration
-  if (user_merge_operator_) {
-    Log(logger_, "SeekToLast not supported yet if merge_operator is provided");
-    throw std::logic_error("DBIter::SeekToLast: backward iteration not"
-                           " supported if merge_operator is provided");
+  // Don't use iter_::Seek() if we set a prefix extractor
+  // because prefix seek wiil be used.
+  if (prefix_extractor_ != nullptr) {
+    max_skip_ = std::numeric_limits<uint64_t>::max();
   }
-
   direction_ = kReverse;
   ClearSavedValue();
-  PERF_TIMER_AUTO(seek_internal_seek_time);
-  iter_->SeekToLast();
-  PERF_TIMER_STOP(seek_internal_seek_time);
-  FindPrevUserEntry();
+
+  {
+    PERF_TIMER_GUARD(seek_internal_seek_time);
+    iter_->SeekToLast();
+  }
+
+  PrevInternal();
 }
 
-Iterator* NewDBIterator(Env* env, const Options& options,
+Iterator* NewDBIterator(Env* env, const ImmutableCFOptions& ioptions,
                         const Comparator* user_key_comparator,
                         Iterator* internal_iter,
-                        const SequenceNumber& sequence) {
-  return new DBIter(env, options, user_key_comparator, internal_iter, sequence,
-                    false);
+                        const SequenceNumber& sequence,
+                        uint64_t max_sequential_skip_in_iterations,
+                        const Slice* iterate_upper_bound) {
+  return new DBIter(env, ioptions, user_key_comparator, internal_iter, sequence,
+                    false, max_sequential_skip_in_iterations,
+                    iterate_upper_bound);
 }
 
 ArenaWrappedDBIter::~ArenaWrappedDBIter() { db_iter_->~DBIter(); }
@@ -503,14 +672,20 @@ void ArenaWrappedDBIter::RegisterCleanup(CleanupFunction function, void* arg1,
 }
 
 ArenaWrappedDBIter* NewArenaWrappedDbIterator(
-    Env* env, const Options& options, const Comparator* user_key_comparator,
-    const SequenceNumber& sequence) {
+    Env* env, const ImmutableCFOptions& ioptions,
+    const Comparator* user_key_comparator,
+    const SequenceNumber& sequence,
+    uint64_t max_sequential_skip_in_iterations,
+    const Slice* iterate_upper_bound) {
   ArenaWrappedDBIter* iter = new ArenaWrappedDBIter();
   Arena* arena = iter->GetArena();
   auto mem = arena->AllocateAligned(sizeof(DBIter));
-  DBIter* db_iter = new (mem)
-      DBIter(env, options, user_key_comparator, nullptr, sequence, true);
+  DBIter* db_iter = new (mem) DBIter(env, ioptions, user_key_comparator,
+      nullptr, sequence, true, max_sequential_skip_in_iterations,
+      iterate_upper_bound);
+
   iter->SetDBIter(db_iter);
+
   return iter;
 }
 

db/db_iter.h

@@ -24,10 +24,12 @@ class DBIter;
 // into appropriate user keys.
 extern Iterator* NewDBIterator(
     Env* env,
-    const Options& options,
+    const ImmutableCFOptions& options,
     const Comparator *user_key_comparator,
     Iterator* internal_iter,
-    const SequenceNumber& sequence);
+    const SequenceNumber& sequence,
+    uint64_t max_sequential_skip_in_iterations,
+    const Slice* iterate_upper_bound = nullptr);
 
 // A wrapper iterator which wraps DB Iterator and the arena, with which the DB
 // iterator is supposed be allocated. This class is used as an entry point of
@@ -67,7 +69,9 @@ class ArenaWrappedDBIter : public Iterator {
 
 // Generate the arena wrapped iterator class.
 extern ArenaWrappedDBIter* NewArenaWrappedDbIterator(
-    Env* env, const Options& options, const Comparator* user_key_comparator,
-    const SequenceNumber& sequence);
+    Env* env, const ImmutableCFOptions& options,
+    const Comparator* user_key_comparator,
+    const SequenceNumber& sequence, uint64_t max_sequential_skip_in_iterations,
+    const Slice* iterate_upper_bound = nullptr);
 
 }  // namespace rocksdb

db/db_stats_logger.cc

@@ -1,95 +0,0 @@
-//  Copyright (c) 2013, 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_impl.h"
-#include <string>
-#include <stdint.h>
-#include <stdio.h>
-#include "db/version_set.h"
-#include "rocksdb/db.h"
-#include "rocksdb/env.h"
-#include "port/port.h"
-#include "util/mutexlock.h"
-
-namespace rocksdb {
-
-void DBImpl::MaybeScheduleLogDBDeployStats() {
-// we did say maybe
-#ifndef ROCKSDB_LITE
-  // There is a lock in the actual logger.
-  if (!logger_ || options_.db_stats_log_interval < 0
-      || host_name_.empty()) {
-    return;
-  }
-
-  if(bg_logstats_scheduled_ || shutting_down_.Acquire_Load()) {
-    // Already scheduled
-  } else {
-    int64_t current_ts = 0;
-    Status st = env_->GetCurrentTime(&current_ts);
-    if (!st.ok()) {
-      return;
-    }
-    if ((current_ts - last_log_ts) < options_.db_stats_log_interval) {
-      return;
-    }
-    last_log_ts = current_ts;
-    bg_logstats_scheduled_ = true;
-    env_->Schedule(&DBImpl::BGLogDBDeployStats, this);
-  }
-}
-
-void DBImpl::BGLogDBDeployStats(void* db) {
-  DBImpl* db_inst = reinterpret_cast<DBImpl*>(db);
-  db_inst->LogDBDeployStats();
-}
-
-void DBImpl::LogDBDeployStats() {
-  mutex_.Lock();
-
-  if (shutting_down_.Acquire_Load()) {
-    bg_logstats_scheduled_ = false;
-    bg_cv_.SignalAll();
-    mutex_.Unlock();
-    return;
-  }
-
-  char tmp_ver[100];
-  sprintf(tmp_ver, "%d.%d", kMajorVersion, kMinorVersion);
-  std::string version_info(tmp_ver);
-
-  uint64_t file_total_size = 0;
-  uint32_t file_total_num = 0;
-  Version* current = default_cf_handle_->cfd()->current();
-  for (int i = 0; i < current->NumberLevels(); i++) {
-    file_total_num += current->NumLevelFiles(i);
-    file_total_size += current->NumLevelBytes(i);
-  }
-
-  Version::LevelSummaryStorage scratch;
-  const char* file_num_summary = current->LevelSummary(&scratch);
-  std::string file_num_per_level(file_num_summary);
-  std::string data_size_per_level(file_num_summary);
-
-  mutex_.Unlock();
-
-  int64_t unix_ts;
-  env_->GetCurrentTime(&unix_ts);
-
-  logger_->Log_Deploy_Stats(version_info, host_name_,
-      db_absolute_path_, file_total_size, file_total_num, file_num_per_level,
-      data_size_per_level, unix_ts);
-
-  mutex_.Lock();
-  bg_logstats_scheduled_ = false;
-  bg_cv_.SignalAll();
-  mutex_.Unlock();
-#endif
-}
-}

db/dbformat.cc

@@ -127,26 +127,6 @@ void InternalKeyComparator::FindShortSuccessor(std::string* key) const {
   }
 }
 
-const char* InternalFilterPolicy::Name() const {
-  return user_policy_->Name();
-}
-
-void InternalFilterPolicy::CreateFilter(const Slice* keys, int n,
-                                        std::string* dst) const {
-  // We rely on the fact that the code in table.cc does not mind us
-  // adjusting keys[].
-  Slice* mkey = const_cast<Slice*>(keys);
-  for (int i = 0; i < n; i++) {
-    mkey[i] = ExtractUserKey(keys[i]);
-    // TODO(sanjay): Suppress dups?
-  }
-  user_policy_->CreateFilter(keys, n, dst);
-}
-
-bool InternalFilterPolicy::KeyMayMatch(const Slice& key, const Slice& f) const {
-  return user_policy_->KeyMayMatch(ExtractUserKey(key), f);
-}
-
 LookupKey::LookupKey(const Slice& user_key, SequenceNumber s) {
   size_t usize = user_key.size();
   size_t needed = usize + 13;  // A conservative estimate

db/dbformat.h

@@ -9,6 +9,7 @@
 
 #pragma once
 #include <stdio.h>
+#include <string>
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/filter_policy.h"
@@ -123,17 +124,6 @@ class InternalKeyComparator : public Comparator {
   int Compare(const ParsedInternalKey& a, const ParsedInternalKey& b) const;
 };
 
-// Filter policy wrapper that converts from internal keys to user keys
-class InternalFilterPolicy : public FilterPolicy {
- private:
-  const FilterPolicy* const user_policy_;
- public:
-  explicit InternalFilterPolicy(const FilterPolicy* p) : user_policy_(p) { }
-  virtual const char* Name() const;
-  virtual void CreateFilter(const Slice* keys, int n, std::string* dst) const;
-  virtual bool KeyMayMatch(const Slice& key, const Slice& filter) const;
-};
-
 // Modules in this directory should keep internal keys wrapped inside
 // the following class instead of plain strings so that we do not
 // incorrectly use string comparisons instead of an InternalKeyComparator.
@@ -254,8 +244,38 @@ class IterKey {
 
   Slice GetKey() const { return Slice(key_, key_size_); }
 
+  size_t Size() { return key_size_; }
+
   void Clear() { key_size_ = 0; }
 
+  // Append "non_shared_data" to its back, from "shared_len"
+  // This function is used in Block::Iter::ParseNextKey
+  // shared_len: bytes in [0, shard_len-1] would be remained
+  // non_shared_data: data to be append, its length must be >= non_shared_len
+  void TrimAppend(const size_t shared_len, const char* non_shared_data,
+                  const size_t non_shared_len) {
+    assert(shared_len <= key_size_);
+
+    size_t total_size = shared_len + non_shared_len;
+    if (total_size <= buf_size_) {
+      key_size_ = total_size;
+    } else {
+      // Need to allocate space, delete previous space
+      char* p = new char[total_size];
+      memcpy(p, key_, shared_len);
+
+      if (key_ != nullptr && key_ != space_) {
+        delete[] key_;
+      }
+
+      key_ = p;
+      key_size_ = total_size;
+      buf_size_ = total_size;
+    }
+
+    memcpy(key_ + shared_len, non_shared_data, non_shared_len);
+  }
+
   void SetKey(const Slice& key) {
     size_t size = key.size();
     EnlargeBufferIfNeeded(size);
@@ -370,4 +390,12 @@ class InternalKeySliceTransform : public SliceTransform {
   const SliceTransform* const transform_;
 };
 
+// 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.
+// Tag is defined as ValueType.
+// 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);
 }  // namespace rocksdb

db/dbformat_test.cc

@@ -110,6 +110,45 @@ TEST(FormatTest, InternalKeyShortestSuccessor) {
             ShortSuccessor(IKey("\xff\xff", 100, kTypeValue)));
 }
 
+TEST(FormatTest, IterKeyOperation) {
+  IterKey k;
+  const char p[] = "abcdefghijklmnopqrstuvwxyz";
+  const char q[] = "0123456789";
+
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string(""));
+
+  k.TrimAppend(0, p, 3);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abc"));
+
+  k.TrimAppend(1, p, 3);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("aabc"));
+
+  k.TrimAppend(0, p, 26);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abcdefghijklmnopqrstuvwxyz"));
+
+  k.TrimAppend(26, q, 10);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abcdefghijklmnopqrstuvwxyz0123456789"));
+
+  k.TrimAppend(36, q, 1);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abcdefghijklmnopqrstuvwxyz01234567890"));
+
+  k.TrimAppend(26, q, 1);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abcdefghijklmnopqrstuvwxyz0"));
+
+  // Size going up, memory allocation is triggered
+  k.TrimAppend(27, p, 26);
+  ASSERT_EQ(std::string(k.GetKey().data(), k.GetKey().size()),
+            std::string("abcdefghijklmnopqrstuvwxyz0"
+              "abcdefghijklmnopqrstuvwxyz"));
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/deletefile_test.cc

@@ -34,6 +34,7 @@ class DeleteFileTest {
   DeleteFileTest() {
     db_ = nullptr;
     env_ = Env::Default();
+    options_.max_background_flushes = 0;
     options_.write_buffer_size = 1024*1024*1000;
     options_.target_file_size_base = 1024*1024*1000;
     options_.max_bytes_for_level_base = 1024*1024*1000;
@@ -147,7 +148,6 @@ class DeleteFileTest {
 TEST(DeleteFileTest, AddKeysAndQueryLevels) {
   CreateTwoLevels();
   std::vector<LiveFileMetaData> metadata;
-  std::vector<int> keysinlevel;
   db_->GetLiveFilesMetaData(&metadata);
 
   std::string level1file = "";

db/file_indexer.cc

@@ -14,21 +14,15 @@
 
 namespace rocksdb {
 
-FileIndexer::FileIndexer(const uint32_t num_levels,
-                         const Comparator* ucmp)
-  : num_levels_(num_levels),
-    ucmp_(ucmp),
-    next_level_index_(num_levels),
-    level_rb_(num_levels, -1) {
-}
-
+FileIndexer::FileIndexer(const Comparator* ucmp)
+    : num_levels_(0), ucmp_(ucmp), level_rb_(nullptr) {}
 
 uint32_t FileIndexer::NumLevelIndex() {
   return next_level_index_.size();
 }
 
 uint32_t FileIndexer::LevelIndexSize(uint32_t level) {
-  return next_level_index_[level].size();
+  return next_level_index_[level].num_index;
 }
 
 void FileIndexer::GetNextLevelIndex(
@@ -46,11 +40,13 @@ void FileIndexer::GetNextLevelIndex(
   assert(level < num_levels_ - 1);
   assert(static_cast<int32_t>(file_index) <= level_rb_[level]);
 
-  const auto& index = next_level_index_[level][file_index];
+  const IndexUnit* index_units = next_level_index_[level].index_units;
+  const auto& index = index_units[file_index];
 
   if (cmp_smallest < 0) {
-    *left_bound = (level > 0 && file_index > 0) ?
-      next_level_index_[level][file_index - 1].largest_lb : 0;
+    *left_bound = (level > 0 && file_index > 0)
+                      ? index_units[file_index - 1].largest_lb
+                      : 0;
     *right_bound = index.smallest_rb;
   } else if (cmp_smallest == 0) {
     *left_bound = index.smallest_lb;
@@ -73,16 +69,25 @@ void FileIndexer::GetNextLevelIndex(
   assert(*right_bound <= level_rb_[level + 1]);
 }
 
-void FileIndexer::ClearIndex() {
-  for (uint32_t level = 1; level < num_levels_; ++level) {
-    next_level_index_[level].clear();
-  }
-}
-
-void FileIndexer::UpdateIndex(std::vector<FileMetaData*>* const files) {
+void FileIndexer::UpdateIndex(Arena* arena, const uint32_t num_levels,
+                              std::vector<FileMetaData*>* const files) {
   if (files == nullptr) {
     return;
   }
+  if (num_levels == 0) {  // uint_32 0-1 would cause bad behavior
+    num_levels_ = num_levels;
+    return;
+  }
+  assert(level_rb_ == nullptr);  // level_rb_ should be init here
+
+  num_levels_ = num_levels;
+  next_level_index_.resize(num_levels);
+
+  char* mem = arena->AllocateAligned(num_levels_ * sizeof(int32_t));
+  level_rb_ = new (mem) int32_t[num_levels_];
+  for (size_t i = 0; i < num_levels_; i++) {
+    level_rb_[i] = -1;
+  }
 
   // L1 - Ln-1
   for (uint32_t level = 1; level < num_levels_ - 1; ++level) {
@@ -93,55 +98,56 @@ void FileIndexer::UpdateIndex(std::vector<FileMetaData*>* const files) {
     if (upper_size == 0) {
       continue;
     }
-    auto& index = next_level_index_[level];
-    index.resize(upper_size);
+    IndexLevel& index_level = next_level_index_[level];
+    index_level.num_index = upper_size;
+    char* mem = arena->AllocateAligned(upper_size * sizeof(IndexUnit));
+    index_level.index_units = new (mem) IndexUnit[upper_size];
 
-    CalculateLB(upper_files, lower_files, &index,
-        [this](const FileMetaData* a, const FileMetaData* b) -> int {
+    CalculateLB(
+        upper_files, lower_files, &index_level,
+        [this](const FileMetaData * a, const FileMetaData * b)->int {
           return ucmp_->Compare(a->smallest.user_key(), b->largest.user_key());
         },
-        [](IndexUnit* index, int32_t f_idx) {
-          index->smallest_lb = f_idx;
-        });
-    CalculateLB(upper_files, lower_files, &index,
-        [this](const FileMetaData* a, const FileMetaData* b) -> int {
+        [](IndexUnit* index, int32_t f_idx) { index->smallest_lb = f_idx; });
+    CalculateLB(
+        upper_files, lower_files, &index_level,
+        [this](const FileMetaData * a, const FileMetaData * b)->int {
           return ucmp_->Compare(a->largest.user_key(), b->largest.user_key());
         },
-        [](IndexUnit* index, int32_t f_idx) {
-          index->largest_lb = f_idx;
-        });
-    CalculateRB(upper_files, lower_files, &index,
-        [this](const FileMetaData* a, const FileMetaData* b) -> int {
+        [](IndexUnit* index, int32_t f_idx) { index->largest_lb = f_idx; });
+    CalculateRB(
+        upper_files, lower_files, &index_level,
+        [this](const FileMetaData * a, const FileMetaData * b)->int {
           return ucmp_->Compare(a->smallest.user_key(), b->smallest.user_key());
         },
-        [](IndexUnit* index, int32_t f_idx) {
-          index->smallest_rb = f_idx;
-        });
-    CalculateRB(upper_files, lower_files, &index,
-        [this](const FileMetaData* a, const FileMetaData* b) -> int {
+        [](IndexUnit* index, int32_t f_idx) { index->smallest_rb = f_idx; });
+    CalculateRB(
+        upper_files, lower_files, &index_level,
+        [this](const FileMetaData * a, const FileMetaData * b)->int {
           return ucmp_->Compare(a->largest.user_key(), b->smallest.user_key());
         },
-        [](IndexUnit* index, int32_t f_idx) {
-          index->largest_rb = f_idx;
-        });
+        [](IndexUnit* index, int32_t f_idx) { index->largest_rb = f_idx; });
   }
+
   level_rb_[num_levels_ - 1] = files[num_levels_ - 1].size() - 1;
 }
 
-void FileIndexer::CalculateLB(const std::vector<FileMetaData*>& upper_files,
-    const std::vector<FileMetaData*>& lower_files,
-    std::vector<IndexUnit>* index,
+void FileIndexer::CalculateLB(
+    const std::vector<FileMetaData*>& upper_files,
+    const std::vector<FileMetaData*>& lower_files, IndexLevel* index_level,
     std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
     std::function<void(IndexUnit*, int32_t)> set_index) {
   const int32_t upper_size = upper_files.size();
   const int32_t lower_size = lower_files.size();
   int32_t upper_idx = 0;
   int32_t lower_idx = 0;
+
+  IndexUnit* index = index_level->index_units;
   while (upper_idx < upper_size && lower_idx < lower_size) {
     int cmp = cmp_op(upper_files[upper_idx], lower_files[lower_idx]);
 
     if (cmp == 0) {
-      set_index(&(*index)[upper_idx], lower_idx);
+      set_index(&index[upper_idx], lower_idx);
       ++upper_idx;
       ++lower_idx;
     } else if (cmp > 0) {
@@ -151,7 +157,7 @@ void FileIndexer::CalculateLB(const std::vector<FileMetaData*>& upper_files,
     } else {
       // Lower level's file becomes larger, update the index, and
       // move to the next upper file
-      set_index(&(*index)[upper_idx], lower_idx);
+      set_index(&index[upper_idx], lower_idx);
       ++upper_idx;
     }
   }
@@ -159,25 +165,27 @@ void FileIndexer::CalculateLB(const std::vector<FileMetaData*>& upper_files,
   while (upper_idx < upper_size) {
     // Lower files are exhausted, that means the remaining upper files are
     // greater than any lower files. Set the index to be the lower level size.
-    set_index(&(*index)[upper_idx], lower_size);
+    set_index(&index[upper_idx], lower_size);
     ++upper_idx;
   }
 }
 
-void FileIndexer::CalculateRB(const std::vector<FileMetaData*>& upper_files,
-    const std::vector<FileMetaData*>& lower_files,
-    std::vector<IndexUnit>* index,
+void FileIndexer::CalculateRB(
+    const std::vector<FileMetaData*>& upper_files,
+    const std::vector<FileMetaData*>& lower_files, IndexLevel* index_level,
     std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
     std::function<void(IndexUnit*, int32_t)> set_index) {
   const int32_t upper_size = upper_files.size();
   const int32_t lower_size = lower_files.size();
   int32_t upper_idx = upper_size - 1;
   int32_t lower_idx = lower_size - 1;
+
+  IndexUnit* index = index_level->index_units;
   while (upper_idx >= 0 && lower_idx >= 0) {
     int cmp = cmp_op(upper_files[upper_idx], lower_files[lower_idx]);
 
     if (cmp == 0) {
-      set_index(&(*index)[upper_idx], lower_idx);
+      set_index(&index[upper_idx], lower_idx);
       --upper_idx;
       --lower_idx;
     } else if (cmp < 0) {
@@ -187,14 +195,14 @@ void FileIndexer::CalculateRB(const std::vector<FileMetaData*>& upper_files,
     } else {
       // Lower level's file becomes smaller, update the index, and move to
       // the next the upper file
-      set_index(&(*index)[upper_idx], lower_idx);
+      set_index(&index[upper_idx], lower_idx);
       --upper_idx;
     }
   }
   while (upper_idx >= 0) {
     // Lower files are exhausted, that means the remaining upper files are
     // smaller than any lower files. Set it to -1.
-    set_index(&(*index)[upper_idx], -1);
+    set_index(&index[upper_idx], -1);
     --upper_idx;
   }
 }

db/file_indexer.h

@@ -12,11 +12,15 @@
 #include <functional>
 #include <limits>
 #include <vector>
+#include "util/arena.h"
+#include "util/autovector.h"
 
 namespace rocksdb {
 
 class Comparator;
 struct FileMetaData;
+struct FdWithKeyRange;
+struct FileLevel;
 
 // The file tree structure in Version is prebuilt and the range of each file
 // is known. On Version::Get(), it uses binary search to find a potential file
@@ -36,7 +40,7 @@ struct FileMetaData;
 // naive approach.
 class FileIndexer {
  public:
-  FileIndexer(const uint32_t num_levels, const Comparator* ucmp);
+  explicit FileIndexer(const Comparator* ucmp);
 
   uint32_t NumLevelIndex();
 
@@ -50,16 +54,15 @@ class FileIndexer {
     const uint32_t level, const uint32_t file_index, const int cmp_smallest,
     const int cmp_largest, int32_t* left_bound, int32_t* right_bound);
 
-  void ClearIndex();
-
-  void UpdateIndex(std::vector<FileMetaData*>* const files);
+  void UpdateIndex(Arena* arena, const uint32_t num_levels,
+                   std::vector<FileMetaData*>* const files);
 
   enum {
     kLevelMaxIndex = std::numeric_limits<int32_t>::max()
   };
 
  private:
-  const uint32_t num_levels_;
+  uint32_t num_levels_;
   const Comparator* ucmp_;
 
   struct IndexUnit {
@@ -110,20 +113,28 @@ class FileIndexer {
     int32_t largest_rb;
   };
 
-  void CalculateLB(const std::vector<FileMetaData*>& upper_files,
-    const std::vector<FileMetaData*>& lower_files,
-    std::vector<IndexUnit>* index,
-    std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
-    std::function<void(IndexUnit*, int32_t)> set_index);
+  // Data structure to store IndexUnits in a whole level
+  struct IndexLevel {
+    size_t num_index;
+    IndexUnit* index_units;
 
-  void CalculateRB(const std::vector<FileMetaData*>& upper_files,
-    const std::vector<FileMetaData*>& lower_files,
-    std::vector<IndexUnit>* index,
-    std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
-    std::function<void(IndexUnit*, int32_t)> set_index);
+    IndexLevel() : num_index(0), index_units(nullptr) {}
+  };
 
-  std::vector<std::vector<IndexUnit>> next_level_index_;
-  std::vector<int32_t> level_rb_;
+  void CalculateLB(
+      const std::vector<FileMetaData*>& upper_files,
+      const std::vector<FileMetaData*>& lower_files, IndexLevel* index_level,
+      std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
+      std::function<void(IndexUnit*, int32_t)> set_index);
+
+  void CalculateRB(
+      const std::vector<FileMetaData*>& upper_files,
+      const std::vector<FileMetaData*>& lower_files, IndexLevel* index_level,
+      std::function<int(const FileMetaData*, const FileMetaData*)> cmp_op,
+      std::function<void(IndexUnit*, int32_t)> set_index);
+
+  autovector<IndexLevel> next_level_index_;
+  int32_t* level_rb_;
 };
 
 }  // namespace rocksdb

db/file_indexer_test.cc

@@ -35,16 +35,13 @@ class IntComparator : public Comparator {
   void FindShortSuccessor(std::string* key) const {}
 };
 
-
 struct FileIndexerTest {
  public:
-  FileIndexerTest() :
-    kNumLevels(4), indexer(kNumLevels, &ucmp),
-    files(new std::vector<FileMetaData*>[kNumLevels]) {
-  }
+  FileIndexerTest()
+      : kNumLevels(4), files(new std::vector<FileMetaData*>[kNumLevels]) {}
 
   ~FileIndexerTest() {
-    Reset();
+    ClearFiles();
     delete[] files;
   }
 
@@ -59,14 +56,13 @@ struct FileIndexerTest {
     return InternalKey(Slice(reinterpret_cast<char*>(&v), 8), 0, kTypeValue);
   }
 
-  void Reset() {
+  void ClearFiles() {
     for (uint32_t i = 0; i < kNumLevels; ++i) {
       for (auto* f : files[i]) {
         delete f;
       }
       files[i].clear();
     }
-    indexer.ClearIndex();
   }
 
   void GetNextLevelIndex(const uint32_t level, const uint32_t file_index,
@@ -74,23 +70,32 @@ struct FileIndexerTest {
       int32_t* right_index) {
     *left_index = 100;
     *right_index = 100;
-    indexer.GetNextLevelIndex(level, file_index, cmp_smallest, cmp_largest,
-        left_index, right_index);
+    indexer->GetNextLevelIndex(level, file_index, cmp_smallest, cmp_largest,
+                               left_index, right_index);
   }
 
+  int32_t left = 100;
+  int32_t right = 100;
   const uint32_t kNumLevels;
   IntComparator ucmp;
-  FileIndexer indexer;
+  FileIndexer* indexer;
 
   std::vector<FileMetaData*>* files;
 };
 
-TEST(FileIndexerTest, next_level_hint) {
-  for (uint32_t i = 0; i < kNumLevels; ++i) {
-    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
-  }
+// Case 0: Empty
+TEST(FileIndexerTest, Empty) {
+  Arena arena;
+  indexer = new FileIndexer(&ucmp);
+  indexer->UpdateIndex(&arena, 0, files);
+  delete indexer;
+}
 
-  // Case 1: no overlap, files are on the left of next level files
+// Case 1: no overlap, files are on the left of next level files
+TEST(FileIndexerTest, no_overlap_left) {
+  Arena arena;
+  uint32_t kNumLevels = 4;
+  indexer = new FileIndexer(&ucmp);
   // level 1
   AddFile(1, 100, 200);
   AddFile(1, 300, 400);
@@ -103,9 +108,7 @@ TEST(FileIndexerTest, next_level_hint) {
   AddFile(3, 2500, 2600);
   AddFile(3, 2601, 2699);
   AddFile(3, 2700, 2800);
-  indexer.UpdateIndex(files);
-  int32_t left = 100;
-  int32_t right = 100;
+  indexer->UpdateIndex(&arena, kNumLevels, files);
   for (uint32_t level = 1; level < 3; ++level) {
     for (uint32_t f = 0; f < 3; ++f) {
       GetNextLevelIndex(level, f, -1, -1, &left, &right);
@@ -125,12 +128,15 @@ TEST(FileIndexerTest, next_level_hint) {
       ASSERT_EQ(2, right);
     }
   }
+  delete indexer;
+  ClearFiles();
+}
 
-  // Case 2: no overlap, files are on the right of next level files
-  Reset();
-  for (uint32_t i = 1; i < kNumLevels; ++i) {
-    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
-  }
+// Case 2: no overlap, files are on the right of next level files
+TEST(FileIndexerTest, no_overlap_right) {
+  Arena arena;
+  uint32_t kNumLevels = 4;
+  indexer = new FileIndexer(&ucmp);
   // level 1
   AddFile(1, 2100, 2200);
   AddFile(1, 2300, 2400);
@@ -143,7 +149,7 @@ TEST(FileIndexerTest, next_level_hint) {
   AddFile(3, 500, 600);
   AddFile(3, 501, 699);
   AddFile(3, 700, 800);
-  indexer.UpdateIndex(files);
+  indexer->UpdateIndex(&arena, kNumLevels, files);
   for (uint32_t level = 1; level < 3; ++level) {
     for (uint32_t f = 0; f < 3; ++f) {
       GetNextLevelIndex(level, f, -1, -1, &left, &right);
@@ -166,11 +172,16 @@ TEST(FileIndexerTest, next_level_hint) {
       ASSERT_EQ(2, right);
     }
   }
+  delete indexer;
+}
 
-  // Case 3: empty L2
-  Reset();
+// Case 3: empty L2
+TEST(FileIndexerTest, empty_L2) {
+  Arena arena;
+  uint32_t kNumLevels = 4;
+  indexer = new FileIndexer(&ucmp);
   for (uint32_t i = 1; i < kNumLevels; ++i) {
-    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
+    ASSERT_EQ(0U, indexer->LevelIndexSize(i));
   }
   // level 1
   AddFile(1, 2100, 2200);
@@ -180,7 +191,7 @@ TEST(FileIndexerTest, next_level_hint) {
   AddFile(3, 500, 600);
   AddFile(3, 501, 699);
   AddFile(3, 700, 800);
-  indexer.UpdateIndex(files);
+  indexer->UpdateIndex(&arena, kNumLevels, files);
   for (uint32_t f = 0; f < 3; ++f) {
     GetNextLevelIndex(1, f, -1, -1, &left, &right);
     ASSERT_EQ(0, left);
@@ -201,13 +212,14 @@ TEST(FileIndexerTest, next_level_hint) {
     ASSERT_EQ(0, left);
     ASSERT_EQ(-1, right);
   }
+  delete indexer;
+  ClearFiles();
+}
 
-
-  // Case 4: mixed
-  Reset();
-  for (uint32_t i = 1; i < kNumLevels; ++i) {
-    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
-  }
+// Case 4: mixed
+TEST(FileIndexerTest, mixed) {
+  Arena arena;
+  indexer = new FileIndexer(&ucmp);
   // level 1
   AddFile(1, 100, 200);
   AddFile(1, 250, 400);
@@ -222,7 +234,7 @@ TEST(FileIndexerTest, next_level_hint) {
   AddFile(3, 0, 50);
   AddFile(3, 100, 200);
   AddFile(3, 201, 250);
-  indexer.UpdateIndex(files);
+  indexer->UpdateIndex(&arena, kNumLevels, files);
   // level 1, 0
   GetNextLevelIndex(1, 0, -1, -1, &left, &right);
   ASSERT_EQ(0, left);
@@ -321,6 +333,8 @@ TEST(FileIndexerTest, next_level_hint) {
     ASSERT_EQ(3, left);
     ASSERT_EQ(2, right);
   }
+  delete indexer;
+  ClearFiles();
 }
 
 }  // namespace rocksdb

db/filename.cc

@@ -6,8 +6,12 @@
 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
+#ifndef __STDC_FORMAT_MACROS
+#define __STDC_FORMAT_MACROS
+#endif
 
 #include "db/filename.h"
+#include <inttypes.h>
 
 #include <ctype.h>
 #include <stdio.h>
@@ -19,14 +23,16 @@
 namespace rocksdb {
 
 // Given a path, flatten the path name by replacing all chars not in
-// {[0-9,a-z,A-Z,-,_,.]} with _. And append '\0' at the end.
+// {[0-9,a-z,A-Z,-,_,.]} with _. And append '_LOG\0' at the end.
 // Return the number of chars stored in dest not including the trailing '\0'.
-static int FlattenPath(const std::string& path, char* dest, int len) {
-  int write_idx = 0;
-  int i = 0;
-  int src_len = path.size();
+static size_t GetInfoLogPrefix(const std::string& path, char* dest, int len) {
+  const char suffix[] = "_LOG";
 
-  while (i < src_len && write_idx < len - 1) {
+  size_t write_idx = 0;
+  size_t i = 0;
+  size_t src_len = path.size();
+
+  while (i < src_len && write_idx < len - sizeof(suffix)) {
     if ((path[i] >= 'a' && path[i] <= 'z') ||
         (path[i] >= '0' && path[i] <= '9') ||
         (path[i] >= 'A' && path[i] <= 'Z') ||
@@ -40,8 +46,10 @@ static int FlattenPath(const std::string& path, char* dest, int len) {
     }
     i++;
   }
-
-  dest[write_idx] = '\0';
+  assert(sizeof(suffix) <= len - write_idx);
+  // "\0" is automatically added by snprintf
+  snprintf(dest + write_idx, len - write_idx, suffix);
+  write_idx += sizeof(suffix) - 1;
   return write_idx;
 }
 
@@ -71,23 +79,29 @@ std::string MakeTableFileName(const std::string& path, uint64_t number) {
   return MakeFileName(path, number, "sst");
 }
 
-std::string TableFileName(const std::vector<std::string> db_paths,
-                          uint64_t number, uint32_t path_id) {
+std::string TableFileName(const std::vector<DbPath>& db_paths, uint64_t number,
+                          uint32_t path_id) {
   assert(number > 0);
   std::string path;
   if (path_id >= db_paths.size()) {
-    path = db_paths.back();
+    path = db_paths.back().path;
   } else {
-    path = db_paths[path_id];
+    path = db_paths[path_id].path;
   }
   return MakeTableFileName(path, number);
 }
 
-std::string FormatFileNumber(uint64_t number, uint32_t path_id) {
+const size_t kFormatFileNumberBufSize = 38;
+
+void FormatFileNumber(uint64_t number, uint32_t path_id, char* out_buf,
+                      size_t out_buf_size) {
   if (path_id == 0) {
-    return std::to_string(number);
+    snprintf(out_buf, out_buf_size, "%" PRIu64, number);
   } else {
-    return std::to_string(number) + "(path " + std::to_string(path_id) + ")";
+    snprintf(out_buf, out_buf_size, "%" PRIu64
+                                    "(path "
+                                    "%" PRIu32 ")",
+             number, path_id);
   }
 }
 
@@ -111,14 +125,26 @@ std::string TempFileName(const std::string& dbname, uint64_t number) {
   return MakeFileName(dbname, number, "dbtmp");
 }
 
+InfoLogPrefix::InfoLogPrefix(bool has_log_dir,
+                             const std::string& db_absolute_path) {
+  if (!has_log_dir) {
+    const char kInfoLogPrefix[] = "LOG";
+    // "\0" is automatically added to the end
+    snprintf(buf, sizeof(buf), kInfoLogPrefix);
+    prefix = Slice(buf, sizeof(kInfoLogPrefix) - 1);
+  } else {
+    size_t len = GetInfoLogPrefix(db_absolute_path, buf, sizeof(buf));
+    prefix = Slice(buf, len);
+  }
+}
+
 std::string InfoLogFileName(const std::string& dbname,
     const std::string& db_path, const std::string& log_dir) {
   if (log_dir.empty())
     return dbname + "/LOG";
 
-  char flatten_db_path[256];
-  FlattenPath(db_path, flatten_db_path, 256);
-  return log_dir + "/" + flatten_db_path + "_LOG";
+  InfoLogPrefix info_log_prefix(true, db_path);
+  return log_dir + "/" + info_log_prefix.buf;
 }
 
 // Return the name of the old info log file for "dbname".
@@ -130,9 +156,8 @@ std::string OldInfoLogFileName(const std::string& dbname, uint64_t ts,
   if (log_dir.empty())
     return dbname + "/LOG.old." + buf;
 
-  char flatten_db_path[256];
-  FlattenPath(db_path, flatten_db_path, 256);
-  return log_dir + "/" + flatten_db_path + "_LOG.old." + buf;
+  InfoLogPrefix info_log_prefix(true, db_path);
+  return log_dir + "/" + info_log_prefix.buf + ".old." + buf;
 }
 
 std::string MetaDatabaseName(const std::string& dbname, uint64_t number) {
@@ -150,8 +175,8 @@ std::string IdentityFileName(const std::string& dbname) {
 //    dbname/IDENTITY
 //    dbname/CURRENT
 //    dbname/LOCK
-//    dbname/LOG
-//    dbname/LOG.old.[0-9]+
+//    dbname/<info_log_name_prefix>
+//    dbname/<info_log_name_prefix>.old.[0-9]+
 //    dbname/MANIFEST-[0-9]+
 //    dbname/[0-9]+.(log|sst)
 //    dbname/METADB-[0-9]+
@@ -160,6 +185,12 @@ bool ParseFileName(const std::string& fname,
                    uint64_t* number,
                    FileType* type,
                    WalFileType* log_type) {
+  return ParseFileName(fname, number, "", type, log_type);
+}
+
+bool ParseFileName(const std::string& fname, uint64_t* number,
+                   const Slice& info_log_name_prefix, FileType* type,
+                   WalFileType* log_type) {
   Slice rest(fname);
   if (fname.length() > 1 && fname[0] == '/') {
     rest.remove_prefix(1);
@@ -173,18 +204,22 @@ bool ParseFileName(const std::string& fname,
   } else if (rest == "LOCK") {
     *number = 0;
     *type = kDBLockFile;
-  } else if (rest == "LOG" || rest == "LOG.old") {
-    *number = 0;
-    *type = kInfoLogFile;
-  } else if (rest.starts_with("LOG.old.")) {
-    uint64_t ts_suffix;
-    // sizeof also counts the trailing '\0'.
-    rest.remove_prefix(sizeof("LOG.old.") - 1);
-    if (!ConsumeDecimalNumber(&rest, &ts_suffix)) {
-      return false;
+  } else if (info_log_name_prefix.size() > 0 &&
+             rest.starts_with(info_log_name_prefix)) {
+    rest.remove_prefix(info_log_name_prefix.size());
+    if (rest == "" || rest == ".old") {
+      *number = 0;
+      *type = kInfoLogFile;
+    } else if (rest.starts_with(".old.")) {
+      uint64_t ts_suffix;
+      // sizeof also counts the trailing '\0'.
+      rest.remove_prefix(sizeof(".old.") - 1);
+      if (!ConsumeDecimalNumber(&rest, &ts_suffix)) {
+        return false;
+      }
+      *number = ts_suffix;
+      *type = kInfoLogFile;
     }
-    *number = ts_suffix;
-    *type = kInfoLogFile;
   } else if (rest.starts_with("MANIFEST-")) {
     rest.remove_prefix(strlen("MANIFEST-"));
     uint64_t num;

db/filename.h

@@ -58,10 +58,14 @@ extern std::string MakeTableFileName(const std::string& name, uint64_t number);
 // Return the name of the sstable with the specified number
 // in the db named by "dbname".  The result will be prefixed with
 // "dbname".
-extern std::string TableFileName(const std::vector<std::string> db_paths,
+extern std::string TableFileName(const std::vector<DbPath>& db_paths,
                                  uint64_t number, uint32_t path_id);
 
-extern std::string FormatFileNumber(uint64_t number, uint32_t path_id);
+// Sufficient buffer size for FormatFileNumber.
+extern const size_t kFormatFileNumberBufSize;
+
+extern void FormatFileNumber(uint64_t number, uint32_t path_id, char* out_buf,
+                             size_t out_buf_size);
 
 // Return the name of the descriptor file for the db named by
 // "dbname" and the specified incarnation number.  The result will be
@@ -82,6 +86,16 @@ extern std::string LockFileName(const std::string& dbname);
 // The result will be prefixed with "dbname".
 extern std::string TempFileName(const std::string& dbname, uint64_t number);
 
+// A helper structure for prefix of info log names.
+struct InfoLogPrefix {
+  char buf[260];
+  Slice prefix;
+  // Prefix with DB absolute path encoded
+  explicit InfoLogPrefix(bool has_log_dir, const std::string& db_absolute_path);
+  // Default Prefix
+  explicit InfoLogPrefix();
+};
+
 // Return the name of the info log file for "dbname".
 extern std::string InfoLogFileName(const std::string& dbname,
     const std::string& db_path="", const std::string& log_dir="");
@@ -103,10 +117,13 @@ extern std::string IdentityFileName(const std::string& dbname);
 // If filename is a rocksdb file, store the type of the file in *type.
 // The number encoded in the filename is stored in *number.  If the
 // filename was successfully parsed, returns true.  Else return false.
-extern bool ParseFileName(const std::string& filename,
-                          uint64_t* number,
-                          FileType* type,
+// info_log_name_prefix is the path of info logs.
+extern bool ParseFileName(const std::string& filename, uint64_t* number,
+                          const Slice& info_log_name_prefix, FileType* type,
                           WalFileType* log_type = nullptr);
+// Same as previous function, but skip info log files.
+extern bool ParseFileName(const std::string& filename, uint64_t* number,
+                          FileType* type, WalFileType* log_type = nullptr);
 
 // Make the CURRENT file point to the descriptor file with the
 // specified number.

db/filename_test.cc

@@ -23,30 +23,50 @@ TEST(FileNameTest, Parse) {
   FileType type;
   uint64_t number;
 
+  char kDefautInfoLogDir = 1;
+  char kDifferentInfoLogDir = 2;
+  char kNoCheckLogDir = 4;
+  char kAllMode = kDefautInfoLogDir | kDifferentInfoLogDir | kNoCheckLogDir;
+
   // Successful parses
   static struct {
     const char* fname;
     uint64_t number;
     FileType type;
+    char mode;
   } cases[] = {
-    { "100.log",            100,   kLogFile },
-    { "0.log",              0,     kLogFile },
-    { "0.sst",              0,     kTableFile },
-    { "CURRENT",            0,     kCurrentFile },
-    { "LOCK",               0,     kDBLockFile },
-    { "MANIFEST-2",         2,     kDescriptorFile },
-    { "MANIFEST-7",         7,     kDescriptorFile },
-    { "METADB-2",           2,     kMetaDatabase },
-    { "METADB-7",           7,     kMetaDatabase },
-    { "LOG",                0,     kInfoLogFile },
-    { "LOG.old",            0,     kInfoLogFile },
-    { "18446744073709551615.log", 18446744073709551615ull, kLogFile },
-  };
-  for (unsigned int i = 0; i < sizeof(cases) / sizeof(cases[0]); i++) {
-    std::string f = cases[i].fname;
-    ASSERT_TRUE(ParseFileName(f, &number, &type)) << f;
-    ASSERT_EQ(cases[i].type, type) << f;
-    ASSERT_EQ(cases[i].number, number) << f;
+        {"100.log", 100, kLogFile, kAllMode},
+        {"0.log", 0, kLogFile, kAllMode},
+        {"0.sst", 0, kTableFile, kAllMode},
+        {"CURRENT", 0, kCurrentFile, kAllMode},
+        {"LOCK", 0, kDBLockFile, kAllMode},
+        {"MANIFEST-2", 2, kDescriptorFile, kAllMode},
+        {"MANIFEST-7", 7, kDescriptorFile, kAllMode},
+        {"METADB-2", 2, kMetaDatabase, kAllMode},
+        {"METADB-7", 7, kMetaDatabase, kAllMode},
+        {"LOG", 0, kInfoLogFile, kDefautInfoLogDir},
+        {"LOG.old", 0, kInfoLogFile, kDefautInfoLogDir},
+        {"LOG.old.6688", 6688, kInfoLogFile, kDefautInfoLogDir},
+        {"rocksdb_dir_LOG", 0, kInfoLogFile, kDifferentInfoLogDir},
+        {"rocksdb_dir_LOG.old", 0, kInfoLogFile, kDifferentInfoLogDir},
+        {"rocksdb_dir_LOG.old.6688", 6688, kInfoLogFile, kDifferentInfoLogDir},
+        {"18446744073709551615.log", 18446744073709551615ull, kLogFile,
+         kAllMode}, };
+  for (char mode : {kDifferentInfoLogDir, kDefautInfoLogDir, kNoCheckLogDir}) {
+    for (unsigned int i = 0; i < sizeof(cases) / sizeof(cases[0]); i++) {
+      InfoLogPrefix info_log_prefix(mode != kDefautInfoLogDir, "/rocksdb/dir");
+      if (cases[i].mode & mode) {
+        std::string f = cases[i].fname;
+        if (mode == kNoCheckLogDir) {
+          ASSERT_TRUE(ParseFileName(f, &number, &type)) << f;
+        } else {
+          ASSERT_TRUE(ParseFileName(f, &number, info_log_prefix.prefix, &type))
+              << f;
+        }
+        ASSERT_EQ(cases[i].type, type) << f;
+        ASSERT_EQ(cases[i].number, number) << f;
+      }
+    }
   }
 
   // Errors
@@ -85,6 +105,22 @@ TEST(FileNameTest, Parse) {
   };
 }
 
+TEST(FileNameTest, InfoLogFileName) {
+  std::string dbname = ("/data/rocksdb");
+  std::string db_absolute_path;
+  Env::Default()->GetAbsolutePath(dbname, &db_absolute_path);
+
+  ASSERT_EQ("/data/rocksdb/LOG", InfoLogFileName(dbname, db_absolute_path, ""));
+  ASSERT_EQ("/data/rocksdb/LOG.old.666",
+            OldInfoLogFileName(dbname, 666u, db_absolute_path, ""));
+
+  ASSERT_EQ("/data/rocksdb_log/data_rocksdb_LOG",
+            InfoLogFileName(dbname, db_absolute_path, "/data/rocksdb_log"));
+  ASSERT_EQ(
+      "/data/rocksdb_log/data_rocksdb_LOG.old.666",
+      OldInfoLogFileName(dbname, 666u, db_absolute_path, "/data/rocksdb_log"));
+}
+
 TEST(FileNameTest, Construction) {
   uint64_t number;
   FileType type;
@@ -108,8 +144,9 @@ TEST(FileNameTest, Construction) {
   ASSERT_EQ(192U, number);
   ASSERT_EQ(kLogFile, type);
 
-  fname = TableFileName({"bar"}, 200, 0);
-  std::string fname1 = TableFileName({"foo", "bar"}, 200, 1);
+  fname = TableFileName({DbPath("bar", 0)}, 200, 0);
+  std::string fname1 =
+      TableFileName({DbPath("foo", 0), DbPath("bar", 0)}, 200, 1);
   ASSERT_EQ(fname, fname1);
   ASSERT_EQ("bar/", std::string(fname.data(), 4));
   ASSERT_TRUE(ParseFileName(fname.c_str() + 4, &number, &type));

db/flush_scheduler.cc

@@ -0,0 +1,63 @@
+//  Copyright (c) 2013, 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 "db/flush_scheduler.h"
+
+#include <cassert>
+
+#include "db/column_family.h"
+
+namespace rocksdb {
+
+void FlushScheduler::ScheduleFlush(ColumnFamilyData* cfd) {
+#ifndef NDEBUG
+  assert(column_families_set_.find(cfd) == column_families_set_.end());
+  column_families_set_.insert(cfd);
+#endif  // NDEBUG
+  cfd->Ref();
+  column_families_.push_back(cfd);
+}
+
+ColumnFamilyData* FlushScheduler::GetNextColumnFamily() {
+  ColumnFamilyData* cfd = nullptr;
+  while (column_families_.size() > 0) {
+    cfd = column_families_.front();
+    column_families_.pop_front();
+    if (cfd->IsDropped()) {
+      if (cfd->Unref()) {
+        delete cfd;
+        cfd = nullptr;
+      }
+    } else {
+      break;
+    }
+  }
+#ifndef NDEBUG
+  if (cfd != nullptr) {
+    auto itr = column_families_set_.find(cfd);
+    assert(itr != column_families_set_.end());
+    column_families_set_.erase(itr);
+  }
+#endif  // NDEBUG
+  return cfd;
+}
+
+bool FlushScheduler::Empty() { return column_families_.empty(); }
+
+void FlushScheduler::Clear() {
+  for (auto cfd : column_families_) {
+#ifndef NDEBUG
+    auto itr = column_families_set_.find(cfd);
+    assert(itr != column_families_set_.end());
+    column_families_set_.erase(itr);
+#endif  // NDEBUG
+    if (cfd->Unref()) {
+      delete cfd;
+    }
+  }
+  column_families_.clear();
+}
+
+}  // namespace rocksdb

db/flush_scheduler.h

@@ -0,0 +1,39 @@
+//  Copyright (c) 2013, 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 <stdint.h>
+#include <deque>
+#include <set>
+#include <vector>
+
+namespace rocksdb {
+
+class ColumnFamilyData;
+
+// This class is thread-compatible. It's should only be accessed from single
+// write thread (between BeginWrite() and EndWrite())
+class FlushScheduler {
+ public:
+  FlushScheduler() = default;
+  ~FlushScheduler() = default;
+
+  void ScheduleFlush(ColumnFamilyData* cfd);
+  // Returns Ref()-ed column family. Client needs to Unref()
+  ColumnFamilyData* GetNextColumnFamily();
+
+  bool Empty();
+
+  void Clear();
+
+ private:
+  std::deque<ColumnFamilyData*> column_families_;
+#ifndef NDEBUG
+  std::set<ColumnFamilyData*> column_families_set_;
+#endif  // NDEBUG
+};
+
+}  // namespace rocksdb

db/forward_iterator.cc

@@ -6,9 +6,10 @@
 #ifndef ROCKSDB_LITE
 #include "db/forward_iterator.h"
 
+#include <limits>
 #include <string>
 #include <utility>
-#include <limits>
+
 #include "db/db_impl.h"
 #include "db/db_iter.h"
 #include "db/column_family.h"
@@ -37,12 +38,16 @@ class LevelIterator : public Iterator {
     assert(file_index < files_.size());
     if (file_index != file_index_) {
       file_index_ = file_index;
-      file_iter_.reset(cfd_->table_cache()->NewIterator(
-          read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
-          files_[file_index_]->fd, nullptr /* table_reader_ptr */, false));
+      Reset();
     }
     valid_ = false;
   }
+  void Reset() {
+    assert(file_index_ < files_.size());
+    file_iter_.reset(cfd_->table_cache()->NewIterator(
+        read_options_, *(cfd_->soptions()), cfd_->internal_comparator(),
+        files_[file_index_]->fd, nullptr /* table_reader_ptr */, false));
+  }
   void SeekToLast() override {
     status_ = Status::NotSupported("LevelIterator::SeekToLast()");
     valid_ = false;
@@ -63,12 +68,15 @@ class LevelIterator : public Iterator {
     assert(file_iter_ != nullptr);
     file_iter_->Seek(internal_key);
     valid_ = file_iter_->Valid();
-    assert(valid_);
   }
   void Next() override {
     assert(valid_);
     file_iter_->Next();
-    while (!file_iter_->Valid()) {
+    for (;;) {
+      if (file_iter_->status().IsIncomplete() || file_iter_->Valid()) {
+        valid_ = !file_iter_->status().IsIncomplete();
+        return;
+      }
       if (file_index_ + 1 >= files_.size()) {
         valid_ = false;
         return;
@@ -76,7 +84,6 @@ class LevelIterator : public Iterator {
       SetFileIndex(file_index_ + 1);
       file_iter_->SeekToFirst();
     }
-    valid_ = file_iter_->Valid();
   }
   Slice key() const override {
     assert(valid_);
@@ -118,16 +125,19 @@ ForwardIterator::ForwardIterator(DBImpl* db, const ReadOptions& read_options,
       mutable_iter_(nullptr),
       current_(nullptr),
       valid_(false),
-      is_prev_set_(false) {}
+      is_prev_set_(false),
+      is_prev_inclusive_(false) {}
 
 ForwardIterator::~ForwardIterator() {
   Cleanup();
 }
 
 void ForwardIterator::Cleanup() {
-  delete mutable_iter_;
+  if (mutable_iter_ != nullptr) {
+    mutable_iter_->~Iterator();
+  }
   for (auto* m : imm_iters_) {
-    delete m;
+    m->~Iterator();
   }
   imm_iters_.clear();
   for (auto* f : l0_iters_) {
@@ -160,6 +170,8 @@ void ForwardIterator::SeekToFirst() {
   if (sv_ == nullptr ||
       sv_ ->version_number != cfd_->GetSuperVersionNumber()) {
     RebuildIterators();
+  } else if (status_.IsIncomplete()) {
+    ResetIncompleteIterators();
   }
   SeekInternal(Slice(), true);
 }
@@ -168,6 +180,8 @@ void ForwardIterator::Seek(const Slice& internal_key) {
   if (sv_ == nullptr ||
       sv_ ->version_number != cfd_->GetSuperVersionNumber()) {
     RebuildIterators();
+  } else if (status_.IsIncomplete()) {
+    ResetIncompleteIterators();
   }
   SeekInternal(internal_key, false);
 }
@@ -211,7 +225,15 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,
         }
         l0_iters_[i]->Seek(internal_key);
       }
-      if (l0_iters_[i]->Valid()) {
+
+      if (l0_iters_[i]->status().IsIncomplete()) {
+        // if any of the immutable iterators is incomplete (no-io option was
+        // used), we are unable to reliably find the smallest key
+        assert(read_options_.read_tier == kBlockCacheTier);
+        status_ = l0_iters_[i]->status();
+        valid_ = false;
+        return;
+      } else if (l0_iters_[i]->Valid()) {
         immutable_min_heap_.push(l0_iters_[i]);
       }
     }
@@ -280,18 +302,29 @@ void ForwardIterator::SeekInternal(const Slice& internal_key,
         level_iters_[level - 1]->SetFileIndex(f_idx);
         seek_to_first ? level_iters_[level - 1]->SeekToFirst() :
                         level_iters_[level - 1]->Seek(internal_key);
-        if (level_iters_[level - 1]->Valid()) {
+
+        if (level_iters_[level - 1]->status().IsIncomplete()) {
+          // see above
+          assert(read_options_.read_tier == kBlockCacheTier);
+          status_ = level_iters_[level - 1]->status();
+          valid_ = false;
+          return;
+        } else if (level_iters_[level - 1]->Valid()) {
           immutable_min_heap_.push(level_iters_[level - 1]);
         }
       }
     }
 
-    if (seek_to_first || immutable_min_heap_.empty()) {
+    if (seek_to_first) {
       is_prev_set_ = false;
     } else {
       prev_key_.SetKey(internal_key);
       is_prev_set_ = true;
+      is_prev_inclusive_ = true;
     }
+  } else if (current_ && current_ != mutable_iter_) {
+    // current_ is one of immutable iterators, push it back to the heap
+    immutable_min_heap_.push(current_);
   }
 
   UpdateCurrent();
@@ -301,7 +334,7 @@ void ForwardIterator::Next() {
   assert(valid_);
 
   if (sv_ == nullptr ||
-      sv_ ->version_number != cfd_->GetSuperVersionNumber()) {
+      sv_->version_number != cfd_->GetSuperVersionNumber()) {
     std::string current_key = key().ToString();
     Slice old_key(current_key.data(), current_key.size());
 
@@ -312,14 +345,34 @@ void ForwardIterator::Next() {
     }
   } else if (current_ != mutable_iter_) {
     // It is going to advance immutable iterator
-    prev_key_.SetKey(current_->key());
-    is_prev_set_ = true;
+
+    bool update_prev_key = true;
+    if (is_prev_set_ && prefix_extractor_) {
+      // advance prev_key_ to current_ only if they share the same prefix
+      update_prev_key =
+        prefix_extractor_->Transform(prev_key_.GetKey()).compare(
+          prefix_extractor_->Transform(current_->key())) == 0;
+    }
+
+    if (update_prev_key) {
+      prev_key_.SetKey(current_->key());
+      is_prev_set_ = true;
+      is_prev_inclusive_ = false;
+    }
   }
 
   current_->Next();
-  if (current_->Valid() && current_ != mutable_iter_) {
-    immutable_min_heap_.push(current_);
+  if (current_ != mutable_iter_) {
+    if (current_->status().IsIncomplete()) {
+      assert(read_options_.read_tier == kBlockCacheTier);
+      status_ = current_->status();
+      valid_ = false;
+      return;
+    } else if (current_->Valid()) {
+      immutable_min_heap_.push(current_);
+    }
   }
+
   UpdateCurrent();
 }
 
@@ -364,8 +417,8 @@ void ForwardIterator::RebuildIterators() {
   Cleanup();
   // New
   sv_ = cfd_->GetReferencedSuperVersion(&(db_->mutex_));
-  mutable_iter_ = sv_->mem->NewIterator(read_options_);
-  sv_->imm->AddIterators(read_options_, &imm_iters_);
+  mutable_iter_ = sv_->mem->NewIterator(read_options_, &arena_);
+  sv_->imm->AddIterators(read_options_, &imm_iters_, &arena_);
   const auto& l0_files = sv_->current->files_[0];
   l0_iters_.reserve(l0_files.size());
   for (const auto* l0 : l0_files) {
@@ -386,6 +439,29 @@ void ForwardIterator::RebuildIterators() {
   is_prev_set_ = false;
 }
 
+void ForwardIterator::ResetIncompleteIterators() {
+  const auto& l0_files = sv_->current->files_[0];
+  for (uint32_t i = 0; i < l0_iters_.size(); ++i) {
+    assert(i < l0_files.size());
+    if (!l0_iters_[i]->status().IsIncomplete()) {
+      continue;
+    }
+    delete l0_iters_[i];
+    l0_iters_[i] = cfd_->table_cache()->NewIterator(
+        read_options_, *cfd_->soptions(), cfd_->internal_comparator(),
+        l0_files[i]->fd);
+  }
+
+  for (auto* level_iter : level_iters_) {
+    if (level_iter && level_iter->status().IsIncomplete()) {
+      level_iter->Reset();
+    }
+  }
+
+  current_ = nullptr;
+  is_prev_set_ = false;
+}
+
 void ForwardIterator::UpdateCurrent() {
   if (immutable_min_heap_.empty() && !mutable_iter_->Valid()) {
     current_ = nullptr;
@@ -414,7 +490,14 @@ void ForwardIterator::UpdateCurrent() {
 }
 
 bool ForwardIterator::NeedToSeekImmutable(const Slice& target) {
-  if (!is_prev_set_) {
+  // We maintain the interval (prev_key_, immutable_min_heap_.top()->key())
+  // such that there are no records with keys within that range in
+  // immutable_min_heap_. Since immutable structures (SST files and immutable
+  // memtables) can't change in this version, we don't need to do a seek if
+  // 'target' belongs to that interval (immutable_min_heap_.top() is already
+  // at the correct position).
+
+  if (!valid_ || !current_ || !is_prev_set_) {
     return true;
   }
   Slice prev_key = prev_key_.GetKey();
@@ -423,13 +506,17 @@ bool ForwardIterator::NeedToSeekImmutable(const Slice& target) {
     return true;
   }
   if (cfd_->internal_comparator().InternalKeyComparator::Compare(
-        prev_key, target) >= 0) {
+        prev_key, target) >= (is_prev_inclusive_ ? 1 : 0)) {
     return true;
   }
-  if (immutable_min_heap_.empty() ||
-      cfd_->internal_comparator().InternalKeyComparator::Compare(
-          target, current_ == mutable_iter_ ? immutable_min_heap_.top()->key()
-                                            : current_->key()) > 0) {
+
+  if (immutable_min_heap_.empty() && current_ == mutable_iter_) {
+    // Nothing to seek on.
+    return false;
+  }
+  if (cfd_->internal_comparator().InternalKeyComparator::Compare(
+        target, current_ == mutable_iter_ ? immutable_min_heap_.top()->key()
+                                          : current_->key()) > 0) {
     return true;
   }
   return false;

db/forward_iterator.h

@@ -14,6 +14,7 @@
 #include "rocksdb/iterator.h"
 #include "rocksdb/options.h"
 #include "db/dbformat.h"
+#include "util/arena.h"
 
 namespace rocksdb {
 
@@ -73,6 +74,7 @@ class ForwardIterator : public Iterator {
  private:
   void Cleanup();
   void RebuildIterators();
+  void ResetIncompleteIterators();
   void SeekInternal(const Slice& internal_key, bool seek_to_first);
   void UpdateCurrent();
   bool NeedToSeekImmutable(const Slice& internal_key);
@@ -99,6 +101,8 @@ class ForwardIterator : public Iterator {
 
   IterKey prev_key_;
   bool is_prev_set_;
+  bool is_prev_inclusive_;
+  Arena arena_;
 };
 
 }  // namespace rocksdb

db/internal_stats.cc

@@ -7,16 +7,105 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "db/internal_stats.h"
-#include "db/column_family.h"
 
+#ifndef __STDC_FORMAT_MACROS
+#define __STDC_FORMAT_MACROS
+#endif
+
+#include <inttypes.h>
 #include <vector>
+#include "db/column_family.h"
+#include "db/db_impl.h"
 
 namespace rocksdb {
 
-DBPropertyType GetPropertyType(const Slice& property) {
+namespace {
+const double kMB = 1048576.0;
+const double kGB = kMB * 1024;
+
+void PrintLevelStatsHeader(char* buf, size_t len, const std::string& cf_name) {
+  snprintf(
+      buf, len,
+      "\n** Compaction Stats [%s] **\n"
+      "Level   Files   Size(MB) Score Read(GB)  Rn(GB) Rnp1(GB) "
+      "Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s)  Rn(cnt) "
+      "Rnp1(cnt) Wnp1(cnt) Wnew(cnt)  Comp(sec) Comp(cnt) Avg(sec) "
+      "Stall(sec) Stall(cnt) Avg(ms) RecordIn RecordDrop\n"
+      "--------------------------------------------------------------------"
+      "--------------------------------------------------------------------"
+      "--------------------------------------------------------------------\n",
+      cf_name.c_str());
+}
+
+void PrintLevelStats(char* buf, size_t len, const std::string& name,
+    int num_files, int being_compacted, double total_file_size, double score,
+    double rw_amp, double w_amp, double stall_us, uint64_t stalls,
+    const InternalStats::CompactionStats& stats) {
+  uint64_t bytes_read = stats.bytes_readn + stats.bytes_readnp1;
+  uint64_t bytes_new = stats.bytes_written - stats.bytes_readnp1;
+  double elapsed = (stats.micros + 1) / 1000000.0;
+
+  snprintf(buf, len,
+    "%4s %5d/%-3d %8.0f %5.1f " /* Level, Files, Size(MB), Score */
+    "%8.1f " /* Read(GB) */
+    "%7.1f " /* Rn(GB) */
+    "%8.1f " /* Rnp1(GB) */
+    "%9.1f " /* Write(GB) */
+    "%8.1f " /* Wnew(GB) */
+    "%6.1f " /* RW-Amp */
+    "%5.1f " /* W-Amp */
+    "%8.1f " /* Rd(MB/s) */
+    "%8.1f " /* Wr(MB/s) */
+    "%8d " /* Rn(cnt) */
+    "%9d " /* Rnp1(cnt) */
+    "%9d " /* Wnp1(cnt) */
+    "%9d " /* Wnew(cnt) */
+    "%10.0f " /* Comp(sec) */
+    "%9d " /* Comp(cnt) */
+    "%8.3f " /* Avg(sec) */
+    "%10.2f " /* Stall(sec) */
+    "%10" PRIu64 " " /* Stall(cnt) */
+    "%7.2f" /* Avg(ms) */
+    "%8d " /* input entries */
+    "%10d\n" /* number of records reduced */,
+    name.c_str(), num_files, being_compacted, total_file_size / kMB, score,
+    bytes_read / kGB,
+    stats.bytes_readn / kGB,
+    stats.bytes_readnp1 / kGB,
+    stats.bytes_written / kGB,
+    bytes_new / kGB,
+    rw_amp,
+    w_amp,
+    bytes_read / kMB / elapsed,
+    stats.bytes_written / kMB / elapsed,
+    stats.files_in_leveln,
+    stats.files_in_levelnp1,
+    stats.files_out_levelnp1,
+    stats.files_out_levelnp1 - stats.files_in_levelnp1,
+    stats.micros / 1000000.0,
+    stats.count,
+    stats.count == 0 ? 0 : stats.micros / 1000000.0 / stats.count,
+    stall_us / 1000000.0,
+    stalls,
+    stalls == 0 ? 0 : stall_us / 1000.0 / stalls,
+    stats.num_input_records,
+    stats.num_dropped_records);
+}
+
+
+}
+
+DBPropertyType GetPropertyType(const Slice& property, bool* is_int_property,
+                               bool* need_out_of_mutex) {
+  assert(is_int_property != nullptr);
+  assert(need_out_of_mutex != nullptr);
   Slice in = property;
   Slice prefix("rocksdb.");
-  if (!in.starts_with(prefix)) return kUnknown;
+  *need_out_of_mutex = false;
+  *is_int_property = false;
+  if (!in.starts_with(prefix)) {
+    return kUnknown;
+  }
   in.remove_prefix(prefix.size());
 
   if (in.starts_with("num-files-at-level")) {
@@ -25,9 +114,16 @@ DBPropertyType GetPropertyType(const Slice& property) {
     return kLevelStats;
   } else if (in == "stats") {
     return kStats;
+  } else if (in == "cfstats") {
+    return kCFStats;
+  } else if (in == "dbstats") {
+    return kDBStats;
   } else if (in == "sstables") {
     return kSsTables;
-  } else if (in == "num-immutable-mem-table") {
+  }
+
+  *is_int_property = true;
+  if (in == "num-immutable-mem-table") {
     return kNumImmutableMemTable;
   } else if (in == "mem-table-flush-pending") {
     return kMemtableFlushPending;
@@ -41,14 +137,37 @@ DBPropertyType GetPropertyType(const Slice& property) {
     return kNumEntriesInMutableMemtable;
   } else if (in == "num-entries-imm-mem-tables") {
     return kNumEntriesInImmutableMemtable;
+  } else if (in == "estimate-num-keys") {
+    return kEstimatedNumKeys;
+  } else if (in == "estimate-table-readers-mem") {
+    *need_out_of_mutex = true;
+    return kEstimatedUsageByTableReaders;
+  } else if (in == "is-file-deletions-enabled") {
+    return kIsFileDeletionEnabled;
   }
   return kUnknown;
 }
 
-bool InternalStats::GetProperty(DBPropertyType property_type,
-                                const Slice& property, std::string* value,
-                                ColumnFamilyData* cfd) {
-  Version* current = cfd->current();
+bool InternalStats::GetIntPropertyOutOfMutex(DBPropertyType property_type,
+                                             Version* version,
+                                             uint64_t* value) const {
+  assert(value != nullptr);
+  if (property_type != kEstimatedUsageByTableReaders) {
+    return false;
+  }
+  if (version == nullptr) {
+    *value = 0;
+  } else {
+    *value = version->GetMemoryUsageByTableReaders();
+  }
+  return true;
+}
+
+bool InternalStats::GetStringProperty(DBPropertyType property_type,
+                                      const Slice& property,
+                                      std::string* value) {
+  assert(value != nullptr);
+  Version* current = cfd_->current();
   Slice in = property;
 
   switch (property_type) {
@@ -76,294 +195,292 @@ bool InternalStats::GetProperty(DBPropertyType property_type,
       for (int level = 0; level < number_levels_; level++) {
         snprintf(buf, sizeof(buf), "%3d %8d %8.0f\n", level,
                  current->NumLevelFiles(level),
-                 current->NumLevelBytes(level) / 1048576.0);
+                 current->NumLevelBytes(level) / kMB);
         value->append(buf);
       }
       return true;
     }
     case kStats: {
-      char buf[1000];
-
-      uint64_t wal_bytes = 0;
-      uint64_t wal_synced = 0;
-      uint64_t user_bytes_written = 0;
-      uint64_t write_other = 0;
-      uint64_t write_self = 0;
-      uint64_t write_with_wal = 0;
-      uint64_t total_bytes_written = 0;
-      uint64_t total_bytes_read = 0;
-      uint64_t micros_up = env_->NowMicros() - started_at_;
-      // Add "+1" to make sure seconds_up is > 0 and avoid NaN later
-      double seconds_up = (micros_up + 1) / 1000000.0;
-      uint64_t total_slowdown = 0;
-      uint64_t total_slowdown_count = 0;
-      uint64_t interval_bytes_written = 0;
-      uint64_t interval_bytes_read = 0;
-      uint64_t interval_bytes_new = 0;
-      double interval_seconds_up = 0;
-
-      if (statistics_) {
-        wal_bytes = statistics_->getTickerCount(WAL_FILE_BYTES);
-        wal_synced = statistics_->getTickerCount(WAL_FILE_SYNCED);
-        user_bytes_written = statistics_->getTickerCount(BYTES_WRITTEN);
-        write_other = statistics_->getTickerCount(WRITE_DONE_BY_OTHER);
-        write_self = statistics_->getTickerCount(WRITE_DONE_BY_SELF);
-        write_with_wal = statistics_->getTickerCount(WRITE_WITH_WAL);
+      if (!GetStringProperty(kCFStats, "rocksdb.cfstats", value)) {
+        return false;
       }
-
-      snprintf(
-          buf, sizeof(buf),
-          "                               Compactions\n"
-          "Level  Files Size(MB) Score Time(sec)  Read(MB) Write(MB)    Rn(MB) "
-          " "
-          "Rnp1(MB)  Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s)      Rn     "
-          "Rnp1 "
-          "    Wnp1     NewW    Count   msComp   msStall  Ln-stall Stall-cnt\n"
-          "--------------------------------------------------------------------"
-          "--"
-          "--------------------------------------------------------------------"
-          "--"
-          "----------------------------------------------------------------\n");
-      value->append(buf);
-      for (int level = 0; level < number_levels_; level++) {
-        int files = current->NumLevelFiles(level);
-        if (compaction_stats_[level].micros > 0 || files > 0) {
-          int64_t bytes_read = compaction_stats_[level].bytes_readn +
-                               compaction_stats_[level].bytes_readnp1;
-          int64_t bytes_new = compaction_stats_[level].bytes_written -
-                              compaction_stats_[level].bytes_readnp1;
-          double amplify =
-              (compaction_stats_[level].bytes_readn == 0)
-                  ? 0.0
-                  : (compaction_stats_[level].bytes_written +
-                     compaction_stats_[level].bytes_readnp1 +
-                     compaction_stats_[level].bytes_readn) /
-                        (double)compaction_stats_[level].bytes_readn;
-
-          total_bytes_read += bytes_read;
-          total_bytes_written += compaction_stats_[level].bytes_written;
-
-          uint64_t stalls = level == 0 ? (stall_counts_[LEVEL0_SLOWDOWN] +
-                                          stall_counts_[LEVEL0_NUM_FILES] +
-                                          stall_counts_[MEMTABLE_COMPACTION])
-                                       : stall_leveln_slowdown_count_[level];
-
-          double stall_us = level == 0 ? (stall_micros_[LEVEL0_SLOWDOWN] +
-                                          stall_micros_[LEVEL0_NUM_FILES] +
-                                          stall_micros_[MEMTABLE_COMPACTION])
-                                       : stall_leveln_slowdown_[level];
-
-          snprintf(buf, sizeof(buf),
-                   "%3d %8d %8.0f %5.1f %9.0f %9.0f %9.0f %9.0f %9.0f %9.0f "
-                   "%10.1f %9.1f %11.1f %8d %8d %8d %8d %8d %8d %9.1f %9.1f "
-                   "%9lu\n",
-                   level, files, current->NumLevelBytes(level) / 1048576.0,
-                   current->NumLevelBytes(level) /
-                       cfd->compaction_picker()->MaxBytesForLevel(level),
-                   compaction_stats_[level].micros / 1e6,
-                   bytes_read / 1048576.0,
-                   compaction_stats_[level].bytes_written / 1048576.0,
-                   compaction_stats_[level].bytes_readn / 1048576.0,
-                   compaction_stats_[level].bytes_readnp1 / 1048576.0,
-                   bytes_new / 1048576.0, amplify,
-                   // +1 to avoid division by 0
-                   (bytes_read / 1048576.0) /
-                       ((compaction_stats_[level].micros + 1) / 1000000.0),
-                   (compaction_stats_[level].bytes_written / 1048576.0) /
-                       ((compaction_stats_[level].micros + 1) / 1000000.0),
-                   compaction_stats_[level].files_in_leveln,
-                   compaction_stats_[level].files_in_levelnp1,
-                   compaction_stats_[level].files_out_levelnp1,
-                   compaction_stats_[level].files_out_levelnp1 -
-                       compaction_stats_[level].files_in_levelnp1,
-                   compaction_stats_[level].count,
-                   (int)((double)compaction_stats_[level].micros / 1000.0 /
-                         (compaction_stats_[level].count + 1)),
-                   (double)stall_us / 1000.0 / (stalls + 1),
-                   stall_us / 1000000.0, (unsigned long)stalls);
-          total_slowdown += stall_leveln_slowdown_[level];
-          total_slowdown_count += stall_leveln_slowdown_count_[level];
-          value->append(buf);
-        }
+      if (!GetStringProperty(kDBStats, "rocksdb.dbstats", value)) {
+        return false;
       }
-
-      interval_bytes_new = user_bytes_written - last_stats_.ingest_bytes_;
-      interval_bytes_read =
-          total_bytes_read - last_stats_.compaction_bytes_read_;
-      interval_bytes_written =
-          total_bytes_written - last_stats_.compaction_bytes_written_;
-      interval_seconds_up = seconds_up - last_stats_.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),
-               "Writes cumulative: %llu total, %llu batches, "
-               "%.1f per batch, %.2f ingest GB\n",
-               (unsigned long long)(write_other + write_self),
-               (unsigned long long)write_self,
-               (write_other + write_self) / (double)(write_self + 1),
-               user_bytes_written / (1048576.0 * 1024));
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "WAL cumulative: %llu WAL writes, %llu WAL syncs, "
-               "%.2f writes per sync, %.2f GB written\n",
-               (unsigned long long)write_with_wal,
-               (unsigned long long)wal_synced,
-               write_with_wal / (double)(wal_synced + 1),
-               wal_bytes / (1048576.0 * 1024));
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "Compaction IO cumulative (GB): "
-               "%.2f new, %.2f read, %.2f write, %.2f read+write\n",
-               user_bytes_written / (1048576.0 * 1024),
-               total_bytes_read / (1048576.0 * 1024),
-               total_bytes_written / (1048576.0 * 1024),
-               (total_bytes_read + total_bytes_written) / (1048576.0 * 1024));
-      value->append(buf);
-
-      snprintf(
-          buf, sizeof(buf),
-          "Compaction IO cumulative (MB/sec): "
-          "%.1f new, %.1f read, %.1f write, %.1f read+write\n",
-          user_bytes_written / 1048576.0 / seconds_up,
-          total_bytes_read / 1048576.0 / seconds_up,
-          total_bytes_written / 1048576.0 / seconds_up,
-          (total_bytes_read + total_bytes_written) / 1048576.0 / seconds_up);
-      value->append(buf);
-
-      // +1 to avoid divide by 0 and NaN
-      snprintf(
-          buf, sizeof(buf),
-          "Amplification cumulative: %.1f write, %.1f compaction\n",
-          (double)(total_bytes_written + wal_bytes) / (user_bytes_written + 1),
-          (double)(total_bytes_written + total_bytes_read + wal_bytes) /
-              (user_bytes_written + 1));
-      value->append(buf);
-
-      uint64_t interval_write_other = write_other - last_stats_.write_other_;
-      uint64_t interval_write_self = write_self - last_stats_.write_self_;
-
-      snprintf(buf, sizeof(buf),
-               "Writes interval: %llu total, %llu batches, "
-               "%.1f per batch, %.1f ingest MB\n",
-               (unsigned long long)(interval_write_other + interval_write_self),
-               (unsigned long long)interval_write_self,
-               (double)(interval_write_other + interval_write_self) /
-                   (interval_write_self + 1),
-               (user_bytes_written - last_stats_.ingest_bytes_) / 1048576.0);
-      value->append(buf);
-
-      uint64_t interval_write_with_wal =
-          write_with_wal - last_stats_.write_with_wal_;
-
-      uint64_t interval_wal_synced = wal_synced - last_stats_.wal_synced_;
-      uint64_t interval_wal_bytes = wal_bytes - last_stats_.wal_bytes_;
-
-      snprintf(buf, sizeof(buf),
-               "WAL interval: %llu WAL writes, %llu WAL syncs, "
-               "%.2f writes per sync, %.2f MB written\n",
-               (unsigned long long)interval_write_with_wal,
-               (unsigned long long)interval_wal_synced,
-               interval_write_with_wal / (double)(interval_wal_synced + 1),
-               interval_wal_bytes / (1048576.0 * 1024));
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "Compaction IO interval (MB): "
-               "%.2f new, %.2f read, %.2f write, %.2f read+write\n",
-               interval_bytes_new / 1048576.0, interval_bytes_read / 1048576.0,
-               interval_bytes_written / 1048576.0,
-               (interval_bytes_read + interval_bytes_written) / 1048576.0);
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "Compaction IO interval (MB/sec): "
-               "%.1f new, %.1f read, %.1f write, %.1f read+write\n",
-               interval_bytes_new / 1048576.0 / interval_seconds_up,
-               interval_bytes_read / 1048576.0 / interval_seconds_up,
-               interval_bytes_written / 1048576.0 / interval_seconds_up,
-               (interval_bytes_read + interval_bytes_written) / 1048576.0 /
-                   interval_seconds_up);
-      value->append(buf);
-
-      // +1 to avoid divide by 0 and NaN
-      snprintf(
-          buf, sizeof(buf),
-          "Amplification interval: %.1f write, %.1f compaction\n",
-          (double)(interval_bytes_written + interval_wal_bytes) /
-              (interval_bytes_new + 1),
-          (double)(interval_bytes_written + interval_bytes_read + interval_wal_bytes) /
-              (interval_bytes_new + 1));
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "Stalls(secs): %.3f level0_slowdown, %.3f level0_numfiles, "
-               "%.3f memtable_compaction, %.3f leveln_slowdown\n",
-               stall_micros_[LEVEL0_SLOWDOWN] / 1000000.0,
-               stall_micros_[LEVEL0_NUM_FILES] / 1000000.0,
-               stall_micros_[MEMTABLE_COMPACTION] / 1000000.0,
-               total_slowdown / 1000000.0);
-      value->append(buf);
-
-      snprintf(buf, sizeof(buf),
-               "Stalls(count): %lu level0_slowdown, %lu level0_numfiles, "
-               "%lu memtable_compaction, %lu leveln_slowdown\n",
-               (unsigned long)stall_counts_[LEVEL0_SLOWDOWN],
-               (unsigned long)stall_counts_[LEVEL0_NUM_FILES],
-               (unsigned long)stall_counts_[MEMTABLE_COMPACTION],
-               (unsigned long)total_slowdown_count);
-      value->append(buf);
-
-      last_stats_.compaction_bytes_read_ = total_bytes_read;
-      last_stats_.compaction_bytes_written_ = total_bytes_written;
-      last_stats_.ingest_bytes_ = user_bytes_written;
-      last_stats_.seconds_up_ = seconds_up;
-      last_stats_.wal_bytes_ = wal_bytes;
-      last_stats_.wal_synced_ = wal_synced;
-      last_stats_.write_with_wal_ = write_with_wal;
-      last_stats_.write_other_ = write_other;
-      last_stats_.write_self_ = write_self;
-
+      return true;
+    }
+    case kCFStats: {
+      DumpCFStats(value);
+      return true;
+    }
+    case kDBStats: {
+      DumpDBStats(value);
       return true;
     }
     case kSsTables:
       *value = current->DebugString();
       return true;
-    case kNumImmutableMemTable:
-      *value = std::to_string(cfd->imm()->size());
-      return true;
-    case kMemtableFlushPending:
-      // Return number of mem tables that are ready to flush (made immutable)
-      *value = std::to_string(cfd->imm()->IsFlushPending() ? 1 : 0);
-      return true;
-    case kCompactionPending:
-      // 1 if the system already determines at least one compacdtion is needed.
-      // 0 otherwise,
-      *value = std::to_string(current->NeedsCompaction() ? 1 : 0);
-      return true;
-    case kBackgroundErrors:
-      // Accumulated number of  errors in background flushes or compactions.
-      *value = std::to_string(GetBackgroundErrorCount());
-      return true;
-    case kCurSizeActiveMemTable:
-      // Current size of the active memtable
-      *value = std::to_string(cfd->mem()->ApproximateMemoryUsage());
-      return true;
-    case kNumEntriesInMutableMemtable:
-      // Current size of the active memtable
-      *value = std::to_string(cfd->mem()->GetNumEntries());
-      return true;
-    case kNumEntriesInImmutableMemtable:
-      // Current size of the active memtable
-      *value = std::to_string(cfd->imm()->current()->GetTotalNumEntries());
-      return true;
     default:
       return false;
   }
 }
 
+bool InternalStats::GetIntProperty(DBPropertyType property_type,
+                                   uint64_t* value, DBImpl* db) const {
+  Version* current = cfd_->current();
+
+  switch (property_type) {
+    case kNumImmutableMemTable:
+      *value = cfd_->imm()->size();
+      return true;
+    case kMemtableFlushPending:
+      // Return number of mem tables that are ready to flush (made immutable)
+      *value = (cfd_->imm()->IsFlushPending() ? 1 : 0);
+      return true;
+    case kCompactionPending:
+      // 1 if the system already determines at least one compacdtion is needed.
+      // 0 otherwise,
+      *value = (current->NeedsCompaction() ? 1 : 0);
+      return true;
+    case kBackgroundErrors:
+      // Accumulated number of  errors in background flushes or compactions.
+      *value = GetBackgroundErrorCount();
+      return true;
+    case kCurSizeActiveMemTable:
+      // Current size of the active memtable
+      *value = cfd_->mem()->ApproximateMemoryUsage();
+      return true;
+    case kNumEntriesInMutableMemtable:
+      // Current size of the active memtable
+      *value = cfd_->mem()->GetNumEntries();
+      return true;
+    case kNumEntriesInImmutableMemtable:
+      // Current size of the active memtable
+      *value = cfd_->imm()->current()->GetTotalNumEntries();
+      return true;
+    case kEstimatedNumKeys:
+      // Estimate number of entries in the column family:
+      // Use estimated entries in tables + total entries in memtables.
+      *value = cfd_->mem()->GetNumEntries() +
+               cfd_->imm()->current()->GetTotalNumEntries() +
+               current->GetEstimatedActiveKeys();
+      return true;
+#ifndef ROCKSDB_LITE
+    case kIsFileDeletionEnabled:
+      *value = db->IsFileDeletionsEnabled();
+      return true;
+#endif
+    default:
+      return false;
+  }
+}
+
+void InternalStats::DumpDBStats(std::string* value) {
+  char buf[1000];
+  // DB-level stats, only available from default column family
+  double seconds_up = (env_->NowMicros() - started_at_ + 1) / 1000000.0;
+  double interval_seconds_up = seconds_up - db_stats_snapshot_.seconds_up;
+  snprintf(buf, sizeof(buf),
+           "\n** DB Stats **\nUptime(secs): %.1f total, %.1f interval\n",
+           seconds_up, interval_seconds_up);
+  value->append(buf);
+  // Cumulative
+  uint64_t user_bytes_written = db_stats_[InternalStats::BYTES_WRITTEN];
+  uint64_t write_other = db_stats_[InternalStats::WRITE_DONE_BY_OTHER];
+  uint64_t write_self = db_stats_[InternalStats::WRITE_DONE_BY_SELF];
+  uint64_t wal_bytes = db_stats_[InternalStats::WAL_FILE_BYTES];
+  uint64_t wal_synced = db_stats_[InternalStats::WAL_FILE_SYNCED];
+  uint64_t write_with_wal = db_stats_[InternalStats::WRITE_WITH_WAL];
+  // Data
+  snprintf(buf, sizeof(buf),
+           "Cumulative writes: %" PRIu64 " writes, %" PRIu64 " batches, "
+           "%.1f writes per batch, %.2f GB user ingest\n",
+           write_other + write_self, write_self,
+           (write_other + write_self) / static_cast<double>(write_self + 1),
+           user_bytes_written / kGB);
+  value->append(buf);
+  // WAL
+  snprintf(buf, sizeof(buf),
+           "Cumulative WAL: %" PRIu64 " writes, %" PRIu64 " syncs, "
+           "%.2f writes per sync, %.2f GB written\n",
+           write_with_wal, wal_synced,
+           write_with_wal / static_cast<double>(wal_synced + 1),
+           wal_bytes / kGB);
+  value->append(buf);
+
+  // Interval
+  uint64_t interval_write_other = write_other - db_stats_snapshot_.write_other;
+  uint64_t interval_write_self = write_self - db_stats_snapshot_.write_self;
+  snprintf(buf, sizeof(buf),
+           "Interval writes: %" PRIu64 " writes, %" PRIu64 " batches, "
+           "%.1f writes per batch, %.1f MB user ingest\n",
+           interval_write_other + interval_write_self,
+           interval_write_self,
+           static_cast<double>(interval_write_other + interval_write_self) /
+               (interval_write_self + 1),
+           (user_bytes_written - db_stats_snapshot_.ingest_bytes) / kMB);
+  value->append(buf);
+
+  uint64_t interval_write_with_wal =
+      write_with_wal - db_stats_snapshot_.write_with_wal;
+  uint64_t interval_wal_synced = wal_synced - db_stats_snapshot_.wal_synced;
+  uint64_t interval_wal_bytes = wal_bytes - db_stats_snapshot_.wal_bytes;
+
+  snprintf(buf, sizeof(buf),
+           "Interval WAL: %" PRIu64 " writes, %" PRIu64 " syncs, "
+           "%.2f writes per sync, %.2f MB written\n",
+           interval_write_with_wal,
+           interval_wal_synced,
+           interval_write_with_wal /
+              static_cast<double>(interval_wal_synced + 1),
+           interval_wal_bytes / kGB);
+  value->append(buf);
+
+  db_stats_snapshot_.seconds_up = seconds_up;
+  db_stats_snapshot_.ingest_bytes = user_bytes_written;
+  db_stats_snapshot_.write_other = write_other;
+  db_stats_snapshot_.write_self = write_self;
+  db_stats_snapshot_.wal_bytes = wal_bytes;
+  db_stats_snapshot_.wal_synced = wal_synced;
+  db_stats_snapshot_.write_with_wal = write_with_wal;
+}
+
+void InternalStats::DumpCFStats(std::string* value) {
+  Version* current = cfd_->current();
+
+  int num_levels_to_check =
+      (cfd_->options()->compaction_style != kCompactionStyleUniversal &&
+       cfd_->options()->compaction_style != kCompactionStyleFIFO)
+          ? current->NumberLevels() - 1
+          : 1;
+  // Compaction scores are sorted base on its value. Restore them to the
+  // level order
+  std::vector<double> compaction_score(number_levels_, 0);
+  for (int i = 0; i < num_levels_to_check; ++i) {
+    compaction_score[current->compaction_level_[i]] =
+      current->compaction_score_[i];
+  }
+  // Count # of files being compacted for each level
+  std::vector<int> files_being_compacted(number_levels_, 0);
+  for (int level = 0; level < num_levels_to_check; ++level) {
+    for (auto* f : current->files_[level]) {
+      if (f->being_compacted) {
+        ++files_being_compacted[level];
+      }
+    }
+  }
+
+  char buf[1000];
+  // Per-ColumnFamily stats
+  PrintLevelStatsHeader(buf, sizeof(buf), cfd_->GetName());
+  value->append(buf);
+
+  CompactionStats stats_sum(0);
+  int total_files = 0;
+  int total_files_being_compacted = 0;
+  double total_file_size = 0;
+  uint64_t total_slowdown_soft = 0;
+  uint64_t total_slowdown_count_soft = 0;
+  uint64_t total_slowdown_hard = 0;
+  uint64_t total_slowdown_count_hard = 0;
+  uint64_t total_stall_count = 0;
+  double total_stall_us = 0;
+  for (int level = 0; level < number_levels_; level++) {
+    int files = current->NumLevelFiles(level);
+    total_files += files;
+    total_files_being_compacted += files_being_compacted[level];
+    if (comp_stats_[level].micros > 0 || files > 0) {
+      uint64_t stalls = level == 0 ?
+        (cf_stats_count_[LEVEL0_SLOWDOWN] +
+         cf_stats_count_[LEVEL0_NUM_FILES] +
+         cf_stats_count_[MEMTABLE_COMPACTION])
+        : (stall_leveln_slowdown_count_soft_[level] +
+           stall_leveln_slowdown_count_hard_[level]);
+
+      double stall_us = level == 0 ?
+         (cf_stats_value_[LEVEL0_SLOWDOWN] +
+          cf_stats_value_[LEVEL0_NUM_FILES] +
+          cf_stats_value_[MEMTABLE_COMPACTION])
+         : (stall_leveln_slowdown_soft_[level] +
+            stall_leveln_slowdown_hard_[level]);
+
+      stats_sum.Add(comp_stats_[level]);
+      total_file_size += current->NumLevelBytes(level);
+      total_stall_us += stall_us;
+      total_stall_count += stalls;
+      total_slowdown_soft += stall_leveln_slowdown_soft_[level];
+      total_slowdown_count_soft += stall_leveln_slowdown_count_soft_[level];
+      total_slowdown_hard += stall_leveln_slowdown_hard_[level];
+      total_slowdown_count_hard += stall_leveln_slowdown_count_hard_[level];
+      int64_t bytes_read = comp_stats_[level].bytes_readn +
+                           comp_stats_[level].bytes_readnp1;
+      double rw_amp = (comp_stats_[level].bytes_readn == 0) ? 0.0
+          : (comp_stats_[level].bytes_written + bytes_read) /
+            static_cast<double>(comp_stats_[level].bytes_readn);
+      double w_amp = (comp_stats_[level].bytes_readn == 0) ? 0.0
+          : comp_stats_[level].bytes_written /
+            static_cast<double>(comp_stats_[level].bytes_readn);
+      PrintLevelStats(buf, sizeof(buf), "L" + std::to_string(level),
+          files, files_being_compacted[level], current->NumLevelBytes(level),
+          compaction_score[level], rw_amp, w_amp, stall_us, stalls,
+          comp_stats_[level]);
+      value->append(buf);
+    }
+  }
+  uint64_t curr_ingest = cf_stats_value_[BYTES_FLUSHED];
+  // Cumulative summary
+  double rw_amp = (stats_sum.bytes_written + stats_sum.bytes_readn +
+      stats_sum.bytes_readnp1) / static_cast<double>(curr_ingest + 1);
+  double w_amp = stats_sum.bytes_written / static_cast<double>(curr_ingest + 1);
+  // Stats summary across levels
+  PrintLevelStats(buf, sizeof(buf), "Sum", total_files,
+      total_files_being_compacted, total_file_size, 0, rw_amp, w_amp,
+      total_stall_us, total_stall_count, stats_sum);
+  value->append(buf);
+  // Interval summary
+  uint64_t interval_ingest =
+      curr_ingest - cf_stats_snapshot_.ingest_bytes + 1;
+  CompactionStats interval_stats(stats_sum);
+  interval_stats.Subtract(cf_stats_snapshot_.comp_stats);
+  rw_amp = (interval_stats.bytes_written +
+      interval_stats.bytes_readn + interval_stats.bytes_readnp1) /
+      static_cast<double>(interval_ingest);
+  w_amp = interval_stats.bytes_written / static_cast<double>(interval_ingest);
+  PrintLevelStats(buf, sizeof(buf), "Int", 0, 0, 0, 0,
+      rw_amp, w_amp, total_stall_us - cf_stats_snapshot_.stall_us,
+      total_stall_count - cf_stats_snapshot_.stall_count, interval_stats);
+  value->append(buf);
+
+  snprintf(buf, sizeof(buf),
+           "Flush(GB): accumulative %.3f, interval %.3f\n",
+           curr_ingest / kGB, interval_ingest / kGB);
+  value->append(buf);
+  snprintf(buf, sizeof(buf),
+           "Stalls(secs): %.3f level0_slowdown, %.3f level0_numfiles, "
+           "%.3f memtable_compaction, %.3f leveln_slowdown_soft, "
+           "%.3f leveln_slowdown_hard\n",
+           cf_stats_value_[LEVEL0_SLOWDOWN] / 1000000.0,
+           cf_stats_value_[LEVEL0_NUM_FILES] / 1000000.0,
+           cf_stats_value_[MEMTABLE_COMPACTION] / 1000000.0,
+           total_slowdown_soft / 1000000.0,
+           total_slowdown_hard / 1000000.0);
+  value->append(buf);
+
+  snprintf(buf, sizeof(buf),
+           "Stalls(count): %" PRIu64 " level0_slowdown, "
+           "%" PRIu64 " level0_numfiles, %" PRIu64 " memtable_compaction, "
+           "%" PRIu64 " leveln_slowdown_soft, "
+           "%" PRIu64 " leveln_slowdown_hard\n",
+           cf_stats_count_[LEVEL0_SLOWDOWN],
+           cf_stats_count_[LEVEL0_NUM_FILES],
+           cf_stats_count_[MEMTABLE_COMPACTION],
+           total_slowdown_count_soft, total_slowdown_count_hard);
+  value->append(buf);
+
+  cf_stats_snapshot_.ingest_bytes = curr_ingest;
+  cf_stats_snapshot_.comp_stats = stats_sum;
+  cf_stats_snapshot_.stall_us = total_stall_us;
+  cf_stats_snapshot_.stall_count = total_stall_count;
+}
+
 }  // namespace rocksdb

db/internal_stats.h

@@ -9,8 +9,6 @@
 //
 
 #pragma once
-#include "rocksdb/statistics.h"
-#include "util/statistics.h"
 #include "db/version_set.h"
 
 #include <vector>
@@ -23,11 +21,15 @@ namespace rocksdb {
 class MemTableList;
 class DBImpl;
 
-enum DBPropertyType {
+enum DBPropertyType : uint32_t {
+  kUnknown,
   kNumFilesAtLevel,  // Number of files at a specific level
   kLevelStats,       // Return number of files and total sizes of each level
-  kStats,            // Return general statitistics of DB
+  kCFStats,          // Return general statitistics of CF
+  kDBStats,          // Return general statitistics of DB
+  kStats,            // Return general statitistics of both DB and CF
   kSsTables,         // Return a human readable string of current SST files
+  kStartIntTypes,    // ---- Dummy value to indicate the start of integer values
   kNumImmutableMemTable,   // Return number of immutable mem tables
   kMemtableFlushPending,   // Return 1 if mem table flushing is pending,
                            // otherwise 0.
@@ -38,45 +40,79 @@ enum DBPropertyType {
                                    // memtable.
   kNumEntriesInImmutableMemtable,  // Return sum of number of entries in all
                                    // the immutable mem tables.
-  kUnknown,
+  kEstimatedNumKeys,  // Estimated total number of keys in the database.
+  kEstimatedUsageByTableReaders,  // Estimated memory by table readers.
+  kIsFileDeletionEnabled,         // Equals disable_delete_obsolete_files_,
+                                  // 0 means file deletions enabled
 };
 
-extern DBPropertyType GetPropertyType(const Slice& property);
+extern DBPropertyType GetPropertyType(const Slice& property,
+                                      bool* is_int_property,
+                                      bool* need_out_of_mutex);
 
 class InternalStats {
  public:
-  enum WriteStallType {
+  enum InternalCFStatsType {
     LEVEL0_SLOWDOWN,
     MEMTABLE_COMPACTION,
     LEVEL0_NUM_FILES,
     WRITE_STALLS_ENUM_MAX,
+    BYTES_FLUSHED,
+    INTERNAL_CF_STATS_ENUM_MAX,
   };
 
-  InternalStats(int num_levels, Env* env, Statistics* statistics)
-      : compaction_stats_(num_levels),
-        stall_micros_(WRITE_STALLS_ENUM_MAX, 0),
-        stall_counts_(WRITE_STALLS_ENUM_MAX, 0),
-        stall_leveln_slowdown_(num_levels, 0),
-        stall_leveln_slowdown_count_(num_levels, 0),
+  enum InternalDBStatsType {
+    WAL_FILE_BYTES,
+    WAL_FILE_SYNCED,
+    BYTES_WRITTEN,
+    WRITE_DONE_BY_OTHER,
+    WRITE_DONE_BY_SELF,
+    WRITE_WITH_WAL,
+    INTERNAL_DB_STATS_ENUM_MAX,
+  };
+
+  InternalStats(int num_levels, Env* env, ColumnFamilyData* cfd)
+      : db_stats_(INTERNAL_DB_STATS_ENUM_MAX),
+        cf_stats_value_(INTERNAL_CF_STATS_ENUM_MAX),
+        cf_stats_count_(INTERNAL_CF_STATS_ENUM_MAX),
+        comp_stats_(num_levels),
+        stall_leveln_slowdown_hard_(num_levels),
+        stall_leveln_slowdown_count_hard_(num_levels),
+        stall_leveln_slowdown_soft_(num_levels),
+        stall_leveln_slowdown_count_soft_(num_levels),
         bg_error_count_(0),
         number_levels_(num_levels),
-        statistics_(statistics),
         env_(env),
-        started_at_(env->NowMicros()) {}
+        cfd_(cfd),
+        started_at_(env->NowMicros()) {
+    for (int i = 0; i< INTERNAL_DB_STATS_ENUM_MAX; ++i) {
+      db_stats_[i] = 0;
+    }
+    for (int i = 0; i< INTERNAL_CF_STATS_ENUM_MAX; ++i) {
+      cf_stats_value_[i] = 0;
+      cf_stats_count_[i] = 0;
+    }
+    for (int i = 0; i < num_levels; ++i) {
+      stall_leveln_slowdown_hard_[i] = 0;
+      stall_leveln_slowdown_count_hard_[i] = 0;
+      stall_leveln_slowdown_soft_[i] = 0;
+      stall_leveln_slowdown_count_soft_[i] = 0;
+    }
+  }
 
-  // Per level compaction stats.  compaction_stats_[level] stores the stats for
+  // Per level compaction stats.  comp_stats_[level] stores the stats for
   // compactions that produced data for the specified "level".
   struct CompactionStats {
     uint64_t micros;
 
     // Bytes read from level N during compaction between levels N and N+1
-    int64_t bytes_readn;
+    uint64_t bytes_readn;
 
     // Bytes read from level N+1 during compaction between levels N and N+1
-    int64_t bytes_readnp1;
+    uint64_t bytes_readnp1;
 
     // Total bytes written during compaction between levels N and N+1
-    int64_t bytes_written;
+    uint64_t bytes_written;
 
     // Files read from level N during compaction between levels N and N+1
     int files_in_leveln;
@@ -87,10 +123,17 @@ class InternalStats {
     // Files written during compaction between levels N and N+1
     int files_out_levelnp1;
 
+    // Total incoming entries during compaction between levels N and N+1
+    int num_input_records;
+
+    // Accumulated diff number of entries
+    // (num input entries - num output entires) for compaction  levels N and N+1
+    int num_dropped_records;
+
     // Number of compactions done
     int count;
 
-    CompactionStats()
+    explicit CompactionStats(int count = 0)
         : micros(0),
           bytes_readn(0),
           bytes_readnp1(0),
@@ -98,7 +141,21 @@ class InternalStats {
           files_in_leveln(0),
           files_in_levelnp1(0),
           files_out_levelnp1(0),
-          count(0) {}
+          num_input_records(0),
+          num_dropped_records(0),
+          count(count) {}
+
+    explicit CompactionStats(const CompactionStats& c)
+        : micros(c.micros),
+          bytes_readn(c.bytes_readn),
+          bytes_readnp1(c.bytes_readnp1),
+          bytes_written(c.bytes_written),
+          files_in_leveln(c.files_in_leveln),
+          files_in_levelnp1(c.files_in_levelnp1),
+          files_out_levelnp1(c.files_out_levelnp1),
+          num_input_records(c.num_input_records),
+          num_dropped_records(c.num_dropped_records),
+          count(c.count) {}
 
     void Add(const CompactionStats& c) {
       this->micros += c.micros;
@@ -108,68 +165,114 @@ class InternalStats {
       this->files_in_leveln += c.files_in_leveln;
       this->files_in_levelnp1 += c.files_in_levelnp1;
       this->files_out_levelnp1 += c.files_out_levelnp1;
-      this->count += 1;
+      this->num_input_records += c.num_input_records;
+      this->num_dropped_records += c.num_dropped_records;
+      this->count += c.count;
+    }
+
+    void Subtract(const CompactionStats& c) {
+      this->micros -= c.micros;
+      this->bytes_readn -= c.bytes_readn;
+      this->bytes_readnp1 -= c.bytes_readnp1;
+      this->bytes_written -= c.bytes_written;
+      this->files_in_leveln -= c.files_in_leveln;
+      this->files_in_levelnp1 -= c.files_in_levelnp1;
+      this->files_out_levelnp1 -= c.files_out_levelnp1;
+      this->num_input_records -= c.num_input_records;
+      this->num_dropped_records -= c.num_dropped_records;
+      this->count -= c.count;
     }
   };
 
   void AddCompactionStats(int level, const CompactionStats& stats) {
-    compaction_stats_[level].Add(stats);
+    comp_stats_[level].Add(stats);
   }
 
-  void RecordWriteStall(WriteStallType write_stall_type, uint64_t micros) {
-    stall_micros_[write_stall_type] += micros;
-    ++stall_counts_[write_stall_type];
+  void RecordLevelNSlowdown(int level, uint64_t micros, bool soft) {
+    if (soft) {
+      stall_leveln_slowdown_soft_[level] += micros;
+      ++stall_leveln_slowdown_count_soft_[level];
+    } else {
+      stall_leveln_slowdown_hard_[level] += micros;
+      ++stall_leveln_slowdown_count_hard_[level];
+    }
   }
 
-  void RecordLevelNSlowdown(int level, uint64_t micros) {
-    stall_leveln_slowdown_[level] += micros;
-    ++stall_leveln_slowdown_count_[level];
+  void AddCFStats(InternalCFStatsType type, uint64_t value) {
+    cf_stats_value_[type] += value;
+    ++cf_stats_count_[type];
+  }
+
+  void AddDBStats(InternalDBStatsType type, uint64_t value) {
+    db_stats_[type] += value;
   }
 
   uint64_t GetBackgroundErrorCount() const { return bg_error_count_; }
 
   uint64_t BumpAndGetBackgroundErrorCount() { return ++bg_error_count_; }
 
-  bool GetProperty(DBPropertyType property_type, const Slice& property,
-                   std::string* value, ColumnFamilyData* cfd);
+  bool GetStringProperty(DBPropertyType property_type, const Slice& property,
+                         std::string* value);
+
+  bool GetIntProperty(DBPropertyType property_type, uint64_t* value,
+                      DBImpl* db) const;
+
+  bool GetIntPropertyOutOfMutex(DBPropertyType property_type, Version* version,
+                                uint64_t* value) const;
 
  private:
-  std::vector<CompactionStats> compaction_stats_;
+  void DumpDBStats(std::string* value);
+  void DumpCFStats(std::string* value);
+
+  // Per-DB stats
+  std::vector<uint64_t> db_stats_;
+  // Per-ColumnFamily stats
+  std::vector<uint64_t> cf_stats_value_;
+  std::vector<uint64_t> cf_stats_count_;
+  // Per-ColumnFamily/level compaction stats
+  std::vector<CompactionStats> comp_stats_;
+  // These count the number of microseconds for which MakeRoomForWrite stalls.
+  std::vector<uint64_t> stall_leveln_slowdown_hard_;
+  std::vector<uint64_t> stall_leveln_slowdown_count_hard_;
+  std::vector<uint64_t> stall_leveln_slowdown_soft_;
+  std::vector<uint64_t> stall_leveln_slowdown_count_soft_;
 
   // Used to compute per-interval statistics
-  struct StatsSnapshot {
-    uint64_t compaction_bytes_read_;     // Bytes read by compaction
-    uint64_t compaction_bytes_written_;  // Bytes written by compaction
-    uint64_t ingest_bytes_;              // Bytes written by user
-    uint64_t wal_bytes_;                 // Bytes written to WAL
-    uint64_t wal_synced_;                // Number of times WAL is synced
-    uint64_t write_with_wal_;            // Number of writes that request WAL
+  struct CFStatsSnapshot {
+    // ColumnFamily-level stats
+    CompactionStats comp_stats;
+    uint64_t ingest_bytes;            // Bytes written to L0
+    uint64_t stall_us;                // Stall time in micro-seconds
+    uint64_t stall_count;             // Stall count
+
+    CFStatsSnapshot()
+        : comp_stats(0),
+          ingest_bytes(0),
+          stall_us(0),
+          stall_count(0) {}
+  } cf_stats_snapshot_;
+
+  struct DBStatsSnapshot {
+    // DB-level stats
+    uint64_t ingest_bytes;            // Bytes written by user
+    uint64_t wal_bytes;               // Bytes written to WAL
+    uint64_t wal_synced;              // Number of times WAL is synced
+    uint64_t write_with_wal;          // Number of writes that request WAL
     // These count the number of writes processed by the calling thread or
     // another thread.
-    uint64_t write_other_;
-    uint64_t write_self_;
-    double seconds_up_;
+    uint64_t write_other;
+    uint64_t write_self;
+    double seconds_up;
 
-    StatsSnapshot()
-        : compaction_bytes_read_(0),
-          compaction_bytes_written_(0),
-          ingest_bytes_(0),
-          wal_bytes_(0),
-          wal_synced_(0),
-          write_with_wal_(0),
-          write_other_(0),
-          write_self_(0),
-          seconds_up_(0) {}
-  };
-
-  // Counters from the previous time per-interval stats were computed
-  StatsSnapshot last_stats_;
-
-  // These count the number of microseconds for which MakeRoomForWrite stalls.
-  std::vector<uint64_t> stall_micros_;
-  std::vector<uint64_t> stall_counts_;
-  std::vector<uint64_t> stall_leveln_slowdown_;
-  std::vector<uint64_t> stall_leveln_slowdown_count_;
+    DBStatsSnapshot()
+        : ingest_bytes(0),
+          wal_bytes(0),
+          wal_synced(0),
+          write_with_wal(0),
+          write_other(0),
+          write_self(0),
+          seconds_up(0) {}
+  } db_stats_snapshot_;
 
   // Total number of background errors encountered. Every time a flush task
   // or compaction task fails, this counter is incremented. The failure can
@@ -178,10 +281,10 @@ class InternalStats {
   // or compaction will cause the counter to increase too.
   uint64_t bg_error_count_;
 
-  int number_levels_;
-  Statistics* statistics_;
+  const int number_levels_;
   Env* env_;
-  uint64_t started_at_;
+  ColumnFamilyData* cfd_;
+  const uint64_t started_at_;
 };
 
 }  // namespace rocksdb

db/log_and_apply_bench.cc

@@ -9,6 +9,7 @@
 #include "util/testharness.h"
 #include "util/benchharness.h"
 #include "db/version_set.h"
+#include "db/write_controller.h"
 #include "util/mutexlock.h"
 
 namespace rocksdb {
@@ -21,6 +22,7 @@ std::string MakeKey(unsigned int num) {
 
 void BM_LogAndApply(int iters, int num_base_files) {
   VersionSet* vset;
+  WriteController wc;
   ColumnFamilyData* default_cfd;
   uint64_t fnum = 1;
   port::Mutex mu;
@@ -42,7 +44,12 @@ void BM_LogAndApply(int iters, int num_base_files) {
 
     Options options;
     EnvOptions sopt;
-    vset = new VersionSet(dbname, &options, sopt, nullptr);
+    // Notice we are using the default options not through SanitizeOptions().
+    // We might want to initialize some options manually if needed.
+    options.db_paths.emplace_back(dbname, 0);
+    // The parameter of table cache is passed in as null, so any file I/O
+    // operation is likely to fail.
+    vset = new VersionSet(dbname, &options, sopt, nullptr, &wc);
     std::vector<ColumnFamilyDescriptor> dummy;
     dummy.push_back(ColumnFamilyDescriptor());
     ASSERT_OK(vset->Recover(dummy));
@@ -53,7 +60,8 @@ void BM_LogAndApply(int iters, int num_base_files) {
       InternalKey limit(MakeKey(2 * fnum + 1), 1, kTypeDeletion);
       vbase.AddFile(2, ++fnum, 0, 1 /* file size */, start, limit, 1, 1);
     }
-    ASSERT_OK(vset->LogAndApply(default_cfd, &vbase, &mu));
+    ASSERT_OK(vset->LogAndApply(default_cfd,
+          *default_cfd->GetLatestMutableCFOptions(), &vbase, &mu));
   }
 
   for (int i = 0; i < iters; i++) {
@@ -62,8 +70,10 @@ void BM_LogAndApply(int iters, int num_base_files) {
     InternalKey start(MakeKey(2 * fnum), 1, kTypeValue);
     InternalKey limit(MakeKey(2 * fnum + 1), 1, kTypeDeletion);
     vedit.AddFile(2, ++fnum, 0, 1 /* file size */, start, limit, 1, 1);
-    vset->LogAndApply(default_cfd, &vedit, &mu);
+    vset->LogAndApply(default_cfd, *default_cfd->GetLatestMutableCFOptions(),
+                      &vedit, &mu);
   }
+  delete vset;
 }
 
 BENCHMARK_NAMED_PARAM(BM_LogAndApply, 1000_iters_1_file, 1000, 1)

db/memtable.cc

@@ -31,41 +31,57 @@
 
 namespace rocksdb {
 
-MemTable::MemTable(const InternalKeyComparator& cmp, const Options& options)
+MemTableOptions::MemTableOptions(
+    const MutableCFOptions& mutable_cf_options, const Options& options)
+  : write_buffer_size(mutable_cf_options.write_buffer_size),
+    arena_block_size(mutable_cf_options.arena_block_size),
+    memtable_prefix_bloom_bits(mutable_cf_options.memtable_prefix_bloom_bits),
+    memtable_prefix_bloom_probes(
+        mutable_cf_options.memtable_prefix_bloom_probes),
+    memtable_prefix_bloom_huge_page_tlb_size(
+        mutable_cf_options.memtable_prefix_bloom_huge_page_tlb_size),
+    inplace_update_support(options.inplace_update_support),
+    inplace_update_num_locks(options.inplace_update_num_locks),
+    inplace_callback(options.inplace_callback),
+    max_successive_merges(mutable_cf_options.max_successive_merges),
+    filter_deletes(mutable_cf_options.filter_deletes) {}
+
+MemTable::MemTable(const InternalKeyComparator& cmp,
+                   const ImmutableCFOptions& ioptions,
+                   const MemTableOptions& moptions)
     : comparator_(cmp),
+      ioptions_(ioptions),
+      moptions_(moptions),
       refs_(0),
-      kArenaBlockSize(OptimizeBlockSize(options.arena_block_size)),
-      kWriteBufferSize(options.write_buffer_size),
-      arena_(options.arena_block_size),
-      table_(options.memtable_factory->CreateMemTableRep(
-          comparator_, &arena_, options.prefix_extractor.get(),
-          options.info_log.get())),
+      kArenaBlockSize(OptimizeBlockSize(moptions.arena_block_size)),
+      arena_(moptions.arena_block_size),
+      table_(ioptions.memtable_factory->CreateMemTableRep(
+          comparator_, &arena_, ioptions.prefix_extractor, ioptions.info_log)),
       num_entries_(0),
       flush_in_progress_(false),
       flush_completed_(false),
       file_number_(0),
       first_seqno_(0),
       mem_next_logfile_number_(0),
-      locks_(options.inplace_update_support ? options.inplace_update_num_locks
-                                            : 0),
-      prefix_extractor_(options.prefix_extractor.get()),
-      should_flush_(ShouldFlushNow()) {
+      locks_(moptions.inplace_update_support ? moptions.inplace_update_num_locks
+                                             : 0),
+      prefix_extractor_(ioptions.prefix_extractor),
+      should_flush_(ShouldFlushNow()),
+      flush_scheduled_(false) {
   // if should_flush_ == true without an entry inserted, something must have
   // gone wrong already.
   assert(!should_flush_);
-  if (prefix_extractor_ && options.memtable_prefix_bloom_bits > 0) {
+  if (prefix_extractor_ && moptions.memtable_prefix_bloom_bits > 0) {
     prefix_bloom_.reset(new DynamicBloom(
         &arena_,
-        options.memtable_prefix_bloom_bits, options.bloom_locality,
-        options.memtable_prefix_bloom_probes, nullptr,
-        options.memtable_prefix_bloom_huge_page_tlb_size,
-        options.info_log.get()));
+        moptions.memtable_prefix_bloom_bits, ioptions.bloom_locality,
+        moptions.memtable_prefix_bloom_probes, nullptr,
+        moptions.memtable_prefix_bloom_huge_page_tlb_size,
+        ioptions.info_log));
   }
 }
 
-MemTable::~MemTable() {
-  assert(refs_ == 0);
-}
+MemTable::~MemTable() { assert(refs_ == 0); }
 
 size_t MemTable::ApproximateMemoryUsage() {
   size_t arena_usage = arena_.ApproximateMemoryUsage();
@@ -97,14 +113,16 @@ bool MemTable::ShouldFlushNow() const {
   // if we can still allocate one more block without exceeding the
   // over-allocation ratio, then we should not flush.
   if (allocated_memory + kArenaBlockSize <
-      kWriteBufferSize + kArenaBlockSize * kAllowOverAllocationRatio) {
+      moptions_.write_buffer_size +
+      kArenaBlockSize * kAllowOverAllocationRatio) {
     return false;
   }
 
-  // if user keeps adding entries that exceeds kWriteBufferSize, we need to
-  // flush earlier even though we still have much available memory left.
-  if (allocated_memory >
-      kWriteBufferSize + kArenaBlockSize * kAllowOverAllocationRatio) {
+  // if user keeps adding entries that exceeds moptions.write_buffer_size,
+  // we need to flush earlier even though we still have much available
+  // memory left.
+  if (allocated_memory > moptions_.write_buffer_size +
+      kArenaBlockSize * kAllowOverAllocationRatio) {
     return true;
   }
 
@@ -174,13 +192,13 @@ const char* EncodeKey(std::string* scratch, const Slice& target) {
 
 class MemTableIterator: public Iterator {
  public:
-  MemTableIterator(const MemTable& mem, const ReadOptions& options,
-                   bool enforce_total_order, Arena* arena)
+  MemTableIterator(
+      const MemTable& mem, const ReadOptions& read_options, Arena* arena)
       : bloom_(nullptr),
         prefix_extractor_(mem.prefix_extractor_),
         valid_(false),
         arena_mode_(arena != nullptr) {
-    if (prefix_extractor_ != nullptr && !enforce_total_order) {
+    if (prefix_extractor_ != nullptr && !read_options.total_order_seek) {
       bloom_ = mem.prefix_bloom_.get();
       iter_ = mem.table_->GetDynamicPrefixIterator(arena);
     } else {
@@ -248,15 +266,10 @@ class MemTableIterator: public Iterator {
   void operator=(const MemTableIterator&);
 };
 
-Iterator* MemTable::NewIterator(const ReadOptions& options,
-                                bool enforce_total_order, Arena* arena) {
-  if (arena == nullptr) {
-    return new MemTableIterator(*this, options, enforce_total_order, nullptr);
-  } else {
-    auto mem = arena->AllocateAligned(sizeof(MemTableIterator));
-    return new (mem)
-        MemTableIterator(*this, options, enforce_total_order, arena);
-  }
+Iterator* MemTable::NewIterator(const ReadOptions& read_options, Arena* arena) {
+  assert(arena != nullptr);
+  auto mem = arena->AllocateAligned(sizeof(MemTableIterator));
+  return new (mem) MemTableIterator(*this, read_options, arena);
 }
 
 port::RWMutex* MemTable::GetLock(const Slice& key) {
@@ -390,7 +403,16 @@ static bool SaveValue(void* arg, const char* entry) {
         return false;
       }
       case kTypeMerge: {
-        std::string merge_result;  // temporary area for merge results later
+        if (!merge_operator) {
+          *(s->status) = Status::InvalidArgument(
+              "merge_operator is not properly initialized.");
+          // Normally we continue the loop (return true) when we see a merge
+          // operand.  But in case of an error, we should stop the loop
+          // immediately and pretend we have found the value to stop further
+          // seek.  Otherwise, the later call will override this error status.
+          *(s->found_final_value) = true;
+          return false;
+        }
         Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
         *(s->merge_in_progress) = true;
         merge_context->PushOperand(v);
@@ -407,8 +429,13 @@ static bool SaveValue(void* arg, const char* entry) {
 }
 
 bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
-                   MergeContext& merge_context, const Options& options) {
-  PERF_TIMER_AUTO(get_from_memtable_time);
+                   MergeContext* merge_context) {
+  // The sequence number is updated synchronously in version_set.h
+  if (IsEmpty()) {
+    // Avoiding recording stats for speed.
+    return false;
+  }
+  PERF_TIMER_GUARD(get_from_memtable_time);
 
   Slice user_key = key.user_key();
   bool found_final_value = false;
@@ -426,11 +453,11 @@ bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
     saver.value = value;
     saver.status = s;
     saver.mem = this;
-    saver.merge_context = &merge_context;
-    saver.merge_operator = options.merge_operator.get();
-    saver.logger = options.info_log.get();
-    saver.inplace_update_support = options.inplace_update_support;
-    saver.statistics = options.statistics.get();
+    saver.merge_context = merge_context;
+    saver.merge_operator = ioptions_.merge_operator;
+    saver.logger = ioptions_.info_log;
+    saver.inplace_update_support = moptions_.inplace_update_support;
+    saver.statistics = ioptions_.statistics;
     table_->Get(key, &saver, SaveValue);
   }
 
@@ -438,7 +465,6 @@ bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
   if (!found_final_value && merge_in_progress) {
     *s = Status::MergeInProgress("");
   }
-  PERF_TIMER_STOP(get_from_memtable_time);
   PERF_COUNTER_ADD(get_from_memtable_count, 1);
   return found_final_value;
 }
@@ -503,8 +529,7 @@ void MemTable::Update(SequenceNumber seq,
 
 bool MemTable::UpdateCallback(SequenceNumber seq,
                               const Slice& key,
-                              const Slice& delta,
-                              const Options& options) {
+                              const Slice& delta) {
   LookupKey lkey(key, seq);
   Slice memkey = lkey.memtable_key();
 
@@ -539,8 +564,8 @@ bool MemTable::UpdateCallback(SequenceNumber seq,
 
           std::string str_value;
           WriteLock wl(GetLock(lkey.user_key()));
-          auto status = options.inplace_callback(prev_buffer, &new_prev_size,
-                                                    delta, &str_value);
+          auto status = moptions_.inplace_callback(prev_buffer, &new_prev_size,
+                                                   delta, &str_value);
           if (status == UpdateStatus::UPDATED_INPLACE) {
             // Value already updated by callback.
             assert(new_prev_size <= prev_size);
@@ -553,12 +578,12 @@ bool MemTable::UpdateCallback(SequenceNumber seq,
                 memcpy(p, prev_buffer, new_prev_size);
               }
             }
-            RecordTick(options.statistics.get(), NUMBER_KEYS_UPDATED);
+            RecordTick(ioptions_.statistics, NUMBER_KEYS_UPDATED);
             should_flush_ = ShouldFlushNow();
             return true;
           } else if (status == UpdateStatus::UPDATED) {
             Add(seq, kTypeValue, key, Slice(str_value));
-            RecordTick(options.statistics.get(), NUMBER_KEYS_WRITTEN);
+            RecordTick(ioptions_.statistics, NUMBER_KEYS_WRITTEN);
             should_flush_ = ShouldFlushNow();
             return true;
           } else if (status == UpdateStatus::UPDATE_FAILED) {

db/memtable.h

@@ -10,14 +10,18 @@
 #pragma once
 #include <string>
 #include <memory>
+#include <functional>
 #include <deque>
+#include <vector>
 #include "db/dbformat.h"
 #include "db/skiplist.h"
 #include "db/version_edit.h"
 #include "rocksdb/db.h"
 #include "rocksdb/memtablerep.h"
+#include "rocksdb/immutable_options.h"
 #include "util/arena.h"
 #include "util/dynamic_bloom.h"
+#include "util/mutable_cf_options.h"
 
 namespace rocksdb {
 
@@ -26,6 +30,25 @@ class Mutex;
 class MemTableIterator;
 class MergeContext;
 
+struct MemTableOptions {
+  explicit MemTableOptions(
+      const MutableCFOptions& mutable_cf_options,
+      const Options& options);
+  size_t write_buffer_size;
+  size_t arena_block_size;
+  uint32_t memtable_prefix_bloom_bits;
+  uint32_t memtable_prefix_bloom_probes;
+  size_t memtable_prefix_bloom_huge_page_tlb_size;
+  bool inplace_update_support;
+  size_t inplace_update_num_locks;
+  UpdateStatus (*inplace_callback)(char* existing_value,
+                                   uint32_t* existing_value_size,
+                                   Slice delta_value,
+                                   std::string* merged_value);
+  size_t max_successive_merges;
+  bool filter_deletes;
+};
+
 class MemTable {
  public:
   struct KeyComparator : public MemTableRep::KeyComparator {
@@ -40,7 +63,8 @@ class MemTable {
   // MemTables are reference counted.  The initial reference count
   // is zero and the caller must call Ref() at least once.
   explicit MemTable(const InternalKeyComparator& comparator,
-                    const Options& options);
+                    const ImmutableCFOptions& ioptions,
+                    const MemTableOptions& moptions);
 
   ~MemTable();
 
@@ -67,7 +91,11 @@ class MemTable {
 
   // This method heuristically determines if the memtable should continue to
   // host more data.
-  bool ShouldFlush() const { return should_flush_; }
+  bool ShouldScheduleFlush() const {
+    return flush_scheduled_ == false && should_flush_;
+  }
+
+  void MarkFlushScheduled() { flush_scheduled_ = true; }
 
   // Return an iterator that yields the contents of the memtable.
   //
@@ -81,9 +109,7 @@ class MemTable {
   // arena: If not null, the arena needs to be used to allocate the Iterator.
   //        Calling ~Iterator of the iterator will destroy all the states but
   //        those allocated in arena.
-  Iterator* NewIterator(const ReadOptions& options,
-                        bool enforce_total_order = false,
-                        Arena* arena = nullptr);
+  Iterator* NewIterator(const ReadOptions& read_options, Arena* arena);
 
   // Add an entry into memtable that maps key to value at the
   // specified sequence number and with the specified type.
@@ -101,7 +127,7 @@ class MemTable {
   //   store MergeInProgress in s, and return false.
   // Else, return false.
   bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext& merge_context, const Options& options);
+           MergeContext* merge_context);
 
   // Attempts to update the new_value inplace, else does normal Add
   // Pseudocode
@@ -125,8 +151,7 @@ class MemTable {
   //   else return false
   bool UpdateCallback(SequenceNumber seq,
                       const Slice& key,
-                      const Slice& delta,
-                      const Options& options);
+                      const Slice& delta);
 
   // Returns the number of successive merge entries starting from the newest
   // entry for the key up to the last non-merge entry or last entry for the
@@ -139,6 +164,9 @@ class MemTable {
   // Returns the edits area that is needed for flushing the memtable
   VersionEdit* GetEdits() { return &edit_; }
 
+  // Returns if there is no entry inserted to the mem table.
+  bool IsEmpty() const { return first_seqno_ == 0; }
+
   // Returns the sequence number of the first element that was inserted
   // into the memtable
   SequenceNumber GetFirstSequenceNumber() { return first_seqno_; }
@@ -171,8 +199,11 @@ class MemTable {
 
   const Arena& TEST_GetArena() const { return arena_; }
 
+  const ImmutableCFOptions* GetImmutableOptions() const { return &ioptions_; }
+  const MemTableOptions* GetMemTableOptions() const { return &moptions_; }
+
  private:
-  // Dynamically check if we can add more incoming entries.
+  // Dynamically check if we can add more incoming entries
   bool ShouldFlushNow() const;
 
   friend class MemTableIterator;
@@ -180,9 +211,10 @@ class MemTable {
   friend class MemTableList;
 
   KeyComparator comparator_;
+  const ImmutableCFOptions& ioptions_;
+  const MemTableOptions moptions_;
   int refs_;
   const size_t kArenaBlockSize;
-  const size_t kWriteBufferSize;
   Arena arena_;
   unique_ptr<MemTableRep> table_;
 
@@ -215,6 +247,9 @@ class MemTable {
 
   // a flag indicating if a memtable has met the criteria to flush
   bool should_flush_;
+
+  // a flag indicating if flush has been scheduled
+  bool flush_scheduled_;
 };
 
 extern const char* EncodeKey(std::string* scratch, const Slice& target);

db/memtable_list.cc

@@ -62,10 +62,9 @@ int MemTableList::size() const {
 // Return the most recent value found, if any.
 // Operands stores the list of merge operations to apply, so far.
 bool MemTableListVersion::Get(const LookupKey& key, std::string* value,
-                              Status* s, MergeContext& merge_context,
-                              const Options& options) {
+                              Status* s, MergeContext* merge_context) {
   for (auto& memtable : memlist_) {
-    if (memtable->Get(key, value, s, merge_context, options)) {
+    if (memtable->Get(key, value, s, merge_context)) {
       return true;
     }
   }
@@ -73,9 +72,10 @@ bool MemTableListVersion::Get(const LookupKey& key, std::string* value,
 }
 
 void MemTableListVersion::AddIterators(const ReadOptions& options,
-                                       std::vector<Iterator*>* iterator_list) {
+                                       std::vector<Iterator*>* iterator_list,
+                                       Arena* arena) {
   for (auto& m : memlist_) {
-    iterator_list->push_back(m->NewIterator(options));
+    iterator_list->push_back(m->NewIterator(options, arena));
   }
 }
 
@@ -160,7 +160,8 @@ void MemTableList::RollbackMemtableFlush(const autovector<MemTable*>& mems,
 
 // Record a successful flush in the manifest file
 Status MemTableList::InstallMemtableFlushResults(
-    ColumnFamilyData* cfd, const autovector<MemTable*>& mems, VersionSet* vset,
+    ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
+    const autovector<MemTable*>& mems, VersionSet* vset,
     port::Mutex* mu, Logger* info_log, uint64_t file_number,
     FileNumToPathIdMap* pending_outputs, autovector<MemTable*>* to_delete,
     Directory* db_directory, LogBuffer* log_buffer) {
@@ -197,7 +198,7 @@ Status MemTableList::InstallMemtableFlushResults(
                 cfd->GetName().c_str(), (unsigned long)m->file_number_);
 
     // this can release and reacquire the mutex.
-    s = vset->LogAndApply(cfd, &m->edit_, mu, db_directory);
+    s = vset->LogAndApply(cfd, mutable_cf_options, &m->edit_, mu, db_directory);
 
     // we will be changing the version in the next code path,
     // so we better create a new one, since versions are immutable

db/memtable_list.h

@@ -46,10 +46,10 @@ class MemTableListVersion {
   // Search all the memtables starting from the most recent one.
   // Return the most recent value found, if any.
   bool Get(const LookupKey& key, std::string* value, Status* s,
-           MergeContext& merge_context, const Options& options);
+           MergeContext* merge_context);
 
   void AddIterators(const ReadOptions& options,
-                    std::vector<Iterator*>* iterator_list);
+                    std::vector<Iterator*>* iterator_list, Arena* arena);
 
   void AddIterators(const ReadOptions& options,
                     MergeIteratorBuilder* merge_iter_builder);
@@ -113,7 +113,8 @@ class MemTableList {
 
   // Commit a successful flush in the manifest file
   Status InstallMemtableFlushResults(
-      ColumnFamilyData* cfd, const autovector<MemTable*>& m, VersionSet* vset,
+      ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
+      const autovector<MemTable*>& m, VersionSet* vset,
       port::Mutex* mu, Logger* info_log, uint64_t file_number,
       FileNumToPathIdMap* pending_outputs, autovector<MemTable*>* to_delete,
       Directory* db_directory, LogBuffer* log_buffer);

db/merge_helper.cc

@@ -24,10 +24,12 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
                              bool at_bottom, Statistics* stats, int* steps) {
   // Get a copy of the internal key, before it's invalidated by iter->Next()
   // Also maintain the list of merge operands seen.
+  assert(HasOperator());
   keys_.clear();
   operands_.clear();
   keys_.push_front(iter->key().ToString());
   operands_.push_front(iter->value().ToString());
+  assert(user_merge_operator_);
 
   success_ = false;   // Will become true if we hit Put/Delete or bottom
 

db/merge_helper.h

@@ -78,13 +78,16 @@ class MergeHelper {
   //   IMPORTANT 2: The entries were traversed in order from BACK to FRONT.
   //                So keys().back() was the first key seen by iterator.
   // TODO: Re-style this comment to be like the first one
-  bool IsSuccess() { return success_; }
-  Slice key() { assert(success_); return Slice(keys_.back()); }
-  Slice value() { assert(success_); return Slice(operands_.back()); }
-  const std::deque<std::string>& keys() { assert(!success_); return keys_; }
-  const std::deque<std::string>& values() {
+  bool IsSuccess() const { return success_; }
+  Slice key() const { assert(success_); return Slice(keys_.back()); }
+  Slice value() const { assert(success_); return Slice(operands_.back()); }
+  const std::deque<std::string>& keys() const {
+    assert(!success_); return keys_;
+  }
+  const std::deque<std::string>& values() const {
     assert(!success_); return operands_;
   }
+  bool HasOperator() const { return user_merge_operator_ != nullptr; }
 
  private:
   const Comparator* user_comparator_;

db/merge_test.cc

@@ -12,12 +12,12 @@
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/merge_operator.h"
+#include "rocksdb/utilities/db_ttl.h"
 #include "db/dbformat.h"
 #include "db/db_impl.h"
 #include "db/write_batch_internal.h"
 #include "utilities/merge_operators.h"
 #include "util/testharness.h"
-#include "utilities/db_ttl.h"
 
 using namespace std;
 using namespace rocksdb;
@@ -212,7 +212,7 @@ class Counters {
   void assert_add(const string& key, uint64_t value) {
     int result = add(key, value);
     assert(result);
-    if (result == 0) exit(1); // Disable unused variable warning. 
+    if (result == 0) exit(1); // Disable unused variable warning.
   }
 };
 
@@ -460,6 +460,41 @@ void runTest(int argc, const string& dbname, const bool use_ttl = false) {
       }
     }
   }
+
+  {
+    cout << "Test merge-operator not set after reopen\n";
+    {
+      auto db = OpenDb(dbname);
+      MergeBasedCounters counters(db, 0);
+      counters.add("test-key", 1);
+      counters.add("test-key", 1);
+      counters.add("test-key", 1);
+      db->CompactRange(nullptr, nullptr);
+    }
+
+    DB* reopen_db;
+    ASSERT_OK(DB::Open(Options(), dbname, &reopen_db));
+    std::string value;
+    ASSERT_TRUE(!(reopen_db->Get(ReadOptions(), "test-key", &value).ok()));
+    delete reopen_db;
+    DestroyDB(dbname, Options());
+  }
+
+  /* Temporary remove this test
+  {
+    cout << "Test merge-operator not set after reopen (recovery case)\n";
+    {
+      auto db = OpenDb(dbname);
+      MergeBasedCounters counters(db, 0);
+      counters.add("test-key", 1);
+      counters.add("test-key", 1);
+      counters.add("test-key", 1);
+    }
+
+    DB* reopen_db;
+    ASSERT_TRUE(DB::Open(Options(), dbname, &reopen_db).IsInvalidArgument());
+  }
+  */
 }
 }  // namespace
 

db/perf_context_test.cc

@@ -6,7 +6,6 @@
 #include <algorithm>
 #include <iostream>
 #include <vector>
-#include "/usr/include/valgrind/callgrind.h"
 
 #include "rocksdb/db.h"
 #include "rocksdb/perf_context.h"
@@ -29,7 +28,7 @@ const std::string kDbName = rocksdb::test::TmpDir() + "/perf_context_test";
 
 namespace rocksdb {
 
-std::shared_ptr<DB> OpenDb() {
+std::shared_ptr<DB> OpenDb(bool read_only = false) {
     DB* db;
     Options options;
     options.create_if_missing = true;
@@ -39,12 +38,16 @@ std::shared_ptr<DB> OpenDb() {
       FLAGS_min_write_buffer_number_to_merge;
 
     if (FLAGS_use_set_based_memetable) {
-      auto prefix_extractor = rocksdb::NewFixedPrefixTransform(0);
-      options.memtable_factory.reset(
-          NewHashSkipListRepFactory(prefix_extractor));
+      options.prefix_extractor.reset(rocksdb::NewFixedPrefixTransform(0));
+      options.memtable_factory.reset(NewHashSkipListRepFactory());
     }
 
-    Status s = DB::Open(options, kDbName,  &db);
+    Status s;
+    if (!read_only) {
+      s = DB::Open(options, kDbName, &db);
+    } else {
+      s = DB::OpenForReadOnly(options, kDbName, &db);
+    }
     ASSERT_OK(s);
     return std::shared_ptr<DB>(db);
 }
@@ -76,7 +79,8 @@ TEST(PerfContextTest, SeekIntoDeletion) {
     std::string value;
 
     perf_context.Reset();
-    StopWatchNano timer(Env::Default(), true);
+    StopWatchNano timer(Env::Default());
+    timer.Start();
     auto status = db->Get(read_options, key, &value);
     auto elapsed_nanos = timer.ElapsedNanos();
     ASSERT_TRUE(status.IsNotFound());
@@ -149,11 +153,12 @@ TEST(PerfContextTest, StopWatchNanoOverhead) {
 TEST(PerfContextTest, StopWatchOverhead) {
   // profile the timer cost by itself!
   const int kTotalIterations = 1000000;
+  uint64_t elapsed = 0;
   std::vector<uint64_t> timings(kTotalIterations);
 
-  StopWatch timer(Env::Default());
+  StopWatch timer(Env::Default(), nullptr, 0, &elapsed);
   for (auto& timing : timings) {
-    timing = timer.ElapsedMicros();
+    timing = elapsed;
   }
 
   HistogramImpl histogram;
@@ -166,7 +171,7 @@ TEST(PerfContextTest, StopWatchOverhead) {
   std::cout << histogram.ToString();
 }
 
-void ProfileKeyComparison() {
+void ProfileQueries(bool enabled_time = false) {
   DestroyDB(kDbName, Options());    // Start this test with a fresh DB
 
   auto db = OpenDb();
@@ -175,11 +180,21 @@ void ProfileKeyComparison() {
   ReadOptions read_options;
 
   HistogramImpl hist_put;
+
   HistogramImpl hist_get;
   HistogramImpl hist_get_snapshot;
   HistogramImpl hist_get_memtable;
+  HistogramImpl hist_get_files;
   HistogramImpl hist_get_post_process;
   HistogramImpl hist_num_memtable_checked;
+
+  HistogramImpl hist_mget;
+  HistogramImpl hist_mget_snapshot;
+  HistogramImpl hist_mget_memtable;
+  HistogramImpl hist_mget_files;
+  HistogramImpl hist_mget_post_process;
+  HistogramImpl hist_mget_num_memtable_checked;
+
   HistogramImpl hist_write_pre_post;
   HistogramImpl hist_write_wal_time;
   HistogramImpl hist_write_memtable_time;
@@ -187,8 +202,13 @@ void ProfileKeyComparison() {
   std::cout << "Inserting " << FLAGS_total_keys << " key/value pairs\n...\n";
 
   std::vector<int> keys;
+  const int kFlushFlag = -1;
   for (int i = 0; i < FLAGS_total_keys; ++i) {
     keys.push_back(i);
+    if (i == FLAGS_total_keys / 2) {
+      // Issuing a flush in the middle.
+      keys.push_back(kFlushFlag);
+    }
   }
 
   if (FLAGS_random_key) {
@@ -196,27 +216,54 @@ void ProfileKeyComparison() {
   }
 
   for (const int i : keys) {
+    if (i == kFlushFlag) {
+      FlushOptions fo;
+      db->Flush(fo);
+      continue;
+    }
     std::string key = "k" + std::to_string(i);
     std::string value = "v" + std::to_string(i);
 
+    std::vector<Slice> keys = {Slice(key)};
+    std::vector<std::string> values;
+
     perf_context.Reset();
     db->Put(write_options, key, value);
     hist_write_pre_post.Add(perf_context.write_pre_and_post_process_time);
     hist_write_wal_time.Add(perf_context.write_wal_time);
     hist_write_memtable_time.Add(perf_context.write_memtable_time);
     hist_put.Add(perf_context.user_key_comparison_count);
+  }
+
+  for (const int i : keys) {
+    std::string key = "k" + std::to_string(i);
+    std::string value = "v" + std::to_string(i);
+
+    std::vector<Slice> keys = {Slice(key)};
+    std::vector<std::string> values;
 
     perf_context.Reset();
     db->Get(read_options, key, &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);
     hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
     hist_get_post_process.Add(perf_context.get_post_process_time);
     hist_get.Add(perf_context.user_key_comparison_count);
+
+    perf_context.Reset();
+    db->MultiGet(read_options, keys, &values);
+    hist_mget_snapshot.Add(perf_context.get_snapshot_time);
+    hist_mget_memtable.Add(perf_context.get_from_memtable_time);
+    hist_mget_files.Add(perf_context.get_from_output_files_time);
+    hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
+    hist_mget_post_process.Add(perf_context.get_post_process_time);
+    hist_mget.Add(perf_context.user_key_comparison_count);
   }
 
   std::cout << "Put uesr key comparison: \n" << hist_put.ToString()
-            << "Get uesr key comparison: \n" << hist_get.ToString();
+            << "Get uesr key comparison: \n" << hist_get.ToString()
+            << "MultiGet uesr key comparison: \n" << hist_get.ToString();
   std::cout << "Put(): Pre and Post Process Time: \n"
             << hist_write_pre_post.ToString()
             << " Writing WAL time: \n"
@@ -224,25 +271,139 @@ void ProfileKeyComparison() {
             << " Writing Mem Table time: \n"
             << hist_write_memtable_time.ToString() << "\n";
 
-  std::cout << "Get(): Time to get snapshot: \n"
+  std::cout << "Get(): Time to get snapshot: \n" << hist_get_snapshot.ToString()
+            << " Time to get value from memtables: \n"
+            << hist_get_memtable.ToString() << "\n"
+            << " Time to get value from output files: \n"
+            << hist_get_files.ToString() << "\n"
+            << " Number of memtables checked: \n"
+            << hist_num_memtable_checked.ToString() << "\n"
+            << " Time to post process: \n" << hist_get_post_process.ToString()
+            << "\n";
+
+  std::cout << "MultiGet(): Time to get snapshot: \n"
+            << hist_mget_snapshot.ToString()
+            << " Time to get value from memtables: \n"
+            << hist_mget_memtable.ToString() << "\n"
+            << " Time to get value from output files: \n"
+            << hist_mget_files.ToString() << "\n"
+            << " Number of memtables checked: \n"
+            << hist_mget_num_memtable_checked.ToString() << "\n"
+            << " Time to post process: \n" << hist_mget_post_process.ToString()
+            << "\n";
+
+  if (enabled_time) {
+    ASSERT_GT(hist_get.Average(), 0);
+    ASSERT_GT(hist_get_snapshot.Average(), 0);
+    ASSERT_GT(hist_get_memtable.Average(), 0);
+    ASSERT_GT(hist_get_files.Average(), 0);
+    ASSERT_GT(hist_get_post_process.Average(), 0);
+    ASSERT_GT(hist_num_memtable_checked.Average(), 0);
+
+    ASSERT_GT(hist_mget.Average(), 0);
+    ASSERT_GT(hist_mget_snapshot.Average(), 0);
+    ASSERT_GT(hist_mget_memtable.Average(), 0);
+    ASSERT_GT(hist_mget_files.Average(), 0);
+    ASSERT_GT(hist_mget_post_process.Average(), 0);
+    ASSERT_GT(hist_mget_num_memtable_checked.Average(), 0);
+  }
+
+  db.reset();
+  db = OpenDb(true);
+
+  hist_get.Clear();
+  hist_get_snapshot.Clear();
+  hist_get_memtable.Clear();
+  hist_get_files.Clear();
+  hist_get_post_process.Clear();
+  hist_num_memtable_checked.Clear();
+
+  hist_mget.Clear();
+  hist_mget_snapshot.Clear();
+  hist_mget_memtable.Clear();
+  hist_mget_files.Clear();
+  hist_mget_post_process.Clear();
+  hist_mget_num_memtable_checked.Clear();
+
+  for (const int i : keys) {
+    std::string key = "k" + std::to_string(i);
+    std::string value = "v" + std::to_string(i);
+
+    std::vector<Slice> keys = {Slice(key)};
+    std::vector<std::string> values;
+
+    perf_context.Reset();
+    db->Get(read_options, key, &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);
+    hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
+    hist_get_post_process.Add(perf_context.get_post_process_time);
+    hist_get.Add(perf_context.user_key_comparison_count);
+
+    perf_context.Reset();
+    db->MultiGet(read_options, keys, &values);
+    hist_mget_snapshot.Add(perf_context.get_snapshot_time);
+    hist_mget_memtable.Add(perf_context.get_from_memtable_time);
+    hist_mget_files.Add(perf_context.get_from_output_files_time);
+    hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
+    hist_mget_post_process.Add(perf_context.get_post_process_time);
+    hist_mget.Add(perf_context.user_key_comparison_count);
+  }
+
+  std::cout << "ReadOnly Get uesr key comparison: \n" << hist_get.ToString()
+            << "ReadOnly MultiGet uesr key comparison: \n"
+            << hist_mget.ToString();
+
+  std::cout << "ReadOnly Get(): Time to get snapshot: \n"
             << hist_get_snapshot.ToString()
             << " Time to get value from memtables: \n"
             << hist_get_memtable.ToString() << "\n"
+            << " Time to get value from output files: \n"
+            << hist_get_files.ToString() << "\n"
             << " Number of memtables checked: \n"
             << hist_num_memtable_checked.ToString() << "\n"
-            << " Time to post process: \n"
-            << hist_get_post_process.ToString() << "\n";
+            << " Time to post process: \n" << hist_get_post_process.ToString()
+            << "\n";
+
+  std::cout << "ReadOnly MultiGet(): Time to get snapshot: \n"
+            << hist_mget_snapshot.ToString()
+            << " Time to get value from memtables: \n"
+            << hist_mget_memtable.ToString() << "\n"
+            << " Time to get value from output files: \n"
+            << hist_mget_files.ToString() << "\n"
+            << " Number of memtables checked: \n"
+            << hist_mget_num_memtable_checked.ToString() << "\n"
+            << " Time to post process: \n" << hist_mget_post_process.ToString()
+            << "\n";
+
+  if (enabled_time) {
+    ASSERT_GT(hist_get.Average(), 0);
+    ASSERT_GT(hist_get_memtable.Average(), 0);
+    ASSERT_GT(hist_get_files.Average(), 0);
+    ASSERT_GT(hist_num_memtable_checked.Average(), 0);
+    // In read-only mode Get(), no super version operation is needed
+    ASSERT_EQ(hist_get_post_process.Average(), 0);
+    ASSERT_EQ(hist_get_snapshot.Average(), 0);
+
+    ASSERT_GT(hist_mget.Average(), 0);
+    ASSERT_GT(hist_mget_snapshot.Average(), 0);
+    ASSERT_GT(hist_mget_memtable.Average(), 0);
+    ASSERT_GT(hist_mget_files.Average(), 0);
+    ASSERT_GT(hist_mget_post_process.Average(), 0);
+    ASSERT_GT(hist_mget_num_memtable_checked.Average(), 0);
+  }
 }
 
 TEST(PerfContextTest, KeyComparisonCount) {
   SetPerfLevel(kEnableCount);
-  ProfileKeyComparison();
+  ProfileQueries();
 
   SetPerfLevel(kDisable);
-  ProfileKeyComparison();
+  ProfileQueries();
 
   SetPerfLevel(kEnableTime);
-  ProfileKeyComparison();
+  ProfileQueries(true);
 }
 
 // make perf_context_test

db/plain_table_db_test.cc

@@ -23,6 +23,7 @@
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/table.h"
 #include "table/meta_blocks.h"
+#include "table/bloom_block.h"
 #include "table/plain_table_factory.h"
 #include "table/plain_table_reader.h"
 #include "util/hash.h"
@@ -61,8 +62,20 @@ class PlainTableDBTest {
   // Return the current option configuration.
   Options CurrentOptions() {
     Options options;
-    options.table_factory.reset(NewPlainTableFactory(0, 2, 0.8, 3, 0, kPrefix));
-    options.memtable_factory.reset(NewHashLinkListRepFactory(4, 0, 3, true, 3));
+
+    PlainTableOptions plain_table_options;
+    plain_table_options.user_key_len = 0;
+    plain_table_options.bloom_bits_per_key = 2;
+    plain_table_options.hash_table_ratio = 0.8;
+    plain_table_options.index_sparseness = 3;
+    plain_table_options.huge_page_tlb_size = 0;
+    plain_table_options.encoding_type = kPrefix;
+    plain_table_options.full_scan_mode = false;
+    plain_table_options.store_index_in_file = false;
+
+    options.table_factory.reset(NewPlainTableFactory(plain_table_options));
+    options.memtable_factory.reset(NewHashLinkListRepFactory(4, 0, 3, true));
+
     options.prefix_extractor.reset(NewFixedPrefixTransform(8));
     options.allow_mmap_reads = true;
     return options;
@@ -175,17 +188,21 @@ TEST(PlainTableDBTest, Empty) {
   ASSERT_EQ("NOT_FOUND", Get("0000000000000foo"));
 }
 
+extern const uint64_t kPlainTableMagicNumber;
+
 class TestPlainTableReader : public PlainTableReader {
  public:
-  TestPlainTableReader(const EnvOptions& storage_options,
+  TestPlainTableReader(const EnvOptions& env_options,
                        const InternalKeyComparator& icomparator,
                        EncodingType encoding_type, uint64_t file_size,
                        int bloom_bits_per_key, double hash_table_ratio,
                        size_t index_sparseness,
                        const TableProperties* table_properties,
                        unique_ptr<RandomAccessFile>&& file,
-                       const Options& options, bool* expect_bloom_not_match)
-      : PlainTableReader(options, std::move(file), storage_options, icomparator,
+                       const ImmutableCFOptions& ioptions,
+                       bool* expect_bloom_not_match,
+                       bool store_index_in_file)
+      : PlainTableReader(ioptions, std::move(file), env_options, icomparator,
                          encoding_type, file_size, table_properties),
         expect_bloom_not_match_(expect_bloom_not_match) {
     Status s = MmapDataFile();
@@ -195,6 +212,19 @@ class TestPlainTableReader : public PlainTableReader {
                       bloom_bits_per_key, hash_table_ratio, index_sparseness,
                       2 * 1024 * 1024);
     ASSERT_TRUE(s.ok());
+
+    TableProperties* props = const_cast<TableProperties*>(table_properties);
+    if (store_index_in_file) {
+      auto bloom_version_ptr = props->user_collected_properties.find(
+          PlainTablePropertyNames::kBloomVersion);
+      ASSERT_TRUE(bloom_version_ptr != props->user_collected_properties.end());
+      ASSERT_EQ(bloom_version_ptr->second, std::string("1"));
+      if (ioptions.bloom_locality > 0) {
+        auto num_blocks_ptr = props->user_collected_properties.find(
+            PlainTablePropertyNames::kNumBloomBlocks);
+        ASSERT_TRUE(num_blocks_ptr != props->user_collected_properties.end());
+      }
+    }
   }
 
   virtual ~TestPlainTableReader() {}
@@ -202,7 +232,11 @@ class TestPlainTableReader : public PlainTableReader {
  private:
   virtual bool MatchBloom(uint32_t hash) const override {
     bool ret = PlainTableReader::MatchBloom(hash);
-    ASSERT_TRUE(!*expect_bloom_not_match_ || !ret);
+    if (*expect_bloom_not_match_) {
+      ASSERT_TRUE(!ret);
+    } else {
+      ASSERT_TRUE(ret);
+    }
     return ret;
   }
   bool* expect_bloom_not_match_;
@@ -212,27 +246,38 @@ extern const uint64_t kPlainTableMagicNumber;
 class TestPlainTableFactory : public PlainTableFactory {
  public:
   explicit TestPlainTableFactory(bool* expect_bloom_not_match,
-                                 uint32_t user_key_len, int bloom_bits_per_key,
-                                 double hash_table_ratio,
-                                 size_t index_sparseness,
-                                 size_t huge_page_tlb_size,
-                                 EncodingType encoding_type)
-      : PlainTableFactory(user_key_len, bloom_bits_per_key, hash_table_ratio,
-                          index_sparseness, huge_page_tlb_size, encoding_type),
-        bloom_bits_per_key_(bloom_bits_per_key),
-        hash_table_ratio_(hash_table_ratio),
-        index_sparseness_(index_sparseness),
+                                 const PlainTableOptions& options)
+      : 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) {}
 
-  Status NewTableReader(const Options& options, const EnvOptions& soptions,
+  Status NewTableReader(const ImmutableCFOptions& ioptions,
+                        const EnvOptions& env_options,
                         const InternalKeyComparator& internal_comparator,
                         unique_ptr<RandomAccessFile>&& file, uint64_t file_size,
                         unique_ptr<TableReader>* table) const override {
     TableProperties* props = nullptr;
     auto s = ReadTableProperties(file.get(), file_size, kPlainTableMagicNumber,
-                                 options.env, options.info_log.get(), &props);
+                                 ioptions.env, ioptions.info_log, &props);
     ASSERT_TRUE(s.ok());
 
+    if (store_index_in_file_) {
+      BlockHandle bloom_block_handle;
+      s = FindMetaBlock(file.get(), file_size, kPlainTableMagicNumber,
+                        ioptions.env, BloomBlockBuilder::kBloomBlock,
+                        &bloom_block_handle);
+      ASSERT_TRUE(s.ok());
+
+      BlockHandle index_block_handle;
+      s = FindMetaBlock(
+          file.get(), file_size, kPlainTableMagicNumber, ioptions.env,
+          PlainTableIndexBuilder::kPlainTableIndexBlock, &index_block_handle);
+      ASSERT_TRUE(s.ok());
+    }
+
     auto& user_props = props->user_collected_properties;
     auto encoding_type_prop =
         user_props.find(PlainTablePropertyNames::kEncodingType);
@@ -241,9 +286,10 @@ class TestPlainTableFactory : public PlainTableFactory {
         DecodeFixed32(encoding_type_prop->second.c_str()));
 
     std::unique_ptr<PlainTableReader> new_reader(new TestPlainTableReader(
-        soptions, internal_comparator, encoding_type, file_size,
+        env_options, internal_comparator, encoding_type, file_size,
         bloom_bits_per_key_, hash_table_ratio_, index_sparseness_, props,
-        std::move(file), options, expect_bloom_not_match_));
+        std::move(file), ioptions, expect_bloom_not_match_,
+        store_index_in_file_));
 
     *table = std::move(new_reader);
     return s;
@@ -253,6 +299,7 @@ class TestPlainTableFactory : public PlainTableFactory {
   int bloom_bits_per_key_;
   double hash_table_ratio_;
   size_t index_sparseness_;
+  bool store_index_in_file_;
   bool* expect_bloom_not_match_;
 };
 
@@ -262,40 +309,84 @@ TEST(PlainTableDBTest, Flush) {
     for (EncodingType encoding_type : {kPlain, kPrefix}) {
     for (int bloom_bits = 0; bloom_bits <= 117; bloom_bits += 117) {
       for (int total_order = 0; total_order <= 1; total_order++) {
-        Options options = CurrentOptions();
-        options.create_if_missing = true;
-        // Set only one bucket to force bucket conflict.
-        // Test index interval for the same prefix to be 1, 2 and 4
-        if (total_order) {
-          options.prefix_extractor.reset();
-          options.table_factory.reset(NewPlainTableFactory(
-              0, bloom_bits, 0, 2, huge_page_tlb_size, encoding_type));
-        } else {
-          options.table_factory.reset(NewPlainTableFactory(
-              0, bloom_bits, 0.75, 16, huge_page_tlb_size, encoding_type));
+        for (int store_index_in_file = 0; store_index_in_file <= 1;
+             ++store_index_in_file) {
+          if (!bloom_bits && store_index_in_file) {
+            continue;
+          }
+
+          Options options = CurrentOptions();
+          options.create_if_missing = true;
+          // Set only one bucket to force bucket conflict.
+          // Test index interval for the same prefix to be 1, 2 and 4
+          if (total_order) {
+            options.prefix_extractor.reset();
+
+            PlainTableOptions plain_table_options;
+            plain_table_options.user_key_len = 0;
+            plain_table_options.bloom_bits_per_key = bloom_bits;
+            plain_table_options.hash_table_ratio = 0;
+            plain_table_options.index_sparseness = 2;
+            plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+            plain_table_options.encoding_type = encoding_type;
+            plain_table_options.full_scan_mode = false;
+            plain_table_options.store_index_in_file = store_index_in_file;
+
+            options.table_factory.reset(
+                NewPlainTableFactory(plain_table_options));
+          } else {
+            PlainTableOptions plain_table_options;
+            plain_table_options.user_key_len = 0;
+            plain_table_options.bloom_bits_per_key = bloom_bits;
+            plain_table_options.hash_table_ratio = 0.75;
+            plain_table_options.index_sparseness = 16;
+            plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+            plain_table_options.encoding_type = encoding_type;
+            plain_table_options.full_scan_mode = false;
+            plain_table_options.store_index_in_file = store_index_in_file;
+
+            options.table_factory.reset(
+                NewPlainTableFactory(plain_table_options));
+          }
+          DestroyAndReopen(&options);
+          uint64_t int_num;
+          ASSERT_TRUE(dbfull()->GetIntProperty(
+              "rocksdb.estimate-table-readers-mem", &int_num));
+          ASSERT_EQ(int_num, 0U);
+
+          ASSERT_OK(Put("1000000000000foo", "v1"));
+          ASSERT_OK(Put("0000000000000bar", "v2"));
+          ASSERT_OK(Put("1000000000000foo", "v3"));
+          dbfull()->TEST_FlushMemTable();
+
+          ASSERT_TRUE(dbfull()->GetIntProperty(
+              "rocksdb.estimate-table-readers-mem", &int_num));
+          ASSERT_GT(int_num, 0U);
+
+          TablePropertiesCollection ptc;
+          reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
+          ASSERT_EQ(1U, ptc.size());
+          auto row = ptc.begin();
+          auto tp = row->second;
+
+          if (!store_index_in_file) {
+            ASSERT_EQ(total_order ? "4" : "12",
+                      (tp->user_collected_properties)
+                          .at("plain_table_hash_table_size"));
+            ASSERT_EQ("0", (tp->user_collected_properties)
+                               .at("plain_table_sub_index_size"));
+          } else {
+            ASSERT_EQ("0", (tp->user_collected_properties)
+                               .at("plain_table_hash_table_size"));
+            ASSERT_EQ("0", (tp->user_collected_properties)
+                               .at("plain_table_sub_index_size"));
+          }
+          ASSERT_EQ("v3", Get("1000000000000foo"));
+          ASSERT_EQ("v2", Get("0000000000000bar"));
+        }
         }
-        DestroyAndReopen(&options);
-
-        ASSERT_OK(Put("1000000000000foo", "v1"));
-        ASSERT_OK(Put("0000000000000bar", "v2"));
-        ASSERT_OK(Put("1000000000000foo", "v3"));
-        dbfull()->TEST_FlushMemTable();
-
-        TablePropertiesCollection ptc;
-        reinterpret_cast<DB*>(dbfull())->GetPropertiesOfAllTables(&ptc);
-        ASSERT_EQ(1U, ptc.size());
-        auto row = ptc.begin();
-        auto tp = row->second;
-        ASSERT_EQ(total_order ? "4" : "12", (tp->user_collected_properties).at(
-                                                "plain_table_hash_table_size"));
-        ASSERT_EQ("0", (tp->user_collected_properties)
-                           .at("plain_table_sub_index_size"));
-
-        ASSERT_EQ("v3", Get("1000000000000foo"));
-        ASSERT_EQ("v2", Get("0000000000000bar"));
       }
     }
-    }
   }
 }
 
@@ -305,7 +396,15 @@ TEST(PlainTableDBTest, Flush2) {
     for (EncodingType encoding_type : {kPlain, kPrefix}) {
     for (int bloom_bits = 0; bloom_bits <= 117; bloom_bits += 117) {
       for (int total_order = 0; total_order <= 1; total_order++) {
-        if (encoding_type == kPrefix && total_order == 1) {
+        for (int store_index_in_file = 0; store_index_in_file <= 1;
+             ++store_index_in_file) {
+          if (encoding_type == kPrefix && total_order) {
+            continue;
+          }
+          if (!bloom_bits && store_index_in_file) {
+            continue;
+          }
+          if (total_order && store_index_in_file) {
           continue;
         }
         bool expect_bloom_not_match = false;
@@ -313,16 +412,23 @@ TEST(PlainTableDBTest, Flush2) {
         options.create_if_missing = true;
         // Set only one bucket to force bucket conflict.
         // Test index interval for the same prefix to be 1, 2 and 4
+        PlainTableOptions plain_table_options;
         if (total_order) {
           options.prefix_extractor = nullptr;
-          options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, 0, bloom_bits, 0, 2, huge_page_tlb_size,
-              encoding_type));
+          plain_table_options.hash_table_ratio = 0;
+          plain_table_options.index_sparseness = 2;
         } else {
-          options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, 0, bloom_bits, 0.75, 16,
-              huge_page_tlb_size, encoding_type));
+          plain_table_options.hash_table_ratio = 0.75;
+          plain_table_options.index_sparseness = 16;
         }
+        plain_table_options.user_key_len = kPlainTableVariableLength;
+        plain_table_options.bloom_bits_per_key = bloom_bits;
+        plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+        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));
+
         DestroyAndReopen(&options);
         ASSERT_OK(Put("0000000000000bar", "b"));
         ASSERT_OK(Put("1000000000000foo", "v1"));
@@ -350,7 +456,6 @@ TEST(PlainTableDBTest, Flush2) {
           // Neither key nor value should exist.
           expect_bloom_not_match = true;
           ASSERT_EQ("NOT_FOUND", Get("5_not00000000bar"));
-
           // Key doesn't exist any more but prefix exists.
           if (total_order) {
             ASSERT_EQ("NOT_FOUND", Get("1000000000000not"));
@@ -359,6 +464,7 @@ TEST(PlainTableDBTest, Flush2) {
           expect_bloom_not_match = false;
         }
       }
+      }
     }
     }
   }
@@ -380,13 +486,28 @@ TEST(PlainTableDBTest, Iterator) {
         // Test index interval for the same prefix to be 1, 2 and 4
         if (total_order) {
           options.prefix_extractor = nullptr;
+
+          PlainTableOptions plain_table_options;
+          plain_table_options.user_key_len = 16;
+          plain_table_options.bloom_bits_per_key = bloom_bits;
+          plain_table_options.hash_table_ratio = 0;
+          plain_table_options.index_sparseness = 2;
+          plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+          plain_table_options.encoding_type = encoding_type;
+
           options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, 16, bloom_bits, 0, 2, huge_page_tlb_size,
-              encoding_type));
+              &expect_bloom_not_match, plain_table_options));
         } else {
+          PlainTableOptions plain_table_options;
+          plain_table_options.user_key_len = 16;
+          plain_table_options.bloom_bits_per_key = bloom_bits;
+          plain_table_options.hash_table_ratio = 0.75;
+          plain_table_options.index_sparseness = 16;
+          plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+          plain_table_options.encoding_type = encoding_type;
+
           options.table_factory.reset(new TestPlainTableFactory(
-              &expect_bloom_not_match, 16, bloom_bits, 0.75, 16,
-              huge_page_tlb_size, encoding_type));
+              &expect_bloom_not_match, plain_table_options));
         }
         DestroyAndReopen(&options);
 
@@ -485,7 +606,13 @@ std::string MakeLongKey(size_t length, char c) {
 
 TEST(PlainTableDBTest, IteratorLargeKeys) {
   Options options = CurrentOptions();
-  options.table_factory.reset(NewPlainTableFactory(0, 0, 0));
+
+  PlainTableOptions plain_table_options;
+  plain_table_options.user_key_len = 0;
+  plain_table_options.bloom_bits_per_key = 0;
+  plain_table_options.hash_table_ratio = 0;
+
+  options.table_factory.reset(NewPlainTableFactory(plain_table_options));
   options.create_if_missing = true;
   options.prefix_extractor.reset();
   DestroyAndReopen(&options);
@@ -529,7 +656,16 @@ std::string MakeLongKeyWithPrefix(size_t length, char c) {
 
 TEST(PlainTableDBTest, IteratorLargeKeysWithPrefix) {
   Options options = CurrentOptions();
-  options.table_factory.reset(NewPlainTableFactory(16, 0, 0.8, 3, 0, kPrefix));
+
+  PlainTableOptions plain_table_options;
+  plain_table_options.user_key_len = 16;
+  plain_table_options.bloom_bits_per_key = 0;
+  plain_table_options.hash_table_ratio = 0.8;
+  plain_table_options.index_sparseness = 3;
+  plain_table_options.huge_page_tlb_size = 0;
+  plain_table_options.encoding_type = kPrefix;
+
+  options.table_factory.reset(NewPlainTableFactory(plain_table_options));
   options.create_if_missing = true;
   DestroyAndReopen(&options);
 
@@ -665,8 +801,16 @@ TEST(PlainTableDBTest, HashBucketConflict) {
       options.create_if_missing = true;
       // Set only one bucket to force bucket conflict.
       // Test index interval for the same prefix to be 1, 2 and 4
-      options.table_factory.reset(
-          NewPlainTableFactory(16, 0, 0, 2 ^ i, huge_page_tlb_size));
+
+      PlainTableOptions plain_table_options;
+      plain_table_options.user_key_len = 16;
+      plain_table_options.bloom_bits_per_key = 0;
+      plain_table_options.hash_table_ratio = 0;
+      plain_table_options.index_sparseness = 2 ^ i;
+      plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+
+      options.table_factory.reset(NewPlainTableFactory(plain_table_options));
+
       DestroyAndReopen(&options);
       ASSERT_OK(Put("5000000000000fo0", "v1"));
       ASSERT_OK(Put("5000000000000fo1", "v2"));
@@ -752,8 +896,15 @@ TEST(PlainTableDBTest, HashBucketConflictReverseSuffixComparator) {
       options.comparator = &comp;
       // Set only one bucket to force bucket conflict.
       // Test index interval for the same prefix to be 1, 2 and 4
-      options.table_factory.reset(
-          NewPlainTableFactory(16, 0, 0, 2 ^ i, huge_page_tlb_size));
+
+      PlainTableOptions plain_table_options;
+      plain_table_options.user_key_len = 16;
+      plain_table_options.bloom_bits_per_key = 0;
+      plain_table_options.hash_table_ratio = 0;
+      plain_table_options.index_sparseness = 2 ^ i;
+      plain_table_options.huge_page_tlb_size = huge_page_tlb_size;
+
+      options.table_factory.reset(NewPlainTableFactory(plain_table_options));
       DestroyAndReopen(&options);
       ASSERT_OK(Put("5000000000000fo0", "v1"));
       ASSERT_OK(Put("5000000000000fo1", "v2"));
@@ -833,7 +984,13 @@ TEST(PlainTableDBTest, NonExistingKeyToNonEmptyBucket) {
   options.create_if_missing = true;
   // Set only one bucket to force bucket conflict.
   // Test index interval for the same prefix to be 1, 2 and 4
-  options.table_factory.reset(NewPlainTableFactory(16, 0, 0, 5));
+  PlainTableOptions plain_table_options;
+  plain_table_options.user_key_len = 16;
+  plain_table_options.bloom_bits_per_key = 0;
+  plain_table_options.hash_table_ratio = 0;
+  plain_table_options.index_sparseness = 5;
+
+  options.table_factory.reset(NewPlainTableFactory(plain_table_options));
   DestroyAndReopen(&options);
   ASSERT_OK(Put("5000000000000fo0", "v1"));
   ASSERT_OK(Put("5000000000000fo1", "v2"));

db/repair.cc

@@ -31,7 +31,10 @@
 
 #ifndef ROCKSDB_LITE
 
+#ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
+#endif
+
 #include <inttypes.h>
 #include "db/builder.h"
 #include "db/db_impl.h"
@@ -46,6 +49,9 @@
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
+#include "rocksdb/options.h"
+#include "rocksdb/immutable_options.h"
+#include "util/scoped_arena_iterator.h"
 
 namespace rocksdb {
 
@@ -57,8 +63,8 @@ class Repairer {
       : dbname_(dbname),
         env_(options.env),
         icmp_(options.comparator),
-        ipolicy_(options.filter_policy),
-        options_(SanitizeOptions(dbname, &icmp_, &ipolicy_, options)),
+        options_(SanitizeOptions(dbname, &icmp_, options)),
+        ioptions_(options_),
         raw_table_cache_(
             // TableCache can be small since we expect each table to be opened
             // once.
@@ -66,7 +72,7 @@ class Repairer {
                         options_.table_cache_remove_scan_count_limit)),
         next_file_number_(1) {
     table_cache_ =
-        new TableCache(&options_, storage_options_, raw_table_cache_.get());
+        new TableCache(ioptions_, env_options_, raw_table_cache_.get());
     edit_ = new VersionEdit();
   }
 
@@ -108,9 +114,9 @@ class Repairer {
 
   std::string const dbname_;
   Env* const env_;
-  InternalKeyComparator const icmp_;
-  InternalFilterPolicy const ipolicy_;
-  Options const options_;
+  const InternalKeyComparator icmp_;
+  const Options options_;
+  const ImmutableCFOptions ioptions_;
   std::shared_ptr<Cache> raw_table_cache_;
   TableCache* table_cache_;
   VersionEdit* edit_;
@@ -120,13 +126,14 @@ class Repairer {
   std::vector<uint64_t> logs_;
   std::vector<TableInfo> tables_;
   uint64_t next_file_number_;
-  const EnvOptions storage_options_;
+  const EnvOptions env_options_;
 
   Status FindFiles() {
     std::vector<std::string> filenames;
     bool found_file = false;
     for (uint32_t path_id = 0; path_id < options_.db_paths.size(); path_id++) {
-      Status status = env_->GetChildren(options_.db_paths[path_id], &filenames);
+      Status status =
+          env_->GetChildren(options_.db_paths[path_id].path, &filenames);
       if (!status.ok()) {
         return status;
       }
@@ -191,7 +198,7 @@ class Repairer {
     // Open the log file
     std::string logname = LogFileName(dbname_, log);
     unique_ptr<SequentialFile> lfile;
-    Status status = env_->NewSequentialFile(logname, &lfile, storage_options_);
+    Status status = env_->NewSequentialFile(logname, &lfile, env_options_);
     if (!status.ok()) {
       return status;
     }
@@ -212,7 +219,8 @@ class Repairer {
     std::string scratch;
     Slice record;
     WriteBatch batch;
-    MemTable* mem = new MemTable(icmp_, options_);
+    MemTable* mem = new MemTable(icmp_, ioptions_,
+        MemTableOptions(MutableCFOptions(options_, ioptions_), options_));
     auto cf_mems_default = new ColumnFamilyMemTablesDefault(mem, &options_);
     mem->Ref();
     int counter = 0;
@@ -237,11 +245,15 @@ class Repairer {
     // since ExtractMetaData() will also generate edits.
     FileMetaData meta;
     meta.fd = FileDescriptor(next_file_number_++, 0, 0);
-    ReadOptions ro;
-    Iterator* iter = mem->NewIterator(ro, true /* enforce_total_order */);
-    status = BuildTable(dbname_, env_, options_, storage_options_, table_cache_,
-                        iter, &meta, icmp_, 0, 0, kNoCompression);
-    delete iter;
+    {
+      ReadOptions ro;
+      ro.total_order_seek = true;
+      Arena arena;
+      ScopedArenaIterator iter(mem->NewIterator(ro, &arena));
+      status = BuildTable(dbname_, env_, ioptions_, env_options_, table_cache_,
+                          iter.get(), &meta, icmp_, 0, 0, kNoCompression,
+                          CompressionOptions());
+    }
     delete mem->Unref();
     delete cf_mems_default;
     mem = nullptr;
@@ -264,9 +276,10 @@ class Repairer {
       if (!status.ok()) {
         std::string fname = TableFileName(
             options_.db_paths, t.meta.fd.GetNumber(), t.meta.fd.GetPathId());
-        Log(options_.info_log, "Table #%s: ignoring %s",
-            FormatFileNumber(t.meta.fd.GetNumber(), t.meta.fd.GetPathId())
-                .c_str(),
+        char file_num_buf[kFormatFileNumberBufSize];
+        FormatFileNumber(t.meta.fd.GetNumber(), t.meta.fd.GetPathId(),
+                         file_num_buf, sizeof(file_num_buf));
+        Log(options_.info_log, "Table #%s: ignoring %s", file_num_buf,
             status.ToString().c_str());
         ArchiveFile(fname);
       } else {
@@ -285,7 +298,7 @@ class Repairer {
                                 file_size);
     if (status.ok()) {
       Iterator* iter = table_cache_->NewIterator(
-          ReadOptions(), storage_options_, icmp_, t->meta.fd);
+          ReadOptions(), env_options_, icmp_, t->meta.fd);
       bool empty = true;
       ParsedInternalKey parsed;
       t->min_sequence = 0;
@@ -325,7 +338,7 @@ class Repairer {
     std::string tmp = TempFileName(dbname_, 1);
     unique_ptr<WritableFile> file;
     Status status = env_->NewWritableFile(
-        tmp, &file, env_->OptimizeForManifestWrite(storage_options_));
+        tmp, &file, env_->OptimizeForManifestWrite(env_options_));
     if (!status.ok()) {
       return status;
     }

db/simple_table_db_test.cc

@@ -1,800 +0,0 @@
-// 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.
-//  Copyright (c) 2013, 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 <algorithm>
-#include <set>
-
-#include "rocksdb/db.h"
-#include "rocksdb/filter_policy.h"
-#include "db/db_impl.h"
-#include "db/filename.h"
-#include "db/version_set.h"
-#include "db/write_batch_internal.h"
-#include "rocksdb/statistics.h"
-#include "rocksdb/cache.h"
-#include "rocksdb/compaction_filter.h"
-#include "rocksdb/env.h"
-#include "rocksdb/table.h"
-#include "rocksdb/table_properties.h"
-#include "table/table_builder.h"
-#include "util/hash.h"
-#include "util/logging.h"
-#include "util/mutexlock.h"
-#include "util/testharness.h"
-#include "util/testutil.h"
-#include "utilities/merge_operators.h"
-
-using std::unique_ptr;
-
-// IS THIS FILE STILL NEEDED?
-namespace rocksdb {
-
-// SimpleTable is a simple table format for UNIT TEST ONLY. It is not built
-// as production quality.
-// SimpleTable requires the input key size to be fixed 16 bytes, value cannot
-// be longer than 150000 bytes and stored data on disk in this format:
-// +--------------------------------------------+  <= key1 offset
-// | key1            | value_size (4 bytes) |   |
-// +----------------------------------------+   |
-// | value1                                     |
-// |                                            |
-// +----------------------------------------+---+  <= key2 offset
-// | key2            | value_size (4 bytes) |   |
-// +----------------------------------------+   |
-// | value2                                     |
-// |                                            |
-// |        ......                              |
-// +-----------------+--------------------------+   <= index_block_offset
-// | key1            | key1 offset (8 bytes)    |
-// +-----------------+--------------------------+
-// | key2            | key2 offset (8 bytes)    |
-// +-----------------+--------------------------+
-// | key3            | key3 offset (8 bytes)    |
-// +-----------------+--------------------------+
-// |        ......                              |
-// +-----------------+------------+-------------+
-// | index_block_offset (8 bytes) |
-// +------------------------------+
-
-// SimpleTable is a simple table format for UNIT TEST ONLY. It is not built
-// as production quality.
-class SimpleTableReader: public TableReader {
-public:
-  // Attempt to open the table that is stored in bytes [0..file_size)
-  // of "file", and read the metadata entries necessary to allow
-  // retrieving data from the table.
-  //
-  // If successful, returns ok and sets "*table" to the newly opened
-  // table.  The client should delete "*table" when no longer needed.
-  // If there was an error while initializing the table, sets "*table"
-  // to nullptr and returns a non-ok status.  Does not take ownership of
-  // "*source", but the client must ensure that "source" remains live
-  // for the duration of the returned table's lifetime.
-  //
-  // *file must remain live while this Table is in use.
-  static Status Open(const Options& options, const EnvOptions& soptions,
-                     unique_ptr<RandomAccessFile> && file, uint64_t file_size,
-                     unique_ptr<TableReader>* table_reader);
-
-  Iterator* NewIterator(const ReadOptions&, Arena* arena) override;
-
-  Status Get(const ReadOptions&, const Slice& key, void* arg,
-             bool (*handle_result)(void* arg, const ParsedInternalKey& k,
-                                   const Slice& v),
-             void (*mark_key_may_exist)(void*) = nullptr) override;
-
-  uint64_t ApproximateOffsetOf(const Slice& key) override;
-
-  void SetupForCompaction() override;
-
-  std::shared_ptr<const TableProperties> GetTableProperties() const override;
-
-  ~SimpleTableReader();
-
-private:
-  struct Rep;
-  Rep* rep_;
-
-  explicit SimpleTableReader(Rep* rep) {
-    rep_ = rep;
-  }
-  friend class TableCache;
-  friend class SimpleTableIterator;
-
-  Status GetOffset(const Slice& target, uint64_t* offset);
-
-  // No copying allowed
-  explicit SimpleTableReader(const TableReader&) = delete;
-  void operator=(const TableReader&) = delete;
-};
-
-// Iterator to iterate SimpleTable
-class SimpleTableIterator: public Iterator {
-public:
-  explicit SimpleTableIterator(SimpleTableReader* table);
-  ~SimpleTableIterator();
-
-  bool Valid() const;
-
-  void SeekToFirst();
-
-  void SeekToLast();
-
-  void Seek(const Slice& target);
-
-  void Next();
-
-  void Prev();
-
-  Slice key() const;
-
-  Slice value() const;
-
-  Status status() const;
-
-private:
-  SimpleTableReader* table_;
-  uint64_t offset_;
-  uint64_t next_offset_;
-  Slice key_;
-  Slice value_;
-  char tmp_str_[4];
-  char* key_str_;
-  char* value_str_;
-  int value_str_len_;
-  Status status_;
-  // No copying allowed
-  SimpleTableIterator(const SimpleTableIterator&) = delete;
-  void operator=(const Iterator&) = delete;
-};
-
-struct SimpleTableReader::Rep {
-  ~Rep() {
-  }
-  Rep(const EnvOptions& storage_options, uint64_t index_start_offset,
-      int num_entries) :
-      soptions(storage_options), index_start_offset(index_start_offset),
-      num_entries(num_entries) {
-  }
-
-  Options options;
-  const EnvOptions& soptions;
-  Status status;
-  unique_ptr<RandomAccessFile> file;
-  uint64_t index_start_offset;
-  int num_entries;
-  std::shared_ptr<TableProperties> table_properties;
-
-  const static int user_key_size = 16;
-  const static int offset_length = 8;
-  const static int key_footer_len = 8;
-
-  static int GetInternalKeyLength() {
-    return user_key_size + key_footer_len;
-  }
-};
-
-SimpleTableReader::~SimpleTableReader() {
-  delete rep_;
-}
-
-Status SimpleTableReader::Open(const Options& options,
-                               const EnvOptions& soptions,
-                               unique_ptr<RandomAccessFile> && file,
-                               uint64_t size,
-                               unique_ptr<TableReader>* table_reader) {
-  char footer_space[Rep::offset_length];
-  Slice footer_input;
-  Status s = file->Read(size - Rep::offset_length, Rep::offset_length,
-                        &footer_input, footer_space);
-  if (s.ok()) {
-    uint64_t index_start_offset = DecodeFixed64(footer_space);
-
-    int num_entries = (size - Rep::offset_length - index_start_offset)
-        / (Rep::GetInternalKeyLength() + Rep::offset_length);
-    SimpleTableReader::Rep* rep = new SimpleTableReader::Rep(soptions,
-                                                             index_start_offset,
-                                                             num_entries);
-
-    rep->file = std::move(file);
-    rep->options = options;
-    table_reader->reset(new SimpleTableReader(rep));
-  }
-  return s;
-}
-
-void SimpleTableReader::SetupForCompaction() {
-}
-
-std::shared_ptr<const TableProperties> SimpleTableReader::GetTableProperties()
-    const {
-  return rep_->table_properties;
-}
-
-Iterator* SimpleTableReader::NewIterator(const ReadOptions& options,
-                                         Arena* arena) {
-  if (arena == nullptr) {
-    return new SimpleTableIterator(this);
-  } else {
-    auto mem = arena->AllocateAligned(sizeof(SimpleTableIterator));
-    return new (mem) SimpleTableIterator(this);
-  }
-}
-
-Status SimpleTableReader::GetOffset(const Slice& target, uint64_t* offset) {
-  uint32_t left = 0;
-  uint32_t right = rep_->num_entries - 1;
-  char key_chars[Rep::GetInternalKeyLength()];
-  Slice tmp_slice;
-
-  uint32_t target_offset = 0;
-  while (left <= right) {
-    uint32_t mid = (left + right + 1) / 2;
-
-    uint64_t offset_to_read = rep_->index_start_offset
-        + (Rep::GetInternalKeyLength() + Rep::offset_length) * mid;
-    Status s = rep_->file->Read(offset_to_read, Rep::GetInternalKeyLength(),
-                                &tmp_slice, key_chars);
-    if (!s.ok()) {
-      return s;
-    }
-
-    InternalKeyComparator ikc(rep_->options.comparator);
-    int compare_result = ikc.Compare(tmp_slice, target);
-
-    if (compare_result < 0) {
-      if (left == right) {
-        target_offset = right + 1;
-        break;
-      }
-      left = mid;
-    } else {
-      if (left == right) {
-        target_offset = left;
-        break;
-      }
-      right = mid - 1;
-    }
-  }
-
-  if (target_offset >= (uint32_t) rep_->num_entries) {
-    *offset = rep_->index_start_offset;
-    return Status::OK();
-  }
-
-  char value_offset_chars[Rep::offset_length];
-
-  int64_t offset_for_value_offset = rep_->index_start_offset
-      + (Rep::GetInternalKeyLength() + Rep::offset_length) * target_offset
-      + Rep::GetInternalKeyLength();
-  Status s = rep_->file->Read(offset_for_value_offset, Rep::offset_length,
-                              &tmp_slice, value_offset_chars);
-  if (s.ok()) {
-    *offset = DecodeFixed64(value_offset_chars);
-  }
-  return s;
-}
-
-Status SimpleTableReader::Get(const ReadOptions& options, const Slice& k,
-                              void* arg,
-                              bool (*saver)(void*, const ParsedInternalKey&,
-                                            const Slice&),
-                              void (*mark_key_may_exist)(void*)) {
-  Status s;
-  SimpleTableIterator* iter = new SimpleTableIterator(this);
-  for (iter->Seek(k); iter->Valid(); iter->Next()) {
-    ParsedInternalKey parsed_key;
-    if (!ParseInternalKey(iter->key(), &parsed_key)) {
-      return Status::Corruption(Slice());
-    }
-
-    if (!(*saver)(arg, parsed_key, iter->value())) {
-      break;
-    }
-  }
-  s = iter->status();
-  delete iter;
-  return s;
-}
-
-uint64_t SimpleTableReader::ApproximateOffsetOf(const Slice& key) {
-  return 0;
-}
-
-SimpleTableIterator::SimpleTableIterator(SimpleTableReader* table) :
-    table_(table) {
-  key_str_ = new char[SimpleTableReader::Rep::GetInternalKeyLength()];
-  value_str_len_ = -1;
-  SeekToFirst();
-}
-
-SimpleTableIterator::~SimpleTableIterator() {
- delete[] key_str_;
- if (value_str_len_ >= 0) {
-   delete[] value_str_;
- }
-}
-
-bool SimpleTableIterator::Valid() const {
-  return offset_ < table_->rep_->index_start_offset;
-}
-
-void SimpleTableIterator::SeekToFirst() {
-  next_offset_ = 0;
-  Next();
-}
-
-void SimpleTableIterator::SeekToLast() {
-  assert(false);
-}
-
-void SimpleTableIterator::Seek(const Slice& target) {
-  Status s = table_->GetOffset(target, &next_offset_);
-  if (!s.ok()) {
-    status_ = s;
-  }
-  Next();
-}
-
-void SimpleTableIterator::Next() {
-  offset_ = next_offset_;
-  if (offset_ >= table_->rep_->index_start_offset) {
-    return;
-  }
-  Slice result;
-  int internal_key_size = SimpleTableReader::Rep::GetInternalKeyLength();
-
-  Status s = table_->rep_->file->Read(next_offset_, internal_key_size, &result,
-                                      key_str_);
-  next_offset_ += internal_key_size;
-  key_ = result;
-
-  Slice value_size_slice;
-  s = table_->rep_->file->Read(next_offset_, 4, &value_size_slice, tmp_str_);
-  next_offset_ += 4;
-  uint32_t value_size = DecodeFixed32(tmp_str_);
-
-  Slice value_slice;
-  if ((int) value_size > value_str_len_) {
-    if (value_str_len_ >= 0) {
-      delete[] value_str_;
-    }
-    value_str_ = new char[value_size];
-    value_str_len_ = value_size;
-  }
-  s = table_->rep_->file->Read(next_offset_, value_size, &value_slice,
-                               value_str_);
-  next_offset_ += value_size;
-  value_ = value_slice;
-}
-
-void SimpleTableIterator::Prev() {
-  assert(false);
-}
-
-Slice SimpleTableIterator::key() const {
-  Log(table_->rep_->options.info_log, "key!!!!");
-  return key_;
-}
-
-Slice SimpleTableIterator::value() const {
-  return value_;
-}
-
-Status SimpleTableIterator::status() const {
-  return status_;
-}
-
-class SimpleTableBuilder: public TableBuilder {
-public:
-  // Create a builder that will store the contents of the table it is
-  // building in *file.  Does not close the file.  It is up to the
-  // caller to close the file after calling Finish(). The output file
-  // will be part of level specified by 'level'.  A value of -1 means
-  // that the caller does not know which level the output file will reside.
-  SimpleTableBuilder(const Options& options, WritableFile* file,
-                     CompressionType compression_type);
-
-  // REQUIRES: Either Finish() or Abandon() has been called.
-  ~SimpleTableBuilder();
-
-  // Add key,value to the table being constructed.
-  // REQUIRES: key is after any previously added key according to comparator.
-  // REQUIRES: Finish(), Abandon() have not been called
-  void Add(const Slice& key, const Slice& value) override;
-
-  // Return non-ok iff some error has been detected.
-  Status status() const override;
-
-  // Finish building the table.  Stops using the file passed to the
-  // constructor after this function returns.
-  // REQUIRES: Finish(), Abandon() have not been called
-  Status Finish() override;
-
-  // Indicate that the contents of this builder should be abandoned.  Stops
-  // using the file passed to the constructor after this function returns.
-  // If the caller is not going to call Finish(), it must call Abandon()
-  // before destroying this builder.
-  // REQUIRES: Finish(), Abandon() have not been called
-  void Abandon() override;
-
-  // Number of calls to Add() so far.
-  uint64_t NumEntries() const override;
-
-  // Size of the file generated so far.  If invoked after a successful
-  // Finish() call, returns the size of the final generated file.
-  uint64_t FileSize() const override;
-
-private:
-  struct Rep;
-  Rep* rep_;
-
-  // No copying allowed
-  SimpleTableBuilder(const SimpleTableBuilder&) = delete;
-  void operator=(const SimpleTableBuilder&) = delete;
-};
-
-struct SimpleTableBuilder::Rep {
-  Options options;
-  WritableFile* file;
-  uint64_t offset = 0;
-  Status status;
-
-  uint64_t num_entries = 0;
-
-  bool closed = false;  // Either Finish() or Abandon() has been called.
-
-  const static int user_key_size = 16;
-  const static int offset_length = 8;
-  const static int key_footer_len = 8;
-
-  static int GetInternalKeyLength() {
-    return user_key_size + key_footer_len;
-  }
-
-  std::string index;
-
-  Rep(const Options& opt, WritableFile* f) :
-      options(opt), file(f) {
-  }
-  ~Rep() {
-  }
-};
-
-SimpleTableBuilder::SimpleTableBuilder(const Options& options,
-                                       WritableFile* file,
-                                       CompressionType compression_type) :
-    rep_(new SimpleTableBuilder::Rep(options, file)) {
-}
-
-SimpleTableBuilder::~SimpleTableBuilder() {
-  delete (rep_);
-}
-
-void SimpleTableBuilder::Add(const Slice& key, const Slice& value) {
-  assert((int ) key.size() == Rep::GetInternalKeyLength());
-
-  // Update index
-  rep_->index.append(key.data(), key.size());
-  PutFixed64(&(rep_->index), rep_->offset);
-
-  // Write key-value pair
-  rep_->file->Append(key);
-  rep_->offset += Rep::GetInternalKeyLength();
-
-  std::string size;
-  int value_size = value.size();
-  PutFixed32(&size, value_size);
-  Slice sizeSlice(size);
-  rep_->file->Append(sizeSlice);
-  rep_->file->Append(value);
-  rep_->offset += value_size + 4;
-
-  rep_->num_entries++;
-}
-
-Status SimpleTableBuilder::status() const {
-  return Status::OK();
-}
-
-Status SimpleTableBuilder::Finish() {
-  Rep* r = rep_;
-  assert(!r->closed);
-  r->closed = true;
-
-  uint64_t index_offset = rep_->offset;
-  Slice index_slice(rep_->index);
-  rep_->file->Append(index_slice);
-  rep_->offset += index_slice.size();
-
-  std::string index_offset_str;
-  PutFixed64(&index_offset_str, index_offset);
-  Slice foot_slice(index_offset_str);
-  rep_->file->Append(foot_slice);
-  rep_->offset += foot_slice.size();
-
-  return Status::OK();
-}
-
-void SimpleTableBuilder::Abandon() {
-  rep_->closed = true;
-}
-
-uint64_t SimpleTableBuilder::NumEntries() const {
-  return rep_->num_entries;
-}
-
-uint64_t SimpleTableBuilder::FileSize() const {
-  return rep_->offset;
-}
-
-class SimpleTableFactory: public TableFactory {
-public:
-  ~SimpleTableFactory() {
-  }
-  SimpleTableFactory() {
-  }
-  const char* Name() const override {
-    return "SimpleTable";
-  }
-  Status NewTableReader(const Options& options, const EnvOptions& soptions,
-                        const InternalKeyComparator& internal_key,
-                        unique_ptr<RandomAccessFile>&& file, uint64_t file_size,
-                        unique_ptr<TableReader>* table_reader) const;
-
-  TableBuilder* NewTableBuilder(const Options& options,
-                                const InternalKeyComparator& internal_key,
-                                WritableFile* file,
-                                CompressionType compression_type) const;
-};
-
-Status SimpleTableFactory::NewTableReader(
-    const Options& options, const EnvOptions& soptions,
-    const InternalKeyComparator& internal_key,
-    unique_ptr<RandomAccessFile>&& file, uint64_t file_size,
-    unique_ptr<TableReader>* table_reader) const {
-
-  return SimpleTableReader::Open(options, soptions, std::move(file), file_size,
-                                 table_reader);
-}
-
-TableBuilder* SimpleTableFactory::NewTableBuilder(
-    const Options& options, const InternalKeyComparator& internal_key,
-    WritableFile* file, CompressionType compression_type) const {
-  return new SimpleTableBuilder(options, file, compression_type);
-}
-
-class SimpleTableDBTest {
-protected:
-public:
-  std::string dbname_;
-  Env* env_;
-  DB* db_;
-
-  Options last_options_;
-
-  SimpleTableDBTest() :
-      env_(Env::Default()) {
-    dbname_ = test::TmpDir() + "/simple_table_db_test";
-    ASSERT_OK(DestroyDB(dbname_, Options()));
-    db_ = nullptr;
-    Reopen();
-  }
-
-  ~SimpleTableDBTest() {
-    delete db_;
-    ASSERT_OK(DestroyDB(dbname_, Options()));
-  }
-
-  // Return the current option configuration.
-  Options CurrentOptions() {
-    Options options;
-    options.table_factory.reset(new SimpleTableFactory());
-    return options;
-  }
-
-  DBImpl* dbfull() {
-    return reinterpret_cast<DBImpl*>(db_);
-  }
-
-  void Reopen(Options* options = nullptr) {
-    ASSERT_OK(TryReopen(options));
-  }
-
-  void Close() {
-    delete db_;
-    db_ = nullptr;
-  }
-
-  void DestroyAndReopen(Options* options = nullptr) {
-    //Destroy using last options
-    Destroy(&last_options_);
-    ASSERT_OK(TryReopen(options));
-  }
-
-  void Destroy(Options* options) {
-    delete db_;
-    db_ = nullptr;
-    ASSERT_OK(DestroyDB(dbname_, *options));
-  }
-
-  Status PureReopen(Options* options, DB** db) {
-    return DB::Open(*options, dbname_, db);
-  }
-
-  Status TryReopen(Options* options = nullptr) {
-    delete db_;
-    db_ = nullptr;
-    Options opts;
-    if (options != nullptr) {
-      opts = *options;
-    } else {
-      opts = CurrentOptions();
-      opts.create_if_missing = true;
-    }
-    last_options_ = opts;
-
-    return DB::Open(opts, dbname_, &db_);
-  }
-
-  Status Put(const Slice& k, const Slice& v) {
-    return db_->Put(WriteOptions(), k, v);
-  }
-
-  Status Delete(const std::string& k) {
-    return db_->Delete(WriteOptions(), k);
-  }
-
-  std::string Get(const std::string& k, const Snapshot* snapshot = nullptr) {
-    ReadOptions options;
-    options.snapshot = snapshot;
-    std::string result;
-    Status s = db_->Get(options, k, &result);
-    if (s.IsNotFound()) {
-      result = "NOT_FOUND";
-    } else if (!s.ok()) {
-      result = s.ToString();
-    }
-    return result;
-  }
-
-
-  int NumTableFilesAtLevel(int level) {
-    std::string property;
-    ASSERT_TRUE(
-        db_->GetProperty("rocksdb.num-files-at-level" + NumberToString(level),
-                         &property));
-    return atoi(property.c_str());
-  }
-
-  // Return spread of files per level
-  std::string FilesPerLevel() {
-    std::string result;
-    int last_non_zero_offset = 0;
-    for (int level = 0; level < db_->NumberLevels(); level++) {
-      int f = NumTableFilesAtLevel(level);
-      char buf[100];
-      snprintf(buf, sizeof(buf), "%s%d", (level ? "," : ""), f);
-      result += buf;
-      if (f > 0) {
-        last_non_zero_offset = result.size();
-      }
-    }
-    result.resize(last_non_zero_offset);
-    return result;
-  }
-
-  std::string IterStatus(Iterator* iter) {
-    std::string result;
-    if (iter->Valid()) {
-      result = iter->key().ToString() + "->" + iter->value().ToString();
-    } else {
-      result = "(invalid)";
-    }
-    return result;
-  }
-};
-
-TEST(SimpleTableDBTest, Empty) {
-  ASSERT_TRUE(db_ != nullptr);
-  ASSERT_EQ("NOT_FOUND", Get("0000000000000foo"));
-}
-
-TEST(SimpleTableDBTest, ReadWrite) {
-  ASSERT_OK(Put("0000000000000foo", "v1"));
-  ASSERT_EQ("v1", Get("0000000000000foo"));
-  ASSERT_OK(Put("0000000000000bar", "v2"));
-  ASSERT_OK(Put("0000000000000foo", "v3"));
-  ASSERT_EQ("v3", Get("0000000000000foo"));
-  ASSERT_EQ("v2", Get("0000000000000bar"));
-}
-
-TEST(SimpleTableDBTest, Flush) {
-  ASSERT_OK(Put("0000000000000foo", "v1"));
-  ASSERT_OK(Put("0000000000000bar", "v2"));
-  ASSERT_OK(Put("0000000000000foo", "v3"));
-  dbfull()->TEST_FlushMemTable();
-  ASSERT_EQ("v3", Get("0000000000000foo"));
-  ASSERT_EQ("v2", Get("0000000000000bar"));
-}
-
-TEST(SimpleTableDBTest, Flush2) {
-  ASSERT_OK(Put("0000000000000bar", "b"));
-  ASSERT_OK(Put("0000000000000foo", "v1"));
-  dbfull()->TEST_FlushMemTable();
-
-  ASSERT_OK(Put("0000000000000foo", "v2"));
-  dbfull()->TEST_FlushMemTable();
-  ASSERT_EQ("v2", Get("0000000000000foo"));
-
-  ASSERT_OK(Put("0000000000000eee", "v3"));
-  dbfull()->TEST_FlushMemTable();
-  ASSERT_EQ("v3", Get("0000000000000eee"));
-
-  ASSERT_OK(Delete("0000000000000bar"));
-  dbfull()->TEST_FlushMemTable();
-  ASSERT_EQ("NOT_FOUND", Get("0000000000000bar"));
-
-  ASSERT_OK(Put("0000000000000eee", "v5"));
-  dbfull()->TEST_FlushMemTable();
-  ASSERT_EQ("v5", Get("0000000000000eee"));
-}
-
-static std::string Key(int i) {
-  char buf[100];
-  snprintf(buf, sizeof(buf), "key_______%06d", i);
-  return std::string(buf);
-}
-
-static std::string RandomString(Random* rnd, int len) {
-  std::string r;
-  test::RandomString(rnd, len, &r);
-  return r;
-}
-
-TEST(SimpleTableDBTest, CompactionTrigger) {
-  Options options = CurrentOptions();
-  options.write_buffer_size = 100 << 10; //100KB
-  options.num_levels = 3;
-  options.max_mem_compaction_level = 0;
-  options.level0_file_num_compaction_trigger = 3;
-  Reopen(&options);
-
-  Random rnd(301);
-
-  for (int num = 0; num < options.level0_file_num_compaction_trigger - 1;
-      num++) {
-    std::vector<std::string> values;
-    // Write 120KB (12 values, each 10K)
-    for (int i = 0; i < 12; i++) {
-      values.push_back(RandomString(&rnd, 10000));
-      ASSERT_OK(Put(Key(i), values[i]));
-    }
-    dbfull()->TEST_WaitForFlushMemTable();
-    ASSERT_EQ(NumTableFilesAtLevel(0), num + 1);
-  }
-
-  //generate one more file in level-0, and should trigger level-0 compaction
-  std::vector<std::string> values;
-  for (int i = 0; i < 12; i++) {
-    values.push_back(RandomString(&rnd, 10000));
-    ASSERT_OK(Put(Key(i), values[i]));
-  }
-  dbfull()->TEST_WaitForCompact();
-
-  ASSERT_EQ(NumTableFilesAtLevel(0), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(1), 1);
-}
-
-}  // namespace rocksdb
-
-int main(int argc, char** argv) {
-  return rocksdb::test::RunAllTests();
-}

db/snapshot.h

@@ -71,7 +71,7 @@ class SnapshotList {
   }
 
   // get the sequence number of the most recent snapshot
-  const SequenceNumber GetNewest() {
+  SequenceNumber GetNewest() {
     if (empty()) {
       return 0;
     }

db/table_cache.cc

@@ -15,6 +15,7 @@
 #include "rocksdb/statistics.h"
 #include "table/iterator_wrapper.h"
 #include "table/table_reader.h"
+#include "table/get_context.h"
 #include "util/coding.h"
 #include "util/stop_watch.h"
 
@@ -36,12 +37,10 @@ static Slice GetSliceForFileNumber(const uint64_t* file_number) {
                sizeof(*file_number));
 }
 
-TableCache::TableCache(const Options* options,
-                       const EnvOptions& storage_options, Cache* const cache)
-    : env_(options->env),
-      db_paths_(options->db_paths),
-      options_(options),
-      storage_options_(storage_options),
+TableCache::TableCache(const ImmutableCFOptions& ioptions,
+                       const EnvOptions& env_options, Cache* const cache)
+    : ioptions_(ioptions),
+      env_options_(env_options),
       cache_(cache) {}
 
 TableCache::~TableCache() {
@@ -55,7 +54,7 @@ void TableCache::ReleaseHandle(Cache::Handle* handle) {
   cache_->Release(handle);
 }
 
-Status TableCache::FindTable(const EnvOptions& toptions,
+Status TableCache::FindTable(const EnvOptions& env_options,
                              const InternalKeyComparator& internal_comparator,
                              const FileDescriptor& fd, Cache::Handle** handle,
                              const bool no_io) {
@@ -68,24 +67,24 @@ Status TableCache::FindTable(const EnvOptions& toptions,
       return Status::Incomplete("Table not found in table_cache, no_io is set");
     }
     std::string fname =
-        TableFileName(db_paths_, fd.GetNumber(), fd.GetPathId());
+        TableFileName(ioptions_.db_paths, fd.GetNumber(), fd.GetPathId());
     unique_ptr<RandomAccessFile> file;
     unique_ptr<TableReader> table_reader;
-    s = env_->NewRandomAccessFile(fname, &file, toptions);
-    RecordTick(options_->statistics.get(), NO_FILE_OPENS);
+    s = ioptions_.env->NewRandomAccessFile(fname, &file, env_options);
+    RecordTick(ioptions_.statistics, NO_FILE_OPENS);
     if (s.ok()) {
-      if (options_->advise_random_on_open) {
+      if (ioptions_.advise_random_on_open) {
         file->Hint(RandomAccessFile::RANDOM);
       }
-      StopWatch sw(env_, options_->statistics.get(), TABLE_OPEN_IO_MICROS);
-      s = options_->table_factory->NewTableReader(
-          *options_, toptions, internal_comparator, std::move(file),
+      StopWatch sw(ioptions_.env, ioptions_.statistics, TABLE_OPEN_IO_MICROS);
+      s = ioptions_.table_factory->NewTableReader(
+          ioptions_, env_options, internal_comparator, std::move(file),
           fd.GetFileSize(), &table_reader);
     }
 
     if (!s.ok()) {
       assert(table_reader == nullptr);
-      RecordTick(options_->statistics.get(), NO_FILE_ERRORS);
+      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 {
@@ -97,7 +96,7 @@ Status TableCache::FindTable(const EnvOptions& toptions,
 }
 
 Iterator* TableCache::NewIterator(const ReadOptions& options,
-                                  const EnvOptions& toptions,
+                                  const EnvOptions& env_options,
                                   const InternalKeyComparator& icomparator,
                                   const FileDescriptor& fd,
                                   TableReader** table_reader_ptr,
@@ -109,7 +108,7 @@ Iterator* TableCache::NewIterator(const ReadOptions& options,
   Cache::Handle* handle = nullptr;
   Status s;
   if (table_reader == nullptr) {
-    s = FindTable(toptions, icomparator, fd, &handle,
+    s = FindTable(env_options, icomparator, fd, &handle,
                   options.read_tier == kBlockCacheTier);
     if (!s.ok()) {
       return NewErrorIterator(s, arena);
@@ -134,34 +133,33 @@ Iterator* TableCache::NewIterator(const ReadOptions& options,
 
 Status TableCache::Get(const ReadOptions& options,
                        const InternalKeyComparator& internal_comparator,
-                       const FileDescriptor& fd, const Slice& k, void* arg,
-                       bool (*saver)(void*, const ParsedInternalKey&,
-                                     const Slice&),
-                       void (*mark_key_may_exist)(void*)) {
+                       const FileDescriptor& fd, const Slice& k,
+                       GetContext* get_context) {
   TableReader* t = fd.table_reader;
   Status s;
   Cache::Handle* handle = nullptr;
   if (!t) {
-    s = FindTable(storage_options_, internal_comparator, fd, &handle,
+    s = FindTable(env_options_, internal_comparator, fd, &handle,
                   options.read_tier == kBlockCacheTier);
     if (s.ok()) {
       t = GetTableReaderFromHandle(handle);
     }
   }
   if (s.ok()) {
-    s = t->Get(options, k, arg, saver, mark_key_may_exist);
+    s = t->Get(options, k, get_context);
     if (handle != nullptr) {
       ReleaseHandle(handle);
     }
   } else if (options.read_tier && s.IsIncomplete()) {
     // Couldnt find Table in cache but treat as kFound if no_io set
-    (*mark_key_may_exist)(arg);
+    get_context->MarkKeyMayExist();
     return Status::OK();
   }
   return s;
 }
+
 Status TableCache::GetTableProperties(
-    const EnvOptions& toptions,
+    const EnvOptions& env_options,
     const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
     std::shared_ptr<const TableProperties>* properties, bool no_io) {
   Status s;
@@ -174,7 +172,7 @@ Status TableCache::GetTableProperties(
   }
 
   Cache::Handle* table_handle = nullptr;
-  s = FindTable(toptions, internal_comparator, fd, &table_handle, no_io);
+  s = FindTable(env_options, internal_comparator, fd, &table_handle, no_io);
   if (!s.ok()) {
     return s;
   }
@@ -185,6 +183,29 @@ Status TableCache::GetTableProperties(
   return s;
 }
 
+size_t TableCache::GetMemoryUsageByTableReader(
+    const EnvOptions& env_options,
+    const InternalKeyComparator& internal_comparator,
+    const FileDescriptor& fd) {
+  Status s;
+  auto table_reader = fd.table_reader;
+  // table already been pre-loaded?
+  if (table_reader) {
+    return table_reader->ApproximateMemoryUsage();
+  }
+
+  Cache::Handle* table_handle = nullptr;
+  s = FindTable(env_options, internal_comparator, fd, &table_handle, true);
+  if (!s.ok()) {
+    return 0;
+  }
+  assert(table_handle);
+  auto table = GetTableReaderFromHandle(table_handle);
+  auto ret = table->ApproximateMemoryUsage();
+  ReleaseHandle(table_handle);
+  return ret;
+}
+
 void TableCache::Evict(Cache* cache, uint64_t file_number) {
   cache->Erase(GetSliceForFileNumber(&file_number));
 }

db/table_cache.h

@@ -19,6 +19,7 @@
 #include "rocksdb/cache.h"
 #include "rocksdb/env.h"
 #include "rocksdb/table.h"
+#include "rocksdb/options.h"
 #include "table/table_reader.h"
 
 namespace rocksdb {
@@ -26,11 +27,12 @@ namespace rocksdb {
 class Env;
 class Arena;
 struct FileDescriptor;
+class GetContext;
 
 class TableCache {
  public:
-  TableCache(const Options* options, const EnvOptions& storage_options,
-             Cache* cache);
+  TableCache(const ImmutableCFOptions& ioptions,
+             const EnvOptions& storage_options, Cache* cache);
   ~TableCache();
 
   // Return an iterator for the specified file number (the corresponding
@@ -51,10 +53,8 @@ class TableCache {
   // it returns false.
   Status Get(const ReadOptions& options,
              const InternalKeyComparator& internal_comparator,
-             const FileDescriptor& file_fd, const Slice& k, void* arg,
-             bool (*handle_result)(void*, const ParsedInternalKey&,
-                                   const Slice&),
-             void (*mark_key_may_exist)(void*) = nullptr);
+             const FileDescriptor& file_fd, const Slice& k,
+             GetContext* get_context);
 
   // Evict any entry for the specified file number
   static void Evict(Cache* cache, uint64_t file_number);
@@ -80,14 +80,19 @@ class TableCache {
                             std::shared_ptr<const TableProperties>* properties,
                             bool no_io = false);
 
+  // Return total memory usage of the table reader of the file.
+  // 0 of table reader of the file is not loaded.
+  size_t GetMemoryUsageByTableReader(
+      const EnvOptions& toptions,
+      const InternalKeyComparator& internal_comparator,
+      const FileDescriptor& fd);
+
   // Release the handle from a cache
   void ReleaseHandle(Cache::Handle* handle);
 
  private:
-  Env* const env_;
-  const std::vector<std::string> db_paths_;
-  const Options* options_;
-  const EnvOptions& storage_options_;
+  const ImmutableCFOptions& ioptions_;
+  const EnvOptions& env_options_;
   Cache* const cache_;
 };
 

db/table_properties_collector_test.cc

@@ -11,6 +11,7 @@
 #include "db/dbformat.h"
 #include "db/table_properties_collector.h"
 #include "rocksdb/table.h"
+#include "rocksdb/immutable_options.h"
 #include "table/block_based_table_factory.h"
 #include "table/meta_blocks.h"
 #include "table/plain_table_factory.h"
@@ -85,12 +86,14 @@ class DumbLogger : public Logger {
 // Utilities test functions
 namespace {
 void MakeBuilder(const Options& options,
+                 const ImmutableCFOptions& ioptions,
                  const InternalKeyComparator& internal_comparator,
                  std::unique_ptr<FakeWritableFile>* writable,
                  std::unique_ptr<TableBuilder>* builder) {
   writable->reset(new FakeWritableFile);
-  builder->reset(options.table_factory->NewTableBuilder(
-      options, internal_comparator, writable->get(), options.compression));
+  builder->reset(ioptions.table_factory->NewTableBuilder(
+      ioptions, internal_comparator, writable->get(),
+      options.compression, options.compression_opts));
 }
 }  // namespace
 
@@ -153,7 +156,8 @@ void TestCustomizedTablePropertiesCollector(
   // -- Step 1: build table
   std::unique_ptr<TableBuilder> builder;
   std::unique_ptr<FakeWritableFile> writable;
-  MakeBuilder(options, internal_comparator, &writable, &builder);
+  const ImmutableCFOptions ioptions(options);
+  MakeBuilder(options, ioptions, internal_comparator, &writable, &builder);
 
   for (const auto& kv : kvs) {
     if (encode_as_internal) {
@@ -213,7 +217,14 @@ TEST(TablePropertiesTest, CustomizedTablePropertiesCollector) {
   Options options;
   options.table_properties_collector_factories.emplace_back(
       new RegularKeysStartWithAFactory());
-  options.table_factory = std::make_shared<PlainTableFactory>(8, 8, 0);
+
+  PlainTableOptions plain_table_options;
+  plain_table_options.user_key_len = 8;
+  plain_table_options.bloom_bits_per_key = 8;
+  plain_table_options.hash_table_ratio = 0;
+
+  options.table_factory =
+      std::make_shared<PlainTableFactory>(plain_table_options);
   test::PlainInternalKeyComparator ikc(options.comparator);
   TestCustomizedTablePropertiesCollector(kPlainTableMagicNumber, true, options,
                                          ikc);
@@ -250,16 +261,17 @@ void TestInternalKeyPropertiesCollector(
     // SanitizeOptions().
     options.info_log = std::make_shared<DumbLogger>();
     options = SanitizeOptions("db",            // just a place holder
-                              &pikc, nullptr,  // don't care filter policy
+                              &pikc,
                               options);
     options.comparator = comparator;
   } else {
     options.table_properties_collector_factories = {
         std::make_shared<InternalKeyPropertiesCollectorFactory>()};
   }
+  const ImmutableCFOptions ioptions(options);
 
   for (int iter = 0; iter < 2; ++iter) {
-    MakeBuilder(options, pikc, &writable, &builder);
+    MakeBuilder(options, ioptions, pikc, &writable, &builder);
     for (const auto& k : keys) {
       builder->Add(k.Encode(), "val");
     }
@@ -299,11 +311,15 @@ TEST(TablePropertiesTest, InternalKeyPropertiesCollector) {
       true /* not sanitize */,
       std::make_shared<BlockBasedTableFactory>()
   );
+
+  PlainTableOptions plain_table_options;
+  plain_table_options.user_key_len = 8;
+  plain_table_options.bloom_bits_per_key = 8;
+  plain_table_options.hash_table_ratio = 0;
+
   TestInternalKeyPropertiesCollector(
-      kPlainTableMagicNumber,
-      false /* not sanitize */,
-      std::make_shared<PlainTableFactory>(8, 8, 0)
-  );
+      kPlainTableMagicNumber, false /* not sanitize */,
+      std::make_shared<PlainTableFactory>(plain_table_options));
 }
 
 }  // namespace rocksdb

db/tailing_iter.cc

@@ -1,221 +0,0 @@
-//  Copyright (c) 2013, 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.
-
-#ifndef ROCKSDB_LITE
-#include "db/tailing_iter.h"
-
-#include <string>
-#include <utility>
-#include <vector>
-#include "db/db_impl.h"
-#include "db/db_iter.h"
-#include "db/column_family.h"
-#include "rocksdb/env.h"
-#include "rocksdb/slice.h"
-#include "rocksdb/slice_transform.h"
-#include "table/merger.h"
-
-namespace rocksdb {
-
-TailingIterator::TailingIterator(Env* const env, DBImpl* db,
-    const ReadOptions& read_options, ColumnFamilyData* cfd)
-    : env_(env),
-      db_(db),
-      read_options_(read_options),
-      cfd_(cfd),
-      super_version_(nullptr),
-      current_(nullptr),
-      status_(Status::InvalidArgument("Seek() not called on this iterator")) {}
-
-TailingIterator::~TailingIterator() {
-  Cleanup();
-}
-
-bool TailingIterator::Valid() const {
-  return current_ != nullptr;
-}
-
-void TailingIterator::SeekToFirst() {
-  if (!IsCurrentVersion()) {
-    CreateIterators();
-  }
-
-  mutable_->SeekToFirst();
-  immutable_->SeekToFirst();
-  UpdateCurrent();
-}
-
-void TailingIterator::Seek(const Slice& target) {
-  if (!IsCurrentVersion()) {
-    CreateIterators();
-  }
-
-  mutable_->Seek(target);
-
-  // We maintain the interval (prev_key_, immutable_->key()] such that there
-  // are no records with keys within that range in immutable_ other than
-  // immutable_->key(). Since immutable_ can't change in this version, we don't
-  // need to do a seek if 'target' belongs to that interval (i.e. immutable_ is
-  // already at the correct position)!
-  //
-  // If prefix seek is used and immutable_ is not valid, seek if target has a
-  // different prefix than prev_key.
-  //
-  // prev_key_ is updated by Next(). SeekImmutable() sets prev_key_ to
-  // 'target' -- in this case, prev_key_ is included in the interval, so
-  // prev_inclusive_ has to be set.
-
-  const Comparator* cmp = cfd_->user_comparator();
-  if (!is_prev_set_ || cmp->Compare(prev_key_, target) >= !is_prev_inclusive_ ||
-      (immutable_->Valid() && cmp->Compare(target, immutable_->key()) > 0) ||
-      (cfd_->options()->prefix_extractor != nullptr && !IsSamePrefix(target))) {
-    SeekImmutable(target);
-  }
-
-  UpdateCurrent();
-}
-
-void TailingIterator::Next() {
-  assert(Valid());
-
-  if (!IsCurrentVersion()) {
-    // save the current key, create new iterators and then seek
-    std::string current_key = key().ToString();
-    Slice key_slice(current_key.data(), current_key.size());
-
-    CreateIterators();
-    Seek(key_slice);
-
-    if (!Valid() || key().compare(key_slice) != 0) {
-      // record with current_key no longer exists
-      return;
-    }
-
-  } else if (current_ == immutable_.get()) {
-    // immutable iterator is advanced -- update prev_key_
-    prev_key_ = key().ToString();
-    is_prev_inclusive_ = false;
-    is_prev_set_ = true;
-  }
-
-  current_->Next();
-  UpdateCurrent();
-}
-
-Slice TailingIterator::key() const {
-  assert(Valid());
-  return current_->key();
-}
-
-Slice TailingIterator::value() const {
-  assert(Valid());
-  return current_->value();
-}
-
-Status TailingIterator::status() const {
-  if (!status_.ok()) {
-    return status_;
-  } else if (!mutable_->status().ok()) {
-    return mutable_->status();
-  } else {
-    return immutable_->status();
-  }
-}
-
-void TailingIterator::Prev() {
-  status_ = Status::NotSupported("This iterator doesn't support Prev()");
-}
-
-void TailingIterator::SeekToLast() {
-  status_ = Status::NotSupported("This iterator doesn't support SeekToLast()");
-}
-
-void TailingIterator::Cleanup() {
-  // Release old super version if necessary
-  mutable_.reset();
-  immutable_.reset();
-  if (super_version_ != nullptr && super_version_->Unref()) {
-    DBImpl::DeletionState deletion_state;
-    db_->mutex_.Lock();
-    super_version_->Cleanup();
-    db_->FindObsoleteFiles(deletion_state, false, true);
-    db_->mutex_.Unlock();
-    delete super_version_;
-    if (deletion_state.HaveSomethingToDelete()) {
-      db_->PurgeObsoleteFiles(deletion_state);
-    }
-  }
-}
-
-void TailingIterator::CreateIterators() {
-  Cleanup();
-  super_version_= cfd_->GetReferencedSuperVersion(&(db_->mutex_));
-
-  Iterator* mutable_iter = super_version_->mem->NewIterator(read_options_);
-  // create a DBIter that only uses memtable content; see NewIterator()
-  mutable_.reset(
-      NewDBIterator(env_, *cfd_->options(), cfd_->user_comparator(),
-                    mutable_iter, kMaxSequenceNumber));
-
-  std::vector<Iterator*> list;
-  super_version_->imm->AddIterators(read_options_, &list);
-  super_version_->current->AddIterators(
-      read_options_, *cfd_->soptions(), &list);
-  Iterator* immutable_iter =
-      NewMergingIterator(&cfd_->internal_comparator(), &list[0], list.size());
-
-  // create a DBIter that only uses memtable content; see NewIterator()
-  immutable_.reset(
-      NewDBIterator(env_, *cfd_->options(), cfd_->user_comparator(),
-                    immutable_iter, kMaxSequenceNumber));
-
-  current_ = nullptr;
-  is_prev_set_ = false;
-}
-
-void TailingIterator::UpdateCurrent() {
-  current_ = nullptr;
-
-  if (mutable_->Valid()) {
-    current_ = mutable_.get();
-  }
-  const Comparator* cmp = cfd_->user_comparator();
-  if (immutable_->Valid() &&
-      (current_ == nullptr ||
-       cmp->Compare(immutable_->key(), current_->key()) < 0)) {
-    current_ = immutable_.get();
-  }
-
-  if (!status_.ok()) {
-    // reset status that was set by Prev() or SeekToLast()
-    status_ = Status::OK();
-  }
-}
-
-bool TailingIterator::IsCurrentVersion() const {
-  return super_version_ != nullptr &&
-         super_version_->version_number == cfd_->GetSuperVersionNumber();
-}
-
-bool TailingIterator::IsSamePrefix(const Slice& target) const {
-  const SliceTransform* extractor = cfd_->options()->prefix_extractor.get();
-
-  assert(extractor);
-  assert(is_prev_set_);
-
-  return extractor->Transform(target)
-    .compare(extractor->Transform(prev_key_)) == 0;
-}
-
-void TailingIterator::SeekImmutable(const Slice& target) {
-  prev_key_ = target.ToString();
-  is_prev_inclusive_ = true;
-  is_prev_set_ = true;
-
-  immutable_->Seek(target);
-}
-
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE

db/tailing_iter.h

@@ -1,97 +0,0 @@
-//  Copyright (c) 2013, 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
-
-#ifndef ROCKSDB_LITE
-
-#include <string>
-
-#include "rocksdb/db.h"
-#include "rocksdb/iterator.h"
-#include "rocksdb/options.h"
-
-namespace rocksdb {
-
-class DBImpl;
-class Env;
-struct SuperVersion;
-class ColumnFamilyData;
-
-/**
- * TailingIterator is a special type of iterator that doesn't use an (implicit)
- * snapshot. In other words, it can be used to read data that was added to the
- * db after the iterator had been created.
- *
- * TailingIterator is optimized for sequential reading. It doesn't support
- * Prev() and SeekToLast() operations.
- */
-class TailingIterator : public Iterator {
- public:
-  TailingIterator(Env* const env, DBImpl* db, const ReadOptions& read_options,
-                  ColumnFamilyData* cfd);
-  virtual ~TailingIterator();
-
-  virtual bool Valid() const override;
-  virtual void SeekToFirst() override;
-  virtual void SeekToLast() override;
-  virtual void Seek(const Slice& target) override;
-  virtual void Next() override;
-  virtual void Prev() override;
-  virtual Slice key() const override;
-  virtual Slice value() const override;
-  virtual Status status() const override;
-
- private:
-  void Cleanup();
-
-  Env* const env_;
-  DBImpl* const db_;
-  const ReadOptions read_options_;
-  ColumnFamilyData* const cfd_;
-  SuperVersion* super_version_;
-
-  // TailingIterator merges the contents of the two iterators below (one using
-  // mutable memtable contents only, other over SSTs and immutable memtables).
-  // See DBIter::GetTailingIteratorPair().
-  std::unique_ptr<Iterator> mutable_;
-  std::unique_ptr<Iterator> immutable_;
-
-  // points to either mutable_ or immutable_
-  Iterator* current_;
-
-  // key that precedes immutable iterator's current key
-  std::string prev_key_;
-
-  // unless prev_set is true, prev_key/prev_head is not valid and shouldn't be
-  // used; reset by createIterators()
-  bool is_prev_set_;
-
-  // prev_key_ was set by SeekImmutable(), which means that the interval of
-  // keys covered by immutable_ is [prev_key_, current], i.e. it includes the
-  // left endpoint
-  bool is_prev_inclusive_;
-
-  // internal iterator status
-  Status status_;
-
-  // check if this iterator's version matches DB's version
-  bool IsCurrentVersion() const;
-
-  // check if SeekImmutable() is needed due to target having a different prefix
-  // than prev_key_ (used when in prefix seek mode)
-  bool IsSamePrefix(const Slice& target) const;
-
-  // creates mutable_ and immutable_ iterators and updates version_number_
-  void CreateIterators();
-
-  // set current_ to be one of the iterators with the smallest key
-  void UpdateCurrent();
-
-  // seek on immutable_ and update prev_key
-  void SeekImmutable(const Slice& target);
-};
-
-}  // namespace rocksdb
-#endif  // ROCKSDB_LITE

db/version_edit.cc

@@ -64,7 +64,7 @@ void VersionEdit::Clear() {
   column_family_name_.clear();
 }
 
-void VersionEdit::EncodeTo(std::string* dst) const {
+bool VersionEdit::EncodeTo(std::string* dst) const {
   if (has_comparator_) {
     PutVarint32(dst, kComparator);
     PutLengthPrefixedSlice(dst, comparator_);
@@ -98,6 +98,9 @@ void VersionEdit::EncodeTo(std::string* dst) const {
 
   for (size_t i = 0; i < new_files_.size(); i++) {
     const FileMetaData& f = new_files_[i].second;
+    if (!f.smallest.Valid() || !f.largest.Valid()) {
+      return false;
+    }
     if (f.fd.GetPathId() == 0) {
       // Use older format to make sure user can roll back the build if they
       // don't config multiple DB paths.
@@ -131,6 +134,7 @@ void VersionEdit::EncodeTo(std::string* dst) const {
   if (is_column_family_drop_) {
     PutVarint32(dst, kColumnFamilyDrop);
   }
+  return true;
 }
 
 static bool GetInternalKey(Slice* input, InternalKey* dst) {

db/version_edit.h

@@ -14,6 +14,8 @@
 #include <string>
 #include "rocksdb/cache.h"
 #include "db/dbformat.h"
+#include "util/arena.h"
+#include "util/autovector.h"
 
 namespace rocksdb {
 
@@ -79,6 +81,8 @@ struct FileMetaData {
   uint64_t num_deletions;          // the number of deletion entries.
   uint64_t raw_key_size;           // total uncompressed key size.
   uint64_t raw_value_size;         // total uncompressed value size.
+  bool init_stats_from_file;   // true if the data-entry stats of this file
+                               // has initialized from file.
 
   FileMetaData()
       : refs(0),
@@ -88,7 +92,8 @@ struct FileMetaData {
         num_entries(0),
         num_deletions(0),
         raw_key_size(0),
-        raw_value_size(0) {}
+        raw_value_size(0),
+        init_stats_from_file(false) {}
 };
 
 // A compressed copy of file meta data that just contain
@@ -158,13 +163,13 @@ class VersionEdit {
   // Add the specified file at the specified number.
   // REQUIRES: This version has not been saved (see VersionSet::SaveTo)
   // REQUIRES: "smallest" and "largest" are smallest and largest keys in file
-  void AddFile(int level, uint64_t file, uint64_t file_size,
-               uint64_t file_path_id, const InternalKey& smallest,
+  void AddFile(int level, uint64_t file, uint64_t file_path_id,
+               uint64_t file_size, const InternalKey& smallest,
                const InternalKey& largest, const SequenceNumber& smallest_seqno,
                const SequenceNumber& largest_seqno) {
     assert(smallest_seqno <= largest_seqno);
     FileMetaData f;
-    f.fd = FileDescriptor(file, file_size, file_path_id);
+    f.fd = FileDescriptor(file, file_path_id, file_size);
     f.smallest = smallest;
     f.largest = largest;
     f.smallest_seqno = smallest_seqno;
@@ -207,7 +212,8 @@ class VersionEdit {
     is_column_family_drop_ = true;
   }
 
-  void EncodeTo(std::string* dst) const;
+  // return true on success.
+  bool EncodeTo(std::string* dst) const;
   Status DecodeFrom(const Slice& src);
 
   std::string DebugString(bool hex_key = false) const;

db/version_edit_test.cc

@@ -44,6 +44,16 @@ TEST(VersionEditTest, EncodeDecode) {
   TestEncodeDecode(edit);
 }
 
+TEST(VersionEditTest, EncodeEmptyFile) {
+  VersionEdit edit;
+  edit.AddFile(0, 0, 0, 0,
+               InternalKey(),
+               InternalKey(),
+               0, 0);
+  std::string buffer;
+  ASSERT_TRUE(!edit.EncodeTo(&buffer));
+}
+
 TEST(VersionEditTest, ColumnFamilyTest) {
   VersionEdit edit;
   edit.SetColumnFamily(2);

db/version_set.h

@@ -34,6 +34,7 @@
 #include "db/column_family.h"
 #include "db/log_reader.h"
 #include "db/file_indexer.h"
+#include "db/write_controller.h"
 
 namespace rocksdb {
 
@@ -53,14 +54,6 @@ class ColumnFamilySet;
 class TableCache;
 class MergeIteratorBuilder;
 
-
-// Return the smallest index i such that files[i]->largest >= key.
-// Return files.size() if there is no such file.
-// REQUIRES: "files" contains a sorted list of non-overlapping files.
-extern int FindFile(const InternalKeyComparator& icmp,
-                    const std::vector<FileMetaData*>& files,
-                    const Slice& key);
-
 // Return the smallest index i such that file_level.files[i]->largest >= key.
 // Return file_level.num_files if there is no such file.
 // REQUIRES: "file_level.files" contains a sorted list of
@@ -82,13 +75,18 @@ extern bool SomeFileOverlapsRange(
     const Slice* smallest_user_key,
     const Slice* largest_user_key);
 
+// Generate FileLevel from vector<FdWithKeyRange*>
+// Would copy smallest_key and largest_key data to sequential memory
+// arena: Arena used to allocate the memory
+extern void DoGenerateFileLevel(FileLevel* file_level,
+        const std::vector<FileMetaData*>& files,
+        Arena* arena);
+
 class Version {
  public:
   // Append to *iters a sequence of iterators that will
   // yield the contents of this Version when merged together.
   // REQUIRES: This version has been saved (see VersionSet::SaveTo)
-  void AddIterators(const ReadOptions&, const EnvOptions& soptions,
-                    std::vector<Iterator*>* iters);
 
   void AddIterators(const ReadOptions&, const EnvOptions& soptions,
                     MergeIteratorBuilder* merger_iter_builder);
@@ -105,14 +103,18 @@ class Version {
   // We use compaction scores to figure out which compaction to do next
   // REQUIRES: If Version is not yet saved to current_, it can be called without
   // a lock. Once a version is saved to current_, call only with mutex held
-  void ComputeCompactionScore(std::vector<uint64_t>& size_being_compacted);
+  void ComputeCompactionScore(
+      const MutableCFOptions& mutable_cf_options,
+      std::vector<uint64_t>& size_being_compacted);
 
   // Generate file_levels_ from files_
   void GenerateFileLevels();
 
   // Update scores, pre-calculated variables. It needs to be called before
   // applying the version to the version set.
-  void PrepareApply(std::vector<uint64_t>& size_being_compacted);
+  void PrepareApply(
+      const MutableCFOptions& mutable_cf_options,
+      std::vector<uint64_t>& size_being_compacted);
 
   // Reference count management (so Versions do not disappear out from
   // under live iterators)
@@ -171,7 +173,8 @@ class Version {
 
   // Return the level at which we should place a new memtable compaction
   // result that covers the range [smallest_user_key,largest_user_key].
-  int PickLevelForMemTableOutput(const Slice& smallest_user_key,
+  int PickLevelForMemTableOutput(const MutableCFOptions& mutable_cf_options,
+                                 const Slice& smallest_user_key,
                                  const Slice& largest_user_key);
 
   int NumberLevels() const { return num_levels_; }
@@ -180,15 +183,15 @@ class Version {
   int NumLevelFiles(int level) const { return files_[level].size(); }
 
   // Return the combined file size of all files at the specified level.
-  int64_t NumLevelBytes(int level) const;
+  uint64_t NumLevelBytes(int level) const;
 
   // Return a human-readable short (single-line) summary of the number
   // of files per level.  Uses *scratch as backing store.
   struct LevelSummaryStorage {
-    char buffer[100];
+    char buffer[1000];
   };
   struct FileSummaryStorage {
-    char buffer[1000];
+    char buffer[3000];
   };
   const char* LevelSummary(LevelSummaryStorage* scratch) const;
   // Return a human-readable short (single-line) summary of files
@@ -233,6 +236,10 @@ class Version {
   // tables' propertis, represented as shared_ptr.
   Status GetPropertiesOfAllTables(TablePropertiesCollection* props);
 
+  uint64_t GetEstimatedActiveKeys();
+
+  size_t GetMemoryUsageByTableReaders();
+
   // used to sort files by size
   struct Fsize {
     int index;
@@ -243,17 +250,19 @@ class Version {
   friend class Compaction;
   friend class VersionSet;
   friend class DBImpl;
+  friend class CompactedDBImpl;
   friend class ColumnFamilyData;
   friend class CompactionPicker;
   friend class LevelCompactionPicker;
   friend class UniversalCompactionPicker;
   friend class FIFOCompactionPicker;
   friend class ForwardIterator;
+  friend class InternalStats;
 
   class LevelFileNumIterator;
   class LevelFileIteratorState;
 
-  bool PrefixMayMatch(const ReadOptions& options, Iterator* level_iter,
+  bool PrefixMayMatch(const ReadOptions& read_options, Iterator* level_iter,
                       const Slice& internal_prefix) const;
 
   // Update num_non_empty_levels_.
@@ -284,6 +293,7 @@ class Version {
   int num_levels_;              // Number of levels
   int num_non_empty_levels_;    // Number of levels. Any level larger than it
                                 // is guaranteed to be empty.
+  FileIndexer file_indexer_;
   VersionSet* vset_;            // VersionSet to which this Version belongs
   Arena arena_;                 // Used to allocate space for file_levels_
   Version* next_;               // Next version in linked list
@@ -294,7 +304,6 @@ class Version {
   // in increasing order of keys
   std::vector<FileMetaData*>* files_;
 
-
   // A list for the same set of files that are stored in files_,
   // but files in each level are now sorted based on file
   // size. The file with the largest size is at the front.
@@ -319,15 +328,14 @@ class Version {
   // These are used to pick the best compaction level
   std::vector<double> compaction_score_;
   std::vector<int> compaction_level_;
-  double max_compaction_score_; // max score in l1 to ln-1
-  int max_compaction_score_level_; // level on which max score occurs
+  double max_compaction_score_ = 0.0;   // max score in l1 to ln-1
+  int max_compaction_score_level_ = 0;  // level on which max score occurs
 
   // A version number that uniquely represents this version. This is
   // used for debugging and logging purposes only.
   uint64_t version_number_;
 
   Version(ColumnFamilyData* cfd, VersionSet* vset, uint64_t version_number = 0);
-  FileIndexer file_indexer_;
 
   // total file size
   uint64_t total_file_size_;
@@ -337,7 +345,8 @@ class Version {
   uint64_t total_raw_value_size_;
   // total number of non-deletion entries
   uint64_t num_non_deletions_;
-
+  // total number of deletion entries
+  uint64_t num_deletions_;
 
   ~Version();
 
@@ -354,8 +363,9 @@ class Version {
 
 class VersionSet {
  public:
-  VersionSet(const std::string& dbname, const DBOptions* options,
-             const EnvOptions& storage_options, Cache* table_cache);
+  VersionSet(const std::string& dbname, const DBOptions* db_options,
+             const EnvOptions& env_options, Cache* table_cache,
+             WriteController* write_controller);
   ~VersionSet();
 
   // Apply *edit to the current version to form a new descriptor that
@@ -364,7 +374,9 @@ class VersionSet {
   // column_family_options has to be set if edit is column family add
   // REQUIRES: *mu is held on entry.
   // REQUIRES: no other thread concurrently calls LogAndApply()
-  Status LogAndApply(ColumnFamilyData* column_family_data, VersionEdit* edit,
+  Status LogAndApply(ColumnFamilyData* column_family_data,
+                     const MutableCFOptions& mutable_cf_options,
+                     VersionEdit* edit,
                      port::Mutex* mu, Directory* db_directory = nullptr,
                      bool new_descriptor_log = false,
                      const ColumnFamilyOptions* column_family_options =
@@ -393,7 +405,7 @@ class VersionSet {
   // among [4-6] contains files.
   static Status ReduceNumberOfLevels(const std::string& dbname,
                                      const Options* options,
-                                     const EnvOptions& storage_options,
+                                     const EnvOptions& env_options,
                                      int new_levels);
 
   // printf contents (for debugging)
@@ -502,14 +514,14 @@ class VersionSet {
   bool ManifestContains(uint64_t manifest_file_number,
                         const std::string& record) const;
 
-  ColumnFamilyData* CreateColumnFamily(const ColumnFamilyOptions& options,
+  ColumnFamilyData* CreateColumnFamily(const ColumnFamilyOptions& cf_options,
                                        VersionEdit* edit);
 
   std::unique_ptr<ColumnFamilySet> column_family_set_;
 
   Env* const env_;
   const std::string dbname_;
-  const DBOptions* const options_;
+  const DBOptions* const db_options_;
   uint64_t next_file_number_;
   uint64_t manifest_file_number_;
   uint64_t pending_manifest_file_number_;
@@ -530,12 +542,12 @@ class VersionSet {
 
   std::vector<FileMetaData*> obsolete_files_;
 
-  // storage options for all reads and writes except compactions
-  const EnvOptions& storage_options_;
+  // env options for all reads and writes except compactions
+  const EnvOptions& env_options_;
 
-  // storage options used for compactions. This is a copy of
-  // storage_options_ but with readaheads set to readahead_compactions_.
-  const EnvOptions storage_options_compactions_;
+  // env options used for compactions. This is a copy of
+  // env_options_ but with readaheads set to readahead_compactions_.
+  const EnvOptions env_options_compactions_;
 
   // No copying allowed
   VersionSet(const VersionSet&);

db/version_set_test.cc

@@ -14,14 +14,15 @@
 
 namespace rocksdb {
 
-class FindFileTest {
+class GenerateFileLevelTest {
  public:
   std::vector<FileMetaData*> files_;
-  bool disjoint_sorted_files_;
+  FileLevel file_level_;
+  Arena arena_;
 
-  FindFileTest() : disjoint_sorted_files_(true) { }
+  GenerateFileLevelTest() { }
 
-  ~FindFileTest() {
+  ~GenerateFileLevelTest() {
     for (unsigned int i = 0; i < files_.size(); i++) {
       delete files_[i];
     }
@@ -37,55 +38,44 @@ class FindFileTest {
     files_.push_back(f);
   }
 
-  int Find(const char* key) {
-    InternalKey target(key, 100, kTypeValue);
-    InternalKeyComparator cmp(BytewiseComparator());
-    return FindFile(cmp, files_, target.Encode());
+  int Compare() {
+    int diff = 0;
+    for (size_t i = 0; i < files_.size(); i++) {
+      if (file_level_.files[i].fd.GetNumber() != files_[i]->fd.GetNumber()) {
+        diff++;
+      }
+    }
+    return diff;
   }
 };
 
-TEST(FindFileTest, Empty) {
-  ASSERT_EQ(0, Find("foo"));
+TEST(GenerateFileLevelTest, Empty) {
+  DoGenerateFileLevel(&file_level_, files_, &arena_);
+  ASSERT_EQ(0u, file_level_.num_files);
+  ASSERT_EQ(0, Compare());
 }
 
-TEST(FindFileTest, Single) {
+TEST(GenerateFileLevelTest, Single) {
   Add("p", "q");
-  ASSERT_EQ(0, Find("a"));
-  ASSERT_EQ(0, Find("p"));
-  ASSERT_EQ(0, Find("p1"));
-  ASSERT_EQ(0, Find("q"));
-  ASSERT_EQ(1, Find("q1"));
-  ASSERT_EQ(1, Find("z"));
+  DoGenerateFileLevel(&file_level_, files_, &arena_);
+  ASSERT_EQ(1u, file_level_.num_files);
+  ASSERT_EQ(0, Compare());
 }
 
 
-TEST(FindFileTest, Multiple) {
+TEST(GenerateFileLevelTest, Multiple) {
   Add("150", "200");
   Add("200", "250");
   Add("300", "350");
   Add("400", "450");
-  ASSERT_EQ(0, Find("100"));
-  ASSERT_EQ(0, Find("150"));
-  ASSERT_EQ(0, Find("151"));
-  ASSERT_EQ(0, Find("199"));
-  ASSERT_EQ(0, Find("200"));
-  ASSERT_EQ(1, Find("201"));
-  ASSERT_EQ(1, Find("249"));
-  ASSERT_EQ(1, Find("250"));
-  ASSERT_EQ(2, Find("251"));
-  ASSERT_EQ(2, Find("299"));
-  ASSERT_EQ(2, Find("300"));
-  ASSERT_EQ(2, Find("349"));
-  ASSERT_EQ(2, Find("350"));
-  ASSERT_EQ(3, Find("351"));
-  ASSERT_EQ(3, Find("400"));
-  ASSERT_EQ(3, Find("450"));
-  ASSERT_EQ(4, Find("451"));
+  DoGenerateFileLevel(&file_level_, files_, &arena_);
+  ASSERT_EQ(4u, file_level_.num_files);
+  ASSERT_EQ(0, Compare());
 }
 
 class FindLevelFileTest {
  public:
-  FileLevel level_files_;
+  FileLevel file_level_;
   bool disjoint_sorted_files_;
   Arena arena_;
 
@@ -96,8 +86,8 @@ class FindLevelFileTest {
 
   void LevelFileInit(size_t num = 0) {
     char* mem = arena_.AllocateAligned(num * sizeof(FdWithKeyRange));
-    level_files_.files = new (mem)FdWithKeyRange[num];
-    level_files_.num_files = 0;
+    file_level_.files = new (mem)FdWithKeyRange[num];
+    file_level_.num_files = 0;
   }
 
   void Add(const char* smallest, const char* largest,
@@ -115,27 +105,27 @@ class FindLevelFileTest {
     memcpy(mem + smallest_slice.size(), largest_slice.data(),
         largest_slice.size());
 
-    // add compressd_level_
-    size_t num = level_files_.num_files;
-    auto& file = level_files_.files[num];
+    // add to file_level_
+    size_t num = file_level_.num_files;
+    auto& file = file_level_.files[num];
     file.fd = FileDescriptor(num + 1, 0, 0);
     file.smallest_key = Slice(mem, smallest_slice.size());
     file.largest_key = Slice(mem + smallest_slice.size(),
         largest_slice.size());
-    level_files_.num_files++;
+    file_level_.num_files++;
   }
 
   int Find(const char* key) {
     InternalKey target(key, 100, kTypeValue);
     InternalKeyComparator cmp(BytewiseComparator());
-    return FindFile(cmp, level_files_, target.Encode());
+    return FindFile(cmp, file_level_, target.Encode());
   }
 
   bool Overlaps(const char* smallest, const char* largest) {
     InternalKeyComparator cmp(BytewiseComparator());
     Slice s(smallest != nullptr ? smallest : "");
     Slice l(largest != nullptr ? largest : "");
-    return SomeFileOverlapsRange(cmp, disjoint_sorted_files_, level_files_,
+    return SomeFileOverlapsRange(cmp, disjoint_sorted_files_, file_level_,
                                  (smallest != nullptr ? &s : nullptr),
                                  (largest != nullptr ? &l : nullptr));
   }

db/write_batch.cc

@@ -23,10 +23,10 @@
 //    data: uint8[len]
 
 #include "rocksdb/write_batch.h"
-#include "rocksdb/options.h"
 #include "rocksdb/merge_operator.h"
 #include "db/dbformat.h"
 #include "db/db_impl.h"
+#include "db/column_family.h"
 #include "db/memtable.h"
 #include "db/snapshot.h"
 #include "db/write_batch_internal.h"
@@ -80,6 +80,58 @@ int WriteBatch::Count() const {
   return WriteBatchInternal::Count(this);
 }
 
+Status ReadRecordFromWriteBatch(Slice* input, char* tag,
+                                uint32_t* column_family, Slice* key,
+                                Slice* value, Slice* blob) {
+  assert(key != nullptr && value != nullptr);
+  *tag = (*input)[0];
+  input->remove_prefix(1);
+  *column_family = 0;  // default
+  switch (*tag) {
+    case kTypeColumnFamilyValue:
+      if (!GetVarint32(input, column_family)) {
+        return Status::Corruption("bad WriteBatch Put");
+      }
+    // intentional fallthrough
+    case kTypeValue:
+      if (!GetLengthPrefixedSlice(input, key) ||
+          !GetLengthPrefixedSlice(input, value)) {
+        return Status::Corruption("bad WriteBatch Put");
+      }
+      break;
+    case kTypeColumnFamilyDeletion:
+      if (!GetVarint32(input, column_family)) {
+        return Status::Corruption("bad WriteBatch Delete");
+      }
+    // intentional fallthrough
+    case kTypeDeletion:
+      if (!GetLengthPrefixedSlice(input, key)) {
+        return Status::Corruption("bad WriteBatch Delete");
+      }
+      break;
+    case kTypeColumnFamilyMerge:
+      if (!GetVarint32(input, column_family)) {
+        return Status::Corruption("bad WriteBatch Merge");
+      }
+    // intentional fallthrough
+    case kTypeMerge:
+      if (!GetLengthPrefixedSlice(input, key) ||
+          !GetLengthPrefixedSlice(input, value)) {
+        return Status::Corruption("bad WriteBatch Merge");
+      }
+      break;
+    case kTypeLogData:
+      assert(blob != nullptr);
+      if (!GetLengthPrefixedSlice(input, blob)) {
+        return Status::Corruption("bad WriteBatch Blob");
+      }
+      break;
+    default:
+      return Status::Corruption("unknown WriteBatch tag");
+  }
+  return Status::OK();
+}
+
 Status WriteBatch::Iterate(Handler* handler) const {
   Slice input(rep_);
   if (input.size() < kHeader) {
@@ -91,57 +143,33 @@ Status WriteBatch::Iterate(Handler* handler) const {
   int found = 0;
   Status s;
   while (s.ok() && !input.empty() && handler->Continue()) {
-    char tag = input[0];
-    input.remove_prefix(1);
+    char tag = 0;
     uint32_t column_family = 0;  // default
+
+    s = ReadRecordFromWriteBatch(&input, &tag, &column_family, &key, &value,
+                                 &blob);
+    if (!s.ok()) {
+      return s;
+    }
+
     switch (tag) {
       case kTypeColumnFamilyValue:
-        if (!GetVarint32(&input, &column_family)) {
-          return Status::Corruption("bad WriteBatch Put");
-        }
-      // intentional fallthrough
       case kTypeValue:
-        if (GetLengthPrefixedSlice(&input, &key) &&
-            GetLengthPrefixedSlice(&input, &value)) {
-          s = handler->PutCF(column_family, key, value);
-          found++;
-        } else {
-          return Status::Corruption("bad WriteBatch Put");
-        }
+        s = handler->PutCF(column_family, key, value);
+        found++;
         break;
       case kTypeColumnFamilyDeletion:
-        if (!GetVarint32(&input, &column_family)) {
-          return Status::Corruption("bad WriteBatch Delete");
-        }
-      // intentional fallthrough
       case kTypeDeletion:
-        if (GetLengthPrefixedSlice(&input, &key)) {
-          s = handler->DeleteCF(column_family, key);
-          found++;
-        } else {
-          return Status::Corruption("bad WriteBatch Delete");
-        }
+        s = handler->DeleteCF(column_family, key);
+        found++;
         break;
       case kTypeColumnFamilyMerge:
-        if (!GetVarint32(&input, &column_family)) {
-          return Status::Corruption("bad WriteBatch Merge");
-        }
-      // intentional fallthrough
       case kTypeMerge:
-        if (GetLengthPrefixedSlice(&input, &key) &&
-            GetLengthPrefixedSlice(&input, &value)) {
-          s = handler->MergeCF(column_family, key, value);
-          found++;
-        } else {
-          return Status::Corruption("bad WriteBatch Merge");
-        }
+        s = handler->MergeCF(column_family, key, value);
+        found++;
         break;
       case kTypeLogData:
-        if (GetLengthPrefixedSlice(&input, &blob)) {
-          handler->LogData(blob);
-        } else {
-          return Status::Corruption("bad WriteBatch Blob");
-        }
+        handler->LogData(blob);
         break;
       default:
         return Status::Corruption("unknown WriteBatch tag");
@@ -186,17 +214,6 @@ void WriteBatchInternal::Put(WriteBatch* b, uint32_t column_family_id,
   PutLengthPrefixedSlice(&b->rep_, value);
 }
 
-namespace {
-inline uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family) {
-  uint32_t column_family_id = 0;
-  if (column_family != nullptr) {
-    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
-    column_family_id = cfh->GetID();
-  }
-  return column_family_id;
-}
-}  // namespace
-
 void WriteBatch::Put(ColumnFamilyHandle* column_family, const Slice& key,
                      const Slice& value) {
   WriteBatchInternal::Put(this, GetColumnFamilyID(column_family), key, value);
@@ -281,17 +298,17 @@ class MemTableInserter : public WriteBatch::Handler {
  public:
   SequenceNumber sequence_;
   ColumnFamilyMemTables* cf_mems_;
-  bool recovery_;
+  bool ignore_missing_column_families_;
   uint64_t log_number_;
   DBImpl* db_;
   const bool dont_filter_deletes_;
 
   MemTableInserter(SequenceNumber sequence, ColumnFamilyMemTables* cf_mems,
-                   bool recovery, uint64_t log_number, DB* db,
-                   const bool dont_filter_deletes)
+                   bool ignore_missing_column_families, uint64_t log_number,
+                   DB* db, const bool dont_filter_deletes)
       : sequence_(sequence),
         cf_mems_(cf_mems),
-        recovery_(recovery),
+        ignore_missing_column_families_(ignore_missing_column_families),
         log_number_(log_number),
         db_(reinterpret_cast<DBImpl*>(db)),
         dont_filter_deletes_(dont_filter_deletes) {
@@ -303,12 +320,18 @@ class MemTableInserter : public WriteBatch::Handler {
 
   bool SeekToColumnFamily(uint32_t column_family_id, Status* s) {
     bool found = cf_mems_->Seek(column_family_id);
-    if (recovery_ && (!found || log_number_ < cf_mems_->GetLogNumber())) {
-      // if in recovery envoronment:
-      // * 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.
+    if (!found) {
+      if (ignore_missing_column_families_) {
+        *s = Status::OK();
+      } else {
+        *s = Status::InvalidArgument(
+            "Invalid column family specified in write batch");
+      }
+      return false;
+    }
+    if (log_number_ != 0 && log_number_ < cf_mems_->GetLogNumber()) {
+      // This is true only in recovery environment (log_number_ is always 0 in
+      // non-recovery, regular write code-path)
       // * If log_number_ < cf_mems_->GetLogNumber(), this means that column
       // family already contains updates from this log. We can't apply updates
       // twice because of update-in-place or merge workloads -- ignore the
@@ -316,18 +339,8 @@ class MemTableInserter : public WriteBatch::Handler {
       *s = Status::OK();
       return false;
     }
-    if (!found) {
-      assert(!recovery_);
-      // If the column family was not found in non-recovery enviornment
-      // (client's write code-path), we have to fail the write and return
-      // the failure status to the client.
-      *s = Status::InvalidArgument(
-          "Invalid column family specified in write batch");
-      return false;
-    }
     return true;
   }
-
   virtual Status PutCF(uint32_t column_family_id, const Slice& key,
                        const Slice& value) {
     Status seek_status;
@@ -336,14 +349,15 @@ class MemTableInserter : public WriteBatch::Handler {
       return seek_status;
     }
     MemTable* mem = cf_mems_->GetMemTable();
-    const Options* options = cf_mems_->GetOptions();
-    if (!options->inplace_update_support) {
+    auto* ioptions = mem->GetImmutableOptions();
+    auto* moptions = mem->GetMemTableOptions();
+    if (!moptions->inplace_update_support) {
       mem->Add(sequence_, kTypeValue, key, value);
-    } else if (options->inplace_callback == nullptr) {
+    } else if (moptions->inplace_callback == nullptr) {
       mem->Update(sequence_, key, value);
-      RecordTick(options->statistics.get(), NUMBER_KEYS_UPDATED);
+      RecordTick(ioptions->statistics, NUMBER_KEYS_UPDATED);
     } else {
-      if (mem->UpdateCallback(sequence_, key, value, *options)) {
+      if (mem->UpdateCallback(sequence_, key, value)) {
       } else {
         // key not found in memtable. Do sst get, update, add
         SnapshotImpl read_from_snapshot;
@@ -362,17 +376,17 @@ class MemTableInserter : public WriteBatch::Handler {
 
         char* prev_buffer = const_cast<char*>(prev_value.c_str());
         uint32_t prev_size = prev_value.size();
-        auto status = options->inplace_callback(s.ok() ? prev_buffer : nullptr,
-                                                s.ok() ? &prev_size : nullptr,
-                                                value, &merged_value);
+        auto status = moptions->inplace_callback(s.ok() ? prev_buffer : nullptr,
+                                                 s.ok() ? &prev_size : nullptr,
+                                                 value, &merged_value);
         if (status == UpdateStatus::UPDATED_INPLACE) {
           // prev_value is updated in-place with final value.
           mem->Add(sequence_, kTypeValue, key, Slice(prev_buffer, prev_size));
-          RecordTick(options->statistics.get(), NUMBER_KEYS_WRITTEN);
+          RecordTick(ioptions->statistics, NUMBER_KEYS_WRITTEN);
         } else if (status == UpdateStatus::UPDATED) {
           // merged_value contains the final value.
           mem->Add(sequence_, kTypeValue, key, Slice(merged_value));
-          RecordTick(options->statistics.get(), NUMBER_KEYS_WRITTEN);
+          RecordTick(ioptions->statistics, NUMBER_KEYS_WRITTEN);
         }
       }
     }
@@ -380,6 +394,7 @@ class MemTableInserter : public WriteBatch::Handler {
     // sequence number. Even if the update eventually fails and does not result
     // in memtable add/update.
     sequence_++;
+    cf_mems_->CheckMemtableFull();
     return Status::OK();
   }
 
@@ -391,17 +406,18 @@ class MemTableInserter : public WriteBatch::Handler {
       return seek_status;
     }
     MemTable* mem = cf_mems_->GetMemTable();
-    const Options* options = cf_mems_->GetOptions();
+    auto* ioptions = mem->GetImmutableOptions();
+    auto* moptions = mem->GetMemTableOptions();
     bool perform_merge = false;
 
-    if (options->max_successive_merges > 0 && db_ != nullptr) {
+    if (moptions->max_successive_merges > 0 && db_ != nullptr) {
       LookupKey lkey(key, sequence_);
 
       // Count the number of successive merges at the head
       // of the key in the memtable
       size_t num_merges = mem->CountSuccessiveMergeEntries(lkey);
 
-      if (num_merges >= options->max_successive_merges) {
+      if (num_merges >= moptions->max_successive_merges) {
         perform_merge = true;
       }
     }
@@ -425,19 +441,19 @@ class MemTableInserter : public WriteBatch::Handler {
       Slice get_value_slice = Slice(get_value);
 
       // 2) Apply this merge
-      auto merge_operator = options->merge_operator.get();
+      auto merge_operator = ioptions->merge_operator;
       assert(merge_operator);
 
       std::deque<std::string> operands;
       operands.push_front(value.ToString());
       std::string new_value;
       if (!merge_operator->FullMerge(key, &get_value_slice, operands,
-                                     &new_value, options->info_log.get())) {
+                                     &new_value, ioptions->info_log)) {
           // Failed to merge!
-        RecordTick(options->statistics.get(), NUMBER_MERGE_FAILURES);
+        RecordTick(ioptions->statistics, NUMBER_MERGE_FAILURES);
 
-          // Store the delta in memtable
-          perform_merge = false;
+        // Store the delta in memtable
+        perform_merge = false;
       } else {
         // 3) Add value to memtable
         mem->Add(sequence_, kTypeValue, key, new_value);
@@ -450,6 +466,7 @@ class MemTableInserter : public WriteBatch::Handler {
     }
 
     sequence_++;
+    cf_mems_->CheckMemtableFull();
     return Status::OK();
   }
 
@@ -460,8 +477,9 @@ class MemTableInserter : public WriteBatch::Handler {
       return seek_status;
     }
     MemTable* mem = cf_mems_->GetMemTable();
-    const Options* options = cf_mems_->GetOptions();
-    if (!dont_filter_deletes_ && options->filter_deletes) {
+    auto* ioptions = mem->GetImmutableOptions();
+    auto* moptions = mem->GetMemTableOptions();
+    if (!dont_filter_deletes_ && moptions->filter_deletes) {
       SnapshotImpl read_from_snapshot;
       read_from_snapshot.number_ = sequence_;
       ReadOptions ropts;
@@ -472,12 +490,13 @@ class MemTableInserter : public WriteBatch::Handler {
         cf_handle = db_->DefaultColumnFamily();
       }
       if (!db_->KeyMayExist(ropts, cf_handle, key, &value)) {
-        RecordTick(options->statistics.get(), NUMBER_FILTERED_DELETES);
+        RecordTick(ioptions->statistics, NUMBER_FILTERED_DELETES);
         return Status::OK();
       }
     }
     mem->Add(sequence_, kTypeDeletion, key, Slice());
     sequence_++;
+    cf_mems_->CheckMemtableFull();
     return Status::OK();
   }
 };
@@ -485,10 +504,12 @@ class MemTableInserter : public WriteBatch::Handler {
 
 Status WriteBatchInternal::InsertInto(const WriteBatch* b,
                                       ColumnFamilyMemTables* memtables,
-                                      bool recovery, uint64_t log_number,
-                                      DB* db, const bool dont_filter_deletes) {
+                                      bool ignore_missing_column_families,
+                                      uint64_t log_number, DB* db,
+                                      const bool dont_filter_deletes) {
   MemTableInserter inserter(WriteBatchInternal::Sequence(b), memtables,
-                            recovery, log_number, db, dont_filter_deletes);
+                            ignore_missing_column_families, log_number, db,
+                            dont_filter_deletes);
   return b->Iterate(&inserter);
 }
 

db/write_batch_internal.h

@@ -28,6 +28,7 @@ class ColumnFamilyMemTables {
   virtual MemTable* GetMemTable() const = 0;
   virtual const Options* GetOptions() const = 0;
   virtual ColumnFamilyHandle* GetColumnFamilyHandle() = 0;
+  virtual void CheckMemtableFull() = 0;
 };
 
 class ColumnFamilyMemTablesDefault : public ColumnFamilyMemTables {
@@ -54,6 +55,8 @@ class ColumnFamilyMemTablesDefault : public ColumnFamilyMemTables {
 
   ColumnFamilyHandle* GetColumnFamilyHandle() override { return nullptr; }
 
+  void CheckMemtableFull() override {}
+
  private:
   bool ok_;
   MemTable* mem_;
@@ -106,18 +109,18 @@ class WriteBatchInternal {
   // Inserts batch entries into memtable
   // If dont_filter_deletes is false AND options.filter_deletes is true,
   // then --> Drops deletes in batch if db->KeyMayExist returns false
-  // If recovery == true, this means InsertInto is executed on a recovery
-  // code-path. WriteBatch referencing a dropped column family can be
-  // found on a recovery code-path and should be ignored (recovery should not
-  // fail). Additionally, the memtable will be updated only if
+  // If ignore_missing_column_families == true. WriteBatch referencing
+  // non-existing column family should be ignored.
+  // However, if ignore_missing_column_families == false, any WriteBatch
+  // referencing non-existing column family will return a InvalidArgument()
+  // failure.
+  //
+  // If log_number is non-zero, the memtable will be updated only if
   // memtables->GetLogNumber() >= log_number
-  // However, if recovery == false, any WriteBatch referencing
-  // non-existing column family will return a failure. Also, log_number is
-  // ignored in that case
   static Status InsertInto(const WriteBatch* batch,
                            ColumnFamilyMemTables* memtables,
-                           bool recovery = false, uint64_t log_number = 0,
-                           DB* db = nullptr,
+                           bool ignore_missing_column_families = false,
+                           uint64_t log_number = 0, DB* db = nullptr,
                            const bool dont_filter_deletes = true);
 
   static void Append(WriteBatch* dst, const WriteBatch* src);

db/write_batch_test.cc

@@ -15,8 +15,10 @@
 #include "db/write_batch_internal.h"
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
+#include "rocksdb/utilities/write_batch_with_index.h"
 #include "util/logging.h"
 #include "util/testharness.h"
+#include "util/scoped_arena_iterator.h"
 
 namespace rocksdb {
 
@@ -25,13 +27,16 @@ static std::string PrintContents(WriteBatch* b) {
   auto factory = std::make_shared<SkipListFactory>();
   Options options;
   options.memtable_factory = factory;
-  MemTable* mem = new MemTable(cmp, options);
+  ImmutableCFOptions ioptions(options);
+  MemTable* mem = new MemTable(cmp, ioptions,
+      MemTableOptions(MutableCFOptions(options, ioptions), options));
   mem->Ref();
   std::string state;
   ColumnFamilyMemTablesDefault cf_mems_default(mem, &options);
   Status s = WriteBatchInternal::InsertInto(b, &cf_mems_default);
   int count = 0;
-  Iterator* iter = mem->NewIterator(ReadOptions());
+  Arena arena;
+  ScopedArenaIterator iter(mem->NewIterator(ReadOptions(), &arena));
   for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
     ParsedInternalKey ikey;
     memset((void *)&ikey, 0, sizeof(ikey));
@@ -66,7 +71,6 @@ static std::string PrintContents(WriteBatch* b) {
     state.append("@");
     state.append(NumberToString(ikey.sequence));
   }
-  delete iter;
   if (!s.ok()) {
     state.append(s.ToString());
   } else if (count != WriteBatchInternal::Count(b)) {
@@ -286,6 +290,9 @@ class ColumnFamilyHandleImplDummy : public ColumnFamilyHandleImpl {
   explicit ColumnFamilyHandleImplDummy(int id)
       : ColumnFamilyHandleImpl(nullptr, nullptr, nullptr), id_(id) {}
   uint32_t GetID() const override { return id_; }
+  const Comparator* user_comparator() const override {
+    return BytewiseComparator();
+  }
 
  private:
   uint32_t id_;
@@ -316,6 +323,88 @@ TEST(WriteBatchTest, ColumnFamiliesBatchTest) {
       handler.seen);
 }
 
+TEST(WriteBatchTest, ColumnFamiliesBatchWithIndexTest) {
+  WriteBatchWithIndex batch;
+  ColumnFamilyHandleImplDummy zero(0), two(2), three(3), eight(8);
+  batch.Put(&zero, Slice("foo"), Slice("bar"));
+  batch.Put(&two, Slice("twofoo"), Slice("bar2"));
+  batch.Put(&eight, Slice("eightfoo"), Slice("bar8"));
+  batch.Delete(&eight, Slice("eightfoo"));
+  batch.Merge(&three, Slice("threethree"), Slice("3three"));
+  batch.Put(&zero, Slice("foo"), Slice("bar"));
+  batch.Merge(Slice("omom"), Slice("nom"));
+
+  std::unique_ptr<WBWIIterator> iter;
+
+  iter.reset(batch.NewIterator(&eight));
+  iter->Seek("eightfoo");
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kPutRecord, iter->Entry().type);
+  ASSERT_EQ("eightfoo", iter->Entry().key.ToString());
+  ASSERT_EQ("bar8", iter->Entry().value.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kDeleteRecord, iter->Entry().type);
+  ASSERT_EQ("eightfoo", iter->Entry().key.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(!iter->Valid());
+
+  iter.reset(batch.NewIterator());
+  iter->Seek("gggg");
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kMergeRecord, iter->Entry().type);
+  ASSERT_EQ("omom", iter->Entry().key.ToString());
+  ASSERT_EQ("nom", iter->Entry().value.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(!iter->Valid());
+
+  iter.reset(batch.NewIterator(&zero));
+  iter->Seek("foo");
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kPutRecord, iter->Entry().type);
+  ASSERT_EQ("foo", iter->Entry().key.ToString());
+  ASSERT_EQ("bar", iter->Entry().value.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kPutRecord, iter->Entry().type);
+  ASSERT_EQ("foo", iter->Entry().key.ToString());
+  ASSERT_EQ("bar", iter->Entry().value.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ(WriteType::kMergeRecord, iter->Entry().type);
+  ASSERT_EQ("omom", iter->Entry().key.ToString());
+  ASSERT_EQ("nom", iter->Entry().value.ToString());
+
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(!iter->Valid());
+
+  TestHandler handler;
+  batch.GetWriteBatch()->Iterate(&handler);
+  ASSERT_EQ(
+      "Put(foo, bar)"
+      "PutCF(2, twofoo, bar2)"
+      "PutCF(8, eightfoo, bar8)"
+      "DeleteCF(8, eightfoo)"
+      "MergeCF(3, threethree, 3three)"
+      "Put(foo, bar)"
+      "Merge(omom, nom)",
+      handler.seen);
+}
+
 }  // namespace rocksdb
 
 int main(int argc, char** argv) {

db/write_controller.cc

@@ -0,0 +1,37 @@
+//  Copyright (c) 2013, 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 "db/write_controller.h"
+
+#include <cassert>
+
+namespace rocksdb {
+
+std::unique_ptr<WriteControllerToken> WriteController::GetStopToken() {
+  ++total_stopped_;
+  return std::unique_ptr<WriteControllerToken>(new StopWriteToken(this));
+}
+
+std::unique_ptr<WriteControllerToken> WriteController::GetDelayToken(
+    uint64_t delay_us) {
+  total_delay_us_ += delay_us;
+  return std::unique_ptr<WriteControllerToken>(
+      new DelayWriteToken(this, delay_us));
+}
+
+bool WriteController::IsStopped() const { return total_stopped_ > 0; }
+uint64_t WriteController::GetDelay() const { return total_delay_us_; }
+
+StopWriteToken::~StopWriteToken() {
+  assert(controller_->total_stopped_ >= 1);
+  --controller_->total_stopped_;
+}
+
+DelayWriteToken::~DelayWriteToken() {
+  assert(controller_->total_delay_us_ >= delay_us_);
+  controller_->total_delay_us_ -= delay_us_;
+}
+
+}  // namespace rocksdb

db/write_controller.h

@@ -0,0 +1,78 @@
+//  Copyright (c) 2013, 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 <stdint.h>
+
+#include <memory>
+
+namespace rocksdb {
+
+class WriteControllerToken;
+
+// WriteController is controlling write stalls in our write code-path. Write
+// stalls happen when compaction can't keep up with write rate.
+// All of the methods here (including WriteControllerToken's destructors) need
+// to be called while holding DB mutex
+class WriteController {
+ public:
+  WriteController() : total_stopped_(0), total_delay_us_(0) {}
+  ~WriteController() = default;
+
+  // When an actor (column family) requests a stop token, all writes will be
+  // stopped until the stop token is released (deleted)
+  std::unique_ptr<WriteControllerToken> GetStopToken();
+  // When an actor (column family) requests a delay token, total delay for all
+  // writes will be increased by delay_us. The delay will last until delay token
+  // is released
+  std::unique_ptr<WriteControllerToken> GetDelayToken(uint64_t delay_us);
+
+  // these two metods are querying the state of the WriteController
+  bool IsStopped() const;
+  uint64_t GetDelay() const;
+
+ private:
+  friend class WriteControllerToken;
+  friend class StopWriteToken;
+  friend class DelayWriteToken;
+
+  int total_stopped_;
+  uint64_t total_delay_us_;
+};
+
+class WriteControllerToken {
+ public:
+  explicit WriteControllerToken(WriteController* controller)
+      : controller_(controller) {}
+  virtual ~WriteControllerToken() {}
+
+ protected:
+  WriteController* controller_;
+
+ private:
+  // no copying allowed
+  WriteControllerToken(const WriteControllerToken&) = delete;
+  void operator=(const WriteControllerToken&) = delete;
+};
+
+class StopWriteToken : public WriteControllerToken {
+ public:
+  explicit StopWriteToken(WriteController* controller)
+      : WriteControllerToken(controller) {}
+  virtual ~StopWriteToken();
+};
+
+class DelayWriteToken : public WriteControllerToken {
+ public:
+  DelayWriteToken(WriteController* controller, uint64_t delay_us)
+      : WriteControllerToken(controller), delay_us_(delay_us) {}
+  virtual ~DelayWriteToken();
+
+ private:
+  uint64_t delay_us_;
+};
+
+}  // namespace rocksdb

db/write_controller_test.cc

@@ -0,0 +1,40 @@
+//  Copyright (c) 2013, 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 "db/write_controller.h"
+
+#include "util/testharness.h"
+
+namespace rocksdb {
+
+class WriteControllerTest {};
+
+TEST(WriteControllerTest, SanityTest) {
+  WriteController controller;
+  auto stop_token_1 = controller.GetStopToken();
+  auto stop_token_2 = controller.GetStopToken();
+
+  ASSERT_EQ(true, controller.IsStopped());
+  stop_token_1.reset();
+  ASSERT_EQ(true, controller.IsStopped());
+  stop_token_2.reset();
+  ASSERT_EQ(false, controller.IsStopped());
+
+  auto delay_token_1 = controller.GetDelayToken(5);
+  ASSERT_EQ(static_cast<uint64_t>(5), controller.GetDelay());
+  auto delay_token_2 = controller.GetDelayToken(8);
+  ASSERT_EQ(static_cast<uint64_t>(13), controller.GetDelay());
+
+  delay_token_2.reset();
+  ASSERT_EQ(static_cast<uint64_t>(5), controller.GetDelay());
+  delay_token_1.reset();
+  ASSERT_EQ(static_cast<uint64_t>(0), controller.GetDelay());
+  delay_token_1.reset();
+  ASSERT_EQ(false, controller.IsStopped());
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) { return rocksdb::test::RunAllTests(); }

db/write_thread.cc

@@ -0,0 +1,147 @@
+//  Copyright (c) 2013, 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 "db/write_thread.h"
+
+namespace rocksdb {
+
+Status WriteThread::EnterWriteThread(WriteThread::Writer* w,
+                                     uint64_t expiration_time) {
+  // the following code block pushes the current writer "w" into the writer
+  // queue "writers_" and wait until one of the following conditions met:
+  // 1. the job of "w" has been done by some other writers.
+  // 2. "w" becomes the first writer in "writers_"
+  // 3. "w" timed-out.
+  writers_.push_back(w);
+
+  bool timed_out = false;
+  while (!w->done && w != writers_.front()) {
+    if (expiration_time == 0) {
+      w->cv.Wait();
+    } else if (w->cv.TimedWait(expiration_time)) {
+      if (w->in_batch_group) {
+        // then it means the front writer is currently doing the
+        // write on behalf of this "timed-out" writer.  Then it
+        // should wait until the write completes.
+        expiration_time = 0;
+      } else {
+        timed_out = true;
+        break;
+      }
+    }
+  }
+
+  if (timed_out) {
+#ifndef NDEBUG
+    bool found = false;
+#endif
+    for (auto iter = writers_.begin(); iter != writers_.end(); iter++) {
+      if (*iter == w) {
+        writers_.erase(iter);
+#ifndef NDEBUG
+        found = true;
+#endif
+        break;
+      }
+    }
+#ifndef NDEBUG
+    assert(found);
+#endif
+    // writers_.front() might still be in cond_wait without a time-out.
+    // As a result, we need to signal it to wake it up.  Otherwise no
+    // one else will wake him up, and RocksDB will hang.
+    if (!writers_.empty()) {
+      writers_.front()->cv.Signal();
+    }
+    return Status::TimedOut();
+  }
+  return Status::OK();
+}
+
+void WriteThread::ExitWriteThread(WriteThread::Writer* w,
+                                  WriteThread::Writer* last_writer,
+                                  Status status) {
+  // Pop out the current writer and all writers being pushed before the
+  // current writer from the writer queue.
+  while (!writers_.empty()) {
+    Writer* ready = writers_.front();
+    writers_.pop_front();
+    if (ready != w) {
+      ready->status = status;
+      ready->done = true;
+      ready->cv.Signal();
+    }
+    if (ready == last_writer) break;
+  }
+
+  // Notify new head of write queue
+  if (!writers_.empty()) {
+    writers_.front()->cv.Signal();
+  }
+}
+
+// This function will be called only when the first writer succeeds.
+// All writers in the to-be-built batch group will be processed.
+//
+// REQUIRES: Writer list must be non-empty
+// REQUIRES: First writer must have a non-nullptr batch
+void WriteThread::BuildBatchGroup(WriteThread::Writer** last_writer,
+                                  autovector<WriteBatch*>* write_batch_group) {
+  assert(!writers_.empty());
+  Writer* first = writers_.front();
+  assert(first->batch != nullptr);
+
+  size_t size = WriteBatchInternal::ByteSize(first->batch);
+  write_batch_group->push_back(first->batch);
+
+  // Allow the group to grow up to a maximum size, but if the
+  // original write is small, limit the growth so we do not slow
+  // down the small write too much.
+  size_t max_size = 1 << 20;
+  if (size <= (128<<10)) {
+    max_size = size + (128<<10);
+  }
+
+  *last_writer = first;
+  std::deque<Writer*>::iterator iter = writers_.begin();
+  ++iter;  // Advance past "first"
+  for (; iter != writers_.end(); ++iter) {
+    Writer* w = *iter;
+    if (w->sync && !first->sync) {
+      // Do not include a sync write into a batch handled by a non-sync write.
+      break;
+    }
+
+    if (!w->disableWAL && first->disableWAL) {
+      // Do not include a write that needs WAL into a batch that has
+      // WAL disabled.
+      break;
+    }
+
+    if (w->timeout_hint_us < first->timeout_hint_us) {
+      // Do not include those writes with shorter timeout.  Otherwise, we might
+      // execute a write that should instead be aborted because of timeout.
+      break;
+    }
+
+    if (w->batch == nullptr) {
+      // Do not include those writes with nullptr batch. Those are not writes,
+      // those are something else. They want to be alone
+      break;
+    }
+
+    size += WriteBatchInternal::ByteSize(w->batch);
+    if (size > max_size) {
+      // Do not make batch too big
+      break;
+    }
+
+    write_batch_group->push_back(w->batch);
+    w->in_batch_group = true;
+    *last_writer = w;
+  }
+}
+
+}  // namespace rocksdb

db/write_thread.h

@@ -0,0 +1,80 @@
+//  Copyright (c) 2013, 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 <stdint.h>
+#include <deque>
+#include <limits>
+#include "rocksdb/status.h"
+#include "db/write_batch_internal.h"
+#include "util/autovector.h"
+#include "port/port.h"
+
+namespace rocksdb {
+
+class WriteThread {
+ public:
+  static const uint64_t kNoTimeOut = std::numeric_limits<uint64_t>::max();
+  // Information kept for every waiting writer
+  struct Writer {
+    Status status;
+    WriteBatch* batch;
+    bool sync;
+    bool disableWAL;
+    bool in_batch_group;
+    bool done;
+    uint64_t timeout_hint_us;
+    port::CondVar cv;
+
+    explicit Writer(port::Mutex* mu)
+        : batch(nullptr),
+          sync(false),
+          disableWAL(false),
+          in_batch_group(false),
+          done(false),
+          timeout_hint_us(kNoTimeOut),
+          cv(mu) {}
+  };
+
+  WriteThread() = default;
+  ~WriteThread() = default;
+
+  // Before applying write operation (such as DBImpl::Write, DBImpl::Flush)
+  // thread should grab the mutex_ and be the first on writers queue.
+  // EnterWriteThread is used for it.
+  // Be aware! Writer's job can be done by other thread (see DBImpl::Write
+  // for examples), so check it via w.done before applying changes.
+  //
+  // Writer* w:                writer to be placed in the queue
+  // uint64_t expiration_time: maximum time to be in the queue
+  // See also: ExitWriteThread
+  // REQUIRES: db mutex held
+  Status EnterWriteThread(Writer* w, uint64_t expiration_time);
+
+  // After doing write job, we need to remove already used writers from
+  // writers_ queue and notify head of the queue about it.
+  // ExitWriteThread is used for this.
+  //
+  // Writer* w:           Writer, that was added by EnterWriteThread function
+  // Writer* last_writer: Since we can join a few Writers (as DBImpl::Write
+  //                      does)
+  //                      we should pass last_writer as a parameter to
+  //                      ExitWriteThread
+  //                      (if you don't touch other writers, just pass w)
+  // Status status:       Status of write operation
+  // See also: EnterWriteThread
+  // REQUIRES: db mutex held
+  void ExitWriteThread(Writer* w, Writer* last_writer, Status status);
+
+  void BuildBatchGroup(Writer** last_writer,
+                       autovector<WriteBatch*>* write_batch_group);
+
+ private:
+  // Queue of writers.
+  std::deque<Writer*> writers_;
+};
+
+}  // namespace rocksdb

include/rocksdb/c.h

@@ -6,8 +6,8 @@
   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.
 
-  C bindings for leveldb.  May be useful as a stable ABI that can be
-  used by programs that keep leveldb in a shared library, or for
+  C bindings for rocksdb.  May be useful as a stable ABI that can be
+  used by programs that keep rocksdb in a shared library, or for
   a JNI api.
 
   Does not support:
@@ -61,6 +61,10 @@ typedef struct rocksdb_compactionfiltercontext_t
     rocksdb_compactionfiltercontext_t;
 typedef struct rocksdb_compactionfilterfactory_t
     rocksdb_compactionfilterfactory_t;
+typedef struct rocksdb_compactionfilterv2_t
+    rocksdb_compactionfilterv2_t;
+typedef struct rocksdb_compactionfilterfactoryv2_t
+    rocksdb_compactionfilterfactoryv2_t;
 typedef struct rocksdb_comparator_t      rocksdb_comparator_t;
 typedef struct rocksdb_env_t             rocksdb_env_t;
 typedef struct rocksdb_fifo_compaction_options_t rocksdb_fifo_compaction_options_t;
@@ -71,6 +75,8 @@ typedef struct rocksdb_iterator_t        rocksdb_iterator_t;
 typedef struct rocksdb_logger_t          rocksdb_logger_t;
 typedef struct rocksdb_mergeoperator_t   rocksdb_mergeoperator_t;
 typedef struct rocksdb_options_t         rocksdb_options_t;
+typedef struct rocksdb_block_based_table_options_t
+    rocksdb_block_based_table_options_t;
 typedef struct rocksdb_randomfile_t      rocksdb_randomfile_t;
 typedef struct rocksdb_readoptions_t     rocksdb_readoptions_t;
 typedef struct rocksdb_seqfile_t         rocksdb_seqfile_t;
@@ -342,6 +348,34 @@ extern void rocksdb_writebatch_iterate(
     void (*deleted)(void*, const char* k, size_t klen));
 extern const char* rocksdb_writebatch_data(rocksdb_writebatch_t*, size_t *size);
 
+/* Block based table options */
+
+extern rocksdb_block_based_table_options_t*
+    rocksdb_block_based_options_create();
+extern void rocksdb_block_based_options_destroy(
+    rocksdb_block_based_table_options_t* options);
+extern void rocksdb_block_based_options_set_block_size(
+    rocksdb_block_based_table_options_t* options, size_t block_size);
+extern void rocksdb_block_based_options_set_block_size_deviation(
+    rocksdb_block_based_table_options_t* options, int block_size_deviation);
+extern void rocksdb_block_based_options_set_block_restart_interval(
+    rocksdb_block_based_table_options_t* options, int block_restart_interval);
+extern void rocksdb_block_based_options_set_filter_policy(
+    rocksdb_block_based_table_options_t* options,
+    rocksdb_filterpolicy_t* filter_policy);
+extern void rocksdb_block_based_options_set_no_block_cache(
+    rocksdb_block_based_table_options_t* options,
+    unsigned char no_block_cache);
+extern void rocksdb_block_based_options_set_block_cache(
+    rocksdb_block_based_table_options_t* options, rocksdb_cache_t* block_cache);
+extern void rocksdb_block_based_options_set_block_cache_compressed(
+    rocksdb_block_based_table_options_t* options,
+    rocksdb_cache_t* block_cache_compressed);
+extern void rocksdb_block_based_options_set_whole_key_filtering(
+    rocksdb_block_based_table_options_t*, unsigned char);
+extern void rocksdb_options_set_block_based_table_factory(
+    rocksdb_options_t *opt, rocksdb_block_based_table_options_t* table_options);
+
 /* Options */
 
 extern rocksdb_options_t* rocksdb_options_create();
@@ -349,7 +383,7 @@ extern void rocksdb_options_destroy(rocksdb_options_t*);
 extern void rocksdb_options_increase_parallelism(
     rocksdb_options_t* opt, int total_threads);
 extern void rocksdb_options_optimize_for_point_lookup(
-    rocksdb_options_t* opt);
+    rocksdb_options_t* opt, uint64_t block_cache_size_mb);
 extern void rocksdb_options_optimize_level_style_compaction(
     rocksdb_options_t* opt, uint64_t memtable_memory_budget);
 extern void rocksdb_options_optimize_universal_style_compaction(
@@ -359,18 +393,19 @@ extern void rocksdb_options_set_compaction_filter(
     rocksdb_compactionfilter_t*);
 extern void rocksdb_options_set_compaction_filter_factory(
     rocksdb_options_t*, rocksdb_compactionfilterfactory_t*);
+extern void rocksdb_options_set_compaction_filter_factory_v2(
+    rocksdb_options_t*,
+    rocksdb_compactionfilterfactoryv2_t*);
 extern void rocksdb_options_set_comparator(
     rocksdb_options_t*,
     rocksdb_comparator_t*);
-extern void rocksdb_options_set_merge_operator(rocksdb_options_t*,
-                                               rocksdb_mergeoperator_t*);
+extern void rocksdb_options_set_merge_operator(
+    rocksdb_options_t*,
+    rocksdb_mergeoperator_t*);
 extern void rocksdb_options_set_compression_per_level(
   rocksdb_options_t* opt,
   int* level_values,
   size_t num_levels);
-extern void rocksdb_options_set_filter_policy(
-    rocksdb_options_t*,
-    rocksdb_filterpolicy_t*);
 extern void rocksdb_options_set_create_if_missing(
     rocksdb_options_t*, unsigned char);
 extern void rocksdb_options_set_create_missing_column_families(
@@ -384,13 +419,8 @@ extern void rocksdb_options_set_info_log(rocksdb_options_t*, rocksdb_logger_t*);
 extern void rocksdb_options_set_info_log_level(rocksdb_options_t*, int);
 extern void rocksdb_options_set_write_buffer_size(rocksdb_options_t*, size_t);
 extern void rocksdb_options_set_max_open_files(rocksdb_options_t*, int);
-extern void rocksdb_options_set_cache(rocksdb_options_t*, rocksdb_cache_t*);
-extern void rocksdb_options_set_cache_compressed(rocksdb_options_t*, rocksdb_cache_t*);
-extern void rocksdb_options_set_block_size(rocksdb_options_t*, size_t);
-extern void rocksdb_options_set_block_restart_interval(rocksdb_options_t*, int);
 extern void rocksdb_options_set_compression_options(
     rocksdb_options_t*, int, int, int);
-extern void rocksdb_options_set_whole_key_filtering(rocksdb_options_t*, unsigned char);
 extern void rocksdb_options_set_prefix_extractor(
     rocksdb_options_t*, rocksdb_slicetransform_t*);
 extern void rocksdb_options_set_num_levels(rocksdb_options_t*, int);
@@ -441,8 +471,6 @@ extern void rocksdb_options_set_arena_block_size(
     rocksdb_options_t*, size_t);
 extern void rocksdb_options_set_use_fsync(
     rocksdb_options_t*, int);
-extern void rocksdb_options_set_db_stats_log_interval(
-    rocksdb_options_t*, int);
 extern void rocksdb_options_set_db_log_dir(
     rocksdb_options_t*, const char*);
 extern void rocksdb_options_set_wal_dir(
@@ -485,7 +513,6 @@ extern void rocksdb_options_set_max_sequential_skip_in_iterations(
     rocksdb_options_t*, uint64_t);
 extern void rocksdb_options_set_disable_data_sync(rocksdb_options_t*, int);
 extern void rocksdb_options_set_disable_auto_compactions(rocksdb_options_t*, int);
-extern void rocksdb_options_set_disable_seek_compaction(rocksdb_options_t*, int);
 extern void rocksdb_options_set_delete_obsolete_files_period_micros(
     rocksdb_options_t*, uint64_t);
 extern void rocksdb_options_set_source_compaction_factor(rocksdb_options_t*, int);
@@ -510,8 +537,6 @@ extern void rocksdb_options_set_min_partial_merge_operands(
     rocksdb_options_t*, uint32_t);
 extern void rocksdb_options_set_bloom_locality(
     rocksdb_options_t*, uint32_t);
-extern void rocksdb_options_set_allow_thread_local(
-    rocksdb_options_t*, unsigned char);
 extern void rocksdb_options_set_inplace_update_support(
     rocksdb_options_t*, unsigned char);
 extern void rocksdb_options_set_inplace_update_num_locks(
@@ -571,6 +596,35 @@ extern rocksdb_compactionfilterfactory_t*
 extern void rocksdb_compactionfilterfactory_destroy(
     rocksdb_compactionfilterfactory_t*);
 
+/* Compaction Filter V2 */
+
+extern rocksdb_compactionfilterv2_t* rocksdb_compactionfilterv2_create(
+    void* state,
+    void (*destructor)(void*),
+    // num_keys specifies the number of array entries in every *list parameter.
+    // New values added to the new_values_list should be malloc'd and will be
+    // freed by the caller. Specify true in the to_delete_list to remove an
+    // entry during compaction; false to keep it.
+    void (*filter)(
+        void*, int level, size_t num_keys,
+        const char* const* keys_list, const size_t* keys_list_sizes,
+        const char* const* existing_values_list, const size_t* existing_values_list_sizes,
+        char** new_values_list, size_t* new_values_list_sizes,
+        unsigned char* to_delete_list),
+    const char* (*name)(void*));
+extern void rocksdb_compactionfilterv2_destroy(rocksdb_compactionfilterv2_t*);
+
+/* Compaction Filter Factory V2 */
+
+extern rocksdb_compactionfilterfactoryv2_t* rocksdb_compactionfilterfactoryv2_create(
+    void* state,
+    rocksdb_slicetransform_t* prefix_extractor,
+    void (*destructor)(void*),
+    rocksdb_compactionfilterv2_t* (*create_compaction_filter_v2)(
+        void*, const rocksdb_compactionfiltercontext_t* context),
+    const char* (*name)(void*));
+extern void rocksdb_compactionfilterfactoryv2_destroy(rocksdb_compactionfilterfactoryv2_t*);
+
 /* Comparator */
 
 extern rocksdb_comparator_t* rocksdb_comparator_create(
@@ -642,6 +696,10 @@ extern void rocksdb_readoptions_set_fill_cache(
 extern void rocksdb_readoptions_set_snapshot(
     rocksdb_readoptions_t*,
     const rocksdb_snapshot_t*);
+extern void rocksdb_readoptions_set_iterate_upper_bound(
+    rocksdb_readoptions_t*,
+    const char* key,
+    size_t keylen);
 extern void rocksdb_readoptions_set_read_tier(
     rocksdb_readoptions_t*, int);
 extern void rocksdb_readoptions_set_tailing(

include/rocksdb/cache.h

@@ -127,9 +127,6 @@ class Cache {
   void LRU_Append(Handle* e);
   void Unref(Handle* e);
 
-  struct Rep;
-  Rep* rep_;
-
   // No copying allowed
   Cache(const Cache&);
   void operator=(const Cache&);

include/rocksdb/compaction_filter.h

@@ -9,6 +9,7 @@
 #ifndef STORAGE_ROCKSDB_INCLUDE_COMPACTION_FILTER_H_
 #define STORAGE_ROCKSDB_INCLUDE_COMPACTION_FILTER_H_
 
+#include <memory>
 #include <string>
 #include <vector>
 

include/rocksdb/comparator.h

@@ -62,6 +62,10 @@ class Comparator {
 // must not be deleted.
 extern const Comparator* BytewiseComparator();
 
+// Return a builtin comparator that uses reverse lexicographic byte-wise
+// ordering. 
+extern const Comparator* ReverseBytewiseComparator();
+
 }  // namespace rocksdb
 
 #endif  // STORAGE_ROCKSDB_INCLUDE_COMPARATOR_H_

include/rocksdb/db.h

@@ -123,7 +123,7 @@ class DB {
 
   // Open DB with column families.
   // db_options specify database specific options
-  // column_families is the vector of all column families in the databse,
+  // column_families is the vector of all column families in the database,
   // containing column family name and options. You need to open ALL column
   // families in the database. To get the list of column families, you can use
   // ListColumnFamilies(). Also, you can open only a subset of column families
@@ -160,6 +160,7 @@ class DB {
   virtual Status DropColumnFamily(ColumnFamilyHandle* column_family);
 
   // Set the database entry for "key" to "value".
+  // If "key" already exists, it will be overwritten.
   // Returns OK on success, and a non-OK status on error.
   // Note: consider setting options.sync = true.
   virtual Status Put(const WriteOptions& options,
@@ -306,6 +307,14 @@ class DB {
     return GetProperty(DefaultColumnFamily(), property, value);
   }
 
+  // Similar to GetProperty(), but only works for a subset of properties whose
+  // return value is an integer. Return the value by integer.
+  virtual bool GetIntProperty(ColumnFamilyHandle* column_family,
+                              const Slice& property, uint64_t* value) = 0;
+  virtual bool GetIntProperty(const Slice& property, uint64_t* value) {
+    return GetIntProperty(DefaultColumnFamily(), property, value);
+  }
+
   // For each i in [0,n-1], store in "sizes[i]", the approximate
   // file system space used by keys in "[range[i].start .. range[i].limit)".
   //
@@ -338,15 +347,25 @@ class DB {
   // hosting all the files. In this case, client could set reduce_level
   // to true, to move the files back to the minimum level capable of holding
   // the data set or a given level (specified by non-negative target_level).
+  // Compaction outputs should be placed in options.db_paths[target_path_id].
+  // Behavior is undefined if target_path_id is out of range.
   virtual Status CompactRange(ColumnFamilyHandle* column_family,
                               const Slice* begin, const Slice* end,
-                              bool reduce_level = false,
-                              int target_level = -1) = 0;
+                              bool reduce_level = false, int target_level = -1,
+                              uint32_t target_path_id = 0) = 0;
   virtual Status CompactRange(const Slice* begin, const Slice* end,
-                              bool reduce_level = false,
-                              int target_level = -1) {
+                              bool reduce_level = false, int target_level = -1,
+                              uint32_t target_path_id = 0) {
     return CompactRange(DefaultColumnFamily(), begin, end, reduce_level,
-                        target_level);
+                        target_level, target_path_id);
+  }
+  virtual bool SetOptions(ColumnFamilyHandle* column_family,
+      const std::unordered_map<std::string, std::string>& new_options) {
+    return true;
+  }
+  virtual bool SetOptions(
+      const std::unordered_map<std::string, std::string>& new_options) {
+    return SetOptions(DefaultColumnFamily(), new_options);
   }
 
   // Number of levels used for this DB.

include/rocksdb/env.h

@@ -201,7 +201,7 @@ class Env {
   // Priority for scheduling job in thread pool
   enum Priority { LOW, HIGH, TOTAL };
 
-  // Priority for scheduling job in thread pool
+  // Priority for requesting bytes in rate limiter scheduler
   enum IOPriority {
     IO_LOW = 0,
     IO_HIGH = 1,
@@ -272,6 +272,9 @@ class Env {
   // default number: 1
   virtual void SetBackgroundThreads(int number, Priority pri = LOW) = 0;
 
+  // Lower IO priority for threads from the specified pool.
+  virtual void LowerThreadPoolIOPriority(Priority pool = LOW) {}
+
   // Converts seconds-since-Jan-01-1970 to a printable string
   virtual std::string TimeToString(uint64_t time) = 0;
 
@@ -779,6 +782,9 @@ class EnvWrapper : public Env {
   void SetBackgroundThreads(int num, Priority pri) {
     return target_->SetBackgroundThreads(num, pri);
   }
+  void LowerThreadPoolIOPriority(Priority pool = LOW) override {
+    target_->LowerThreadPoolIOPriority(pool);
+  }
   std::string TimeToString(uint64_t time) {
     return target_->TimeToString(time);
   }

include/rocksdb/filter_policy.h

@@ -21,11 +21,52 @@
 #define STORAGE_ROCKSDB_INCLUDE_FILTER_POLICY_H_
 
 #include <string>
+#include <memory>
 
 namespace rocksdb {
 
 class Slice;
 
+// A class that takes a bunch of keys, then generates filter
+class FilterBitsBuilder {
+ public:
+  virtual ~FilterBitsBuilder() {}
+
+  // Add Key to filter, you could use any way to store the key.
+  // Such as: storing hashes or original keys
+  // Keys are in sorted order and duplicated keys are possible.
+  virtual void AddKey(const Slice& key) = 0;
+
+  // Generate the filter using the keys that are added
+  // The return value of this function would be the filter bits,
+  // The ownership of actual data is set to buf
+  virtual Slice Finish(std::unique_ptr<const char[]>* buf) = 0;
+};
+
+// A class that checks if a key can be in filter
+// It should be initialized by Slice generated by BitsBuilder
+class FilterBitsReader {
+ public:
+  virtual ~FilterBitsReader() {}
+
+  // Check if the entry match the bits in filter
+  virtual bool MayMatch(const Slice& entry) = 0;
+};
+
+// We add a new format of filter block called full filter block
+// This new interface gives you more space of customization
+//
+// For the full filter block, you can plug in your version by implement
+// the FilterBitsBuilder and FilterBitsReader
+//
+// There are two sets of interface in FilterPolicy
+// Set 1: CreateFilter, KeyMayMatch: used for blockbased filter
+// Set 2: GetFilterBitsBuilder, GetFilterBitsReader, they are used for
+// full filter.
+// Set 1 MUST be implemented correctly, Set 2 is optional
+// RocksDB would first try using functions in Set 2. if they return nullptr,
+// it would use Set 1 instead.
+// You can choose filter type in NewBloomFilterPolicy
 class FilterPolicy {
  public:
   virtual ~FilterPolicy();
@@ -51,11 +92,28 @@ class FilterPolicy {
   // This method may return true or false if the key was not on the
   // list, but it should aim to return false with a high probability.
   virtual bool KeyMayMatch(const Slice& key, const Slice& filter) const = 0;
+
+  // Get the FilterBitsBuilder, which is ONLY used for full filter block
+  // It contains interface to take individual key, then generate filter
+  virtual FilterBitsBuilder* GetFilterBitsBuilder() const {
+    return nullptr;
+  }
+
+  // Get the FilterBitsReader, which is ONLY used for full filter block
+  // It contains interface to tell if key can be in filter
+  // The input slice should NOT be deleted by FilterPolicy
+  virtual FilterBitsReader* GetFilterBitsReader(const Slice& contents) const {
+    return nullptr;
+  }
 };
 
 // Return a new filter policy that uses a bloom filter with approximately
-// the specified number of bits per key.  A good value for bits_per_key
+// the specified number of bits per key.
+//
+// bits_per_key: bits per key in bloom filter. A good value for bits_per_key
 // is 10, which yields a filter with ~ 1% false positive rate.
+// use_block_based_builder: use block based filter rather than full fiter.
+// If you want to builder full filter, it needs to be set to false.
 //
 // Callers must delete the result after any database that is using the
 // result has been closed.
@@ -67,8 +125,8 @@ class FilterPolicy {
 // ignores trailing spaces, it would be incorrect to use a
 // FilterPolicy (like NewBloomFilterPolicy) that does not ignore
 // trailing spaces in keys.
-extern const FilterPolicy* NewBloomFilterPolicy(int bits_per_key);
-
+extern const FilterPolicy* NewBloomFilterPolicy(int bits_per_key,
+    bool use_block_based_builder = true);
 }
 
 #endif  // STORAGE_ROCKSDB_INCLUDE_FILTER_POLICY_H_

include/rocksdb/flush_block_policy.h

@@ -6,6 +6,7 @@
 #pragma once
 
 #include <string>
+#include "rocksdb/table.h"
 
 namespace rocksdb {
 
@@ -37,7 +38,8 @@ class FlushBlockPolicyFactory {
   // Callers must delete the result after any database that is using the
   // result has been closed.
   virtual FlushBlockPolicy* NewFlushBlockPolicy(
-      const Options& options, const BlockBuilder& data_block_builder) const = 0;
+      const BlockBasedTableOptions& table_options,
+      const BlockBuilder& data_block_builder) const = 0;
 
   virtual ~FlushBlockPolicyFactory() { }
 };
@@ -51,7 +53,7 @@ class FlushBlockBySizePolicyFactory : public FlushBlockPolicyFactory {
   }
 
   virtual FlushBlockPolicy* NewFlushBlockPolicy(
-      const Options& options,
+      const BlockBasedTableOptions& table_options,
       const BlockBuilder& data_block_builder) const override;
 };
 

include/rocksdb/immutable_options.h

@@ -0,0 +1,87 @@
+// Copyright (c) 2013, 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 <vector>
+#include "rocksdb/options.h"
+
+namespace rocksdb {
+
+// ImmutableCFOptions is a data struct used by RocksDB internal. It contains a
+// subset of Options that should not be changed during the entire lifetime
+// of DB. You shouldn't need to access this data structure unless you are
+// implementing a new TableFactory. Raw pointers defined in this struct do
+// not have ownership to the data they point to. Options contains shared_ptr
+// to these data.
+struct ImmutableCFOptions {
+  explicit ImmutableCFOptions(const Options& options);
+
+  CompactionStyle compaction_style;
+
+  CompactionOptionsUniversal compaction_options_universal;
+  CompactionOptionsFIFO compaction_options_fifo;
+
+  const SliceTransform* prefix_extractor;
+
+  const Comparator* comparator;
+
+  MergeOperator* merge_operator;
+
+  const CompactionFilter* compaction_filter;
+
+  CompactionFilterFactory* compaction_filter_factory;
+
+  CompactionFilterFactoryV2* compaction_filter_factory_v2;
+
+  Logger* info_log;
+
+  Statistics* statistics;
+
+  InfoLogLevel info_log_level;
+
+  Env* env;
+
+  // Allow the OS to mmap file for reading sst tables. Default: false
+  bool allow_mmap_reads;
+
+  // Allow the OS to mmap file for writing. Default: false
+  bool allow_mmap_writes;
+
+  std::vector<DbPath> db_paths;
+
+  MemTableRepFactory* memtable_factory;
+
+  TableFactory* table_factory;
+
+  Options::TablePropertiesCollectorFactories
+    table_properties_collector_factories;
+
+  bool advise_random_on_open;
+
+  // This options is required by PlainTableReader. May need to move it
+  // to PlainTalbeOptions just like bloom_bits_per_key
+  uint32_t bloom_locality;
+
+  bool purge_redundant_kvs_while_flush;
+
+  uint32_t min_partial_merge_operands;
+
+  bool disable_data_sync;
+
+  bool use_fsync;
+
+  CompressionType compression;
+
+  std::vector<CompressionType> compression_per_level;
+
+  CompressionOptions compression_opts;
+
+  Options::AccessHint access_hint_on_compaction_start;
+
+  int num_levels;
+};
+
+}  // namespace rocksdb

include/rocksdb/iostats_context.h

@@ -27,7 +27,9 @@ struct IOStatsContext {
   uint64_t bytes_read;
 };
 
+#ifndef IOS_CROSS_COMPILE
 extern __thread IOStatsContext iostats_context;
+#endif  // IOS_CROSS_COMPILE
 
 }  // namespace rocksdb
 

include/rocksdb/memtablerep.h

@@ -186,12 +186,23 @@ class MemTableRepFactory {
 };
 
 // This uses a skip list to store keys. It is the default.
+//
+// Parameters:
+//   lookahead: If non-zero, each iterator's seek operation will start the
+//     search from the previously visited record (doing at most 'lookahead'
+//     steps). This is an optimization for the access pattern including many
+//     seeks with consecutive keys.
 class SkipListFactory : public MemTableRepFactory {
  public:
+  explicit SkipListFactory(size_t lookahead = 0) : lookahead_(lookahead) {}
+
   virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator&,
                                          Arena*, const SliceTransform*,
                                          Logger* logger) override;
   virtual const char* Name() const override { return "SkipListFactory"; }
+
+ private:
+  const size_t lookahead_;
 };
 
 #ifndef ROCKSDB_LITE

include/rocksdb/options.h

@@ -14,6 +14,7 @@
 #include <memory>
 #include <vector>
 #include <stdint.h>
+#include <unordered_map>
 
 #include "rocksdb/version.h"
 #include "rocksdb/universal_compaction.h"
@@ -39,7 +40,6 @@ class Slice;
 class SliceTransform;
 class Statistics;
 class InternalKeyComparator;
-class RateLimiter;
 
 // DB contents are stored in a set of blocks, each of which holds a
 // sequence of key,value pairs.  Each block may be compressed before
@@ -58,6 +58,7 @@ enum CompactionStyle : char {
   kCompactionStyleFIFO = 0x2,       // FIFO compaction style
 };
 
+
 struct CompactionOptionsFIFO {
   // once the total sum of table files reaches this, we will delete the oldest
   // table file
@@ -83,6 +84,14 @@ enum UpdateStatus {    // Return status For inplace update callback
   UPDATED         = 2, // No inplace update. Merged value set
 };
 
+struct DbPath {
+  std::string path;
+  uint64_t target_size;  // Target size of total files under the path, in byte.
+
+  DbPath() : target_size(0) {}
+  DbPath(const std::string& p, uint64_t t) : path(p), target_size(t) {}
+};
+
 struct Options;
 
 struct ColumnFamilyOptions {
@@ -90,7 +99,8 @@ struct ColumnFamilyOptions {
 
   // Use this if you don't need to keep the data sorted, i.e. you'll never use
   // an iterator, only Put() and Get() API calls
-  ColumnFamilyOptions* OptimizeForPointLookup();
+  ColumnFamilyOptions* OptimizeForPointLookup(
+      uint64_t block_cache_size_mb);
 
   // Default values for some parameters in ColumnFamilyOptions are not
   // optimized for heavy workloads and big datasets, which means you might
@@ -199,34 +209,6 @@ struct ColumnFamilyOptions {
   // individual write buffers.  Default: 1
   int min_write_buffer_number_to_merge;
 
-  // Control over blocks (user data is stored in a set of blocks, and
-  // a block is the unit of reading from disk).
-
-  // If non-NULL use the specified cache for blocks.
-  // If NULL, rocksdb will automatically create and use an 8MB internal cache.
-  // Default: nullptr
-  std::shared_ptr<Cache> block_cache;
-
-  // If non-NULL use the specified cache for compressed blocks.
-  // If NULL, rocksdb will not use a compressed block cache.
-  // Default: nullptr
-  std::shared_ptr<Cache> block_cache_compressed;
-
-  // Approximate size of user data packed per block.  Note that the
-  // block size specified here corresponds to uncompressed data.  The
-  // actual size of the unit read from disk may be smaller if
-  // compression is enabled.  This parameter can be changed dynamically.
-  //
-  // Default: 4K
-  size_t block_size;
-
-  // Number of keys between restart points for delta encoding of keys.
-  // This parameter can be changed dynamically.  Most clients should
-  // leave this parameter alone.
-  //
-  // Default: 16
-  int block_restart_interval;
-
   // Compress blocks using the specified compression algorithm.  This
   // parameter can be changed dynamically.
   //
@@ -244,29 +226,17 @@ struct ColumnFamilyOptions {
   CompressionType compression;
 
   // Different levels can have different compression policies. There
-  // are cases where most lower levels would like to quick compression
-  // algorithm while the higher levels (which have more data) use
+  // are cases where most lower levels would like to use quick compression
+  // algorithms while the higher levels (which have more data) use
   // compression algorithms that have better compression but could
-  // be slower. This array, if non nullptr, should have an entry for
-  // each level of the database. This array, if non nullptr, overides the
-  // value specified in the previous field 'compression'. The caller is
-  // reponsible for allocating memory and initializing the values in it
-  // before invoking Open(). The caller is responsible for freeing this
-  // array and it could be freed anytime after the return from Open().
-  // This could have been a std::vector but that makes the equivalent
-  // java/C api hard to construct.
+  // be slower. This array, if non-empty, should have an entry for
+  // each level of the database; these override the value specified in
+  // the previous field 'compression'.
   std::vector<CompressionType> compression_per_level;
 
   // different options for compression algorithms
   CompressionOptions compression_opts;
 
-  // If non-nullptr, use the specified filter policy to reduce disk reads.
-  // Many applications will benefit from passing the result of
-  // NewBloomFilterPolicy() here.
-  //
-  // Default: nullptr
-  const FilterPolicy* filter_policy;
-
   // If non-nullptr, use the specified function to determine the
   // prefixes for keys.  These prefixes will be placed in the filter.
   // Depending on the workload, this can reduce the number of read-IOP
@@ -283,12 +253,6 @@ struct ColumnFamilyOptions {
   // Default: nullptr
   std::shared_ptr<const SliceTransform> prefix_extractor;
 
-  // If true, place whole keys in the filter (not just prefixes).
-  // This must generally be true for gets to be efficient.
-  //
-  // Default: true
-  bool whole_key_filtering;
-
   // Number of levels for this database
   int num_levels;
 
@@ -324,7 +288,7 @@ struct ColumnFamilyOptions {
   // and each file on level-3 will be 200MB.
 
   // by default target_file_size_base is 2MB.
-  int target_file_size_base;
+  uint64_t target_file_size_base;
   // by default target_file_size_multiplier is 1, which means
   // by default files in different levels will have similar size.
   int target_file_size_multiplier;
@@ -368,18 +332,6 @@ struct ColumnFamilyOptions {
   // stop building a single file in a level->level+1 compaction.
   int max_grandparent_overlap_factor;
 
-  // We decided to remove seek compaction from RocksDB because:
-  // 1) It makes more sense for spinning disk workloads, while RocksDB is
-  // primarily designed for flash and memory,
-  // 2) It added some complexity to the important code-paths,
-  // 3) None of our internal customers were really using it.
-  //
-  // Since we removed seek compaction, this option is now obsolete.
-  // We left it here for backwards compatiblity (otherwise it would break the
-  // build), but we'll remove it at some point.
-  // Default: true
-  bool disable_seek_compaction;
-
   // Puts are delayed 0-1 ms when any level has a compaction score that exceeds
   // soft_rate_limit. This is ignored when == 0.0.
   // CONSTRAINT: soft_rate_limit <= hard_rate_limit. If this constraint does not
@@ -392,17 +344,9 @@ struct ColumnFamilyOptions {
   // Default: 0 (disabled)
   double hard_rate_limit;
 
-  // Max time a put will be stalled when hard_rate_limit is enforced. If 0, then
-  // there is no limit.
-  // Default: 1000
+  // DEPRECATED -- this options is no longer used
   unsigned int rate_limit_delay_max_milliseconds;
 
-  // Disable block cache. If this is set to true,
-  // then no block cache should be used, and the block_cache should
-  // point to a nullptr object.
-  // Default: false
-  bool no_block_cache;
-
   // size of one block in arena memory allocation.
   // If <= 0, a proper value is automatically calculated (usually 1/10 of
   // writer_buffer_size).
@@ -426,14 +370,6 @@ struct ColumnFamilyOptions {
   // Default: true
   bool purge_redundant_kvs_while_flush;
 
-  // This is used to close a block before it reaches the configured
-  // 'block_size'. If the percentage of free space in the current block is less
-  // than this specified number and adding a new record to the block will
-  // exceed the configured block size, then this block will be closed and the
-  // new record will be written to the next block.
-  // Default is 10.
-  int block_size_deviation;
-
   // The compaction style. Default: kCompactionStyleLevel
   CompactionStyle compaction_style;
 
@@ -468,10 +404,24 @@ struct ColumnFamilyOptions {
   std::shared_ptr<MemTableRepFactory> memtable_factory;
 
   // This is a factory that provides TableFactory objects.
-  // Default: a factory that provides a default implementation of
-  // Table and TableBuilder.
+  // Default: a block-based table factory that provides a default
+  // implementation of TableBuilder and TableReader with default
+  // BlockBasedTableOptions.
   std::shared_ptr<TableFactory> table_factory;
 
+  // Block-based table related options are moved to BlockBasedTableOptions.
+  // Related options that were originally here but now moved include:
+  //   no_block_cache
+  //   block_cache
+  //   block_cache_compressed
+  //   block_size
+  //   block_size_deviation
+  //   block_restart_interval
+  //   filter_policy
+  //   whole_key_filtering
+  // If you'd like to customize some of these options, you will need to
+  // use NewBlockBasedTableFactory() to construct a new table factory.
+
   // This option allows user to to collect their own interested statistics of
   // the tables.
   // Default: empty vector -- no user-defined statistics collection will be
@@ -480,7 +430,9 @@ struct ColumnFamilyOptions {
       TablePropertiesCollectorFactories;
   TablePropertiesCollectorFactories table_properties_collector_factories;
 
-  // Allows thread-safe inplace updates.
+  // Allows thread-safe inplace updates. If this is true, there is no way to
+  // achieve point-in-time consistency using snapshot or iterator (assuming
+  // concurrent updates).
   // If inplace_callback function is not set,
   //   Put(key, new_value) will update inplace the existing_value iff
   //   * key exists in current memtable
@@ -627,6 +579,7 @@ struct DBOptions {
 
   // Use to control write rate of flush and compaction. Flush has higher
   // priority than compaction. Rate limiting is disabled if nullptr.
+  // If rate limiter is enabled, bytes_per_sync is set to 1MB by default.
   // Default: nullptr
   std::shared_ptr<RateLimiter> rate_limiter;
 
@@ -659,7 +612,7 @@ struct DBOptions {
   // it does not use any locks to prevent concurrent updates.
   std::shared_ptr<Statistics> statistics;
 
-  // If true, then the contents of data files are not synced
+  // If true, then the contents of manifest and data files are not synced
   // to stable storage. Their contents remain in the OS buffers till the
   // OS decides to flush them. This option is good for bulk-loading
   // of data. Once the bulk-loading is complete, please issue a
@@ -674,18 +627,31 @@ struct DBOptions {
   // Default: false
   bool use_fsync;
 
-  // This number controls how often a new scribe log about
-  // db deploy stats is written out.
-  // -1 indicates no logging at all.
-  // Default value is 1800 (half an hour).
-  int db_stats_log_interval;
-
-  // A list paths where SST files can be put into. A compaction style can
-  // determine which of those paths it will put the file to.
+  // A list of paths where SST files can be put into, with its target size.
+  // Newer data is placed into paths specified earlier in the vector while
+  // older data gradually moves to paths specified later in the vector.
+  //
+  // For example, you have a flash device with 10GB allocated for the DB,
+  // as well as a hard drive of 2TB, you should config it to be:
+  //   [{"/flash_path", 10GB}, {"/hard_drive", 2TB}]
+  //
+  // The system will try to guarantee data under each path is close to but
+  // not larger than the target size. But current and future file sizes used
+  // by determining where to place a file are based on best-effort estimation,
+  // which means there is a chance that the actual size under the directory
+  // is slightly more than target size under some workloads. User should give
+  // some buffer room for those cases.
+  //
+  // If none of the paths has sufficient room to place a file, the file will
+  // be placed to the last path anyway, despite to the target size.
+  //
+  // Placing newer data to ealier paths is also best-efforts. User should
+  // expect user files to be placed in higher levels in some extreme cases.
+  //
   // If left empty, only one path will be used, which is db_name passed when
   // opening the DB.
   // Default: empty
-  std::vector<std::string> db_paths;
+  std::vector<DbPath> db_paths;
 
   // This specifies the info LOG dir.
   // If it is empty, the log files will be in the same dir as data.
@@ -818,12 +784,13 @@ struct DBOptions {
   // Specify the file access pattern once a compaction is started.
   // It will be applied to all input files of a compaction.
   // Default: NORMAL
-  enum {
-    NONE,
-    NORMAL,
-    SEQUENTIAL,
-    WILLNEED
-  } access_hint_on_compaction_start;
+  enum AccessHint {
+      NONE,
+      NORMAL,
+      SEQUENTIAL,
+      WILLNEED
+  };
+  AccessHint access_hint_on_compaction_start;
 
   // Use adaptive mutex, which spins in the user space before resorting
   // to kernel. This could reduce context switch when the mutex is not
@@ -832,22 +799,22 @@ struct DBOptions {
   // Default: false
   bool use_adaptive_mutex;
 
-  // Allows OS to incrementally sync files to disk while they are being
-  // written, asynchronously, in the background.
-  // Issue one request for every bytes_per_sync written. 0 turns it off.
-  // Default: 0
-  uint64_t bytes_per_sync;
-
-  // Allow RocksDB to use thread local storage to optimize performance.
-  // Default: true
-  bool allow_thread_local;
-
   // Create DBOptions with default values for all fields
   DBOptions();
   // Create DBOptions from Options
   explicit DBOptions(const Options& options);
 
   void Dump(Logger* log) const;
+
+  // Allows OS to incrementally sync files to disk while they are being
+  // written, asynchronously, in the background.
+  // Issue one request for every bytes_per_sync written. 0 turns it off.
+  // Default: 0
+  //
+  // You may consider using rate_limiter to regulate write rate to device.
+  // When rate limiter is enabled, it automatically enables bytes_per_sync
+  // to 1MB.
+  uint64_t bytes_per_sync;
 };
 
 // Options to control the behavior of a database (passed to DB::Open)
@@ -928,6 +895,18 @@ struct ReadOptions {
   // ! DEPRECATED
   // const Slice* prefix;
 
+  // "iterate_upper_bound" defines the extent upto which the forward iterator
+  // can returns entries. Once the bound is reached, Valid() will be false.
+  // "iterate_upper_bound" is exclusive ie the bound value is
+  // not a valid entry.  If iterator_extractor is not null, the Seek target
+  // and iterator_upper_bound need to have the same prefix.
+  // This is because ordering is not guaranteed outside of prefix domain.
+  // There is no lower bound on the iterator. If needed, that can be easily
+  // implemented
+  //
+  // Default: nullptr
+  const Slice* iterate_upper_bound;
+
   // Specify if this read request should process data that ALREADY
   // resides on a particular cache. If the required data is not
   // found at the specified cache, then Status::Incomplete is returned.
@@ -942,18 +921,27 @@ struct ReadOptions {
   // Not supported in ROCKSDB_LITE mode!
   bool tailing;
 
+  // Enable a total order seek regardless of index format (e.g. hash index)
+  // used in the table. Some table format (e.g. plain table) may not support
+  // this option.
+  bool total_order_seek;
+
   ReadOptions()
       : verify_checksums(true),
         fill_cache(true),
         snapshot(nullptr),
+        iterate_upper_bound(nullptr),
         read_tier(kReadAllTier),
-        tailing(false) {}
+        tailing(false),
+        total_order_seek(false) {}
   ReadOptions(bool cksum, bool cache)
       : verify_checksums(cksum),
         fill_cache(cache),
         snapshot(nullptr),
+        iterate_upper_bound(nullptr),
         read_tier(kReadAllTier),
-        tailing(false) {}
+        tailing(false),
+        total_order_seek(false) {}
 };
 
 // Options that control write operations
@@ -991,7 +979,17 @@ struct WriteOptions {
   // Default: 0
   uint64_t timeout_hint_us;
 
-  WriteOptions() : sync(false), disableWAL(false), timeout_hint_us(0) {}
+  // If true and if user is trying to write to column families that don't exist
+  // (they were dropped),  ignore the write (don't return an error). If there
+  // are multiple writes in a WriteBatch, other writes will succeed.
+  // Default: false
+  bool ignore_missing_column_families;
+
+  WriteOptions()
+      : sync(false),
+        disableWAL(false),
+        timeout_hint_us(0),
+        ignore_missing_column_families(false) {}
 };
 
 // Options that control flush operations
@@ -1003,20 +1001,21 @@ struct FlushOptions {
   FlushOptions() : wait(true) {}
 };
 
+// Get options based on some guidelines. Now only tune parameter based on
+// flush/compaction and fill default parameters for other parameters.
+// total_write_buffer_limit: budget for memory spent for mem tables
+// read_amplification_threshold: comfortable value of read amplification
+// write_amplification_threshold: comfortable value of write amplification.
+// target_db_size: estimated total DB size.
+extern Options GetOptions(size_t total_write_buffer_limit,
+                          int read_amplification_threshold = 8,
+                          int write_amplification_threshold = 32,
+                          uint64_t target_db_size = 68719476736 /* 64GB */);
 
-// Create a RateLimiter object, which can be shared among RocksDB instances to
-// control write rate of flush and compaction.
-// @rate_bytes_per_sec: desired total write rate in bytes per second.
-// @refill_period_us: token refill interval in micro-second.
-// @fairness: RateLimiter accepts high-pri requests and low-pri requests.
-// low-pri request is usually blocked in favor of hi-pri request. To prevent
-// low-pri request from being blocked for too long, it can get processed first
-// by 1/fairness chance.
-extern RateLimiter* NewRateLimiter(
-    int64_t rate_bytes_per_sec,
-    int64_t refill_period_us = 100 * 1000,
-    int32_t fairness = 10);
-
+bool GetOptionsFromStrings(
+    const Options& base_options,
+    const std::unordered_map<std::string, std::string>& options_map,
+    Options* new_options);
 
 }  // namespace rocksdb